arXiv Daily Digest

1099

Papers

RoboHarness: A Simple Harness Outperforms VLA and World Actions Models

RoboHarness：一种简单的机器人束带超越了VLA和世界动作模型

Qu, Bince, Chen, Wei, Zhang, Bo

Abstract

The rise of large language models (LLMs) has opened a path toward generalist embodied AI, raising the question of how LLMs can act in the physical world. One straightforward approach is training the model to generate actions: collect robot-action data and video demonstration, refine LLM architecture with an action head, and train a vision-language-action (VLA) model at scale. However, we argue that coding agents have the potential to realize generalist embodied manipulation. All they need is a simple yet effective robot harness. We present RoboHarness: a robotic harness that gives LLM agents a visual-geometric control panel so that they can directly understand and invoke embodied tasks. With RoboHarness, Qwen3.8-Flash-Next outperforms state-of-the-art VLA models on BEHAVIOR Challenge 2025 tasks. The implementation will be released as open source.

Chinese Translation

大型语言模型（LLMs）的兴起为通用化的具身人工智能开辟了道路，提出了LLMs如何在物理世界中行动的问题。一种直接的方法是训练模型生成动作：收集机器人动作数据和视频演示，优化LLM架构并添加动作头，并大规模训练视觉-语言-动作（VLA）模型。然而，我们认为编码代理有潜力实现通用的具身操控。它们所需的只是一个简单而有效的机器人束带。我们提出了RoboHarness：一种机器人束带，为LLM代理提供了一个视觉-几何控制面板，使它们能够直接理解和调用具身任务。使用RoboHarness，Qwen3.8-Flash-Next在BEHAVIOR Challenge 2025任务中超越了最先进的VLA模型。该实现将作为开源发布。

View on arXiv Download PDF AI Translation

cs.RO / 2 / 2610.14020

LabInstruct: Benchmarking Situated Instructional Video Generation for Lab Procedures

LabInstruct：实验室程序的情境教学视频生成基准测试

Fu, Yuming, Wu, Weijia, Chen, Jing, Tang, Jiahao, Chen, Feifei, Zhu, Hongyu, Jin, Xin, Wang, Alex Jinpeng

Abstract

Self-driving laboratories (SDLs) aim to automate the full experimental loop, from scientific decision-making to physical execution. Ideally, AI-generated plans could be carried out directly by robotic systems, but reliable automation remains difficult in complex, open-world laboratory environments, where experiments often involve fine-grained manipulation, long-horizon procedures, and substantial variation across tasks and setups. Humans therefore remain an important execution interface between AI-generated plans and physical experiments, creating a need for clear and effective human-facing experimental guidance. Because laboratory procedures are inherently visual, spatial, and dynamic, video is particularly well suited to communicating apparatus configurations, manipulation actions, temporal dependencies, and state changes. Recent advances in video generation now make it possible to synthesize experimental demonstrations directly from an initial workspace image and a natural-language instruction. However, whether such models can reliably communicate real laboratory procedures has not been systematically studied. We introduce LabInstruct, a benchmark for situated instructional video generation in real laboratories. LabInstruct contains 204 tasks across 5 scientific disciplines, with real reference executions and structured annotations of objects, actions, contacts, and state transitions. Evaluating 8 frontier image-to-video models, we find that visually plausible generations frequently remain procedurally incorrect, revealing a substantial gap between visual realism and the reliability required for experimental instruction.

Chinese Translation

自驱动实验室（SDL）旨在自动化完整的实验循环，从科学决策到物理执行。理想情况下，AI生成的计划可以由机器人系统直接执行，但在复杂的开放世界实验室环境中，可靠的自动化仍然困难，因为实验通常涉及细致的操作、长时间的程序以及任务和设置之间的显著变化。因此，人类仍然是AI生成计划与物理实验之间的重要执行接口，这就需要清晰有效的人机实验指导。由于实验室程序本质上是视觉的、空间的和动态的，视频特别适合于传达设备配置、操作动作、时间依赖性和状态变化。最近在视频生成方面的进展使得可以直接从初始工作区图像和自然语言指令合成实验演示。然而，这些模型是否能够可靠地传达真实的实验室程序尚未系统地研究。我们介绍了LabInstruct，这是一个针对真实实验室的情境教学视频生成的基准测试。LabInstruct包含来自5个科学学科的204个任务，具有真实的参考执行和对象、动作、接触及状态转变的结构化注释。在评估8个前沿的图像到视频模型时，我们发现视觉上合理的生成往往在程序上不正确，揭示了视觉现实主义与实验指导所需可靠性之间的显著差距。

View on arXiv Download PDF AI Translation

cs.RO / 3 / 2610.14033

OIL: An Object Interaction Layer for Humanoid Whole-Body Control

OIL：一种用于类人机器人全身控制的对象交互层

Dugar, Pranay, Gadde, Mohitvishnu S., Mangannavar, Rajesh, Shrestha, Aayam Kumar, Fern, Alan

Abstract

Recent progress in learned whole-body controllers for humanoid robots has enabled robust free-space motion by tracking kinematic targets such as joint or end-effector trajectories. This interface provides a useful abstraction for upstream systems, including motion capture, teleoperation, or generative models, which can specify desired motion without reasoning about forces or balance. However, this abstraction has largely been limited to robot motion, requiring task-specific controllers or retraining for object interaction. We introduce OIL, an Object Interaction Layer that extends the kinematic command interface of whole-body control to include desired object motion and contact. OIL augments a standard controller with object-centric commands and translates them into coordinated whole-body behavior while preserving locomotion and balance. This design enables a single learned policy to execute non-prehensile interactions, including pushing, pulling, two-arm pick-and-place, and locomotion-coupled transport, directly from object-level kinematic targets without task-specific retraining. In simulation, we demonstrate that object-level commands improve performance over robot-centric controllers while matching task-specific policies across variations in object properties such as mass and size. We further demonstrate zero-shot generalization to novel behavior combinations and sim-to-real transfer on the Digit humanoid robot.

Chinese Translation

近年来，针对类人机器人的学习型全身控制器的进展，使得通过跟踪关节或末端执行器轨迹等运动学目标实现稳健的自由空间运动成为可能。该接口为上游系统（包括运动捕捉、遥操作或生成模型）提供了一种有用的抽象，这些系统可以在不考虑力或平衡的情况下指定期望的运动。然而，这种抽象在很大程度上仅限于机器人运动，要求针对特定任务的控制器或重新训练以实现对象交互。我们提出了OIL（对象交互层），它扩展了全身控制的运动学指令接口，以包括期望的对象运动和接触。OIL通过对象中心的指令增强标准控制器，并将其转化为协调的全身行为，同时保持运动和稳定性。这种设计使得单一学习策略能够直接从对象级运动学目标执行非抓取交互，包括推、拉、双臂拾取和放置以及与运动结合的运输，而无需针对特定任务的重新训练。在仿真中，我们证明了对象级指令在性能上优于以机器人为中心的控制器，同时在对象属性（如质量和大小）变化下与特定任务策略相匹配。我们进一步展示了对新行为组合的零样本泛化以及在Digit类人机器人上的仿真到现实转移。

View on arXiv Download PDF AI Translation

cs.RO / 4 / 2610.14034

Building Trustworthy Household Robots through Open-World Safety

通过开放世界安全构建可信赖的家庭机器人

Tsai, Yu-Lin, Lyu, Muxi, Herzig, Roei, Sen, Koushik, Popa, Raluca Ada, Wagner, David

Abstract

Foundation models are expanding robot capabilities, enabling robots to handle more diverse objects, follow sophisticated instructions, and operate in complex environments. These advances bring us to a future in which robots become part of everyday household life. However, cluttered, dynamic household environments also expose robots to unfamiliar objects, tasks, and interactions, making reliable zero-shot task completion an ongoing challenge for robotics research. From a security perspective, we argue for open-world safety: household robots should preserve specified safety requirements in unseen situations, including when they cannot complete the requested task. Safety protections must account for context and interactions that persist across actions and tasks. In this position paper, we examine why existing safety methods can fail to provide this protection. Protection can fail through misinterpreted context, overlooked past interactions, or stopping and recovery strategies that become unsafe under unfamiliar conditions. We therefore propose a research agenda to specify, learn, and ground safety requirements, enforce them with reliable recovery, and evaluate the generalization of such protections. We call for household robots whose safety protections remain reliable beyond the situations, context, and history used to train and develop them, to achieve open-world safety.

Chinese Translation

基础模型正在扩展机器人的能力，使其能够处理更多样化的物体、遵循复杂的指令，并在复杂的环境中操作。这些进展使我们走向一个机器人成为日常家庭生活一部分的未来。然而，杂乱且动态的家庭环境也使机器人面临不熟悉的物体、任务和互动，这使得可靠的零样本任务完成成为机器人研究中的一个持续挑战。从安全的角度来看，我们主张开放世界安全：家庭机器人应在未见过的情况下保持特定的安全要求，包括在无法完成请求任务时。安全保护必须考虑到在行动和任务之间持续存在的上下文和互动。在这篇立场论文中，我们探讨了现有安全方法为何可能无法提供这种保护。保护可能因误解上下文、忽视过去的互动或在不熟悉条件下变得不安全的停止和恢复策略而失败。因此，我们提出了一项研究议程，以明确、学习和落实安全要求，通过可靠的恢复来执行这些要求，并评估此类保护的泛化能力。我们呼吁开发家庭机器人，使其安全保护在超出用于训练和开发的情境、上下文和历史的情况下仍然可靠，以实现开放世界安全。

View on arXiv Download PDF AI Translation

cs.RO / 5 / 2610.14035

Residual Compliant Policy: Learning Compliant Policies from Few Demonstrations

残余顺应策略：从少量示范中学习顺应策略

Jiang, Chunyu, Cai, Yichen, Li, Shuai, Sun, Wenbin, Carvalho, João, Daniilidis, Kostas, Arenz, Oleg, de Farias, Cristiana, Peters, Jan

Abstract

Contact-rich manipulation requires a robot to be precise in free space to establish the intended contacts, and compliant once in contact to avoid excessive forces. Imitation learning has proven effective for simpler manipulation tasks, but the compliance underlying a demonstration is not directly observable and is therefore difficult to infer from demonstration data alone. We present a framework that makes compliance explicit at every stage. During data collection, the operator teleoperates the real robot while a synchronized digital twin visualizes the error between the commanded reference and the achieved pose, making the current compliance visible, and selects between two discrete compliance modes. To generalize across object poses, we combine Laplacian editing, which retargets trajectories efficiently but is not physics-consistent during contact, with simulation-based trajectory optimization, which enforces physical plausibility but offers no manual guidance. Finally, a residual reinforcement learning policy refines the interaction forces by adjusting the reference pose of the underlying admittance controller. We evaluate on three flipping and pushing tasks on a Franka FR3 based on a total of 7 demonstrations across tasks. Our method is successful on all 40 out-of-distribution trials, compared to 33/40 and 24/40 for two baselines, with our base policy alone reaching 35/40.

Chinese Translation

接触丰富的操作要求机器人在自由空间中精确定位以建立预期的接触，并在接触后保持顺应以避免过大的力。模仿学习在较简单的操作任务中已被证明有效，但示范中所蕴含的顺应性并不可直接观察，因此仅从示范数据中推断顺应性是困难的。我们提出了一个框架，使得顺应性在每个阶段都变得明确。在数据收集过程中，操作员通过遥控真实机器人，同时一个同步的数字双胞胎可视化命令参考与实际姿态之间的误差，使当前的顺应性可见，并在两种离散的顺应模式之间进行选择。为了在物体姿态之间进行泛化，我们结合了拉普拉斯编辑（Laplacian editing），该方法能够高效地重新定向轨迹但在接触时不符合物理规律，以及基于仿真的轨迹优化，该方法强制执行物理合理性但不提供手动指导。最后，残余强化学习策略通过调整基础顺应控制器的参考姿态来优化交互力。我们在基于Franka FR3的三个翻转和推送任务上进行了评估，总共使用了7个示范。与两个基线方法相比，我们的方法在所有40个超出分布的试验中均取得成功，基线方法的成功率分别为33/40和24/40，而我们的基础策略单独达到了35/40。

View on arXiv Download PDF AI Translation

cs.RO / 6 / 2610.14038

From bits to atoms: A survey of cross-layer safety in Embodied AI

从比特到原子：具身人工智能跨层安全的调查

Feng, Tuo, Zhang, Yue, Zhou, Shiji, Fan, Zhaoxin, Wang, Wenguan, Chi, Xiaowei, Shen, Tianyu, Li, Xiangtai, Bai, Xuesong, Zhu, Zheng, Han, Sirui, Zhou, Boyu, Wang, Zhaosong, Chen, Shizhe, Zheng, Zilong, Bai, Chenjia, Pang, Jiangmiao, Yang, Yaodong, Zhao, Hao, Hu, Di, Yi, Li, Huang, Yan, Wu, Qi, Zhang, Hanwang, Gao, Fei, Zhang, Shanghang, Li, Hongsheng, Hao, Jianye, Zhang, Yuan, Xia, Yingjie, Shen, Jian, Lu, Cewu, Zhang, Hui, Chen, Kai, Yu, Wenwu, Chen, Badong, Han, Jungong, Guo, Yike

Abstract

Embodied Artificial Intelligence (AI) introduces cyber-physical risks because digital failures and attacks can cause physical harm. This survey organizes embodied AI safety through a four-layer reference architecture -- comprising the physical interaction, system and middleware, algorithm and decision, and cloud-edge and data layers -- and uses it to connect vulnerabilities, propagation paths, and safeguards across the system stack. It traces the evolution of safety paradigms and examines how faults and attacks cross architectural boundaries to produce physical consequences. Furthermore, the survey systematizes composite attack chains, outlines multi-layered defense strategies, and reviews representative embodied safety benchmarks. Ultimately, we advocate for unifying software alignment with rigorous hardware constraints to build trustworthy systems capable of operating safely within human-centric environments.

Chinese Translation

具身人工智能（AI）引入了网络物理风险，因为数字故障和攻击可能导致物理伤害。本调查通过一个四层参考架构组织具身AI的安全性——包括物理交互、系统与中间件、算法与决策，以及云边缘与数据层——并利用该架构连接系统堆栈中的脆弱性、传播路径和防护措施。它追溯了安全范式的演变，并考察了故障和攻击如何跨越架构边界产生物理后果。此外，本调查系统化了复合攻击链，概述了多层防御策略，并回顾了具有代表性的具身安全基准。最终，我们倡导将软件对齐与严格的硬件约束相结合，以构建能够在以人为本的环境中安全运行的可信系统。

View on arXiv Download PDF AI Translation

cs.RO / 7 / 2610.14039

RAIN: Robotic Task Generalization with Region-Aware Interaction Networks

RAIN：基于区域感知交互网络的机器人任务泛化

Lee, Seunghun, Shim, Kyuhwan, Song, Yeogyeong, Kim, Seung Hwan, Lee, Sihaeng, Zhang, Byoung-Tak, Im, Sunghoon, Lee, Gunhee

Abstract

Existing vision-language-action (VLA) models benefit from large-scale vision-language pretraining, but tend to memorize training trajectories, limiting their ability to adapt learned interactions to new task and scene configurations. To address this problem, we propose \textbf{Region-Aware Interaction Networks (RAIN)}, which formulates a task as a sequence of goal-conditioned subtasks, each specified by an action type and a target-region mask. This formulation allows learned skills to be reused across different objects and scene layouts and composed into task sequences not seen during training. Its Target-adaptive Cross-View Encoder (TCE) conditions multi-view features on target masks and exchanges information across views, while PlanDiT predicts target-directed trajectories and actions. A Transition Head estimates subtask completion to enable automatic transitions. We also introduce a \textbf{Reference-based Retargeting Strategy (RRS)}, which uses recorded trajectories to synthesize smooth approaches to reference interactions with alternative target regions, without collecting additional data. Experiments on \textbf{LIBERO-Analogy}, our new benchmark for robotic task generalization, and real-world tabletop tasks demonstrate that RAIN generalizes more robustly than existing VLA baselines.

Chinese Translation

现有的视觉-语言-动作（VLA）模型受益于大规模的视觉-语言预训练，但往往倾向于记忆训练轨迹，从而限制了它们将学习到的交互适应于新任务和场景配置的能力。为了解决这个问题，我们提出了 extbf{区域感知交互网络（RAIN）}，将任务表述为一系列以目标为条件的子任务，每个子任务由一个动作类型和一个目标区域掩码指定。这种表述方式允许在不同的物体和场景布局中重用学习到的技能，并将其组合成在训练期间未见过的任务序列。其目标自适应跨视图编码器（TCE）根据目标掩码对多视图特征进行条件处理，并在视图之间交换信息，而PlanDiT则预测目标导向的轨迹和动作。过渡头（Transition Head）估计子任务的完成情况，以实现自动过渡。我们还引入了一种 extbf{基于参考的重定向策略（RRS）}，该策略利用记录的轨迹合成平滑的方法，以实现与替代目标区域的参考交互，而无需收集额外的数据。在我们的新基准 extbf{LIBERO-Analogy}和现实世界的桌面任务上的实验表明，RAIN的泛化能力优于现有的VLA基线。

View on arXiv Download PDF AI Translation

cs.RO / 8 / 2610.14042

Reinforcement Learning from Real-Robot Interaction without High-Quality State Estimates

无需高质量状态估计的真实机器人交互强化学习

Gao, Ryan, Anilkumar, Ashvin, Corrado, Nicholas, Cong, William, Hanna, Josiah

Abstract

A longstanding goal for robot reinforcement learning (RL) is that robots learn from their own physical experience, as humans and animals do. A key obstacle to learning directly from physical experience, especially outside instrumented lab environments, is that the state of the world is usually unobservable to the learner and must be estimated from on-board sensing. Errors in that estimate corrupt both the policy input and potentially the reward signal. We study this challenge using the contact-rich task of a bipedal robot learning to score soccer goals. We estimate all key state variables on-board with an unscented Kalman filter (UKF) and no external instrumentation (e.g., motion capture or overhead cameras). Because the UKF may incorrectly estimate state when the ball is unobserved, we learn a new state representation with a recurrent state-space model (RSSM) that, critically, only corrects the RSSM's latent state when the ball is observed. Building on RLPD, we introduce RSSM-RLPD and apply it to the task of a NAO robot learning to dribble a soccer ball into a goal. We show that the robot learns to score from multiple starting positions after 30 minutes of real-world interaction, reaching 100% success against 0% for a feed-forward policy under the same training budget. We complement our physical robot experiments with a simulation study and show that RSSM-RLPD learns policies that achieve a higher success rate than behavior cloning, a feed-forward, and GRU baseline. Taken together, our results show that the addition of a learned state representation to existing on-robot learning methods enables RL from real-robot data even with a high degree of partial observability.

Chinese Translation

机器人强化学习（RL）的一个长期目标是让机器人像人类和动物一样从自身的物理经验中学习。直接从物理经验中学习的一个关键障碍是，世界的状态通常对学习者是不可观察的，必须通过机载传感器进行估计。该估计中的错误会影响策略输入，并可能影响奖励信号。我们通过研究一项双足机器人学习进球的接触丰富任务来探讨这一挑战。我们使用无外部仪器（例如运动捕捉或顶视摄像头）的无味卡尔曼滤波器（UKF）在机载上估计所有关键状态变量。由于当球未被观察时，UKF可能会错误估计状态，我们使用递归状态空间模型（RSSM）学习一种新的状态表示，关键在于仅在观察到球时纠正RSSM的潜在状态。在RLPD的基础上，我们引入了RSSM-RLPD，并将其应用于NAO机器人学习将足球运球进球的任务。我们展示了机器人在30分钟的真实世界交互后能够从多个起始位置学习进球，成功率达到100%，而在相同训练预算下，前馈策略的成功率为0%。我们通过模拟研究补充了物理机器人实验，显示RSSM-RLPD学习的策略比行为克隆、前馈和GRU基线实现了更高的成功率。综合来看，我们的结果表明，将学习的状态表示添加到现有的机器人学习方法中，使得即使在高度部分可观察的情况下，也能够从真实机器人数据中进行强化学习。

View on arXiv Download PDF AI Translation

cs.RO / 9 / 2610.14061

Terrain Traversal Without Motion Priors: Expert-Routed Humanoid Locomotion at Deployment Fidelity

无需运动先验的地形穿越：部署精度下的专家引导类人步态

Ly, Thanh, Le, Chien, Dang, Truong-Duy, Nguyen, Duy M. H., Le, Duy-Dung, Ngo, Vien Anh, Le, An T.

Abstract

State-of-the-art perceptive humanoid locomotion is mostly built on retargeted motion capture, a generative motion prior, and a distillation stage. We report a complete system that does not require this pipeline for a humanoid to cross varied terrain. In its place, we provide a five-expert mixture routed by a terrain classifier over a sensed elevation grid, and target every foothold with a model-based planner, refined by six closed-form-gradient steps on an interpolated terrain-cost field instead of picked from a fixed grid. Driven through the same perception-to-actuation software a robot would run, inside a second, deployment-fidelity simulator, it crosses an $85$\,m five-segment waypoint course $18$ of $20$ times end-to-end ($90\%$, Wilson $95\%$ CI $70$--$97$) and succeeds $20/20$ on four of five segments scored alone, where an ablated configuration without the mixture and without refinement reaches $0\%$ on both the stair and the hurdle segment. The continuous refinement moves $72.4\%$ of foot placements across at least one grid-cell boundary, which produces real sub-grid correction. We also ran matched component studies to report what does not account for it: one expert equals five on the training distribution, four label-free routers collapse to near-uniform activation, and a self-supervised terrain embedding loses to the privileged label.

Chinese Translation

最先进的感知类人步态主要基于重定向的动作捕捉、生成的动作先验和蒸馏阶段。我们报告了一个完整的系统，该系统不需要这一流程即可使类人机器人跨越多样的地形。取而代之的是，我们提供了一个由地形分类器引导的五个专家混合体，通过感知的高程网格，并使用基于模型的规划器针对每个足迹进行目标定位，该规划器通过在插值的地形成本场上进行六个封闭形式梯度步骤进行优化，而不是从固定网格中选择。该系统通过与机器人运行的相同感知到驱动软件，在一个部署精度的模拟器中，在一个 $85$ 米的五段航点课程中成功完成 $18$ 次中的 $20$ 次（$90 ext{ extperthousand}$，Wilson $95 ext{ extperthousand}$ 置信区间 $70$--$97$），并在单独评分的五段中的四段上成功率达到 $20/20$，而没有混合和优化的消融配置在楼梯和障碍段的成功率均为 $0 ext{ extperthousand}$。连续的优化使 $72.4 ext{ extperthousand}$ 的足部放置跨越至少一个网格单元边界，从而产生真实的子网格修正。我们还进行了匹配组件研究，以报告哪些因素不影响结果：一个专家在训练分布中相当于五个，四个无标签路由器接近均匀激活，自监督的地形嵌入不如特权标签。

View on arXiv Download PDF AI Translation

cs.RO / 10 / 2610.14064

Do My Samples Help? Compute Budget Allocation for Real-World Sampling-Based MPC in Contact-Rich Manipulation

我的样本有帮助吗？基于采样的真实世界接触丰富操控的计算预算分配

Dierking, Magnus, Carvalho, João, Kurtz, Vince, Peters, Jan, Dengler, Nils, Chalvatzaki, Georgia

Abstract

Sampling-based model predictive control (SMPC) is increasingly practical for contact-rich manipulation because GPU-based physics simulators enable real-time evaluation of large numbers of non-differentiable contact rollouts. However, this computation budget is finite and can be spent in competing ways, including denser control exploration, representing dynamics or state uncertainty, replanning more frequently, or aligning the simulation better with the physical state. Yet, the resulting trade-offs are not well understood in real-world sim-to-real control. In this work, we address this question through a systematic study of how to allocate a fixed computation budget for SMPC in contact-rich manipulation. To enable controlled comparisons, we develop a GPU-based framework named Budget-Allocating Model Predictive Control (BAMPC), which runs on MuJoCo Warp and exposes dynamics uncertainty, state uncertainty, and temporal alignment as independent dimensions of the computation budget while using the same control stack in simulation and on a real robot. Across three contact-rich manipulation tasks, we find that the best budget allocation depends on the dominant sim-to-real error, with dynamics randomization not universally the most effective one.

Chinese Translation

基于采样的模型预测控制（SMPC）在接触丰富的操控中越来越实用，因为基于GPU的物理模拟器能够实时评估大量不可微分的接触展开。然而，这一计算预算是有限的，可以通过多种方式进行支配，包括更密集的控制探索、表示动力学或状态不确定性、更频繁的重新规划，或更好地将模拟与物理状态对齐。然而，现实世界的模拟到现实控制中，这些权衡的结果尚不清楚。在本研究中，我们通过系统研究如何为接触丰富的操控分配固定的计算预算来解决这一问题。为了实现受控比较，我们开发了一个名为预算分配模型预测控制（BAMPC）的基于GPU的框架，该框架在MuJoCo Warp上运行，并将动力学不确定性、状态不确定性和时间对齐作为计算预算的独立维度，同时在模拟和真实机器人上使用相同的控制堆栈。在三个接触丰富的操控任务中，我们发现最佳预算分配取决于主导的模拟到现实误差，动力学随机化并不总是最有效的选择。

View on arXiv Download PDF AI Translation

cs.RO / 11 / 2610.14066

Safety-Oriented Trajectory Ranking from BEV Perception Outputs

基于安全导向的鸟瞰视角感知输出轨迹排名

Pohl, Leon, Frauendorfer, Martin, Schmitt, Corinna, Maehlisch, Mirko

Abstract

Bird's-Eye-View (BEV) representations are increasingly used as compact inputs for autonomous driving perception and planning. However, lightweight and interpretable methods for evaluating candidate ego trajectories directly on BEV outputs remain useful for vehicular safety workflows, rapid prototyping, and human-in-the-loop validation. This paper presents a modular trajectory ranking pipeline that consumes BEVFusion object detections and map segmentations derived from nuScenes sensor data and ranks manually specified candidate trajectories according to safety-first criteria. The ranking prioritizes avoiding collisions and walkway violations, then smaller end distance to the target, and finally path efficiency. To study whether these rule-based rankings resemble human judgment, we created seven BEV traffic scenarios and compared the script output with preferences from a survey of 21 participants. Across 146 pairwise votes, 124 matched the trajectory ranked first by the evaluator, corresponding to an overall agreement of 84.9 %. The results indicate that simple BEV-centric safety heuristics can approximate human preferences in constrained trajectory choice tasks while remaining transparent enough for analysis, dataset curation, and autonomous driving studies.

Chinese Translation

鸟瞰视角（BEV）表示法越来越多地被用作自动驾驶感知和规划的紧凑输入。然而，针对BEV输出直接评估候选自我轨迹的轻量级和可解释的方法仍然对车辆安全工作流程、快速原型设计和人机协同验证具有重要意义。本文提出了一种模块化轨迹排名管道，该管道使用来自nuScenes传感器数据的BEVFusion物体检测和地图分割，并根据安全优先标准对手动指定的候选轨迹进行排名。排名优先考虑避免碰撞和人行道违规，其次是与目标的较小最终距离，最后是路径效率。为了研究这些基于规则的排名是否类似于人类判断，我们创建了七个BEV交通场景，并将脚本输出与21名参与者的偏好进行了比较。在146对投票中，有124票与评估者排名第一的轨迹相匹配，总体一致性为84.9%。结果表明，简单的以BEV为中心的安全启发式方法可以在受限轨迹选择任务中近似人类偏好，同时足够透明以便于分析、数据集整理和自动驾驶研究。

View on arXiv Download PDF AI Translation

cs.RO / 12 / 2610.14067

From Hitter to Player: Separating Motor Execution and Strategy in Competitive Humanoid Table Tennis

从击球者到选手：在竞争性类人乒乓球中分离运动执行与策略

Dong, Fanrong, Sun, Ziyi, Han, Ziqi, Sun, Junhan, Ye, Lei, Jing, Zetong, Zhao, Hao, Liu, Shaoxun, Wang, Rongrong

Abstract

Competitive humanoid table tennis requires two distinct abilities: executing a desired return and deciding where to return the ball. We separate these two problems, with the ball landing target serving as an interface between motor control and strategy. (1) We first train a whole-body motor policy conditioned on the desired landing target, which maps the target to an executable humanoid stroke through a model-based strike planner and a learned controller. (2) We then improve the motor policy through self-play in closed-loop rallies, where the robot learns to handle incoming balls generated by another player rather than only prescribed incoming balls. At this stage, the landing targets are still sampled without considering the opponent, so the policy improves its execution without learning game strategy. (3) Finally, we freeze the entire motor stack and train a stroke-level strategic policy to choose landing targets through competitive self-play. Motor self-play improves both landing accuracy and legal-return rate during rallies. More importantly, with the entire motor stack fixed, changing only the target-selection policy leads to large differences in game outcomes: the learned strategy wins 89%, 82%, and 73% of points against center, random, and away-from-opponent baselines, respectively. The learned strategy also changes where it sends the ball according to the opponent's position and the previous target, showing that self-play learns more than simply aiming at fixed "good" locations. On hardware, the same frozen execution stack and strategic interface support complete 11-point games against a recreational human opponent, including multi-stroke rallies with both forehand and backhand returns.

Chinese Translation

竞争性类人乒乓球需要两种不同的能力：执行所需的回球和决定将球回到哪里。我们将这两个问题分开，球落点目标作为运动控制与策略之间的接口。(1) 我们首先训练一个全身运动策略，该策略以所需落点目标为条件，通过基于模型的击球规划器和学习控制器将目标映射到可执行的类人击球。(2) 然后，我们通过闭环对打中的自我对弈来改进运动策略，在这一过程中，机器人学习处理由另一个选手生成的来球，而不仅仅是预设的来球。在这个阶段，落点目标仍然是在不考虑对手的情况下进行采样，因此策略在不学习游戏策略的情况下改善了其执行。(3) 最后，我们冻结整个运动堆栈，并训练一个击球级别的策略，以通过竞争性自我对弈选择落点目标。运动自我对弈在对打中提高了落点准确性和合法回球率。更重要的是，在整个运动堆栈固定的情况下，仅改变目标选择策略会导致游戏结果的巨大差异：学习到的策略在对抗中心、随机和远离对手的基线时分别赢得了89%、82%和73%的得分。学习到的策略还会根据对手的位置和先前的目标改变击球方向，表明自我对弈学习的不仅仅是瞄准固定的“好”位置。在硬件上，相同的冻结执行堆栈和战略接口支持与休闲人类对手进行完整的11分比赛，包括多击球的正手和反手回球对打。

View on arXiv Download PDF AI Translation

cs.RO / 13 / 2610.14075

Assessing the Marginal Utility of Exocentric Head Views in UMI-Style Large-FoV Wrist-Camera Manipulation

评估UMI风格大视场腕部摄像头操作中外向头视图的边际效用

Duan, Yuhui, Jin, Shi, Wang, Yuntian, Wang, Jin, Liu, Siao, Gao, Wenchang, Zhaxizhuoma, Ding, Yan

Abstract

Multi-view perception is often treated as a safe design choice for robot manipulation: if one camera is useful, more cameras should be better. We revisit that assumption in tabletop bimanual manipulation by comparing policies trained with two large-field-of-view (FoV) wrist cameras against policies that additionally receive a head view, using controlled-variable experiments on a real bimanual platform. Across three tabletop tasks, adding a head view does not consistently improve task completion or execution quality: when the wrist cameras already observe the objects, grippers, and goal regions, wrist-only policies form a strong baseline. Restricting the wrist-camera FoV increases the potential value of a head view, but performance improves most reliably when the added view supplies concentrated, task-relevant information: region-of-interest (ROI) cropping yields more consistent head-view gains than the uncropped image, which is dominated by background and task-irrelevant regions. The data-scaling results further suggest an interaction between FoV and demonstration count: wide observations improve coverage but may require more data to stabilize, and they may reach a higher performance ceiling once enough demonstrations are available. These results suggest that camera design should be framed as a task-relevance and information-density problem rather than as a simple count of input views.

Chinese Translation

多视角感知通常被视为机器人操作的安全设计选择：如果一个摄像头有用，那么更多的摄像头应该更好。我们在桌面双手操作中重新审视这一假设，通过比较使用两个大视场（FoV）腕部摄像头训练的策略与额外接收头视图的策略，采用在真实双手平台上的控制变量实验。在三个桌面任务中，添加头视图并未始终改善任务完成度或执行质量：当腕部摄像头已经观察到物体、夹具和目标区域时，仅使用腕部摄像头的策略形成了强有力的基线。限制腕部摄像头的视场增加了头视图的潜在价值，但当添加的视图提供集中且与任务相关的信息时，性能最可靠地提升：感兴趣区域（ROI）裁剪比未裁剪的图像（后者被背景和与任务无关的区域主导）带来了更一致的头视图增益。数据缩放结果进一步表明视场与演示次数之间的相互作用：宽视场观察改善了覆盖范围，但可能需要更多数据来稳定，并且一旦有足够的演示可用，它们可能达到更高的性能上限。这些结果表明，摄像头设计应被视为任务相关性和信息密度问题，而不仅仅是输入视图数量的简单计数。

View on arXiv Download PDF AI Translation

cs.RO / 14 / 2610.14082

SenseWeather: Multimodal Sensing for Robust Robotic Perception under Extreme Conditions

SenseWeather：极端条件下稳健机器人感知的多模态感知

Ke, Fucai, Huang, Sukai, Hao, Junyi, Backman, Kal, Rastgoo, Naim, Haffari, Gholamreza, Rezatofighi, Hamid

Abstract

Robotic perception must remain reliable when extreme environmental conditions severely degrade the primary RGB observation. Although robotic platforms commonly carry complementary sensors such as depth and infrared, much of the modern perception stack remains RGB-centric, creating a mismatch between multimodal onboard sensing and pretrained downstream models. We study whether complementary sensing can be used upstream to recover a task-compatible RGB observation that preserves existing perception capabilities without modifying or retraining the downstream stack. To this end, we introduce SenseWeather, a controlled multimodal benchmark for robotic perception under severe rain, snow, dust, fog, and darkness, with registered RGB, depth, and infrared observations, clean references, and grounded high-level perception tasks. We further develop Fusion-to-RGB, with two parameter-efficient sensor-fusion realizations based on token-space fusion for a diffusion transformer and residual latent conditioning for an inpainting UNet. Across frozen open-vocabulary grounding and vision-language models, RGB+D+IR fusion improves all paired downstream comparisons over the corresponding RGB-only restorers, including AP50 gains across all ten detector comparisons. Importantly, these perception gains do not consistently follow conventional image-fidelity metrics, and increasing weather severity does not induce monotonic downstream degradation. These results show that complementary sensing can preserve reusable robotic perception under extreme conditions, while highlighting task-level perception robustness as an objective distinct from visual restoration quality.

Chinese Translation

在极端环境条件严重降低主要RGB观测的情况下，机器人感知必须保持可靠性。尽管机器人平台通常配备深度和红外等互补传感器，但现代感知系统仍然以RGB为中心，这导致多模态机载感知与预训练的下游模型之间存在不匹配。我们研究了互补感知是否可以在上游用于恢复与任务兼容的RGB观测，从而在不修改或重新训练下游系统的情况下保留现有的感知能力。为此，我们引入了SenseWeather，这是一个针对严重降雨、降雪、尘土、雾霾和黑暗条件下的机器人感知的受控多模态基准，包含注册的RGB、深度和红外观测、干净的参考数据以及基础的高层感知任务。我们进一步开发了Fusion-to-RGB，基于令牌空间融合的扩散变换器和用于修复UNet的残差潜在条件，提出了两种参数高效的传感器融合实现。在冻结的开放词汇基础和视觉-语言模型中，RGB+D+IR融合在所有配对下游比较中优于相应的RGB单一恢复器，包括在所有十个检测器比较中的AP50增益。重要的是，这些感知增益并不总是遵循传统的图像保真度指标，且天气严重程度的增加并未导致下游性能单调下降。这些结果表明，互补感知能够在极端条件下保留可重用的机器人感知，同时强调任务级感知稳健性是一个与视觉恢复质量不同的目标。

View on arXiv Download PDF AI Translation

cs.RO / 15 / 2610.14090

When pose graphs are not enough: a failure analysis of pose-graph-based camera estimation for Gaussian splatting

当姿态图不足时：基于姿态图的相机估计在高斯点云中的失败分析

Palanisamy, Senthil, Sabik, Azwad, Dolhasz, Alan, Slavcheva, Mira

Abstract

Exterior Gaussian-splat twins for real-estate listings are captured to a fixed recipe: a consumer drone orbits a single property in several overlapping passes at stepped altitudes, with occasional vertical ascents, so the imagery is dense, sequential, GPS- and time-tagged, and revisits each facade from many angles. This structure is tantalizing: it suggests camera poses could be recovered by a dense pose-graph optimization (PGO) whose only unknowns are the poses, far fewer than the joint pose-and-point problem bundle adjustment (BA) solves. PGO is mature in robotics, but almost always as post-processing: it redistributes the drift of an already-decent trajectory once loop closures tie it together. Splatting captures push that regime to its limit, every frame closing loops with dozens of others, raising a clean question: does pose-only optimization reach BA accuracy? We built a system specialized to this capture structure, not a general-purpose SfM tool, to test it. It beat COLMAP on speed but never on quality; this paper is the candid post-mortem. The shortfall arises from several interacting causes, hand-set weights, greedy merging and a noisy GPS ruler, together with two limits inherent to the formulation: SE(3) edges carry no scale degree of freedom, so independently scaled sub-maps drift when merged, and a two-view edge stops short of the coupling one 3D point imposes on every frame that sees it. A viable design must add back Sim(3) constraints, absolute GPS priors, retained structure, and uncertainty-aware weighting

Chinese Translation

外部高斯点云双胞胎用于房地产列表的捕获遵循固定的流程：消费级无人机在多个重叠的高度上围绕单一物业进行多次飞行，并偶尔进行垂直上升，因此图像密集、顺序、带有GPS和时间标签，并从多个角度重新访问每个立面。这种结构引人注目：它暗示通过一个密集的姿态图优化（PGO）可以恢复相机姿态，其唯一的未知数是姿态，远少于联合姿态和点问题的束调整（BA）所解决的未知数。PGO在机器人领域已经成熟，但几乎总是作为后处理：它在环路闭合将轨迹联系在一起后重新分配已经不错的轨迹的漂移。高斯点云捕获将这一机制推向极限，每帧与其他数十帧闭合环路，提出了一个清晰的问题：仅基于姿态的优化是否能够达到BA的精度？我们构建了一个专门针对这一捕获结构的系统，而不是通用的结构从运动（SfM）工具，以进行测试。它在速度上超越了COLMAP，但在质量上从未超过；本文是对这一现象的坦诚事后分析。短缺的原因源于几个相互作用的因素，包括手动设置的权重、贪婪合并和噪声GPS标尺，以及该公式固有的两个限制：SE(3)边缘不携带尺度自由度，因此独立缩放的子地图在合并时会漂移，以及双视图边缘未能完全体现一个3D点对每个看到它的帧施加的耦合。一个可行的设计必须重新引入Sim(3)约束、绝对GPS先验、保留的结构以及考虑不确定性的加权。

View on arXiv Download PDF AI Translation

cs.RO / 16 / 2610.14106

Embodied Zero-Shot Task Planning via Hardware-Aware LLM-VLM Integration for Grounded Robotic Manipulation

基于硬件感知的LLM-VLM集成的具身零-shot任务规划用于基础机器人操作

Tankersley, Lonnie, Golilarz, Noorbakhsh Amiri, Korivand, Soroush

Abstract

Large language models can translate natural-language instructions into plausible manipulation plans, but robotic execution requires physically grounded decisions that respect robot embodiment, workspace geometry, reachability, collision constraints, and grasp feasibility. This paper presents a hardware-aware large language model-vision-language model (LLM-VLM) manipulation framework for a simulated Franka Research 3 robot in NVIDIA Isaac Sim. The framework separates symbolic reasoning from metric execution: an API-hosted LLM planner receives a structured scene summary and generates only symbolic task information, including the object to manipulate, target relation, approach preference, and objects to avoid. A local Qwen3-VL 8B VLM verifier supports semantic confirmation of ambiguous objects and open-container surrogates. Both the LLM planner and the VLM verifier are prohibited from generating robot coordinates, grasp poses, keepout dimensions, gripper commands, trajectory waypoints, or success labels. Instead, these decisions are handled by deterministic grounding and execution modules that use the robot model, scene geometry, MoveIt 2 collision checking, inverse kinematics, grasp validation, release verification, and final visual assessment. Execution feedback is returned only as symbolic context for replanning. The framework is evaluated through simulated digital-twin trials involving pickup, occluded-source access, keepout reasoning, relation placement, and sequential manipulation tasks.

Chinese Translation

大型语言模型能够将自然语言指令转化为合理的操作计划，但机器人执行需要尊重机器人具身性、工作空间几何、可达性、碰撞约束和抓取可行性的物理基础决策。本文提出了一种硬件感知的大型语言模型-视觉语言模型（LLM-VLM）操作框架，适用于NVIDIA Isaac Sim中的模拟Franka Research 3机器人。该框架将符号推理与度量执行分开：一个托管API的LLM规划器接收结构化场景摘要，并仅生成符号任务信息，包括要操作的对象、目标关系、接近偏好和需要避免的对象。一个本地的Qwen3-VL 8B VLM验证器支持对模糊对象和开放容器替代物的语义确认。LLM规划器和VLM验证器均被禁止生成机器人坐标、抓取姿态、禁入尺寸、夹持器命令、轨迹路径点或成功标签。相反，这些决策由确定性基础和执行模块处理，这些模块使用机器人模型、场景几何、MoveIt 2碰撞检测、逆向运动学、抓取验证、释放验证和最终视觉评估。执行反馈仅作为符号上下文返回，以便重新规划。该框架通过模拟数字双胞胎试验进行评估，涉及拾取、遮挡源访问、禁入推理、关系放置和顺序操作任务。

View on arXiv Download PDF AI Translation

cs.RO / 17 / 2610.14117

Make {\pi}0.5 Learn to Call Astra for Reliable Robot Manipulation

让 {}.5 学会调用 Astra 以实现可靠的机器人操作

Sun, Jianli, Li, Chengxiang, Tian, Yonglin

Abstract

GPT-6 Astra provides robust embodied reasoning, but frequent external queries increase inference time and token costs. We equip a frozen $\pi_{0.5}$ policy with a lightweight CALL head that learns to predict when assistance is necessary. CALL reuses features from the standard policy forward pass and operates locally without consulting external models. Astra is activated solely for critical recovery events and returns control to $\pi_{0.5}$ upon stabilization. Evaluated across 40 LIBERO tasks (200 episodes), our system achieves a 97.5% success rate, outperforming the 95.5% baseline of pure $\pi_{0.5}$. In the Object, Goal, and Spatial suites, Astra executes only 5.49% of environment steps, consuming approximately 5,027 tokens per test episode. This selective intervention preserves the fast ordinary execution of $\pi_{0.5}$ while seamlessly integrating Astra's reasoning on demand. Our project page is available at https://kykgeek.github.io/Pi05_CALL_Astra/.

Chinese Translation

GPT-6 Astra 提供了强大的具身推理能力，但频繁的外部查询会增加推理时间和令牌成本。我们为冻结的 $_{0.5}$ 策略配备了一个轻量级的 CALL 头，该头学习预测何时需要帮助。CALL 重用标准策略前向传播中的特征，并在本地操作，无需咨询外部模型。Astra 仅在关键恢复事件中被激活，并在稳定后将控制权返回给 $_{0.5}$。在 40 个 LIBERO 任务（200 个回合）的评估中，我们的系统实现了 97.5% 的成功率，超越了纯 $_{0.5}$ 的 95.5% 基线。在对象、目标和空间套件中，Astra 仅执行了 5.49% 的环境步骤，每个测试回合消耗约 5,027 个令牌。这种选择性干预保留了 $_{0.5}$ 的快速普通执行，同时无缝集成了 Astra 的按需推理。我们的项目页面可在 https://kykgeek.github.io/Pi05_CALL_Astra/ 获取。

View on arXiv Download PDF AI Translation

cs.RO / 18 / 2610.14142

Refine Where It Matters: Contact-Adaptive Meshing for Deformable Object Manipulation

在关键区域进行细化：用于可变形物体操作的接触自适应网格划分

Johnson, Gabriel, Olesch, Tom, Liu, Fei

Abstract

Deformable-object manipulation requires accurate modeling of highly localized contact while maintaining simulation speeds suitable for interactive use. Uniformly fine tetrahedral meshes can resolve these interactions, but they introduce unnecessary degrees of freedom throughout the object. Coarse meshes are more efficient, yet they may under-resolve contact-induced deformation. We present a contact-adaptive mesh refinement framework that introduces spatial resolution only where it is most useful. The method uses geometric, elastic, and contact quantities already available from the deformable-body solver to score candidate edges. A fully GPU-parallel procedure then selects a conflict-free subset and updates both mesh topology and physical state directly within the simulation loop. We evaluate the approach on real deformable-object manipulation sequences. Adaptive refinement reduces simulation-to-real error, improves agreement with a fine-mesh reference, and remains substantially more efficient than uniformly fine simulation. These results show that local refinement can improve physical fidelity without incurring the cost of a globally fine discretization. The framework provides a practical basis for contact-rich simulation in interactive manipulation, sim-to-real modeling, and online physical digital twins.

Chinese Translation

可变形物体的操作需要对高度局部化的接触进行准确建模，同时保持适合交互使用的仿真速度。均匀细致的四面体网格能够解决这些交互，但它们在整个物体中引入了不必要的自由度。粗糙网格更为高效，但可能无法充分解析接触引起的变形。我们提出了一种接触自适应网格细化框架，仅在最有用的地方引入空间分辨率。该方法利用来自可变形体求解器的几何、弹性和接触量来对候选边进行评分。然后，完全基于GPU的并行程序选择一个无冲突的子集，并在仿真循环中直接更新网格拓扑和物理状态。我们在真实的可变形物体操作序列上评估了该方法。自适应细化减少了仿真与真实之间的误差，提高了与细网格参考的吻合度，并且在效率上仍显著优于均匀细致的仿真。这些结果表明，局部细化可以提高物理真实性，而无需承担全局细致离散化的成本。该框架为交互操作中的接触丰富仿真、仿真到真实建模以及在线物理数字双胞胎提供了实用基础。

View on arXiv Download PDF AI Translation

cs.RO / 19 / 2610.14155

Axiom-0: Hierarchical Self-Refinement for Vision-Language-Action Policies

公理-0：视觉-语言-动作策略的层次自我精炼

Shu, Renbin, Yang, Jing, You, Xiangyuan, Qiu, Guolin, Xia, Yufei, Zhu, Yiming, Dai, Nanjun, Zhang, Fei, Wu, Feng, Tang, Keke

Abstract

Reliable manipulation requires policies that generalize across tasks, layouts, and object interactions. Vision-language-action (VLA) models combine pretrained multimodal understanding with learned control, but their performance remains limited on unfamiliar task compositions. We present Axiom-0, a VLA policy that addresses both the quality of its representations and their use in action generation through hierarchical self-refinement. Intra-VLM Representation Refinement (IVRR) strengthens selected intermediate features through consistency supervision with learned feature projections during training. VLM-Action Interface Refinement (VAIR) recursively updates the resulting conditioning before action generation. Joint optimization of the supervised VLM layers, VAIR, and the action generator connects the two components. On RoboCasa365, Axiom-0 achieves 62.32% overall success in our evaluation, exceeding the highest published leaderboard score (Paimon-0, 58.1%) by 4.22 percentage points and the Xiaomi-Robotics-1 reference by 4.92 points. Under the evaluated profiling setup, mean action-chunk generation latency increases by 1.7% over Xiaomi-Robotics-1. These results demonstrate improved task completion with modest inference overhead.

Chinese Translation

可靠的操控需要能够在任务、布局和物体交互之间进行泛化的策略。视觉-语言-动作（VLA）模型结合了预训练的多模态理解与学习的控制，但在不熟悉的任务组合上，其性能仍然有限。我们提出了公理-0（Axiom-0），这是一种VLA策略，通过层次自我精炼来解决其表征质量及其在动作生成中的应用。内部VLM表征精炼（Intra-VLM Representation Refinement, IVRR）通过在训练过程中与学习的特征投影进行一致性监督，增强所选的中间特征。VLM-动作接口精炼（VLM-Action Interface Refinement, VAIR）在动作生成之前递归更新生成的条件。对监督VLM层、VAIR和动作生成器的联合优化将这两个组件连接起来。在RoboCasa365上，公理-0在我们的评估中实现了62.32%的整体成功率，超过了最高已发布排行榜分数（Paimon-0，58.1%）4.22个百分点，以及小米机器人1（Xiaomi-Robotics-1）参考值4.92点。在评估的配置设置下，平均动作块生成延迟比小米机器人1增加了1.7%。这些结果表明，在适度的推理开销下，任务完成度得到了改善。

View on arXiv Download PDF AI Translation

cs.RO / 20 / 2610.14159

Disentangling Task-Relevant Latent Dynamics for Reward Modeling

解耦任务相关的潜在动态以进行奖励建模

Li, Dayou, Wang, Jionghao, Wang, Peihao, Lei, Jiuzhou, Wang, Hao, Chai, Jichen, Hu, Hezhen, Zheng, Minghui, Wang, Zhangyang, Li, Xin, Wang, Wenping, Fan, Zhiwen

Abstract

Reward design remains a major bottleneck for reinforcement learning in robotic manipulation. Handcrafted dense rewards can compensate for the limited guidance provided by sparse rewards, but require substantial task-specific reward engineering. Learning rewards from videos provides a scalable, data-driven alternative to manual reward design. However, existing video-based rewards are often computed in entangled visual spaces where task-relevant state changes are confounded with extraneous visual variation. As a result, the learned reward may respond to visual distractors instead of actual task execution. To address this limitation, we propose DiTaR (Disentangling Task-Relevant Latent Dynamics for Reward Modeling), a framework for learning dense rewards from heterogeneous videos. DiTaR learns a continuous task-relevant representation from multi-frame video latents through an asymmetric two-stage procedure that suppresses task-irrelevant visual variation. In the task-relevant latent space, we train a history- and language-conditioned rectified-flow model to capture task dynamics and derive a directional dense reward from temporal changes in dynamics compatibility. Together, these components enable reward modeling directly over task-relevant dynamics rather than generic visual variation. DiTaR substantially improves policy learning on MetaWorld-v3 and ManiSkill over video-based reward baselines, and transfers to real-world offline RL through reward relabeling. Reward analyses further show that DiTaR provides more reliable rewards.

Chinese Translation

奖励设计仍然是机器人操作中强化学习的主要瓶颈。手工设计的密集奖励可以弥补稀疏奖励提供的有限指导，但需要大量特定任务的奖励工程。从视频中学习奖励提供了一种可扩展的数据驱动替代手动奖励设计的方法。然而，现有的视频基础奖励通常是在纠缠的视觉空间中计算的，其中任务相关的状态变化与外部视觉变化混淆。因此，学习到的奖励可能对视觉干扰物做出反应，而不是实际的任务执行。为了解决这一局限性，我们提出了DiTaR（Disentangling Task-Relevant Latent Dynamics for Reward Modeling），一个从异构视频中学习密集奖励的框架。DiTaR通过一种不对称的两阶段程序，从多帧视频潜在表示中学习连续的任务相关表示，抑制与任务无关的视觉变化。在任务相关的潜在空间中，我们训练了一个历史和语言条件的修正流模型，以捕捉任务动态，并从动态兼容性的时间变化中推导出方向性密集奖励。这些组件共同使得奖励建模能够直接针对任务相关的动态，而不是通用的视觉变化。DiTaR在MetaWorld-v3和ManiSkill上显著改善了策略学习，相较于基于视频的奖励基线，并通过奖励重新标记转移到真实世界的离线强化学习。奖励分析进一步表明，DiTaR提供了更可靠的奖励。

View on arXiv Download PDF AI Translation

cs.RO / 21 / 2610.14191

Observation-Referenced Tissue Deformation Forecasting with Tool-Driven and Residual Dynamics

基于观察的组织变形预测：工具驱动与残差动态

Olesch, Tom, Xiao, Nan, Xiao, Xiongye, DeLong, Jonathan C., Liu, Fei

Abstract

Forecasting tissue deformation from future instrument motion could support predictive planning and increasingly autonomous surgical systems. We study this problem using a scene-specific differentiable surface model fitted to an observed tool-tissue interaction window. Rather than forecasting absolute simulated positions, we introduce an \textit{observation-referenced} formulation that anchors each tissue point to its last valid observation and adds only the deformation induced by the known, preplanned future tool trajectory. This reduces sensitivity to reconstruction, initialization, and persistent simulation offsets while preserving the predicted deformation response. The physical model uses differentiable, mesh-free position-based dynamics with scene-specific rigidity, attachment, region scale, and grasp coupling. We further introduce an optional \textit{observation-referenced residual model} combining graph-spectral spatial modes with periodic harmonic dynamics for structured motion not explained by the tool input. We evaluate the framework retrospectively on 12 surgical manipulation clips from three datasets. Observation referencing provides the largest gain, with a median within-clip error ratio of $0.598$ relative to the corresponding absolute simulation forecast. The optional periodic residual further improves forecasting when the remaining motion exhibits periodic structure. These results support observation-referenced deformation prediction as a practical building block for future surgical world models, digital twins, and generative simulators that must predict how tissue evolves under instrument actions.

Chinese Translation

从未来仪器运动中预测组织变形可以支持预测性规划和日益自主的外科系统。我们使用一个针对观察到的工具-组织交互窗口拟合的场景特定可微表面模型来研究这个问题。我们引入了一种 extit{基于观察的}公式，将每个组织点锚定到其最后有效观察，并仅添加由已知的、预先计划的未来工具轨迹引起的变形，而不是预测绝对的模拟位置。这减少了对重建、初始化和持续模拟偏差的敏感性，同时保留了预测的变形响应。物理模型使用可微的无网格基于位置的动态，具有场景特定的刚度、附着、区域尺度和抓取耦合。我们进一步引入了一个可选的 extit{基于观察的残差模型}，将图谱谱空间模式与周期谐波动态相结合，用于解释工具输入无法解释的结构化运动。我们在三个数据集的12个外科操作剪辑上对该框架进行了回顾性评估。基于观察的引用提供了最大的增益，相对于相应的绝对模拟预测，剪辑内的中位误差比为$0.598$。当剩余运动表现出周期结构时，可选的周期残差进一步改善了预测。这些结果支持基于观察的变形预测作为未来外科世界模型、数字双胞胎和生成模拟器的实用构建块，这些模型必须预测组织在仪器作用下的演变。

View on arXiv Download PDF AI Translation

cs.RO / 22 / 2610.14200

ClothCode: Generative Modeling for Garment State Estimation and Dynamics Prediction

ClothCode：用于服装状态估计和动态预测的生成建模

Ru, Yingdong, Zhuang, Lipeng, Fan, Shiyu, He, Zhuo, Audonnet, Florent P., Camarasa, Gerardo Aragon

Abstract

We introduce ClothCode, a unified generative framework for garment state estimation and action-conditioned dynamics prediction in a shared discrete configuration space. A material-aligned tokenizer encodes garment configuration as residual vector-quantized patch tokens, while task-specific Masked Generative Transformers predict these tokens through parallel confidence-guided refinement. Across three garment templates, ClothCode reduces mean displacement error by 26.6% for state estimation and 37.0% for dynamics prediction relative to the strongest baselines. Dynamics prediction requires only 16.7 ms per window, compared with 556.2 ms for diffusion. Integrated into the same model-predictive controller, ClothCode improves real-world folding and flattening success from 56.7% to 78.9%. Further results and videos are available at: https://anonymous.4open.science/r/ClothCode-DE32

Chinese Translation

我们介绍了ClothCode，这是一个统一的生成框架，用于在共享的离散配置空间中进行服装状态估计和基于动作的动态预测。一个与材料对齐的标记器将服装配置编码为残差向量量化的补丁标记，而任务特定的掩蔽生成变换器通过并行的置信度引导细化来预测这些标记。在三个服装模板中，ClothCode在状态估计方面将平均位移误差降低了26.6%，在动态预测方面降低了37.0%，相较于最强基线。动态预测每个窗口仅需16.7毫秒，而扩散方法则需556.2毫秒。集成到同一模型预测控制器中，ClothCode将现实世界中折叠和平整的成功率从56.7%提高到78.9%。更多结果和视频可在以下链接查看：https://anonymous.4open.science/r/ClothCode-DE32

View on arXiv Download PDF AI Translation

cs.RO / 23 / 2610.14235

Mapping Tire-Road Friction Coefficients Along Driven Trajectories

沿驱动轨迹映射轮胎-路面摩擦系数

Ojala, Risto

Abstract

Information of the tire-road friction coefficient is critical for intelligent vehicle systems, as friction establishes the limits at which the vehicle can be manoeuvred without loss of control. While there is extensive literature on online friction estimation, no previous studies have explored post-processing and mapping friction coefficients along a driven trajectory based on recorded sensor data. Mapped friction information could be shared by connected vehicles, enabling proactive measures in low-friction areas. This paper presents an offline post-processing method, FrictionMap, for mapping friction coefficients from recorded driving data. The method operates on commonly available on-board vehicle sensor information, and requires only a basic vehicle parameter set and a slip threshold as prior knowledge. FrictionMap runs an extended Kalman filter combined with a Rauch-Tung-Striebel smoother to estimate vehicle dynamics. Data points where the friction limit is observable are identified by filtering points based on tire slip. A Gaussian process is fitted to these points to acquire a continuous estimate of the friction coefficient along the trajectory. FrictionMap is validated via a simulation study, and shown to outperform commonly used online friction estimation techniques. The code implementation and the simulation benchmark are released as open-source after paper acceptance to enable further research.

Chinese Translation

轮胎-路面摩擦系数的信息对于智能车辆系统至关重要，因为摩擦决定了车辆在不失控的情况下可以操控的极限。尽管关于在线摩擦估计的文献广泛，但之前的研究并未探讨基于记录的传感器数据对驱动轨迹上的摩擦系数进行后处理和映射。映射的摩擦信息可以被连接车辆共享，从而在低摩擦区域采取主动措施。本文提出了一种离线后处理方法FrictionMap，用于从记录的驾驶数据中映射摩擦系数。该方法基于常见的车载传感器信息，仅需基本的车辆参数集和滑移阈值作为先验知识。FrictionMap运行扩展卡尔曼滤波器，并结合Rauch-Tung-Striebel平滑器来估计车辆动态。通过基于轮胎滑移过滤点，识别出摩擦极限可观察的数据点。对这些点拟合高斯过程，以获得沿轨迹的摩擦系数的连续估计。FrictionMap通过仿真研究进行验证，并显示出优于常用的在线摩擦估计技术。代码实现和仿真基准将在论文接受后作为开源发布，以促进进一步研究。

View on arXiv Download PDF AI Translation

cs.RO / 24 / 2610.14237

Composable Dynamics for Cooperative Humanoids: A RoboCompiler Extension

可组合动态的协作人形机器人：RoboCompiler扩展

Shahna, Mehdi Heydari, Mattila, Jouni

Abstract

Physically connected humanoids need consistent dynamics for internal actuator loops, inter-robot forces, and changing ground supports. We extend RoboCompiler with member-wise dynamics assembly, compiling reduced dynamics from rigid bodies, moving actuator parts, and reflected rotor inertias. Compiled body Jacobians and analytic constraint curvature yield an explicit Coriolis factor satisfying the power identity for floating-base quasi-velocities. A two-level reduction retains reusable robot models while composing their supports and hand connections; local inertial updates have rank at most six, and prescribed-acceleration constraint wrenches are affine in actuator efforts. These interfaces support actuator-aware tracking, coupled force allocation, payload updates, and support-mode switching for five hand-linked humanoids. Across 264 feasible states in eight humanoid configurations, normalised acceleration discrepancies remain within $1.6\!\times\!10^{-10}$ against a hard-constraint reference and $1.4\!\times\!10^{-10}$ against a compiler-independent reference; MuJoCo simulations achieve 0.73~mm root-mean-square (RMS) pelvis error in the ring dance and complete side steps through 20 support-mode changes.

Chinese Translation

物理连接的人形机器人需要一致的动态以支持内部执行器环路、机器人间的力以及变化的地面支撑。我们通过成员级动态组装扩展了RoboCompiler，从刚体、移动执行器部件和反射转子惯性中编译出简化的动态。编译的身体雅可比矩阵和解析约束曲率产生了一个显式的科里奥利因子，满足浮动基准准速度的功率恒等式。两级简化保留了可重用的机器人模型，同时组合它们的支撑和手部连接；局部惯性更新的秩最多为六，规定加速度约束扭矩在执行器努力中是仿射的。这些接口支持执行器感知跟踪、耦合力分配、有效载荷更新以及五个手部连接的人形机器人之间的支撑模式切换。在八种人形配置的264个可行状态中，归一化加速度差异在硬约束参考下保持在$1.6 imes10^{-10}$以内，在编译器无关参考下保持在$1.4 imes10^{-10}$以内；MuJoCo模拟在环舞中实现了0.73~mm的根均方(RMS)骨盆误差，并通过20次支撑模式变化完成了完整的侧步。

View on arXiv Download PDF AI Translation

cs.RO / 25 / 2610.14245

From LLM Plans to Authorized Motion: Execution-Bound Authorization for Excavators

从大语言模型计划到授权运动：挖掘机的执行绑定授权

Shahna, Mehdi Heydari, Kim, Seihun, Jung, Soyi, Park, Soohyun, Mattila, Jouni, Kim, Joongheon

Abstract

An LLM-directed excavator can execute mechanically feasible motion whose destination or authority no longer matches the operator's approval. We propose execution-bound authorization for the approval--dispatch--execution interval. A one-use lease binds the action, generated motion, mission contract and admission tolerance; atomic dispatch rechecks current authority and measured joint drift. Periodic supervision detects revocation and requests a quintic stop with a carry-pitch endpoint, selected through sampled joint, stroke, flow and reference-acceleration checks. The framework connects exact motion identity to dispatch integrity and load-aware interruption. Evaluation combines a 672-case interface suite, four loaded digging cases in synchronous simulation, unloaded MuJoCo/Pinocchio transfers and two command-level LLM planners. The gateway admitted all 120 benign interface cases and produced zero unauthorized dispatches in the attack suite. Loaded dispatch-window tests produced zero unauthorized deposition. During simulated dump-phase revocation, maximum closed-site deposition decreased from 1.23 to 0.33\,m$^3$; loaded-slew stops used less than 75\% of nominal swing torque. These results demonstrate motion-bound authorization and quantify the retention and stopping behavior of supervised excavation skills.

Chinese Translation

一个由大语言模型（LLM）指导的挖掘机可以执行机械上可行的运动，但其目的地或权限不再与操作员的批准相匹配。我们提出了一种执行绑定授权的方法，适用于批准-调度-执行的时间间隔。一次性租赁绑定了动作、生成的运动、任务合同和接纳容忍度；原子调度重新检查当前权限和测量的关节漂移。定期监督检测撤销并请求一个五次停止（quintic stop），其终点通过采样的关节、行程、流量和参考加速度检查进行选择。该框架将精确的运动身份与调度完整性和负载感知中断连接起来。评估结合了672个案例的接口套件、四个同步仿真的负载挖掘案例、未加载的MuJoCo/Pinocchio转移以及两个命令级LLM规划器。该网关接受了所有120个良性接口案例，并在攻击套件中产生了零个未经授权的调度。负载调度窗口测试产生了零个未经授权的沉积。在模拟的倾倒阶段撤销期间，最大闭合现场沉积量从1.23降至0.33 m³；负载回转停止使用的名义摆动扭矩不到75%。这些结果展示了运动绑定授权，并量化了监督挖掘技能的保持和停止行为。

View on arXiv Download PDF AI Translation

cs.RO / 26 / 2610.14248

Akule: Fast and Scalable Multi-Robot Motion Planning via Sparse Interaction Diffusion

Akule：通过稀疏交互扩散实现快速且可扩展的多机器人运动规划

Vora, Manav, Puthumanaillam, Gokul, Penumarti, Aditya, Shin, Jane, Ornik, Melkior

Abstract

Diffusion models provide expressive trajectory priors for multi-robot motion planning (MRMP), but scaling them to large teams requires coordinating individual motions without expensive joint computation. We introduce Akule, a sparse interaction diffusion framework that augments a frozen single-robot prior with three learned modules: a reusable pairwise correction model $R_{\phi}$, a contextual selector $U_{\psi}$ that determines which corrections to evaluate, and a composer $G_{\omega}$ that combines their effects through context-dependent signed weights. This decomposition separates which interactions merit computation from how their effects should be combined into coordinated denoising updates. We train $R_{\phi}$ using matched pairwise trajectory optimizations, then learn $G_{\omega}$ and $U_{\psi}$ from physically optimized joint trajectories without requiring a prescribed ground-truth interaction graph. The resulting joint proposals are passed to a downstream search-based planner for validation and repair. Across diverse MRMP benchmarks, Akule achieves competitive planning success, complete-planner speedups of up to $25\times$, zero-shot population transfer, and successful planning with up to $256$ robots. Ablations further demonstrate that learned selection and composition reduce pre-repair conflicts and downstream repair effort relative to heuristic selection and uniform summation.

Chinese Translation

扩散模型为多机器人运动规划（MRMP）提供了丰富的轨迹先验，但将其扩展到大规模团队需要协调个体运动而不进行昂贵的联合计算。我们提出了Akule，一种稀疏交互扩散框架，它通过三个学习模块增强了冻结的单机器人先验：一个可重用的成对修正模型 $R_{ heta}$、一个上下文选择器 $U_{ ho}$，用于确定评估哪些修正，以及一个组合器 $G_{ au}$，通过上下文相关的加权组合它们的效果。这种分解将哪些交互值得计算与如何将其效果组合成协调的去噪更新分开。我们使用匹配的成对轨迹优化训练 $R_{ heta}$，然后从物理优化的联合轨迹中学习 $G_{ au}$ 和 $U_{ ho}$，而不需要预设的真实交互图。生成的联合提案被传递给下游基于搜索的规划器进行验证和修复。在多样化的MRMP基准测试中，Akule实现了竞争性的规划成功率、最高达 $25 imes$ 的完整规划器加速、零样本种群迁移，以及成功规划多达 $256$ 个机器人。消融实验进一步表明，相较于启发式选择和均匀求和，学习的选择和组合减少了修复前的冲突和下游修复工作量。

View on arXiv Download PDF AI Translation

cs.RO / 27 / 2610.14335

PILOT: Guiding Long-Horizon World-Model Planning for Robot Manipulation with Visual Subgoals and Policy Priors

PILOT：通过视觉子目标和策略先验引导机器人操控的长时间世界模型规划

Liu, Yang, Cui, Hanchen, Desingh, Karthik

Abstract

Latent world-model planning has shown promising results in goal-conditioned robotic manipulation, but remains challenging for long-horizon planning due to limited temporal guidance and inefficient action search. A distant visual goal provides little guidance for preparatory actions, while uninformed action sampling makes local optimization inefficient. We propose PILOT (Policy-Initialized Local Optimization with Temporal subgoals), which uses a video model to generate intermediate visual subgoals and a learned policy to initialize the cross-entropy method (CEM). CEM refines candidate action sequences toward each subgoal using an action-conditioned world model in a receding-horizon control loop. In our primary real-robot evaluation across four UR5e manipulation tasks, PILOT achieves 68.8% success, compared with 55.0% without policy initialization, 13.8% for policy-only execution, and 0.0% for direct final-goal planning. Simulation results show that these gains persist over longer horizons: PILOT exceeds direct execution of the same policy by 30 percentage points on long Push-T tasks and by 22 percentage points on LIBERO-10. Policy initialization also accelerates local optimization: on Push-T, CEM matches the cost reached by generic initialization after 10 iterations in only 4 iterations and reduces the final planning cost by 24.1% under the same sampling budget. These results show that combining temporal subgoals with policy-guided action search improves both long-horizon task execution and local planning efficiency. Videos are available on the webpage: https://pilot-planner.github.io/.

Chinese Translation

潜在世界模型规划在目标条件下的机器人操控中显示出良好的结果，但由于时间指导有限和行动搜索效率低下，长时间规划仍然具有挑战性。遥远的视觉目标对准备行动提供的指导有限，而无信息的行动采样使得局部优化效率低下。我们提出了PILOT（基于策略初始化的局部优化与时间子目标），该方法使用视频模型生成中间视觉子目标，并利用学习到的策略初始化交叉熵方法（CEM）。CEM在递归控制循环中使用行动条件的世界模型，针对每个子目标优化候选行动序列。在我们对四个UR5e操控任务的主要真实机器人评估中，PILOT的成功率达到68.8%，而没有策略初始化时为55.0%，仅使用策略执行时为13.8%，直接进行最终目标规划时为0.0%。仿真结果表明，这些收益在更长的时间范围内仍然存在：在长时间的Push-T任务中，PILOT的表现比直接执行相同策略高出30个百分点，在LIBERO-10任务中高出22个百分点。策略初始化还加速了局部优化：在Push-T任务中，CEM在仅用4次迭代后就达到了通用初始化在10次迭代后达到的成本，并在相同的采样预算下将最终规划成本降低了24.1%。这些结果表明，将时间子目标与策略引导的行动搜索相结合，能够提高长时间任务执行和局部规划的效率。视频可在网页上查看：https://pilot-planner.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 28 / 2610.14341

CoPlanVLM: Coordinated Path Planning for a Robot Team using Vision Language Models

CoPlanVLM：基于视觉语言模型的机器人团队协调路径规划

Martin, David R., Paquet, Jared, Faruque, Mohammad Abdullah Al, Khargonekar, Pramod, Volle, Kyle, Ramos, J. Humberto, Nivison, Scott

Abstract

We present CoPlanVLM, which uses a Vision Language Model (VLM) to translate a single natural-language instruction and image into coordinated paths for a team of two robots. While VLMs offer a flexible, natural-language interface for commanding robot teams, their weak spatial reasoning has greatly limited VLM-based path planning. CoPlanVLM improves the VLM's spatial reasoning and path planning abilities through three components: (1) an image overlay marking both free space and obstacles, (2) task classification that selects the most appropriate low-level planner and reasoning prompt for each instruction, and (3) task-specific chain-of-thought reasoning that prompts the VLM to identify key objects, allocate roles, and verify routes are collision-free before committing to a plan. Together, these components enable the team to complete complex coordinated maneuvers requiring both robots and divide large tasks between robots to increase efficiency. CoPlanVLM also supports detailed maneuvers involving specific routes, loops, and coverage of large areas, as well as tasks requiring directional reasoning such as north/south or left/right. We validate CoPlanVLM in both Gazebo simulation and on physical Turtlebot 4 hardware. CoPlanVLM attains an average success rate of 84.4% compared to 50.0% for the closest baseline. On trials where both methods succeed, CoPlanVLM's paths are 27% shorter on average. Project code is available at https://github.com/AICPS/CoPlanVLM.

Chinese Translation

我们提出了CoPlanVLM，它利用视觉语言模型（VLM）将单一自然语言指令和图像转换为两个机器人团队的协调路径。虽然VLM为指挥机器人团队提供了灵活的自然语言接口，但其空间推理能力的不足极大限制了基于VLM的路径规划。CoPlanVLM通过三个组件提升了VLM的空间推理和路径规划能力：（1）图像叠加标记自由空间和障碍物；（2）任务分类，选择最合适的低级规划器和推理提示以应对每个指令；（3）特定任务的思维链推理，促使VLM识别关键物体、分配角色，并在确定计划之前验证路径是否无碰撞。这些组件共同使团队能够完成复杂的协调动作，要求两个机器人协同工作，并在机器人之间分配大型任务以提高效率。CoPlanVLM还支持涉及特定路线、循环和大面积覆盖的详细操作，以及需要方向推理的任务，如北/南或左/右。我们在Gazebo仿真和物理Turtlebot 4硬件上验证了CoPlanVLM。与最接近的基线相比，CoPlanVLM的平均成功率为84.4%，而基线为50.0%。在两种方法均成功的试验中，CoPlanVLM的路径平均缩短了27%。项目代码可在https://github.com/AICPS/CoPlanVLM获取。

View on arXiv Download PDF AI Translation

cs.RO / 29 / 2610.14372

DexFLEX: Contact-Aware Foundation Controller for Command-Guided Dexterity

DexFLEX：一种接触感知的基础控制器，用于命令引导的灵巧性

Wan, Weikang, Lai, Zeyu, Hou, Jianyu, Zhou, Yuchen, Fu, Jiawei, Christensen, Henrik I, Su, Hao

Abstract

Dexterous robot hands can receive useful motion intent from teleoperation or learned policies, but successful execution also depends on contact decisions that these commands rarely specify. We introduce DexFLEX, a contact-aware foundation controller that turns upstream fingertip-motion drafts into contact-consistent joint commands. Instead of treating a draft as a trajectory to copy, DexFLEX treats it as evidence about intent: from the current tactile-proprioceptive state, it proposes feasible short-horizon motion chunks, predicts their contact consequences, and selects the candidate that best follows the command while preserving future contact stability. At inference time, a simple Draft-Dream-Select procedure combines pure-prior proposals for recovery with draft-seeded proposals for responsiveness, then decodes the selected motion into executable joint-space targets. Across degraded-command simulation, real-world shared control, and visuomotor policy learning, the same trained controller improves robustness without retraining, increasing real-world teleoperation success from 29.2% to 78.3% and reaching 46.7% policy-learning success, 3.5 times direct joint-action prediction, and 1.56 times prior-only correction. Videos are available at dex-flex.github.io.

Chinese Translation

灵巧的机器人手可以从远程操作或学习的策略中接收有用的运动意图，但成功执行还依赖于这些命令很少具体说明的接触决策。我们提出了DexFLEX，一种接触感知的基础控制器，它将上游的指尖运动草图转化为接触一致的关节命令。DexFLEX并不将草图视为需要复制的轨迹，而是将其视为关于意图的证据：从当前的触觉-本体感知状态出发，它提出可行的短期运动片段，预测其接触后果，并选择最符合命令且保持未来接触稳定性的候选方案。在推理时，一个简单的草图-梦想-选择程序将纯先验提案与基于草图的响应性提案结合起来，然后将所选运动解码为可执行的关节空间目标。在降级命令模拟、现实世界共享控制和视觉运动策略学习中，同一训练的控制器在不重新训练的情况下提高了鲁棒性，使现实世界的远程操作成功率从29.2%提高到78.3%，策略学习成功率达到46.7%，直接关节动作预测提高了3.5倍，先验纠正提高了1.56倍。视频可在dex-flex.github.io查看。

View on arXiv Download PDF AI Translation

cs.RO / 30 / 2610.14379

OATH: Object-Agnostic Tracking for Dexterous Hands, Enabling Rapid Physics-based Retargeting

OATH：面向灵巧手的物体无关跟踪，实现快速基于物理的重定向

Sigurdson, Solvin, Sriganesh, Prasanna, Borbon, Juan, Burdick, Joel, Zhang, Dingqi, Wang, Jiuguang

Abstract

Retargeting human demonstrations into dynamically feasible dexterous hand trajectories requires resolving complex contact physics, yet current methods either omit physical constraints or rely on computationally expensive sampling-based optimization in simulation. Reinforcement learning has previously been used to achieve planning and control through contact dynamics; however, standard policies require object-specific training and fail to generalize across arbitrary geometries. This work reframes physics-based retargeting as a residual learning problem, where we demonstrate that zero-shot generalization across objects relies on structured state representations. Specifically, formulating the observation space to include a localized contact-error signal guiding each robot link to its target coordinate on the object surface enables a single, unified residual policy to successfully grasp novel, unseen object geometries. Rolling out this policy in simulation on a kinematic reference trajectory yields a physics-based retargeter, OATH. Operating at 0.4 seconds of compute per second of reference, OATH achieves 2.4x faster-than-real-time performance and a 60-350x reduction in computational cost compared to sampling-based methods. The system recovers 58% of held-out and 54% of unseen-object DexYCB trajectories while maintaining a sub-2 cm tracking error.

Chinese Translation

将人类示范重定向为动态可行的灵巧手轨迹需要解决复杂的接触物理问题，然而当前的方法要么忽略物理约束，要么依赖于计算成本高昂的基于采样的优化仿真。强化学习曾被用于通过接触动态实现规划和控制；然而，标准策略需要针对特定物体的训练，且无法在任意几何形状间进行泛化。本研究将基于物理的重定向重新构架为一个残差学习问题，我们展示了跨物体的零样本泛化依赖于结构化的状态表示。具体而言，将观察空间构建为包含局部接触误差信号，指导每个机器人关节到达物体表面的目标坐标，使得一个统一的残差策略能够成功抓取新颖且未见过的物体几何形状。在运动学参考轨迹上在仿真中展开该策略，产生了一个基于物理的重定向器OATH。OATH在每秒参考时间的计算下以0.4秒的速度运行，实现了2.4倍于真实时间的性能，并与基于采样的方法相比，计算成本降低了60-350倍。该系统在保持小于2厘米的跟踪误差的同时，成功恢复了58%的保留和54%的未见物体DexYCB轨迹。

View on arXiv Download PDF AI Translation

cs.RO / 31 / 2610.14406

Memory Without Curation: Learning Multi-Stream Latent Memory for Robot Policies

无策展的记忆：学习多流潜在记忆以优化机器人策略

Wu, Yilin, Shah, Rutav, Huang, Shuaiyi, Sontakke, Sumedh A., Bharadhwaj, Homanga, Bajcsy, Andrea, Simchowitz, Max, Shi, Guanya

Abstract

It is widely appreciated that robotic policies require long-term memory to perform complex tasks. However, directly conditioning policies on all past observations leaves policy learning susceptible to learning spurious shortcuts, and introduces large latency at policy deployment. As such, today's memory-augmented policies curate what to remember, either by retaining only recent frames, selected keyframes, or textual summaries, or by learning latent representations of the visual stream alone. In this work, we propose an alternative approach IMPRINT, a policy-agnostic latent memory framework that augments vision-language-action (VLA) models with multi-stream memory without explicitly constraining what the policy should remember. IMPRINT compresses different streams into a latent memory that is updated incrementally at inference, enabling low latency. To avoid shortcut learning, IMPRINT adopts a two-stage training procedure, first supervising memory to reconstruct multiple streams (observations, actions, proprioception) of past information, as well as estimate current task progress, followed by end-to-end finetuning, jointly with the policy, on action prediction. Across both simulation and real-world tasks demanding memory, IMPRINT improves average success over the base VLA by at least 26%, outperforming popular alternatives, and maintaining lowest inference latency among memory-augmented VLAs. Through careful ablations, we demonstrate the utility of incorporating diverse information streams into robotic memory, validating the advantages of foregoing history curation. Project website: https://yilin-wu98.github.io/imprint_website/.

Chinese Translation

人们普遍认为，机器人策略在执行复杂任务时需要长期记忆。然而，直接将策略与所有过去的观察结果进行条件化，会使策略学习容易受到虚假捷径的影响，并在策略部署时引入较大的延迟。因此，当前的增强记忆策略会策划记忆内容，通常只保留最近的帧、选定的关键帧或文本摘要，或者仅学习视觉流的潜在表示。在本研究中，我们提出了一种替代方法IMPRINT，这是一种与策略无关的潜在记忆框架，通过多流记忆增强视觉-语言-动作（VLA）模型，而不明确限制策略应记住的内容。IMPRINT将不同的流压缩为潜在记忆，并在推理时逐步更新，从而实现低延迟。为了避免捷径学习，IMPRINT采用了两阶段训练程序，首先监督记忆重建过去信息的多个流（观察、动作、自我感知），并估计当前任务进展，随后与策略共同进行端到端微调，专注于动作预测。在需要记忆的模拟和现实任务中，IMPRINT在基础VLA的平均成功率上提高了至少26%，超越了流行的替代方案，并在增强记忆的VLA中保持最低的推理延迟。通过细致的消融实验，我们展示了将多样化信息流纳入机器人记忆的实用性，验证了放弃历史策展的优势。项目网站：https://yilin-wu98.github.io/imprint_website/.

View on arXiv Download PDF AI Translation

cs.RO / 32 / 2610.14415

Red-Teaming of Vision-Language-Action Models via Quality Diversity Environment Generation

通过质量多样性环境生成对视觉-语言-行动模型进行红队测试

Zhao, Shihan, Chen, Henry, Hsu, Ya-Chuan, Hwang, Minjune, Srikanth, Siddharth, Liang, Freddie, Saran, Akanksha, Tabrez, Aaquib, Nikolaidis, Stefanos, Seita, Daniel

Abstract

A long-standing goal in robotics is to develop generalist robots that can perform tasks across a wide range of settings, flexibly specified with images and text. Vision-Language-Action (VLA) models offer a promising path toward this goal by training on large collections of robot demonstrations. However, while VLAs perform well in-distribution, they often fail under minor test-time shifts, such as changes in object positions, lighting, textures, and camera poses. To address this, we propose Quality Diversity for Diverse Environment Generation (Q-DEG), a method for improving VLA robustness by generating adversarial environments and fine-tuning the VLA on its own successful rollouts. Q-DEG uses Quality Diversity (QD) optimization to find a set of high-failure environments, while promoting diversity using policy embeddings extracted from the VLA. These environments expose vulnerabilities that can be mitigated through self-imitation fine-tuning, thereby improving the VLA's robustness. In simulation, we conduct experiments across three representative VLA models, and our results indicate that Q-DEG improves robustness to out-of-distribution (OOD) adversarial environments compared to the base VLA and VLAs fine-tuned using data collected from two red-teaming baselines. Furthermore, we demonstrate that improving VLA robustness through Q-DEG facilitates sim-to-real transfer. Our anonymous project website is at: https://qd-env.github.io/qdeg/.

Chinese Translation

机器人技术的一个长期目标是开发能够在广泛环境中执行任务的通用机器人，这些任务可以通过图像和文本灵活指定。视觉-语言-行动（VLA）模型通过在大量机器人演示上进行训练，为实现这一目标提供了有希望的途径。然而，尽管VLA在分布内表现良好，但在测试时的轻微变化（例如物体位置、光照、纹理和相机姿态的变化）下，它们往往会失败。为了解决这个问题，我们提出了质量多样性用于多样化环境生成（Q-DEG）的方法，通过生成对抗性环境并在VLA自身成功的回放上进行微调，从而提高VLA的鲁棒性。Q-DEG利用质量多样性（QD）优化来寻找一组高失败率的环境，同时使用从VLA提取的策略嵌入促进多样性。这些环境暴露了可以通过自我模仿微调来缓解的脆弱性，从而提高VLA的鲁棒性。在仿真中，我们对三个代表性的VLA模型进行了实验，结果表明，与基础VLA和使用从两个红队基线收集的数据进行微调的VLA相比，Q-DEG提高了对分布外（OOD）对抗性环境的鲁棒性。此外，我们证明，通过Q-DEG提高VLA鲁棒性有助于实现从仿真到现实的转移。我们的匿名项目网站地址为：https://qd-env.github.io/qdeg/.

View on arXiv Download PDF AI Translation

cs.RO / 33 / 2610.14478

Shared-Control Teleoperation of Omnidirectional Aerial Robots for Continuous Valve Rotation

全向无人机的共享控制遥操作用于连续阀门旋转

Li, Jinjie, Chen, Yicheng, Liu, Haokun, Zhao, Moju

Abstract

Omnidirectional aerial robots offer full 6-DoF independent control over position and orientation, making them popular for aerial manipulation. Although there have been advancements in robotic autonomy, human operation remains essential in complex aerial environments. Previously, we proposed an aerial teleoperation system that brings the rotational flexibility of human hands into the unbounded aerial workspace, which was evaluated through a vertically mounted valve-turning task. However, due to the limited range of human forearm's motion, the robot could not continuously rotate the valve, hindering its application in industry. In this work, we propose an approach to initiate continuous rotation when the forearm angle exceeds a threshold. This switch of control authority exemplifies the concept shared control from human-robot interaction, marking a further step in aerial teleoperation.

Chinese Translation

全向无人机提供对位置和方向的完全6自由度独立控制，使其在空中操作中备受欢迎。尽管机器人自主性已有所提升，但在复杂的空中环境中，人类操作仍然至关重要。此前，我们提出了一种空中遥操作系统，将人类手部的旋转灵活性引入无界的空中工作空间，并通过垂直安装的阀门旋转任务进行了评估。然而，由于人类前臂运动范围的限制，机器人无法持续旋转阀门，限制了其在工业中的应用。在本研究中，我们提出了一种方法，当前臂角度超过阈值时启动连续旋转。这种控制权的切换体现了人机交互中的共享控制概念，标志着空中遥操作的进一步发展。

View on arXiv Download PDF AI Translation

cs.RO / 34 / 2610.14482

From Field to Planning: Learning Temporal Polar Gradient Fields for Vision-based AAV Trajectory Planning in Dynamic Environments

从场到规划：在动态环境中基于视觉的自主空中载具轨迹规划的时间极坐标梯度场学习

Gao, Haowen, Chen, Zhan, Xu, Yinuo, Niu, Yifeng, Wang, Mengyun, Yan, Chao

Abstract

Autonomous Aerial Vehicle (AAV) trajectory planning requires real-time and accurate safety gradient guidance, especially in dynamic environments. Existing methods typically rely on static maps and dynamic obstacle tracking to construct gradient fields, imposing heavy computational burdens and yielding conservative trajectories. To address these limitations, we propose a two-stage Field-to-Planning (F2P) learning framework to provide accurate and efficient safety gradients for trajectory planning in dynamic environments. In the first stage, we introduce an Ego-centric Polar Occupancy (EPO) representation that naturally matches the geometry of the onboard camera frustum, while supporting unstructured scene modeling. Subsequently, we formulate a Temporal Polar Gradient Field (TPGF) via multi-kernel separable Gaussian convolution to provide unified safety gradients, without explicitly distinguishing between static and dynamic obstacles. To mitigate the inherent spatial ambiguity introduced by applying convolutions within the polar grid, we propose an ESDF-guided offline learning strategy that incorporates geometric priors into TPGF. In the second stage, we employ the calibrated TPGF to guide the trajectory planning policy in a self-supervised learning manner. Comparative results with other approaches demonstrate the superiority of our method. Extensive real-world experiments within various cluttered dynamic environments validate the real-time performance and robustness on resource-constrained onboard systems.

Chinese Translation

自主空中载具（AAV）轨迹规划需要实时且准确的安全梯度指导，尤其是在动态环境中。现有方法通常依赖静态地图和动态障碍物跟踪来构建梯度场，这带来了较大的计算负担，并产生保守的轨迹。为了解决这些局限性，我们提出了一种两阶段的场到规划（Field-to-Planning，F2P）学习框架，以提供准确且高效的安全梯度用于动态环境中的轨迹规划。在第一阶段，我们引入了一种以自我为中心的极坐标占用（Ego-centric Polar Occupancy，EPO）表示，该表示自然匹配机载摄像头视锥的几何形状，并支持非结构化场景建模。随后，我们通过多核可分离高斯卷积构建了时间极坐标梯度场（Temporal Polar Gradient Field，TPGF），以提供统一的安全梯度，而无需明确区分静态和动态障碍物。为了减轻在极坐标网格中应用卷积所引入的固有空间模糊性，我们提出了一种基于扩展距离场（ESDF）引导的离线学习策略，将几何先验融入TPGF。在第二阶段，我们利用校准后的TPGF以自我监督学习的方式指导轨迹规划策略。与其他方法的比较结果表明了我们方法的优越性。在各种杂乱的动态环境中进行的大量真实世界实验验证了在资源受限的机载系统上的实时性能和鲁棒性。

View on arXiv Download PDF AI Translation

cs.RO / 35 / 2610.14492

Multi-modal Representation Fusion for Long-term Microsurgical Anastomosis Trajectory Anticipation

多模态表示融合用于长期显微外科吻合术轨迹预测

Liu, Yuxuan, Luo, Yating, Luan, Yunfei, Zhou, Xinyao, Zhuge, Yuyang, Guo, Yao, Yang, Guang-Zhong

Abstract

Microsurgical anastomosis is one of the most challenging and sophisticated procedures for surgical robotics, requiring accurate manipulation of surgical instruments within a limited space to achieve precise operations. Accurate surgical trajectory anticipation is essential for surgical scene understanding, intraoperative planning, robotic manipulation, and reliable behaviour analysis. However, existing approaches mainly focus on features extracted from surgical videos and historical trajectories, while overlooking rich semantic and motion information that can provide more discriminative multi-modal representations for complex surgical instrument-tissue interactions. In this work, we propose a cross-modal feature fusion framework for long-term 2D and 3D microsurgical anastomosis trajectory anticipation. The proposed framework consists of a multi-modal representation extractor and a cross-modal feature fusion network. Semantic recognition of surgical objects and consecutive motion estimation are incorporated as multi-modal visual cues, providing complementary spatial and temporal guidance for instrument-tissue interaction characterization. A coarse-to-fine progressive fusion strategy with a long-short-term memory mechanism is introduced to capture temporal dependencies step by step. The training and evaluation are conducted on our established robotic platform and a comprehensive dataset containing both phantom and ex-vivo surgical scenarios is constructed. Extensive quantitative and qualitative experiments demonstrate that the proposed framework consistently outperforms existing methods and exhibits improved robustness across different surgical scenes, highlighting its potential for both intelligent robot-assisted and autonomous surgical systems.

Chinese Translation

显微外科吻合术是外科机器人中最具挑战性和复杂性的手术之一，需要在有限的空间内准确操作外科器械以实现精确的手术。准确的手术轨迹预测对于手术场景理解、术中规划、机器人操作和可靠的行为分析至关重要。然而，现有的方法主要集中于从手术视频和历史轨迹中提取特征，而忽视了丰富的语义和运动信息，这些信息可以为复杂的外科器械与组织的交互提供更具区分性的多模态表示。在本研究中，我们提出了一种用于长期二维和三维显微外科吻合术轨迹预测的跨模态特征融合框架。该框架由多模态表示提取器和跨模态特征融合网络组成。手术物体的语义识别和连续运动估计被纳入作为多模态视觉线索，为器械与组织交互特征提供互补的空间和时间指导。引入了一种粗到细的渐进融合策略，并结合长短期记忆机制，逐步捕捉时间依赖关系。训练和评估在我们建立的机器人平台上进行，并构建了一个包含幻影和体外手术场景的综合数据集。大量的定量和定性实验表明，所提出的框架在不同的手术场景中始终优于现有方法，并表现出更强的鲁棒性，突显其在智能机器人辅助和自主外科系统中的潜力。

View on arXiv Download PDF AI Translation

cs.RO / 36 / 2610.14522

Safety-Critical Quadrotor Navigation Using Fourth-Order Control Barrier Functions with Full Nonlinear Quadrotor Dynamics

基于四阶控制障碍函数的全非线性四旋翼动力学安全关键导航

Liu, Yanze, İdikut, Yaşar, Agrawal, Dhruv Madhusudan, Xiao, Wei, Li, Guanrui

Abstract

Safety-critical quadrotor navigation in cluttered environments typically relies on hierarchical architectures in which a high-level safety filter modifies position references that are then tracked by a separate attitude controller. However, this separation can create a mismatch between translational safety objectives and attitude evolution. To address this gap, we propose a safety controller based on fourth-order control barrier functions (CBFs) that explicitly incorporates the full nonlinear quadrotor dynamics, including attitude dynamics and torque-level actuation. Furthermore, we apply the integral high-order CBF technique to resolve the mixed-relative degree issue in the derived constraint formulation. The resulting formulation enforces obstacle-avoidance constraints while optimizing control torques, granting the safety layer direct authority over attitude behavior and eliminating the hierarchical disconnect inherent in reference-tracking safety filters. We validate the approaches in both high-fidelity simulation and real-world flight experiments, demonstrating reliable constraint satisfaction and successful obstacle avoidance. These results suggest that torque-level high-order CBFs provide a practical pathway to provably safer quadrotor navigation under full rigid-body dynamics.

Chinese Translation

在复杂环境中，安全关键的四旋翼导航通常依赖于分层架构，其中高层安全过滤器修改位置参考，然后由单独的姿态控制器进行跟踪。然而，这种分离可能导致平移安全目标与姿态演变之间的不匹配。为了解决这一问题，我们提出了一种基于四阶控制障碍函数（CBFs）的安全控制器，该控制器明确地结合了完整的非线性四旋翼动力学，包括姿态动力学和扭矩级驱动。此外，我们应用积分高阶 CBF 技术来解决所推导约束公式中的混合相对度问题。最终的公式在优化控制扭矩的同时强制执行避障约束，使安全层对姿态行为拥有直接控制权，消除了参考跟踪安全过滤器固有的分层断开。我们在高保真仿真和实际飞行实验中验证了这些方法，证明了可靠的约束满足和成功的避障。这些结果表明，扭矩级高阶 CBF 提供了一条在完整刚体动力学下实现可证明更安全的四旋翼导航的实用途径。

View on arXiv Download PDF AI Translation

cs.RO / 37 / 2610.14530

Who to Trust in Multi-Robot Systems: Joint Robot and Object Hypothesis Trust Inference via Neural-Symbolic Cross-Validation

在多机器人系统中信任谁：通过神经符号交叉验证进行机器人与对象假设信任推断

Wang, Ruiyang, Hsu, Hao-Lun, Meng, Haocheng, Xie, Songqiao, Pajic, Miroslav

Abstract

Multi-robot systems (MRS) offer significant advantages over single-robot platforms by enabling information sharing and coordination for different tasks in complex environments. However, reliable cooperation critically depends on inferring trust over both information sources and their reported hypotheses, particularly under partial observability and potential adversarial interference. Existing approaches frequently depend on pre-established trust relationships or side-channel information, which may not be available or reliable in realistic MRS deployments. This work introduces NeST-Bayes, a neural-symbolic framework that jointly infers robot and hypothesis trust using only inter-robot observation consistency, without requiring preassigned robot trust, reputation information, or side-channel identity signals. NeST-Bayes infers instantaneous trust evidence scores for robots and object hypotheses by reasoning over ego-centric relational graphs constructed from cross-robot observation consistency. These evidence scores are accumulated over time through Bayesian updates of Beta-distributed trust beliefs to maintain temporally consistent trust estimates. Extensive evaluations show that NeST-Bayes achieves the highest mean object precision and robot recall across all benchmark settings, and the highest robot precision and object recall in most settings, compared to trust-based methods. Under aggressive optimized attacks, it achieves 98.8\% robot precision with 93.8\% recall and 98.3\% object precision with 95.4\% recall. In an occlusion-rich Webots warehouse, it maintains 92.5\% robot precision with 92.0\% recall and 97.3\% object precision with 85.0\% recall. These results demonstrate joint improvements in the reliability and coverage of retained information.

Chinese Translation

多机器人系统（MRS）相较于单机器人平台具有显著优势，能够在复杂环境中实现信息共享和任务协调。然而，可靠的合作在很大程度上依赖于对信息源及其报告假设的信任推断，特别是在部分可观测性和潜在对抗干扰的情况下。现有方法通常依赖于预先建立的信任关系或侧信道信息，而这些在现实的 MRS 部署中可能不可用或不可靠。本研究提出了 NeST-Bayes，这是一种神经符号框架，通过仅利用机器人间观察一致性来联合推断机器人和假设信任，而无需预先分配的机器人信任、声誉信息或侧信道身份信号。NeST-Bayes 通过对从跨机器人观察一致性构建的自我中心关系图进行推理，推断出机器人和对象假设的瞬时信任证据分数。这些证据分数通过对 Beta 分布的信任信念进行贝叶斯更新，随着时间的推移进行累积，以保持时间上的一致信任估计。广泛的评估表明，与基于信任的方法相比，NeST-Bayes 在所有基准设置中实现了最高的平均对象精度和机器人召回率，并在大多数设置中实现了最高的机器人精度和对象召回率。在激进的优化攻击下，它实现了 98.8\% 的机器人精度和 93.8\\% 的召回率，以及 98.3\\% 的对象精度和 95.4\\% 的召回率。在一个遮挡丰富的 Webots 仓库中，它保持了 92.5\\% 的机器人精度和 92.0\\% 的召回率，以及 97.3\\% 的对象精度和 85.0\\% 的召回率。这些结果表明在保留信息的可靠性和覆盖率方面的联合改进。

View on arXiv Download PDF AI Translation

cs.RO / 38 / 2610.14541

GG-Nav: Goal-oriented Grounding Chain-of-Thought Elicits Reasoning in Navigation Foundation Models

GG-Nav：目标导向的推理链条激发导航基础模型中的推理能力

Zhu, Shaohao, Huang, Wensi, Zhang, Siqi, Yang, Yuqiang, Wei, Meng, Wan, Chenyang, Liu, Tianle, Wang, Tai, Xu, Jinming

Abstract

Emerging strong Vision-Language Models (VLMs) have advanced navigation foundation models. However, these models still have unsatisfactory performance on goal-oriented navigation settings. In this paper, we first conduct a motivating study revealing that the key issue lies in the insufficient utilization of streaming observations and imprecise target localization. To tackle these challenges, we propose GG-Nav, a simple yet effective navigation framework that learns a structured chain-of-thought (CoT) mechanism via supervised fine-tuning to explicitly localize goal-relevant instances, accumulate grounding evidence, and support more reliable decision-making. Specifically, the structured CoT is designed with two key stages: (i) Observation-to-Evidence, which reasons over all intermediate frames between adjacent decision steps to retain useful goal-related evidence; and (ii) Evidence-to-Decision, which integrates the current observation with accumulated evidence to ground the target and support the generation of navigation actions or clarification questions. Extensive experiments show that GG-Nav outperforms the state-of-the-art methods in terms of success rates on the HM3D ObjectNav, VL-LN, and CoIN benchmarks by 3.1%, 16.6%, and 5.3%, respectively. Ablation studies further show that accumulating evidence from intermediate observations reduces evidence-missing failures by 11.6%, while explicit grounding reduces stopping errors by 49.4%. We also deploy GG-Nav on a Unitree Go2 robot, demonstrating its zero-shot generalization capability in unseen real-world environments.

Chinese Translation

新兴的强大视觉-语言模型（VLMs）推动了导航基础模型的发展。然而，这些模型在目标导向的导航设置中仍表现不佳。本文首先进行了一项激励研究，揭示了关键问题在于对流式观测的利用不足和目标定位不准确。为了解决这些挑战，我们提出了GG-Nav，一个简单而有效的导航框架，通过监督微调学习结构化的推理链条（CoT）机制，以明确定位与目标相关的实例，积累基础证据，并支持更可靠的决策。具体而言，结构化的CoT设计包含两个关键阶段：（i）观察到证据（Observation-to-Evidence），该阶段对相邻决策步骤之间的所有中间帧进行推理，以保留有用的与目标相关的证据；（ii）证据到决策（Evidence-to-Decision），该阶段将当前观测与积累的证据整合，以确定目标并支持生成导航动作或澄清问题。大量实验表明，GG-Nav在HM3D ObjectNav、VL-LN和CoIN基准测试中的成功率分别比最先进的方法提高了3.1%、16.6%和5.3%。消融研究进一步表明，从中间观测中积累证据将证据缺失的失败率降低了11.6%，而明确的基础定位则将停顿错误降低了49.4%。我们还在Unitree Go2机器人上部署了GG-Nav，展示了其在未见过的真实环境中的零样本泛化能力。

View on arXiv Download PDF AI Translation

cs.RO / 39 / 2610.14551

Before We Act: Learning Action-Conditioned Team-Response Beliefs for Decentralized Multi-Robot Manipulation

行动前的思考：学习去中心化多机器人操作的行动条件团队响应信念

Li, Zihao, Zhou, Yanan, Qiu, Ranpeng, Ren, Guoqiang, Zhi, Weiming

Abstract

In multi-robot manipulation, an action's value depends on how teammates respond. We present \method{} (Cooperative Action--Response Belief Estimation), a framework for learning \emph{team-response beliefs} from local history. A predictive belief encoder represents the interaction and conditions a nominal action. Matched reactive and replay continuations separate each candidate's direct advantage from the utility contributed by teammate adaptation. A candidate-conditioned readout learns this response distribution alongside total advantage, and calibrated selection chooses an alternative or retains the nominal action. Our analysis characterizes when response changes the preferred action and how prediction and source errors limit the recoverable gain. With the representation, proposals, and data fixed, explicit response supervision improves candidate ranking and reduces selection regret relative to learning the same information through branch utilities. The benefit is concentrated in decisions where teammate adaptation changes the preferred action. Across RoboFactory, the MARS Control Track, and physical collaboration tasks, CARE achieves the highest average success among the evaluated policies, combining a capable nominal policy with selective coordination adjustments. Together, these results show that CARE turns predictions of teammate adaptation into effective decentralized action selection. Project page: https://jeong-zju.github.io/care/.

Chinese Translation

在多机器人操作中，行动的价值取决于队友的响应。我们提出了 extit{方法}（Cooperative Action--Response Belief Estimation），这是一个从局部历史中学习 extit{团队响应信念}的框架。预测信念编码器表示交互并对名义行动进行条件化。匹配的反应和重放延续将每个候选者的直接优势与队友适应所贡献的效用分开。候选条件的读出学习这一响应分布以及总优势，而校准选择则选择替代方案或保留名义行动。我们的分析描述了何时响应会改变首选行动，以及预测和源错误如何限制可恢复的收益。在固定表示、提案和数据的情况下，显式响应监督提高了候选排名，并减少了相对于通过分支效用学习相同信息的选择遗憾。这一好处集中在队友适应改变首选行动的决策上。在RoboFactory、MARS控制轨道和物理协作任务中，CARE在评估的策略中实现了最高的平均成功率，结合了强大的名义策略与选择性协调调整。这些结果表明，CARE将队友适应的预测转化为有效的去中心化行动选择。项目页面：https://jeong-zju.github.io/care/

View on arXiv Download PDF AI Translation

cs.RO / 40 / 2610.14561

From Video to Vision: Learning Generalizable Humanoid Carrying from Human Motion

从视频到视觉：学习可推广的人形机器人物体搬运能力

Ding, Ziluo

Abstract

Most existing approaches rely on video or motion-capture data for reinforcement learning (RL) training; however, all of these methods require an upper-level module to generate reference trajectories in real time. In object carrying tasks, the humanoid can only lift objects from designated locations along the reference trajectory, resulting in poor spatial generalization. Moreover, existing policies can only be evaluated inside a motion-capture studio. To address these limitations, we propose \textbf{MOVER}, a multi-stage, Teacher-Assistant-Student, RL training framework for humanoid object carrying. We first collect data from videos and employ GENMO and GMR to retarget the video data, combining this with data augmentation techniques to generalize the data across objects of varying shapes and sizes. In the \textbf{first stage}, building upon the DeepMimic \cite{peng2018deepmimic} motion-training framework, we train the humanoid to transport objects from specified locations along designated trajectories. In the \textbf{second stage}, we relax the tracking of the reference motion and adjust the reward so that the humanoid learns to transport objects from different positions. In the \textbf{third stage}, we apply a Dagger-PPO distillation method to distill away the proprioceptive position observations and reference joint position of the humanoid. This allows the deployed policy to operate without relying on global position information from a motion-capture system; instead, the humanoid can directly perceive the object's pose from its onboard RGB-D camera to complete the carrying task. We design a series of simulation experiments to compare the success rates across stages (Stages 1 vs. 2) and before versus after distillation. We further transfer the policy to a real humanoid, successfully transporting objects at varying positions in a motion-capture-free setting.

Chinese Translation

大多数现有方法依赖于视频或动作捕捉数据进行强化学习（RL）训练；然而，所有这些方法都需要一个上层模块实时生成参考轨迹。在物体搬运任务中，人形机器人只能沿着参考轨迹从指定位置提起物体，这导致空间泛化能力差。此外，现有策略只能在动作捕捉工作室内进行评估。为了解决这些限制，我们提出了 extbf{MOVER}，一个多阶段的教师-助手-学生强化学习训练框架，用于人形机器人物体搬运。我们首先从视频中收集数据，并采用GENMO和GMR对视频数据进行重新定向，结合数据增强技术，使数据能够在不同形状和大小的物体之间进行泛化。在 extbf{第一阶段}，基于DeepMimic extit{(peng2018deepmimic)} 动作训练框架，我们训练人形机器人沿着指定轨迹从特定位置搬运物体。在 extbf{第二阶段}，我们放宽对参考动作的跟踪，并调整奖励，使人形机器人学习从不同位置搬运物体。在 extbf{第三阶段}，我们应用Dagger-PPO蒸馏方法，去除对人形机器人本体位置观察和参考关节位置的依赖。这使得部署的策略能够在不依赖动作捕捉系统的全局位置信息的情况下运行；相反，人形机器人可以通过其机载RGB-D相机直接感知物体的姿态以完成搬运任务。我们设计了一系列仿真实验，以比较各阶段（阶段1与阶段2）之间的成功率，以及蒸馏前后的表现。我们进一步将策略转移到真实的人形机器人上，成功地在无动作捕捉环境中从不同位置搬运物体。

View on arXiv Download PDF AI Translation

cs.RO / 41 / 2610.14564

Chunk-wise Model-based Policy Optimization

基于模型的分块策略优化

Song, Weixi, Wang, Donglin, Li, Yong-Lu

Abstract

Model-based reinforcement learning has shown superior sample efficiency by leveraging the learned latent dynamics either in the form of derivative-free planning/imagined rollout training or dynamics gradient-based policy updates. While these approaches achieve comparatively fast learning, they all suffer, to varying degrees, from issues of recurrent compounding errors, high inference-time latency, and instability in backward propagation through time (BPTT). In this work, we propose a model-based RL approach that uses the chunk world model as a differentiable training-time evaluator to optimize the chunk actor without recurrent model invocation. Through the collaborative chunk-wise prediction of the world model and policy, we can generate H-step hypothetical rollouts in a single forward pass and directly optimize value expansion, avoiding the costly and unstable BPTT. We evaluate our method on 19 challenging high-dimensional humanoid tasks from HumanoidBench and deploy our method to the real-world for quadruped bipedal walking. Project Website: https://song-wx.github.io/cmpo.github.io/

Chinese Translation

基于模型的强化学习通过利用学习到的潜在动态，展示了优越的样本效率，具体表现为无导数规划/想象回滚训练或基于动态梯度的策略更新。尽管这些方法实现了相对快速的学习，但它们在不同程度上都面临着递归复合误差、高推理时延和时间反向传播（BPTT）不稳定性的问题。在本研究中，我们提出了一种基于模型的强化学习方法，该方法使用分块世界模型作为可微分的训练时评估器，以优化分块演员，而无需递归调用模型。通过世界模型和策略的协作分块预测，我们能够在单次前向传递中生成H步假设回滚，并直接优化价值扩展，避免了代价高昂且不稳定的BPTT。我们在HumanoidBench的19个具有挑战性的高维人形任务上评估了我们的方法，并将其应用于真实世界的四足双足行走。项目网站：https://song-wx.github.io/cmpo.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 42 / 2610.14577

In View or Out of View? A Vision-Aware Filter that Infers and Reacts to the Human's Field of View

在视野内还是视野外？一种能够推断并响应人类视野的视觉感知过滤器

Fu, Mengxue, Hsu, Ya-Chuan, Dai, Yinlong, Bıyık, Erdem, Losey, Dylan P.

Abstract

During human-robot interaction, sight plays an important role. When the human sees a robot, they can observe the robot's actions and better predict the robot's future behaviors. Conversely, when the human cannot see the robot, they become uncertain about the robot's intent and plans. In this paper, we focus on the human's field of view. Existing human-aware planners either ignore this field of view and assume the human can observe the entire scene, or intentionally design their interactions so the robot is always visible to the human. By contrast, we enable robots to infer what the human has and has not seen during interaction, i.e., the human's field of view. We then apply this estimate as a vision-aware filter on the robot's actions. Given a nominal robot policy and a model of the interaction, we roll out the policy over a short horizon to see how the robot's state updates. We then reweight the actions based on how those actions impact the robot's predicted visibility: higher weights are assigned to actions that bring the robot into (or away from) the human's line of sight. In practice, our framework provides an online estimation and filtering process that repeatedly updates its understanding of the human's field of view and then locally perturbs robot actions based on the human's vision. We evaluate our vision-aware filter on robots that assist humans in collaborative cooking and coordinate with them in driving environments. Across simulations and a 12-participant user study, we find that our vision-aware filter changes both performance and preference: participants completed over three times as many subtasks with in-view assistance while cooking, and had fewer collisions with cars that purposely stayed out of their view while driving.

Chinese Translation

在人机交互中，视觉发挥着重要作用。当人类看到机器人时，他们可以观察机器人的动作，并更好地预测机器人的未来行为。相反，当人类看不见机器人时，他们对机器人的意图和计划感到不确定。本文重点关注人类的视野。现有的人类感知规划者要么忽略这一视野，假设人类可以观察到整个场景，要么故意设计其交互，使机器人始终对人类可见。相比之下，我们使机器人能够推断人类在交互过程中所见与未见的内容，即人类的视野。然后，我们将这一估计应用于机器人的动作作为一种视觉感知过滤器。给定一个标准的机器人策略和交互模型，我们在短期内展开该策略，以观察机器人的状态如何更新。然后，我们根据这些动作对机器人的预测可见性影响重新加权：对那些使机器人进入（或远离）人类视线的动作赋予更高的权重。实际上，我们的框架提供了一种在线估计和过滤过程，持续更新对人类视野的理解，并根据人类的视觉局部扰动机器人动作。我们在协作烹饪中帮助人类的机器人以及在驾驶环境中与他们协调的机器人上评估了我们的视觉感知过滤器。在模拟和12名参与者的用户研究中，我们发现我们的视觉感知过滤器改变了性能和偏好：参与者在烹饪时完成了三倍以上的视野内协助子任务，并在驾驶时与故意保持在其视野之外的汽车发生的碰撞更少。

View on arXiv Download PDF AI Translation

cs.RO / 43 / 2610.14583

Remembering What Matters for Long-Horizon Robotic World Models

记住长时间范围机器人世界模型中的重要信息

Wu, Hao, Shao, Tianyu, Qian, Shengju, Zhao, Yizhou, Zhang, Fan, Zhao, Penghao, Wang, Weiyan, Huang, Yufei, Xu, Fan, Wang, Kun, Wang, Xin, Yang, Ming-Hsuan, Liang, Yuxuan

Abstract

Autoregressive robot video world models based on discrete visual tokens remain unreliable over long rollout horizons. Retaining the entire generated history does not necessarily improve prediction quality. We investigate a possible memory-role conflict in Full Context decoding: persistent scene information, recent visual-action states, and stale generated history compete for attention despite their different predictive roles. This perspective motivates ReCAP (Receding Context with Anchor Preservation), a training-free context-scheduling method for long-horizon robot video world models. ReCAP keeps the pretrained tokenizer, world model, and visual decoder frozen. It preserves the initial scene anchor and a contiguous window of recent complete visual-action blocks while evicting older generated history. The schedule preserves within-block structure and temporal order, requires no additional modules or parameter updates, and bounds active-context growth independently of rollout length. We evaluate ReCAP on RT-1, BridgeV2, LIBERO-90, and CALVIN using frozen autoregressive world models. Across evaluated environments and horizons, ReCAP improves visual fidelity, action-conditioned consistency, and robot structure preservation relative to Full Context. On BridgeV2, it reduces LPIPS by 63.3%. In the system benchmark at H=64, it reduces the logical KV footprint by 73.3% and p50 latency by 6.51%. Attention analyses show increased attention to retained recent states, while history interventions associate larger amounts of stale context with higher prediction error. These observations are consistent with the memory-role-conflict hypothesis, although they do not isolate it from context-length and position-handling effects. Our results support inference-time context scheduling as a practical approach to improving long-horizon robot video prediction and computational efficiency without retraining.

Chinese Translation

基于离散视觉标记的自回归机器人视频世界模型在长时间展开的情况下仍然不可靠。保留整个生成历史并不一定能提高预测质量。我们研究了全上下文解码中的可能记忆角色冲突：持久的场景信息、最近的视觉-动作状态和陈旧的生成历史在不同的预测角色下竞争注意力。这一视角激发了ReCAP（递减上下文与锚点保留），这是一种用于长时间范围机器人视频世界模型的无训练上下文调度方法。ReCAP保持预训练的标记器、世界模型和视觉解码器不变。它保留初始场景锚点和一段连续的最近完整视觉-动作块，同时驱逐较旧的生成历史。该调度保留了块内结构和时间顺序，不需要额外的模块或参数更新，并且独立于展开长度限制活动上下文的增长。我们在RT-1、BridgeV2、LIBERO-90和CALVIN上评估了ReCAP，使用冻结的自回归世界模型。在评估的环境和时间范围内，ReCAP相较于全上下文提高了视觉保真度、动作条件一致性和机器人结构保留。在BridgeV2上，它将LPIPS降低了63.3%。在H=64的系统基准测试中，它将逻辑KV占用减少了73.3%，p50延迟减少了6.51%。注意力分析显示对保留的最近状态的注意力增加，而历史干预则将较大数量的陈旧上下文与更高的预测误差相关联。这些观察结果与记忆角色冲突假说一致，尽管它们并未将其与上下文长度和位置处理效应分离。我们的结果支持推理时上下文调度作为一种实用方法，以提高长时间范围机器人视频预测和计算效率，而无需重新训练。

View on arXiv Download PDF AI Translation

cs.RO / 44 / 2610.14586

PC7 Model: Dynamics of Expanding and Contracting Soft Robots using Piece-wise Constant 7D Deformation Based on Similarity Transformation Group

PC7模型：基于相似变换群的分段常数7D变形的扩展与收缩软机器人动力学

Kudo, Yamato, Ishigaki, Taiki, Ayusawa, Ko, Yamamoto, Ko

Abstract

This paper proposes a novel soft robot model named piece-wise constant 7D deformation (PC7) model that can represent expanding and contracting deformation of, for example, artificial muscle actuator. In the proposed model, we newly define the 7D deformation vector based on SIM(3) - 3D similarity transformation group - whereas a 6D deformation vector was defined for Cosserat rod based on SE(3) - the special Euclid group - in the piece-wise constant strain (PCS) model proposed in the previous study. The additional dimension of the 7D deformation vector represents expanding and contracting deformation, which comes from the scaling parameter in SIM(3). Common mathematical properties between SE(3) and SIM(3) allow us to directly extend the kinematics and dynamics of the PCS model to those of the PC7 model. In particular, we derive the dynamical equation of the PC7 model based on Euler-Poincare equation in a mathematically precise manner. We validate the proposed model through dynamical simulations assuming a thin cylinder structure or McKibben-type artificial muscle actuator.

Chinese Translation

本文提出了一种新颖的软机器人模型，称为分段常数7D变形（PC7）模型，该模型能够表示例如人工肌肉驱动器的扩展和收缩变形。在所提出的模型中，我们基于SIM(3) - 3D相似变换群重新定义了7D变形向量，而在之前的研究中，基于SE(3) - 特殊欧几里得群，为Cosserat杆定义了6D变形向量。7D变形向量的额外维度表示扩展和收缩变形，这源于SIM(3)中的缩放参数。SE(3)和SIM(3)之间的共同数学性质使我们能够将PCS模型的运动学和动力学直接扩展到PC7模型。特别地，我们以数学精确的方式基于欧拉-庞加莱方程推导了PC7模型的动力学方程。我们通过假设薄圆柱结构或McKibben型人工肌肉驱动器的动力学仿真验证了所提出的模型。

View on arXiv Download PDF AI Translation

cs.RO / 45 / 2610.14588

Flow-Support-Weighted Proposals for Test-Time Verified VLA Policies

测试时验证的流支持加权提议用于变换学习策略

Zhao, Chen, Ye, Zipeng, Shen, Haitao, Li, Guanlin, Xie, Yuchen, Zhang, Jing

Abstract

Test-time sampling promises to turn one action prediction into a set of distinct executable candidates that a verifier can choose among. Downstream fine-tuning works against that promise. A vision-language-action model trained to solve one task distribution concentrates its flow-matching decoder around the dominant demonstrated action, so repeated samples differ only marginally and the verifier receives many copies of one decision. We call this proposal contraction and identify it as a key constraint on test-time scaling: repeated samples can remain highly redundant as the candidate budget grows. Action Dynamic Sampling (ADS) alleviates the constraint on the proposal side, during training. At each training step, ADS integrates $K$ additional trajectories from the current policy under fresh base noise, scores each by the flow support along its path with a Hutchinson trace estimator, and corrects the policy's local action distribution toward this weighted set through a detached finite-set KL term. Training cost rises; deployment cost is unchanged. Across simulation and four physical-robot tasks, ADS raises proposal spread without the success penalty that Gaussian perturbation incurs, lifts oracle best-of-8 success on RoboCasa from 57.6% to 63.0%, improves fixed verifiers by 2.4 to 4.9 points in simulation, and raises TokenPB by up to 13.0 points on real-world manipulation.

Chinese Translation

测试时采样承诺将一个动作预测转化为一组可供验证者选择的不同可执行候选项。然而，下游微调与这一承诺相悖。训练以解决一个任务分布的视觉-语言-动作模型，其流匹配解码器集中在主导的示范动作上，因此重复样本之间的差异仅微乎其微，验证者接收到的则是同一决策的多个副本。我们称这种现象为提议收缩，并将其识别为测试时扩展的一个关键约束：随着候选预算的增加，重复样本可能保持高度冗余。动作动态采样（Action Dynamic Sampling, ADS）在训练过程中缓解了提议方面的约束。在每个训练步骤中，ADS从当前策略下的全新基础噪声中整合$K$条额外轨迹，通过Hutchinson迹估计器根据路径上的流支持对每条轨迹进行评分，并通过一个独立的有限集KL项修正策略的局部动作分布，以朝向这一加权集合进行调整。训练成本上升；部署成本保持不变。在模拟和四个物理机器人任务中，ADS在不引入高斯扰动所带来的成功惩罚的情况下，提高了提议的分散性，将RoboCasa的最佳8次成功率从57.6%提升至63.0%，在模拟中使固定验证者的表现提高了2.4到4.9个百分点，并在真实世界操作中将TokenPB提高了最多13.0个百分点。

View on arXiv Download PDF AI Translation

cs.RO / 46 / 2610.14590

ToF-Wheel Odometry: Navigating Degeneration of Localization With Limited Sensing Capabilities

ToF-轮式里程计：在有限感知能力下应对定位退化

Yang, Xu, Li, Wenhao, Wei, Zhengyu, Li, Zening, Huang, Longxiang, Wang, Bo

Abstract

Wheel odometry provides an efficient motion prior for ground robots but is prone to transient violations such as wheel slip or terrain-induced tilt. In filtering-based odometry, these violations frequently trigger divergence when they coincide with constrained geometric observability (e.g., featureless corridors). We address this challenge by proposing ToF-Wheel Odometry (TWO), a tightly-coupled filtering framework that integrates wheel kinematics with geometric constraints from Time-of-Flight (ToF) cameras. The primary contribution of this work is a degeneracy-aware, directional covariance-shaping strategy that replaces heuristic online covariance tuning with geometry-conditioned inflation. The core idea is to infer process noise from geometric corrections with a strict reliability check. We decompose the posterior-to-prior correction into a wheel-explainable subspace and an orthogonal remainder. The process covariance is anisotropically adjusted according to the observability of the ToF update: if the geometry is degenerate in non-kinematic directions, the remainder is suppressed as sensor noise; if the geometry is distinct, it is admitted as a valid kinematic violation. Additionally, we introduce an intra-frame feedback loop that retrospectively repairs the process covariance within the current frame to reduce structural lag during bursty disturbances. Extensive experiments on challenging scenarios demonstrate that TWO improves trajectory accuracy and robustness compared to static and adaptive Gaussian filtering, robustly navigating the trade-off between kinematic reliability and limited sensing capabilities.

Chinese Translation

轮式里程计为地面机器人提供了高效的运动先验，但容易受到瞬态干扰的影响，例如轮滑或地形引起的倾斜。在基于滤波的里程计中，这些干扰常常在与受限几何可观测性（例如，无特征走廊）重合时引发发散。我们通过提出ToF-轮式里程计（TWO）来应对这一挑战，这是一种紧密耦合的滤波框架，将轮子运动学与时间飞行（ToF）相机的几何约束相结合。本文的主要贡献是提出了一种考虑退化的方向性协方差调整策略，替代了启发式的在线协方差调节，采用几何条件的膨胀。核心思想是通过严格的可靠性检查，从几何校正中推断过程噪声。我们将后验到先验的校正分解为一个可由轮子解释的子空间和一个正交余量。过程协方差根据ToF更新的可观测性进行各向异性调整：如果几何在非运动学方向上是退化的，则余量被抑制为传感器噪声；如果几何是独特的，则被视为有效的运动学干扰。此外，我们引入了一个帧内反馈回路，回溯性地修复当前帧内的过程协方差，以减少在突发干扰期间的结构滞后。在具有挑战性的场景中进行的大量实验表明，与静态和自适应高斯滤波相比，TWO提高了轨迹的准确性和鲁棒性，稳健地在运动学可靠性与有限感知能力之间进行权衡。

View on arXiv Download PDF AI Translation

cs.RO / 47 / 2610.14596

Load-Aware Quadruped Navigation with Online Contact-based Geometry Estimation

基于负载感知的四足机器人导航与在线接触几何估计

Yin, Qi, Zhang, XiaoLi, Shi, Yangming

Abstract

Navigating cluttered environments while carrying objects remains a challenging problem for quadruped robots. Compared with works that focused on load-aware locomotion adaptation to maintain gait stability, existing autonomous navigation systems generally ignore the influence of carried objects' geometries and use fixed footprint inflation, leading to overly conservative navigation. To fill this knowledge gap, we propose a two-phase load-aware navigation learning framework. In Phase 1, we train LOADNav, a load-aware navigation policy via reinforcement learning, enabling the robot to adapt its trajectory and clearance behavior according to the geometric size of the carried object. In Phase 2, we train CoGE, an online geometry predictor via imitation learning, which estimates unknown payload geometry from occasional contact events. We evaluate the framework in cluttered environments with loads of varying sizes and shapes. The results show that our navigation policy exhibits clear load-aware behaviors, and the online geometry predictor estimates object geometry from occasional contact events, refining its estimate and adapting the navigation behavior accordingly. Without requiring any prior knowledge of the payload, it narrows the gap to geometry-aware policies.

Chinese Translation

在携带物体的情况下，四足机器人在杂乱环境中导航仍然是一个具有挑战性的问题。与专注于负载感知的运动适应以维持步态稳定性的研究相比，现有的自主导航系统通常忽视了携带物体几何形状的影响，并使用固定的足迹膨胀，这导致了过于保守的导航。为填补这一知识空白，我们提出了一种两阶段的负载感知导航学习框架。在第一阶段，我们通过强化学习训练了LOADNav，一个负载感知导航策略，使机器人能够根据携带物体的几何尺寸调整其轨迹和间隙行为。在第二阶段，我们通过模仿学习训练了CoGE，一个在线几何预测器，它通过偶尔的接触事件估计未知的负载几何。我们在具有不同大小和形状负载的杂乱环境中评估了该框架。结果表明，我们的导航策略表现出明显的负载感知行为，而在线几何预测器能够从偶尔的接触事件中估计物体几何，精细化其估计并相应地调整导航行为。在不需要任何关于负载的先验知识的情况下，它缩小了与几何感知策略之间的差距。

View on arXiv Download PDF AI Translation

cs.RO / 48 / 2610.14601

StateSkill-Agent: An Effective and Efficient Embodied Agent via State-conditioned Skill

StateSkill-Agent：通过状态条件技能实现有效且高效的具身智能体

Wei, Changxu, Zhang, Jingyi, Zheng, Zhiyu, Lin, Tianyi, Zhang, Yixian, Qing, Yunpeng, Liu, Jingyuan, Hou, Zhi, Huang, Jiaxing

Abstract

In this work, we aim to develop an embodied MLLM agent that understands and solves embodied tasks by learning, abstracting, and reusing transferrable and generalizable behavioral patterns. To this end, we propose StateSkill, a new skill representation for embodied agents, which introduces the concept of "finite-state machines" into skill modeling for effective and efficient skill construction and learning. The core idea of StateSkill is to model embodied interaction experiences as a finite-state machine, where a skill is organized around a set of task-progress states, state transitions, and state-conditioned guidance, enabling embodied agents to dynamically adjust their skills as the task progresses while keeping skill contexts compact and stage-relevant. We further develop StateSkill-Evolve, an experience-driven evolution mechanism that iteratively refines StateSkill through retrospective analysis of new successful and failed state-guided trajectories. Upon this, we introduce StateGRPO, which integrates StateSkill evolution and model optimization via state-conditioned GRPO into a unified iterative learning loop, enabling state-conditioned skills and model capabilities to co-evolve and reinforce each other. Extensive experiments demonstrate the effectiveness of our proposed methods and trained model (i.e., StateSkill-Agent) across various embodied tasks.

Chinese Translation

在本研究中，我们旨在开发一个具身的多模态语言模型（MLLM）智能体，该智能体通过学习、抽象和重用可转移和可推广的行为模式来理解和解决具身任务。为此，我们提出了StateSkill，一种新的具身智能体技能表示，它将“有限状态机”的概念引入技能建模，以实现有效和高效的技能构建与学习。StateSkill的核心思想是将具身交互体验建模为有限状态机，其中技能围绕一组任务进展状态、状态转换和状态条件指导进行组织，使得具身智能体能够在任务进展过程中动态调整其技能，同时保持技能上下文的紧凑性和阶段相关性。我们进一步开发了StateSkill-Evolve，这是一种基于经验的进化机制，通过对新的成功和失败的状态引导轨迹进行回顾性分析，迭代地优化StateSkill。在此基础上，我们引入了StateGRPO，它将StateSkill的进化与通过状态条件的GRPO进行模型优化整合到一个统一的迭代学习循环中，使得状态条件技能和模型能力能够共同进化并相互增强。大量实验表明，我们提出的方法和训练模型（即StateSkill-Agent）在各种具身任务中具有有效性。

View on arXiv Download PDF AI Translation

cs.RO / 49 / 2610.14603

AnyRank: Reliable Anytime Policy Ranking via Sequential Testing

AnyRank：通过序贯测试实现可靠的任意时刻策略排名

Badithela, Apurva, Snyder, David, Yin, Tenny, Sceats, Jade, Pappas, George, Matni, Nikolai, Itkina, Masha, Nishimura, Haruki, Majumdar, Anirudha

Abstract

Due to the high cost of hardware experiments, robot policies are predominantly evaluated by comparing their empirical performance on a limited number of hardware trials. However, finite-sample statistical uncertainty can lead to erroneous conclusions, especially when ranking multiple policies using a fixed batch of evaluation data. We propose AnyRank: a sequential method for multi-policy ranking that (i) adaptively allocates trials to hard-to-separate policies, (ii) provides (partial) rankings at any stopping time while bounding the probability of an incorrect ranking, and (iii) optionally exploits prior information to prioritize hardware evaluations more efficiently. This prior information could be informed from expert intuition, a small set of held-out hardware evaluations, or from policy rollouts in a simulator or an action-conditioned video world model. Notably, the prior can be imperfect; while a strong prior can provide efficiency gains on hardware, any inaccuracies in prior policy rankings do not invalidate the statistical assurances of AnyRank. We achieve the aforementioned properties through novel theoretical contributions that combine sequential testing with multiple-hypothesis graphical testing approaches. Experiments in simulation and real-world policy evaluation benchmarks show that AnyRank requires up to 50% fewer hardware evaluations compared to existing statistically rigorous approaches, while guarding against the incorrect rankings produced by heuristic methods used by state-of-the-art benchmarks.

Chinese Translation

由于硬件实验的高成本，机器人策略主要通过比较其在有限数量硬件试验中的经验表现来进行评估。然而，有限样本的统计不确定性可能导致错误的结论，特别是在使用固定批次的评估数据对多个策略进行排名时。我们提出了AnyRank：一种用于多策略排名的序贯方法，该方法(i) 自适应地将试验分配给难以区分的策略，(ii) 在任何停止时间提供（部分）排名，同时限制错误排名的概率，以及(iii) 可选地利用先验信息以更有效地优先进行硬件评估。这些先验信息可以来自专家直觉、一小组保留的硬件评估，或来自模拟器中的策略滚动或动作条件视频世界模型。值得注意的是，先验信息可以是不完美的；虽然强先验可以在硬件上提供效率提升，但先验策略排名中的任何不准确性并不会使AnyRank的统计保证失效。我们通过将序贯测试与多假设图形测试方法相结合的新理论贡献，实现了上述属性。在模拟和真实世界策略评估基准中的实验表明，与现有的统计严谨方法相比，AnyRank所需的硬件评估次数最多减少50%，同时防范了当前最先进基准所使用的启发式方法产生的错误排名。

View on arXiv Download PDF AI Translation

cs.RO / 50 / 2610.14634

VbotNav-0: Asynchronous Cloud-Edge Dual-Policy Navigation across Indoor and Outdoor Environments

VbotNav-0：跨室内和室外环境的异步云边双策略导航

Tan, Shiqi, Zhang, Yujun, Li, Keyu, Pan, Shuaixi, Zhou, Siyu, Qin, Hailong, Yu, Yinan

Abstract

Continuous indoor-outdoor navigation connects destination-level reasoning with motion decisions that must respond to current observations. We present VbotNav-0, a cloud-edge dual-policy framework that links these processes through asynchronous policy-to-policy conditioning. The cloud policy produces subgoal guidance from visual history, navigation instructions, and route tokens; the edge policy combines available guidance with current RGB and LiDAR observations to generate local trajectories. This formulation allows the local perception-action loop to continue between cloud updates. A common route representation connects indoor paths and outdoor map routes, and the training recipe combines navigation demonstrations, corrective data, and grounded visual question answering (VQA). In the reported closed-loop evaluation of 258 episodes across 145 scenes, VbotNav-0 achieves 55.04% success rate and 55.03% success weighted by path length.

Chinese Translation

连续的室内外导航将目的地层级推理与必须响应当前观察的运动决策连接起来。我们提出了VbotNav-0，一种通过异步策略到策略的条件化将这些过程连接起来的云边双策略框架。云策略基于视觉历史、导航指令和路径标记生成子目标指导；边缘策略将可用指导与当前的RGB和LiDAR观察结合，以生成局部轨迹。这种构造允许在云更新之间继续进行局部感知-行动循环。一个通用的路径表示连接室内路径和室外地图路线，训练方案结合了导航示范、纠正数据和基于视觉的问题回答（VQA）。在对145个场景的258个回合的闭环评估中，VbotNav-0实现了55.04%的成功率和按路径长度加权的55.03%的成功率。

View on arXiv Download PDF AI Translation

cs.RO / 51 / 2610.14655

Global Optimality of Pose Graph Optimization

姿态图优化的全局最优性

Ding, Shuan, Chen, Ruishuang, Cao, Mengfan, Yang, Zaiyue

Abstract

Pose Graph Optimization (PGO) is a classical and widely applied problem in state estimation for robotics, playing a fundamental role in localization, mapping, and navigation. Yet, it is nonconvex in nature and may have many local minimizers, thus difficult to analyze. In this paper, we investigate the optimization landscape and establish a sufficient condition of global optimality for PGO. The ultimate goal of this paper is to theoretically confirm that the optimization landscape is benign and attractive around the global minimizer, if noise is smaller than an explicit and a priori bound. To reveal such an insight, this paper develops three main technical results. First, we identify a proper basin of attraction (BOA) containing global minimizer, and characterize its geometric property by showing that its radius is controlled by noise in an explicit way. Then, by exploring the first order optimality conditions of PGO, we prove that a local minimizer is globally optimal if a noise-controlled criterion is satisfied. Finally, by combining above two results, we prove that one and only one local minimizer exists in the BOA and this minimizer is globally optimal, when the noise is smaller than an explicit bound that can be verified a priori. Numerical simulations validate the proposed theory under various random cases, showing that the obtained conditions provide an effective criterion for certifying the global optimality of PGO.

Chinese Translation

姿态图优化（Pose Graph Optimization, PGO）是机器人状态估计中的一个经典且广泛应用的问题，在定位、地图构建和导航中发挥着基础性作用。然而，由于其本质上是非凸的，可能存在许多局部最小值，因此难以分析。本文研究了优化景观，并建立了PGO全局最优性的充分条件。本文的最终目标是理论上确认，当噪声小于一个明确且先验的界限时，优化景观在全局最小值附近是良性且具有吸引力的。为了揭示这一见解，本文发展了三个主要技术结果。首先，我们识别了一个包含全局最小值的适当吸引盆地（Basin of Attraction, BOA），并通过显示其半径以明确的方式受噪声控制来表征其几何属性。然后，通过探索PGO的一阶最优性条件，我们证明了如果满足一个噪声控制标准，则局部最小值是全局最优的。最后，通过结合上述两个结果，我们证明了在噪声小于一个可以先验验证的明确界限时，BOA中存在且仅存在一个局部最小值，并且该最小值是全局最优的。数值仿真在各种随机情况下验证了所提出的理论，显示所获得的条件为证明PGO的全局最优性提供了有效标准。

View on arXiv Download PDF AI Translation

cs.RO / 52 / 2610.14682

Successful but Unsafe: An Analytical Benchmark for Revealing Collisions in Vision-and-Language Navigation

成功但不安全：揭示视觉与语言导航中碰撞的分析基准

Pham, Hieu D., Luong, Quoc-An, Le, Dung D.

Abstract

Vision-and-language navigation (VLN) agents are typically evaluated by whether they reach a target and how efficiently they follow an instructed route, yet these metrics provide little information about collision behavior. Moreover, continuous VLN evaluation commonly uses a sliding regime, where an agent can slide along obstacles after a collision, potentially masking the effect of unsafe actions on navigation performance. We introduce VLN-Safe, a benchmark for analyzing collision safety under two complementary evaluation regimes: sliding, reflecting common VLN evaluation practice, and blocking, where collision-inducing actions prevent forward motion and more directly expose their consequences. Across six published agents on R2R-CE, 38.5-56.8% of successful episodes contain contact under sliding, while switching to blocking decreases success rate by 11.7-36.1 percentage points. We further analyze signals for anticipating imminent collisions and find that depth geometry and recent movement history are more informative than policy-derived signals, while safe alternatives near the policy's preferred action are rarely available. Motivated by these findings, we introduce a training-free collision guard that vetoes risky forward actions, executes an escape maneuver, and returns control to the original policy. Under blocking, the guard reduces collision steps by 62-98% and improves success rate by 6.9-29.3 percentage points across five recurrent, topological, and vision-language-action agents. Under sliding, it leaves success rate unchanged while raising the share of collision-free successes by 4.1-12.4 points. These results reveal a substantial gap between task success and collision-safe navigation and show that lightweight collision-aware intervention can improve both safety and navigation performance.

Chinese Translation

视觉与语言导航（VLN）代理通常通过是否到达目标以及遵循指令路线的效率来评估，然而这些指标对碰撞行为提供的信息有限。此外，持续的VLN评估通常采用滑动模式，在碰撞后，代理可以沿着障碍物滑动，这可能掩盖不安全行为对导航性能的影响。我们引入了VLN-Safe，一个用于分析碰撞安全性的基准，采用两种互补的评估模式：滑动，反映了常见的VLN评估实践，以及阻挡，在这种模式下，导致碰撞的行为会阻止前进，直接暴露其后果。在R2R-CE上对六个已发布的代理进行评估时，38.5%-56.8%的成功事件在滑动模式下存在接触，而切换到阻挡模式则使成功率下降11.7%-36.1个百分点。我们进一步分析了预测即将发生的碰撞的信号，发现深度几何和近期运动历史比基于策略的信号更具信息量，而在策略偏好的行为附近，安全的替代方案很少可用。基于这些发现，我们引入了一种无训练的碰撞保护机制，该机制否决风险较大的前进行为，执行逃逸操作，并将控制权返回给原始策略。在阻挡模式下，该保护机制将碰撞步骤减少62%-98%，并在五个递归、拓扑和视觉-语言-行动代理中提高成功率6.9%-29.3个百分点。在滑动模式下，它保持成功率不变，同时将无碰撞成功的比例提高4.1-12.4个百分点。这些结果揭示了任务成功与碰撞安全导航之间的显著差距，并表明轻量级的碰撞感知干预可以同时改善安全性和导航性能。

View on arXiv Download PDF AI Translation

cs.RO / 53 / 2610.14691

DexCast: Learning Real-World Dexterous Control from Human Videos

DexCast：从人类视频中学习真实世界的灵巧控制

Gao, Zihan, Zhang, Yuhong, Wang, Mengzhe, Liu, Kaisheng, Li, Shengpeng, Xiao, Siyao, Cheng, Runqing, Lin, Xiao, Liu, Zhifang, Wang, Haoqian

Abstract

Recent advances in 3D hand-object interaction (HOI) perception have opened new opportunities for learning dexterous robotic skills from human videos. However, existing dexterous imitation learning methods often rely on costly and manually teleoperated robot demonstrations, requiring substantial human effort. Therefore, we build DEXCAST, a framework that learns dexterous robotic hand control directly from monocular RGB-D videos by leveraging joint HOI reconstructions, without requiring any teleoperation. Furthermore, how to model high-dimensional robotic hand motion and extract temporal context to inform future action generation remains a key challenge. To this end, we propose a DexHand Motion Perception Module (DMPM) to capture inter-joint coordination and temporal dependencies in compact latent representations that condition our diffusion policy. Extensive real-world experiments demonstrate that DEXCAST achieves exceptional performance across diverse dexterous manipulation tasks.

Chinese Translation

近年来，3D 手-物体交互（HOI）感知的进展为从人类视频中学习灵巧的机器人技能开辟了新的机会。然而，现有的灵巧模仿学习方法通常依赖于昂贵且手动遥控的机器人演示，需耗费大量人力。因此，我们构建了 DEXCAST，一个直接从单目 RGB-D 视频中学习灵巧机器人手控制的框架，利用关节 HOI 重建，而无需任何遥控操作。此外，如何建模高维机器人手运动并提取时间上下文以指导未来的动作生成仍然是一个关键挑战。为此，我们提出了 DexHand 运动感知模块（DMPM），以捕捉关节间的协调性和时间依赖性，并在紧凑的潜在表示中进行条件化，以指导我们的扩散策略。大量的现实世界实验表明，DEXCAST 在各种灵巧操作任务中表现出色。

View on arXiv Download PDF AI Translation

cs.RO / 54 / 2610.14702

BGC-Nav: Cross-Altitude Multi-UAV Collaborative Navigation via Belief-Guided Coupling

BGC-Nav：通过信念引导耦合实现跨高度多无人机协同导航

Hao, Jiuwu, Wu, Yueyang, Sun, Liguo, Song, Haolin, Zhang, Zao, Lv, Pin

Abstract

Despite recent advances in Aerial Vision-and-Language Navigation (VLN), single Unmanned Aerial Vehicle (UAV) search-and-rescue in outdoor scenarios remains challenging due to partial observability, making cross-altitude collaboration a promising direction. However, existing collaborative navigation methods rely on precise target localization and rigid goal-dispatching paradigms, ignoring high-altitude observation uncertainty and suppressing low-altitude exploration autonomy. To address these issues, we introduce BGC-Nav, a training-free cross-altitude collaboration framework that enables robust navigation via Belief-Guided Coupling (BGC). Specifically, the high-altitude UAV maintains a Global Belief Memory encoding task-conditioned priors, which the low-altitude agent incorporates into decision-making via Local Exhaustion Memory (LEM), Local Attraction Augmentation (LAA), and Directional Belief Prior (DBP) as an advisory bias rather than a rigid command. Extensive experiments on ESARBench show that BGC-Nav consistently outperforms state-of-the-art single-agent and collaborative methods. Code is available at https://github.com/uestchjw/BGC-Nav.

Chinese Translation

尽管最近在空中视觉与语言导航（VLN）方面取得了进展，但在户外场景中单一无人机（UAV）的搜索与救援仍然面临挑战，这主要是由于部分可观测性，使得跨高度协作成为一个有前景的方向。然而，现有的协同导航方法依赖于精确的目标定位和严格的目标调度范式，忽视了高空观察的不确定性，并抑制了低空探索的自主性。为了解决这些问题，我们提出了BGC-Nav，这是一种无训练的跨高度协作框架，通过信念引导耦合（BGC）实现稳健的导航。具体而言，高空无人机维护一个全局信念记忆，编码任务条件的先验信息，低空代理则通过局部耗尽记忆（LEM）、局部吸引增强（LAA）和方向信念先验（DBP）将其纳入决策过程，作为一种建议性偏差，而非严格的指令。在ESARBench上的大量实验表明，BGC-Nav在性能上始终优于最先进的单代理和协同方法。代码可在 https://github.com/uestchjw/BGC-Nav 获取。

View on arXiv Download PDF AI Translation

cs.RO / 55 / 2610.14704

PRISM: Precision Forcing for Generalist Vision-Language-Action Policies

PRISM：用于通用视觉-语言-动作策略的精确强制

Yue, Chenghao, Li, Peiyan, Wang, Hao, Lv, Lei, Chen, Jiahao, Liu, Huaping, Luo, Yu, Zhang, Yuanlong

Abstract

Coupling a pretrained vision-language model (VLM) with a diffusion or flow-matching action expert has become the mainstream vision-language-action (VLA) paradigm for embodied policies, yet they remain unreliable once a task demands precise manipulation, e.g., submillimeter cable insertion. We attribute this weakness to a training objective that averages flow-matching errors across contexts and flow times, allocating little supervision to the few decisive moments determining task success. Along the generative path, inference integrates a locally supervised velocity field, so equally penalized errors at different flow times can affect the executed action unequally. Consequently, precision-task performance deteriorates as physical tolerance tightens by orders of magnitude. We call this supervision imbalance across contexts and flow times Flow-Time Starvation. The remedy is to jointly align supervision allocation and representation learning with context-dependent precision demand, enabling the policy to recognize when greater precision is needed. We instantiate this precision-forcing strategy as Precision-aware Reweighting and Internalization for Skilled Manipulation (PRISM), which scores precision demand offline, strengthens flow supervision where errors are consequential and action signals are visible, and internalizes that demand in the VLM condition representation. Simulation and real-robot experiments show that PRISM preserves or improves general-task performance while raising mean end-to-end success from 38.3% to 86.7% across drive connector insertion, bimanual connector mating, and screw tightening, substantially extending the precision boundary of existing embodied policies without new sensors or observation modalities.

Chinese Translation

将预训练的视觉-语言模型（VLM）与扩散或流匹配动作专家相结合，已成为具身策略的主流视觉-语言-动作（VLA）范式，然而一旦任务需要精确操作（例如亚毫米级电缆插入），它们仍然不可靠。我们将这种弱点归因于一个训练目标，该目标在不同上下文和流动时间中平均流匹配误差，几乎没有对决定任务成功的少数关键时刻进行监督。在生成路径中，推理集成了一个局部监督的速度场，因此在不同流动时间下同样受到惩罚的错误可能对执行的动作产生不均等的影响。因此，随着物理容忍度的显著收紧，精确任务的表现恶化。我们将这种在上下文和流动时间之间的监督不平衡称为流动时间饥饿（Flow-Time Starvation）。解决方案是将监督分配和表示学习与上下文相关的精确需求联合对齐，使策略能够识别何时需要更高的精度。我们将这种精确强制策略实例化为精确感知重加权与熟练操作内化（Precision-aware Reweighting and Internalization for Skilled Manipulation，PRISM），该策略离线评分精确需求，在错误影响显著且动作信号可见的地方增强流动监督，并在VLM条件表示中内化该需求。模拟和真实机器人实验表明，PRISM在提升驱动连接器插入、双手连接器配合和螺丝紧固等任务的平均端到端成功率从38.3%提高到86.7%的同时，保持或改善了通用任务的表现，显著扩展了现有具身策略的精确边界，而无需新的传感器或观察模式。

View on arXiv Download PDF AI Translation

cs.RO / 56 / 2610.14715

C2Flight: A Closed-Loop Co-Simulation Framework for Early-Stage Evaluation of Heterogeneous UAV Computing Systems

C2Flight：一种用于异构无人机计算系统早期评估的闭环协同仿真框架

Chen, Xizhu, Zhang, Shuyuan, Li, Yixuan, Yang, Huazhong, Wang, Yu, Yu, Jincheng, Jia, Hongyang

Abstract

Early-stage design of heterogeneous UAV computing systems requires evaluating how algorithm, hardware, communication, and system topology affect closed-loop flight before the complete system is available. However, design choices interact through the robotic feedback loop, making task performance difficult to predict from isolated component metrics, while existing execution-driven approaches postpone system-level design feedback until the computing system is integrated. This work presents C2Flight, a closed-loop co-simulation framework that represents target systems as parameterized dataflow graphs and integrates models of hardware under design with independently accessible physical hardware through a unified node interface. Its graph runtime captures the combined effects of buffering, activation, resource sharing, communication, and local timing on closed-loop flight behavior. Validation against onboard and wireless offboard physical computing pipelines confirms its timing fidelity and closed-loop prediction accuracy. Two case studies demonstrate early-stage exploration of deployment and accelerator bit-width choices. These results show that C2Flight exposes system-level design trade-offs before full-system integration, enabling rapid early-stage design refinement.

Chinese Translation

异构无人机计算系统的早期设计需要评估算法、硬件、通信和系统拓扑如何影响闭环飞行，尤其是在完整系统尚未可用的情况下。然而，设计选择通过机器人反馈循环相互作用，使得从孤立组件指标预测任务性能变得困难，而现有的执行驱动方法则将系统级设计反馈推迟到计算系统集成之后。本文提出了C2Flight，一种闭环协同仿真框架，它将目标系统表示为参数化的数据流图，并通过统一节点接口将正在设计的硬件模型与独立可访问的物理硬件集成。其图形运行时捕捉了缓冲、激活、资源共享、通信和局部时序对闭环飞行行为的综合影响。与机载和无线离线物理计算管道的验证确认了其时序保真度和闭环预测准确性。两个案例研究展示了部署和加速器位宽选择的早期探索。这些结果表明，C2Flight在全系统集成之前揭示了系统级设计权衡，从而实现快速的早期设计优化。

View on arXiv Download PDF AI Translation

cs.RO / 57 / 2610.14720

CAG-WAM: Causal Action-Gain World Action Models for Robot Manipulation

CAG-WAM：用于机器人操作的因果行动增益世界行动模型

Guo, Peizheng, Zhang, Cheng, Zhang, Xingyu, Zheng, Changwen, Qiang, Wenwen

Abstract

World Action Models (WAMs) combine action generation with future prediction, allowing predicted futures to support action evaluation and robot decision making. However, the future associated with a candidate action may contain action-induced and environment-driven changes. Directly evaluating the predicted future may not fully characterize the future change associated with the candidate action. For example, a candidate action that worsens a favorable environmental future may still yield a favorable outcome, while one that improves an unfavorable environmental future may still yield an unfavorable outcome. To analyze this issue, we formulate a Structural Causal Model (SCM) for action evaluation in WAMs. The SCM formalizes how action-induced and environment-driven changes jointly shape future evaluation, and reveals a backdoor path through which environmental evolution can also affect the future evaluation associated with a candidate action. Intervention and backdoor adjustment account for this path, providing a causal basis for characterizing the future change associated with each candidate action. Based on this analysis, we propose Causal Action-Gain World Action Model (CAG-WAM). For each candidate action, CAG-WAM predicts an action-conditioned future and an environmental future under the same current observation, and defines their evaluation difference as causal action gain (CAG) for characterizing the future change associated with the candidate action. During training, we incorporate CAG into Group Relative Policy Optimization (GRPO) to guide robot policy learning. During inference, we evaluate the initial action with CAG and, when needed, revise it using predicted future information. Experiments on multiple benchmarks support the effectiveness of CAG-WAM.

Chinese Translation

世界行动模型（WAMs）将行动生成与未来预测相结合，使得预测的未来能够支持行动评估和机器人决策。然而，与候选行动相关的未来可能包含由行动引起的变化和环境驱动的变化。直接评估预测的未来可能无法充分表征与候选行动相关的未来变化。例如，一个使有利环境未来恶化的候选行动仍可能产生有利结果，而一个改善不利环境未来的候选行动仍可能产生不利结果。为了解析这一问题，我们为WAMs中的行动评估构建了一个结构因果模型（SCM）。SCM形式化了行动引起的变化和环境驱动的变化如何共同塑造未来评估，并揭示了一个后门路径，通过该路径环境演变也可以影响与候选行动相关的未来评估。干预和后门调整考虑了这一路径，为表征与每个候选行动相关的未来变化提供了因果基础。基于这一分析，我们提出了因果行动增益世界行动模型（CAG-WAM）。对于每个候选行动，CAG-WAM在相同的当前观察下预测一个基于行动的未来和一个环境未来，并将它们的评估差异定义为因果行动增益（CAG），以表征与候选行动相关的未来变化。在训练过程中，我们将CAG纳入群体相对策略优化（GRPO）中，以指导机器人策略学习。在推理过程中，我们使用CAG评估初始行动，并在需要时利用预测的未来信息对其进行修正。在多个基准上的实验支持了CAG-WAM的有效性。

View on arXiv Download PDF AI Translation

cs.RO / 58 / 2610.14733

B-Human '26 - Scalable Team Play up to 11 vs. 11

B-Human '26 - 可扩展的团队比赛至11对11

Hasselbring, Arne, Laue, Tim, Röfer, Thomas, Reichenberg, Philip

Abstract

B-Human won multiple competitions in the Middle Division of the new Humanoid Soccer League in 2026. In addition to regular matches with three or five players per team, we ventured into uncharted territory by playing demo matches with eleven humanoid robots, highlighting the tactical flexibility of our behavior and the overall robustness of our system. As B-Human formerly competed in the Standard Platform League with robots half the size and less than a third the weight, we begin by outlining the adjustments needed for this transition. The central section of this paper details our behavior architecture, drawing on concepts from RoboCup leagues that have used 11 players for much longer.

Chinese Translation

B-Human在2026年新的人形足球联赛中赢得了多个中级别比赛的冠军。除了每队三人或五人的常规比赛外，我们还通过与十一台人形机器人进行演示比赛，探索了未知领域，突显了我们行为的战术灵活性和系统的整体稳健性。由于B-Human之前在标准平台联赛中与体型仅为一半、重量不足三分之一的机器人竞争，我们首先概述了这一转变所需的调整。本文的核心部分详细介绍了我们的行为架构，借鉴了在RoboCup联赛中使用11名球员的概念，这些联赛的历史更为悠久。

View on arXiv Download PDF AI Translation

cs.RO / 59 / 2610.14746

BridgeWAM: Bridging Video and Action Experts via Latent Bridge Queries

BridgeWAM：通过潜在桥查询连接视频和动作专家

Fang, Jiaxin, Liu, Ruitong, Lai, Zeyu, Wang, Xingyu, Jia, Mingyuan, Sheng, Pingyue, Wan, Weikang, Xu, Mengdi, Gao, Yuxiang

Abstract

World Action Models (WAMs) commonly couple video and action experts through dense, layer-wise Mixture-of-Transformers (MoT) attention, exposing action prediction to irrelevant visual features while enforcing matched expert depths. This coupling hinders selective visual-feature use and introduces redundant computation. We introduce Bridge-of-Experts (BoE), instantiated as BridgeWAM, which aggregates visual, language, and proprioceptive priors through 32 Latent Bridge Queries (LBQs) into a shared, limited latent space. Joint video-action supervision lets action gradients guide the Video Expert toward representations better suited for action prediction, while the Action Expert learns to selectively retrieve task-relevant information through cross-attention. Decoupling expert depths and reusing LBQs during action denoising yield a 5.06B model with a 30-layer Video Expert and a two-layer Action Expert, achieving 98.25% success on LIBERO and 72.24% pooled success on LIBERO-Plus, demonstrating effective task execution and strong generalization with an ultralight action head. Ablations support joint representation adaptation and unified LBQ conditioning. Relative to Fast-WAM, BridgeWAM reduces single-sample inference FLOPs by 28.72% in simulation profiling and real-robot latency to 77.1 ms (a 2.47-fold speedup). Qualitative experimental results further indicate preferential attention to target objects and task destinations across perturbations.

Chinese Translation

世界动作模型（WAMs）通常通过密集的逐层混合变换器（Mixture-of-Transformers, MoT）注意力将视频和动作专家结合在一起，这使得动作预测受到无关视觉特征的影响，同时强制匹配专家的深度。这种结合阻碍了选择性视觉特征的使用，并引入了冗余计算。我们提出了专家桥（Bridge-of-Experts, BoE），具体实现为BridgeWAM，它通过32个潜在桥查询（Latent Bridge Queries, LBQs）将视觉、语言和本体感知先验聚合到一个共享的、有限的潜在空间中。联合视频-动作监督使得动作梯度能够引导视频专家朝向更适合动作预测的表示，而动作专家则学习通过交叉注意力选择性地检索与任务相关的信息。解耦专家深度并在动作去噪过程中重用LBQs，产生了一个包含30层视频专家和两层动作专家的5.06B模型，在LIBERO上取得了98.25%的成功率，在LIBERO-Plus上获得了72.24%的汇总成功率，展示了有效的任务执行和强大的泛化能力，配备超轻量级的动作头。消融实验支持联合表示适应和统一的LBQ条件。相较于Fast-WAM，BridgeWAM在仿真分析中将单样本推理的FLOPs减少了28.72%，在真实机器人延迟中降低至77.1毫秒（速度提升2.47倍）。定性实验结果进一步表明在扰动下对目标物体和任务目的地的优先关注。

View on arXiv Download PDF AI Translation

cs.RO / 60 / 2610.14753

Beyond Fixed Visual History: Learning Motion-Adaptive Temporal Receptive Fields for Complex-Terrain Quadruped Locomotion

超越固定视觉历史：学习运动自适应时间感受野以实现复杂地形四足运动

Zhang, Hao, Yan, Yuping, Hong, Yitian, Feng, Luying, Wang, Tianhong, Dai, Fengqi, Chen, Wei, Jin, Yaochu

Abstract

Versatile quadruped locomotion depends on a tight coupling between what the robot perceives and what its body can execute. Yet existing visual controllers typically organize depth history using fixed temporal patterns, overlooking that the same temporal span corresponds to different spatial ranges as the robot's speed, posture, and gait phase change. Consequently, the same visual history may be insufficient in one motion state but outdated or misaligned in another. To address this limitation, we propose TempoLoco, a body-aware dynamic visual perception framework that organizes visual memory according to the robot's current locomotion state. TempoLoco learns multiple complementary temporal views from a long depth buffer, encodes them within a shared feature space, and dynamically aggregates them using proprioceptive history and multi-scale terrain evidence. This enables a single policy to align visual anticipation with the robot's current motion capabilities across complex terrains, including stairs, platforms, gaps, discrete obstacles, rough ground, and slopes. From the experiment analysis, TempoLoco achieves a 78.4% success rate, compared with 58.2% for the strongest fixed-history baseline and 62.1% for uniform multi-scale fusion. In separate capability evaluations, it completes 41/45 simulated trials across nine terrain types and 23/25 real-world trials on a Unitree B2 with a fixed arm, demonstrating robust and deployable locomotion through body-aware dynamic visual perception.

Chinese Translation

多功能四足运动依赖于机器人感知与其身体执行能力之间的紧密耦合。然而，现有的视觉控制器通常使用固定的时间模式来组织深度历史，忽视了同一时间跨度在机器人速度、姿态和步态阶段变化时对应不同空间范围的事实。因此，在一种运动状态下相同的视觉历史可能不足，而在另一种状态下则可能过时或不对齐。为了解决这一局限性，我们提出了TempoLoco，一个基于身体感知的动态视觉感知框架，根据机器人的当前运动状态组织视觉记忆。TempoLoco从长深度缓冲区学习多个互补的时间视图，将其编码在共享特征空间中，并利用本体历史和多尺度地形证据动态聚合这些视图。这使得单一策略能够在复杂地形（包括楼梯、平台、间隙、离散障碍物、粗糙地面和坡道）中将视觉预期与机器人的当前运动能力对齐。实验分析表明，TempoLoco的成功率达到78.4%，而最强的固定历史基线为58.2%，均匀多尺度融合为62.1%。在单独的能力评估中，它在九种地形类型的模拟试验中完成了41/45次试验，并在固定臂的Unitree B2上完成了23/25次真实世界试验，展示了通过基于身体感知的动态视觉感知实现的稳健和可部署的运动能力。

View on arXiv Download PDF AI Translation

cs.RO / 61 / 2610.14833

Scaling Latent Motor Adaptation to Coordinated Bimanual Manipulation

将潜在运动适应扩展到协调的双手操作

Joo, Seung-Bin, Albini, Alessandro, Wiehe, Luca, Maiolino, Perla, Reino, Daniel Olmeda, Henriques, João F.

Abstract

Robust manipulation under uncertain object dynamics and geometry remains a core challenge in real-world robotics. Policies trained under fixed assumptions often degrade when object mass, friction, scale, or shape vary. Embedding-based adaptation methods such as Rapid Motor Adaptation (RMA) address this by conditioning actions on latent representations of environment dynamics inferred from interaction history. We investigate whether this paradigm scales from single arm settings to coordinated bimanual manipulation, where higher dimensional control, coupled contacts, and shared object forces introduce additional complexity. We evaluate bimanual RMA on three contact-rich ManiSkill tasks with randomized YCB object properties. Bimanual RMA, using observation-action history and dual wrist depth images, retains privileged policy performance on bimanual picking and lifting, while an otherwise matched stationary PPO policy without latent conditioning achieves low success; all methods remain limited on long-horizon stacking. We further demonstrate direct sim-to-real transfer on a dual Franka lifting task without real-world finetuning. Together, these results suggest that latent conditioning provides useful structure for scaling robust control to coordinated bimanual systems, while longer horizon tasks require additional temporal structure.

Chinese Translation

在不确定的物体动力学和几何形状下进行稳健操作仍然是现实世界机器人技术中的核心挑战。在固定假设下训练的策略在物体质量、摩擦、尺度或形状变化时往往会退化。基于嵌入的适应方法，如快速运动适应（Rapid Motor Adaptation, RMA），通过将动作条件化于从交互历史推断出的环境动态的潜在表示来解决这一问题。我们研究这一范式是否能够从单臂设置扩展到协调的双手操作，在这种情况下，更高维度的控制、耦合接触和共享物体力引入了额外的复杂性。我们在三个接触丰富的 ManiSkill 任务上评估了双手 RMA，这些任务具有随机化的 YCB 物体属性。双手 RMA 使用观察-动作历史和双腕深度图像，在双手拾取和提升任务中保持了优越的策略性能，而没有潜在条件的匹配静态 PPO 策略则成功率较低；所有方法在长时间堆叠任务中仍然受到限制。我们进一步展示了在没有现实世界微调的情况下，双 Franka 提升任务的直接仿真到现实转移。这些结果表明，潜在条件化为将稳健控制扩展到协调的双手系统提供了有用的结构，而较长的时间任务则需要额外的时间结构。

View on arXiv Download PDF AI Translation

cs.RO / 62 / 2610.14840

Push the Limit: Learning High-Speed Sprinting and Aggressive Cornering on a Full-Scale Humanoid

突破极限：在全尺度人形机器人上学习高速冲刺和激进转弯

Huang, Huaxing, Li, Ruidong, Han, Jinchao, Dong, Ziyang, Cui, Wenhao

Abstract

We present a learning-based framework that enables a full-scale bipedal humanoid to sprint at over 7 m/s with motion-prior-guided gait and perform aggressive cornering at high speed with a pronounced trunk lean of up to $25^\circ$. Achieving this combination is challenging: above approximately 4 m/s, periodic foot impacts induce severe, high-frequency yaw-rate oscillations intrinsic to high-speed gaits. This high-frequency temporal structure corrupts the history of proprioceptive feedback that flat-MLP policies rely on, while rendering conventional yaw-rate tracking rewards counterproductive, as they fight against the robot's natural oscillatory dynamics. Moreover, a single style prior shared across skills cannot tell which skill is commanded and inherently compromises the low-speed skill fidelity. To overcome these limitations, we adopt a causal multi-head attention (MHA) module to explicitly model the high-frequency temporal feedback. In place of the oscillatory yaw rate, we design a heading-direction reward that supervises only the low-frequency velocity direction, allowing agile turning at high speed without suppressing natural gait oscillations. We further introduce skill-conditioned multi-adversarial-motion-prior (MAMP) discriminators that condition the style reward on the commanded skill. Ablation experiments confirm that each mechanism is essential: replacing structured temporal encoding with a flat-history MLP caps the top speed; reverting to yaw-rate tracking prevents high-speed turning; and a single shared discriminator makes in-place turning collapse into a quasi-static stance. Our framework bridges the gap between raw sprinting speed and reference-motion fidelity, demonstrating stable high-speed sprinting and aggressive cornering in both simulation and real-robot deployment.

Chinese Translation

我们提出了一种基于学习的框架，使全尺度双足人形机器人能够以超过7米/秒的速度冲刺，并在高速下以高达$25^ ext{°}$的显著躯干倾斜进行激进转弯。实现这种组合具有挑战性：在约4米/秒以上，周期性的脚部冲击会引发与高速步态固有相关的严重高频偏航率振荡。这种高频时间结构会破坏平坦多层感知机（flat-MLP）策略所依赖的本体反馈历史，同时使得传统的偏航率跟踪奖励适得其反，因为它们与机器人的自然振荡动态相抵触。此外，跨技能共享的单一风格先验无法指示具体的技能命令，固有地妨碍了低速技能的保真度。为了克服这些限制，我们采用了因果多头注意力（MHA）模块，明确建模高频时间反馈。我们设计了一种朝向方向奖励，监督仅低频速度方向，允许在高速下灵活转弯而不抑制自然步态振荡。我们进一步引入了技能条件的多对抗运动先验（MAMP）鉴别器，使风格奖励依赖于所命令的技能。消融实验确认了每个机制的必要性：用平坦历史的MLP替代结构化的时间编码会限制最高速度；回归到偏航率跟踪会阻碍高速转弯；而单一共享的鉴别器会使原地转弯崩溃为准静态姿态。我们的框架弥合了原始冲刺速度与参考运动保真度之间的差距，展示了在模拟和真实机器人部署中稳定的高速冲刺和激进转弯。

View on arXiv Download PDF AI Translation

cs.RO / 63 / 2610.14848

Learning Dynamic Grasping

学习动态抓取

Ho, Nga Kiu Kelly, Liu, Yilin, Xu, Hao, Mitra, Niloy J.

Abstract

Synthesizing believable 4D hand-object grasping sequences is important for character animation, extended reality, and embodied agents. Grasping requires a hand to approach, contact, and lift or manipulate an object coherently. Existing synthesis methods often model only the final static grasp or rely on additional conditions, such as language, target grasps, or object trajectories, to specify the interaction. We instead study the underconstrained problem of generating a complete grasping motion from a single initial hand-object state. Our framework decouples hand-motion generation from object-motion recovery. Given an object point cloud in the initial wrist frame and the initial parameterized hand configuration, conditional flow matching predicts a sequence-level contact map and a hand pose at each normalized sequence time. This time-indexed formulation avoids the error accumulation of autoregressive rollout while representing the complete motion. We then infer an attachment frame and recover rigid-object motion through training-free, contact-guided weighted registration. We evaluate on held-out DexYCB objects and unseen object geometries reconstructed from in-the-wild videos by Open-CHOIR. Compared with LatentHOI, our method produces smoother hand and object motion in every setting, matches or improves on its contact and penetration metrics in the wild under a shared initialization, and is about 30x faster per generated frame. In a 15-participant study, our motions were preferred in roughly two thirds of (forced-choice pairwise) comparisons. Precise contact closure on novel geometry, however, remains a key open challenge.

Chinese Translation

合成可信的四维手-物体抓取序列对于角色动画、扩展现实和具身智能体至关重要。抓取需要手部以连贯的方式接近、接触并提升或操控物体。现有的合成方法通常仅建模最终的静态抓取，或依赖于额外条件，如语言、目标抓取或物体轨迹来指定交互。相反，我们研究从单一初始手-物体状态生成完整抓取动作的欠约束问题。我们的框架将手部运动生成与物体运动恢复解耦。给定初始腕部坐标系中的物体点云和初始参数化手部配置，条件流匹配预测每个归一化序列时间的序列级接触图和手部姿态。这种时间索引的表述避免了自回归展开中的误差累积，同时表示完整的运动。然后，我们推断附着框架，并通过无训练的接触引导加权配准恢复刚性物体运动。我们在保留的DexYCB物体和从野外视频中重建的未见物体几何上进行评估，后者由Open-CHOIR提供。与LatentHOI相比，我们的方法在每种设置中产生更平滑的手部和物体运动，在共享初始化下匹配或改善其在野外的接触和穿透指标，并且每生成一帧的速度约快30倍。在一项包含15名参与者的研究中，我们的运动在大约三分之二的（强制选择成对）比较中被偏好。然而，在新几何体上实现精确的接触闭合仍然是一个关键的开放挑战。

View on arXiv Download PDF AI Translation

cs.RO / 64 / 2610.14854

DEFT-VLA: Stage-Localized Force Specialization for Contact-Rich Robotic Manipulation

DEFT-VLA：针对接触丰富的机器人操作的阶段局部化力专业化

Huang, Zan, Qiu, Yibo, Ding, Kan, Zhang, Xuanping, Tian, Ao, Xu, Ronald X, Sun, Mingzhai

Abstract

Contact-rich manipulation poses a stage-dependent sensory conflict: vision and proprioception are essential for localization and approach, when force is largely uninformative and potentially distracting, whereas force becomes critical after contact, when irrelevant visual details and correlated robot states may become shortcuts. We propose DEFT-VLA (Dual-Expert Force-specialized Training for Vision-Language-Action Manipulation), which retains the task-adapted base expert for vision-dominated non-contact behavior and introduces a force-aware contact expert for critical contact stages, localizing force-conditioned adaptation to where it is needed. We further introduce a contact-localized force-attending curriculum that applies force-attending post-training only to the contact expert, progressively restoring vision and masking state tokens to suppress shortcuts. Real-robot experiments on pipette tip attachment, bimanual pipette aspiration, and whiteboard wiping show that DEFT-VLA effectively exploits force feedback to improve contact-stage performance while maintaining strong non-contact behavior and robustness to held-out grasp heights. It improves overall task success by 20-40 percentage points over the strongest baseline on each task.

Chinese Translation

接触丰富的操作面临着阶段依赖的感知冲突：在定位和接近时，视觉和本体感觉对于任务至关重要，而此时力信息往往不具备信息性且可能造成干扰；而在接触后，力变得至关重要，此时无关的视觉细节和相关的机器人状态可能成为捷径。我们提出了DEFT-VLA（双专家力专业化训练用于视觉-语言-动作操作），该方法保留了适应任务的基础专家以应对以视觉为主的非接触行为，并引入了一个力感知的接触专家以应对关键接触阶段，将力条件适应局部化到所需的地方。我们进一步提出了一种接触局部化的力关注课程，该课程仅在接触专家上应用力关注的后训练，逐步恢复视觉并屏蔽状态标记以抑制捷径。在移液管尖端附着、双手移液管吸取和白板擦拭的真实机器人实验中，DEFT-VLA有效利用力反馈提高了接触阶段的表现，同时保持了强大的非接触行为和对保持高度的鲁棒性。与每项任务中最强基线相比，它提高了整体任务成功率20-40个百分点。

View on arXiv Download PDF AI Translation

cs.RO / 65 / 2610.14864

Predict in Chunks, Revise in Spans: History-conditioned Action Regeneration for Robotic Manipulation

分块预测，跨度修正：基于历史条件的动作再生用于机器人操作

Wang, Chenyang, Wang, Yuntian, Tang, Jiaying, Li, Chunzheng, Liu, Yang, Liu, Siao

Abstract

Action chunking improves temporal coherence in visuomotor imitation learning by jointly predicting sequences of future actions. However, many existing policies regenerate entire action chunks even when errors are localized, potentially replacing otherwise valid predictions. We introduce Contextual Masked Action Policy (CMAP), a visuomotor policy that refines continuous action chunks through history-conditioned local regeneration. Specifically, CMAP introduces a history-conditioned action tape that combines executed commands with masked future positions to produce coarse action predictions and per-step uncertainty scores. An uncertainty-guided span regeneration mechanism then selects a high-uncertainty contiguous span and reconstructs it using the executed history and retained future predictions as context. By jointly reconstructing neighboring actions, CMAP models local temporal dependencies while preserving all predictions outside the selected span. Extensive experiments on nine simulated and four real-world manipulation tasks demonstrate that CMAP achieves superior overall performance over several competitive baselines. Ablation studies validate the complementary contributions of its key components, while held-out trajectory analysis shows that local regeneration reduces prediction errors within selected spans.

Chinese Translation

动作分块通过共同预测未来动作序列来提高视觉运动模仿学习中的时间一致性。然而，许多现有策略在错误局部化时仍会再生整个动作块，这可能会替换掉本应有效的预测。我们提出了上下文掩蔽动作策略（Contextual Masked Action Policy, CMAP），这是一种通过历史条件局部再生来细化连续动作块的视觉运动策略。具体而言，CMAP引入了一种历史条件动作带，将已执行的指令与掩蔽的未来位置结合，以生成粗略的动作预测和每步的不确定性评分。然后，一个不确定性引导的跨度再生机制选择一个高不确定性的连续跨度，并利用已执行的历史和保留的未来预测作为上下文进行重构。通过共同重构相邻动作，CMAP建模局部时间依赖性，同时保留选定跨度外的所有预测。在九个模拟和四个真实世界操作任务上的广泛实验表明，CMAP在多个竞争基线中实现了优越的整体性能。消融研究验证了其关键组件的互补贡献，而保留轨迹分析显示局部再生减少了选定跨度内的预测错误。

View on arXiv Download PDF AI Translation

cs.RO / 66 / 2610.14885

Verify, Act, Distill: Turning Test-Time Compute into Test-Time Adaptation for Vision-Language-Action Models

验证、行动、提炼：将测试时计算转化为视觉-语言-行动模型的测试时适应

Hiratsuka, Kaneyoshi, Yamamori, Satoshi, Morimoto, Jun

Abstract

Test-time compute (TTC), which scores multiple candidate actions at inference time, is a powerful way to improve the execution performance of vision-language-action (VLA) models, but existing methods keep the policy frozen, so the gain must be recomputed at every execution. We propose Verify, Act, Distill, a framework that consolidates the gains of TTC into the VLA itself. A fixed verifier scores candidates generated by a flow-matching VLA, the weighted aggregate is executed, and the observations and aggregated chunks from successful adaptation episodes are distilled directly into LoRA parameters with the standard flow-matching loss. This reuses supervision that TTC has already produced during execution, without additional reinforcement learning or rollouts dedicated to teacher generation. On the 40 LIBERO tasks with a SmolVLA 5k checkpoint, distilling from only 10 episodes per task raises the mean success rate of verifier-free single-candidate generation from 54.1% to 70.6%, retaining about 74% of the improvement obtained by aggregating eight candidates. Applying the same aggregation to the distilled policy further raises the success rate from 76.4% to 82.3%, with a particularly large gain on LIBERO-Long (44.0% to 57.8%), and the single-candidate policy also improves on LIBERO-plus, which contains distribution shifts. These results show that using verification as a learning signal, rather than only as an execution-time selector, both retains the improvements of TTC and re-amplifies the effect of subsequent TTC.

Chinese Translation

测试时计算（TTC）在推理时对多个候选动作进行评分，是提高视觉-语言-行动（VLA）模型执行性能的有效方法，但现有方法保持策略不变，因此每次执行时都必须重新计算增益。我们提出了验证、行动、提炼（Verify, Act, Distill）框架，将TTC的增益整合到VLA模型中。一个固定的验证器对由流匹配VLA生成的候选项进行评分，执行加权聚合，并将成功适应阶段的观察结果和聚合块直接提炼为LoRA参数，使用标准的流匹配损失。这种方法重用了TTC在执行过程中已经产生的监督，而无需额外的强化学习或专门用于教师生成的回滚。在使用SmolVLA 5k检查点的40个LIBERO任务中，仅从每个任务的10个阶段提炼，验证器无候选生成的平均成功率从54.1%提高到70.6%，保留了通过聚合八个候选项获得的约74%的改善。将相同的聚合应用于提炼后的策略，进一步将成功率从76.4%提高到82.3%，在LIBERO-Long任务上获得了特别大的增益（从44.0%提高到57.8%），而在包含分布转移的LIBERO-plus任务上，单候选策略也有所改善。这些结果表明，将验证作为学习信号，而不仅仅是作为执行时选择器，既保留了TTC的改进，又重新放大了后续TTC的效果。

View on arXiv Download PDF AI Translation

cs.RO / 67 / 2610.14889

LETS: Low-Overhead External-Torque Sensing for Force-Feedback Bilateral Teleoperation

LETS：用于力反馈双向遥操作的低开销外部扭矩传感

Guo, Zonghao, Guan, Dongchen, Wang, Yubo, Yang, Peng

Abstract

Bilateral teleoperation with force feedback is becoming increasingly important, and the quality of external-torque estimation is critical to this process. However, manufacturer-provided disturbance-torque estimates on commercial robotic arms are often inaccurate, while the use of end-effector force/torque sensors adds cost and payload. We propose LETS, a physics-guided learning approach that assigns rigid-body dynamics to recursive Newton-Euler inverse dynamics and reserves learning for only the robot-specific residual. LETS uses only the robot's URDF model and joint feedback, deliberately excluding control commands and thereby decoupling external-torque estimation from control-command inputs. It trains on 12 minutes of free-motion data in under one minute, without requiring a dedicated end-effector six-axis force/torque sensor. In bilateral teleoperation, LETS achieves real-time external-torque estimation that enables stable force feedback. Its free-space error is 69% lower than the manufacturer-provided estimate, substantially reducing coupled oscillations between leader and follower; its contact error is 28% lower, yielding more accurate haptic feedback.

Chinese Translation

带有力反馈的双向遥操作变得越来越重要，而外部扭矩估计的质量对这一过程至关重要。然而，商业机器人臂提供的制造商干扰扭矩估计往往不准确，而使用末端执行器的力/扭矩传感器则增加了成本和负载。我们提出了LETS（低开销外部扭矩传感），这是一种物理引导的学习方法，将刚体动力学分配给递归的牛顿-欧拉逆动力学，仅将学习保留给机器人特定的残余部分。LETS仅使用机器人的URDF模型和关节反馈，故意排除了控制命令，从而将外部扭矩估计与控制命令输入解耦。它在不到一分钟的时间内对12分钟的自由运动数据进行训练，无需专用的末端执行器六轴力/扭矩传感器。在双向遥操作中，LETS实现了实时外部扭矩估计，使得稳定的力反馈成为可能。其自由空间误差比制造商提供的估计低69%，显著减少了领导者和跟随者之间的耦合振荡；其接触误差低28%，提供了更准确的触觉反馈。

View on arXiv Download PDF AI Translation

cs.RO / 68 / 2610.14896

Test-Time Compute for Humanoid Behavior Foundation Models

人形行为基础模型的测试时计算

Xie, Weiji, Wu, Junsong, Zhou, Zibo, Chen, Yueru, Wang, Xingyi, Xiong, Zhenhua, Zhang, Weinan, Bai, Chenjia

Abstract

Humanoid Behavior Foundation Models (BFMs) can execute impressive motions, but downstream applications such as loco-manipulation demand high whole-body positional accuracy. Existing BFMs suffer from a systematic command-execution gap, where executed motions deviate from desired ones. We show that additional test-time compute can improve execution without retraining the BFM. First, we introduce Test-Time Command Optimization (TTCO), which adapts the input command to the specific BFM to produce an executed trajectory closer to the desired motion. TTCO finds such commands by sampling-based optimization through closed-loop simulations of the BFM and robot dynamics. Across three BFMs on the AMASS dataset, TTCO reduces mean global body-position error by 79-84% in simulation. Second, we analyze test-time compute scaling and derive an approximate compute-optimal strategy that can sustain performance gains as the compute budget increases. Third, TTCO can generate BFM-specific action labels from recorded robot trajectories without requiring the original teleoperation body-motion commands. Simulation and real-robot experiments show that these labels are effective for training high-level loco-manipulation policies. This method also supports reusing existing demonstrations across BFMs.

Chinese Translation

人形行为基础模型（BFMs）能够执行令人印象深刻的动作，但下游应用如运动操控（loco-manipulation）需要高的全身位置精度。现有的BFMs存在系统性的指令执行差距，执行的动作与期望的动作偏离。我们表明，额外的测试时计算可以在不重新训练BFM的情况下改善执行效果。首先，我们引入了测试时指令优化（Test-Time Command Optimization, TTCO），它将输入指令调整为特定的BFM，以生成更接近期望动作的执行轨迹。TTCO通过对BFM和机器人动力学的闭环仿真进行基于采样的优化来寻找这样的指令。在AMASS数据集上的三个BFMs中，TTCO在仿真中将平均全局身体位置误差减少了79-84%。其次，我们分析了测试时计算的扩展性，并推导出一种近似的计算最优策略，能够在计算预算增加时维持性能提升。第三，TTCO可以从记录的机器人轨迹中生成特定于BFM的动作标签，而无需原始的遥操作身体动作指令。仿真和真实机器人实验表明，这些标签对于训练高层次的运动操控策略是有效的。这种方法还支持在不同BFMs之间重用现有的演示。

View on arXiv Download PDF AI Translation

cs.RO / 69 / 2610.14911

Scalable Multi-Human Cooperative Manipulation via Commander-Executor Coordination

通过指挥者-执行者协调实现可扩展的多人体协作操控

Huang, Buzhen, Wu, Xingchen, Yang, Yuanwang, Xie, Chenyi, Li, Kun

Abstract

Multi-human cooperative object manipulation is a highly challenging topic in humanoid control. While existing methods can handle simple collaborative carrying tasks, they struggle with role-dependent, long-horizon, and sequential cooperative manipulation. To address this issue, we propose a hierarchical commander-executor framework that enables multiple agents to accomplish complex collaborative tasks in distinct ways. Specifically, a commander combines instructions and scene information for joint semantic and geometric planning. It then decomposes the task into role-specific subtasks and generates shared object-centric coordination plans, specifying execution timing, target positions, and manipulation actions for each agent. Conditioned on the commander's high-level strategy and each agent's local observations, the executor learns a low-level physics-based policy through a solo-to-cooperative curriculum to realize coordinated human-object interactions. The planned object trajectory provides a shared coordination target, while root positions and grasping instructions specify role-dependent execution cues. We evaluate our method on a variety of object manipulation tasks, demonstrating its ability to execute long-horizon tasks that combine different types of collaboration and to scale from two-agent to multi-agent cooperation. Extensive experiments further show that our approach achieves higher success rates compared to state-of-the-art methods. Code will be made publicly available at https://github.com/TJU-3DV/Co-Manipulation.

Chinese Translation

多人体协作物体操控是人形控制领域中的一个极具挑战性的课题。尽管现有方法能够处理简单的协作搬运任务，但在角色依赖、长时间跨度和顺序协作操控方面却显得力不从心。为了解决这一问题，我们提出了一种分层的指挥者-执行者框架，使多个智能体能够以不同的方式完成复杂的协作任务。具体而言，指挥者结合指令和场景信息进行联合语义和几何规划。然后，它将任务分解为角色特定的子任务，并生成共享的以物体为中心的协调计划，指定每个智能体的执行时机、目标位置和操控动作。在指挥者的高层策略和每个智能体的局部观察的条件下，执行者通过单独到协作的课程学习一种基于物理的低层策略，以实现协调的人物-物体交互。规划的物体轨迹提供了一个共享的协调目标，而根位置和抓取指令则指定了角色依赖的执行提示。我们在多种物体操控任务上评估了我们的方法，展示了其执行结合不同类型协作的长时间跨度任务的能力，并能够从双智能体扩展到多智能体合作。大量实验进一步表明，我们的方法相比于最先进的方法实现了更高的成功率。代码将公开发布在 https://github.com/TJU-3DV/Co-Manipulation。

View on arXiv Download PDF AI Translation

cs.RO / 70 / 2610.14917

Funnel planning and control for manipulators using neural Signed Configuration Distance Functions

基于神经签名配置距离函数的操纵器漏斗规划与控制

Wullt, Bernhard, Verginis, Christos K., Mattsson, Per, Norrlöf, Mikael, Schön, Thomas B.

Abstract

Robust motion planning and control for manipulators is the problem of guaranteeing safe execution and collision-free motion under disturbances. The Kinodynamic Motion Planning via Funnel Control (KDF) is a recently proposed planner that comes with robustness guarantees and is particularly attractive for collision-free control as it guarantees closed-loop evolution inside a funnel, i.e., a collision-free region around a nominal trajectory. Despite its attractive theoretical properties, the current approach for computing funnels is computationally demanding, which limits its expressiveness and performance. We address its current limitations by proposing a novel funnel planning and control solution. Our key tool to enable this is a neural Signed Configuration Distance Function (nSCDF) which allows us to estimate collision-free balls and gradients in a computationally efficient way. We use the nSCDF in our planner to compute a nominal trajectory, where path length and clearance are balanced. To track the trajectory with collision-avoidance guarantees, we design a funnel controller that exploits the geometric information from the nSCDF. From numerical experiments, we observe that our planner produces nominal paths $2$ times faster, while at the same time obtaining $20 \%$ shorter path length and $5$ times the clearance. Finally, we also observe improved motion performance and robustness properties when using the nSCDF for funnel control.

Chinese Translation

操纵器的鲁棒运动规划与控制是确保在干扰下安全执行和无碰撞运动的问题。基于漏斗控制的运动规划（Kinodynamic Motion Planning via Funnel Control, KDF）是一种最近提出的规划方法，具有鲁棒性保证，特别适用于无碰撞控制，因为它保证了在漏斗内的闭环演变，即在名义轨迹周围的无碰撞区域。尽管其理论性质吸引人，但当前计算漏斗的方法计算开销较大，限制了其表现力和性能。我们通过提出一种新颖的漏斗规划与控制解决方案来应对其当前的局限性。我们实现这一目标的关键工具是神经签名配置距离函数（neural Signed Configuration Distance Function, nSCDF），它使我们能够以计算高效的方式估计无碰撞球体和梯度。我们在规划器中使用nSCDF来计算名义轨迹，其中路径长度和间隙得到了平衡。为了在避免碰撞的保证下跟踪轨迹，我们设计了一种利用nSCDF几何信息的漏斗控制器。通过数值实验，我们观察到我们的规划器生成名义路径的速度提高了2倍，同时路径长度减少了20%，间隙增加了5倍。最后，我们还观察到在使用nSCDF进行漏斗控制时，运动性能和鲁棒性得到了改善。

View on arXiv Download PDF AI Translation

cs.RO / 71 / 2610.14926

Edge-Native Temporal--Graph Dual-Expert Inference for Dexterous Grasp Classification

边缘原生时序-图双专家推理用于灵巧抓取分类

Zheng, Jiajian, Zhang, Bin, Fan, Wen, Liu, Yanan

Abstract

Multi-finger hands contact objects at different times, so a global close command can overload early-contact fingers while others are still moving. Dense tactile arrays address this problem but add hardware, calibration, bandwidth, and computation. We present an edge-native pipeline that uses only built-in joint-angle and pre-use-calibrated actuator-load feedback to recognize the grasped object and predict an independent Stop event for each finger. Four samples and their first differences form a 4-by-20 contact-evolution window. At its core is a heterogeneous dual-expert architecture, termed DualExpert, that assigns complementary tasks to a Temporal Convolutional Network (TCN) and a Finger Graph Filter Network (FGFN). The TCN models short-term contact evolution and supplies all five Stop probabilities, while the FGFN encodes finger topology and contributes only to object classification. Fixed 0.7/0.3 class-probability fusion reaches 93.78\% trajectory-level classification accuracy on 225 held-out trials, improving the standalone TCN and FGFN by 4.00 and 2.22 percentage points, respectively. The TCN Stop path attains 95.25\% F1. The dataset contains 1,500 automatically collected trajectories from five object instances. In Flash deployment, mixed accelerator execution retains 93.78\% classification accuracy and 94.84\% Stop F1 with 2.890~ms mean inference time and 9.172~ms mean sample-to-stop latency. Across 50 physical trials, the system achieves 96.00\% recognition and 98.00\% grasp success while the complete embedded board draws approximately 0.80~W.

Chinese Translation

多指手在不同时间接触物体，因此全局关闭命令可能会导致早期接触的手指过载，而其他手指仍在移动。密集触觉阵列解决了这个问题，但增加了硬件、校准、带宽和计算的需求。我们提出了一种边缘原生管道，仅使用内置的关节角度和预先使用校准的执行器负载反馈来识别抓取的物体，并为每个手指预测独立的停止事件。四个样本及其首次差异形成一个4乘20的接触演变窗口。其核心是一个异构双专家架构，称为DualExpert，它将互补任务分配给时序卷积网络（Temporal Convolutional Network, TCN）和手指图滤波网络（Finger Graph Filter Network, FGFN）。TCN建模短期接触演变并提供所有五个停止概率，而FGFN编码手指拓扑，仅对物体分类做出贡献。固定的0.7/0.3类别概率融合在225个保留试验中达到了93.78%的轨迹级分类准确率，分别提高了独立TCN和FGFN 4.00和2.22个百分点。TCN停止路径达到了95.25%的F1得分。数据集包含来自五个物体实例的1,500条自动收集的轨迹。在闪存部署中，混合加速器执行保持了93.78%的分类准确率和94.84%的停止F1，平均推理时间为2.890毫秒，平均样本到停止延迟为9.172毫秒。在50个物理试验中，该系统实现了96.00%的识别率和98.00%的抓取成功率，而完整的嵌入式板大约消耗0.80瓦特。

View on arXiv Download PDF AI Translation

cs.RO / 72 / 2610.14936

CFG-Policy: Unified Diffusion Policy Learning Framework via Classifier-free Guidance

CFG-Policy：通过无分类器引导的统一扩散策略学习框架

Ding, Shutong, Liu, Zhaoyang, Zhou, Yimiao, Wang, Jingya, Gu, Jiayuan, Shi, Ye

Abstract

Diffusion policies are increasingly used in robotic control and manipulation, vision-language-action models, and world-model-based agents. Their ability to represent multimodal action distributions offers rich opportunities for exploration in reinforcement learning (RL). However, optimizing diffusion policies with RL can reduce the behavioral diversity and further limit their exploration capability. In RL post-training, directly fine-tuning a pretrained policy may also disrupt its acquired capabilities, while action-space residuals may lack the expressive power for multimodal exploration. Together, these issues reveal a common challenge across online and offline-to-online diffusion RL: improving task performance while retaining the behavioral diversity and capabilities needed for continued adaptation. We introduce CFG-Policy, a unified, plug-and-play framework that separates exploration and exploitation into two generative policies and combines their score or velocity fields during sampling through classifier-free guidance. For online RL, a frozen exploration branch sustains behavioral diversity while an exploitation branch learns from task rewards; for offline-to-online RL, a pretrained exploitation branch remains frozen while a trainable branch steers the composed behavior. CFG-Policy preserves the underlying RL objectives and improves performance with QSM and QVPO on 5 MuJoCo tasks and with Flow-SDE+PPO on 4 RoboTwin manipulation tasks. Project page is at https://zhy-liu.github.io/cfg-policy-page.

Chinese Translation

扩散策略在机器人控制与操作、视觉-语言-动作模型以及基于世界模型的智能体中得到了越来越多的应用。它们能够表示多模态动作分布，为强化学习（RL）中的探索提供了丰富的机会。然而，使用RL优化扩散策略可能会降低行为多样性，并进一步限制其探索能力。在RL后训练中，直接微调预训练策略可能会破坏其获得的能力，而动作空间残差可能缺乏多模态探索所需的表达能力。这些问题共同揭示了在线和离线到在线扩散RL中的一个共同挑战：在保持行为多样性和适应能力的同时提高任务性能。我们提出了CFG-Policy，一个统一的即插即用框架，它将探索与利用分为两个生成策略，并通过无分类器引导在采样过程中结合它们的评分或速度场。对于在线RL，一个冻结的探索分支维持行为多样性，而一个利用分支则从任务奖励中学习；对于离线到在线RL，一个预训练的利用分支保持冻结，而一个可训练的分支则引导组合行为。CFG-Policy保留了基础RL目标，并在5个MuJoCo任务上通过QSM和QVPO以及在4个RoboTwin操作任务上通过Flow-SDE+PPO提高了性能。项目页面为 https://zhy-liu.github.io/cfg-policy-page。

View on arXiv Download PDF AI Translation

cs.RO / 73 / 2610.14951

FoundationVIO: Enforcing Metric Scale in Feed-forward Visual-Inertial Odometry

FoundationVIO：在前馈视觉惯性里程计中强制执行度量尺度

Rhee, Jin, Zhang, Yuhao, Dellaert, Frank, Fallon, Maurice

Abstract

Feed-forward geometric foundation models (GFMs) directly predict dense geometry and camera poses from batches of images. Emerging visual SLAM methods chain together consecutive GFM batches to form dense point cloud maps. A core challenge is how to address the ambiguous geometric scale of GFM predictions: independent predictions are locally consistent up-to-scale but not constrained by any direct physical measurements. Models trained on data with known scale improve the consistency of GFM reconstructions but again do not directly measure it. Consequently, trajectories and 3D reconstructions from GFM-enabled SLAM methods inherit this issue. Motivated by parallels with classical visual-inertial odometry, we introduce FoundationVIO which integrates inertial sensing with GFM predictions to recover metric scale reconstruction using a factor-graph based approach. We demonstrate odometry estimates competitive with classical state-of-the-art VIO methods and accurate dense reconstructions that outperform GFM-enabled SLAM methods. We also add a minimal GFM-based loop-closure mechanism to create a full Visual-Inertial SLAM system capable of creating globally accurate reconstructions and pose estimates. We evaluate our method on three datasets with well known sensor systems and demonstrate its compatibility on ubiquitous camera and IMU sensors. The method runs at approximately real-time on a consumer-grade laptop GPU. Code: https://github.com/ori-drs/FoundationVIO.

Chinese Translation

前馈几何基础模型（GFMs）直接从一批图像中预测稠密几何和相机姿态。新兴的视觉SLAM方法将连续的GFM批次串联在一起，以形成稠密点云地图。一个核心挑战是如何解决GFM预测的模糊几何尺度：独立预测在局部上是一致的，但并未受到任何直接物理测量的约束。在已知尺度的数据上训练的模型提高了GFM重建的一致性，但同样没有直接进行测量。因此，基于GFM的SLAM方法的轨迹和3D重建继承了这个问题。受到经典视觉惯性里程计的启发，我们引入了FoundationVIO，它将惯性传感与GFM预测相结合，通过基于因子图的方法恢复度量尺度重建。我们展示了与经典最先进的VIO方法竞争的里程计估计和超越GFM支持的SLAM方法的准确稠密重建。我们还添加了一个最小的基于GFM的回环闭合机制，以创建一个完整的视觉-惯性SLAM系统，能够生成全球准确的重建和姿态估计。我们在三个具有知名传感器系统的数据集上评估了我们的方法，并展示了其与普遍存在的相机和IMU传感器的兼容性。该方法在消费级笔记本GPU上以近实时速度运行。代码：https://github.com/ori-drs/FoundationVIO。

View on arXiv Download PDF AI Translation

cs.RO / 74 / 2610.14977

EnactOpt: Diagnosis-Guided Harness Evolution for Embodied Agents

EnactOpt：面向具身智能体的诊断引导式执行工具演化

Feng, ZhiYuan, Pan, Zhuoshi, Kang, Zhaolu, Chen, Zhongxia, Yang, Shuo, Ye, Sen, Wang, Heng, Li, Han

Abstract

Embodied agents built on foundation models rely on an execution harness to organize observations, guide actions, and incorporate execution feedback into subsequent decisions. Failures can arise not only from limited model capabilities, but also from how the harness presents evidence and guides its use. These model-external mechanisms provide an adaptation target without updating model parameters, but task-specific interaction experience can be costly to collect. Task-level outcomes alone do not reveal whether relevant evidence was missing from the model input or was available but inadequately used. Moreover, these two aspects are coupled: effective action guidance may require evidence that the current context omits, while exposing that evidence alone does not ensure an appropriate response. We introduce EnactOpt, which uses limited execution feedback to jointly evolve reusable textual skills and context-construction code while keeping the underlying models and runtime fixed. It diagnoses adjacent decisions by relating action feedback to the actual model input and subsequent action. Structured episode summaries guide joint revisions through prompt-level edit operators, and candidate harnesses undergo compatibility checks and validation-based retention. Across VLABench, ALFWorld, and Embodied Agent Interface (EAI), EnactOpt improves task success by approximately 10-27 percentage points over the initial harnesses and outperforms the evaluated GEPA, TextGrad, and Trace2Skill baselines. These gains extend to task conditions not used for harness evolution or candidate selection. The results support a complementary route to embodied adaptation: turning limited execution experience into reusable behavioral improvements by jointly refining how evidence is presented and acted upon.

Chinese Translation

基于基础模型的具身智能体依赖执行工具来组织观察、指导行动以及将执行反馈纳入后续决策。失败的原因不仅可能源于模型能力的局限，还可能与执行工具如何呈现证据和指导其使用有关。这些模型外部机制提供了一个适应目标，而无需更新模型参数，但任务特定的交互经验收集成本可能较高。仅凭任务级结果无法揭示相关证据是缺失于模型输入，还是可用但未被充分利用。此外，这两个方面是相互关联的：有效的行动指导可能需要当前上下文中缺失的证据，而单独暴露该证据并不能确保适当的响应。我们提出了EnactOpt，它利用有限的执行反馈共同演化可重用的文本技能和上下文构建代码，同时保持底层模型和运行时不变。它通过将行动反馈与实际模型输入和后续行动相关联来诊断相邻决策。结构化的事件摘要通过提示级编辑操作指导联合修订，候选执行工具经过兼容性检查和基于验证的保留。在VLABench、ALFWorld和具身智能体接口（EAI）上，EnactOpt使任务成功率比初始执行工具提高了约10-27个百分点，并且优于评估的GEPA、TextGrad和Trace2Skill基线。这些增益扩展到未用于执行工具演化或候选选择的任务条件。结果支持了一条补充的具身适应路径：将有限的执行经验转化为可重用的行为改进，通过共同优化证据的呈现和行动方式。

View on arXiv Download PDF AI Translation

cs.RO / 75 / 2610.14997

Barrier-Constrained Guidance For Real-time Asynchronous VLA Execution

实时异步 VLA 执行的障碍约束引导

Yu, Wenbo, Li, Jifan, Xia, Wenke, Li, Yuchen, Zhang, Yizhuo, Ren, Pei, Che, Zhengping, hu, Di

Abstract

While asynchronous execution enables VLA models to operate in real-time environments by predicting the next chunk during execution, it inevitably introduces boundary misalignment with the committed action stream. Existing delay-aware solutions address this at the prohibitive cost of extensive retraining, architectural modifications, or fixed-prefix inpainting that inflates computation when low latency is vital. To address these problems, we propose FreeBar, a training-free real-time inference plugin that enables robust asynchronous execution with only marginal additional inference cost by casting chunk alignment in flow-based VLAs as barrier-guided action-space constraint. Specifically, FreeBar introduces a lightweight Control Barrier Function (CBF)-inspired consistency constraint between consecutive action chunks and solves the resulting correction in closed form, avoiding costly backpropagation through the VLA action head. By progressively tightening the consistency bound during generation, FreeBar aligns generated chunks with the committed execution stream while preserving early-stage generative and within-chunk future action flexibility,. Across Kinetix, RoboCasa, and diverse real-world robotic platforms with multiple VLA backbones, FreeBar improves success rates over naive asynchronous execution by 10%--15% on average, with larger gains in highly dynamic scenarios. It also preserves real-time computational efficiency, decreasing inference latency per action chunk over 45% than RTC. Additionally, as a universal plugin, FreeBar remains robust across distinct latency environments, base policies, and hyperparameter settings.

Chinese Translation

虽然异步执行使 VLA 模型能够通过在执行过程中预测下一个数据块而在实时环境中运行，但它不可避免地引入了与已承诺动作流的边界错位。现有的延迟感知解决方案以广泛的再训练、架构修改或固定前缀填充为代价来解决这个问题，这在低延迟至关重要时会增加计算负担。为了解决这些问题，我们提出了 FreeBar，这是一种无训练的实时推理插件，它通过将流基 VLA 中的数据块对齐视为障碍引导的动作空间约束，从而以仅有的边际额外推理成本实现稳健的异步执行。具体而言，FreeBar 引入了一种轻量级的控制障碍函数（Control Barrier Function, CBF）启发的一致性约束，用于连续动作块之间，并以封闭形式解决所产生的修正，避免了通过 VLA 动作头的高成本反向传播。通过在生成过程中逐步收紧一致性界限，FreeBar 将生成的数据块与已承诺的执行流对齐，同时保留早期生成和块内未来动作的灵活性。在 Kinetix、RoboCasa 以及多个具有不同 VLA 主干的现实世界机器人平台上，FreeBar 在高度动态场景中平均提高了 10%--15% 的成功率，相较于简单的异步执行，且保持了实时计算效率，每个动作块的推理延迟比 RTC 下降超过 45%。此外，作为一个通用插件，FreeBar 在不同的延迟环境、基础策略和超参数设置下仍然保持稳健性。

View on arXiv Download PDF AI Translation

cs.RO / 76 / 2610.15001

Design and Evaluation of a Collaborative Teleoperation Interface for Multi-Operator Multi-Robot Systems in Challenging Environments

在挑战性环境中多操作员多机器人系统的协作遥操作界面的设计与评估

Kenan, Alperen, Cárdenas, Juan José García, Tapus, Adriana, Bremner, Paul, Giuliani, Manuel

Abstract

Multi-robot systems that combine the complementary capabilities of specialised platforms are critical for complex missions in challenging, inaccessible environments. Current deployments assign an operator or operator team to each robot, with collaboration relying on operator-to-operator communication to manually coordinate heterogeneous platforms with incompatible interfaces. This operator-dependent approach increases cognitive load and limits scalability across robots from different manufacturers. This study designs and evaluates a shared-visualisation collaborative teleoperation interface to improve operator coordination, situational awareness, and the task performance in multi-operator, multi-robot systems. A comparative user study with 16 participants compared three teleoperation conditions: co-located operators with a shared interface, remote operators with a shared interface, and co-located operators with individual interfaces. Participants controlled two simulated robots in a nuclear power plant digital twin to complete disaster response tasks. Performance metrics, physiological data, and behavioural measures were recorded to assess task performance, cognitive load, and operator awareness across conditions. The co-located shared-interface condition achieved the fastest mission completion (292.8 s), outperforming remote operation (351.6 s) and individual interfaces (385.1 s), with the fewest collisions (0.9 per trial) and highest situational awareness accuracy (91.1 %). Physical co-presence was associated with reduced cognitive load, with NASA-TLX scores increasing from 38.2 in co-located to 57.7 in remote operation, supported by consistent physiological indicators. These results demonstrate that co-presence and shared interfaces provide complementary benefits, supporting workload regulation and mutual awareness respectively in multi-operator, multi-robot teleoperation.

Chinese Translation

结合专业平台互补能力的多机器人系统对于在复杂、难以接近的环境中执行任务至关重要。目前的部署方式是为每个机器人分配一个操作员或操作员团队，协作依赖于操作员之间的沟通，以手动协调具有不兼容接口的异构平台。这种依赖操作员的方法增加了认知负担，并限制了来自不同制造商的机器人之间的可扩展性。本研究设计并评估了一种共享可视化的协作遥操作界面，以改善多操作员多机器人系统中的操作员协调、情境意识和任务表现。通过对16名参与者进行的比较用户研究，比较了三种遥操作条件：共处操作员使用共享接口、远程操作员使用共享接口，以及共处操作员使用各自接口。参与者在核电厂数字双胞胎中控制两个模拟机器人以完成灾难响应任务。记录了性能指标、生理数据和行为测量，以评估不同条件下的任务表现、认知负担和操作员意识。共处共享接口条件下的任务完成时间最快（292.8秒），优于远程操作（351.6秒）和个体接口（385.1秒），且碰撞次数最少（每次试验0.9次），情境意识准确率最高（91.1%）。身体共在与认知负担降低相关，NASA-TLX评分从共处操作的38.2增加到远程操作的57.7，生理指标一致支持这一点。这些结果表明，共在和共享接口提供了互补的好处，分别支持多操作员多机器人遥操作中的工作负载调节和相互意识。

View on arXiv Download PDF AI Translation

cs.RO / 77 / 2610.15038

Training-Free Multi-Agent Cooperative Embodied Search from Uncertain Vision-Language Priors

无训练的多智能体协作具身搜索：基于不确定视觉-语言先验

Pu, Fan, Wu, Keshu, Jiang, Jiwan, Shen, Shiyu, Fan, Zhiwen, Tu, Zhengzhong, Li, Jiachen, Zhou, Yang

Abstract

Real-world Vision-Language Navigation (VLN) requires agents to act under partial observability, where map structure, semantic cues, and target locations may be incomplete or incorrectly inferred by foundation models. This paper presents a lightweight, training-free, and uncertainty-aware framework for cooperative multi-agent multi-target search in mapless environments. Instead of learning an end-to-end navigation policy, the framework separates high-level semantic reasoning from low-level execution through a scalable graph-optimization backbone. Multimodal Large Language Models (MLLMs) serve only as hypothesis generators, proposing semantic zones, target-location cues, and possible navigable connections. These outputs are treated as uncertain priors and further refined using online observations and graph-context evidence. Cooperative navigation is then formulated as a rolling-horizon multi-agent mixed-integer linear program with a lexicographic objective that first maximizes target-directed reward and then reduces travel distance, uncertain-edge use, and repeated visits. Experiments derived from 100 Matterport3D test cases show that the proposed framework improves target coverage and search efficiency over representative baseline methods, while shared information and joint planning enable more balanced workloads across agents. By operating on an abstract graph without relying on embodiment-specific policy parameters, the framework scales across varying numbers of agents and targets, generalizes to unseen environments, and accommodates different perception modules and low-level controllers without task-specific retraining.

Chinese Translation

现实世界中的视觉-语言导航（VLN）要求智能体在部分可观测的环境中行动，其中地图结构、语义线索和目标位置可能不完整或被基础模型错误推断。本文提出了一种轻量级、无训练且具有不确定性意识的框架，用于在无地图环境中进行协作多智能体多目标搜索。该框架通过可扩展的图优化骨干，将高层语义推理与低层执行分离，而不是学习端到端的导航策略。多模态大型语言模型（MLLMs）仅作为假设生成器，提出语义区域、目标位置线索和可能的可导航连接。这些输出被视为不确定的先验，并通过在线观察和图上下文证据进一步精炼。协作导航随后被构建为一个滚动视野的多智能体混合整数线性规划，具有词典序目标，首先最大化目标导向奖励，然后减少旅行距离、不确定边缘使用和重复访问。基于100个Matterport3D测试案例的实验表明，所提出的框架在目标覆盖率和搜索效率上优于代表性基线方法，而共享信息和联合规划使得智能体之间的工作负载更加平衡。通过在一个抽象图上操作，而不依赖于具身特定的策略参数，该框架能够在不同数量的智能体和目标之间扩展，推广到未见环境，并适应不同的感知模块和低层控制器，而无需特定任务的重新训练。

View on arXiv Download PDF AI Translation

cs.RO / 78 / 2610.15073

Modeling and Control of a Rigid-soft Hybrid-link Humanoid Robot

刚性-柔性混合连杆类人机器人建模与控制

He, Zewen, Ishigaki, Taiki, Yamamoto, Ko

Abstract

Physical flexibility plays a significant role not only in human motions but also in robot motions. For example, a humanoid robot with passive and flexible elements, which can be regarded as a rigid--soft hybrid-link system, has the advantages of storing energy and mitigating impact in a dynamic motion. However, the motion design and control of a hybrid-link system are challenging to achieve and have only been discussed in the literature. In this study, we propose the framework of inverse kinematics and compliance optimization-based control for a hybrid-link humanoid system that represents the continuous deformation of a flexible structure using the piece-wise constant strain (PCS) model. The proposed method was validated using forward dynamics simulations for standing, jumping, and walking motions. Then, we built an experimental setup using a biped humanoid robot with leaf-spring type prosthesis leg, in which the prosthesis deformation was measured by a motion capture system. Using this setup, we experimentally validated the proposed method for standing balance control in both double- and single-support cases.

Chinese Translation

物理柔性在人体运动和机器人运动中都扮演着重要角色。例如，具有被动和柔性元素的类人机器人可以被视为刚性-柔性混合连杆系统，具有储存能量和减轻动态运动冲击的优势。然而，混合连杆系统的运动设计与控制难以实现，相关讨论在文献中较为有限。在本研究中，我们提出了一种基于逆向运动学和柔顺性优化的混合连杆类人系统控制框架，该框架使用分段常量应变（PCS）模型表示柔性结构的连续变形。通过对站立、跳跃和行走运动的前向动力学仿真验证了所提方法。随后，我们建立了一个实验装置，使用具有叶片弹簧型假肢腿的双足类人机器人，并通过运动捕捉系统测量假肢的变形。利用该装置，我们在双支撑和单支撑情况下对站立平衡控制的提议方法进行了实验验证。

View on arXiv Download PDF AI Translation

cs.RO / 79 / 2610.15122

Targeting the Right Situations: Mutation Testing of Fuzzy-Powered Situation Hyperspace

针对正确情境的目标：模糊驱动情境超空间的变异测试

Proma, Nawshin Mannan

Abstract

Ensuring the safety of autonomous Unmanned Aerial Vehicles (UAVs) operating in underground mines is challenging due to the vast and uncertain Operational Design Domain (ODD). Traditional validation approaches, such as random testing, often result in inefficient situation coverage, allocating computational resources to low-risk scenarios while failing to expose critical edge cases. Building on prior work in risk-prioritized Situation Coverage, this paper investigates whether Fuzzy Logic-based risk estimation effectively correlates with real system failures. To this end, we conduct a mutation testing study within our SCALOFT simulation framework. A fixed set of software mutants is systematically injected and evaluated under two testing strategies: (i) uniformly random situation generation and (ii) Fuzzy-Guided risk-prioritized situation generation. Experimental results show that the Fuzzy-Guided approach achieves a higher mutation score and is more effective at exposing difficult-to-detect faults compared to random testing.

Chinese Translation

确保在地下矿井中操作的自主无人机（UAV）的安全性具有挑战性，因为其操作设计域（ODD）广阔且不确定。传统的验证方法，如随机测试，往往导致情境覆盖效率低下，将计算资源分配给低风险场景，同时未能暴露关键的边缘案例。基于先前在风险优先情境覆盖方面的研究，本文探讨了基于模糊逻辑的风险估计是否与真实系统故障有效相关。为此，我们在SCALOFT仿真框架内进行了一项变异测试研究。我们系统地注入并评估了一组固定的软件变异体，采用两种测试策略：（i）均匀随机情境生成和（ii）模糊引导的风险优先情境生成。实验结果表明，模糊引导的方法实现了更高的变异得分，并且在暴露难以检测的故障方面比随机测试更有效。

View on arXiv Download PDF AI Translation

cs.RO / 80 / 2610.15157

RootSLAM: Square Root Bundle Adjustment for Visual-Inertial SLAM

RootSLAM：用于视觉惯性SLAM的平方根束调整

Santucci, Tomás, de Mayo, Mateo, Cremers, Daniel, Pire, Taihú

Abstract

Efficient perception is a prerequisite for the small, low-power devices targeted by robotics, mixed reality, autonomous driving, and drone applications, all of which rely on Simultaneous Localization and Mapping (SLAM) to estimate their motion within their surroundings. Learning-based SLAM systems achieve state-of-the-art accuracy but are demanding in compute, whereas classical geometry-based methods retain a strong efficiency advantage with little to no loss in accuracy. In this work we present RootSLAM, a fully-featured visual-inertial SLAM system for multi-camera setups built on Basalt. RootSLAM achieves accuracy competitive with the state of the art while running up to 20 times faster, facilitating the deployment of perception systems in resource constrained environments. We adapt the square-root bundle adjustment formulation from structure from motion into the multi-camera visual-inertial SLAM setting through a novel reformulation that leverages the system's pose-graph, letting the global optimizer run in single-precision floating point, thus doubling its performance. We provide a thorough benchmark against the state of the art together with an ablation study of our contributions. Our new implementation, including its map management and loop detection module, pose-graph optimizer, and global mapping components, will be open-sourced.

Chinese Translation

高效的感知是机器人、混合现实、自动驾驶和无人机应用等小型低功耗设备的前提，这些应用依赖于同时定位与地图构建（SLAM）来估计其在环境中的运动。基于学习的SLAM系统实现了最先进的精度，但对计算资源的需求较高，而经典的基于几何的方法则在效率上保持了显著优势，并且几乎没有精度损失。在本研究中，我们提出了RootSLAM，一个基于Basalt构建的多摄像头设置的全功能视觉惯性SLAM系统。RootSLAM在精度上与最先进的技术相媲美，同时运行速度快达20倍，从而促进了在资源受限环境中感知系统的部署。我们通过一种新颖的重构方法，将结构光束调整中的平方根束调整公式适应于多摄像头视觉惯性SLAM设置，利用系统的位姿图，使得全局优化器能够以单精度浮点数运行，从而将其性能翻倍。我们提供了与最先进技术的全面基准测试，以及对我们贡献的消融研究。我们的新实现，包括地图管理和回环检测模块、位姿图优化器和全局映射组件，将开源发布。

View on arXiv Download PDF AI Translation

cs.RO / 81 / 2610.15215

nanoflann: A Header-Only KD-Tree Library for Exact Nearest-Neighbor Search on Compile-Time Product Manifolds

nanoflann：一个仅包含头文件的 KD-树库，用于在编译时产品流形上进行精确的最近邻搜索

Blanco-Claraco, José Luis

Abstract

Nearest-neighbor (NN) search underlies countless robotics algorithms, but many state spaces are not Euclidean: orientations live on $SO(2)$ or $SO(3)$ and poses on $SE(2)$ or $SE(3)$, and a Euclidean KD-tree on their raw coordinates returns wrong neighbors. We present a header-only, dependency-free C++ KD-tree library for exact NN search on user-defined product manifolds declared at compile time: the space is composed from base manifolds ($\mathbb{R}^n$, $SO(2)$, $SO(3)$, $S^n$; products $SE(2)$, $SE(3)$, tori) with a single type alias, and the library synthesizes a metric whose pruning is provably admissible, wrap-aware for circular coordinates and double-cover-aware for unit quaternions. Experiments confirm exactness (recall $1.0$ vs an exhaustive scan under the same product metric), where the Euclidean-on-raw shortcut collapses to recall $\approx 0.5^{m}$ for $m$ rotation blocks, with up to three orders of magnitude speedup over brute force. We further contribute an incremental, self-balancing single-tree index for points under heavy churn (e.g. sliding-window LiDAR maps) or $SE(3)$ trajectories: bounded memory and faster updates than prior work, with an optional asynchronous tail-latency-bounding mode. The implementation is open source: https://github.com/jlblancoc/nanoflann

Chinese Translation

最近邻（NN）搜索是无数机器人算法的基础，但许多状态空间并非欧几里得空间：方向位于 $SO(2)$ 或 $SO(3)$，姿态位于 $SE(2)$ 或 $SE(3)$，而在其原始坐标上使用欧几里得 KD-树会返回错误的邻居。我们提出了一个仅包含头文件、无依赖的 C++ KD-树库，用于在编译时声明的用户定义产品流形上进行精确的最近邻搜索：该空间由基础流形（$ ext{R}^n$、$SO(2)$、$SO(3)$、$S^n$；产品 $SE(2)$、$SE(3)$、环面）通过单一类型别名组合而成，库合成了一种度量，其剪枝是可证明的可接受的，且对圆形坐标是感知的，对单位四元数是双覆盖感知的。实验确认了精确性（在相同产品度量下回忆率为 $1.0$ 对比全面扫描），而在原始坐标上使用欧几里得快捷方式的回忆率约为 $0.5^{m}$，其中 $m$ 为旋转块的数量，相较于暴力搜索速度提升可达三个数量级。我们进一步贡献了一个增量的、自平衡的单树索引，用于处理重负载下的点（例如滑动窗口 LiDAR 地图）或 $SE(3)$ 轨迹：内存受限且更新速度比之前的工作更快，且提供可选的异步尾延迟限制模式。该实现是开源的： https://github.com/jlblancoc/nanoflann

View on arXiv Download PDF AI Translation

cs.RO / 82 / 2610.15237

Reinforcement Learning Based Heterogeneous Task Allocation Under Spatio-Temporal Environmental Uncertainty for Multi-Robot Teams

基于强化学习的异构任务分配在时空环境不确定性下的多机器人团队

Lin, Xi, Li, Yaxuan, Szenher, Paul, Englot, Brendan

Abstract

Autonomous Surface Vehicles (ASVs) have been increasingly deployed to perform different kinds of maritime tasks in recent years, which motivates the need for a reliable task allocation policy that can adapt to heterogeneous tasks and maritime disturbances. In this work, we formulate a multi-robot task allocation problem in which a team of ASVs with heterogeneous capabilities needs to complete tasks of different types and appearance times under unknown stochastic spatio-temporal uncertainties from maritime disturbances. We propose a decentralized Reinforcement Learning (RL) based policy network that is trained to be robust towards uncertainty-induced observation deviations, and is able to perform individual decision making that leads to efficient execution behaviors toward heterogeneous task requirements. We evaluate the proposed policy and four baseline methods in simulated experiments with varied proportions of different tasks types, numbers of tasks, robot species and robots per species. Experiment results show that our proposed policy achieves the lowest makespan and can be deployed to address scenarios with 100 robots and 400 tasks.

Chinese Translation

近年来，自主水面车辆（ASVs）被越来越多地部署以执行不同类型的海洋任务，这促使我们需要一种能够适应异构任务和海洋干扰的可靠任务分配策略。在本研究中，我们构建了一个多机器人任务分配问题，其中一组具有异构能力的ASVs需要在来自海洋干扰的未知随机时空不确定性下，完成不同类型和出现时间的任务。我们提出了一种去中心化的基于强化学习（RL）的策略网络，该网络经过训练，能够对不确定性引起的观察偏差具有鲁棒性，并能够进行个体决策，从而有效地执行异构任务要求。我们在模拟实验中评估了所提议的策略和四种基线方法，实验中不同类型任务的比例、任务数量、机器人种类及每种机器人数量均有所变化。实验结果表明，我们提出的策略实现了最低的完工时间，并能够在100个机器人和400个任务的场景中部署。

View on arXiv Download PDF AI Translation

cs.RO / 83 / 2610.15274

Embodiment-Aware Inference-Time Search for Human Reaction Selection in Humanoid Retargeting

考虑具身性的推理时反应选择搜索在类人重定向中的应用

Cao, Yongpeng, Hirano, Masahiro, Yamakawa, Yuji

Abstract

Training humanoid robots to learn human behaviors requires expensive data collection, and human-robot interaction data can be even more difficult and costly to collect. Training a specialized reaction model for each humanoid embodiment can also be time-consuming and resource-intensive. Human-to-humanoid motion retargeting provides a more scalable alternative, allowing human-human reaction data to serve as a common motion representation for different humanoid embodiments. However, due to morphological and kinematic differences between humans and humanoid robots, a reaction preferred in human motion space may not be the most suitable one for execution on a particular robot embodiment. In this work, we propose an inference-time reaction search framework that leverages stochastic sampling from a generative reaction model and jointly considers reaction quality and humanoid embodiment feasibility. Given the observed interaction, our framework generates multiple plausible reaction candidates. Each candidate is retargeted to the target humanoid and evaluated using complementary interaction and embodiment-specific feasibility criteria, allowing candidate selection under the physical constraints of the target embodiment. Since the reaction generator remains frozen, the framework can adapt to different humanoid embodiments at inference time without retraining. Experimental results demonstrate that the proposed framework maintains overall human reaction quality while improving the feasibility of retargeted humanoid reactions across different embodiments. The proposed pipeline improves humanoid retargeting feasibility by 1.61%. Moreover, human-space and G1 best-candidate selections disagree in 78.98% of samples, demonstrating substantial embodiment dependence. Code will be released.

Chinese Translation

训练类人机器人学习人类行为需要昂贵的数据收集，而人机交互数据的收集则更加困难且成本高昂。为每个类人具身体训练专门的反应模型也可能耗时且资源密集。人类到类人的动作重定向提供了一种更具可扩展性的替代方案，使人类-人类反应数据能够作为不同类人具身体的共同动作表征。然而，由于人类与类人机器人之间的形态和运动学差异，在人类运动空间中偏好的反应可能并不适合在特定机器人具身体上执行。在本研究中，我们提出了一种推理时反应搜索框架，该框架利用来自生成反应模型的随机采样，并共同考虑反应质量和类人具身体的可行性。基于观察到的交互，我们的框架生成多个合理的反应候选。每个候选都被重定向到目标类人机器人，并使用互补的交互和具身体特定的可行性标准进行评估，从而在目标具身体的物理约束下选择候选。由于反应生成器保持不变，该框架能够在推理时适应不同的类人具身体，而无需重新训练。实验结果表明，所提出的框架在保持整体人类反应质量的同时，提高了不同具身体之间重定向类人反应的可行性。所提出的管道将类人重定向的可行性提高了1.61%。此外，人类空间和G1最佳候选选择在78.98%的样本中存在不一致，显示出显著的具身依赖性。代码将会发布。

View on arXiv Download PDF AI Translation

cs.RO / 84 / 2610.15294

ProWorld: Multi-View World Model with Proprioceptive Feedback for Closed-Loop Policy Evaluation

ProWorld：具有本体反馈的多视角世界模型用于闭环策略评估

Wang, Zijun, Yan, Tianyi, Lin, Haokun, Hu, Panwen, He, Bohang, Ren, Siyu, Guo, Xiaoyu, Liufu, Weijia, Xue, Shuoshuo, Chen, Fengjiao, Ye, Zipeng, Cao, Xuezhi, Jiang, Zutao, Li, Hanhui, Liang, Xiaodan

Abstract

Evaluating robot manipulation policies through repeated physical trials is costly. We study whether an action-conditioned world model can serve as a closed-loop test environment whose policy success rates track those on real robots. Beyond realistic video, the model must return the camera views and proprioceptive state needed for the policy's next action and preserve failed-action outcomes to avoid overestimating policy capability. We introduce ProWorld, a world model that jointly predicts multi-view video and robot state, then feeds both back to a frozen policy for repeated interaction. ProWorld comprises a shared transformer for joint video-state flow prediction and Shared-Time View Encoding to distinguish cameras while preserving temporal synchronization. To reduce interaction hallucinations that turn missed grasps into successful lifts, we augment training with failed interactions collected in high-fidelity simulation, reducing inflated success rates in the generated environment. ProWorld achieves the highest overall TriWorldBench score among the compared methods. Across three policies and four local manipulation tasks, it attains a Pearson correlation of 0.95 and a Spearman correlation of 0.94 between generated and real success rates. These results support joint visual-proprioceptive feedback and failure-aware training as a basis for world-model-based policy evaluation.

Chinese Translation

通过重复的物理试验评估机器人操作策略成本高昂。我们研究了一个基于动作条件的世界模型是否可以作为一个闭环测试环境，其策略成功率能够与真实机器人上的成功率相匹配。除了提供真实的视频外，该模型还必须返回政策下一个动作所需的相机视图和本体状态，并保留失败动作的结果，以避免高估策略能力。我们引入了ProWorld，这是一种世界模型，它联合预测多视角视频和机器人状态，然后将两者反馈给一个冻结的策略以进行重复交互。ProWorld包含一个共享的变换器，用于联合视频-状态流预测，以及共享时间视图编码，以区分相机同时保持时间同步。为了减少将错过的抓取转变为成功提升的交互幻觉，我们通过在高保真模拟中收集的失败交互增强训练，从而降低生成环境中虚高的成功率。ProWorld在比较方法中获得了最高的TriWorldBench总分。在三个策略和四个局部操作任务中，它在生成和真实成功率之间达到了0.95的Pearson相关性和0.94的Spearman相关性。这些结果支持联合视觉-本体反馈和关注失败的训练作为基于世界模型的策略评估的基础。

View on arXiv Download PDF AI Translation

cs.RO / 85 / 2610.15305

RM-WAM: Reliable Memory-Augmented World Action Models for Long-Horizon Robot Manipulation

RM-WAM：用于长时间跨度机器人操作的可靠记忆增强世界动作模型

Miao, Cui, Yu, Junhao, Liu, Ping, Yang, Ning, Chen, Xuhua, Zheng, He, Wang, Xiaodong, Zhang, Tao

Abstract

World Action Models (WAMs) improve robotic manipulation by jointly predicting future visual states and actions, yet most existing methods make decisions primarily from the current observation. This limits their reliability in long-horizon tasks, where visually similar states may correspond to different interaction histories and task stages. Moreover, directly incorporating historical observations can introduce stale, stage-mismatched, or error-contaminated information, leading to inaccurate world prediction and accumulated closed-loop errors. We propose RM-WAM, a reliable memory-augmented World Action Model that treats memory as a context-dependent predictive resource rather than universally useful historical context. RM-WAM represents history using complementary short-term visual state memory and latent event memory, which respectively capture recent interaction dynamics and longer-term state transitions. A counterfactual memory learning scheme varies the accessible history while keeping the current inputs and prediction targets fixed, enabling the model to learn a context-dependent utility estimate for historical memory. Based on this estimate, a utility-aware gating mechanism adaptively controls memory injection into the world-action backbone. Experiments demonstrate that RM-WAM highlights the importance of reliable and selective memory utilization in long-horizon tasks.

Chinese Translation

世界动作模型（WAMs）通过共同预测未来的视觉状态和动作来改善机器人操作，然而大多数现有方法主要基于当前观察做出决策。这限制了它们在长时间跨度任务中的可靠性，因为视觉上相似的状态可能对应于不同的交互历史和任务阶段。此外，直接纳入历史观察可能会引入过时、阶段不匹配或受错误污染的信息，导致世界预测不准确和累积的闭环误差。我们提出了RM-WAM，一种可靠的记忆增强世界动作模型，将记忆视为依赖上下文的预测资源，而不是普遍有用的历史上下文。RM-WAM使用互补的短期视觉状态记忆和潜在事件记忆来表示历史，分别捕捉最近的交互动态和更长期的状态转变。反事实记忆学习方案在保持当前输入和预测目标不变的情况下，改变可访问的历史，使模型能够学习对历史记忆的上下文相关效用估计。基于该估计，效用感知的门控机制自适应地控制记忆注入到世界动作主干中。实验表明，RM-WAM强调了在长时间跨度任务中可靠和选择性记忆利用的重要性。

View on arXiv Download PDF AI Translation

cs.RO / 86 / 2610.15308

An Open MuJoCo Simulation Framework for a Quadrotor Carrying a Tendon-Driven Soft Continuum Arm

一个开放的MuJoCo仿真框架用于携带腱驱动软连续臂的四旋翼飞行器

El-Hussieny, Haitham

Abstract

Combining aerial robots with soft continuum manipulators promises safe, compliant interaction with the environment, but the coupled dynamics of a free-flying base and a highly deformable, underactuated arm are difficult to study without a suitable simulator. This paper describes an open-source simulation framework, built on the MuJoCo physics engine, for a quadrotor that carries a tendon-driven soft continuum arm underneath its body. The arm is modelled as a pseudo-rigid-body chain of twelve segments joined by pairs of orthogonal compliant hinges. Three tendons, spaced \SI{120}{\degree} apart and routed through every segment, bend it in any direction. A geometric tracking controller on $SE(3)$ flies the vehicle; its rotor allocation matrix is generated directly from the model description. A closed-form tendon-tension allocation keeps every tension within its limits and gives the same bending moment in every direction. We report the full set of modelling choices and parameters, and we evaluate the framework in four simulated scenarios: a \SI{36}{\second} mission with a figure-eight and a tendon-driven tip sweep, an attitude-gain study, a quasi-static characterization of the bending workspace, and an impulsive disturbance applied to the arm tip. With the nominal gains the vehicle tracks the mission with \SI{3.1}{\centi\meter} RMS position error while the arm tip swings up to \SI{37}{\centi\meter} from the body axis. The gain study shows that a low-bandwidth attitude loop lets the arm excite sustained oscillations. The simulation runs about $15\times$ faster than real time on a laptop CPU. All models, controllers and scripts that reproduce the figures are released.

Chinese Translation

将空中机器人与软连续操作器相结合，能够实现与环境的安全、柔性交互，但自由飞行基座与高度可变形、欠驱动臂的耦合动力学在没有合适的仿真器的情况下难以研究。本文描述了一个基于MuJoCo物理引擎的开源仿真框架，适用于在其机身下方携带腱驱动软连续臂的四旋翼飞行器。该臂被建模为由十二个段组成的伪刚体链，这些段通过一对对正交的柔性铰链连接。三根腱以120度间隔分布，并穿过每个段，使其可以向任意方向弯曲。一个在$SE(3)$上的几何跟踪控制器负责飞行器的飞行；其转子分配矩阵直接从模型描述生成。一个封闭形式的腱张力分配保持每个张力在其限制内，并在每个方向上提供相同的弯矩。我们报告了完整的建模选择和参数，并在四个仿真场景中评估了该框架：一个包含八字形轨迹和腱驱动尖端扫动的36秒任务，一个姿态增益研究，一个弯曲工作空间的准静态表征，以及施加于臂尖的冲击干扰。在名义增益下，飞行器在任务中跟踪时的均方根位置误差为3.1厘米，而臂尖摆动至机身轴线的37厘米。增益研究表明，低带宽的姿态环使得臂部激发持续振荡。该仿真在笔记本电脑CPU上运行速度约为实时的15倍。所有模型、控制器和重现图形的脚本均已发布。

View on arXiv Download PDF AI Translation

cs.RO / 87 / 2610.15337

CoLA: Multi-VLA Coordination via Learned Adapters

CoLA：通过学习适配器实现多视觉-语言-动作协调

Kashyap, Ahaskar, Zong, Jiawen, Ma, Boyu, Zhou, Chuhao, Li, Jingliang, Albrecht, Stefano V., Yang, Jianfei

Abstract

Vision-Language-Action (VLA) models are typically deployed as isolated single-agent policies. When several such agents share a workspace, they can coordinate implicitly, by observing one another, or explicitly, by exchanging messages, but it remains unclear when explicit communication becomes necessary. We introduce CoLA (Coordination via Learned Adapters), which keeps each agent's VLA backbone frozen and trains a lightweight message channel and a per-arm diffusion action head on top, with fully decentralised execution. Across three tasks that differ in how much each agent can observe, the channel matters only when information is private. On an occluded tray handover, where the box's colour marker is visible to one agent only, CoLA reaches 76.0% success versus 1.3% with messages zeroed throughout training. On two-arm and three-arm handover tasks, removing messages has no measurable effect with a shared overhead camera (90.6% vs. 88.3% at two arms; 63.5% vs. 60.0% at three), but lowers success on the two-arm task's wrist-only variant (74.8% vs. 64.3%). This architecture generalises across different VLA backbones as well. With a frozen $\pi_{0.5}$ backbone on the same occluded tray task, CoLA matches the overall success rate of a LoRA-fine-tuned baseline (37.3% vs. 34.7%) using a fraction of the trainable parameters. When CoLA places the box, it selects the correct tray 71.8% of the time, whereas the independently fine-tuned baseline, whose placing arm never sees the marker, selects it at chance level (34.7% against 33.3%).

Chinese Translation

视觉-语言-动作（VLA）模型通常作为孤立的单一代理策略进行部署。当多个此类代理共享一个工作空间时，它们可以通过观察彼此隐式协调，或通过交换信息显式协调，但何时需要显式沟通仍不清楚。我们提出了CoLA（通过学习适配器进行协调），该方法保持每个代理的VLA主干网络不变，并在其上训练一个轻量级消息通道和一个每臂扩散动作头，执行方式完全去中心化。在三个任务中，这些任务在每个代理可以观察到的信息量上有所不同，只有在信息是私有时，通道才显得重要。在一个遮挡的托盘交接任务中，只有一个代理可以看到盒子的颜色标记，CoLA的成功率达到了76.0%，而在整个训练过程中消息被归零时成功率仅为1.3%。在双臂和三臂交接任务中，移除消息对共享的上方摄像头没有可测量的影响（双臂任务中为90.6%对88.3%；三臂任务中为63.5%对60.0%），但在双臂任务的仅手腕变体中降低了成功率（74.8%对64.3%）。该架构也能在不同的VLA主干网络之间进行泛化。在同一遮挡托盘任务中，使用冻结的$ ext{π}_{0.5}$主干，CoLA的整体成功率与经过LoRA微调的基线相匹配（37.3%对34.7%），且使用的可训练参数比例更小。当CoLA放置盒子时，正确选择托盘的概率为71.8%，而独立微调的基线（其放置臂从未看到标记）则以偶然水平选择（34.7%对33.3%）。

View on arXiv Download PDF AI Translation

cs.RO / 88 / 2610.15362

LERR-Nav: Learning to Coordinate Exploration, Recovery, and Reminiscing for Object Navigation

LERR-Nav：学习协调探索、恢复和回忆以实现物体导航

Huang, Jingzhi, Huang, Junkai, Peng, Yuhang, Mu, Keyu, Huang, Hailong, Yang, Haoyang, Luo, Renbo, Wang, Lujia, Wang, Yi

Abstract

Object Navigation in unseen environments remains a fundamental challenge in embodied AI, as it requires agents to generalize visual and spatial reasoning to novel layouts. While recent advances in large language model (LLM) and vision-language model (VLM) have enhanced semantic mapping, spatial memory, and waypoint selecting for navigation, successful navigation requires more than selecting the next promising waypoint. A robot must also recognize stalled progress, escape unproductive behaviors, and revisit overlooked target cues. Addressing these challenges requires both specialized navigation strategies and a high-level policy that coordinates their execution. Motivated by the analysis of failed navigation episodes, we propose LERR-Nav, an object navigation framework that coordinates exploration, recovery, and reminiscing through a learned high-level policy. The framework comprises four complementary navigation strategies: value-function-guided exploration, LLM-guided exploration, recovery from stalled or repetitive behavior, and memory-guided reminiscing. To support history-aware coordination, we maintain the execution memory that preserves textual summaries of all strategy executions alongside a sliding window of map sequences. To learn the coordination policy, we initialize a high-level vision-language agent from Qwen2.5-VL-7B-Instruct and train it through a two-stage curriculum. The first stage uses map-centric supervised fine-tuning (SFT) to establish spatial grounding. The second stage performs SFT on variable-length navigation segments to learn which navigation strategy to invoke next based on current maps and the hybrid execution memory. Experiments on HM3D and MP3D demonstrate state-of-the-art navigation performance, while ablation studies validate the contributions of individual components.

Chinese Translation

在未见环境中的物体导航仍然是具身人工智能中的一个基本挑战，因为它要求智能体将视觉和空间推理推广到新颖的布局。尽管最近在大型语言模型（LLM）和视觉-语言模型（VLM）方面的进展增强了语义映射、空间记忆和导航的路径点选择，但成功的导航不仅仅依赖于选择下一个有前景的路径点。机器人还必须识别停滞的进展，摆脱无效的行为，并重新审视被忽视的目标线索。解决这些挑战需要专门的导航策略和协调其执行的高层策略。受到失败导航事件分析的启发，我们提出了 LERR-Nav，一个通过学习的高层策略协调探索、恢复和回忆的物体导航框架。该框架包括四种互补的导航策略：基于价值函数的探索、基于 LLM 的探索、从停滞或重复行为中恢复，以及基于记忆的回忆。为了支持历史感知的协调，我们维护执行记忆，保存所有策略执行的文本摘要，以及地图序列的滑动窗口。为了学习协调策略，我们从 Qwen2.5-VL-7B-Instruct 初始化一个高层视觉-语言智能体，并通过两阶段课程进行训练。第一阶段使用以地图为中心的监督微调（SFT）来建立空间基础。第二阶段对可变长度的导航片段进行 SFT，以学习根据当前地图和混合执行记忆调用下一个导航策略。对 HM3D 和 MP3D 的实验展示了最先进的导航性能，而消融研究验证了各个组件的贡献。

View on arXiv Download PDF AI Translation

cs.RO / 89 / 2610.15375

Have You Seen It Clearly? Uncertainty-Guided Belief Refinement for Zero-Shot Object Navigation

你看得清楚吗？基于不确定性的信念细化用于零-shot物体导航

Huang, Jingzhi, Huang, Junkai, Peng, Yuhang, Mu, Keyu, Zhong, Jiaru, Wang, Yi, Wang, Lujia, Chau, Lap-Pui, Gao, Siyang

Abstract

Zero-Shot Object Navigation (ZSON) requires an agent to locate target objects in unfamiliar environments without additional training for the test categories or environments. Recent approaches leverage large language models and vision-language models to construct spatial representations and guide exploration. However, whether the available evidence sufficiently supports a navigation choice, and how new evidence changes the search value of previously visited locations, remain insufficiently addressed. We propose RefBeliefMap, an uncertainty-guided mapping framework that distinguishes target-location belief from spatial and semantic evidence uncertainty. By jointly modeling intra-observation uncertainty and inter-observation disagreement over a sliding memory window, RefBeliefMap assesses the reliability of both evidence channels. When competing frontiers remain insufficiently resolved, these uncertainty estimates guide active observation and semantic reasoning to refine the evidence supporting target-location beliefs before navigation commitment. Building on the refined map, we introduce RefBeliefNav, which jointly evaluates unexplored frontiers and historical waypoints according to their target-finding potential, expected information gain, and navigation cost. This unified objective allows historical locations to be reconsidered as new evidence changes their search value, coordinating exploration and revisitation. Experiments on HM3D and MP3D demonstrate state-of-the-art performance among the evaluated zero-shot baselines, with ablation studies examining the contributions of modules. Real-world experiments on a Unitree Go2 quadruped robot further demonstrate the sim-to-real transferability of our method.

Chinese Translation

零-shot物体导航（Zero-Shot Object Navigation, ZSON）要求代理在不熟悉的环境中定位目标物体，而无需针对测试类别或环境进行额外训练。最近的方法利用大型语言模型和视觉-语言模型构建空间表示并指导探索。然而，现有证据是否足够支持导航选择，以及新证据如何改变先前访问位置的搜索价值，仍然未得到充分解决。我们提出了RefBeliefMap，一个基于不确定性的映射框架，它将目标位置信念与空间和语义证据的不确定性区分开来。通过在滑动记忆窗口内联合建模观察内不确定性和观察间不一致性，RefBeliefMap评估两个证据通道的可靠性。当竞争前沿仍然不足以解决时，这些不确定性估计指导主动观察和语义推理，以在导航承诺之前细化支持目标位置信念的证据。在经过细化的地图基础上，我们引入了RefBeliefNav，它根据目标发现潜力、预期信息增益和导航成本共同评估未探索的前沿和历史航点。这个统一的目标允许在新证据改变其搜索价值时重新考虑历史位置，从而协调探索和重访。在HM3D和MP3D上的实验表明，我们的方法在评估的零-shot基线中表现出最先进的性能，消融研究考察了各模块的贡献。对Unitree Go2四足机器人进行的真实世界实验进一步展示了我们方法的仿真到现实的可转移性。

View on arXiv Download PDF AI Translation

cs.RO / 90 / 2610.15422

A Reinforcement Learning Framework for Safety-Aware Haptic Shared Control in Surgical Robotics

一种用于手术机器人安全感知触觉共享控制的强化学习框架

Zhang, Zihao, Handwerker, Karl, Varga, Balint

Abstract

Haptic shared control in surgical robotics combines human-side and robot-side forces at a shared end effector while preserving operator involvement. However, the effects of geometric safety choices on task execution and force coordination remain unclear. We model the interaction as a two-agent cooperative Markov game under centralized training and decentralized execution, comparing learned human control with a task-tuned fixed-impedance baseline. This model-free multi-agent reinforcement learning framework supports joint policy adaptation without requiring a complete analytical interaction model. It compares MADDPG and rMAPPO with an epigraph-constrained approach under common geometric task definitions. In simulation, rMAPPO generally achieved the highest return and tracking performance, whereas the epigraph approach produced more conservative motions and lower geometric exposure, including no observed collision or dangerous-region exposure after geometry transfer. With real operator input, fixed-impedance-trained policies were faster and successful in all recorded trials. Policies trained with a learned partner exhibited greater command opposition, while the fixed epigraph condition produced the most aligned commands. This work provides a benchmark for studying human-control representation and constraint-handling trade-offs in haptic shared control.

Chinese Translation

手术机器人中的触觉共享控制结合了人机双方在共享末端执行器上的力量，同时保持操作员的参与。然而，几何安全选择对任务执行和力量协调的影响仍不明确。我们将这种交互建模为一个在集中训练和分散执行下的双代理合作马尔可夫游戏，比较学习到的人类控制与任务调优的固定阻抗基线。该无模型多代理强化学习框架支持联合策略适应，而无需完整的分析交互模型。它在共同几何任务定义下比较了 MADDPG 和 rMAPPO 以及一种受上界约束的方法。在仿真中，rMAPPO 通常实现了最高的回报和跟踪性能，而上界方法则产生了更保守的运动和较低的几何暴露，包括在几何转移后未观察到碰撞或危险区域暴露。在真实操作员输入下，固定阻抗训练的策略在所有记录的试验中都更快且成功。与学习到的伙伴训练的策略相比，固定上界条件下的策略产生了最一致的指令。这项工作为研究触觉共享控制中的人类控制表示和约束处理权衡提供了基准。

View on arXiv Download PDF AI Translation

cs.RO / 91 / 2610.15425

Exploring the Limits of Combinatorial Generalization in Embodied Manipulation

探索具身操控中组合泛化的极限

Zhang, Xupeng, Yang, Yantai, Guo, Chang, Yin, Zhaokai, Zhang, Zhipeng

Abstract

In embodied manipulation, Vision Language Action (VLA) models can execute atomic skills, but long-horizon manipulation requires recombining these skills without demonstrations for every skill combination. The full instruction route treats skill chaining as long-horizon imitation. In a controlled study, full instruction training on limited skill combination sequences reaches 40.0% ordered success on covered combinations but only 7.7% on recombination, suggesting memorization of sequence templates rather than robust skill recombination. We introduce Hierarchical Subtask Execution (HSE), a diagnostic framework that exposes skill recombination through a shared modular interface over the planner, trigger, and policy, decomposing long-horizon failures into separable modes. We introduce RoboCombine, a benchmark for skill recombination without combined skill training trajectories, with metrics for ordered success and transition conversion across subtask boundaries. RoboCombine reveals that atomic competence alone does not ensure ordered execution. Failures concentrate at transitions where downstream subtasks launch from handoff states from preceding subtasks rather than canonical atomic starts. We identify this mismatch as the start state gap. To address it, we introduce Pose Robust Task Initialization (PRTI), a lightweight atomic task post-training strategy that broadens start state coverage beyond canonical starts. Under oracle timed HSE on RoboCombine, PRTI reaches 18.16% ordered success, compared with 4.21% under random perturbation and 9.47% under a robot-reset oracle, without combined skill training trajectories. These results show that combinatorial generalization requires handoff robust atomic skills that convert first stage progress into reliable downstream execution, not merely decomposition into known subtasks. Code: https://github.com/AutoLab-SAI-SJTU/RoboCombine.

Chinese Translation

在具身操控中，视觉语言行动（Vision Language Action, VLA）模型能够执行原子技能，但长时间跨度的操控需要在没有每个技能组合演示的情况下重新组合这些技能。完整指令路径将技能链视为长时间跨度的模仿。在一项受控研究中，针对有限技能组合序列的完整指令训练在覆盖的组合上达到了40.0%的有序成功率，但在重新组合上仅为7.7%，这表明模型更倾向于记忆序列模板而非实现稳健的技能重新组合。我们引入了层次子任务执行（Hierarchical Subtask Execution, HSE），这是一个诊断框架，通过规划器、触发器和策略的共享模块接口揭示技能重新组合，将长时间跨度的失败分解为可分离的模式。我们还推出了RoboCombine，这是一个不依赖于组合技能训练轨迹的技能重新组合基准，提供了有序成功率和跨子任务边界的转换度量。RoboCombine揭示，仅具备原子能力并不能确保有序执行。失败主要集中在下游子任务从前一子任务的交接状态启动，而非从标准的原子起始状态启动的过渡阶段。我们将这种不匹配称为起始状态差距。为了解决这一问题，我们引入了姿态鲁棒任务初始化（Pose Robust Task Initialization, PRTI），这是一种轻量级的原子任务后训练策略，旨在扩大起始状态的覆盖范围，超越标准起始状态。在RoboCombine上，使用oracle定时HSE，PRTI达到了18.16%的有序成功率，而在随机扰动下仅为4.21%，在机器人重置oracle下为9.47%，且不依赖于组合技能训练轨迹。这些结果表明，组合泛化需要具备交接鲁棒性的原子技能，将第一阶段的进展转化为可靠的下游执行，而不仅仅是分解为已知的子任务。代码链接： https://github.com/AutoLab-SAI-SJTU/RoboCombine.

View on arXiv Download PDF AI Translation

cs.RO / 92 / 2610.15486

Learning Safety-Constrained Whole-Body Loco-Manipulation for Lunar Surface Operations

学习安全约束的全身运动操作用于月球表面作业

Belmonte-Baeza, Alvaro, Cazorla, Miguel, García, Gabriel J., Pérez-Del-Pulgar, Carlos J., Pomares, Jorge

Abstract

Robotic systems for lunar surface operations must act autonomously and preserve their own integrity in an environment where failures are unrecoverable and continuous supervision is impossible. This paper presents a Constrained Reinforcement Learning framework for quadrupedal mobile manipulators that learns a single whole-body policy for tracking a 6D end-effector pose specified in task space, with operational limits enforced during learning through a constrained learning mechanism. We introduce a reward formulation in which the transition between locomotion and manipulation emerges from the reward geometry rather than from an explicit gating mechanism, a stability term derived from the Gravito-Inertial Inclination Margin and motivated by the reduced-gravity dynamics of the lunar surface, and a design rule for reward regularization that follows from the multiplicative structure of the termination-based objective. The framework is evaluated in a lunar-analog simulation spanning three terrain morphologies at five difficulty levels under lunar gravity, with regolith contact properties calibrated to the JSC-1a simulant and varied from loose to dense. The learned policy attains mean position and orientation errors of 3.22 cm and 6.83 deg. under nominal conditions, retains this precision across regolith densities and over the operationally representative range of terrain difficulty, and reduces soft-constraint violation time by up to an order of magnitude relative to an unconstrained baseline while incurring no hard-constraint violations.

Chinese Translation

用于月球表面作业的机器人系统必须能够自主行动，并在无法恢复故障且无法进行持续监督的环境中保持自身完整性。本文提出了一种约束强化学习框架，适用于四足移动操作手，旨在学习一个单一的全身策略，以跟踪在任务空间中指定的6D末端执行器姿态，同时在学习过程中通过约束学习机制施加操作限制。我们引入了一种奖励公式，其中运动与操作之间的过渡源自奖励几何结构，而非显式的门控机制，此外还包括一个基于重力惯性倾斜边际（Gravito-Inertial Inclination Margin）推导的稳定性项，受到月球表面低重力动力学的启发，以及一个基于终止目标的乘法结构的奖励正则化设计规则。该框架在一个涵盖三种地形形态和五个难度级别的月球类比模拟中进行了评估，模拟在月球重力下进行，风化层接触特性校准为JSC-1a模拟物，并从松散到致密进行变化。在名义条件下，学习到的策略在位置和方向误差上分别达到了3.22厘米和6.83度，并在不同的风化层密度和操作上具有代表性的地形难度范围内保持了这一精度，同时相较于无约束基线将软约束违反时间减少了一个数量级，而没有发生硬约束违反。

View on arXiv Download PDF AI Translation

cs.RO / 93 / 2610.15491

Phase-Gated Constrained Reinforcement Learning for End-to-End Capture and Detumbling of Non-Cooperative Targets

相位门控约束强化学习用于非合作目标的端到端捕获与去旋转

Belmonte-Baeza, Álvaro, Redondo-Verdú, Celia, Ramón, José L., Cazorla, Miguel, Felicetti, Leonard, Pomares, Jorge

Abstract

Active debris removal requires a servicer spacecraft to approach a non-cooperative tumbling object, capture it, and arrest its rotation. This sequence is conventionally divided among phase-specific controllers, each depending on knowledge of the target's inertia, a dedicated grappling fixture, or a hand-designed switch at the contact-rich seam where such pipelines are most fragile. We show that a single constrained reinforcement learning policy can execute the entire mission, driving a dual-arm free-flying servicer from free-space closing through friction-based fixture-free capture to full detumbling of a target whose mass and inertia are unknown and whose state is observed only through noisy estimates. The enabling mechanism is a smooth proximity gate applied at three loci of a constrained Markov decision process. In the reward it induces the phase structure rather than imposing it; in the action space it forecloses a degenerate grab-and-pull exploit as a hard guarantee; and in the constraint set it resolves a normalization failure of constraint-based learning, in which a constraint spanning phases that load it unequally is rendered nearly inert in the phase where precision matters most. Both the failure mode and the remedy apply to any multi-phase problem of this kind. Over 1000 randomized trials the policy succeeds in 99.5% while holding its contact-force, actuator, base-reaction and collision limits throughout, transfers zero-shot to unseen geometries and centre-of-mass offsets, and extrapolates across a tenfold range of target mass.

Chinese Translation

主动碎片清除要求服务航天器接近一个非合作的翻滚物体，捕获它并停止其旋转。这个过程通常被分为多个相位特定的控制器，每个控制器依赖于对目标惯性的知识、专用的抓取装置或在接触丰富的接缝处设计的开关。我们展示了一个单一的约束强化学习策略可以执行整个任务，驱动一个双臂自由飞行的服务航天器从自由空间接近，通过基于摩擦的无夹具捕获到完全去旋转一个质量和惯性未知且状态仅通过噪声估计观察的目标。启用机制是在约束马尔可夫决策过程的三个位置应用平滑的接近门。在奖励中，它诱导相位结构而不是强加它；在动作空间中，它作为硬性保证排除了退化的抓取和拉动策略；在约束集内，它解决了基于约束学习的归一化失败，其中跨越相位的约束不均匀加载，使得在精度最重要的相位中几乎无效。这种失败模式和补救措施适用于任何此类多相位问题。在超过1000次随机试验中，该策略成功率达到99.5%，同时保持其接触力、执行器、基础反应和碰撞限制，能够零样本转移到未见几何形状和质心偏移，并在目标质量的十倍范围内进行外推。

View on arXiv Download PDF AI Translation

cs.RO / 94 / 2610.15522

GNFormer: Gaussian Neural Field Transformer for Efficient Camera-based Occupancy Prediction

GNFormer：高斯神经场变换器用于高效的基于摄像头的占用预测

Chen, Guangcheng, He, Li, Zhang, Hong

Abstract

Camera-based occupancy prediction provides geometric and semantic scene understanding for robot perception and has been adopted in autonomous vehicle and UAV perception systems. Gaussian primitives offer a promising alternative by potentially representing only occupied regions, thereby concentrating computation and representational capacity where needed and reducing wasted computation and memory in empty space. However, existing Gaussian-based methods suffer from poor primitive utilization, as ineffective training objectives leave some Gaussians in empty space, while the limited expressiveness of Gaussian functions restricts their ability to model complex shapes. To address these limitations, we propose a novel 3D representation, Gaussian Neural Field (GNF), which can accurately map voxel positions to semantic occupancies. In GNF, Gaussians and a neural network jointly parameterize the field, where the Gaussians define its spatial support and provide features for neural decoding, while the neural network enables more expressive occupancy modeling than Gaussian functions alone. We further develop the Gaussian Neural Field Transformer (GNFormer), a streamlined and flexible method that predicts GNFs for camera-based occupancy prediction. Across Occ-ScanNet, SSCBench-KITTI-360, and nuScenes-SurroundOcc benchmarks, GNFormer achieves the highest IoU and mIoU and the fastest inference among the compared single-frame methods. In the experiments, GNFormer can represent indoor and outdoor scenes with only 256 and 1024 Gaussians, respectively, substantially fewer primitives than prior Gaussian-based methods. The code will be released.

Chinese Translation

基于摄像头的占用预测为机器人感知提供了几何和语义场景理解，并已被应用于自主车辆和无人机感知系统。高斯原语通过潜在地仅表示占用区域，提供了一种有前景的替代方案，从而将计算和表示能力集中在需要的地方，减少在空白区域的计算和内存浪费。然而，现有的基于高斯的方法在原语利用率方面表现不佳，因为无效的训练目标使得一些高斯留在空白区域，而高斯函数的有限表达能力限制了它们建模复杂形状的能力。为了解决这些限制，我们提出了一种新颖的3D表示方法——高斯神经场（Gaussian Neural Field, GNF），该方法可以准确地将体素位置映射到语义占用。在GNF中，高斯和神经网络共同参数化该场，其中高斯定义其空间支持并为神经解码提供特征，而神经网络则使得占用建模比单独使用高斯函数更具表现力。我们进一步开发了高斯神经场变换器（GNFormer），这是一种简化且灵活的方法，用于预测基于摄像头的占用预测的GNF。在Occ-ScanNet、SSCBench-KITTI-360和nuScenes-SurroundOcc基准测试中，GNFormer在比较的单帧方法中实现了最高的IoU和mIoU，以及最快的推理速度。在实验中，GNFormer可以分别用仅256和1024个高斯表示室内和室外场景，显著少于之前基于高斯的方法。代码将会发布。

View on arXiv Download PDF AI Translation

cs.RO / 95 / 2610.15548

PointWalker: Mapless Quadruped Locomotion Directly From Raw LiDAR Point Clouds

PointWalker：直接从原始LiDAR点云进行无地图四足运动

Li, Sicen, Chu, Zhen, Li, Chao, Zheng, Jun, Zhu, Qiuguo, Wu, Jun

Abstract

Quadruped robots deployed for inspection and field missions must cross heterogeneous terrain, from stairs, slopes, and gaps to high platforms, stepping stones, and balance beams, where safe support is sparse or see-through. Existing methods interpose a terrain map or a depth image between sensing and control. Maps are costly to maintain, corrupted by odometry drift and outliers, and misled by see-through gratings, while depth cameras degrade with lighting and surface material. In this paper, we present PointWalker, a mapless framework whose single reinforcement learning policy directly consumes raw point clouds from four light detection and ranging (LiDAR) sensors. Voxelized points are encoded by linear self-attention and fused with proprioception through cross-attention, so terrain geometry reaches the policy without maps or foothold estimation. One proximal policy optimization (PPO) run, with terrain-aware rewards and dual-channel simulation of see-through gratings, yields one policy for all seven terrain classes with a 90.1% mean success rate in simulation. Deployed zero-shot on a Deep Robotics X30 at 50 Hz onboard, it crosses 0.8-m gaps, climbs 0.8-m platforms and 45{\deg} steel-grating stairs, carries a 50-kg payload, and completes a 2.5-km mountain route.

Chinese Translation

部署用于检查和现场任务的四足机器人必须跨越异质地形，包括楼梯、坡道、间隙、高平台、踏脚石和平衡梁等，在这些地方安全支撑稀少或透明。现有方法在感知和控制之间插入地形地图或深度图像。地图维护成本高，受里程计漂移和异常值影响，并且在透明栅格的情况下容易产生误导，而深度相机在光照和表面材料变化时性能下降。本文提出了PointWalker，一个无地图框架，其单一的强化学习策略直接处理来自四个激光雷达（LiDAR）传感器的原始点云。体素化点通过线性自注意力进行编码，并通过交叉注意力与自我感知融合，从而使地形几何信息无需地图或支撑点估计即可到达策略。一次近端策略优化（PPO）运行，结合地形感知奖励和透明栅格的双通道仿真，为所有七种地形类别生成了一种策略，在仿真中实现了90.1%的平均成功率。在Deep Robotics X30上以50 Hz的频率零-shot部署，能够跨越0.8米的间隙，攀爬0.8米的平台和45度的钢格栅楼梯，承载50公斤的有效载荷，并完成2.5公里的山地路线。

View on arXiv Download PDF AI Translation

cs.RO / 96 / 2610.15558

Analytic Robot Self-Model: Learning Screw-Axis Kinematics for Rendering and Model-Based Control

解析机器人自我模型：学习螺旋轴运动学以实现渲染和基于模型的控制

Xu, Chen, Tan, Ning

Abstract

Visual self-modelling lets robots learn body models from observations, but such models usually encode kinematics implicitly rather than exposing metric joint parameters for computation and control. Moreover, a good fit to images does not by itself establish accurate kinematics. Here we present an explicit analytic robot self-model that requires neither deformation networks nor URDF/CAD templates. Multi-view images and joint readings establish Gaussian appearance and provide motion evidence for rigid link assignments, whereas sparse end-effector poses independently identify metric screw axes. The resulting closed-form forward kinematics and analytic Jacobian can be evaluated beyond the fitting ranges. We validate the model on a simulated seven-axis arm, a simulated twelve-joint quadruped and a real six-axis arm. In simulation, it achieves millimetre-level reaching and pick-and-place, as well as closed-loop quadruped walking. On the real arm, it renders held-out configurations, predicts end-effector positions with a median error of 6.4 mm, compared with 11.7 mm for a calibrated nominal model, and supports reaching with external marker feedback. A controlled comparison shows that image fit alone is not enough: two Panda models differing only in their joint axes are 0.54 dB apart in PSNR, yet their end-effector prediction errors differ 13.7-fold.

Chinese Translation

视觉自我建模使机器人能够从观察中学习身体模型，但此类模型通常隐式编码运动学，而不是暴露用于计算和控制的度量关节参数。此外，良好的图像拟合本身并不能建立准确的运动学。在此，我们提出了一种显式的解析机器人自我模型，该模型既不需要变形网络，也不需要URDF/CAD模板。多视角图像和关节读数建立高斯外观，并为刚性链接分配提供运动证据，而稀疏的末端执行器姿态独立识别度量螺旋轴。由此产生的封闭形式前向运动学和解析雅可比矩阵可以在拟合范围之外进行评估。我们在一个模拟的七轴臂、一个模拟的十二关节四足动物和一个真实的六轴臂上验证了该模型。在模拟中，它实现了毫米级的到达和抓取，以及闭环四足行走。在真实的机械臂上，它渲染了保留的配置，预测末端执行器位置的中位误差为6.4毫米，而经过校准的名义模型为11.7毫米，并支持带有外部标记反馈的到达。受控比较显示，仅靠图像拟合是不够的：两个仅在关节轴上不同的Panda模型在PSNR上相差0.54 dB，但它们的末端执行器预测误差相差13.7倍。

View on arXiv Download PDF AI Translation

cs.RO / 97 / 2610.15560

Stereo-Planner: Learning Stereo-to-Trajectory Planning for Zero-Shot Sim-to-Real Quadrotor Navigation

立体规划器：学习立体到轨迹规划的零样本仿真到真实四旋翼导航

Shen, Hongming, Lu, Junjie, Chen, Xun, Wang, Wei, Xu, Xinhang, Tian, Bailing, Xie, Lihua

Abstract

Existing navigation systems commonly rely on dedicated depth sensors or explicit depth reconstruction, while learning to generate trajectories directly from RGB images remains challenging due to the sim-to-real domain gap. In this work, we propose Stereo-Planner, an end-to-end vision-to-trajectory framework that enables zero-shot sim-to-real planning from stereo RGB observations. The key idea is to formulate trajectory generation as state- and scale-conditioned querying of latent stereo geometry. A Scene Representor incorporates a pretrained vision grounding foundation model to extract geometric cues shared across simulated and real-world environments, supporting generalization across visual domains. A lightweight Trajectory Decoder then translates these representations into trajectories through interleaved attention, separating trajectory-state interaction from scene-feature retrieval and incorporating the stereo baseline as a metric-scale cue. With these designs, Stereo-Planner is trained entirely on simulated data by minimizing differentiable trajectory costs and deployed directly in real-world environments without fine-tuning. Benchmark results demonstrate strong cross-scene generalization, with a 95% success rate in simulation environments at a commanded speed of 8 m/s, and competitive performance against depth-based planners using only stereo RGB observations for scene perception. Furthermore, deployment on a compact quadrotor equipped with an NVIDIA Orin NX enables autonomous flight at speeds of up to 7 m/s, with a mean onboard vision-to-action inference time of 24.27 ms. The software, simulation environments, and hardware designs will be open-sourced at https://github.com/Stereo-Planner/Stereo-Planner.

Chinese Translation

现有的导航系统通常依赖于专用深度传感器或显式深度重建，而直接从 RGB 图像生成轨迹仍然面临着仿真到真实领域差距的挑战。在本研究中，我们提出了立体规划器（Stereo-Planner），这是一个端到端的视觉到轨迹框架，能够从立体 RGB 观测中实现零样本的仿真到真实规划。其关键思想是将轨迹生成表述为对潜在立体几何的状态和尺度条件查询。场景表示器（Scene Representor）结合了一个预训练的视觉基础模型，以提取在仿真和真实环境中共享的几何线索，从而支持跨视觉领域的泛化。一个轻量级的轨迹解码器（Trajectory Decoder）通过交错注意力将这些表示转换为轨迹，分离轨迹状态交互与场景特征检索，并将立体基线作为度量尺度线索。通过这些设计，立体规划器完全在仿真数据上进行训练，最小化可微分的轨迹成本，并直接在真实环境中部署，无需微调。基准结果表明其在跨场景泛化方面表现出色，在命令速度为 8 m/s 的仿真环境中成功率达到 95%，并且在仅使用立体 RGB 观测进行场景感知时，与基于深度的规划器相比表现竞争。此外，在配备 NVIDIA Orin NX 的紧凑型四旋翼上部署，实现了高达 7 m/s 的自主飞行，平均机载视觉到动作推理时间为 24.27 毫秒。软件、仿真环境和硬件设计将开源于 https://github.com/Stereo-Planner/Stereo-Planner。

View on arXiv Download PDF AI Translation

cs.RO / 98 / 2610.15563

Anticipatory Textile MMG-Strain Sensing for Confirm-Then-Assist Soft Knee Exosuits

预期纺织品机械肌电应变传感用于确认-再辅助软膝外骨骼

Ma, Zheng, Jin, Xiao, Yu, Zhenhua

Abstract

Soft wearable robots must commit to assistance around movement onset, when the available signals are ambiguous. Surface electromyography (sEMG) detects neuromuscular drive early but degrades under sweat and electrode shift, and kinematic sensors respond only once the limb moves. This letter presents a knitted textile interface that co-samples a three-channel mechanomyography (MMG) array and an anterior strain sensor on a single synchronous 500 Hz clock. Across five participants, MMG activation preceded strain-detected knee motion during sit-to-stand by per-channel medians of 322-344 ms. Sampled on the same clock, the sign of the strain slope resolved movement direction on 649 of 650 completed transitions with no training or per-participant calibration, raising four-class balanced accuracy from 90.3% to 95.6%. The fused signals drove a two-stage confirm-then-assist policy for a soft pneumatic knee. Across 240 trials in four participants, the onboard controller entered the assist state on 78 of 80 sit-to-stand trials and on none of the 80 isometric co-contractions. It also entered the assist state on 27 of 80 stand-to-sit trials and left it at the hold timeout, motivating an antagonistic flexor gauge to resolve this direction at onset. Co-sampling muscle acoustics with textile strain on one clock lets the inflation command be issued around movement onset.

Chinese Translation

软穿戴机器人必须在运动开始时承诺提供辅助，此时可用信号往往模糊不清。表面肌电图（sEMG）能够早期检测神经肌肉驱动，但在出汗和电极移动的情况下会降低效果，而运动学传感器仅在肢体移动后才会响应。本文提出了一种编织的纺织界面，该界面在单一同步500 Hz时钟下共同采样三通道机械肌电（MMG）阵列和前向应变传感器。在五名参与者中，MMG激活在坐立过程中，按通道中位数比应变检测到的膝关节运动提前322-344毫秒。由于在同一时钟上采样，应变斜率的符号在650次完成的过渡中解决了运动方向，其中649次没有进行训练或参与者的单独校准，使四类平衡准确率从90.3%提高到95.6%。融合信号驱动了一种两阶段的确认-再辅助策略，用于软气动膝关节。在四名参与者的240次试验中，车载控制器在80次坐立试验中进入辅助状态78次，而在80次等长共同收缩中则没有进入辅助状态。它还在80次站坐试验中进入辅助状态27次，并在保持超时后离开，促使对抗性屈肌测量在开始时解决此方向。将肌肉声学与纺织应变在一个时钟上共同采样，使得在运动开始时能够发出充气指令。

View on arXiv Download PDF AI Translation

cs.RO / 99 / 2610.15572

Look Before You Move: Agent-Interface-Aligned Supervision for Open-World Mobile Manipulation

动手之前先观察：面向开放世界移动操控的智能体-接口对齐监督

Qiu, Ronghe, Ma, Teli, Zhou, Jiaming, Gong, Zeying, He, Yiting, Huang, Qiang, Liang, Junwei

Abstract

Open-world mobile manipulation requires a robot to find an object, approach it, and move it to a target receptacle from a language instruction. Training vision-language model (VLM) agents for these tasks requires demonstrations of both action selection and active vision. Existing pipelines often convert planned trajectories into high-level action labels, but shortest-path progress provides limited supervision for deciding where to look. We introduce Agent-Interface Data Alignment (AIDA), a data collection pipeline in which a teacher directly selects and executes actions through the deployed agent's skill interface. For mobile manipulation, we design teachers that maintain target visibility, recover lost targets, and approach for picking or placing. We add template-based spatial reasoning supervision and use a shared harness to organize observations, history, and outputs during training and inference. The resulting agent, AIDA-OWMM, achieves 71.76% full-task success on 170 held-out OWMM episodes. Teacher sensitivity analysis shows that active-vision demonstrations improve student success. The simulation-trained agent achieves 38% end-to-end success on a real mobile robot without additional training. Applying AIDA to a navigation subset of EmbodiedBench further demonstrates the pipeline's utility for navigation tasks.

Chinese Translation

开放世界移动操控要求机器人根据语言指令找到物体、接近物体并将其移动到目标容器。训练视觉-语言模型（VLM）智能体以完成这些任务需要对动作选择和主动视觉的演示。现有的流程通常将规划的轨迹转换为高级动作标签，但最短路径进展为决定观察位置提供的监督有限。我们提出了智能体-接口数据对齐（AIDA），这是一种数据收集流程，其中教师通过部署的智能体的技能接口直接选择和执行动作。针对移动操控，我们设计了能够保持目标可见、恢复丢失目标并进行拾取或放置的教师。我们增加了基于模板的空间推理监督，并使用共享的框架在训练和推理过程中组织观察、历史和输出。最终生成的智能体AIDA-OWMM在170个保留的开放世界移动操控（OWMM）剧集中实现了71.76%的完整任务成功率。教师敏感性分析表明，主动视觉演示提高了学生的成功率。经过模拟训练的智能体在真实移动机器人上实现了38%的端到端成功率，且无需额外训练。将AIDA应用于EmbodiedBench的导航子集进一步展示了该流程在导航任务中的实用性。

View on arXiv Download PDF AI Translation

cs.RO / 100 / 2610.15577

SIGN-MAPPO: State-Intent-Aware Graph Multi-Agent Reinforcement Learning for Cooperative Driving under Imperfect V2X

SIGN-MAPPO：在不完美V2X环境下的状态-意图感知图多智能体强化学习用于协同驾驶

Fan, Zide, Yuan, Jiahui, Zhang, Kaixin, Fan, Junhao, Li, Xiaohe, Wang, Yiru

Abstract

Cooperative driving at mixed-traffic unsignalized intersections is challenged by imperfect vehicle-to-everything (V2X) communication, human-driven vehicle (HV) intentions that cannot be directly communicated, and the potential incompatibility of independently generated crossing decisions. This paper proposes State-Intent-Aware Graph Multi-Agent Reinforcement Learning for Cooperative Driving under Imperfect V2X (SIGN-MAPPO) to address these challenges through three coordinated mechanisms. Each decision-eligible connected and automated vehicle (CAV) first reconstructs a receiver-specific traffic graph from its ego state and the peer-state records available to it, from which a shared graph-attention policy generates a raw Stop/Go intention augmented with predicted HV-intent features. CAV intentions are then exchanged through timestamped vehicle-to-vehicle (V2V) messages, and a local conflict-aware arbitration layer converts each CAV's intention into a safe executable action despite potentially inconsistent peer information. In the principal Simulation of Urban MObility (SUMO) evaluation, SIGN-MAPPO achieves a 97.53% completion rate, a 0.10% collision rate, and a mean intersection traversal time of 35.28 s. Ablation studies further confirm the complementary contributions of HV-intent prediction and CAV-intent exchange to this performance. The trained policy also remains robust under the evaluated communication-delay and packet-loss ranges, and cross-scenario tests show complete traffic clearance across the tested settings, though safety margins narrow under compound unpredictable HV behavior.

Chinese Translation

在混合交通无信号交叉口，协同驾驶面临着不完美的车对一切（V2X）通信、无法直接传达的人驾驶车辆（HV）意图以及独立生成的过路决策可能不兼容等挑战。本文提出了一种状态-意图感知图多智能体强化学习方法（SIGN-MAPPO），通过三种协调机制来应对这些挑战。每个具有决策资格的连接和自动化车辆（CAV）首先从其自我状态和可用的同行状态记录中重构一个特定接收者的交通图，然后通过共享的图注意力策略生成一个原始的停车/行驶意图，并增强预测的HV意图特征。CAV意图通过带时间戳的车对车（V2V）消息进行交换，局部冲突感知仲裁层将每个CAV的意图转换为安全可执行的动作，尽管同行信息可能不一致。在主要的城市交通模拟（SUMO）评估中，SIGN-MAPPO实现了97.53%的完成率，0.10%的碰撞率，以及35.28秒的平均交叉口通行时间。消融研究进一步确认了HV意图预测和CAV意图交换对这一性能的互补贡献。训练后的策略在评估的通信延迟和数据包丢失范围内仍然保持稳健，跨场景测试显示在测试设置中实现了完全的交通清理，尽管在复杂不可预测的HV行为下安全边际缩小。

View on arXiv Download PDF AI Translation

cs.RO / 101 / 2610.15585

StreamingWAM: Streaming World Action Modeling via Chunk-Wise Staircase Flow Matching

StreamingWAM：通过分块阶梯流匹配实现流式世界动作建模

Ren, Shaojie, Sun, Jingwen, Liu, Zezhi, Chen, Hongjin, Jin, Xin, Zou, Wei, Chen, Zhibo

Abstract

World Action Models (WAMs), often built upon pretrained video-generation world models, have emerged as a promising framework for general-purpose and open-ended robot control. However, their real-world deployment often suffers from discontinuous and jerky execution. Existing methods typically address this issue by either improving inference efficiency or enhancing action continuity across prediction chunks, without jointly addressing both challenges. To bridge this gap, we propose StreamingWAM, a streaming-diffusion framework for WAM-based robot control. StreamingWAM combines a noise ladder for chunk-wise action prediction, a prefix rollout module for cross-chunk continuity, and dense video-history memory for temporally consistent conditioning. Experiments on LIBERO, LIBERO-Plus, RoboTwin 2.0, and real-world robotic tasks demonstrate that StreamingWAM improves action continuity across static and dynamic settings, achieving competitive performance while reducing steady-state replanning latency to 72.9 ms in simulation and 57.3 ms on the real robot. The method also produces smoother action transitions, while the streaming staircase reduces sensitivity to the execution horizon K.

Chinese Translation

世界动作模型（WAMs）通常基于预训练的视频生成世界模型，已成为一种有前景的通用和开放式机器人控制框架。然而，它们在现实世界中的部署往往受到不连续和颠簸执行的影响。现有方法通常通过提高推理效率或增强预测块之间的动作连续性来解决这一问题，但未能同时解决这两个挑战。为了解决这一问题，我们提出了StreamingWAM，一种基于WAM的机器人控制流式扩散框架。StreamingWAM结合了用于分块动作预测的噪声阶梯、用于跨块连续性的前缀展开模块，以及用于时间一致性条件的密集视频历史记忆。在LIBERO、LIBERO-Plus、RoboTwin 2.0和现实世界机器人任务上的实验表明，StreamingWAM在静态和动态环境中改善了动作连续性，实现了具有竞争力的性能，同时将稳态重新规划延迟降低到模拟中的72.9毫秒和真实机器人上的57.3毫秒。该方法还产生了更平滑的动作过渡，而流式阶梯减少了对执行时间范围K的敏感性。

View on arXiv Download PDF AI Translation

cs.RO / 102 / 2610.15639

Imitation Learning via Discrete and Rhythmic Primitive Composition

通过离散和节奏原语组合的模仿学习

Saif, Rafid, Ijspeert, Auke, Lin, Hsiu-Chin

Abstract

Long-horizon robotic manipulation is inherently composed of semantically meaningful sub-tasks. While Movement Primitives (MPs) can model these components, their composition typically relies on manual engineering or sample-inefficient reinforcement learning. Existing supervised methods are limited by the lack of MP labels in standard datasets and often fail to generalize across discrete and rhythmic movements. We propose Neural Dynamic Composition (NDC), a supervised framework to compose Dynamic Movement Primitives (DMPs) using only raw trajectories for supervision, given a single example of each primitive. Our approach exploits parallel combinations of DMPs as a differentiable relaxation of categorical primitive selection, and utilizes a warp-invariant loss formulation to identify primitives despite spatial or temporal variations. A staged training curriculum is employed to learn the selection policy and DMP parameters, unifying discrete and rhythmic movements in one framework. Evaluated on 2D handwriting and 3D robotic tasks, our method outperforms baselines on most tasks and metrics. An ablation study demonstrates how each design choice contributes to the system's performance.

Chinese Translation

长时间范围的机器人操作本质上由语义上有意义的子任务组成。虽然运动原语（Movement Primitives, MPs）可以建模这些组件，但它们的组合通常依赖于手动工程或样本效率低的强化学习。现有的监督方法受到标准数据集中缺乏运动原语标签的限制，且往往无法在离散和节奏运动之间进行泛化。我们提出了神经动态组合（Neural Dynamic Composition, NDC），这是一个监督框架，旨在仅使用原始轨迹进行动态运动原语（Dynamic Movement Primitives, DMPs）的组合，前提是每个原语有一个示例。我们的方法利用DMP的并行组合作为类别原语选择的可微松弛，并采用一种不变于扭曲的损失公式来识别原语，尽管存在空间或时间的变化。我们采用分阶段的训练课程来学习选择策略和DMP参数，将离散和节奏运动统一在一个框架中。在二维手写和三维机器人任务上的评估表明，我们的方法在大多数任务和指标上优于基线。消融研究展示了每个设计选择如何对系统性能产生贡献。

View on arXiv Download PDF AI Translation

cs.RO / 103 / 2610.15675

Recovering Low-Bit Failures in Vision-Language-Action Policies via Quantization-Conditioned Reinforcement Learning

通过量化条件强化学习恢复视觉-语言-动作策略中的低比特故障

Park, Seongmin, Lee, Jaewon, Heo, Yujun, Kim, Yoonsik, Choi, Seheon, Kim, Hyoung-rock, Choi, Jungwook

Abstract

Quantization lowers deployment costs of vision-language-action (VLA) policies but can compromise task completion. Across five VLA models and 4,050 real-world trials, we systematically reveal that quantization-aware training (QAT) generally improves on post-training quantization (PTQ) but leaves substantial gaps to full precision (FP) on long-horizon tasks and under visual and semantic shifts, while its imitation-based objective does not explicitly use rollout success or partial completion. To exploit these outcome signals, we introduce Quantization-Conditioned Reinforcement Learning (QCRL), an offline RL method coupling progress-based advantage estimation with quantization-aware source conditioning. FP rollouts supply reference experience beyond action-level distillation, progress-scaled terminal penalty distinguishes partial completion, and quantization-aware conditioning selects advantageous reference behavior within the quantized policy. We evaluate weight-only INT4 W4A16 on $\pi_{0.5}$ and GR00T N1.5 and joint weight-activation NVFP4 W4A4 on $\pi_{0.5}$. Using only in-distribution data, QCRL reaches 99.3% of FP's mean score across three visual-shift conditions for $\pi_{0.5}$ with INT4 W4A16, alongside 42.9% model-memory savings. With NVFP4 W4A4, it reaches 100.7% of the same FP baseline, alongside a $1.73\times$ speedup using NVFP4/FP8 deployment. Hardware gains are relative to FP16 on Jetson Thor.

Chinese Translation

量化降低了视觉-语言-动作（VLA）策略的部署成本，但可能会影响任务完成。在五个VLA模型和4050个真实世界试验中，我们系统地揭示了量化感知训练（QAT）通常优于后训练量化（PTQ），但在长时间任务和视觉及语义变化下仍然与全精度（FP）存在显著差距，同时其基于模仿的目标并未明确利用回滚成功或部分完成。为了利用这些结果信号，我们引入了量化条件强化学习（QCRL），这是一种离线强化学习方法，将基于进展的优势估计与量化感知源条件结合。FP回滚提供了超越动作级蒸馏的参考经验，基于进展的终端惩罚区分部分完成，量化感知条件选择量化策略中的有利参考行为。我们在$ ext{π}_{0.5}$和GR00T N1.5上评估了仅权重的INT4 W4A16，以及在$ ext{π}_{0.5}$上的联合权重-激活NVFP4 W4A4。仅使用分布内数据，QCRL在$ ext{π}_{0.5}$上使用INT4 W4A16达到了FP在三个视觉变化条件下的平均得分的99.3%，同时节省了42.9%的模型内存。使用NVFP4 W4A4时，它达到了相同FP基线的100.7%，并在使用NVFP4/FP8部署时实现了$1.73 imes$的加速。硬件增益相对于Jetson Thor上的FP16。

View on arXiv Download PDF AI Translation

cs.RO / 104 / 2610.15735

Geometry-Aware Predictive Control for Walk-to-Run Bipedal Obstacle Crossing

基于几何感知的双足障碍跨越行走到奔跑的预测控制

Sahu, Aashish, Kumar, R. Prasanth

Abstract

This paper presents a geometry-aware predictive control framework for bipedal obstacle crossing across walking and running regimes. The proposed approach integrates a phase-structured trajectory generator with a constrained model predictive control (MPC) tracking layer to generate coordinated pelvis and swing-foot motions during obstacle traversal. The crossing process is represented through five interpretable phases: pre-crossing, leading-leg clearance, straddle, trailing-leg clearance, and recovery. Within this structure, swing-foot elevation is adapted as a function of obstacle height, while lateral stance widening and pelvis repositioning are introduced as functions of obstacle depth to improve transverse clearance and body alignment. A continuous gait-parameter evolution enables smooth walk-to-run transition without discrete controller switching. The generated references are tracked using MPC to regulate motion smoothness, clearance feasibility, touchdown behavior, and lateral admissibility. The framework is evaluated in reduced-order simulation over five progressively difficult obstacles under walking and running conditions. The robot successfully crosses the first four obstacles, achieving minimum body-obstacle clearances of 0.036, 0.036, 0.059, and 0.017 m, respectively. The fifth and most demanding obstacle leads to collision, identifying the current operating boundary of the proposed reduced-order formulation. Overall, the method provides a compact, interpretable, and computationally efficient basis for intelligent obstacle-aware bipedal locomotion.

Chinese Translation

本文提出了一种基于几何感知的双足障碍跨越预测控制框架，适用于行走和奔跑状态下的障碍物跨越。所提出的方法将相位结构的轨迹生成器与约束模型预测控制（MPC）跟踪层相结合，以在障碍物穿越过程中生成协调的骨盆和摆动脚运动。跨越过程通过五个可解释的阶段表示：预跨越、领先腿清除、跨越、后腿清除和恢复。在这一结构中，摆动脚的抬升高度根据障碍物高度进行调整，同时引入侧向支撑宽度和骨盆重新定位作为障碍物深度的函数，以改善横向清除和身体对齐。连续的步态参数演变使得行走到奔跑的过渡平滑，无需离散的控制器切换。生成的参考轨迹通过MPC进行跟踪，以调节运动的平滑性、清除的可行性、着陆行为和侧向可接受性。该框架在减少阶数的仿真中评估了五个逐渐增加难度的障碍物，在行走和奔跑条件下进行。机器人成功跨越前四个障碍，分别实现了0.036、0.036、0.059和0.017米的最小身体-障碍物间隙。第五个也是最具挑战性的障碍物导致了碰撞，确定了所提出的减少阶数公式的当前操作边界。总体而言，该方法为智能障碍感知的双足运动提供了一个紧凑、可解释且计算高效的基础。

View on arXiv Download PDF AI Translation

cs.RO / 105 / 2610.15739

Distributed area coverage path planning subject to the inclusion of a headland path

考虑岬角路径的分布式区域覆盖路径规划

Plessen, Mogens

Abstract

A deterministic method for the problem of distributed area coverage path planning subject to the inclusion of a headland path is presented. Given a, in general, non-convex area to be covered that may, in general, include obstacle areas, multiple robots, at least one robot depot, and constraints on each machinery such as a working width, work capacity limits, travel capacity limits, nonholonomic dynamical constraints and, possibly, a robot start and end depot assignment, distributed paths for each machinery are calculated that collectively enable full area coverage. Variations are discussed, in particular, a scenario where unconstrained depots assignment can attain a global optimal solution to the problem. The importance of inclusion of a headland path for area coverage, as well as the benefits in comparison to an alternative Boustrophedon-pattern based approach are highlighted.

Chinese Translation

本文提出了一种确定性方法，用于解决考虑岬角路径的分布式区域覆盖路径规划问题。给定一个一般情况下为非凸的待覆盖区域，该区域可能包含障碍区域，多个机器人，至少一个机器人仓库，以及对每台机械的约束，例如工作宽度、工作能力限制、行驶能力限制、非完整动力约束，以及可能的机器人起始和结束仓库分配，计算出每台机械的分布式路径，以共同实现全面的区域覆盖。讨论了不同的变体，特别是一个场景，其中不受约束的仓库分配可以获得问题的全局最优解。强调了岬角路径在区域覆盖中的重要性，以及与替代的Boustrophedon模式方法相比的优势。

View on arXiv Download PDF AI Translation

cs.RO / 106 / 2610.15740

Multimodal Sensor-Gated Mid-Air Launch and Vision-Based Recovery of a Child UAV on a Hovering Mothership Drone

多模态传感器引导的空中发射与基于视觉的儿童无人机在悬停母船上的回收

Sahu, Aashish, Kumar, R. Prasanth

Abstract

Launching a micro unmanned aerial vehicle (UAV) from a hovering carrier and recovering it in flight requires reliable confirmation that the child is securely seated before the carrier is allowed to descend. This is challenging because vision provides relative guidance but does not directly indicate load transfer, and a downward-facing camera can lose its fiducial near touchdown. This letter presents a mothership-child quadrotor system in which mission transitions are gated by complementary nonvisual sensing. An upward-facing single-beam LiDAR on the recovery deck provides a binary occupancy state using hysteresis and a debounce interval derived from the child's occluding footprint and crossing speed; real-time kinematic (RTK) positioning conditions the carrier hover; and carrier thrust provides a proprioceptive indication of load transfer. In ten outdoor missions using an F450-class mothership and a Q250-class child with camera-guided recovery, nine completed successfully while RTK-fixed positioning was maintained through release and capture. A complementary MuJoCo study of fully autonomous fiducial-based recovery achieved touchdown 0.077 m from the deck center, while marker detection was lost within the final approximately 0.2 m above the docked configuration. In simulation, carrier thrust changed by -17.0 N at release and +17.3 N at capture, agreeing within 6% with the child's 18.0 N hover thrust. The results support a sensing architecture in which vision guides the approach, while range and proprioceptive information corroborate the final docking decision.

Chinese Translation

从悬停的载体上发射微型无人机（UAV）并在飞行中回收它需要可靠确认儿童在发射前安全就座。这一过程具有挑战性，因为视觉提供相对引导但并不直接指示负载转移，而向下的摄像头在接近着陆时可能会失去其基准。本文提出了一种母船-儿童四旋翼系统，其中任务转换由互补的非视觉传感引导。回收平台上的向上单束激光雷达（LiDAR）利用滞后和基于儿童遮挡足迹及其穿越速度的消抖间隔提供二元占用状态；实时动态定位（RTK）条件控制载体悬停；而载体推力则提供了负载转移的本体感知指示。在十次户外任务中，使用F450级母船和Q250级儿童无人机进行摄像头引导的回收，其中九次成功完成，同时在释放和捕获过程中保持RTK固定定位。对完全自主基于基准的回收的互补MuJoCo研究在距离平台中心0.077米处实现了着陆，而在最后约0.2米的高度内失去了标记检测。在仿真中，载体在释放时推力变化为-17.0 N，在捕获时为+17.3 N，与儿童的18.0 N悬停推力相符，误差在6%以内。结果支持了一种传感架构，其中视觉引导接近，而距离和本体感知信息证实了最终对接决策。

View on arXiv Download PDF AI Translation

cs.RO / 107 / 2610.15747

Model-Based Trailer-Sway Suppression for Autonomous Passenger Vehicles

基于模型的自动驾驶乘用车拖车摆动抑制

McFadyen, Alec, De La Torre, Pablo Covarrubias, McGovern, Connor

Abstract

Trailer sway is a persistent contributor to severe crashes among light passenger vehicles towing trailers, yet it remains largely unaddressed by the autonomous vehicle systems now entering this new era of automation. Improper cargo placement rearward of the trailer axle reduces, and can invert, the tongue load transmitted through the hitch. The resulting loss of vertical load at the tow vehicle's rear axle diminishes the lateral tire force available there and degrades yaw stability of the coupled system, admitting a divergent lateral oscillation of the trailer known as trailer sway. Under adverse combinations of speed and loading, this oscillation can progress to jackknifing, in which the articulation angle diverges rapidly and directional control is lost. This work develops a linear state-space model of the yaw-plane dynamics of a light vehicle-trailer combination, parameterized to expose tongue load as an explicit stability-determining quantity. From this model a linear quadratic regulator is synthesized to suppress articulation-angle divergence. Because consumer trailers carry no sensors, the two trailer states the controller needs, trailer yaw rate and articulation angle, are estimated by a Kalman filter from measurements taken only on the tow vehicle. The controller is then implemented and evaluated in a MuJoCo multibody simulation across a range of longitudinal speeds and tongue-load conditions, including the negative-tongue-load regime. Results show a significant decrease in trailer oscillation when controlling a given vehicle's steering with an LQR using tow-vehicle IMU and LiDAR data. These findings indicate that LQR controlled anti-sway systems are a viable control method for future autonomous vehicles.

Chinese Translation

拖车摆动是轻型乘用车拖曳拖车时导致严重碰撞的一个持续因素，但在当前进入自动化新纪元的自动驾驶系统中，这一问题仍然未得到充分解决。拖车轴后方不当的货物放置会减少甚至反转通过拖钩传递的舌头负载。随之而来的拖车后轴的垂直负载损失降低了可用的侧向轮胎力，并削弱了耦合系统的偏航稳定性，从而引发了被称为拖车摆动的侧向振荡。在不利的速度和负载组合下，这种振荡可能会发展为折叠（jackknifing），其中关节角度迅速发散，导致方向控制丧失。本研究开发了轻型车辆-拖车组合的偏航平面动力学的线性状态空间模型，参数化以明确舌头负载作为稳定性决定量。从该模型合成了线性二次调节器（LQR），以抑制关节角度的发散。由于消费者拖车不配备传感器，控制器所需的两个拖车状态（拖车偏航率和关节角度）通过卡尔曼滤波器从仅在拖车上测得的拖曳车辆数据中进行估计。然后在MuJoCo多体仿真中实现并评估该控制器，涵盖了多种纵向速度和舌头负载条件，包括负舌头负载状态。结果表明，当使用拖曳车辆的IMU和激光雷达数据通过LQR控制给定车辆的转向时，拖车振荡显著减少。这些发现表明，LQR控制的抗摆动系统是未来自动驾驶车辆的一种可行控制方法。

View on arXiv Download PDF AI Translation

cs.RO / 108 / 2610.15756

Kinematic Reward, Measured Effort: Hardware-Validated RL Morphology Optimization of a Planar 2R Manipulator

运动学奖励与测量努力：基于硬件验证的平面2R操纵器的强化学习形态优化

Gupta, Gunesh, Mishra, Arvind Kumar, Chakrabarty, Sohom

Abstract

Yoshikawa's Manipulability Index is generally used as a classical measure to define an optimal morphology of a manipulator. However, any such optimization is almost always studied as a purely kinematic quantity in simulation and is rarely validated on hardware. We present a reinforcement-learning framework with RSL-RL PPO that optimizes the morphology of a 2R SCARA-type manipulator arm with Yoshikawa's kinematic manipulability index as a core reward, and the analytically derived closed-form optimum for the circular task is used to verify that our method recovers it. Then the same method is applied to custom elliptical and rectangular paths having no analytical optimum. The RL-optimized morphologies are then transferred to a 2R hardware manipulator to validate its effect in reduction of joint load in the hardware. Compared to the worst morphology, the optimum morphology delivered by our method reduces second joint load by 38.4 +/- 0.6% for the ellipse path and 41.3 +/- 0.3% for the rectangle, and first joint load by 15%.

Chinese Translation

吉川的可操作性指数通常被用作定义操纵器最佳形态的经典度量。然而，任何此类优化几乎总是作为纯粹的运动学量在仿真中进行研究，且很少在硬件上进行验证。我们提出了一个强化学习框架，使用RSL-RL PPO优化2R SCARA型操纵臂的形态，以吉川的运动学可操作性指数作为核心奖励，并使用解析推导的圆形任务的闭式最优解来验证我们的方法是否能够恢复该最优解。然后，将相同的方法应用于没有解析最优解的自定义椭圆和矩形路径。经过强化学习优化的形态随后被转移到2R硬件操纵器上，以验证其在减少硬件关节负载方面的效果。与最差形态相比，我们的方法提供的最佳形态在椭圆路径上将第二关节负载减少了38.4 +/- 0.6%，在矩形路径上减少了41.3 +/- 0.3%，并将第一关节负载减少了15%。

View on arXiv Download PDF AI Translation

cs.RO / 109 / 2610.15763

Competitive Task Allocation with Temporal Proposal Windows and Improved Artificial Potential Fields for Dual-Arm Industrial Manipulation

具有时间提议窗口和改进人工势场的双臂工业操作的竞争任务分配

K, Surya Prakash S., Narula, Bhuvan, Sahoo, Jagannath Prasad, VK, Jayakumar, Hambarde, Praful, Shukla, Amit

Abstract

Modern industrial automation demands autonomous collaborative systems capable of coordinated manipulation in workspaces with heterogeneous object arrangements. Existing systems rely on predetermined trajectories and rigid allocation strategies, causing suboptimal workspace utilization and collision risks. This paper presents a distance-based competitive task allocation framework for dual-arm systems, integrating real-time vision perception with collision-free motion planning. A temporal window mechanism enables each arm to evaluate object detections over five consecutive vision frames before committing to optimal distance-based selection, featuring dynamic reallocation and spatial conflict resolution at 0.01m threshold accuracy. An improved Artificial Potential Field (iAPF) framework provides coordinated dual-arm movements through exponential goal attraction, inter-arm repulsion, home-seeking, and velocity damping forces. Simulation results validate superiority over market-based, round-robin, and fixed allocation strategies in distance efficiency, collision risk, and workspace coverage. Hardware experiments using two 6-DOF Kinova Gen3 Lite manipulators demonstrate approximately 47% average velocity improvement over fixed allocation in homogeneous scenarios with zero collisions, while mixed object scenarios confirm universal handling with approximately 11% time overhead offset by superior safety margins. The framework integrates YOLOv8 OBB detection at 30 FPS with 10 Hz control loops, demonstrating practical deployment capability for industrial assembly and sorting operations.

Chinese Translation

现代工业自动化要求自主协作系统能够在异构物体排列的工作空间中进行协调操作。现有系统依赖于预先确定的轨迹和刚性分配策略，导致工作空间利用率低下和碰撞风险。本文提出了一种基于距离的双臂系统竞争任务分配框架，结合实时视觉感知与无碰撞运动规划。时间窗口机制使每个手臂能够在承诺进行基于距离的最优选择之前，评估连续五帧视觉图像中的物体检测，具有动态重新分配和0.01米阈值精度的空间冲突解决能力。改进的人工势场（iAPF）框架通过指数目标吸引、臂间排斥、寻家行为和速度阻尼力实现协调的双臂运动。仿真结果验证了该框架在距离效率、碰撞风险和工作空间覆盖方面优于市场基础、轮询和固定分配策略。在使用两个6自由度Kinova Gen3 Lite机械手的硬件实验中，在同质场景中实现了约47%的平均速度提升，且无碰撞，而混合物体场景则确认了普遍处理能力，时间开销约11%被更优的安全边际所抵消。该框架在30 FPS下集成了YOLOv8 OBB检测与10 Hz控制循环，展示了在工业组装和分拣操作中的实际部署能力。

View on arXiv Download PDF AI Translation

cs.RO / 110 / 2610.15766

SurgiCLAW: An Interpretable Atomic-Skill Agent for Precision Robotic Micromanipulation

SurgiCLAW：一种可解释的原子技能代理用于精确的机器人微操控

Ye, Angen, xie, junjie, Song, Yujia, Huang, Junkai, Zhang, Dapeng

Abstract

Recent multimodal large language models (MLLMs) and vision--language--action models have provided robots with new capabilities for long-horizon task understanding and planning. In surgical robotics, however, end-to-end prediction of low-level actions remains limited by insufficient precision, opaque decisions, and scarce data. This paper presents SurgiCLAW, a hierarchical skill-based framework for precision surgical manipulation. SurgiCLAW confines an MLLM to intraoperative visual understanding, structured explanation, and atomic surgical-skill selection, while delegating physical execution to verifiable deterministic control programs. For brain--machine-interface implantation, a representative high-precision surgical task, we construct a state-to-skill invocation dataset and expand it through visual and state perturbations. The model first learns structured skill invocation through supervised fine-tuning and is then optimized using three rewards for action correctness, operation prerequisites, and format validity, improving action selection, operational safety, and output parseability. We evaluate subtask success, offline model comparison, real-robot performance, and ablations. On a fixed test set, SurgiCLAW achieves 94.6% skill-selection accuracy and 93.0% macro-averaged F1 (Macro-F1). Across 50 closed-loop real-robot implantation trials, it achieves an 88.0% task success rate.

Chinese Translation

最近的多模态大型语言模型（MLLMs）和视觉-语言-动作模型为机器人提供了新的能力，以理解和规划长时间跨度的任务。然而，在外科机器人领域，低级动作的端到端预测仍然受到精度不足、决策不透明和数据稀缺的限制。本文提出了SurgiCLAW，一个基于层次技能的精确外科操控框架。SurgiCLAW将MLLM限制在术中视觉理解、结构化解释和原子外科技能选择，同时将物理执行委托给可验证的确定性控制程序。对于脑机接口植入这一具有代表性的高精度外科任务，我们构建了一个状态到技能调用的数据集，并通过视觉和状态扰动进行扩展。该模型首先通过监督微调学习结构化技能调用，然后使用针对动作正确性、操作前提和格式有效性的三个奖励进行优化，从而提高动作选择、操作安全性和输出可解析性。我们评估了子任务成功率、离线模型比较、真实机器人性能和消融实验。在固定测试集上，SurgiCLAW实现了94.6%的技能选择准确率和93.0%的宏平均F1（Macro-F1）。在50次闭环真实机器人植入试验中，它达到了88.0%的任务成功率。

View on arXiv Download PDF AI Translation

cs.RO / 111 / 2610.15769

Hand2Robot: Object-Centric 4D Hand Reconstruction from Human Demonstrations for Robotics

Hand2Robot：基于人类示范的面向对象的4D手部重建框架用于机器人学

Du, Zetao, Wang, Weiqing, Zhang, Hanbing, Zhou, Yimiao, Shi, Ye, Wang, Liang, Wang, Jingya

Abstract

Human demonstrations can capture task-relevant manipulation without using the target robot, but the recorded hand motions cannot directly serve as robot action supervision. Hand reconstruction and motion retargeting provide a direct route to such supervision, yet recovering hand motion alone does not ensure physically consistent hand-object interactions. To address this issue, we introduce Hand2Robot, an object-centric 4D hand reconstruction framework for robot policy post-training. Given videos of humans performing the target task, Hand2Robot extracts interaction clips, reconstructs object geometry, and generates robot action supervision through hand-object interaction optimization and motion retargeting. For object tracking, we combine observation confidence with temporal motion constraints to select a globally consistent object pose trajectory from per-frame pose candidates. For hand refinement, we use object geometry and motion as a reference and combine staged 2D alignment with object-local contact constraints to refine hand trajectories and finger joint configurations. Experiments on DexYCB and HO3D-v3 show improved object pose accuracy and hand-object interaction quality. Experiments on three real-robot dexterous manipulation tasks further show that reconstructed and retargeted human demonstrations can effectively supplement limited robot data for target-task policy post-training.

Chinese Translation

人类示范能够捕捉与任务相关的操作，而无需使用目标机器人，但记录的手部动作无法直接作为机器人动作的监督。手部重建和动作重定向提供了一条直接的监督途径，然而，仅仅恢复手部动作并不能确保手与物体之间的物理一致性交互。为了解决这一问题，我们提出了Hand2Robot，一个面向对象的4D手部重建框架，用于机器人策略的后训练。给定人类执行目标任务的视频，Hand2Robot提取交互片段，重建物体几何形状，并通过手与物体的交互优化和动作重定向生成机器人动作监督。对于物体跟踪，我们将观察置信度与时间运动约束相结合，从每帧姿态候选中选择一个全局一致的物体姿态轨迹。对于手部精细化，我们使用物体几何形状和运动作为参考，并结合分阶段的2D对齐与物体局部接触约束来精细化手部轨迹和手指关节配置。在DexYCB和HO3D-v3上的实验表明，物体姿态的准确性和手与物体的交互质量得到了改善。在三个真实机器人灵巧操作任务上的实验进一步表明，重建和重定向的人类示范可以有效补充有限的机器人数据，以支持目标任务的策略后训练。

View on arXiv Download PDF AI Translation

cs.RO / 112 / 2610.15794

$\Delta$ Flow Matching for Robot Foundation Models

机器人基础模型的$ riangle$流匹配

Zuo, Kuangji, An, Tuo, Guo, Xinying, Li, Gen, Zhao, Mengfei, Bai, Jiaqi, Lyu, Bofan, Yu, Zhuoyuan, Yang, Jinghan, Li, Jiayi, Jia, Jindou, Yang, Jianfei

Abstract

Flow matching is a common generative recipe for robot foundation models. Vision-language-action models (VLAs) generate action chunks, while world-action models (WAMs) additionally generate future visual states. Both start generation from a fixed Gaussian prior, despite temporal continuity between past and future states and actions. This requires costly training and iterative sampling, especially for WAMs with an additional visual branch. We introduce $\Delta$-flow, which constructs the source from recent history with mild noise rather than from a fixed Gaussian prior. This moves the source closer to the target, reducing transport to a small residual $\Delta$ that can be learned efficiently and enables single-step inference. We integrate $\Delta$-flow into the VLA policy $\pi_{0.5}$ and the WAM policies ImageWAM and Fast-WAM. On LIBERO, ImageWAM reaches 90% success in one epoch rather than six, and Fast-WAM reaches 95% in two epochs rather than eight. With LoRA fine-tuning, $\pi_{0.5}$ exceeds its baseline's five-epoch performance after one epoch. Single-step inference gives $1.54\times$, $2.49\times$, and $4.45\times$ speedups on $\pi_{0.5}$, ImageWAM, and Fast-WAM, respectively. It preserves action smoothness, achieving a $4.5\times$ reduction in action jerk compared with one-step Gaussian sampling. $\Delta$-flow reaches state-of-the-art generalization performance on LIBERO-Plus. Action-history sources can be applied to pretrained WAM policies without retraining. Replacing the Gaussian action source at inference largely preserves task success with a single sampling step and can slightly improve it. We further study how action representation and visual information in action learning affect history-to-future prediction in $\Delta$-flow. These results show that $\Delta$-flow is simple yet effective, improving the efficiency and generalization of robot foundation models.

Chinese Translation

流匹配是一种常见的生成方法，用于机器人基础模型。视觉-语言-动作模型（VLA）生成动作片段，而世界-动作模型（WAM）则额外生成未来的视觉状态。尽管过去和未来的状态与动作之间存在时间连续性，但两者都从固定的高斯先验开始生成。这需要昂贵的训练和迭代采样，尤其是对于具有额外视觉分支的WAM。我们提出了$ riangle$-流，它从近期历史构建源，而不是从固定的高斯先验中构建，且加入了轻微的噪声。这使得源更接近目标，减少了传输至小的残差$ riangle$，可以高效学习，并实现单步推理。我们将$ riangle$-流集成到VLA策略$ heta_{0.5}$和WAM策略ImageWAM及Fast-WAM中。在LIBERO上，ImageWAM在一个周期内达到90%的成功率，而不是六个周期，Fast-WAM在两个周期内达到95%的成功率，而不是八个周期。通过LoRA微调，$ heta_{0.5}$在一个周期后超过其基线的五个周期性能。单步推理在$ heta_{0.5}$、ImageWAM和Fast-WAM上分别实现了$1.54 imes$、$2.49 imes$和$4.45 imes$的加速。它保持了动作的平滑性，与一步高斯采样相比，动作的抖动减少了$4.5 imes$。$ riangle$-流在LIBERO-Plus上达到了最先进的泛化性能。动作历史源可以在不重新训练的情况下应用于预训练的WAM策略。在推理时替换高斯动作源在单次采样步骤中大致保持任务成功率，并且可能略有改善。我们进一步研究了动作表示和动作学习中的视觉信息如何影响$ riangle$-流中的历史到未来预测。这些结果表明，$ riangle$-流简单而有效，提高了机器人基础模型的效率和泛化能力。

View on arXiv Download PDF AI Translation

cs.RO / 113 / 2610.15814

ReTrace: Cross-Stage Failure Attribution for Self-Improving Robotic Manipulation

ReTrace：用于自我改进机器人操作的跨阶段故障归因

Yu, Yue, Zhu, Bin, Li, Junhui, Wang, Jiayu, Zhou, Shijie, Chen, Jingjing

Abstract

General-purpose multimodal foundation models are increasingly capable of directly controlling robots, yet improving their manipulation capabilities through autonomous interaction remains challenging. A key challenge to such self-improvement is learning from failures in complex manipulation tasks involving cross-stage dependencies, where an action may succeed locally but cause a later stage to fail. To address this challenge, we introduce ReTrace, a self-improvement framework for robotic manipulation through autonomous physical interaction and retrospective reasoning, without provided demonstrations or model-parameter updates. ReTrace retrospectively traces downstream failures to earlier decisions, proposes targeted corrections, and evaluates them through subsequent interactions. Across attempts, ReTrace progressively consolidates validated corrections into a reusable manipulation skill that guide subsequent execution. We evaluate ReTrace on four real-world manipulation tasks using a Franka Research 3 robot. On the two tasks with strong cross-stage dependencies, GPT-6 Astra-low improves from 0% to 80% success on plug insertion and from 0% to 100% on ball pouring, while GPT-6 Sol-xhigh improves from 0% to 10% and 0% to 90%, respectively. The learned skills also transfer across model and reasoning configurations and reduce inference-time decision cost in most evaluated settings.

Chinese Translation

通用多模态基础模型越来越能够直接控制机器人，但通过自主交互提升其操作能力仍然具有挑战性。自我改进的一个关键挑战是从涉及跨阶段依赖的复杂操作任务中的失败中学习，其中一个动作可能在局部成功但导致后续阶段失败。为了解决这一挑战，我们提出了ReTrace，一个通过自主物理交互和回溯推理实现机器人操作自我改进的框架，无需提供演示或模型参数更新。ReTrace回溯追踪下游失败到早期决策，提出针对性的修正，并通过后续交互进行评估。在多次尝试中，ReTrace逐步将验证过的修正整合为可重用的操作技能，以指导后续执行。我们在四个真实世界的操作任务上评估了ReTrace，使用的是Franka Research 3机器人。在两个具有强跨阶段依赖的任务中，GPT-6 Astra-low在插头插入任务中的成功率从0%提升至80%，在球倒入任务中的成功率从0%提升至100%；而GPT-6 Sol-xhigh在这两个任务中的成功率分别从0%提升至10%和从0%提升至90%。所学习的技能也能够跨模型和推理配置进行迁移，并在大多数评估设置中降低推理时的决策成本。

View on arXiv Download PDF AI Translation

cs.RO / 114 / 2610.15833

P-BFM: Perceptive Behavioral Foundation Models for Humanoid Control via Unsupervised Reinforcement Learning

P-BFM：通过无监督强化学习实现类人控制的感知行为基础模型

Wang, Xue, Li, Yitang, Zhang, Yiming, Wang, Siqin, Sun, Junhan, Liu, Zhihao, Shen, Yaojie, Yuan, Jingbo, Zhao, Hao, Li, Qingdu

Abstract

Behavioral foundation models (BFMs) provide humanoids with reusable repertoires of whole-body behaviors; in real-world humanoid control, the same behavioral intent may require fundamentally different motions under different terrain constraints. Covering such variation with terrain-matched demonstrations requires enumerating a combinatorial space of behaviors and terrain geometries. We introduce P-BFM, a perceptive behavioral foundation model that factorizes humanoid control into terrain-invariant intent and terrain-conditioned realization. A forward-backward representation maps motion references, target states, and reward functions into a shared intent space without terrain information, while a perceptive policy learns how to realize each intent under the surrounding terrain geometry. We further identify a complementary limitation of purely latent control: behavioral latents can preserve local motion patterns while losing precise global progress. We therefore introduce Explicit Root Instructions (ERI), which separates global transport (e.g., heading and planar velocity) from behavioral intent, while leaving local whole-body adaptation to the perceptive policy. This decomposition leads to a characteristic capability absent from conventional BFMs: one fixed intent induces different, terrain-appropriate behaviors, including changes in step height, stride, posture, foothold placement, and contacts, without terrain-specific demonstrations. Across challenging terrains, P-BFM improves motion fidelity, root-control precision, and perturbation recovery over strong baselines. A single frozen policy further supports motion tracking, goal reaching, reward inference, teleoperation, and terrain-aware recovery, and transfers zero-shot to a Unitree G1 using onboard depth sensing.

Chinese Translation

行为基础模型（BFMs）为类人机器人提供可重用的全身行为库；在现实世界的类人控制中，相同的行为意图在不同的地形约束下可能需要根本不同的动作。通过与地形匹配的示范来覆盖这种变化需要枚举行为和地形几何的组合空间。我们提出了P-BFM，一种感知行为基础模型，它将类人控制分解为与地形无关的意图和受地形影响的实现。前向-后向表示将运动参考、目标状态和奖励函数映射到一个不包含地形信息的共享意图空间，而感知策略则学习如何在周围地形几何下实现每个意图。我们进一步识别出纯潜在控制的一个互补限制：行为潜在可以保留局部运动模式，但会失去精确的全局进展。因此，我们引入了显式根指令（Explicit Root Instructions, ERI），它将全局传输（例如，朝向和平面速度）与行为意图分离，同时将局部全身适应留给感知策略。这种分解导致了一种传统BFMs所缺乏的特征能力：一个固定的意图可以诱导出不同的、适合地形的行为，包括步高、步幅、姿态、脚步放置和接触的变化，而无需地形特定的示范。在具有挑战性的地形上，P-BFM在运动保真度、根控制精度和扰动恢复方面优于强基线。一个固定的策略进一步支持运动跟踪、目标到达、奖励推断、遥操作和地形感知恢复，并能够零样本迁移到使用板载深度传感器的Unitree G1上。

View on arXiv Download PDF AI Translation

cs.RO / 115 / 2610.15867

REFLEX: A Learned On-Board Reflex for Robust Autonomous Collision Recovery of Soft Quadrotors

REFLEX：一种用于软四旋翼稳健自主碰撞恢复的学习型机载反射机制

Hermeler, Jakob, Girardi, Luca, Maquignaz, Gabriel, Leutenegger, Stefan, Mintchev, Stefano

Abstract

Collision-resilient airframes allow quadrotors to survive impacts, but do not ensure autonomous post-collision recovery. Innovation gating in stock onboard estimators often rejects valid external velocity measurements after high-speed impacts, compromising recovery. We present REFLEX, a collision-recovery flight mode that runs entirely on the flight controller, requiring neither contact sensors nor a companion computer. Its proprioceptive recurrent policy stabilizes the vehicle without position, velocity, heading, or altitude estimates. The translational estimate is reinitialized while REFLEX is active, and a trust gate checks estimator consistency before returning control to the stock autopilot. The policy is trained in simulation using proximal policy optimization, with episodes initialized from post-impact states measured in 47 real collisions. In hardware experiments, REFLEX recovered 19 of 22 wall impacts at 5.03 $\pm$ 0.09 m/s, compared with 1 of 14 for the matched stock autopilot. These included 15 consecutive recoveries in a single flight. A separate exploratory flight demonstrated two additional recoveries at 6.02 and 6.05 m/s within the same flight, exceeding the previously reported velocity for autonomous wall-collision recovery by 64%.

Chinese Translation

抗碰撞机架使四旋翼能够承受冲击，但并不能确保自主的碰撞后恢复。现有的机载估计器中的创新门控机制往往在高速冲击后拒绝有效的外部速度测量，从而妨碍恢复。我们提出了REFLEX，一种完全在飞行控制器上运行的碰撞恢复飞行模式，既不需要接触传感器，也不需要伴随计算机。其自我感知的递归策略在没有位置、速度、航向或高度估计的情况下稳定飞行器。在REFLEX激活期间，平移估计会被重新初始化，并且信任门会在将控制权返回给标准自动驾驶仪之前检查估计器的一致性。该策略通过近端策略优化在仿真中进行训练，训练集由47次真实碰撞后的状态初始化。在硬件实验中，REFLEX在5.03 ± 0.09 m/s的速度下成功恢复了22次墙面冲击中的19次，而匹配的标准自动驾驶仪仅恢复了14次中的1次。这些恢复中包括在一次飞行中连续的15次恢复。一次单独的探索性飞行展示了在同一次飞行中以6.02和6.05 m/s的速度实现的另外两次恢复，超过了之前报告的自主墙面碰撞恢复的速度64%。

View on arXiv Download PDF AI Translation

cs.RO / 116 / 2610.15934

Efficient Hybrid Hard--Soft Signal Temporal Logic MPC for Driving under Multimodal Uncertainty

针对多模态不确定性的高效混合硬-软信号时序逻辑模型预测控制

Manas, Kumar, Paschke, Adrian

Abstract

Motion planning for automated vehicles in interactive traffic has to simultaneously ensure safety, respect traffic rules, and maintain comfortable progress under multi-modal behavior uncertainty. Imposing all these requirements as hard constraints reduces feasibility, whereas treating them all as costs allows safety-critical requirements to be traded against progress. We propose a hybrid hard--soft Signal Temporal Logic (STL) MPC framework that optimizes a single shared ego trajectory against every retained behavior hypothesis: minimum-distance and speed predicates are imposed as explicit hard constraints, while soft STL robustness terms shape clearance, comfort, and interaction quality within the resulting feasible driving corridor. This retains robustness to discrete interaction modes without branch-specific contingent ego policies. We further identify substantial symbolic graph-construction overhead and introduce reachability-certified relevance screening and vectorized symbolic construction to reduce computational cost while preserving the feasible set and objective. Closed-loop Monte Carlo evaluation over six highway and intersection scenarios shows the highest strict success rate among the compared planners with no observed collisions or rule violations, at roughly a quarter of the planning latency of the unoptimized formulation. An ablation study separates the two layers: the hard constraints define the safety envelope, while the soft terms improve clearance and motion quality within it.

Chinese Translation

在交互式交通中，自动驾驶车辆的运动规划必须同时确保安全、遵守交通规则，并在多模态行为不确定性下保持舒适的行驶进度。将所有这些要求作为硬约束施加会降低可行性，而将它们视为成本则允许在安全关键要求与进度之间进行权衡。我们提出了一种混合硬-软信号时序逻辑（Signal Temporal Logic, STL）模型预测控制（MPC）框架，该框架针对每个保留的行为假设优化单一共享的自我轨迹：最小距离和速度谓词作为显式硬约束施加，而软STL鲁棒性项则在生成的可行行驶通道内塑造间隙、舒适度和交互质量。这种方法保持了对离散交互模式的鲁棒性，而无需特定于分支的应急自我策略。我们进一步识别出显著的符号图构建开销，并引入可达性认证的相关性筛选和向量化符号构建，以降低计算成本，同时保持可行集和目标。对六个高速公路和交叉口场景的闭环蒙特卡洛评估显示，在比较的规划者中具有最高的严格成功率，没有观察到碰撞或规则违反，且规划延迟约为未优化公式的四分之一。消融研究将这两个层次分开：硬约束定义了安全边界，而软项则在此边界内改善了间隙和运动质量。

View on arXiv Download PDF AI Translation

cs.RO / 117 / 2610.15936

Jacobian State Augmented Viable Kinematic Safety Filter

雅可比状态增强可行运动安全滤波器

Singh, Siddharth, Vaughan, Joshua

Abstract

Safety Filters (SF) are leveraged to design certifiably safe controllers, which ensures forward invariance on a defined safe set when the filter is feasible. Predictive Safety Filters (PSF) embed these constraints to a closed-loop model-based optimization method to improve constraint satisfaction, and reduce feasibility loss caused by myopic, one-step filtering. Related viability-based methods emphasize approximating a viability kernel which ensures admissible control for all future time, but requires significant computational expense. This work introduces a Jacobian-based kinematic safety filter leveraging an augmented state variable with task-space velocity and filters task-space acceleration to preserve the feasibility of a low-level filter while enforcing joint position and velocity. The proposed method introduces a nested Quadratic Program (QP) formulation, defines a conservative inner approximation of viability envelope and computes a viable planner online. The method is validated in a simplified 2R planar manipulator simulation environment and shows significant improvement compared to a baseline Control Barrier Function Quadratic Program (CBF-QP) based filter. The experiments highlights the improvements in terms of computation time compared to traditional PSF and Reference Governor approach while also showing significant reduction in feasibility violations. Lastly, a simplified demonstration on a simulated 6-DoF manipulator highlights the potential scalability of the method to higher degree-of-freedom setups.

Chinese Translation

安全滤波器（Safety Filters, SF）被用于设计可证明安全的控制器，确保在滤波器可行时在定义的安全集上保持前向不变性。预测安全滤波器（Predictive Safety Filters, PSF）将这些约束嵌入到基于闭环模型的优化方法中，以提高约束满足度，并减少因短视的单步滤波而导致的可行性损失。相关的基于可行性的研究方法强调近似可行核，以确保在未来所有时间内都能进行可接受的控制，但这需要显著的计算开销。本研究提出了一种基于雅可比的运动安全滤波器，利用增强状态变量和任务空间速度，并滤波任务空间加速度，以保持低级滤波器的可行性，同时强制执行关节位置和速度。所提出的方法引入了嵌套的二次规划（Quadratic Program, QP）形式，定义了可行性包络的保守内近似，并在线计算可行规划器。该方法在简化的2R平面操纵器仿真环境中进行了验证，并显示出与基线控制障碍函数二次规划（Control Barrier Function Quadratic Program, CBF-QP）滤波器相比的显著改进。实验结果突出了与传统的PSF和参考调节器方法相比，在计算时间方面的改进，同时也显示出可行性违规的显著减少。最后，在模拟的6自由度操纵器上进行的简化演示突显了该方法在更高自由度设置中的潜在可扩展性。

View on arXiv Download PDF AI Translation

cs.RO / 118 / 2610.15966

A Hardware-Partitioned micro-ROS Mini-Vehicle for Operator-Supervised Neural Driving

一种硬件分区的微型ROS迷你车辆用于操作员监督的神经驾驶

Couto, Gustavo Claudio Karl, Antonelo, Eric Aislan, Zipperer, Gabriel George

Abstract

We present a low-cost miniature Ackermann vehicle for embedded neural-driving research, organized as a hardware-partitioned perceive-decide-act agent: an ESP32 camera node streams frames over Wi-Fi to a host computer, where neural inference and state estimation run, while an ESP32-S3 chassis controller executes safety-critical actuation in a 100 Hz per-wheel proportional-integral-derivative (PID) loop beneath a micro-ROS client. The partition isolates real-time control and safety on the vehicle while learning-based components run off-board, supervised by an operator from a gamepad. We characterize the complete pipeline during real closed-loop neural driving. Wireless sensing, rather than neural inference, dominates latency, with a 59.5 ms median capture-to-actuation delay; the platform absorbed wireless variability by holding the last safe command under a firmware watchdog, operating across twelve autonomous sessions with zero collisions or manual interventions; and encoder-inertial-measurement-unit (IMU) fusion cut odometry loop-closure error by 22%. These results indicate that carefully partitioned microcontroller-class vehicles support reproducible experiments in embedded AI, wireless control, and operator-supervised autonomous driving. The platform software is open source.

Chinese Translation

我们提出了一种低成本的微型阿克曼车辆，用于嵌入式神经驾驶研究，组织为一个硬件分区的感知-决策-执行代理：一个ESP32摄像头节点通过Wi-Fi向主计算机传输帧，在主计算机上进行神经推理和状态估计，而ESP32-S3底盘控制器以每轮100 Hz的比例-积分-微分（PID）循环执行安全关键的驱动。该分区将车辆上的实时控制和安全隔离，同时基于学习的组件在板外运行，由操作员通过游戏手柄进行监督。我们在真实的闭环神经驾驶过程中对整个流程进行了表征。无线传感器的延迟高于神经推理，捕获到驱动的中位延迟为59.5毫秒；该平台通过在固件看门狗下保持最后一个安全命令来吸收无线变异，跨越十二个自主会话实现零碰撞或手动干预；编码器-惯性测量单元（IMU）融合将里程计闭环误差降低了22%。这些结果表明，经过精心分区的微控制器级车辆支持嵌入式人工智能、无线控制和操作员监督的自主驾驶中的可重复实验。该平台软件是开源的。

View on arXiv Download PDF AI Translation

cs.RO / 119 / 2610.16103

PathLift: Depth-Conditioned Metric Lifting for Visual-Navigation Priors

PathLift：基于深度条件的度量提升用于视觉导航先验

Shin, Gwanjun, Suh, Hyeongseok, Jung, Hongmo, Lim, Jongwoo

Abstract

Generalist visual-navigation policies tell a robot which direction to go but not how far. A separate lifting step is needed at deployment to convert their unitless trajectory into a metric path that a controller can track; existing scalar lifters apply uniform scaling that misses per-waypoint corrections, while in-policy approaches that inject monocular depth into the policy inherit the prediction errors of the depth model near obstacles and at close range. We present PathLift, a per-waypoint depth-conditioned adapter that lifts a pre-trained visual-navigation policy by consuming an explicit, sensor-grade depth source (RGB-D, stereo, or projected LiDAR) and fitting per-waypoint magnitude and lateral residuals to a canonicalized prior. PathLift consistently improves closed-loop navigation across diverse visual-navigation policies in simulation and on a real wheeled robot under both stereo and LiDAR depth. By turning the policy's unitless intent into a depth-feasible metric path in a single feed-forward pass, the adapter lifts real-world success without test-time optimization.

Chinese Translation

通用视觉导航策略告诉机器人应该朝哪个方向前进，但并未指明前进的距离。在部署时需要一个单独的提升步骤，将无单位的轨迹转换为控制器可以跟踪的度量路径；现有的标量提升方法应用均匀缩放，未能考虑每个航点的修正，而在策略中注入单目深度的方式则会继承深度模型在障碍物附近和近距离的预测误差。我们提出了PathLift，这是一种基于每个航点的深度条件适配器，通过消耗明确的传感器级深度源（RGB-D、立体视觉或投影激光雷达），并将每个航点的幅度和横向残差拟合到规范化的先验中，从而提升预训练的视觉导航策略。PathLift在模拟环境和真实轮式机器人上，在立体视觉和激光雷达深度下，始终改善闭环导航。通过在单次前馈传递中将策略的无单位意图转化为深度可行的度量路径，该适配器在不进行测试时优化的情况下提升了现实世界的成功率。

View on arXiv Download PDF AI Translation

cs.RO / 120 / 2610.16119

Constrained Differentiable Policy Learning for Safe Quadrotor Navigation

用于安全四旋翼导航的约束可微政策学习

Tan, Yan Rui, Krishnapur, Praveen, Feng, Zeyu, Bao, Han, Ong, Yew-Soon, Shi, Fan

Abstract

Differentiable simulation enables efficient policy optimization through analytic gradients, yet safety is commonly incorporated as a weighted penalty, requiring manual tuning of the trade-off between task performance and constraint satisfaction. We instead formulate obstacle avoidance as an explicit constraint and train quadrotor policies using primal--dual gradient descent--ascent through a differentiable simulator. We introduce two key components to enable effective constrained training. First, we propose a bounded clearance-based constraint function that provides informative gradients around the safety boundary while preventing large violations from destabilizing primal--dual optimization. Second, we introduce log-mean-exp (LME) aggregation to smoothly emphasize critical per-step safety violations across a trajectory and training batch within a single constraint governed by one dual variable. Together, these components enable the desired safety level to be specified directly while balancing constraint satisfaction and task completion. We theoretically prove that the proposed optimization with differentiable simulation converges to feasible policy under mild assumptions. Experiments demonstrate improved safety--performance trade-offs over fixed-penalty and existing policy-learning baselines, as well as generalization to denser out-of-distribution environments. Finally, the learned policy transfers zero-shot to Gazebo/PX4 and a physical quadrotor without additional fine-tuning.

Chinese Translation

可微仿真通过解析梯度实现高效的政策优化，然而安全性通常作为加权惩罚纳入，这需要手动调整任务性能与约束满足之间的权衡。我们将障碍物规避形式化为显式约束，并通过可微仿真使用原始-对偶梯度下降-上升训练四旋翼政策。我们引入两个关键组件以实现有效的约束训练。首先，我们提出了一种基于边界间隙的约束函数，该函数在安全边界附近提供有信息的梯度，同时防止大幅违规导致原始-对偶优化的不稳定。其次，我们引入对数均值指数（log-mean-exp, LME）聚合，以平滑地强调轨迹和训练批次中单个约束下由一个对偶变量支配的关键每步安全违规。结合这些组件，可以直接指定所需的安全水平，同时平衡约束满足与任务完成。我们理论证明了在温和假设下，采用可微仿真的优化方法收敛于可行政策。实验表明，与固定惩罚和现有政策学习基线相比，安全-性能权衡得到了改善，并且在更密集的分布外环境中具有良好的泛化能力。最后，学习到的政策在不进行额外微调的情况下，能够零-shot迁移到Gazebo/PX4和物理四旋翼上。

View on arXiv Download PDF AI Translation

cs.RO / 121 / 2610.16139

Directional Pullback Regularization for Long-Horizon Rollout of Latent World Models in Autonomous Driving

自主驾驶中潜在世界模型的长时间展望的方向性回拉正则化

Mishra, Shatadal, Moradipari, Ahmadreza, Ammar, Nejib

Abstract

Learned world models for autonomous driving operate in unconstrained latent spaces, yet the physical systems they represent are governed by strict non-holonomic constraints: ground vehicles cannot realize instantaneous pure lateral motion. We study long-horizon latent rollout modeling for autonomous driving and ask whether rollout quality can be improved by assigning a motion-aware geometry to latent transitions, rather than by redesigning the predictor architecture. Starting from an unconstrained latent rollout baseline, we introduce Directional Pullback Regularization (DPR), a training time method that defines a kinematics-aware anisotropic Riemannian metric in decoded state space and pulls it back through the decoder Jacobian to the latent manifold. The resulting latent-space metric assigns higher cost to latent directions that decode to lateral vehicle motion, while preserving the original encoder-dynamics-decoder backbone. This yields a clean three-way comparison among an Unconstrained Latent Rollout model (ULR), an Euclidean Pullback Regularization ablation (EPR), and the full directional method (DPR). We evaluate all models on open-loop ego-motion rollout over a common 10-second test horizon, using training rollout lengths of 1, 3, 5, and 10 steps. Across all four training regimes, DPR achieves lower 10-second final displacement error than both ULR and EPR. The strongest relative accuracy gain appears for models trained with 3-step rollout supervision, where DPR reduces the 10-second final displacement error by approximately 44 percent relative to ULR. The strongest overall control-relevant tradeoff is obtained for longer rollout-trained models, especially at 10 steps, where DPR simultaneously improves displacement error, lateral motion consistency, and forward motion preservation relative to both ULR and EPR.

Chinese Translation

自主驾驶的学习型世界模型在无约束的潜在空间中运行，但它们所代表的物理系统受到严格的非完整约束的支配：地面车辆无法实现瞬时的纯横向运动。我们研究了自主驾驶的长时间潜在展望建模，并探讨通过为潜在转移分配运动感知几何体，是否可以改善展望质量，而不是重新设计预测器架构。从无约束潜在展望基线出发，我们引入了方向性回拉正则化（Directional Pullback Regularization, DPR），这是一种训练时方法，在解码状态空间中定义了一种运动学感知的各向异性黎曼度量，并通过解码器雅可比矩阵将其回拉到潜在流形。得到的潜在空间度量对解码为横向车辆运动的潜在方向赋予更高的成本，同时保留原有的编码器-动力学-解码器骨架。这使得无约束潜在展望模型（Unconstrained Latent Rollout, ULR）、欧几里得回拉正则化消融（Euclidean Pullback Regularization, EPR）和完整方向性方法（DPR）之间形成了清晰的三方比较。我们在一个共同的10秒测试时间范围内，对所有模型进行了开放环路自我运动展望的评估，使用的训练展望长度为1、3、5和10步。在所有四种训练方案中，DPR的10秒最终位移误差均低于ULR和EPR。相对准确性的最大增益出现在使用3步展望监督训练的模型中，DPR相对于ULR将10秒最终位移误差降低了约44%。对于经过较长展望训练的模型，特别是在10步时，获得了最强的整体控制相关权衡，此时DPR在相对于ULR和EPR的情况下，同时改善了位移误差、横向运动一致性和前向运动保持。

View on arXiv Download PDF AI Translation

cs.RO / 122 / 2610.16160

FlowVLA: Enabling World Dynamics Understanding through Pluggable 3D Flow Learning

FlowVLA：通过可插拔的3D流学习实现世界动态理解

Go, Junhyeong, Jung, Mingyu, Im, Woobin, Ryu, Jongbin

Abstract

Recent advances in 3D reconstruction enable the extraction of temporally consistent 3D motion from RGB observations; however, current Vision-Language-Action (VLA) models still predominantly rely on 2D image-text priors. Therefore, they do not accurately represent the fundamental dynamics of the 3D world associated with physical actions. This over-reliance on static 2D visual cues limits the capacity of VLA models in achieving robust robot manipulation. To address these limitations, we introduce FlowVLA, a reconstruction-guided, pluggable 3D flow learning framework aimed at empowering VLA models with an inherent understanding of action-relevant dynamics. Instead of requiring explicit 3D sensor inputs or altering the inference-time architecture, our method is designed to utilize temporally aligned 3D flow only during training. This allows the model to learn the dynamic evolution of the physical scene while preserving the computational efficiency of conventional 2D VLA backbones. To bridge scene-level dynamics with robotic manipulation, we propose a modular 3D flow learning framework with dual-level flow supervision. Specifically, point-level flow guides the model to forecast the future displacements of task-critical 3D points, while TCP-level flow aggregates these motions to align with the robot's TCP trajectories. This dual-level flow supervision encourages the model to focus on semantically meaningful dynamic regions, effectively coupling scene-level motion with precise control signals. Extensive evaluations across simulation and real-world experiments demonstrate that FlowVLA consistently improves manipulation performance, zero-shot robustness, and optimizes data efficiency. FlowVLA adds only lightweight auxiliary heads during training and discards them at inference. Therefore, FlowVLA introduces no additional 3D sensor inputs, architectural modifications, or inference-time computation.

Chinese Translation

最近的3D重建进展使得从RGB观测中提取时间一致的3D运动成为可能；然而，当前的视觉-语言-动作（VLA）模型仍然主要依赖于2D图像-文本先验。因此，它们无法准确表示与物理动作相关的3D世界的基本动态。这种对静态2D视觉线索的过度依赖限制了VLA模型在实现稳健机器人操作方面的能力。为了解决这些局限性，我们提出了FlowVLA，这是一种重建引导的、可插拔的3D流学习框架，旨在赋予VLA模型对与动作相关的动态的内在理解。我们的方法不需要显式的3D传感器输入或改变推理时的架构，而是设计为在训练期间仅利用时间对齐的3D流。这使得模型能够学习物理场景的动态演变，同时保持传统2D VLA骨干网络的计算效率。为了将场景级动态与机器人操作相结合，我们提出了一种具有双层流监督的模块化3D流学习框架。具体而言，点级流引导模型预测任务关键3D点的未来位移，而TCP级流则聚合这些运动以与机器人的TCP轨迹对齐。这种双层流监督鼓励模型关注语义上有意义的动态区域，有效地将场景级运动与精确控制信号耦合。通过在模拟和现实世界实验中的广泛评估，FlowVLA始终提高了操作性能、零-shot鲁棒性，并优化了数据效率。FlowVLA在训练期间仅增加轻量级辅助头，并在推理时丢弃它们。因此，FlowVLA没有引入额外的3D传感器输入、架构修改或推理时计算。

View on arXiv Download PDF AI Translation

cs.RO / 123 / 2610.16173

Extending Wire-Driven Parallel Robots to Multi-Floating-Link Systems with an Inter-Body Wire

将线驱动并联机器人扩展至具有体间线的多浮动连杆系统

Inoue, Shintaro, Kawaharazuka, Kento, Suzuki, Temma, Yuzaki, Sota, Yoneda, Keita, Okada, Kei

Abstract

Parallel wire-driven systems are a framework in which multiple wires are connected to an end effector, and the end effector is driven by the tensions of those wires. Because they can transmit large forces, up to the breaking strength of the wires, over lengths on the order of several tens of meters, they can realize powerful actuation in large workspaces. In recent years, studies have also investigated connecting wires to external environments such as trees and columns and forming parallel wire actuation on site. However, conventional parallel wire-driven systems often assume a single end effector, and because of interference between wires and links in a shared workspace, it has been difficult to handle multiple floating links simultaneously. In this study, we realize multi-floating-link parallel wire actuation by introducing an inter-body wire. In real-robot experiments, the two bodies supported and lifted a mannequin and transferred it onto a stretcher as a rescue motion. This shows that parallel wire actuation can be extended from a single end effector to a multi-floating-link system, and indicates its potential for broader tasks such as the manipulation of large objects and rescue support.

Chinese Translation

并联线驱动系统是一种框架，其中多个线缆连接到末端执行器，末端执行器由这些线缆的张力驱动。由于它们能够在数十米的长度上传递大于线缆断裂强度的巨大力量，因此可以在大工作空间中实现强大的驱动。近年来，研究还探讨了将线缆连接到外部环境（如树木和柱子）并在现场形成并联线驱动。然而，传统的并联线驱动系统通常假设只有一个末端执行器，并且由于在共享工作空间中线缆和连杆之间的干扰，同时处理多个浮动连杆变得困难。在本研究中，我们通过引入体间线实现了多浮动连杆的并联线驱动。在真实机器人实验中，两个支撑体抬起了一个假人并将其转移到担架上，作为一种救援动作。这表明并联线驱动可以从单一末端执行器扩展到多浮动连杆系统，并显示出其在处理大物体和救援支持等更广泛任务中的潜力。

View on arXiv Download PDF AI Translation

cs.RO / 124 / 2610.16179

One-Shot Learning for Double-Shot Latte Art: Adapting Cobot Pouring Skills with Dynamic Movement Primitives

一次性学习双重拉花艺术：通过动态运动原语适应协作机器人倾倒技能

Khan, Zain, Moreno, Victor Hernandez, Le, Dinh Tung

Abstract

Latte art is a complex task requiring dexterity and precision, making it a challenging and commercially relevant task. A one-shot learning system is employed in which a single barista demonstration is segmented into five functionally distinct phases, each encoded as an independent Cartesian Dynamic Movement Primitive (CDMP) with its own parameters. Against a single-CDMP baseline, per-phase encoding reduces transition endpoint error by 3.2-4.5 times and art-stroke shape error by 2.7 times, reproducing jug tilt to below 1 degree RMS in the pouring phases. The learnt pour scales from a 280mL to a 400mL cup through a single spatial offset and a scaled recipe, with no re-demonstration. The phase-based CDMP approach generalises beyond latte art to any demonstration task whose phases have distinct spatial or temporal requirements whilst demonstrating the ability to adapt reproductions to suit different task requirements.

Chinese Translation

拉花艺术是一项复杂的任务，要求灵巧和精确，因此成为一项具有挑战性且商业相关的任务。本文采用了一种一次性学习系统，其中单个咖啡师的演示被分割为五个功能上独立的阶段，每个阶段被编码为一个独立的笛卡尔动态运动原语（CDMP），并具有其自身的参数。与单一CDMP基线相比，基于阶段的编码将过渡端点误差降低了3.2-4.5倍，艺术笔画形状误差降低了2.7倍，在倾倒阶段中将壶的倾斜度再现至1度均方根以下。所学习的倾倒技术能够通过单一的空间偏移和缩放的配方，从280毫升的杯子扩展到400毫升的杯子，无需重新演示。基于阶段的CDMP方法超越了拉花艺术，适用于任何具有独特空间或时间要求的演示任务，同时展示了适应不同任务需求的再现能力。

View on arXiv Download PDF AI Translation

cs.RO / 125 / 2610.16199

MSC: Multi-Stage Caster-aware Control Framework with Uncertainty-Aware NMPC and RL-based Execution Layer

MSC：具有不确定性感知的非线性模型预测控制和基于强化学习的执行层的多阶段轮子感知控制框架

Park, Jinho, Lee, Jeongtae, Lim, Jeonghyeok, Choi, Keun Ha, Kim, Kyung-Soo

Abstract

Freely swiveling casters are widely adopted in industrial AMRs/SDVs for their mechanical simplicity and load capacity. However, caster misalignment during low-speed or high-curvature maneuvers induces strong contact nonlinearities, widening the planning-execution gap between commanded $(v, \omega)$ and realized motion. Existing caster-aware strategies are highly sensitive to state-estimation reliability, causing unstable execution in uncertain transition regimes. To address this, we present Multi-Stage Caster-aware Control (MSC), a planning-execution integrated hierarchical framework featuring two core mechanisms. First, an uncertainty-aware NMPC (CU-MPC) dynamically schedules caster-aware costs and input regularization using predictive uncertainty to suppress overly aggressive maneuvers. Second, a reinforcement learning-based execution layer estimates and predicts caster states under partial observability, transforming high-level velocity commands into physically executable wheel speeds adapted to current contact conditions. Validations in Isaac Sim and real robots demonstrate that MSC offers a clear comparative advantage over existing decoupled compensation approaches by significantly improving caster-state estimation accuracy, minimizing misalignment, and enhancing tracking robustness in caster-sensitive maneuvers.

Chinese Translation

自由旋转的轮子在工业自主移动机器人（AMRs）和自主驾驶车辆（SDVs）中被广泛采用，因其机械结构简单且承载能力强。然而，在低速或高曲率机动过程中，轮子错位会引发强烈的接触非线性，扩大指令 $(v, heta)$ 与实际运动之间的规划-执行差距。现有的轮子感知策略对状态估计的可靠性高度敏感，导致在不确定的过渡状态下执行不稳定。为了解决这一问题，我们提出了多阶段轮子感知控制（MSC），这是一个规划-执行集成的分层框架，具有两个核心机制。首先，不确定性感知的非线性模型预测控制（CU-MPC）动态调度轮子感知成本和输入正则化，利用预测不确定性抑制过于激进的机动。其次，基于强化学习的执行层在部分可观测性下估计和预测轮子状态，将高层速度指令转化为适应当前接触条件的物理可执行轮速。通过在Isaac Sim和真实机器人中的验证，MSC显示出相较于现有的解耦补偿方法具有明显的比较优势，显著提高了轮子状态估计的准确性，最小化了错位，并增强了在轮子敏感机动中的跟踪鲁棒性。

View on arXiv Download PDF AI Translation

cs.RO / 126 / 2610.16238

Embodied Skill Bench: Benchmarking How Well Agents with Embodied Skills Perform on Physical Tasks

具身技能基准：评估具身技能的智能体在物理任务中的表现

Han, Zhi, Guo, Zihan, Zeng, Chenxi, Cui, Zicai, Xin, Yue, Zhou, Ming, Zhou, Dongzhan, Zhang, Weinan, Li, Yang

Abstract

LLM agents increasingly control robots using VLA policies, perception models, and procedural knowledge. These modules are difficult to reuse because their interfaces, embodiment assumptions, and conditions of use differ, and their benefits on physical tasks remain unclear. We package them as embodied skills that extend agent skills with explicit embodiment assumptions and conditions of use, building 250 action, perception, and experience skills from 1,052 resources. On this basis we introduce EmbodiedSkillBench, which runs an agent in a closed observation-action loop and records success, actions, tokens, and wall-clock time. It supports three modes: native control, skill-only execution, and Agent-Directed Reuse (ADR), a mode we propose in which the agent keeps its native controls and decides for each skill whether to invoke, modify, or ignore it. Across four agents on 100 perturbed LIBERO-Pro tasks, we find that skills help an agent only as far as they exceed what it can already do. Skill-only execution lifts GPT-5.6-Sol from 11% to 63% and Kimi-K3 and Claude-Opus-5 by about 20 points, while cutting their token usage by 58% to 85%. Yet all four agents then end up between 61% and 64%, and GPT-6-Astra falls from 92% under native control to 61%. This suggests that as agents grow stronger, the role of embodied skills shifts from supplying missing capability to reducing cost: GPT-6-Astra already solves most tasks through native control, so restricting it to skills removes more than it adds. Under ADR, which lets it keep native control and call skills selectively, GPT-6-Astra recovers to 85%, within 7 points of native control, while using 14.1% fewer tokens and 11.9% less time. With Jev as a skill choosing motion directions and gripper actions, it reaches 81% with 50% fewer tokens and 57% less time. The code is available at https://github.com/HanZhi306/EmbodiedSkill-Bench.

Chinese Translation

大型语言模型（LLM）智能体越来越多地使用变换学习策略（VLA）、感知模型和程序知识来控制机器人。这些模块由于其接口、具身假设和使用条件的不同，难以重用，其在物理任务上的效果仍不明确。我们将这些模块打包为具身技能，扩展智能体技能，明确具身假设和使用条件，从1,052个资源中构建了250个动作、感知和经验技能。在此基础上，我们引入了EmbodiedSkillBench，它在一个封闭的观察-动作循环中运行智能体，并记录成功率、动作、令牌和实际时间。它支持三种模式：原生控制、仅技能执行和智能体主导重用（Agent-Directed Reuse, ADR），我们提出的这种模式允许智能体保留其原生控制，并决定对每个技能是调用、修改还是忽略。在100个扰动的LIBERO-Pro任务中，我们发现技能仅在超出智能体已有能力的情况下才对其有帮助。仅技能执行将GPT-5.6-Sol的成功率从11%提升至63%，Kimi-K3和Claude-Opus-5的成功率提高约20个百分点，同时令牌使用量减少58%至85%。然而，所有四个智能体的最终表现均在61%至64%之间，而GPT-6-Astra在原生控制下的成功率从92%降至61%。这表明，随着智能体能力的增强，具身技能的角色从提供缺失能力转变为降低成本：GPT-6-Astra已经通过原生控制解决了大多数任务，因此将其限制于技能反而减少了其能力。在ADR模式下，GPT-6-Astra保留了原生控制并选择性调用技能，成功率恢复至85%，与原生控制相差7个百分点，同时令牌使用量减少14.1%，时间减少11.9%。使用Jev作为选择运动方向和抓取动作的技能时，其成功率达到81%，令牌使用量减少50%，时间减少57%。代码可在https://github.com/HanZhi306/EmbodiedSkill-Bench获取。

View on arXiv Download PDF AI Translation

cs.RO / 127 / 2610.16242

Same Goal, Different Words: Spatial Instruction Sensitivity in World-Action Models

相同目标，不同表述：世界-动作模型中的空间指令敏感性

Abbas, Ali-Adeeb, Seraj, Esmaeil, Toghi, Behrad, Padir, Taskin

Abstract

World-action models should preserve the intended manipulation across equivalent descriptions of a spatial goal. We evaluate this consistency in 864 rollouts from three models across four RoboLab scenes, holding initial states and task goals fixed. Describing relations through the reference object rather than the object-to-move lowers the rate of reaching a stable placement by 12.2 percentage points on average. The effect of instruction rephrasing varies across spatial relations, so aggregate scores can hide sensitivity to wording. Standalone tests of the policies' language understanding identify relation reversals and show that adding images can disrupt otherwise correct text-only interpretations. Rollout analysis also distinguishes spatial placement from instruction following: moving the wrong object can satisfy the requested relation, and rates of stable placement fall from 43.2% at any time to 21.8% at episode end. These findings motivate a more thorough evaluation of world-action models' steerability, language understanding, and grounding. Evaluations should test consistency across equivalent instructions, verify that the intended object moved, and check whether the requested arrangement persists to the end of the rollout.

Chinese Translation

世界-动作模型应在空间目标的等效描述中保持预期的操作一致性。我们在固定初始状态和任务目标的情况下，评估了来自三个模型在四个RoboLab场景中的864次回合的这种一致性。通过参考对象而非待移动对象来描述关系，平均降低了达到稳定放置的比例12.2个百分点。指令重述的效果在不同空间关系中有所不同，因此汇总得分可能掩盖对措辞的敏感性。对策略语言理解的独立测试识别了关系反转，并表明添加图像可能会干扰原本正确的仅文本解释。回合分析还区分了空间放置与指令遵循：移动错误的对象可以满足请求的关系，而稳定放置的比例从任何时刻的43.2%下降到剧集结束时的21.8%。这些发现促使我们对世界-动作模型的可引导性、语言理解和基础进行更全面的评估。评估应测试等效指令之间的一致性，验证移动的是预期对象，并检查请求的排列是否在回合结束时仍然存在。

View on arXiv Download PDF AI Translation

cs.RO / 128 / 2610.16308

PACE: Temporal-Role Tactile Conditioning for Chunk-Based Insertion Policies

PACE：基于块的插入策略的时间角色触觉条件化

Chen, Zhiyuan, Xu, Shixiong, Chen, Shuai, You, Yangwei

Abstract

Contact-rich insertion requires robots to combine visual alignment with tactile adjustment. However, existing visuotactile chunk-based policies collapse tactile feedback into a single generation-time fusion pathway, leaving persistent contact states, abrupt contact events, and future contact consequences implicitly modeled in a single representation. We present PACE (Phase-Aware Contact-Event Conditioning), a temporal-role tactile conditioning framework for chunk-based insertion policies. PACE organizes a shared tactile stream into three temporal roles: persistent contact states condition action-chunk generation, abrupt contact events correct the generated chunk, and contact-consequence prediction shapes action representations during training. This design preserves chunk-level action continuity while assigning distinct pathways to different temporal roles of contact feedback. Across five real-robot insertion tasks, PACE achieves an average success rate of 90.7%, outperforming vision-only and generation-time tactile-fusion baselines by over 44 percentage points; on the ManipulationNet benchmark, PACE maintains 88.7% success across six geometry-clearance configurations, including an industrial-grade 0.1-mm clearance. Ablations confirm the complementary contributions of the three temporal roles, mechanistic analyses show that PACE separates normal insertion progress from contact events requiring immediate correction, and cross-signal and cross-architecture experiments indicate that this temporal-role principle extends across different physical feedback signals and policy backbones. Videos and materials are available at https://pace-insertion.github.io/.

Chinese Translation

接触丰富的插入要求机器人将视觉对齐与触觉调整相结合。然而，现有的视觉-触觉基于块的策略将触觉反馈压缩为单一的生成时间融合路径，导致持续接触状态、突发接触事件和未来接触后果在单一表示中隐式建模。我们提出了PACE（相位感知接触事件条件化），这是一个用于基于块的插入策略的时间角色触觉条件化框架。PACE将共享的触觉流组织为三种时间角色：持续接触状态条件化动作块的生成，突发接触事件修正生成的块，接触后果预测在训练过程中塑造动作表示。该设计在保持块级动作连续性的同时，为接触反馈的不同时间角色分配了独特的路径。在五个真实机器人插入任务中，PACE实现了90.7%的平均成功率，超过了仅使用视觉和生成时间触觉融合基线44个百分点；在ManipulationNet基准测试中，PACE在六种几何清除配置下保持了88.7%的成功率，包括工业级0.1毫米的清除。消融实验确认了三种时间角色的互补贡献，机制分析显示PACE将正常插入进展与需要立即修正的接触事件分开，交叉信号和交叉架构实验表明这一时间角色原则扩展到不同的物理反馈信号和策略骨干。视频和材料可在https://pace-insertion.github.io/获取。

View on arXiv Download PDF AI Translation

cs.RO / 129 / 2610.16309

ReAP: From Visual Failure Recognition to Action-Conditioned Risk Prediction for Runtime Failure Monitoring

ReAP：从视觉故障识别到基于动作的风险预测，用于运行时故障监测

Liu, Jingyuan, Qing, Yunpeng, Wei, Changxu, Hou, Zhi

Abstract

Despite rapid progress, generalist robot policies remain brittle during execution. Reliable deployment therefore requires runtime monitoring that provides timely warnings of failure. Existing monitors commonly estimate risk from current observations or policy signals without explicitly predicting the future consequences of proposed actions. To anticipate such risks before action execution, we introduce ReAP, a future-aware runtime monitoring framework that combines visual failure recognition with action-conditioned failure prediction. Specifically, ReAP first trains a visual detector on successful and failed rollouts to learn a failure-relevant representation space and estimate current risk. An action-conditioned predictor then forecasts the resulting future representation in this space to estimate future failure risk. At runtime, the detector and predictor provide complementary signals for current-state failure recognition and future-aware risk assessment. Experiments across diverse robot policies in simulation and the real world demonstrate that ReAP effectively balances failure-detection accuracy and warning timeliness.

Chinese Translation

尽管取得了快速进展，但通用机器人策略在执行过程中仍然脆弱。因此，可靠的部署需要运行时监测，以提供及时的故障警告。现有的监测器通常根据当前观察或策略信号估计风险，而没有明确预测拟议动作的未来后果。为了在动作执行之前预测这些风险，我们引入了ReAP，一个未来感知的运行时监测框架，它将视觉故障识别与基于动作的故障预测相结合。具体而言，ReAP首先在成功和失败的回放上训练一个视觉检测器，以学习与故障相关的表示空间并估计当前风险。然后，基于动作的预测器在该空间中预测未来的表示，以估计未来的故障风险。在运行时，检测器和预测器为当前状态的故障识别和未来感知的风险评估提供互补信号。在模拟和现实世界中对多种机器人策略的实验表明，ReAP有效地平衡了故障检测的准确性和警告的及时性。

View on arXiv Download PDF AI Translation

cs.RO / 130 / 2610.16313

Listen Before Act: Audio-Conditioned Robot Manipulation with ListenVLA

先听后动：基于音频条件的机器人操作与 ListenVLA

Fan, Jiaqi, Xu, Rui, Xia, Jiakun, Fan, Xuyu, Ma, Shuailei, Wang, Yuefeng, Wang, Qing, Zhang, Jiarui, Fu, Yili, Li, Xu

Abstract

Vision-language-action (VLA) models have advanced robotic manipulation by connecting visual perception and language instructions to action generation. However, many manipulation tasks depend on auditory cues that cannot be reliably inferred from visual observations and may need to be elicited through physical interaction. To address this gap, we present ListenVLA, an audio-conditioned VLA policy built on GR00T N1.7 for sound-guided manipulation. ListenVLA combines semantic audio embeddings with waveform-derived acoustic cues over a causal audio history and conditions continuous action generation through audio tokens and feature modulation. Moreover, we introduce an audio-action binding objective that uses context-matched demonstrations to link auditory differences to demonstrated action differences. The objective supervises short-horizon differences between flow-velocity predictions under different audio conditions, with shared pure-noise action inputs and fixed non-audio observations, alongside full-chunk action supervision. Furthermore, we develop LIBERO-Audio, an eight-task benchmark covering sound-based target selection, responses to auditory events, and physical interactions that reveal acoustic evidence. Experiments on LIBERO-Audio and three real-world manipulation tasks show excellent performance. Ablation studies support the contributions of binding supervision and audio modulation, while audio swap tests show that sound influences action predictions. Our code and videos are available at https://anonymous.4open.science/r/listenvla-F18D

Chinese Translation

视觉-语言-动作（VLA）模型通过将视觉感知与语言指令连接到动作生成，推动了机器人操作的发展。然而，许多操作任务依赖于听觉线索，这些线索无法通过视觉观察可靠推断，并可能需要通过物理交互来引发。为了解决这一问题，我们提出了 ListenVLA，这是一种基于音频条件的 VLA 策略，建立在 GR00T N1.7 之上，用于声音引导的操作。ListenVLA 将语义音频嵌入与基于波形的声学线索结合在一起，通过因果音频历史条件化连续动作生成，使用音频标记和特征调制。此外，我们引入了一种音频-动作绑定目标，该目标利用上下文匹配的示范将听觉差异与示范的动作差异联系起来。该目标监督在不同音频条件下流速预测的短期差异，使用共享的纯噪声动作输入和固定的非音频观察，同时进行完整的动作监督。此外，我们开发了 LIBERO-Audio，这是一个涵盖基于声音的目标选择、对听觉事件的响应以及揭示声学证据的物理交互的八任务基准。在 LIBERO-Audio 和三个真实世界的操作任务上的实验显示了出色的性能。消融研究支持绑定监督和音频调制的贡献，而音频交换测试表明声音影响动作预测。我们的代码和视频可在 https://anonymous.4open.science/r/listenvla-F18D 获取。

View on arXiv Download PDF AI Translation

cs.RO / 131 / 2610.16316

Constraint-Decomposed Impedance Control from Generative Policy for Articulated Object Manipulation

基于生成策略的约束分解阻抗控制用于关节物体操作

Park, Kyungseo, Lim, Byeongdo, Lim, Jungbin, Choi, Minseok, Park, Frank C.

Abstract

Articulated object manipulation tasks such as opening doors, tightening bolts, and pulling drawers impose kinematic constraints on a robot end-effector's motion that neural policies may fail to satisfy. Unlike hybrid force-position control that strictly separates force and position control into orthogonal subspaces, impedance control modulates a mechanical impedance system in the form of a virtual inertia, damping, and stiffness. However, selecting task-appropriate impedance gains remains challenging for robot policies. In this paper, we address this problem for the case of manipulating articulated objects. Our constraint-decomposed impedance control method uses the sampling diversity of a generative policy to estimate the feasible motion directions in twist space. The key idea is to infer the feasible and constrained subspaces of twists from an ensemble of trajectories generated by the policy from the same observation. Principal component analysis (PCA) of the end-effector twists determines the feasible twist subspace, which is then used to form 6x6 impedance gain matrices. For many common articulated objects, the feasible twist subspace is independent of the object configuration, so the subspace and resulting gains are estimated once and held fixed during manipulation, minimizing computational overhead. For less common cases where this independence does not hold, we instead estimate a configuration-independent twist subspace of higher dimension that contains the local feasible subspaces, allowing the same fixed-gain strategy to be used. Experiments across simulated and real-world single- and multi-DoF objects show that the proposed method consistently reduces constraint-violating contact loads relative to isotropic impedance control while maintaining task execution.

Chinese Translation

关节物体操作任务，如开门、拧紧螺栓和拉出抽屉，对机器人末端执行器的运动施加了运动学约束，神经策略可能无法满足这些约束。与严格将力和位置控制分离到正交子空间的混合力-位置控制不同，阻抗控制以虚拟惯性、阻尼和刚度的形式调节机械阻抗系统。然而，为机器人策略选择适当的阻抗增益仍然具有挑战性。本文针对关节物体操作的情况解决了这一问题。我们的约束分解阻抗控制方法利用生成策略的采样多样性来估计扭转空间中的可行运动方向。关键思想是通过从相同观测生成的轨迹集推断扭转的可行和受限子空间。末端执行器扭转的主成分分析（PCA）确定可行的扭转子空间，然后用于形成6x6的阻抗增益矩阵。对于许多常见的关节物体，可行的扭转子空间与物体配置无关，因此该子空间及其结果增益在操作过程中估计一次并保持不变，从而最小化计算开销。对于那些不满足独立性的较少见情况，我们则估计一个包含局部可行子空间的更高维度的配置独立扭转子空间，从而允许使用相同的固定增益策略。针对模拟和真实世界的单自由度和多自由度物体的实验表明，所提出的方法在保持任务执行的同时，始终减少了相对于各向同性阻抗控制的约束违反接触载荷。

View on arXiv Download PDF AI Translation

cs.RO / 132 / 2610.16326

Imitation Learning of Avoidance Modes for Efficient Multimodal Model Predictive Control

高效多模态模型预测控制的规避模式模仿学习

Lee, Changyu, Lee, Byung-Kwan, Yoon, Kanghoon

Abstract

Multimodal model predictive control (MM-MPC) avoids local minima among moving obstacles by solving one optimal control problem (OCP) per combination of avoidance sides, so its computational cost grows exponentially with the number of decision-critical obstacles. We propose SideNet-MPC, which replaces this enumeration with a mode decision learned by imitating MM-MPC as an expert. A graph neural network is trained to predict the expert's minimum-cost mode, one avoidance side per obstacle. Its edges encode, for each obstacle pair, the gap predicted for the moment the vehicle first reaches either obstacle, so a neighbor that blocks a preferred passage can reverse the proposed side. The expert's selection rule compares its minimum-cost mode with the previously executed mode; SideNet-MPC keeps this rule and substitutes the network proposal for the minimum-cost mode. Hence at most two OCPs are solved per control period, and every applied input still comes from an OCP with control barrier function constraints rather than from the network. In $100$ seed-matched closed-loop Monte Carlo runs of mixed traffic, SideNet-MPC reaches $93\%$ of the expert's success rate and needs \SI{18.6}{ms} per control period on a single thread, 48\% less than the expert's \SI{35.5}{ms} on sixteen parallel threads. In a controlled gap sweep, its proposed side reverses in the same gap interval as the expert's, whereas proposers without pairwise edges, trained on the same data, never reverse and all deadlock at the blocked gates of a deterministic scene.

Chinese Translation

多模态模型预测控制（MM-MPC）通过为每种规避侧的组合解决一个最优控制问题（OCP），以避免在移动障碍物之间的局部最小值，因此其计算成本随着决策关键障碍物数量的增加而呈指数增长。我们提出了SideNet-MPC，它用模仿MM-MPC作为专家学习的模式决策替代了这种枚举。一个图神经网络被训练来预测专家的最低成本模式，为每个障碍物提供一个规避侧。其边缘编码了每对障碍物在车辆首次到达任一障碍物时预测的间隙，因此一个阻塞优选通道的邻居可以逆转提议的侧面。专家的选择规则将其最低成本模式与先前执行的模式进行比较；SideNet-MPC保留了这一规则，并用网络提议替代最低成本模式。因此，每个控制周期最多解决两个OCP，并且每个应用的输入仍然来自具有控制障碍函数约束的OCP，而不是来自网络。在$100$个种子匹配的闭环蒙特卡洛混合交通运行中，SideNet-MPC达到了专家成功率的$93 ext{%}$，在单线程上每个控制周期需要 ext{18.6 ms}，比专家在十六个并行线程上的 ext{35.5 ms}减少了48 ext{%}。在一个受控间隙扫描中，其提议的侧面在与专家相同的间隙区间内逆转，而没有成对边缘的提议者在相同数据上训练时从不逆转，并且在确定性场景的阻塞门口全部死锁。

View on arXiv Download PDF AI Translation

cs.RO / 133 / 2610.16334

Toward Study on the Effects of Robot's Humorous Apology on Human's Trust Recovery

关于机器人幽默道歉对人类信任恢复影响的研究

Kim, Gayoung, Cha, Go-Eum, Jo, Wonse

Abstract

Robot errors can disrupt human-robot collaboration by reducing user trust and willingness to rely on the robot. Robot apologies can support trust recovery after such errors, but their effects may vary with apology strategy and error severity. This study therefore presents a user experiment based on a 2x2 between-subjects factorial design to examine the effect of humorous robot apologies on user trust recovery and the moderating role of error severity. The design crosses robot error severity (low vs. high) with apology strategy (generic vs. humorous). An online experiment will be conducted using the open-source Warehouse Robot Interaction Simulator. Participants will complete a collaborative task, experience a robot error followed by an apology, and report trust, trust recovery, interaction experience and reuse intention. We hypothesize that humorous apologies will produce greater trust recovery than generic apologies after low-severity errors, but that the advantage of humor will weaken or reverse after high-severity errors.

Chinese Translation

机器人错误可能通过降低用户信任和依赖机器人的意愿来破坏人机协作。机器人道歉可以在此类错误发生后支持信任恢复，但其效果可能因道歉策略和错误严重性而异。因此，本研究基于2x2被试间设计进行了一项用户实验，以考察幽默机器人道歉对用户信任恢复的影响及错误严重性的调节作用。该设计将机器人错误严重性（低 vs. 高）与道歉策略（通用 vs. 幽默）相结合。将使用开源的仓库机器人交互模拟器进行在线实验。参与者将完成一项协作任务，经历机器人错误后进行道歉，并报告信任、信任恢复、互动体验和重用意图。我们假设在低严重性错误后，幽默道歉将比通用道歉产生更大的信任恢复，但在高严重性错误后，幽默的优势将减弱或逆转。

View on arXiv Download PDF AI Translation

cs.RO / 134 / 2610.16355

PINT: A Unified Model with Shared Predictive Intent for Robot Manipulation

PINT：一种具有共享预测意图的统一机器人操作模型

Guo, Xinyu, Jing, Chao, Yi, Li, Chen, Xingyu, Wu, Zhengxing

Abstract

Generalist robot manipulation can benefit from pretrained vision-language models (VLMs) for semantic reasoning and video generation models (VGMs) for world modeling. However, existing unified models that directly fuse them through shared attention at multiple layers can make it harder to draw on their specialized capabilities. We introduce PINT, a unified model that coordinates semantic reasoning, world modeling, and action generation through a shared chunk-level Predictive Intent while preserving expert-specific pathways. PINT inserts learnable Intent Tokens into the VLM sequence to form this representation, conditioning both the VGM and an Action Expert (AE). It replaces only the VGM's prompt conditioning and retains the VGM at inference, where the AE reads its representations through one-way attention. Rather than prescribing what Predictive Intent should encode, PINT learns it jointly from the native prediction objectives of the VLM, VGM, and AE, which provide complementary views of the same upcoming task execution. Training uses two stages: Stage 1 freezes the VGM so that future video prediction shapes Predictive Intent with task-relevant future dynamics, while Stage 2 fixes the learned Intent and adapts the VGM and AE. Without embodied pretraining, PINT achieves strong in-distribution (ID) performance and substantially improves out-of-distribution (OOD) generalization, reaching 15.5% on RoboTwin 2.0-Hard compared with 1.8% for Fast-WAM and 3.2% for StarVLA, and 79.5% on LIBERO-Plus compared with 51.5% and 75.0%, respectively. Across five real-world bimanual tasks, PINT reaches 68% ID and 24% OOD success, compared with 35%/9% for pretrained Motus and 67%/35% for pretrained $\pi_{0.5}$.

Chinese Translation

通用机器人操作可以从预训练的视觉-语言模型（VLMs）中受益，以进行语义推理，并从视频生成模型（VGMs）中受益，以进行世界建模。然而，现有的通过多个层次的共享注意力直接融合这两者的统一模型，可能会使得利用它们的专业能力变得更加困难。我们提出了PINT，这是一种通过共享块级预测意图协调语义推理、世界建模和动作生成的统一模型，同时保留专家特定的路径。PINT将可学习的意图标记（Intent Tokens）插入到VLM序列中以形成这种表示，条件化VGM和动作专家（AE）。它仅替换VGM的提示条件，并在推理时保留VGM，其中AE通过单向注意力读取其表示。PINT并不规定预测意图应编码什么，而是从VLM、VGM和AE的原生预测目标中共同学习，这些目标提供了同一即将执行任务的互补视角。训练分为两个阶段：第一阶段冻结VGM，以便未来的视频预测塑造与任务相关的未来动态的预测意图，而第二阶段固定学习到的意图并适应VGM和AE。在没有具身预训练的情况下，PINT在分布内（ID）表现出色，并显著提高了分布外（OOD）泛化能力，在RoboTwin 2.0-Hard上达到15.5%，而Fast-WAM为1.8%，StarVLA为3.2%；在LIBERO-Plus上达到79.5%，而分别为51.5%和75.0%。在五个真实世界的双手任务中，PINT的ID成功率达到68%，OOD成功率达到24%，而预训练的Motus为35%/9%，预训练的$ ext{π}_{0.5}$为67%/35%。

View on arXiv Download PDF AI Translation

cs.RO / 135 / 2610.16369

RoboBugBench: Benchmarking Automated Repair of Real Robot Software Under an Audited Behavioral Oracle

RoboBugBench：在经过审计的行为oracle下对真实机器人软件的自动修复进行基准测试

Tuzikov, Alexey

Abstract

Language agents increasingly modify robot software, but whether they can repair it is unmeasured, because the domain supplies no oracle to inherit: a robot bug manifests as behavior, and robot repositories ship almost no tests that capture it. We present RoboBugBench, a benchmark of 33 real, execution-validated defects in a pinned ROS 2 navigation stack, in which a candidate patch is graded by compiling it and driving the robot through a physics simulation rather than by running a unit test. Three frontier models, each given 1 attempt per instance, repair 24.2%-36.4%, with no pairwise difference significant; difficulty tracks the oracle rather than the domain, at 66.7% on instances graded by process survival against 22.2% on those graded by what the robot did (instance-level p = 0.0027). Because the oracle is constructed rather than inherited we audit it adversarially, and report the decomposition that audit yields: counterfeit patches which suppress a failure without repairing the defect are accepted by the shipped rule 11/11, while a distributional rule over the same data rejects 4 at no cost in false rejects. The residual is an observability bound (what the observation channel discards, no predicate over it can recover), so discrimination must be bought with instrumentation rather than stricter criteria, and our repair rates are bounds: inflated by what the oracle cannot distinguish, deflated by unanimity's rejection of correct fixes under run-to-run noise. We release the corpus, criterion batteries, counterfeit suite and cached telemetry.

Chinese Translation

语言代理越来越多地修改机器人软件，但它们是否能够修复软件尚未被测量，因为该领域没有可继承的oracle：机器人缺陷表现为行为，而机器人代码库几乎没有捕捉到这些行为的测试。我们提出了RoboBugBench，这是一个基于33个真实的、经过执行验证的缺陷的基准，缺陷存在于一个固定的ROS 2导航栈中，其中候选补丁的评分是通过编译补丁并在物理仿真中驱动机器人来进行，而不是通过运行单元测试。三个前沿模型，每个实例给出1次尝试，修复率为24.2%-36.4%，且没有成对差异显著；难度与oracle相关，而非与领域相关，在通过过程存活进行评分的实例中为66.7%，而在通过机器人行为进行评分的实例中为22.2%（实例级p = 0.0027）。由于oracle是构建而非继承的，我们以对抗性方式对其进行审计，并报告审计所产生的分解：伪造补丁在不修复缺陷的情况下抑制故障，按照已发布的规则被接受11/11，而对同一数据的分布规则在没有虚假拒绝成本的情况下拒绝了4个。剩余部分是可观测性界限（观察通道丢弃的内容，任何谓词都无法恢复），因此区分必须通过仪器化而非更严格的标准来获得，我们的修复率是界限：由于oracle无法区分而被夸大，因一致性拒绝在运行到运行的噪声下正确修复而被缩小。我们发布了该语料库、标准电池、伪造套件和缓存的遥测数据。

View on arXiv Download PDF AI Translation

cs.RO / 136 / 2610.16398

LABIT: Long-Horizon Robotic Assembly Benchmark for Industrial Tasks

LABIT：面向工业任务的长时间机器人装配基准

Schempp, Constantin, Friedrich, Christian, Kofler, Luca, Suhm, Laura

Abstract

Robot learning has undergone significant advancements in the domain of compliant control and robotic assembly. The development of novel methodologies has transcended the capacity to address discrete tasks, such as grasping or inserting, and has evolved to encompass the comprehensive construction of entire assemblies. Conventional benchmarks, which are predominantly focused on single tasks, are no longer adequate to address the complexity and interconnected nature of modern manufacturing processes. Consequently, the necessity for benchmarks to encompass a broader spectrum of assembly and subassembly steps is becoming apparent. In light of these developments, we propose LABIT, an industrial robotic assembly benchmark for long-horizon manipulation. This benchmark encompasses DIN-norm assembly tasks, including screwing, placing, inserting, and snapping, under both high and low tolerance conditions. The successive nature of these assembly steps implies that prior errors can influence subsequent manipulation. Consequently, this benchmark is particularly relevant for assessing the robustness of intelligent robotic systems. The proposed benchmarking protocols and novel performance metric, based on a taxonomy of assembly complexity in conjunction with the comprehensive list of parts and their corresponding CAD models, serve as essential tools for research and development in the field. To provide a baseline, we evaluate the benchmark using traditional force-position control and report reliability and score metrics per primitive, subtask, assembly category, and subassembly. Accompanying data is available at https://irp-hka.github.io/robotic-assembly-benchmark/.

Chinese Translation

机器人学习在顺应控制和机器人装配领域取得了显著进展。新方法的开发超越了处理离散任务（如抓取或插入）的能力，发展到涵盖整个装配的全面构建。传统基准主要集中于单一任务，已无法满足现代制造过程的复杂性和相互关联性。因此，基准需要涵盖更广泛的装配和子装配步骤的必要性日益明显。鉴于这些发展，我们提出了LABIT，一个用于长时间操控的工业机器人装配基准。该基准涵盖DIN标准装配任务，包括在高低公差条件下的拧紧、放置、插入和卡合。这些装配步骤的连续性意味着先前的错误可能影响后续的操控。因此，该基准特别适用于评估智能机器人系统的鲁棒性。我们提出的基准协议和基于装配复杂性分类法的新性能指标，以及相应零件及其CAD模型的综合列表，成为该领域研究与开发的重要工具。为了提供基准线，我们使用传统的力-位置控制对基准进行评估，并报告每个原始任务、子任务、装配类别和子装配的可靠性和评分指标。相关数据可在 https://irp-hka.github.io/robotic-assembly-benchmark/ 获取。

View on arXiv Download PDF AI Translation

cs.RO / 137 / 2610.16411

Estimating the Causal Impact of Social Navigation Signals in Corridor Scenarios

估计走廊场景中社会导航信号的因果影响

Baldo, Francesco, Castri, Luca, Beraldo, Gloria, Bellotto, Nicola

Abstract

Deploying robots in human-shared environments requires deciding not only how to navigate, but also when dedicated social signals are worth emitting. Classic observation-based inference can be misleading in such settings: confounding factors may bias the estimated relationship between a signal and the task outcome, leading to incorrect decisions. Causal inference provides a framework for disentangling these relationships by modelling the mechanisms linking observations, actions, and outcomes, and by estimating interventional effects after accounting for confounding paths. We study a human-robot interaction scenario in a corridor and use a causal model to estimate whether a social navigation signal causally improves transit success. We represent the interaction as a causal DAG involving robot action, pedestrian occupancy, available space, task outcome, and a controlled confounder (static obstacles). We compare the interventional estimate with the observational estimate obtained by conditioning only on the robot action, which does not explicitly model interventions. Our results show that accounting for confounding variables can reverse the conclusions drawn from naive observational analysis: the observational estimate wrongly suggests that the social signal is detrimental, whereas the causal estimate shows that it is beneficial. This discrepancy highlights how failing to account for confounding can lead to incorrect conclusions and demonstrates the importance of causal models for context-aware signalling decisions in social robot navigation.

Chinese Translation

在与人类共享的环境中部署机器人不仅需要决定如何导航，还需要判断何时发出专门的社会信号。在这种环境中，基于经典观察的推断可能会产生误导：混杂因素可能会偏倚信号与任务结果之间的估计关系，从而导致错误的决策。因果推断提供了一种框架，通过建模连接观察、行动和结果的机制，并在考虑混杂路径后估计干预效应，从而解开这些关系。我们研究了走廊中的人机交互场景，并使用因果模型来估计社会导航信号是否在因果上改善了通行成功率。我们将交互表示为一个因果有向无环图（DAG），涉及机器人行动、行人占用、可用空间、任务结果和一个受控混杂因素（静态障碍物）。我们将干预估计与仅基于机器人行动进行条件化的观察估计进行比较，后者并未明确建模干预。我们的结果表明，考虑混杂变量可能会推翻从天真的观察分析中得出的结论：观察估计错误地表明社会信号是有害的，而因果估计则显示其是有益的。这一差异突显了未能考虑混杂因素可能导致错误结论的重要性，并展示了因果模型在社会机器人导航中的上下文感知信号决策中的重要性。

View on arXiv Download PDF AI Translation

cs.RO / 138 / 2610.16457

SHOES: Sim-to-real Hardware-adaptation via On-policy Expansion of Skills for Legged Robots

SHOES：通过对腿部机器人技能的在线扩展实现仿真到现实的硬件适应

Tsuruhara, Kota, Chen, Tsung-Yuan, Kawaharazuka, Kento

Abstract

Humans expand their locomotion capability by wearing appropriate footwear, such as hiking boots for mountains or skates for ice. Footwear changes the ground-contact interface and enables traversal of otherwise difficult terrain. In contrast, locomotion controllers are typically trained for a fixed morphology, and their performance can degrade when shoes alter foot geometry, mass, and friction. We propose SHOES, a sample-efficient framework for adapting locomotion policies to footwear-induced morphology changes. SHOES freezes a simulation-trained nominal policy and fine-tunes only lightweight, identity-initialized modulation modules on the robot, preserving the initial behavior while enabling efficient hardware adaptation. We also introduce SafeBox, a workspace-aware waypoint sampler that constrains online adaptation within a user-defined workspace. The simulation results with a shoe-equipped Unitree Go2 show that SHOES substantially recovers locomotion performance after footwear-induced morphology changes using limited additional adaptation. Under the same adaptation budget, SHOES outperforms FiLM-only, Linear Adapter-only, and residual-RL baselines. Real-world experiments on a Unitree Go2 fitted with human shoes further evaluate the effectiveness of SHOES for hardware deployment.

Chinese Translation

人类通过穿戴适当的鞋类来扩展其运动能力，例如在山地穿登山靴或在冰面上穿滑冰鞋。鞋类改变了与地面的接触界面，使得在原本困难的地形上行走成为可能。相比之下，运动控制器通常是在固定的形态下进行训练的，当鞋子改变脚的几何形状、质量和摩擦时，它们的性能可能会下降。我们提出了SHOES，这是一个高效的框架，用于将运动策略适应于鞋类引起的形态变化。SHOES冻结一个经过仿真训练的标准策略，仅对机器人上轻量级、身份初始化的调制模块进行微调，从而保持初始行为，同时实现高效的硬件适应。我们还引入了SafeBox，一个感知工作空间的路径点采样器，它限制在线适应在用户定义的工作空间内进行。与配备鞋子的Unitree Go2的仿真结果表明，SHOES在鞋类引起的形态变化后，能够在有限的额外适应下显著恢复运动性能。在相同的适应预算下，SHOES的表现优于仅使用FiLM、仅使用线性适配器和残差强化学习的基线。对配备人类鞋子的Unitree Go2的实际实验进一步评估了SHOES在硬件部署中的有效性。

View on arXiv Download PDF AI Translation

cs.RO / 139 / 2610.16463

DeTac: Dual-branch Decoupling of Texture and Force Signals in Vision-Based Tactile Sensors

DeTac：基于视觉的触觉传感器中纹理与力信号的双分支解耦

Zhang, Zhirui, Tang, Yunkai, Li, Jiahang, Fei, Fan, Zhang, Jinnuo, Ding, Kairan, Ren, Jieji, Guo, Yandong, Shi, Boxin

Abstract

Vision-based tactile sensors enable fine-grained perception for dexterous robotic manipulation, but current marker-based designs tightly couple surface texture and contact force signals within a single observation. Existing data-driven decoupling methods rely on serial mask-based inpainting, which suffers from mask-sizing dilemmas caused by physical indentation effects, error retention, and lack of real paired data for supervision. To address these challenges, we propose DeTac, a novel dual-branch generative framework that concurrently extracts clean shading and marker images from a single mixed tactile image. DeTac leverages full-image synthesis to naturally bypass mask dependency and introduces a cross-branch synergistic optimization that enables mutual supervision, effectively preventing error retention. Furthermore, we construct a strictly aligned, real-world paired tactile dataset using custom hardware to provide high-quality supervision. Extensive experiments demonstrate that DeTac achieves state-of-the-art decoupling accuracy, significantly facilitating downstream surface reconstruction and force estimation tasks while exhibiting cross-sensor generalization.

Chinese Translation

基于视觉的触觉传感器能够实现灵巧机器人操作的精细感知，但当前基于标记的设计将表面纹理和接触力信号紧密耦合在单一观测中。现有的数据驱动解耦方法依赖于串行的基于掩码的修复，这种方法受到物理压痕效应导致的掩码尺寸困境、错误保留以及缺乏真实配对数据进行监督的影响。为了解决这些挑战，我们提出了DeTac，一种新颖的双分支生成框架，能够同时从单一混合触觉图像中提取干净的阴影和标记图像。DeTac利用全图像合成自然绕过掩码依赖，并引入了跨分支协同优化，使得相互监督成为可能，有效防止错误保留。此外，我们使用定制硬件构建了一个严格对齐的真实配对触觉数据集，以提供高质量的监督。大量实验表明，DeTac实现了最先进的解耦精度，显著促进了下游表面重建和力估计任务，同时展现了跨传感器的泛化能力。

View on arXiv Download PDF AI Translation

cs.RO / 140 / 2610.16517

Trust-Calibrated VLA Policy Post-Training with Evolving Imagination

信任校准的VLA策略后训练与演变想象

Li, Wenzhan, Qin, Yiran, Zhou, Heng, Wang, Huirui, Guo, Yulan, Zhang, Ruimao

Abstract

Vision-Language-Action (VLA) models are often pretrained via imitation learning, making reinforcement learning (RL) post-training important when deployment moves beyond demonstration coverage. Existing refinement paradigms expose a reliability--efficiency tension: model-free refinement avoids simulator bias but is interaction-intensive, whereas model-based refinement can exploit unreliable imagination even as its simulator adapts. We propose Model-Evolving Reinforcement Learning (MERL), a trust-calibrated framework that couples policy optimization with an online-updated simulator comprising a world model and a reward proxy head. Across stages, simulator reliability controls the imagined-data ratio and rollout horizon; within rollouts, chunk-level trust controls replay selection and optimization weight. For recursive imagination without exact grounded continuations, a residual predictor estimates trust without querying future ground truth. On perturbed LIBERO-PRO tasks, MERL reaches 79.7% average success, improves high-threshold learning speed by 10.9% over the matched model-free control, and outperforms same-protocol external baselines across three training seeds. Fixed-data adaptation from 50 demonstrations per task achieves 80.0% average success on two physical-robot tasks.

Chinese Translation

视觉-语言-行动（VLA）模型通常通过模仿学习进行预训练，因此在部署超出演示覆盖范围时，强化学习（RL）后训练显得尤为重要。现有的精炼范式暴露了可靠性与效率之间的紧张关系：无模型的精炼避免了模拟器偏差，但交互密集；而基于模型的精炼则可以利用不可靠的想象，即使其模拟器在适应。我们提出了模型演变强化学习（MERL），这是一个信任校准的框架，将策略优化与一个在线更新的模拟器结合，该模拟器包括一个世界模型和一个奖励代理头。在不同阶段，模拟器的可靠性控制想象数据的比例和展开视野；在展开过程中，块级信任控制重放选择和优化权重。为了实现递归想象而不需要精确的基础延续，一个残差预测器在不查询未来真实情况的情况下估计信任。在扰动的LIBERO-PRO任务中，MERL达到了79.7%的平均成功率，相较于匹配的无模型控制提高了10.9%的高阈值学习速度，并在三个训练种子中超越了相同协议的外部基准。每个任务从50个演示的固定数据适应在两个物理机器人任务上达到了80.0%的平均成功率。

View on arXiv Download PDF AI Translation

cs.RO / 141 / 2610.16522

Learning Reachability-Based Safety Representations for Real-Time Robot Navigation in Unknown and Dynamic Environments

基于可达性学习的安全表示方法用于未知动态环境中的实时机器人导航

Derajić, Bojan

Abstract

Safe robot navigation in unknown, dynamic environments requires more than geometric collision checking: a robot still far from an obstacle may be unable to avoid it because of its speed, orientation, or the obstacle's motion. Hamilton--Jacobi (HJ) reachability analysis captures this in a value function whose sign separates states that admit a safe continuation from those where collision is unavoidable, but computing it online for every newly observed environment is too expensive. This dissertation trains neural implicit representations offline on reachability solutions to give optimization-based controllers this information in real time. Since the value function never exceeds the constraint, it equals a directly computed constraint, such as a signed distance, minus a nonnegative dynamics correction. Learning only this correction keeps represented obstacles outside the predicted safe set despite approximation errors. The first two of five methods use the learned function as a terminal constraint in model predictive control, for static local environments and predicted obstacle motion. The next two use it as a barrier candidate in control barrier function safety filters, for static environments and, through pairwise barriers in relative coordinates, for a varying number of moving obstacles. Simulations and hardware experiments on ground and aerial robots show better collision avoidance than geometric and hand-designed baselines, especially for short horizons and moving obstacles. The fifth method replaces the grid-based offline solver, which limits the first four to low-dimensional models. For control-affine dynamics with polytopic inputs, the HJ Hamiltonian is maximized at a vertex, so a parallelizable tree search over vertex inputs can generate training labels. With a loss from the HJ variational inequality, these labels yield barrier candidates for systems with up to 13 states.

Chinese Translation

在未知动态环境中安全的机器人导航不仅仅依赖几何碰撞检测：即使机器人距离障碍物较远，由于其速度、方向或障碍物的运动，仍可能无法避免碰撞。汉密尔顿-雅可比（Hamilton-Jacobi, HJ）可达性分析通过一个值函数捕捉这一点，该值函数的符号将允许安全继续的状态与无法避免碰撞的状态分开，但在每个新观察到的环境中在线计算这一函数的代价过高。本论文离线训练神经隐式表示，以可达性解为基础，实时为基于优化的控制器提供这一信息。由于值函数从不超过约束，它等于直接计算的约束（如带符号的距离）减去非负动态修正。仅学习这一修正可以使表示的障碍物保持在预测安全集之外，尽管存在近似误差。前两种方法将学习到的函数用作模型预测控制中的终端约束，适用于静态局部环境和预测的障碍物运动。接下来的两种方法将其用作控制障碍函数安全过滤器中的障碍候选，适用于静态环境，并通过相对坐标中的成对障碍物处理变化数量的移动障碍物。针对地面和空中机器人的仿真和硬件实验显示，与几何和手工设计的基线相比，碰撞避免效果更佳，尤其是在短时间范围和移动障碍物的情况下。第五种方法替代了基于网格的离线求解器，这限制了前四种方法只能应用于低维模型。对于具有多面体输入的控制仿射动态，HJ哈密尔顿量在一个顶点处达到最大值，因此可以通过对顶点输入进行可并行化的树搜索来生成训练标签。通过HJ变分不等式的损失，这些标签为最多13个状态的系统提供了障碍候选。

View on arXiv Download PDF AI Translation

cs.RO / 142 / 2610.16604

Design and Analysis of a Nonlinear Parallel Mechanism Coupling for Dynamic Humanoid Legs

动态人形腿部的非线性并联机制耦合设计与分析

Kumar, Rohit, Peters, Heiner, Wiebe, Felix, Stark, Franek, Janzen, Janne, Mronga, Dennis, Vyas, Shubham, Kirchner, Frank

Abstract

Humanoids increasingly demand lightweight leg mechanisms for agile, dynamic, and energy-efficient locomotion. Reducing distal inertia through proximal actuation has thus become a common design goal. Yet, compact transmissions that preserve torque capability and efficient force transmission remain hard to achieve. This paper presents a novel series-parallel leg design combining a knee parallel mechanism, an ankle parallel mechanism, and a nonlinear knee-ankle coupling where all actuators placed closer to the hip. As the knee flexes, the coupling passively assists ankle dorsiflexion for foot clearance and, as it extends it assists plantarflexion for push-off, redistributing ankle load through the knee transmission. We derive the coupling's nonlinear transmission models and validate them in simulation and on hardware. The design shows that the shared transmission lets both joints draw on each other's actuator capacity by easing the knee's peak torque demand under load and the ankle's peak speed demand during fast motion. %Together With the leg's proximal actuator placement, the leg achieves an effective reflected mass of 0.37kg at the foot for a 6kg leg, demonstrating very low distal leg inertia for dynamic humanoid robots.

Chinese Translation

人形机器人日益需要轻量化的腿部机制，以实现敏捷、动态和高效能的运动。因此，通过近端驱动来减少远端惯性已成为一个普遍的设计目标。然而，保持扭矩能力和高效力传输的紧凑传动系统仍然难以实现。本文提出了一种新颖的串并联腿部设计，结合了膝关节并联机制、踝关节并联机制以及一种非线性膝踝耦合，所有驱动器均靠近髋关节布置。当膝关节屈曲时，耦合被动地辅助踝关节背屈以提高足部离地高度；而当膝关节伸展时，它则辅助跖屈以实现助推，通过膝部传动重新分配踝关节的负载。我们推导了耦合的非线性传动模型，并在仿真和硬件上进行了验证。设计表明，共享传动使得两个关节能够相互利用对方的驱动能力，通过减轻膝关节在负载下的峰值扭矩需求和踝关节在快速运动中的峰值速度需求，从而实现更好的性能。结合腿部近端驱动器的布置，该腿部在6kg的总重下，脚部有效反射质量为0.37kg，展示了动态人形机器人极低的远端腿部惯性。

View on arXiv Download PDF AI Translation

cs.RO / 143 / 2610.16634

Language2Drive: Do Vision-Language-Action Models Really Follow the Instructions in Autonomous Driving?

Language2Drive：视觉-语言-动作模型在自动驾驶中真的遵循指令吗？

You, Zihan, Fan, Zehua, Lyu, Wenqi, Wang, Aoqi, Wang, Ni, Yang, Yifei, Zhao, Linge, Li, Huan, Ang, Sining, Yang, Yuguang, Chen, Jueyu, Li, Xiangyu, Wei, Qianshan, Pei, Jieyuan, Wang, Yuxin, Chen, Zhiwen, Li, Jiaqi, Song, Wenxuan, Qiao, Yanyuan, Zhu, Benjin, Li, Haoang, Wang, Yan

Abstract

Recent Vision-Language-Action (VLA) models increasingly use natural-language instructions to condition autonomous driving behavior. Yet existing closed-loop benchmarks primarily measure route completion, safety, and rule compliance, leaving a fundamental question unresolved: does language actually determine which feasible behavior the vehicle executes? A trajectory may be safe and successful while still violating or even ignoring the passenger's instruction. To address this, we introduce \ltd, a benchmark designed to directly test this capability. Its key evaluation unit, the paired counterfactual unit (PCU), holds the initial driving scene fixed while changing only the instruction. This design enables us to distinguish isolated instruction success from counterfactual paired success across four instruction-following scopes. To test whether this distinction persists across evaluation regimes, we instantiate Language2Drive in complementary closed-loop CARLA simulation, reconstructed real-world scenes, and open-loop driving benchmarks, including our released CARLA-Language2Drive simulator with purpose-built maps. Evaluating representative driving VLAs across these settings reveals a striking mismatch between strong driving competence and weak language control: 93.2% of successful rollouts do not have a successful counterpart under the alternative feasible instruction. This failure is dominated by one-sided completion, where a model repeatedly executes one preferred behavior but fails when the same scene requires the counterfactual alternative. These results expose two fundamental limitations of current evaluation: isolated success and driving scores, which are widely used by conventional driving benchmarks, can substantially overstate language-following capability, while current driving VLAs themselves lack reliable instruction following under matched counterfactual tests.

Chinese Translation

近年来，视觉-语言-动作（VLA）模型越来越多地使用自然语言指令来调节自动驾驶行为。然而，现有的闭环基准主要测量路线完成情况、安全性和规则遵循，未能解决一个根本性问题：语言是否真正决定了车辆执行的可行行为？一条轨迹可能是安全和成功的，但仍然可能违反或忽视乘客的指令。为了解决这个问题，我们引入了Language2Drive（ extit{ltd}），这是一个旨在直接测试这一能力的基准。其关键评估单元是配对反事实单元（PCU），在固定初始驾驶场景的同时，仅改变指令。该设计使我们能够区分四个指令遵循范围内的孤立指令成功与反事实配对成功。为了测试这种区分是否在不同评估机制中持续存在，我们在互补的闭环CARLA仿真、重建的真实场景和开放环驾驶基准中实例化了Language2Drive，包括我们发布的具有专用地图的CARLA-Language2Drive仿真器。在这些设置中评估代表性的驾驶VLA，揭示了强大的驾驶能力与弱语言控制之间的显著不匹配：93.2%的成功执行在替代可行指令下没有成功的对应项。这种失败主要表现为单方面完成，即模型反复执行一种首选行为，但在同一场景需要反事实替代时却失败。这些结果揭示了当前评估的两个根本局限性：孤立成功和驾驶评分，这些评分在传统驾驶基准中被广泛使用，可能大幅夸大语言遵循能力，而当前的驾驶VLA在匹配反事实测试下缺乏可靠的指令遵循。

View on arXiv Download PDF AI Translation

cs.RO / 144 / 2610.16641

What Limits Cross-Task Generalization in World Action Models?

是什么限制了世界行动模型中的跨任务泛化？

Zhao, Chen, Ye, Zipeng, Lin, Yihan, Xie, Yuchen, Zhang, Jing

Abstract

Cross-task generalization remains difficult for world action models (WAMs), even when their predicted futures depict plausible manipulation. This gap can be attributed directly: we make this gap directly testable by decoupling a WAM into a world model and an inverse-dynamics model (IDM), then holding the IDM fixed while replacing the source of future images. We introduce Align Once, Evaluate Everywhere (AoE), a camera-space inverse-dynamics model pretrained on large-scale robot data, which provides this recovery interface for different future sources across tasks and environments. Through it, we find that actions recovered from recorded futures of unseen tasks execute almost as well as replayed demonstrations, whereas futures predicted by world models collapse across benchmarks. The failures are not generic: they concentrate on the gripper and interaction geometry while generation capacity is spent on task-irrelevant background. This diagnosis suggests that supervision should target exactly these regions. We therefore introduce the vision-to-action (VITA) loss, which passes generated futures through the frozen AoE and penalizes discrepancies in the recovered actions, redirecting generation toward action-relevant detail. Across two world models and three manipulation benchmarks, this simple change raises the execution success of generated futures from near zero to 47.0%. The same loss also transfers to WAMs and yields consistent gains in native control.

Chinese Translation

尽管世界行动模型（WAMs）所预测的未来场景表现出合理的操作，但跨任务泛化仍然困难。这一差距可以直接归因于：我们通过将WAM解耦为世界模型和逆动力学模型（IDM），并在替换未来图像源的同时保持IDM固定，使这一差距变得可直接测试。我们引入了Align Once, Evaluate Everywhere（AoE），这是一个在大规模机器人数据上预训练的相机空间逆动力学模型，它为不同任务和环境中的未来源提供了恢复接口。通过这一接口，我们发现从未见任务的记录未来中恢复的动作几乎与重放演示一样有效，而世界模型预测的未来在基准测试中则崩溃。这些失败并非普遍存在：它们集中在抓取器和交互几何上，而生成能力则被用于与任务无关的背景。这一诊断表明，监督应准确针对这些区域。因此，我们引入了视觉到动作（VITA）损失，该损失通过冻结的AoE传递生成的未来，并惩罚恢复动作中的差异，从而将生成重定向到与动作相关的细节。在两个世界模型和三个操作基准测试中，这一简单的改变使生成未来的执行成功率从接近零提高到47.0%。相同的损失也可以转移到WAMs，并在本地控制中产生一致的提升。

View on arXiv Download PDF AI Translation

cs.RO / 145 / 2610.16667

FEM-Validated Real-Time Modelling of Soft Tactile Deformation

有限元验证的软触觉变形实时建模

Terrile, Silvia, Shi, Kaichen, Kwon, Taejun, Lee, Jiseok, Lepora, Nathan, Lee, Yeejin, Nam, Saekwang, Psomopoulou, Efi

Abstract

Accurate modelling of soft tactile sensors remains difficult because finite element methods (FEM) are computa- tionally expensive, while real-time simulators often lack fidelity for hyperelastic deformation. Large physical datasets are also impractical to collect because repeated testing can damage fragile sensors. We present a data-efficient framework that combines limited real-world validation, high-fidelity FEM, and graph neural networks (GNNs) for accurate real-time prediction of two-dimensional tactile deformation in a planar soft tactile structure. The key idea is to decouple validation from training: 82 camera-tracked contact configurations are used only to tune and validate an ANSYS FEM model, which then generates 768 synthetic contact scenarios for GNN training. We evaluate both forward prediction from marker deformation to contact localisation and inverse inference from contact configuration to marker deformation. The validated FEM achieves sub-0.5 mm error in more than 90% of validation cases, while the learned GNN surrogate reaches a mean absolute error of 0.32 mm with millisecond-scale inference, compared with minute-scale FEM simulation. Models trained entirely on synthetic FEM data retain sub-millimetre accuracy on real measurements from the 2D tactile structure, demonstrating strong sim-to-real transfer within the tested planar setting. We further integrate the learned model into a real-time visualisation pipeline, enabling rapid deformation prediction for virtual prototyping and simulator-based design exploration without extensive physical testing.

Chinese Translation

软触觉传感器的准确建模仍然困难，因为有限元方法（FEM）计算成本高，而实时模拟器在超弹性变形方面往往缺乏真实性。由于重复测试可能损坏脆弱的传感器，收集大量物理数据集也不切实际。我们提出了一种数据高效的框架，结合有限的现实世界验证、高保真FEM和图神经网络（GNN），以准确实时预测平面软触觉结构中的二维触觉变形。关键思想是将验证与训练解耦：82个相机跟踪的接触配置仅用于调整和验证ANSYS FEM模型，随后生成768个合成接触场景用于GNN训练。我们评估从标记变形到接触定位的正向预测，以及从接触配置到标记变形的逆向推断。经过验证的FEM在超过90%的验证案例中实现了小于0.5毫米的误差，而学习到的GNN代理在毫秒级推断下达到了0.32毫米的平均绝对误差，相比之下，FEM模拟需要分钟级时间。完全基于合成FEM数据训练的模型在2D触觉结构的实际测量中保持了亚毫米的准确性，展示了在测试的平面设置中强大的仿真到现实的转移。我们进一步将学习到的模型集成到实时可视化管道中，实现了快速变形预测，以便于虚拟原型设计和基于模拟器的设计探索，而无需大量的物理测试。

View on arXiv Download PDF AI Translation

cs.RO / 146 / 2610.16696

Belief Space Planning for Risk Signal Temporal Logic via Mixed-Integer Convex Programming

基于混合整数凸编程的风险信号时序逻辑信念空间规划

Han, Shaohang, Verhagen, Joris, Liang, Kaier, Tumova, Jana

Abstract

Signal Temporal Logic (STL) specifies robot tasks with spatio-temporal requirements, and its robustness semantics allow planning by trajectory optimization. In practice, robots estimate their state from noisy measurements, and when the information gained from a measurement depends on the state, the robot can actively reduce its uncertainty by planning in the belief space. The resulting problem is, however, generally non-convex, for which gradient-based methods are neither complete nor globally optimal. In this paper, we consider Risk STL (RiSTL) planning in Gaussian belief space and show that an information-form parameterization of the belief, together with a convex relaxation of the belief dynamics, yields a mixed-integer convex program (MICP). We prove that the MICP is sound and complete, and that the relaxation preserves the globally optimal input trajectories. We validate the effectiveness of our method in numerical simulations.

Chinese Translation

信号时序逻辑（STL）通过时空要求来指定机器人任务，其鲁棒性语义允许通过轨迹优化进行规划。在实际应用中，机器人从噪声测量中估计其状态，当测量所获得的信息依赖于状态时，机器人可以通过在信念空间中规划来主动减少不确定性。然而，所产生的问题通常是非凸的，因此基于梯度的方法既不完整也不全局最优。在本文中，我们考虑在高斯信念空间中进行风险STL（RiSTL）规划，并展示信念的信息形式参数化与信念动态的凸松弛结合，产生一个混合整数凸规划（MICP）。我们证明了MICP是健全且完整的，并且松弛保持了全局最优输入轨迹。我们在数值模拟中验证了我们方法的有效性。

View on arXiv Download PDF AI Translation

cs.RO / 147 / 2610.16737

CopyQuad: Quadrupedal Skill Transfer from a Single Side-View Video Without a Depth Model

CopyQuad：从单侧视图视频中进行四足动物技能转移，无需深度模型

Danesh, Mohamad H., Hu, Raina, Xu, Mengxue, Lin, Hsiu-Chin

Abstract

Video-to-robot pipelines for quadrupedal imitation insert a learned lifting stage, either a monocular depth model or a per-clip 3D reconstruction, to recover the one axis a side-view camera cannot see. We present CopyQuad, which omits that stage and pins the missing axis to the robot's own stance width instead. The setting is side-view footage of sagittally dominated motion, where limb motion is almost entirely in the image plane. A frozen 2D keypoint detector and a fixed analytic retargeting then turn a single uncalibrated clip into a robot reference whose foot targets are reachable by construction, so no per-clip trajectory optimization is needed. One phase-indexed reward structure, whose only per-behavior freedom is three scalars from a shared sweep, learns a dog walk, a horse walk, a horse trot and two non-periodic gestures, all deployed zero-shot on hardware. We then measure what omitting the axis costs. In a training-free rigidity test, a video depth model varies the distance between the animal's hips by $40$ to $107\%$ of its own mean and a single-image metric model by $22$ to $58\%$, though that distance cannot move. Motion capture puts the true lateral paw travel of a dog walk at $4.5$ cm peak to peak, a mean deviation near $1.4$ cm against the $3.5$ cm of lateral foot error our policies already leave, so the discarded axis sits below the pipeline's resolution. Depth repaired to feasibility never beats the constant, and the unrepaired estimates put the robot on the ground.

Chinese Translation

四足模仿的视频到机器人管道插入了一个学习的提升阶段，使用单目深度模型或每个剪辑的3D重建，以恢复侧视相机无法看到的一个轴。我们提出了CopyQuad，省略了该阶段，并将缺失的轴固定在机器人的自身站立宽度上。该设置为侧视镜头，主要表现出矢状面运动，其中肢体运动几乎完全在图像平面内。一个冻结的2D关键点检测器和一个固定的解析重定向将单个未校准的剪辑转变为一个机器人参考，其足部目标在构造上是可达的，因此不需要每个剪辑的轨迹优化。一个相位索引奖励结构，其唯一的行为自由度是来自共享扫描的三个标量，学习了狗步、马步、马小跑和两个非周期性手势，所有这些都在硬件上零-shot 部署。然后我们测量省略该轴的代价。在一个无训练的刚性测试中，视频深度模型使动物的臀部之间的距离变化为其自身均值的40%到107%，而单图像度量模型则为22%到58%，尽管该距离无法移动。运动捕捉将狗步的真实侧向爪子移动量定为4.5厘米的峰值到峰值，平均偏差接近1.4厘米，而我们的策略已经留下了3.5厘米的侧向足部误差，因此被丢弃的轴低于管道的分辨率。修复到可行性的深度模型从未超过常数，而未修复的估计则将机器人置于地面上。

View on arXiv Download PDF AI Translation

cs.RO / 148 / 2610.16855

Human-Like Redundancy Resolution for Real-Time Bimanual Humanoid Teleoperation

类人冗余解决方案用于实时双手类人机器人遥操作

Tan, Chao, Sung, Ming Shin Cindy, Liang, Shuxian, Wan, Chaoqun, Huang, Zhen, Hua, Xian-Sheng

Abstract

Real-time bimanual humanoid teleoperation requires accurate wrist tracking while keeping arm configurations predictable. Because a 7-DoF arm admits multiple elbow configurations for the same wrist pose, wrist-only tracking can cause elbow drift or abrupt configuration changes. Fixed anthropomorphic constraints can stabilize elbow posture, but they can also block the redundant adjustments needed near singular configurations. We propose a human-like redundancy-resolution framework that combines adaptive elbow guidance with singularity-robust wrist tracking. Shoulder-wrist geometric retargeting generates robot-scale wrist targets. An adaptive elbow-offset criterion uses the offset direction to guide posture and scales this guidance by the offset magnitude, so the redundant degree of freedom is progressively released as the human arm straightens. A hierarchical velocity solver prioritizes wrist tracking and projects the adaptive elbow guidance into the wrist-task null space. In a fixed-wrist-target ablation, adaptive guidance reduces mean elbow-direction error by 54.4% compared with removing this guidance. In a workspace-boundary stress test, singularity-robust tracking reduces peak raw joint-velocity demand by 97.9% compared with the undamped pseudoinverse. Real-world bimanual tasks and cross-platform retargeting show applicability across robot morphologies.

Chinese Translation

实时双手类人机器人遥操作需要准确的手腕跟踪，同时保持手臂配置的可预测性。由于7自由度（7-DoF）手臂对于相同的手腕姿态允许多种肘部配置，仅依赖手腕跟踪可能导致肘部漂移或突变的配置变化。固定的人体工程学约束可以稳定肘部姿态，但也可能阻碍在奇异配置附近所需的冗余调整。我们提出了一种类人冗余解决框架，结合了自适应肘部引导和抗奇异性手腕跟踪。肩部-手腕几何重定向生成机器人尺度的手腕目标。自适应肘部偏移标准利用偏移方向引导姿态，并根据偏移幅度调整这一引导，因此随着人类手臂的伸展，冗余自由度逐渐释放。分层速度求解器优先考虑手腕跟踪，并将自适应肘部引导投影到手腕任务的零空间。在固定手腕目标的消融实验中，与去除该引导相比，自适应引导将平均肘部方向误差降低了54.4%。在工作空间边界压力测试中，抗奇异性跟踪将峰值原始关节速度需求降低了97.9%，与未阻尼的伪逆相比。实际的双手任务和跨平台重定向展示了该方法在不同机器人形态中的适用性。

View on arXiv Download PDF AI Translation

cs.RO / 149 / 2610.16969

UniLAP: A Unified Latent Action Policy for Cross-Embodiment Manipulation

UniLAP：一种用于跨体现操作的统一潜在动作策略

Yao, Lei, Feng, Qiao, Tong, Mutian, Yun, Qihao, Chau, Lap-Pui, Liu, Lingjie

Abstract

Latent Action Models (LAMs) can capture manipulation priors from action-free videos, but discrete latents usually impose a granularity bottleneck, limiting their ability to represent fine-grained action. Recent continuous alternatives mitigate this, but they are typically used as proxies for world modeling and require a further planning step to produce executable actions. We thus introduce UniLAP, a unified continuous latent-action policy that models the latent space via flow matching and bypasses reliance on future frames during inference. By distilling a LAM teacher with a flow-adaptive objective, UniLAP inherits the teacher's action representation while retaining the multimodal distribution of feasible behaviors. To decouple latent representations from task-irrelevant appearance, we extract a label-free motion target to isolate relevant kinematic cues from spurious visual dynamics. With the established embodiment-agnostic latent manifold, UniLAP can be easily adapted to novel platforms. Pretrained on multi-source videos at a compact 1.1B parameter scale, UniLAP outperforms state-of-the-art latent-action and Vision-Language-Action (VLA) baselines across hand pose forecasting, single-arm and humanoid manipulation, and real-world deployment.

Chinese Translation

潜在动作模型（LAMs）能够从无动作视频中捕捉操作先验，但离散潜在变量通常会造成粒度瓶颈，限制了其表示细粒度动作的能力。最近的连续替代方案缓解了这一问题，但它们通常被用作世界建模的代理，并且需要进一步的规划步骤来生成可执行的动作。因此，我们提出了UniLAP，一种统一的连续潜在动作策略，通过流匹配对潜在空间建模，并在推理过程中绕过对未来帧的依赖。通过使用流自适应目标对LAM教师进行蒸馏，UniLAP继承了教师的动作表示，同时保留了可行行为的多模态分布。为了将潜在表示与任务无关的外观解耦，我们提取了一个无标签的运动目标，以隔离相关的运动学线索与虚假的视觉动态。凭借建立的与体现无关的潜在流形，UniLAP可以轻松适应新平台。在以紧凑的11亿参数规模对多源视频进行预训练后，UniLAP在手部姿态预测、单臂和类人操作以及现实世界部署等任务中超越了最先进的潜在动作和视觉-语言-动作（VLA）基线。

View on arXiv Download PDF AI Translation

cs.RO / 150 / 2610.16978

Joint-Feasibility Margin Feedback for Bounded-Input Second-Order Control Barrier Functions

有界输入二阶控制障碍函数的联合可行性边际反馈

Yang, Siyuan, He, Shaoming

Abstract

Multiple second-order control barrier function constraints can exhaust their common control authority under bounded acceleration, even when the current state remains feasible. We define a signed local joint-feasibility margin whose sign exactly identifies whether the current barrier half-spaces intersect the admissible input set. For unit-scaled distance constraints, it has acceleration units and a support-function dual that identifies weighted constraint conflicts. We predict the margin one sample ahead for nine task-directed candidates, selecting either the least-deviating candidate that attains a desired margin or the candidate with the largest predicted margin before solving the same hard HOCBF quadratic program. A controlled one-step case shows that, at the state predicted under the hard-QP command, two constraints remain individually feasible but become jointly infeasible, whereas the margin-feedback controller retains a positive next margin. Separately, across 20 paired initial conditions, the margin-feedback controller retains the hard-QP command in a low-pressure task and acts at positive margin in a critical task before the hard-QP problem becomes infeasible; its trajectories remain feasible at that time and complete the task. Source code is publicly available. https://github.com/xirhxq/cbf/tree/robio2026-preventive-regulation

Chinese Translation

在有界加速度的情况下，多个二阶控制障碍函数约束可能会耗尽其共同的控制权，即使当前状态仍然可行。我们定义了一个带符号的局部联合可行性边际，其符号准确地识别当前障碍半空间是否与可接受输入集相交。对于单位缩放的距离约束，它具有加速度单位，并且有一个支持函数对偶，能够识别加权约束冲突。我们预测九个任务导向候选者的边际，选择要么是实现所需边际的最小偏差候选者，要么是预测边际最大的候选者，然后解决相同的困难高阶控制障碍函数（HOCBF）二次规划。一个受控的一步案例表明，在硬二次规划命令下预测的状态下，两个约束仍然各自可行，但变得联合不可行，而边际反馈控制器保持正的下一个边际。此外，在20对初始条件中，边际反馈控制器在低压任务中保持硬二次规划命令，并在关键任务中以正边际行动，直到硬二次规划问题变得不可行；此时其轨迹仍然可行并完成任务。源代码已公开可用。https://github.com/xirhxq/cbf/tree/robio2026-preventive-regulation

View on arXiv Download PDF AI Translation

cs.RO / 151 / 2610.17001

MOSAIC: Mode- and Support-Aware Inference-Time Control of Robot Diffusion Models

MOSAIC：基于模式和支持的机器人扩散模型推理时控制

Bodmer, Sabrina, Rickenbach, Rahel, Zurbrügg, René, Zeilinger, Melanie N., Jones, Colin

Abstract

Generative policies such as diffusion models provide expressive data-driven priors for robot behavior, yet the situations a robot faces at execution time rarely match its training data and impose constraints and objectives the policy never learned. Existing inference-time guidance methods either rely on gradients that can push samples away from the learned prior, or on sampling procedures that can be inefficient, collapse modes, or average them into out-of-distribution predictions. We introduce MOSAIC, a mode- and support-aware framework for gradient-free inference-time control of robotic diffusion models. MOSAIC treats the initial noise and all stochastic reverse-step perturbations as a joint input to the pretrained denoising process and adapts a distribution over this input using importance weights computed from fully denoised outputs. Physical feasibility and consistency with the learned support are balanced using adaptive penalties, while mode-wise importance-weighted updates preserve modes and avoid destructive averaging. The same formulation can jointly steer multiple independently trained diffusion models through shared coupling costs, enabling coordinated behavior without requiring a joint model. We evaluate MOSAIC on navigation, dexterous grasping, motion planning with unseen geometry, and collaborative box manipulation with independently trained robot priors. Across these settings, MOSAIC adapts and composes frozen diffusion priors under new deployment constraints without retraining.

Chinese Translation

生成策略，如扩散模型，为机器人行为提供了表达丰富的数据驱动先验，然而，机器人在执行时面临的情况很少与其训练数据匹配，并且施加了政策从未学习过的约束和目标。现有的推理时指导方法要么依赖于可能将样本推离学习先验的梯度，要么依赖于可能效率低下、导致模式崩溃或将其平均为分布外预测的采样过程。我们提出了MOSAIC，这是一个无梯度的推理时控制框架，专为机器人扩散模型设计，关注模式和支持。MOSAIC将初始噪声和所有随机反向步骤扰动视为预训练去噪过程的联合输入，并使用从完全去噪输出计算的重要性权重来调整该输入的分布。通过自适应惩罚平衡物理可行性与学习支持的一致性，而模式加权更新则保持模式并避免破坏性平均。相同的公式可以通过共享耦合成本共同引导多个独立训练的扩散模型，实现协调行为而无需联合模型。我们在导航、灵巧抓取、具有未见几何形状的运动规划以及与独立训练的机器人先验的协作箱体操作等场景中评估了MOSAIC。在这些设置中，MOSAIC在不重新训练的情况下，适应并组合冻结的扩散先验，以应对新的部署约束。

View on arXiv Download PDF AI Translation

cs.RO / 152 / 2610.17069

Task-Space Motion-Force Nonlinear Model Predictive Control for Precise and Compliant Control under Uncertainties

针对不确定性下精确和柔性控制的任务空间运动-力非线性模型预测控制

Nguyen, Quang-Nam, Righetti, Ludovic

Abstract

In Nonlinear Model Predictive Control (NMPC), the control law is implicitly determined by the solution of an Optimal Control Problem (OCP). Consequently, OCP design is often tailored to specific platforms and tasks, demanding extensive expert knowledge and tuning effort. To address this, we propose a unified NMPC framework that directly plans and controls the robot joints subject to task-space motion and force requirements and is capable of seamlessly generalizing across multiple robotic platforms and diverse tasks. Furthermore, optimization-based methods often treat interactions as rigid contacts, making the torque solution sensitive to the modeling errors of the robot dynamics and the environment. We study the intrinsic task-space impedance of our proposed controller, its partial contraction for differential stability and differential passivity under unmodeled interactions. As a result, our framework enables closing the force-feedback loop in the high-level planning layer to achieve precise motion-force tracking performance, addressing the long-standing challenge of force-feedback NMPC. Finally, we validate our framework on physical hardware, demonstrating hybrid motion-force control on a redundant manipulator and compliant locomotion on a quadruped, while preserving core NMPC advantages such as hard constraint satisfaction, optimal redundancy exploitation, and real-time planning under unforeseen interactions.

Chinese Translation

在非线性模型预测控制（NMPC）中，控制律是通过最优控制问题（OCP）的解隐式确定的。因此，OCP设计通常针对特定平台和任务进行定制，要求大量的专家知识和调整工作。为了解决这个问题，我们提出了一个统一的NMPC框架，该框架直接规划和控制机器人关节，以满足任务空间的运动和力要求，并能够在多个机器人平台和多样化任务之间无缝泛化。此外，基于优化的方法通常将相互作用视为刚性接触，这使得扭矩解对机器人动力学和环境的建模误差敏感。我们研究了所提控制器的内在任务空间阻抗，以及在未建模相互作用下的部分收缩以实现微分稳定性和微分被动性。因此，我们的框架能够在高层规划层中闭合力反馈环路，以实现精确的运动-力跟踪性能，解决了力反馈NMPC长期以来面临的挑战。最后，我们在物理硬件上验证了我们的框架，展示了在冗余操纵器上的混合运动-力控制和在四足机器人上的柔性运动，同时保持了NMPC的核心优势，如硬约束满足、最优冗余利用和在不可预见的相互作用下的实时规划。

View on arXiv Download PDF AI Translation

cs.RO / 153 / 2610.17083

Declarative Reactive Robot Behaviors from Imperative Language Models

从命令式语言模型生成声明式反应式机器人行为

Berenz, Vincent, Kim, Tasha, Schölkopf, Bernhard

Abstract

Large Language Models (LLMs) are predominantly trained on imperative programming languages, and existing work on LLM-based robot programming reflects this: it focuses on goal-directed tasks expressed as plans. However, many robotic applications require not tasks but behaviors--open-ended, continuous patterns of sensory-motor engagement with no termination condition (e.g., greeting visitors, patrolling an area). Such behaviors are better served by declarative and reactive programming, on which LLMs have virtually no training. We investigate whether this gap can be bridged. Using Playful, a reactive scripting language, we show that an LLM-based coding agent can generate correct declarative and reactive programs for robotic behaviors. These programs are produced from under-specified natural language descriptions, given suitable context and a set of reusable programming guidelines. The guidelines encode design principles that redirect the agent from imperative defaults toward proper reactive patterns. Developed once by an expert, they generalize across robots and behaviors. Starting from only the generic robot operating system API, the agent jointly designs both the imperative component (Python modules for sensing, acting, and evaluating) and the declarative component (the reactive script that composes them into behavior logic). The declarative nature of reactive programming supports both incremental refinement through conversational turns and one-shot generation of complete behaviors.

Chinese Translation

大型语言模型（LLMs）主要是在命令式编程语言上进行训练的，而基于LLM的机器人编程现有研究也反映了这一点：它集中于以计划形式表达的目标导向任务。然而，许多机器人应用需要的不是任务，而是行为——开放式的、持续的感知-运动交互模式，没有终止条件（例如，迎接访客、巡逻区域）。这样的行为更适合使用声明式和反应式编程，而LLMs在这方面几乎没有训练。我们探讨是否可以弥补这一差距。通过使用Playful，一种反应式脚本语言，我们展示了基于LLM的编码代理可以为机器人行为生成正确的声明式和反应式程序。这些程序是从不完全指定的自然语言描述中生成的，前提是提供合适的上下文和一套可重用的编程指南。这些指南编码了设计原则，指导代理从命令式默认模式转向适当的反应式模式。由专家一次性开发的这些指南可以在不同的机器人和行为之间进行推广。代理从通用的机器人操作系统API开始，联合设计命令式组件（用于感知、行动和评估的Python模块）和声明式组件（将它们组合成行为逻辑的反应式脚本）。反应式编程的声明性特征支持通过对话轮次进行增量细化以及一次性生成完整行为。

View on arXiv Download PDF AI Translation

cs.RO / 154 / 2610.17089

Interactive Navigation Beyond Reachability: Benchmark and Agentic Framework

超越可达性的互动导航：基准与代理框架

Li, Delun, Wei, Xueyao, Chen, Rui, Wang, Yifei, Yu, Junzhi, Song, Wenjie, Chen, Xingyu

Abstract

Existing embodied navigation tasks commonly assume that traversability and object states remain fixed during an episode. Yet reaching a goal may require opening a door to access a room and then opening a container to reveal a hidden target. We formulate interactive navigation beyond reachability, unifying channel-induced changes in spatial connectivity, container-induced changes in target visibility, and their sequential dependencies. We introduce MultINav-Bench, a 3K-episode ObjectGoal benchmark in MolmoSpaces spanning channel, container, and mixed tasks, with interaction-necessity annotations covering required and unnecessary episodes. Its data generation pipeline automates task construction across multiple task types. The evaluation protocol separately measures navigation success, interaction effectiveness, and cost. We further propose MultINav-Agent, a three-stage agentic framework built on a persistent, state-conditioned interaction graph. Semantic interactive perception grounds entities and interaction states; a vision-language model selects task-relevant actions based on expected changes in connectivity and target observability; and post-interaction verification updates the graph to guide subsequent navigation, exploration, or interaction. Experiments on MultINav-Bench show that MultINav-Agent consistently outperforms adapted baselines across channel, container, and mixed tasks.

Chinese Translation

现有的具身导航任务通常假设在一个回合中可通行性和物体状态保持不变。然而，达到目标可能需要打开一扇门以进入一个房间，然后打开一个容器以揭示一个隐藏的目标。我们提出超越可达性的互动导航，统一了通道引起的空间连通性变化、容器引起的目标可见性变化及其顺序依赖关系。我们引入了MultINav-Bench，这是一个包含3000个回合的ObjectGoal基准，涵盖了通道、容器和混合任务，且包含了对所需和不必要回合的互动必要性注释。其数据生成管道自动化了多种任务类型的任务构建。评估协议分别测量导航成功率、互动有效性和成本。我们进一步提出了MultINav-Agent，这是一个基于持久状态条件互动图的三阶段代理框架。语义互动感知为实体和互动状态提供基础；视觉-语言模型根据预期的连通性和目标可观察性的变化选择与任务相关的动作；而后互动验证则更新图以指导后续的导航、探索或互动。在MultINav-Bench上的实验表明，MultINav-Agent在通道、容器和混合任务中始终优于适应的基线。

View on arXiv Download PDF AI Translation

cs.RO / 155 / 2610.17114

D-World: An Explicit World-Language-Action Model with Recursive Self-Improvement under Dual Control

D-World：一种在双重控制下具有递归自我改进的显式世界-语言-行动模型

Wen, Xuan, Wang, Baichao

Abstract

Frontier large language models (LLMs) now reason about space well enough to program robots, yet the world models behind robot policies are implicit, held in network weights or predicted video, and cannot be inspected or corrected directly. We present D-World, an explicit world-language-action model built on the Sense, Plan and Act decomposition. A perception agent writes the scene as an executable digital twin using generated vision and geometry code. A planning agent searches for actions in simulation and, after each executed action chunk, revises the twin's object states and named physical parameters against what the robot measured. A parameter revision is accepted only if replaying the latest chunk does not increase any measured residual beyond its tolerance, and every chunk is checked in the updated twin before it runs. Structural mismatches return to perception for reconstruction. We call this measurement-grounded cycle of correction and revision recursive self-improvement (RSI), in which the LLM stays frozen while the world it acts in improves. We model the choice of actions that both advance the task and expose model error as a dual control problem and show that the cost of incorporating evidence into the twin sets the threshold at which probing pays, so cheap local revision counteracts the turn-off of cautious controllers. Each episode runs a real-to-sim-to-real loop, and its recorded twins can be refitted and re-rendered offline into higher-fidelity environments for post-training. Because perception and planning communicate only through the twin and language, either can be replaced by a learned model, and since inference dominates the loop, the loop speeds up with faster models. On a dual-arm robot, towel and toaster tasks show local revisions, such as one reducing the bread's rendered-depth residual from 22.4 to 1.5 mm, that let execution continue without reconstruction.

Chinese Translation

前沿的大型语言模型（LLMs）现在能够足够好地推理空间以编程机器人，但支撑机器人策略的世界模型是隐式的，存储在网络权重或预测视频中，无法直接检查或纠正。我们提出了D-World，这是一种基于感知、规划和行动分解的显式世界-语言-行动模型。感知代理使用生成的视觉和几何代码将场景写成可执行的数字双胞胎。规划代理在模拟中搜索行动，并在每个执行的行动块后，依据机器人测量的结果修正双胞胎的对象状态和命名物理参数。只有当重放最新的块不会使任何测量残差超出其容忍度时，参数修正才被接受，并且每个块在运行之前都在更新的双胞胎中进行检查。结构不匹配返回感知进行重建。我们将这种基于测量的纠正和修正循环称为递归自我改进（RSI），在此过程中，LLM保持不变，而其所处的世界不断改进。我们将推进任务和暴露模型错误的行动选择建模为双重控制问题，并展示将证据纳入双胞胎的成本设定了探测的收益阈值，因此廉价的局部修正抵消了谨慎控制器的关闭。每个回合运行一个真实-模拟-真实的循环，其记录的双胞胎可以离线重新拟合和重新渲染到更高保真度的环境中以进行后训练。由于感知和规划仅通过双胞胎和语言进行通信，因此任一方都可以被学习模型替代，并且由于推理主导循环，循环随着更快的模型而加速。在一个双臂机器人上，毛巾和烤面包机任务展示了局部修正，例如将面包的渲染深度残差从22.4毫米减少到1.5毫米，使得执行可以在不重建的情况下继续进行。

View on arXiv Download PDF AI Translation

cs.RO / 156 / 2610.17161

Force-Modulated Dexterity: Unified Compliance for Contact-Rich Manipulation

力调制灵巧性：接触丰富操作的统一顺应性

Yuan, Ying, Held, David, Malik, Jitendra, Qi, Haozhi

Abstract

Teleoperation is widely used to collect high-quality demonstrations for robot learning. Dexterous manipulation, however, requires coordinated control of motion and contact forces. Operators can specify motion, but struggle to reproduce the subtle force adjustments of human manipulation without feeling the robot's contacts. We introduce Force-Modulated Dexterity (FOMO-Dex), a unified compliance formulation for contact-rich manipulation. FOMO-Dex expresses nominal actions in coordinates defined by task geometry and contact structure. It applies force-driven motion corrections along selected directions while leaving the remaining command components unchanged. Across card picking, paper cutting, screwdriving, and nut-and-bolt assembly, FOMO-Dex reduces excessive contact forces compared with stiff control. A user study shows that operators collect more successful demonstration data per hour after brief practice. For policy learning, imitating the corrected actions improves card-picking success over imitating raw operator commands. Policies trained on FOMO-Dex demonstrations also achieve more rotation in nut-and-bolt manipulation and screwdriving than those trained on demonstrations collected with the stiff controller. Videos are shown on https://yingyuan0414.github.io/fomo-dex/ .

Chinese Translation

远程操作广泛用于收集高质量的机器人学习示范。然而，灵巧操作需要运动和接触力的协调控制。操作员可以指定运动，但在没有感知机器人接触的情况下，难以重现人类操作的微妙力调整。我们提出了力调制灵巧性（Force-Modulated Dexterity, FOMO-Dex），这是一种用于接触丰富操作的统一顺应性公式。FOMO-Dex在由任务几何和接触结构定义的坐标系中表达名义动作。它沿选定方向施加基于力的运动修正，同时保持其余指令组件不变。在卡片拾取、纸张切割、螺丝拧紧和螺母与螺栓组装等任务中，与刚性控制相比，FOMO-Dex减少了过大的接触力。一项用户研究表明，经过简短的练习后，操作员每小时收集到的成功示范数据更多。在策略学习中，模仿修正后的动作比模仿原始操作员命令提高了卡片拾取的成功率。基于FOMO-Dex示范训练的策略在螺母与螺栓操作和螺丝拧紧中也实现了比基于刚性控制器收集的示范训练的策略更多的旋转。相关视频可在 https://yingyuan0414.github.io/fomo-dex/ 查看。

View on arXiv Download PDF AI Translation

cs.RO / 157 / 2610.17174

Generative Flow on Symbolic Abstractions for Embodiment Capability Discovery

基于符号抽象的生成流用于体现能力发现

Yu, Youwei, Liu, Lantao

Abstract

Achieving complex, long-horizon robot behavior remains challenging because successful execution depends on skills grounded in a robot's physical affordances. Existing training-task distributions are often handcrafted or sampled from large language models (LLMs), relying on semantic plausibility rather than embodiment-specific feasibility. This mismatch can produce curricula that are too difficult, redundant, or blind to non-obvious but learnable behaviors. We propose Generative Flow on Symbolic Abstractions (FOSA), a framework for discovering embodiment-specific capabilities by learning a grounded task generator. FOSA parameterizes an abstract policy as a Generative Flow Network that constructs symbolic task skeletons. Each skeleton is compiled into a simulation environment, where low-level policy-learning performance provides the reward signal for the abstract policy. Across heterogeneous robot platforms, FOSA discovers diverse and executable tasks near the robot's empirical capability frontier. After distilling the discovered task policies into a single language-conditioned policy, FOSA improves performance on unseen complex and long-horizon tasks. The code and demonstration videos are available at https://www.youwei-yu.com/fosa.

Chinese Translation

实现复杂的、长时间跨度的机器人行为仍然具有挑战性，因为成功执行依赖于基于机器人物理可供性（affordances）的技能。现有的训练任务分布通常是手工制作的或从大型语言模型（LLMs）中抽样的，这些模型依赖于语义上的合理性，而非特定于体现的可行性。这种不匹配可能导致课程设计过于困难、冗余，或对非显而易见但可学习的行为视而不见。我们提出了基于符号抽象的生成流（FOSA），这是一个通过学习一个基础任务生成器来发现特定于体现的能力的框架。FOSA将抽象策略参数化为生成流网络（Generative Flow Network），该网络构建符号任务骨架。每个骨架被编译到一个仿真环境中，在该环境中，低级策略学习性能提供了抽象策略的奖励信号。在异构机器人平台上，FOSA发现了多样且可执行的任务，接近机器人的经验能力边界。在将发现的任务策略提炼为单一的语言条件策略后，FOSA在未见过的复杂和长时间跨度任务上提高了性能。代码和演示视频可在 https://www.youwei-yu.com/fosa 获取。

View on arXiv Download PDF AI Translation

cs.RO / 158 / 2610.17208

DexNinja: Learning Robust Dexterous Cutting Policy with a Real-to-Sim-to-Real Data Engine

DexNinja：通过真实-模拟-真实数据引擎学习稳健的灵巧切割策略

Lou, Haozhe, Yang, Runyi, Zhong, Wanqi, Liu, Caiyun, Liu, Yicheng, Liu, Wenhao, Ma, Wenlong, Xia, Jiawei, Wu, Kanzhi, Paudel, Danda Pani, Van Gool, Luc, Zhao, Hang, Li, Yiming

Abstract

Cutting a piece of food looks effortless in human hands. For a robot, however, every cut is a high-contact interaction with a deformable, elasto-plastic object whose behavior changes with pose, contact mode, and cutting stage. Collecting real-world data is expensive and wasteful: each trial typically contaminates or damages the food, so obtaining scalable real world data for dexterous cutting is expensive and difficult. Simulation is an appealing alternative, but a large sim-to-real gap exists in both physics and perception. This gap is especially severe for topology-changing interactions like cutting, where small errors in contact, friction, or material and elasticity responses quickly compound. We propose \textsc{DexNinja}, a real2sim2real framework that turns a handful of real demonstrations into a dense and realistic training distribution. \textsc{DexNinja} (i) reconstructs object instances from real captures, (ii) randomizes physically meaningful parameters under category-level constraints, and (iii) produces augmented dexterous manipulation episodes using a custom differentiable simulator that couples robot dynamics, tactile contact, and deformable cutting. We evaluate \textsc{DexNinja} on a dexterous food-slicing task and show that (1) with tens of real demonstrations and hundreds of augmented simulator data, the success rate could reach more than 60\%. (2) the augmented data consistently improves success rates when only using real data with more than 20\% via sim-to-real, and (3) the resulting policies generalize to out-of-distribution objects with varying shapes and sizes.

Chinese Translation

在人的手中，切割一块食物看似毫不费力。然而，对于机器人来说，每一次切割都是与可变形的弹塑性物体进行高接触交互，其行为会随着姿态、接触模式和切割阶段而变化。收集真实世界的数据既昂贵又浪费：每次试验通常会污染或损坏食物，因此获取可扩展的真实世界灵巧切割数据既昂贵又困难。模拟是一个有吸引力的替代方案，但在物理和感知上存在着巨大的模拟到真实的差距。对于像切割这样的拓扑变化交互，这种差距尤其严重，因为接触、摩擦或材料和弹性响应中的小错误会迅速累积。我们提出了 extsc{DexNinja}，一个真实-模拟-真实框架，将少量真实演示转化为密集且真实的训练分布。 extsc{DexNinja} (i) 从真实捕获中重建物体实例，(ii) 在类别级别约束下随机化物理上有意义的参数，以及 (iii) 使用一个自定义的可微分模拟器生成增强的灵巧操作情节，该模拟器耦合了机器人动力学、触觉接触和可变形切割。我们在灵巧食品切割任务上评估了 extsc{DexNinja}，结果表明 (1) 通过数十个真实演示和数百个增强的模拟器数据，成功率可以超过 60%。 (2) 增强数据在仅使用真实数据时，成功率通过模拟到真实的方式始终提高超过 20%。 (3) 生成的策略能够推广到形状和大小各异的分布外物体。

View on arXiv Download PDF AI Translation

cs.RO / 159 / 2610.17232

LiDAR-based Humanoid Navigation in Dynamic, Non-Convex Environments via Spatio-Temporal Safe Corridors

基于LiDAR的人形机器人在动态非凸环境中的导航：时空安全通道方法

Kumbhar, Shubham S., Artemiadis, Panagiotis

Abstract

Navigating humanoid robots through dynamic, non-convex environments requires tightly coupling onboard perception with footstep planning and dynamic locomotion control. Existing methods frequently discard complex free space through simplified convex obstacle bounds and fail to account for real-time dynamic occupancy predictions. This paper presents a real-time framework for humanoid navigation that directly constructs spatio-temporal safe corridors from LiDAR-based particle representations and couples them with Linear Inverted Pendulum (LIP)-based planning and Whole Body Control (WBC) to enable navigation around and within moving non-convex structures without restrictive geometric assumptions. Simulations on the Digit humanoid yield a 96% success rate in navigating around and inside moving non-convex obstacles in a 24 by 24 m^2 arena. Ablation studies confirm that the proposed iterative polytope generation significantly outperforms single-iteration baselines in narrow corridors, while evaluations demonstrate robustness to clutter, sensor noise, control delays, and force perturbations.

Chinese Translation

在动态非凸环境中导航人形机器人需要将机载感知与步态规划和动态运动控制紧密结合。现有方法常常通过简化的凸障碍边界丢弃复杂的自由空间，并未考虑实时动态占用预测。本文提出了一种实时框架，通过基于LiDAR的粒子表示直接构建时空安全通道，并将其与基于线性倒立摆（Linear Inverted Pendulum, LIP）规划和全身控制（Whole Body Control, WBC）相结合，使得在不受限几何假设的情况下，能够在动态非凸结构周围和内部进行导航。在24平方米的数字人形机器人（Digit）模拟实验中，成功率达到96%，能够在动态非凸障碍物周围和内部导航。消融研究确认，所提出的迭代多面体生成方法在狭窄通道中显著优于单次迭代基线，而评估结果表明该方法对杂乱环境、传感器噪声、控制延迟和力扰动具有良好的鲁棒性。

View on arXiv Download PDF AI Translation

cs.RO / 160 / 2610.17252

Value-Preserving Action Quantization

价值保留的动作量化

Sun, Max Muchen, Kwon, Jueun, Riviere, Benjamin, Murphey, Todd

Abstract

Search-based policy learning, such as Monte Carlo tree search (MCTS), offers the flexibility to trade runtime computation for data efficiency and adaptation to novel environments. For robot learning, however, applying search requires discretizing the continuous action space and associated dynamics, typically through action quantization. While reconstruction-based quantization is sufficient for representing the continuous action space alone, policy learning additionally requires the quantized actions to preserve their effects on future states and values. We show that such requirement is not necessarily captured by action reconstruction error. Instead, we prove that the error in the value function resulting from action quantization is tightly bounded by the reconstruction error of state reachability under the quantized actions. Based on these results, we propose value-preserving action quantization (VAQ), which quantizes actions by preserving state reachability. We further combine VAQ with MCTS (VAQ-MCTS) as a value-preserving, search-based policy learning framework for behavioral cloning. Using a wide range of object pushing tasks as a controlled benchmark, we evaluate VAQ-MCTS against state-of-the-art behavioral cloning methods. VAQ-MCTS matches or exceeds their performance under standard settings, while substantially improving data efficiency and robustness to heterogeneous demonstrations, while enabling zero-shot adaptation to novel environments. Hardware experiments further demonstrate the effectiveness of VAQ-MCTS, completing all 30 real-world object pushing trials with only 10 minutes of demonstration data.

Chinese Translation

基于搜索的策略学习，如蒙特卡洛树搜索（Monte Carlo tree search, MCTS），提供了在运行时计算与数据效率及适应新环境之间进行权衡的灵活性。然而，对于机器人学习而言，应用搜索需要对连续动作空间及其相关动态进行离散化，通常通过动作量化来实现。虽然基于重建的量化足以表示连续动作空间，但策略学习还要求量化后的动作能够保留其对未来状态和价值的影响。我们表明，这一要求并不一定通过动作重建误差来捕捉。相反，我们证明了由于动作量化导致的价值函数误差与量化动作下状态可达性的重建误差之间存在紧密的界限。基于这些结果，我们提出了价值保留的动作量化（Value-Preserving Action Quantization, VAQ），该方法通过保留状态可达性来量化动作。我们进一步将VAQ与MCTS结合（VAQ-MCTS），形成一个价值保留的基于搜索的策略学习框架，用于行为克隆。通过使用广泛的物体推动任务作为受控基准，我们评估了VAQ-MCTS与最先进的行为克隆方法的表现。VAQ-MCTS在标准设置下的性能与这些方法相匹配或超过，同时显著提高了数据效率和对异构演示的鲁棒性，并实现了对新环境的零样本适应。硬件实验进一步验证了VAQ-MCTS的有效性，在仅用10分钟的演示数据的情况下，完成了所有30个真实世界物体推动试验。

View on arXiv Download PDF AI Translation

cs.RO / 161 / 2610.17260

Physical SLAM: Physics-aware Simultaneous Localisation and Mapping

物理SLAM：物理感知的同时定位与地图构建

Wang, Guangming, Liu, Zhishuo, Wu, Haibing, Ma, Wanli, Jing, Yixiong, Wysocki, Olaf, Sheil, Brian

Abstract

To reconstruct a predictive 4D representation of the physical world from monocular video captured by a moving camera, we jointly reconstruct dynamic Gaussian radiance fields and infer their underlying physical dynamics. This paper proposes \emph{Physical SLAM}, a physics-aware method for constructing 4D Gaussian fields of deformable objects and predicting their future state when subjected to external actions, given a sequence of RGB-D images. Our method separates the scene into a static background Gaussian field for camera localization and a dynamic object Gaussian field driven by a spring--mass physical model of elastic deformation, inertia, and damping. To integrate visual observations with physical dynamics, we introduce virtual observation forces that gradually pull simulated surface nodes toward observed depth points, avoiding abrupt replacement of simulated node positions. Meanwhile, the physical parameters of the spring--mass model are updated from simulation errors aggregated over the full reconstruction observation window, enabling the reconstructed dynamics to align with observations. To couple appearance, geometry, and physics, we further design a physics-aware Gaussian transformation and surface alignment strategy, where dynamic Gaussians are transformed along the deforming physical surface of the spring--mass model and constrained to remain aligned with the surface during rendering-based optimization. The resulting representation supports observed-frame rendering and action-conditioned simulation, extending SLAM maps to describe how deformable objects respond to manipulation. On the PhysTwin benchmark, our method improves reconstruction and prediction PSNR over PhysTwin by 10.4\% and 11.0\%, respectively. On the TUM RGB-D benchmark, it reduces average camera tracking error by 20.6\%, improves SSIM by 3.6\%, and reduces LPIPS by 8.3\% compared with 4DGS-SLAM.

Chinese Translation

为了从移动相机捕获的单目视频中重建物理世界的预测性四维表示，我们联合重建动态高斯辐射场并推断其潜在的物理动态。本文提出了 extit{物理SLAM}，一种物理感知的方法，用于构建可变形物体的四维高斯场，并在外部作用下预测其未来状态，基于一系列RGB-D图像。我们的方法将场景分离为用于相机定位的静态背景高斯场和由弹簧-质量物理模型驱动的动态物体高斯场，该模型考虑了弹性变形、惯性和阻尼。为了将视觉观测与物理动态相结合，我们引入了虚拟观测力，逐渐将模拟表面节点拉向观察到的深度点，避免了模拟节点位置的突然替换。同时，弹簧-质量模型的物理参数根据在整个重建观测窗口中聚合的模拟误差进行更新，使重建的动态与观测结果对齐。为了耦合外观、几何和物理，我们进一步设计了一种物理感知的高斯变换和表面对齐策略，其中动态高斯沿着弹簧-质量模型的变形物理表面进行变换，并在基于渲染的优化过程中被约束保持与表面对齐。最终的表示支持观察帧渲染和基于动作的模拟，扩展了SLAM地图以描述可变形物体如何响应操控。在PhysTwin基准测试中，我们的方法在重建和预测的PSNR上分别提高了10.4 ext{%}和11.0 ext{%}。在TUM RGB-D基准测试中，与4DGS-SLAM相比，平均相机跟踪误差降低了20.6 ext{%}，SSIM提高了3.6 ext{%}，LPIPS降低了8.3 ext{%}。

View on arXiv Download PDF AI Translation

cs.RO / 162 / 2610.17279

ReForesight: Occlusion-Robust Re-Identification and Future Occupancy Forecasting for Dynamic Semantic Mapping

ReForesight：针对动态语义映射的遮挡鲁棒性再识别与未来占用预测

Seyidov, Didar, Ogunwoye, Faith, Amjad, Hajira, Tolkachev, Danil, Sarkisov, Yuri, Tsetserukou, Dzmitry

Abstract

Robots sharing walkways need to know where people will be, yet semantic occupancy maps show only where they are. We present ReForesight, an open-source ROS 2 toolkit that adds instance-level future occupancy and re-identification after occlusion to a particle-based semantic mapping pipeline. A 36.6k-parameter MLP predicts each tracked person or vehicle up to 3 s ahead, guarded by a Kalman fallback and a stationary gate, and the object's own points are moved to the forecast pose, forming a "ghost" point cloud in the map frame without an occupancy decoder. An optional tracker re-associates objects after occlusion using keypoint descriptors the pipeline already computes and a metric 3D gate. On four held-out CODa sequences, the MLP lowers the mean 3 s final displacement error from 0.728 m (Kalman) to 0.662 m and beats a similar-size GRU (0.692 m) on every sequence, at 0.04 ms per instance on one CPU core. Code, all models, and a ROS-free demo are released; field demonstrations use an outdoor cleaning robot.

Chinese Translation

共享人行道的机器人需要知道人们将会在哪里，但语义占用地图仅显示他们当前的位置。我们提出了ReForesight，这是一个开源的ROS 2工具包，它在基于粒子的语义映射流程中增加了实例级的未来占用预测和遮挡后的再识别。一个具有36.6k参数的多层感知器（MLP）可以预测每个跟踪的人员或车辆在未来3秒内的位置，辅以卡尔曼滤波器和静态门的保护，并将物体自身的点移动到预测位置，形成一个在地图坐标系中没有占用解码器的“幽灵”点云。一个可选的跟踪器使用管道已经计算的关键点描述符和一个度量3D门在遮挡后重新关联物体。在四个保留的CODa序列上，MLP将平均3秒最终位移误差从0.728米（卡尔曼）降低到0.662米，并在每个序列中都超越了类似规模的GRU（0.692米），每个实例在一个CPU核心上耗时0.04毫秒。代码、所有模型和一个无ROS的演示已发布；现场演示使用了一台户外清扫机器人。

View on arXiv Download PDF AI Translation

cs.RO / 163 / 2610.17281

Phase-Conditioned Affordance Fields for SE(3)-Equivariant Visuomotor Policies

相位条件的可供性场用于 SE(3) 等变视觉运动策略

Chen, Xinzhe, Huang, Liqi, Ren, Sihua, Sun, Haowen, Li, Mingyang, Liu, Zeyang, Chen, Xingyu, Lan, Xuguang

Abstract

SE(3)-equivariance provides a useful geometric inductive bias for visuomotor policies, encoding that actions covary with rigid transformations of the observation. However, existing SE(3)-equivariant visuomotor policies build this bias on a global equivariance assumption that treats the entire observation as governed by a single rigid transformation, which breaks down in real-world manipulation tasks where the task-relevant region changes across different objects. We address this by learning a phase-conditioned affordance field (PAF) over the observed point cloud, a per-point score that the SE(3)-equivariant policy uses in two complementary ways. Its score-weighted centroid serves as a learned, phase-conditioned anchor that canonicalizes the observation and the action around the active region, replacing the fixed gripper-frame reference, and the score itself is appended as a per-point feature. This introduces a local geometric inductive bias, while preserving the policy's global SE(3) equivariance. We validate on multiple MimicGen tasks under varying spatial randomization, and PAF consistently improves the success rate of the underlying policy. On multi-object real-robot tasks, PAF outperforms baselines and keeps its lead under lighting variations.

Chinese Translation

SE(3) 等变性为视觉运动策略提供了一种有用的几何归纳偏置，编码了动作与观察的刚性变换之间的协变关系。然而，现有的 SE(3) 等变视觉运动策略基于一个全球等变假设，该假设将整个观察视为由单一刚性变换控制，这在实际操作任务中会失效，因为任务相关区域在不同物体之间会发生变化。我们通过在观察到的点云上学习一个相位条件的可供性场（PAF）来解决这一问题，该场为每个点提供一个分数，SE(3) 等变策略以两种互补的方式使用该分数。其分数加权的质心作为一个学习到的、相位条件的锚点，规范化观察和动作围绕活跃区域，取代了固定的抓取框架参考，而分数本身则作为每个点的特征附加。这引入了局部几何归纳偏置，同时保持了策略的全球 SE(3) 等变性。我们在多个 MimicGen 任务下进行验证，结果表明 PAF 一致地提高了基础策略的成功率。在多物体真实机器人任务中，PAF 超越了基线，并在光照变化下保持其优势。

View on arXiv Download PDF AI Translation

cs.RO / 164 / 2610.17298

Improving Policy Learning from Delayed Interventions

改善延迟干预下的政策学习

Zhu, Xiru, Meger, David, Dudek, Gregory

Abstract

In this work, we explore the idea of expert intervention delays in human-supervised robotic systems for imitation learning system. Previous works in imitation learning have primarily examined the impact of suboptimal experts, where the expert policy makes mistakes, but assumed expert interventions occur immediately after errors are made. Although delayed expert intervention can be viewed as a form of noise that might improve policy robustness, our results show that learning from delayed expert interventions is overall less effective than learning from non-delayed expert interventions; delayed expert interventions are less effective at demonstrating error avoidance behaviors. To address this, we propose Learning from Delayed Intervention (LfDI), which learns a cost function based on where and when the expert intervenes, combined with limited information about the extent of intervention delays provided by the expert. This cost function is then used to infer additional supervision for states without expert data, guiding the policy to avoid states that would likely trigger interventions. We evaluate our method, Learning from Delayed Intervention (LfDI), on simulated environments and show that it outperforms other methods that do not account for intervention delays.

Chinese Translation

在本研究中，我们探讨了在模仿学习系统中人类监督的机器人系统中专家干预延迟的概念。以往的模仿学习研究主要考察了次优专家的影响，即专家政策会犯错，但假设专家干预是在错误发生后立即进行的。尽管延迟的专家干预可以被视为一种噪声，可能提高政策的鲁棒性，但我们的结果表明，从延迟的专家干预中学习的效果总体上不如从非延迟的专家干预中学习的效果；延迟的专家干预在展示错误避免行为方面的效果较差。为了解决这个问题，我们提出了基于延迟干预学习（Learning from Delayed Intervention, LfDI），该方法根据专家干预的时间和地点学习成本函数，并结合专家提供的关于干预延迟程度的有限信息。然后，这个成本函数用于推断没有专家数据的状态的额外监督，引导政策避免可能触发干预的状态。我们在模拟环境中评估了我们的方法（LfDI），并显示其优于不考虑干预延迟的其他方法。

View on arXiv Download PDF AI Translation

cs.RO / 165 / 2610.17326

Evaluating Long-Horizon Robot Planning under Uncertainty

在不确定性下评估长时间范围的机器人规划

Muriki, Harsh, Aravindan, Gayathri, Zheng, Yuchen, Chernova, Sonia

Abstract

Household robots must successfully navigate im-perfect circumstances, both in their internal models and the external world. We introduce a benchmark of seven long-horizon household tasks in Habitat that vary initial robot memory and environment configurations. We evaluate adaptations of ReAct, VLM-TAMP, and Tru-POMDP within a shared execution framework across 90 tasks-environment-memory combinations, totaling 1,350 episodes. Our evaluation measures task success, partial completion, and action efficiency, along with rule-based analysis of execution traces. ReAct achieves the highest overall task success (45.3%) and completion (71.7%), while having accurate initial memory does not consistently yield the highest performance for every planner. We identify recurring placement errors, precondition errors, and redundant actions through trace analysis. These findings highlight differences in how the evaluated planners use initial knowledge and respond to uncertainty.

Chinese Translation

家庭机器人必须在其内部模型和外部世界中成功应对不完美的情况。我们在Habitat中引入了七个长时间范围的家庭任务基准，这些任务在初始机器人记忆和环境配置上有所不同。我们在一个共享执行框架下评估了ReAct、VLM-TAMP和Tru-POMDP的适应性，共涉及90个任务-环境-记忆组合，总计1,350个实验回合。我们的评估指标包括任务成功率、部分完成率和行动效率，以及对执行轨迹的基于规则的分析。ReAct实现了最高的总体任务成功率（45.3%）和完成率（71.7%），而准确的初始记忆并不总是为每个规划器带来最高的性能。通过轨迹分析，我们识别出反复出现的放置错误、前置条件错误和冗余动作。这些发现突显了被评估的规划器在使用初始知识和应对不确定性方面的差异。

View on arXiv Download PDF AI Translation

cs.RO / 166 / 2610.17350

Can a Frontier VLM Serve as a Robot Policy for Memory-Dependent Manipulation?

前沿视觉语言模型能否作为记忆依赖性操作的机器人策略？

Jedlicki, Jack B., Sethi, Amish, Yang, Heng

Abstract

We use a frozen frontier vision-language model, GPT-6-Astra, directly as the robot policy on RoboMME, a benchmark of 16 memory-dependent manipulation tasks spanning temporal, spatial, object, and procedural memory. The model keeps the whole episode in context and outputs every end-effector waypoint and gripper command, with no RoboMME-specific training. On the 800 test episodes it succeeds in 92.1%, above the previously reported human score of 90.5%, which was obtained with an oracle skill interface, and far ahead of the best leaderboard entry (60.8%) and of a hierarchical system that pairs the same model with a RoboMME-trained policy (79.1%). It matches or exceeds the human score on 13 of the 16 tasks. Its 63 failures concentrate in two tasks and are mostly premature completions and mistimed or unintended button presses. These results blur the usual division between long-horizon planning, memory and end-effector control, and raise the question of how much of this stack a single general-purpose model can take over in embodied systems. Project website: https://embodied-astra.github.io. Code, prompts and all episode logs: https://github.com/jackbjed/embodied-astra-robomme.

Chinese Translation

我们直接使用一个冻结的前沿视觉语言模型GPT-6-Astra作为RoboMME上的机器人策略，RoboMME是一个涵盖时间、空间、物体和程序记忆的16个记忆依赖性操作任务的基准。该模型在上下文中保持整个情节，并输出每个末端执行器的路径点和夹持器命令，且无需针对RoboMME进行特定训练。在800个测试情节中，其成功率为92.1%，高于先前报告的人类得分90.5%（该得分是通过一个神谕技能接口获得的），远超最佳排行榜条目（60.8%）以及将同一模型与RoboMME训练策略配对的分层系统（79.1%）。在16个任务中，有13个任务的得分与人类得分相匹配或超过。其63次失败主要集中在两个任务中，且大多数为过早完成和时机不当或意外的按钮按压。这些结果模糊了长时间规划、记忆和末端执行器控制之间的通常划分，并提出了一个问题：在具身系统中，一个通用模型能够接管多少这一堆栈的功能。项目网站：https://embodied-astra.github.io。代码、提示和所有情节日志：https://github.com/jackbjed/embodied-astra-robomme。

View on arXiv Download PDF AI Translation

cs.RO / 167 / 2610.17419

ClearNav: Perception-Aligned Quadrotor Flight Using Differential-Physics

ClearNav：基于差异物理的感知对齐四旋翼飞行

Azad, Aditya, Huang, Arthur, Li, Guanrui

Abstract

Autonomous quadrotor flight in cluttered environment is an key capability for tasks like inspection and search. Navigation policies trained in simulation and transferred to hardware can fly at high speed in open space, but they can still collide as obstacle clearance shrinks. We attribute most of these collisions to a single mechanism, namely motion in directions the onboard camera does not observe. Heading that is not aligned with the direction of travel and lateral excursion during obstacle avoidance both carry the vehicle toward obstacles outside its field of view. We introduce ClearNav, a navigation policy trained with two reward terms that constrain motion to the observed direction, a forward-looking heading target and a penalty on lateral velocity. A third term penalizes speed near obstacles so that an obstacle entering the field of view remains avoidable. In simulation, across 18 held-out scenarios with 2304 paired flights per policy unseen in training, ClearNav improves success by 6.8 points, cuts the collision rate by 65% relative to the state of the arts, and flies at maximum 5m/s. For the proximity term alone under matched commanded speeds, the collision reduction still holds at 70%. The effect is significant on static scenes and consistent in direction on moving obstacles, though not significant there at the suite level, showing that safer flight in clutter is possible even on the limited sensors and compute of small drones. We also validate our navigation policy in a real-world cluttered environment, showing that it enables the quadrotor to navigate through the cluttered environment successfully and safely.

Chinese Translation

在复杂环境中自主四旋翼飞行是执行检查和搜索等任务的关键能力。在模拟中训练并转移到硬件的导航策略可以在开阔空间中以高速飞行，但在障碍物间隙缩小时仍可能发生碰撞。我们将大多数碰撞归因于一个单一机制，即在机载摄像头未观察到的方向上的运动。航向与行进方向不一致以及在避障过程中横向偏移都会使飞行器朝向其视野之外的障碍物移动。我们提出了ClearNav，这是一种通过两个奖励项训练的导航策略，约束运动朝向观察到的方向，包括一个前视航向目标和对横向速度的惩罚。第三个项对靠近障碍物的速度进行惩罚，以确保进入视野的障碍物仍然可以被避开。在模拟中，在18个未在训练中见过的持出场景中，每个策略有2304次配对飞行，ClearNav的成功率提高了6.8个百分点，碰撞率相较于现有技术降低了65%，最大飞行速度为5m/s。仅在匹配指令速度下，接近项的碰撞减少仍保持在70%。在静态场景中效果显著，对移动障碍物的方向一致性也表现良好，尽管在整体测试中未显著，但显示出即使在小型无人机有限的传感器和计算能力下，在复杂环境中实现更安全的飞行是可能的。我们还在真实的复杂环境中验证了我们的导航策略，显示其能够成功且安全地引导四旋翼穿越复杂环境。

View on arXiv Download PDF AI Translation

cs.RO / 168 / 2610.17422

MIRROR: Reflecting Hidden Symmetries from Offline Robot Interaction Data

MIRROR：从离线机器人交互数据中反映隐藏的对称性

Liu, Shiqi, Yao, Yihang, Huang, Peide, Xu, Mengdi, Zhao, Ding

Abstract

Symmetry-aware robot policy learning requires coupled observation-action transformations. For pixel observations, even simple physical symmetries can induce complex, viewpoint-dependent image transformations that are difficult to specify by hand. Existing symmetry-aware methods often rely on representations with predefined transformations, requiring preprocessing tailored to the sensing setup and task. We introduce MIRROR, a framework that learns how a known symmetry group operates jointly on observations and actions in partially observable Markov decision processes (POMDPs) from offline interaction data. MIRROR couples these transformations through adversarial tuple matching over one-step transition tuples, while explicit group-consistency regularization penalizes violations of the group-operation axioms. On four Robosuite tabletop manipulation tasks, MIRROR achieves the lowest aggregate error on three of four symmetry-quality metrics, including a 61% reduction in compatibility error (which measures consistency under transformation composition) relative to an adapted LaLiGAN baseline. Using the learned transformations in frame-averaged behavior cloning (BC) yields higher mean success rates than vanilla BC on all four tasks, increasing average success from 80.48% to 85.12%, with per-task gains up to 9.66 percentage points. We also repeat symmetry discovery and policy training in simulation with heightmap and segmentation-mask observations, then deploy the resulting policy zero-shot on a real Kinova GEN3 arm. Project page: https://pomdp-mirror.github.io.

Chinese Translation

对称感知的机器人策略学习需要耦合的观察-动作转换。对于像素观察，即使是简单的物理对称性也会引发复杂的、视角依赖的图像转换，这些转换难以手动指定。现有的对称感知方法通常依赖于具有预定义转换的表示，要求针对传感设置和任务进行定制的预处理。我们提出了MIRROR，一个框架，它从离线交互数据中学习已知对称群在部分可观察马尔可夫决策过程（POMDPs）中的观察和动作上的联合操作。MIRROR通过对一步转移元组进行对抗性元组匹配来耦合这些转换，同时显式的群一致性正则化惩罚对群操作公理的违反。在四个Robosuite桌面操作任务中，MIRROR在四个对称质量指标中的三个上实现了最低的总误差，包括相对于调整后的LaLiGAN基线在兼容性误差（衡量转换组合下的一致性）上减少了61%。在帧平均行为克隆（BC）中使用学习到的转换，在所有四个任务上都比普通BC获得了更高的平均成功率，将平均成功率从80.48%提高到85.12%，每个任务的增益最高达到9.66个百分点。我们还在模拟中重复对称发现和策略训练，使用高度图和分割掩码观察，然后将得到的策略零-shot部署在真实的Kinova GEN3臂上。项目页面：https://pomdp-mirror.github.io。

View on arXiv Download PDF AI Translation

cs.RO / 169 / 2610.17427

Smoothness Ranks Skill, Not Success: An Audit of Published Trajectory-Smoothness Curation Metrics Under Episode-Length Controls

平滑性排名技能，而非成功：对在情节长度控制下发布的轨迹平滑性策展指标的审计

Bassi, Jaiveer

Abstract

Trajectory smoothness has been proposed as a cheap, training-free signal for curating imitation-learning demonstrations, most recently by rinse, which introduces Spectral Arc Length (SAL) and a contact-aware Trajectory-Envelope Distance (TED). We test whether these published metrics survive episode-length controls on real robomimic data. On can/mh, SAL separates better from worse operators with AUROC 0.952, but episode length alone achieves 0.957, and inside a duration-matched band SAL falls to 0.628 while raw mean jerk, the baseline rinse argues against, rises to 0.896. The confound replicates on square/mh. On can/mg, where every episode is exactly 150 steps, SAL is at chance (0.454) and TED is inverted (0.376) against task success, while jerk reaches 0.706. Training GMM behavior-cloning policies on each metric's curated subset (10 seeds), curating by TED costs roughly a third of downstream success relative to random selection of the same size (0.240 vs. 0.360, p=0.0001, surviving Bonferroni correction), while SAL and jerk are indistinguishable from not curating. Under a diffusion policy TED still ranks worst, though its deficit is not resolvable at five seeds, and jerk becomes significantly better than both baselines, so the curation decision does not transfer unchanged across the two policy configurations we tested. Inspecting trajectories shows why TED inverts: successful manipulation requires sustained high-deviation motion through contact, while failure is low-amplitude hovering that a smoothed envelope follows closely.

Chinese Translation

轨迹平滑性被提出作为一种廉价的、无需训练的信号，用于策划模仿学习示范，最近由 rinse 引入，提出了谱弧长（Spectral Arc Length, SAL）和一种考虑接触的轨迹包络距离（Trajectory-Envelope Distance, TED）。我们测试这些已发布的指标在真实的 robomimic 数据上是否能在情节长度控制下生存。在 can/mh 数据集上，SAL 能够以 0.952 的 AUROC 更好地区分优秀与劣质操作员，但仅凭情节长度就能达到 0.957，而在持续时间匹配的范围内，SAL 降至 0.628，而原始平均冲击（mean jerk），即 rinse 反对的基线，升至 0.896。这个混淆在 square/mh 数据集上也得到了复制。在 can/mg 数据集上，每个情节恰好为 150 步，SAL 的表现处于随机水平（0.454），而 TED 对任务成功的表现则呈现反转（0.376），而冲击达到了 0.706。在每个指标策划的子集上（10 个种子）训练 GMM 行为克隆策略，使用 TED 策划的成本大约是随机选择相同大小的下游成功的三分之一（0.240 对比 0.360，p=0.0001，经过 Bonferroni 校正仍然有效），而 SAL 和冲击的表现与不策划无异。在扩散策略下，TED 仍然排名最差，尽管其劣势在五个种子下无法解决，而冲击的表现显著优于两个基线，因此策划决策在我们测试的两种策略配置中并未保持不变。检查轨迹显示了 TED 反转的原因：成功的操作需要通过接触保持高偏差运动，而失败则是低幅度悬停，平滑的包络紧随其后。

View on arXiv Download PDF AI Translation

cs.RO / 170 / 2610.17462

Act4R: Action-conditioned 4D reconstruction for kinematics-free control

Act4R：无运动学控制的动作条件4D重建

Zhang, Zhaoliang, Yu, Chenhao, Shao, Zhimin, Liu, Xijun, Peng, Cheng

Abstract

Precise control of movement across various robotic embodiments is a longstanding challenge. Modeling complex robotic structures requires significant time, dedicated sensors, and manual labeling. Recent advancements in 4D representation learning offer an interesting alternative: can we leverage 4D features lifted from 2D observations to densely map a robot's spatial movement based on the action signal? To this end, we propose Act4R, a framework that builds on 3D foundation model to predict a robot's 3D scene flow with any given action. Act4R introduces a Spatial-Action Encoder, which encodes the current 2D robot state and its action signals into embeddings that represent future robot state in 3D. Then, both the current and future spatial embeddings are decoded into point maps to estimate action-conditioned scene flow. To effectively learn the mapping between action and future state, Act4R introduces spatial-action fusion in its transformer-based encoder, and a spatial alignment loss that guides the encoder for future flow prediction. After training on some samples of a given embodiment, Act4R can perform inverse control by optimizing the signal to achieve the desired scene flow. We evaluate Act4R on a public real-world dataset and synthetic data collected in simulator. Experiments on various robotic platforms show that Act4R outperforms the previous SoTA in scene flow prediction. Our method also performs better in inverse control, both on the action accuracy metric and in closed-loop control in simulation.

Chinese Translation

在各种机器人形态中精确控制运动一直是一个长期挑战。建模复杂的机器人结构需要大量时间、专用传感器和手动标注。最近在4D表示学习方面的进展提供了一种有趣的替代方案：我们能否利用从2D观察中提取的4D特征，根据动作信号密集映射机器人的空间运动？为此，我们提出了Act4R，一个基于3D基础模型的框架，用于预测机器人在给定动作下的3D场景流。Act4R引入了一种空间-动作编码器，它将当前的2D机器人状态及其动作信号编码为表示未来机器人状态的嵌入。然后，当前和未来的空间嵌入被解码为点图，以估计动作条件的场景流。为了有效学习动作与未来状态之间的映射，Act4R在其基于变换器的编码器中引入了空间-动作融合，并采用空间对齐损失来指导编码器进行未来流的预测。在对给定形态的一些样本进行训练后，Act4R可以通过优化信号来实现逆控制，以达到期望的场景流。我们在一个公共真实世界数据集和在模拟器中收集的合成数据上评估了Act4R。在各种机器人平台上的实验表明，Act4R在场景流预测方面优于之前的最先进技术（SoTA）。我们的算法在逆控制方面也表现更佳，无论是在动作准确性指标上，还是在模拟中的闭环控制中。

View on arXiv Download PDF AI Translation

cs.RO / 171 / 2610.17468

From Orbit Matching to Docking: A Hierarchical Reinforcement-Learning Approach to Multiphase Rendezvous Trajectory Design

从轨道匹配到对接：一种层次化强化学习方法用于多相会合轨迹设计

Storey-Matsutani, Mariko, Zlotnik, Anatoly, Holguin, Lee, Hampel-Arias, Zigfried, Linares, Richard

Abstract

As the complexity of space missions increases, autonomous spacecraft guidance must accommodate transitions between mission phases with distinct dynamics, objectives, and constraints. Rendezvous and proximity operations (RPO) are particularly challenging because they span multiple operating regimes, from orbital-scale matching and far-range approach to close-range maneuvering and terminal docking. Existing reinforcement-learning approaches generally address only a subset of this sequence or rely on manually defined switching conditions between controllers. This study presents a hierarchical reinforcement-learning framework for autonomous, three-degree-of-freedom RPO that spans the complete trajectory from orbit matching to docking. Five specialized policies are trained for orbit matching, far-range approach, mid-range approach, close-range approach, and docking. A learned high-level manager policy then selects among these phase-specific controllers based on the evolving spacecraft state. The hierarchical approach is evaluated over the full end-to-end rendezvous sequence and compared with a fixed sequential-invocation heuristic to assess trajectory performance and transition behavior. Training and inference metrics are also analyzed for the individual phase policies and the complete hierarchical system.

Chinese Translation

随着太空任务复杂性的增加，自主航天器的引导必须适应具有不同动力学、目标和约束的任务阶段之间的转换。会合与接近操作（RPO）尤其具有挑战性，因为它们跨越多个操作模式，从轨道级匹配和远程接近到近距离机动和终端对接。现有的强化学习方法通常仅解决这一序列的一个子集，或依赖于手动定义的控制器之间的切换条件。本研究提出了一种层次化强化学习框架，用于自主的三自由度RPO，涵盖从轨道匹配到对接的完整轨迹。为轨道匹配、远程接近、中程接近、近程接近和对接训练了五个专门的策略。然后，学习到的高层管理策略根据不断变化的航天器状态在这些特定阶段的控制器中进行选择。该层次化方法在完整的端到端会合序列上进行了评估，并与固定顺序调用启发式方法进行了比较，以评估轨迹性能和转换行为。同时，还分析了各个阶段策略和完整层次系统的训练和推理指标。

View on arXiv Download PDF AI Translation

cs.RO / 172 / 2610.17511

An Optimal Algorithm for Changing from Longitudinal to Latitudinal Formation of Autonomous Aircraft Squadrons

自主飞行器编队从纵向到横向转换的最优算法

Giacomin, Paulo André Sperandio, Hemerly, Elder Moreira

Abstract

This work presents an algorithm for changing from longitudinal to latitudinal formation of autonomous aircraft squadrons. The maneuvers are defined dynamically by using a predefined set of 3D basic maneuvers. This formation change is necessary when the squadron has to perform tasks demanding both formations, such as landing, lift-off, obstacle avoidance and georeferencing. The time complexity analysis of the transformation algorithm reveals that its efficiency is optimal. Simulations show that the formation change is done without collision, and the proof of correctness ensures its latitudinal formation features.

Chinese Translation

本研究提出了一种将自主飞行器编队从纵向转换为横向的算法。该机动通过使用预定义的三维基本机动集动态定义。当编队需要执行同时要求两种编队的任务时，例如着陆、起飞、避障和地理参考时，这种编队转换是必要的。对转换算法的时间复杂度分析表明其效率是最优的。仿真结果显示编队转换在没有碰撞的情况下完成，并且正确性证明确保了其横向编队特性。

View on arXiv Download PDF AI Translation

cs.RO / 173 / 2610.17521

YUBI-Real2Sim: 3DGS Scene Verification for Reliable Policy Evaluation

YUBI-Real2Sim：用于可靠策略评估的3DGS场景验证

Nishizawa, Hiroki, Ohkawa, Takehiko, Scianca, Nicola, Matsushima, Tatsuya, Morishima, Shigeo, Ota, Kei

Abstract

Evaluating VLA policies on real robots is costly, time-consuming, and difficult to reproduce, making real-to-sim simulation essential for scalable evaluation. While 3D Gaussian Splatting (3DGS) enables rapid real-to-sim scene generation, the reconstruction quality required for reliable policy evaluation remains unquantified. In particular, policies relying on wrist-mounted fisheye cameras (e.g., YUBI handheld gripper data) are heavily exposed to visual artifacts (holes, floaters) in wide-angle views, directly degrading task performance. To address this, we present YUBI-Real2Sim, a unified framework for 3DGS-based scene reconstruction and real-to-sim policy evaluation under dual fisheye wrist views. We develop an integrated pipeline combining 200-degree fisheye camera capture, 3DGS training, 3D foundation models for object modeling, and MuJoCo-based physical simulation, reproducing the observed real-world policy ordering in eight of nine scene-task comparisons. Using this framework, we construct a benchmark across two bimanual robot platforms, three scenes, three tasks, and dual pi0.5-based policies. This systematically measures real-to-sim rank agreement and evaluates policy sensitivity against counterfactual 3DGS artifacts (ceiling holes, floaters, warping) and capture disturbances. We further analyze whether novel-view quality scores can indicate policy-ranking reliability under a scene-held-out evaluation protocol. The results show task-dependent performance and limitations of quality-based reliability prediction, highlighting the need to assess reconstruction quality when using Real2Sim for policy evaluation. These results do not yet establish a generally reliable scene-acceptance rule. Project Page: https://yubi-real2sim.airoa.io.

Chinese Translation

在真实机器人上评估VLA策略既昂贵又耗时，且难以复制，这使得真实到仿真的模拟对于可扩展评估至关重要。虽然3D高斯点云（3DGS）能够快速生成真实到仿真的场景，但用于可靠策略评估所需的重建质量仍未量化。特别是，依赖于手腕安装鱼眼相机的策略（例如，YUBI手持抓取器数据）在广角视图中严重暴露于视觉伪影（孔洞、漂浮物），直接降低了任务性能。为了解决这一问题，我们提出了YUBI-Real2Sim，这是一个基于3DGS的场景重建和在双鱼眼手腕视图下进行真实到仿真策略评估的统一框架。我们开发了一个集成管道，结合了200度鱼眼相机捕获、3DGS训练、用于物体建模的3D基础模型以及基于MuJoCo的物理模拟，在九个场景任务比较中重现了观察到的真实世界策略排序。利用该框架，我们在两个双手机器人平台、三个场景、三个任务和双pi0.5基础策略之间构建了一个基准。这系统性地测量了真实到仿真的排名一致性，并评估了策略对反事实3DGS伪影（天花板孔、漂浮物、变形）和捕获干扰的敏感性。我们进一步分析了新视角质量评分是否可以指示在场景保留评估协议下的策略排名可靠性。结果显示任务依赖的性能和基于质量的可靠性预测的局限性，强调在使用Real2Sim进行策略评估时评估重建质量的必要性。这些结果尚未建立普遍可靠的场景接受规则。项目页面：https://yubi-real2sim.airoa.io。

View on arXiv Download PDF AI Translation

cs.RO / 174 / 2610.17526

Box-WAM: Box Trajectories as Compact State Representations for World Action Models

Box-WAM：将盒子轨迹作为世界行动模型的紧凑状态表示

Yu, Qinxi, Ding, Wenzhou, Wang, Gloria X., Zhai, Albert J., Wang, Shenlong

Abstract

World action models (WAMs) are a category of policy models that ex- plicitly predict both actions and future world states. They have been shown to outperform action-only models across various robotic manipulation tasks. How- ever, representing the future with full-frame video prediction can be computation- ally expensive during both training and inference. This raises a central question: what compact state representation preserves the planning and supervision bene- fits of future prediction? In this paper, we propose BoxWAM, which replaces dense future frames with oriented bounding box trajectories of task-relevant ob- jects. By explicitly imagining object-centric future motion, our method provides a lightweight prediction target that captures the key physical changes induced by robot actions. We show that box trajectory prediction substantially reduces com- putational cost while improving action prediction performance, suggesting that object-centric compact states can serve as an efficient alternative to video-based world action modeling for robotic manipulation. Result videos can be found on https://box-wam.github.io

Chinese Translation

世界行动模型（WAMs）是一类政策模型，明确预测动作和未来世界状态。研究表明，它们在各种机器人操作任务中优于仅基于动作的模型。然而，使用全帧视频预测来表示未来在训练和推理过程中可能会消耗大量计算资源。这引出了一个核心问题：什么样的紧凑状态表示能够保留未来预测的规划和监督优势？在本文中，我们提出了BoxWAM，它用与任务相关对象的定向边界框轨迹替代了密集的未来帧。通过明确想象以对象为中心的未来运动，我们的方法提供了一种轻量级的预测目标，捕捉机器人动作引起的关键物理变化。我们展示了盒子轨迹预测显著降低了计算成本，同时提高了动作预测性能，表明以对象为中心的紧凑状态可以作为机器人操作中视频基础世界行动建模的高效替代方案。结果视频可以在 https://box-wam.github.io 找到。

View on arXiv Download PDF AI Translation

cs.RO / 175 / 2610.17534

Sim4RAD: A MATLAB Toolkit for Design-Phase Sizing, Actuator Selection and URDF-Model Verification of Serial Robotic Arms

Sim4RAD：用于串联机器人臂设计阶段尺寸计算、执行器选择和URDF模型验证的MATLAB工具包

Li, Jingchen

Abstract

Actuator sizing is an early and consequential decision: joint torque and velocity specifications fix the motors, gearboxes and brakes before any CAD model exists. Sim4RAD is a MATLAB toolkit for that phase and for the verification that follows it. It builds a rigid-body model from a handful of parameters, and four analyses supply the specification: a static analysis combining lever-arm sampling with a posture search, giving a rated torque at a chosen payload envelope and a reproducible peak torque; a dynamics pipeline that reshapes a trajectory under one of four constraints, including the actuator torque-speed envelope; a stochastic analysis over many random motions, so a specification rests on a class of motions, not one trajectory; and a workspace and dexterity analysis that separates where the arm can reach from how it can reach there. Replacing the synthesized model by a URDF file turns the same analyses to verification. A 7-DOF anthropomorphic arm is compared in three layouts differing in the shoulder axis order and in whether the last wrist joint is coaxial with the payload. The static demand at half arm length is nearly the same in all three, and it sizes the joints rather than the sampled motion; reordering the shoulder axes leaves the worst-case torque unchanged at 55 N m and the reachable volume within 8 percent. A last wrist joint coaxial with the payload removes its gravity torque, reduces the adjacent wrist joint by almost a third and cuts the last joint demand tenfold in a 5 m/s scan, at the price of dexterity (mean 0.570 to 0.348). The same arm needs 20 to 37 times more torque for a 5 m/s target than for a 2 s motion: the scenario, not the trajectory, must be fixed first. Verification should use the URDF model of the detailed design; the toolkit is available at https://github.com/LasterLi/Sim4RAD.

Chinese Translation

执行器尺寸计算是一个早期且重要的决策：关节扭矩和速度规格在任何CAD模型存在之前就已确定了电机、齿轮箱和制动器。Sim4RAD是一个用于该阶段及其后续验证的MATLAB工具包。它根据少量参数构建刚体模型，并通过四种分析提供规格：静态分析结合了杠杆臂采样与姿态搜索，给出了在选定负载包络下的额定扭矩和可重复的峰值扭矩；动态管道在四种约束之一下重塑轨迹，包括执行器扭矩-速度包络；对许多随机运动进行的随机分析，使得规格基于一类运动，而不是单一轨迹；以及工作空间和灵活性分析，将手臂的可达区域与其到达方式分开。用URDF文件替换合成模型使得相同的分析可以用于验证。对一个7自由度的人形臂在三种不同的布局中进行比较，这些布局在肩部轴顺序和最后一个腕关节是否与负载同轴方面有所不同。在手臂长度的一半处，三者的静态需求几乎相同，且其尺寸是基于关节而非采样运动；重新排列肩部轴顺序使得最坏情况下的扭矩保持不变，为55 N·m，并且可达体积在8%以内。同轴于负载的最后一个腕关节消除了其重力扭矩，使得相邻腕关节的需求减少近三分之一，并在5 m/s扫描中将最后一个关节的需求降低十倍，但代价是灵活性降低（平均值从0.570降至0.348）。同样的手臂在5 m/s目标下所需的扭矩是2 s运动的20到37倍：必须首先固定场景，而不是轨迹。验证应使用详细设计的URDF模型；该工具包可在https://github.com/LasterLi/Sim4RAD获取。

View on arXiv Download PDF AI Translation

cs.RO / 176 / 2610.17555

ATOM: Abstract Transition and Option Modeling - Experience-Informed Hierarchical Navigation

ATOM：抽象转移与选项建模 - 基于经验的分层导航

Yao, Changxuan, Tsiotras, Panagiotis

Abstract

We propose ATOM (Abstract Transition and Option Modeling), a hierarchical navigation framework that combines higher-level graph planning with reusable lower-level option policies for efficient repeated navigation in large-scale 3D environments represented by hierarchical structures such as 3D scene graphs (3DSGs). ATOM operates on a hierarchical graph representation, implemented using, for instance, the Rooms and Places layers of a 3DSG, where each directed abstract transition is associated with a navigation option. For each transition, ATOM computes and stores a reusable option policy along with its state-dependent cost-to-go. Across repeated navigation tasks, ATOM maintains a smoothed empirical distribution over option initiation states and uses it to progressively refine the corresponding abstract transition cost. Ablation studies show that ATOM reduces cumulative repeated-planning time by 97.5% relative to a monolithic MDP baseline while maintaining trajectories within approximately 7% of a Global Places A* shortest-path reference. Experience-informed refinement further reduces abstract transition-cost estimation error by 10.3%. Physical experiments validate ATOM on an F1Tenth vehicle, successfully completing all five navigation queries with 43% of inter-region option executions reusing previously generated solutions.

Chinese Translation

我们提出了ATOM（抽象转移与选项建模），这是一个分层导航框架，结合了高层次图规划与可重用的低层次选项策略，以实现对大型3D环境的高效重复导航，这些环境通过层次结构如3D场景图（3DSG）表示。ATOM在层次图表示上运行，例如使用3DSG的房间和地点层，其中每个有向抽象转移都与一个导航选项相关联。对于每个转移，ATOM计算并存储一个可重用的选项策略及其状态依赖的成本。通过重复导航任务，ATOM维持一个平滑的选项启动状态的经验分布，并利用它逐步优化相应的抽象转移成本。消融研究表明，ATOM相对于单一MDP基线将累计重复规划时间减少了97.5%，同时保持轨迹与全球地点A*最短路径参考的偏差约为7%。基于经验的优化进一步将抽象转移成本估计误差降低了10.3%。物理实验在F1Tenth车辆上验证了ATOM，成功完成了所有五个导航查询，其中43%的区域间选项执行重用了先前生成的解决方案。

View on arXiv Download PDF AI Translation

cs.RO / 177 / 2610.17562

Mobility-Budget-Constrained Submodular Maximization for Distributed Semantic Multi-Robot Exploration

受移动预算约束的子模最大化用于分布式语义多机器人探索

Yao, Changxuan, Garimella, Sandilya Sai, Gan, Lu, Tsiotras, Panagiotis

Abstract

Multi-robot exploration under communication and mobility constraints requires balancing information gain with reachability. We present a distributed semantic exploration framework that combines Dirichlet mapping, budget-constrained trajectory selection, and Resource-Aware distributed Greedy (RAG) coordination. Each robot clusters frontiers, computes terrain- and heading-aware least-cost paths with a single budget-limited Dijkstra search, and selects feasible paths by marginal information gain, represented by the expected reduction in Dirichlet entropy. Each robot's per-round selection is exact, the team's joint per-round information gain is at least half the centralized optimum, and the bound degrades gracefully under limited communication. Ten-run gridworld ablations show improved coverage, semantic mIoU, and map certainty with lower entropy and traversal cost per unit distance than shortest-distance frontier selection. Budget-sensitivity experiments show that an intermediate budget best balances path reach and receding-horizon execution. In photorealistic full-system simulation against five coordination rules run in their native published forms, the proposed method attains the highest mean coverage and semantic accuracy while being the only one that respects the per-trajectory mobility budget.

Chinese Translation

在通信和移动约束下的多机器人探索需要在信息增益与可达性之间取得平衡。我们提出了一种分布式语义探索框架，该框架结合了Dirichlet映射、预算约束的轨迹选择和资源感知的分布式贪婪（Resource-Aware Greedy, RAG）协调。每个机器人对前沿进行聚类，通过单一预算限制的Dijkstra搜索计算地形和朝向感知的最小成本路径，并通过边际信息增益选择可行路径，该增益由Dirichlet熵的预期减少表示。每个机器人的每轮选择是精确的，团队的联合每轮信息增益至少是集中最优的一半，并且在有限通信下界限优雅下降。十次运行的网格世界消融实验显示，与最短距离前沿选择相比，所提出的方法在覆盖率、语义mIoU和地图确定性方面表现更佳，同时每单位距离的熵和遍历成本更低。预算敏感性实验表明，中等预算最佳平衡路径可达性和后退视野执行。在与五种协调规则进行的光真实全系统仿真中，所提出的方法在平均覆盖率和语义准确性方面达到了最高，同时也是唯一一个遵循每条轨迹移动预算的方法。

View on arXiv Download PDF AI Translation

cs.RO / 178 / 2610.17572

EgoCPT: Learning Dexterous Manipulation from a Single Ego-Centric Video with Contact-Aware Optimization

EgoCPT：从单个自我中心视频中学习灵巧操作的接触感知优化

Sun, Zhixin, Shi, Kairui, Ren, Tianao, Yang, Ming, Cutkosky, Mark, Xu, James Jingxi

Abstract

While expert demonstrations on real robots are difficult to collect, human ego-centric videos are easy to collect and abundant on the internet. However, learning dexterous manipulation policies from human videos remains challenging---one of the most important challenges being the difficulty of reconstructing physically grounded robot trajectories due to the large embodiment gap. In this paper, we propose EgoCPT, a novel paradigm for learning high-quality retargeted dexterous hand trajectories directly from an ego-centric video, which can then be deployed directly on real robots. The only inputs to our method are an ego-centric RGB video and a reference hand pose trajectory, which can be easily obtained through off-the-shelf software or specialized cameras during recording. Our method is particularly robust to noisy hand poses attributed to our novel contact-aware optimization framework, which iteratively optimizes the contact points between the robot hands and objects. Across five manipulation tasks, our method achieves overall success rates of 67.3% on noisy EgoDex hand-pose reconstructions and 92.0% on high-quality TACO reconstructions collected using motion-capture hardware, outperforming multiple baselines on both benchmarks.

Chinese Translation

尽管在真实机器人上收集专家演示是困难的，但人类自我中心视频的收集相对容易且在互联网上丰富。然而，从人类视频中学习灵巧操作策略仍然面临挑战——其中一个最重要的挑战是由于巨大的体现差距，重建物理基础的机器人轨迹的难度。在本文中，我们提出了EgoCPT，一种新颖的范式，用于直接从自我中心视频中学习高质量的重定向灵巧手轨迹，这些轨迹可以直接部署在真实机器人上。我们方法的唯一输入是一个自我中心的RGB视频和一个参考手势轨迹，这可以通过现成的软件或在录制过程中使用专用摄像头轻松获得。由于我们新颖的接触感知优化框架，我们的方法对噪声手势具有特别的鲁棒性，该框架迭代优化机器人手与物体之间的接触点。在五个操作任务中，我们的方法在噪声EgoDex手势重建上实现了67.3%的整体成功率，在使用动作捕捉硬件收集的高质量TACO重建上实现了92.0%的成功率，超越了多个基准的多个基线。

View on arXiv Download PDF AI Translation

cs.RO / 179 / 2610.17574

TacSpin: Tactile Tracking of In-Gripper Object Rotation on SO(3) via Axis-Magnitude Estimation

TacSpin：通过轴-幅度估计在SO(3)上进行夹持器内物体旋转的触觉跟踪

Wu, Yuhao, Song, Linhao, Yin, Huan

Abstract

Tracking object motion within a gripper provides essential feedback for robotic manipulation. However, limited contact coverage and uneven deformation make tactile estimation challenging. In this study, we present TacSpin, a method that tracks in-gripper object rotation on SO(3) using only two vision-based tactile sensors. The key idea is to estimate the axis and magnitude of the rotation response separately, rather than fitting a rigid-body model by least squares. Each finger's response is further refined with contact-boundary cues. The two fingers are then fused according to their agreement. On standard objects, TacSpin achieves lower mean checkpoint errors. Qualitative demonstrations on real-world objects, using calibration transferred from a standard sphere, illustrate tracking under different contact geometries.

Chinese Translation

在夹持器内跟踪物体运动为机器人操作提供了重要反馈。然而，有限的接触覆盖和不均匀的变形使得触觉估计变得具有挑战性。在本研究中，我们提出了TacSpin，一种仅使用两个基于视觉的触觉传感器来跟踪夹持器内物体在SO(3)上的旋转的方法。其关键思想是分别估计旋转响应的轴和幅度，而不是通过最小二乘法拟合刚体模型。每个手指的响应进一步通过接触边界线索进行细化。然后根据它们的一致性融合两个手指的响应。在标准物体上，TacSpin实现了更低的平均检查点误差。使用从标准球体转移的标定对真实物体进行的定性演示，展示了在不同接触几何下的跟踪效果。

View on arXiv Download PDF AI Translation

cs.RO / 180 / 2610.17593

Training Generalist Navigation Policies via Action Self-Distillation

通过动作自蒸馏训练通用导航策略

Feng, Bowen, Wang, Keqin, Yang, Yulong, Joshi, Amogh, Allen-Blanchette, Christine, Shah, Dhruv, Heide, Felix

Abstract

Language provides a natural interface for humans to specify and reason about goals, but training language-conditioned robotic policies and reasoning typically requires expensive manual annotations. We propose Action Self-Distillation (ASD), a method for learning language-conditioned reasoning for robotic navigation without human-annotated language or reasoning labels. ASD uses an open-source VLM, which contains broad world knowledge and reasoning capabilities, as both a teacher network and a student policy. During training, the self-teacher receives privileged trajectory feedback and infers the underlying navigation goal together with goal-oriented reasoning. The student is then trained to condition on the inferred navigation goal, predict future ego-frame trajectory, and provide a description of the scene semantics, plan, and reasoning, from onboard observations alone. Across 13 challenging indoor and outdoor real-world navigation tasks, our method improves success rate over the strongest baseline by $30\%$ in indoor settings, $30\%$ on multi-stage tasks, and achieves $95\%$ success rate on outdoor tasks, validating that privileged trajectory feedback can enable self-distillation for scalable generalist navigation policies without human annotations.

Chinese Translation

语言为人类指定和推理目标提供了自然的接口，但训练语言条件的机器人策略和推理通常需要昂贵的人工标注。我们提出了一种方法——动作自蒸馏（Action Self-Distillation, ASD），用于在没有人工标注语言或推理标签的情况下学习语言条件的机器人导航推理。ASD使用一个开源的视觉语言模型（VLM），该模型包含广泛的世界知识和推理能力，作为教师网络和学生策略。在训练过程中，自我教师接收特权的轨迹反馈，并推断潜在的导航目标以及目标导向的推理。然后，学生被训练为基于推断出的导航目标进行条件化，预测未来的自我框架轨迹，并仅根据机载观察提供场景语义、计划和推理的描述。在13个具有挑战性的室内和室外真实世界导航任务中，我们的方法在室内环境中成功率比最强基线提高了30%，在多阶段任务中提高了30%，并在室外任务中达到了95%的成功率，验证了特权轨迹反馈可以实现自蒸馏，从而在没有人工标注的情况下实现可扩展的通用导航策略。

View on arXiv Download PDF AI Translation

cs.RO / 181 / 2610.17630

From Pixels to Executable Physics: Behavior-Grounded World Models for Embodied Action Consequence Prediction

从像素到可执行物理：基于行为的具身行动后果预测世界模型

Chen, Ruo, Pang, Jiahua, Niu, Yingjie, Lin, Hongyi, Tian, Jun, Zhu, Yang, Li, Chao

Abstract

In outdoor environments, varying terrain and contact conditions can cause identical action commands to produce different motion outcomes for quadruped robots. Predicting these outcomes is essential for reliable navigation and informed path selection. Existing video world models can generate visually plausible futures, yet often struggle to capture robot dynamics and terrain interactions, leading to overly optimistic predictions of action outcomes. We present DR-WM, a physically consistent world model that incorporates robot-terrain contact dynamics to predict the consequences of quadruped actions. Starting from an initial RGB image, our method reconstructs scene geometry and integrates dense, dimensionless priors over terrain contact properties, calibrated using real-world action-response data. The resulting terrain model supports the simulation of physics with spatially varying contact properties such as friction. Given the initial state of the robot and candidate action sequences, we simulate its future motion and use the resulting camera trajectories and rendered geometry to guide the training of a video world model. The generated future observations are then used to update the terrain reconstruction, enabling iterative model rollouts without additional real-world images. Extensive experiments demonstrate that DR-WM substantially outperforms geometry-only baselines and baselines with uniform physical parameters on key physical metrics characterizing action outcomes, while achieving the best performance among the evaluated baselines. These physically grounded predictions support more informed action selection: across 90 challenging scenarios, an 8B vision-language model augmented with DR-WM outperforms an unassisted 32B vision-language model on five of six evaluation metrics.

Chinese Translation

在户外环境中，变化的地形和接触条件可能导致相同的动作指令产生不同的运动结果，这对四足机器人来说尤为明显。预测这些结果对于可靠的导航和明智的路径选择至关重要。现有的视频世界模型能够生成视觉上可信的未来场景，但往往难以捕捉机器人动力学和地形交互，导致对动作结果的过于乐观的预测。我们提出了DR-WM，这是一种物理一致的世界模型，结合了机器人与地形接触的动力学，以预测四足机器人动作的后果。我们的算法从初始的RGB图像出发，重建场景几何，并整合关于地形接触特性的密集无量纲先验，这些先验通过真实世界的动作-反应数据进行校准。最终生成的地形模型支持具有空间变化接触特性的物理模拟，例如摩擦力。给定机器人的初始状态和候选动作序列，我们模拟其未来运动，并利用生成的相机轨迹和渲染几何体来指导视频世界模型的训练。生成的未来观测结果随后用于更新地形重建，使得在没有额外真实世界图像的情况下实现模型的迭代推演。大量实验表明，DR-WM在关键物理指标上显著优于仅基于几何的基线和具有均匀物理参数的基线，同时在评估的基线中实现了最佳性能。这些基于物理的预测支持更明智的动作选择：在90个具有挑战性的场景中，增强了DR-WM的8B视觉-语言模型在六个评估指标中的五个上超越了未辅助的32B视觉-语言模型。

View on arXiv Download PDF AI Translation

cs.RO / 182 / 2610.17638

Acting Through Registers

通过寄存器进行动作

Zhang, Yiwei, Yang, Yantai, Gao, Jin, Lou, Jin, Chen, Andong, Liu, Yufei, Xing, Boyang, Zhu, Yuchen, Hu, Weiming, Zhang, Zhipeng

Abstract

World Action Models (WAMs) use generative visual representations to condition robot actions, yet how these representations organize action-relevant information remains poorly understood. We identify persistent high-norm spatial outliers in WAM representations and find that their action relevance persists even after normalization and K/V projection attenuate their extreme magnitude. Motivated by this observation, we introduce RegisterWAM, which adds learnable non-spatial registers to the world model and uses them directly for action conditioning. Registers alone do not fully eliminate spatial outliers; pairing them with observation flow-matching (FM) loss supervision suppresses spatial outliers while concentrating high-norm responses in registers. Beyond this reorganization, observation FM loss supervision also improves action learning without registers, motivating an additional noised-observation FM objective that learns current-scene recovery from corrupted inputs. The resulting model achieves 81.64\% success on LIBERO-Plus without additional embodied pretraining data or foundation models, reaches 93.0\% average success on RoboTwin 2.0, and improves real-world manipulation performance.

Chinese Translation

世界动作模型（World Action Models, WAMs）利用生成的视觉表示来调节机器人动作，但这些表示如何组织与动作相关的信息仍然不够清楚。我们识别出WAM表示中的持续高范数空间异常值，并发现即使在归一化和K/V投影减弱其极端幅度后，这些异常值的动作相关性仍然存在。基于这一观察，我们引入了RegisterWAM，它在世界模型中增加了可学习的非空间寄存器，并直接用于动作调节。仅使用寄存器并不能完全消除空间异常值；将其与观察流匹配（flow-matching, FM）损失监督结合使用，可以抑制空间异常值，同时将高范数响应集中在寄存器中。除了这种重组，观察FM损失监督还在没有寄存器的情况下改善了动作学习，这促使我们提出了一个额外的噪声观察FM目标，从损坏的输入中学习当前场景的恢复。最终模型在LIBERO-Plus上取得了81.64%的成功率，无需额外的具身预训练数据或基础模型，在RoboTwin 2.0上达到了93.0%的平均成功率，并改善了现实世界的操作性能。

View on arXiv Download PDF AI Translation

cs.RO / 183 / 2610.17651

Code2Worlds-2: Physically Plausible Agent Harness for Embodied World Simulation

Code2Worlds-2：用于具身世界模拟的物理可信代理工具

Zhang, Haoyu, Li, Siming, Zhang, Zeyu, Zhang, Yi, Chua, Tat-Seng, Tang, Hao

Abstract

Recent coding large language models (LLMs) can synthesize dynamic 4D worlds from natural language, but the resulting content remains a visual artifact. It lacks physical semantics such as the collision geometry, mass, friction, inertia, and joint structure, required to enable it to simulate the physical world, and its dynamics are scripted to look plausible rather than designed to execute. What is needed is an executable world: one that a physics engine can load and step, resolving contact physically. We present Code2Worlds-2, an extension of Code2Worlds that turns 4D world generation into an embodied world simulation. We keep the original dual-stream generation pipeline intact and append two modules. First, a SimReady Scene Compiler that turns a generated visual scene into a simulator-compatible representation: it assigns collision meshes via convex decomposition, estimates physical attributes with a vision-language model, infers a lightweight support graph, instantiates the engine's contact, friction, and dynamics constraints, and applies a gravity settling pass, before exporting the scene as MuJoCo MJCF. Second, a Simulator-in-the-Loop 4D Refiner that implements an automated agent harness-a compile, simulate, critique and refine loop-that converts visual dynamics into executable rigid and articulated dynamics and evolves the physical settings rather than only the code, replacing the purely visual motion critic with a simulator critic that inspects penetration depth, collision pairs, stability, contact forces, joint limits, and robot reachability. We further propose SimCode4D, a state-based benchmark with five task templates for evaluating embodied simulation. Experiments on SimCode4D and across natural, indoor, industrial, urban, and service scenes show that Code2Worlds-2 produces physically plausible, simulator-compatible, and robot-interactable scenes.

Chinese Translation

最近的大型语言模型（LLMs）能够从自然语言合成动态的四维世界，但生成的内容仍然是视觉伪影。它缺乏模拟物理世界所需的物理语义，例如碰撞几何、质量、摩擦、惯性和关节结构，其动态是为了看起来可信而编写的，而不是为了执行而设计的。我们需要一个可执行的世界：一个物理引擎可以加载并逐步处理的世界，能够物理地解决接触问题。我们提出了Code2Worlds-2，这是Code2Worlds的扩展，将四维世界生成转变为具身世界模拟。我们保持原有的双流生成管道不变，并附加了两个模块。首先是SimReady场景编译器，它将生成的视觉场景转换为与模拟器兼容的表示：通过凸分解分配碰撞网格，利用视觉-语言模型估计物理属性，推断轻量级支持图，实例化引擎的接触、摩擦和动态约束，并应用重力沉降处理，最后将场景导出为MuJoCo MJCF。其次是环中模拟器的四维精炼器，它实现了一个自动化的代理工具——一个编译、模拟、评估和精炼的循环——将视觉动态转换为可执行的刚体和关节动态，并演化物理设置，而不仅仅是代码，用模拟器评估替代纯视觉运动评估，检查穿透深度、碰撞对、稳定性、接触力、关节限制和机器人可达性。我们进一步提出了SimCode4D，这是一个基于状态的基准，包含五个任务模板，用于评估具身模拟。在SimCode4D以及自然、室内、工业、城市和服务场景中的实验表明，Code2Worlds-2生成了物理可信、与模拟器兼容且可与机器人交互的场景。

View on arXiv Download PDF AI Translation

cs.RO / 184 / 2610.17657

Towards a Systematic Study of Embodied System 2 over Full Task Execution Cycles

朝着对完整任务执行周期的具身系统2的系统研究

Wei, Yuxi, Fang, Ye, Yang, Shuai, Wu, Tong, Wang, Tai, Qi, Xiaojuan

Abstract

An embodied brain, or System 2, must turn an abstract goal into actions while discovering relevant scene conditions and checking whether executed skills achieved their intended effects. Open-loop evaluations cannot test observations shaped by an agent's decisions, while some closed-loop evaluations provide limited evidence during execution or assume idealized skill outcomes. We introduce SCOPER, a suite for studying observation-grounded planning, execution, and recovery. SCOPER-bench combines high-level goals, ongoing visual observations, a shared rule-based System 1 for low-level actions, and controlled execution failures. Its construction and validation pipeline yields 870 tasks across BEHAVIOR and ProcTHOR, complemented by about 3,500 situated questions tied to the evaluated agent's execution. Benchmark results reveal difficulties in retaining scene evidence, monitoring progress, and recovering from unexpected outcomes. To address these difficulties, SCOPER-harness combines Evidence Memory and Progress Verification with a dynamic task agenda and bounded exploration, without additional model training. Across two evaluated open-source backbones and two simulators, the harness improves task success by 11.1-28.6%; a harnessed 9B model outperforms a directly executed 27B model on both simulators. In an offline next-action evaluation using observations from 200 real-robot demonstrations, the harness also improves full-call accuracy by 14.7% and 15.3%. SCOPER provides a setting for diagnosing and improving embodied reasoning throughout a complete task execution cycle.

Chinese Translation

具身大脑或系统2必须将抽象目标转化为行动，同时发现相关的场景条件并检查执行的技能是否达到了预期效果。开放式评估无法测试由代理决策形成的观察，而一些闭环评估在执行过程中提供的证据有限或假设理想化的技能结果。我们引入了SCOPER，一个用于研究基于观察的规划、执行和恢复的工具包。SCOPER-bench结合了高层目标、持续的视觉观察、一个共享的基于规则的系统1用于低层次行动，以及控制的执行失败。其构建和验证流程产生了870个任务，涵盖BEHAVIOR和ProcTHOR，并辅以约3500个与被评估代理执行相关的情境问题。基准结果揭示了在保留场景证据、监控进展和从意外结果中恢复方面的困难。为了解决这些困难，SCOPER-harness结合了证据记忆和进展验证，配备动态任务议程和有限探索，而无需额外的模型训练。在两个评估的开源基础模型和两个模拟器上，harness提高了任务成功率11.1%-28.6%；一个经过harness处理的9B模型在两个模拟器上均优于直接执行的27B模型。在使用200个真实机器人演示的观察进行的离线下一步行动评估中，harness还提高了完整调用的准确率14.7%和15.3%。SCOPER提供了一个设置，用于诊断和改善整个任务执行周期中的具身推理。

View on arXiv Download PDF AI Translation

cs.RO / 185 / 2610.17667

PrT-Flow: Probability-Tractable Flow Matching for Reinforcement Learning

PrT-Flow：用于强化学习的概率可处理流匹配

Sun, Congzhong, Tan, Ning

Abstract

Diffusion models and Flow Matching can represent flexible distributions over robot actions, but many reinforcement learning algorithms also require the probability of each action. We introduce PrT-Flow (Probability-Tractable Flow Matching), which computes probability density while retaining multimodal, multi-step action generation. Its Jacobian-Tractable Transformer (JTT) makes each sampling step invertible and enables efficient computation of its Jacobian determinant. We train these finite steps with Conditional Euler-step Training (CET), which supervises their displacement over one step during sampling rather than their instantaneous velocity. We prove that, under certain assumptions, CET has the same minimizers as fitting the Euler-solver updates of Flow Matching. PrT-Flow achieves 96% success on Robomimic Lift after 300 epochs of imitation learning, matching the diffusion policy reference, and 96% success on OGBench Cube-Triple after 600k online steps. These results demonstrate that invertible networks with tractable probability densities can learn effective robot policies and support SAC-style actor-critic training. Our code is available at https://github.com/Sunlight-zero/PrT-Flow .

Chinese Translation

扩散模型和流匹配可以表示机器人动作的灵活分布，但许多强化学习算法也需要每个动作的概率。我们提出了PrT-Flow（概率可处理流匹配），它在保留多模态和多步动作生成的同时计算概率密度。其雅可比可处理变换器（Jacobian-Tractable Transformer, JTT）使每个采样步骤可逆，并能够高效计算其雅可比行列式。我们通过条件欧拉步训练（Conditional Euler-step Training, CET）来训练这些有限步骤，该方法在采样过程中监督它们在一步内的位移，而不是瞬时速度。我们证明，在某些假设下，CET的最小化器与流匹配的欧拉求解器更新拟合相同。PrT-Flow在300个模仿学习周期后，在Robomimic Lift上取得了96%的成功率，匹配了扩散策略参考，并在600k在线步骤后在OGBench Cube-Triple上取得了96%的成功率。这些结果表明，具有可处理概率密度的可逆网络能够学习有效的机器人策略，并支持SAC风格的演员-评论家训练。我们的代码可在https://github.com/Sunlight-zero/PrT-Flow获取。

View on arXiv Download PDF AI Translation

cs.RO / 186 / 2610.17688

OccWAM: Occupancy World-Action Models with Joint Scene-to-Ego Flow Matching

OccWAM：具有联合场景到自我流匹配的占用世界-动作模型

Huang, Qihang, Huang, Wenhui, Chua, Collister, Fang, Shiyu, Chen, Zhan, Zhang, Songyan, Wang, Zhidong, Tian, Daiying, Lv, Chen

Abstract

Occupancy world models provide a structured, metric representation of future driving scenes, making them a promising basis for end-to-end autonomous driving. However, existing occupancy-aware planners typically couple occupancy prediction and action generation through either hierarchical coupling or parallel multi-task learning. Hierarchical methods use occupancy either as a fully predicted future preceding action generation or as an action-conditioned rollout for trajectory evaluation, whereas parallel multi-task methods predict occupancy alongside planning through a separate branch or auxiliary objective. These interfaces retain either an explicit generation order or a role asymmetry, limiting reciprocal interaction between world prediction and action generation during trajectory formation. We propose OccWAM, an occupancy-native world-action model that jointly generates dense future occupancy and continuous ego trajectories via flow matching conditioned on scene representations. Building on this joint formulation, we introduce scene-to-ego world-action flow matching, a two-stage formulation. The first stage predicts scene-centric occupancy together with an absolute ego trajectory, providing a stable representation of future scene evolution. The same predicted world is then re-indexed into the future ego frames induced by the initial trajectory, where ego motion is represented through adjacent pose increments and a second joint flow refines ego-centric occupancy and motion for action-aligned planning. Unlike video-based world-action models, OccWAM preserves explicit semantic geometry for direct planning reasoning without pixel synthesis. Extensive experiments on Bench2Drive and nuScenes demonstrate state-of-the-art performance in closed-loop driving, open-loop planning, and future occupancy prediction.

Chinese Translation

占用世界模型提供了未来驾驶场景的结构化度量表示，使其成为端到端自动驾驶的有希望基础。然而，现有的占用感知规划器通常通过层次耦合或并行多任务学习将占用预测与动作生成结合在一起。层次方法将占用作为完全预测的未来，先于动作生成，或作为动作条件的展开用于轨迹评估，而并行多任务方法则通过单独的分支或辅助目标在规划过程中同时预测占用。这些接口保留了明确的生成顺序或角色不对称，限制了在轨迹形成过程中世界预测与动作生成之间的相互作用。我们提出了OccWAM，一种占用原生的世界-动作模型，通过基于场景表示的流匹配联合生成密集的未来占用和连续的自我轨迹。在这一联合公式的基础上，我们引入了场景到自我世界-动作流匹配的两阶段公式。第一阶段预测以场景为中心的占用以及绝对自我轨迹，提供未来场景演变的稳定表示。然后，将相同预测的世界重新索引到由初始轨迹引发的未来自我帧中，其中自我运动通过相邻姿态增量表示，第二个联合流则细化自我中心的占用和运动，以便进行与动作对齐的规划。与基于视频的世界-动作模型不同，OccWAM保留了明确的语义几何形状，以便在不进行像素合成的情况下进行直接规划推理。在Bench2Drive和nuScenes上的大量实验表明，在闭环驾驶、开环规划和未来占用预测方面，OccWAM实现了最先进的性能。

View on arXiv Download PDF AI Translation

cs.RO / 187 / 2610.17694

HelpMeOut: A Neuropsychology-Grounded Benchmark for Error-Aware Monitoring of Household Task Execution

HelpMeOut：基于神经心理学的家庭任务执行错误监测基准

Wang, Haoran, Huang, Jingwei, Duan, Anqing, Ahn, Ho Seok, Macdonald, Bruce

Abstract

We introduce HelpMeOut, an embodied AI benchmark for error-aware monitoring of household task execution. Existing embodied household benchmarks typically assume that humans execute tasks correctly, but this assumption does not hold for cognitively impaired persons or older adults whose everyday activities produce structured execution errors. To address this gap, HelpMeOut operationalizes research on everyday action errors into 20000 paired activity episodes, covering nine neuropsychologically motivated error categories. In each episode, agents observe a streaming human activity trajectory and must monitor whether the execution is unfolding correctly, diagnosing any structured errors as early as possible. We evaluate learned-monitor, symbolic, change-point, and frontier LLM prompting baselines. Results show that these methods perform poorly as error monitors, often missing when execution goes wrong. These results suggest a gap left by standard goal-inference and task-completion benchmarks, making HelpMeOut a testbed for error-aware assistive agents.

Chinese Translation

我们介绍了HelpMeOut，这是一个针对家庭任务执行错误监测的具身人工智能基准。现有的具身家庭基准通常假设人类能够正确执行任务，但这一假设对于认知障碍者或老年人并不成立，他们的日常活动会产生结构化的执行错误。为了解决这一问题，HelpMeOut将日常行动错误的研究转化为20000个成对的活动情节，涵盖九个神经心理学驱动的错误类别。在每个情节中，代理观察一个流式的人类活动轨迹，并必须监测执行是否正确展开，尽早诊断出任何结构化错误。我们评估了学习监测、符号、变更点和前沿大语言模型（LLM）提示的基线。结果表明，这些方法作为错误监测器的表现较差，常常未能发现执行中的错误。这些结果表明，标准的目标推断和任务完成基准存在缺口，使得HelpMeOut成为错误意识辅助代理的测试平台。

View on arXiv Download PDF AI Translation

cs.RO / 188 / 2610.17746

The Human--VLA Communication Gap: What Humans Leave Implicit and What Policies Cannot Recover

人类与VLA之间的沟通差距：人类隐含的内容与政策无法恢复的内容

Xu, Yiqing, Wu, Jiajun

Abstract

A robot may know how to perform a task yet fail when a human asks for it naturally. We study this human--VLA communication gap by separating an \emph{information gap}, where task semantics, grounding, or event structure remain unresolved, from a residual \emph{execution gap}, where even a fully specified task depends on how it is organized and invoked. To do this, we characterize how people naturally express demonstrated objectives and evaluate controlled variants of the same physical tasks, with detailed analysis on $\pi_{0.5}$ and consistent trends on $\pi_0$-FAST and Cosmos3-Edge. Our study reveals that humans increasingly compress instructions as tasks grow, relying more on contextual grounding and implicit execution structure. Current policies are misaligned with this behavior: they tolerate grounding dependence relatively well, but struggle when task semantics or event structure must be reconstructed. Even after the information gap is resolved, execution remains sensitive to command granularity and invocation state. These findings identify concrete targets for human-centered policy learning and interfaces that better bridge natural human objectives to robot behavior. Project website: https://commgap.github.io/.

Chinese Translation

机器人可能知道如何执行某项任务，但在人类自然请求时却失败。我们通过将 extit{信息差距}（即任务语义、基础或事件结构未得到解决）与剩余的 extit{执行差距}（即即使任务完全指定，仍然依赖于其组织和调用方式）分开，来研究这种人类与VLA之间的沟通差距。为此，我们描述了人们如何自然地表达示范目标，并评估相同物理任务的受控变体，详细分析了$ ext{π}_{0.5}$和$ ext{π}_0$-FAST以及Cosmos3-Edge的一致趋势。我们的研究揭示，随着任务的增加，人类越来越倾向于压缩指令，更多依赖于上下文基础和隐含的执行结构。目前的政策与这种行为不一致：它们对基础依赖的容忍度相对较高，但在任务语义或事件结构必须重建时却面临困难。即使在信息差距得到解决后，执行仍然对命令粒度和调用状态敏感。这些发现为以人为中心的政策学习和接口提供了具体目标，以更好地将自然的人类目标与机器人行为桥接。项目网站：https://commgap.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 189 / 2610.17748

I3L: Iterative Interactive Imitation Learning for Whole-Body Mobile Manipulation

I3L：用于全身移动操控的迭代交互模仿学习

Ai, Wensi, Luey, Christopher, Dharmarajan, Karthik, Han, Evans, Jiang, Yunfan, Ren, Stef, Xie, Jianwen, Wu, Jiajun, Fei-Fei, Li, Zhang, Ruohan

Abstract

Policies trained offline from teleoperated demonstrations can degrade outside their training distribution. This gap is particularly acute in long-horizon whole-body mobile manipulation, where long horizons amplify compounding errors and coordinated arm, torso, and base motion makes exhaustive demonstration coverage impractical. We introduce I3L, an integrated hardware and learning system that combines whole-body correction with iterative policy refinement to address this last-mile reliability gap. Its low-cost, open-source JoyLo+ interface uses policy-tracking leader arms and touch-triggered takeover to enable coordinated corrections across the body parts, while iterative deployment and retraining allow operators to target failures of the updated policy. We evaluate I3L across three robot embodiments and eight simulated and real-world tasks, including four long-horizon whole-body mobile manipulation tasks. Controlled simulation experiments show that four-round HG-DAgger achieves the highest final success on both tasks. Under this protocol, I3L improves autonomous success over the base policies by an average of 68.3 percentage points and achieves at least 90% success on every task, outperforming demonstration-only training by 35.6 percentage points on average with the same number of training trajectories. Averaged across tasks, takeover duration per episode decreases by 72% over four rounds. Together, these results show that I3L helps close the last-mile gap between offline training and reliable deployment and provides practical insights into collecting and using corrections for long-horizon whole-body mobile manipulation. Videos, visualizations, and appendix are at https://i-3l.github.io/.

Chinese Translation

从遥控演示中离线训练的策略在其训练分布之外可能会退化。这种差距在长时间跨度的全身移动操控中尤为明显，因为长时间跨度会加剧累积误差，而协调的手臂、躯干和底座运动使得全面演示覆盖变得不切实际。我们提出了I3L，这是一种集成硬件和学习系统，结合了全身校正与迭代策略优化，以解决这一最后一公里的可靠性差距。其低成本、开源的JoyLo+接口使用策略跟踪的领导臂和触发接管的方式，使得身体各部位之间能够进行协调校正，同时迭代部署和再训练使操作员能够针对更新策略的失败进行调整。我们在三种机器人实现和八个模拟及现实世界任务中评估了I3L，包括四个长时间跨度的全身移动操控任务。受控的模拟实验表明，四轮HG-DAgger在这两个任务上达到了最高的最终成功率。在该协议下，I3L使自主成功率比基础策略平均提高了68.3个百分点，并在每个任务上都达到了至少90%的成功率，平均比仅依赖演示的训练提高了35.6个百分点，且训练轨迹数量相同。跨任务平均来看，每集的接管持续时间在四轮中减少了72%。这些结果表明，I3L有助于缩小离线训练与可靠部署之间的最后一公里差距，并为收集和使用长时间跨度全身移动操控的校正提供了实用的见解。视频、可视化和附录可在 https://i-3l.github.io/ 查阅。

View on arXiv Download PDF AI Translation

cs.RO / 190 / 2610.17750

Learning Robot Policies for Whole-Body Mobility Assistance via Dynamic Human Motion Generation

通过动态人类动作生成学习全身移动辅助的机器人策略

Ban, Shikun, Nagai, Shohei, Song, Qingyang, Wan, Hanxi, Zhao, Eric, Wang, Thomas, Erickson, Zackory

Abstract

Robotic assistance for human mobility could benefit care recipients and reduce caregiver burden during physically demanding tasks such as Sit-to-Stand (STS) and Lift-and-Transfer (L&T). Learning robot policies for these whole-body assistive tasks is challenging: real-world data collection for human-robot interaction is costly and safety-critical, making simulation an attractive alternative. Prior work, however, typically treats the human as static or minimally responsive and thus fails to capture the cooperative nature of mobility assistance. We present a simulation-based framework for learning whole-body assistive robot policies. At its core is a human policy trained with an adversarial motion prior that produces realistic care-recipient motion in response to the robot's actions. A human-adaptive scripted controller interacts with this policy to collect diverse robot demonstrations at scale. From these demonstrations, we train 3D diffusion policies that map segmented human point clouds and robot joint positions to whole-body actions for a mobile manipulator, without relying on privileged human state at deployment. The resulting robot policies successfully perform both STS and L&T assistance in simulation. Preliminary experiments further demonstrate sim-to-real transfer for STS after fine-tuning on a small amount of real-world data. Videos are available on our project page: https://rchi-lab.github.io/human-gen-mobility-assistance/

Chinese Translation

机器人对人类移动的辅助可以惠及护理对象，并在诸如坐立（Sit-to-Stand, STS）和搬运转移（Lift-and-Transfer, L&T）等体力要求高的任务中减轻护理人员的负担。学习这些全身辅助任务的机器人策略具有挑战性：人机交互的真实数据收集成本高且安全关键，使得模拟成为一种有吸引力的替代方案。然而，先前的研究通常将人类视为静态或反应最小的存在，因此未能捕捉到移动辅助的合作性质。我们提出了一种基于模拟的框架，用于学习全身辅助机器人策略。其核心是一个经过对抗性动作先验训练的人类策略，该策略能够根据机器人的动作生成真实的护理对象动作。一个人类自适应脚本控制器与该策略互动，以大规模收集多样化的机器人演示。从这些演示中，我们训练了3D扩散策略，将分割的人类点云和机器人关节位置映射到移动操控器的全身动作，而不依赖于部署时的特权人类状态。最终生成的机器人策略成功地在模拟中执行了STS和L&T辅助。初步实验进一步展示了在少量真实数据上微调后，STS的模拟到真实转移。视频可在我们的项目页面上查看：https://rchi-lab.github.io/human-gen-mobility-assistance/

View on arXiv Download PDF AI Translation

cs.RO / 191 / 2610.17789

DexStitch: Learning and Composing Dexterous Skills from Human Videos via Contact-Aware Robotic Inpainting

DexStitch：通过接触感知机器人修复从人类视频中学习和组合灵巧技能

Choi, Byeonggyeol, Jung, Seongmin, Nam, Unghui, Song, Youngbo, Min, Sunhong, Lee, MinHyung, Shim, Byonghyo, Lee, Yongseok

Abstract

A long-horizon manipulation task consists of primitive skills and the transitions connecting them. Learning skills that can be executed in sequence requires supervision for both, which is costly to collect on a robot. Human videos offer a scalable source of demonstrations, but transferring them to robots requires extracting robot-compatible observations and actions together with their skill structure. We present DexStitch, a framework that cuts long-horizon human videos into skills and transitions, converts them into aligned robot observation-action-skill data, and stitches the learned skills into a requested sequence. A skill classifier first segments the videos into manipulation skills and transitions. DexStitch then converts the human demonstrations into robot observations and actions while retaining their corresponding skill labels. During conversion, per-finger human-object contacts guide dexterous hand retargeting and robotic inpainting, preserving contact geometry and finger-object occlusion ordering. To stitch the learned skills into a sequence, a shared skill-conditioned policy is pretrained on the resulting synthetic data and finetuned on a mixture of synthetic data and a small set of single-skill teleoperation episodes. Conditioning on the current and next skills guides each transition toward the following skill, while predicted skill progression enables closed-loop switching. On a held-out session, the classifier agrees with manual annotations on 92.4% of frames. Across four household skills, DexStitch achieves 80.0% single-skill success and 77.5% sequence success.

Chinese Translation

长时间的操作任务由基本技能及其连接的过渡组成。学习可以按顺序执行的技能需要对这两者进行监督，而在机器人上收集这些监督是成本高昂的。人类视频提供了一种可扩展的示范来源，但将其转移到机器人上需要提取与机器人兼容的观察和动作，以及它们的技能结构。我们提出了DexStitch，一个框架，它将长时间的人类视频切割成技能和过渡，将其转换为对齐的机器人观察-动作-技能数据，并将学习到的技能缝合成所请求的序列。技能分类器首先将视频分割为操作技能和过渡。然后，DexStitch将人类示范转换为机器人观察和动作，同时保留相应的技能标签。在转换过程中，每个手指与物体的接触引导灵巧手的重新定向和机器人修复，保持接触几何形状和手指-物体遮挡顺序。为了将学习到的技能缝合成一个序列，一个共享的技能条件策略在生成的合成数据上进行预训练，并在合成数据和一小组单一技能遥操作集的混合上进行微调。对当前和下一个技能的条件指导每个过渡朝向下一个技能，而预测的技能进展使闭环切换成为可能。在一个保留的会话中，分类器与手动注释的帧一致率达到92.4%。在四项家庭技能中，DexStitch实现了80.0%的单一技能成功率和77.5%的序列成功率。

View on arXiv Download PDF AI Translation

cs.RO / 192 / 2610.17799

Beyond Safe Execution: Safe Online Reinforcement Learning for Autonomous Driving with Value Calibration and Adaptive Guidance

超越安全执行：具有价值校准和自适应指导的自主驾驶安全在线强化学习

Li, Zeqiao, Wang, Yijing, Wang, Haoyu, Li, Zheng, Zuo, Zhiqiang

Abstract

Online reinforcement learning (RL) enables driving policies to adapt to changing deployment conditions, but requires safe exploration throughout training. Safety filtering alone does not ensure stable policy improvement or reduced dependence on corrections. We propose Safe Online Adaptive Refinement (SOAR-Flow), a framework for safe and stable online fine-tuning of offline-initialized Flow-based driving policies. A control barrier function (CBF)--model predictive control (MPC) filter corrects potentially unsafe actions before execution. To support stable policy updates, we introduce Gaussian-return-aware value calibration based on calibrated Q-learning (Cal-QL), using recorded behavior returns and predicted return variability to smooth conservative critic updates and mitigate policy degradation. Paired corrective learning teaches feasible action adjustments, while an intervention-to-go predictor encourages decisions requiring fewer future corrections. Its predictions also regulate guidance strength, allowing recovery supervision to weaken as predicted dependence decreases. Together with gradually reduced calibration regularization, this supports continued task improvement while preserving hard execution constraints. We analyze conditions for safe execution at control steps and diminishing auxiliary learning influences. We evaluate SOAR-Flow in simulation and on an uncrewed ground vehicle (UGV). The results demonstrate improved driving performance and reduced reliance on safety corrections during protected online fine-tuning. The demonstration video and project overview are available at https://lzqw.github.io/SOAR-Flow/.

Chinese Translation

在线强化学习（RL）使驾驶策略能够适应不断变化的部署条件，但在训练过程中需要安全探索。仅依靠安全过滤并不能确保稳定的策略改进或减少对修正的依赖。我们提出了安全在线自适应精细化（Safe Online Adaptive Refinement，SOAR-Flow）框架，用于安全且稳定的在线微调离线初始化的基于流的驾驶策略。控制障碍函数（Control Barrier Function，CBF）-模型预测控制（Model Predictive Control，MPC）过滤器在执行前纠正潜在的不安全动作。为了支持稳定的策略更新，我们引入了基于校准Q学习（Calibrated Q-learning，Cal-QL）的高斯回报感知价值校准，利用记录的行为回报和预测的回报变异性来平滑保守的评论家更新并减轻策略退化。配对的修正学习教授可行的动作调整，而干预预测器鼓励减少未来修正需求的决策。其预测还调节指导强度，使得随着预测依赖性的降低，恢复监督的强度减弱。结合逐渐减少的校准正则化，这支持了任务的持续改进，同时保持严格的执行约束。我们分析了控制步骤下安全执行的条件以及辅助学习影响的减弱。我们在仿真和无人地面车辆（Uncrewed Ground Vehicle，UGV）上评估了SOAR-Flow。结果表明，在受保护的在线微调过程中，驾驶性能得到了改善，并且对安全修正的依赖减少。演示视频和项目概述可在 https://lzqw.github.io/SOAR-Flow/ 获取。

View on arXiv Download PDF AI Translation

cs.RO / 193 / 2610.17802

Beyond the End-Effector: Learning Whole-Arm Contact Manipulation from Human Demonstrations

超越末端执行器：从人类示范中学习全臂接触操控

Huang, Bo-Ruei, Seita, Daniel, Bıyık, Erdem

Abstract

Robot manipulation has traditionally relied on end-effector-centric interactions. In contrast, humans routinely exploit whole-arm contacts to improve efficiency, robustness, and adaptability, especially when the hands are occupied or when precise grasping is unnecessary. Enabling similar capabilities in robots remains challenging: the robot must select where on its arm to make contact, generate effective non-prehensile interactions, and satisfy task-specific pose or safety constraints on the end-effector. We propose Whole-Arm Contact Learning (WACL), a hierarchical framework that describes each whole-arm contact task with three robot-agnostic representations and hands them to a quasi-static surface-contact optimizer: semantic keypoints that summarize the scene, a reference contact trajectory that specifies where and how the arm presses on the object, and safety-aware path constraints that a vision-language model generates from the task description to protect the end-effector's task. The robot's kinematic chain enters only in this optimizer, which grounds these representations onto the arm by jointly searching over which link makes contact, where on that link, and the joint configuration that realizes it. This decomposition lets WACL learn the contact trajectory from 10 bare-hand human demonstrations per task collected without teleoperation, while new end-effector constraints change only the optimizer's input rather than any trained component. Real-world experiments on a 7-DoF arm across four whole-arm tasks show that WACL sustains contact under end-effector constraints where a strong link-pose baseline fails (78% vs. 33% success) and generalizes to new constraints without retraining. Website: https://wacl-robot.github.io

Chinese Translation

机器人操控传统上依赖于以末端执行器为中心的交互。相比之下，人类常常利用全臂接触来提高效率、鲁棒性和适应性，特别是在双手被占用或不需要精确抓取的情况下。在机器人中实现类似的能力仍然具有挑战性：机器人必须选择其臂部的接触位置，生成有效的非抓取交互，并满足末端执行器的任务特定姿态或安全约束。我们提出了全臂接触学习（Whole-Arm Contact Learning, WACL），这是一个分层框架，用于用三种与机器人无关的表示描述每个全臂接触任务，并将其交给一个准静态表面接触优化器：总结场景的语义关键点、指定臂部如何以及何处施压于物体的参考接触轨迹，以及由视觉-语言模型从任务描述中生成的安全意识路径约束，以保护末端执行器的任务。机器人的运动链仅在此优化器中介入，该优化器通过共同搜索哪个关节与物体接触、在该关节的哪个位置接触以及实现该接触的关节配置，将这些表示与臂部相结合。这种分解使得WACL能够从每个任务收集的10个不通过遥操作的赤手人类示范中学习接触轨迹，而新的末端执行器约束仅改变优化器的输入，而不是任何训练组件。在四个全臂任务上对一个7自由度臂的实际实验表明，WACL在末端执行器约束下保持接触，而强链接姿态基线则失败（成功率78%对33%），并且在不重新训练的情况下能够推广到新的约束。网站：https://wacl-robot.github.io

View on arXiv Download PDF AI Translation

cs.RO / 194 / 2610.17806

Which Trajectory Is Safer? Combining Rule and Learning for Safety Evaluation

哪种轨迹更安全？结合规则与学习进行安全评估

Luo, Xuewen, Jiang, Linlian, Yu, Bo, Ding, Fan, Zuo, Xinxin, Wang, Yang

Abstract

Collision rate remains a dominant metric for evaluating autonomous driving trajectory safety, but its binary nature obscures substantial safety differences among collision-free trajectories. Existing surrogate-safety metrics provide finer-grained assessments, yet often trade off physical interpretability against adaptivity to interaction context. We introduce FutureRisk, a continuous metric that quantifies the latent future risk of a trajectory by how closely and persistently its realized interactions approach unsafe states. At its core, our Context-Adaptive Risk Model (CARM) combines a context-dependent safety-time distribution with an independently supervised critical-event-time distribution. Its safety criterion is bounded around a closed-form physical prior, while a bounded residual provides additional adaptation without abandoning this physical anchor. Experiments on DeepAccident and Bench2Drive show that FutureRisk achieves stronger agreement with native collision labels than existing surrogate-safety measures, discriminates risk among collision-free trajectories, and transfers zero-shot across datasets. Comparisons across four planners further show that FutureRisk reveals safety differences hidden by binary collision outcomes. Together, these results establish FutureRisk as a fine-grained, interpretable, and context-adaptive metric for trajectory safety evaluation. Code is available at https://github.com/Xuewen2025/safety-evaluation.

Chinese Translation

碰撞率仍然是评估自动驾驶轨迹安全性的主要指标，但其二元性质掩盖了无碰撞轨迹之间的实质性安全差异。现有的替代安全指标提供了更细致的评估，但往往在物理可解释性与适应交互上下文之间存在权衡。我们引入了FutureRisk，这是一种连续指标，通过其实现的交互如何紧密且持续地接近不安全状态来量化轨迹的潜在未来风险。我们的上下文自适应风险模型（Context-Adaptive Risk Model, CARM）核心结合了上下文依赖的安全时间分布与独立监督的关键事件时间分布。其安全标准围绕一个封闭形式的物理先验进行界定，而一个有界的残差则提供了额外的适应性，而不放弃这一物理锚点。在DeepAccident和Bench2Drive上的实验表明，FutureRisk在与原生碰撞标签的一致性上优于现有的替代安全措施，能够区分无碰撞轨迹之间的风险，并在数据集间实现零样本迁移。对四个规划器的比较进一步表明，FutureRisk揭示了二元碰撞结果所隐藏的安全差异。这些结果共同确立了FutureRisk作为一种细致、可解释且上下文自适应的轨迹安全评估指标。代码可在 https://github.com/Xuewen2025/safety-evaluation 获取。

View on arXiv Download PDF AI Translation

cs.RO / 195 / 2610.17843

Physics-Aware Object-Centric SE(3) Forecasting for Real-Time Proactive Dynamic Grasping

基于物理知识的物体中心SE(3)预测用于实时主动动态抓取

Zhou, Yirui, Xu, Hongli, Zheng, Ying, Wang, Yi, Chau, Lap-Pui

Abstract

Proactive dynamic grasping requires robots to predict the future state at which physical contact will occur, rather than reacting to the currently observed object pose. But what future must the robot predict, and at what level of abstraction? We argue that contact-relevant rigid-body state in SE(3) is a compact task-sufficient abstraction for this problem: it directly determines where and when the robot should establish contact, without requiring visually complete future scenes. This insight leads to a lightweight object-centric forecast-then-select framework that predicts future rigid-body trajectories from short pose histories and uses a reachability-aware interception selector to choose feasible contact states under kinematic and timing constraints. Trained with randomized physics simulation over primitive shapes, the forecaster transfers to real objects once their metric geometry and SE(3) state histories are recovered. Experiments across simulation and real-world dynamic grasping tasks show that our framework improves the average success rate from $24.5\%$ to $62.0\%$ in simulation, and from $15.3\%$ to $67.3\%$ in real-world catching, conveyor, and turntable interception. The SE(3) forecaster runs in $15\,\mathrm{ms}$, orders of magnitude faster than dense visual future prediction, enabling real-time contact decision making. Our results suggest that effective proactive grasping does not require imagining the entire future scene. Instead, predicting the compact physical future that determines contact provides an effective and practical abstraction for real-time dynamic grasping.

Chinese Translation

主动动态抓取要求机器人预测将发生物理接触的未来状态，而不是对当前观察到的物体姿态做出反应。但机器人需要预测什么样的未来，以及在什么抽象层次上进行预测？我们认为，SE(3)中与接触相关的刚体状态是解决此问题的紧凑且足够的抽象：它直接决定了机器人何时何地应建立接触，而无需完整的未来场景视觉信息。这一见解促成了一种轻量级的物体中心预测-选择框架，该框架从短的姿态历史中预测未来的刚体轨迹，并使用考虑可达性的拦截选择器在运动学和时间约束下选择可行的接触状态。通过对基本形状进行随机物理仿真训练，预测器在恢复真实物体的度量几何和SE(3)状态历史后可以迁移到真实物体上。跨越仿真和真实世界动态抓取任务的实验表明，我们的框架在仿真中的平均成功率从24.5%提高到62.0%，在真实世界的接球、传送带和转盘拦截中从15.3%提高到67.3%。SE(3)预测器的运行时间为15毫秒，比密集视觉未来预测快几个数量级，使得实时接触决策成为可能。我们的结果表明，有效的主动抓取并不需要想象整个未来场景。相反，预测决定接触的紧凑物理未来为实时动态抓取提供了一种有效且实用的抽象。

View on arXiv Download PDF AI Translation

cs.RO / 196 / 2610.17870

LAG-IO: Multi-IMU State Estimation for Legged Robots Using Factor Graphs

LAG-IO：基于因子图的多IMU状态估计用于腿式机器人

Zhou, Yicheng, Chen, Longwen, Yu, Qinghua, Xiao, Junhao, Lu, Huimin, Chen, Xieyuanli

Abstract

Robust proprioceptive state estimation is crucial for legged robots, particularly in challenging environments where exteroceptive sensors are unreliable. While traditional leg odometry typically relies on a single body-mounted inertial measurement unit (IMU) fused with leg kinematics via an extended Kalman filter (EKF), such methods are vulnerable to linearization errors and hard-contact assumptions, leading to severe drift during slippage or dynamic locomotion.In this paper, we propose LAG-IO, a factor graph-based multi-IMU state estimation framework that tightly couples measurements from a main body IMU, multiple leg-mounted IMUs, and joint encoders within a sliding-window optimization. To handle non-stationary foot-ground interactions, we introduce a continuous probabilistic contact detection module based on kernel density estimation (KDE), which adaptively scales the covariance of zero-velocity update (ZUPT) factors. Furthermore, kinematic relative position factors are formulated to structurally anchor the independent leg inertial states to the robot's main body. Experimental results on a quadruped robot across various unstructured terrains demonstrate that, compared to existing baselines, LAG-IO achieves higher planar trajectory accuracy, reduces unobservable vertical divergence, and provides smoother trajectory transitions during unstable contacts, while maintaining competitive attitude observability.

Chinese Translation

稳健的本体状态估计对腿式机器人至关重要，尤其是在外部传感器不可靠的挑战性环境中。传统的腿部里程计通常依赖于一个安装在主体上的惯性测量单元（IMU），通过扩展卡尔曼滤波器（EKF）与腿部运动学进行融合，但这种方法容易受到线性化误差和硬接触假设的影响，导致在滑动或动态运动过程中出现严重漂移。本文提出了LAG-IO，一种基于因子图的多IMU状态估计框架，它在滑动窗口优化中紧密耦合来自主体IMU、多个腿部IMU和关节编码器的测量数据。为了处理非平稳的足部与地面的交互，我们引入了一个基于核密度估计（KDE）的连续概率接触检测模块，该模块自适应地调整零速度更新（ZUPT）因子的协方差。此外，构建了运动学相对位置因子，以结构性地将独立的腿部惯性状态锚定到机器人的主体上。在各种非结构化地形上对四足机器人进行的实验结果表明，与现有基线相比，LAG-IO实现了更高的平面轨迹精度，减少了不可观测的垂直发散，并在不稳定接触期间提供了更平滑的轨迹过渡，同时保持了竞争力的姿态可观测性。

View on arXiv Download PDF AI Translation

cs.RO / 197 / 2610.17873

Information-based Unified Design Framework for NMR Magnetic Tracking System

基于信息的核磁共振磁追踪系统统一设计框架

Yetim, Seçkin Eren, Tiryaki, Mehmet Efe

Abstract

Nuclear magnetic resonance (NMR) magnetic tracking offers a promising alternative to magnetic resonance imaging (MRI)-based localization for medical robotic guidance. However, scalar field measurements and field-aligned marker moments constrain position and orientation tracking, requiring careful configuration design. We present a unified information-based framework that uses local covariance analysis to evaluate receiver geometry, marker separation, and external source configuration in on/off-board NMR tracking. Numerical evaluation shows that planar-array precision depends jointly on receiver count and geometry, with triangular lattice arrays achieving the lowest best-case position uncertainty among the evaluated layouts for 9--20 receivers. Two spatially separated, field-aligned markers enable device-axis tracking, with predicted angular uncertainty below $2^\circ$ at the evaluated workspace center for suitable marker separations. We further establish point-dipole source--receiver duality, unifying the analysis of on/off-board NMR tracking. These

Chinese Translation

核磁共振（NMR）磁追踪为医学机器人引导提供了一种有前景的替代方案，取代基于磁共振成像（MRI）的定位。然而，标量场测量和场对齐标记的矩限制了位置和方向的追踪，需谨慎设计配置。我们提出了一种统一的信息基础框架，利用局部协方差分析评估在板上/板下NMR追踪中的接收器几何形状、标记间距和外部源配置。数值评估表明，平面阵列的精度共同依赖于接收器数量和几何形状，其中三角晶格阵列在评估的9至20个接收器布局中实现了最低的最佳位置不确定性。两个空间分离的、场对齐的标记能够实现设备轴的追踪，预测在适当的标记间距下，在评估的工作空间中心的角度不确定性低于$2^ heta$。我们进一步建立了点偶极子源-接收器的对偶性，统一了板上/板下NMR追踪的分析。

View on arXiv Download PDF AI Translation

cs.RO / 198 / 2610.17881

Agentic Robot Creation through Generation, Verification, and Revision

通过生成、验证和修订创造自主机器人

Li, Xuebing, He, Yibo, Liu, Sitong, Ji, Ran, Gu, Jiayuan

Abstract

Can an AI agent turn high-level user intent into a functioning robot and teach it how to behave? We study this question from two complementary aspects: embodiment and behavior creation. For embodiment creation, we investigate how well current multimodal models capture mechanical requirements, and introduce computational metrics for evaluating generated designs before fabrication. For behavior creation, we present Roboforge, an agentic framework that turns task specifications into executable policies through motion generation, policy learning, and execution-grounded revision. Given a robot embodiment and task, the agent generates a reference motion, configures the learning objective, and trains a policy. Crucially, motion remains in the loop: rollout feedback can revise the reference motion, reward, or both. We evaluate a generated embodiment across multiple MLLMs and behavior creation across four diverse robots against reward-refinement and one-shot motion baselines. Our computational metrics expose recurring mechanical failures in generated designs, while behavior experiments show that execution-grounded refinement improves task-level behavior and execution quality. Finally, we fabricate an agent-generated Gorilla and directly deploy learned behaviors on the physical robot.

Chinese Translation

一个人工智能代理能否将高层次的用户意图转化为一个功能性机器人并教会它如何行为？我们从两个互补的方面研究这个问题：体现和行为创造。在体现创造方面，我们研究当前多模态模型在捕捉机械需求方面的表现，并引入计算指标用于在制造前评估生成的设计。在行为创造方面，我们提出了Roboforge，一个自主框架，通过运动生成、策略学习和执行基础的修订将任务规范转化为可执行策略。给定一个机器人体现和任务，代理生成参考运动，配置学习目标，并训练策略。关键是，运动始终处于循环中：回滚反馈可以修订参考运动、奖励或两者。我们在多个多语言学习模型（MLLMs）上评估生成的体现，并在四种不同的机器人上评估行为创造，比较奖励细化和单次运动基线。我们的计算指标揭示了生成设计中反复出现的机械故障，而行为实验表明，执行基础的细化改善了任务级行为和执行质量。最后，我们制造了一个代理生成的猩猩，并直接在物理机器人上部署学习到的行为。

View on arXiv Download PDF AI Translation

cs.RO / 199 / 2610.17922

WoVR+: World Model as Reliable Simulator for Versatile VLA Policy Optimization with RL

WoVR+: 作为可靠模拟器的世界模型在多功能视觉-语言-动作策略优化中的应用

Zhou, Shangqing, Jiang, Zhennan, Ma, Ning, Fu, Xin, Zheng, Yupeng, Tian, Shuai, Li, Huangrui, Liu, Jiahao, Yu, Chao, Xia, Zhongpu, Li, Haoran, Zhao, Dongbin

Abstract

Action-conditioned world models can reduce costly real-robot interaction in reinforcement learning for Vision-Language-Action models (VLA-RL), but hallucinations can mislead policy optimization. Extending them to richer VLA interfaces and long-horizon tasks raises challenges in spatiotemporal and cross-modal consistency, exploration-reliability balance in RL, and compounding rollout errors. We present WoVR+, extending WoVR's hallucination-aware WM-RL framework by redesigning the learned environment and imagined interaction protocol: A VLA-interface-aligned world model provides multi-view prediction and proprioceptive updates while mitigating hallucinations. Chunk-wise noise preserves flow-based policy exploration while limiting prediction-degrading action jitter. For long-horizon tasks, stage-local imagined training combines stage-entry resets, stage-specific rewards, local horizons, and controllable sampling to shorten generated prefixes and balance training across stages. WoVR+ improves success over SFT by 21.09% and 16.7% on RoboTwin and 6 real-robot tasks. Long-horizon phone-packaging success rises from 30% to 75%. Videos are available at https://youmo445.github.io/WoVR_Plus.github.io/.

Chinese Translation

基于动作的世界模型可以减少视觉-语言-动作模型（VLA-RL）中强化学习的高成本真实机器人交互，但幻觉可能会误导策略优化。将其扩展到更丰富的VLA接口和长时间任务面临时空和跨模态一致性、强化学习中的探索-可靠性平衡以及累积回放错误等挑战。我们提出了WoVR+，通过重新设计学习的环境和想象的交互协议，扩展了WoVR的幻觉感知WM-RL框架：一个与VLA接口对齐的世界模型提供多视角预测和本体感知更新，同时减轻幻觉影响。分块噪声保留基于流的策略探索，同时限制预测退化的动作抖动。对于长时间任务，阶段局部的想象训练结合了阶段入口重置、阶段特定奖励、局部视野和可控采样，以缩短生成前缀并平衡各阶段的训练。WoVR+在RoboTwin和6个真实机器人任务上相比SFT提高了21.09%和16.7%的成功率。长时间手机包装的成功率从30%上升至75%。视频可在https://youmo445.github.io/WoVR_Plus.github.io/观看。

View on arXiv Download PDF AI Translation

cs.RO / 200 / 2610.17927

GLST: Global-Local Spatio-Temporal Memory Decomposition Framework for Online Chained-Goal Navigation

GLST：用于在线链式目标导航的全球-局部时空记忆分解框架

Liang, Yousef, Ardalan, Ryan, Goraly, Yarden, Zhu, Chenhao, Yin, Yue, Frey, Jonas, Tomlin, Claire, Pavone, Marco, Sastry, Shankar

Abstract

Autonomous navigation in real-world environments often requires long-horizon operation under partial observability and evolving goals. Existing mapless navigation approaches predominantly assume fixed goals throughout an episode, limiting applicability to practical scenarios where goals change online during execution. We study online chained-goal navigation, where goals evolve through existing-goal redirection or new-goal emergence without explicit transition signals. Under this setting, we find navigation may require either continued exploration or strategic revisitation, creating conflicting memory requirements. We propose Global-Local Spatio-Temporal Belief Learning (GLST), a memory-decomposition framework that separates latent beliefs into global-local spatial and long-short temporal representations for long-horizon reasoning. We further introduce goal-conditioned rewards to balance exploration and exploitation under evolving goals. Extensive experiments in simulation across diverse terrains demonstrate improved chained-goal navigation performance over baselines. Real-robot experiments using the Spot quadruped further validate our memory decomposition framework.

Chinese Translation

在现实环境中，自主导航通常需要在部分可观测性和不断变化的目标下进行长时间的操作。现有的无地图导航方法主要假设在整个过程中目标是固定的，这限制了其在执行过程中目标在线变化的实际应用场景。我们研究了在线链式目标导航，其中目标通过现有目标的重定向或新目标的出现而演变，而没有明确的过渡信号。在这种情况下，我们发现导航可能需要持续探索或战略性重访，这导致了相互矛盾的记忆需求。我们提出了全球-局部时空信念学习（Global-Local Spatio-Temporal Belief Learning，GLST），这是一种记忆分解框架，将潜在信念分离为全球-局部空间和长-短时态表示，以便进行长时间的推理。我们进一步引入了目标条件奖励，以平衡在不断变化的目标下的探索与利用。在多种地形下的广泛模拟实验表明，我们的方法在链式目标导航性能上优于基线。使用Spot四足机器人进行的真实机器人实验进一步验证了我们的记忆分解框架。

View on arXiv Download PDF AI Translation

cs.RO / 201 / 2610.17930

Locomotion with Online Awareness of Dynamics for Precision Footholds

动态在线感知下的精确踏点运动

Ren, Kunzhao, Zhu, Zhishuo, Xiong, Xiaobin

Abstract

Unknown payloads alter robot dynamics and make precise contact execution difficult when foothold margins are small. We present LOAD, a perceptive locomotion framework that combines explicit inertial estimates with implicit dynamics features for narrow-beam traversal. Separate recurrent modules infer torso inertial changes and task-relevant history from proprioception, while a height-map teacher is distilled into a depth-based student. We extensively evaluate LOAD and its variants in simulation and on hardware across beam widths and payload configurations. The results show that implicit dynamics features accelerate training, while explicit inertial estimation further improves traversal success; combining both enables LOAD to approach the performance of a privileged policy supplied with ground-truth inertial parameters. Hardware experiments demonstrate traversal with unilateral payloads of 6 kg on a 10 cm beam and 9 kg on a 15 cm beam, and an evenly distributed 11 kg payload on a 15 cm beam. These results support combining physical estimates and history features for precision locomotion under unknown loads.

Chinese Translation

未知的负载会改变机器人动态，并在踏点边际较小时使精确接触执行变得困难。我们提出了LOAD，一个感知运动框架，它将显式的惯性估计与隐式动态特征结合，用于狭窄梁的穿越。独立的递归模块通过本体感觉推断躯干的惯性变化和与任务相关的历史信息，同时一个高度图教师被提炼成一个基于深度的学生。我们在模拟和硬件上对LOAD及其变体进行了广泛评估，涵盖了不同的梁宽和负载配置。结果表明，隐式动态特征加速了训练，而显式惯性估计进一步提高了穿越成功率；两者结合使LOAD的性能接近于提供真实惯性参数的特权策略。硬件实验展示了在10厘米梁上以6公斤的单侧负载和在15厘米梁上以9公斤的负载进行穿越，以及在15厘米梁上均匀分布的11公斤负载。这些结果支持在未知负载下结合物理估计和历史特征以实现精确运动。

View on arXiv Download PDF AI Translation

cs.RO / 202 / 2610.17977

ReEnvision: Future-Guided Selective Action Repair for Long-Horizon Manipulation

ReEnvision：面向未来的选择性动作修复框架用于长时间跨度的操控

Zeng, Xiyin, Huang, Xuewei, Lu, Yi, Yuan, Weihao, Wang, Hao

Abstract

Long-horizon robotic manipulation requires robust execution over sequences of interdependent actions. Existing methods typically handle execution deviations by replanning the remaining action chunk or correcting actions step by step as new observations arrive. However, a local deviation may affect only a few actions under distribution shift, yet replanning regenerates the entire chunk from an unfamiliar state, potentially introducing errors into otherwise valid actions. Step-wise correction may react too late, as it does not explicitly anticipate how the deviation affects subsequent actions. We propose ReEnvision, a future-guided selective action repair framework that updates a joint future-action belief using execution feedback. It then learns how the resulting belief revision affects each unexecuted action. The resulting belief revision enables selective repair of only the affected actions within a cached action chunk, preserving valid actions and reserving full replanning for severe deviations. On CALVIN, L-CALVIN, and RLBench, ReEnvision achieves 86.6% 5-step success, improves 10-step success from 56% to 63%, and reaches an average success rate of 88.8%, respectively. On five real-world manipulation tasks, it achieves an average success rate of 82.8%. Controlled repair experiments further demonstrate the benefits of localizing affected actions and preserving valid ones during execution.

Chinese Translation

长时间跨度的机器人操控需要在一系列相互依赖的动作中进行稳健的执行。现有的方法通常通过重新规划剩余的动作块或在新观察到达时逐步纠正动作来处理执行偏差。然而，局部偏差可能仅影响少数几个动作，在分布变化下，重新规划会从一个不熟悉的状态重新生成整个块，可能会将错误引入到原本有效的动作中。逐步纠正可能反应过晚，因为它并未明确预测偏差如何影响后续动作。我们提出了ReEnvision，一个面向未来的选择性动作修复框架，通过执行反馈更新联合未来动作信念。然后，它学习信念修正如何影响每个未执行的动作。最终的信念修正使得仅对受影响的动作进行选择性修复，从而保留有效的动作，并将全面重新规划留给严重偏差。在CALVIN、L-CALVIN和RLBench上，ReEnvision实现了86.6%的5步成功率，将10步成功率从56%提高到63%，并达到了88.8%的平均成功率。在五个真实世界的操控任务中，它实现了82.8%的平均成功率。控制修复实验进一步证明了在执行过程中局部化受影响动作和保留有效动作的好处。

View on arXiv Download PDF AI Translation

cs.RO / 203 / 2610.17983

Breaking the Average: Overcoming Convergent Defaults in Robotic Manipulation via Diagnostic Chain-of-Thought and Structured Guidance

打破平均：通过诊断思维链和结构化指导克服机器人操作中的收敛默认行为

Daba, Mohammed, AlAkwa, Waleed Abdulrahman Saleh

Abstract

Recent Vision-Language-Action (VLA) models, such as $\pi_{0.5}$, have demonstrated impressive generalization capabilities by scaling behavior cloning across massive datasets. However, these models frequently suffer from distributional convergence, a phenomenon where the policy regresses to "safe" but suboptimal mean behaviors due to mode averaging in the training data. This often results in generic grasp strategies, Convergent Defaults, that fail in constrained or long-tail scenarios. In this work, we propose a novel framework to identify and correct these behaviors without parameter updates. We introduce a three-stage methodology: (1) Probing for Defaults, identifying task-specific mode collapse; (2) Diagnostic Root Cause Analysis, utilizing the VLA's latent reasoning capabilities to verbalize the logic behind suboptimal actions; and (3) Structured Guidance Injection, where concrete, distinct alternatives are encoded into reusable "Lens Prompts" that shift the sampling distribution. We validate this approach using $\pi_{0.5}$ in ManiSkill3 simulation environments. We hypothesize that our method outperforms vanilla $\pi_{0.5}$ and standard Chain-of-Thought prompting by effectively breaking convergent defaults in complex manipulation tasks.

Chinese Translation

近期的视觉-语言-动作（VLA）模型，如$ ext{π}_{0.5}$，通过在大规模数据集上扩展行为克隆，展现了令人印象深刻的泛化能力。然而，这些模型常常面临分布收敛的问题，即由于训练数据中的模式平均，策略回归到“安全”但次优的平均行为。这通常导致在受限或长尾场景中失败的通用抓取策略，即收敛默认行为。在本研究中，我们提出了一种新颖的框架，以识别和纠正这些行为，而无需更新参数。我们引入了一个三阶段的方法论：（1）默认行为探测，识别任务特定的模式崩溃；（2）诊断根本原因分析，利用VLA的潜在推理能力来阐述次优行为背后的逻辑；（3）结构化指导注入，将具体、不同的替代方案编码为可重用的“透镜提示”，以改变采样分布。我们在ManiSkill3仿真环境中使用$ ext{π}_{0.5}$验证了该方法。我们假设我们的方法在复杂操作任务中有效打破收敛默认行为，优于普通的$ ext{π}_{0.5}$和标准的思维链提示。

View on arXiv Download PDF AI Translation

cs.RO / 204 / 2610.17999

TRACE: Efficient Transfer Learning for Neural Whole-Body Control Across Humanoid Embodiments

TRACE：跨人形体的神经全身控制的高效迁移学习

Lan, Qingyu, Tian, Zhihao, Wang, Zifan, Liang, Junwei

Abstract

Humanoid behavior foundation models (BFMs) enable diverse whole-body control, but adapting a pretrained controller to a new embodiment remains challenging because of kinematic and dynamic mismatch. We present TRACE, a framework for efficient cross-embodiment adaptation that preserves pretrained motion representations while specializing feedback control to the target robot.TRACE first injects kinematic priors into input and output projection layers and aligns nominal joint dynamics to reduce embodiment mismatch and support stable fine-tuning. It then combines encoder distillation with low-rank reinforcement learning on a compact, quality-screened motion set, using a two-phase curriculum to establish stable tracking before refining difficult motions. We apply TRACE to transfer SONIC from Unitree G1 to Unitree H2 using 10 million motion frames and approximately eight hours of training on a single RTX PRO 6000 GPU, at less than 0.1% of the GPU-hour cost of training SONIC from scratch. The adapted controller achieves 96.08\% success on 67,953 evaluation motions under a height-scaled tracking criterion. Real-robot demonstrations span dynamic motion tracking and whole-body teleoperation, with collected demonstrations supporting downstream vision-language-action policy post-training. These results demonstrate practical reuse of a pretrained humanoid controller across robots with different body scales and dynamics.

Chinese Translation

人形行为基础模型（BFMs）能够实现多样化的全身控制，但将预训练控制器适配到新的人形体仍然面临运动学和动力学不匹配的挑战。我们提出了TRACE，一个高效的跨人形体适配框架，能够在专注于目标机器人反馈控制的同时保留预训练的运动表示。TRACE首先将运动学先验注入到输入和输出投影层，并对标称关节动力学进行对齐，以减少人形体不匹配并支持稳定的微调。然后，它结合编码器蒸馏与低秩强化学习，在一个紧凑且经过质量筛选的运动集上，使用两阶段课程建立稳定的跟踪，然后再精炼困难的运动。我们将TRACE应用于将SONIC从Unitree G1迁移到Unitree H2，使用了1000万帧运动数据，并在单个RTX PRO 6000 GPU上进行了大约八小时的训练，其GPU小时成本不到从零开始训练SONIC的0.1%。适配后的控制器在67,953个评估运动中，在高度缩放的跟踪标准下达到了96.08%的成功率。真实机器人演示涵盖了动态运动跟踪和全身遥操作，收集的演示支持下游视觉-语言-动作策略的后续训练。这些结果展示了预训练人形控制器在不同体型和动力学的机器人之间的实际重用。

View on arXiv Download PDF AI Translation

cs.RO / 205 / 2610.18038

Data-Free Continual Learning for World Action Model with Rectified Generative Replay

无数据持续学习的世界动作模型与修正生成重放

Fang, Kaipeng, Liu, Hangxu, Shi, Chaojian, Zhou, Yunsong, Zhu, Hongzi, Guo, Minyi, Song, Jingkuan, Gao, Lianli

Abstract

World action models (WAMs) built on pretrained video generation models adapt quickly to new robotic tasks, but they forget previously learned tasks under sequential fine-tuning. Experience replay (ER) mitigates such forgetting by rehearsing data of previous tasks, yet this data is often unavailable in robotics, where a policy is deployed as weights only and its training data is not shared along with it. In this work, we ask whether a WAM can achieve data-free continual learning by regenerating its own past. Our key observation is that a WAM can indeed autoregressively regenerate complete rollouts of previously learned tasks and replay them to mitigate forgetting. However, replaying such raw rollouts still lags far behind replaying real data, which we attribute to two issues. First, the WAM is unable to recognize task completion and keeps generating purposeless motion after the task is done. Second, under open-loop generation, the generated actions gradually drift away from the generated observations. To address both issues, we present rectified generative replay (R$^2$-WAM), which truncates generated rollouts at task completion with an agentic evaluator and filters out rollouts whose actions are physically inconsistent with the observations. On the four LIBERO suites and five real-world tasks, R$^2$-WAM reduces forgetting by over 90% relative to sequential fine-tuning and nearly matches ER without accessing any data from previous tasks. Our analysis further reveals that forgetting in WAMs concentrates in inverse dynamics modeling: a forgotten WAM still produces visually plausible observations of previous tasks while its generated actions fail.

Chinese Translation

基于预训练视频生成模型构建的世界动作模型（WAMs）能够快速适应新的机器人任务，但在顺序微调过程中会遗忘先前学习的任务。经验重放（ER）通过重演先前任务的数据来减轻这种遗忘，然而在机器人领域，这些数据通常不可用，因为策略仅以权重的形式部署，而其训练数据并未随之共享。在本研究中，我们探讨了WAM是否能够通过重新生成自身的过去来实现无数据的持续学习。我们的关键观察是，WAM确实能够自回归地重新生成先前学习任务的完整回放并重放它们以减轻遗忘。然而，重放这些原始回放仍然远远落后于重放真实数据，我们将其归因于两个问题。首先，WAM无法识别任务完成，任务完成后仍持续生成无目的的动作。其次，在开环生成下，生成的动作逐渐偏离生成的观察。为了解决这两个问题，我们提出了修正生成重放（R$^2$-WAM），该方法通过一个代理评估器在任务完成时截断生成的回放，并过滤掉与观察不一致的动作。针对四个LIBERO套件和五个真实世界任务，R$^2$-WAM相较于顺序微调减少了超过90%的遗忘，并且在不访问任何先前任务数据的情况下，几乎达到了ER的效果。我们的分析进一步揭示，WAM中的遗忘主要集中在逆动力学建模上：一个被遗忘的WAM仍然能够生成先前任务的视觉上合理的观察，而其生成的动作却失败了。

View on arXiv Download PDF AI Translation

cs.RO / 206 / 2610.18040

Contractor Filters

承包商滤波器

Rohou, Simon

Abstract

We introduce Contractor Filters (CFs), a set-membership architecture for nonlinear continuous-time state estimation with bounded uncertainties and sporadic outliers. CFs represent feasible states symbolically by compositions of contractors rather than rematerializing a geometric enclosure after each update. This provides an abstraction layer between the set-theoretic filtering equations and their numerical realization, so prediction, correction, ambiguity, and outlier handling can be expressed within one algebra without prescribing a state geometry. We derive an exact nominal recursion and a guaranteed robust enclosure, with validated flow inclusions preserving soundness. Optional sub-paving compression then provides finite outer representations. CFs are evaluated against a robust particle filter on range-only and terrain-based localization under outliers, probabilistic model mismatch, association ambiguity, and multimodality. CFs retain the simulated truth in all tested configurations, while particle-filter reliability degrades under persistent bias and spatial aliasing. Contractor strength and compression budget control the trade-off between paving tightness and computational cost, while preserving the enclosure guarantee.

Chinese Translation

我们介绍了承包商滤波器（Contractor Filters, CFs），这是一种用于具有界限不确定性和偶发异常值的非线性连续时间状态估计的集合成员架构。CFs通过承包商的组合以符号方式表示可行状态，而不是在每次更新后重新构建几何包络。这为集合论滤波方程与其数值实现之间提供了一个抽象层，使得预测、修正、模糊性和异常值处理可以在一个代数中表达，而无需规定状态几何。我们推导出一个精确的名义递归和一个保证的鲁棒包络，并验证了保持健全性的流包含。可选的子铺装压缩随后提供有限的外部表示。在异常值、概率模型不匹配、关联模糊性和多模态情况下，CFs与鲁棒粒子滤波器进行了评估，应用于仅基于范围和地形的定位。CFs在所有测试配置中保持了模拟真实值，而粒子滤波器的可靠性在持续偏差和空间混叠下下降。承包商的强度和压缩预算控制了铺装紧密度与计算成本之间的权衡，同时保持包络保证。

View on arXiv Download PDF AI Translation

cs.RO / 207 / 2610.18053

LADDER: A Long-Horizon Robot Manipulation Benchmark

LADDER：一个长时间跨度的机器人操作基准

Petkov, Petko, Albanese, Giuliano, Nikolov, Nikolay, Van Gool, Luc, Tremblay, Jonathan, Paudel, Danda Pani

Abstract

General-purpose robot manipulation policies learn short-horizon skills effectively, but degrade rapidly on long-horizon tasks. Existing benchmarks cannot isolate failures because task horizon is either too short or entangled with scene and object variations. We propose LADDER, a long-horizon robot manipulation benchmark scaling from 1 to 10 subgoals within the same environment over three suites (SimplerEnv-Long, RobotArena-$\infty$-Long, and RoboLab-Long), 145 tasks, eight categories, two embodiments, and 2,024 simulated episodes per system. We evaluate two orchestration paradigms: text subgoals for learned low-level policies (VoLo) and tool selection orchestration (CaP-X, GPT-Policy, PSP-Agent, TiPToP). Given the full instruction directly, low-level policies achieve 85.4% on single-subgoal tasks, but drop to near zero on extended tasks. While orchestration remedies this, it remains bounded by physical execution errors of the low-level policy rather than high-level planning and achieves only ~25%. Tool selection orchestration with GPT-Policy on GPT-6 Astra significantly outperforms learned-policy methods, but incurs significant time and cost budget and succeeds only ~62%. Real-world experiments across two embodiments confirm the collapse of the low-level policy and the gain from orchestration. Project website: https://ladder.insait.ai

Chinese Translation

通用机器人操作策略能够有效学习短时间跨度的技能，但在长时间跨度任务中迅速退化。现有基准无法隔离失败，因为任务的时间跨度要么太短，要么与场景和物体的变化交织在一起。我们提出了LADDER，一个长时间跨度的机器人操作基准，在同一环境中扩展1到10个子目标，涵盖三个套件（SimplerEnv-Long、RobotArena-$ ext{∞}$-Long和RoboLab-Long），145个任务，八个类别，两种实现，以及每个系统2,024个模拟实验。我们评估了两种协调范式：用于学习的低级策略的文本子目标（VoLo）和工具选择协调（CaP-X、GPT-Policy、PSP-Agent、TiPToP）。在直接给出完整指令的情况下，低级策略在单子目标任务上达到了85.4%的成功率，但在扩展任务中降至接近零。虽然协调可以改善这一情况，但仍受限于低级策略的物理执行错误，而非高级规划，仅实现了约25%的成功率。使用GPT-Policy进行工具选择协调在GPT-6 Astra上显著优于学习策略方法，但耗费了大量时间和成本预算，仅成功率约为62%。跨两种实现的现实世界实验确认了低级策略的崩溃以及协调带来的收益。项目网站：https://ladder.insait.ai

View on arXiv Download PDF AI Translation

cs.RO / 208 / 2610.18093

LynxScenes: A Large-Scale Real-World Benchmark for Outdoor Wheel-Legged Robot Perception, Navigation, Imagination and Reconstruction

LynxScenes：一个大规模真实世界基准，用于户外轮腿机器人感知、导航、想象和重建

Tian, Jun, Wu, Hao, Chen, Ruo, Liao, Yili, Lin, Hongyi, Yu, Jincheng, Li, Chao, Niu, Yingjie

Abstract

Wheel-legged robots operating in outdoor spaces shared with humans require accurate perception, socially compliant navigation, future-scene reasoning, and scene reconstruction. However, existing datasets typically support these capabilities in isolation and rarely provide dense spatial and behavior-level supervision on real wheel-legged platforms. We introduce LynxScenes, a large-scale real-world dataset and benchmark collected with a LYNX M20 wheel-legged robot. It contains approximately 100K 30-s navigation clips spanning 833.6 h and 3,294.7 km, with 3.0M annotated frames. The synchronized recordings include four fisheye cameras, dual LiDARs, IMU, odometry, and robot states. Annotations comprise 20.82M oriented 3D boxes across 27 categories, 13.89M road-structure annotations, and 20 navigation instructions organized into three layers and aligned with continuous future trajectories. LynxScenes supports four complementary tasks: instruction-conditioned trajectory prediction, BEV-based perception of 3D objects and road structures, future scene prediction, and multi-view scene reconstruction. For trajectory prediction, we further introduce a log-replay-based pseudo-closed-loop evaluator that measures safety, progress, and instruction compliance beyond trajectory imitation. Extensive evaluations across representative navigation, perception, world-model, and reconstruction methods reveal substantial remaining challenges in close-range outdoor robot navigation. The dataset and evaluation toolkit are planned for public release.

Chinese Translation

在与人类共享的户外空间中操作的轮腿机器人需要准确的感知、社会适应的导航、未来场景推理和场景重建。然而，现有的数据集通常仅支持这些能力的孤立实现，且很少在真实的轮腿平台上提供密集的空间和行为级别的监督。我们介绍了LynxScenes，这是一个大规模的真实世界数据集和基准，使用LYNX M20轮腿机器人收集。该数据集包含大约100K个30秒的导航片段，覆盖833.6小时和3,294.7公里，标注帧数达到3.0M。同步录制的数据包括四个鱼眼摄像头、双激光雷达、惯性测量单元（IMU）、里程计和机器人状态。标注包括27个类别的20.82M个定向3D框、13.89M个道路结构标注，以及20个导航指令，这些指令分为三层，并与连续的未来轨迹对齐。LynxScenes支持四个互补任务：基于指令的轨迹预测、基于鸟瞰视图（BEV）的3D物体和道路结构感知、未来场景预测以及多视角场景重建。对于轨迹预测，我们进一步引入了一种基于日志重放的伪闭环评估器，该评估器在轨迹模仿之外测量安全性、进展和指令遵从性。在代表性的导航、感知、世界模型和重建方法上的广泛评估揭示了近距离户外机器人导航中仍然存在的重大挑战。该数据集和评估工具包计划公开发布。

View on arXiv Download PDF AI Translation

cs.RO / 209 / 2610.18118

ForeTac: Predictive Contact Guidance for Generative Robot Policies

ForeTac：生成机器人策略的预测接触指导

Chen, Zhiyuan, Chen, Shuai, Zhou, Haoran, Zhao, Yizhe, Yang, Kaiyuan, Duan, Zhixiang, Xu, Shaoqing, Cao, Ran, Xiang, Diyun, You, Yangwei

Abstract

Visually similar actions can produce contact loss, jamming, slippage, or irreversible damage, while tactile feedback may arrive after a critical transition. Contact-rich manipulation therefore requires generative policies to anticipate and evaluate whether candidate actions will produce desirable contact outcomes, then adjust action generation before execution. We present ForeTac, a predictive contact-guidance framework organized as a represent-predict-evaluate-guide loop. ForeTac encodes the current tactile state, predicts the tactile consequences of a candidate action sequence, and evaluates the predicted trajectory using a contact-quality objective learned from successful and failed physical interactions. At deployment, its gradient propagates through the tactile predictor to guide action generation while the pretrained model parameters remain frozen. We evaluate ForeTac on four real-robot tasks (planar and curved-surface wiping, card swiping, and fragile-object grasping) with diffusion and flow-matching policies. Across these tasks, ForeTac raises the average task success rate from 36.3% to 71.3% and the average contact success rate (CSR) from 20.0% to 66.3% compared with vision-only Diffusion Policy. Component analyses support the tactile representation and action-conditioned foresight, and show that predicted touch preserves contact-quality ordering. Project page: https://foretac.github.io/.

Chinese Translation

视觉上相似的动作可能导致接触丢失、卡滞、滑动或不可逆损伤，而触觉反馈可能在关键转变后才到达。因此，接触丰富的操控需要生成策略来预测和评估候选动作是否会产生理想的接触结果，然后在执行前调整动作生成。我们提出了ForeTac，一种组织为表示-预测-评估-指导循环的预测接触指导框架。ForeTac编码当前的触觉状态，预测候选动作序列的触觉后果，并使用从成功和失败的物理交互中学习的接触质量目标评估预测轨迹。在部署时，其梯度通过触觉预测器传播，以指导动作生成，同时预训练的模型参数保持不变。我们在四个真实机器人任务（平面和曲面擦拭、卡片刷卡和易碎物体抓取）上评估了ForeTac，使用扩散和流匹配策略。在这些任务中，ForeTac将平均任务成功率从36.3%提高到71.3%，将平均接触成功率（CSR）从20.0%提高到66.3%，相较于仅使用视觉的扩散策略。组件分析支持触觉表示和基于动作的前瞻性，并显示预测的触觉保持接触质量排序。项目页面：https://foretac.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 210 / 2610.18123

MARTS: A Modular Agricultural Robot for Targeted Spraying of Diseased Tomato Plants in Greenhouses

MARTS：一种用于温室中针对性喷洒病害番茄植物的模块化农业机器人

Singh, Rajmeet, Kaur, Manveen, Adamu, Yusuf, Kshetrimayum, Lochan, Hussain, Irfan

Abstract

Broadcast spraying remains the dominant method of disease control in greenhouse tomato production, yet it treats healthy and diseased tissue alike, exposes workers to hazardous aerosols, and depends on operator skill. This paper presents MARTS (Modular Agricultural Robot for Targeted Spraying), a low-cost robotic platform that closes the loop from plant-level disease detection to leaf-localised spray delivery in commercial greenhouses. MARTS integrates a Clearpath Husky A200 mobile base, a UR5 manipulator, a novel modular spraying end-effector built around a single 12 V micro-diaphragm pump, and a wrist-mounted Intel RealSense D435 RGB-D camera. Disease detection reuses our previously published Tomato Leaf Disease Vision Language Model (TLDVLM; Grounding DINO, SAM-2 and LoRA-tuned BLIP-2, 97.27% accuracy over 10 classes); this work focuses on actuation, control and system integration. A ROS pipeline lifts detections into 3D targets, generates collision-aware nozzle poses at a 120 mm stand-off using fitted leaf normals, and scales pump pulse duration to lesion area. Water-sensitive paper calibration showed a linear pulse-volume relationship (R^2 = 0.994). In MuJoCo simulation, an adaptive base-repositioning strategy achieved 100% coverage of 18 waypoints with a maximum orientation error of 1.1 deg. Deployed at the SILAL commercial greenhouse (Al Ain, UAE) on cherry tomatoes under natural infection, MARTS treated 312 diseased leaves in 12 autonomous trials with 92.4% targeting success, 18.7 mm mean spray-centre error, 6.1% adjacent-leaf overspray, and a 71% reduction in pesticide volume versus broadcast spraying. Mean cycle time was 4.8 s per leaf, and the spraying subsystem costs USD 412, 6-10x lower than comparable systems, showing that accurate leaf-level targeting is achievable on commodity hardware.

Chinese Translation

在温室番茄生产中，广播喷洒仍然是主要的病害控制方法，但它对健康和病变组织一视同仁，暴露工人于有害气溶胶中，并依赖操作员的技能。本文介绍了MARTS（模块化农业机器人用于针对性喷洒），这是一个低成本的机器人平台，旨在实现从植物级别病害检测到叶片局部喷洒的闭环，适用于商业温室。MARTS集成了Clearpath Husky A200移动底盘、UR5机械手、围绕单个12 V微型膜泵构建的新型模块化喷洒末端执行器，以及一个腕部安装的Intel RealSense D435 RGB-D相机。病害检测重用了我们之前发布的番茄叶病视觉语言模型（TLDVLM；Grounding DINO、SAM-2和LoRA调优的BLIP-2，10类的准确率为97.27%）；本研究重点在于驱动、控制和系统集成。ROS管道将检测结果提升为3D目标，生成在120 mm距离下使用拟合叶法线的碰撞感知喷嘴姿态，并根据病变面积调整泵脉冲持续时间。水敏纸校准显示出线性脉冲体积关系（R^2 = 0.994）。在MuJoCo仿真中，自适应底盘重定位策略实现了18个航点的100%覆盖，最大方向误差为1.1度。在阿联酋阿布扎比的SILAL商业温室中，MARTS在自然感染的樱桃番茄上进行了12次自主试验，处理了312片病叶，目标成功率为92.4%，喷洒中心平均误差为18.7 mm，邻近叶片的过喷率为6.1%，与广播喷洒相比，农药用量减少了71%。每片叶子的平均循环时间为4.8秒，喷洒子系统成本为412美元，比可比系统低6-10倍，显示出在商品硬件上实现准确的叶级目标是可行的。

View on arXiv Download PDF AI Translation

cs.RO / 211 / 2610.18135

SkipWorld: Latent Skip Planning with Distilled World Models for Efficient Visual Navigation

SkipWorld：基于提炼世界模型的潜在跳跃规划用于高效视觉导航

Yang, Yifei, Fan, Zehua, Yan, Zike, Li, Zhongyu, Wang, Yan

Abstract

World-model-based visual navigation enables action planning by predicting future states under candidate trajectories. However, long-horizon rollout remains challenging: models are often trained with ground-truth visual histories but deployed in closed-loop prediction, and evaluating many candidates with step-by-step rollout is computationally expensive. We propose SkipWorld, a latent skip planning framework for efficient visual navigation. SkipWorld introduces three components: (i) Latent AutoRoll, which trains a world model to recursively roll out future states in latent space, reducing the mismatch between training and closed-loop inference; (ii) SkipDistill, which distills the latent autoregressive model into a skip world model that directly predicts long-stride future latent states; and (iii) Skip-then-Refine Planning, a dual-model coarse-to-fine planner that first filters candidate trajectories with the skip model and then reranks selected candidates with the stronger autoregressive model. This design preserves accurate trajectory evaluation while reducing redundant rollout computation. Experiments on RECON, HuRoN, SCAND, and TartanDrive show that, compared with NWM, SkipWorld reduces the average ATE by 4.2% and RPE by 2.7%, while decreasing planning time by 62.4% with a 2.7-fold speedup.

Chinese Translation

基于世界模型的视觉导航通过预测候选轨迹下的未来状态来实现行动规划。然而，长时间跨度的展开仍然具有挑战性：模型通常在真实视觉历史上进行训练，但在闭环预测中部署，并且逐步展开评估多个候选的计算成本高昂。我们提出了SkipWorld，一种用于高效视觉导航的潜在跳跃规划框架。SkipWorld引入了三个组件：（i）潜在自回归展开（Latent AutoRoll），它训练一个世界模型在潜在空间中递归展开未来状态，从而减少训练与闭环推理之间的不匹配；（ii）跳跃提炼（SkipDistill），它将潜在自回归模型提炼为一个跳跃世界模型，直接预测长步长的未来潜在状态；（iii）跳跃后精炼规划（Skip-then-Refine Planning），一个双模型粗到细的规划器，首先使用跳跃模型过滤候选轨迹，然后使用更强的自回归模型对选定候选进行重新排序。该设计在减少冗余展开计算的同时保持了准确的轨迹评估。在RECON、HuRoN、SCAND和TartanDrive上的实验表明，与NWM相比，SkipWorld将平均绝对误差（ATE）降低了4.2%，相对位置误差（RPE）降低了2.7%，同时将规划时间减少了62.4%，实现了2.7倍的加速。

View on arXiv Download PDF AI Translation

cs.RO / 212 / 2610.18147

Walk Like This: Stylized Locomotion for Physics-Based Characters through Example-Driven Diffusion

像这样走：通过示例驱动扩散实现基于物理角色的风格化运动

Aurand, Joshua, Serifi, Agon, Christen, Sammy, Grandia, Ruben, Knoop, Espen, Hutter, Marco, Bächer, Moritz

Abstract

We present a real-time framework for expressive legged locomotion that enables independent control over velocity and style. Unlike existing data-driven controllers, which typically require explicit style labels for training, our method efficiently leverages large-scale unstructured motion data and generalizes to unseen styles. We disentangle style from functional locomotion through a tailored sampling strategy used to train a style encoder whose output conditions an autoregressive diffusion model. During inference, the user provides a style reference clip and a joystick velocity command. The diffusion model then generates a kinematic reference tracked by a reinforcement learning controller. To improve user controllability and stylization strength, we propose an attention biasing mechanism and a velocity command correction strategy. We demonstrate our method on a physics-based humanoid character in simulation and on a physical robot, showcasing robust real-time control and continuous style modulation. The results can be best seen in our supplemental video (https://www.youtube.com/watch?v=roFnS6N7nJ4).

Chinese Translation

我们提出了一种实时框架，用于表现力丰富的腿部运动，能够独立控制速度和风格。与现有的数据驱动控制器不同，后者通常需要明确的风格标签进行训练，我们的方法有效利用大规模非结构化运动数据，并能够推广到未见过的风格。我们通过一种定制的采样策略将风格与功能性运动解耦，该策略用于训练一个风格编码器，其输出条件化一个自回归扩散模型。在推理过程中，用户提供一个风格参考片段和一个操纵杆速度指令。然后，扩散模型生成一个由强化学习控制器跟踪的运动学参考。为了提高用户的可控性和风格化强度，我们提出了一种注意力偏置机制和速度指令修正策略。我们在仿真中的基于物理的人形角色和物理机器人上展示了我们的方法，展示了强大的实时控制和连续的风格调节。结果可以在我们的补充视频中看到（https://www.youtube.com/watch?v=roFnS6N7nJ4）。

View on arXiv Download PDF AI Translation

cs.RO / 213 / 2610.18160

Force Sensor-Based Picking Pattern Analysis and Two-Jaw Gripper Design for Cherry Tomato Robotic Harvesting in Greenhouse

基于力传感器的樱桃番茄机器人采摘模式分析及双爪夹具设计

Singh, Rajmeet, Kaur, Manveen, Adamu, Yusuf, Hussain, Irfan

Abstract

Robotic harvesting of cherry tomatoes in greenhouses remains challenging due to dense fruit clusters, delicate skin, and the need for gentle yet effective detachment. This study presents a force-data-driven framework for autonomous cherry tomato harvesting that integrates six-axis force/torque sensing, a bio-inspired compliant gripper, and deployment on a mobile robotic platform. A wearable ATI six-axis force/torque sensor mounted on the index finger captured kinetic signatures of three elemental picking patterns (gripping, twisting, and lateral pulling) across 150 manual trials (50 per pattern) in laboratory and greenhouse settings. Gripping required a peak normal force of 3.92 and 0.45 N, twisting produced a dominant torque of 19.20 2.10 Nmm about the pedicel axis with minimal truss disturbance, and pure lateral pulling alone was insufficient for reliable detachment. These measurements informed a novel fingertip-localised Finray gripper, with compliant TPU 95A Finray geometry restricted to the fruit-contact region and a rigid PLA body, emulating the dual-stiffness human fingertip. A Dynamixel XM430-W350 servomotor under current-based torque control applies the measured gripping force and a 45{\deg} wrist twist. Finite element analysis showed a maximum von Mises stress of 0.24 MPa under the 3.92 N load, below the bruising threshold of cherry tomato skin. Integrated on a UR5 arm and Clearpath Husky A200 base, the system was validated in NVIDIA Isaac Sim and in 50 autonomous trials at the SILAL commercial greenhouse, Abu Dhabi, UAE. The hybrid grip-twist-pull strategy achieved an 86.0% picking success rate, 2.3% fruit damage, and an 8.4 s cycle time per fruit, outperforming vacuum (66.3%), rotating (70.1%), clamping (76.0%), and suction-cutting (52.4%) end-effectors. These results demonstrate preliminary feasibility under the tested facility and cultivar conditions

Chinese Translation

在温室中，樱桃番茄的机器人采摘仍然面临挑战，原因在于果实簇的密集、果皮的脆弱，以及需要温和而有效的分离。本研究提出了一种基于力数据的自主樱桃番茄采摘框架，该框架集成了六轴力/扭矩传感器、生物启发的柔性夹具，并部署在移动机器人平台上。佩戴在食指上的ATI六轴力/扭矩传感器捕获了150次手动试验（每种模式50次）中三种基本采摘模式（抓取、扭转和侧向拉动）的运动特征，试验在实验室和温室环境中进行。抓取需要的峰值法向力为3.92 N和0.45 N，扭转在果梗轴上产生了19.20 ± 2.10 Nmm的主导扭矩，且对支架的干扰最小，而单纯的侧向拉动不足以实现可靠的分离。这些测量结果为一种新型的指尖局部Finray夹具提供了依据，该夹具在果实接触区域采用柔性TPU 95A Finray几何形状，并具有刚性PLA主体，模拟了双刚度的人类指尖。在基于电流的扭矩控制下，Dynamixel XM430-W350伺服电机施加测得的抓取力和45°的手腕扭转。有限元分析显示，在3.92 N负载下，最大von Mises应力为0.24 MPa，低于樱桃番茄果皮的挫伤阈值。该系统集成在UR5机械臂和Clearpath Husky A200底座上，在阿布扎比的SILAL商业温室中经过NVIDIA Isaac Sim验证，并进行了50次自主试验。混合抓取-扭转-拉动策略实现了86.0%的采摘成功率，2.3%的果实损伤率，以及每个果实8.4秒的循环时间，优于真空（66.3%）、旋转（70.1%）、夹持（76.0%）和吸附切割（52.4%）末端执行器。这些结果表明在测试的设施和品种条件下的初步可行性。

View on arXiv Download PDF AI Translation

cs.RO / 214 / 2610.18163

RESTA: Time-Scaling Augmentation for Data-Efficient Robot Imitation Learning

RESTA：用于数据高效的机器人模仿学习的时间缩放增强

Jiang, Yunpeng, Hu, Jianshu, Fang, Han, Guan, Zhiyuan, Weng, Paul, Ban, Yutong

Abstract

Robot manipulation policies trained through imitation learning commonly rely on costly demonstrations recorded in one task direction and at one temporal resolution. We propose Reversal and Stride-based Trajectory Augmentation (RESTA), which exploits time-scaling data augmentation to improve data efficiency through trajectory reversal and temporal resampling. Trajectory-reversal augmentation reverses and relabels demonstrations between paired tasks with opposite objectives, reusing recorded rollouts without additional environment interaction. Because contact-rich manipulation is not exactly reversible, we treat reversal as an approximate paired-task prior rather than a physical symmetry. Temporal-stride augmentation varies observation and action sampling strides and conditions the policy on the resulting offsets, enabling fixed or randomized temporal schedules. Experiments across 16 paired bimanual manipulation tasks show that trajectory reversal is particularly effective in the low-data regime and that temporal-stride augmentation improves success under randomized temporal schedules. With 25 and 50 independently collected demonstrations per task, RESTA achieves 78.4% and 85.2% average success, respectively, matching or exceeding baselines trained with twice as many demonstrations (77.3% and 85.4%). These results show that RESTA can improve data efficiency in robot imitation learning.

Chinese Translation

通过模仿学习训练的机器人操作策略通常依赖于在一个任务方向和一个时间分辨率下记录的高成本演示。我们提出了基于反转和步幅的轨迹增强（RESTA），利用时间缩放数据增强通过轨迹反转和时间重采样来提高数据效率。轨迹反转增强在具有相反目标的配对任务之间反转和重新标记演示，重用记录的轨迹而无需额外的环境交互。由于接触丰富的操作并不完全可逆，我们将反转视为一种近似的配对任务先验，而不是物理对称性。时间步幅增强改变观察和动作采样的步幅，并根据产生的偏移条件化策略，从而实现固定或随机的时间调度。在16个配对的双手操作任务中的实验表明，轨迹反转在低数据条件下特别有效，而时间步幅增强在随机时间调度下提高了成功率。在每个任务中独立收集25和50个演示时，RESTA分别达到了78.4%和85.2%的平均成功率，匹配或超过了使用两倍演示训练的基线（77.3%和85.4%）。这些结果表明，RESTA可以提高机器人模仿学习中的数据效率。

View on arXiv Download PDF AI Translation

cs.RO / 215 / 2610.18167

Beyond Planning: Differentiable Control-in-the-loop End-to-End Autonomous Driving

超越规划：可微分控制环路的端到端自主驾驶

Wang, Zhidong, Li, Zirui, Shan, Zitong, Mao, Zhiqi, Huang, Qihang, Chen, Zhan, Liang, Jingsong, Yu, Han, Lv, Chen

Abstract

End-to-end autonomous driving has achieved remarkable progress by planning trajectories directly from raw sensor data. Yet the step from planning to control still relies on hand-tuned tracking controllers that block the gradient path from vehicle motion back to the network. This gap stems from non-differentiable vehicle dynamics, aliased supervision from coarsely sampled expert labels, rate mismatch between high-rate control and low-rate re-planning, and vehicle model discrepancies between training and deployment. This paper proposes DIVA (Differentiable Identified Vehicle dynamics for end-to-end Autonomous driving) to close this gap. An identified actuator-level vehicle model is embedded in the training loop of the end-to-end driving network, forming a differentiable chain from perception to vehicle states. A learning-based control decoder replaces the external controller and is supervised on the rolled-out vehicle states, bypassing the aliased control-input labels. A rate-consistent control-input parameterization reconciles the heterogeneous sampling rates and enables asynchronous execution with a 3.8 ms control-only cycle. On-policy re-identification and an online disturbance observer address training-time distribution shift and deployment-time model mismatch respectively. On the closed-loop Bench2Drive benchmark, DIVA achieves a driving score of 86.7, surpassing the widely adopted state-of-the-art approach by 3.3 points while nearly halving mean jerk and reducing brake switching by 64%, and maintains superior path tracking accuracy under significant vehicle model mismatch at deployment.

Chinese Translation

端到端自主驾驶通过直接从原始传感器数据规划轨迹取得了显著进展。然而，从规划到控制的步骤仍然依赖于手动调节的跟踪控制器，这阻碍了从车辆运动到网络的梯度传递。这一差距源于不可微分的车辆动力学、来自粗采样专家标签的别名监督、高频控制与低频重新规划之间的速率不匹配，以及训练与部署之间的车辆模型差异。本文提出了DIVA（可微分识别车辆动力学用于端到端自主驾驶）以填补这一差距。一个识别的执行器级车辆模型被嵌入到端到端驾驶网络的训练循环中，形成了从感知到车辆状态的可微分链。基于学习的控制解码器替代了外部控制器，并在展开的车辆状态上进行监督，从而绕过了别名控制输入标签。速率一致的控制输入参数化调和了异构采样速率，并使得以3.8毫秒的控制周期实现异步执行。在线策略再识别和在线干扰观测器分别解决了训练时分布转移和部署时模型不匹配的问题。在闭环Bench2Drive基准测试中，DIVA的驾驶得分达到86.7，超越了广泛采用的最先进方法3.3分，同时将平均抖动几乎减半，并将刹车切换减少了64%，在显著的车辆模型不匹配情况下仍保持优越的路径跟踪精度。

View on arXiv Download PDF AI Translation

cs.RO / 216 / 2610.18188

OneDemo: Unified Policy Learning and Data Generation from a Single Demonstration

OneDemo：从单一示范中统一学习策略和数据生成

Qian, Ying, Hu, Jianshu, Fang, Han, Li, Shuguang, Ban, Yutong

Abstract

Learning robot policies from demonstrations is often limited by the cost of collecting diverse, high-quality expert data. While an appealing alternative is to expand a single successful demonstration into a diverse set of successful trajectories, existing augmentation methods typically rely on manually designed transformation rules or complex trajectory planning, requiring substantial task-specific engineering and human effort. We introduce OneDemo, a reinforcement learning framework that learns to edit a single successful demonstration for new task configurations. We formulate demonstration editing as a structured reinforcement learning problem, in which an editor learns, from sparse task-success rewards, both when to transition between object-centric trajectory segments and how to geometrically adapt each segment. The learned editor serves both as a task policy and as a reusable demonstration generator. Across 15 Meta-World tasks, OneDemo reaches a 92.7% success rate within 53.7k environment interactions on average. Across 9 DexMimicGen tasks, it generates successful demonstrations with a 98.4% average success rate, and BC-RNN policies trained on the generated data achieve 99.6% average success at matched dataset scale. These results show that learning to edit a single successful demonstration provides a structured and effective basis for both policy learning and scalable demonstration generation.

Chinese Translation

从示范中学习机器人策略通常受到收集多样化、高质量专家数据成本的限制。一个有吸引力的替代方案是将单一成功示范扩展为一组多样化的成功轨迹，但现有的增强方法通常依赖于手动设计的变换规则或复杂的轨迹规划，这需要大量特定任务的工程和人力投入。我们提出了OneDemo，一个强化学习框架，旨在学习如何编辑单一成功示范以适应新的任务配置。我们将示范编辑形式化为一个结构化的强化学习问题，在这个问题中，编辑器从稀疏的任务成功奖励中学习何时在以对象为中心的轨迹段之间过渡，以及如何在几何上调整每个段落。所学习的编辑器既作为任务策略，也作为可重用的示范生成器。在15个Meta-World任务中，OneDemo在平均53.7k环境交互中达到了92.7%的成功率。在9个DexMimicGen任务中，它生成的成功示范平均成功率为98.4%，而在生成数据上训练的BC-RNN策略在匹配数据集规模下达到了99.6%的平均成功率。这些结果表明，学习编辑单一成功示范为策略学习和可扩展示范生成提供了一个结构化且有效的基础。

View on arXiv Download PDF AI Translation

cs.RO / 217 / 2610.18194

Cross-Embodiment Affordance Transfer for Robotic Manipulation

跨体现赋能转移用于机器人操作

Yaseen, Sawera, Thill, Serge, Raja, Rekha

Abstract

Robotic manipulation systems commonly learn object affordances from human-annotated datasets, which reflect interaction possibilities for a human hand rather than the physical constraints of a specific robot. For example, a surface region of an object identified as graspable by a human annotator may be geometrically unreachable or mechanically infeasible for a robot, yet this mismatch is rarely addressed explicitly. To tackle this challenge, we proposed a framework that translates human affordance annotations into robot-executable affordance predictions conditioned on the physical properties of the robot. The network predicts which surface regions of an object are physically executable by the robot, given a 3D point cloud of an object and a descriptor encoding the robot's geometry and constraints. The model was trained using human-annotated affordances together with robotic execution outcomes from multiple sampled approach trials in simulation, where execution outcomes determined which regions are executable for the robot. The evaluation was performed on a household objects dataset using two parallel-jaw gripper configurations on a UR5e robot arm. The framework outperformed a model that does not receive information about the robot. The experimental results demonstrated that affordance prediction must be conditioned on the robot and that human annotations alone are insufficient for learning robot-executable affordance representations.

Chinese Translation

机器人操作系统通常从人类标注的数据集中学习物体赋能，这些数据集反映了人类手部的交互可能性，而非特定机器人所面临的物理约束。例如，人类标注者识别为可抓取的物体表面区域，可能在几何上对机器人不可达或在机械上不可行，但这种不匹配很少被明确解决。为了解决这一挑战，我们提出了一个框架，该框架将人类赋能标注转换为基于机器人物理属性的可执行赋能预测。该网络预测在给定物体的3D点云和描述机器人几何形状及约束的描述符的情况下，哪些物体表面区域是机器人可以物理执行的。该模型使用人类标注的赋能以及来自多个采样接近试验的机器人执行结果进行训练，其中执行结果决定了哪些区域对机器人是可执行的。评估在一个家庭物体数据集上进行，使用UR5e机器人臂的两种平行夹爪配置。该框架的表现优于未接收机器人信息的模型。实验结果表明，赋能预测必须以机器人为条件，而单靠人类标注不足以学习机器人可执行的赋能表示。

View on arXiv Download PDF AI Translation

cs.RO / 218 / 2610.18307

I2O2I-Nav: A Benchmark with a Modular Agent for Long-Horizon Indoor-to-Outdoor-to-Indoor Navigation

I2O2I-Nav：一个具有模块化智能体的长时间室内到室外再到室内导航基准

Shu, Zhengjie, Li, Wanyu, Xiao, Di, Jiang, Lingxin, Jiang, Lai, Hu, Kaidi, Tang, Xincheng, Hu, Wenkang, Wang, Haoran, Zhang, Jinchuan, Chen, Chang, Li, Wei, Pan, Jia, Yang, Ruigang

Abstract

Service and delivery robots often need to complete long-horizon navigation missions spanning indoor and outdoor environments. Existing embodied navigation benchmarks primarily evaluate isolated stages or partial cross-domain journeys, providing limited support for assessing how heterogeneous navigation capabilities compose under continuous execution. We introduce I2O2I-Nav, a physics-enabled benchmark of 20 real-world composite scenes spanning over 115,000 square meters across multiple buildings and floors, with aligned 3DGS rendering and collision geometry. Its three-level evaluation progresses from independent primitives to two-stage compositions and complete four-stage I2O2I missions. Each stage starts from its predecessor's actual endpoint, while stage-conditional metrics track completion and localize failures along the mission. We further propose the MODULAR-Agent that coordinates frozen navigation policies through completion verification, stage handoffs, and bounded recovery. Evaluation across 1,374 episodes highlights the difficulty of continuous composition. The MODULAR-Agent achieves 40.5-76% success on individual primitives, but full I2O2I success remains low. Compared with a fixed policy chain using the same underlying policies, it more than doubles success on both Indoor-to-Outdoor (I2O) and Outdoor-to-Indoor (O2I), while improving full-mission success from 1.15% to 4.02%. Stage-wise analysis shows that the MODULAR-Agent enables more missions to reach the final object-navigation stage, where conditional success remains low, revealing challenges in both long-horizon stage execution and completion decisions.

Chinese Translation

服务和配送机器人通常需要完成跨越室内和室外环境的长时间导航任务。现有的具身导航基准主要评估孤立的阶段或部分跨域旅程，提供的支持有限，难以评估异构导航能力在连续执行中的组合效果。我们提出了I2O2I-Nav，这是一个物理启用的基准，涵盖了20个真实世界的复合场景，跨越多个建筑和楼层，总面积超过115,000平方米，并具有对齐的3DGS渲染和碰撞几何。其三层评估从独立原语进展到两阶段组合和完整的四阶段I2O2I任务。每个阶段都从其前一个阶段的实际终点开始，而阶段条件指标则跟踪完成情况并定位任务中的失败。我们进一步提出了MODULAR-Agent，该智能体通过完成验证、阶段交接和有界恢复来协调冻结的导航策略。对1,374个回合的评估突显了连续组合的难度。MODULAR-Agent在单个原语上的成功率为40.5%-76%，但完整的I2O2I成功率仍然较低。与使用相同基础策略的固定策略链相比，它在室内到室外（I2O）和室外到室内（O2I）上的成功率均翻倍，同时将完整任务的成功率从1.15%提高到4.02%。阶段分析表明，MODULAR-Agent使更多任务能够达到最终的目标导航阶段，但条件成功率仍然较低，揭示了长时间阶段执行和完成决策中的挑战。

View on arXiv Download PDF AI Translation

cs.RO / 219 / 2610.18321

CoSMO: Conformal Semantic Mapping for Open-World Safe Robot Navigation

CoSMO：开放世界安全机器人导航的符合语义映射

Wang, Shuqi, Zheng, Yiwei, Zhang, Pingrui, Gao, Yue, Yin, Xiang

Abstract

Safe autonomous navigation requires not only geometric obstacle avoidance but also semantic awareness. Different object categories demand different safety clearances, and the set of relevant categories may expand as the robot explores novel environments. Vision-language models (VLMs) offer rich semantic representations for building maps of unknown scenes, yet their confidence scores are poorly calibrated, making raw predictions unreliable for safety-critical planning. Conformal prediction (CP) can provide coverage guarantees, but existing methods assume a fixed label set and stationary data, limiting their use in open-world settings. We propose Conformal Semantic Mapping for Open-World Safe Robot Navigation (CoSMO), a semantic safety mapping framework that transforms VLM embeddings into calibrated coverage sets and safety-aware obstacle dilations over a dynamically expanding hazard vocabulary, even under distribution shift. Experiments on the Matterport3D (MP3D) dataset show that CoSMO reduces semantic collisions by 61.6% relative to a geometry-only baseline, while maintaining formal coverage guarantees even under delayed expert feedback. Real-world experiments on a quadruped robot further demonstrate generalization by achieving long-run empirical coverage of 90.73% against the 90% target. Project website: https://yangming911.github.io/cosmo-page/.

Chinese Translation

安全的自主导航不仅需要几何障碍物规避，还需要语义意识。不同的物体类别要求不同的安全间距，并且随着机器人探索新环境，相关类别的集合可能会扩展。视觉-语言模型（VLMs）为构建未知场景的地图提供了丰富的语义表示，但它们的置信度评分校准不佳，使得原始预测在安全关键的规划中不可靠。符合预测（CP）可以提供覆盖保证，但现有方法假设固定的标签集和静态数据，限制了它们在开放世界环境中的使用。我们提出了开放世界安全机器人导航的符合语义映射（CoSMO），这是一个语义安全映射框架，它将VLM嵌入转换为校准的覆盖集和安全意识障碍物膨胀，甚至在分布变化的情况下也能动态扩展危险词汇。对Matterport3D（MP3D）数据集的实验表明，CoSMO相较于仅使用几何的基线减少了61.6%的语义碰撞，同时在延迟专家反馈的情况下仍保持正式的覆盖保证。在四足机器人上的真实世界实验进一步展示了其泛化能力，实现了90.73%的长期经验覆盖，接近90%的目标。项目网站：https://yangming911.github.io/cosmo-page/

View on arXiv Download PDF AI Translation

cs.RO / 220 / 2610.18352

Let the Mind Conceive, Let the Hand Achieve: Following Multimodal Instructions for Long-Horizon Robotic Manipulation

让思维构思，让双手实现：遵循多模态指令进行长时间机器人操控

Wen, Youpeng, Wang, Gang, Huang, Tianyu, Zhang, Weiwei, Huang, Zhijian, Chen, Wei, Chen, Xieyuanli, Liu, Yunhui

Abstract

A robot following a multi-step instruction needs to act on the right objects in the right order, even as the scene changes. It must also maintain awareness of task progress and determine when to advance. To address these challenges, we introduce Orchestra, an agent harness that separates task planning and execution management from action generation. Orchestra decomposes a complete multimodal instruction into ordered subtasks, tracks completed steps, and updates the locations of relevant objects and destinations during execution. It maintains a shared context containing the active subtask, spatial grounding, and recent observations. An action policy uses this context to generate actions, while a separately trained progress evaluator estimates when to advance to the next subtask. To evaluate whether robots trained on atomic instructions and short trajectories can execute longer subtask sequences, we introduce a benchmark named RMBench-Compose. Orchestra achieves a mean success rate of $68.22\%$ across nine RMBench tasks, outperforming Mem-0 by $26.22$ percentage points. On RMBench-Compose, it achieves an Average Success Length (ASL) of $7.35$ subtasks, $2.46$ times that of HarnessVLA. Across four real-world multi-step tasks, it improves mean ASL by $25.0\%$ over Orchestra using the base $\pi_{0.5}$ policy. These results show that Orchestra helps robots carry out longer sequences of subtasks while preserving the requested subtask order and acting on the specified objects. Our code and project page are available at https://youpengwen.github.io/Orchestra.

Chinese Translation

一个遵循多步骤指令的机器人需要在场景变化时，按照正确的顺序对正确的物体进行操作。它还必须保持对任务进展的意识，并判断何时推进。为了解决这些挑战，我们提出了Orchestra，一个将任务规划与执行管理从动作生成中分离的代理框架。Orchestra将完整的多模态指令分解为有序的子任务，跟踪已完成的步骤，并在执行过程中更新相关物体和目标的位置。它维护一个共享上下文，包含当前活动的子任务、空间定位和最近的观察。一个动作策略利用这个上下文生成动作，而一个单独训练的进度评估器则估计何时推进到下一个子任务。为了评估在原子指令和短轨迹上训练的机器人是否能够执行更长的子任务序列，我们引入了一个名为RMBench-Compose的基准测试。Orchestra在九个RMBench任务中实现了$68.22\%$的平均成功率，超越了Mem-0 $26.22$个百分点。在RMBench-Compose中，它的平均成功长度（ASL）为$7.35$个子任务，是HarnessVLA的$2.46$倍。在四个真实世界的多步骤任务中，它的平均ASL比使用基础$ ext{π}_{0.5}$策略的Orchestra提高了$25.0\\%$。这些结果表明，Orchestra帮助机器人执行更长的子任务序列，同时保持所请求的子任务顺序并对指定的物体进行操作。我们的代码和项目页面可在https://youpengwen.github.io/Orchestra获取。

View on arXiv Download PDF AI Translation

cs.RO / 221 / 2610.18363

Visual Sevoing from Lines

基于直线的视觉伺服控制

Andreff, Nicolas, Espiau, Bernard, Horaud, Radu

Abstract

In this work, we present a fundamentally new approach to visual servoing using lines. It is based on a theoretical and geometrical study of the main line representations which allowed us to define a new representation, the so-called binormalized Plucker coordinates. They are particularly well suited to visual servoing. Indeed, they allow the definition of a proper image line alignment concept. The control law which realizes such an alignment has moreover several properties: partial decoupling between rotation and translation, analytical inversion of the motion equations and global asymptotic stability conditions. This control law was validated both in simulation and experimentally in the specific case of an orthogonal trihedron.

Chinese Translation

在本研究中，我们提出了一种基于直线的视觉伺服控制的新方法。该方法基于对主要直线表示的理论和几何研究，使我们能够定义一种新的表示法，即所谓的双法线普克坐标（binormalized Plucker coordinates）。这种表示法特别适合于视觉伺服控制。实际上，它允许定义一个合适的图像直线对齐概念。此外，实现这种对齐的控制律具有多个特性：旋转与平移之间的部分解耦、运动方程的解析逆解以及全局渐近稳定性条件。该控制律在仿真和实验中均得到了验证，特别是在正交三棱柱的特定情况下。

View on arXiv Download PDF AI Translation

cs.RO / 222 / 2610.18373

DR-VLA: Learning Human-Like Quadruped Navigation from Egocentric Web Videos via Domain-Invariant Scene Representations

DR-VLA：通过领域不变场景表示从自我中心网络视频学习类人四足动物导航

Tian, Jun, Niu, Yingjie, Chen, Ruo, Lin, Hongyi, Li, Chao

Abstract

Quadruped robots performing outdoor deliveries must interpret complex scenes, anticipate changes, and navigate in socially compliant ways. Robot demonstrations are costly to collect at the scale needed for generalization. Egocentric web videos offer abundant human navigation experience, but differences in viewpoint and motion make this experience difficult to transfer to robots. We introduce DR-VLA, a Mixture-of-Transformers framework that transfers human navigation experience through a domain-invariant scene representation (DISR). The framework combines vision-language understanding and latent world modeling with a flow-matching action expert. We introduce EgoLynx, comprising 10,000 hours of curated egocentric web video (EgoLynx-Web) and 2,000 clips of teleoperated robot navigation collected during real-world deliveries (EgoLynx-Robot). We pretrain the understanding and world experts on EgoLynx-Web without action labels, then adapt the model to robot observations and train the action expert on EgoLynx-Robot to predict executable trajectories. We also propose the Quadruped Navigation Score (QNS), a pseudo-closed-loop metric combining safety, instruction compliance, comfort, and progress. On the held-out EgoLynx-Robot test split, DR-VLA improves QNS by approximately 5% and reduces open-loop final heading error (FHE) by 14.6%, each relative to the best competing result for that metric. Ablations show that web-video pretraining improves generalization and that the model's key components contribute to performance. As a future-prediction target, DISR outperforms five generic visual representations and yields performance close to task-specific 3D perception without dedicated geometric labels or point clouds for target construction. Trained separately on NAVSIM v1 without web pretraining, DR-VLA achieves 92.1 PDMS, suggesting that the framework also applies to autonomous driving.

Chinese Translation

进行户外配送的四足机器人必须解读复杂场景、预测变化，并以符合社会规范的方式进行导航。收集机器人演示所需的规模成本高昂，而自我中心的网络视频提供了丰富的人类导航经验，但视角和运动的差异使得这一经验难以转移到机器人上。我们提出了DR-VLA，一个通过领域不变场景表示（DISR）转移人类导航经验的混合变换器框架。该框架结合了视觉-语言理解和潜在世界建模，并配备了流匹配动作专家。我们引入了EgoLynx，包含10,000小时精心策划的自我中心网络视频（EgoLynx-Web）和2,000个在真实配送过程中收集的遥控机器人导航片段（EgoLynx-Robot）。我们在没有动作标签的情况下对EgoLynx-Web上的理解和世界专家进行预训练，然后将模型适应于机器人观察，并在EgoLynx-Robot上训练动作专家以预测可执行轨迹。我们还提出了四足导航评分（QNS），这是一个伪闭环指标，结合了安全性、指令合规性、舒适性和进展。在保留的EgoLynx-Robot测试集上，DR-VLA将QNS提高了约5%，并将开放环最终航向误差（FHE）降低了14.6%，相对于该指标的最佳竞争结果。消融实验表明，网络视频预训练提高了泛化能力，并且模型的关键组件对性能有贡献。作为未来预测目标，DISR的表现优于五种通用视觉表示，并且在没有专用几何标签或点云用于目标构建的情况下，其性能接近任务特定的3D感知。在没有网络预训练的情况下，单独在NAVSIM v1上训练的DR-VLA达到了92.1 PDMS，表明该框架也适用于自动驾驶。

View on arXiv Download PDF AI Translation

cs.RO / 223 / 2610.18381

Benchmarking Interaction-Oriented Aerial Mobility across Heterogeneous Environments

跨异构环境的交互导向空中移动基准测试

Duan, Zhibo, Kong, Hanlin, Ma, Tianran, Wu, Yuze, Xu, Yuxuan, Hoi, Kenghou, Xie, Changheng, Zhou, Xin, Gao, Jianzhe, Gao, Fei, Wang, Wenguan

Abstract

Aerial mobile manipulation requires unmanned aerial vehicles (UAVs) to traverse heterogeneous environments such as indoor and outdoor domains, and connect navigation with subsequent physical interaction within the same trajectories. Existing aerial benchmarks are typically confined to single-environment settings, limiting evaluation of agents across continuous environmental transitions. When interaction is evaluated, agents are often initialized near the target, leaving the transition from navigation to interaction largely unassessed. This work introduces CrossFly, a benchmark for long-horizon interaction-oriented aerial mobility across heterogeneous environments. CrossFly contains 15K trajectories across more than 20 reconstructed scenes under a unified simulation and evaluation framework, covering Indoor, Outdoor, and Indoor$\leftrightarrow$Outdoor trajectories. It further introduces Interaction Readiness Rate (IRR) to assess whether terminal navigation states provide suitable conditions for downstream interaction without assuming a specific manipulation policy. Experiments with five representative aerial vision-language-action (VLA) agents and the proposed CrossFly-Agent show that Indoor$\leftrightarrow$Outdoor settings are more challenging than single-environment settings, with the highest success rate reaching 9.7% on unseen scenes. A downstream manipulation case study further validates the relevance of interaction-oriented arrival to subsequent physical interaction. We hope CrossFly helps advance aerial embodied intelligence toward unified mobility and physical interaction in heterogeneous environments.

Chinese Translation

空中移动操控要求无人机（UAV）在室内和室外等异构环境中穿行，并将导航与后续的物理交互连接在同一轨迹中。现有的空中基准测试通常局限于单一环境设置，限制了对智能体在连续环境过渡中的评估。当评估交互时，智能体通常在目标附近初始化，使得从导航到交互的过渡大部分未被评估。本研究提出了CrossFly，一个用于跨异构环境的长时间交互导向空中移动的基准测试。CrossFly包含超过20个重建场景下的15K条轨迹，采用统一的仿真和评估框架，涵盖室内、室外以及室内$ ightarrow$室外的轨迹。它进一步引入了交互准备率（Interaction Readiness Rate, IRR），用于评估终端导航状态是否为下游交互提供了合适的条件，而不假设特定的操控策略。与五个代表性的空中视觉-语言-动作（VLA）智能体及所提出的CrossFly-Agent进行的实验表明，室内$ ightarrow$室外设置比单一环境设置更具挑战性，未见场景的最高成功率达到9.7%。下游操控案例研究进一步验证了交互导向到达与后续物理交互的相关性。我们希望CrossFly能够推动空中具身智能朝向异构环境中的统一移动和物理交互发展。

View on arXiv Download PDF AI Translation

cs.RO / 224 / 2610.18386

Intent-guided Adaptive Stiffening for Visuomotor Policies under Unknown Resistance

基于意图的自适应加固在未知阻力下的视觉运动策略

Cufino, Francesco, Mariniello, Carmela, Selvaggio, Mario, Sochopoulos, Andreas, Ruggiero, Fabio

Abstract

Manipulation tasks such as unscrewing a cap or sliding a part in contact with a surface require imposing force along a high-friction direction: the robot must overcome a resistance rather than yield to it. Generative visuomotor policies predict the motion for such tasks but not the authority needed to realize it, and that authority is bounded by the impedance controller beneath them: when the resistance exceeds what a compliant controller can exert under a bounded tracking error, the task is unattainable however accurate the prediction is. The required amount cannot be anticipated either, since it depends on a contact state the observation does not reveal. In this paper, we propose a residual stiffening layer placed underneath an unmodified policy. Resistance is recognized, without force sensing, from the mismatch between the motion the policy commands and the motion the robot realizes, and a four-mode supervisor raises the stiffness along a single direction read off the policy's own action chunk, retains it while the task progresses and releases it afterwards. The layer needs no compliance labels, no retraining and no re-collected demonstrations. On two real tasks, unscrewing a bottle cap and pushing a T-nut along a T-slot groove against non-homogeneous friction, evaluated at six resistance levels of which two were withheld from the demonstrations, the layer raises the success rate from 52% to 90% while producing 61% less peak off-axis force than a uniformly stiff controller at the same resistance.

Chinese Translation

操控任务，如拧开瓶盖或在表面上滑动部件，需要沿着高摩擦方向施加力量：机器人必须克服阻力，而不是屈服于它。生成的视觉运动策略预测此类任务的运动，但无法预测实现该运动所需的权威性，而这种权威性受到其下方阻抗控制器的限制：当阻力超过合规控制器在有限跟踪误差下能够施加的力量时，无论预测多么准确，任务都是无法实现的。所需的力量也无法预见，因为它依赖于观察无法揭示的接触状态。本文提出了一种残余加固层，置于未修改策略的下方。通过策略命令的运动与机器人实际实现的运动之间的不匹配，识别出阻力，而无需力传感器，并且一个四模式的监督器沿着从策略自身动作块读取的单一方向提高刚度，在任务进行时保持该刚度，并在之后释放。该层不需要合规标签、不需要重新训练，也不需要重新收集演示。在两个真实任务中，拧开瓶盖和在非均匀摩擦下沿T型槽推送T型螺母，在六个阻力水平下进行评估，其中两个未包含在演示中，该层将成功率从52%提高到90%，同时在相同阻力下产生的峰值离轴力比均匀刚性控制器减少了61%。

View on arXiv Download PDF AI Translation

cs.RO / 225 / 2610.18425

CALM: Decentralized Control of Cooperative Aerial Manipulation with Multi-Agent Reinforcement Learning

CALM：基于多智能体强化学习的协作空中操控的去中心化控制

Huo, Jiaxi, Ricioppo, Petre, Lizzio, Fausto Francesco

Abstract

Cooperative aerial manipulation is difficult under wind, payload motion, and noisy goal broadcasts because multiple Unmanned Aerial Vehicles (UAVs) are coupled through a shared payload. We present CALM (Cooperative Aerial Learned Manipulation), a decentralized controller for aerial manipulators carrying a basket while tracking a moving ground robot. CALM combines a parameter-shared Multi-Agent Reinforcement Learning (MARL) policy, trained using Multi-Agent Proximal Policy Optimization (MAPPO), with goal estimation, per-UAV Extended State Observers (ESOs), basket-mass estimation, and anticipatory formation-yaw feedforward. The anticipatory formation-yaw feedforward uses estimated goal motion to rotate the formation before position error develops, reducing basket-yaw lag without adding net commanded force. A disturbance-conditioned critic supports centralized training while execution remains decentralized. Across held-out trajectories, CALM achieves lower mean position and orientation errors than all evaluated baselines, including the matched arm baseline (7.68 versus 10.95 cm and 3.3 versus 4.7$^\circ$), the cable-suspended realization, and two model-based controllers. Under unseen full-information-path degradation, CALM remains statistically comparable or superior to the arm baseline. Videos are available at https://petreric.github.io/CALM/.

Chinese Translation

在风力、负载运动和噪声目标广播的影响下，协作空中操控面临挑战，因为多个无人机（UAV）通过共享负载相互耦合。我们提出了CALM（协作空中学习操控），这是一种用于携带篮子的空中操控器的去中心化控制器，同时跟踪移动的地面机器人。CALM结合了使用多智能体近端策略优化（MAPPO）训练的参数共享多智能体强化学习（MARL）策略、目标估计、每个无人机的扩展状态观察器（ESO）、篮子质量估计和预期的队形偏航前馈。预期的队形偏航前馈利用估计的目标运动在位置误差发展之前旋转队形，从而减少篮子偏航滞后，而不增加净指令力。一个基于干扰的评论者支持集中训练，同时执行保持去中心化。在保留的轨迹中，CALM实现了比所有评估的基线更低的平均位置和方向误差，包括匹配臂基线（7.68厘米对比10.95厘米，3.3度对比4.7度）、电缆悬挂实现和两个基于模型的控制器。在未见的全信息路径退化下，CALM在统计上与臂基线相当或更优。视频可在 https://petreric.github.io/CALM/ 查看。

View on arXiv Download PDF AI Translation

cs.RO / 226 / 2610.18443

PACT: Posture-Aligned Co-Design Technique for Hand Configuration and Retargeting in Multi-Finger Imitation Learning

PACT：用于多指模仿学习中的手部配置与重定向的姿态对齐协同设计技术

Sasazaki, Koyo, Funabashi, Satoshi, Ponnivalavan, Pranav, Schmitz, Alexander, Ogata, Tetsuya, Sugano, Shigeki

Abstract

The fidelity of teleoperated demonstrations used for imitation learning is determined by how faithfully a robot hand can replicate the operator's posture. The posture retargeting law that the robot will execute is typically selected only once the hardware has been finalized, and conventional hand designs often neglect human-like grasping postures. We introduce PACT, an optimization based on human grasping postures that jointly optimizes the mechanical parameters of a 17-servo robot hand (link lengths, joint offsets, mounting angles, and base placements) and its nonlinear posture retargeter (a linear human-to-robot joint map, per-joint posture-IK weights, and damping). The optimization enforces full-finger correspondence (MCP, PIP, DIP, and tip of each finger) to human poses from the DexYCB dataset, and is constrained by servo interference, assembly clearances, and flat-hand protrusion, so the resulting design can be physically assembled. We fabricate the PACT Hand and validate it through teleoperation experiments and an imitation-learning demonstration. The hand reproduces all human hand keypoints with a kinematic RMSE of 3.9 mm, compared to 12.8 mm for the LEAP Hand baseline, and keeps all fingertips within 10 mm on every grasping posture, while the baseline achieves this for only 21.6% of the poses. Under matched per-joint torque limits, it lifts twice the baseline payload, 2 kg against 1 kg, and deforms a held paper cup seven times less, reaching a deformation level comparable to a human hand. It also achieves an 86% grasping success rate on YCB daily objects, compared to 64% for the baseline. Finally, as a demonstration, an imitation-learning policy trained on demonstrations collected with the PACT Hand achieved a 75% success rate at held-out placements in a can pick-and-place task. Project page: https://pact-hand-2027.github.io

Chinese Translation

用于模仿学习的远程操作演示的真实性取决于机器人手能够多大程度上忠实地复制操作员的姿势。机器人将执行的姿势重定向法则通常在硬件最终确定后才会选择，而传统的手部设计往往忽视类人抓握姿势。我们提出了PACT，这是一种基于人类抓握姿势的优化方法，联合优化17伺服机器人手的机械参数（连杆长度、关节偏移、安装角度和基座位置）及其非线性姿势重定向器（线性人机关节映射、每关节姿势逆运动学权重和阻尼）。该优化强制实现手指的全指对应（每个手指的MCP、PIP、DIP和指尖）与DexYCB数据集中人类姿势的对应，并受到伺服干扰、装配间隙和手掌平面突出等因素的限制，从而确保所得到的设计可以物理组装。我们制造了PACT手，并通过远程操作实验和模仿学习演示对其进行了验证。与LEAP手基线相比，该手以3.9毫米的运动学均方根误差（RMSE）重现了所有人类手部关键点，而LEAP手的RMSE为12.8毫米，并且在每个抓握姿势中，所有指尖保持在10毫米以内，而基线仅在21.6%的姿势中实现这一点。在匹配的每关节扭矩限制下，它能够提升两倍于基线的负载，2千克对比1千克，并且对所持纸杯的变形程度减少了七倍，达到了与人类手相当的变形水平。它在YCB日常物品上的抓取成功率为86%，而基线为64%。最后，作为演示，使用PACT手收集的演示训练的模仿学习策略在罐子拾取与放置任务中的保留位置上达到了75%的成功率。项目页面：https://pact-hand-2027.github.io

View on arXiv Download PDF AI Translation

cs.RO / 227 / 2610.18478

MemoryVLN: From Ego-Aware Memory Representation to Cross-Context Memory Reuse in Vision-Language Navigation

MemoryVLN：从自我意识的记忆表征到视觉-语言导航中的跨上下文记忆重用

Wei, Xueyao, Wang, Yifei, Li, Delun, Wang, Shaoan, Yu, Junzhi, Song, Wenjie, Chen, Xingyu

Abstract

Large vision-language models (LVLMs) have shown strong potential for embodied vision-language navigation (VLN), yet their performance degrades on long-horizon trajectories due to context-length limits, redundant historical observations, and unstable spatial alignment across time. These issues make it difficult to both compress long navigation histories and reliably reuse past spatial experience. To address these limitations, we propose \textbf{MemoryVLN}, a unified memory-driven navigation system built around memory construction and management. During memory construction, \emph{Long Sequence Compression} employs forgetting-curve-inspired temporal sampling to preserve dense recent evidence and sparse yet representative distant observations under a limited token budget. \emph{Ego-Aware Memory Representation} further integrates 2D semantic tokens, 3D geometric tokens, and 6D agent poses, allowing each memory to encode both what was observed and where it was observed. During memory management, \emph{Hierarchical Memory Bank} maintains in-episode and context-level memories through cross-context warm-start and confidence-controlled write-back. \emph{Memory Enhancement} then retrieves the Top-$K$ relevant memories according to the current navigation state and fuses them with the current visual tokens. Together, these mechanisms transform raw navigation history into compact, spatially grounded, selectively retrievable, and reusable navigation experience. Extensive experiments on R2R-CE and RxR-CE demonstrate consistent improvements over the state-of-the-art baselines by a substantial margin (i.e., 2\% and 6.5\% in successful rate).

Chinese Translation

大型视觉-语言模型（LVLMs）在具身视觉-语言导航（VLN）中展现出强大的潜力，但由于上下文长度限制、冗余的历史观察以及随时间变化的不稳定空间对齐，其在长时间轨迹上的表现有所下降。这些问题使得压缩长导航历史和可靠地重用过去的空间经验变得困难。为了解决这些局限性，我们提出了 extbf{MemoryVLN}，一个围绕记忆构建和管理的统一记忆驱动导航系统。在记忆构建过程中， extit{长序列压缩}采用受遗忘曲线启发的时间采样，以在有限的标记预算下保留密集的近期证据和稀疏但具有代表性的远期观察。 extit{自我意识记忆表征}进一步整合了2D语义标记、3D几何标记和6D智能体姿态，使每个记忆能够编码观察到的内容及其观察位置。在记忆管理过程中， extit{层次记忆库}通过跨上下文的热启动和信心控制的写回，维护剧集内和上下文级别的记忆。 extit{记忆增强}则根据当前导航状态检索Top-$K$相关记忆，并将其与当前视觉标记融合在一起。这些机制共同将原始导航历史转化为紧凑的、空间上扎根的、可选择检索和重用的导航经验。在R2R-CE和RxR-CE上的大量实验表明，相较于最先进的基线，MemoryVLN在成功率上有显著的提升（即2 ext{%}和6.5 ext{%}）。

View on arXiv Download PDF AI Translation

cs.RO / 228 / 2610.18479

RTISSLAM: In-air Imaging Sonar SLAM using a Robust Pose Graph and Occupancy Grid Mapping

RTISSLAM：基于稳健位姿图和占据网格映射的空中成像声纳SLAM

Steckel, Jan

Abstract

Sonar was one of the first sensors with which mobile robots built maps, and in-air sonar keeps working where cameras and lidar fail, in darkness, smoke, dust and next to glass. Classical sonar SLAM, however, relied on geometric features or occupancy grids built from sparse range readings, and both struggle with the specular reflections that dominate indoor sonar. In this letter, we introduce RTISSLAM, a novel sonar-only SLAM system for an embedded 3D imaging sonar, which combines the robust pose graph of BatSLAM 2.0 as back-end with an occupancy grid as map. Energyscapes from delay-multiply-and-sum beamforming with a range-wise CFAR serve as place descriptors for sequence-verified loop closures, acoustic odometry calibrates the wheel odometry online, recognized revisits are registered metrically, and the grid is stored per keyframe and anchored to the pose-graph nodes, so that it follows every loop closure. In simulated worlds with specular walls, RTISSLAM reduces the trajectory error of the wheel odometry from \SI{8.2}{\meter} to \SI{0.31}{\meter} in an office, while lidar-style grid SLAM on the same sonar data fails.

Chinese Translation

声纳是移动机器人构建地图的第一批传感器之一，而空中声纳在黑暗、烟雾、灰尘和靠近玻璃的环境中仍然有效，而相机和激光雷达则无法工作。然而，经典的声纳SLAM依赖于几何特征或从稀疏距离读数构建的占据网格，这两者在室内声纳中都面临着以镜面反射为主的挑战。在本文中，我们介绍了RTISSLAM，一种针对嵌入式3D成像声纳的全新声纳SLAM系统，它将BatSLAM 2.0的稳健位姿图作为后端，与占据网格结合作为地图。通过延迟乘法和求和波束形成得到的能量景观与范围方向的CFAR一起，作为序列验证回环闭合的位置信息描述符，声学里程计在线校准轮子里程计，识别的重访以度量方式注册，网格按关键帧存储并锚定到位姿图节点，以便在每次回环闭合时跟随。在具有镜面墙的模拟世界中，RTISSLAM将轮子里程计的轨迹误差从8.2米降低到0.31米，而在相同声纳数据上，激光雷达风格的网格SLAM则失败。

View on arXiv Download PDF AI Translation

cs.RO / 229 / 2610.18523

APTree: Extended Behavior Trees for Adaptive and Scalable Automated Robotic Task Planning in Construction

APTree：用于建筑领域自适应和可扩展自动化机器人任务规划的扩展行为树

Sherkat, Shermin, Wortmann, Thomas, Wortmann, Andreas

Abstract

Task planning for robots in off-site construction prefabrication is primarily manual, leading to planning inefficiencies and reluctance to adopt robotic solutions. Existing approaches provide limited support for scalable, modular automated task planning and localized runtime replanning. This paper proposes APTree, a model-driven framework that integrates automated task planning with modular task execution using behavior trees as the execution model. APTree repeatedly applies and coordinates heterogeneous symbolic planners to automatically generate and compose modular task plans at runtime. To support this approach, we develop a domain-specific language for the formal specification and validation of task plans, together with a compatible execution engine. We evaluate APTree on the real-world livMatS Biomimetic Shell pavilion, where it models the construction robotic fabrication domain, detects injected specification and planning errors at design time and runtime, and automatically generates large-scale task plans by coordinating three symbolic planners across numerous planning modules. A machine-checked formal analysis further shows that composed execution and local replanning are safe under explicitly stated assumptions, including planner soundness and non-interference of generated plans, providing practitioners with explicit conditions for dependable plan composition. APTree thus enables construction practitioners to leverage automated task planning without requiring detailed knowledge of symbolic planning techniques or APTree's internal implementation.

Chinese Translation

在离线建筑预制中的机器人任务规划主要是手动进行，这导致了规划效率低下和对机器人解决方案的采纳不情愿。现有的方法对可扩展的模块化自动任务规划和局部运行时重新规划的支持有限。本文提出了APTree，一个模型驱动的框架，它将自动任务规划与模块化任务执行相结合，使用行为树作为执行模型。APTree重复应用并协调异构符号规划器，以自动生成和组合运行时的模块化任务计划。为了支持这一方法，我们开发了一种领域特定语言，用于任务计划的形式化规范和验证，并配备了兼容的执行引擎。我们在真实世界的livMatS仿生壳亭上评估了APTree，该亭模型化了建筑机器人制造领域，在设计时和运行时检测注入的规范和规划错误，并通过协调三个符号规划器跨多个规划模块自动生成大规模任务计划。机器检查的形式分析进一步表明，在明确声明的假设下，包括规划器的可靠性和生成计划的非干扰，组合执行和局部重新规划是安全的，为从业者提供了可靠计划组合的明确条件。因此，APTree使建筑从业者能够利用自动任务规划，而无需详细了解符号规划技术或APTree的内部实现。

View on arXiv Download PDF AI Translation

cs.RO / 230 / 2610.18548

Behavioural self-monitoring for autonomous earth observation agents: The attribute reliability monitor (ARM)

自主地球观测代理的行为自我监测：属性可靠性监测器（ARM）

Mousist, Alejandro D.

Abstract

Language and vision language models can support autonomous systems, but changing operating conditions may compromise the reliability of their semantic outputs. This work introduces the Attribute Reliability Monitor (ARM), a framework that accumulates evidence of suspicion over time to assess individual semantic attributes. ARM accommodates different detectors and temporal monitoring methods, while leaving intervention policies to the consuming system. An initial implementation was evaluated using a specialized vision language model developed for satellite thermal infrared imagery. Monitoring followed a black box approach, using only observable responses without access to internal model states, confidence scores, or external reference answers. Under controlled input degradation, disagreement between responses provided evidence for selective suspension of attributes. The results illustrate how external behavioral monitoring can inform continued use of model outputs, while distinguishing operational suspicion from semantic correctness.

Chinese Translation

语言和视觉语言模型可以支持自主系统，但变化的操作条件可能会影响其语义输出的可靠性。本研究介绍了属性可靠性监测器（Attribute Reliability Monitor, ARM），这是一个通过时间积累怀疑证据以评估个别语义属性的框架。ARM 适应不同的检测器和时间监测方法，同时将干预政策留给消费系统。初步实现使用了一种专门为卫星热红外图像开发的视觉语言模型进行评估。监测采用黑箱方法，仅使用可观察的响应，而不访问内部模型状态、置信度分数或外部参考答案。在控制输入退化的情况下，响应之间的不一致提供了选择性暂停属性的证据。结果表明，外部行为监测如何能够为模型输出的持续使用提供信息，同时区分操作怀疑与语义正确性。

View on arXiv Download PDF AI Translation

cs.RO / 231 / 2610.18568

WAVE-0: Bridging VLA Policies and WAMs with Lightweight Latent Future Prediction

WAVE-0：通过轻量级潜在未来预测连接视觉-语言-动作政策与世界动作模型

Wang, Haomin, Ji, Anqiong, Wang, Ke, Liu, Zhongshan, Ying, Ru, Zheng, Ran, Zhang, Zhipeng

Abstract

Vision-language-action (VLA) policies excel at generalist robotic manipulation but generate actions without explicitly modeling their environmental consequences. World action models (WAMs) incorporate future prediction, yet most rely on costly video generation, while latent alternatives either use predicted futures only as auxiliary targets or infer latent actions from suboptimal visual transitions through costly large-scale video pretraining. We introduce WAVE-0 (World-model-Augmented VLA Execution), a framework that bridges VLAs and WAMs through a lightweight latent world model. By connecting the semantic understanding of pretrained VLAs with explicit future prediction, WAVE-0 allows anticipated action consequences to directly guide action generation. Built on a pretrained VLA, its 257M world model translates VLM-generated latent action plans into predicted future representations that condition the action expert, unifying latent planning, future prediction, and action generation. The world model is trained directly on ground-truth robot actions with a counterfactual ranking objective, requiring neither pixel reconstruction nor large-scale video pretraining. WAVE-0 achieves success rates of 91.47% on RoboTwin 2.0 and 98.7% on LIBERO, along with 88.0%, 61.4%, and 75.0% on LIBERO-Plus, LIBERO-Pro, and LIBERO-Para, respectively. On real-world tasks, it improves the success rate of $\pi_{0.5}$ from 38.3% to 73.3%. These results demonstrate strong manipulation performance and generalization across diverse distribution shifts.

Chinese Translation

视觉-语言-动作（VLA）政策在通用机器人操作方面表现出色，但在生成动作时并未明确建模其环境后果。世界动作模型（WAMs）结合了未来预测，但大多数依赖于昂贵的视频生成，而潜在替代方案要么仅将预测的未来作为辅助目标，要么通过成本高昂的大规模视频预训练从次优视觉过渡中推断潜在动作。我们提出了WAVE-0（世界模型增强的VLA执行），这是一个通过轻量级潜在世界模型连接VLA和WAM的框架。通过将预训练VLA的语义理解与明确的未来预测相结合，WAVE-0使得预期的动作后果能够直接指导动作生成。基于预训练的VLA，其257M的世界模型将VLM生成的潜在动作计划转化为条件化动作专家的预测未来表示，实现潜在规划、未来预测和动作生成的统一。该世界模型直接在真实机器人动作上进行训练，采用反事实排名目标，既不需要像素重建，也不需要大规模视频预训练。WAVE-0在RoboTwin 2.0上的成功率达到91.47%，在LIBERO上为98.7%，在LIBERO-Plus、LIBERO-Pro和LIBERO-Para上的成功率分别为88.0%、61.4%和75.0%。在真实世界任务中，它将$ ext{π}_{0.5}$的成功率从38.3%提高到73.3%。这些结果展示了强大的操作性能和在多样化分布变化中的泛化能力。

View on arXiv Download PDF AI Translation

cs.RO / 232 / 2610.18596

ACCRUE: Continual Off-Policy Reinforcement Learning for Robot Foundation Models

ACCRUE：用于机器人基础模型的持续离线强化学习

Kim, Changyeon, Lee, Kimin, Narang, Yashraj, Guo, Yijie

Abstract

Reinforcement learning (RL) post-training improves robot foundation models (RFMs) beyond their demonstrations, but real-world deployment demands more: learning new tasks efficiently while retaining previously acquired behaviors. Existing continual RL methods on top of VLAs mitigate forgetting but rely on on-policy updates with limited forward transfer, whereas off-policy methods improve single-task sample efficiency without addressing learner persistence across tasks. We propose ACCRUE, a continual off-policy RL framework that carries lightweight actor and critic networks across tasks, allowing each stage to reuse behavioral and value knowledge acquired through prior RL. The inherited critic adapts only on current-task transitions and guides current-task actor learning, while prior behavior is preserved by distilling the actor toward cached targets on a small number of retained episodes. Fixed visual features keep the inherited learner and its replay compatible with base-policy updates. Across contact-rich assembly sequences in simulation and on real-robot hardware, ACCRUE accelerates subsequent-task learning with near-zero forgetting and significantly reduces real-robot training time by 31.0\% to 68.5\%.

Chinese Translation

强化学习（RL）后训练能够在机器人基础模型（RFM）上超越其演示效果，但实际部署要求更多：在保留先前获得的行为的同时高效学习新任务。现有的基于变换学习算法（VLA）的持续强化学习方法虽然减轻了遗忘，但依赖于策略更新，前向迁移有限，而离线方法则提高了单任务样本效率，但未能解决跨任务学习者的持续性问题。我们提出了ACCRUE，一个持续的离线强化学习框架，它在任务之间携带轻量级的演员和评论家网络，使每个阶段能够重用通过先前强化学习获得的行为和价值知识。继承的评论家仅在当前任务的转移上进行适应，并指导当前任务的演员学习，而通过将演员蒸馏到少量保留的剧集中的缓存目标来保留先前的行为。固定的视觉特征使得继承的学习者及其重放与基础策略更新兼容。在模拟和真实机器人硬件上的接触丰富的装配序列中，ACCRUE加速了后续任务的学习，几乎没有遗忘，并显著减少了真实机器人训练时间，降低幅度为31.0 ext{%}至68.5 ext{%}。

View on arXiv Download PDF AI Translation

cs.RO / 233 / 2610.18637

Grid or Nodes? Probing Object Memory in World Models Under Partial Observability

网格还是节点？在部分可观测性下探究世界模型中的物体记忆

McMillen, Justin, Yilmaz, Yasin

Abstract

A robot's world model must remember objects that leave the camera's view. FloWM, a learned world model, keeps them in a spatial grid of features that it moves with five built-in velocities. A node memory instead keeps each object's position and velocity as numbers. We compare them in FloWM's own block world, with an oracle-fed Kalman tracker as the node memory. How the grid is read changes its score: a single-cell probe finds 66% of out-of-view blocks, while FloWM's own probe 89%. FloWM also scores 8 points higher in view although no frame is shown, which we attribute to its write-back of its own predictions. Out of view, the tracker reaches FloWM's range in 14KB against 5.6MB, with no probe. On diagonal motion, outside its five velocities, FloWM finds 11% of blocks even after fine-tuning, while the unchanged tracker keeps 93%. Faster blocks defeat both until one more allowed speed takes the tracker to 94%, where FloWM's built-in shifts would need 25 copies of its grid. A learned node memory must still replace the oracles and handle velocity under noise, long gaps, and the bounce.

Chinese Translation

机器人的世界模型必须记住离开摄像头视野的物体。FloWM是一种学习的世界模型，它将物体保存在一个特征的空间网格中，并以五种内置速度移动。节点记忆则将每个物体的位置和速度以数字形式保存。我们在FloWM自己的块世界中对这两者进行了比较，使用一个由oracle提供数据的卡尔曼跟踪器作为节点记忆。网格的读取方式改变了其评分：单个单元探测器能够找到66%的视野外块，而FloWM自己的探测器则能找到89%。尽管没有显示任何帧，FloWM在视野内的得分也高出8分，这归因于其对自身预测的回写。在视野外，跟踪器在14KB的情况下达到了FloWM的范围，而在5.6MB的情况下则没有探测。在对角运动中，超出其五种速度范围，FloWM即使经过微调也只能找到11%的块，而未改变的跟踪器保持93%。更快的块击败了两者，直到允许一种额外速度使跟踪器达到94%，而FloWM的内置移动需要25个网格副本。学习的节点记忆仍然必须替代oracle，并在噪声、长时间间隔和反弹情况下处理速度。

View on arXiv Download PDF AI Translation

cs.RO / 234 / 2610.18650

ForkCast: Predictive Recovery for Vision-Language-Action Policies via Counterfactual Futures

ForkCast：通过反事实未来进行视觉-语言-动作策略的预测性恢复

Wen, Chenru, Zeng, Huajian, Ding, Haokai, Kang, Jian, Duan, Anqing, Zhou, Dawei, Song, Dezhen, Zuo, Xingxing

Abstract

Vision-language-action (VLA) policies perform manipulation tasks using camera images and language instructions, but object slip or visual occlusion can disrupt execution. Recovery is typically learned during policy training or delegated to a monitor that triggers a scripted repair after detecting a failure. We introduce ForkCast, a predictive recovery framework that compares counterfactual futures to determine whether and how to intervene in a frozen VLA policy. At each decision point, ForkCast evaluates three candidates: executing the policy's nominal action chunk (Commit), adjusting the camera viewpoint (Reobserve), and replanning the arm trajectory (Replan). A candidate-conditioned world model forecasts the latent outcome of each candidate at a common future horizon, and a shared task-progress head scores the predicted outcomes. By evaluating the policy's own continuation alongside the repair candidates, a single comparison based on task progress determines both whether and how to intervene, without a failure detector, an intervention threshold, or updates to the VLA backbone. On five simulated RoboTwin 2.0 tasks with injected occlusion and object slip, ForkCast increases pooled success from 13.3% for the frozen $\pi_{0.5}$ policy to 30.0% and transfers to two additional VLA backbones without updating ForkCast. After adaptation on real-robot candidate rollouts, ForkCast also improves success over the base policy under both clean and visibility-limited conditions. Code will be released to benefit the research community.

Chinese Translation

视觉-语言-动作（VLA）策略利用摄像头图像和语言指令执行操作任务，但物体滑动或视觉遮挡可能会干扰执行。恢复通常在策略训练期间学习，或委托给监控器，在检测到失败后触发脚本修复。我们提出了ForkCast，这是一种预测性恢复框架，通过比较反事实未来来确定是否以及如何干预冻结的VLA策略。在每个决策点，ForkCast评估三个候选方案：执行策略的名义动作块（Commit）、调整摄像头视角（Reobserve）和重新规划手臂轨迹（Replan）。候选条件的世界模型在共同的未来视野下预测每个候选方案的潜在结果，而共享的任务进展头对预测结果进行评分。通过评估策略自身的延续与修复候选方案，基于任务进展的单一比较决定了是否以及如何干预，而无需失败检测器、干预阈值或对VLA骨干的更新。在五个注入了遮挡和物体滑动的模拟RoboTwin 2.0任务中，ForkCast将冻结的$ ext{π}_{0.5}$策略的成功率从13.3%提高到30.0%，并在不更新ForkCast的情况下转移到另外两个VLA骨干上。在对真实机器人候选方案的适应后，ForkCast在干净和可见性受限的条件下也提高了基础策略的成功率。代码将发布以惠及研究社区。

View on arXiv Download PDF AI Translation

cs.RO / 235 / 2610.18670

Perception Beyond the Visible: Incompleteness-Aware Scene Understanding for Robot Autonomy

超越可见的感知：面向机器人自主性的意识不完全场景理解

Mohan, Rohit

Abstract

Reliable robot autonomy requires reasoning about the world from partial and noisy observations. However, many perception systems model only visible scene regions, rely on fixed semantic categories, and assume reliable sensor measurements. These assumptions break under occlusion, novel object encounters, and sensor degradation. In this paper, I present a research vision for incompleteness-aware scene understanding. I organize this vision around three fundamental challenges: incomplete evidence, incomplete knowledge, and incomplete sensor reliability. I discuss my research contributions toward addressing these challenges through amodal perception, semantic occupancy forecasting, open-world scene understanding, label-efficient learning, and uncertainty-guided multi-modal fusion. Building on these contributions, I outline future research directions toward belief-centric 4D scene representations. These representations aim to jointly model geometry, semantics, motion, and uncertainty. I also discuss active sensing and safe adaptation strategies that allow robots to reduce uncertainty through interaction. The goal is to develop calibrated and actionable scene representations that support reliable autonomy in dynamic, open-world environments.

Chinese Translation

可靠的机器人自主性需要基于部分和噪声观察进行世界推理。然而，许多感知系统仅建模可见场景区域，依赖固定的语义类别，并假设传感器测量可靠。这些假设在遮挡、新物体遇见和传感器退化的情况下失效。本文提出了一种关于意识不完全场景理解的研究愿景。我围绕三个基本挑战组织这一愿景：不完整的证据、不完整的知识和不完整的传感器可靠性。我讨论了我在通过无模态感知、语义占用预测、开放世界场景理解、标签高效学习和不确定性引导的多模态融合等方面应对这些挑战的研究贡献。在这些贡献的基础上，我概述了未来研究方向，旨在实现以信念为中心的4D场景表示。这些表示旨在共同建模几何、语义、运动和不确定性。我还讨论了主动感知和安全适应策略，使机器人能够通过交互减少不确定性。目标是开发经过校准和可操作的场景表示，以支持在动态开放世界环境中的可靠自主性。

View on arXiv Download PDF AI Translation

cs.RO / 236 / 2610.18697

UniW(A)M: A Unified Mixture-of-Experts Framework for World and World-Action Models

UniW(A)M：一个统一的混合专家框架用于世界模型和世界-动作模型

Xu, Jingbo, Fang, Jianjie, Li, Zhuohang, Song, Zihao, Wang, Zhaolu, Liu, Xvyuan, Zhang, Xin, Su, Haisheng, Tang, Yinzhou, Gao, Chen, Wu, Wei, Li, Yong

Abstract

World models (WMs) and world-action models (WAMs) focus on future world prediction and executable action generation, respectively. Although these models learn shared physical dynamics and interaction semantics from embodied trajectories, they are typically trained in isolation, which restricts knowledge reuse between prediction and control. The central challenge is therefore to develop a unified learning framework that shares these dynamics while preserving the distinct modeling capabilities required by each paradigm. We propose UniW(A)M, a mixture-of-experts world-(action) model built upon diffusion Transformers that unifies action-conditioned world modeling and diverse WAM variants. Shared experts capture common dynamics, while atomic capability experts are reused across interfaces so that each interface keeps its own conditional pathway. At deployment, a plug-in adaptive interface router (PAIR) learns from a small subset of target-task chunk scenarios to select a suitable WAM interface for each execution chunk, balancing task success and execution efficiency while keeping the backbone frozen. We evaluate UniW(A)M on LIBERO, RoboTwin 2.0, and three real-world manipulation tasks, where it achieves 88.3% average real-world success versus 76.7% for the strongest baseline. On WorldArena, it leads action-conditioned world modeling on all three tasks. AC-WM co-training contributes to action generation on every action interface and PAIR raises OOD success from 63.3% to 76.7%.

Chinese Translation

世界模型（WMs）和世界-动作模型（WAMs）分别关注未来世界的预测和可执行动作的生成。尽管这些模型从具身轨迹中学习共享的物理动态和交互语义，但它们通常是孤立训练的，这限制了预测与控制之间的知识重用。因此，中心挑战在于开发一个统一的学习框架，既能共享这些动态，又能保留每种范式所需的独特建模能力。我们提出了UniW(A)M，这是一个基于扩散变换器的混合专家世界（动作）模型，统一了基于动作的世界建模和多样的WAM变体。共享专家捕捉共同的动态，而原子能力专家在不同接口间重用，以便每个接口保持其自身的条件路径。在部署时，一个插件自适应接口路由器（PAIR）从一小部分目标任务块场景中学习，以为每个执行块选择合适的WAM接口，在保持主干不变的同时平衡任务成功率和执行效率。我们在LIBERO、RoboTwin 2.0和三个真实世界的操作任务上评估了UniW(A)M，其在真实世界的平均成功率达到88.3%，而最强基线为76.7%。在WorldArena上，它在所有三个任务中引领了基于动作的世界建模。AC-WM共同训练促进了每个动作接口上的动作生成，PAIR将OOD成功率从63.3%提高到76.7%。

View on arXiv Download PDF AI Translation

cs.RO / 237 / 2610.18699

RoboFQE: Rollout-Free Evaluation of Visual Robot Policies from Offline Trajectories

RoboFQE：基于离线轨迹的视觉机器人策略无回滚评估

Deng, Yunfu, Shin, Sangwoo, Pavse, Brahma S., Hanna, Josiah P.

Abstract

Evaluating robot policies before deployment typically requires producing new trajectories through real-world execution, simulators, or learned world models. These rollout-based approaches impose infrastructure requirements; simulator and world-model evaluators additionally face prediction errors that can compound over the evaluation horizon. In this paper, we study off-policy evaluation as an alternative to rollout-based robot-policy evaluation methods. Specifically, we introduce RoboFQE, an evaluator that estimates the discounted expected return of a target policy from offline robot trajectories without estimating a behavior-policy density, combining fitted Q-evaluation with a pretrained vision foundation model and learned terminal success annotation to bridge the gap between standard OPE benchmarks and continuous-control visual manipulation. Controlled experiments across simulated and real-world manipulation tasks show that RoboFQE produces accurate discounted-return estimates and preserves policy rankings even on long-horizon manipulation tasks where video world-model evaluators are degraded by autoregressive prediction errors. Ablation experiments show that pretrained vision encoders substantially outperform from-scratch alternatives, particularly on long-horizon tasks, and that learned terminal success annotation enables evaluation without ground-truth reward labels with only slight reduction of accuracy.

Chinese Translation

在部署之前评估机器人策略通常需要通过实际执行、模拟器或学习的世界模型生成新的轨迹。这些基于回滚的方法对基础设施有要求；模拟器和世界模型评估者还面临预测误差，这些误差可能在评估范围内累积。在本文中，我们研究了离策略评估作为一种替代基于回滚的机器人策略评估方法。具体而言，我们介绍了RoboFQE，这是一种评估器，它从离线机器人轨迹中估计目标策略的折扣期望回报，而无需估计行为策略密度，结合了拟合Q评估、预训练的视觉基础模型和学习的终端成功注释，以弥合标准OPE基准与连续控制视觉操控之间的差距。在模拟和现实世界操控任务中的受控实验表明，RoboFQE能够生成准确的折扣回报估计，并在长时间范围的操控任务中保持策略排名，即使在视频世界模型评估者因自回归预测误差而退化的情况下。消融实验表明，预训练的视觉编码器在长时间范围任务中显著优于从头开始的替代方案，并且学习的终端成功注释使得在没有真实奖励标签的情况下进行评估成为可能，且仅略微降低了准确性。

View on arXiv Download PDF AI Translation

cs.RO / 238 / 2610.18711

Steertuning Policies from a Few Human Videos

从少量人类视频中调整策略

Yan, Billy, Tu, Isaac, Lee, Kelly, Chang, Blake, Ma, Kenny, Patel, Kanad, Kabra, Dhawal, Patel, Neer, Wu, Kevin, Pinto, Lerrel, Guzey, Irmak

Abstract

Current recipes for robot learning require massive amounts of robot demonstrations to obtain robust and adaptive policies. In this work, we introduce Steertuning, a system that enables robots to learn new tasks in new environments from 5 human demonstrations in under 5 Steertuning consists of two parts. (1) We pretrain a 400M-parameter diffusion policy on our egocentric human dataset, which consists of 100 hours of densely annotated random interactions. This pretraining makes Steertuning robust to background differences and enables spatial and object generalization when deployed on new tasks. (2) Given 5 demonstrations of a new task, we finetune our model for only 2 minutes (2000 training updates) and steer the policy's actions during reverse sampling using the demonstration whose observation is closest to that of the deployment environment. Together with a carefully designed data processing pipeline for embodiment transfer, this enables Steertuning to achieve a 71% success rate across 12 tasks in 12 new environments, ranging from lab settings to kitchens in new homes. The entire process, from data collection for a new task to robot policy deployment, takes less than 5 minutes. Our full dataset, model weights, codebase, and robot videos will be made available at our website: https://steertuning.github.io.

Chinese Translation

当前的机器人学习方法需要大量的机器人演示以获得稳健且适应性强的策略。在本研究中，我们提出了Steertuning，一个系统使机器人能够从5个人工演示中在新环境中学习新任务，且时间不超过5分钟。Steertuning由两个部分组成：(1) 我们在自我中心的人类数据集上预训练了一个包含4亿参数的扩散策略，该数据集包含100小时密集注释的随机交互。这一预训练使Steertuning对背景差异具有鲁棒性，并在部署到新任务时能够实现空间和对象的泛化。(2) 在获得5个新任务的演示后，我们仅用2分钟（2000次训练更新）对模型进行微调，并在反向采样过程中使用与部署环境观察最接近的演示来引导策略的动作。结合精心设计的数据处理流程以实现体现转移，Steertuning在12个新环境中的12个任务上达到了71%的成功率，这些环境从实验室设置到新家中的厨房不等。从新任务的数据收集到机器人策略的部署，整个过程不超过5分钟。我们的完整数据集、模型权重、代码库和机器人视频将会在我们的网站上发布：https://steertuning.github.io。

View on arXiv Download PDF AI Translation

cs.RO / 239 / 2610.18774

GreenWAM: A World Foundation Model for Robotics

GreenWAM：一种用于机器人技术的世界基础模型

Sidorov, Gennady, Kornilov, Valerii, Grankin, Denis, Fedotova, Polina, Postnikov, Aleksey

Abstract

A robot needs a world model that predicts how its actions will change the scene and a policy that chooses those actions. Models are usually built and evaluated for only one of these roles, and every robot records its actions in its own format. We ask whether one robot world model can serve both roles across different bodies, and what makes this work. We present GreenWAM, a world foundation model for robotics, together with the recipe that produces it: (1) a shared anatomical action space that describes 20 robot families, from bimanual arms to dexterous humanoids, in common joint coordinates; (2) joint training on these families in visual, action-conditioned and world-action modes; and (3) policies that split an action expert off the generator. As a world model, GreenWAM attains top-tier trajectory, depth and 3D accuracy on WorldArena 1.0, ahead of every world-action model and video generator in our comparison. As a policy, it scores among the top entries of the RoboTwin 2.0 leaderboard, reaches the highest zero-shot LIBERO-plus success among the compared methods and is competitive with dedicated vision-language-action and world-action policies on CALVIN and LIBERO. We further adapt GreenWAM to a real humanoid, where it succeeds more often than policies adapted from $\pi_{0.5}$ and from Cosmos 3 Edge, and distill it into a single-step policy for real-time control. Controlled experiments yield three findings: robot specialization improves world prediction and control together; an action expert helps when split off at the start of adaptation; and keeping world-model modes during policy adaptation preserves robustness to scene changes.

Chinese Translation

机器人需要一个世界模型来预测其动作如何改变场景，以及一个选择这些动作的策略。模型通常仅为这两个角色中的一个构建和评估，并且每个机器人以自己的格式记录其动作。我们探讨是否可以使用一个机器人世界模型在不同的机体上同时承担这两个角色，以及是什么使其能够实现。我们提出了GreenWAM，一种用于机器人技术的世界基础模型，并提供了其生成的配方：(1) 一个共享的解剖动作空间，描述了20个机器人家族，从双手臂到灵巧的人形机器人，采用共同的关节坐标；(2) 在这些家族上进行视觉、动作条件和世界-动作模式的联合训练；(3) 将动作专家从生成器中分离的策略。作为一个世界模型，GreenWAM在WorldArena 1.0上达到了顶级的轨迹、深度和三维精度，领先于我们比较中的每个世界-动作模型和视频生成器。作为一个策略，它在RoboTwin 2.0排行榜中名列前茅，在比较方法中实现了最高的零-shot LIBERO-plus成功率，并且在CALVIN和LIBERO上与专用的视觉-语言-动作和世界-动作策略具有竞争力。我们进一步将GreenWAM适配到一个真实的人形机器人上，其成功率超过了从$ ext{π}_{0.5}$和Cosmos 3 Edge适配的策略，并将其提炼为一个用于实时控制的单步策略。控制实验得出了三个发现：机器人专业化共同改善了世界预测和控制；在适应开始时分离的动作专家有助于提升效果；在策略适应过程中保持世界模型模式能够保持对场景变化的鲁棒性。

View on arXiv Download PDF AI Translation

cs.RO / 240 / 2610.18790

EqC: Equilibrium Energy Control for Introspective Lifelong Robot Learning

EqC：内省终身机器人学习的平衡能量控制

Yang, Haodong, Huang, Runhan, Zhang, Bowen, Lei, Iofai, Zhang, Zunzhe, Zhang, Xiangcheng, Zhu, Shaoting, Mou, Linzhan, Liu, Yicheng, Zhao, Hang

Abstract

Long-term robot autonomy depends on more than effective action generation: a policy must recognize when its behavior is no longer sufficiently supported and use assistance to improve. We introduce Equilibrium Energy Control (EqC), a framework for introspective lifelong robot learning built around an explicit conditional energy landscape. EqC casts generative action refinement as equilibrium-seeking optimization, while the geometry of the refinement process reveals whether the generated behavior remains consistent with the learned landscape. This competence signal enables Energy-Gated DAgger, where persistent energy-consistency violations trigger expert takeover and the resulting recovery trajectories are aggregated for subsequent policy updates. EqC integrates naturally into both Diffusion Transformer policies and flow-matching-based vision-language-action (VLA) policies. Across simulation and real-world bimanual manipulation, EqC matches or improves control performance and robustness while enabling timely intervention near the policy's competence boundary. A real-world study further shows how these interventions progressively expand the range of situations handled autonomously, providing a practical path from static generative controllers toward self-improving robot systems.

Chinese Translation

长期机器人自主性不仅依赖于有效的动作生成：策略必须识别其行为何时不再得到充分支持，并利用辅助来改善。我们提出了平衡能量控制（EqC），这是一个围绕显式条件能量景观构建的内省终身机器人学习框架。EqC将生成性动作优化视为寻求平衡的优化过程，同时优化过程的几何形状揭示生成的行为是否与学习到的景观保持一致。这一能力信号使得能量门控的DAgger（Energy-Gated DAgger）成为可能，其中持续的能量一致性违规会触发专家接管，随后生成的恢复轨迹会被聚合用于后续的策略更新。EqC自然地融入了扩散变换器（Diffusion Transformer）策略和基于流匹配的视觉-语言-动作（VLA）策略。在模拟和真实世界的双手操作中，EqC的控制性能和鲁棒性与现有方法相当或有所提升，同时在策略的能力边界附近实现及时干预。一项真实世界的研究进一步表明，这些干预逐步扩展了自主处理的情况范围，为从静态生成控制器向自我改进的机器人系统提供了切实可行的路径。

View on arXiv Download PDF AI Translation

cs.RO / 241 / 2610.18792

SEAR: Self-Evolving Agentic Robotics

SEAR：自我进化的代理机器人技术

Wang, Xinyi, Li, Bangzheng, Wang, Hanyu, Zhang, Lin, Cui, Andrea Yaoyun, Tang, Zhenggang, Buch, Shyamal, Xu, Dejia

Abstract

Frontier language-model agents can now improve robots on their own: they write control code, test it, and keep what works as skills and memories for the next task. We introduce SEAR (Self-Evolving Agentic Robotics), a framework and benchmark built around two questions. Is the agent really learning, or just trying more often? And does what it learns still help on new tasks, in a new simulator, and on a real robot? The framework runs an agent in three modes: it writes code that becomes the policy (code-as-policy), acts live in the control loop (agent-as-policy), or works as a researcher that fine-tunes a vision-language-action model (auto-research). The benchmark adds controls that separate retained experience from repeated attempts, familiar tasks, privileged simulator state, larger budgets, and luck. Across seven frontier language models, 330 simulated and five physical tasks, and over four years of agent runtime, we highlight three findings. First, the learning curve of a robotic agent depends both on the overall time budget and the task difficulty relative to model capability. Harder tasks make longer sessions worthwhile; on easier tasks, re-rolling short attempts does just as well. Second, control code and shared routines transfer across tasks and simulators, while task-specific solutions do not. On a real robot, we observe a performance gap between agent-designed practice and novel human-designed scenes. Third, different models evolve more effectively in different ways. Some do best acting live, others writing code first. And a text-only model keeps pace with a comparable multimodal one: perception helps with error assessment, but agentic reasoning and tool use can matter more. We release the benchmark and codebase to support future research.

Chinese Translation

前沿语言模型代理现在可以自主改进机器人：它们编写控制代码，测试代码，并将有效的部分保留为技能和记忆，以便用于下一个任务。我们介绍了SEAR（Self-Evolving Agentic Robotics），这是一个围绕两个问题构建的框架和基准。代理是否真的在学习，还是仅仅在尝试更多的次数？它所学到的知识在新任务、新模拟器和真实机器人上是否仍然有效？该框架以三种模式运行代理：编写成为策略的代码（code-as-policy），在控制循环中实时行动（agent-as-policy），或作为研究者微调视觉-语言-动作模型（auto-research）。基准增加了控制因素，以区分保留的经验与重复尝试、熟悉任务、特权模拟器状态、更大的预算和运气。在七种前沿语言模型、330个模拟任务和五个物理任务，以及超过四年的代理运行时间中，我们突出了三个发现。首先，机器人代理的学习曲线既依赖于整体时间预算，也依赖于任务难度与模型能力的相对关系。更困难的任务使得较长的会话更有价值；在较简单的任务中，重新进行短期尝试的效果同样良好。其次，控制代码和共享例程可以在任务和模拟器之间转移，而特定任务的解决方案则不能。在真实机器人上，我们观察到代理设计的练习与新的人类设计场景之间存在性能差距。第三，不同模型以不同的方式更有效地进化。有些模型在实时行动时表现最佳，而另一些则更适合先编写代码。而一个仅文本模型的表现与一个可比的多模态模型相当：感知有助于错误评估，但代理推理和工具使用可能更为重要。我们发布了基准和代码库，以支持未来的研究。

View on arXiv Download PDF AI Translation

cs.RO / 242 / 2610.18822

EgoMatrix: Physically Faithful Dexterous Robot Demonstrations from Egocentric Human Videos

EgoMatrix：基于自我中心人类视频的物理真实灵巧机器人演示

Wang, Ziheng, Shao, Yanming, Mao, Ting, Chen, Kaihan, Wang, Yiqun, Wu, Xuanye, Mu, Yao

Abstract

Egocentric human videos offer a scalable alternative to robot teleoperation for collecting dexterous manipulation data. However, two challenges limit their use as reliable training data: accurate hand-object interaction (HOI) reconstruction from monocular RGB and high-quality robot demonstrations, especially visual observations. We present EgoMatrix, an end-to-end data engine that converts monocular egocentric human videos into bimanual dexterous trajectories and paired robot observation videos, with the trajectories directly replayable on real robots. It reconstructs HOI with emphasis on relative geometry and contact rather than exact metric scene recovery, then uses retargeting and reinforcement learning (RL) to reproduce the demonstrated tasks in simulation. We jointly render robot hands and objects from these simulated trajectories to produce physically faithful observations synchronized with state and control records. Rendering the same trajectories in diverse, task-compatible backgrounds expands paired training data while preserving robot-object motion and contact. Across TACO, OakInk2, and self-collected data, EgoMatrix achieves 54.84% end-to-end success in simulation versus 14.52% for the strongest baseline. Reconstruction reaches state-of-the-art performance on most metrics of the two public benchmarks. Qualitative results show more faithful hand-object interactions than prior methods, while feature-space analysis shows closer alignment with real-robot observations than GPT-6 Astra.Our project page is available at https://spikew726.github.io/EgoMatrix/.

Chinese Translation

自我中心的人类视频为收集灵巧操作数据提供了一种可扩展的替代方案，取代了机器人遥操作。然而，有两个挑战限制了它们作为可靠训练数据的使用：从单目RGB中准确重建手-物体交互（HOI）以及高质量的机器人演示，尤其是视觉观察。我们提出了EgoMatrix，这是一个端到端的数据引擎，将单目自我中心的人类视频转换为双手灵巧轨迹和配对的机器人观察视频，这些轨迹可以直接在真实机器人上重放。它重建HOI，强调相对几何和接触，而非精确的度量场景恢复，然后使用重定向和强化学习（RL）在仿真中再现演示的任务。我们从这些仿真轨迹中共同渲染机器人手和物体，以生成与状态和控制记录同步的物理真实观察。在多样化、任务兼容的背景中渲染相同的轨迹扩展了配对训练数据，同时保持了机器人-物体的运动和接触。在TACO、OakInk2和自收集的数据中，EgoMatrix在仿真中的端到端成功率达到54.84%，而最强基线为14.52%。在两个公共基准的绝大多数指标上，重建达到了最先进的性能。定性结果显示出比以往方法更真实的手-物体交互，而特征空间分析则显示出与真实机器人观察的更紧密对齐。我们的项目页面可访问 https://spikew726.github.io/EgoMatrix/。

View on arXiv Download PDF AI Translation

cs.RO / 243 / 2610.18826

WorldOracle: Benchmarking Action-Conditioned World Models against Real-World Executions

WorldOracle：针对真实世界执行的动作条件世界模型基准评估

Duan, Haoyi, Chen, Liliang, Wang, Nan, Zhang, Yang, Lu, Yuhang, Chen, Xiaotong, Tseng, Wei-Cheng, Gao, Shenyuan, Li, Zhaoshuo, Jin, Jingyi, Jannaty, Pooya, Shah, Dhruv, Zhu, Yuke, Liu, Ming-Yu, Yao, Maoqing, Ren, Guanghui

Abstract

We introduce WorldOracle, a benchmark specifically designed to evaluate action-conditioned world models (ACWMs) against paired real-world executions. Given the same initial observation and low-level action sequence, WorldOracle asks whether a predicted rollout reproduces what happens in the real world, including physical consequences and task-state transitions. We evaluate models along four complementary dimensions: physical state correctness, interaction spatial alignment, key interaction area fidelity, and multi-view action coherence. To support this evaluation, we curate real-world training and evaluation suites that cover different interaction dynamics and embodiments. The evaluation suite contains long, multi-segment episodes with both successful and failed executions, enabling evaluation across varied action outcomes and behavior distributions. Experiments on seven representative ACWMs reveal distinct model strengths and failure modes. WorldOracle provides a systematic foundation for measuring and advancing action-conditioned world modeling for manipulation. We will release the datasets, evaluation code, and leaderboard.

Chinese Translation

我们介绍了WorldOracle，这是一个专门设计用于评估动作条件世界模型（ACWMs）与配对真实世界执行的基准。给定相同的初始观察和低级动作序列，WorldOracle询问预测的展开是否再现了真实世界中发生的情况，包括物理后果和任务状态转变。我们从四个互补维度评估模型：物理状态正确性、交互空间对齐、关键交互区域保真度和多视角动作一致性。为了支持这一评估，我们策划了涵盖不同交互动态和表现形式的真实世界训练和评估套件。评估套件包含长时间的多段剧集，既包括成功执行也包括失败执行，从而能够在不同的动作结果和行为分布中进行评估。对七个代表性ACWM的实验揭示了不同模型的优势和失败模式。WorldOracle为测量和推动操控的动作条件世界建模提供了系统基础。我们将发布数据集、评估代码和排行榜。

View on arXiv Download PDF AI Translation

cs.RO / 244 / 2610.18835

RoboSimStudio: An Interactive Toolkit for Rapid Real2Sim of Robot Manipulation Tasks

RoboSimStudio：一个用于快速真实到仿真机器人操作任务的互动工具包

Chen, Kaiyuan, Xie, Shuangyu, Yoo, Uksang, Zhu, Xinyue, Magner, Cael, Zhu, Alvin, Kim, Daehwa, Heiden, Eric, Coumans, Erwin, Macklin, Miles, Huang, Spencer, Goldberg, Ken

Abstract

It is difficult to accurately model the physics of the interaction between a robot and the objects it manipulates, including contact mechanics, coupled contact dynamics, deformations, stiffness and friction. We present RoboSimStudio (RSS), an interactive toolkit that makes choices of solver, contact model, and physical parameters explicit and editable, allowing both humans and coding agents to iteratively refine simulations based on observed physical behavior. We implement RSS with the open Newton physics simulation system and describe experiments with human-guided real2sim using (1) visual evaluation, (2) force-feedback and (3) agentic Real2Sim2Real. Code, video and additional information can be found on the project website.

Chinese Translation

准确建模机器人与其操作物体之间的相互作用的物理特性是困难的，包括接触力学、耦合接触动力学、变形、刚度和摩擦。我们提出了RoboSimStudio (RSS)，这是一个互动工具包，使得求解器、接触模型和物理参数的选择变得明确且可编辑，从而允许人类和编码代理根据观察到的物理行为迭代地优化仿真。我们使用开放的Newton物理仿真系统实现了RSS，并描述了通过（1）视觉评估、（2）力反馈和（3）代理式Real2Sim2Real进行的人机引导真实到仿真的实验。代码、视频和更多信息可以在项目网站上找到。

View on arXiv Download PDF AI Translation

cs.RO / 245 / 2610.18843

LIBERO-RIG: A Hardware-Setup Robustness Benchmark for Vision-Language-Action Models

LIBERO-RIG：一种针对视觉-语言-动作模型的硬件设置鲁棒性基准

Cui, Andrea Yaoyun, Wang, Hanyu, Wang, Xinyi, Li, Bangzheng, Zhang, Lin, Tang, Zhenggang, Buch, Shyamal, Xu, Dejia, Tancik, Matthew

Abstract

Robot policies are expected to generalize across deployment environments, including variations in the robot itself: cameras are bumped and remounted, calibrations drift, controllers run at different rates, and end-effector pads wear down. We present LIBERO-RIG, a benchmark that perturbs only the robot's hardware and setup while holding every LIBERO task fixed. The benchmark spans eleven perturbation axes, including camera mounting and optics, camera-to-arm calibration, actuation gain, control rate, proprioceptive noise, base pose, table height, and gripper pads. Each axis has five severity levels, yielding 210 conditions across the four LIBERO suites. We evaluate nine recent vision-language-action models using 500 episodes per condition and substantiate each finding with rigorous statistical tests. Pooled success falls from 0.932 to 0.767 at the middle severity and 0.509 at the highest severity. Performance degradation varies substantially across perturbation axes and action interfaces, and training on visual perturbations does not mitigate faults that leave camera images unchanged. We release the benchmark together with manifests specifying the fixed perturbation parameters for each episode.

Chinese Translation

机器人策略预计能够在不同的部署环境中进行泛化，包括机器人本身的各种变化：摄像头可能会被碰撞并重新安装，校准可能会漂移，控制器的运行速率可能不同，末端执行器的垫片可能会磨损。我们提出了LIBERO-RIG，这是一个仅扰动机器人硬件和设置的基准，同时保持每个LIBERO任务不变。该基准涵盖了十一条扰动轴，包括摄像头安装和光学、摄像头与机械臂的校准、驱动增益、控制速率、自我感知噪声、基座姿态、桌面高度和夹持器垫片。每条轴有五个严重程度级别，共产生210种条件，涵盖四个LIBERO套件。我们使用每个条件500个实验评估了九个最新的视觉-语言-动作模型，并通过严格的统计测试验证了每一项发现。在中等严重程度下，整体成功率从0.932下降到0.767，在最高严重程度下降至0.509。性能下降在不同的扰动轴和动作接口之间差异显著，且在视觉扰动上训练并未减轻那些使摄像头图像保持不变的故障。我们发布了该基准，并附上了指定每个实验固定扰动参数的清单。

View on arXiv Download PDF AI Translation

cs.RO / 246 / 2610.18848

Contact Trajectory Prompting: In-Context Transfer of Contact-Rich Behaviors from a Single Demonstration

接触轨迹提示：从单一示范中进行接触丰富行为的上下文转移

Nie, Xian, Zang, Yujie, Zheng, Yuhang, Zheng, Yupeng, Gu, Songen, Chen, Wendi, Wen, Chuan, Lu, Cewu, Ding, Wenchao, Yan, Junchi, Yan, Shuicheng

Abstract

In-context robot learning enables robots to acquire new behaviors from demonstrations without test-time parameter updates. However, visual and kinematic specifications alone may not distinguish contact-rich behaviors that share similar motions but require different interaction intensities or patterns. We propose Contact Trajectory Prompting (CTP), a framework for one-shot transfer of contact-rich manipulation behaviors. CTP encodes motion and contact information from a single demonstration into an ordered contact trajectory prompt that captures their temporal evolution and specifies the desired physical interaction behavior. During execution, the policy uses online sensory feedback to retrieve interaction-relevant information from the prompt, conditioning closed-loop action generation on the demonstrated contact behavior. Across real-world wiping, rotation, and insertion tasks on objects excluded from training, CTP achieves 78.7% overall success, outperforming the strongest evaluated baseline by 13.0 percentage points. Changing only the reference demonstration enables CTP to follow contact specifications for both intensity and direction, including an intermediate wiping force level and clockwise or counterclockwise cap rotation. These results demonstrate that contact trajectories serve as effective behavioral specifications, extending in-context robot learning beyond motion imitation to physical interaction transfer. Project website: https://ctp-robotics.github.io/project_page.

Chinese Translation

上下文机器人学习使机器人能够在不进行测试时参数更新的情况下，从示范中获取新行为。然而，仅凭视觉和运动学规范可能无法区分那些共享相似运动但需要不同交互强度或模式的接触丰富行为。我们提出了接触轨迹提示（Contact Trajectory Prompting, CTP），这是一个用于一次性转移接触丰富操作行为的框架。CTP将来自单一示范的运动和接触信息编码为有序的接触轨迹提示，该提示捕捉了它们的时间演变并指定所需的物理交互行为。在执行过程中，策略利用在线传感反馈从提示中检索与交互相关的信息，将闭环动作生成条件化于示范的接触行为。在排除训练对象的真实世界擦拭、旋转和插入任务中，CTP实现了78.7%的整体成功率，超越了最强基线评估结果13.0个百分点。仅更改参考示范即可使CTP遵循接触规范，包括强度和方向，涵盖中等擦拭力水平和顺时针或逆时针的瓶盖旋转。这些结果表明，接触轨迹作为有效的行为规范，扩展了上下文机器人学习，从运动模仿到物理交互转移。项目网站：https://ctp-robotics.github.io/project_page.

View on arXiv Download PDF AI Translation

cs.RO / 247 / 2610.18850

Collaborative Planning for UAV-UGV Teams under Operational Constraints via Masked Error-bounded Neuro-Symbolic Diffusion

基于受限操作条件的无人机-无人地面车辆团队协同规划：通过有界误差的神经符号扩散

Deng, Zihao, Li, Qianhuang, Wigness, Maggie, Rogers, John, Kim, Donghyun, Zhang, Hao

Abstract

Collaborative planning under operational constraints is essential for heterogeneous robot teams in large-scale missions. Unmanned Aerial Vehicles (UAVs) provide rapid area coverage but are limited by onboard energy, while Unmanned Ground Vehicles (UGVs) can serve as mobile recharging stations yet are constrained by road-network traversability. Effective collaboration must enforce operational constraints during execution while enabling parallel, coordinated actions. Masked Diffusion Models (MDMs) have shown promise through token unmasking but have two key limitations: entropy-guided scheduling assumes distributional alignment and reliance on differentiable constraints for diffusion-time optimization. We propose Masked Error-bounded Neuro-Symbolic Diffusion (MEND), which formulates collaborative planning as a conditional masked generative process to generate context-aware action sequences for UAV-UGV teams. MEND introduces a neuro-symbolic diffusion paradigm that constrains conditional masked diffusion policies at execution by reweighting the learned action distribution with a logic-induced admissibility indicator, thereby enforcing compliance with operational constraints. MEND introduces an error-bounded inference scheduling at execution that constructs the unmasking schedule by selecting positions with minimal predictive entropy over the constraint-aware action distribution, enhancing collaborative planning optimality under operational constraints. Extensive experimental results in simulation, on real-world maps, and with physical robot teams demonstrate that MEND advances UAV-UGV collaborative planning under operational constraints.

Chinese Translation

在操作约束下进行协同规划对于在大规模任务中异构机器人团队至关重要。无人机（UAV）提供快速的区域覆盖，但受限于机载能量，而无人地面车辆（UGV）可以充当移动充电站，但受到道路网络可通行性的限制。有效的协作必须在执行过程中强制执行操作约束，同时实现并行的协调行动。掩蔽扩散模型（MDMs）通过令牌解掩蔽显示出潜力，但存在两个主要限制：熵引导调度假设分布对齐，并依赖于可微分约束进行扩散时间优化。我们提出了有界误差的神经符号扩散（MEND），将协同规划公式化为条件掩蔽生成过程，以生成针对无人机-无人地面车辆团队的上下文感知行动序列。MEND引入了一种神经符号扩散范式，通过使用逻辑诱导的可接受性指标重新加权学习的行动分布，在执行时约束条件掩蔽扩散策略，从而强制遵守操作约束。MEND在执行时引入了一种有界误差推理调度，通过选择在约束感知行动分布上具有最小预测熵的位置来构建解掩蔽调度，从而增强在操作约束下的协同规划最优性。在模拟、真实地图以及物理机器人团队的广泛实验结果表明，MEND推动了在操作约束下的无人机-无人地面车辆协同规划。

View on arXiv Download PDF AI Translation

cs.RO / 248 / 2610.18852

ForceWAM: Decoupling Future and Response Learning from Demo and Play Data for Contact-Rich Manipulation

ForceWAM：从示范和游戏数据中解耦未来与响应学习以实现接触丰富的操作

Xu, Zijun, Chen, Wendi, Wang, Xinkai, Zou, Jiachen, Chen, Hongjin, Liu, Hairuo, Lin, Bokai, Ma, Shihao, Ma, Ke, Zhang, Zhiwei, Yan, Zhuohao, Wu, Ke, Song, Haoming, Li, Xueqi, Wen, Chuan, Zhao, Jieru, Ding, Wenchao, Lu, Cewu, Yang, Lixin

Abstract

Expert demonstrations show how to complete contact-rich tasks, but provide limited coverage of how interaction forces respond to different actions. Task-agnostic play provides broader experience of contact dynamics without specifying task goals. To fully leverage these two types of data, we introduce ForceWAM, a force world action model that decouples knowledge of task-compatible interaction outcomes from knowledge of how local actions change contact. A world model learns from successful demonstrations to jointly predict future video and six-axis force/torque trajectories. Together, these futures describe the expected evolution of motion and contact during the task. A separate inverse dynamics model (IDM) learns from both task and play, mapping current observations and recorded visual-wrench futures to their associated robot actions. Play data thus provide knowledge of force responses for goal-directed execution. At deployment, the IDM reactively translates generated futures into actions, reusing them with fresh wrench feedback between world-model updates. Across real-robot force-regulated sliding, box flipping with and without disturbances, and ring placement, ForceWAM achieves $85$-$95\%$ mean success, surpassing world-modeling and reactive baselines. Our experiments further show that adding task-agnostic play data improves success by $15$-$45$ percentage points, demonstrating its value for learning contact-rich execution. Project page and code are available at https://force-wam.github.io/.

Chinese Translation

专家示范展示了如何完成接触丰富的任务，但对交互力如何响应不同动作的覆盖有限。与任务无关的游戏提供了更广泛的接触动态经验，而无需指定任务目标。为了充分利用这两种数据，我们引入了ForceWAM，一种力世界动作模型，它将任务兼容的交互结果知识与局部动作如何改变接触的知识解耦。一个世界模型从成功的示范中学习，以共同预测未来视频和六轴力/扭矩轨迹。这些未来共同描述了任务期间运动和接触的预期演变。一个独立的逆动力学模型（IDM）从任务和游戏中学习，将当前观察和记录的视觉-扭矩未来映射到其相关的机器人动作。因此，游戏数据提供了目标导向执行的力响应知识。在部署时，IDM 反应性地将生成的未来转化为动作，在世界模型更新之间重用这些动作并结合新的扭矩反馈。在真实机器人力调节滑动、带有和不带有干扰的箱子翻转以及环形放置等任务中，ForceWAM 实现了 85%-95% 的平均成功率，超越了世界建模和反应基线。我们的实验进一步表明，添加与任务无关的游戏数据可将成功率提高 15-45 个百分点，证明其在学习接触丰富执行中的价值。项目页面和代码可在 https://force-wam.github.io/ 获取。

View on arXiv Download PDF AI Translation

cs.RO / 249 / 2610.18855

Interaction Latents as Actions for Contact-Rich Manipulation

将交互潜变量作为接触丰富操作的动作

Zheng, Yuhang, Zang, Yujie, Liu, Yuchen, Zheng, Yupeng, Jiang, Zhennan, Xing, Zebin, Tian, Shuai, Gu, Songen, Guo, Jingxiang, Gao, Chen, Xia, Zhongpu, Yan, Shuicheng

Abstract

Contact-rich manipulation requires robots to coordinate not only where to move, but also how contact and force should evolve throughout an interaction. Yet most robot policies represent actions primarily as motion trajectories, using tactile signals as inputs or auxiliary prediction targets without explicitly encoding contact dynamics in the action representation. We propose ILA (Interaction Latents as Actions), which uses decodable interaction latents that jointly capture the evolution of motion, contact, and force application as policy action units, providing a shared representation for action generation, evaluation, and offline improvement. We learn these latents through visually conditioned joint action-tactile reconstruction with a time-conditioned implicit decoder. Using them as flow-matching generation targets supports post-training of pretrained vision-language-action models (VLAs) and world action models (WAMs), as well as training flow-matching policies from scratch. A shared decoder maps generated latents to executable actions. For offline improvement, we learn a tactile-conditioned expert reference model from interaction latents encoded from expert demonstrations. The resulting expert-consistency rewards guide weighted learning from successful rollouts, while observation-compatible expert references guide correction of low-reward predictions from failed rollouts. Across three policy families, ILA improves success rates averaged over four real-world tasks by 21.0-36.5 percentage points relative to tactile-conditioned direct-action baselines. Ablations highlight the benefit of encoding contact evolution in action targets beyond tactile conditioning, while the same latent space supports expert-guided offline improvement. Project website: https://mrsecant.github.io/ILA/.

Chinese Translation

接触丰富的操作要求机器人不仅协调移动的位置，还要协调在交互过程中接触和施加力的演变。然而，大多数机器人策略主要将动作表示为运动轨迹，使用触觉信号作为输入或辅助预测目标，而没有在动作表示中明确编码接触动态。我们提出了 ILA（将交互潜变量作为动作），它使用可解码的交互潜变量，这些潜变量共同捕捉运动、接触和施加力的演变，作为策略动作单元，为动作生成、评估和离线改进提供共享表示。我们通过视觉条件下的联合动作-触觉重建与时间条件隐式解码器学习这些潜变量。将其作为流匹配生成目标支持对预训练的视觉-语言-动作模型（VLA）和世界动作模型（WAM）的后训练，以及从头训练流匹配策略。一个共享解码器将生成的潜变量映射到可执行的动作。为了离线改进，我们从专家演示中编码的交互潜变量学习一个触觉条件的专家参考模型。生成的专家一致性奖励指导从成功的回放中加权学习，而与观察兼容的专家参考则指导对失败回放中低奖励预测的修正。在三个策略家族中，ILA 相对于触觉条件的直接动作基线提高了四个真实世界任务的成功率，平均提高了 21.0-36.5 个百分点。消融实验突显了在动作目标中编码接触演变的好处，超越了触觉条件，而同一潜空间支持专家指导的离线改进。项目网站：https://mrsecant.github.io/ILA/

View on arXiv Download PDF AI Translation

cs.RO / 250 / 2610.18866

AutoAgent0: An Agentic Runtime for Safe Closed-Loop Driving

AutoAgent0：一种安全闭环驾驶的智能运行时

Liu, Jiageng, Zhao, Seth Z., Zhou, Zewei, Huerta, Alejandro, Bayer-Calvert, Aidan, Bao, Yuxin, Lin, Haotian, Spataru, Robert, Cui, Zijun, Hong, Yining, Zhou, Bolei, Ma, Jiaqi

Abstract

Safe autonomous driving requires more than accurate planning at a single frame: it demands reliable closed-loop execution in dynamic environments. Although end-to-end driving policies have advanced rapidly, the accumulated closed-loop errors, especially in unseen scenarios, can push vehicles into states from which the same policy struggles to recover. We address the challenge by treating closed-loop driving safety as a test-time execution problem. We introduce AutoAgent0, an agentic runtime and harness framework that augments existing driving experts with complementary detection skills, rule-based action verification, and feedback-driven recovery. To address inaccurate scene perception, an independent detection module constructs scene context using complementary perception skills, and a rule-based verifier checks proposed trajectories against this context before execution. When the nominal trajectories are unreliable, the orchestrator of the runtime activates Recovery Planning, where an agentic loop constructs alternative action proposals from action primitives and iteratively refines them using verification feedback. This design combines independent scene assessment with feasibility checks and online recovery, addressing both perception errors and the limited recovery capabilities of individual experts. Evaluations on Fail2Drive and NavSafe-$\infty$ demonstrate improved closed-loop robustness over single-expert execution in long-tail and safety-critical scenarios. These results demonstrate that runtime supervision can improve closed-loop robustness across multiple driving experts without retraining the underlying policies.

Chinese Translation

安全的自动驾驶不仅需要在单帧下进行准确的规划，还要求在动态环境中可靠地执行闭环控制。尽管端到端的驾驶策略发展迅速，但在未见场景中积累的闭环误差可能会使车辆陷入同一策略难以恢复的状态。我们通过将闭环驾驶安全视为测试时执行问题来应对这一挑战。我们引入了AutoAgent0，一个智能运行时和框架，旨在增强现有驾驶专家的补充检测能力、基于规则的动作验证和反馈驱动的恢复能力。为了解决场景感知不准确的问题，一个独立的检测模块利用补充感知技能构建场景上下文，而一个基于规则的验证器在执行前检查提议的轨迹与该上下文的一致性。当名义轨迹不可靠时，运行时的协调者会激活恢复规划，其中智能循环从动作原语构建替代动作提议，并利用验证反馈进行迭代优化。该设计结合了独立的场景评估、可行性检查和在线恢复，解决了感知误差和单一专家有限恢复能力的问题。在Fail2Drive和NavSafe-$ ext{∞}$上的评估显示，在长尾和安全关键场景中，闭环的鲁棒性优于单一专家执行。这些结果表明，运行时监督可以在不重新训练基础策略的情况下，提升多个驾驶专家的闭环鲁棒性。

View on arXiv Download PDF AI Translation

cs.RO / 251 / 2610.18867

EmbodiedEvo: Self-improving Coding Agents for Physical Intelligence

EmbodiedEvo：自我提升的物理智能编码代理

Li, Wenyi, Chi, Yizhe, Zheng, Youjie, Liu, Chunyu, Wang, Xiaoqiu, Gao, Mingju, Jiang, Zheng, Hong, Deyao, Xiao, Calvin, Yang, Kaisen, Na, Qinhuai

Abstract

Bringing general-purpose agents into the physical world calls for an executable representation of physical intelligence. Compared with vision-language-action (VLA) models, which map observations and language instructions directly to actions, code makes control logic explicit and reusable. Code can serve not only as a policy representation but also as the workspace in which the agent solves a task. Rather than following a prescribed workflow of planning, execution, and verification, the agent can autonomously inspect the state, revise and execute code, and use feedback to choose the next step. The success of this SWE-style loop in digital environments motivates its extension to physical tasks. Yet physical execution remains prone to recurring failures. To retain and refine experience for later scenes, we introduce EmbodiedEvo, where a coding agent evolves reusable Skills and Lessons through feedback on a single Playground scene, then freezes the resulting harness for diverse new scenes. On RoboDojo-Sim, EmbodiedEvo achieves an average score of 44.48 and a success rate of 37.82\%, surpassing all officially verified results. Paired tests show the largest gains on memory-dependent tasks and successful reuse of coordinated procedures, while fine contact control remains difficult. Experience from stronger models also improves weaker executors: Astra's experience raises Sol and Luna's success rates by 8.86 and 5.50 percentage points over no experience. Compared with raw interaction history, Skills and Lessons achieve comparable task performance at 53.0\% lower execution cost.

Chinese Translation

将通用代理引入物理世界需要一种可执行的物理智能表示。与直接将观察和语言指令映射到动作的视觉-语言-动作（VLA）模型相比，代码使控制逻辑变得明确且可重用。代码不仅可以作为策略表示，还可以作为代理解决任务的工作空间。代理可以自主检查状态、修订和执行代码，并利用反馈选择下一步，而不是遵循预定的规划、执行和验证工作流程。在数字环境中，这种软件工程（SWE）风格的循环取得了成功，促使其扩展到物理任务。然而，物理执行仍然容易出现反复的失败。为了保留和改进后续场景的经验，我们引入了EmbodiedEvo，在这里，一个编码代理通过对单一游乐场场景的反馈演化出可重用的技能和经验，然后将生成的框架冻结以适应多样的新场景。在RoboDojo-Sim上，EmbodiedEvo的平均得分为44.48，成功率为37.82%，超越了所有官方验证的结果。配对测试显示，在依赖记忆的任务上获得了最大的提升，并成功重用协调程序，而精细接触控制仍然困难。来自更强模型的经验也改善了较弱执行者的表现：Astra的经验使Sol和Luna的成功率分别提高了8.86和5.50个百分点。与原始交互历史相比，技能和经验在执行成本降低53.0%的情况下实现了可比的任务表现。

View on arXiv Download PDF AI Translation

cs.RO / 252 / 2610.18869

What Matters in Quality-Aware VLA Pretraining: A Benchmark and Empirical Study

质量感知的视觉-语言-动作（VLA）预训练中的关键因素：基准测试与实证研究

Yan, Mingxuan, Gong, Litian, Huang, Dihong, Li, Zhixu, Zhang, Tianyu, Zhang, Wenqian, Jiang, Hefeifei, Li, Peihao, Zhang, Jian, Zhao, Mengfei, Tang, Yikai, Zhai, Hai, Wang, Zehao, Liang, Ruijian, Huang, Yanjia, Shao, Lin, Liu, Changliu, Hang, Kaiyu, Fan, Zhiwen, Tomizuka, Masayoshi, Yang, Jianfei, Li, Jiachen

Abstract

How should demonstration quality be used during vision-language-action (VLA) pretraining to improve performance after target-task fine-tuning? We compare weighting, filtering, and conditioning across reward definitions and label granularities using 5,000 naturally suboptimal human demonstrations and 24 target tasks. Within each VLA, we match demonstrations and training budgets. Additional comparisons examine pretraining data size and the roles of quality tags during fine-tuning and inference. Across three VLAs, we find that approach-aware episode conditioning improves mean success over plain behavioral cloning; conditioning and filtering favor episode and chunk labels, respectively, on average, with model-dependent reversals; and constant-tag fine-tuning improves mean success after conditioned pretraining with either tagged or untagged inference. We introduce AXIS-Bench to support these controlled comparisons. Built on crowdsourced simulation demonstrations, it provides recorded object and contact states for recomputing quality labels on shared trajectories. Replayable trajectories support workspace-aligned rendering and user-defined real-world target tasks. In our real-world experiment, simulation pretraining raises success from 9.6\% to 31.7\%. Project page: https://axis-bench.github.io/.

Chinese Translation

在视觉-语言-动作（VLA）预训练中，如何利用示范质量以提高目标任务微调后的性能？我们使用5000个自然次优的人类示范和24个目标任务，对奖励定义和标签粒度下的加权、过滤和条件化进行了比较。在每个VLA中，我们匹配示范和训练预算。额外的比较考察了预训练数据规模以及质量标签在微调和推理过程中的作用。在三个VLA中，我们发现基于方法的情节条件化在平均成功率上优于简单的行为克隆；条件化和过滤分别在平均上更倾向于情节和块标签，但存在模型依赖的反转；而常量标签微调在条件预训练后，无论是带标签还是不带标签的推理，均提高了平均成功率。我们引入了AXIS-Bench来支持这些受控比较。AXIS-Bench建立在众包的模拟示范基础上，提供了记录的物体和接触状态，以便在共享轨迹上重新计算质量标签。可重放的轨迹支持与工作空间对齐的渲染和用户定义的现实世界目标任务。在我们的现实世界实验中，模拟预训练将成功率从9.6\%提高到31.7\%。项目页面：https://axis-bench.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 253 / 2610.18872

EgoExplorer: A Vision-Motion-Touch Observatory for Human Loco-Manipulation In Situ

EgoExplorer：用于人类运动操控的视觉-运动-触觉观察系统

Zhou, Yunsong, Jiang, Xuekun, Shen, Haiqi, Liu, Hangxu, Shi, Chaojian, Liu, Yujie, Li, Jingyuan, Yang, Haoran, Cao, Jiafei, Zhou, Dongzhan, Wang, Dong, Mu, Yao, Shen, Chunhua, Zhang, Weinan

Abstract

Learning humanoid loco-manipulation from human demonstrations rests on observing how locomotion, posture, and object interaction evolve together as tasks unfold. Scaling collection across diverse environments, in turn, favors lightweight, portable capture systems. Yet, existing datasets provide only partial views of this coordination. Thus, we introduce EgoExplorer, a 100-hour vision-motion-touch observatory of human loco-manipulation built with a wearable system. With an emphasis on professional activities, its continuous recordings align egocentric video with reconstructed whole-body motion, fine-grained finger articulation, and tactile observations, accompanied by step-level descriptions. To achieve this coverage with sparse body sensing, we develop the Calibrate-Complete-Constrain pipeline. Concretely, reusable calibration fixes participant geometry and sensor attachments before motion inference to disentangle mounting offsets from articulation. A temporal prior then completes unobserved joint motion, while layered constraints promote measurement agreement, temporal continuity, and geometric consistency, properties that motion plausibility alone does not guarantee. Against tracker-fitted references, the method reduces body-point error by 71.9% relative to the strongest baseline; matched evaluations also show gains in hand stability and contact estimation. Preliminary studies explore multimodal generation and conditioning alongside humanoid retargeting with whole-body control in simulation.

Chinese Translation

从人类示范中学习类人运动操控依赖于观察在任务展开过程中，运动、姿态和物体交互如何共同演变。为了在多样化环境中扩大数据收集，轻量便携的捕捉系统显得尤为重要。然而，现有的数据集仅提供了这种协调性的部分视角。因此，我们推出了EgoExplorer，这是一个基于可穿戴系统构建的100小时人类运动操控的视觉-运动-触觉观察系统。该系统强调专业活动，其连续记录将自我中心视频与重建的全身运动、精细的手指关节运动和触觉观察相结合，并附有逐步描述。为了在稀疏的身体传感下实现这一覆盖，我们开发了校准-完成-约束管道。具体而言，可重用的校准在运动推断之前固定参与者的几何形状和传感器附件，以便将安装偏差与关节运动分离。然后，时间先验完成未观察到的关节运动，而分层约束促进测量一致性、时间连续性和几何一致性，这些特性是运动合理性所无法保证的。与跟踪器拟合的参考相比，该方法相对于最强基线将身体点误差降低了71.9%；匹配评估也显示出手部稳定性和接触估计的提升。初步研究探索了多模态生成和条件化，以及在仿真中与全身控制的类人重定向。

View on arXiv Download PDF AI Translation

cs.RO / 254 / 2610.18873

LoopWAM: Closed-Loop Generalist Policy Improvement from Deployment Rollouts

LoopWAM：从部署回滚中进行闭环通用策略改进

Guo, Wenxuan, Liang, Yingping, Wang, Zhaoguo, Li, Ziyuan, Ma, Angyuan, Wu, Zhenyu, Wang, Chengkun, Yu, Bingyao, Chen, Ze, Zhang, Borui, Xu, Xiuwei, Zhang, Meng, Zhou, Erjin, Lu, Jiwen

Abstract

Scaling expert demonstrations alone does not enable generalist robot policies to improve from deployment. Real-world rollouts contain successes, suboptimal behavior, and failures: imitating every recorded action can reinforce poor behavior, while filtering out failures discards evidence of physical interactions and their consequences. We introduce \textbf{LoopWAM}, a closed-loop world-action model that learns directly from this full quality spectrum, using rollouts from both itself and other deployed policies. Source-dependent noise sampling unifies behavior learning and action-conditioned world modeling, while quality steering and quality-weighted supervision retain full video supervision across qualities and regulate action imitation. Video and action tokens interact throughout parallel denoising, allowing learned consequences to inform action generation directly. On RoboChallenge Table30-V2, three deployment-and-training rounds raise success from 23.67\% to 43.33\%, a 19.7-percentage-point gain, with a final score of 56.80. LoopWAM also achieves 94.5\% average success on RoboTwin~2.0 and 49.5\% overall success on RoboCasa365 without additional embodied pretraining, obtaining the best overall results among the compared methods across all three benchmarks. On challenging shirt folding task, two rounds of deployment-driven improvement after single-task post-training raise success from 13.20\% to 55.56\%. These results demonstrate closed-loop improvement from mixed-quality experience under both multi-task learning and single-task post-training. \href{https://loop-wam.github.io/}{\textcolor{orange}{Project Page}}

Chinese Translation

仅仅扩展专家演示并不能使通用机器人策略在部署中得到改进。现实世界的回滚包含成功、次优行为和失败：模仿每一个记录的动作可能会强化不良行为，而过滤掉失败则会丢弃物理交互及其后果的证据。我们引入了 extbf{LoopWAM}，一种闭环世界-动作模型，直接从这一完整质量谱中学习，利用自身及其他已部署策略的回滚。源依赖噪声采样统一了行为学习和基于动作的世界建模，而质量引导和质量加权监督则在不同质量下保留了完整的视频监督，并调节动作模仿。在并行去噪过程中，视频和动作标记相互作用，使得学习到的后果能够直接影响动作生成。在 RoboChallenge Table30-V2 上，经过三轮部署和训练，成功率从 23.67\% 提升至 43.33\%，增幅为 19.7 个百分点，最终得分为 56.80。LoopWAM 在 RoboTwin~2.0 上也达到了 94.5\\% 的平均成功率，在 RoboCasa365 上的整体成功率为 49.5\\%，且无需额外的具身预训练，在所有三个基准中获得了比较方法中的最佳整体结果。在具有挑战性的衬衫折叠任务中，经过单任务后训练后的两轮部署驱动改进，成功率从 13.20\\% 提升至 55.56\\%。这些结果展示了在多任务学习和单任务后训练下，基于混合质量经验的闭环改进。 \\href{https://loop-wam.github.io/}{ extcolor{orange}{项目页面}}

View on arXiv Download PDF AI Translation

cs.RO / 255 / 2610.18874

ChunkAct: Action Chunking for Agentic Robotics

ChunkAct：用于自主机器人行动分块

Xu, Bingxin, Bao, Wenrui, Liu, Xinxin, Zhou, Jiawei, Shang, Yuzhang, Ferrara, Emilio

Abstract

Agentic robotics places a large language model (LLM) agent in the robot control loop, where it composes perception, motion, and vision-language-action (VLA) tools under the guidance of a memory and thereby generalizes to new tasks without retraining. This generality comes at a steep price in efficiency. Every primitive requires a round of LLM reasoning over the full interaction history, so the cost of an episode grows with the number of primitives. We find this reasoning largely redundant: once a sub-goal is chosen, its primitives follow a routine already prescribed by memory. We present ChunkAct, which introduces action chunking to agentic robotics. Online, a single call launches a chunk of chained primitives, which executes autonomously and returns control only upon failure. Offline, ChunkAct compiles the memory into these chunks, so execution is memory-free at test time and the planner is reserved for decisions that require reasoning. LLM computation therefore scales with the number of decisions rather than the number of primitives. On LIBERO-Pro, ChunkAct improves task success by 5.0 points over the state of the art in agentic robotics while requiring 2.9x fewer LLM calls, 4.4x fewer input tokens, and 2.3x fewer output tokens.

Chinese Translation

自主机器人将大型语言模型（LLM）代理置于机器人控制循环中，在记忆的指导下组合感知、运动和视觉-语言-行动（VLA）工具，从而在不重新训练的情况下对新任务进行泛化。这种泛化的代价是效率的显著降低。每个原始动作都需要对完整的交互历史进行一次LLM推理，因此一个回合的成本随着原始动作数量的增加而增加。我们发现这种推理在很大程度上是多余的：一旦选择了子目标，其原始动作遵循记忆中已经规定的例程。我们提出了ChunkAct，它将行动分块引入自主机器人。在在线模式下，单次调用启动一组链式原始动作，这些动作自主执行，仅在失败时返回控制。在离线模式下，ChunkAct将记忆编译成这些分块，因此在测试时执行不需要记忆，规划器仅用于需要推理的决策。因此，LLM计算的规模与决策数量相关，而非原始动作数量。在LIBERO-Pro上，ChunkAct将任务成功率提高了5.0个百分点，相较于自主机器人领域的现有技术，同时减少了2.9倍的LLM调用、4.4倍的输入标记和2.3倍的输出标记。

View on arXiv Download PDF AI Translation

计算机视觉 (Computer Vision)

300

cs.CV / 1 / 2610.13992

How Vision Transformers Represent Position: A Mechanistic Comparison of Absolute and Rotary Positional Encodings

视觉变换器如何表示位置：绝对位置编码与旋转位置编码的机制比较

Mannes, Mahmoud, Maturi, Nikhil, Mehta, Parva, Jeong, Jiwon, Kannappan, Aravind Singaram

Abstract

Positional embeddings encode the spatial position of each image patch in a Vision Transformer (ViT), allowing the model to distinguish different spatial arrangements of visual content. However, we know little about how the choice of embedding shapes a vision transformer's internal representation. In this work, we show that Absolute Positional Embeddings (APE) and Rotary Positional Embeddings (RoPE) encode spatial position through different internal mechanisms: APE stores position in explicit features, while RoPE builds it inside attention and rebuilds it when it is withheld. Specifically, in APE models, sparse autoencoders recover causal row- and column-specific features, whereas interventions on RoPE reveal distinct rotary components that separately encode row and column information. Additionally, causal interventions selectively disabling RoPE's row or column rotations show that decodable row and column information can be independently removed and, once rotations are restored, partly reconstructed, in part by axis-specialized attention heads. Both mechanisms are causally important for classification: permuting APE's positional features across patches reduces top-1 accuracy by 24--31 points, and ablating the rotations of RoPE's primary rebuilding head reduces it more than ablating those of any other head. Our findings show that the choice of positional encoding fundamentally changes how spatial information is organized and processed inside Vision Transformers. More broadly, our work demonstrates how mechanistic analysis can reveal how models construct, reconstruct, and maintain structured representations, providing a basis for targeted interventions in vision models.

Chinese Translation

位置嵌入编码了视觉变换器（Vision Transformer, ViT）中每个图像块的空间位置，使模型能够区分视觉内容的不同空间排列。然而，我们对嵌入的选择如何塑造视觉变换器的内部表示知之甚少。在本研究中，我们展示了绝对位置嵌入（Absolute Positional Embeddings, APE）和旋转位置嵌入（Rotary Positional Embeddings, RoPE）通过不同的内部机制编码空间位置：APE在显式特征中存储位置，而RoPE则在注意力内部构建位置，并在被抑制时重新构建。具体而言，在APE模型中，稀疏自编码器恢复因果行和列特定特征，而对RoPE的干预揭示了分别编码行和列信息的不同旋转成分。此外，选择性禁用RoPE的行或列旋转的因果干预表明，可解码的行和列信息可以独立移除，并且一旦旋转被恢复，部分信息可以通过轴专用的注意力头重新构建。这两种机制对分类具有因果重要性：在图像块之间重新排列APE的位置信息会使top-1准确率降低24到31个百分点，而消融RoPE主要重建头的旋转会使准确率降低得比消融其他任何头更严重。我们的研究结果表明，位置编码的选择从根本上改变了空间信息在视觉变换器内部的组织和处理方式。更广泛地说，我们的工作展示了机制分析如何揭示模型构建、重建和维护结构化表示的方式，为视觉模型中的有针对性干预提供了基础。

View on arXiv Download PDF AI Translation

cs.CV / 2 / 2610.13997

SharedSAE: a unified sparse dictionary across u-net blocks and denoising time for diffusion models

SharedSAE：跨U-Net块和去噪时间的统一稀疏字典用于扩散模型

Lei, Longhao, Yao, Lan

Abstract

Concept representations in diffusion models drift across network depth and denoising time, yet existing sparse autoencoders offer only per-block or per-timestep dictionaries, leaving no single feature dictionary for localization and intervention in the network. We present Shared Sparse Autoencoder (SharedSAE), which learns one sparse dictionary shared across four U-Net attention blocks and all 50 denoising steps of SDXL 1.0. A block-specific low-rank input adapter aligns the residual distribution of each block before the shared encoder, and a temporal conditioning module with a shared timestep branch and a block-timestep correction branch absorbs temporal non-stationarity. Despite spanning both axes, Shared SAE preserves a coherent dictionary structure. Built on the shared dictionary, localized feature suppression reduces the number of NudeNet-positive generations in the I2P detector-positive subset from 139 to 1. Tracing sparse features across the trajectory further reveals that concept representations are redistributed across U-Net blocks during generation, shifting from the earliest downsampling block toward intermediate and later blocks as denoising proceeds. SharedSAE thus provides a unified, interpretable, and intervention-ready representation space for diffusion models, enabling concept analysis and erasure at any covered block and denoising step.

Chinese Translation

在扩散模型中，概念表示在网络深度和去噪时间上漂移，然而现有的稀疏自编码器仅提供每个块或每个时间步的字典，未能为网络中的定位和干预提供单一特征字典。我们提出了共享稀疏自编码器（Shared Sparse Autoencoder，SharedSAE），它学习一个跨四个U-Net注意力块和SDXL 1.0的所有50个去噪步骤共享的稀疏字典。一个块特定的低秩输入适配器在共享编码器之前对齐每个块的残差分布，而一个具有共享时间步分支和块-时间步修正分支的时间条件模块则吸收时间上的非平稳性。尽管跨越了这两个轴，SharedSAE仍然保持一致的字典结构。基于共享字典，本地化特征抑制将I2P检测器阳性子集中的NudeNet阳性生成数量从139减少到1。追踪稀疏特征的轨迹进一步揭示，在生成过程中，概念表示在U-Net块之间重新分配，随着去噪的进行，从最早的下采样块向中间和后续块转移。因此，SharedSAE为扩散模型提供了一个统一、可解释且易于干预的表示空间，使得在任何覆盖的块和去噪步骤中进行概念分析和抹除成为可能。

View on arXiv Download PDF AI Translation

cs.CV / 3 / 2610.14091

Pristine Models

原始模型

Rajan, Anirudh Sundara, Singh, Krishna Kumar, Zhang, Richard, Lee, Yong Jae

Abstract

Training on broad, unfiltered data enables high-quality image generation, but can also teach models concepts that should not be generated; restricting the training data ensures safety, but reduces quality because examples containing unsafe concepts also provide useful knowledge that is not itself unsafe. Existing approaches largely address this trade-off post hoc through concept unlearning, which can lead to catastrophic forgetting as the number of concepts to remove grows. We instead ask whether unsafe concept knowledge can be compartmentalized during pretraining so that access to it can be disabled at inference. We introduce Pristine Models, a novel text-to-image generation architecture with an attention mechanism that encourages the separation of safe and unsafe knowledge into independently parameterized branches. The branch-specific parameters are learned during training, while the unsafe branch can be detached at inference. On ImageNet, Pristine Models enable safer generation while retaining much of the utility of full-data pretraining, achieving higher image quality and better text alignment than training only on safe data or applying post-hoc concept removal. We further scale our approach to CC12M, demonstrating that it remains effective in a substantially larger and more diverse open-world pretraining setting.

Chinese Translation

在广泛且未经过滤的数据上进行训练可以实现高质量的图像生成，但也可能使模型学习到不应生成的概念；限制训练数据虽然能确保安全性，但会降低质量，因为包含不安全概念的示例也提供了有用的知识，而这些知识本身并不不安全。现有的方法主要通过概念遗忘事后处理这一权衡，这可能导致随着需要移除的概念数量增加而出现灾难性遗忘。我们则提出在预训练过程中是否可以将不安全的概念知识进行隔离，以便在推理时可以禁用对其的访问。我们介绍了原始模型（Pristine Models），这是一种新颖的文本到图像生成架构，采用注意力机制，鼓励将安全和不安全知识分离到独立参数化的分支中。特定分支的参数在训练过程中学习，而不安全分支可以在推理时被断开。在ImageNet上，原始模型实现了更安全的生成，同时保留了全数据预训练的许多实用性，图像质量和文本对齐度均优于仅在安全数据上训练或应用事后概念移除。我们进一步将该方法扩展到CC12M，证明其在一个大得多且更具多样性的开放世界预训练环境中仍然有效。

View on arXiv Download PDF AI Translation

cs.CV / 4 / 2610.14093

Human Pose Estimation by Probabilistic Mixtures of Gaussian and Uniform Distributions

通过高斯和均匀分布的概率混合进行人体姿态估计

Alguacil, Santiago Reina, Rubio, Ezequiel López

Abstract

Despite advancements in sub-pixel decoding, modern Human Pose Estimation (HPE) architectures remain constrained by deterministic coordinate extraction from discrete heatmaps, which systematically fails under dense occlusions and Out-of-Distribution (OoD) ambiguities. To address these limitations, we propose a post-hoc continuous probabilistic framework that intercepts raw heatmaps and models them as a Gaussian Mixture Model (GMM) augmented with a spatially agnostic uniform distribution, requiring no architectural modifications or retraining. This continuous representation allows us to comprehensively explore the distinct contributions of density modeling to HPE. We show that the spatial expectation of our fitted GMM achieves precision parity with leading sub-pixel decoders without heuristic post-processing. Leveraging this continuous space, our test-time augmentation aggregates probability distributions rather than discrete activations, mitigating multi-scale heatmap poisoning. Furthermore, because our multimodal GMM parameterization isolates competing spatial hypotheses, it naturally interfaces with geometric regularizers such as Markov Random Fields. However, our analysis exposes the inherent vulnerabilities of 2D kinematic priors under perspective foreshortening and visual degradation. Crucially, this density formulation yields significant improvements in uncertainty quantification: our calibrated geometric covariance volume outperforms standard peak-intensity heuristics in OoD anomaly detection under severe occlusion. For in-distribution calibration, dynamically fusing our geometric metric via Adaptive Softmax reduces or matches baseline Area Under the Sparsification Error (AUSE) across benchmarks and degradation levels. Finally, we establish that this continuous parameterization is extensible to multi-view 3D architectures, bypassing the $\mathcal{O}(N^3)$ voxel memory bottleneck.

Chinese Translation

尽管在亚像素解码方面取得了进展，现代人体姿态估计（HPE）架构仍然受到从离散热图中提取确定性坐标的限制，这在密集遮挡和分布外（OoD）模糊情况下系统性失效。为了解决这些局限性，我们提出了一种后处理的连续概率框架，该框架拦截原始热图，并将其建模为增强了空间无关均匀分布的高斯混合模型（GMM），无需进行架构修改或重新训练。这种连续表示使我们能够全面探索密度建模对HPE的不同贡献。我们展示了我们拟合的GMM的空间期望在没有启发式后处理的情况下达到了与领先的亚像素解码器相当的精度。利用这一连续空间，我们的测试时增强聚合概率分布而非离散激活，从而减轻多尺度热图污染。此外，由于我们的多模态GMM参数化隔离了竞争的空间假设，它自然与几何正则化器（如马尔可夫随机场）接口。然而，我们的分析揭示了在透视缩短和视觉退化下，二维运动先验的固有脆弱性。关键是，这种密度公式在不确定性量化方面带来了显著改善：我们校准的几何协方差体积在严重遮挡下的OoD异常检测中超越了标准峰值强度启发式方法。对于分布内校准，通过自适应Softmax动态融合我们的几何度量，在基准测试和退化水平上减少或匹配基线稀疏化误差下的面积（AUSE）。最后，我们确定这种连续参数化可以扩展到多视角三维架构，绕过$ ext{O}(N^3)$体素内存瓶颈。

View on arXiv Download PDF AI Translation

cs.CV / 5 / 2610.14094

Scenes as an Attack Surface: Uncovering Scene-Grounded Safety Failures in MLLMs

场景作为攻击面：揭示多模态大型语言模型中的场景基础安全失败

Wu, Tongxi, Cheng, Yuyang, Xian, Jiacheng, Huang, Hu, Wu, Zhihao, Yang, Menglin, Zhang, Jian, Gao, Yang

Abstract

Existing jailbreak attacks have exposed serious safety weaknesses in multimodal large language models (MLLMs). Most prior attacks rely on specially constructed visual inputs to carry or support harmful intent. In this paper, we uncover a different vulnerability: task-relevant scene semantics alone can shift an MLLM from refusal to compliance, even when the visual input introduces no additional attack instructions and the original task intent remains unchanged. We formalize this failure as a scene-grounded visual safety vulnerability and introduce SceneSafetyBench to investigate it across eight risk categories and nine white- and black-box MLLMs. Our analysis reveals that this failure is associated with how task-relevant visual evidence influences downstream safety representations, suggesting that visual grounding alone cannot fully explain the observed safety shift. Building on this finding, we propose Scene-Grounded Attacks (SGA), which preserve the original task while constructing validated scene-grounded queries and combining their independent responses. Across nine target MLLMs, SGA achieves an average attack success rate (ASR) of 86.17%, outperforming the strongest baseline by 8.55 percentage points and achieving the highest ASR on eight models. Our results identify scene semantics as an overlooked attack surface in MLLMs and highlight a growing tension between stronger visual grounding and robust multimodal safety alignment.

Chinese Translation

现有的越狱攻击暴露了多模态大型语言模型（MLLMs）中的严重安全弱点。大多数先前的攻击依赖于特意构造的视觉输入来携带或支持有害意图。本文揭示了一种不同的脆弱性：仅凭与任务相关的场景语义就可以使MLLM从拒绝转变为遵从，即使视觉输入没有引入额外的攻击指令，且原始任务意图保持不变。我们将这种失败形式化为场景基础的视觉安全脆弱性，并引入SceneSafetyBench以在八个风险类别和九个白盒及黑盒MLLM中进行调查。我们的分析显示，这种失败与任务相关的视觉证据如何影响下游安全表示有关，表明单靠视觉基础无法完全解释观察到的安全转变。在此基础上，我们提出了场景基础攻击（SGA），该攻击在构建经过验证的场景基础查询的同时保留原始任务，并结合其独立响应。在九个目标MLLM中，SGA实现了86.17%的平均攻击成功率（ASR），比最强基线高出8.55个百分点，并在八个模型上达到了最高的ASR。我们的结果表明，场景语义是MLLM中一个被忽视的攻击面，并突显出更强的视觉基础与稳健的多模态安全对齐之间日益增长的紧张关系。

View on arXiv Download PDF AI Translation

cs.CV / 6 / 2610.14153

TrajCFG: Token-Wise Classifier-Free Guidance via Denoising Trajectory Feedback

TrajCFG：通过去噪轨迹反馈实现逐标记无分类器引导

Liu, Jingyu, Jiang, Hao, Ding, Yikang, Shi, Huafeng, Yang, Xu, Liu, Xiaoqiang, Zhang, Guoxin, Wan, Pengfei, Gai, Kun

Abstract

Classifier-free guidance (CFG) typically applies a shared scale to all latent tokens, although their velocity predictions evolve differently during sampling. We introduce Trajectory-Adaptive Classifier-Free Guidance (TrajCFG), a training-free method for flow-matching models that adapts token-wise guidance using historical changes in guided velocity direction. TrajCFG maps this feedback to bounded scales, assigning stronger guidance to larger directional changes and weaker guidance to more consistent directions. It combines adjacent-step and interval-normalized longer-span feedback, reusing standard CFG predictions without additional model evaluations. Controlled diagnostics and ablations examine the roles of directional feedback, token-wise allocation, and history-span fusion. Experiments on text-to-video, image-to-video, text-to-audio-video, text-to-image, and image-to-3D generation show improvements over base samplers across multiple backbones. For text-to-video generation, TrajCFG improves VBench Total scores by 0.30-0.75 points over the strongest evaluated baseline on each of three backbones. These results support historical directional feedback as a broadly applicable signal for fine-grained guidance allocation. The project page is available at https://jyliu-98.github.io/trajcfg-project-page/.

Chinese Translation

无分类器引导（Classifier-free guidance, CFG）通常对所有潜在标记应用共享尺度，尽管它们在采样过程中速度预测的演变各不相同。我们提出了轨迹自适应无分类器引导（Trajectory-Adaptive Classifier-Free Guidance, TrajCFG），这是一种针对流匹配模型的无训练方法，利用历史变化的引导速度方向来调整逐标记引导。TrajCFG将这种反馈映射到有界尺度上，对较大的方向变化赋予更强的引导，而对更一致的方向则赋予较弱的引导。它结合了相邻步骤和区间归一化的长跨度反馈，重用标准CFG预测而无需额外的模型评估。控制诊断和消融实验考察了方向反馈、逐标记分配和历史跨度融合的作用。在文本到视频、图像到视频、文本到音频视频、文本到图像以及图像到3D生成的实验中，TrajCFG在多个基础模型上显示出相较于基础采样器的改进。在文本到视频生成中，TrajCFG在三个基础模型上分别提高了VBench总分0.30-0.75分，超越了每个模型上评估的最强基线。这些结果支持历史方向反馈作为一种广泛适用的信号，用于细粒度的引导分配。项目页面可访问 https://jyliu-98.github.io/trajcfg-project-page/。

View on arXiv Download PDF AI Translation

cs.CV / 7 / 2610.14158

Video-Arena: A Scalable Agentic Video Benchmark for Spatiotemporal Capability

视频竞技场：一个可扩展的代理视频基准用于时空能力评估

Zhang, Kuan, Chen, Yukun, Yang, Zhihao, Su, Yue, Wu, Xiangnan, Chen, Zirong, Luo, Run, Tan, Tao, Zheng, Yinhe, Li, Yiming

Abstract

Understanding and tracking a dynamic world are fundamental to visual intelligence. Existing video benchmarks provide limited insight into whether models can maintain a coherent understanding of evolving world state. Building such a benchmark from real-world videos is challenging because the underlying world state is partially observable, making annotations hard to obtain. To close this gap, we introduce Video-Arena, a benchmark that turns competitive games into scalable, verifiable tests of how well models understand evolving world states from videos. Specifically, inspired by how humans understand dynamic events, we evaluate spatiotemporal world state understanding and tracking through three task categories: perceiving critical events, interpreting state updates, and tracking the resulting states. Our pipeline can generate videos and aligned reference labels from game records at scale. To keep evaluation affordable and reproducible, we curate a golden set of 180 tasks across 9 games and run evaluations under two protocols: native visual understanding and tool-assisted agentic analysis. The best-performing systems achieve F1 scores of and under the two protocols, respectively. The result highlights that native models struggle to perceive events reliably, while tool-assisted systems improve event recovery but still struggle to interpret state updates and track the resulting states consistently over time. Our project page is in https://videoarena-bench.github.io

Chinese Translation

理解和追踪动态世界是视觉智能的基础。现有的视频基准对模型是否能够保持对不断变化的世界状态的连贯理解提供的见解有限。从真实世界视频构建这样的基准具有挑战性，因为潜在的世界状态是部分可观察的，这使得注释难以获得。为了解决这一问题，我们提出了视频竞技场（Video-Arena），这是一个将竞争游戏转化为可扩展、可验证的测试，以评估模型从视频中理解不断变化的世界状态的能力。具体而言，受人类理解动态事件的启发，我们通过三个任务类别评估时空世界状态的理解和追踪：感知关键事件、解释状态更新和追踪结果状态。我们的流程可以从游戏记录中大规模生成视频和对齐的参考标签。为了保持评估的可负担性和可重复性，我们策划了一套包含9个游戏的180个任务的黄金集，并在两种协议下进行评估：原生视觉理解和工具辅助的代理分析。最佳表现的系统在这两种协议下分别达到了F1分数。结果表明，原生模型在可靠感知事件方面存在困难，而工具辅助系统虽然改善了事件恢复，但在解释状态更新和持续追踪结果状态方面仍然面临挑战。我们的项目页面为 https://videoarena-bench.github.io

View on arXiv Download PDF AI Translation

cs.CV / 8 / 2610.14207

Token-Budgeted Candidate Retrieval for 3D Scene-Grounded VLMs: Category Names vs. Pairwise Geometry

基于令牌预算的候选检索用于3D场景基础的视觉语言模型：类别名称与成对几何

Zakrzewski, Michał, Shpir, Mariia, Rumezhak, Taras

Abstract

Under a fixed context budget, which objects and relations should a 3D scene-grounded VLM see first? On ScanNet indoor referring benchmarks (Dai et al., 2017), matching the query to object category names outperforms pairwise geometry scoring at every reported K. The gaps are largest at high recall. Analytic pairwise cues beat per-object cue scoring and query-agnostic kNN graphs, including on paraphrases. MiniLM and BGE-small still score higher on those splits. We evaluate token-budgeted candidate retrieval with axis-aligned boxes, a token-ratio metric, training-free baselines, two small encoders that match q to each object's category name, and QCRR (Query-Conditioned Relation Retrieval), a deterministic pairwise-geometry scorer. Across ScanRefer gt_val plus a custom 20% scan-id split of Nr3D and Sr3D (35,504 queries; not official ReferIt3D val), additive noun matching reaches R@10 0.864 on ScanRefer against QCRR's 0.626; MiniLM and BGE-small reach 0.911 to 0.913 on ScanRefer and 0.941 to 0.947 on Nr3D. A corrected 3DGraphLLM harness (native [email protected] 0.67 on the full graph, n=300 ScanRefer queries) shows that context selection changes downstream accuracy under a matched object budget (Random 0.44, QCRR 0.56, MiniLM 0.67, oracle 0.70). Semantic retrieval beats hand-designed geometry on that harness. We will release the protocol, retriever implementations, and per-query outputs so new selectors can be swapped in without rewriting the analysis.

Chinese Translation

在固定的上下文预算下，3D场景基础的视觉语言模型（VLM）应首先看到哪些对象和关系？在ScanNet室内参考基准测试（Dai et al., 2017）中，将查询与对象类别名称匹配的表现优于成对几何评分，在所有报告的K值中均是如此。在高召回率时，差距最大。分析性的成对线索在每个对象线索评分和与查询无关的kNN图上表现更好，包括在同义句中。MiniLM和BGE-small在这些分割上仍然得分更高。我们评估了使用轴对齐框的令牌预算候选检索、令牌比率度量、无训练基线、两个小编码器（将q与每个对象的类别名称匹配）以及QCRR（查询条件关系检索），这是一个确定性的成对几何评分器。在ScanRefer gt_val以及Nr3D和Sr3D的自定义20%扫描ID分割（35,504个查询；非官方ReferIt3D验证集）中，附加名词匹配在ScanRefer上达到R@10 0.864，而QCRR为0.626；MiniLM和BGE-small在ScanRefer上达到0.911到0.913，在Nr3D上达到0.941到0.947。一个修正的3DGraphLLM工具（在完整图上的原生[email protected]为0.67，n=300个ScanRefer查询）显示，在匹配对象预算下，上下文选择改变了下游准确性（随机0.44，QCRR 0.56，MiniLM 0.67，oracle 0.70）。语义检索在该工具上优于手工设计的几何。我们将发布协议、检索器实现和每个查询的输出，以便新的选择器可以在不重写分析的情况下进行替换。

View on arXiv Download PDF AI Translation

cs.CV / 9 / 2610.14226

CAPTOR: a vision-language foundation model for Chest-Abdomen-Pelvis Oncologic CT

CAPTOR：用于胸部-腹部-盆腔肿瘤CT的视觉-语言基础模型

Jean-Mactoux, Séléna, Loizillon, Sophie, Jardali, Ghina, Crombé, Amandine, Lassau, Nathalie, Bône, Alexandre

Abstract

Purpose : To develop and evaluate CAPTOR, a deep-learning vision-language foundation model for oncologic chest-abdomen-pelvis (CAP) CT interpretation. Materials & Methods : CAPTOR (Chest-Abdomen-Pelvis foundaTion model for Oncologic Radiology) is a contrastive vision-language model, trained using a pretraining strategy for oncologic CAP CT (in single- or multiple-series acquisitions), on a multicenter retrospective dataset from North America, Europe and Asia, predominantly from tumor lesion examinations. CAPTOR was compared with state-of-the-art CT foundation models on three downstream tasks: disease detection, survival prediction using TumorImagingBench and image retrieval. Classification and retrieval were both evaluated on held-out internal and public test sets, stratified by anatomical coverage and indication. Bootstrap resampling was used to estimate confidence intervals, and pairwise comparisons were performed using permutation tests. Results : CAPTOR was trained and tested on 25,512 and 10,794 CT-report pairs respectively. TumorImagingBench datasets comprised between 134 and 421 studies each. CAPTOR outperformed all baseline models, on the classification task, on the CAP subset (AUC 0.767, 95%CI: 0.734-0.793, vs. 0.695, 95%CI: 0.645-0.736 for the next-best, p<0.001) and oncologic subset (AUC 0.784, 95%CI: 0.759-0.805, vs. 0.734, 95%CI: 0.703-0.767 p<0.001), ranked first on three of the four oncologic prognostic benchmarks, and achieved superior retrieval performance on CAP and oncologic subsets. Conclusion : A foundation model trained on oncology-enriched CAP CT data outperformed relevant state-of-the-art models in its target clinical context, underscoring the importance of clinically aligned pretraining. We release our preprocessing, training and inference code : https://github.com/Guerbet-AI/CAPTOR and model weights : https://huggingface.co/Guerbet-AI/CAPTOR

Chinese Translation

目的：开发和评估CAPTOR，一个用于肿瘤胸部-腹部-盆腔（CAP）CT解读的深度学习视觉-语言基础模型。材料与方法：CAPTOR（Chest-Abdomen-Pelvis foundaTion model for Oncologic Radiology）是一个对比视觉-语言模型，采用针对肿瘤CAP CT（单系列或多系列采集）的预训练策略，在来自北美、欧洲和亚洲的多中心回顾性数据集中进行训练，数据主要来源于肿瘤病变检查。CAPTOR在三个下游任务上与最先进的CT基础模型进行了比较：疾病检测、使用TumorImagingBench进行生存预测和图像检索。分类和检索均在按解剖覆盖和指征分层的内部保留测试集和公共测试集上进行了评估。使用自助重抽样法估计置信区间，并使用置换检验进行成对比较。结果：CAPTOR分别在25,512对和10,794对CT报告上进行了训练和测试。TumorImagingBench数据集各包含134至421个研究。CAPTOR在分类任务中优于所有基线模型，在CAP子集上（AUC 0.767，95%CI：0.734-0.793，vs. 0.695，95%CI：0.645-0.736，p<0.001）和肿瘤子集上（AUC 0.784，95%CI：0.759-0.805，vs. 0.734，95%CI：0.703-0.767，p<0.001），在四个肿瘤预后基准中排名前三，并在CAP和肿瘤子集上实现了优越的检索性能。结论：在肿瘤丰富的CAP CT数据上训练的基础模型在其目标临床背景中优于相关的最先进模型，强调了临床对齐预训练的重要性。我们发布了我们的预处理、训练和推理代码：https://github.com/Guerbet-AI/CAPTOR 和模型权重：https://huggingface.co/Guerbet-AI/CAPTOR

View on arXiv Download PDF AI Translation

cs.CV / 10 / 2610.14228

Latent World Action Model for GUI Agents

图形用户界面代理的潜在世界动作模型

Yu, Seungjun, Choi, Hojun, Park, Seojeong, Shin, Jaeyo, Shim, Hyunjung

Abstract

Pixel-space world models have recently emerged as a promising approach for GUI agents, predicting future screens prior to execution and subsequently re-perceiving them for action selection. However, this two-stage paradigm faces two key challenges in GUI environments: (L1) the high computational cost of pixel reconstruction and (L2) the need to reconstruct large and diverse appearance-level details beyond what is required for action selection. In practice, this screen-generation objective can induce an identity shortcut problem, where future-state prediction largely preserves the current state rather than capturing action-induced changes. In this work, we reformulate GUI world modeling by shifting the prediction target from full-screen reconstruction to action-induced future information in latent space. Concretely, we leverage a large multimodal model as the latent world model, whose shared representations support two dedicated heads for both action selection and predictive transition modeling. The action head directly translates these latent representations into action selection, while the transition head captures action-conditioned future information. To mitigate the identity shortcut, we further introduce an action-aware contrastive objective that encourages the transition representations to capture discriminative action-induced changes. Extensive experiments across offline and online GUI settings show that our method achieves up to 31.9% and 40.6% relative gains in success rate over pixel-space world models, respectively, while substantially reducing inference latency by up to 98.9%. Our code is available at https://github.com/sjyu001/latent2world.

Chinese Translation

像素空间世界模型最近作为图形用户界面（GUI）代理的一种有前景的方法出现，能够在执行之前预测未来屏幕，并随后重新感知这些屏幕以进行动作选择。然而，这种两阶段范式在GUI环境中面临两个主要挑战：（L1）像素重建的高计算成本和（L2）需要重建超出动作选择所需的大量多样化外观级细节。在实践中，这一屏幕生成目标可能会引发身份捷径问题，即未来状态预测在很大程度上保留当前状态，而不是捕捉由动作引起的变化。在本研究中，我们通过将预测目标从全屏重建转变为潜在空间中的动作引起的未来信息，重新构建了GUI世界建模。具体而言，我们利用一个大型多模态模型作为潜在世界模型，其共享表示支持两个专用头，分别用于动作选择和预测转移建模。动作头直接将这些潜在表示转换为动作选择，而转移头则捕捉基于动作的未来信息。为了减轻身份捷径问题，我们进一步引入了一种动作感知对比目标，鼓励转移表示捕捉具有区分性的动作引起的变化。在离线和在线GUI设置下的广泛实验表明，我们的方法在成功率上相较于像素空间世界模型分别实现了高达31.9%和40.6%的相对提升，同时将推理延迟显著降低了高达98.9%。我们的代码可在 https://github.com/sjyu001/latent2world 获取。

View on arXiv Download PDF AI Translation

cs.CV / 11 / 2610.14253

SliderEdit++: Improved Slider Trajectories in Instruction-Based Image Editing

SliderEdit++：基于指令的图像编辑中的改进滑块轨迹

Zarei, Arman, Nag, Sayan, Wang, Bryan, Dernoncourt, Franck, Manjunatha, Varun, Lipka, Nedim, Feizi, Soheil, Basu, Samyadeep

Abstract

Slider-based control aims to give users continuous, fine-grained control over the intensity of individual edits in instruction-based image editing. However, existing approaches fail to deliver reliable continuous controllability: traversing a slider can produce abrupt visual jumps and non-monotonic changes, while jointly controlled sliders can interfere with one another. To systematically study these limitations, we introduce SlideBench, a comprehensive benchmark for continuous image editing spanning diverse editing categories and equipped with a unified suite of metrics quantifying monotonic progression, perceptual continuity, and inter-slider disentanglement. Building on these observations, we propose SliderEdit++, which explicitly regularizes the intermediate editing trajectory rather than supervising only its endpoints. Our method combines velocity- and semantic-space trajectory regularization with spectral control to encourage smooth, predictable slider traversal, together with a disentanglement objective that reduces interference between independently controlled edits. On SlideBench, SliderEdit++ achieves state-of-the-art performance, improving the overall score by 18.5% relative to the strongest baseline. At the same time, using a few-step editing backbone makes our method approximately 3x to 20x faster than existing slider-based approaches, enabling substantially more responsive interactive editing.

Chinese Translation

基于滑块的控制旨在为用户提供对基于指令的图像编辑中单个编辑强度的连续、细粒度控制。然而，现有的方法未能提供可靠的连续可控性：滑块的移动可能会产生突兀的视觉跳跃和非单调变化，而共同控制的滑块可能会相互干扰。为了系统地研究这些局限性，我们引入了SlideBench，这是一个全面的连续图像编辑基准，涵盖多种编辑类别，并配备了一套统一的指标，量化单调进展、感知连续性和滑块间解耦。基于这些观察，我们提出了SliderEdit++，该方法明确地对中间编辑轨迹进行正则化，而不仅仅是监督其端点。我们的方法结合了速度和语义空间轨迹正则化与谱控制，以鼓励平滑、可预测的滑块移动，同时设定了解耦目标，以减少独立控制编辑之间的干扰。在SlideBench上，SliderEdit++实现了最先进的性能，相较于最强基线提高了18.5%的总体得分。同时，使用少量步骤的编辑骨干使我们的方法比现有的基于滑块的方法快约3到20倍，从而实现了显著更具响应性的交互编辑。

View on arXiv Download PDF AI Translation

cs.CV / 12 / 2610.14254

Image Diffusion Models Are Good 3D Human Motion Generators

图像扩散模型是优秀的3D人类运动生成器

Yang, Siyuan, Wang, Chong, Cheng, Hao, Liu, Jun, Wen, Bihan, Kot, Alex, Kjellstrom, Hedvig

Abstract

Text-to-motion generation typically relies on dataset-specific skeletal representations, with models trained separately for benchmarks with different joint layouts. We investigate whether image diffusion architectures can model 3D human motion through a common image-compatible representation that accommodates heterogeneous skeletal layouts while preserving motion-specific kinematic structure. To this end, we organize root-centered joint trajectories into topology-informed joint--time images and model global root dynamics through a complementary latent stream. A motion autoencoder conditions local-motion decoding on the root latent, while a text-conditioned Scalable Interpolant Transformer (SiT) jointly generates both latent streams, with supervision on decoded 3D joint positions further constraining generation in motion space. In addition to dataset-specific models, we train a single motion autoencoder on HumanML3D and KIT-ML, followed by a single text-conditioned SiT shared across both datasets, preserving their respective skeletal definitions without skeleton retargeting to a common joint layout. Experiments on HumanML3D and KIT-ML show that our dataset-specific models achieve state-of-the-art performance, while shared training yields further gains on both benchmarks, demonstrating the benefits of shared modeling across heterogeneous skeletal layouts.

Chinese Translation

文本到运动生成通常依赖于特定数据集的骨骼表示，模型为不同关节布局的基准单独训练。我们研究图像扩散架构是否能够通过一种通用的图像兼容表示来建模3D人类运动，该表示能够适应异构骨骼布局，同时保持运动特定的运动学结构。为此，我们将以根关节为中心的关节轨迹组织成拓扑信息驱动的关节-时间图像，并通过一个互补的潜在流来建模全局根动态。运动自编码器将局部运动解码条件化于根潜在，而文本条件的可扩展插值变换器（Scalable Interpolant Transformer, SiT）共同生成两个潜在流，并在解码的3D关节位置上进行监督，从而进一步约束运动空间中的生成。除了特定数据集的模型外，我们在HumanML3D和KIT-ML上训练了一个单一的运动自编码器，随后在两个数据集上共享一个文本条件的SiT，保留各自的骨骼定义，而无需将骨骼重新定向到一个共同的关节布局。在HumanML3D和KIT-ML上的实验表明，我们的数据集特定模型达到了最先进的性能，而共享训练在两个基准上都带来了进一步的提升，展示了在异构骨骼布局中共享建模的优势。

View on arXiv Download PDF AI Translation

cs.CV / 13 / 2610.14261

BigCell: Generating Gigapixel Whole-Slide Images

BigCell：生成千兆像素全切片图像

Yellapragada, Srikar, Graikos, Alexandros, Li, Zilinghan, Triaridis, Kostas, Nandi, Tarak Nath, Bai, Ji Dong K, Knudsen, Beatrice, Kurc, Tahsin, Prasanna, Prateek, Gupta, Rajarsi R., Madduri, Ravi K, Saltz, Joel, Samaras, Dimitris

Abstract

Whole-slide images (WSIs) are gigapixel-scale digital scans of histopathology tissue and serve as the primary basis for cancer diagnosis. Existing histopathology generative models operate at the patch scale, producing fixed-size tiles of at most 1024$\times$1024 pixels which encompass only a small fraction of the WSI. In contrast, text reports for clinical diagnoses and slide-level diagnostic labels -- the supervision that exists at scale -- operate at the whole-slide level. We present BigCell, the first approach to generate coherent whole-slide histopathology images at native gigapixel resolution. BigCell trains a flow-matching transformer at 2048$\times$2048 pixels on over 120,000 publicly available WSIs spanning multiple organ types. Conditioning on embeddings from pathology foundation models and using an inference-time sliding-window pipeline, BigCell generates coherent images of up to 32,768$\times$32,768 pixels, outperforming prior tile generators on both tile-level and slide-level fidelity metrics. To make gigapixel synthesis practical, we introduce an adaptive scheduling strategy that prioritizes important slide regions, speeding up image generation by over 3\ttimes with minimal quality degradation. Furthermore, we propose the first text-to-image synthesis at 8k resolution by training a separate model that produces dense conditioning grids directly from pathology text reports. We demonstrate that augmenting with synthetic samples improves slide-level few-shot classification performance by up to $18 \%$ over baselines trained on real data.

Chinese Translation

全切片图像（WSIs）是组织病理学的千兆像素级数字扫描，作为癌症诊断的主要依据。现有的组织病理学生成模型在补丁尺度上运行，生成的固定大小图块最多为1024×1024像素，仅涵盖WSI的一小部分。相比之下，临床诊断的文本报告和切片级诊断标签——在大规模上存在的监督——则在全切片级别上进行。我们提出了BigCell，这是首个以原生千兆像素分辨率生成连贯全切片组织病理图像的方法。BigCell在超过120,000个涵盖多种器官类型的公开WSI上，以2048×2048像素训练了一个流匹配变换器。通过对病理基础模型的嵌入进行条件化，并使用推理时的滑动窗口管道，BigCell生成高达32,768×32,768像素的连贯图像，在图块级和切片级的保真度指标上均优于先前的图块生成器。为了使千兆像素合成更具实用性，我们引入了一种自适应调度策略，优先处理重要的切片区域，使图像生成速度提高了超过3倍，且质量下降最小。此外，我们提出了首个以8k分辨率进行的文本到图像合成，通过训练一个单独的模型，直接从病理文本报告生成密集条件网格。我们证明，使用合成样本进行增强可以使切片级少样本分类性能比基于真实数据训练的基线提高多达18%。

View on arXiv Download PDF AI Translation

cs.CV / 14 / 2610.14265

Thinking with Images without Coordinates

无坐标的图像思维

Cao, Meng, Zhao, Haoze, Wang, Yi, Reid, Ian, Liang, Xiaodan

Abstract

Recent advances in multi-modal large language models have shifted toward the "thinking with images" paradigm to improve reasoning faithfulness, yet existing methods rely heavily on explicit coordinate- based grounding. We identify two fundamental limitations in this approach, i.e., a semantic gap where sparse regressions miss high-dimensional latent structures, and a resolution gap due to the granularity mismatch between patch-level tokens and pixel-level coordinates. To address these, we propose TiCO (Thinking with images without COordinates). It reformulates visual reasoning as a semantic evidence discovery process within the latent space, and evidence is injected as token-level regions without pixel coordinates. Specifically, TiCO utilizes a dedicated probe token to estimate similarity distributions and performs importance sampling over spatially connected components to generate diverse, semantically dominant evidence rollouts. Extensive experiments across multiple high-resolution and general visual reasoning benchmarks demonstrate that our TiCO consistently outperforms state-of-the-art coordinate-based methods with significantly improved inference efficiency.

Chinese Translation

近期多模态大语言模型的进展已转向“图像思维”范式，以提高推理的可信度，但现有方法过于依赖显式的基于坐标的基础。我们识别出该方法的两个基本局限性，即稀疏回归错过高维潜在结构的语义差距，以及由于补丁级别标记与像素级别坐标之间的粒度不匹配而导致的分辨率差距。为了解决这些问题，我们提出了TiCO（无坐标的图像思维）。它将视觉推理重新表述为潜在空间中的语义证据发现过程，并将证据作为不带像素坐标的标记级区域注入。具体而言，TiCO利用专用探测标记来估计相似性分布，并在空间连接组件上执行重要性采样，以生成多样化且语义主导的证据展开。在多个高分辨率和一般视觉推理基准上的广泛实验表明，我们的TiCO在推理效率显著提高的情况下，始终优于最先进的基于坐标的方法。

View on arXiv Download PDF AI Translation

cs.CV / 15 / 2610.14269

RoboFake: Are Deepfake Detectors Ready for Humanoid Face Robots?

RoboFake：深度伪造检测器是否准备好应对类人面部机器人？

Chen, Hao, Xu, Junnan

Abstract

Humanoid face robots can now move their lips in sync with speech and show detailed facial expressions. Give such a robot the face of a real person and drive it with generated speech, and you get a face forgery that happens in front of the camera: the camera records a real object, and no pixel is edited. We call this a RoboFake. In this paper we ask what current deepfake detectors do when they see a robot face. We test six public detectors trained on FaceForensics++ on video of a physical lip-syncing face robot, on a synthesized video of the same robot, and on rendered robot faces. All thresholds are fixed on FaceForensics++, and every result is compared with real human faces that went through the same processing as the robot videos. At a threshold that flags 5% of real faces, three detectors flag almost every physical robot frame, while two flag fewer than 7%. A matched human control and masking show that the response comes from the face, not from the processing or the background. However, it does not track synthesis: with only the face visible, three of the six detectors give the physical robot a higher fake score than its synthesized video. The response is also fragile, and light JPEG compression is enough to push most robot frames below the threshold for some detectors. Our results suggest that current detectors flag today's robots because their faces do not look human, not because they recognize a forgery. We release our evaluation protocol and code, and state three testable hypotheses about realistic, interactive RoboFakes.

Chinese Translation

类人面部机器人现在能够与语音同步移动嘴唇，并展示细致的面部表情。给这样的机器人一个真实人物的面孔，并用生成的语音驱动它，就会在摄像头前产生面孔伪造：摄像头记录的是一个真实物体，没有任何像素被编辑。我们称之为RoboFake。在本文中，我们探讨当前的深度伪造检测器在看到机器人面孔时的表现。我们测试了六个在FaceForensics++上训练的公共检测器，使用了一个物理的同步唇动面部机器人视频、同一机器人的合成视频以及渲染的机器人面孔。所有阈值均固定在FaceForensics++上，每个结果都与经过与机器人视频相同处理的真实人脸进行比较。在标记5%真实人脸的阈值下，三个检测器几乎标记了每一帧物理机器人，而两个检测器标记的少于7%。匹配的人类对照和遮罩显示，反应来自面孔，而不是来自处理或背景。然而，它并未跟踪合成：在仅显示面孔的情况下，六个检测器中的三个给物理机器人一个比其合成视频更高的伪造评分。该反应也很脆弱，轻微的JPEG压缩就足以使大多数机器人帧低于某些检测器的阈值。我们的结果表明，当前的检测器标记今天的机器人是因为它们的面孔看起来不人类，而不是因为它们识别出伪造。我们发布了我们的评估协议和代码，并提出了关于现实互动RoboFakes的三个可测试假设。

View on arXiv Download PDF AI Translation

cs.CV / 16 / 2610.14281

TGSR: Text-Grounded Diffusion for Faithful Image Super-Resolution

TGSR：基于文本的扩散模型用于真实图像超分辨率

Hu, Zhiming, Zhang, Ran, Ye, Angela, Goyal, Raghav, Ali, Faraz, Aumentado-Armstrong, Tristan Ty, Rangrej, Samrudhdhi B., Derpanis, Konstantinos G., Mohomed, Iqbal, Levinshtein, Alex

Abstract

Diffusion models have achieved remarkable success in image super-resolution (SR) by leveraging strong generative priors to enhance perceptual quality. However, for textual content, these priors often produce text hallucinations: non-existent or distorted text, which alter scene semantics and compromise model reliability. While recent methods focus on enhancing text generation, they do so indiscriminately, leaving hallucinations fundamentally unresolved. We argue the core problem is a lack of selective control. Therefore, we reformulate text SR as a selective generation problem: text is resolved only when reliable generation is expected, while other scene content is fully super-resolved. We introduce TGSR (Text-Grounded Super-Resolution), a test-time framework that progressively identifies and suppresses likely hallucinated text regions during generation. To enable this, we train dedicated detectors on a large-scale dataset curated from diverse diffusion-based SR outputs. TGSR consistently improves text fidelity across multiple backbones, increasing recognition precision by up to 17.68 points without degrading perceptual quality, underscoring the value of selective control in generative SR.

Chinese Translation

扩散模型通过利用强大的生成先验在图像超分辨率（SR）中取得了显著成功，从而增强了感知质量。然而，对于文本内容，这些先验往往会产生文本幻觉：即不存在或扭曲的文本，这会改变场景语义并影响模型的可靠性。尽管近期的方法专注于增强文本生成，但它们的处理方式往往是无差别的，根本没有解决幻觉问题。我们认为核心问题在于缺乏选择性控制。因此，我们将文本超分辨率重新表述为一个选择性生成问题：仅在预期可靠生成时解决文本，而其他场景内容则完全进行超分辨率处理。我们提出了TGSR（基于文本的超分辨率），这是一个在生成过程中逐步识别和抑制可能幻觉文本区域的测试时框架。为此，我们在一个从多样化的基于扩散的超分辨率输出中策划的大规模数据集上训练了专门的检测器。TGSR在多个基础模型中持续提高了文本的保真度，识别精度提高了多达17.68个百分点，而没有降低感知质量，突显了在生成超分辨率中选择性控制的价值。

View on arXiv Download PDF AI Translation

cs.CV / 17 / 2610.14293

GeoFix: Geometry Fixing for 4D Consistent Generation

GeoFix：用于4D一致性生成的几何修正

Hong, Yeobin, Go, Hyojun, Lee, Dohun, Lee, Jeongjae, Hong, Sunghwan, Tombari, Federico, An, Zhaochong, Ye, Jong Chul, Schindler, Konrad, Narnhofer, Dominik

Abstract

Video diffusion models increasingly serve as priors for 4D generation, but the generated videos often lack geometric consistency. Recent approaches address this through post-training by designing geometric rewards, yet this can degrade texture and fine detail, requiring auxiliary appearance rewards and careful reward balancing. To address this challenge, we propose GeoFix, a unified framework that improves geometric consistency and visual quality with a single geometric reward, applicable to both training and inference-time reward alignment. Our reward measures cross-view depth consistency in static scene regions, targeting structural inconsistencies without directly constraining appearance or penalizing valid object motion. Crucially, depth agreement increases with camera field-of-view overlap even in real videos and 4D datasets, allowing a smaller camera movement to be mistaken for better geometry. We account for this effect by calibrating the reward against the overlap-dependent trend. GeoFix achieves a 39% improvement in WorldScore 3D consistency and a 4% increase in VBench Quality Score.

Chinese Translation

视频扩散模型越来越多地作为4D生成的先验，但生成的视频往往缺乏几何一致性。近期的方法通过设计几何奖励进行后训练来解决这一问题，但这可能会降低纹理和细节，需依赖辅助外观奖励和仔细的奖励平衡。为了解决这一挑战，我们提出了GeoFix，一个统一框架，通过单一的几何奖励来提高几何一致性和视觉质量，适用于训练和推理时的奖励对齐。我们的奖励衡量静态场景区域的视角深度一致性，针对结构不一致性而不直接限制外观或惩罚有效的物体运动。关键是，即使在真实视频和4D数据集中，深度一致性也随着相机视场重叠的增加而提高，这使得较小的相机移动可能被误认为是更好的几何。我们通过将奖励与重叠依赖的趋势进行校准来考虑这一效应。GeoFix在WorldScore 3D一致性上实现了39%的提升，并在VBench质量评分上提高了4%。

View on arXiv Download PDF AI Translation

cs.CV / 18 / 2610.14315

MvFFN: Multi-view Floor plan Feed-forward Network for Unposed Wide-Baseline Panorama Layout Reconstruction

MvFFN：用于无姿态宽基线全景布局重建的多视角平面前馈网络

Li, Yuguang, Deng, Yichuan, Liu, Zixuan, Boyadzhiev, Ivaylo, Wu, Xiuyuan, Zou, Xinyu, Shapiro, Linda, Colburn, Alex

Abstract

Reconstructing accurate camera poses and floor-plan layouts from sparse, unposed, wide-baseline RGB panoramas remains a challenging open problem. Direct multi-room floor-plan prediction is especially difficult in this setting: pose, geometry, semantics, and global layout structure are tightly coupled, yet a complete floor-plan is hard to learn as a single global output. Our key insight is to keep the model's predictions strictly local. Multi-view Ordered Wall Instance Segmentation (MOIS) is designed around this concept. Each panorama column predicts only the wall it observes plus the next two adjacent walls in clockwise order. The global wall partition, room layouts, and shared coordinates emerge from deterministic aggregation of overlapping local predictions. Building on MOIS, we propose Multi-view floor-plan Feed-Forward Network (MvFFN), which jointly predicts coarse camera poses, dense global geometry, and per-wall semantic outputs. A downstream Layout Chaining (LC) post processing pipeline then assembles these predictions into multi-room floor-plans with shared global coordinates, consistent wall identities, and connected room structure. The dense MOIS predictions recover cross-view wall identity and local within-room connectivity more accurately than composed modular baselines, and the assembled layouts substantially outperform prior methods on wall and junction level accuracy. We also introduce the ZInD-CrossView dataset, which augments ZInD with globally unique wall instances and cross-view segmentation labels for this task.

Chinese Translation

从稀疏的无姿态宽基线RGB全景中重建准确的相机姿态和楼层平面布局仍然是一个具有挑战性的开放问题。在这种情况下，直接进行多房间楼层平面预测尤其困难：姿态、几何、语义和全局布局结构紧密耦合，但作为单一全局输出学习完整的楼层平面非常困难。我们的关键见解是保持模型的预测严格局部化。多视角有序墙实例分割（Multi-view Ordered Wall Instance Segmentation, MOIS）正是围绕这一概念设计的。每个全景列仅预测它观察到的墙以及顺时针方向上相邻的两面墙。全局墙体划分、房间布局和共享坐标通过重叠局部预测的确定性聚合而产生。在MOIS的基础上，我们提出了多视角平面前馈网络（Multi-view floor-plan Feed-Forward Network, MvFFN），该网络联合预测粗略的相机姿态、密集的全局几何和每面墙的语义输出。下游的布局链式（Layout Chaining, LC）后处理管道随后将这些预测组装成具有共享全局坐标、一致墙体身份和连接房间结构的多房间楼层平面。密集的MOIS预测比组合模块基线更准确地恢复跨视图的墙体身份和房间内的局部连接性，而组装后的布局在墙体和交汇点级别的准确性上显著优于先前的方法。我们还引入了ZInD-CrossView数据集，该数据集为此任务增强了ZInD，提供了全局唯一的墙实例和跨视图分割标签。

View on arXiv Download PDF AI Translation

cs.CV / 19 / 2610.14324

EV-JEPA: Elastic V-JEPA for Any Device and Cross-Device Inference

EV-JEPA：适用于任何设备和跨设备推理的弹性 V-JEPA

Diko, Anxhelo, Lin, Jieyu, Ambrogio, Stefano, Li, Ziyun, Assran, Mahmoud, Moutakanni, Théo, De Salvo, Barbara, Liu, Chiao

Abstract

Foundation models like V-JEPA are increasingly deployed across a heterogeneous edge-cloud continuum spanning smart glasses, phones, robots, and servers. Yet, they are typically trained as rigid, fixed-capacity networks bound to one device, supporting neither reconfigurable compute budgets nor collaborative execution. Existing approaches address only one axis at a time: slimmable and NAS-based methods provide width elasticity but assume single-device inference, whereas split-computing methods enable cross-device collaboration only for fixed partitions and static capacities. We introduce Elastic V-JEPA (EV-JEPA), a single model trained once and deployable everywhere along the continuum. Its width is reconfigurable, and it is natively partitionable across heterogeneous devices. At its core is Transition-Aware Triplet Training (T3). At each step we pick a narrow and a wide width, and train three paths: each width alone, plus a hybrid path that runs the early layers narrow and the rest wide. The hybrid switch point is where one device hands off to another, so the same weights learn both single-device and cross-device execution. Across various benchmarks, EV-JEPA falls within 1 accuracy point of independently trained baselines while improving over previous elastic training approaches by up to 11 accuracy points. Narrow-to-wide cross-device hybrid execution recovers to within 1.1 points of the highest-capacity standalone model and supports many-to-one, any-to-any, and variable-split configurations. Finally, we show that the elastic principle extends beyond the backbone to an elastic token compression mechanism that exposes controllable trade-offs among on-device computation, latency, and accuracy.

Chinese Translation

基础模型如 V-JEPA 正在越来越多地部署在跨越智能眼镜、手机、机器人和服务器的异构边缘-云连续体中。然而，它们通常被训练为固定容量的刚性网络，绑定于单一设备，既不支持可重配置的计算预算，也不支持协同执行。现有的方法仅解决一个维度的问题：可缩减和基于神经架构搜索（NAS）的方法提供宽度弹性，但假设单设备推理，而分割计算方法仅支持固定分区和静态容量的跨设备协作。我们提出了弹性 V-JEPA（EV-JEPA），这是一个训练一次后可以在整个连续体中部署的单一模型。它的宽度是可重新配置的，并且可以在异构设备之间原生分区。其核心是过渡感知三重训练（T3）。在每一步中，我们选择一个窄宽度和一个宽宽度，并训练三条路径：每个宽度单独训练，以及一条混合路径，该路径在早期层使用窄宽度，后续层使用宽宽度。混合切换点是一个设备向另一个设备交接的地方，因此相同的权重同时学习单设备和跨设备执行。在各种基准测试中，EV-JEPA 的准确率与独立训练的基线相差不超过 1 个百分点，同时在之前的弹性训练方法上提高了多达 11 个百分点的准确率。窄到宽的跨设备混合执行恢复到与最高容量独立模型相差 1.1 个百分点，并支持多对一、任意对任意和可变分割配置。最后，我们展示了弹性原理超越了骨干网络，扩展到一种弹性令牌压缩机制，该机制揭示了设备内计算、延迟和准确性之间可控的权衡。

View on arXiv Download PDF AI Translation

cs.CV / 20 / 2610.14325

Manifold Navigation Learning Using Recurrent Networks: Efficient Training Techniques

使用递归网络的流形导航学习：高效训练技术

Dayton, Sarah Harkins, Rengaswami, Sathya, Maroulas, Vasileios, Boothe, David L., Schizas, Ioannis D.

Abstract

Position tracking is a well-studied problem in Euclidean navigation, yet comparatively little attention has been given to navigation on general manifolds, where motion is governed by the underlying geometry of the space. In this work, we formalize the problem of manifold navigation as a velocity-driven tracking task and propose a coordinate-chart-based learning framework in which learning models operate entirely in a Euclidean representation. Predicted trajectories are mapped to the manifold via an immersion function prior to error computation, allowing geometric consistency to be evaluated without explicitly enforcing manifold constraints during learning. Within this framework, we focus on Gated Recurrent Units (GRUs) for sequential position prediction on curved manifolds after comparisons with alternative recurrent architectures. We evaluate our GRU models, trained with pertinent custom loss functions, against both a standard MSE loss function and a naive vector-addition baseline. We evaluate our approach on both analytically (and numerically) defined manifolds, as well as a surface reconstructed from a LiDAR point cloud. Our results show that recurrent models can learn to approximate manifold-constrained dynamics from velocity sequences and consistently outperform Euclidean baselines. Our results demonstrate that certain custom loss functions significantly improve long-horizon prediction accuracy compared to conventional point-wise loss functions.

Chinese Translation

位置跟踪是欧几里得导航中一个研究较为深入的问题，然而对于一般流形上的导航，关注相对较少，其中运动受空间底层几何的支配。在本研究中，我们将流形导航问题形式化为一个基于速度的跟踪任务，并提出了一种基于坐标图的学习框架，在该框架中，学习模型完全在欧几里得表示中操作。预测的轨迹通过浸入函数映射到流形上，以便在计算误差之前进行几何一致性的评估，而无需在学习过程中明确施加流形约束。在这个框架内，我们专注于门控递归单元（Gated Recurrent Units, GRUs）在曲面流形上的顺序位置预测，并与其他递归架构进行了比较。我们评估了使用相关自定义损失函数训练的GRU模型，相较于标准均方误差（MSE）损失函数和简单的向量加法基线。我们在解析（和数值）定义的流形以及从LiDAR点云重建的表面上评估了我们的方法。我们的结果表明，递归模型能够从速度序列中学习近似流形约束的动力学，并且在性能上始终优于欧几里得基线。我们的结果还表明，与传统的逐点损失函数相比，某些自定义损失函数显著提高了长时间预测的准确性。

View on arXiv Download PDF AI Translation

cs.CV / 21 / 2610.14326

Real-Time Traffic Density Estimation: A Hybrid Computer Vision and Predictive Machine Learning Pipeline

实时交通密度估计：一种混合计算机视觉与预测机器学习的管道

Boadu, Prince Kwabena Appiah

Abstract

Traffic density estimation from camera feeds requires connecting visual perception with short-horizon forecasting. We present a two-stage pipeline that combines vehicle detection and tracking with a spatial logic layer, followed by a temporal estimator. The spatial layer filters bounding boxes by their overlap with a precomputed road mask and extracts per-frame traffic features. LSTM, GRU, and Temporal Convolutional Network (TCN) estimators use these features to predict five future steps from a ten-frame input window. We benchmark ten detector variants from six families on a dataset of 5,716 annotated frames using a single NVIDIA GPU. YOLOv5l achieves the highest mAP@50:95 of 0.524 at 56 frames per second, while YOLOv8n reaches 258 frames per second at 0.487 mAP@50:95; these rates measure detector inference without tracking or spatial filtering. Temporal evaluation on 29,243 held-out windows yields aggregate R-squared values of 0.858 for GRU, 0.856 for LSTM, and 0.840 for TCN. These metrics measure forecasts of detector-derived telemetry and are dominated by direction variance, rather than establishing absolute traffic-density accuracy. No road mask was available for the evaluation stream, limiting validation of the spatial filtering component. A live run on a third camera stream examines forecasting behavior within the complete pipeline. The results characterize detector speed-accuracy trade-offs and the limitations of short-window traffic telemetry forecasting.

Chinese Translation

从摄像头视频流中进行交通密度估计需要将视觉感知与短期预测相结合。我们提出了一种两阶段的管道，结合了车辆检测与跟踪以及空间逻辑层，随后是时间估计器。空间层通过与预先计算的道路掩膜的重叠情况过滤边界框，并提取每帧的交通特征。长短期记忆网络（LSTM）、门控循环单元（GRU）和时间卷积网络（TCN）估计器利用这些特征从十帧输入窗口中预测未来五个时间步。我们在一个包含5,716帧标注数据集上，使用单个NVIDIA GPU对六个家族的十种检测器变体进行了基准测试。YOLOv5l在56帧每秒的速度下达到了最高的mAP@50:95为0.524，而YOLOv8n在0.487 mAP@50:95的情况下达到了258帧每秒；这些速率是测量检测器推理时不考虑跟踪或空间过滤的结果。对29,243个保留窗口的时间评估得到了GRU的聚合R平方值为0.858，LSTM为0.856，TCN为0.840。这些指标测量了基于检测器推导的遥测的预测，主要受方向方差的影响，而不是建立绝对的交通密度准确性。评估流中没有可用的道路掩膜，限制了空间过滤组件的验证。在第三个摄像头流上的实时运行考察了整个管道内的预测行为。结果表征了检测器的速度-准确性权衡及短窗口交通遥测预测的局限性。

View on arXiv Download PDF AI Translation

cs.CV / 22 / 2610.14332

TraceSampler: Making Sampling Errors Predictable and Correctable for Accelerated Generative Sampling

TraceSampler：使加速生成采样中的采样误差可预测和可纠正

Matthew, George, Zhang, Guoqiang, Niwa, Kenta

Abstract

Accurate sampling from continuous-time generative models can require many model evaluations. Low-step sampling reduces cost, but coarse updates introduce endpoint errors that can accumulate and degrade generation quality. We introduce TraceSampler, a post-training method that corrects these errors without modifying the pretrained model. Rather than optimising low-step updates directly, TraceSampler models endpoint error as a sample-specific prediction problem. A higher-step reference integration over the same interval provides the supervised target, while a compact representation makes high-dimensional error predictable from the sampling trace. This representation combines shared residual structure, trace-derived directions, and a low-rank remainder, so shallow per-step MLP heads predict sample-dependent coefficients. These coefficients drive a two-stage learned residual correction: the first stage predicts the full endpoint residual, while an intermediate backbone evaluation provides new trace information to predict the remaining residual; this evaluation is reused by the following step. TraceSampler is evaluated with Rectified Flow on CIFAR-10, SiT-XL/2 on ImageNet-256, and SD3.5 Medium for text-to-image generation. At 10 steps, it reduces trajectory MSE relative to matched-step Euler by 90.5% on CIFAR-10 and 85.6% on ImageNet-256, while achieving FID scores of 2.692 and 2.189, close to 120-step Heun references of 2.601 and 2.056 with 24$\times$ fewer backbone evaluations. On SD3.5 Medium, TraceSampler improves image quality and prompt following; at 10 steps, ImageReward rises from 0.752 to 0.906 while trajectory MSE falls by 70.7%. These gains preserve Euler's backbone-evaluation budget, add less than 0.4% inference FLOPs, and incur negligible wall-clock overhead, showing that coarse-step sampling error can be made predictable from the sampling trace and efficiently corrected.

Chinese Translation

从连续时间生成模型中进行准确采样可能需要多次模型评估。低步采样降低了成本，但粗糙的更新引入了端点误差，这些误差可能会累积并降低生成质量。我们提出了TraceSampler，这是一种后训练方法，可以在不修改预训练模型的情况下纠正这些误差。TraceSampler并不是直接优化低步更新，而是将端点误差建模为一个样本特定的预测问题。在同一时间间隔内进行更高步数的参考积分提供了监督目标，而紧凑的表示使得高维误差可以从采样轨迹中预测。该表示结合了共享的残差结构、基于轨迹的方向和低秩余量，因此浅层每步多层感知器（MLP）头可以预测样本依赖的系数。这些系数驱动了一个两阶段的学习残差校正：第一阶段预测完整的端点残差，而中间的主干评估提供新的轨迹信息以预测剩余的残差；该评估在后续步骤中被重用。TraceSampler在CIFAR-10上的Rectified Flow、在ImageNet-256上的SiT-XL/2以及在文本到图像生成的SD3.5 Medium上进行了评估。在10步时，相较于匹配步数的欧拉方法，CIFAR-10上的轨迹均方误差（MSE）减少了90.5%，ImageNet-256上减少了85.6%，同时FID得分分别为2.692和2.189，接近120步Heun参考的2.601和2.056，且主干评估次数减少了24倍。在SD3.5 Medium上，TraceSampler提高了图像质量和提示遵循度；在10步时，ImageReward从0.752上升至0.906，而轨迹MSE下降了70.7%。这些提升保持了欧拉方法的主干评估预算，增加的推理FLOPs不足0.4%，且几乎没有时钟开销，表明粗步采样误差可以从采样轨迹中变得可预测并高效纠正。

View on arXiv Download PDF AI Translation

cs.CV / 23 / 2610.14351

Learning Transferable World-Action Models from Task-Paired Human Videos

从任务配对的人类视频中学习可转移的世界-动作模型

Park, Jeongeun, Kim, Taekyung, Han, Dongyoon, Yun, Sangdoo

Abstract

World-action models inherit broad visual and physical priors from video pre-training. However, standard within-video next-frame prediction allows them to rely on appearance shared between the context and future, rather than learning interactions that transfer across scenes and embodiments. We introduce WATCH, a cross-video pre-training that learns from task-paired human demonstrations collected across diverse environments. Conditioned on the video and actions of one demonstration, the model jointly predicts another demonstration's actions and future frames from its current observation. Predicting one demonstration from another encourages the model to learn task-relevant interactions that transfer across scenes and demonstrators. Our pre-training shows that these capabilities can be learned from readily scalable human-human pairs alone, without requiring matched human-robot demonstrations. Analyses before robot post-training show improved motion alignment across embodiments and better action and video prediction on unseen robots, with representations closer to those learned from robot pairs. After robot post-training, WATCH improves MimicDroid in-context success from 40.9% to 68.3% with Cosmos Policy and from 59.8% to 77.7% with Cosmos3, with the largest gains on the unseen levels. Gains persist under context-free post-training across robot embodiments. WATCH outperforms within-video pre-training and co-training baselines, supporting cross-video prediction as a scalable approach to learning transferable world-action representations.

Chinese Translation

世界-动作模型从视频预训练中继承了广泛的视觉和物理先验。然而，标准的场内下一帧预测使得它们依赖于上下文和未来之间共享的外观，而不是学习跨场景和体现的交互。我们提出了WATCH，一种跨视频预训练方法，从在多样环境中收集的任务配对人类演示中学习。在一个演示的视频和动作的条件下，该模型共同预测另一个演示的动作和未来帧。通过从一个演示预测另一个演示，鼓励模型学习跨场景和演示者的任务相关交互。我们的预训练表明，这些能力可以仅通过易于扩展的人类-人类配对学习，而无需匹配的人类-机器人演示。在机器人后训练之前的分析显示，在不同体现之间的运动对齐得到了改善，并且在未见过的机器人上，动作和视频预测表现更佳，表示更接近于从机器人配对中学习到的表示。在机器人后训练后，WATCH使MimicDroid在上下文成功率上从40.9%提高到68.3%（使用Cosmos Policy），从59.8%提高到77.7%（使用Cosmos3），在未见过的关卡上获得了最大的提升。在跨机器人体现的无上下文后训练下，提升依然存在。WATCH的表现优于场内预训练和共同训练基线，支持跨视频预测作为学习可转移世界-动作表示的可扩展方法。

View on arXiv Download PDF AI Translation

cs.CV / 24 / 2610.14397

RealityBench: Simulated Environments to Evaluate a Vision Language Model's Ability to Document Human Physical Work

RealityBench：评估视觉语言模型记录人类物理工作的能力的模拟环境

Perrault, Andrew, Takiar, Anmol, Wienecki, Dominik, Nandi, Arnab

Abstract

Augmenting worker capabilities in the physical world through guidance has been a decades-long goal for AI agents. One key supporting capability is to understand what the worker is doing, seeing through the human's camera or smart glasses, and assist them in documentation. Despite advances in vision language models, their capabilities in these settings remain uncertain and are challenging to benchmark because most physical work data is proprietary, and involves too much variability to study in a principled manner. Unlike structured settings such as kitchens and warehouses that have predictable lighting, geometry, and objects, field-facing tasks such as rooftop solar installation and vehicle repair are unconstrained and high variance. Here, we model real-world physical work in Blender across eleven environments, incorporating variants in both image acceptability and documentation. The acceptability task asks whether a particular view would be suitable proof-of-work and documentation asks whether an agent can understand the task-relevant state of the environment from that view. Across 26 vision-language models, we find that top models largely solve documentation when the answer is text in frame, but are weak at reading measurements and at judging small parts and spatial relations, such as whether a tape measure leans. Acceptability is harder to pin down because human raters disagree with each other. Measured against a scorer that agrees with the raters' majority about as often as a single rater does, the best models cannot be distinguished from a single non-author rater, at a cost within the range of a human pass. In the three environments where we can trace acceptability errors, most wrong rejections come from misreading the scene rather than the requirement. On the three documentation questions for which we have real images, model rankings on rendered frames transfer to real images.

Chinese Translation

通过指导增强工人在物理世界中的能力一直是人工智能代理数十年来的目标。一个关键的支持能力是理解工人在做什么，通过人类的相机或智能眼镜观察，并协助他们进行文档记录。尽管视觉语言模型取得了进展，但它们在这些环境中的能力仍然不确定，并且难以进行基准测试，因为大多数物理工作数据是专有的，并且涉及太多变异，无法以原则性方式进行研究。与厨房和仓库等具有可预测照明、几何形状和物体的结构化环境不同，屋顶太阳能安装和车辆维修等面向现场的任务是不受限制且变异性高的。在这里，我们在Blender中模拟了现实世界的物理工作，涵盖了十一种环境，结合了图像可接受性和文档记录的变体。可接受性任务询问特定视图是否适合作为工作证明，而文档记录则询问代理是否能从该视图理解与任务相关的环境状态。在26个视觉语言模型中，我们发现当答案是框内文本时，顶级模型在文档记录方面表现良好，但在读取测量值和判断小部件及空间关系（例如卷尺是否倾斜）方面较弱。可接受性更难以确定，因为人类评审者之间存在分歧。与一个与评审者大多数意见一致的评分者相比，最佳模型无法与单个非作者评审者区分开，且成本在人工通过的范围内。在我们能够追踪可接受性错误的三个环境中，大多数错误拒绝源于对场景的误读，而非要求。在我们拥有真实图像的三个文档记录问题中，模型在渲染帧上的排名可以转移到真实图像上。

View on arXiv Download PDF AI Translation

cs.CV / 25 / 2610.14403

VisualCoding: Learning Executable Skills for Reliable Visual Reasoning

视觉编码：学习可执行技能以实现可靠的视觉推理

Jiaqi, Liu, Ruinan, Jin, Xiaoxiao, Li

Abstract

Learning reusable skills can improve visual reasoning agents, yet applying a skill still requires the agent to implement or invoke a procedure and produce the final answer. Errors and hallucinations at these steps can prevent acquired experience from translating into reliable single-run performance. We propose VisualCoding, which combines execution-guided acquisition of reusable visual programs with direct adoption of their answers on applicable inputs. Execution-Guided Program Acquisition composes visual primitives and additional code into family-level solvers, then iteratively tests and revises them on labeled training instances. Scope-Aware Direct Execution runs a retained solver and returns its valid in-scope answer. We evaluate VisualCoding on four models across four benchmarks spanning 69 task families. VisualCoding exceeds the strongest evaluated baseline in mean accuracy by 4.45 percentage points with Gemini-3.1-Pro and by 14.78 points when reusing the frozen Pro-acquired artifacts with Flash-Lite. Ablations support execution-guided acquisition and show that direct execution outperforms textual or callable reuse of the same programs. Code is available at https://github.com/jiaq-liu/VisualCoding.

Chinese Translation

学习可重用的技能可以提升视觉推理代理的能力，但应用技能仍然要求代理实施或调用一个过程并产生最终答案。这些步骤中的错误和幻觉可能会阻碍获得的经验转化为可靠的单次运行性能。我们提出了视觉编码（VisualCoding），它结合了可重用视觉程序的执行引导获取与在适用输入上直接采用其答案。执行引导程序获取（Execution-Guided Program Acquisition）将视觉原语和附加代码组合成家族级求解器，然后在标记的训练实例上进行迭代测试和修订。范围感知直接执行（Scope-Aware Direct Execution）运行保留的求解器并返回其有效的范围内答案。我们在四个模型上评估视觉编码，涵盖69个任务家族的四个基准。视觉编码在平均准确率上超越了最强的评估基线，使用Gemini-3.1-Pro提高了4.45个百分点，而在重用冻结的Pro获取的工件与Flash-Lite时提高了14.78个百分点。消融实验支持执行引导获取，并表明直接执行优于文本或可调用的相同程序的重用。代码可在 https://github.com/jiaq-liu/VisualCoding 获取。

View on arXiv Download PDF AI Translation

cs.CV / 26 / 2610.14439

SceneForesee: Learning Scene Evolution for Spatiotemporal Reasoning

场景预见：学习场景演变以进行时空推理

Li, Chaoyue, Han, Shoudong, Wu, Haoyang, Yang, Boxue, Chen, Sijia, Chen, Yufan, Zhang, Ruping

Abstract

Large multimodal models (LMMs) need to understand how entities move and interact as a scene develops. This requires following the same entities over time and relating their observed motion to subsequent changes. We formulate Scene Evolution Understanding (SEU), a task for predicting and interpreting this change. Given observed video, a model estimates entity histories and predicts one scene continuation through future trajectories and Events, accompanied by Visual Evidence from observed motion and a scene description. We construct SceneForesee, a benchmark for evaluating entity continuity and scene evolution reasoning through paired observed and future scenes with shared entity identities. It contains 144.57K temporal pairs, 1.05M entity correspondences, and 272.49K Events paired with observed evidence. We propose a spatiotemporal reasoning framework, EvoSee, that learns future geometry and Event semantics through shared candidate states and joint assignment. It selects one continuation per entity from observation and uses the selected changes for entity refinement, retention, and scene representation. On SceneForesee, EvoSee outperforms Qwen3.5-9B by 10.45 points in Combined Point-F1 and 17.24 points in Event Global F1. On the 4D reasoning benchmark Dyn-Bench, it improves question answering accuracy and grounding J&F by 4.8 and 3.58 points over their respective base systems. These results support scene evolution as both a prediction target and a dynamic representation that brings predicted entity changes into spatiotemporal reasoning.The project is available at https://github.com/mikubaka88/SceneForesee-Spatiotemporal-Reasoning.

Chinese Translation

大型多模态模型（LMMs）需要理解实体在场景发展过程中如何移动和互动。这要求随着时间的推移跟踪相同的实体，并将其观察到的运动与随后的变化关联起来。我们提出了场景演变理解（Scene Evolution Understanding, SEU），这是一个用于预测和解释这种变化的任务。给定观察到的视频，模型估计实体历史，并通过未来轨迹和事件预测一个场景的延续，同时提供来自观察到的运动和场景描述的视觉证据。我们构建了场景预见（SceneForesee），这是一个通过配对的观察场景和未来场景（共享实体身份）来评估实体连续性和场景演变推理的基准。该基准包含144.57K个时间对、1.05M个实体对应关系和272.49K个与观察证据配对的事件。我们提出了一种时空推理框架EvoSee，通过共享候选状态和联合分配学习未来几何和事件语义。它从观察中为每个实体选择一个延续，并使用所选的变化进行实体细化、保留和场景表示。在场景预见基准上，EvoSee在综合点-F1上超越Qwen3.5-9B 10.45分，在事件全局F1上超越17.24分。在4D推理基准Dyn-Bench上，它在问题回答准确性和基础系统的J&F定位上分别提高了4.8和3.58分。这些结果支持场景演变作为一个预测目标和动态表示，将预测的实体变化引入时空推理。该项目可在 https://github.com/mikubaka88/SceneForesee-Spatiotemporal-Reasoning 获取。

View on arXiv Download PDF AI Translation

cs.CV / 27 / 2610.14445

FreeCHROMA: Generative Chroma Keying from Unaligned Background References

FreeCHROMA：基于未对齐背景参考的生成性色度键控

Wu, Jiayi, Wang, Tianfu, Zhao, Rui, Schiffers, Florian, Chen, Zhimin, Fermuller, Cornelia, Yang, Yezhou, Metzler, Christopher, Aloimonos, Yiannis

Abstract

In film production, chroma keying composites an approved foreground performance into a new environment, but conventional workflows require coordinated background plates and dedicated harmonization. Recent generative methods often require pre-aligned backgrounds, focus on background-conditioned foreground relighting, or represent the environment only through an edited first frame. These restrictions limit background selection, scene composition, and camera motion while adding preparation or editing effort. We present FreeCHROMA, which adapts a pretrained video diffusion model in a parameter-efficient manner for generative video compositing from one to three unaligned background references. The model integrates the foreground into the referenced environment, preserving foreground identity, motion, and fine visual details while generating a spatiotemporally coherent background. Foreground-background harmonization occurs within the same generation process, without a separate relighting stage. Optional text instructions help disambiguate complex spatial relationships and camera behavior. To train the model, we construct pseudo-paired data from readily available real monocular videos through foreground-background separation and augmentation with existing foundation models. The original videos serve as training targets, while automated processing and quality verification enable scalable data construction. Across reference fidelity, foreground preservation, scene integration, harmonization, and overall video quality, FreeCHROMA consistently outperforms recent compositing and conditional video-generation baselines, with particularly clear gains under camera motion. FreeCHROMA relaxes the constraints of existing tools to enable end-to-end generative chroma keying from unaligned inputs, with exploratory experiments demonstrating its potential in robotics.

Chinese Translation

在电影制作中，色度键控将经过批准的前景表演合成到新的环境中，但传统工作流程需要协调的背景画面和专门的和谐处理。最近的生成方法通常要求预对齐的背景，专注于基于背景的前景重光照，或仅通过编辑的第一帧来表示环境。这些限制限制了背景选择、场景构图和摄像机运动，同时增加了准备或编辑的工作量。我们提出了FreeCHROMA，它以参数高效的方式适应了预训练的视频扩散模型，从一个到三个未对齐的背景参考中进行生成性视频合成。该模型将前景整合到参考环境中，保持前景的身份、运动和细微的视觉细节，同时生成时空一致的背景。前景与背景的和谐处理在同一生成过程中进行，无需单独的重光照阶段。可选的文本指令有助于消除复杂空间关系和摄像机行为的歧义。为了训练模型，我们通过前景与背景分离以及使用现有基础模型进行增强，从现成的真实单目视频中构建伪配对数据。原始视频作为训练目标，而自动化处理和质量验证使得数据构建具有可扩展性。在参考保真度、前景保留、场景整合、和谐处理和整体视频质量方面，FreeCHROMA始终优于最近的合成和条件视频生成基线，尤其在摄像机运动下表现出明显的提升。FreeCHROMA放宽了现有工具的限制，使得从未对齐输入中实现端到端的生成性色度键控成为可能，探索性实验展示了其在机器人技术中的潜力。

View on arXiv Download PDF AI Translation

cs.CV / 28 / 2610.14499

DSR-Net: Learning Dynamic Skeletal Relations for 3D Human Pose Estimation

DSR-Net：学习动态骨骼关系以进行3D人体姿态估计

Li, Fan, Xiang, Jian, Xue, LinLin, Gao, Yuwei

Abstract

3D human pose estimation requires accurate modeling of motion-dependent skeletal structures and inter-joint dependencies. Existing methods typically rely on fixed skeletal topologies in Graph Convolutional Networks (GCNs) or implicitly capture joint relations through attention mechanisms, limiting their adaptability to diverse motion patterns. To address this issue,we proposeDSR-Net, a dynamic structural representation learning framework for 3D human pose estimation. Specifically, a Dynamic Skeletal Relation (DSR) module is introduced to learn adaptive inter-joint dependencies beyond predefined skeletal connections. Based on the learned relations, a hierarchical relation-aware fusion strategy is developed to facilitate effective interaction between structural and temporal representations. In addition, a lightweight Residual Skeletal Recalibration Module (RSRM) is designed to enhance the stability of structure-aware features with minimal computational overhead. Extensive experiments on Human3.6M and MPI-INF-3DHP demonstrate that DSR-Net achieves MPJPEs of 37.9 mm and 14.9 mm, respectively, consistently improving over strong baselines. The results validate the effectiveness of dynamic skeletal relation modeling for robust 3D human pose estimation.

Chinese Translation

3D人体姿态估计需要准确建模依赖于运动的骨骼结构和关节间的依赖关系。现有方法通常依赖于图卷积网络（Graph Convolutional Networks, GCNs）中的固定骨骼拓扑结构，或通过注意力机制隐式捕捉关节关系，这限制了它们对多样化运动模式的适应性。为了解决这一问题，我们提出了DSR-Net，一种用于3D人体姿态估计的动态结构表示学习框架。具体而言，引入了动态骨骼关系（Dynamic Skeletal Relation, DSR）模块，以学习超越预定义骨骼连接的自适应关节间依赖关系。基于学习到的关系，开发了一种层次关系感知融合策略，以促进结构表示和时间表示之间的有效交互。此外，设计了一种轻量级残差骨骼重校准模块（Residual Skeletal Recalibration Module, RSRM），以在最小计算开销下增强结构感知特征的稳定性。在Human3.6M和MPI-INF-3DHP上的大量实验表明，DSR-Net分别实现了37.9毫米和14.9毫米的MPJPE，始终优于强基线。结果验证了动态骨骼关系建模在稳健的3D人体姿态估计中的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 29 / 2610.14512

Spatiotemporal Recognition of Abdominal Wall Layers During Optical Trocar Entry

光学穿刺器进入过程中腹壁层的时空识别

Athreya, Shreeram, Melehy, Andrew, Prentice, Carter, Robinson, Michael, Arnold, Corey, Wu, James, Speier, William

Abstract

Objective. To evaluate whether spatial context and temporal modeling improve tissue recognition during optical trocar entry, where multiple abdominal-wall tissues may be visible simultaneously. Methods. We retrospectively analyzed 71 recordings with 1,516 annotated keyframes and 29,151 clinician-annotated sectors. Each frame was divided into 28 polar-grid sectors assigned to nine tissue classes. Color, neighboring-sector, and positional features were used by a gradient-boosted emission classifier. The full pipeline added learned temporal smoothing, spatial regularization, an anatomically ordered hidden Markov model, and posterior blending. Five-fold cross-validation was grouped by recording, with cross-fitting to prevent within-recording leakage. Primary evaluation used clinician-derived targets; ablations assessed classification, probability quality, spatial consistency, and peritoneum-or-deeper onset timing. Results. The complete pipeline achieved accuracy 0.646, macro-F1 0.507, and macro-AUROC 0.882, versus 0.600, 0.440, and 0.861 for emissions alone. Macro-F1 improved by 0.066 across all five folds. Label changes decreased from 66.2 to 19.5 per sector-minute, and median onset error decreased from 15.58 to 2.45 seconds. Frames with a tissue-depth-code span of at least three decreased from 53% to 26%, versus 22% for clinician annotations. Performance remained limited for preperitoneal fat and bowel, and the full model retained a 44% premature-onset proportion. Conclusions. Combining appearance, spatial context, and temporal constraints improved retrospective tissue recognition and onset localization. Tissue-specific limitations, residual timing errors, and reliance on future observations warrant external validation and prospective evaluation of causal models before clinical use.

Chinese Translation

目的。评估空间上下文和时间建模是否能改善光学穿刺器进入过程中的组织识别，在此过程中，多个腹壁组织可能会同时可见。方法。我们回顾性分析了71个录音，包含1,516个标注关键帧和29,151个临床标注的区域。每帧被划分为28个极坐标网格区域，分配给九个组织类别。使用梯度提升发射分类器提取颜色、邻近区域和位置特征。完整的处理流程增加了学习的时间平滑、空间正则化、解剖顺序的隐马尔可夫模型和后验融合。采用五折交叉验证，按录音分组，并进行交叉拟合以防止录音内泄漏。主要评估使用临床派生目标；消融评估分类、概率质量、空间一致性和腹膜或更深层次的起始时机。结果。完整的处理流程实现了0.646的准确率，0.507的宏观F1值和0.882的宏观AUROC，而仅发射的结果分别为0.600、0.440和0.861。所有五折的宏观F1值提高了0.066。每个区域每分钟的标签变化从66.2减少到19.5，中位起始误差从15.58秒减少到2.45秒。具有至少三层组织深度代码跨度的帧比例从53%降至26%，而临床标注为22%。在前腹膜脂肪和肠道的表现仍然有限，完整模型保留了44%的早期起始比例。结论。结合外观、空间上下文和时间约束改善了回顾性组织识别和起始定位。组织特定的局限性、残余时间误差以及对未来观察的依赖，要求在临床使用之前进行外部验证和因果模型的前瞻性评估。

View on arXiv Download PDF AI Translation

cs.CV / 30 / 2610.14513

TartanMatch: Towards Universal Dense Correspondence Across Modalities

TartanMatch：迈向跨模态的通用稠密对应

Kwon, Hyeokjoon, Cai, Jiting, Li, Ruogu, Bhandari, Kritan, Hemkumar, Geethika, Maheshwari, Parv, Qiu, Yuheng, Zhang, Yuchen, Scherer, Sebastian, Wang, Wenshan

Abstract

Robotic platforms combine multi-modal sensors to operate across diverse conditions, but most dense correspondence methods are limited to RGB-RGB or a small set of modality pairs. Training a model for each modality pair scales poorly with the number of sensors and forgoes potential positive transfer across modality pairs. We introduce TartanMatch, a unified model for dense correspondence across all 25 ordered pairs among RGB, event, thermal, depth, and projected LiDAR. TartanMatch adapts an RGB-pretrained correspondence model to all modality pairs within one shared network using lightweight modality adaptation and a mixture of real and synthetic data. On zero-shot real-world benchmarks, TartanMatch reduces Endpoint Error (EPE) over the strongest baselines by 61.9% for cross-modal matching and 49.6% for same-modal matching. We further show that joint training across all modality pairs improves average matching accuracy over pair-specific training, especially on challenging cross-modal pairs. These results demonstrate that a single model achieves broad multi-modal coverage and state-of-the-art matching accuracy.

Chinese Translation

机器人平台结合多模态传感器以在多样化条件下操作，但大多数稠密对应方法仅限于RGB-RGB或少数模态对。为每对模态训练一个模型在传感器数量增加时扩展性较差，并且放弃了跨模态对的潜在正向迁移。我们提出了TartanMatch，这是一个统一的模型，能够在RGB、事件、热成像、深度和投影LiDAR之间的25个有序对中实现稠密对应。TartanMatch通过轻量级模态适配和真实与合成数据的混合，调整一个RGB预训练的对应模型，以在一个共享网络中适应所有模态对。在零样本真实世界基准测试中，TartanMatch在跨模态匹配上将端点误差（EPE）减少了61.9%，在同模态匹配上减少了49.6%。我们进一步展示了跨所有模态对的联合训练在平均匹配准确性上优于特定对训练，尤其是在具有挑战性的跨模态对上。这些结果表明，单一模型实现了广泛的多模态覆盖和最先进的匹配准确性。

View on arXiv Download PDF AI Translation

cs.CV / 31 / 2610.14514

Compositional Concepts Erasing: Disentangling Object-Style via Domain Manifold Unlearning

组合概念消除：通过领域流形遗忘解耦对象风格

Xiao, Yang, Deng, Kaiyuan, Li, Gen, Ling, Zinan, Fan, Zhiwen, Ma, Xiaolong, Hui, Bo

Abstract

Diffusion models pose significant risks regarding the generation of illicit content and copyright violations. This has catalyzed the development of machine unlearning to erase harmful concepts or copyrighted styles from a model's generative repertoire without compromising its general performance. However, previous unlearning approaches on diffusion models have largely focused on either object-level or style-level removal and failed to address the challenge of object-style combination unlearning. At the same time, multiple unlearning requests often lead to catastrophic forgetting. To address these challenges, we propose Domain Manifold Unlearning (DMU), which uses an EMA-updated feature proxy extracted from intermediate diffusion features to represent the retained target-object direction. As a lightweight estimating representation of an object, DMU enables selective unlearning of domain-specific components in a specific object while preserving other semantic representations, thereby significantly enhancing robustness in multiple combination unlearning scenarios. Extensive experiments show that DMU significantly mitigates catastrophic forgetting, outperforming state-of-the-art methods of batch unlearning and sequential unlearning.

Chinese Translation

扩散模型在生成非法内容和侵犯版权方面存在重大风险。这促使了机器遗忘的发展，以便在不影响模型整体性能的情况下，从模型的生成库中抹去有害概念或版权风格。然而，之前针对扩散模型的遗忘方法主要集中在对象级别或风格级别的去除，未能解决对象风格组合遗忘的挑战。同时，多次遗忘请求往往导致灾难性遗忘。为了解决这些挑战，我们提出了领域流形遗忘（Domain Manifold Unlearning, DMU），它使用从中间扩散特征中提取的EMA更新特征代理来表示保留的目标对象方向。作为对象的轻量级估计表示，DMU能够在保留其他语义表示的同时，有选择性地遗忘特定对象中的领域特定组件，从而显著增强在多重组合遗忘场景中的鲁棒性。大量实验表明，DMU显著减轻了灾难性遗忘，超越了当前批量遗忘和顺序遗忘的最先进方法。

View on arXiv Download PDF AI Translation

cs.CV / 32 / 2610.14517

We Have Been Interpreting Diffusion Transformer Backbone Behaviors Wrong

我们一直错误地解读扩散变换器骨干网络的行为

Meng, Benyuan, Huang, Longtao, Liu, Yang, Lin, Liang

Abstract

During a network forward call, diffusion backbones naturally produce many intermediate activations. If we evaluate each activation against an identical downstream benchmarking protocol, we can plot an inverted-U-shaped curve as the depth profile: the activation's performance goes up first and then down as we go into deeper layers. This is common knowledge in the diffusion model community, which inspires many applications including the widely-adopted REPA. However, the common interpretation of this depth profile may be inaccurate and incomplete: (i) The cause of this curve has long been treated as signal granularity changing from coarse to fine, but there has been little evidence in DiT backbones. (ii) The community has been mainly studying this phenomenon in a static way, without investigating how it changes across different training setups. In this work, we show a different mechanism picture for the profile, where the prediction target (e.g., $\epsilon$, $u$, $x_0$) sits at the center. The semantically meaningful signal is only an intermediate by-product for predicting the target, while the prediction target is the real goal for the backbone. Furthermore, the choice of prediction targets as part of the training setup can predictably affect the curve's shape. With this updated interpretation, many applications of this depth profile can be better grounded and predicted. As an example, we show how to predict the near-optimal placement of REPA based on the prediction target of the training setup. The code is at https://github.com/Darkbblue/dit-compare.

Chinese Translation

在网络前向调用过程中，扩散骨干网络自然会产生许多中间激活。如果我们根据相同的下游基准协议评估每个激活，我们可以绘制出一个倒U型曲线作为深度轮廓：激活的性能先上升，然后在更深的层次下降。这在扩散模型社区中是常识，激发了许多应用，包括广泛采用的REPA。然而，对这一深度轮廓的常见解读可能是不准确和不完整的：(i) 这一曲线的原因长期以来被视为信号粒度从粗到细的变化，但在DiT骨干网络中几乎没有证据支持这一点。(ii) 社区主要以静态方式研究这一现象，而没有探讨其在不同训练设置下的变化。在本研究中，我们展示了这一轮廓的不同机制图景，其中预测目标（例如，$ ext{ε}$、$u$、$x_0$）处于中心位置。语义上有意义的信号仅仅是预测目标的一个中间副产品，而预测目标才是骨干网络的真正目标。此外，作为训练设置一部分的预测目标的选择可以可预测地影响曲线的形状。通过这种更新的解读，许多深度轮廓的应用可以得到更好的基础和预测。作为一个例子，我们展示了如何根据训练设置的预测目标来预测REPA的近似最佳位置。代码可在 https://github.com/Darkbblue/dit-compare 找到。

View on arXiv Download PDF AI Translation

cs.CV / 33 / 2610.14544

InfraWorld: Motion-Aware World Modeling for Roadside 4D Occupancy Forecasting

InfraWorld：基于运动感知的路边4D占用预测世界建模

Wang, Songkai, Bai, Xiaokai, Yang, Lei, Shen, Hui-liang

Abstract

Roadside 4D occupancy forecasting predicts future semantic occupancy in a fixed infrastructure coordinate system. Roadside scenes are globally structured yet locally evolving, requiring comprehensive scene-level context while precisely modeling sparse localized motion. However, existing forecasting paradigms struggle to tackle both requirements. To address this limitation, we propose InfraWorld, a world model formulating roadside occupancy forecasting as selective state evolution. InfraWorld maintains a dense BEV state for the complete scene and persistent motion queries for moving regions, assigning complementary states to global context and local dynamics. Specifically, we propose Dynamic Query Initiation (DQI) to initialize queries in regions supported by motion in the observed history. During rollout, the dense state is updated through Long--Short Temporal Fusion (LST), which combines short-term adjacent-frame continuity with long-term motion trends to preserve stable context while capturing scene evolution. Motion-Aware Query Injection (MAQI) then refines the queries and injects confidence-weighted query information into corresponding BEV tokens, enabling targeted dynamic updates without perturbing global context. Together, InfraWorld concentrates predictive capacity on evolving traffic regions. Extensive experiments on the InfraOcc and Occ3D--nus datasets demonstrate that InfraWorld achieves state-of-the-art performance for roadside 4D occupancy forecasting while retaining its effectiveness under ego-centric settings. Code will be released.

Chinese Translation

路边4D占用预测在固定基础设施坐标系统中预测未来的语义占用。路边场景在全球上是结构化的，但在局部上是不断演变的，这要求在精确建模稀疏局部运动的同时，提供全面的场景级上下文。然而，现有的预测范式在满足这两个要求方面存在困难。为了解决这一局限性，我们提出了InfraWorld，一个将路边占用预测表述为选择性状态演变的世界模型。InfraWorld为完整场景维护一个密集的鸟瞰图（BEV）状态，并为运动区域保持持久的运动查询，将互补状态分配给全球上下文和局部动态。具体而言，我们提出了动态查询初始化（Dynamic Query Initiation, DQI），以在历史观察中支持运动的区域初始化查询。在推理过程中，通过长短期时间融合（Long-Short Temporal Fusion, LST）更新密集状态，该方法结合了短期相邻帧的连续性和长期运动趋势，以保持稳定的上下文，同时捕捉场景演变。运动感知查询注入（Motion-Aware Query Injection, MAQI）则进一步细化查询，并将置信加权的查询信息注入相应的BEV标记，从而实现针对性的动态更新，而不扰动全球上下文。总的来说，InfraWorld将预测能力集中于不断演变的交通区域。在InfraOcc和Occ3D-nus数据集上的大量实验表明，InfraWorld在路边4D占用预测方面实现了最先进的性能，同时在自我中心设置下保持其有效性。代码将会发布。

View on arXiv Download PDF AI Translation

cs.CV / 34 / 2610.14553

Adversarial Gradient Forcing for Robust Few-Step Autoregressive Video Generation

对抗梯度强制用于鲁棒的少步自回归视频生成

Xia, Chenxi, Xu, Yixuan, Liu, Shan, Wang, Jiapeng, Li, Maohua, Tang, Hanlin, Liu, Kan, Lan, Tao, Lu, Weiyi, Liu, Xinggao

Abstract

Autoregressive video generation models generate videos sequentially, with each new block conditioned on previously generated blocks. By training on the model's own rollouts, Self Forcing and its follow-up methods substantially mitigate exposure bias caused by the mismatch between training- and inference-time history distributions. However, these models still suffer from error accumulation during long-horizon generation. We identify a key underlying limitation: the model remains highly sensitive to errors in its generated history, allowing small deviations to be amplified through subsequent generations. We refer to this underexplored problem as the local history robustness gap. To address this gap, we introduce Adversarial Gradient Forcing (AGF), a framework designed to improve the robustness of few-step autoregressive video models to perturbations and accumulated errors in their own generated histories. AGF uses distribution matching gradients from future block predictions to select adversarial perturbations of self-generated histories. It then jointly trains the generator on the original and perturbed histories to improve the robustness of future generation. Extensive experiments show that AGF achieves the highest VBench Total score on 60-second video generation and the best visual quality on four-minute video generation. Notably, AGF achieves such long-horizon robustness while requiring only 5-second rollouts for training.

Chinese Translation

自回归视频生成模型按顺序生成视频，每个新块都以先前生成的块为条件。通过在模型自身的生成结果上进行训练，自我强制（Self Forcing）及其后续方法显著减轻了由于训练和推理时历史分布不匹配而导致的曝光偏差。然而，这些模型在长时间生成过程中仍然面临错误累积的问题。我们识别出一个关键的潜在限制：模型对其生成历史中的错误高度敏感，导致小的偏差在后续生成中被放大。我们将这一未被充分探讨的问题称为局部历史鲁棒性差距。为了解决这一差距，我们引入了对抗梯度强制（Adversarial Gradient Forcing, AGF），这是一个旨在提高少步自回归视频模型对自身生成历史中的扰动和累积错误鲁棒性的框架。AGF利用来自未来块预测的分布匹配梯度来选择自生成历史的对抗扰动。然后，它在原始历史和扰动历史上共同训练生成器，以提高未来生成的鲁棒性。大量实验表明，AGF在60秒视频生成中达到了最高的VBench总分，并在四分钟视频生成中实现了最佳视觉质量。值得注意的是，AGF在仅需5秒生成结果进行训练的情况下，实现了如此长时间的鲁棒性。

View on arXiv Download PDF AI Translation

cs.CV / 35 / 2610.14568

FrameFold: Response-Guided KV Cache Consolidation for Autoregressive Long Video Generation

FrameFold：自回归长视频生成中的响应引导键值缓存整合

Liu, Muyu, Liu, Mingming, Gu, Hong, Zhang, Yuyao, Lu, Qin

Abstract

In autoregressive video generation, retaining all historical key-value (KV) states increases memory usage and per-step attention costs as videos grow longer. A fixed cache budget bounds these costs but raises the challenge of preserving historical content that supports consistent, high-quality generation. Existing methods control cache size through history selection or consolidation, yet deciding what to retain and how to limit the loss of useful content remains a shared challenge. Our analysis shows that spatial queries depend on different historical content and that higher current attention coverage does not necessarily yield better long-video quality. We introduce FrameFold, a training-free KV cache consolidation method that compares the content retrieved by the same recent historical queries before and after merging to guide both representative construction and merge selection. Exploiting temporal redundancy between adjacent historical entries, FrameFold merges their KV states into shared representatives while retaining complete spatial grids. It fits value-merging coefficients in closed form by matching conditional attention responses, then selects the merge with the lowest remaining response distortion. Under matched KV cache budgets, experiments on Self Forcing and Causal Forcing at 30 and 60 seconds and on LongLive at 120 seconds demonstrate improved consistency and visual quality. Continuous eight-minute Self Forcing rollouts further show less quality degradation than the baselines.

Chinese Translation

在自回归视频生成中，保留所有历史键值（KV）状态会随着视频长度的增加而增加内存使用和每步的注意力成本。固定的缓存预算限制了这些成本，但带来了保持支持一致、高质量生成的历史内容的挑战。现有方法通过历史选择或整合来控制缓存大小，但决定保留什么以及如何限制有用内容的损失仍然是一个共同的挑战。我们的分析表明，空间查询依赖于不同的历史内容，且较高的当前注意力覆盖并不一定能产生更好的长视频质量。我们提出了FrameFold，这是一种无训练的KV缓存整合方法，它比较合并前后相同最近历史查询所检索的内容，以指导代表构建和合并选择。通过利用相邻历史条目之间的时间冗余，FrameFold将它们的KV状态合并为共享代表，同时保留完整的空间网格。它通过匹配条件注意力响应以封闭形式拟合值合并系数，然后选择剩余响应失真最低的合并。在匹配的KV缓存预算下，在30秒和60秒的Self Forcing和Causal Forcing实验以及120秒的LongLive实验中，展示了更好的一致性和视觉质量。连续八分钟的Self Forcing展开进一步表明，其质量下降低于基线。

View on arXiv Download PDF AI Translation

cs.CV / 36 / 2610.14589

Attribute-Centric Vision-Language Alignment for Fine-Grained Gait Understanding

基于属性的视觉-语言对齐用于细粒度步态理解

Hou, Saihui, Wang, Chenye, Lang, Wenpeng, Guo, Qiannan, Lan, Zhengxiang, Fan, Chao, Yu, Shiqi, Huang, Yongzhen

Abstract

Walking sequences reveal human-interpretable attributes related to body structure, posture, and motion. Gait attribute recognition provides a concrete route to fine-grained gait understanding by predicting these attributes. However, existing benchmarks provide limited fine-grained annotations, while classifier-based gait attribute recognition methods typically represent candidate classes through learned output weights rather than directly matching visual evidence with textual class representations. To address the annotation gap, we present RA-GAR, a large-scale, richly annotated multimodal benchmark for gait attribute recognition, containing 1,001 subjects, 242,729 walking sequences, and annotations for 15 static and motion gait attributes. We further propose Gait-VL, an attribute-centric vision-language alignment framework that recognizes gait attributes by matching attribute-specific multi-frame visual evidence with textual class prototypes. On the visual side, attribute-specific learnable queries aggregate RGB features across spatial locations and frames into per-attribute visual embeddings. On the language side, diverse descriptions expanded by a large language model are combined with context about relevant body regions to construct textual class prototypes. Beyond attribute recognition, we evaluate the transferability of the attribute-pretrained visual encoder by using it to replace generic upstream large vision models in identityoriented gait recognition pipelines; attribute pre-training uses no identity classification or metric-learning objectives. Experiments demonstrate state-of-the-art attribute recognition performance on RA-GAR and consistent improvements in downstream identity recognition across multiple pipelines and benchmarks, supporting the effectiveness and transferability of attribute-centric gait representation learning.

Chinese Translation

步态序列揭示了与身体结构、姿势和运动相关的人类可解释属性。步态属性识别通过预测这些属性为细粒度步态理解提供了一条具体途径。然而，现有基准提供的细粒度注释有限，而基于分类器的步态属性识别方法通常通过学习的输出权重来表示候选类别，而不是直接将视觉证据与文本类别表示进行匹配。为了解决注释差距，我们提出了RA-GAR，这是一个大规模、丰富注释的多模态基准，用于步态属性识别，包含1,001个受试者、242,729个步态序列，以及15个静态和运动步态属性的注释。我们进一步提出了Gait-VL，一个基于属性的视觉-语言对齐框架，通过将属性特定的多帧视觉证据与文本类别原型进行匹配来识别步态属性。在视觉方面，属性特定的可学习查询将RGB特征在空间位置和帧之间聚合为每个属性的视觉嵌入。在语言方面，由大型语言模型扩展的多样化描述与相关身体区域的上下文结合，以构建文本类别原型。除了属性识别外，我们通过在身份导向的步态识别管道中使用属性预训练的视觉编码器替换通用的上游大型视觉模型来评估其可迁移性；属性预训练不使用身份分类或度量学习目标。实验表明，在RA-GAR上实现了最先进的属性识别性能，并在多个管道和基准上实现了下游身份识别的一致性提升，支持基于属性的步态表征学习的有效性和可迁移性。

View on arXiv Download PDF AI Translation

cs.CV / 37 / 2610.14598

Bidirectional Adaptive Transformers for Multimodal Anomaly Detection

双向自适应变换器在多模态异常检测中的应用

Jiang, Yuxin, Cao, Yunkang, Shen, Weiming

Abstract

The effective integration of features derived from RGB images and 3D point clouds presents a substantial challenge in the field of multimodal anomaly detection (MAD). Existing MAD methods typically rely on static fusion paradigms built upon frozen pre-trained encoders. While these frameworks facilitate spatial alignment, they lack the mechanism to establish latent semantic correspondence, thereby inducing cross-modal interference that degrades anomaly discriminability. To mitigate these limitations, this study proposes a Bidirectional Adaptive Transformers (BAT) framework, which constructs comprehensive and semantically aligned multimodal representations by leveraging complementary attributes across modalities. The BAT framework introduces a Bidirectional Feature Alignment (BFA) module, which employs a mutual predictive mechanism to bridge the gap between 2D textural patterns and 3D geometric structures. This bidirectional interaction explicitly enhances the textural granularity of point cloud descriptors while augmenting the spatial contextual awareness of RGB features. To ensure that the latent representations possess sufficient informational density for alignment, BAT incorporates a parameter-efficient adaptation architecture. By utilizing lightweight image and point cloud adapters, the framework facilitates specialized fine-tuning of pre-trained models, mitigating the information loss inherent in frozen encoders. Extensive experiments conducted on MVTec 3D-AD and Eyecandies confirm that BAT achieves state-of-the-art anomaly detection performance, significantly outperforming existing MAD methods.

Chinese Translation

有效整合来自RGB图像和3D点云的特征在多模态异常检测（MAD）领域中面临重大挑战。现有的MAD方法通常依赖于基于冻结的预训练编码器构建的静态融合范式。虽然这些框架促进了空间对齐，但缺乏建立潜在语义对应关系的机制，从而导致跨模态干扰，降低了异常的可区分性。为了解决这些局限性，本研究提出了一种双向自适应变换器（Bidirectional Adaptive Transformers, BAT）框架，该框架通过利用跨模态的互补属性构建全面且语义对齐的多模态表示。BAT框架引入了一个双向特征对齐（Bidirectional Feature Alignment, BFA）模块，该模块采用互预测机制来弥合2D纹理模式与3D几何结构之间的差距。这种双向交互显著增强了点云描述符的纹理细粒度，同时增强了RGB特征的空间上下文感知。为了确保潜在表示具有足够的信息密度以进行对齐，BAT结合了一种参数高效的适应架构。通过利用轻量级的图像和点云适配器，该框架促进了对预训练模型的专业微调，减轻了冻结编码器固有的信息损失。在MVTec 3D-AD和Eyecandies上进行的广泛实验确认，BAT在异常检测性能上达到了最先进的水平，显著优于现有的MAD方法。

View on arXiv Download PDF AI Translation

cs.CV / 38 / 2610.14609

EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment

EviScene：超越“它是人工智能吗？”的代理图像事实评估

Wang, ZiAn, Liu, Ruiqi, Zhang, Zhengbo, Han, Yi, Shen, Jialiang, Chen, Ruoxin, Cui, Manni, Zhu, Ximo, Jiang, Jin, Liu, Mingzhe, Wang, Linhao, Wang, Wenhao, Wu, Shu

Abstract

Determining whether an image is AI-generated does not establish whether the situation it depicts is factually supported. We formulate Image Factuality Assessment: given an image and a generic verification instruction, a system returns a supported or refuted verdict with an evidence-grounded explanation. Without a supplied textual claim, the system must progressively refine factual questions during investigation and align external evidence with the specific situation depicted. We introduce EviLens, an expert-reviewed benchmark of 10,174 supported and refuted images from controlled construction and web collection, covering twelve factual error types and evaluating both verdict accuracy and evidence sufficiency. We develop EviScene, a multimodal ReAct agent that interleaves visual inspection, external retrieval, and evidence analysis, using new findings to refine its questions and revisit relevant image details. We train a compact policy with supervised fine-tuning on reviewed teacher trajectories, followed by privileged on-policy self-distillation applied to verified decisions from the policy's own investigations. On the 1,527-image test set, the trained Qwen3.5-9B agent achieves 79.05% balanced accuracy and a Strict Evidence Sufficiency Rate (SESR) of 52.26%, improving over the same-backbone base agent by 6.71 and 9.37 percentage points, respectively. Its SESR is the highest among the evaluated agents.Code, models, and data will be released at https://github.com/wangzian828/EVISCENE and https://huggingface.co/ZiAnwang/EviScene.

Chinese Translation

确定一幅图像是否由人工智能生成并不能证明其所描绘的情况是否有事实支持。我们提出了图像事实评估：给定一幅图像和一个通用的验证指令，系统返回一个支持或反驳的裁决，并提供基于证据的解释。在没有提供文本声明的情况下，系统必须在调查过程中逐步细化事实问题，并将外部证据与所描绘的具体情况对齐。我们引入了EviLens，这是一个经过专家审查的基准，包含10,174幅支持和反驳的图像，来源于控制构建和网络收集，涵盖十二种事实错误类型，并评估裁决的准确性和证据的充分性。我们开发了EviScene，一个多模态ReAct代理，交替进行视觉检查、外部检索和证据分析，利用新发现来细化其问题并重新审视相关图像细节。我们在经过审查的教师轨迹上进行监督微调，训练出一个紧凑的策略，随后对该策略自身调查中验证的决策应用特权在线自蒸馏。在1,527幅图像的测试集上，训练后的Qwen3.5-9B代理实现了79.05%的平衡准确率和52.26%的严格证据充分性率（SESR），分别比同骨干基础代理提高了6.71和9.37个百分点。其SESR在评估的代理中最高。代码、模型和数据将发布在https://github.com/wangzian828/EVISCENE和https://huggingface.co/ZiAnwang/EviScene。

View on arXiv Download PDF AI Translation

cs.CV / 39 / 2610.14624

EarthAlign: Learning Language-Aligned Representations of Earth System Data

EarthAlign：学习与地球系统数据对齐的语言表示

Maani, Fadillah Adamsyah, Saeed, Numan, Ma, Jiaqi, Nguyen, Dat Tien, Wehbe, Youssef, Watson-Parris, Duncan, Khan, Salman

Abstract

Interpreting Earth system conditions is central to understanding environmental risks and informing decisions across weather- and climate-sensitive sectors. Existing multimodal weather systems primarily target task-specific prediction, generation, or question answering, while general-purpose alignment between Earth system data and language remains underexplored due to the limited availability of diverse language supervision grounded in meteorological conditions. To fill this gap, we introduce EarthAlign, a large-scale dataset with broad geographic coverage spanning 2000-2024 that combines 10 expert-curated natural event and disaster data sources with climatology-derived normal-weather samples. Each sample includes temporal and geospatial information and four grounded caption styles aligned with 63-variable ERA5 meteorological data. Building on EarthAlign, we introduce MeteoLIP, a dual-encoder framework for learning aligned meteorology-language representations. MeteoLIP incorporates MetTok, which uses a geolocation- and time-conditioned query to aggregate variable-specific patch embeddings into a compact sequence of meteorology tokens. To comprehensively assess Earth system multimodal representations, we introduce three downstream tasks: cross-modal retrieval, weather-condition classification, and physical variable estimation. Our experiments show that MeteoLIP improves upon the corresponding CLIP and SigLIP baselines across these tasks. Further analyses highlight the importance of meteorological input, temporal and geospatial conditioning, and multi-style supervision for meteorology-language alignment. Thus, EarthAlign and MeteoLIP provide a foundation for learning and evaluating language-aligned representations of Earth system data. Our code, dataset, and weights are publicly available at https://github.com/mbzuai-oryx/EarthAlign.

Chinese Translation

解读地球系统状况对于理解环境风险以及为气候和天气敏感领域的决策提供信息至关重要。现有的多模态天气系统主要针对特定任务的预测、生成或问答，而地球系统数据与语言之间的通用对齐仍未得到充分探索，这主要是由于基于气象条件的多样化语言监督的有限可用性。为填补这一空白，我们引入了EarthAlign，这是一个覆盖广泛的、涵盖2000年至2024年的大规模数据集，结合了10个专家策划的自然事件和灾害数据源以及气候学派生的正常天气样本。每个样本包括时间和地理空间信息，以及与63个变量的ERA5气象数据对齐的四种基础描述风格。在此基础上，我们提出了MeteoLIP，一个用于学习对齐气象-语言表示的双编码器框架。MeteoLIP结合了MetTok，它使用基于地理位置和时间条件的查询，将特定变量的补丁嵌入聚合为紧凑的气象标记序列。为了全面评估地球系统的多模态表示，我们引入了三个下游任务：跨模态检索、天气条件分类和物理变量估计。我们的实验表明，MeteoLIP在这些任务上优于相应的CLIP和SigLIP基线。进一步的分析强调了气象输入、时间和地理空间条件以及多样化监督在气象-语言对齐中的重要性。因此，EarthAlign和MeteoLIP为学习和评估与地球系统数据对齐的语言表示提供了基础。我们的代码、数据集和权重已公开发布在https://github.com/mbzuai-oryx/EarthAlign。

View on arXiv Download PDF AI Translation

cs.CV / 40 / 2610.14630

DuSRFlow: Dual-Lens Super-Resolution via Parallax-Aware Flow Estimation and Complementary Reference Warping

DuSRFlow：通过视差感知流估计和互补参考扭曲的双镜头超分辨率

Zhang, Fangpu, Yue, Huanjing, Cui, Zifan, Tu, Shaoxiong, Yang, Qirui, Yang, Jingyu

Abstract

Dual-lens super-resolution (DuSR) aims to enhance a wide-angle low-resolution (LR) image by exploiting a telephoto reference (Ref) image. A central challenge in DuSR is reliable reference utilization under large parallax, occlusion, and cross-camera imaging discrepancies. Existing methods often employ off-the-shelf monocular optical flow models to provide geometric priors. However, their assumptions of photometric consistency and smooth motion are violated in dual-lens scenarios, leading to unreliable alignment and degraded reference-based reconstruction. To address these issues, we propose DuSRFlow, a dual-lens super-resolution framework that improves reference utilization from both correspondence learning and reference aggregation. First, we introduce DuFlowNet, a dedicated dual-lens flow estimator for robust cross-view alignment. To train DuFlowNet without ground-truth flow annotations, we develop a real--synthetic data fusion strategy that composites synthetic foreground objects onto real dual-lens image pairs, generating controllable large-parallax samples while preserving realistic cross-camera imaging characteristics. Second, we observe that reference reliability is highly spatially uneven in DuSR, where unreliable local reference responses may suppress the utilization of reliable cues. We therefore propose a Complementary Reference Warping strategy that coordinates patch-based warping and kernel-free-guided deformable warping according to local reference reliability, producing a more reliable reference representation before LR--Ref fusion. In addition, we construct DuSR-RealV2, a challenging real-world benchmark with complex depth variations, large parallax, and occlusion. Extensive experiments on multiple DuSR benchmarks demonstrate that DuSRFlow achieves state-of-the-art performance in reconstruction quality, perceptual quality, and cross-dataset generalization.

Chinese Translation

双镜头超分辨率（DuSR）旨在通过利用长焦参考（Ref）图像来增强广角低分辨率（LR）图像。DuSR中的一个核心挑战是在大视差、遮挡和跨相机成像差异下可靠地利用参考。现有方法通常采用现成的单目光流模型来提供几何先验。然而，在双镜头场景中，它们对光度一致性和光滑运动的假设被破坏，导致对齐不可靠和基于参考的重建质量下降。为了解决这些问题，我们提出了DuSRFlow，一个双镜头超分辨率框架，通过对应学习和参考聚合来改善参考的利用。首先，我们引入了DuFlowNet，一个专门的双镜头流估计器，用于稳健的跨视图对齐。为了在没有真实光流标注的情况下训练DuFlowNet，我们开发了一种真实-合成数据融合策略，将合成前景对象合成到真实的双镜头图像对上，生成可控的大视差样本，同时保持真实的跨相机成像特性。其次，我们观察到在DuSR中，参考的可靠性在空间上高度不均匀，不可靠的局部参考响应可能抑制可靠线索的利用。因此，我们提出了一种互补参考扭曲策略，根据局部参考可靠性协调基于块的扭曲和无核引导的可变形扭曲，在LR-Ref融合之前生成更可靠的参考表示。此外，我们构建了DuSR-RealV2，这是一个具有复杂深度变化、大视差和遮挡的挑战性真实世界基准。在多个DuSR基准上的广泛实验表明，DuSRFlow在重建质量、感知质量和跨数据集泛化方面达到了最先进的性能。

View on arXiv Download PDF AI Translation

cs.CV / 41 / 2610.14633

Automatic assembly modeling through attention-based assembly feature mate prediction and assembly feature pattern recognition

基于注意力的装配特征配对预测与装配特征模式识别的自动装配建模

Dai, Ziheng, Pan, Wanbin, Tang, Zhihong, Gao, Shuming

Abstract

Assembly modeling plays an important role in constructing product models, yet it is still mainly performed via interactive operations that are experience-dependent and labor-intensive. Data-driven strategies hold promise for automating this process; however, existing approaches either predict poses that cannot directly generate editable B-Rep assembly constraints or require prescribed part pairs or predefined assembly faces. To reduce the large candidate space of constraint prediction without predefined assembly faces, we propose a data-driven framework based on assembly feature mate prediction. The framework performs automatic assembly modeling from unassembled B-Rep parts, using assembly features to group the geometric entities of local mating interfaces and restrict candidate formation. Given the lack of suitable assembly feature annotations in existing datasets, we construct a large-scale dataset through an automatic annotation pipeline. Using the annotated dataset, we train an attention-based network to predict assembly feature mates, with Engineering Compatibility Fusion (ECF) incorporating explicit engineering compatibility into learned matching scores. To improve the accuracy of automatic assembly modeling, we use assembly feature patterns (AFPs) to guide the matching process. Finally, a post-processing search selects assembly feature mate and AFP mate candidates, whose associated constraints are generated and solved to complete assembly modeling. Experimental results show that the framework achieves assembly constraint accuracies of 74.79% on the six-assembly comparison set and 64.09% on the extended test set of 16 assemblies, which includes the comparison set. With ground-truth face labels supplied as additional input, the corresponding accuracies increase to 81.78% and 80.93%, respectively.

Chinese Translation

装配建模在构建产品模型中扮演着重要角色，但目前仍主要依赖于经验性和劳动密集型的交互操作。数据驱动的策略有望实现这一过程的自动化；然而，现有的方法要么预测的姿态无法直接生成可编辑的 B-Rep 装配约束，要么需要预定义的零件对或预设的装配面。为了在没有预定义装配面的情况下减少约束预测的大候选空间，我们提出了一种基于装配特征配对预测的数据驱动框架。该框架从未装配的 B-Rep 部件中执行自动装配建模，利用装配特征对局部配合接口的几何实体进行分组并限制候选形成。鉴于现有数据集中缺乏合适的装配特征注释，我们通过自动注释管道构建了一个大规模数据集。使用注释数据集，我们训练了一个基于注意力的网络来预测装配特征配对，其中工程兼容性融合（Engineering Compatibility Fusion, ECF）将显式的工程兼容性纳入学习的匹配分数中。为了提高自动装配建模的准确性，我们使用装配特征模式（Assembly Feature Patterns, AFPs）来指导匹配过程。最后，后处理搜索选择装配特征配对和 AFP 配对候选，其相关约束被生成并求解以完成装配建模。实验结果表明，该框架在六个装配比较集上的装配约束准确率达到 74.79%，在包含比较集的 16 个装配的扩展测试集上的准确率为 64.09%。在提供真实面标签作为额外输入的情况下，相应的准确率分别提高到 81.78% 和 80.93%。

View on arXiv Download PDF AI Translation

cs.CV / 42 / 2610.14647

Can Synthetic UAV Benchmarks Be Trusted? Real-Synthetic Validation for UAV Model Selection

合成无人机基准是否可信？无人机模型选择的真实-合成验证

Zhi, Yuyi, Du, Min, Shi, Bill, Yuan, Guodong, Liang, Junheng, Liao, Peifan, Guan, Xin

Abstract

Synthetic UAV imagery can expand pre-deployment test evidence when task-specific Real data are scarce, but visual plausibility or semantic label alignment alone does not establish validity for model selection. We study when Synthetic evaluation preserves the model-selection conclusions obtained from Real UAV data through two complementary cases under fixed evaluation protocols. In a requirement-specific road-debris benchmark, four language-conditioned perception systems obtain lower absolute F1 on Synthetic data, yet the point-estimate ordering and top-ranked model are preserved; bootstrap analysis yields no supported pairwise reversal, and size-matched resampling preserves the top-model decision. In contrast, a closed-set stress test on a separately materialized business corpus, using four VisDrone-compatible detectors over eight empirically supported classes, changes the point-estimate top-ranked detector and reverses three of six pairwise point-estimate preference directions. A complementary audit of five UAV business scenarios further reveals substantial Real-Synthetic mismatch in class composition, object density, and target scale, including systematic under-representation of small targets. These results show that Synthetic UAV benchmark trust is conditional: semantic coverage is insufficient, and validity must be established empirically for the task, operating regime, and model panel of interest.

Chinese Translation

合成无人机图像可以在任务特定的真实数据稀缺时扩展部署前测试证据，但仅凭视觉可信度或语义标签对齐并不能确立模型选择的有效性。我们研究了在固定评估协议下，合成评估何时能够保持从真实无人机数据中获得的模型选择结论。在一个特定需求的道路碎片基准中，四个语言条件感知系统在合成数据上获得了较低的绝对 F1 分数，但点估计排序和排名最高的模型得以保留；自助法分析未发现支持的成对反转，且大小匹配的重采样保持了顶级模型的决策。相比之下，在一个独立构建的商业语料库上进行的封闭集压力测试，使用四个与 VisDrone 兼容的检测器对八个经验支持的类别进行评估，改变了点估计的排名最高检测器，并反转了六个成对点估计偏好方向中的三个。对五个无人机商业场景的补充审计进一步揭示了类组成、物体密度和目标规模方面的实质性真实-合成不匹配，包括小目标的系统性低估。这些结果表明，合成无人机基准的可信度是有条件的：语义覆盖不足，且有效性必须针对特定任务、操作模式和感兴趣的模型面板进行实证建立。

View on arXiv Download PDF AI Translation

cs.CV / 43 / 2610.14651

TAG: Calibrating SAM3 for Universal Image Segmentation without Dense Supervision

TAG：为无密集监督的通用图像分割校准SAM3

Zhu, Hao, Sun, Heyu, Jin, Shuo, Xiao, Jiayu, Dai, Feng

Abstract

The recent Segment Anything Model 3 (SAM3) extends powerful promptable segmentation from spatial cues to concepts, raising the prospect of a single foundation model for universal image segmentation. However, this promise often fails to translate into reliable performance on downstream segmentation tasks. In this work, we empirically identify concept recognition, rather than mask quality, as the primary bottleneck: SAM3 accurately delineates the objects it recognizes yet struggles to robustly determine which concepts are present. Driven by this, we present \textit{\underline{T}ext \underline{A}nchor re-\underline{G}rounding (\TAG{})}, a simple yet effective framework that rectifies the concept recognition of SAM3 on the target domain without dense annotations. Specifically, \TAG{} introduces an \textit{Anchor Residual Pathway} comprising image-conditioned, low-rank adapters to re-ground text anchors in the target distribution and modulate subsequent text-conditioned interactions. Guided solely by image-level category labels, \TAG{} jointly optimizes binary cross-entropy with \textit{Soft Boundary Ranking} objectives to calibrate presence confidence and distinguish present concepts from confusable negatives. Building upon this, we devise \textit{Confidence-Weighted Distillation} to couple rich semantic context of a multimodal large language model with fine-grained spatial priors of SAM3, thereby achieving effective calibration in the absence of manual annotations. Extensive experiments demonstrate that \TAG{} unlocks SAM3's segmentation potential with only 3.1M trainable parameters, achieving impressive performance across diverse scenarios, and notably, even surpassing leading mask-supervised methods on several metrics. \href{https://github.com/MiSsU-HH/TAG_SAM3}{Our code will be publicly available at GitHub \faGithub}.

Chinese Translation

最近的Segment Anything Model 3 (SAM3) 将强大的可提示分割从空间线索扩展到概念，提出了一个通用图像分割的单一基础模型的前景。然而，这一承诺往往未能转化为下游分割任务的可靠性能。在本研究中，我们实证地识别出概念识别，而非掩膜质量，作为主要瓶颈：SAM3能够准确描绘其识别的对象，但在稳健地确定哪些概念存在方面却存在困难。基于此，我们提出了 extit{ extbf{T}ext extbf{A}nchor re- extbf{G}rounding ( extbf{TAG})}，这是一个简单而有效的框架，能够在没有密集注释的情况下纠正SAM3在目标领域的概念识别。具体而言， extbf{TAG}引入了一个 extit{Anchor Residual Pathway}，该路径由图像条件的低秩适配器组成，用于在目标分布中重新定位文本锚点并调节后续的文本条件交互。在仅由图像级类别标签引导下， extbf{TAG}联合优化二元交叉熵和 extit{Soft Boundary Ranking}目标，以校准存在置信度并区分存在的概念与易混淆的负样本。在此基础上，我们设计了 extit{Confidence-Weighted Distillation}，将多模态大型语言模型的丰富语义上下文与SAM3的细粒度空间先验相结合，从而在缺乏手动注释的情况下实现有效校准。大量实验表明， extbf{TAG}以仅3.1M的可训练参数释放了SAM3的分割潜力，在多种场景中取得了令人印象深刻的性能，特别是在多个指标上甚至超过了领先的掩膜监督方法。我们的代码将公开发布在GitHub上。

View on arXiv Download PDF AI Translation

cs.CV / 44 / 2610.14654

Prospective Forcing: Learning to Look Ahead for Autoregressive Long Video Diffusion

前瞻性强制：学习前瞻性自回归长视频扩散

Shen, Yifan, Xia, Yewei, Luo, Gongxu, Zhu, Peiyuan, Sun, Boyang, Chen, Zhenhao, Liu, Guangyi, Chen, Guangyi, Zhang, Kun

Abstract

Autoregressive video generators synthesize long videos through chunk-wise rollouts, yet they are typically trained to predict only the immediate future. This limited prediction horizon makes it challenging to maintain video quality over extended rollouts. To address this problem, we introduce Prospective Forcing, a multi-head prediction framework that injects long-range predictive signals into autoregressive video representations. Built on a rolling-window generator, Prospective Forcing attaches lightweight, offset-specific heads that share the same conditioning information and independently predict chunks at different future temporal offsets. By extending supervision beyond the next chunk, these heads encourage the shared history representation to retain information useful for longer-horizon prediction, supporting temporally consistent generation. At inference time, we accelerate sampling by using accepted future-chunk predictions to skip full-window backbone executions. We equip the head with learnable embeddings to generate multiple candidates for the next chunk, and a learned verifier either selects a candidate or rejects all and falls back to standard autoregressive generation.

Chinese Translation

自回归视频生成器通过分块展开合成长视频，但它们通常仅被训练以预测最近的未来。这种有限的预测视野使得在长时间展开中保持视频质量变得具有挑战性。为了解决这个问题，我们引入了前瞻性强制（Prospective Forcing），一种多头预测框架，它将长距离预测信号注入自回归视频表示中。基于滚动窗口生成器，前瞻性强制附加了轻量级的、特定偏移的头部，这些头部共享相同的条件信息，并独立地预测不同未来时间偏移的块。通过将监督扩展到下一个块之外，这些头部鼓励共享历史表示保留对长时间预测有用的信息，从而支持时间一致的生成。在推理时，我们通过使用接受的未来块预测来加速采样，以跳过完整窗口的主干执行。我们为头部配备了可学习的嵌入，以生成下一个块的多个候选项，而一个学习的验证器则选择一个候选项，或拒绝所有候选项并回退到标准的自回归生成。

View on arXiv Download PDF AI Translation

cs.CV / 45 / 2610.14657

TGuide: Faster Autoregressive Video Diffusion via Historical Trajectory Guidance

TGuide：通过历史轨迹指导加速自回归视频扩散

Chen, Zhuo, Zou, Bin, Wei, Fanyue, Duan, Lixin, Yao, Angela, Li, Wen

Abstract

Can autoregressive video diffusion be accelerated further when only four denoising steps remain? Existing feature caching approaches struggle in this regime, as large feature changes between denoising timesteps undermine both direct reuse and lightweight predictors. We observe that features across video chunks remain strongly correlated at matching denoising timesteps, offering a complementary source of guidance. Building on this insight, we introduce TGuide, a framework that accelerates autoregressive video diffusion through a lightweight Anchor-Transport-Correct (ATC) predictor guided by historical sampling trajectories. Our ATC combines two complementary references: the current chunk's preceding-step features preserve its spatial structure, while the previous chunk's same-timestep features provide information at the target denoising stage. By integrating these references through spatial retrieval and adaptive correction, ATC improves velocity prediction accuracy despite the large gaps between denoising steps. To mitigate error accumulation across both denoising steps and video chunks, we train the predictor on its own autoregressive rollouts, allowing it to adapt to states shaped by its own earlier predictions. An optional confidence-aware scheduler further adjusts the quality-efficiency trade-off. Experiments on Self Forcing, Causal Forcing, and HY-WorldPlay show that TGuide preserves generation quality while achieving denoising speedups of up to $1.89\times$ over the original 4-step generators. On Self Forcing, TGuide generalizes from 81-frame training videos to 333-frame generation, approximately four times the training length, while exceeding the original model's VBench score.

Chinese Translation

当仅剩四个去噪步骤时，自回归视频扩散是否可以进一步加速？现有的特征缓存方法在这一情况下面临挑战，因为去噪时间步之间的特征变化较大，削弱了直接重用和轻量级预测器的效果。我们观察到，在匹配的去噪时间步之间，视频片段的特征保持强相关性，这提供了一种互补的指导来源。基于这一见解，我们提出了TGuide，一个通过历史采样轨迹指导的轻量级Anchor-Transport-Correct (ATC)预测器加速自回归视频扩散的框架。我们的ATC结合了两个互补的参考：当前片段的前一步特征保持其空间结构，而前一个片段的同时间步特征则在目标去噪阶段提供信息。通过空间检索和自适应校正整合这些参考，ATC提高了速度预测的准确性，尽管去噪步骤之间存在较大间隔。为了减轻在去噪步骤和视频片段之间的误差累积，我们在预测器的自回归展开上进行训练，使其能够适应由自身早期预测所形成的状态。一个可选的基于置信度的调度器进一步调整质量与效率的权衡。在Self Forcing、Causal Forcing和HY-WorldPlay上的实验表明，TGuide在保持生成质量的同时，实现了相较于原始4步生成器的去噪速度提升，最高可达1.89倍。在Self Forcing上，TGuide能够从81帧的训练视频推广到333帧的生成，约为训练长度的四倍，同时超越了原始模型的VBench评分。

View on arXiv Download PDF AI Translation

cs.CV / 46 / 2610.14663

TETRIS-4DGS: Efficient Post-Training Compression of 4D Gaussian Splatting

TETRIS-4DGS：高效的4D高斯点云后训练压缩

Wu, Chengzhen, Liu, Kelei, Li, Sicheng, Yu, Lu, Liao, Yiyi

Abstract

4D Gaussian Splatting (4DGS) has emerged as a powerful dynamic scene representation, supporting real-time high quality rendering but suffers from prohibitive storage overhead. Existing methods typically attempt to solve this by treating compression as a per-scene optimization problem, limiting generalizability and applicability where low-latency compression is required. In this work, we leverage existing image and video codecs and propose a generalized post-training compression framework for 4DGS to address these issues. Specifically, we apply importance-based adaptive quantization to Gaussians according to their importance levels, effectively refining bitrate allocation while remaining fully compatible with existing codecs. To further improve compression efficiency, we utilize the statistical properties of 4DGS and propose motion- and importance-aware packing to reorganize attributes into codec-friendly 2D maps. Our approach can achieve over 40x compression rate in seconds while maintaining high quality, facilitating the practical deployment of 4DGS.

Chinese Translation

4D高斯点云（4DGS）作为一种强大的动态场景表示方法，支持实时高质量渲染，但面临着巨大的存储开销。现有方法通常将压缩视为每个场景的优化问题，从而限制了其在需要低延迟压缩的场景中的通用性和适用性。在本研究中，我们利用现有的图像和视频编码器，提出了一种针对4DGS的通用后训练压缩框架，以解决这些问题。具体而言，我们根据高斯的显著性水平对其应用基于重要性的自适应量化，有效地优化比特率分配，同时与现有编码器完全兼容。为了进一步提高压缩效率，我们利用4DGS的统计特性，提出了运动和重要性感知的打包方法，将属性重新组织为适合编码器的二维图。我们的方法能够在几秒钟内实现超过40倍的压缩率，同时保持高质量，从而促进4DGS的实际应用。

View on arXiv Download PDF AI Translation

cs.CV / 47 / 2610.14671

SPARE: Shared Perception and Adaptive Retrieval of Evidence for Streaming Video Understanding

SPARE：用于流媒体视频理解的共享感知与自适应证据检索

Xu, Senbin, Zhang, Zijun, Wu, Xinhua, He, Li, Feng, Jubo, Qiao, Li, Gao, Zhen, Zhang, Yan, Luo, Jixiang, Ji, Rongrong

Abstract

Streaming video understanding requires answering questions from the observed video, with evidence drawn from recent observations or earlier events. Historical context can improve recall but may interfere with judgments based on recent frames. Its usefulness depends on both when it is introduced and which clips are selected, while relevance alone does not guarantee sufficient evidence for an answer. We propose SPARE, a framework that combines adaptive historical retrieval with visual encoder sharing. A lightweight gate predicts whether recent observations provide sufficient evidence to answer the question. When additional evidence is predicted to be necessary, a multimodal embedding model retrieves relevant historical clips to supplement the current context. LoRA instruction tuning trains the understanding model to use recent observations and retrieved history jointly and abstain when the evidence remains insufficient. To reduce redundant visual parameters, the understanding and retrieval branches share a visual encoder while retaining separate language backbones. Four linear adapters align the shared features with the retrieval model through distillation at the visual token and final retrieval embedding levels. With visual-only inputs, SPARE achieves 77.78% on the evaluated Real-time and Backward task groups of OVO-Bench and 73.31% on StreamingBench. Compared with Qwen3-VL-8B-Instruct evaluated at 2 fps, SPARE improves accuracy by 20.68 percentage points on OVO-Bench and 14.15 percentage points on StreamingBench.

Chinese Translation

流媒体视频理解需要回答来自观察视频的问题，证据来自于最近的观察或早期事件。历史背景可以提高回忆能力，但可能会干扰基于最近帧的判断。其有效性取决于引入的时机和选择的剪辑，而相关性本身并不能保证提供足够的证据来回答问题。我们提出了SPARE，一个结合自适应历史检索与视觉编码器共享的框架。一个轻量级的门控机制预测最近的观察是否提供足够的证据来回答问题。当预测需要额外证据时，多模态嵌入模型检索相关的历史剪辑以补充当前的背景。LoRA指令调优训练理解模型共同使用最近的观察和检索的历史，并在证据仍然不足时选择放弃。为了减少冗余的视觉参数，理解和检索分支共享一个视觉编码器，同时保留独立的语言骨干网络。四个线性适配器通过在视觉标记和最终检索嵌入层的蒸馏，将共享特征与检索模型对齐。在仅使用视觉输入的情况下，SPARE在评估的OVO-Bench的实时和回溯任务组上达到了77.78%的准确率，在StreamingBench上达到了73.31%。与在2 fps下评估的Qwen3-VL-8B-Instruct相比，SPARE在OVO-Bench上提高了20.68个百分点，在StreamingBench上提高了14.15个百分点。

View on arXiv Download PDF AI Translation

cs.CV / 48 / 2610.14672

Competition Domains Matter: Selective Head Globalization for Training-Free Sparse Attention in Video Diffusion Transformers

竞争领域的重要性：无训练稀疏注意力在视频扩散变换器中的选择性头部全球化

Liu, Tiantian, Wan, Meng, Liang, Zhiqiang, Wang, Jue, Nie, Ningming, Wang, Yangang

Abstract

Training-free sparse attention is a practical route to efficient video diffusion transformer inference, yet existing methods typically fix the competition domain at each query row for every attention head. Prior work primarily addresses how to estimate token-level importance within this fixed domain; whether the domain itself should adapt to head heterogeneity remains under-specified. We introduce EG-Sparse, a training-free framework that routes low-entropy heads to a grouped cross-head pool while retaining per-row local selection for high-entropy heads, using only scores already materialized by the sparse pipeline. A controlled diagnostic fixes the ranking score and varies only the competition rule: grouped cross-head selection reduces low-entropy-head reconstruction error by 9.8%, whereas forcing high-entropy heads into an all-head global pool increases it by 7.9%. Across four configurations, EG-Sparse achieves higher video quality while reducing attention density and GPU denoising latency compared with strong training-free sparse-attention baselines. Mechanism ablations further show that selective domain adaptation outperforms uniform, random, and fixed domain-assignment policies. These results support competition-domain adaptation as a distinct design axis within centroid-guided training-free sparse attention.

Chinese Translation

无训练稀疏注意力是实现高效视频扩散变换器推理的实际途径，但现有方法通常在每个注意力头的每个查询行中固定竞争领域。之前的研究主要关注如何在这个固定领域内估计令牌级别的重要性；而领域本身是否应该适应头部异质性仍未明确。我们提出了EG-Sparse，一个无训练框架，它将低熵头路由到分组的跨头池，同时对高熵头保留每行的局部选择，仅使用稀疏管道中已生成的分数。一个受控诊断固定了排名分数，仅改变竞争规则：分组跨头选择将低熵头的重建误差降低了9.8%，而将高熵头强制进入全头全球池则增加了7.9%。在四种配置中，EG-Sparse在降低注意力密度和GPU去噪延迟的同时，获得了更高的视频质量，相较于强大的无训练稀疏注意力基线。机制消融进一步表明，选择性领域适应优于均匀、随机和固定领域分配策略。这些结果支持竞争领域适应作为一种独特的设计轴，在以质心为指导的无训练稀疏注意力中。

View on arXiv Download PDF AI Translation

cs.CV / 49 / 2610.14676

CAST-ASD: Causal Attention over Speakers and Time for Online Active Speaker Detection

CAST-ASD：针对在线主动说话者检测的因果注意力模型

Gupta, Pratham, Qi, Haotian, Sandhan, Tushar, Skantze, Gabriel

Abstract

We present CAST-ASD (Causal Attention over Speakers and Time), a causal joint model for active speaker detection (ASD). Recent ASD models improve performance by jointly processing all faces in a scene, but operate offline, with the full clip available before any frame is labelled. Low-latency causal detectors instead process each face independently. CAST-ASD combines joint face reasoning with fully causal inference. It applies scene-level attention across faces within the current frame and causal temporal attention over each face's history, enabling information sharing across speakers with zero lookahead. Its parameter count does not grow with the number of faces. On MSDWild, where most negative face-frames contain a visible competing speaker, CAST-ASD reduces competing-speaker false alarms by 32% relative to the causal per-face detector it extends, with larger gains as scene occupancy increases. It outperforms all causal baselines on both validation splits while using only one-tenth the inter-speaker parameters of a three-block fixed-roster convolution. Code and models are available at https://github.com/prathamg007/CAST-ASD.

Chinese Translation

我们提出了CAST-ASD（因果注意力模型），这是一个用于主动说话者检测（ASD）的因果联合模型。近期的ASD模型通过联合处理场景中的所有面孔来提高性能，但它们是离线操作的，在标记任何帧之前需要完整的视频片段。低延迟的因果检测器则独立处理每个面孔。CAST-ASD结合了联合面孔推理与完全因果推断。它在当前帧内的面孔之间应用场景级注意力，并对每个面孔的历史应用因果时间注意力，从而实现了在零前瞻情况下的说话者之间的信息共享。其参数数量不会随着面孔数量的增加而增长。在MSDWild数据集上，大多数负面面孔帧包含一个可见的竞争说话者，CAST-ASD相较于其扩展的因果每面孔检测器减少了32%的竞争说话者误报，且随着场景占用率的增加，增益更为显著。在两个验证集上，它在所有因果基准模型中表现优异，同时仅使用了三块固定名册卷积的一成跨说话者参数。代码和模型可在https://github.com/prathamg007/CAST-ASD获取。

View on arXiv Download PDF AI Translation

cs.CV / 50 / 2610.14679

HumanPhys: A Human-Centric Physical-World Perception Benchmark for Real-World Visual Reasoning

HumanPhys：一个以人为中心的现实世界物理感知基准，用于真实世界视觉推理

Zhang, Yiming, Zhou, Enzhi, Lin, Junan, Zhou, Dulan, Wang, Weiqiang, Huang, Yuge, Liu, Jian

Abstract

We introduce HumanPhys, a human-centric benchmark for real-world physical perception, and a training corpus emitted by the same pipeline. Each question targets a physically grounded quantity related to people, including depth, inter-person distance, gaze angle, displacement, velocity, heading-step trajectory, blink count, and contact-deformation strength. The rendered probes additionally test illumination inference from body and face cues and camera ego-motion, both with and without a person in the frame. All labels are derived programmatically from sensor calibration, source metadata, or renderer state, rather than written by annotators or produced by estimators on the benchmark frames. HumanPhys systematically probes six dimensions of human-centric physical perception: spatial state, kinematics, blink events, deformation, illumination, and camera pose. To the best of our knowledge, no existing MLLM benchmark-and-training corpus jointly supervises this set of human-centered physical quantities under direct metadata- or renderer-derived labels. The six dimensions appear only partially dissociable across the tested MLLMs. We benchmark a range of open- and closed-source MLLMs on HumanPhys and find that all six capability dimensions remain challenging for current models. We hope HumanPhys will support future research on multimodal systems that perceive and reason about people in the physical world.

Chinese Translation

我们介绍了HumanPhys，这是一个以人为中心的现实世界物理感知基准，以及由同一管道生成的训练语料库。每个问题针对与人相关的物理量，包括深度、人与人之间的距离、视线角度、位移、速度、朝向-步态轨迹、眨眼次数和接触-变形强度。渲染的探针还测试了从身体和面部线索以及相机自我运动中推断照明的能力，无论画面中是否有人。所有标签均通过传感器校准、源元数据或渲染器状态程序生成，而非由注释员手动编写或通过基准帧上的估计器生成。HumanPhys系统性地探测了以人为中心的物理感知的六个维度：空间状态、运动学、眨眼事件、变形、照明和相机姿态。根据我们所知，目前没有现有的MLLM基准和训练语料库能够在直接的元数据或渲染器派生标签下共同监督这一组以人为中心的物理量。这六个维度在测试的MLLM中仅部分可分离。我们在HumanPhys上对一系列开源和闭源的MLLM进行了基准测试，发现当前模型在所有六个能力维度上仍然面临挑战。我们希望HumanPhys能够支持未来关于感知和推理物理世界中人类的多模态系统的研究。

View on arXiv Download PDF AI Translation

cs.CV / 51 / 2610.14685

Demiurge: Forging Navigable 4D Worlds via Autoregressive Video Rerendering

造物主：通过自回归视频重渲染铸造可导航的四维世界

Zhu, Hanxin, Ma, Jianqi, Wang, Cong, Hu, Zhihao, Gao, Junyao, He, Tianyu, Lin, Junyan, Xiao, Jiayin, Shou, Jiateng, Liang, Sen, Zhang, Weidong, Chen, Zhibo, Zhang, Jun

Abstract

Recent advances in video diffusion enable novel-view video synthesis from a single input video, supporting world modeling, embodied intelligence, and robotic simulation. However, existing methods process camera trajectories independently, lacking the cross-trajectory consistency needed for coherent, navigable 4D worlds. We introduce Demiurge, an autoregressive video rerendering pipeline built on a dynamic 4D point-cloud memory cache. It promotes cross-trajectory consistency through four components: (1) an appearance video diffusion model that synthesizes photorealistic novel-view videos conditioned on the 4D memory; (2) a memory-consistent video depth completion model that aligns known-region depths with cached geometry and completes missing regions, mitigating scale ambiguity and discontinuities during memory updates; (3) memory-aware dual-stream warping that provides high-fidelity visual guidance and occlusion cues for novel-view synthesis; and (4) geometry-aware video refinement that enhances perceptual quality. Extensive experiments demonstrate state-of-the-art performance and three key capabilities: coherent spatiotemporal lattice construction over dense multi-view trajectories; large-parallax video rerendering under challenging camera motions, including 180-degree rotations with no input-view overlap; and geometry-faithful outputs that can be directly optimized into high-quality 3D Gaussian Splatting representations with minimal artifacts. Project page: https://hanxinzhu-lab.github.io/demiurge_project_page/.

Chinese Translation

最近在视频扩散领域的进展使得从单个输入视频合成新视角视频成为可能，这支持了世界建模、具身智能和机器人仿真。然而，现有方法独立处理相机轨迹，缺乏构建连贯、可导航的四维世界所需的跨轨迹一致性。我们提出了Demiurge，一个基于动态四维点云内存缓存的自回归视频重渲染管道。它通过四个组件促进跨轨迹一致性：(1) 一个外观视频扩散模型，基于四维内存合成光真实感的新视角视频；(2) 一个内存一致的视频深度补全模型，将已知区域的深度与缓存几何体对齐，并补全缺失区域，减轻内存更新过程中的尺度模糊和不连续性；(3) 具有内存感知的双流变形技术，为新视角合成提供高保真视觉指导和遮挡线索；(4) 几何感知的视频精炼，提升感知质量。大量实验表明，我们的方法在性能上达到了最先进水平，并展现了三个关键能力：在密集多视角轨迹上构建连贯的时空格子；在具有挑战性的相机运动下进行大视差视频重渲染，包括180度旋转且没有输入视图重叠；以及生成几何忠实的输出，这些输出可以直接优化为高质量的三维高斯溅射表示，且几乎没有伪影。项目页面：https://hanxinzhu-lab.github.io/demiurge_project_page/

View on arXiv Download PDF AI Translation

cs.CV / 52 / 2610.14698

PURE: Refining Feed-forward 3D Reconstruction with Unlabeled Post-training

PURE：通过无标签后训练精炼前馈3D重建

Zhu, Yaolong, Li, Peng, Lu, Jiahao, Zhao, Chengfeng, Li, Mengfei, Wang, Chao, Chen, Wenyue, Gu, Zekai, Ren, Chengwei, Wang, Ronggang, Liu, Yuan

Abstract

Feed-forward 3D reconstruction models have achieved remarkable progress through large-scale training and strong learned priors. However, they remain fragile when applied to dynamic in-the-wild videos, where non-rigid motion, occlusions, and illumination changes are prevalent. A key limitation lies in their reliance on expensive depth and camera pose annotations, which hinders scaling to diverse real-world dynamic data and restricts generalization. To address this issue, we propose \textbf{PURE}, an unlabeled post-training refinement framework that improves depth and camera estimation without requiring ground-truth supervision. Instead of relying on annotations, PURE elegantly distills robust structural knowledge from state-of-the-art 2D priors into 4D feed-forward architectures. PURE leverages relative depth constraints, affine-invariant geometry distillation, and pose refinement based on feature correspondences to stabilize geometric predictions under challenging dynamic conditions. Extensive experiments on both synthetic and real-world dynamic benchmarks demonstrate that PURE delivers consistent improvements across multiple state-of-the-art architectures. Furthermore, our data-efficient paradigm effectively mitigates domain shifts, successfully unlocking the architectural limits of these base models in challenging, in-the-wild dynamic scenes.

Chinese Translation

前馈3D重建模型通过大规模训练和强大的学习先验取得了显著进展。然而，当应用于动态野外视频时，它们仍然脆弱，因为非刚性运动、遮挡和光照变化普遍存在。一个关键的限制在于它们依赖于昂贵的深度和相机姿态标注，这阻碍了对多样化真实世界动态数据的扩展，并限制了泛化能力。为了解决这个问题，我们提出了 extbf{PURE}，一种无标签后训练精炼框架，能够在不需要真实标签监督的情况下改善深度和相机估计。PURE不依赖于标注，而是优雅地从最先进的2D先验中提炼出稳健的结构知识，融入4D前馈架构中。PURE利用相对深度约束、仿射不变几何提炼以及基于特征对应的姿态精炼，以稳定在挑战性动态条件下的几何预测。在合成和真实世界动态基准上的大量实验表明，PURE在多种最先进架构中提供了一致的改进。此外，我们的数据高效范式有效缓解了领域转移，成功解锁了这些基础模型在挑战性野外动态场景中的架构极限。

View on arXiv Download PDF AI Translation

cs.CV / 53 / 2610.14713

PixARG: Pixel-Asynchronous Retrieval-Augmented Diffusion for Fast and Realistic Image Generation

PixARG：用于快速和真实图像生成的像素异步检索增强扩散

Xin, Zewei, Li, Qinya, Niu, Chaoyue, Wu, Fan

Abstract

While diffusion models achieve impressive image generation capabilities, they suffer from high computational costs, implausible artifacts, and limited adaptability to dynamic or private data due to static, centralized training. To address these, we propose PixARG, a novel retrieval-augmented generative framework that integrates dynamic real-world image retrieval into the diffusion pipeline to simultaneously enhance efficiency and visual realism. Specifically, PixARG constructs a pixel-level affinity map between the user's text prompt and the retrieved image to guide a heterogeneous noise initialization. By modulating noise intensity across spatial regions, this mechanism preserves semantically relevant structural priors while perturbing less relevant areas, thereby significantly accelerating convergence. To accommodate this spatially non-uniform noise and ensure synchronized generation, we introduce a pixel-wise asynchronous denoising scheduler and a spatiotemporal denoiser conditioned on the affinity map. Furthermore, we derive an affinity-adaptive training objective that elegantly balances deterministic restoration with stochastic regeneration across pixels. Experiments on LAION and COYO datasets across offline, online, and user-specific settings show PixARG enhances realism and semantic alignment while drastically reducing diffusion steps.

Chinese Translation

尽管扩散模型在图像生成方面取得了令人瞩目的能力，但由于静态的集中训练，它们面临着高计算成本、不合理的伪影以及对动态或私有数据的适应性有限等问题。为了解决这些问题，我们提出了PixARG，这是一种新颖的检索增强生成框架，它将动态真实世界图像检索集成到扩散流程中，以同时提高效率和视觉真实感。具体而言，PixARG在用户的文本提示与检索到的图像之间构建了一个像素级的亲和力图，以指导异质噪声初始化。通过在空间区域中调节噪声强度，该机制在扰动不太相关的区域的同时，保留了语义相关的结构先验，从而显著加速收敛。为了适应这种空间非均匀噪声并确保同步生成，我们引入了一个像素级异步去噪调度器和一个基于亲和力图的时空去噪器。此外，我们推导出一个亲和力自适应训练目标，优雅地平衡了像素间的确定性恢复与随机再生。在LAION和COYO数据集上的离线、在线和用户特定设置的实验表明，PixARG在显著减少扩散步骤的同时，提高了真实感和语义对齐。

View on arXiv Download PDF AI Translation

cs.CV / 54 / 2610.14726

LightVAE: Towards Compact and Efficient Video Autoencoders

LightVAE：朝着紧凑高效的视频自编码器迈进

Gu, Shiqiao, Yong, Yang, Zhong, Fengjun, Gong, Ruihao, Lv, Chengtao, Fan, Xiangyu, Yang, Lei, Liu, Xianglong

Abstract

Latent diffusion models enable high-quality video generation, but video VAE encoding and decoding remain computationally expensive. Existing methods use stage-specific lightweight operators, yet either constrain spatial and temporal computation jointly or remove temporal computation altogether, without distinguishing their compression sensitivities in pretrained Conv3D. To address these limitations, we present LightVAE, a compression framework that exploits redundancy in temporal modeling and network depth. LightVAE exactly reparameterizes eligible causal Conv3D layers into an aggregate spatial response and a difference-driven temporal correction. Based on their distinct compression sensitivities, temporal low-rank compression (TLR) keeps the aggregate spatial kernel dense and applies layer-specific low-rank approximation only to the temporal correction. An algebraically equivalent folded implementation converts the arithmetic savings into practical acceleration. LightVAE further uses width-preserving block pruning to reduce depth and probe-guided feature alignment, guided by feedback from a frozen teacher suffix, to recover quality after compression. The framework also extends to video encoders. Across three video VAE backbones and two GPU platforms, LightVAE consistently accelerates decoding. For the Wan VAE family, it achieves up to $2.24\times$ and $1.11\times$ the decoding throughput of state-of-the-art efficient VAE decoders on NVIDIA H100 and RTX~5090, respectively, with higher DAVIS PSNR and lower LPIPS. On MiniMax-H3, it achieves up to $2.97\times$ the original decoding throughput while retaining 99.51\% of the original model's overall VBench-2.0 score. Our code is available at https://github.com/ModelTC/LightVAE.

Chinese Translation

潜在扩散模型能够实现高质量的视频生成，但视频变分自编码器（VAE）的编码和解码仍然计算开销较大。现有方法使用特定阶段的轻量级操作符，但要么共同限制空间和时间计算，要么完全去除时间计算，而未能区分其在预训练的Conv3D中的压缩敏感性。为了解决这些限制，我们提出了LightVAE，一个利用时间建模和网络深度冗余的压缩框架。LightVAE准确地将合适的因果Conv3D层重新参数化为聚合空间响应和基于差异的时间修正。基于它们不同的压缩敏感性，时间低秩压缩（TLR）保持聚合空间核的稠密性，并仅对时间修正应用层特定的低秩近似。一个代数上等效的折叠实现将算术节省转化为实际加速。LightVAE进一步使用宽度保持的块剪枝来减少深度，并通过来自冻结教师后缀的反馈引导特征对齐，以在压缩后恢复质量。该框架还扩展到视频编码器。在三个视频VAE骨干网络和两个GPU平台上，LightVAE始终加速解码。对于Wan VAE家族，它在NVIDIA H100和RTX~5090上分别实现了高达$2.24 imes$和$1.11 imes$的解码吞吐量，相较于最先进的高效VAE解码器，具有更高的DAVIS PSNR和更低的LPIPS。在MiniMax-H3上，它实现了高达$2.97 imes$的原始解码吞吐量，同时保留了原始模型整体VBench-2.0得分的99.51\%。我们的代码可在https://github.com/ModelTC/LightVAE获取。

View on arXiv Download PDF AI Translation

cs.CV / 55 / 2610.14734

Align-VAE: Aligning Attention and Geometry in VecSet VAEs for 3D Reconstruction

Align-VAE：在 VecSet VAE 中对齐注意力与几何以进行 3D 重建

Qu, Huaizhi, Yang, Mingxin, Han, Xintong, Zhan, Fangneng, Guo, Chunchao

Abstract

Set-based variational autoencoders (VecSet VAEs) represent 3D shapes as compact sets of latent tokens. Although each token originates from a query at a sampled surface location, whether these latents truly align with geometry remains unclear. We investigate this alignment and identify four issues: (i) attention-sink-like tokens that attract queries from across the shape; (ii) spatially periodic repetitions in decoder attention responses; (iii) mismatched normal information in encoder and decoder query features; and (iv) missing supervision of the intrinsic relation between signed distance fields and surface normals. To address these issues, we introduce Align-VAE with four corresponding modifications: (i) a shared bank of learnable decoder tokens provides a workspace for global computation; (ii) distance-biased cross-attention guides retrieval toward nearby predicted spatial anchors; (iii) normal-free encoder queries align the two stages' access to normal information while retaining normals in the encoder keys and values; and (iv) a screened Poisson-inspired regularizer supervises surface values and normal-direction derivatives. These changes align latent retrieval and field prediction with geometry without enlarging the per-shape code. On the ABO dataset, Align-VAE reduces Chamfer distance by 26.6% compared to the standard VecSet VAE.

Chinese Translation

基于集合的变分自编码器（VecSet VAE）将 3D 形状表示为潜在标记的紧凑集合。尽管每个标记源自于在采样表面位置的查询，但这些潜在标记是否真正与几何对齐仍不清楚。我们研究了这种对齐，并识别出四个问题：（i）类似注意力汇的标记吸引来自整个形状的查询；（ii）解码器注意力响应中的空间周期性重复；（iii）编码器和解码器查询特征中的法线信息不匹配；以及（iv）缺乏对有符号距离场与表面法线之间内在关系的监督。为了解决这些问题，我们引入了 Align-VAE，并进行了四项相应的修改：（i）共享的可学习解码器标记库为全局计算提供了工作空间；（ii）距离偏置的交叉注意力引导检索靠近预测的空间锚；（iii）无法线的编码器查询对齐了两个阶段对法线信息的访问，同时在编码器的键和值中保留法线；（iv）受限的类似泊松的正则化器监督表面值和法线方向导数。这些变化在不增加每个形状编码的情况下，将潜在检索和场预测与几何对齐。在 ABO 数据集上，Align-VAE 相较于标准的 VecSet VAE 将 Chamfer 距离减少了 26.6%。

View on arXiv Download PDF AI Translation

cs.CV / 56 / 2610.14743

Japanese Handwriting Generation Based on Human Preference Feedback Using Direct Preference Optimization

基于人类偏好反馈的日本手写生成框架：直接偏好优化方法

Kuwata, Wakana, Mibayashi, Ryota, Tani, Masanori, Ohshima, Hiroaki

Abstract

We propose a Japanese handwriting generation framework that integrates human preference feedback using Direct Preference Optimization (DPO) to align generative models with human cognitive and stylistic preferences. While font generation technologies leveraging Generative Adversarial Networks and diffusion models have advanced significantly, applying these approaches to handwriting generation presents unique challenges. Font generation is a deterministic process yielding consistent styles. Handwriting generation, on the other hand, is inherently non-deterministic. In handwriting, glyph shapes never match exactly even when the same writer writes the same character. We define this shape difference as handwriting variability. Although this variability is essential for conveying the natural quality of human writing, it complicates model training and substantially increases data collection costs to capture diverse distribution patterns. To address these limitations, we incorporate human perceptual feedback into the training loop of handwriting generation. While collecting extensive handwritten data is costly, evaluating pairwise preferences between two generated samples remains very easy for humans. Applying DPO, an efficient RL-free preference optimization method, to our generation model enables high-quality handwriting generation without relying on large-scale dataset collection. Experimental results across four distinct Japanese handwriting styles demonstrate that DPO-based training enhances overall structural accuracy. However, in highly complex styles with significant variability, such as brush pen writing, the model fails to capture style-specific features. These findings highlight a critical challenge in extending DPO to complex handwriting styles.

Chinese Translation

我们提出了一种日本手写生成框架，该框架通过直接偏好优化（Direct Preference Optimization, DPO）整合人类偏好反馈，以使生成模型与人类的认知和风格偏好相一致。尽管利用生成对抗网络（Generative Adversarial Networks）和扩散模型（diffusion models）的字体生成技术已取得显著进展，但将这些方法应用于手写生成却面临独特的挑战。字体生成是一个确定性过程，产生一致的风格。而手写生成则本质上是非确定性的。在手写中，即使同一作者书写同一字符，字形也从未完全匹配。我们将这种形状差异定义为手写变异性。尽管这种变异性对于传达人类书写的自然特质至关重要，但它使模型训练变得复杂，并显著增加了捕捉多样分布模式的数据收集成本。为了解决这些限制，我们将人类感知反馈纳入手写生成的训练循环中。虽然收集大量手写数据成本高昂，但评估两个生成样本之间的成对偏好对于人类来说非常简单。将DPO这一高效的无强化学习偏好优化方法应用于我们的生成模型，使得高质量的手写生成得以实现，而无需依赖大规模数据集的收集。针对四种不同的日本手写风格的实验结果表明，基于DPO的训练提高了整体结构准确性。然而，在具有显著变异性的高度复杂风格中，如毛笔书写，模型未能捕捉到特定风格的特征。这些发现突显了将DPO扩展到复杂手写风格中的一个关键挑战。

View on arXiv Download PDF AI Translation

cs.CV / 57 / 2610.14802

YOUDrive: Driving Style as a Composable Direction for End-to-End Autonomous Driving

YOUDrive：将驾驶风格作为端到端自主驾驶的可组合方向

Zheng, Guantian, Yu, Lidong, Gao, Tianyu, Chen, Jiaxing, Li, Jiashu, Lv, Chen

Abstract

End-to-end driving policies can achieve strong task performance while offering limited control over their driving style. Driving style describes how a vehicle accelerates, brakes, and maintains distance. Policies trained without explicit preference conditioning lack a direct mechanism for adjusting these behaviors at inference. We propose YOUDrive, a vision--language--action planner for continuous and composable driving-style control. Each teacher's behavior is distilled into a task vector, a learned parameter update that induces a style direction in the planner's representation. These directions can be scaled and combined at inference without retraining the backbone. Conditioned on these directions, a flow-matching decoder models a multimodal trajectory distribution and generates kinematically feasible motion alternatives. We introduce the Style Alignment Score (SAS) to describe human-referenced kinematic behavior after safety and task gating. On NAVSIM navtest, a single coefficient monotonically controls assertiveness across six teacher-derived personas, while PDMS remains above 0.90 across the tested individual and blended configurations, reaching 0.954. Further analyses reveal systematic policy-specific kinematics after controlling for scene context, whereas the evaluated discrete style labels provide limited additional explanatory power. Code, checkpoints, and the metric will be released.

Chinese Translation

端到端驾驶策略能够实现强大的任务性能，但对驾驶风格的控制有限。驾驶风格描述了车辆加速、刹车和保持距离的方式。未经过显式偏好条件训练的策略在推理时缺乏直接调整这些行为的机制。我们提出了YOUDrive，一种用于连续和可组合驾驶风格控制的视觉-语言-动作规划器。每个教师的行为被提炼为任务向量，这是一种学习的参数更新，能够在规划器的表示中引入风格方向。这些方向可以在推理时进行缩放和组合，而无需重新训练主干网络。在这些方向的条件下，流匹配解码器建模了多模态轨迹分布，并生成运动学上可行的运动替代方案。我们引入了风格对齐评分（Style Alignment Score, SAS）来描述在安全和任务门控后的人类参考运动行为。在NAVSIM navtest上，单一系数单调控制六个教师派生角色的果断性，同时PDMS在测试的个体和混合配置中保持在0.90以上，达到0.954。进一步分析显示，在控制场景上下文后，政策特定的运动学表现出系统性，而评估的离散风格标签提供的额外解释能力有限。代码、检查点和度量将会发布。

View on arXiv Download PDF AI Translation

cs.CV / 58 / 2610.14832

[Re] Boosting the Visual Interpretability of CLIP via Adversarial Fine-Tuning

[重] 通过对抗微调提升 CLIP 的视觉可解释性

Nuzhdin, Anton, Gruzitski, Andrei, Egyed, Balázs, Raudvee, Aleksandr

Abstract

This paper presents a reproducibility study of ``Boosting the Visual Interpretability of CLIP via Adversarial Fine-Tuning'', published at ICLR 2025, which proposes an unsupervised adversarial fine-tuning (AFT) method with norm regularisation to enhance the visual interpretability of CLIP's image encoder. We attempt to reproduce the key claims regarding improved saliency map quality, increased concept alignment, transferability to out-of-distribution datasets, and the trade-off with zero-shot accuracy. Beyond reproduction, we propose a saliency-guided regularisation extension that introduces an Energy Pointing Game loss, directly supervising the spatial alignment of Simple Gradient saliency maps with target objects. We evaluate our extension across a range of saliency-loss weights and show that explicit saliency supervision \rev{substantially} improves localisation metrics at a modest cost: the reduction in zero-shot accuracy and adversarial robustness is small at the lowest saliency-loss weight ($w_s=0.2$) and moderate at the strongest setting ($w_s=0.6$). We restructure the codebase for configurable, systematic experimentation.

Chinese Translation

本文呈现了对 ``通过对抗微调提升 CLIP 的视觉可解释性'' 的可重复性研究，该论文发表于 ICLR 2025，提出了一种带有范数正则化的无监督对抗微调 (AFT) 方法，以增强 CLIP 图像编码器的视觉可解释性。我们尝试重现关于提高显著性图质量、增加概念对齐、向分布外数据集的可迁移性以及与零-shot 准确性之间权衡的关键论点。除了重现，我们还提出了一种显著性引导的正则化扩展，引入了一种能量指向游戏损失，直接监督简单梯度显著性图与目标对象的空间对齐。我们在一系列显著性损失权重下评估我们的扩展，结果表明，显式的显著性监督在适度成本下显著改善了定位指标：在最低显著性损失权重 ($w_s=0.2$) 下，零-shot 准确性和对抗鲁棒性的下降较小，而在最强设置 ($w_s=0.6$) 下则为适度下降。我们重构了代码库，以便进行可配置的系统实验。

View on arXiv Download PDF AI Translation

cs.CV / 59 / 2610.14846

Towards Privacy-Preserving Thermal Human Perception from Dataset to Deployment

面向隐私保护的热人类感知：从数据集到部署

Sun, Kailai, Cui, Yue, Wang, Xinwei, Gao, Zijie, Guo, Baoshen, Hong, Zhiqing, Qaisar, Irfan, Yang, Xu, Li, Yueting, Fan, Dian, Zhao, Qianchuan

Abstract

Indoor human perception is the foundation for smart homes and healthcare. However, RGB cameras raise privacy concerns in private indoor environments, while low-resolution thermal cameras provide limited spatial and semantic information. Large-scale indoor thermal datasets and real-world deployments remain limited. Here, we introduce Indoor RGB-T, a large-scale paired RGB--thermal dataset for indoor human perception. The dataset covers 20 indoor rooms and contains 90,576 80x62 thermal frames, 87,016 temporally matched RGB--thermal pairs, and 115,557 annotated human bounding boxes. We establish benchmarks for robust thermal occupant-state detection with 0.75 mAP and occupancy estimation with an accuracy of 0.847 in unseen rooms. We further evaluate RGB-to-thermal and thermal-to-RGB translation, achieving a temperature MAE of 0.654 C and an R2 of 0.923. Finally, we deploy the thermal perception model on a commercial sleep-monitoring system in the real world. Field testing with 100 participants and deployment of approximately 1,000 devices further demonstrate its practical feasibility. We identify practical challenges that motivate future research directions, including environmental thermal interference, sensing range and multi-person occlusion. The code, trained models and partial dataset are available at https://github.com/kailaisun/RGBT and https://huggingface.co/skl24/RGBT.

Chinese Translation

室内人类感知是智能家居和医疗保健的基础。然而，RGB摄像头在私密室内环境中引发隐私问题，而低分辨率热成像摄像头提供的空间和语义信息有限。大规模室内热成像数据集和实际部署仍然有限。在此，我们介绍了Indoor RGB-T，这是一个用于室内人类感知的大规模配对RGB-热成像数据集。该数据集涵盖20个室内房间，包含90,576帧80x62的热成像图像、87,016对时间匹配的RGB-热成像配对以及115,557个标注的人类边界框。我们建立了鲁棒的热占用状态检测基准，mAP为0.75，并在未见房间中的占用估计准确率为0.847。我们进一步评估了RGB到热成像和热成像到RGB的转换，获得了温度平均绝对误差为0.654°C，R²为0.923。最后，我们在现实世界中将热感知模型部署在商业睡眠监测系统上。对100名参与者的现场测试和约1,000个设备的部署进一步证明了其实际可行性。我们识别出了一些实际挑战，这些挑战激励了未来的研究方向，包括环境热干扰、感知范围和多人遮挡。代码、训练模型和部分数据集可在https://github.com/kailaisun/RGBT和https://huggingface.co/skl24/RGBT获取。

View on arXiv Download PDF AI Translation

cs.CV / 60 / 2610.14852

Getting Space Right: Agentic Rewards for Image Generation

正确获取空间：图像生成的自主奖励

Tan, Xiaofeng, Lin, Jiaxin, Yuan, Haobo, Shen, Tiancheng, Su, Zihan, Chen, Rui, Wang, Hongsong, Yang, Ming-Hsuan

Abstract

Reinforcement learning (RL) is promising for spatial text-to-image generation, but existing work largely focuses on policy optimization, assuming accurate rewards and effective online rollouts. We observe in-domain gains with out-of-domain degradation, exposing two limitations: unreliable spatial rewards in open-world settings and rollouts lacking basic-to-complex progression. We propose a curriculum-guided RL framework. First, we introduce SpatialCritic, an agentic reward model equipped with a spatial perception harness that combines reusable verification guidance with specialized vision tools. We then construct spatial preference datasets through targeted counterfactual perturbations and perform agent-harness alignment to improve tool selection and evidence integration efficiently. Second, curriculum learning organizes shared foundational capabilities and their compositional dependencies in a capability graph for mastery-guided rollouts, periodic review, and targeted remediation. On SpaRW-Eval, SpatialCritic achieves 95.1% verification accuracy, outperforming 320B GLM-5.3-Flash with a 40x smaller 8B backbone and raising average baseline scores from 78.4 to 83.9 on SD3.5-Medium and 83.2 to 86.5 on BAGEL. Project page: https://xiaofeng-tan.github.io/projects/spatialcritic/

Chinese Translation

强化学习（RL）在空间文本到图像生成方面展现出良好的前景，但现有研究主要集中在策略优化上，假设奖励准确且在线回放有效。我们观察到在领域内的增益与领域外的退化，暴露出两个局限性：在开放世界环境中空间奖励的不可靠性以及缺乏从基础到复杂的回放进展。我们提出了一种课程引导的强化学习框架。首先，我们引入了SpatialCritic，一个配备空间感知工具的自主奖励模型，该工具结合了可重用的验证指导和专业的视觉工具。然后，我们通过针对性的反事实扰动构建空间偏好数据集，并进行代理-工具对齐，以高效改善工具选择和证据整合。其次，课程学习在能力图中组织共享的基础能力及其组合依赖关系，以便进行掌握引导的回放、定期复习和针对性补救。在SpaRW-Eval上，SpatialCritic实现了95.1%的验证准确率，超越了320B GLM-5.3-Flash，后者的基础模型规模为40倍更小的8B，并将SD3.5-Medium的平均基线分数从78.4提高到83.9，以及将BAGEL的分数从83.2提高到86.5。项目页面：https://xiaofeng-tan.github.io/projects/spatialcritic/

View on arXiv Download PDF AI Translation

cs.CV / 61 / 2610.14856

CAST: Real-Time Motion Capture for Any Skeleton Topology

CAST：适用于任意骨架拓扑的实时动作捕捉

Pang, Pu, Qing, Zhongfei, Chen, Ling-Hao, Li, Songnan, Huang, Hua, Hu, Bo

Abstract

This paper studies the problem of recovering motion from monocular video, even when the target skeleton topology differs from the subject observed in the video. Although previous methods have made efforts to resolve this problem, they remain vulnerable to joint-rotation errors that are coupled across the skeleton and can propagate along its kinematic hierarchy. To mitigate these errors, we present CAST, a topology-aware motion capture framework that combines learned local rotations with geometric correction guided by predicted global bone directions. Our observability analysis characterizes which rotational components can be constrained by outgoing bone directions. Guided by this analysis, GRACE uses reliability-gated alignment to correct composed orientations in parent-to-child order, leaving unconstrained components to the learned rotation predictor. The motion predictor and correction module are trained jointly with motion-derived supervision, using a frozen image encoder. Experiments under the TopoCap evaluation protocol show that the proposed framework improves motion recovery across different target topologies, including unseen structures. Reference-based cross-skeleton evaluation on a curated Mixamo benchmark and a user preference study further assess transfer to different target structures. A causal CAST-B variant achieves over 51 FPS on a single NVIDIA A100 GPU, with timing covering image encoding and motion prediction. Code repository: https://github.com/fankewen/CAST.

Chinese Translation

本文研究了从单目视频中恢复动作的问题，即使目标骨架拓扑与视频中观察到的主体不同。尽管之前的方法在解决这一问题上做出了努力，但它们仍然容易受到关节旋转误差的影响，这些误差在骨架中是耦合的，并且可以沿其运动学层次结构传播。为了减轻这些误差，我们提出了CAST，一个拓扑感知的动作捕捉框架，它结合了学习到的局部旋转与由预测的全局骨骼方向引导的几何校正。我们的可观测性分析描述了哪些旋转分量可以通过外发骨骼方向进行约束。在此分析的指导下，GRACE使用可靠性门控对齐以父子顺序校正组合方向，将不受约束的分量留给学习的旋转预测器。动作预测器和校正模块通过运动导向的监督共同训练，使用一个冻结的图像编码器。在TopoCap评估协议下的实验表明，所提出的框架在不同目标拓扑下改善了动作恢复，包括未见结构。基于参考的跨骨架评估在一个精心挑选的Mixamo基准和用户偏好研究中进一步评估了向不同目标结构的转移。一个因果CAST-B变体在单个NVIDIA A100 GPU上实现了超过51帧每秒的速度，时间覆盖了图像编码和动作预测。代码库： https://github.com/fankewen/CAST。

View on arXiv Download PDF AI Translation

cs.CV / 62 / 2610.14866

ScalLoc: Continuous Oblique Thermal UAV Geo-Localization with a 2D Satellite Map

ScalLoc：基于2D卫星地图的连续斜向热成像无人机地理定位

Zhang, Xiaoran, Liu, Yu, Liang, Jinyu, Xiao, Huaxin

Abstract

Cross-modal Thermal Geo-localization (TG) provides an illumination-robust positioning solution for Unmanned Aerial Vehicles (UAVs) in Global Navigation Satellite System (GNSS)-denied environments. However, existing thermal geo-localization methods mainly focus on near-nadir observations or depend on map-side elevation priors, limiting their applicability to continuous oblique thermal localization with only geo-referenced 2D satellite maps. To address this problem, we propose ScalLoc, a scale-aligned local-to-map localization framework for continuous oblique thermal UAV geo-localization. ScalLoc follows a reconstruction--refinement--anchoring--propagation design: it reconstructs local window geometry from thermal sequences, refines camera poses using track-based pose optimization, adaptively anchors each window to the satellite map, and propagates validated states to recover frame-level global trajectories. To support systematic evaluation, we construct Thermal-Oblique, a real thermal UAV sequence dataset containing 216 urban and rural day--night trajectories and 28,112 thermal frames with frame-level annotations. Extensive experiments demonstrate that ScalLoc reduces the mean localization error from 112.69 m to 26.97 m compared with the VGGT-only configuration, corresponding to a 76.07% reduction, and achieves the lowest frame-weighted mean position error among the compared methods. These results demonstrate the feasibility of continuous oblique thermal geo-localization using only lightweight 2D satellite maps, without requiring digital surface models or preconstructed 3D maps. Code and dataset are available at https://github.com/FloralHercules/ScalLoc.

Chinese Translation

跨模态热成像地理定位（TG）为在全球导航卫星系统（GNSS）无法使用的环境中提供了一种抗光照干扰的无人机（UAV）定位解决方案。然而，现有的热成像地理定位方法主要集中在近天顶观测或依赖地图侧的高程先验，限制了它们在仅使用地理参考的2D卫星地图进行连续斜向热成像定位的适用性。为了解决这一问题，我们提出了ScalLoc，一个用于连续斜向热成像无人机地理定位的尺度对齐局部到地图定位框架。ScalLoc遵循重建-精炼-锚定-传播的设计：它从热成像序列中重建局部窗口几何，使用基于轨迹的姿态优化精炼相机姿态，自适应地将每个窗口锚定到卫星地图，并传播验证的状态以恢复帧级全局轨迹。为了支持系统评估，我们构建了Thermal-Oblique，一个包含216条城市和乡村昼夜轨迹及28,112帧带有帧级注释的真实热成像无人机序列数据集。大量实验表明，与仅使用VGGT配置相比，ScalLoc将平均定位误差从112.69米降低到26.97米，减少幅度达到76.07%，并在比较方法中实现了最低的帧加权平均位置误差。这些结果证明了仅使用轻量级2D卫星地图进行连续斜向热成像地理定位的可行性，而无需数字表面模型或预构建的3D地图。代码和数据集可在https://github.com/FloralHercules/ScalLoc获取。

View on arXiv Download PDF AI Translation

cs.CV / 63 / 2610.14871

MORE: Learning Motion-Oriented Registers for Driving World Models and Planning

MORE：学习面向运动的寄存器以驱动世界模型和规划

Gu, Yingdong, Wang, Haoyu, Yan, Shaocheng, Zhao, Zhenjun, Han, Hao, Shi, Pengcheng, Li, Xianfei, Tan, Qinghua, Liao, Wenlong, He, Tao, Li, Jiayuan, Peng, Pai

Abstract

By predicting future states, latent world models provide rich self-supervision for end-to-end driving, but often inherit dense features from pretrained visual encoders as their world-state interface. This creates a mismatch: visual pretraining produces information-rich, spatially dense features, but their adaptation for motion-conditioned world modeling and planning is deferred to downstream training. The dense grid also makes future prediction more computationally expensive as visual resolution increases. We therefore argue that video pretraining should directly learn state representations suited to downstream world modeling and planning, rather than relying on downstream training to adapt inherited visual features. We propose ***MORE***, a V-JEPA framework that learns **M**otion-**O**riented **RE**gisters during video pretraining as a shared state interface for driving world models and planning. MORE injects temporally aligned ego motion into a small set of registers and uses them as the sole context for masked patch prediction, channeling dense patch-level supervision through a compact state bottleneck. A complementary blockwise register-masking objective further shapes predictive structure in the register space. During downstream training, the same register space defines the future states predicted by the world model and provides shared scene memory for trajectory generation and scoring. On NAVSIM-v2, MORE outperforms the state of the art by 1.6 in EPDMS, while achieving superior efficiency with only 1/10 the inference latency of DriveLaW. These results support learning the state interface for world modeling and planning directly during video pretraining.

Chinese Translation

通过预测未来状态，潜在的世界模型为端到端驾驶提供了丰富的自我监督，但通常将预训练视觉编码器的密集特征作为其世界状态接口。这造成了不匹配：视觉预训练产生了信息丰富、空间密集的特征，但其在运动条件下的世界建模和规划的适应性则推迟到下游训练。因此，我们认为视频预训练应直接学习适合下游世界建模和规划的状态表示，而不是依赖下游训练来适应继承的视觉特征。我们提出了***MORE***，一个V-JEPA框架，在视频预训练期间学习**M**otion-**O**riented **RE**gisters，作为驱动世界模型和规划的共享状态接口。MORE将时间对齐的自我运动注入一小组寄存器，并将其作为掩蔽补丁预测的唯一上下文，通过紧凑的状态瓶颈引导密集的补丁级监督。一个互补的块状寄存器掩蔽目标进一步塑造了寄存器空间中的预测结构。在下游训练期间，相同的寄存器空间定义了世界模型预测的未来状态，并为轨迹生成和评分提供共享场景记忆。在NAVSIM-v2上，MORE在EPDMS上超越了现有技术1.6的表现，同时以仅为DriveLaW的1/10的推理延迟实现了更高的效率。这些结果支持在视频预训练期间直接学习用于世界建模和规划的状态接口。

View on arXiv Download PDF AI Translation

cs.CV / 64 / 2610.14940

Engram Neuron Inspired Adaptive Memorization for Lifelong Learning

基于记忆痕迹神经元的自适应记忆机制用于终身学习

Jiang, Suiyang, Wang, Bomin, Zhou, Hangqi, Feng, Jianfeng, Zhuang, Xiahai

Abstract

Lifelong learning enables artificial systems to adapt to changing environments. To achieve this, an effective artificial learner should dynamically reorganize past and new experiences, maintaining stability while enabling flexible knowledge integration. However, existing efforts primarily mitigate catastrophic forgetting, lacking the ability to adaptively update memory akin to biological systems. Here, by integrating the full memory life cycle of engram neurons into artificial learners, we propose a framework that constructs a structured memory from gradient representations, which regulates parameter updates and guides memory management. Through comprehensive evaluations on four benchmarks, our approach shows promising lifelong learning capability, maintaining stability over long task sequences and supporting flexible adaptation. Beyond the performance gains, our approach offers a computational perspective on robust and flexible lifelong learning, echoing mechanisms observed in biological memory systems.

Chinese Translation

终身学习使人工系统能够适应不断变化的环境。为了实现这一目标，一个有效的人工学习者应动态重组过去和新的经验，保持稳定性，同时实现灵活的知识整合。然而，现有的努力主要是缓解灾难性遗忘，缺乏类似生物系统的自适应更新记忆的能力。在此，我们通过将记忆痕迹神经元的完整记忆生命周期整合到人工学习者中，提出了一个框架，该框架从梯度表示中构建结构化记忆，调节参数更新并指导记忆管理。通过在四个基准测试上的全面评估，我们的方法展示了有希望的终身学习能力，能够在长任务序列中保持稳定性并支持灵活适应。除了性能提升，我们的方法还提供了关于稳健和灵活的终身学习的计算视角，呼应了生物记忆系统中观察到的机制。

View on arXiv Download PDF AI Translation

cs.CV / 65 / 2610.14944

Text-Guided Medical Image Segmentation via Residual Aggregation and Dense Alignment Representations

通过残差聚合和密集对齐表示的文本引导医学图像分割

Gui, Jie, Tu, Hang, Dou, Yunfeng, Sha, Wen, Du, Xiuquan, Li, Shuo

Abstract

Specifying segmentation targets with natural language provides flexible semantic conditioning for medical image segmentation. However, existing methods still struggle to reliably ground high-level textual semantics into accurate pixel-level spatial prompts, especially under low-contrast lesions, ambiguous boundaries, and visually similar anatomical contexts. In this paper, we propose RADAR (Residual Aggregation and Dense Alignment Representations), a parameter-efficient framework for text-guided medical image segmentation. RADAR explicitly decouples and synergizes the medical semantic anchoring capability of BiomedCLIP with the prompt-driven mask decoding ability of MedSAM2, while enabling low-cost task adaptation via LoRA. Specifically, Guided Residual Depth Aggregation adaptively aggregates cross-layer features under semantic guidance to alleviate delayed textual injection and deep semantic dilution. Structured Dense Representation Alignment leverages a frozen BiomedCLIP to construct crop-level target, boundary, and context anchors, improving the local discriminability of dense features and the reliability of automatic prompts. Multi-modal Prompt Generation further converts cross-modal semantic responses into composite text, point, and mask prompts for MedSAM2 decoding. Experiments on three public medical image segmentation datasets show that RADAR achieves competitive performance with only 0.74M trainable parameters, demonstrating its effectiveness in parameter-efficient adaptation, semantic grounding, and dense representation enhancement. The source code is available at https://github.com/JGuiidev/RADAR.

Chinese Translation

使用自然语言指定分割目标为医学图像分割提供了灵活的语义条件。然而，现有方法在将高层次文本语义可靠地转化为准确的像素级空间提示方面仍然面临挑战，尤其是在低对比度病变、模糊边界和视觉上相似的解剖背景下。本文提出了RADAR（Residual Aggregation and Dense Alignment Representations），这是一个参数高效的文本引导医学图像分割框架。RADAR明确地解耦并协同BiomedCLIP的医学语义锚定能力与MedSAM2的提示驱动掩膜解码能力，同时通过LoRA实现低成本的任务适应。具体而言，Guided Residual Depth Aggregation在语义指导下自适应地聚合跨层特征，以减轻文本注入延迟和深层语义稀释。Structured Dense Representation Alignment利用冻结的BiomedCLIP构建作物级目标、边界和上下文锚点，提高了密集特征的局部可辨别性和自动提示的可靠性。多模态提示生成进一步将跨模态语义响应转换为复合文本、点和掩膜提示，以供MedSAM2解码。在三个公共医学图像分割数据集上的实验表明，RADAR在仅有0.74M可训练参数的情况下实现了竞争性能，证明了其在参数高效适应、语义锚定和密集表示增强方面的有效性。源代码可在 https://github.com/JGuiidev/RADAR 获取。

View on arXiv Download PDF AI Translation

cs.CV / 66 / 2610.14945

DSM-Free UAV Visual Localization via Latent-Depth Variable Projection and Structure-Adaptive Coupling

基于潜在深度变量投影和结构自适应耦合的无DSM无人机视觉定位

Ye, Yibin, Teng, Xichao, Chen, Shuo, Song, Xiaokai, Sun, Yuli, Yu, Qifeng, Li, Zhang

Abstract

Visual localization enables unmanned aerial vehicles (UAVs) to maintain absolute positioning in GNSS-denied environments. A common approach matches UAV images with geo-tagged orthographic maps and estimates camera pose using Perspective-n-Point (PnP) with elevations from Digital Surface Models (DSMs). However, accurate and up-to-date DSMs are often unavailable in practice. Orthographic-perspective geometry (OPG) enables DSM-free localization in scenes with varying elevations, yet existing methods typically estimate pose indirectly through intermediate matrix estimation and decomposition. To address this limitation, we propose Latent-Depth Variable Projection (LDVP), which treats the unknown scales along perspective viewing rays as latent depths. By analytically eliminating these depths and solving for the horizontal camera center in closed form given the rotation, LDVP reduces nonlinear estimation to a rotation-only problem, enabling direct recovery of metric horizontal position without assuming a planar scene. To improve localization robustness across different scene structures, we further introduce Structure-Adaptive Coupling (SAC), which uses relative scene structure inferred from monocular depth estimation to select LDVP when the matched points are not dominated by a single plane, and Planar-PnP otherwise. Our local observability analysis reveals how scene elevation variation affects the geometric constraints of OPG, providing a basis for this adaptive strategy. Extensive experiments show that LDVP consistently outperforms existing OPG-based DSM-free methods, while SAC further improves robustness across diverse scenes and substantially narrows the gap to DSM-based localization. Code will be released at https://github.com/UAV-AVL/LDVP-SAC.

Chinese Translation

视觉定位使无人机（UAV）能够在无GNSS环境中保持绝对定位。常见的方法是将无人机图像与地理标记的正射地图进行匹配，并使用带有数字表面模型（DSM）高程的透视n点（PnP）估计相机姿态。然而，准确且最新的DSM在实际应用中往往不可用。正射透视几何（OPG）使得在具有不同高程的场景中实现无DSM定位成为可能，但现有方法通常通过中间矩阵估计和分解间接估计姿态。为了解决这一限制，我们提出了潜在深度变量投影（LDVP），该方法将透视视线上的未知尺度视为潜在深度。通过解析消除这些深度，并在给定旋转的情况下以封闭形式求解水平相机中心，LDVP将非线性估计简化为仅旋转的问题，从而实现直接恢复度量水平位置，而无需假设平面场景。为了提高在不同场景结构下的定位鲁棒性，我们进一步引入了结构自适应耦合（SAC），该方法利用从单目深度估计推断的相对场景结构，在匹配点不被单一平面主导时选择LDVP，否则选择平面PnP。我们的局部可观测性分析揭示了场景高程变化如何影响OPG的几何约束，为这一自适应策略提供了基础。大量实验表明，LDVP始终优于现有的基于OPG的无DSM方法，而SAC进一步提高了在多样场景中的鲁棒性，并显著缩小了与基于DSM定位的差距。代码将发布在 https://github.com/UAV-AVL/LDVP-SAC。

View on arXiv Download PDF AI Translation

cs.CV / 67 / 2610.14957

CatPortrait: Taming Video Generative Models for Portrait Animation via In-context Conditioning

CatPortrait：通过上下文条件驯化视频生成模型以实现肖像动画

Ni, Ziqi, Fang, Zhou, Wang, Shijian, Li, Rui, Zhou, Yi

Abstract

Recent video generative models have shown strong capability in synthesizing high-quality and temporally coherent videos. Yet, adapting them for controllable portrait animation remains nontrivial, as existing methods typically rely on architectural changes or extra conditioning modules, increasing complexity and reducing compatibility with pretrained models. We present CatPortrait, a simple and effective framework that enables controllable portrait animation via 3D face guidance as in-context conditioning. CatPortrait preserves the original video generation backbone and injects structured 3D facial guidance into the input space through spatial concatenation, allowing pretrained image-to-video models to animate portraits without sacrificing their pretrained priors. CatPortrait supports both video-driven and audio-driven portrait animation within a unified framework. Extensive experiments demonstrate that CatPortrait achieves competitive or superior results across both settings.

Chinese Translation

近期的视频生成模型在合成高质量和时间一致性视频方面表现出强大的能力。然而，将其适应于可控的肖像动画仍然不是一件简单的事情，因为现有方法通常依赖于架构变化或额外的条件模块，这增加了复杂性并降低了与预训练模型的兼容性。我们提出了CatPortrait，这是一个简单而有效的框架，通过3D面部引导作为上下文条件，实现可控的肖像动画。CatPortrait保留了原始的视频生成主干，并通过空间拼接将结构化的3D面部引导注入输入空间，使得预训练的图像到视频模型能够在不牺牲其预训练先验的情况下动画化肖像。CatPortrait在统一框架内支持视频驱动和音频驱动的肖像动画。大量实验表明，CatPortrait在这两种设置下均实现了具有竞争力或更优的结果。

View on arXiv Download PDF AI Translation

cs.CV / 68 / 2610.14963

AD-GS: Anatomical Density-guided 3D Gaussian Splatting for Sparse-View Dental CBCT Reconstruction

AD-GS：基于解剖密度引导的稀疏视图牙科CBCT重建的三维高斯喷溅

Li, Zonglin, Xie, Kexian, Li, Feiran, Zhou, Chu, Huang, Yixing, Chen, Ruiyang, Ma, Zhanyu, Guo, Heng

Abstract

Sparse-view Cone-Beam Computed Tomography (CBCT) reconstruction is essential for reducing radiation dose and scan effort, but remains underconstrained due to limited projection information. Recent 3D Gaussian Splatting (3DGS)-based CBCT reconstruction methods rely on gradient-driven densification, which becomes unstable under sparse acquisition and often leads to over-smoothed structures. We propose AD-GS, an Anatomical Density-guided 3D Gaussian Splatting framework that formulates Gaussian densification as an attenuation-aware Markov Chain Monte Carlo (MCMC) sampling process. By interpreting Gaussians as samples from an underlying anatomical attenuation distribution, AD-GS adaptively concentrates representation capacity in high-attenuation regions, enabling sharper structural boundaries under sparse projections. To further constrain reconstruction, we incorporate a semantic regularization based on zero-shot features from Segment Anything in Medical Images, enforcing multi-view consistency through semantic priors. Experiments on the ToothFairy3 dental CBCT dataset demonstrate that AD-GS achieves superior structural fidelity and boundary preservation compared with existing 3DGS-based approaches under sparse-view settings within the dental and maxillofacial setting. Code is available at https://github.com/yulinjiaxue/AD-GS-BMVC2026.

Chinese Translation

稀疏视图锥形束计算机断层扫描（CBCT）重建对于降低辐射剂量和扫描工作量至关重要，但由于投影信息有限，仍然存在约束不足的问题。近期基于三维高斯喷溅（3DGS）的CBCT重建方法依赖于梯度驱动的密度化，这在稀疏采集下变得不稳定，常常导致结构过度平滑。我们提出了AD-GS，一个基于解剖密度引导的三维高斯喷溅框架，将高斯密度化公式化为一个考虑衰减的马尔可夫链蒙特卡洛（MCMC）采样过程。通过将高斯视为来自潜在解剖衰减分布的样本，AD-GS自适应地将表示能力集中在高衰减区域，从而在稀疏投影下实现更清晰的结构边界。为了进一步约束重建，我们结合了基于“Segment Anything in Medical Images”的零样本特征的语义正则化，通过语义先验强制多视图一致性。在ToothFairy3牙科CBCT数据集上的实验表明，AD-GS在稀疏视图设置下，相较于现有的基于3DGS的方法，能够实现更优的结构保真度和边界保留。代码可在 https://github.com/yulinjiaxue/AD-GS-BMVC2026 获得。

View on arXiv Download PDF AI Translation

cs.CV / 69 / 2610.14976

Metric-Driven Reinforcement Learning for Object Hallucination Reduction in LVLM-Based Image Captioning

基于度量驱动的强化学习在LVLM图像描述中的物体幻觉减少

Nguyen, Thanh, Yoon, Hee Suk, Luu, Tung Minh, Yoon, Eunseop, Yoo, Chang D.

Abstract

Large Vision-Language Models (LVLMs) have excelled in joint visual and language understanding, particularly in image captioning. However, LVLMs remain prone to object hallucination, describing non-existent entities in the given image. Previous remedies, including dataset-enhanced supervised training and reinforcement learning from human feedback, show promise but rely heavily on costly human involvement, limiting scalability. To address this, we propose a reinforcement learning framework to mitigate object hallucination in LVLM-based image captioning, driven entirely by natural language processing (NLP) metrics. While optimizing reinforcement learning solely on NLP metrics offers scalability, it also introduces pitfalls: reward hacking, sparse feedback, and costly training. Our framework sidesteps these by introducing an Object Consistency Score (OCS) capturing both correctness and coverage of objects, incorporating KL-regularization for denser token-level guidance, and designing a lightweight PPO variant that fine-tunes adapters atop a frozen LVLM backbone, achieving efficiency and stability. Extensive experiments on the InstructBLIP baseline demonstrate the effectiveness and generalizability of our framework, achieving a substantial hallucination reduction of up to 41\% on the COCO dataset and 29\% on the Visual Genome. Furthermore, the framework can enhance caption quality by incorporating complementary evaluation metrics such as BERTScore and METEOR. These results reveal a broader insight in multimodal learning: LVLMs can effectively self-correct hallucinations using metric feedback, motivating deeper exploration into scalable reinforcement frameworks that significantly reduce human effort.

Chinese Translation

大型视觉语言模型（LVLMs）在视觉与语言的联合理解方面表现出色，特别是在图像描述任务中。然而，LVLMs仍然容易出现物体幻觉，即描述图像中不存在的实体。以往的解决方案，包括增强数据集的监督训练和基于人类反馈的强化学习，虽然显示出潜力，但过于依赖昂贵的人力参与，限制了其可扩展性。为了解决这一问题，我们提出了一种强化学习框架，以完全基于自然语言处理（NLP）度量来减轻LVLM图像描述中的物体幻觉。尽管仅基于NLP度量优化强化学习提供了可扩展性，但也引入了奖励黑客、稀疏反馈和高成本训练等陷阱。我们的框架通过引入物体一致性评分（Object Consistency Score, OCS）来规避这些问题，该评分同时捕捉物体的正确性和覆盖率，结合KL正则化以提供更密集的标记级指导，并设计了一种轻量级的PPO变体，在冻结的LVLM主干上微调适配器，以实现效率和稳定性。在InstructBLIP基准上的大量实验表明，我们的框架有效且具有良好的泛化能力，在COCO数据集上实现了高达41 ext{%}的显著幻觉减少，在Visual Genome上为29 ext{%}。此外，该框架通过引入互补的评估指标，如BERTScore和METEOR，能够提升描述质量。这些结果揭示了多模态学习中的更广泛见解：LVLMs可以有效地利用度量反馈自我纠正幻觉，激励我们深入探索可扩展的强化学习框架，以显著减少人力投入。

View on arXiv Download PDF AI Translation

cs.CV / 70 / 2610.15005

Retraining-Free Depth Pruning of Vision Transformers: A Cross Domain Study of Layer Redundancy

无重训练的视觉变换器深度剪枝：层冗余的跨领域研究

Viespoli, Alessandro, Nanni, Loris

Abstract

Transformer architectures have achieved strong results in natural language processing and computer vision, but their deployment is limited by high computational, memory and storage costs, motivating research into model compression techniques, such as pruning, while preserving performance. Prior work has shown substantial redundancy in large language models (LLMs), and redundancy in vision transformers has also been investigated, but it remains less systematically analyzed than in LLMs. In this work, we extend similarity-based redundancy analysis to vision models by measuring cosine similarity between layer inputs and outputs to identify transformations with minimal impact. We study redundancy patterns across several transformer architectures and find that deep vision transformers such as DINOv2 and SwinV2, like language models, exhibit significant redundancy. We then evaluate how this redundancy can be exploited to improve efficiency, analyzing the trade-off between inference speed and performance degradation under different pruning strategies. Our results show that moderate pruning can yield substantial acceleration with limited impact on accuracy, and that 20% of the DINOv2 blocks can be removed without any retraining, for a 1.24x throughput gain and a 20% lower energy per image at a relative accuracy loss below 1% on ImageNet-1K. We further propose a model selection pipeline that automatically identifies the best pruned configuration for a given dataset, reducing FLOPs by 13% to 27% on average across architectures with an average accuracy change between -0.3 and +1.0 points.

Chinese Translation

变换器架构在自然语言处理和计算机视觉领域取得了显著成果，但其部署受到高计算、内存和存储成本的限制，这促使研究模型压缩技术，如剪枝，同时保持性能。先前的研究表明，大型语言模型（LLMs）中存在显著的冗余，而视觉变换器中的冗余也得到了研究，但系统分析程度仍低于LLMs。在本研究中，我们通过测量层输入和输出之间的余弦相似度，将基于相似性的冗余分析扩展到视觉模型，以识别对性能影响最小的变换。我们研究了几种变换器架构中的冗余模式，发现深度视觉变换器如DINOv2和SwinV2与语言模型一样，表现出显著的冗余。然后，我们评估了如何利用这种冗余来提高效率，分析了不同剪枝策略下推理速度与性能下降之间的权衡。我们的结果表明，适度剪枝可以在对准确性影响有限的情况下实现显著加速，并且可以在不进行任何重训练的情况下移除20%的DINOv2块，从而实现1.24倍的吞吐量提升，并在ImageNet-1K上以低于1%的相对准确性损失实现每张图像能耗降低20%。我们进一步提出了一种模型选择管道，自动识别给定数据集的最佳剪枝配置，在不同架构上平均减少13%至27%的FLOPs，平均准确性变化在-0.3至+1.0点之间。

View on arXiv Download PDF AI Translation

cs.CV / 71 / 2610.15012

CowCare: A Unified Visual Framework for Viewpoint-Robust Cattle Tracking and Spatiotemporal Heat-Stress Analysis

CowCare：一个统一的视觉框架，用于视角鲁棒的牛只追踪和时空热应激分析

Liu, Xin, Li, Weiran, Zhu, Minghao, Han, Mina, Li, Zhenbo

Abstract

Reliable individual monitoring is essential for dairy cattle welfare and early heat stress detection in Precision Livestock Farming (PLF). However, translating raw multi-camera videos into automated diagnostics remains challenging due to disconnected tracking and recognition pipelines and complex physical environments. Dense mutual occlusions and cross-camera viewpoint variations fragment identities, while environmental noise, including camera vibrations, and limited agricultural data obscure subtle heat stress cues such as rapid breathing and flank undulations. To bridge low-level visual features and high-level behavioral diagnostics, we propose CowCare, a unified framework combining viewpoint-robust tracking with spatiotemporal heat stress recognition. Its tracking module, CowTrack, maintains cross-camera identity consistency through a View-Decoupled Aggregation Network (VDA-Net). Instead of relying on static global representations, VDA-Net aggregates localized textures under heavy occlusion and uses a Viewpoint Decoupling Transformer (VDT) to suppress transient viewpoint variations. CowTrack further incorporates a causal online self-supervision mechanism for continuous adaptation to dynamic farm scenes. The resulting tracklets are subsequently processed by TSDF-Net, a fine-grained video understanding network. Its Temporal Stabilization and Aggregation Module (TSAM) suppresses high-frequency jitter, while a Diffusion Semantic Prior (DSP) uses frozen pre-trained diffusion semantics to enhance subtle behavioral signals and mitigate overfitting. Extensive evaluations on two real-world farm benchmarks demonstrate the framework's effectiveness: CowTrack achieves 83.48% HOTA and 87.72% IDF1 for global multi-object tracking on MVCow, and TSDF-Net achieves 87.72% Top-1 accuracy on HSCow. These results establish CowCare as a robust end-to-end framework for automated livestock health monitoring.

Chinese Translation

可靠的个体监测对于乳牛福利和精准畜牧业（Precision Livestock Farming, PLF）中早期热应激检测至关重要。然而，由于追踪和识别管道的断开以及复杂的物理环境，将原始多摄像头视频转化为自动诊断仍然具有挑战性。密集的相互遮挡和跨摄像头视角变化使身份碎片化，而环境噪声（包括摄像头震动）和有限的农业数据则模糊了快速呼吸和腹部波动等微妙的热应激线索。为了解决低级视觉特征与高级行为诊断之间的桥梁问题，我们提出了CowCare，一个将视角鲁棒追踪与时空热应激识别相结合的统一框架。其追踪模块CowTrack通过视角解耦聚合网络（View-Decoupled Aggregation Network, VDA-Net）保持跨摄像头的身份一致性。VDA-Net不依赖于静态全局表示，而是在重度遮挡下聚合局部纹理，并使用视角解耦变换器（Viewpoint Decoupling Transformer, VDT）来抑制瞬态视角变化。CowTrack进一步结合了一种因果在线自监督机制，以持续适应动态农场场景。随后，生成的追踪片段由TSDF-Net处理，这是一种细粒度视频理解网络。其时间稳定与聚合模块（Temporal Stabilization and Aggregation Module, TSAM）抑制高频抖动，而扩散语义先验（Diffusion Semantic Prior, DSP）利用冻结的预训练扩散语义来增强微妙的行为信号并减轻过拟合。在两个真实农场基准上的广泛评估证明了该框架的有效性：CowTrack在MVCow上实现了83.48%的HOTA和87.72%的IDF1，TSDF-Net在HSCow上实现了87.72%的Top-1准确率。这些结果确立了CowCare作为一个强大的端到端框架，用于自动化的牲畜健康监测。

View on arXiv Download PDF AI Translation

cs.CV / 72 / 2610.15030

Appearance Compensation in Volume Rendering

体积渲染中的外观补偿

Zhou, Jianghong

Abstract

Can a volume that reproduces every external RGB view still have an incorrect density? We study this question in emission-absorption rendering with an independently parameterized, physically bounded effective source. Appearance compensates a density change by replacing the radiance that it removes from the rest of each ray. We characterize the surviving density information with a shared spatial appearance field. Our first result is an optimal density-only gain after finite-degree spherical-harmonic source refitting. A coupled tensor gauge and physical localized packets identify the sharp limiting nonlinear inverse coefficient as its reciprocal. Our second result realizes positive gain at every fixed finite degree for a residual set of isotropic spatial sources at one fixed density. At the same references, growing angular capacity hides a fixed nonzero density error while retaining physicality and any prescribed finite set of smoothness bounds. Local identifiability and the absence of a capacity-uniform recovery modulus therefore coexist. Our third result constructs an exact physical SH2 ambiguity: central density doubles and relative density $L^2$ error exceeds $14.52\%$, yet every external RGB ray is unchanged. We connect these continuum results to shared finite representations, continuous pixel measurements and density-sensitive acquisition. Complete-class certificates separate SH0, SH1 and SH2 fits at one learned density. Finite-camera results expose the distinct role of spatial capacity, and transmission measurements separate otherwise identical RGB fields. Controlled training tests and captured-smoke profiles supply mechanism evidence with explicit observation scopes. The results show why rendering accuracy alone cannot certify density: recovery also depends on appearance capacity, spatial resolution and measurement design.

Chinese Translation

一个能够重现每个外部 RGB 视图的体积是否仍然可能具有不正确的密度？我们在发射-吸收渲染中研究了这个问题，采用独立参数化、物理有界的有效光源。外观通过替换从每条光线的其余部分移除的辐射来补偿密度变化。我们用共享的空间外观场来表征存活的密度信息。我们的第一个结果是在有限度的球谐光源重新拟合后获得的最佳仅密度增益。耦合的张量量规和物理局部化包识别出尖锐的限制非线性逆系数作为其倒数。我们的第二个结果在一个固定密度下，为一组各向同性空间光源的每个固定有限度实现了正增益。在相同的参考下，增长的角度容量隐藏了固定的非零密度误差，同时保持物理性和任何规定的有限平滑度界限。因此，局部可识别性和缺乏容量均匀恢复模量共存。我们的第三个结果构建了一个精确的物理 SH2 模糊性：中心密度翻倍，相对密度 $L^2$ 误差超过 $14.52\%$，但每条外部 RGB 光线保持不变。我们将这些连续结果与共享有限表示、连续像素测量和密度敏感获取联系起来。完整类证书在一个学习的密度下分离 SH0、SH1 和 SH2 拟合。有限相机结果揭示了空间容量的独特作用，而传输测量则分离了其他相同的 RGB 场。受控训练测试和捕获烟雾轮廓提供了机制证据，具有明确的观察范围。结果表明，仅仅依靠渲染精度无法证明密度的正确性：恢复还依赖于外观容量、空间分辨率和测量设计。

View on arXiv Download PDF AI Translation

cs.CV / 73 / 2610.15035

What Really Works for Universal Multimodal Embeddings? A Controlled Study of Contrastive Training Strategies

通用多模态嵌入的有效策略是什么？对比训练策略的控制研究

Li, Duo, Pan, Haojun, Yang, Zuhao, Wang, Xintong, Huo, Chengfu, Lu, Shijian, Ding, Liang

Abstract

Contrastive learning has become the standard paradigm for multimodal representation learning, aligning semantically matched pairs while separating mismatched ones in a shared embedding space. Although various contrastive training strategies have been proposed across negative selection, batch construction, and gradient modulation, their actual effectiveness under controlled, reproducible conditions remains poorly understood. This motivates a fundamental question: Do sophisticated contrastive strategies consistently bring meaningful downstream gains in multimodal embedding learning? To address this, we systematically evaluate representative strategies under a unified benchmarking framework on MMEB. Surprisingly, direct evaluations reveal that most existing strategies offer marginal gains over a standard aligned baseline, with several even degrading overall performance. To uncover the underlying failure modes, we introduce an MLLM-as-a-Judge diagnostic framework that dissects top-1 retrieval behaviors into labeled positives, false negatives (unlabeled valid matches), and hard/easy negative errors. Our systematic analysis yields three key insights: (1) Gradient distortion: Upweighting gradients from high-similarity negatives consistently hurts downstream generalization; (2) Distribution illusion: Wider separation between positive and negative similarity distributions does not necessarily translate to better retrieval performance; and (3) Synergistic training: While false-negative mitigation or hard-batch construction yields limited gains in isolation, integrating hard-negative mining, false-negative mitigation, and hard-batch construction achieves optimal performance, delivering an average improvement of 6.0% on MMEB with Qwen2-VL-2B. Overall, our findings demystify popular contrastive strategies and provide actionable recipes for training universal multimodal embedding models.

Chinese Translation

对比学习已成为多模态表示学习的标准范式，通过在共享嵌入空间中对语义匹配的对进行对齐，同时分离不匹配的对。尽管在负样本选择、批次构建和梯度调节等方面提出了多种对比训练策略，但在可控、可重复的条件下，它们的实际有效性仍然不甚明了。这引发了一个根本性的问题：复杂的对比策略是否始终能在多模态嵌入学习中带来有意义的下游收益？为了解决这个问题，我们在统一的基准框架下系统性地评估了代表性的策略。令人惊讶的是，直接评估显示大多数现有策略相较于标准对齐基线仅提供了微小的收益，甚至有几种策略导致整体性能下降。为了揭示潜在的失败模式，我们引入了MLLM-as-a-Judge诊断框架，将top-1检索行为分解为标记的正样本、假阴性（未标记的有效匹配）和困难/简单的负错误。我们的系统分析得出了三个关键见解：（1）梯度扭曲：来自高相似度负样本的梯度加权会持续损害下游泛化；（2）分布错觉：正负相似度分布之间的更大分离并不一定转化为更好的检索性能；（3）协同训练：虽然假阴性缓解或困难批次构建在孤立情况下收益有限，但将困难负样本挖掘、假阴性缓解和困难批次构建结合起来可以实现最佳性能，在MMEB上平均提高6.0%（使用Qwen2-VL-2B）。总体而言，我们的发现揭示了流行的对比策略的本质，并为训练通用多模态嵌入模型提供了可行的方案。

View on arXiv Download PDF AI Translation

cs.CV / 74 / 2610.15049

Use Of Convolutional Neural Networks In Detection Of Branch Dieback Disease In Pequi Trees

卷积神经网络在佩基树枝枯萎病检测中的应用

Pinho, Ronilson Rodrigues, Coutinho, Elua Ramos, Sales, Nilza de Lima Pereira, Ribeiro, Sildenir Aalves, da Cruz, Sergio Manuel Serra

Abstract

This study aims to evaluate and compare convolutional neural network (CNN) architectures, specifically AlexNet, Inception, LeNet, VGG16, and MobileNet, for diagnosing branch dieback (seca-dos-ponteiros) in pequi trees (Caryocar brasiliense). This disease represents a severe threat to tree development and fruit yield, causing substantial economic and ecological losses. The methodology implements computer vision techniques using Python libraries, including Scikit-learn, NumPy, TensorFlow, and Matplotlib, to process and evaluate visual data. The CNN models analyze images of leaves, branches, and flower spikes to detect pathological patterns, such as alterations in texture and coloration. Ultimately, this comparative analysis identifies the most effective architecture for accurately distinguishing healthy plants from diseased ones.

Chinese Translation

本研究旨在评估和比较卷积神经网络（CNN）架构，特别是AlexNet、Inception、LeNet、VGG16和MobileNet，用于诊断佩基树（Caryocar brasiliense）的枝枯萎病（seca-dos-ponteiros）。该病对树木生长和果实产量构成了严重威胁，造成了显著的经济和生态损失。该方法采用计算机视觉技术，使用Python库，包括Scikit-learn、NumPy、TensorFlow和Matplotlib，处理和评估视觉数据。CNN模型分析叶子、树枝和花序的图像，以检测病理模式，如纹理和颜色的变化。最终，这一比较分析确定了最有效的架构，以准确区分健康植物与病害植物。

View on arXiv Download PDF AI Translation

cs.CV / 75 / 2610.15061

Deep Learning-based Semantic Segmentation of Dental Radiographs: Comparative Evaluation of PSPNet, DeepLabv3, UNet, and UNet++

基于深度学习的牙科放射影像语义分割：PSPNet、DeepLabv3、UNet 和 UNet++ 的比较评估

Nawarathne, Malithi, Sangeetha, Nadeesha Herath Mudiyanselage Lihini

Abstract

Accurate and efficient analysis of dental radiographs has become increasingly important for disease diagnosis and clinical decision making. In particular, dental radiographs segmentation plays a critical role as it is challenging due to several reasons, such as low contrast, noise in the images, variations in the image quality, etc. Previous studies have primarily addressed these challenges using convolutional neural networks (CNN)-based models. However, direct comparison of these segmentation models is still challenging due to the variations among the experimental settings. Therefore, we mainly compared four segmentation models, including Pyramid Scene Parsing Network (PSPNet), DeepLabV3, UNet, and UNet++, through a fair, controlled experimentation setting under varying conditions such as image sizes and batch sizes. For this, we used 1269 dental radiography images, and prior to training, we implemented a series of preprocessing techniques. To assess the models, we used several metrics, including Dice coefficient (DICE), Intersection over Union (IoU), accuracy, precision, training and validation losses. According to our results, UNet++ demonstrated the highest IoU score of 0.6719, and a DICE value of 0.7869, showing superior results in terms of segmentation, specifically when configured with an image size of 300x300 and a batch size of 16. In addition, we also explored the effect of image resolutions and batch sizes on model performance and accuracy under different architectures. Therefore, this study provides insights into the strengths of different segmentation models and the potential of deep learning-based semantic segmentation methods for improving automated dental radiograph analysis.

Chinese Translation

准确高效地分析牙科放射影像在疾病诊断和临床决策中变得越来越重要。特别是，牙科放射影像的分割由于多种原因（如低对比度、图像噪声、图像质量变化等）而具有挑战性。以往的研究主要通过基于卷积神经网络（CNN）的模型来应对这些挑战。然而，由于实验设置之间的差异，直接比较这些分割模型仍然具有挑战性。因此，我们主要通过在不同条件下（如图像大小和批量大小）进行公平、受控的实验设置，比较了四种分割模型，包括金字塔场景解析网络（PSPNet）、DeepLabV3、UNet 和 UNet++。为此，我们使用了1269张牙科放射影像，并在训练之前实施了一系列预处理技术。为了评估模型，我们使用了多个指标，包括Dice系数（DICE）、交并比（IoU）、准确率、精确率、训练和验证损失。根据我们的结果，UNet++展示了最高的IoU得分为0.6719，DICE值为0.7869，在分割方面表现优越，特别是在图像大小为300x300和批量大小为16的配置下。此外，我们还探讨了图像分辨率和批量大小对模型性能和准确性的影响。因此，本研究为不同分割模型的优势以及基于深度学习的语义分割方法在改善自动化牙科放射影像分析中的潜力提供了见解。

View on arXiv Download PDF AI Translation

cs.CV / 76 / 2610.15064

Global Disaster Risk Assessment Using World Risk Index Data: A Geospatial and Visual Analytics Approach

利用世界风险指数数据进行全球灾害风险评估：一种地理空间与视觉分析方法

Nawarathne, Malithi, Mudiyanselage, Nadeesha Herath, Sangeetha, Lihini

Abstract

The world has seen several threats that include resource depletion, food security, poverty and inequality, difficulties associated with urbanization, migration, and environmental effects. The World Risk Index (WRI) is a metric that is used to measure the susceptibility of nations against natural disasters and related risks that is reported by the Humanitarian Data Exchange. The WRI is comprised of factors such as susceptibility, vulnerability to natural hazards, and coping mechanisms. The WRI index has allowed to recognize regions in advance in order to face threats and situations beforehand. This study was conducted by analyzing the data that was related to years from 2000 to 2025 using data visualization techniques. Python related libraries of Matplotlib, Seaborn, and Geopandas were used due to their efficiency and usability in analyzing WRI data. As per the analysis, it was identified that countries of Philippines, Mexico, Indonesia, and India were recorded with high WRI index values that showed these nations should be prepared for natural disasters and situations. Hence, this study has provided insights using data visualization techniques in understanding natural disasters and threats so that necessary measures can be taken

Chinese Translation

世界面临多种威胁，包括资源枯竭、粮食安全、贫困与不平等、城市化相关的困难、迁徙以及环境影响。世界风险指数（World Risk Index, WRI）是一种用于衡量国家对自然灾害及相关风险的脆弱性的指标，由人道主义数据交换平台提供。WRI包括脆弱性、对自然灾害的易感性以及应对机制等因素。WRI指数使得能够提前识别出面临威胁和情况的地区。本研究通过分析2000年至2025年间的数据，采用数据可视化技术进行研究。由于其在分析WRI数据中的高效性和可用性，本研究使用了Python相关库Matplotlib、Seaborn和Geopandas。分析结果显示，菲律宾、墨西哥、印度尼西亚和印度等国的WRI指数值较高，表明这些国家应为自然灾害和相关情况做好准备。因此，本研究通过数据可视化技术提供了对自然灾害和威胁的理解，以便采取必要的措施。

View on arXiv Download PDF AI Translation

cs.CV / 77 / 2610.15077

BrainAid: Efficient Brain Tumor Detection Using Lightweight CNNs and Explainable AI for Resource-Constrained Environments

BrainAid：在资源受限环境中使用轻量级卷积神经网络和可解释人工智能进行高效脑肿瘤检测

Gonthina, Sai Sriram, Gore, Ross, Singh, Bikash Chandra, Bhattacharya, Pronaya, Shetty, Sachin, Phadke, Abhishek, Roy, Sandip

Abstract

A brain tumor is an unusual growth of cells in the brain that can be cancerous or not, affecting normal brain function and causing serious risk to health. Existing solutions for brain tumor classification, such as expert manual evaluations of magnetic resonance imaging (MRI) scans, are generally inconsistent as different professionals have different criteria and can be resource-intensive, leading to delayed diagnoses. Additionally, while deep learning models have demonstrated potential, they typically require significant computational resources, motivating the need for more efficient and accessible diagnostic tools that can be deployed in real-time healthcare settings. To address these limitations, this study presents BrainAid, which makes efficient use of Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) for effective classification of four-class classification of glioma, meningioma, pituitary tumor, and no-tumor MRI scans, outperforming all the standard models such as VGG16, Inception-V3, ResNet-50 and so, while integrating Explainable AI (XAI) to improve interpretability. The BrainAid is trained to classify tumors with a high precision of 98%, achieving an F1 score of 98% and an accuracy of 98% using publicly available dataset. The model is deployed on a Raspberry Pi, demonstrating efficiency in low-resource environments. This system shows the feasibility of real-time, edge-based tumor classification, providing quick diagnostic support for medical professionals and enhancing patient care in resource-limited healthcare facilities and underserved areas.

Chinese Translation

脑肿瘤是脑内细胞异常生长的表现，可能是癌性或非癌性，影响正常脑功能并对健康造成严重风险。现有的脑肿瘤分类解决方案，如对磁共振成像（MRI）扫描的专家手动评估，通常不一致，因为不同的专业人员有不同的标准，并且可能资源消耗较大，导致诊断延迟。此外，尽管深度学习模型显示出潜力，但它们通常需要大量计算资源，这促使我们需要更高效且可访问的诊断工具，以便在实时医疗环境中部署。为了解决这些局限性，本研究提出了BrainAid，该系统高效利用卷积神经网络（CNNs）和视觉变换器（ViTs）对胶质瘤、脑膜瘤、垂体肿瘤和无肿瘤的MRI扫描进行四类分类，超越了所有标准模型，如VGG16、Inception-V3、ResNet-50等，同时集成了可解释人工智能（XAI）以提高可解释性。BrainAid经过训练，能够以98%的高精度分类肿瘤，F1得分达到98%，准确率为98%，使用的是公开可用的数据集。该模型部署在树莓派（Raspberry Pi）上，展示了在低资源环境中的高效性。该系统展示了实时边缘肿瘤分类的可行性，为医疗专业人员提供快速的诊断支持，并提升资源有限的医疗设施和服务不足地区的患者护理质量。

View on arXiv Download PDF AI Translation

cs.CV / 78 / 2610.15079

Region Aware Squeeze-and-Excitation Networks for Image Segmentation

区域感知的挤压与激励网络用于图像分割

Roberts, Jack, Tummala, Praneeth, Sarna, Kavi, Neto, Jeova Farias Sales Rocha

Abstract

Squeeze-and-Excitation (SE) modules improve convolutional networks by modeling channel-wise dependencies, but global average pooling collapses spatial structure into a single per-channel descriptor, limiting their effectiveness for dense prediction tasks such as image segmentation. We generalize SE by replacing global pooling with spatially guided pooling, enabling region-aware channel recalibration through four modules with complementary cost profiles. BASE formalizes Tiled Squeeze-and-Excite using a regular grid; SPASE pools over SLIC superpixels for image-adaptive regions; MAPSE obtains regions from semantic maps produced by a frozen, general-purpose network; and PLEASE learns soft region assignments end-to-end through a lightweight assignment network. Each channel receives one gate per region, so attention at each pixel reflects its region membership, bridging channel and spatial attention. PLEASE also produces its soft assignments as a byproduct of the forward pass, providing structured spatial interpretability without post-hoc analysis. The modules offer complementary computational tradeoffs: some add essentially no parameters, while others add no measurable latency. Across binary segmentation on CUB, DUTS, and ECSSD, semantic segmentation on PASCAL VOC 2012 and ADE20K, and classification on mini-ImageNet, our modules outperform SE and related spatial-channel attention variants on most benchmarks, achieving the majority of best results.

Chinese Translation

挤压与激励（Squeeze-and-Excitation, SE）模块通过建模通道间依赖关系来改善卷积网络，但全局平均池化将空间结构压缩为单一的每通道描述符，这限制了其在图像分割等密集预测任务中的有效性。我们通过用空间引导池化替代全局池化来推广SE，从而实现区域感知的通道重校准，采用四个具有互补成本特征的模块。BASE使用规则网格形式化了平铺挤压与激励（Tiled Squeeze-and-Excite）；SPASE对SLIC超像素进行池化，以适应图像的区域；MAPSE从由冻结的通用网络生成的语义图中获取区域；而PLEASE通过轻量级分配网络端到端学习软区域分配。每个通道为每个区域接收一个门控，因此每个像素的注意力反映其区域归属，桥接了通道和空间注意力。PLEASE还在前向传播过程中生成其软分配，提供结构化的空间可解释性，而无需事后分析。这些模块提供互补的计算权衡：有些几乎不增加参数，而另一些则不会增加可测量的延迟。在CUB、DUTS和ECSSD的二元分割、PASCAL VOC 2012和ADE20K的语义分割，以及mini-ImageNet的分类任务中，我们的模块在大多数基准测试中超越了SE及相关的空间-通道注意力变体，取得了大多数最佳结果。

View on arXiv Download PDF AI Translation

cs.CV / 79 / 2610.15092

VLCQ: Post-training quantization for deep neural networks using variable length coding

VLCQ：基于可变长度编码的深度神经网络后训练量化

Abdel-Salam, Reem, Abdel-Gawad, Ahmed H., Wassal, Amr G.

Abstract

Quantization plays a crucial role in efficiently deploying deep learning models on resource-constrained devices. Post-training quantization does not require access to the original dataset or retraining the full model. Current methods that achieve high performance close to baseline results typically require INT8 fixed-point integers. However, achieving higher model compression with lower bit-widths often leads to significant performance degradation. In this paper, we propose VLCQ, a post-training quantization method that relaxes the fixed-point encoding constraint, enabling more flexible exploration of quantization techniques through variable-length encoding. This approach achieves competitive or improved performance while reducing bit-widths, without requiring access to training data or model fine-tuning. Extensive experiments on image classification, segmentation, and object detection demonstrate that VLCQ improves model compression and accuracy compared with state-of-the-art post-training quantization approaches. For per-channel quantization, VLCQ surpasses FP32 accuracy and Piece-Wise Linear Quantization (PWLQ) on most evaluated models, achieving up to 6x compression compared with FP32 and up to 1.7x compared with PWLQ. When prioritizing compression with limited accuracy loss, VLCQ achieves up to 12.25x compression compared with FP32 with less than 4% performance degradation. For per-tensor quantization, VLCQ is competitive with Data-Free Quantization (DFQ), while providing greater flexibility to achieve lower bit rates across different tasks and models.

Chinese Translation

量化在高效部署深度学习模型于资源受限设备中扮演着至关重要的角色。后训练量化不需要访问原始数据集或重新训练完整模型。目前实现接近基线结果的高性能方法通常需要使用 INT8 固定点整数。然而，使用更低位宽实现更高模型压缩通常会导致显著的性能下降。本文提出了一种后训练量化方法 VLCQ，它放宽了固定点编码的约束，通过可变长度编码实现对量化技术的更灵活探索。该方法在降低位宽的同时，能够实现具有竞争力或更好的性能，而无需访问训练数据或进行模型微调。在图像分类、分割和目标检测的广泛实验中，VLCQ 相较于最先进的后训练量化方法在模型压缩和准确性上均有所提升。对于每通道量化，VLCQ 在大多数评估模型上超越了 FP32 的准确性和分段线性量化（Piece-Wise Linear Quantization, PWLQ），与 FP32 相比实现了高达 6 倍的压缩，与 PWLQ 相比实现了高达 1.7 倍的压缩。当优先考虑压缩而有限制准确性损失时，VLCQ 在与 FP32 相比的情况下实现了高达 12.25 倍的压缩，且性能下降不足 4%。对于每张量量化，VLCQ 在与无数据量化（Data-Free Quantization, DFQ）竞争的同时，提供了更大的灵活性，以在不同任务和模型中实现更低的比特率。

View on arXiv Download PDF AI Translation

cs.CV / 80 / 2610.15098

TIDE: Trajectory-Informed Skill Distillation for Efficient Text-to-Video Prompt Optimization

TIDE：基于轨迹的信息技能蒸馏用于高效文本到视频提示优化

Zhou, Chengxing, Huang, Yuliang, Song, Jingwei

Abstract

Text-to-video (T2V) models are highly sensitive to prompt specificity, while iterative prompt optimization typically improves generation quality only through repeated video generation and vision-language model evaluation at inference time. We propose TIDE, a framework that distills this costly search process into an interpretable and reusable prompt-rewriting Skill. TIDE formulates prompt rewriting as a set of source-conditioned rule modules, each targeting a distinct semantic dimension. Once distilled, the frozen Skill activates the relevant rules and generates both positive and negative prompts in a single LLM call, followed by only one T2V generation, without test-time visual search or backbone fine-tuning. Evolved using only 100 prompts over 14 hours, TIDE achieves prompt-optimization performance comparable to computationally expensive search-based methods. Moreover, when transferred to previously unseen video-generation models, TIDE outperforms existing approaches specifically designed or trained for those models. Code is available at https://github.com/xiningin/TIDE_T2V.

Chinese Translation

文本到视频（T2V）模型对提示的具体性高度敏感，而迭代提示优化通常仅通过在推理时反复生成视频和评估视觉-语言模型来提高生成质量。我们提出了TIDE，一个将这一昂贵搜索过程蒸馏为可解释且可重用的提示重写技能的框架。TIDE将提示重写形式化为一组源条件规则模块，每个模块针对一个不同的语义维度。一旦蒸馏完成，冻结的技能激活相关规则，并在一次大型语言模型（LLM）调用中生成正向和负向提示，随后仅需一次T2V生成，无需测试时的视觉搜索或主干微调。TIDE仅使用100个提示经过14小时的演化，便实现了与计算成本高昂的基于搜索的方法相当的提示优化性能。此外，当转移到之前未见过的视频生成模型时，TIDE的表现优于专门为这些模型设计或训练的现有方法。代码可在 https://github.com/xiningin/TIDE_T2V 获取。

View on arXiv Download PDF AI Translation

cs.CV / 81 / 2610.15112

Drive3R: Block-wise Streaming Geometric Foundation Models for Autonomous Driving

Drive3R：用于自主驾驶的块级流式几何基础模型

Yang, Tianyi, Yang, Haojun, Xiao, Jipeng, Chen, Xuesong, Shi, Shaoshuai, Jiang, Li, Zhang, Zhipeng

Abstract

High-fidelity closed-loop simulation for autonomous driving increasingly relies on dense multi-view 3D reconstruction. While recent feed-forward geometric foundation models have revolutionized per-scene optimization, applying them to continuous, long-horizon driving sequences is computationally intractable due to GPU memory limits and the quadratic complexity of global attention. Existing scaling strategies are fundamentally limited, as causal streaming suffers from compounding scale drift and trajectory deviation, while rigid "chunk-then-align" heuristics fragment global context and produce trajectory discontinuities. To address these limitations, we introduce Drive3R, a block-wise streaming geometric foundation model. Its block-wise causal attention resolves the scalability bottleneck of fully bidirectional attention while avoiding the loss of local future context in purely causal streaming. Its confidence-aware re-anchoring further mitigates long-range drift by detecting unreliable streaming states, re-initializing the KV cache, and stitching refreshed segments through SE(3) alignment. Extensive experiments demonstrate that Drive3R establishes a new state-of-the-art on nuScenes, KITTI, and Waymo. Specifically, on nuScenes, it improves over the recent DVGT-2 baseline from 0.1806 to 0.1081 in AbsRel, from 0.6750 to 0.5344 in point-cloud Accuracy, and from 1.5812 to 0.9092 in ATE.

Chinese Translation

高保真闭环仿真在自主驾驶中越来越依赖于密集的多视角三维重建。尽管近期的前馈几何基础模型已经彻底改变了每个场景的优化，但由于GPU内存限制和全局注意力的平方复杂性，将其应用于连续的长时间驾驶序列在计算上是不可行的。现有的扩展策略存在根本性限制，因为因果流式处理面临着累积的尺度漂移和轨迹偏差，而刚性的“块后对齐”启发式方法则会破坏全局上下文并产生轨迹不连续性。为了解决这些限制，我们提出了Drive3R，一种块级流式几何基础模型。其块级因果注意力解决了完全双向注意力的可扩展性瓶颈，同时避免了在纯因果流式处理中丢失局部未来上下文。其基于置信度的重新锚定进一步通过检测不可靠的流式状态、重新初始化KV缓存以及通过SE(3)对齐拼接更新的段落来减轻长距离漂移。大量实验表明，Drive3R在nuScenes、KITTI和Waymo上建立了新的最先进水平。具体而言，在nuScenes上，它在AbsRel上将近期的DVGT-2基线从0.1806提升至0.1081，在点云精度上从0.6750提升至0.5344，在ATE上从1.5812提升至0.9092。

View on arXiv Download PDF AI Translation

cs.CV / 82 / 2610.15127

RIECNN: Real-Time Image Enhanced CNN for Traffic Sign Recognition

RIECNN：用于交通标志识别的实时图像增强卷积神经网络

Abdel-Salam, Reem, Mostafa, Rana, Abdel-Gawad, Ahmed H.

Abstract

Traffic sign recognition plays a crucial role in the development of autonomous cars to reduce the accident rate and promote road safety. It has been a necessity to address traffic signs that are affected significantly by the environment as well as poor real-time performance for deep-learning state-of-the-art algorithms. In this paper, we introduce Real-Time Image Enhanced CNN ({\our}) for Traffic Sign Recognition. {\our} is a real-time, novel approach that tackles multiple, diverse traffic sign datasets, and out-performs the state-of-the-art architectures in terms of recognition rate and execution time. Experiments are conducted using the German Traffic Sign Benchmark (GTSRB), the Belgium Traffic Sign Classification (BTSC), and the Croatian Traffic Sign (rMASTIF) benchmark. Experimental results show that our approach has achieved the highest recognition rate for all Benchmarks, achieving a recognition accuracy of 99.75\% for GTSRB, 99.25\% for BTSC and 99.55\% for rMASTIF. In terms of latency and meeting the real-time constraint, the pre-processing time and inference time together do not exceed 1.3 ms per image. Not only have our proposed approach achieved remarkably high accuracy with real-time performance, but it also demonstrated robustness against traffic sign recognition challenges such as brightness and contrast variations in the environment.

Chinese Translation

交通标志识别在自动驾驶汽车的发展中发挥着至关重要的作用，能够降低事故率并促进道路安全。解决受环境显著影响的交通标志以及深度学习最先进算法的实时性能不足已成为一项必要任务。本文介绍了一种用于交通标志识别的实时图像增强卷积神经网络（RIECNN）。RIECNN是一种实时的新方法，能够处理多种多样的交通标志数据集，并在识别率和执行时间上超越最先进的架构。实验使用德国交通标志基准（GTSRB）、比利时交通标志分类（BTSC）和克罗地亚交通标志（rMASTIF）基准进行。实验结果表明，我们的方法在所有基准中均达到了最高的识别率，GTSRB的识别准确率为99.75%，BTSC为99.25%，rMASTIF为99.55%。在延迟和满足实时约束方面，预处理时间和推理时间总和不超过每张图像1.3毫秒。我们提出的方法不仅实现了显著高的准确率和实时性能，还在面对亮度和对比度变化等交通标志识别挑战时表现出良好的鲁棒性。

View on arXiv Download PDF AI Translation

cs.CV / 83 / 2610.15164

TRIG: Toward Measuring Texture Realism in Image Generation

TRIG：朝着测量图像生成中的纹理真实感

Xu, Zhiwei, Cui, Jiali, Lao, Zhiqiang, Yu, Heather, Peng, Liang

Abstract

This paper studies the backbone used by metrics for evaluating the quality of generated images. Although generative models have advanced rapidly, some outputs still exhibit unrealistic texture characteristics, and these differences are difficult to measure. Widely used evaluation metrics provide limited insight into texture realism, since they often rely on a backbone trained for another objective, such as image classification or image-text alignment. Such objectives typically encourage invariance to low-level variations that carry texture, so the resulting features primarily encode semantic and global structure. As a result, texture-specific discrepancies may not be explicitly captured. We propose TRIG, a metric directed at the texture realism of generated images. Its backbone is instead the discriminator of a generative adversarial network pre-trained on a texture dataset, which provides a learned prior for the local statistics of natural textures. A plausibility head is fine-tuned to identify local texture degradations, while preserving natural defocus and sensor noise as positive cases. On the features of this backbone, we compute a standard distributional distance between a reference set of real images and a set of generated ones. We apply TRIG to several generative models and observe closer agreement with human realness ratings than baseline metrics.

Chinese Translation

本文研究了用于评估生成图像质量的指标所依赖的基础架构。尽管生成模型迅速发展，但某些输出仍表现出不真实的纹理特征，而这些差异难以测量。广泛使用的评估指标对纹理真实感的洞察有限，因为它们通常依赖于为其他目标（如图像分类或图像-文本对齐）训练的基础架构。这些目标通常鼓励对承载纹理的低级变化的不变性，因此生成的特征主要编码语义和全局结构。结果，纹理特定的差异可能未被明确捕捉。我们提出了TRIG，这是一种针对生成图像纹理真实感的指标。其基础架构是一个在纹理数据集上预训练的生成对抗网络的判别器，提供了自然纹理局部统计的学习先验。一个合理性头部经过微调，以识别局部纹理退化，同时将自然散焦和传感器噪声作为正例。在该基础架构的特征上，我们计算了一组真实图像和一组生成图像之间的标准分布距离。我们将TRIG应用于多个生成模型，并观察到其与人类真实感评分的吻合度优于基线指标。

View on arXiv Download PDF AI Translation

cs.CV / 84 / 2610.15166

ExeWorldBench: Benchmarking Code-Based Executable 3D World Modeling

ExeWorldBench：基于代码的可执行3D世界建模基准测试

Peng, Xinge, Fang, Zhen, Wang, Ruibing, Lv, Yongxuan, Li, Haodong, Zhang, Guohui, Ma, Xiaoxiao, Lu, Yiting, Chen, Lin, Che, Zehui, Zhao, Feng

Abstract

Coding agents are increasingly evolving beyond reproducing what a world looks like toward constructing executable worlds that capture how a world works. This shift calls for executable world modeling, which integrates two core capabilities: world understanding, reasoning about the underlying structural, mechanical, and causal principles of a world, and executable realization, translating that understanding into a code-based system whose behavior emerges from physical interactions. However, existing benchmarks provide only fragmented evaluations centered on physical reasoning, visual fidelity, or behavioral correctness, leaving the integration of these capabilities insufficiently assessed. To bridge this gap, we introduce ExeWorldBench, to our knowledge the first benchmark to comprehensively evaluate world understanding and executable realization through code-based 3D world construction. ExeWorldBench comprises three progressively challenging tasks: Mechanism Executability, which evaluates the realization of individual mechanisms from visual observations; Compositional Executability, which extends to multi-stage systems with spatial, mechanical, and causal dependencies; and Intent-Driven Executable World Synthesis, which requires constructing executable worlds from high-level goals and physical requirements. Extensive experiments on frontier models reveal substantial limitations in executable world modeling, offering insights into the development of more physically grounded and causally coherent world construction systems.

Chinese Translation

编码代理正在逐步超越仅仅重现世界的外观，向构建可执行世界发展，以捕捉世界的运作方式。这一转变呼唤可执行世界建模，它整合了两个核心能力：世界理解，即推理世界的基础结构、机械和因果原则；以及可执行实现，即将这种理解转化为一个基于代码的系统，其行为源于物理交互。然而，现有基准测试仅提供了围绕物理推理、视觉真实感或行为正确性的片段评估，导致这些能力的整合评估不足。为了填补这一空白，我们引入了ExeWorldBench，作为我们所知的第一个全面评估世界理解和可执行实现的基准，通过基于代码的3D世界构建。ExeWorldBench包括三个逐渐具有挑战性的任务：机制可执行性（Mechanism Executability），评估从视觉观察中实现单个机制的能力；组合可执行性（Compositional Executability），扩展到具有空间、机械和因果依赖的多阶段系统；以及意图驱动的可执行世界合成（Intent-Driven Executable World Synthesis），要求根据高层目标和物理要求构建可执行世界。在前沿模型上的广泛实验揭示了可执行世界建模的重大局限性，为开发更具物理基础和因果一致性的世界构建系统提供了见解。

View on arXiv Download PDF AI Translation

cs.CV / 85 / 2610.15280

GeoBridge: Evolving Transferable Decoder Topologies for Geometric Vision

GeoBridge：用于几何视觉的可迁移解码器拓扑的演化

Wang, Rui, Zhang, Haoyu, Cheng, Ran

Abstract

Depth estimation and Gaussian reconstruction share a need for multi-scale feature processing, yet use different prediction interfaces and training objectives. Whether a decoder's internal computation can be optimized once and reused across these systems remains a structural design question. Search over primitive configurations alone leaves the organization of transformations fixed. To address this gap, we propose GeoBridge, an evolutionary framework for transferable decoder topologies. First, GeoBridge makes stage-internal topology searchable through residual, sequential, and parallel composition of configurable primitives. Shape-preserving composition retains the stage interfaces of a fixed multi-scale scaffold. Second, a functional-tree encoding supports stage-aware crossover and hierarchical mutation, while a gradient sensitivity-alignment proxy evaluates candidates without training. Multi-objective evolution balances this proxy with computation and parameter cost. Third, selected topologies are instantiated and trained in target systems with task-specific prediction interfaces. A single source search takes approximately 12 GPU hours. GeoBridgeNet-A3 achieves the best depth accuracy among four adapted decoder baselines under a common encoder and training protocol on NYU Depth V2 and KITTI. The same topology reaches 27.11 dB PSNR in Gaussian novel-view synthesis on RealEstate10K, and its refined representations support open-vocabulary segmentation on LERF. These results support decoder topology as a reusable design object across the evaluated geometric vision systems. Code and pretrained weights are available at https://github.com/EMI-Group/GeoBridge-NAS.

Chinese Translation

深度估计和高斯重建都需要多尺度特征处理，但它们使用不同的预测接口和训练目标。解码器的内部计算是否可以一次优化并在这些系统中重复使用仍然是一个结构设计问题。仅对原始配置进行搜索会使变换的组织保持固定。为了解决这一问题，我们提出了GeoBridge，一个用于可迁移解码器拓扑的演化框架。首先，GeoBridge通过可配置原件的残差、顺序和并行组合，使阶段内部拓扑可搜索。保持形状的组合保留了固定多尺度支架的阶段接口。其次，功能树编码支持阶段感知的交叉和层次突变，而梯度敏感性对齐代理在不进行训练的情况下评估候选者。多目标演化在计算和参数成本之间平衡这一代理。第三，选定的拓扑在具有任务特定预测接口的目标系统中实例化并训练。单次源搜索大约需要12个GPU小时。GeoBridgeNet-A3在NYU Depth V2和KITTI的共同编码器和训练协议下，在四个适应解码器基线中实现了最佳深度准确性。同一拓扑在RealEstate10K的高斯新视图合成中达到了27.11 dB PSNR，其精细化表示支持在LERF上的开放词汇分割。这些结果支持解码器拓扑作为在评估的几何视觉系统中可重用的设计对象。代码和预训练权重可在https://github.com/EMI-Group/GeoBridge-NAS获取。

View on arXiv Download PDF AI Translation

cs.CV / 86 / 2610.15286

Protein Crystallization Segmentation and Classification Using Subordinate Color Channel in Fluorescence Microscopy Images

利用荧光显微镜图像中的从属颜色通道进行蛋白质结晶的分割与分类

Tran, Truong X., Pusey, Marc L., Aygun, Ramazan S.

Abstract

The accuracy of detecting protein crystals for fluorescence microscopy images is very critical for high throughput and automated systems. Although the trace fluorescent labeling method could highlight protein crystals, reflection and emission from the fluorescence dye is not always due to crystal regions. Therefore, the analysis of the peak wavelength in the emission spectra of a fluorophore may not always yield effective results. In this paper, we show that using the subordinate color intensity corresponding to longer wavelengths than the peak wavelength of the emission spectra could improve the accuracy of protein crystal detection. Hence, we have built a segmentation method based on the percentile intensity of the subordinate color for trace fluorescently labeled (TFL'd) protein crystallization trial images. Compared to using the dominant color channel, our segmentation method on subordinate color channel was able to reduce the misclassification rate of likely-leads or crystals as non-crystals by the percentage of from 9.71% to 2.02% depending on the classifier. Similarly, the accuracy of classifiers were increased by the percentage of from 1.77% to 5.53%. Our method reached around 94% accuracy while keeping misclassification of likely-leads and crystals as non-crystals below 1%. Moreover, to evaluate the generalizability of our method, we have conducted new wet lab experiments on two proteins, Concanavalin A (Con A) and Ab inorganic pyrophosphate (AbIPPase), and the misclassification rate was below 1%. Our experiments show that using the subordinate channel may be more helpful for TFL'd protein trial image classification.

Chinese Translation

在荧光显微镜图像中检测蛋白质晶体的准确性对于高通量和自动化系统至关重要。尽管痕量荧光标记方法可以突出显示蛋白质晶体，但荧光染料的反射和发射并不总是来自晶体区域。因此，荧光体发射光谱中的峰值波长分析可能并不总能产生有效结果。本文展示了使用对应于发射光谱峰值波长更长波长的从属颜色强度可以提高蛋白质晶体检测的准确性。因此，我们建立了一种基于从属颜色百分位强度的分割方法，适用于痕量荧光标记（TFL'd）蛋白质结晶实验图像。与使用主色通道相比，我们的从属颜色通道分割方法能够将可能的晶体或线索误分类为非晶体的比例降低，从9.71%降至2.02%，具体取决于分类器。同样，分类器的准确性提高了1.77%至5.53%。我们的方法在保持可能的线索和晶体误分类为非晶体的比例低于1%的情况下，达到了约94%的准确率。此外，为了评估我们方法的普适性，我们对两种蛋白质（凝集素A（Concanavalin A, Con A）和无机焦磷酸酶（Ab inorganic pyrophosphate, AbIPPase））进行了新的湿实验，误分类率低于1%。我们的实验表明，使用从属通道可能对TFL'd蛋白质实验图像分类更为有利。

View on arXiv Download PDF AI Translation

cs.CV / 87 / 2610.15287

ASM-6D: Fast Certified Pose-Shape Inference with Complete Correspondence Flow Maps

ASM-6D：快速认证的姿态-形状推断与完整对应流图

Choi, Ho Jin, Kim, Minku, Figueroa, Nadia

Abstract

Category-level 6D pose and shape estimation requires recovering an object's pose and geometry from partial observations without an exact instance model. We present ASM-6D, which couples generative complete-geometry prediction with explicit category-level pose-shape inference through shared correspondences. Given an RGB-D image, ASM-6D generates alternative complete 3D correspondence hypotheses whose point identities are shared with a category-level Active Shape Model (ASM), including occluded points. Each hypothesis is fitted directly for rotation, translation, metric scale, and instance shape without a separate correspondence-assignment stage. We analytically eliminate translation and shape coefficients, reducing fitting to a nine-variable scaled-rotation optimization problem that enables efficient batched evaluation of multiple hypotheses in real time. ASM-6D achieves state-of-the-art performance on ROPE and HouseCat6D benchmarks. Ablations demonstrate the importance of generative correspondence prediction and active shape variation. The representation also supports shape completion, shape-dependent pose ambiguity, tracking, and robotic grasping at 6-38 Hz, depending on the inference budget.

Chinese Translation

类别级别的6D姿态和形状估计需要从部分观测中恢复物体的姿态和几何形状，而无需精确的实例模型。我们提出了ASM-6D，它通过共享对应关系将生成的完整几何预测与显式的类别级别姿态-形状推断相结合。给定一幅RGB-D图像，ASM-6D生成替代的完整3D对应假设，其点的身份与类别级别的主动形状模型（Active Shape Model, ASM）共享，包括被遮挡的点。每个假设直接拟合旋转、平移、度量尺度和实例形状，而无需单独的对应分配阶段。我们通过分析消除平移和形状系数，将拟合简化为一个九变量的缩放旋转优化问题，从而实现对多个假设的高效批量实时评估。ASM-6D在ROPE和HouseCat6D基准测试中达到了最先进的性能。消融实验表明生成对应预测和主动形状变化的重要性。该表示还支持形状补全、形状依赖的姿态歧义、跟踪和机器人抓取，频率在6-38 Hz之间，具体取决于推断预算。

View on arXiv Download PDF AI Translation

cs.CV / 88 / 2610.15297

HGA-Quant: Heterogeneity-Aware Quantization for AutoRegressive Visual Generation Models

HGA-Quant：针对自回归视觉生成模型的异质性感知量化

Liu, Xinmeng, Yan, Xianglong, Yang, Kaicheng, Dong, Weisheng, Zhang, Yulun

Abstract

Recently, AutoRegressive Visual Generation (ARVG) models have emerged as a competitive paradigm for high-quality image synthesis. However, their substantial computational and memory demands make them difficult to deploy on resource-constrained edge devices. Post-Training Quantization (PTQ) provides a promising solution to reduce the computational and memory costs of ARVG models, yet existing methods often suffer from substantial image quality degradation under low-bit quantization. In this paper, we propose a training-free PTQ framework for ARVG, consisting of Tail-Aware Bounded-init Grid Refinement (T-BGR) and Coarse-to-Fine Activation Precision Scheduling (CAPS). T-BGR introduces a tail-aware reconstruction objective for weight quantizer calibration, assigning greater importance to distribution-tail coefficients and preserving fine-grained information without introducing additional parameters or high-precision branches. CAPS performs coarse-to-fine activation precision scheduling by jointly considering the importance and computational cost of different generation stages, enabling efficient mixed-precision allocation under a given bit budget. Extensive experiments on multiple ARVG backbones demonstrate that our framework consistently improves robustness under aggressive low-bit quantization. Notably, on Infinity, our framework achieves near full-precision image generation quality at an effective W4A5.6 activation precision, while existing methods at W4A6 still exhibit noticeable quality degradation or inconsistent performance. Code is available at https://github.com/lxm-lc/HGA-Quant.

Chinese Translation

近年来，自回归视觉生成（ARVG）模型已成为高质量图像合成的竞争性范式。然而，它们巨大的计算和内存需求使得在资源受限的边缘设备上部署变得困难。后训练量化（PTQ）为减少ARVG模型的计算和内存成本提供了一种有前景的解决方案，但现有方法在低位量化下往往面临显著的图像质量下降。在本文中，我们提出了一种针对ARVG的无训练PTQ框架，包括尾部感知的有界初始化网格细化（T-BGR）和粗到细的激活精度调度（CAPS）。T-BGR为权重量化器校准引入了尾部感知的重建目标，赋予分布尾部系数更大的重要性，并在不引入额外参数或高精度分支的情况下保留细粒度信息。CAPS通过共同考虑不同生成阶段的重要性和计算成本，执行粗到细的激活精度调度，从而在给定的位预算下实现高效的混合精度分配。在多个ARVG主干网络上的大量实验表明，我们的框架在激进的低位量化下始终提高了鲁棒性。值得注意的是，在Infinity上，我们的框架在有效的W4A5.6激活精度下实现了接近全精度的图像生成质量，而现有方法在W4A6下仍表现出明显的质量下降或不一致的性能。代码可在https://github.com/lxm-lc/HGA-Quant获取。

View on arXiv Download PDF AI Translation

cs.CV / 89 / 2610.15321

Channel Match: Calibrating Shape Latents for Few-Step 3D Generation

通道匹配：为少步3D生成校准形状潜变量

Yang, Di, Li, Zhihao, Xiong, Yanhai, Wang, Yufei

Abstract

Recent native 3D generators achieve high-quality asset creation by modeling structured 3D latents with flow-based transformers, but their high-resolution generation remains computationally expensive due to costly iterative sampling. In this work, we introduce a few-step distillation framework for native 3D generation, enabled by continuous self-consistency training with a custom JVP FlashAttention kernel and an analytic velocity regression objective for sparse structured 3D latents. To stabilize joint optimization, we further introduce an adaptive loss weighting strategy that automatically balances them without manual tuning. Beyond acceleration, we reveal a previously overlooked failure mode of few-step native 3D generation: sampling-induced latent distribution drift preserves geometric correctness but disrupts downstream PBR texture synthesis, causing texture artifacts on geometrically valid assets. To address this issue, we propose Channel Match, a training-free adaptive per-channel latent calibration method that restores the latent distribution without modifying either generator. We validate our approach on TRELLIS.2 at the $1024^3$ resolution setting. Our two-step model achieves $4.5\times$ faster geometry sampling while maintaining geometry quality comparable to the many-step teacher, and Channel Match further improves rendering quality by recovering the latent interface between geometry and appearance generation.

Chinese Translation

近期的原生3D生成器通过基于流的变换器建模结构化3D潜变量，实现了高质量资产的创建，但由于迭代采样成本高，其高分辨率生成仍然计算开销巨大。在本研究中，我们提出了一种少步蒸馏框架，用于原生3D生成，该框架通过使用自定义的JVP FlashAttention内核和针对稀疏结构化3D潜变量的解析速度回归目标，得以实现连续的自一致性训练。为了稳定联合优化，我们进一步引入了一种自适应损失加权策略，能够在无需手动调节的情况下自动平衡损失。除了加速外，我们还揭示了少步原生3D生成中一个之前被忽视的失败模式：采样引起的潜变量分布漂移虽然保持了几何正确性，但却干扰了下游的PBR纹理合成，导致几何有效资产上的纹理伪影。为了解决这一问题，我们提出了通道匹配（Channel Match），这是一种无训练的自适应每通道潜变量校准方法，能够在不修改生成器的情况下恢复潜变量分布。我们在$1024^3$分辨率设置下对TRELLIS.2验证了我们的方法。我们的两步模型在保持与多步教师相当的几何质量的同时，实现了$4.5 imes$的几何采样加速，而通道匹配进一步通过恢复几何与外观生成之间的潜变量接口来提高渲染质量。

View on arXiv Download PDF AI Translation

cs.CV / 90 / 2610.15324

TempoKV: Predictive KV Cache Quantization for Autoregressive Video Generation

TempoKV：用于自回归视频生成的预测性KV缓存量化

Hu, Jiahui, Mo, Weile, Cai, Peiliang, Liu, Jiacheng, Ren, Jinkui, Zhang, Xiantao, Tu, Xiaobing, Zhang, Linfeng

Abstract

Autoregressive (AR) video diffusion models enable streaming generation of long videos but suffer from significant memory overhead due to the rapidly growing KV cache. Existing KV cache quantization methods are primarily designed for LLMs and transfer poorly to video models, where activations exhibit different statistics and strong spatiotemporal redundancy. We observe that the video KV cache is largely predictable over time, with a clear asymmetry. Keys are highly predictable from their temporal history, while values depend primarily on the current key and need a short history of preceding values. Based on these observations, we propose \textbf{\method}, a training-free predictive KV cache quantization framework for AR video generation. \method predicts keys and values according to their distinct dependencies and stores only the residuals at low precision. Both predictors are lightweight, calibrated once offline by linear regression on a small set of BF16 samples, and remain fixed across prompts and generation runs. Experiments on Self-Forcing, Causal Forcing, and LongCat-Video-13B show that \method consistently achieves higher generation quality than existing KV cache quantization methods at comparable or higher compression ratios, reaching up to $6.71\times$ KV cache compression with the best overall quality--efficiency tradeoff. Our code will be released on GitHub.

Chinese Translation

自回归（AR）视频扩散模型能够实现长视频的流式生成，但由于KV缓存的快速增长，导致显著的内存开销。现有的KV缓存量化方法主要针对大型语言模型（LLMs）设计，且在视频模型中效果不佳，因为视频激活展现出不同的统计特性和强烈的时空冗余。我们观察到，视频KV缓存在时间上具有很大的可预测性，并且存在明显的不对称性。键的时间历史具有高度可预测性，而值主要依赖于当前键，并需要短期的前值历史。基于这些观察，我们提出了 extbf{ extit{TempoKV}}，一个无训练的预测性KV缓存量化框架，用于AR视频生成。 extit{TempoKV}根据键和值的不同依赖关系进行预测，并仅以低精度存储残差。两个预测器都是轻量级的，通过对一小组BF16样本进行线性回归在离线状态下进行一次性校准，并在提示和生成运行中保持固定。在Self-Forcing、Causal Forcing和LongCat-Video-13B上的实验表明， extit{TempoKV}在可比或更高的压缩比下，始终实现了比现有KV缓存量化方法更高的生成质量，达到最高$6.71 imes$的KV缓存压缩，同时实现最佳的整体质量-效率权衡。我们的代码将发布在GitHub上。

View on arXiv Download PDF AI Translation

cs.CV / 91 / 2610.15330

BaT-SAM: Introducing Inductive Bias into SAM via a Mixture of Convolutional Experts for Brain Tumor Segmentation

BaT-SAM：通过卷积专家的混合引入归纳偏置以实现脑肿瘤分割

Jannat, Sumiaya, Hasan, Md Mahedi, Islam, Md Shamimul, Hossain, Md Nur

Abstract

Accurate brain tumor segmentation from magnetic resonance imaging (MRI) is essential for diagnosis and treatment planning; however, manual delineation is labor-intensive and time-consuming. Existing approaches are either task-specific with limited generalization or rely on user-provided prompts during inference. We present BaT-SAM, a parameter-efficient adaptation framework that enables fully automatic brain tumor segmentation with the Segment Anything Model (SAM). BaT-SAM injects local inductive bias into SAM's frozen Vision Transformer (ViT) encoder via Conv-LoRA, where image features are processed by a sparsely gated mixture of multi-scale convolutional experts. This design compensates for the limited translation-equivariant inductive bias of vanilla ViTs while fine-tuning fewer than 5M parameters (less than 1% of the full model). To remove the need for user prompts, we further introduce a lightweight self-prompting box regressor that is jointly optimized with the adapter using scheduled sampling and predicts tumor locations directly from image embeddings. On the BRISC2025 benchmark, BaT-SAM achieves a weighted mean IoU of 0.832 while outperforming state-of-the-art benchmarks. It further demonstrates strong cross-dataset generalization, achieving Dice scores of 0.702 on Figshare and 0.784 on the LGG MRI Segmentation dataset.

Chinese Translation

从磁共振成像（MRI）中准确分割脑肿瘤对于诊断和治疗计划至关重要；然而，手动勾画既费时又耗力。现有方法要么是特定任务的，泛化能力有限，要么在推理过程中依赖用户提供的提示。我们提出了BaT-SAM，这是一种参数高效的适应框架，能够通过Segment Anything Model（SAM）实现完全自动的脑肿瘤分割。BaT-SAM通过Conv-LoRA将局部归纳偏置注入到SAM的冻结视觉变换器（ViT）编码器中，其中图像特征由稀疏门控的多尺度卷积专家混合处理。该设计弥补了普通ViT有限的平移等变归纳偏置，同时微调的参数少于500万（不到完整模型的1%）。为了消除对用户提示的需求，我们进一步引入了一种轻量级自提示框回归器，该回归器与适配器共同优化，使用调度采样直接从图像嵌入预测肿瘤位置。在BRISC2025基准测试中，BaT-SAM实现了0.832的加权平均IoU，同时超越了最先进的基准。它还展示了强大的跨数据集泛化能力，在Figshare上获得了0.702的Dice分数，在LGG MRI分割数据集上获得了0.784的Dice分数。

View on arXiv Download PDF AI Translation

cs.CV / 92 / 2610.15341

StimFlow: Synthesizing Brain Activity from Images via Stimulus-conditioned Flow Matching

StimFlow：通过刺激条件流匹配从图像合成脑活动

Van Hung, Bui, Vu, Dang Thanh, Vo, Hoang-Son, Huy, Mai Duc Minh, Dung, Mai Tien, Kim, Soo-Hyung

Abstract

The inherent variability of fMRI responses across trials encourages a generative view of the stimulus-to-response mapping. Prior generative encoders often operate in compressed latent spaces and condition predominantly on semantic features, which provide limited low-level cues for early visual cortex. We propose StimFlow, which synthesizes fMRI directly in native voxel space using conditional flow matching with a linear path. This design aligns training and inference distributions, enabling competitive performance with a single sample and reducing reliance on averaging. To match the visual hierarchy, StimFlow combines structural DINOv2 features with semantic CLIP features, which prove complementary along the cortical processing stream. With a single sample, StimFlow achieves a per-voxel Pearson correlation of 0.371, matching baselines that average five draws. Generation in voxel space allows direct ROI analysis and concept localization without a decoder. Furthermore, it recovers over half of full-data performance from just one hour of scans, supporting efficient cross-subject adaptation.

Chinese Translation

功能性磁共振成像（fMRI）在不同试验中的反应固有的变异性促使我们采用生成视角来理解刺激与反应之间的映射。以往的生成编码器通常在压缩的潜在空间中操作，并主要依赖语义特征，这为早期视觉皮层提供的低级线索有限。我们提出了StimFlow，它通过条件流匹配和线性路径直接在原始体素空间中合成fMRI。这一设计使训练和推断分布对齐，能够在单个样本下实现竞争性的性能，并减少对平均化的依赖。为了匹配视觉层次结构，StimFlow将结构性DINOv2特征与语义CLIP特征相结合，这在皮层处理流中被证明是互补的。在单个样本下，StimFlow实现了每个体素的Pearson相关系数为0.371，匹配了平均五次抽样的基线。体素空间中的生成允许直接进行感兴趣区域（ROI）分析和概念定位，而无需解码器。此外，它在仅一小时的扫描中恢复了超过一半的全数据性能，支持高效的跨受试者适应。

View on arXiv Download PDF AI Translation

cs.CV / 93 / 2610.15343

Across Viewpoints and Time: Trajectory-Generalizable Physical Attacks on World Models and Video Understanding Models

跨视角与时间：对世界模型和视频理解模型的轨迹通用物理攻击

Xia, Song, Ding, Meiwen, Kong, Chenqi, Yang, Wenhan, Jiang, Xudong

Abstract

World models and video understanding models are increasingly deployed in dynamic visual environments, yet their robustness to physical adversarial threats remains largely underexplored. We study whether a single physical adversarial texture can mislead the victim model across different real-world scenarios where observing viewpoints vary along uncontrolled object--camera trajectories. We identify that heterogeneous video semantics and trajectories induce poorly aligned gradients for optimizing the adversarial texture. To address that, we propose the Trajectory-aware adversarial Alignment and Refinement Attack (TARA), which incorporates trajectory-aware optimization to capture the evolving geometric variations of a printed texture under real-world scenarios. To improve cross-instance optimization compatibility, TARA incorporates trajectory-aware contrastive alignment to encourage attacked representations from diverse video--trajectory instances to approach a shared adversarial region and then performs trajectory-aware task refinement to translate the aligned representation into task-level prediction errors. For reproducible evaluation, we develop a geometry-grounded tracking pipeline that provides physically plausible adversarial regions with temporally consistent trajectories on EPIC-KITCHENS-100 and Something-Something V2. We further collect real-world print-and-recapture videos to validate attack effectiveness under physical deployment conditions. Together, the annotated benchmarks and real-world captured data form a reproducible testbed for future evaluation. Notably, our results reveal that a single printable adversarial texture can persistently disrupt world models and video understanding models under physical conditions, highlighting the fundamental security challenge.

Chinese Translation

世界模型和视频理解模型在动态视觉环境中的应用日益增多，但它们对物理对抗威胁的鲁棒性仍然未得到充分探索。我们研究了单一物理对抗纹理是否能够在不同的真实场景中误导受害模型，这些场景中观察视角沿着不受控制的物体-相机轨迹变化。我们发现异构视频语义和轨迹导致优化对抗纹理时梯度对齐不佳。为了解决这一问题，我们提出了轨迹感知对抗对齐与精炼攻击（Trajectory-aware adversarial Alignment and Refinement Attack, TARA），该方法结合轨迹感知优化，以捕捉在真实场景下打印纹理的几何变化。为了提高跨实例优化的兼容性，TARA引入轨迹感知对比对齐，鼓励来自不同视频-轨迹实例的攻击表示接近共享的对抗区域，然后进行轨迹感知任务精炼，将对齐的表示转换为任务级预测误差。为了实现可重复的评估，我们开发了一个基于几何的跟踪管道，提供在EPIC-KITCHENS-100和Something-Something V2上具有时间一致轨迹的物理可信对抗区域。我们进一步收集了真实世界的打印与重捕视频，以验证在物理部署条件下攻击的有效性。综上所述，注释基准和真实捕获数据构成了未来评估的可重复测试平台。值得注意的是，我们的结果揭示了单一可打印的对抗纹理在物理条件下可以持续干扰世界模型和视频理解模型，突显了基本的安全挑战。

View on arXiv Download PDF AI Translation

cs.CV / 94 / 2610.15354

Video-Based Sexual Harassment Detection Using Fine-Tuned Video Transformers

基于视频的性骚扰检测：使用微调的视频变换器

Mahmud, Jubayer- Al-, Islam, Md Shamimul, Hossain, Md Nur

Abstract

Sexual harassment represents a pervasive societal challenge, with a substantial proportion of incidents remaining unreported due to insufficient evidence and delayed intervention mechanisms. Smart surveillance systems capable of detecting such incidents in real-time could assist law enforcement and serve as a deterrent. However, video-based harassment detection poses significant challenges, including variations in clothing, illumina- tion, partial occlusions, low resolution, and diverse camera view- points. Despite its social importance, this problem has received limited research attention, primarily due to the scarcity of large- scale annotated video datasets. To address this gap, we extend the Sexual Harassment Video (SHV) dataset to 600 short video clips sourced from YouTube and annotated into harassment and non-harassment categories using consistent labeling guidelines. We fine-tune three state-of-the-art video transformer models, VideoMAE, ViViT, and TimeSformer, pretrained on Kinetics- 400, for binary harassment classification. Experimental results show that all three models achieve competitive classification performance, with TimeSformer attaining the highest test accu- racy of 92.27%, outperforming VideoMAE 91.71% and ViViT 91.16% on the extended SHV test set. Our analysis reveals that performance is notably influenced by dataset size, video quality, frame sampling strategy, and model architecture. This work presents a comprehensive evaluation of modern video transformers for this sensitive real-world task.

Chinese Translation

性骚扰是一个普遍存在的社会挑战，许多事件因证据不足和干预机制延迟而未被报告。能够实时检测此类事件的智能监控系统可以协助执法并起到威慑作用。然而，基于视频的骚扰检测面临重大挑战，包括服装变化、光照条件、部分遮挡、低分辨率和多样的摄像机视角。尽管这一问题具有重要的社会意义，但由于缺乏大规模标注视频数据集，相关研究相对有限。为了解决这一问题，我们将性骚扰视频（Sexual Harassment Video, SHV）数据集扩展至600个来自YouTube的短视频片段，并使用一致的标注指南将其标注为骚扰和非骚扰类别。我们对三种最先进的视频变换器模型——VideoMAE、ViViT和TimeSformer进行了微调，这些模型是在Kinetics-400上预训练的，用于二分类骚扰检测。实验结果表明，所有三种模型都达到了竞争力的分类性能，其中TimeSformer在扩展的SHV测试集上获得了最高的测试准确率92.27%，优于VideoMAE的91.71%和ViViT的91.16%。我们的分析显示，性能显著受到数据集规模、视频质量、帧采样策略和模型架构的影响。本研究对现代视频变换器在这一敏感现实任务中的应用进行了全面评估。

View on arXiv Download PDF AI Translation

cs.CV / 95 / 2610.15357

UniForge3D: Joint 3D Reconstruction and Generation in a Shared Anchor-Guided Space

UniForge3D：在共享锚点引导空间中的联合3D重建与生成

Chen, Wenjue, Chen, Wenyue, Zheng, Heliang, Jia, Rongfei, Liu, Yuan, Li, Meng

Abstract

Generating complete 3D assets from sparse, unposed images requires preserving observed geometry while completing unseen regions. Joint reconstruction and generation offers a way to address both goals: reconstruction recovers the geometry observed in the input images, while generation contributes learned shape priors. However, combining camera-relative reconstruction with canonical-space generation introduces a coordinate mismatch that makes direct geometric interaction more difficult. Moreover, conditioning each 3D location on all views without accounting for their relevance may mix useful geometric evidence with unrelated image content, hindering local detail recovery. We present UniForge3D, which jointly models reconstruction and generation in a shared anchor-guided space, resolving their coordinate mismatch. This space follows the anchor view's azimuth while keeping the object upright. The two branches exchange features bidirectionally to predict per-view point maps and a complete sparse voxel structure. We further condition point-map prediction on the generated structure to establish pixel-to-voxel correspondences and select the most informative input view for each voxel. Features from the selected views then guide detailed shape generation, helping preserve observed geometric details. Our method achieves state-of-the-art appearance and geometry results on Toys4K and GSO, generating complete 3D assets with improved fidelity. Project webpage: https://xishuxishu.github.io/uniforge3d-project/

Chinese Translation

从稀疏、未定姿态的图像生成完整的3D资产需要在完成未观察区域的同时保留观察到的几何形状。联合重建与生成提供了一种解决这两个目标的方法：重建恢复输入图像中观察到的几何形状，而生成则贡献学习到的形状先验。然而，将相机相对重建与典范空间生成结合起来会引入坐标不匹配，使得直接的几何交互变得更加困难。此外，在没有考虑相关性的情况下将每个3D位置与所有视图进行条件化，可能会将有用的几何证据与无关的图像内容混合，从而妨碍局部细节的恢复。我们提出了UniForge3D，它在共享的锚点引导空间中联合建模重建与生成，解决了它们的坐标不匹配问题。该空间遵循锚点视图的方位角，同时保持物体直立。两个分支双向交换特征，以预测每视图的点图和完整的稀疏体素结构。我们进一步将点图预测条件化于生成的结构，以建立像素与体素之间的对应关系，并为每个体素选择最具信息量的输入视图。然后，来自所选视图的特征引导详细的形状生成，帮助保留观察到的几何细节。我们的方法在Toys4K和GSO上实现了最先进的外观和几何结果，生成了具有更高保真度的完整3D资产。项目网页：https://xishuxishu.github.io/uniforge3d-project/

View on arXiv Download PDF AI Translation

cs.CV / 96 / 2610.15389

VideoEvoHarness: Self Evolving Agent Harness from Unlabeled Videos for Video Understanding

VideoEvoHarness：从未标记视频中自我演化的代理工具以实现视频理解

Xu, Weixin, Liu, Yifan, Wang, Yi, Rong, Mingfei, Yang, Zhenyu, Zhong, Zhisheng, Qian, Shengsheng, Xu, Changsheng

Abstract

Video understanding agents rely on a harness to acquire evidence, use tools, and reason, but improving it requires substantial effort. Recent self evolution methods seek to optimize harnesses through execution feedback to reduce human intervention. However, existing methods often rely on annotated video questions and answers or focus on updating model parameters, leaving harness evolution from unlabeled videos underexplored. VideoEvoHarness jointly evolves a Questioner and a Solver harness from unlabeled videos with frozen model parameters. The Questioner constructs questions and computes reference answers directly from video memory records. Question quality assessments and Solver execution trajectories guide question generation policy revisions. The Solver supports evolving skills, strategies, hyperparameters, and tools using execution feedback. Executable strategies define workflows in code, reducing repeated interpretation of natural language instructions. Each round, candidate harnesses are compared on identical validation questions, with validated improvements retained for later use. Experiments on benchmarks and a dedicated benchmark question set, using videos disjoint from evolution videos, show higher overall accuracy than initially and transfer gains across backbones. Overall accuracy on this set also improves progressively across rounds. This supports our foundational framework for harness evolution from unlabeled videos, opening a promising direction for self evolving video agents. Code is available at https://github.com/Wei-Xin-Xu/VideoEvoHarness.

Chinese Translation

视频理解代理依赖于一个工具来获取证据、使用工具和进行推理，但改进这一工具需要大量的努力。最近的自我演化方法试图通过执行反馈来优化工具，以减少人工干预。然而，现有方法通常依赖于带注释的视频问题和答案，或者专注于更新模型参数，导致从未标记视频中演化工具的研究尚未得到充分探索。VideoEvoHarness 通过冻结模型参数，从未标记视频中共同演化一个提问者和一个解决者工具。提问者直接从视频记忆记录中构建问题并计算参考答案。问题质量评估和解决者执行轨迹指导问题生成策略的修订。解决者支持利用执行反馈演化技能、策略、超参数和工具。可执行策略在代码中定义工作流程，减少对自然语言指令的重复解释。在每一轮中，候选工具在相同的验证问题上进行比较，经过验证的改进将保留以供后续使用。在基准测试和一个专门的基准问题集上的实验，使用与演化视频不重叠的视频，显示出整体准确性高于初始水平，并在不同骨干网络之间实现转移增益。该数据集的整体准确性也在各轮中逐步提高。这支持了我们从未标记视频中演化工具的基础框架，为自我演化视频代理开辟了一个有前景的方向。代码可在 https://github.com/Wei-Xin-Xu/VideoEvoHarness 获取。

View on arXiv Download PDF AI Translation

cs.CV / 97 / 2610.15414

OmniRSI: Taming Latent Diffusion for Unified Multimodal Remote Sensing Image Generation

OmniRSI：驯化潜在扩散以实现统一的多模态遥感图像生成

Wang, Hebaixu, Zhang, Jing, Guo, Haonan, Wang, Di, Ma, Jiayi, Du, Bo

Abstract

Multimodal remote sensing image generation emerges as a promising paradigm for recovering complete and complementary observations from heterogeneous and partially observed remote sensing data. However, existing methods predominantly depend on sensor-specific fine-tuning of natural image generation models, limiting unified multimodal generation capacity due to the absence of shared representations and robust conditioning mechanisms. To address these challenges, we propose OmniRSI, a unified multimodal remote sensing image generation model. Specifically, we develop a remote sensing-specific variational autoencoder that captures multimodal priors and promotes a locally smooth latent space for unified representation and robust reconstruction across diverse observed data. Building upon that, we fully adapt Stable Diffusion 3.5 with a ControlNet branch to support unified multimodal generation and translation conditioned on textual and visual cues. Besides, we devise a training-free cyclic latent alignment mechanism that iteratively enriches incomplete visual conditions with generated multimodal information, progressively improving guidance reliability while preserving observed measurements. We also construct RSI-MM, a large-scale dataset comprising 11.5M multi-resolution images with text annotations and partially paired data for training and evaluation. Extensive experiments verify that OmniRSI achieves superior performance in multimodal reconstruction, generation and translation.

Chinese Translation

多模态遥感图像生成作为一种有前景的范式，能够从异构和部分观测的遥感数据中恢复完整且互补的观测。然而，现有方法主要依赖于针对特定传感器的自然图像生成模型的微调，这限制了统一多模态生成能力，因为缺乏共享表示和稳健的条件机制。为了解决这些挑战，我们提出了OmniRSI，一个统一的多模态遥感图像生成模型。具体而言，我们开发了一种遥感特定的变分自编码器，捕捉多模态先验，并促进统一表示和在多样化观测数据中稳健重建的局部平滑潜在空间。在此基础上，我们完全适配了Stable Diffusion 3.5，并增加了ControlNet分支，以支持基于文本和视觉线索的统一多模态生成和翻译。此外，我们设计了一种无训练的循环潜在对齐机制，迭代丰富不完整的视觉条件，结合生成的多模态信息，逐步提高引导的可靠性，同时保留观测测量。我们还构建了RSI-MM，一个包含1150万张多分辨率图像及文本注释和部分配对数据的大规模数据集，用于训练和评估。大量实验证实，OmniRSI在多模态重建、生成和翻译方面表现优越。

View on arXiv Download PDF AI Translation

cs.CV / 98 / 2610.15431

Harnessing SPAD Noise: Test-Time Gated Attention for Indoor Depth Estimation

利用SPAD噪声：用于室内深度估计的测试时门控注意力

Gao, Yiquan

Abstract

Monocular depth estimation (MDE) has emerged as a cost-effective alternative to hardware-dependent 3D sensing modalities, yet predicting precise metric depth remains a persistent challenge due to intrinsic scale ambiguity, edge bleeding, and limited zero-shot generalization in unseen environments. Recent multi-modal approaches incorporate 1D/3D Single-Photon Avalanche Diode (SPAD) data to resolve these ambiguities, but they mainly treat SPAD noise as a harmful artifact required for denoising or leave the noise ignored. In this paper, we present an essentially different insight: native SPAD noise is an informative prior that implicitly encodes deterministic environmental characteristics. Driven by this paradigm shift, we introduce a novel, lightweight Test-Time Gated Attention framework by Harnessing SPAD Noise, therefore dubbed HSN. Tailored for indoor scenarios, HSN leverages the synergistic combination of 1D SPAD histograms and Gate Adapters (GAs). Through three complementary optimization phases, it generates channel-wise gated attention vectors to dynamically adjust encoder feature maps at inference time, effectively resolving ambiguous scale and boundary challenges without the heavy computational overhead or dense 3D modalities that other TTA methods suffer from. Extensive experiments on challenging real-world indoor benchmarks show that HSN achieves state-of-the-art metric depth accuracy in in-domain settings, while exhibiting even greater potential under zero-shot generalization to unfamiliar environments. Furthermore, interpretability analysis reveals that the proposed gated attention mechanism adaptively shifts attention toward deeper semantic layers under intensifying SPAD noise. Code will be made publicly available.

Chinese Translation

单目深度估计（MDE）已成为一种成本效益高的替代方案，取代依赖硬件的3D感知模式，但由于固有的尺度模糊、边缘渗透和在未见环境中的有限零-shot泛化，精确的度量深度预测仍然是一个持续的挑战。近期的多模态方法结合了一维/三维单光子雪崩二极管（SPAD）数据来解决这些模糊性，但它们主要将SPAD噪声视为需要去噪的有害伪影，或忽略噪声。在本文中，我们提出了一个本质上不同的见解：原生SPAD噪声是一种信息丰富的先验，隐含地编码了确定性的环境特征。在这一范式转变的驱动下，我们引入了一种新颖的轻量级测试时门控注意力框架，称为利用SPAD噪声（Harnessing SPAD Noise），简称HSN。HSN专为室内场景量身定制，利用一维SPAD直方图和门适配器（Gate Adapters, GAs）的协同组合。通过三个互补的优化阶段，它生成通道级门控注意力向量，以在推理时动态调整编码器特征图，有效解决模糊的尺度和边界挑战，而无需其他测试时注意力（TTA）方法所面临的高计算开销或密集的3D模态。在具有挑战性的真实室内基准上的广泛实验表明，HSN在领域内设置中实现了最先进的度量深度精度，同时在对不熟悉环境的零-shot泛化中展现出更大的潜力。此外，解释性分析表明，所提出的门控注意力机制在增强的SPAD噪声下自适应地将注意力转向更深的语义层次。代码将公开发布。

View on arXiv Download PDF AI Translation

cs.CV / 99 / 2610.15445

Rethinking Supervision in Document Unwarping: A Self-consistent Flow-free Approach

重新思考文档去扭曲中的监督：一种自洽的无流方法

Liu, Shaokai, Feng, Hao, Zhou, Wengang

Abstract

In recent years, the proliferation of smartphones has led to an upsurge in the digitization of document files via these portable devices. However, images captured by smartphones often suffer from distortions, thereby negatively affecting digital preservation and downstream applications. To address this issue, we introduce DRNet, a novel deep network for document image rectification. Our approach is based on three key designs. Firstly, we exploit the intrinsic geometric consistency inherent in document images to guide the learning process of distortion rectification. Secondly, we design a coarse-to-fine rectification network to leverage the representations derived from the distorted document image, thereby enhancing the rectification result. Thirdly, we propose a unique perspective for supervising the learning of rectification networks, where undistorted document images are employed for supervision, which is free of warping mesh as ground truth in existing methods. Technically, both low-level pixel alignment and high-level semantic alignment jointly contribute to the learning of the mapping relationship between deformed document images and distortion-free ones. We evaluate our method on the challenging DocUNet Benchmark dataset, where it sets a series of state-of-the-art records, demonstrating the superiority of our approach compared to existing learning-based solutions. Additionally, we conduct a comprehensive series of ablation experiments to further validate the effectiveness and merits of our method.

Chinese Translation

近年来，智能手机的普及导致了通过这些便携设备对文档文件进行数字化的激增。然而，智能手机拍摄的图像常常受到扭曲的影响，从而对数字保存和下游应用产生负面影响。为了解决这一问题，我们提出了DRNet，一种新颖的文档图像校正深度网络。我们的方法基于三个关键设计。首先，我们利用文档图像固有的几何一致性来指导扭曲校正的学习过程。其次，我们设计了一个粗到细的校正网络，以利用从扭曲文档图像中提取的表示，从而增强校正结果。第三，我们提出了一种独特的监督校正网络学习的视角，其中使用未扭曲的文档图像进行监督，这种方法不依赖于现有方法中的扭曲网格作为真实值。在技术上，低级像素对齐和高级语义对齐共同促进了变形文档图像与无扭曲图像之间映射关系的学习。我们在具有挑战性的DocUNet基准数据集上评估了我们的方法，并在该数据集上创下了一系列最先进的记录，展示了我们的方法相较于现有基于学习的解决方案的优越性。此外，我们还进行了全面的消融实验，以进一步验证我们方法的有效性和优点。

View on arXiv Download PDF AI Translation

cs.CV / 100 / 2610.15452

EmoTrace-Bench: Diagnosing Emotional Expression in Text-to-Video Generation through Controlled Cues and Intensity

EmoTrace-Bench：通过控制线索和强度诊断文本到视频生成中的情感表达

Wang, Yeheng, Shi, Liudan, Zheng, Yi, Li, Jingke, Liu, Shiyuan, Tao, Xin, Zhang, Yuanxing, Sun, Guoyu

Abstract

Text-to-video (T2V) models are increasingly used to generate videos with targeted emotional expressions, yet existing evaluation methods mainly rely on holistic video-text alignment and provide limited insight into whether failures arise from target-emotion expression, intensity control, or specific affective cues. We introduce EmoTrace-Bench, a fine-grained diagnostic benchmark that controls emotion intensity and selected affective cues in prompts and evaluates generated videos at multiple levels. EmoTrace-Bench contains 448 controlled prompts covering eight emotions, two intensity levels, and diverse cue conditions. Using 12 T2V models, we generate 5,376 videos and evaluate target-emotion presence, perceived intensity, and ten audiovisual dimensions of emotional expression with an automated evaluator calibrated and validated against human annotations on a subset of the data. Results reveal substantial differences across models and prompt designs. Prompts that explicitly describe a character's emotional reaction are followed more reliably than prompts that describe only events or scenes, whereas adding more affective cues or using more explicit cinematic audiovisual language does not consistently improve emotional generation. These results illustrate how aggregate alignment scores may overlook localized differences in target-emotion expression, intensity control, and cue realization. EmoTrace-Bench provides a traceable and interpretable framework for diagnosing emotional video generation without modifying existing generation pipelines, and we release the prompts and the automated evaluation rules.

Chinese Translation

文本到视频（T2V）模型越来越多地用于生成具有特定情感表达的视频，但现有的评估方法主要依赖于整体视频与文本的对齐，提供的洞察有限，无法明确失败是由于目标情感表达、强度控制还是特定情感线索造成的。我们提出了EmoTrace-Bench，这是一个细粒度的诊断基准，控制提示中的情感强度和选定的情感线索，并在多个层面上评估生成的视频。EmoTrace-Bench包含448个控制提示，涵盖八种情感、两个强度水平和多样的线索条件。使用12个T2V模型，我们生成了5,376个视频，并利用一个经过校准和验证的自动评估器评估目标情感的存在、感知强度以及情感表达的十个视听维度，该评估器是基于对数据子集的人类注释进行校准和验证的。结果显示模型和提示设计之间存在显著差异。明确描述角色情感反应的提示比仅描述事件或场景的提示更可靠，而增加更多情感线索或使用更明确的电影视听语言并不总是能一致地改善情感生成。这些结果表明，聚合对齐分数可能忽视目标情感表达、强度控制和线索实现中的局部差异。EmoTrace-Bench提供了一个可追踪和可解释的框架，用于诊断情感视频生成，而无需修改现有的生成流程，我们发布了提示和自动评估规则。

View on arXiv Download PDF AI Translation

cs.CV / 101 / 2610.15467

Semantic Saliency from Multi-Modal Large Language Model Scene Understanding Maps

基于多模态大语言模型场景理解图的语义显著性

Wen, Ziqi, Murlidaran, Shravan, Skaza, Jonathan, Wang, William, Eckstein, Miguel P.

Abstract

Image-computable low-level image saliency has shaped perceptual psychology and computer science, but is limited by its inability to capture human high-level cognitive factors that influence attention and eye movements. Here, we propose a new fully automated method to compute semantic saliency by quantifying each object's contribution to scene comprehension without fitting any parameters to human data: the Automated Scene Understanding Map (AUTO-SUM). AUTO-SUM automatically segments and removes individual objects from a scene, uses multimodal large language models (MLLMs) to describe the original and modified scenes, and measures the semantic similarity between these descriptions using text embeddings. We show that AUTO-SUM can predict which object is most important in a human-measured scene understanding map (H-SUM) estimated using human scene descriptions. We also show that AUTO-SUM can predict the object most fixated by human observers during free viewing and scene description tasks better than a low-level saliency model (GBVS, Itti-Koch), a saliency model trained based on human meaning ratings (Meaning Map), and comparable to DeepGaze, which is trained on human fixation data. We contend that AUTO-SUM can be used as a semantic saliency model that complements lower-level saliency models. Code are available at https://github.com/starsky77/AUTO-SUM-Source-Code .

Chinese Translation

可通过图像计算的低级图像显著性在感知心理学和计算机科学中发挥了重要作用，但由于无法捕捉影响注意力和眼动的人类高级认知因素，其应用受到限制。在此，我们提出了一种全自动的方法，通过量化每个对象对场景理解的贡献来计算语义显著性，而无需将任何参数拟合到人类数据上：自动场景理解图（Automated Scene Understanding Map, AUTO-SUM）。AUTO-SUM 自动分割并移除场景中的个体对象，利用多模态大语言模型（Multimodal Large Language Models, MLLMs）描述原始和修改后的场景，并使用文本嵌入测量这些描述之间的语义相似性。我们展示了 AUTO-SUM 能够预测在基于人类场景描述估算的人类测量场景理解图（Human-measured Scene Understanding Map, H-SUM）中，哪个对象是最重要的。我们还表明，AUTO-SUM 在自由观看和场景描述任务中能够比低级显著性模型（GBVS, Itti-Koch）、基于人类意义评分训练的显著性模型（Meaning Map）以及与基于人类注视数据训练的 DeepGaze 相媲美地更好地预测人类观察者最关注的对象。我们认为，AUTO-SUM 可以作为一种语义显著性模型，补充低级显著性模型。代码可在 https://github.com/starsky77/AUTO-SUM-Source-Code 获得。

View on arXiv Download PDF AI Translation

cs.CV / 102 / 2610.15512

Structured Stateful Relational Scripts with a High-Quality Dataset for Multi-Shot Audio-Visual Video Generation

具有高质量数据集的结构化状态关系脚本用于多镜头音视频生成

Zhao, Chen, Shen, Lehan, Huang, Qilong, Li, Hongyu, Kang, Zhuoliang, Wei, Xiaoming, Yang, Jian, Tai, Ying

Abstract

Coherent multi-shot audio-visual video generation requires maintaining consistent identities while depicting evolving states and coordinating sounds across shot boundaries. Although dense captions describe individual shots in detail, they rarely specify which entities persist, how their states change, or which entities and actions produce sounds. To address this limitation, we propose Structured Stateful Relational Scripts (SSRS), a textual supervision representation that separates persistent entity and voice identities from shot-specific visual and auditory states. SSRS explicitly encodes sound-source bindings and temporal relations, while a global audio-visual event sequence connects local actions and sounds, including events that continue across shot boundaries. We further introduce MSAV-Script-100K, a large-scale dataset of curated multi-shot audio-visual videos paired with detailed SSRS annotations, enabling models to learn identity persistence, state transitions, and event continuity from structured supervision. Experiments demonstrate improvements in cross-shot consistency, adherence to specified state changes, and audio-visual alignment, highlighting the effectiveness of our SSRS and MSAV-Script-100K for coherent multi-shot audio-visual generation. Our proposed dataset is available at \href{https://huggingface.co/datasets/shenlehan/MSAV-Script-100K}{MSAV-Script-100K}.

Chinese Translation

连贯的多镜头音视频生成需要在描绘不断变化的状态时保持一致的身份，并协调跨镜头边界的声音。尽管密集的字幕详细描述了单个镜头，但它们很少指明哪些实体持续存在、它们的状态如何变化，或哪些实体和动作产生声音。为了解决这一局限性，我们提出了结构化状态关系脚本（Structured Stateful Relational Scripts, SSRS），这是一种文本监督表示，旨在将持久实体和声音身份与特定镜头的视觉和听觉状态分离。SSRS 明确编码了声音源绑定和时间关系，而一个全局音视频事件序列则连接了局部动作和声音，包括跨镜头边界持续的事件。我们进一步引入了 MSAV-Script-100K，这是一个大规模的数据集，包含经过策划的多镜头音视频，配有详细的 SSRS 注释，使模型能够从结构化监督中学习身份持久性、状态转换和事件连续性。实验表明，在跨镜头一致性、遵循指定状态变化和音视频对齐方面都有所改善，突显了我们的 SSRS 和 MSAV-Script-100K 在连贯的多镜头音视频生成中的有效性。我们提出的数据集可在 [MSAV-Script-100K](https://huggingface.co/datasets/shenlehan/MSAV-Script-100K) 获取。

View on arXiv Download PDF AI Translation

cs.CV / 103 / 2610.15515

Semantically Steerable Image Tokenization

语义可引导的图像标记化

de Goede, Danilo, Derakhshani, Mohammad Mahdi, Asano, Yuki M., Snoek, Cees G. M.

Abstract

Adaptive image tokenizers compress an image into a variable number of tokens matched to its complexity, and recent work makes this allocation controllable: conditioned on a target reconstruction error, the tokenizer itself decides how many tokens an image needs. However, the fidelity measure guiding this allocation shapes which aspects of the image are preserved under compression. The pixel-wise error used in prior work is blind to the spatial structure that carries an image's meaning, and under heavy compression such tokenizers are prone to abruptly discarding tokens that encode essential content. We introduce GIST (Guided Image Semantic Tokenizer), an adaptive tokenizer that can be steered toward a requested semantic fidelity. Its token allocation is jointly controlled by pixel-level and semantic target errors, with semantic error measured in the feature space of a frozen DINOv3 encoder. This joint control enables semantic protection: with the semantic target fixed at zero, the reconstruction target can be relaxed to discard pixel detail while preserving what the image depicts. In a controlled comparison on ImageNet-100, GIST outperforms its pixel-controlled counterpart KARL at every token count, with up to $30\%$ lower LPIPS and $37\%$ lower DreamSim. On average, lowering its semantic target reduces LPIPS $2\times$ faster per token than lowering the reconstruction target, and up to $5\times$ in the low-token regime. Semantic fidelity is thus an effective control signal for adaptive tokenization: it lets a tokenizer see, and so preserve, what an image depicts.

Chinese Translation

自适应图像标记器将图像压缩为与其复杂性相匹配的可变数量的标记，最近的研究使得这种分配变得可控：在目标重建误差的条件下，标记器本身决定图像需要多少个标记。然而，指导这种分配的保真度度量影响了在压缩过程中保留图像的哪些方面。先前工作的像素级误差对承载图像意义的空间结构视而不见，在重度压缩下，这种标记器容易突然丢弃编码重要内容的标记。我们引入了GIST（引导图像语义标记器），一种可以朝向请求的语义保真度进行引导的自适应标记器。其标记分配由像素级和语义目标误差共同控制，语义误差在冻结的DINOv3编码器的特征空间中进行测量。这种联合控制实现了语义保护：当语义目标固定为零时，重建目标可以放宽，以丢弃像素细节，同时保留图像所描绘的内容。在对ImageNet-100的受控比较中，GIST在每个标记数量下均优于其像素控制的对应物KARL，LPIPS降低了多达30%，DreamSim降低了37%。平均而言，降低其语义目标的LPIPS每个标记减少的速度是降低重建目标的2倍，在低标记范围内甚至高达5倍。因此，语义保真度是自适应标记化的有效控制信号：它使标记器能够看到并保留图像所描绘的内容。

View on arXiv Download PDF AI Translation

cs.CV / 104 / 2610.15533

UMAS3D: Unified Mask-Free 3D Editing with Region-Adaptive Source Preservation

UMAS3D：统一的无掩膜3D编辑与区域自适应源保留

Zeng, Xianlin, Shen, Xiaolong, Quan, Ruijie, Yang, Yi

Abstract

3D editing aims to modify an existing 3D asset according to user intent while preserving geometry and appearance outside the intended edit region. Existing methods often rely on task-specific training, source-3D inversion, or external 3D masks, and lack a unified framework for diverse editing conditions and tasks. We propose UMAS3D, a unified 3D editing framework that is training-free, inversion-free, and free from external 3D masks. UMAS3D converts text, bounding boxes, reference images, and their combinations into common target-view guidance, after which addition, deletion, modification, appearance, and pose editing share the same downstream 3D generation process. Our key insight is to couple online edit-region estimation with region-aware source preservation. Mask-Free Adaptive Region Modeling (MARM) initializes an Edit/Keep prior from source-target visual differences and progressively refines it as the target structure evolves, while Source-Guided Preservation Steering (SGPS) constrains the current Keep-region states to retain source geometry and appearance without suppressing target-driven changes. Experiments show that UMAS3D supports addition, deletion, modification, appearance, and pose editing while effectively preserving unedited source content. These advances make UMAS3D a unified framework for flexible and source-preserving 3D editing across diverse editing conditions.

Chinese Translation

3D编辑旨在根据用户意图修改现有的3D资产，同时保留意图编辑区域之外的几何形状和外观。现有方法通常依赖于特定任务的训练、源3D反演或外部3D掩膜，缺乏适用于多样化编辑条件和任务的统一框架。我们提出了UMAS3D，一个统一的3D编辑框架，具有无训练、无反演和无外部3D掩膜的特点。UMAS3D将文本、边界框、参考图像及其组合转换为通用的目标视图引导，之后的添加、删除、修改、外观和姿态编辑共享相同的下游3D生成过程。我们的关键见解是将在线编辑区域估计与区域感知源保留相结合。无掩膜自适应区域建模（Mask-Free Adaptive Region Modeling, MARM）从源-目标视觉差异初始化编辑/保留先验，并在目标结构演变的过程中逐步细化，而源引导保留引导（Source-Guided Preservation Steering, SGPS）则约束当前的保留区域状态，以保留源几何形状和外观，同时不抑制目标驱动的变化。实验表明，UMAS3D支持添加、删除、修改、外观和姿态编辑，同时有效保留未编辑的源内容。这些进展使UMAS3D成为一个适用于多样化编辑条件的灵活且源保留的3D编辑统一框架。

View on arXiv Download PDF AI Translation

cs.CV / 105 / 2610.15555

ELECT: Settling Identity Before Rendering for Scalable Story Visualization

ELECT：在渲染前确定身份以实现可扩展的故事可视化

Miraj, Mehrajul Abadin, Hasan, Md. Mahmudul, Farzana, Meherun, Joarder, Aniket, Hasan, Mahmudul, Partha, Abir Chakraborty, Alam, Md. Ahasanul, Rizvee, Redwan Ahmed, Arefin, Md. Fahim, Rahman, Md Mahmudur, Alam, Md. Tanvir, Khan, Md Mosaddek

Abstract

Training-free story visualization without reference images must preserve recurring entities across frames while scaling to long narratives. Existing methods typically maintain consistency by conditioning each frame on other frames of the same story, coupling generation time, VRAM, or both to story length. We introduce ELECT, which instead settles cross-frame identity before final rendering through a language-guided scene-entity mapping and stationary anchors elected from the model's own drafts. This makes the per-frame cost of the final rendering pass independent of story length. Joint attention exceeds 96 GB of VRAM at 32 scenes, whereas ELECT remains at 8.3 s per frame and under 20 GB VRAM from 10 to 1000 scenes. On the Story50 long-story benchmark, ELECT reaches 0.833 average Character Consistency Similarity (aCCS), compared with 0.826 for ten-round refinement, at one seventh of its generation cost, with lower text alignment (0.327 vs. 0.340). At 1000 scenes, to our knowledge, the first evaluation of training-free story visualization at this length, ELECT exceeds ten-round refinement by 0.017 aCCS and the nearest flat-cost baseline by 0.091, while also surpassing refinement in text alignment at one twenty-fourth of its generation time. These results show that long-horizon consistency need not require increasingly expensive inter-frame conditioning as stories grow.

Chinese Translation

无需训练的故事可视化在没有参考图像的情况下必须在帧之间保持重复实体的一致性，同时能够扩展到较长的叙事。现有方法通常通过将每帧的生成条件与同一故事的其他帧相结合来维持一致性，这使得生成时间、显存（VRAM）或两者都与故事长度相关。我们提出了ELECT，它通过语言引导的场景-实体映射和从模型自身草稿中选出的静态锚点，在最终渲染之前确定跨帧身份。这使得最终渲染过程的每帧成本与故事长度无关。在32个场景下，联合注意力超出了96 GB的显存，而ELECT在10到1000个场景中保持在每帧8.3秒和低于20 GB的显存。在Story50长故事基准测试中，ELECT达到了0.833的平均角色一致性相似度（aCCS），相比之下，十轮精炼的结果为0.826，其生成成本仅为前者的七分之一，且文本对齐度较低（0.327对0.340）。在1000个场景中，ELECT在训练无关的故事可视化领域首次进行此长度的评估，超越了十轮精炼0.017 aCCS，并且比最近的平面成本基线高出0.091，同时在文本对齐度上也超过了精炼，生成时间仅为其二十四分之一。这些结果表明，随着故事的增长，长时间一致性并不需要越来越昂贵的帧间条件。

View on arXiv Download PDF AI Translation

cs.CV / 106 / 2610.15562

A Plausible Future Is Not a Causal Future: Can Video Models Generate How the World Unfolds?

合理的未来并非因果未来：视频模型能否生成世界的发展过程？

Li, Jialiang, Xu, Zhengning, Ma, Zongyang, Cui, Shaobo

Abstract

Large-scale video generation models are increasingly viewed as a promising path toward world models, yet their ability to capture causally coherent physical dynamics remains poorly understood. Existing evaluations largely emphasize visual quality, prompt adherence, or broad physical plausibility, leaving open whether a model can faithfully realize a complete causal cascade, in which an initiating event propagates through causally dependent interactions to a terminal outcome. We introduce CausalCascade, a simulator-grounded benchmark centered on 50 event-rich Blender scenes with structured event graphs, graph-grounded diagnostic VQA, three levels of future-event specification, and 2,832 controlled counterfactual variants, and use it to evaluate nine image-to-video models with complementary motion-, event-, and counterfactual-level diagnostics. Across the evaluated models, generated motion is more reliably localized spatially than realized with correct spatiotemporal evolution, while richer future-event specification improves event recovery but leaves physical realization limited. Counterfactual evaluation further reveals weak selective responsiveness to physical interventions, while most audited failures are foundational, arising at scene consistency or event occurrence. Together, these results position causal cascade generation as a process-level test of world modeling and highlight the need to jointly recover the intended future, faithfully realize its physical evolution, selectively respond to physical interventions, and maintain a coherent physical world over time. Code and dataset are available at https://github.com/SJTU-DeepDeltaLab/CausalCascade .

Chinese Translation

大规模视频生成模型越来越被视为构建世界模型的有前景的途径，但它们捕捉因果一致的物理动态的能力仍然不甚清楚。现有的评估主要强调视觉质量、提示遵循或广泛的物理合理性，尚未明确模型是否能够忠实地实现完整的因果链，其中一个起始事件通过因果依赖的交互传播到最终结果。我们引入了CausalCascade，这是一个基于模拟器的基准，集中于50个事件丰富的Blender场景，配有结构化事件图、图基础的诊断视觉问答（VQA）、三个层次的未来事件规范，以及2,832个受控的反事实变体，并利用它来评估九个图像到视频的模型，这些模型具有互补的运动、事件和反事实层面的诊断。在评估的模型中，生成的运动在空间上比正确的时空演变更可靠地被定位，而更丰富的未来事件规范改善了事件恢复，但物理实现仍然有限。反事实评估进一步揭示了对物理干预的选择性响应较弱，而大多数审核失败是基础性的，源于场景一致性或事件发生。综合这些结果，将因果链生成定位为世界建模的过程级测试，并强调需要共同恢复预期的未来，忠实实现其物理演变，选择性响应物理干预，并在时间上保持一致的物理世界。代码和数据集可在 https://github.com/SJTU-DeepDeltaLab/CausalCascade 获取。

View on arXiv Download PDF AI Translation

cs.CV / 107 / 2610.15565

Genesis: Visual Augmentation Primitives for Multimodal Chain-of-Thought Reasoning

Genesis：用于多模态链式思维推理的视觉增强原语

Yang, Fan, Chen, Lifeng, Wang, Yipu, Wang, Han, Chen, Feilong, Ma, Yidong, Chen, Zhiyang, Zhao, Chaoyang, Wu, Lingxiang, Jiang, Hanqi, Chen, Yibo, Tang, Xu, Hu, Yao, Tang, Ming, Jin, Xiaojie, Zheng, Xiaolong, Wang, Jinqiao

Abstract

Recent multimodal large language models achieve promising results on visual reasoning tasks, approaching human-level performance on existing STEM benchmarks. However, these capabilities degrade under complex real-world scenes. We identify that chain-of-thought reasoning remains largely confined to the linguistic space, rarely anchoring each step in visual evidence. The main cause is the absence of such data during training. In this work, we introduce a training data augmentation pipeline, \textbf{Visual Augmentation Primitives} (VAP), which leverages state-of-the-art vision generative models to synthesize diverse photorealistic training samples and corresponding reasoning text. VAP introduces some basic operators and applies them to a seed image and gets a variant. It also constructs coherent multimodal reasoning chains throughout the generation process. Such training data force models to perceive visual content and perform logical reasoning according to them at every step. We further introduce multimodal chain-of-thought rubric-based reinforcement learning to enhance this. It rewards not only text reasoning but also visual reasoning by scoring intermediate reasoning steps for logical validity, completeness, and consistency. Moreover, we construct the UNREAL benchmark to systematically evaluate visual reasoning on hyperrealistic scenes across controllable difficulty levels. Experiments show that a 4B model built on this framework surpasses 8B baselines, with performance improving consistently as the model scales to 32B, validating the scaling law for visual reasoning.

Chinese Translation

最近的多模态大型语言模型在视觉推理任务上取得了令人鼓舞的结果，接近于现有STEM基准测试的人类水平表现。然而，这些能力在复杂的现实场景中会下降。我们发现，链式思维推理在很大程度上仍然局限于语言空间，鲜有将每一步与视觉证据相结合。其主要原因是在训练过程中缺乏此类数据。在本研究中，我们引入了一种训练数据增强管道，称为 extbf{视觉增强原语}（Visual Augmentation Primitives, VAP），该管道利用最先进的视觉生成模型合成多样的照片级真实训练样本及相应的推理文本。VAP引入了一些基本操作符，并将其应用于种子图像以获得变体。它还在生成过程中构建连贯的多模态推理链。这种训练数据迫使模型在每一步都感知视觉内容并根据这些内容进行逻辑推理。我们进一步引入基于多模态链式思维标准的强化学习来增强这一过程。该方法不仅奖励文本推理，还通过对中间推理步骤的逻辑有效性、完整性和一致性进行评分来奖励视觉推理。此外，我们构建了UNREAL基准，以系统性地评估在可控难度水平下的超现实场景中的视觉推理。实验表明，基于该框架构建的4B模型超越了8B基线，随着模型规模扩大到32B，性能持续提升，验证了视觉推理的规模法则。

View on arXiv Download PDF AI Translation

cs.CV / 108 / 2610.15580

Ellipsoids from Tangents

切线生成的椭球体

Dahan, Efrat, Werman, Michael

Abstract

This paper presents an algorithm for detecting and fitting multiple ellipsoids in 2D, 3D and higher-dimensional data. The representation is based on tangents to the ellipsoid together with their contact points, rather than on points alone. A tangent contributes d independent equations where a point contributes one, so each hypothesis needs fewer samples, which in turn reduces the number of iterations required to recover every ellipsoid. The problem is posed as sets of linear equations whose solutions -- the ellipsoid parameters -- are recovered robustly by clustering in parameter space. The algorithm works in all dimensions and outperforms the state of the art in 2D. The approach applies in general to the robust solution of linear equations, and so has utility beyond ellipsoid fitting.

Chinese Translation

本文提出了一种在二维、三维及更高维数据中检测和拟合多个椭球体的算法。该表示方法基于椭球体的切线及其接触点，而不仅仅依赖于点。一个切线提供 d 个独立方程，而一个点仅提供一个方程，因此每个假设所需的样本更少，从而减少了恢复每个椭球体所需的迭代次数。该问题被表述为线性方程组，其解——椭球体参数——通过在参数空间中的聚类方法稳健地恢复。该算法适用于所有维度，并在二维中优于现有技术。该方法一般适用于线性方程的稳健解，因此在椭球体拟合之外也具有实用性。

View on arXiv Download PDF AI Translation

cs.CV / 109 / 2610.15595

FireNav: Decision-Oriented Reconstruction Evidence Routing for Vision-Language Navigation in Indoor Fire Environments

FireNav：面向决策的重建证据路由框架用于室内火灾环境中的视觉-语言导航

Ding, Hongbo, Chen, Xiang, Dong, Jiangxin, Pan, Jinshan

Abstract

In real-world indoor fire environments, visual degradation severely impairs instruction grounding, waypoint prediction, and long-horizon exploration in vision-language navigation (VLN). Although image restoration offers a natural means to mitigate such visual degradation, existing methods primarily optimize reconstruction fidelity, which does not necessarily translate into better navigation decisions and may even corrupt task-critical cues when applied indiscriminately. In this paper, we present FireNav, a decision-oriented reconstruction evidence routing framework that aligns visual restoration with navigation utility in VLN. To advance this task, we first contribute FireVLN-Bench, a paired indoor fire navigation benchmark with clean references preserving task semantics, safety supervision, and matched settings for continuous navigation and robot execution with simulated dynamics. Our FireNav retains native visual tokens and treats latent reconstruction features as auxiliary evidence, where a lightweight router leverages visual features, task context, and prior execution states to determine whether such evidence should be admitted and how strongly it should contribute to the visual tokens. To learn when reconstruction is beneficial, we further introduce decision utility calibration, which compares action prediction losses with and without candidate evidence under the same state, thereby supervising evidence selection according to its contribution to waypoint, heading, and arrival prediction. Experiments demonstrate that FireNav consistently outperforms state-of-the-art baselines in both navigation success and path efficiency across evaluation settings.

Chinese Translation

在现实世界的室内火灾环境中，视觉退化严重影响了视觉-语言导航（VLN）中的指令基础、路径点预测和长时间探索。尽管图像恢复提供了一种自然的手段来缓解这种视觉退化，但现有方法主要优化重建的保真度，这并不一定能转化为更好的导航决策，甚至在不加区分地应用时可能会破坏任务关键线索。在本文中，我们提出了FireNav，一个面向决策的重建证据路由框架，将视觉恢复与VLN中的导航效用对齐。为推进这一任务，我们首先贡献了FireVLN-Bench，这是一个配对的室内火灾导航基准，具有保留任务语义的干净参考、安全监督和匹配的设置，以便于连续导航和机器人执行的模拟动态。我们的FireNav保留了原生视觉标记，并将潜在的重建特征视为辅助证据，其中一个轻量级路由器利用视觉特征、任务上下文和先前执行状态来确定是否应接纳这些证据以及它们对视觉标记的贡献强度。为了学习何时重建是有利的，我们进一步引入了决策效用校准，该方法在相同状态下比较有无候选证据的行动预测损失，从而根据证据对路径点、航向和到达预测的贡献来监督证据选择。实验表明，FireNav在导航成功率和路径效率方面在各评估设置中始终优于最先进的基线。

View on arXiv Download PDF AI Translation

cs.CV / 110 / 2610.15604

Squeeze-Aware-YOLO: Early Channel Recalibration and Multi-Scale Fusion for Real-Time Traffic Sign Detection

压缩感知YOLO：实时交通标志检测的早期通道重校准与多尺度融合

Maruf, Mizbaul Haque, Sadaf, Syed Samin, Bakht, Syed Tamzid, Kabir, Md. Hasanul, Ahmed, Sabbir

Abstract

Traffic sign detection plays a vital role in intelligent transportation systems, supporting the safe operation of advanced driver assistance systems and autonomous vehicles. Despite significant progress in real-time object detection, accurately identifying traffic signs under challenging conditions remains difficult. Variations in lighting, adverse weather, motion blur, and partial occlusion often reduce detection performance, especially for small or visually degraded signs where important spatial information can be lost. To address these challenges, we propose Squeeze-Aware-YOLO, a lightweight traffic sign detection framework that introduces a multi-branch bottleneck module combining convolutional kernels of different sizes with a Squeeze-and-Excitation mechanism to improve feature extraction and adaptive channel weighting. In addition, we integrate the Multi-Branch Auxiliary Feature Pyramid Network and Re-parameterized Heterogeneous Multi-Scale modules to strengthen multi-scale feature representation and feature fusion while preserving real-time inference efficiency for practical deployment. The proposed approach is evaluated on the Bangladeshi Traffic Sign Detection Dataset, the CSUST Chinese Traffic Sign Detection Benchmark 2021, and the German Traffic Sign Detection Benchmark, achieving [email protected] scores of 74.7%, 79.4%, and 84.0%, respectively, while consistently outperforming recent state-of-the-art YOLO-based traffic sign detectors. These findings demonstrate that combining multi-scale receptive fields, channel attention, and enhanced feature fusion enables more accurate and robust traffic sign detection while maintaining the computational efficiency required for real-world deployment.

Chinese Translation

交通标志检测在智能交通系统中发挥着至关重要的作用，支持先进驾驶辅助系统和自动驾驶车辆的安全运行。尽管实时目标检测取得了显著进展，但在复杂条件下准确识别交通标志仍然困难。光照变化、恶劣天气、运动模糊和部分遮挡常常降低检测性能，特别是对于小型或视觉退化的标志，重要的空间信息可能会丢失。为了解决这些挑战，我们提出了压缩感知YOLO（Squeeze-Aware-YOLO），这是一种轻量级的交通标志检测框架，采用多分支瓶颈模块，将不同大小的卷积核与压缩与激励（Squeeze-and-Excitation）机制相结合，以改善特征提取和自适应通道加权。此外，我们集成了多分支辅助特征金字塔网络（Multi-Branch Auxiliary Feature Pyramid Network）和重新参数化异构多尺度模块（Re-parameterized Heterogeneous Multi-Scale），以增强多尺度特征表示和特征融合，同时保持实际部署所需的实时推理效率。所提出的方法在孟加拉国交通标志检测数据集、CSUST中国交通标志检测基准2021和德国交通标志检测基准上进行了评估，分别达到了74.7%、79.4%和84.0%的[email protected]分数，并且始终优于最近的基于YOLO的交通标志检测器。这些发现表明，结合多尺度感受野、通道注意力和增强特征融合能够实现更准确和更稳健的交通标志检测，同时保持实际部署所需的计算效率。

View on arXiv Download PDF AI Translation

cs.CV / 111 / 2610.15612

Lesion-Aware Attention for Mammography Segmentation with Heterogeneous Supervision

关注病灶的注意力机制在异构监督下的乳腺X线摄影分割

Dai, Di, Liu, Bo, Wang, Liwei

Abstract

Accurate lesion segmentation in mammography is important for computer-aided breast cancer diagnosis, but dense pixel-level annotations remain costly and limited. Existing public segmentation datasets usually cover only a small set of lesion types with manually delineated masks, whereas larger mammography datasets often provide only image-level diagnostic labels. This annotation mismatch makes it difficult to train segmentation models that can generalize to rare or weakly annotated lesion categories. To address this problem, we study open-vocabulary mammographic lesion segmentation under a mixed-supervision setting, where pixel-level masks are available for base lesion categories and image-level labels are available for a broader set of categories. We propose a two-branch framework that jointly learns from segmentation data and classification data. The segmentation branch learns lesion boundaries from annotated masks, while the classification branch introduces additional image-level semantic supervision. To better transfer image-level supervision to lesion-relevant regions, we further introduce a Lesion-wise Attentive Classifier that maintains category-specific feature prototypes and generates lesion-aware attention maps for foreground feature pooling. This design reduces the influence of background tissue and encourages more compact lesion representations. Experiments on multiple mammography datasets show that the proposed method consistently improves open-vocabulary segmentation baselines on both base and novel lesion categories, while reducing the gap to fully supervised models that require dense target-domain masks. Qualitative results further indicate that our method produces more complete lesion localization for small, low-contrast, and visually subtle mammographic findings.

Chinese Translation

在乳腺X线摄影中，准确的病灶分割对于计算机辅助乳腺癌诊断至关重要，但密集的像素级标注仍然成本高昂且有限。现有的公共分割数据集通常仅涵盖少量病灶类型，并提供手动勾画的掩膜，而较大的乳腺X线摄影数据集往往仅提供图像级诊断标签。这种标注不匹配使得训练能够推广到稀有或弱标注病灶类别的分割模型变得困难。为了解决这个问题，我们研究了在混合监督设置下的开放词汇乳腺病灶分割，其中基础病灶类别有像素级掩膜可用，而更广泛类别则有图像级标签可用。我们提出了一种双分支框架，联合从分割数据和分类数据中学习。分割分支从标注掩膜中学习病灶边界，而分类分支则引入额外的图像级语义监督。为了更好地将图像级监督转移到与病灶相关的区域，我们进一步引入了一种病灶级注意分类器，该分类器维护类别特定的特征原型，并生成病灶关注的注意力图用于前景特征池化。这种设计减少了背景组织的影响，并鼓励更紧凑的病灶表示。在多个乳腺X线摄影数据集上的实验表明，所提出的方法在基础和新颖病灶类别上均能持续提高开放词汇分割基线，同时缩小与需要密集目标领域掩膜的完全监督模型之间的差距。定性结果进一步表明，我们的方法为小型、低对比度和视觉上微妙的乳腺X线发现提供了更完整的病灶定位。

View on arXiv Download PDF AI Translation

cs.CV / 112 / 2610.15625

1Step-WAM: Your World Action Model but Faster using 1-Step Action Generation

1Step-WAM：更快的世界行动模型，通过一步动作生成

Su, Rundong, Chen, Pengtao, Geng, Haohan, Zhang, Sha, Yuan, Zhihang, Rong, Yicheng, Yu, Chong, Huang, Hao, Chen, Tao

Abstract

World Action Models (WAMs) unify visual world modeling and action generation, yet their real-world deployment is heavily bottlenecked by iterative action denoising and the sequential execution of visual prefilling and action prediction. We demonstrate that the action-denoising trajectories in WAMs are approximately straight, rendering the first-step velocity a strong estimate of the final action direction. Building on this observation, we introduce 1Step-WAM, a framework that transforms multi-step action denoising into one single forward. Specifically, we employ a lightweight calibrator to directly correct the first step velocity, and use bandwidth-constrained action decoding to further reduce errors in the calibrated one-step actions. Compressing action generation into a single step enables it to execute in parallel with visual prefilling, completely breaking the sequential bottleneck of the traditional imagination-then-action paradigm. Evaluations on RoboTwin and LIBERO show that 1Step-WAM largely preserves multi-step task success with only about 13.4M trainable parameters, while achieving inference speedups of $7.44\times$, $5.98\times$, and $24.84\times$ for FastWAM, ImageWAM and Kairos, respectively. Furthermore, real-world robotic deployments using the FastWAM backbone confirm a consistent 7.44$\times$ speedup while maintaining close success rate to the multi-step baseline. Code is avaliable at https://github.com/changdong1687/1Step-WAM.

Chinese Translation

世界行动模型（WAMs）统一了视觉世界建模和动作生成，但其在现实世界中的部署受到迭代动作去噪和视觉预填充与动作预测的顺序执行的严重瓶颈。我们证明了WAMs中的动作去噪轨迹大致是直线的，使得第一步速度成为最终动作方向的强估计。基于这一观察，我们引入了1Step-WAM，一个将多步动作去噪转化为单步前向的框架。具体而言，我们使用轻量级校准器直接修正第一步速度，并采用带宽受限的动作解码进一步减少校准后单步动作的误差。将动作生成压缩为单步使其能够与视觉预填充并行执行，彻底打破了传统的“想象-再行动”范式的顺序瓶颈。在RoboTwin和LIBERO上的评估表明，1Step-WAM在仅约13.4M可训练参数的情况下，基本保持了多步任务的成功率，同时实现了FastWAM、ImageWAM和Kairos的推理速度提升，分别为$7.44 imes$、$5.98 imes$和$24.84 imes$。此外，使用FastWAM骨干的现实世界机器人部署确认了7.44$ imes$的稳定加速，同时保持与多步基线相近的成功率。代码可在https://github.com/changdong1687/1Step-WAM获取。

View on arXiv Download PDF AI Translation

cs.CV / 113 / 2610.15628

Robust Detection of Directional Adversarial Attacks in Deep Neural Networks for Radiological Imaging

深度神经网络在放射影像中对方向性对抗攻击的稳健检测

Malatyński, Tristan, Jaworek-Korjakowska, Joanna

Abstract

Deep learning is now central to radiology, helping detect changes on X-rays, CTs, and MRIs. However, these systems are highly vulnerable to adversarial attacks - small, crafted perturbations that mislead models while appearing unchanged to humans. Such errors risk missed cancers or false positives. Studies show over 90% attack success on medical scans. Almost 30% of US hospitals use AI in imaging in 2023. Thus, an effective method of detecting such attacks and their directionality is crucial to ensure the credibility and reliability of medical systems based on DNNs. We propose a novel detection framework that leverages subsequent attacks using random noise to identify adversarial perturbations. Our approach uses the analysis of variations in the Sigmoid function to distinguish genuine medical images from adversarially manipulated ones. Specifically, we compare prediction differences between clean and attacked images, as well as between different adversarially altered versions, to improve detection accuracy. We evaluated our method on three popular medical datasets including 'Chest X-Ray Images for Classification', 'Retinal Fundus Multi-disease Image Dataset' and 'Brain Tumor Dataset' under various attack scenarios, including random noise. Our framework achieved up to 99.8% ACC in identifying adversarial directional attacks, significantly outperforming existing defense mechanisms. It consistently identified adversarial samples across varying attack strengths while keeping false positives low, showing strong reliability and potential for clinical use. Furthermore, our proposal highlights its potential for real-world deployment.

Chinese Translation

深度学习如今在放射学中占据核心地位，帮助检测X光、CT和MRI上的变化。然而，这些系统对对抗攻击高度脆弱——小的、精心设计的扰动会误导模型，而对人类而言几乎没有变化。这类错误可能导致漏诊癌症或假阳性。研究表明，医疗扫描中攻击成功率超过90%。到2023年，近30%的美国医院在影像学中使用人工智能。因此，开发一种有效的方法来检测此类攻击及其方向性对于确保基于深度神经网络（DNN）的医疗系统的可信性和可靠性至关重要。我们提出了一种新颖的检测框架，利用后续攻击中的随机噪声来识别对抗扰动。我们的方法通过分析Sigmoid函数的变化来区分真实医疗图像与对抗性操控的图像。具体而言，我们比较干净图像与被攻击图像之间的预测差异，以及不同对抗性修改版本之间的差异，以提高检测准确性。我们在三个流行的医疗数据集上评估了我们的方法，包括“胸部X光图像分类”、“视网膜多病种图像数据集”和“脑肿瘤数据集”，在各种攻击场景下进行测试，包括随机噪声。我们的框架在识别对抗性方向攻击方面达到了高达99.8%的准确率，显著优于现有的防御机制。它在不同攻击强度下始终能够识别对抗样本，同时保持低假阳性率，显示出强大的可靠性和临床应用潜力。此外，我们的提案突出了其在现实世界部署中的潜力。

View on arXiv Download PDF AI Translation

cs.CV / 114 / 2610.15645

SurgCast: Action-Conditioned Future Skeletons for Controllable Surgical World Models

SurgCast：用于可控外科世界模型的动作条件未来骨架

Liu, Wanhao, Zhou, Rulin, Shao, Liangjing, Lin, Zhaocheng, Li, Dongyue, Lin, Jinsong, Ma, Guoheng, Yang, Hangheng, Tang, Zhiqing, Chen, Jing, Li, Panshuo, Ren, Hongliang

Abstract

Surgical world models must turn prospective robot actions into visually faithful, structurally consistent instrument motion. We introduce SurgCast, which predicts future instrument skeletons from current structure, robot state, and future actions, then uses the forecast to control video generation. Dual-path injection combines skeleton geometry with derived, instrument-specific motion descriptors through geometric residuals and conditioning modulation. Both paths complement native action conditioning and are retained in a four-step causal student. On SutureBot and SRTH-Porcine-Cholei, the teacher and student improve all seven reported quality and instrument-consistency metrics over their respective direct-action baselines; teacher FVD decreases by 37.6% and 67.3%. Ablations show that future actions improve skeleton prediction and that combining both injection paths performs best across all six metrics under predicted skeletons. These results support using predicted articulated structure and its motion semantics as a control interface for surgical video prediction. Project videos and examples are available at https://surgcast.netlify.app/.

Chinese Translation

外科世界模型必须将预期的机器人动作转化为视觉上真实、结构上一致的器械运动。我们提出了SurgCast，它从当前结构、机器人状态和未来动作中预测未来的器械骨架，然后利用该预测来控制视频生成。双路径注入通过几何残差和条件调制将骨架几何与派生的器械特定运动描述符相结合。这两条路径互补本土动作条件，并在四步因果学生中保留。在SutureBot和SRTH-Porcine-Cholei上，教师和学生在各自的直接动作基线之上提高了所有七个报告的质量和器械一致性指标；教师的FVD分别降低了37.6%和67.3%。消融实验表明，未来动作改善了骨架预测，并且结合两条注入路径在所有六个指标下的预测骨架中表现最佳。这些结果支持使用预测的关节结构及其运动语义作为外科视频预测的控制接口。项目视频和示例可在https://surgcast.netlify.app/获取。

View on arXiv Download PDF AI Translation

cs.CV / 115 / 2610.15660

Sparse All-Layer Connector for Domain Generalised Semantic Segmentation

用于领域泛化语义分割的稀疏全层连接器

Guo, Ruoyu, Lin, Xin Kun, Pagnucco, Maurice, Song, Yang

Abstract

Domain generalised semantic segmentation (DGSS) has recently benefited from combining multiple foundation models, such as vision-language and vision foundation models, to leverage their diverse representations for generalisation. Existing multi-model approaches share a common design choice: cross-model fusion is restricted to depth-aligned layers, implicitly assuming that mutually beneficial information resides at the same depth across models. However, foundation models pretrained under different objectives may develop distinct representation hierarchies, making the optimality of depth-aligned fusion questionable. Moreover, we observe that the foundation models exhibit different depth-wise domain sensitivity. Motivated by these observations, we propose a Sparse All-Layer Connector (SALC) that enables each layer to access information from preceding layers of another model. As the number of accessible layers grows with depth, dense aggregation may hurt generalisation. SALC therefore learns to adaptively select a sparse subset of informative layers. We further introduce a candidate dropout regularisation that strengthens sparsity and encourages SALC to explore diverse layers, leading to more robust selection. Across four foundation model combinations and three DGSS evaluation settings, the proposed design demonstrates stronger generalisation than conventional depth-aligned fusion. The code is available at https://github.com/RuoyuGuo/SALC.

Chinese Translation

领域泛化语义分割（DGSS）最近通过结合多种基础模型（如视觉-语言模型和视觉基础模型）受益，以利用它们多样化的表示进行泛化。现有的多模型方法共享一个共同的设计选择：跨模型融合仅限于深度对齐的层，隐含地假设互惠信息在模型之间的相同深度处存在。然而，在不同目标下预训练的基础模型可能会发展出不同的表示层次结构，这使得深度对齐融合的最优性受到质疑。此外，我们观察到基础模型表现出不同的深度域敏感性。基于这些观察，我们提出了一种稀疏全层连接器（SALC），使每一层能够访问另一个模型前面层的信息。随着可访问层数的增加，密集聚合可能会损害泛化能力。因此，SALC学习自适应地选择一个稀疏的有信息层子集。我们进一步引入了一种候选丢弃正则化，增强稀疏性并鼓励SALC探索多样化的层，从而实现更稳健的选择。在四种基础模型组合和三种DGSS评估设置中，所提出的设计展示了比传统深度对齐融合更强的泛化能力。代码可在 https://github.com/RuoyuGuo/SALC 获取。

View on arXiv Download PDF AI Translation

cs.CV / 116 / 2610.15666

A Persian Vision-Language-Action Dataset and Benchmark for Ambiguity- and Infeasibility-Aware Instruction-to-Plan Grounding

波斯视觉-语言-行动数据集及其在模糊性和不可行性意识的指令到计划基础上的基准测试

Pourasghari, Abdollah, Sadidpour, Saeedeh Sadat, Roustaei, Mohammad

Abstract

Converting natural-language instructions into executable action plans conditioned on a scene image is a core capability for embodied human-robot interaction, but existing benchmarks target English, simulated or video-grounded environments, and largely assume instructions are clear and executable. We introduce the first Persian vision-language-action dataset for this task: 258 instances over 230 real household/office photographs, annotated with reference plans, deliberate ambiguity and infeasibility labels, disambiguating question-answer pairs, and an auto-generated chain-of-thought rationale for training. Using this resource, we ablate retrieval-augmentation, chain-of-thought prompting, and an agentic disambiguation step, evaluated with an LLM-as-judge protocol and a new metric, PGOS-FA, extending prior work (PG2S) with step-order sensitivity, an over-generation penalty, and Persian-adapted embeddings. Paired statistical testing across 36 comparisons shows retrieval-augmentation is a large, robustly significant effect, while finer-grained additions are directionally consistent but not statistically distinguishable from noise at this scale, and the LLM-judge metric has near-zero power where PGOS-FA succeeds. A blind human-correlation study on 79 real outputs shows PGOS-FA tracks human judgment at least as closely as the LLM-judge. Testing generalization beyond the ablation's single backbone, we find that most of the gap between an open-weight and a substantially stronger closed planner is attributable to open-weight model generation rather than to the closed model's capability specifically, and that our best configuration significantly outperforms an independently reproduced external baseline despite a weaker backbone. The dataset, PGOS-FA implementation, and evaluation code will be made available upon reasonable request.

Chinese Translation

将自然语言指令转换为基于场景图像的可执行行动计划是具身人机交互的核心能力，但现有基准主要针对英语、模拟或视频基础环境，并且大多假设指令是清晰且可执行的。我们为此任务引入了首个波斯视觉-语言-行动数据集：包含258个实例，涵盖230张真实的家庭/办公室照片，附有参考计划、故意模糊和不可行性标签、消歧问答对，以及用于训练的自动生成的思维链推理。利用该资源，我们对检索增强、思维链提示和代理消歧步骤进行了消融实验，采用了LLM作为评判者的协议和新指标PGOS-FA，扩展了先前的工作（PG2S），引入了步骤顺序敏感性、过生成惩罚和波斯适配的嵌入。在36个比较中的配对统计测试表明，检索增强具有显著且强大的效果，而更细粒度的添加在此规模下在统计上与噪声无显著区别，且LLM评判指标在PGOS-FA成功的情况下几乎没有效能。对79个真实输出的盲人相关性研究显示，PGOS-FA的表现与人类判断的相关性至少与LLM评判者相当。测试超出消融的单一主干的泛化能力，我们发现开放权重模型生成与强大的封闭规划器之间的大部分差距可归因于开放权重模型生成，而非封闭模型的特定能力，并且我们的最佳配置显著优于独立再现的外部基准，尽管其主干较弱。该数据集、PGOS-FA实现及评估代码将在合理请求下提供。

View on arXiv Download PDF AI Translation

cs.CV / 117 / 2610.15669

ProtoPur: Prototype-Purity Pseudo-Labeling for Weakly Supervised Instance Learning in Pathology

ProtoPur：用于病理学中弱监督实例学习的原型纯度伪标签方法

Shu, Chen, Zhang, Chengyang, Fu, Boyu, Yin, Ting, Zhang, Wenchuan, Chen, Jie, Yi, Yuhao, Bu, Hong

Abstract

Weak bag-level labels are widely available in computational pathology, whereas reliable instance-level annotations for whole-slide images remain expensive and ambiguous. Weakly supervised multiple instance learning (MIL) achieves strong bag-level classification, but instance-level supervision from attention scores, model predictions, negative bags, or spatial refinement may fail to identify class-specific local evidence. We propose ProtoPur, a prototype-purity pseudo-labeling method for weakly supervised instance-level learning in pathology whole-slide images. ProtoPur estimates class-balanced local purity over prototype regions constructed from frozen foundation-model features, and assigns pseudo-labels to high-purity regions strongly associated with a target class. This purity-first design separates class-specific evidence from negative or shared morphology, with coverage used only to ensure sufficient pseudo-label assignment. We provide a PAC-style analysis showing that the risk of an instance classifier trained with ProtoPur pseudo-labels can be decomposed into a standard uniform-convergence term and a ProtoPur-specific pseudo-label noise term, comprising selected-region mismatch and a coverage residual. Experiments on three public whole-slide image datasets with two pathology foundation encoders show that ProtoPur consistently improves instance-level performance across multiple MIL backbones while maintaining competitive bag-level performance. Pseudo-label quality comparisons on two public datasets further demonstrate that prototype-purity selection provides stronger instance supervision. These results suggest that the foundation-model features contain exploitable instance-level supervision signals, and ProtoPur is a way to reveal and use these signals from weak bag labels without requiring dense instance annotations.

Chinese Translation

在计算病理学中，弱包级标签广泛可用，而全切片图像的可靠实例级注释仍然昂贵且模糊。弱监督多实例学习（MIL）能够实现强大的包级分类，但来自注意力分数、模型预测、负包或空间细化的实例级监督可能无法识别特定类别的局部证据。我们提出了ProtoPur，一种用于病理全切片图像的弱监督实例级学习的原型纯度伪标签方法。ProtoPur通过从冻结的基础模型特征构建的原型区域估计类别平衡的局部纯度，并将伪标签分配给与目标类别强相关的高纯度区域。这种以纯度为首的设计将特定类别的证据与负面或共享形态分离，覆盖率仅用于确保足够的伪标签分配。我们提供了一种PAC风格的分析，表明使用ProtoPur伪标签训练的实例分类器的风险可以分解为标准的均匀收敛项和ProtoPur特定的伪标签噪声项，包括所选区域的不匹配和覆盖残差。在三个公共全切片图像数据集上进行的实验表明，ProtoPur在多个MIL基础上始终提高了实例级性能，同时保持了竞争性的包级性能。在两个公共数据集上的伪标签质量比较进一步表明，原型纯度选择提供了更强的实例监督。这些结果表明，基础模型特征包含可利用的实例级监督信号，而ProtoPur是一种从弱包标签中揭示和利用这些信号的方法，无需密集的实例注释。

View on arXiv Download PDF AI Translation

cs.CV / 118 / 2610.15681

Rethinking Human Intent-to-CAD: Parametric CAD Model Generation via Cooperative Multi-Task Alignment and Spatial-Aware Reinforcement Learning

重新思考人类意图到计算机辅助设计（CAD）的转化：通过协作多任务对齐和空间感知强化学习生成参数化CAD模型

Zhang, Qingwang, Li, Jiahao, Zhou, Xiangdong

Abstract

Parametric Computer-Aided-Design (CAD) modeling from human intent remains challenging, particularly during the conceptual design stage, where design goals are expressed through incomplete and unstructured modalities (e.g., hand-drawn sketches and textual descriptions). In this work, we rethink the human intent-to-CAD pipeline and propose a unified method that directly maps multi-level human intents to executable codes, without assuming the prior existence of target CAD models. To support our study, we construct HiCAD, the first large-scale dataset aligning hand-drawn sketches, textual descriptions, and parametric CAD codes. Based on this, we introduce HiCAD, a two-stage framework comprising Cooperative Multi-Task Alignment to bridge the representational gap between heterogeneous inputs, and Spatial-Aware Reinforcement Learning to enforce geometric and topological consistency. Extensive experiments demonstrate that our method significantly outperforms existing baselines across multiple tasks, validating its effectiveness and robustness in transforming heterogeneous human intents into high-fidelity parametric CAD models. Our project page: https://zqwlearning.github.io/HiCAD.

Chinese Translation

从人类意图生成参数化计算机辅助设计（CAD）模型仍然具有挑战性，特别是在概念设计阶段，此时设计目标通过不完整和非结构化的方式表达（例如，手绘草图和文本描述）。在本研究中，我们重新思考人类意图到CAD的流程，并提出一种统一的方法，直接将多层次的人类意图映射到可执行代码，而不假设目标CAD模型的先前存在。为了支持我们的研究，我们构建了HiCAD，这是第一个大规模数据集，将手绘草图、文本描述和参数化CAD代码进行对齐。在此基础上，我们介绍了HiCAD，一个包含协作多任务对齐的两阶段框架，以弥合异构输入之间的表征差距，以及空间感知强化学习，以确保几何和拓扑的一致性。大量实验表明，我们的方法在多个任务上显著优于现有基线，验证了其在将异构人类意图转化为高保真参数化CAD模型方面的有效性和鲁棒性。我们的项目页面：https://zqwlearning.github.io/HiCAD。

View on arXiv Download PDF AI Translation

cs.CV / 119 / 2610.15722

PhySimCode: A Benchmark and Evaluation Method for Physics Video to Code Generation

PhySimCode：物理视频到代码生成的基准与评估方法

Kundu, Souraja, Gupta, Aditya, Julin, Joel, Zhao, Yizhou, Xie, Liuyue, Jeni, Laszlo A.

Abstract

Multimodal large language models (MLLMs) are good at video understanding and code generation, but it remains unclear whether they can recover the physical laws underlying videos, a capability essential for converting videos into executable simulations and for providing structured physical supervision in robotics. Captioning motion video does not imply that a model has inferred its governing dynamics, estimated quantitative parameters, or can regenerate the observed scene from code. Existing video-physics benchmarks evaluate plausibility, property estimation, or future-state prediction, but offer limited insight into whether models can translate visual dynamics into runnable simulation code. We introduce PhySimCode, a benchmark for physics video-to-code generation: given only a raw video, a model must infer the relevant physics, recover parameter values, and generate runnable Python that reproduces the input video. The dataset contains 160.6K procedurally generated samples spanning 162 physics phenomena across analytic SciPy simulations and PyBullet-rendered rigid-body, articulated, fluid, and contact-rich scenes; each sample bundles the simulation code, physical parameters, rendered video, and a structured chain-of-thought covering observations, laws, ODEs, numerical methods, and validation checks. We evaluate ten state-of-the-art closed- and open-source MLLMs along physical-law correctness, simulation equivalence, code executability, and parameter recovery. Models reliably state physics laws (4.27-4.93/5) yet score only 1.51-2.47/5 on inferring them from video, and the best system (Claude Sonnet 4.6) recovers just 7.7% of ground-truth parameters within $\pm 20\%$, exposing a substantial video-to-simulation gap. The 2,430-video evaluation set is available at https://huggingface.co/datasets/SourajaKundu123/PhySimCode and the evaluation code at https://github.com/SourajaKundu/PhySimCode

Chinese Translation

多模态大型语言模型（MLLMs）擅长视频理解和代码生成，但尚不清楚它们是否能够恢复视频背后的物理规律，这一能力对于将视频转换为可执行的模拟以及在机器人技术中提供结构化的物理监督至关重要。对运动视频的描述并不意味着模型已经推断出其支配动力学、估计了定量参数，或能够从代码再生观察到的场景。现有的视频物理基准评估可信度、属性估计或未来状态预测，但对模型是否能够将视觉动态转换为可运行的模拟代码提供的洞见有限。我们引入了PhySimCode，这是一个用于物理视频到代码生成的基准：给定仅一个原始视频，模型必须推断相关的物理规律，恢复参数值，并生成可运行的Python代码以再现输入视频。该数据集包含160.6K个程序生成的样本，涵盖162种物理现象，涉及分析SciPy模拟和PyBullet渲染的刚体、关节、流体和接触丰富场景；每个样本都包含模拟代码、物理参数、渲染视频，以及涵盖观察、规律、常微分方程、数值方法和验证检查的结构化思维链。我们评估了十种最先进的闭源和开源MLLMs，依据物理规律的正确性、模拟等效性、代码可执行性和参数恢复能力。模型可靠地陈述物理规律（4.27-4.93/5），但在从视频推断这些规律时仅得分1.51-2.47/5，而最佳系统（Claude Sonnet 4.6）在$ ext{±}20\%$范围内仅恢复了7.7%的真实参数，暴露了视频到模拟之间的显著差距。2430个视频的评估集可在https://huggingface.co/datasets/SourajaKundu123/PhySimCode获取，评估代码可在https://github.com/SourajaKundu/PhySimCode找到。

View on arXiv Download PDF AI Translation

cs.CV / 120 / 2610.15728

Visually Legible but Adversarial: Watermark-Constrained Attacks on End-to-End VLMs for Document Understanding

可视可读但具对抗性：针对文档理解的端到端视觉语言模型的水印约束攻击

Han, Qingzhe, Dang, Fan, Chen, Shengqi, Sun, Xinyi, Xiong, Haoyu, Li, Lingkun, Zhang, Di

Abstract

Visible watermarks convey document ownership, confidentiality, and status, but can also carry adversarial signals. We introduce Auto-Regional Adversarial Watermarking (AR-AWM), a model-aware white-box attack on end-to-end visual language models (VLMs) for document understanding. AR-AWM constructs a small, spatially distributed watermark-text support and optimizes only its pixels to disrupt visual representations. The attack is untargeted and question-agnostic: generation uses no questions, answers, parsing prompts, language-model gradients, or language-model outputs. Each attacked page is reused across downstream queries. On Qwen2.5-VL-7B, a watermark support covering 7.98% of page pixels under the reporting threshold reduces DocVQA exact match from 0.899 to 0.150 and average normalized Levenshtein similarity (ANLS) from 0.939 to 0.212. Selected qualitative cases retain visually legible source fields. The same support without optimization and with matched random perturbations retains ANLS of 0.932 and 0.933, respectively, identifying adversarial optimization as the primary cause of degradation. Layout-matched optimized rectangles yield ANLS of 0.276, showing an incremental benefit from text-shaped geometry. AR-AWM also degrades all reported text, formula, reading-order, and table metrics on OmniDocBench. Separate white-box optimization on InternVL2.5-8B reduces ANLS from 0.922 to 0.046, while weak cross-model transfer reveals strong encoder specificity. These results identify conventional watermark-shaped supports as an adversarial surface for document VLMs in the evaluated white-box setting.luated white-box setting.

Chinese Translation

可见水印传达文档的所有权、机密性和状态，但也可能携带对抗信号。我们提出了自动区域对抗水印（Auto-Regional Adversarial Watermarking, AR-AWM），这是一种针对端到端视觉语言模型（VLMs）进行文档理解的模型感知白盒攻击。AR-AWM 构建了一个小型、空间分布的水印文本支持，并仅优化其像素以干扰视觉表示。该攻击是无目标的且与问题无关：生成过程中不使用问题、答案、解析提示、语言模型梯度或语言模型输出。每个被攻击的页面在下游查询中被重复使用。在 Qwen2.5-VL-7B 上，覆盖 7.98% 页面像素的水印支持在报告阈值下将 DocVQA 的精确匹配率从 0.899 降低到 0.150，平均归一化 Levenshtein 相似度（ANLS）从 0.939 降低到 0.212。选择的定性案例保留了可视可读的源字段。未优化的相同支持和匹配随机扰动分别保持 ANLS 为 0.932 和 0.933，表明对抗优化是降级的主要原因。与布局匹配的优化矩形产生 ANLS 为 0.276，显示出文本形状几何体的增量效益。AR-AWM 还降低了 OmniDocBench 上所有报告的文本、公式、阅读顺序和表格指标。在 InternVL2.5-8B 上的单独白盒优化将 ANLS 从 0.922 降低到 0.046，而弱跨模型迁移揭示了强编码器特异性。这些结果表明，在评估的白盒设置中，传统的水印形状支持被识别为文档 VLMs 的对抗表面。

View on arXiv Download PDF AI Translation

cs.CV / 121 / 2610.15734

ScenePlacer: Reformulating Scene-Level Object Pose Estimation as Visibility Field Learning

ScenePlacer：将场景级物体姿态估计重构为可见性场学习

Deng, Ken, Liu, Jiarui, Liao, Guibiao, Luo, Kunming, Xiang, Mochu, Yan, Hongyu, Liu, Yebin, Gao, Shenghua, Yu, Yizhou, Tan, Ping

Abstract

We present ScenePlacer, a framework for scene-level pose estimation of objects generated by state-of-the-art native shape generation models, placing generated 3D objects into scenes given a partial geometric observation without modifying the underlying generative pipeline. Existing scene generation approaches either generate objects directly in scene coordinates, compromising their canonical-space priors, or regress poses from complete generated shapes in the shape generation process - the latter being a poorly conditioned problem that is sensitive to scale ambiguity and partial observations. ScenePlacer decouples object generation from pose estimation by reformulating alignment as conditional visibility-field learning. Given an observed partial point cloud from reconstruction and a generated mesh, it predicts a visibility field defined in object space, then estimates the rigid transformation by aligning the partial observation to visible parts on the mesh. We propose a pretrained 3D shape VAE as our backbone, preserving its mesh prior while conditioning it on a separately encoded observation. We further introduce visibility-aware, radius-constrained supervision and synthesize training pairs using only object-level meshes and rendered views, eliminating the need for scene-level pose annotations. On 3D-FUTURE, ScenePlacer consistently outperforms existing placement and pose-refinement baselines under both ground-truth and monocular depth estimation. With generated objects, it achieves a Chamfer distance of 0.0077 and an F-score of 92.40 under estimated depth and improves to 0.0023 and 98.27, respectively, under ground-truth geometry.

Chinese Translation

我们提出了ScenePlacer，这是一个用于场景级物体姿态估计的框架，能够利用最先进的原生形状生成模型生成的物体，将生成的3D物体放置到场景中，前提是给定部分几何观测，而无需修改底层生成管道。现有的场景生成方法要么直接在场景坐标中生成物体，从而妥协其标准空间先验，要么在形状生成过程中从完整生成的形状回归姿态——后者是一个条件较差的问题，对尺度模糊和部分观测非常敏感。ScenePlacer通过将对齐重构为条件可见性场学习，解耦了物体生成与姿态估计。给定从重建中观察到的部分点云和生成的网格，它预测在物体空间中定义的可见性场，然后通过将部分观测与网格上的可见部分对齐来估计刚性变换。我们提出了一种预训练的3D形状变分自编码器（VAE）作为我们的主干，保留其网格先验，同时将其条件化于单独编码的观测上。我们进一步引入了关注可见性的、半径约束的监督，并仅使用物体级网格和渲染视图合成训练对，消除了对场景级姿态注释的需求。在3D-FUTURE数据集上，ScenePlacer在真实值和单目深度估计下始终优于现有的放置和姿态优化基线。在生成的物体上，它在估计深度下达到了0.0077的Chamfer距离和92.40的F-score，而在真实几何下分别提高到0.0023和98.27。

View on arXiv Download PDF AI Translation

cs.CV / 122 / 2610.15742

RSAJEV: Training-Free Abstaining Prototype Memory for Eliminating Silent Hallucination in JEV-Based Remote Sensing Classification

RSAJEV：无训练的弃权原型记忆用于消除基于JEV的遥感分类中的静默幻觉

Chen, Man, Chen, Antai, Xu, Shuai, Gu, Chengjie, Zhou, Huaping, Sun, Kelei, Wu, Tao, Huang, Yongjie

Abstract

Scene classification of remote sensing imagery with multimodal large language models (MLLMs) is formulated as a single-forward-pass decision problem: candidate class names are lettered options in the prompt, a registered answer slot is conditioned on the prompt, and classification is read out by scoring letter tokens with the language-modeling head. We show that although this letter binding is highly effective within its training vocabulary (97.71% OA on UCM), it behaves erratically out of vocabulary: zero-adaptation accuracy across four cross-dataset directions fluctuates between 7.35% and 96.40%, and under out-of-vocabulary weights its confidence fails inversely on open-set evaluation, assigning higher confidence to unseen than to seen classes (AUROC 0.38), a failure we term silent hallucination. We retain the slot representation and replace the readout with RSAJEV, a training-free abstaining prototype memory that performs retrieval-based prototype matching: K reference images per class are cached as prototypes via their slot representations under the same prompt, and classification becomes nearest-neighbor retrieval over the prototypes, refined by a shrinkage-Mahalanobis geometric correction. This readout is mechanistically isomorphic to retrieval-augmented generation (RAG) but operates on discriminative decisions; it is the first retrieval-based readout for MLLM slot representations. The retrieval readout reaches 89.43% on the UCM cross-dataset gate with 16 references per class and zero gradient updates, and 96.90% on AID. Its ID-ness scores attain 0.91-0.99 AUROC on open-set evaluation over two datasets, exceeding symbolic confidence (0.38-0.96) in all four open-set settings, restoring a calibratable abstention channel and supporting class addition/removal at zero extra inference cost via offline rescoring of the cached matrix.

Chinese Translation

遥感影像的场景分类与多模态大语言模型（MLLMs）被表述为一个单次前向传递的决策问题：候选类别名称在提示中作为字母选项，注册的答案槽依赖于提示，分类通过语言建模头对字母标记进行评分来读取。我们展示了尽管这种字母绑定在其训练词汇中非常有效（在UCM上达到97.71%的总体准确率），但在词汇外表现不稳定：在四个跨数据集方向上的零适应准确率波动在7.35%到96.40%之间，并且在词汇外权重下，其信心在开放集评估中反向失败，给未见类别分配的信心高于已见类别（AUROC 0.38），这种失败我们称之为静默幻觉。我们保留槽表示，并用RSAJEV替换读取过程，RSAJEV是一种无训练的弃权原型记忆，执行基于检索的原型匹配：每个类别的K个参考图像通过其在相同提示下的槽表示被缓存为原型，分类变为对原型的最近邻检索，通过收缩-马哈拉诺比斯几何校正进行优化。该读取机制在机制上与检索增强生成（RAG）同构，但在判别决策上操作；这是第一个针对MLLM槽表示的基于检索的读取。该检索读取在UCM跨数据集门上达到89.43%的准确率，使用每个类别16个参考图像且没有梯度更新，在AID上达到96.90%。其ID-ness评分在两个数据集的开放集评估中达到0.91-0.99 AUROC，超越了所有四个开放集设置中的符号信心（0.38-0.96），恢复了可校准的弃权通道，并通过离线重新评分缓存矩阵以零额外推理成本支持类别的添加/删除。

View on arXiv Download PDF AI Translation

cs.CV / 123 / 2610.15745

SpikeOcc: Towards Resilient Multimodal 3D Occupancy Perception via Spiking Experts

SpikeOcc：通过脉冲专家实现更具韧性的多模态三维占用感知

Sun, Yiding, Xu, Zijie, Lin, Hongqiang, Cao, Jiajun, Zhang, Dongxu, Tang, Kai, Zhu, Jihua, Liu, Wenxuan, Yu, Zhaofei, Huang, Tiejun

Abstract

Multimodal 3D occupancy prediction underpins safe autonomous driving through comprehensive scene understanding. However, existing methods have largely pursued higher accuracy through sophisticated fusion, leaving resilience to missing sensors underexplored. In this paper, we investigate spiking neural networks (SNNs) as a means of building more resilient multimodal perception. Still, the discrete nature of spiking neurons makes optimization challenging. To overcome this difficulty, SpikeOcc is introduced with two complementary components, a Spiking Mixture-of-Experts (SMoE) and Sensor Availability Learning (SAL). Specifically, SMoE organizes RGB, LiDAR, and joint fusion into specialized pathways, enabling full-width unimodal representations while retaining the original multimodal features via a bounded residual. Beyond sensor dropout, SAL employs a staged curriculum with explicit routing supervision, progressively improving robustness under severe sensing conditions. Remarkably, SAL also yields notable gains for conventional models, highlighting its broader potential for resilient multimodal learning. Extensive experiments demonstrate that SpikeOcc maintains competitive full-sensor accuracy while markedly raising the performance floor under sensor outages. SpikeOcc illustrates the promise of neuromorphic computing, pointing toward more reliable and energy-efficient autonomous driving.

Chinese Translation

多模态三维占用预测是安全自主驾驶的基础，依赖于全面的场景理解。然而，现有方法主要通过复杂的融合追求更高的准确性，而对缺失传感器的韧性探索不足。本文研究了脉冲神经网络（SNNs）作为构建更具韧性的多模态感知的一种手段。然而，脉冲神经元的离散特性使得优化变得具有挑战性。为了解决这一难题，本文提出了SpikeOcc，包含两个互补组件：脉冲专家混合模型（Spiking Mixture-of-Experts, SMoE）和传感器可用性学习（Sensor Availability Learning, SAL）。具体而言，SMoE将RGB、LiDAR和联合融合组织成专门的路径，能够实现全宽单模态表示，同时通过有界残差保留原始的多模态特征。除了传感器失效，SAL采用分阶段的课程学习和明确的路由监督，逐步提高在严苛感知条件下的鲁棒性。值得注意的是，SAL还为传统模型带来了显著的提升，突显了其在韧性多模态学习中的更广泛潜力。大量实验表明，SpikeOcc在保持竞争性的全传感器准确度的同时，显著提高了在传感器故障情况下的性能底线。SpikeOcc展示了神经形态计算的潜力，指向更可靠和能效更高的自主驾驶。

View on arXiv Download PDF AI Translation

cs.CV / 124 / 2610.15752

Frozen History: When Temporal Losses Stop Accumulation in Learned Real-Time Supersampling

冻结历史：当时间损失停止在学习的实时超采样中积累

Yu, Hao, Zhao, Guoying

Abstract

Learned real-time supersampling accumulates jittered low-resolution samples along motion vectors. Many methods use a temporal consistency loss to reduce flicker. We show that with accurate registration, this loss can suppress accumulation: on static content, the network copies its warped history at almost no temporal cost. Camera jitter in motion vectors raises this cost and can hide the effect. We identify four registration errors that training can conceal: jitter-carrying motion vectors, one motion field for both resolution lattices, jittered depth, and an unregistered skip path. A model-free oracle shows that correct registration makes freezing two orders of magnitude cheaper on static pixels. Controlled training shows that registration reverses the temporal weight's effect on PSNR. Under exact registration, increasing the weight reduces pixel variance at the expense of accuracy and perceived quality, with a 0.7 to 0.9 dB loss on static frames at weight 0.25. This reversal recurs across three network families and two scales; the loss also reduces accuracy on unjittered STSS data. With a small temporal weight and a normalised sample accumulator, a plain registered network improves the best published QRISP PSNR by 0.34 to 0.62 dB at $\times$2, $\times$3 and $\times$4 upscaling to 1080p. At $\times$2, its static pixel standard deviation is one quarter that of DLSS 2. Without retraining, it outperforms classical temporal upsampling by at least 4.4 dB on Arm's jittered Bistro capture. Our most accurate $\times$4 model runs in 2.03 ms per 1080p frame on an H200. Code: https://github.com/CV-AC/Frozen_history

Chinese Translation

学习的实时超采样沿运动向量积累抖动的低分辨率样本。许多方法使用时间一致性损失来减少闪烁。我们展示了在准确配准的情况下，这种损失可以抑制积累：在静态内容上，网络几乎没有时间成本地复制其扭曲的历史。运动向量中的相机抖动提高了这一成本，并可能掩盖这一效果。我们识别出四种训练可以掩盖的配准错误：携带抖动的运动向量、两个分辨率格点共享一个运动场、抖动的深度和未注册的跳过路径。一个无模型的神谕显示，正确的配准使得在静态像素上冻结的成本降低两个数量级。受控训练表明，配准逆转了时间权重对峰值信噪比（PSNR）的影响。在精确配准下，增加权重会减少像素方差，但以牺牲准确性和感知质量为代价，在权重为0.25时，静态帧的损失为0.7到0.9 dB。这种逆转在三个网络家族和两个尺度中反复出现；该损失还降低了未抖动的STSS数据的准确性。使用小的时间权重和归一化样本累加器，一个普通的注册网络在$ imes$2、$ imes$3和$ imes$4放大到1080p时，将最佳已发布的QRISP PSNR提高了0.34到0.62 dB。在$ imes$2时，其静态像素标准差仅为DLSS 2的四分之一。在没有重新训练的情况下，它在Arm的抖动Bistro捕获上至少超越了经典的时间上采样4.4 dB。我们最准确的$ imes$4模型在H200上每1080p帧运行时间为2.03毫秒。代码：https://github.com/CV-AC/Frozen_history

View on arXiv Download PDF AI Translation

cs.CV / 125 / 2610.15764

SlotAVS: Object-Centric Audio-Visual Segmentation via Bi-Modal Slot Attention

SlotAVS：通过双模态槽注意力实现对象中心的音视频分割

Sharma, Arnav, Mishra, Anand, Rameshan, Renu

Abstract

Audio-Visual Segmentation (AVS) requires predicting pixel-wise masks of sounding objects by grounding acoustic signals in visual space. In multi-source scenes, concurrent sound sources often remain entangled within shared visual representations, leading to merged masks or unstable specialization when using standard alignment-focused transformer architectures. To address this, we present SlotAVS, an object-centric framework driven by competitive bi-modal slot attention. While robust foundation models provide rich semantic priors, SlotAVS shifts the paradigm from global cross-modal fusion to explicit competitive binding. By initializing discrete object slots from temporal audio sequences and enforcing strict slot-wise competition over spatial visual tokens, our method structurally mitigates the blending of distinct acoustic entities. Further, to stabilize instance-level learning, we introduce a Hungarian matching strategy that optimally aligns slots with ground-truth objects using spatial and boundary overlap. Coupled with a numerically stable, confidence-aware mask fusion mechanism, SlotAVS robustly aggregates multi-source predictions. Extensive experiments on the AVSBench dataset demonstrate the efficacy of our approach, achieving 88.6% mIoU on the Single-Source (S4) split and 71.7% mIoU on the Multi-Source (MS3) split. These results show that explicit competitive binding is critical for complex audio-visual scene understanding and effectively leveraging foundation-model priors without relying solely on their capacity.

Chinese Translation

音视频分割（AVS）需要通过将声学信号与视觉空间中的对象进行对接，来预测发声对象的像素级掩膜。在多源场景中，多个同时存在的声源往往在共享的视觉表征中交织在一起，导致在使用标准的以对齐为重点的变换器架构时出现掩膜合并或不稳定的专业化。为了解决这一问题，我们提出了SlotAVS，一个由竞争性双模态槽注意力驱动的对象中心框架。虽然强大的基础模型提供了丰富的语义先验，SlotAVS将范式从全局跨模态融合转变为显式的竞争绑定。通过从时间音频序列初始化离散对象槽，并在空间视觉标记上强制执行严格的槽间竞争，我们的方法在结构上减轻了不同声学实体的混合。此外，为了稳定实例级学习，我们引入了一种匈牙利匹配策略，使用空间和边界重叠最优地将槽与真实对象对齐。结合数值稳定的、具有置信度感知的掩膜融合机制，SlotAVS稳健地聚合多源预测。在AVSBench数据集上的大量实验表明我们的方法的有效性，在单源（S4）拆分上实现了88.6%的mIoU，在多源（MS3）拆分上实现了71.7%的mIoU。这些结果表明，显式的竞争绑定对于复杂的音视频场景理解至关重要，并有效利用基础模型的先验，而不单单依赖于其能力。

View on arXiv Download PDF AI Translation

cs.CV / 126 / 2610.15782

Data-Centric Recursive Self-Improvement for Navigation via Policy-Synchronized Language-Action Dual Bootstrapping

基于数据中心的递归自我改进导航方法：策略同步语言-动作双重自举

Li, Songze, Gu, Haoming, Gong, Ziyang, Wang, Zhitong, Zhou, Gengze, Wang, Zun, Zhu, Yuhan, Jiang, Tianxiang, Li, Juanzi, Wu, Qi, Zhang, Weinan, Qiao, Yu, Zhang, Wenwei, Wang, Yi

Abstract

Vision-language navigation requires agents to follow natural-language instructions to reach target locations, making high-quality language supervision essential. Since human annotation is costly, prior work has explored scalable data synthesis methods. However, existing methods typically either use separate models for navigation and instruction generation or couple the two tasks through task-specific components; most also annotate the full trajectory corpus with a fixed model before updating, which can cause generated supervision to lag behind the evolving policy. Therefore, we introduce RSI-Nav, a data-centric recursive self-improvement framework for navigation via policy-synchronized language-action dual bootstrapping, which unifies vision-language-to-action instruction following and vision-action-to-language instruction generation within a shared model. Specifically, after supervised fine-tuning establishes fundamental capabilities, RSI-Nav progressively processes unlabeled trajectories in small shards. The current model generates instructions for these trajectories, executes them to evaluate goal completion and path efficiency, retains qualified samples, and immediately updates itself before processing the next shard. This shard-wise update keeps newly generated supervision aligned with the evolving policy. We also incorporate spatial understanding data, since stronger spatial reasoning benefits navigation. Extensive experiments show that RSI-Nav reaches 68.6% SR on R2R-CE and 72.4% SR on RxR-CE, while providing stronger instruction-generation utility and modestly improved spatial understanding after recursive evolution.

Chinese Translation

视觉-语言导航要求智能体遵循自然语言指令以到达目标位置，这使得高质量的语言监督变得至关重要。由于人工标注成本高昂，之前的研究探索了可扩展的数据合成方法。然而，现有方法通常要么为导航和指令生成使用独立模型，要么通过特定任务的组件将这两项任务结合在一起；大多数方法还在更新之前使用固定模型对完整轨迹语料库进行标注，这可能导致生成的监督信息滞后于不断演变的策略。因此，我们提出了RSI-Nav，一个基于数据中心的递归自我改进框架，通过策略同步的语言-动作双重自举实现导航，该框架在共享模型中统一了视觉-语言到动作的指令跟随和视觉-动作到语言的指令生成。具体而言，在经过监督微调建立基本能力后，RSI-Nav逐步处理未标记的轨迹，分成小块。当前模型为这些轨迹生成指令，执行这些指令以评估目标完成情况和路径效率，保留合格样本，并在处理下一个小块之前立即更新自身。这种按小块更新的方式使新生成的监督与不断演变的策略保持一致。我们还结合了空间理解数据，因为更强的空间推理有利于导航。大量实验表明，RSI-Nav在R2R-CE上达到68.6%的成功率（SR），在RxR-CE上达到72.4%的成功率，同时在递归演化后提供了更强的指令生成效用和适度改善的空间理解能力。

View on arXiv Download PDF AI Translation

cs.CV / 127 / 2610.15806

Reference-Efficient Multimodal Bitstream-Corrupted Video Recovery via Depth-Conditioned Adaptation

基于深度条件适应的参考高效多模态比特流损坏视频恢复

Liu, Tianyi, Wang, Jiaojiao, Lu, Ye, Wu, Kejun, Cai, Chengtao, Li, Zhengguo, Lin, Chia-Wen

Abstract

Recovering visually faithful content from damaged video bitstreams is important for reliable multimedia communication and storage. Existing methods exploit local residuals and non-local RGB references of the input video sequence, but both sources can become unreliable after decoding, while processing additional references increases receiver computation and memory. This paper studies multimodal bitstream-corrupted video recovery (MBSCVR) using RGB and depth information, where pre-transmission depth-based geometry aids reconstruction when the received appearance is corrupted. We propose a novel MBSCVR framework that selects and effectively utilizes a single selected non-local RGB reference per recovery window to achieve higher performance. It integrates a reliability-aware multimodal selector that combines valid appearance with depth compatibility before reference encoding. Depth-Conditioned Reference Adaptation (DCRA) introduces selected RGB evidence into feature completion and content generation, while frame-aligned depth conditions completion queries. On the complete BSCV benchmark with oracle masks, MBSCVR uses one selected reference to achieve 32.30 dB on YouTube-VOS and 28.93 dB on DAVIS, surpassing full-reference B2SCVR by 0.27 and 0.33 dB and improving SSIM on both subsets. Matched receiver-forward measurements show 36.8\% lower latency and 40.8\% lower peak memory. These results demonstrate how compact geometric guidance and adapted appearance processing improve the quality--resource trade-off of multimedia recovery.

Chinese Translation

从损坏的视频比特流中恢复视觉上真实的内容对于可靠的多媒体通信和存储至关重要。现有方法利用输入视频序列的局部残差和非局部RGB参考，但在解码后这两种来源都可能变得不可靠，而处理额外的参考会增加接收端的计算和内存负担。本文研究了使用RGB和深度信息的多模态比特流损坏视频恢复（MBSCVR），其中基于深度的几何信息在接收到的外观被损坏时有助于重建。我们提出了一种新颖的MBSCVR框架，该框架在每个恢复窗口中选择并有效利用一个单一的非局部RGB参考，以实现更高的性能。它集成了一个可靠性感知的多模态选择器，在参考编码之前结合有效的外观与深度兼容性。深度条件参考适应（DCRA）将选定的RGB证据引入特征补全和内容生成，同时帧对齐的深度条件完成查询。在完整的BSCV基准测试中，使用oracle掩码，MBSCVR通过一个选定的参考在YouTube-VOS上达到了32.30 dB，在DAVIS上达到了28.93 dB，分别超越了全参考B2SCVR 0.27 dB和0.33 dB，并在两个子集上提高了SSIM。匹配的接收器前向测量显示出36.8\%的延迟降低和40.8\%的峰值内存降低。这些结果证明了紧凑的几何指导和适应的外观处理如何改善多媒体恢复的质量与资源的权衡。

View on arXiv Download PDF AI Translation

cs.CV / 128 / 2610.15807

Totem: Object-Centric 3D Memory for World-Consistent Video Generation

图腾：面向世界一致性视频生成的对象中心3D记忆

Wan, Zichen, Huang, Tianyu, Li, Fan, Wang, Kai, Zhang, Hongzhi, Zuo, Wangmeng

Abstract

Generating coherent videos grounded in a unified underlying world remains a major challenge, particularly under large viewpoint changes, where existing methods often struggle to maintain stable representations of individual objects. To this end, we introduce Totem, an object-centric 3D memory framework for world-consistent video generation. Totem first represents each selected object as a persistent 3D memory aligned to a shared world frame, assigning it a dedicated state. Based on these memories, we design a dual-stream 3D conditioning architecture that combines object memories with scene geometry for target-view synthesis. Moreover, these object memories constrain the model's attention with 3D structural information, thereby preserving object surfaces and relative spatial layouts. Extensive experiments on world-generation and object-centric benchmarks demonstrate that Totem generates videos depicting a more object-consistent 3D world under challenging camera trajectories, while supporting versatile applications.

Chinese Translation

在统一的基础世界中生成连贯的视频仍然是一个主要挑战，尤其是在大视角变化下，现有方法往往难以保持个体对象的稳定表示。为此，我们提出了图腾（Totem），一种面向世界一致性视频生成的对象中心3D记忆框架。图腾首先将每个选定对象表示为与共享世界框架对齐的持久3D记忆，为其分配专用状态。基于这些记忆，我们设计了一种双流3D条件架构，将对象记忆与场景几何结合，以实现目标视图合成。此外，这些对象记忆通过3D结构信息约束模型的注意力，从而保持对象表面和相对空间布局。在世界生成和对象中心基准测试上的大量实验表明，图腾在具有挑战性的相机轨迹下生成描绘更具对象一致性的3D世界的视频，同时支持多种应用。

View on arXiv Download PDF AI Translation

cs.CV / 129 / 2610.15810

Adaptive Look, Flexible Thought: Learning How and How Long to Think in Multimodal Reasoning

自适应视觉，灵活思维：在多模态推理中学习如何思考及思考的时长

Chen, Quanlin, Chen, Xiuwei, Xiang, Kun, Ma, Zehua, Ye, Chunjin, Guo, Boqiang, Liu, Xianming, Liang, Xiaodan

Abstract

The introduction of long chain-of-thought (CoT) reasoning and external tools has substantially improved the performance of multimodal large language models (MLLMs) on general and reasoning tasks. However, existing fast-slow reasoning paradigms mainly regulate textual reasoning length, while visual computation introduced by external tools or visual tokens is often treated as an auxiliary component. This separation prevents MLLMs from jointly optimizing visual perception and textual reasoning. To address this issue, we propose L2T (adaptive Look, flexible Thought), a unified reasoning framework that jointly optimizes the form and depth of computation. L2T first discretizes tool-produced visual states into visual tokens, bringing visual perception and textual reasoning into a shared autoregressive space. Based on this unified representation, L2T introduces a modality-aware routing reward to select appropriate reasoning modes and a difficulty-aware fast-slow strategy to adapt reasoning depth according to task complexity. Experiments on general perception and reasoning-intensive tasks demonstrate the effectiveness of L2T. Compared with explicit tool-calling approaches, L2T maintains competitive performance while achieving 1.7--3.6 times higher throughput.

Chinese Translation

长链思维（CoT）推理和外部工具的引入显著提高了多模态大型语言模型（MLLMs）在一般任务和推理任务上的表现。然而，现有的快慢推理范式主要调节文本推理的长度，而外部工具或视觉标记引入的视觉计算通常被视为辅助组件。这种分离阻碍了MLLMs在视觉感知和文本推理上的联合优化。为了解决这一问题，我们提出了L2T（自适应视觉，灵活思维），一个统一的推理框架，能够共同优化计算的形式和深度。L2T首先将工具生成的视觉状态离散化为视觉标记，将视觉感知和文本推理引入共享的自回归空间。在这一统一表示的基础上，L2T引入了模态感知的路由奖励，以选择适当的推理模式，并采用难度感知的快慢策略，根据任务复杂性调整推理深度。在一般感知和推理密集型任务上的实验表明L2T的有效性。与显式工具调用方法相比，L2T在保持竞争性性能的同时，实现了1.7至3.6倍的更高吞吐量。

View on arXiv Download PDF AI Translation

cs.CV / 130 / 2610.15813

ChangeT$^3$: Evaluating Vision Test-Time Training for Remote Sensing Change Detection

ChangeT$^3$: 评估视觉测试时训练在遥感变化检测中的应用

Niranjan, Sharukshan, Samarakoon, Oshadha, Wasalathilaka, Nichula, Godaliyadda, Roshan, Ekanayake, Parakrama, Herath, Vijitha

Abstract

Bitemporal remote sensing change detection requires representations that preserve spatial detail while capturing contextual information relevant to changes between acquisitions. Visual test-time training offers an alternative to explicit token-to-token attention by adapting a compact inner model to each image. This letter describes ChangeT$^3$, a vanilla Siamese configuration combining the existing hierarchical ViT$^3$-Tiny encoder with the multiscale ChangeFormer decoder. The two image streams share learned encoder parameters, while their fast weights are computed independently from their respective spatial features. Each encoder block retains the inherited global simplified SwiGLU and local adaptive depthwise-convolution branches. Temporal comparison is performed by the decoder through learned fusion of paired features at four resolutions, followed by coarse-to-fine aggregation and binary prediction. This separation provides a direct setting for examining image-conditioned inner updates without introducing an additional temporal interaction module. Trained from scratch on LEVIR-CD+ with the same decoder, recipe, and budget as a ChangeFormer control, ChangeT$^3$ reaches an F1 of 80.61$\pm$0.90 against 78.84$\pm$0.21 for ChangeFormer (three seeds each); as in the standard protocol for this dataset, checkpoints are selected on the test split. Trained with both inner updates disabled, the same encoder reaches 78.63$\pm$0.20. Neither model transfers zero-shot to SYSU-CD, DSIFN-CD, or WHU-CD, and ChangeT$^3$ is the slower of the two at small inputs despite similar parameter and operation counts.

Chinese Translation

双时相遥感变化检测需要在保留空间细节的同时捕捉与获取之间变化相关的上下文信息。视觉测试时训练提供了一种替代显式的标记到标记注意力的方法，通过将紧凑的内部模型适应于每个图像。本文描述了ChangeT$^3$，一种基础的Siamese配置，结合了现有的层次化ViT$^3$-Tiny编码器与多尺度ChangeFormer解码器。两个图像流共享学习到的编码器参数，而它们的快速权重则独立于各自的空间特征进行计算。每个编码器块保留了继承的全局简化SwiGLU和局部自适应深度卷积分支。解码器通过在四个分辨率下学习融合配对特征来进行时间比较，随后进行粗到细的聚合和二元预测。这种分离提供了一个直接的设置，用于检查图像条件下的内部更新，而无需引入额外的时间交互模块。在LEVIR-CD+上从头开始训练，使用与ChangeFormer控制相同的解码器、配方和预算，ChangeT$^3$达到了80.61$ ext{±}$0.90的F1分数，而ChangeFormer的F1分数为78.84$ ext{±}$0.21（每个模型三次实验）；与该数据集的标准协议一致，检查点在测试集上选择。在禁用内部更新的情况下，相同的编码器达到了78.63$ ext{±}$0.20。两个模型均未能在SYSU-CD、DSIFN-CD或WHU-CD上实现零样本迁移，尽管参数和操作计数相似，ChangeT$^3$在小输入下的速度较慢。

View on arXiv Download PDF AI Translation

cs.CV / 131 / 2610.15846

Mobility Helps Learning: Unsupervised Model Adaptation for Object Recognition via Movement

移动性促进学习：通过运动进行无监督模型适应的物体识别

Ma, Yanan, Tao, Yihang, Fang, Zhengru, Hu, Senkang, Chen, Xianhao, Fang, Yuguang

Abstract

Movable agents, such as autonomous vehicles and robots, should be able to autonomously adapt their pre-trained object recognition models to new environments without human annotations. While existing test-time adaptation (TTA) methods address label-free model adaptation, they typically treat test data as independent snapshots, without constructively exploiting a rich, unique source of supervision: the significant variance in prediction quality as an agent observes the same object from different distances and viewpoints. To capitalize on this, we propose \texttt{MoCaFe}, a model-agnostic and hyperparameter-insensitive paradigm that leverages motion-induced prediction variance for unsupervised adaptation. Specifically, we first develop mobility-calibrated filtering of pseudo-labels, which constructs reliable pseudo-labels by fusing cross-view predictions using inverse-variance weights. The filtered pseudo-labels then guide the selection of ``hard samples'' for effective learning. To further suppress pseudo-label corruption, we introduce \texttt{p-MoCaFe}, an optional extension that uses a public dataset to assign weighting factors to data samples, yielding an estimator of the clean loss with bounded bias. Extensive experiments on autonomous driving (nuScenes, KITTI) and embodied-agent datasets show that \texttt{MoCaFe} significantly outperforms state-of-the-art TTA schemes, achieving up to approximately 20% accuracy gain.

Chinese Translation

可移动的智能体，如自主车辆和机器人，应能够在没有人工标注的情况下，自主地将其预训练的物体识别模型适应于新环境。虽然现有的测试时适应（Test-Time Adaptation, TTA）方法解决了无标签模型适应的问题，但它们通常将测试数据视为独立的快照，而未能有效利用一个丰富且独特的监督源：当智能体从不同距离和视角观察同一物体时，预测质量的显著差异。为了利用这一点，我们提出了 exttt{MoCaFe}，一种与模型无关且对超参数不敏感的范式，利用运动引起的预测方差进行无监督适应。具体而言，我们首先开发了移动性校准的伪标签过滤，通过使用逆方差权重融合跨视图预测来构建可靠的伪标签。过滤后的伪标签随后指导“困难样本”的选择，以实现有效学习。为了进一步抑制伪标签的污染，我们引入了 exttt{p-MoCaFe}，这是一个可选扩展，利用公共数据集为数据样本分配权重因子，从而得到一个具有有界偏差的干净损失估计器。在自主驾驶（nuScenes, KITTI）和具身智能体数据集上进行的广泛实验表明， exttt{MoCaFe}显著优于最先进的TTA方案，准确率提升可达约20%。

View on arXiv Download PDF AI Translation

cs.CV / 132 / 2610.15854

Sharper Frames, Worse Rooms: Viewpoint Coverage Dominates Image Quality in Feed-Forward 3D Reconstruction

更清晰的帧，较差的房间：视点覆盖在前馈3D重建中主导图像质量

Dadajonov, Alibek, Umarhojiyev, Hikmatillo

Abstract

Grounded spatial reasoning about indoor scenes rests on a geometric reconstruction that the multimodal system does not itself produce. Feed-forward multi-view reconstruction models operate under a fixed view budget, making frame selection a critical upstream decision. We show that under this setting the conventional preference for sharp frames can be actively harmful. Across nine handheld room scans, selecting the sharpest 16 of 200 candidate frames is the worst strategy we test, reaching 48\% room dimension error where uniform temporal sampling reaches 3\%. We trace this failure to a coupling between image quality and camera motion: sharp frames tend to occur during slow motion and therefore concentrate the selection in a narrow angular region, leaving $192^\circ$ of the room unobserved in the case we illustrate and over $300^\circ$ in the worst cells. A $2\times2$ factorial experiment that independently manipulates viewpoint spread and per-frame sharpness shows that spread has roughly three times the main effect of quality, and that blurry but well spread selections outperform sharp but clustered selections in 8 of 9 captures. The advantage also diminishes as the frame budget increases and angular coverage saturates, consistent with this mechanism. Finally, we ask whether the failure is detectable at inference: the reconstruction model's multi-view consistency score increases with reconstruction error, while a vision-language model given the same frames reports confidence unrelated to its own spatial estimation error. Under small view budgets, viewpoint coverage is a more important constraint than per-frame image quality, and neither uncertainty signal available at inference reveals this upstream failure.

Chinese Translation

关于室内场景的基础空间推理依赖于几何重建，而多模态系统并不直接生成该重建。前馈多视角重建模型在固定视图预算下运行，使得帧选择成为一个关键的上游决策。我们展示了在这种情况下，传统上对清晰帧的偏好可能会产生积极的危害。在九个手持房间扫描中，选择200个候选帧中最清晰的16帧是我们测试的最差策略，导致房间尺寸误差达到48\%，而均匀时间采样的误差仅为3\%。我们将这一失败归因于图像质量与相机运动之间的耦合：清晰帧往往发生在缓慢运动期间，因此将选择集中在狭窄的角度区域，在我们所示例的情况下，有$192^ ext{°}$的房间未被观察到，而在最糟糕的情况下则超过$300^ ext{°}$。一个$2 imes2$的因子实验独立操控视点分布和每帧清晰度，结果显示视点分布的主要效果大约是质量的三倍，并且在9次捕获中，有8次模糊但分布良好的选择优于清晰但聚集的选择。随着帧预算的增加和角度覆盖的饱和，这一优势也在减小，与这一机制一致。最后，我们探讨这一失败是否在推理阶段可被检测到：重建模型的多视角一致性评分随着重建误差的增加而提高，而在给定相同帧的情况下，视觉-语言模型报告的置信度与其自身的空间估计误差无关。在小视图预算下，视点覆盖是比每帧图像质量更重要的约束，而推理阶段可用的任何不确定性信号都未能揭示这一上游失败。

View on arXiv Download PDF AI Translation

cs.CV / 133 / 2610.15891

Learning Multimodal Sensor Fusion with JEPA Representations for Breathing Activity and Pose Recognition

基于JEPA表示的多模态传感器融合学习用于呼吸活动和姿态识别

Luo, Yunxiang, Casado, Constantino Álvarez, Wu, Xiaoting, Nguyen, Nhi, Nguyen, Le, Cañellas, Manuel Lage, López, Miguel Bordallo

Abstract

Predictive video representations do not by themselves specify how a multimodal system should handle missing sensors. We transfer frozen V-JEPA2 features to a set of heterogeneous sensors comprised by RGB video, depth video, and radar time series. A shared fusion model uses modality queries, fixed sensor slots, and explicit availability to classify breathing activity and pose from all seven non-empty sensor subsets. The contribution is a controlled evaluation of subset supervision, pooling capacity, and representation transfer on 39 OMuSense-23 subjects with five subject-group folds and three seeds. With matched downstream parameter counts, the final model reaches recording-level activity macro-F1 of $75.23%\pm0.60$ averaged across seven subsets. Supervising all subsets in each batch improves the matched query-pooling control by 1.923 percentage points (conditional 95% subject-bootstrap interval [0.647, 3.209]), while query pooling shows no clear advantage over matched mean pooling under balanced supervision. Radar probes support transfer beyond pooled wavelet pixels, 64-token inference approximately preserves mean activity performance, and the tested Consensus-JEPA loss adds no measured benefit against its exact control.

Chinese Translation

预测视频表示本身并未明确说明多模态系统应如何处理缺失的传感器。我们将冻结的V-JEPA2特征转移到由RGB视频、深度视频和雷达时间序列组成的异构传感器集合中。一个共享融合模型利用模态查询、固定传感器槽和显式可用性，从所有七个非空传感器子集中对呼吸活动和姿态进行分类。我们的贡献是对39个OMuSense-23受试者进行子集监督、池化能力和表示转移的控制评估，包含五个受试者组折叠和三个种子。在匹配的下游参数计数下，最终模型在七个子集上达到录音级别活动宏F1值为$75.23% ext{±}0.60$的平均值。每批次监督所有子集将匹配查询池控件提高了1.923个百分点（条件95%受试者自助区间[0.647, 3.209]），而在平衡监督下，查询池化相较于匹配均值池化并未显示出明显优势。雷达探测支持超越池化小波像素的转移，64个标记推理大致保持了均值活动性能，测试的Consensus-JEPA损失相较于其精确控制未增加任何可测益处。

View on arXiv Download PDF AI Translation

cs.CV / 134 / 2610.15900

Gen Through the Lens: Camera-Controllable Video Generation via Cross-Coordinate Conditioning Fusion

透过镜头生成：通过跨坐标条件融合实现相机可控的视频生成

Li, Shuai, Xiang, Jinxu, Li, Xinyuan, Guo, Hanlei, Tan, Xiyang, Zhao, Alan, Zhao, Xiaotong, Zhang, Ran

Abstract

We present Gen Through the Lens, a camera-controllable image-to-video generation framework that synthesizes videos from a given reference image while enabling realistic and physically consistent camera operations. Unlike previous models that treat camera motion as abstract conditioning or rely on a single coordinate system, our approach explicitly decomposes and fuses control signals across three coordinate spaces: the camera coordinate (modeled by intrinsic parameters), the world coordinate (represented by spatial geometry hints), and the object coordinate (described by 3D bounding boxes of target subjects). Through cross-coordinate conditioning fusion, these heterogeneous cues are unified to guide the generative process, enabling precise control of camera motion, focus, and framing while preserving temporal and spatial coherence. Extensive experiments demonstrate that our method produces controllable and visually consistent results across diverse camera operations, outperforming existing approaches in both realism and editability.

Chinese Translation

我们提出了透过镜头生成（Gen Through the Lens），这是一种可通过相机控制的图像到视频生成框架，能够从给定的参考图像合成视频，同时实现真实且物理一致的相机操作。与以前将相机运动视为抽象条件或依赖单一坐标系统的模型不同，我们的方法明确地分解并融合了来自三个坐标空间的控制信号：相机坐标（由内在参数建模）、世界坐标（由空间几何线索表示）和物体坐标（由目标对象的三维边界框描述）。通过跨坐标条件融合，这些异构线索被统一以指导生成过程，从而实现对相机运动、焦点和构图的精确控制，同时保持时间和空间的一致性。大量实验表明，我们的方法在各种相机操作中产生可控且视觉一致的结果，在真实感和可编辑性方面优于现有方法。

View on arXiv Download PDF AI Translation

cs.CV / 135 / 2610.15915

VIEWCRAFT: Evolving What Multimodal Agents See and How They Read It

VIEWCRAFT：进化多模态智能体的视觉与阅读方式

Xu, Binxiao, An, Ruichuan, Zou, Bocheng, Du, Weihua, Wei, Xinyu, Guo, Hangyu, Zhang, YiFan, Yan, Renye, Duan, Chengqi, Hua, Hang

Abstract

Reusable skills improve multimodal agents without updating model weights, but their effectiveness depends on both the evidence presented and the procedure used to read it. We introduce ViewCraft, which jointly evolves an executable evidence interface and its reading procedure. The interface selects, transforms, and organizes observations; the reader interprets the resulting views. Failure traces guide revisions, replay and paired evaluation retain useful changes, and interface projection recovers visual edits from eligible rejected joint proposals. Across eight benchmarks, ViewCraft improves over direct execution by an average of $6.1$ points in task score, with gains on every benchmark. The advantage persists at a matched input-token budget. On all five multi-source tasks, the learned interface uses 8%-15% fewer input tokens than Initial skill. The learned skills also transfer across executors and outperform direct execution in every evaluated transfer, supporting evidence construction and reading as a coupled, reusable skill.

Chinese Translation

可重用的技能在不更新模型权重的情况下提升多模态智能体的性能，但其有效性依赖于所呈现的证据和读取证据的过程。我们提出了ViewCraft，它共同进化一个可执行的证据接口及其阅读过程。该接口选择、转换和组织观察结果；阅读器则解释生成的视图。失败轨迹指导修订，重放和配对评估保留有用的变化，而接口投影则从合格的被拒绝联合提案中恢复视觉编辑。在八个基准测试中，ViewCraft在任务得分上平均提高了6.1分，并在每个基准测试中均有所提升。这一优势在匹配的输入标记预算下依然存在。在所有五个多源任务中，学习到的接口使用的输入标记比初始技能少8%-15%。学习到的技能在不同执行者之间也能迁移，并在每个评估的迁移中优于直接执行，支持证据构建和阅读作为一种耦合的可重用技能。

View on arXiv Download PDF AI Translation

cs.CV / 136 / 2610.15922

Frequency-Step Cache: Separating Synchronization and Repair for Efficient Autoregressive Video Diffusion

频率步缓存：分离同步与修复以实现高效的自回归视频扩散

Xu, Kerry, Misra, Param, Stenglein, Zachary, Karunamurthy, Athri, Ye, Fanjiang, Wang, Yuke

Abstract

Autoregressive (AR) video generation enables long-horizon synthesis but requires repeated forward passes through large diffusion transformers. Training-free caching reduces this cost by reusing intermediate computation, yet existing controllers often couple two decisions: when to synchronize the cache exactly and how much computation to repair between synchronizations. We introduce Frequency-Step Cache (FSC), which separates these roles. FSC schedules exact full-computation anchors on a dense-to-sparse trajectory, while a lightweight latent-space risk based on frequency-resolved drift and token age controls inter-anchor repair through whole-step replay, selective token recomputation, or full repair. On SkyReels-V2, whole-step reuse provides most of the acceleration, while selective token computation is most effective as fidelity repair. A controlled ablation shows that using one reactive signal for both synchronization and repair reduces fidelity at matched latency, with the same separation advantage observed on FramePack and Matrix-Game. Across multiple autoregressive video backbones, FSC improves speed-quality trade-offs. On SkyReels-V2, FSC-Speed reaches 2.36x speedup versus MotionCache's 2.275x while improving SSIM by 0.019 and PSNR by 0.5 dB. At MotionCache-matched speed, FSC-Quality improves SSIM by 0.031 and PSNR by 1.0 dB and lowers LPIPS by 0.024. Without retuning, FSC-Quality transfers to SkyReels-V2 14B at 2.12x speedup versus 1.79x for MotionCache.

Chinese Translation

自回归（AR）视频生成能够实现长时间跨度的合成，但需要通过大型扩散变换器进行重复的前向推理。无训练缓存通过重用中间计算来降低这一成本，然而现有的控制器通常将两个决策耦合在一起：何时精确同步缓存以及在同步之间需要修复多少计算。我们提出了频率步缓存（Frequency-Step Cache, FSC），将这两种角色分离。FSC在密集到稀疏的轨迹上调度精确的全计算锚点，同时基于频率分辨的漂移和标记年龄的轻量级潜在空间风险控制通过全步重放、选择性标记重新计算或完全修复进行锚点间修复。在SkyReels-V2上，全步重用提供了大部分加速，而选择性标记计算在保真度修复方面最为有效。受控消融实验表明，使用一个反应信号同时进行同步和修复会在匹配延迟下降低保真度，而在FramePack和Matrix-Game上观察到相同的分离优势。在多个自回归视频骨干网络上，FSC改善了速度与质量的权衡。在SkyReels-V2上，FSC-Speed相较于MotionCache的2.275倍加速达到了2.36倍，同时SSIM提高了0.019，PSNR提高了0.5 dB。在与MotionCache匹配的速度下，FSC-Quality使SSIM提高了0.031，PSNR提高了1.0 dB，并降低了LPIPS 0.024。无需重新调优，FSC-Quality在SkyReels-V2 14B上以2.12倍的加速转移，相较于MotionCache的1.79倍。

View on arXiv Download PDF AI Translation

cs.CV / 137 / 2610.15929

When Trackers Fail: VLM-Guided Verification and Recovery for Robust Video Object Segmentation

当跟踪器失败时：基于视觉语言模型的验证与恢复方法用于鲁棒的视频目标分割

Bundele, Valay, Agrawal, Susmit, Hosseinzadeh, Mehran, Nguyen, The Nam, Lensch, Hendrik P. A.

Abstract

Recent trackers built on foundation models such as the Segment Anything Model (SAM) achieve strong performance by propagating object masks through memory. However, this strength is also a key source of failure: under occlusion, reappearance, or distractors, an incorrect prediction can enter the tracking loop and be propagated over time, resulting in persistent identity drift. Existing approaches seek to prevent such failures through improved memory and temporal modeling, but once a failure occurs, the tracker has no explicit mechanism to recognize or correct it. We address this gap with Dual-Agent, a failure-aware tracking framework that complements propagation with explicit verification and recovery. We address this gap with Dual-Agent, a failure-aware tracking framework that complements propagation with explicit verification and recovery. A vision-language verification agent first reasons over spatio-temporal context to determine whether the tracker is still following the target. When a failure is detected, a correction agent uses temporal context to recover the target before tracking continues. The correction agent is trained on real SAM3 failures, augmented with targeted identity-switch and distractor perturbations that expose it to challenging ambiguities during recovery. On LVOS-v2 and SecVOS, Dual-Agent improves SAM3 by 1.2 and 2.7 $J\&F$, respectively. Importantly, the proposed framework transfers without retraining to SAM2.1, SeC, and Cutie, as well as to long-term video object tracking benchmarks including LaSOT-ext, TNL2K, and VOT-LT2022. Our results show that explicitly reasoning about when tracking fails and how to recover provides a complementary path to robust tracking beyond improving propagation alone. Project page: valaybundele.github.io/Dual-Agent.

Chinese Translation

近期基于基础模型（如Segment Anything Model，SAM）构建的跟踪器通过在记忆中传播对象掩码实现了强大的性能。然而，这一优势也是失败的主要来源：在遮挡、重新出现或干扰物的情况下，错误的预测可能会进入跟踪循环并随时间传播，导致持续的身份漂移。现有方法试图通过改进记忆和时间建模来防止此类失败，但一旦发生失败，跟踪器没有明确的机制来识别或纠正它。我们通过Dual-Agent解决了这一问题，这是一种关注失败的跟踪框架，通过显式的验证和恢复来补充传播。视觉-语言验证代理首先对时空上下文进行推理，以确定跟踪器是否仍在跟随目标。当检测到失败时，纠正代理利用时间上下文在跟踪继续之前恢复目标。纠正代理在真实的SAM3失败上进行训练，并通过针对性的身份切换和干扰物扰动进行增强，使其在恢复过程中能够应对具有挑战性的模糊情况。在LVOS-v2和SecVOS上，Dual-Agent分别将SAM3的性能提高了1.2和2.7 $J ext{&}F$。重要的是，所提出的框架在不重新训练的情况下可以迁移到SAM2.1、SeC和Cutie，以及长期视频目标跟踪基准，包括LaSOT-ext、TNL2K和VOT-LT2022。我们的结果表明，明确推理跟踪何时失败以及如何恢复，提供了一条超越单纯改进传播的鲁棒跟踪的补充路径。项目页面：valaybundele.github.io/Dual-Agent。

View on arXiv Download PDF AI Translation

cs.CV / 138 / 2610.16009

Meaningful Moments: A Corpus of VLM-Derived Temporal Importance Labels for Video Action Recognition

有意义的时刻：用于视频动作识别的基于视觉语言模型的时间重要性标签语料库

Keith, Cameron, Jin, SouYoung

Abstract

Not every moment in a video is equally useful for recognizing an action. A few decisive moments often carry the evidence: the instant an object moves, the peak of a dive, or one object placed on another. Existing video datasets rarely provide large-scale, action-conditioned labels for which short segments carry that evidence. This work asks whether vision-language models can identify meaningful moments at scale and whether those labels improve action recognition.We introduce Meaningful Moments (MM), a corpus of about 4.58 million VLM-generated per-segment importance scores for nearly 500,000 class-labeled videos spanning Something-Something v2 (SSv2), Kinetics-400 (K400), and Diving-48 across 622 action classes. We freeze the action classifier and vary only temporal sampling. An alpha-parameterized comparison of importance-led selection against an importance-inverted control separates gains from the VLM signal from gains caused by variable-density sampling alone. Across all three datasets, VLM-derived importance is informative: at the hard-cut endpoint, keeping VLM-important segments outperforms keeping VLM-designated filler. Usefulness depends on class structure. On Diving-48, importance-led fast-forwarding preserves full-video recognition while the inverted control collapses by roughly 30 percentage points. On K400, hard-cut selection beats uniform sampling by 1.45 percentage points, rising to 2.05 under threshold selection. On SSv2, the labels carry signal but do not recover full-video performance. Direct scoring matches greedy removal within its noise floor at roughly 25 times fewer oracle calls, and a four-VLM cross-oracle study shows consistent downstream trends. We release the importance labels and evaluation protocol under CC-BY-4.0; SSv2 class strings remain under CC-BY-NC-4.0.

Chinese Translation

并非视频中的每一个时刻在识别动作时都同样有用。一些决定性时刻往往承载着证据：物体移动的瞬间、跳水的顶点，或一个物体放置在另一个物体上。现有的视频数据集很少提供大规模、基于动作的标签，以指明哪些短片段承载着这些证据。本研究探讨视觉语言模型（VLM）是否能够大规模识别有意义的时刻，以及这些标签是否能改善动作识别。我们引入了有意义的时刻（Meaningful Moments, MM），这是一个包含约458万条VLM生成的每个片段重要性评分的语料库，涵盖近50万个带类标签的视频，涉及622个动作类别，数据来源于Something-Something v2 (SSv2)、Kinetics-400 (K400)和Diving-48。我们固定动作分类器，仅改变时间采样。通过对重要性引导选择与重要性反转控制进行α参数化比较，分离出VLM信号带来的增益与仅由变量密度采样引起的增益。在所有三个数据集中，VLM衍生的重要性是有信息量的：在硬切割的终点，保留VLM重要片段的表现优于保留VLM指定的填充片段。有效性依赖于类别结构。在Diving-48上，重要性引导的快进保留了全视频识别能力，而反转控制则下降了约30个百分点。在K400上，硬切割选择比均匀采样高出1.45个百分点，在阈值选择下上升至2.05个百分点。在SSv2上，标签传递信号但未能恢复全视频性能。直接评分在其噪声底线内与贪婪去除相匹配，调用的oracle次数减少了约25倍，而四个VLM交叉oracle研究显示出一致的下游趋势。我们在CC-BY-4.0下发布重要性标签和评估协议；SSv2类字符串仍在CC-BY-NC-4.0下。

View on arXiv Download PDF AI Translation

cs.CV / 139 / 2610.16022

Removing the Compute Bottleneck for Event-based Visual Place Recognition with Spiking Neural Networks

利用脉冲神经网络消除事件驱动视觉位置识别的计算瓶颈

Hussaini, Somayeh, Fischer, Tobias, Milford, Michael

Abstract

Neuromorphic sensing and computing promise low latency and high energy efficiency for robotics, yet most event-based visual place recognition (VPR) pipelines reintroduce dense, frame-like computation downstream of an inherently sparse sensor, forfeiting the sensor's latency advantage. In this paper, we address the problem of resolving an event-based VPR query in tens of microseconds on a commodity CPU. We train a two-layer spiking neural network (SNN) with additive spike-timing-dependent plasticity (STDP), which drives the learned weight matrix towards a strongly bimodal distribution, amenable to aggressive magnitude pruning. At inference, our pipeline exploits this sparsity by reading only the pruned weights corresponding to the query's active pixels. As surviving weights share a near-uniform magnitude, each neuron's score reduces to an integer count of matching synapses instead of a weighted sum, lowering per-query latency. This computation is small enough to remain resident in a CPU's fast on-chip cache. On three event-based VPR benchmarks, our sparse SNN exceeds the recall of LENS, an existing event-based SNN, while reducing per-query latency by 5.5-6.2x to tens of microseconds. It matches the recall of a dense variant of our network and of the sum-of-absolute-differences (SAD) baseline, while reading 10x fewer weights than the dense variant. We further characterise how recall and latency scale with the number of active pixels and reference places. These results suggest that the sparsity of both event-based input and learned weights can already be exploited for low-latency inference on conventional processors.

Chinese Translation

神经形态传感和计算为机器人技术提供了低延迟和高能效的可能性，但大多数事件驱动的视觉位置识别（VPR）管道在一个本质上稀疏的传感器下重新引入了密集的、类似帧的计算，从而放弃了传感器的延迟优势。本文解决了在普通CPU上以微秒级别解析事件驱动VPR查询的问题。我们训练了一个具有加性脉冲时序依赖性可塑性（STDP）的两层脉冲神经网络（SNN），该网络将学习到的权重矩阵驱动向一个强双峰分布，适合进行激进的幅度剪枝。在推理阶段，我们的管道通过仅读取与查询的活跃像素对应的剪枝权重来利用这种稀疏性。由于存活的权重具有近乎均匀的幅度，每个神经元的得分简化为匹配突触的整数计数，而不是加权和，从而降低了每个查询的延迟。这一计算足够小，能够保持在CPU的快速片上缓存中。在三个事件驱动的VPR基准测试中，我们的稀疏SNN超越了现有事件驱动SNN LENS的召回率，同时将每个查询的延迟降低了5.5-6.2倍，达到微秒级别。它的召回率与我们网络的密集变体和绝对差和（SAD）基线相匹配，同时读取的权重比密集变体少10倍。我们进一步描述了召回率和延迟如何随着活跃像素和参考位置的数量而变化。这些结果表明，事件驱动输入和学习权重的稀疏性已经可以在传统处理器上用于低延迟推理。

View on arXiv Download PDF AI Translation

cs.CV / 140 / 2610.16058

From Failure Prediction to Component Selection in Semantic Segmentation

从失败预测到语义分割中的组件选择

Kang, Minhui, Zhang, Yiyao, Long, Peizhe, Qi, Jiafeng, Jin, Hongyu, Zheng, Jiayun

Abstract

Reliable scene understanding requires deciding which predictions to trust, retain, or inspect. In semantic segmentation, component failure predictors support these decisions, but their predictive performance does not fully describe the value of the resulting selection. We develop a decision-centered evaluation framework that connects three questions: what an available score predicts, what can be learned from its existing fields, and what its ranking delivers. Across three segmentation systems on Cityscapes and ACDC at 90% retention, adapted MetaSeg reduces retained failure relative to forecast entropy in all twelve architecture-population comparisons, but does not establish a failure advantage over direct probability ranking. Image-nested source experiments reveal a different opportunity: reusing the full field set improves predictive and retained-loss performance under both learners in all three systems, while the raw OCRNet probe improves prediction with unresolved selective gains. Mapping a fixed prediction pair to conditional log-loss risk produces mixed selection effects, which we examine through the components exchanged by the two rankings. These findings distinguish the performance of a fitted score from the value obtainable from its inputs and the consequences of its use. The framework provides a practical basis for deciding whether to adopt an additional score, relearn existing fields, or revise a selection rule, and for explaining the gains and failures of component-level reliability methods.

Chinese Translation

可靠的场景理解需要决定哪些预测值得信任、保留或检查。在语义分割中，组件失败预测器支持这些决策，但它们的预测性能并不能完全描述结果选择的价值。我们开发了一个以决策为中心的评估框架，连接了三个问题：可用评分预测了什么、可以从其现有领域中学习到什么，以及其排名带来了什么。在Cityscapes和ACDC的三个分割系统中，在90%的保留率下，改进后的MetaSeg在所有十二个架构-人群比较中相对于预测熵减少了保留失败，但并未在直接概率排名上建立失败优势。图像嵌套源实验揭示了一个不同的机会：在所有三个系统中，重用完整的领域集在两个学习者下改善了预测和保留损失性能，而原始的OCRNet探针在未解决的选择性增益下改善了预测。将固定的预测对映射到条件对数损失风险产生了混合选择效果，我们通过两个排名之间交换的组件来检查这些效果。这些发现区分了拟合评分的性能与其输入所能获得的价值及其使用的后果。该框架为决定是否采用额外评分、重新学习现有领域或修订选择规则提供了实用基础，并为解释组件级可靠性方法的增益和失败提供了依据。

View on arXiv Download PDF AI Translation

cs.CV / 141 / 2610.16066

PIGEONS: Incremental Long-Video Captioning with Persistent Identity Gallery and Evolving Overarching Narrative Summary

PIGEONS：基于持久身份库和演变总体叙事摘要的增量长视频字幕生成

Chen, Xinlong, Lin, Weihong, Zhang, Yuanxing, Liu, Xinyu, Ding, Yue, Li, Bozhou, Zeng, Bohan, Shi, Yang, Liu, Qiang, Wan, Pengfei, Wang, Liang, Tan, Tieniu

Abstract

Detailed video captioning has become an essential semantic interface for multi-modal understanding and generation. While existing methods mainly target short videos, extending them to long videos remains challenging, primarily due to limited context windows of current video understanding models. A naive solution is to adopt segment-wise captioning followed by concatenation, yet this paradigm often suffers from inconsistent subject references and fragmented long-range narratives. To mitigate these issues, we introduce PIGEONS, an incremental long-video captioning model built upon a Persistent Identity Gallery and an Evolving Overarching Narrative Summary. Before caption generation, PIGEONS scans the entire video to construct an identity gallery that captures key subjects along with their visual references, thereby providing a stable referential anchor. The model then performs clip-by-clip captioning conditioned on the persistent gallery and an evolving overarching narrative summary, which compactly maintains and updates essential long-range contextual information after each captioning step. To enable these capabilities, we develop a three-stage training strategy that integrates supervised fine-tuning and reinforcement learning. Extensive experiments demonstrate that PIGEONS markedly improves long-video captioning performance, achieving superior accuracy, completeness, and subject referential consistency over existing methods, while maintaining competitive performance on short-video captioning.

Chinese Translation

详细的视频字幕生成已成为多模态理解和生成的重要语义接口。尽管现有方法主要针对短视频，但将其扩展到长视频仍然具有挑战性，主要是由于当前视频理解模型的上下文窗口有限。一个简单的解决方案是采用分段字幕生成后进行拼接，但这种范式往往会遭遇不一致的主题引用和支离破碎的长距离叙事。为了解决这些问题，我们提出了PIGEONS，这是一种基于持久身份库和演变总体叙事摘要的增量长视频字幕生成模型。在生成字幕之前，PIGEONS会扫描整个视频以构建一个身份库，该库捕捉关键主题及其视觉参考，从而提供一个稳定的参考锚点。然后，该模型在持久身份库和演变总体叙事摘要的条件下逐段生成字幕，该摘要紧凑地维护和更新每个字幕生成步骤后的重要长距离上下文信息。为了实现这些功能，我们开发了一种三阶段的训练策略，结合了监督微调和强化学习。大量实验表明，PIGEONS显著提高了长视频字幕生成的性能，在准确性、完整性和主题引用一致性方面优于现有方法，同时在短视频字幕生成上保持了竞争力。

View on arXiv Download PDF AI Translation

cs.CV / 142 / 2610.16082

Regime-Aware Router Learning for Multi-Domain Medical Image Generation with Diffusion Experts

基于政权意识的路由器学习用于多领域医学图像生成与扩散专家

Ma, Yaoshuai, Yao, Wei, Liu, Yongxiang, Wang, Feifan, Chang, Da, Xu, Pengxiang, Zhang, Shixun, Tian, Yonghong

Abstract

Heterogeneous medical image data motivate domain-specialized diffusion experts, but expert specialization alone does not determine how their predictions should be combined during sampling. Learning a global router is particularly challenging when the server has limited, potentially unrepresentative images. We study two router-learning strategies for a fixed expert pool. Server-Only learns input- and timestep-dependent expert assignments from server images with component labels. Anchor-Delta (AD) uses the same images and frozen experts to obtain additional evidence from denoising losses. It distills targets that combine this evidence with the current router's assignments, while a proxy gate controls a subsequent refinement step. Component labels describe data origin; expert losses measure empirical prediction performance. We evaluate these distinct signals through routing analyses and paired medical image generation. An AD-trained router improves generation metrics over Server-Only in a heterogeneous four-domain setting, and a separate server-data study shows gains in held-out routing over a corrected-routing comparator. Additional paired evaluations show that the effect of inference-time evidence mixing varies across benchmarks. These findings establish server-data availability and expert performance as distinct considerations in learning routers for multi-domain diffusion models.

Chinese Translation

异构医学图像数据促使领域专门化的扩散专家的出现，但专家的专业化并不能决定在采样过程中如何组合他们的预测。当服务器拥有有限且可能不具代表性的图像时，学习一个全局路由器尤其具有挑战性。我们研究了两种针对固定专家池的路由器学习策略。Server-Only 从带有组件标签的服务器图像中学习依赖于输入和时间步的专家分配。Anchor-Delta (AD) 使用相同的图像和冻结的专家，从去噪损失中获取额外证据。它提炼出将这些证据与当前路由器的分配相结合的目标，同时一个代理门控控制后续的精炼步骤。组件标签描述数据来源；专家损失测量经验预测性能。我们通过路由分析和配对医学图像生成评估这些不同的信号。经过 AD 训练的路由器在异构四领域设置中改善了生成指标，相较于 Server-Only，另一个服务器数据研究显示在修正路由比较器上获得了保留路由的提升。额外的配对评估表明，推理时证据混合的效果在不同基准中有所不同。这些发现确立了服务器数据可用性和专家性能在多领域扩散模型的路由器学习中的独特考量。

View on arXiv Download PDF AI Translation

cs.CV / 143 / 2610.16118

Elephant in the Fridge: Constant-Memory Frame Packing for Long Video Understanding

冰箱里的大象：用于长视频理解的恒定内存帧打包

Gao, Shuo, Zheng, Chenhao, Ren, Zhongzheng

Abstract

Hour-long videos, streaming feeds, and movie-length content are pushing vision-language models (VLMs) past their breaking point -- the standard recipe of encoding each frame into visual tokens yields sequences that grow linearly with video length, causing prohibitive memory consumption and inference latency. Existing approaches, such as token reduction and keyframe selection, shorten the token sequence but leave the O(n) scaling intact, failing to resolve the fundamental bottleneck. We propose FramePack-U, a strikingly simple, training-free frame packing framework that compresses arbitrarily long videos into a fixed size visual representation at constant memory cost. FramePack-U requires no fine-tuning or architectural modification -- it plugs directly into off-the-shelf VLMs at inference time. Inspired by FramePack from the generation literature, we propose a content-aware compression strategy for visual understanding under a fixed token budget. Specifically, FramePack-U (1) scores each frame by its semantic similarity to the text prompt and ranks frames in a diversity-driven manner, and (2) packs the selected frames at varying resolutions into a fixed-length context window. This concentrates the memory budget on frames that are both instruction-relevant and visually diverse. While FramePack-U can in theory pack arbitrarily many frames into the fixed budget, we identify an empirical instantiation that generalizes robustly across settings. Extensive experiments on LongVideoBench, MLVU, and Video-MME show that FramePack-U consistently outperforms full-resolution baselines and prior token reduction and keyframe selection methods across diverse model families, parameter scales, and frame budgets. Notably, on Qwen3.5-35B-A3B with a 128-frame budget, FramePack-U yields gains of +12.9\%, +10.6\%, and +9.7\% over the official model on the three benchmarks, respectively.

Chinese Translation

长达一小时的视频、流媒体和电影长度的内容正在将视觉-语言模型（VLMs）推向极限——将每帧编码为视觉标记的标准方法导致序列长度与视频长度线性增长，从而造成巨大的内存消耗和推理延迟。现有的方法，如标记减少和关键帧选择，虽然缩短了标记序列，但仍然保持O(n)的扩展性，未能解决根本瓶颈。我们提出了FramePack-U，这是一种极其简单、无需训练的帧打包框架，能够以恒定的内存成本将任意长的视频压缩为固定大小的视觉表示。FramePack-U无需微调或架构修改——它可以直接插入现成的VLMs进行推理。受到生成文献中FramePack的启发，我们提出了一种在固定标记预算下用于视觉理解的内容感知压缩策略。具体而言，FramePack-U (1) 根据每帧与文本提示的语义相似性对其进行评分，并以多样性驱动的方式对帧进行排名；(2) 将选定的帧以不同分辨率打包到固定长度的上下文窗口中。这将内存预算集中在与指令相关且视觉上多样的帧上。虽然理论上FramePack-U可以将任意数量的帧打包到固定预算中，但我们识别出一种在不同设置中具有良好泛化能力的经验实例。在LongVideoBench、MLVU和Video-MME上的大量实验表明，FramePack-U在不同模型系列、参数规模和帧预算下，始终优于全分辨率基线和先前的标记减少及关键帧选择方法。值得注意的是，在具有128帧预算的Qwen3.5-35B-A3B上，FramePack-U在三个基准测试中分别获得了+12.9%、+10.6%和+9.7%的提升。

View on arXiv Download PDF AI Translation

cs.CV / 144 / 2610.16122

Geometry-Guided Semantic Reconstruction for 3D Instance Segmentation

基于几何引导的3D实例分割语义重建

Su, Yuanhao, Zhang, Shaofeng

Abstract

3D instance segmentation requires representations that are both geometrically structured and semantically discriminative. Pure 3D encoders model spatial structure well but often lack strong appearance semantics, while directly lifting 2D priors into 3D by concatenation injects rich semantics without ensuring that they align well with the underlying 3D structure. In this paper, we address this problem by formulating 2D-to-3D semantic transfer as a \textbf{masked semantic reconstruction} task. Specifically, we mask a subset of background lifted 2D priors and train the 3D encoder to recover the missing semantic priors, providing explicit supervision for distilling rich 2D semantics into the 3D backbone. We further propose \textbf{geometry-modulated semantic injection}, which uses raw 3D geometry to modulate lifted 2D priors before sparse encoding, so that geometry explicitly controls semantic injection and alleviates the distribution gap between 2D semantic features and 3D geometric features. Experiments on two baseline frameworks demonstrate that the proposed method achieves state-of-the-art results in both full-data and few-shot settings on ScanNetV2 and ScanNet200. Code is publicly available at https://github.com/yharoldsu0627/Geometry_Guided_Semantic_Reconstruction

Chinese Translation

3D实例分割需要既具有几何结构又具有语义区分能力的表示。纯3D编码器能够很好地建模空间结构，但往往缺乏强大的外观语义，而通过拼接直接将2D先验提升到3D则注入了丰富的语义，但并未确保它们与底层3D结构良好对齐。本文通过将2D到3D的语义转移形式化为 extbf{掩蔽语义重建}任务来解决这一问题。具体而言，我们掩蔽一部分背景提升的2D先验，并训练3D编码器以恢复缺失的语义先验，从而为将丰富的2D语义提炼到3D主干提供明确的监督。我们进一步提出 extbf{几何调制语义注入}，该方法在稀疏编码之前使用原始3D几何体来调制提升的2D先验，使得几何体明确控制语义注入，并缓解2D语义特征与3D几何特征之间的分布差距。在两个基线框架上的实验表明，所提出的方法在ScanNetV2和ScanNet200的全数据和少样本设置中均取得了最先进的结果。代码已公开，地址为https://github.com/yharoldsu0627/Geometry_Guided_Semantic_Reconstruction

View on arXiv Download PDF AI Translation

cs.CV / 145 / 2610.16128

A Concept-informed Interpretable Artificial Intelligence Framework for Breast Ultrasound Diagnosis

一种基于概念的可解释人工智能框架用于乳腺超声诊断

Zhong, Shengzhou, Wang, Ziying, Li, Liang, Yan, Lihua, Chen, Shuochun, Zou, Yudie, Zhang, Jifan, Li, Wenhao, Zhang, Huiling, Lu, Jiayin, Xu, Ziting, Feng, Qianjin, Shi, Zhenwei, Chen, Yan, Li, Yingjia, Zhang, Yu, Ning, Zhenyuan

Abstract

Most existing artificial intelligence systems for breast ultrasound rely primarily on image-driven prediction, with limited incorporation of the knowledge from the Breast Imaging Reporting and Data System (BI-RADS) and the intermediate assessment processes underlying radiologists' workflow, thereby constraining their interpretability and alignment with clinical practice. We curated, to our knowledge, the largest multicenter collection of breast ultrasound images paired with structured BI-RADS descriptors, comprising 9,941 patients across 20 independent datasets. Across 14 testing datasets, IBUS demonstrated consistent generalization performance across different clinical centers and imaging conditions. Incorporating BI-RADS descriptors enabled fine-grained risk stratification, while cross-checking predicted BI-RADS grades against malignancy probabilities identified internally discordant predictions that may warrant additional clinical review. In a reader study involving eight radiologists, IBUS assistance increased diagnostic sensitivity among both junior and senior readers, with the potential of guiding radiologists to make targeted revision in risk stratification. Overall, IBUS exhibits its potential as a second-stage decision-support system that complements radiologists by providing additional risk estimates and interpretable evidence for diagnostically challenging cases. The demo video of IBUS platform is publicly available at https://github.com/zsz1997/IBUS.

Chinese Translation

现有的大多数乳腺超声人工智能系统主要依赖图像驱动的预测，有限地结合了乳腺影像报告和数据系统（BI-RADS）以及放射科医生工作流程中间评估过程的知识，从而限制了它们的可解释性和与临床实践的一致性。我们整理了迄今为止最大的乳腺超声图像多中心集合，配有结构化的BI-RADS描述，共涵盖了来自20个独立数据集的9,941名患者。在14个测试数据集中，IBUS在不同临床中心和成像条件下表现出一致的泛化性能。结合BI-RADS描述使得风险分层更加细致，同时交叉检查预测的BI-RADS等级与恶性概率，识别出内部不一致的预测，这可能需要额外的临床审查。在一项涉及八名放射科医生的读者研究中，IBUS的辅助提高了初级和高级读者的诊断敏感性，有助于指导放射科医生在风险分层中进行针对性的修正。总体而言，IBUS展现出作为第二阶段决策支持系统的潜力，通过提供额外的风险估计和可解释的证据来补充放射科医生，特别是在诊断具有挑战性的病例中。IBUS平台的演示视频可在 https://github.com/zsz1997/IBUS 上公开获取。

View on arXiv Download PDF AI Translation

cs.CV / 146 / 2610.16134

ProMoDD: Dataset Distillation for 3D Human Motion via Prototype-Conditioned Diffusion Models

ProMoDD：通过原型条件扩散模型进行3D人类运动的数据集蒸馏

Wei, Yijin, Zou, Yawen, Cai, Wenqi, Li, Guang, Ge, Hangli, Hotta, Katsuya, Gu, Chunzhi, Zhang, Chao

Abstract

Dataset distillation (DD) aims to retain the training utility of a large dataset in a much smaller set of examples. For skeleton-based action recognition, this requires capturing discriminative motion variation without disrupting the geometry and temporal continuity of articulated bodies. Directly optimizing joint coordinates can violate these constraints, while plausible motion generation alone does not ensure useful training examples. We propose Prototype-Conditioned Motion Dataset Distillation (ProMoDD), which conditions a motion diffusion model on class-wise prototypes of temporally resolved recognition features. We further tailor diffusion training to motion data, enabling the generator to model both skeletal configurations and their temporal evolution. We then construct each compact training set through budget-aware hybrid selection: representative real sequences cover densely populated regions of the pooled recognition space, while selected generated candidates complement them in less-populated regions. This couples prototype-guided generation with feature-space selection, balancing the reliability of observed motions with the variation of synthesis under a fixed per-class sequence budget. Across five difficulty-stratified subsets of the single-person classes in NTU RGB+D 60, ProMoDD achieves its largest gains under tight distillation budgets, reaching 40.32% top-1 accuracy on Full49 with three sequences per class, 4.57 percentage points above class-wise k-means. These gains reflect the complete framework. The same distilled sets train ST-GCN, MS-G3D, and CTR-GCN without regeneration.

Chinese Translation

数据集蒸馏（Dataset Distillation, DD）旨在在一个更小的样本集中保留大型数据集的训练效用。对于基于骨架的动作识别，这需要捕捉具有区分性的运动变化，同时不破坏关节体的几何形状和时间连续性。直接优化关节坐标可能会违反这些约束，而单靠合理的运动生成并不能确保有用的训练样本。我们提出了原型条件运动数据集蒸馏（Prototype-Conditioned Motion Dataset Distillation, ProMoDD），该方法将运动扩散模型以时间解析的识别特征的类别原型为条件。我们进一步针对运动数据定制扩散训练，使生成器能够建模骨架配置及其时间演变。然后，我们通过预算感知的混合选择构建每个紧凑的训练集：代表性的真实序列覆盖了汇聚识别空间的密集区域，而选定的生成候选则在较少人口的区域中对其进行补充。这将原型引导生成与特征空间选择结合起来，在固定的每类序列预算下平衡观察到的运动的可靠性与合成的变化。在NTU RGB+D 60的五个难度分层的单人类别子集中，ProMoDD在严格的蒸馏预算下实现了最大的增益，在Full49上以每类三条序列达到了40.32%的Top-1准确率，比类别k-means高出4.57个百分点。这些增益反映了完整的框架。相同的蒸馏集在不重新生成的情况下训练ST-GCN、MS-G3D和CTR-GCN。

View on arXiv Download PDF AI Translation

cs.CV / 147 / 2610.16143

All-St3R SLAM: Real-time RGB Dynamic SLAM with Pixel-wise Motion Masking

All-St3R SLAM：基于像素级运动掩蔽的实时 RGB 动态 SLAM

Kim, Yewon, Seo, Paul Hongsuck, Ahn, Sang Chul

Abstract

SLAM (Simultaneous Localization and Mapping) is a core technology in autonomous systems that enables them to estimate their own position and build a map of unknown environments simultaneously. SLAM systems in dynamic environments are of great practical importance. However, existing dynamic SLAM approaches still have several limitations like over-removal and high computational costs. To address these limitations, we propose a real-time RGB Dynamic SLAM framework that precisely removes moving objects at the pixel level without relying on predefined labels or segmentation models. Our system combines pixel motion predicted by ego-motion using DUSt3R with point tracking from All-Tracker, removing only actually moving pixels by analyzing discrepancies between the two motion sources. Unlike object-level masking methods, our approach eliminates only currently moving regions, allowing temporarily static parts to be utilized for tracking. Operating at approximately 10 FPS, the system achieves real-time performance on standard 30 FPS video streams by subsampling every 3 frames. Relying solely on RGB inputs, this significantly improves upon state-of-the-art dense dynamic SLAM approaches (0.5-3 FPS), presenting a highly practical solution.

Chinese Translation

SLAM（同时定位与地图构建）是自主系统中的核心技术，使其能够同时估计自身位置并构建未知环境的地图。在动态环境中的 SLAM 系统具有重要的实际意义。然而，现有的动态 SLAM 方法仍然存在过度去除和高计算成本等多个局限性。为了解决这些问题，我们提出了一种实时 RGB 动态 SLAM 框架，该框架在像素级别上精确去除移动物体，而不依赖于预定义标签或分割模型。我们的系统结合了通过自我运动预测的像素运动（使用 DUSt3R）与来自 All-Tracker 的点跟踪，通过分析这两种运动源之间的差异，仅去除实际移动的像素。与基于物体级掩蔽的方法不同，我们的方法仅消除当前移动的区域，允许暂时静止的部分用于跟踪。该系统以约 10 FPS 的速度运行，通过每 3 帧进行下采样，在标准 30 FPS 视频流上实现实时性能。仅依赖 RGB 输入，这显著改善了最先进的稠密动态 SLAM 方法（0.5-3 FPS），提供了一种高度实用的解决方案。

View on arXiv Download PDF AI Translation

cs.CV / 148 / 2610.16146

Unravel: Recovering an Interactive 3D Scene from a Single Image

Unravel：从单幅图像恢复交互式3D场景

Shan, Hui, Luo, Keyang, Zhou, Binjia, Li, Ming, Li, Kai, Wang, Xin, Li, Zhenyang, Yang, Zesong, Chen, Weikai, Zhang, Runze, Wang, Xin, Fu, Yanwei, Chen, Zhen, Huang, Xiangru

Abstract

We present Unravel, a training-free framework that turns a single image into an interactive 3D scene, where every object is an independent and movable asset. Existing generators either produce one entangled geometry in which no object can be moved, or predict all objects in one feed-forward pass, where a wrong guess cannot be detected or repaired. Unravel instead takes a different approach: it first takes the scene apart, and then puts it back together. Starting from the input image, an agentic harness that iteratively peels the image layer by layer, and adopts point cloud as major modality for spatial relationship recovery. Specifically, a vision-language model plans which objects are fully visible, an image editor removes them and restores the occluded regions. Every edit is verified before being incorporated into the reconstruction. This process gradually turns occluded parts of the scene into direct observation, allowing objects to be reconstructed when they are best exposed rather than hallucinated from the original image alone. The scene is then rebuilt in reverse removal order, so that the supporting objects are naturally placed before the objects they support. Since a single-view image only reveals part of an object, we treat the observed points as a lower bound on its size: an asset may grow to cover its points, but can not shrink away from them. Finally, we physically settle the reconstructed scene to remove floating objects and interpenetrations, followed by an agentic audit that corrects remaining placement errors. Extensive experiments on diverse indoor scenes show that Unravel recovers more objects, places them more faithfully, and produces scenes that are more physically-plausible than existing methods.

Chinese Translation

我们提出了Unravel，一个无需训练的框架，将单幅图像转化为交互式3D场景，其中每个物体都是独立且可移动的资产。现有的生成器要么生成一个纠缠的几何体，无法移动任何物体，要么在一次前馈过程中预测所有物体，其中错误的猜测无法被检测或修复。Unravel则采取了不同的方法：它首先将场景拆解，然后再重新组合。从输入图像开始，一个智能的工具逐层剥离图像，并采用点云作为空间关系恢复的主要方式。具体而言，一个视觉-语言模型规划哪些物体是完全可见的，图像编辑器则移除这些物体并恢复被遮挡的区域。每次编辑在纳入重建之前都会经过验证。这个过程逐渐将场景中被遮挡的部分转变为直接观察，使得物体在最佳暴露时被重建，而不是仅仅从原始图像中幻觉出来。然后，场景按照逆移除顺序重建，以便支撑物体自然地放置在它们所支撑的物体之前。由于单视图图像仅揭示了物体的一部分，我们将观察到的点视为其大小的下限：一个资产可以增长以覆盖其点，但不能从中缩小。最后，我们物理地安置重建的场景，以消除漂浮物体和相互穿透，随后进行智能审计以纠正剩余的放置错误。在多样化的室内场景上的大量实验表明，Unravel恢复了更多物体，更忠实地放置它们，并生成了比现有方法更具物理 plausibility 的场景。

View on arXiv Download PDF AI Translation

cs.CV / 149 / 2610.16166

Esports World Model Dataset

电子竞技世界模型数据集

Ma, Weiyu, Gong, Zeyuan, Liu, Xuhui, Zhang, Wenxuan, Ding, Jian, Cui, Xinyu, Zhao, Jian, Elhoseiny, Mohamed

Abstract

Interactive world models must capture how multiple participants jointly change a shared environment while observing different parts of it. Competitive games make this challenge concrete: players cooperate and compete, while fog of war, team vision and camera placement determine what each can see. Modeling these interactions requires data that connect individual observations and actions to shared state and events. We introduce the Esports World Model Dataset (EWM), a collection of human gameplay spanning StarCraft II, Dota 2 and Counter-Strike 2, with 3,053 games and maps and approximately 4,880 player-view hours, including estimated hours for the ongoing CS2 extension. It includes 4,628 completed StarCraft II views, all ten perspectives for 303 Dota 2 matches, and up to ten views per round in CS2. Across these views, replay-clock alignment connects rendered video with game-native actions, structured state and events from the same match. The three games expose complementary observation conditions: private fog of war, shared team vision and first-person viewpoints. In StarCraft II, player-specific entity records further distinguish live observations from remembered snapshots and link co-observed entities across opposing views. Together with temporal alignment, these records allow researchers to distinguish legitimate information differences from disagreement about shared state. By connecting human decisions, shared dynamics and player-specific information, EWM supports research toward jointly controlled multiplayer world models and interactive environments populated by teammates and opponents learned from human play.

Chinese Translation

交互式世界模型必须捕捉多个参与者如何共同改变一个共享环境，同时观察到环境的不同部分。竞技游戏使这一挑战变得具体：玩家在合作与竞争中，战争迷雾、团队视野和摄像机位置决定了每个玩家可以看到的内容。建模这些交互需要将个体观察和行为与共享状态和事件连接起来的数据。我们介绍了电子竞技世界模型数据集（Esports World Model Dataset, EWM），这是一个涵盖《星际争霸 II》、《Dota 2》和《反恐精英 2》的真人游戏数据集，包含3,053场比赛和地图，以及大约4,880小时的玩家视角时长，包括对正在进行的CS2扩展的估计时长。数据集包括4,628个完成的《星际争霸 II》视角、303场《Dota 2》比赛的所有十个视角，以及CS2中每轮最多十个视角。在这些视角中，重放时钟对齐将渲染视频与游戏原生动作、来自同一场比赛的结构化状态和事件连接起来。这三款游戏展现了互补的观察条件：私人战争迷雾、共享团队视野和第一人称视角。在《星际争霸 II》中，特定玩家的实体记录进一步区分了实时观察与记忆快照，并链接了对立视角中的共同观察实体。结合时间对齐，这些记录使研究人员能够区分合法的信息差异与对共享状态的分歧。通过连接人类决策、共享动态和特定玩家信息，EWM支持朝着共同控制的多人世界模型和由队友与对手组成的交互环境的研究，这些环境是从人类游戏中学习而来的。

View on arXiv Download PDF AI Translation

cs.CV / 150 / 2610.16170

Geometric Parrots: Rule Absence vs. Rule Fragility in Spatial and Physical Reasoning for Vision-Language Models

几何鹦鹉：视觉-语言模型中的规则缺失与规则脆弱性在空间和物理推理中的对比

Jawaid, Bilal, Hanif, Hamza, Uddin, Muhammad Maaz, Zafar, Anas, Alqurashi, Inad, Zafar, Aizan, Farooqui, Osama, Khan, Sheheryar, Sadak, Ferhat, Raza, Shaina, Qureshi, Rizwan

Abstract

Vision-language models (VLMs) can answer many geometric and physical reasoning questions correctly, yet still behave inconsistently in interaction, for example, abandoning a correct answer when confronted with a confident but incorrect user assertion. Standard single-turn accuracy cannot distinguish such capitulations from beneficial corrections because it scores each response in isolation. We introduce GRIP, a controlled benchmark with 34 procedurally generated reasoning domains, 100,000 images, and 600,000 questions spanning geometric, spatial, visual, and physical reasoning. Each scene is generated by deterministic code, and every answer is derived directly from the underlying scene state, providing exact ground truth without human annotation. On GRIP, we propose Dual-Loop Evaluation. A closed loop probes each image with a five-step compositional question ladder to localize where reasoning fails. An open loop then re-elicits the model's answer under controlled challenges, including peer-model responses and unsupported user assertions. Because ground truth is known exactly, answer changes can be signed as corrections (wrong-to-right) or capitulations (right-to-wrong), separating two failure modes: rule absence and rule fragility. The dataset is available at https://huggingface.co/datasets/bilaljawaid980/GRIP-Benchmark, with further details at https://bilaljawaid980.github.io/grip-benchmark/.

Chinese Translation

视觉-语言模型（VLMs）能够正确回答许多几何和物理推理问题，但在交互中仍然表现出不一致性，例如，在面对自信但错误的用户主张时放弃正确答案。标准的单轮准确性无法区分这种屈服与有益的纠正，因为它对每个响应的评分是孤立的。我们引入了GRIP，一个受控基准，包含34个程序生成的推理领域、100,000张图像和600,000个问题，涵盖几何、空间、视觉和物理推理。每个场景由确定性代码生成，每个答案直接来源于基础场景状态，提供了没有人工标注的精确真实值。在GRIP上，我们提出了双循环评估。闭环通过五步组合问题梯度探测每张图像，定位推理失败的地方。开放循环则在受控挑战下重新引导模型的答案，包括同行模型的响应和不支持的用户主张。由于真实值是精确已知的，答案变化可以被标记为纠正（错误到正确）或屈服（正确到错误），从而区分两种失败模式：规则缺失和规则脆弱性。数据集可在 https://huggingface.co/datasets/bilaljawaid980/GRIP-Benchmark 获取，更多细节见 https://bilaljawaid980.github.io/grip-benchmark/。

View on arXiv Download PDF AI Translation

cs.CV / 151 / 2610.16172

EventLens: Event-Structure Reinforcement Learning for Video Understanding

EventLens：用于视频理解的事件结构强化学习

Zhang, Guowen, Xu, Boshen, Yue, Zihao, Wang, Ziheng, Liu, Xiaokun, Tao, Xin, Qin, Wenyu, Wan, Pengfei, Jin, Qin

Abstract

Recent multimodal large language models (MLLMs) achieve strong results on video question answering, yet often fail to recover the temporal structure that makes a video coherent. They may merge neighboring events, miss single-event progression, or infer implausible relations across events. We study structured temporal understanding as the ability to recover event structure from visual evidence, including event boundaries, single-event progression, and multi-event relations. We argue that this gap arises from a mismatch between current post-training objectives and temporal reasoning, where caption-style supervision allows models to rely on language shortcuts rather than visual temporal grounding. To address this, we propose EventLens, a vision-centric reinforcement learning framework that learns video understanding through event-structure recovery. Instead of reconstructing captions, EventLens trains models to recover temporal structure under controlled perturbations, such as altered temporal granularity, reversed progression, and shuffled event order. We instantiate this framework with a three-level temporal ontology and derive three verifiable task families: event segmentation, progression discrimination, and multi-event relation reasoning. These tasks admit deterministic rewards, enabling task-specialized GRPO training without human preference labels or learned reward models. We further introduce multi-teacher on-policy distillation (MT-OPD) to consolidate specialized policies into a unified model. Experiments on Qwen3-VL backbones show consistent improvements on temporal grounding and event-sensitive reasoning benchmarks, while preserving general video-QA performance. Learning curves further demonstrate that EventLens achieves comparable downstream transfer with significantly reduced training cost compared to direct mixed RL optimization.

Chinese Translation

近期的多模态大型语言模型（MLLMs）在视频问答方面取得了显著成果，但往往无法恢复视频的时间结构，使其连贯性受到影响。它们可能会合并相邻事件、遗漏单一事件的进展，或推断事件之间不合理的关系。我们研究结构化时间理解的能力，即从视觉证据中恢复事件结构，包括事件边界、单一事件进展和多事件关系。我们认为，这一差距源于当前后训练目标与时间推理之间的不匹配，其中标题风格的监督使模型依赖语言捷径，而非视觉时间基础。为了解决这一问题，我们提出了EventLens，一个以视觉为中心的强化学习框架，通过事件结构恢复来学习视频理解。EventLens并不重建标题，而是训练模型在受控扰动下恢复时间结构，例如改变时间粒度、反向进展和打乱事件顺序。我们用三层时间本体实例化该框架，并推导出三个可验证的任务家族：事件分割、进展辨别和多事件关系推理。这些任务允许确定性奖励，使得任务专用的GRPO训练无需人工偏好标签或学习奖励模型。我们进一步引入多教师在线蒸馏（MT-OPD），将专用策略整合为统一模型。在Qwen3-VL骨干网络上的实验表明，在时间基础和事件敏感推理基准上取得了一致的改善，同时保持了总体视频问答的性能。学习曲线进一步表明，与直接混合强化学习优化相比，EventLens在显著降低训练成本的同时，实现了可比的下游迁移效果。

View on arXiv Download PDF AI Translation

cs.CV / 152 / 2610.16176

EgoAlign: Situated 3D Understanding without Situation Descriptions

EgoAlign：无需情境描述的情境化3D理解

Li, Yiyang, Ma, Tianyi, Wang, Zehong, Ye, Yanfang

Abstract

Situated 3D Question Answering requires models to interpret a 3D scene relative to an agent's pose. Existing methods commonly convey that pose through a natural-language situation description, which may be unavailable or costly to obtain in real-world deployments. While some methods additionally incorporate numerical pose, we find that randomly shuffling the agent pose causes little performance degradation, indicating limited functional dependence on the numerical pose. To address this gap, we introduce EgoAlign, a simple 3D language model that represents a scene in an agent-centric coordinate frame, together with a dedicated egocentric alignment stage comprising four complementary tasks: object category recognition, egocentric direction prediction, metric distance estimation, and spatial relation prediction. This stage teaches the model to consistently interpret scene geometry relative to the agent's pose before task-specific fine-tuning. EgoAlign achieves the strongest performance on SQA3D, Real-3DQA, and MSQA among compared methods while remaining highly efficient. More importantly, removing textual situation descriptions yields comparable or better downstream performance, whereas shuffling EgoAlign's numerical pose causes substantial degradation. These results show that EgoAlign establishes agent-centric spatial context directly from numerical pose without requiring textual situation descriptions.

Chinese Translation

情境化3D问答要求模型相对于代理的姿态解释3D场景。现有方法通常通过自然语言情境描述传达该姿态，而在实际应用中，这种描述可能不可用或获取成本高昂。虽然一些方法还结合了数值姿态，但我们发现随机打乱代理姿态对性能的影响很小，表明对数值姿态的功能依赖有限。为了解决这一问题，我们提出了EgoAlign，这是一种简单的3D语言模型，它在代理中心坐标系中表示场景，并包含一个专门的自我中心对齐阶段，该阶段由四个互补任务组成：物体类别识别、自我中心方向预测、度量距离估计和空间关系预测。该阶段教会模型在任务特定微调之前，始终相对于代理的姿态解释场景几何。EgoAlign在SQA3D、Real-3DQA和MSQA等比较方法中表现最强，同时保持高效性。更重要的是，去除文本情境描述后，模型的下游性能可比或更好，而打乱EgoAlign的数值姿态则会导致显著降级。这些结果表明，EgoAlign能够直接从数值姿态建立代理中心的空间上下文，而无需文本情境描述。

View on arXiv Download PDF AI Translation

cs.CV / 153 / 2610.16177

DIGS: Distribution-Informed Gaussian Splatting for Training-Free Open-Vocabulary 3D Segmentation

DIGS：基于分布信息的高斯点云渲染用于无训练开放词汇的3D分割

Zeng, Zhifan, Xiao, Wenbo, Sowmya, Arcot, Sun, Changming

Abstract

Open-vocabulary 3D Gaussian segmentation typically binds semantic features to the per-Gaussian state via per-scene gradient training (1-4 h) or the more recent closed-form distillation. Both commit each primitive to a single feature; therefore, boundary Gaussians spanning multiple 2D instances are forced into a hard decision that degrades retrieval accuracy and yields unstable mask boundaries at render time. We present DIGS, a training-free framework with two stages, class-agnostic mapping and multi-modal encoding, that decouples instance discovery from linguistic naming. At the Gaussian level, our distribution-informed representation maintains an explicit top-$M$ label distribution per primitive, updated by a tempered Dirichlet rule; preservation of multi-peaked label distributions at boundary primitives allows a late-argmax renderer to recover clean instance boundaries. A bidirectional state machine aggregates these distributions into a persistent class-agnostic instance ontology. The instance-level encoding stage then fuses the CLIP visual embedding of each instance's isolated 3D footprint with the CLIP-text embedding of an LLM-generated attribute description, disambiguating visually similar instances. On a single A100, DIGS completes mapping and instance encoding on LERF-OVS in 1.0-3.2 min per scene, depending on whether LLM fusion is enabled, and serves native 3D queries in 5.4 ms. It reaches 71.95% mean mIoU on LERF-OVS, outperforming both the training-free SFS and training-based LaGa baselines. DIGS also achieves the highest mean mIoU on LERF-Mask and 3D-OVS and surpasses Gaga and ObjectGS on Replica under their respective evaluation protocols. Code is available at https://github.com/ZephyrZos/DIGS.

Chinese Translation

开放词汇的3D高斯分割通常通过每场景的梯度训练（1-4小时）或最近的闭式蒸馏将语义特征绑定到每个高斯状态上。这两者都将每个原始元素固定为单一特征；因此，跨越多个2D实例的边界高斯被迫做出硬决策，这降低了检索精度并在渲染时产生不稳定的掩膜边界。我们提出了DIGS，一个无训练的框架，分为两个阶段：无类别映射和多模态编码，解耦了实例发现与语言命名。在高斯级别，我们的基于分布的信息表示为每个原始元素维护一个明确的前$M$标签分布，通过温和的狄利克雷规则进行更新；在边界原始元素处保留多峰标签分布使得后期最大值渲染器能够恢复干净的实例边界。一个双向状态机将这些分布聚合成一个持久的无类别实例本体。实例级编码阶段随后将每个实例孤立的3D足迹的CLIP视觉嵌入与LLM生成的属性描述的CLIP文本嵌入融合，从而消除视觉上相似实例的歧义。在单个A100上，DIGS在LERF-OVS上完成映射和实例编码的时间为每场景1.0-3.2分钟，具体取决于LLM融合是否启用，并在5.4毫秒内提供原生3D查询。它在LERF-OVS上达到了71.95%的平均mIoU，超越了无训练的SFS和基于训练的LaGa基线。在LERF-Mask和3D-OVS上，DIGS也达到了最高的平均mIoU，并在各自的评估协议下超越了Gaga和ObjectGS。代码可在 https://github.com/ZephyrZos/DIGS 获取。

View on arXiv Download PDF AI Translation

cs.CV / 154 / 2610.16204

VGGA: Geometry-Grounded Autoencoder for Video Generation

VGGA：基于几何的自编码器用于视频生成

Gan, Chaofan, Zhao, Yian, Li, Shijie, Chen, Tieyuan, Zhao, Zicheng, Harandi, Mehrtash, Lin, Weiyao

Abstract

Autoencoders are the core component of latent video generative models, compressing high-dimensional videos into compact latent spaces for downstream generation. However, existing video autoencoders are primarily optimized for pixel-level reconstruction, providing limited supervision for learning structured latent representations. We argue that videos should not be treated merely as signals to be reconstructed, but rather as 2D projections of dynamic 3D worlds that exhibit explicit and well-structured geometric properties. Building on this insight, we propose \textbf{V}ideo \textbf{G}eometry \textbf{G}rounded \textbf{A}utoencoder (\textbf{VGGA}), a geometry-aware video autoencoder that explicitly incorporates geometric constraints into the reconstruction process. Specifically, VGGA grounds the learned representations along three complementary geometric dimensions: \emph{Structural Depth}, \emph{Motion Correspondence}, and \emph{Projective Warping}. These objectives are unified within a Video Geometry Decoder, which employs a shared spatio-temporal upsampling trunk to jointly optimize geometric learning and RGB reconstruction. Through these geometry-grounded objectives, VGGA encourages the latent representations to capture the underlying 3D structure and motion of dynamic scenes. VGGA consistently improves video generation across both geometry-centric and human-motion domains, reducing gFVD by 28.1\% on RealEstate10K and 37.7\% on Taichi-HD. Notably, these substantial generation improvements are achieved while maintaining comparable reconstruction quality, demonstrating that geometry-grounded representation learning yields a more effective latent space for video generation.

Chinese Translation

自编码器是潜在视频生成模型的核心组成部分，它将高维视频压缩为紧凑的潜在空间，以便进行后续生成。然而，现有的视频自编码器主要针对像素级重建进行优化，提供的监督信息有限，无法有效学习结构化的潜在表示。我们认为，视频不应仅被视为需要重建的信号，而应被视为动态三维世界的二维投影，这些世界展现出明确且结构良好的几何特性。基于这一观点，我们提出了 extbf{V}ideo extbf{G}eometry extbf{G}rounded extbf{A}utoencoder（ extbf{VGGA}），这是一种具有几何感知的视频自编码器，明确将几何约束纳入重建过程。具体而言，VGGA在三个互补的几何维度上对学习的表示进行约束： extit{结构深度}、 extit{运动对应}和 extit{投影扭曲}。这些目标在一个视频几何解码器中统一，该解码器采用共享的时空上采样主干，以共同优化几何学习和RGB重建。通过这些基于几何的目标，VGGA鼓励潜在表示捕捉动态场景的基本三维结构和运动。VGGA在几何中心和人类运动领域的一致性视频生成上均有所提升，在RealEstate10K数据集上将gFVD降低了28.1 ext{%}，在Taichi-HD数据集上降低了37.7 ext{%}。值得注意的是，这些显著的生成改进是在保持可比重建质量的同时实现的，证明了基于几何的表示学习为视频生成提供了更有效的潜在空间。

View on arXiv Download PDF AI Translation

cs.CV / 155 / 2610.16212

MP-MNCA: Morphology-Prior Masked Neighbor Cross-Attention for Efficient Spatial Domain Identification

MP-MNCA：基于形态先验的掩蔽邻域交叉注意力框架用于高效空间域识别

Chitra, Sidharthan Sivacoumar, Manivannan, Sanjeev

Abstract

Spatial transcriptomics jointly captures molecular state, tissue morphology, and spatial organization, yet spatial-domain methods are typically compared only at late training checkpoints, obscuring differences in convergence efficiency. We introduce MP-MNCA, a morphology-prior masked neighbor cross-attention framework that preserves the full 3,000 dimensional highly-variable-gene representation while selectively routing molecular information across spatial neighbors, with histological similarity incorporated as a continuous attention prior under a self-supervised masked neighbor-contrastive objective. Across 12 heterogeneous human dorsolateral prefrontal cortex (DLPFC) sections, MP-MNCA is evaluated against established spatial-domain baselines using both convergence trajectories and epoch-wise spatial-domain visualizations. Our results show that MP-MNCA (i) reaches coherent spatial-domain structure in substantially fewer training epochs, as reflected quantitatively by convergence curves and qualitatively by progressive spatial grids, and (ii) maintains strong clustering quality, achieving a mean peak refined ARI of 0.546 versus 0.549 for the strongest baseline (STAIG), while attaining the highest mean refined NMI of 0.657 versus 0.640. Moreover, MP-MNCA outperforms STAIG on 7 of 12 sections in ARI and 11 of 12 in NMI, demonstrating faster convergence without sacrificing spatial-domain quality.

Chinese Translation

空间转录组学共同捕捉分子状态、组织形态和空间组织，然而空间域方法通常仅在训练后期进行比较，掩盖了收敛效率的差异。我们提出了MP-MNCA，一种基于形态先验的掩蔽邻域交叉注意力框架，该框架在选择性地跨空间邻域传递分子信息的同时，保留了完整的3000维高变基因表示，并在自监督的掩蔽邻域对比目标下，将组织学相似性作为连续注意力先验纳入其中。在12个异质的人类背外侧前额叶皮层（DLPFC）切片中，MP-MNCA与已建立的空间域基线进行了评估，使用了收敛轨迹和每个训练周期的空间域可视化。我们的结果表明，MP-MNCA (i) 在显著更少的训练周期内达到了连贯的空间域结构，这在定量上通过收敛曲线反映，定性上通过逐步空间网格体现，以及 (ii) 维持了强大的聚类质量，实现了平均峰值精炼调整Rand指数（ARI）为0.546，相较于最强基线（STAIG）的0.549，同时获得了最高的平均精炼归一化互信息（NMI）为0.657，相较于0.640。此外，MP-MNCA在12个切片中有7个在ARI上优于STAIG，在11个切片中在NMI上优于STAIG，展示了更快的收敛速度而不牺牲空间域质量。

View on arXiv Download PDF AI Translation

cs.CV / 156 / 2610.16224

Look Back to Move Forward: Learning to Revisit for Agentic Compositional Image Generation

回顾以前进：学习重新访问以实现自主组合图像生成

Wang, Zixu, Zhao, Xiaoran, Liu, Yahui, Wang, Feng-ao, Xiao, Junhao, Shu, You, Liu, Bing, Liu, Yiting, Jian, Songlei, Tian, Zhiliang, Li, Dongsheng

Abstract

Text-to-image models have made remarkable progress in visual fidelity and instruction following, yet complex compositional generation remains challenging. Existing agentic approaches address this limitation through iterative generation and editing, adapting their actions based on interaction history. However, they typically fix the continuation state (i.e., the image from which refinement proceeds) to the latest image. Since iterative editing is not necessarily monotonic, the agent may continue refining from the latest image even when it has already degraded. In this work, we introduce GenRevisit, an agentic framework that extends action selection to joint continuation-state and action selection. The agent can not only use interaction history to adjust editing instructions but also select an image candidate to edit, jointly deciding where to continue from and how to refine. We train GenRevisit with SFT followed by agentic reinforcement learning. To address the coarse credit assignment of final outcome reward in GRPO, which assigns the same trajectory-level advantage to all actions, we further propose Normalized Action Credit Assignment (NACA) to redistribute positive advantage according to each action's contribution to reducing the remaining objective. On GenEval2, GenRevisit achieves GM scores of 84.3 and 74.4 with Qwen-Image-2512 and Z-Image-Turbo, surpassing the strongest baselines using the same generators by 12.7 and 10.9 points respectively. GenRevisit achieves state-of-the-art performance on GenEval, GenEval++, and GenEval2 among baselines.

Chinese Translation

文本到图像模型在视觉逼真度和指令遵循方面取得了显著进展，但复杂的组合生成仍然具有挑战性。现有的自主方法通过迭代生成和编辑来解决这一限制，根据交互历史调整其行为。然而，它们通常将继续状态（即从中进行细化的图像）固定为最新图像。由于迭代编辑不一定是单调的，代理可能会继续从最新图像进行细化，即使该图像已经退化。在本研究中，我们引入了GenRevisit，一个自主框架，它将动作选择扩展到联合继续状态和动作选择。代理不仅可以利用交互历史来调整编辑指令，还可以选择要编辑的图像候选，联合决定从哪里继续以及如何进行细化。我们通过SFT训练GenRevisit，随后进行自主强化学习。为了解决GRPO中最终结果奖励的粗略信用分配问题，该方法将相同的轨迹级优势分配给所有动作，我们进一步提出了归一化动作信用分配（NACA），根据每个动作对减少剩余目标的贡献重新分配正优势。在GenEval2上，GenRevisit在使用Qwen-Image-2512和Z-Image-Turbo时分别取得了84.3和74.4的GM分数，超越了使用相同生成器的最强基线，分别提高了12.7和10.9分。GenRevisit在基线中在GenEval、GenEval++和GenEval2上达到了最先进的性能。

View on arXiv Download PDF AI Translation

cs.CV / 157 / 2610.16234

Seg2Click: Learning Pixel-Level Action Masks for GUI Visual Grounding

Seg2Click：学习用于GUI视觉定位的像素级动作掩码

Wang, Jingchao, Huang, Dingjiang

Abstract

Visual grounding is a central challenge for GUI agents, requiring an instruction to be mapped to a target within a crowded screenshot. Coordinate generation with single-point supervision reduces a spatially extended target to one position, despite the presence of many valid click locations. Region-based patch scoring alleviates this ambiguity, but its spatial decisions remain tied to visual-token footprints that may span the target, neighboring controls, and background. To address these limitations, we propose Seg2Click, which reformulates GUI grounding as executable-region prediction: identifying where a click can validly execute the given instruction. Using bounding-box supervision, Seg2Click predicts an executable-region mask and computes the final click point through a deterministic rule. This formulation captures the spatial extent of valid click locations and enables finer localization beyond the visual-token grid, making it well suited to small, densely arranged controls without requiring a verifier or zoom-in pass. Experiments across five backbone configurations from two model families demonstrate the effectiveness of Seg2Click. Seg2Click-9B achieves 92.06\%, 94.10\%, and 55.22\% on ScreenSpot, ScreenSpot-v2, and ScreenSpot-Pro, respectively, while matched-backbone comparisons show improvements over both coordinate generation and patch-based grounding. Code will be available at https://github.com/jcwang0602/Seg2Click.

Chinese Translation

视觉定位是GUI代理面临的一个核心挑战，需要将指令映射到拥挤截图中的目标。单点监督的坐标生成将空间扩展的目标简化为一个位置，尽管存在许多有效的点击位置。基于区域的补丁评分缓解了这种模糊性，但其空间决策仍然与可能跨越目标、相邻控件和背景的视觉标记足迹相关。为了解决这些局限性，我们提出了Seg2Click，它将GUI定位重新构建为可执行区域预测：识别可以有效执行给定指令的点击位置。通过边界框监督，Seg2Click预测可执行区域掩码，并通过确定性规则计算最终点击点。这种表述捕捉了有效点击位置的空间范围，并使得在视觉标记网格之外进行更精细的定位成为可能，适用于小型、密集排列的控件，而无需验证器或放大处理。针对来自两个模型系列的五种主干配置的实验表明，Seg2Click的有效性。Seg2Click-9B在ScreenSpot、ScreenSpot-v2和ScreenSpot-Pro上的准确率分别达到了92.06%、94.10%和55.22%，而匹配主干的比较显示出相对于坐标生成和基于补丁的定位的改进。代码将发布在https://github.com/jcwang0602/Seg2Click。

View on arXiv Download PDF AI Translation

cs.CV / 158 / 2610.16252

MFlowHOI: Whole-Body Human-Object Interaction Generation via Part-Aware MeanFlow

MFlowHOI：通过部件感知的均值流生成全身人机交互

He, Qiang, Zhang, Yonghao, Zhang, Yinda, Meng, Chenyu, Ouyang, Liufei, Hu, Haoxiang, Gao, Cangjun, Deng, Xiaoming, Ma, Cuixia, Wang, Hongan

Abstract

Whole-body human-object interaction (HOI) generation requires coordinating body and finger motion, object trajectories, and time-varying contacts. Existing methods typically rely on iterative diffusion or test-time optimization, resulting in high inference costs. Direct one-step generation is highly desired in downstream applications due to efficient inference. However, it is challenging because these HOI components differ substantially in structure and scale. We present MFlowHOI, an efficient one-step framework for whole-body HOI generation. A Contact-Aware Multipart RVQVAE organizes body, hand, object, and contact signals into a structured latent space, with a shared decoder preserving their global coordination. An ST-DiT then models temporal and cross-part dependencies in this space and generates a complete interaction in a single MeanFlow evaluation. Predicted object-local contact correspondences further enable geometric refinement and interaction-preserving transfer to photorealistic 3D Gaussian avatars with different body proportions. Experiments demonstrate that MFlowHOI outperforms existing base generators in motion quality and text alignment while substantially reducing inference time.

Chinese Translation

全身人机交互（HOI）生成需要协调身体和手指运动、物体轨迹以及时间变化的接触。现有方法通常依赖于迭代扩散或测试时优化，导致高推理成本。由于高效推理，直接一步生成在下游应用中非常受欢迎。然而，由于这些HOI组件在结构和规模上存在显著差异，这一过程具有挑战性。我们提出了MFlowHOI，这是一个高效的一步框架用于全身HOI生成。接触感知的多部件RVQVAE将身体、手、物体和接触信号组织到一个结构化的潜在空间中，共享解码器保持它们的全局协调。接着，ST-DiT在该空间中建模时间和跨部件依赖关系，并在一次均值流评估中生成完整的交互。预测的物体局部接触对应关系进一步实现几何细化，并将交互保持转移到具有不同身体比例的逼真3D高斯头像。实验表明，MFlowHOI在运动质量和文本对齐方面优于现有的基础生成器，同时显著减少了推理时间。

View on arXiv Download PDF AI Translation

cs.CV / 159 / 2610.16256

EgoWholeView: Camera-Only Head-Mounted Capture of Body and Bimanual Hand Motion for Embodied Demonstrations

EgoWholeView：仅使用相机的头戴式身体和双手运动捕捉系统，用于具身演示

Hu, Yaoqing, He, Xuguo, Cui, Daohan, Shi, Yukun, Chen, Kai

Abstract

Head-mounted body and bimanual motion capture supports self-avatars, teleoperation, and embodied demonstrations. However, a single egocentric camera is vulnerable to a limited field of view, scale ambiguity, and hand-object occlusion. We present EgoWholeView, a wearable capture system of six synchronized wide-angle RGB cameras with a role-specialized pipeline. The outward views localize the rig in a world frame, while an overlapping downward pair independently predicts SMPL-X body and MANO hand parameters. Both predictors are adapted to egocentric fisheye data. To combine the two, reliability cues measured from the predictions themselves weight the fusion of body and hand joint rotations, while shape is aggregated across views and frames. The pair's baseline fixes metric scale, and triangulation recovers root translation. Relative to monocular prediction, fusion reduces PA-MPJPE by 3.7% for the body and 11.2% for the hands, and temporal refinement lowers angular-acceleration magnitude by 32.2% and 31.0%, respectively. On our six-camera recording, the pipeline produces body and hand states in a shared world frame using benchmark-fitted gates without retuning, and temporal refinement lowers hand acceleration magnitude by 41.08%. Using only RGB cameras and no third-person view, EgoWholeView offers a low-cost, portable route to egocentric interaction data.

Chinese Translation

头戴式身体和双手运动捕捉支持自我化身、远程操作和具身演示。然而，单一的自我中心相机容易受到视野有限、尺度模糊和手-物体遮挡的影响。我们提出了EgoWholeView，一种可穿戴的捕捉系统，包含六个同步的广角RGB相机，并配备了角色专用的处理流程。外部视角在世界坐标系中定位设备，而重叠的向下视角独立预测SMPL-X身体和MANO手部参数。这两个预测器均适应自我中心的鱼眼数据。为了将两者结合起来，从预测结果中测量的可靠性线索对身体和手部关节旋转的融合进行加权，而形状则在视角和帧之间进行聚合。该对的基线固定了度量尺度，三角测量恢复了根部平移。与单目预测相比，融合在身体上减少了3.7%的PA-MPJPE，在手部上减少了11.2%；而时间精细化分别降低了32.2%和31.0%的角加速度幅度。在我们的六摄像头录制中，该流程使用基准拟合的门控在共享世界坐标系中生成身体和手部状态，无需重新调优，时间精细化将手部加速度幅度降低了41.08%。仅使用RGB相机且没有第三人称视角，EgoWholeView提供了一种低成本、便携的自我中心交互数据获取途径。

View on arXiv Download PDF AI Translation

cs.CV / 160 / 2610.16272

Graph3R: Unifying View Graph Initialization with Feed-Forward Reconstruction

Graph3R：将视图图初始化与前馈重建统一起来

Yu, Jingwen, Wang, Hengli, Jiao, Jianhao, Liu, Zhonghang, Tan, Ping, Zhang, Hong

Abstract

Feed-forward reconstruction recovers camera poses and dense geometry from image sets whose views are assumed to come from a single scene. In practice, uncurated image collections often contain views from disjoint scenes, and reconstructing these views jointly can corrupt camera poses and geometry. Existing pipelines typically resolve view associations in separate stages before final reconstruction, leaving view graph initialization and reconstruction inference only loosely coupled. We introduce Graph3R, which unifies view graph initialization with feed-forward reconstruction through a shared geometric representation. A lightweight verification head predicts pairwise covisibility from the multi-view features of $\pi^3$ to construct a view graph. This graph classifies the input views into scene groups. For each group, we mask the shared features before decoding camera poses and dense geometry. Reusing the multi-view features allows Graph3R to reconstruct disjoint scenes separately with a single backbone pass. Across three benchmarks, Graph3R improves average maximum recall at 100% precision by upto 8.99 percentage points over the baseline, while improving reconstruction quality on mixed-scene inputs. Regarding the multi-session mapping task, Graph3R improves cross-session localization and achieves competitive pose estimation accuracy, while delivering a runtime speedup of approximately $35.9\times$ over the baseline when reconstructing the query group. By integrating learned view association with geometric inference, our approach takes a step toward end-to-end reconstruction from uncurated image collections.

Chinese Translation

前馈重建从假设来自单一场景的图像集恢复相机姿态和密集几何。在实际应用中，未经整理的图像集合通常包含来自不同场景的视图，联合重建这些视图可能会破坏相机姿态和几何。现有的处理流程通常在最终重建之前在不同阶段解决视图关联，使得视图图初始化和重建推断之间的耦合较为松散。我们提出了Graph3R，通过共享几何表示将视图图初始化与前馈重建统一起来。一个轻量级的验证头从$ ext{π}^3$的多视图特征中预测成对的共视性，以构建视图图。该图将输入视图分类为场景组。对于每个组，我们在解码相机姿态和密集几何之前屏蔽共享特征。重用多视图特征使得Graph3R能够通过单个主干传递分别重建不相交的场景。在三个基准测试中，Graph3R在100%精度下提高了平均最大召回率，较基线提高了最多8.99个百分点，同时改善了混合场景输入的重建质量。在多会话映射任务中，Graph3R改善了跨会话定位，并实现了具有竞争力的姿态估计精度，同时在重建查询组时提供了约$35.9 imes$的运行时加速。通过将学习的视图关联与几何推断相结合，我们的方法朝着从未经整理的图像集合进行端到端重建迈出了重要一步。

View on arXiv Download PDF AI Translation

cs.CV / 161 / 2610.16277

FlowAvatar: Real-Time Full-Body Avatars from Sparse Egocentric Inputs on Consumer XR Devices

FlowAvatar：基于消费者XR设备的稀疏自我中心输入实时全身虚拟形象

Gao, Chenfeng, Yeon, Taeyoung, Park, Sungheon, Xu, Vasco, Prabhu, Anish, Gonzalez-Franco, Mar, Ahuja, Karan

Abstract

Prior avatar embodiment research has studied tracking completeness (how faithfully the body, hands, and face are sensed and reproduced) and personalized appearance in isolation, each requiring specialized lab equipment. How these factors interact on consumer hardware remains an open question, because no existing system unifies body pose, face, hands, and personalized appearance on a standalone headset. We present FlowAvatar, a system enabling real-time personalized look-alike avatars with full-body pose, hand articulation, facial expression, and eye gaze tracking from sparse egocentric inputs on consumer XR devices (Meta Quest Pro). Our pose model achieves competitive spatial accuracy and state-of-the-art temporal consistency on the AMASS benchmark against prior works with comprehensive spatial and temporal losses validated through architecture and loss ablations, while running in real-time. Two user studies (N=12, N=10) provide converging evidence that tracking completeness and personalized appearance serve complementary roles: tracking completeness broadly improves perceived motion quality, while only their combination achieves positive total embodiment.

Chinese Translation

以往的虚拟形象体现研究分别研究了跟踪完整性（身体、手部和面部的感知与再现的忠实度）和个性化外观，每个研究都需要专门的实验室设备。在消费者硬件上，这些因素如何相互作用仍然是一个未解之谜，因为目前没有现有系统能够在独立头戴设备上统一身体姿势、面部、手部和个性化外观。我们提出了FlowAvatar，一个系统，能够从消费者XR设备（Meta Quest Pro）上的稀疏自我中心输入中实时生成个性化的类人虚拟形象，具备全身姿势、手部动作、面部表情和眼动跟踪。我们的姿势模型在AMASS基准测试中实现了与先前工作相媲美的空间准确性和最先进的时间一致性，通过架构和损失消融验证了全面的空间和时间损失，同时能够实时运行。两项用户研究（N=12，N=10）提供了趋同证据，表明跟踪完整性和个性化外观发挥互补作用：跟踪完整性广泛提高了感知运动质量，而只有两者的结合才能实现积极的整体体现。

View on arXiv Download PDF AI Translation

cs.CV / 162 / 2610.16278

DepthTile: Depth-Guided Existing-Image Self-Tiling with Input-Side Tileability Assessment

DepthTile：基于深度引导的现有图像自平铺与输入端平铺能力评估

Wang, Rui, Chen, Xi, Dai, Hui, Yu, Piaopiao, Hao, Yixue, Hu, Long, Chen, Min

Abstract

Existing-image self-tiling aims to transform a given image into a seamlessly repeatable tile while preserving its original semantics and local details. This task requires weak editing to retain the source but leave periodic seams, whereas strong editing improves tileability at the large risk of content drift.Existing diffusion-based tiling pipelines alleviate boundary discontinuities, yet lack awareness of which regions are geometrically important to preserve. In this work, we propose DepthTile, a depth-guided self-tiling method whose generation module requires no retraining of the diffusion backbone. It converts monocular depth into a spatial preservation prior and integrates it into the timestep-dependent masking mechanism of Tiled Diffusion. DepthTile converts depth into a spatial editability map and fuses it with the differential diffusion mask. This strengthens source preservation in near-field regions indicated by the monocular depth prior while retaining editing flexibility near compatible background boundaries. To access when self-tiling is feasible, we introduce the Self-Tiling Ability Score (STAS), a lightweight input-side predictor based on RGB, gradient, depth, and depth-gradient cues. On a 335-image benchmark, relative to the default Tiled Diffusion pipeline, DepthTile reduces MAE, LPIPS, and Tiling Score by 12.8%, 26.7%, and 8.4%, respectively. STAS reliably enables difficulty-aware analysis and further supports compatibility-aware pairwise image tiling.

Chinese Translation

现有图像自平铺旨在将给定图像转化为无缝可重复的平铺，同时保留其原始语义和局部细节。该任务需要进行弱编辑以保留源图像，但会留下周期性接缝，而强编辑则在内容漂移的高风险下提高平铺能力。现有的基于扩散的平铺管道缓解了边界不连续性，但缺乏对哪些区域在几何上重要以进行保留的意识。在本研究中，我们提出了DepthTile，一种基于深度引导的自平铺方法，其生成模块不需要重新训练扩散主干。它将单目深度转换为空间保留先验，并将其集成到Tiled Diffusion的时间步依赖掩蔽机制中。DepthTile将深度转换为空间可编辑性图，并将其与差异扩散掩蔽融合。这增强了在单目深度先验指示的近场区域的源图像保留，同时在兼容的背景边界附近保留编辑灵活性。为了评估自平铺的可行性，我们引入了自平铺能力评分（Self-Tiling Ability Score, STAS），这是一种基于RGB、梯度、深度和深度梯度线索的轻量级输入端预测器。在335幅图像的基准测试中，相较于默认的Tiled Diffusion管道，DepthTile分别减少了MAE、LPIPS和Tiling Score的12.8%、26.7%和8.4%。STAS可靠地实现了难度感知分析，并进一步支持兼容性感知的成对图像平铺。

View on arXiv Download PDF AI Translation

cs.CV / 163 / 2610.16287

ProGAN-Based Synthetic Image Generation for Industrial Visual Inspection of Metallic Parts

基于ProGAN的金属部件工业视觉检测合成图像生成

Beiki, Hossein Omid, Asgari, Sepehr, Petrovic, Oliver, Brecher, Christian

Abstract

Automated visual inspection is essential for efficient repair and recycling of metallic components. AI-based defect detection is a strong fit. However, it is often constrained by scarce, low-diversity training data. To address this gap, we study Generative AI-driven synthetic image generation using Progressive Growing GANs (ProGAN). We implement a ProGAN pipeline with Wasserstein loss and gradient penalty and evaluate key hyperparameters on two public datasets, MVTec AD (metal nut) and MPDD (tube). As a result, training remained stable and the model yielded realistic images. These images can augment training data, improve robustness, and reduce the cost of collecting rare defects.

Chinese Translation

自动化视觉检测对于金属组件的高效修复和回收至关重要。基于人工智能的缺陷检测非常适合这一需求。然而，它常常受到稀缺且低多样性的训练数据的限制。为了解决这一问题，我们研究了使用渐进式生长生成对抗网络（Progressive Growing GANs, ProGAN）驱动的合成图像生成。我们实现了一个采用Wasserstein损失和梯度惩罚的ProGAN管道，并在两个公共数据集上评估了关键超参数，分别是MVTec AD（金属螺母）和MPDD（管道）。结果显示，训练过程保持稳定，模型生成了逼真的图像。这些图像可以增强训练数据，提高模型的鲁棒性，并降低收集稀有缺陷的成本。

View on arXiv Download PDF AI Translation

cs.CV / 164 / 2610.16292

3D Gaussian Splatting for Fine-grained Urban Scene Reconstruction via Semantic-guided Density Control and Loss Design

基于语义引导的密度控制与损失设计的细粒度城市场景重建的3D高斯点云技术

Liao, Yueling, Jiang, San, Zhang, Xing, Hu, Zhongwen

Abstract

3D Gaussian Splatting (3DGS) has been becoming a promising 3D reconstruction technique because of its advantage of high-fidelity rendering effect with high efficiency. However, it still suffers from coarse density control and semantically blind loss functions when reconstructing the fine-grained structures in complex urban scenes. Thus, this study proposes SG-GS, a semantic-guided 3DGS method for fine-grained structure reconstruction. It takes the semantic segmentation deduced by the Segment Anything Model (SAM) as a constraint and systematically integrates it into both Gaussian density control and loss function design. On the one hand, a semantic-constrained adaptive density control mechanism is constructed, which dynamically allocates gradient thresholds to different semantic regions of the scene and introduces an additional boundary-enhanced densification strategy, so that new Gaussians are directed to structurally critical regions. On the other hand, a multi-feature guided loss function is designed, and fine-grained supervision signals are gradually introduced through a phased progressive training strategy, alleviating the artifacts caused by the lack of semantic constraints in the basic loss. Experiments on three datasets show that the proposed method consistently improves reconstruction accuracy. Compared with the original 3DGS, the PSNR increases by approximately 2.7% to 5.2%, while artifacts on linear structures and boundary blurring are effectively mitigated. The method can serve the fine-grained 3D reconstruction of urban scenes.

Chinese Translation

3D高斯点云技术（3D Gaussian Splatting, 3DGS）因其高效的高保真渲染效果而成为一种有前景的3D重建技术。然而，在重建复杂城市场景中的细粒度结构时，它仍然面临粗糙的密度控制和缺乏语义约束的损失函数的问题。因此，本研究提出了SG-GS，一种用于细粒度结构重建的语义引导3DGS方法。该方法将由Segment Anything Model（SAM）推导出的语义分割作为约束，系统地将其整合到高斯密度控制和损失函数设计中。一方面，构建了一种语义约束的自适应密度控制机制，动态地将梯度阈值分配给场景的不同语义区域，并引入额外的边界增强密度化策略，使新的高斯点云指向结构上关键的区域。另一方面，设计了一种多特征引导的损失函数，通过分阶段的渐进训练策略逐步引入细粒度监督信号，缓解了基本损失中缺乏语义约束所造成的伪影。在三个数据集上的实验表明，所提方法在重建精度上始终有所提升。与原始的3DGS相比，峰值信噪比（PSNR）提高了约2.7%至5.2%，同时有效减轻了线性结构上的伪影和边界模糊。该方法可用于城市场景的细粒度3D重建。

View on arXiv Download PDF AI Translation

cs.CV / 165 / 2610.16293

ClothAtlas: Joint Garment Identification and State Estimation under Self-Occlusion

ClothAtlas：自遮挡下的服装识别与状态估计

Lundell, Jens, Longhini, Alberta, Shu, Lei, Arora, Hemal, Robert, Evan, Parés-Morlans, Carlota, Büsching, Marcel, Björkman, Mårten, Bohg, Jeannette

Abstract

Garment manipulation requires inferring both garment identity and complete configuration from partial, self-occluded observations. This creates a chicken-and-egg problem: completing a partial observation requires a garment-specific deformation prior, while identifying the garment requires a configuration that explains the observation. We introduce ClothAtlas, which jointly identifies garments and estimates their complete configurations from partial point clouds and a library of canonical meshes. A shared graph-conditioned flow-matching model samples configurations for each candidate mesh and ranks candidates by observation fit. The model accommodates different mesh shapes, vertex counts, and connectivities, allowing new garments to be added from canonical meshes alone without retraining. On held-out DexGarmentLab garments, ClothAtlas ranks the correct mesh first among 221 candidates for all 22 test garments from a single flat observation. From a single self-occluded observation, it identifies 15/22 garments, compared with 8/22 for a deterministic graph estimator. Persistent vertex identities support grasp-target localization on real garments, reducing mean error from 15.4 to 6.4 cm overall and from 17.0 to 8.3 cm on occluded targets compared with dense descriptor matching. Project page: https://clothatlas.github.io/

Chinese Translation

服装操作需要从部分自遮挡的观察中推断服装的身份和完整配置。这造成了一个鸡与蛋的问题：完成部分观察需要特定于服装的变形先验，而识别服装则需要一个能够解释观察结果的配置。我们提出了ClothAtlas，它能够从部分点云和标准网格库中共同识别服装并估计其完整配置。一个共享的图条件流匹配模型为每个候选网格采样配置，并根据观察拟合对候选进行排名。该模型适应不同的网格形状、顶点数量和连接方式，使得可以仅通过标准网格添加新服装而无需重新训练。在保留的DexGarmentLab服装上，ClothAtlas在221个候选中将正确网格的排名置于所有22个测试服装的首位，仅基于一次平面观察。从单次自遮挡观察中，它识别出15/22件服装，而确定性图估计器仅识别出8/22件。持久的顶点身份支持在真实服装上的抓取目标定位，与密集描述符匹配相比，整体平均误差从15.4厘米降低到6.4厘米，在遮挡目标上从17.0厘米降低到8.3厘米。项目页面：https://clothatlas.github.io/

View on arXiv Download PDF AI Translation

cs.CV / 166 / 2610.16299

Learning Temporal Evidence Selection for Video Anomaly Understanding

学习视频异常理解中的时间证据选择

Park, Hangil, Kim, Seungheon, Kim, Tae-Kyun

Abstract

Video anomaly understanding (VAU) aims to generate language-level interpretations of anomalous events in long videos, requiring visual evidence that supports explanation beyond anomaly detection alone. Existing VAU pipelines commonly rely on anomaly-centric or predefined temporal sampling strategies, which efficiently emphasize suspicious moments but overlook contextual cues, object interactions, and temporally distributed evidence needed for reasoning. Motivated by recent progress in learnable temporal selection for long-video question answering, we revisit VAU temporal sampling as a downstream-aligned evidence selection problem. We introduce the semantic-conditioned Temporal Sampling Q-Former that learns to select temporal visual evidence through the VAU generation objective. Since generic VAU prompts provide weak video-specific guidance, we condition the sampler on CLIP-embedded object tags automatically extracted from the input video. These tags are used only for temporal evidence selection and are not appended to the LLM generation prompt, separating sampler conditioning from prompt augmentation. The proposed sampler operates in visual-token space, computes soft temporal selection weights from frame-level representations, and merges duplicated selections before response generation. Experiments on HIVAU-70K show that our model achieves the best performance among the compared fine-tuned models, while ablations demonstrate that learnable temporal sampling improves over fixed anomaly-focused sampling and that CLIP-embedded object-tag conditioning further improves over generic-prompt conditioning. Our results suggest that temporal sampling in VAU should be aligned with explanation-oriented evidence utility rather than anomaly saliency alone.

Chinese Translation

视频异常理解（VAU）旨在生成对长视频中异常事件的语言级解释，这需要支持解释的视觉证据，超越单纯的异常检测。现有的VAU流程通常依赖于以异常为中心或预定义的时间采样策略，这些策略有效地强调可疑时刻，但忽视了推理所需的上下文线索、对象交互和时间分布的证据。受到近期在长视频问答中可学习时间选择的进展的启发，我们将VAU的时间采样重新审视为一个下游对齐的证据选择问题。我们引入了语义条件的时间采样Q-Former，该模型通过VAU生成目标学习选择时间视觉证据。由于通用的VAU提示提供了较弱的视频特定指导，我们将采样器的条件设置为从输入视频自动提取的CLIP嵌入对象标签。这些标签仅用于时间证据选择，而不附加到大语言模型（LLM）生成提示中，从而将采样器条件与提示增强分开。所提出的采样器在视觉标记空间中操作，从帧级表示计算软时间选择权重，并在响应生成之前合并重复的选择。在HIVAU-70K上的实验表明，我们的模型在比较的微调模型中实现了最佳性能，而消融实验表明，可学习的时间采样优于固定的以异常为中心的采样，并且CLIP嵌入对象标签条件进一步优于通用提示条件。我们的结果表明，VAU中的时间采样应与以解释为导向的证据效用对齐，而不仅仅是异常显著性。

View on arXiv Download PDF AI Translation

cs.CV / 167 / 2610.16302

RAR: Rate-Adaptive Rollout for Time-Consistent Interactive World Models

RAR：时间一致的交互式世界模型的速率自适应展开

Liu, Yaoyang, Yan, Dongyu, Du, Yihua, Qian, Shengju, Xie, Zeke, Wang, Xin, Chen, Ying-Cong, Chen, Long

Abstract

Video-based interactive world models aim to enable real-time, user-controllable environments. Yet most existing systems are trained and rolled out at a fixed temporal rate, limiting both runtime-rate adaptation and physical-horizon coverage -- a limitation we term temporal-rate coupling. Specifically, when runtime generation throughput differs from the learned temporal rate, the generated world evolves unintentionally more slowly or quickly. Moreover, using a single temporal rate imposes an inherent trade-off between temporal granularity and physical horizon: higher rates capture finer motion but cover a shorter duration under a fixed context window. To address both limitations, we introduce Rate-Adaptive Rollout (RAR), a framework that enables interactive world models to roll out at controllable temporal rates. At its core, Stride-RoPE aligns latent frames sampled at different rates on a shared physical-time coordinate, enabling consistent multi-rate rollout, unseen-rate generalization, and online rate switching without discarding history. Meanwhile, RAR uses multi-stride video sampling to enable joint learning of fine-grained dynamics and long-horizon evolution. Across action- and camera-controlled environments, RAR achieves temporally consistent multi-rate rollout and online rate switching while preserving high-rate generation quality and extending physical horizons under matched training budgets. RAR also transfers to pretrained world models while preserving perceptual quality.

Chinese Translation

基于视频的交互式世界模型旨在实现实时、用户可控的环境。然而，大多数现有系统在固定的时间速率下进行训练和展开，这限制了运行时速率的适应性和物理视野的覆盖——我们称之为时间速率耦合的限制。具体而言，当运行时生成吞吐量与学习的时间速率不同时，生成的世界会意外地演变得更慢或更快。此外，使用单一时间速率会在时间粒度和物理视野之间施加固有的权衡：更高的速率捕捉更细致的运动，但在固定的上下文窗口下覆盖的持续时间较短。为了解决这两个限制，我们引入了速率自适应展开（Rate-Adaptive Rollout, RAR），这是一个使交互式世界模型能够以可控时间速率展开的框架。其核心是Stride-RoPE，它将以不同速率采样的潜在帧对齐到共享的物理时间坐标上，从而实现一致的多速率展开、未见速率的泛化以及在线速率切换，而无需丢弃历史数据。同时，RAR利用多步长视频采样来实现细粒度动态和长时间演变的联合学习。在动作和相机控制的环境中，RAR实现了时间一致的多速率展开和在线速率切换，同时保持高速率生成质量，并在匹配的训练预算下扩展物理视野。RAR还可以迁移到预训练的世界模型，同时保持感知质量。

View on arXiv Download PDF AI Translation

cs.CV / 168 / 2610.16320

Building age estimation based on widely available multi-source1 geospatial data

基于广泛可用的多源地理空间数据的建筑年龄估计

Sui, Jialu, Zhang, Hongsheng, Liu, Rui, Ling, Jing, Wang, Ruisheng, Wu, Hao, Chen, Gang

Abstract

Building age is fundamental for urban infrastructure safety, operation, and planning, as it directly affects structural safety, and sustainable development. Existing methods mostly rely on detailed building information such as street view, function, or materials, which are harder to obtain than geospatial data like satellite imagery and Digital Surface Models (DSMs). However, existing methods based on those geospatial data mainly rely on optical imagery and local texture features, limiting their ability to capture structural information and distinguish buildings from different periods. Moreover, the inherent temporal ordering among building age categories is often overlooked. To address these limitations, we propose the first deep-learning framework for multi-building age estimation using widely available multi-source geospatial data, including remote sensing data and their derived products, as well as building footprint information that estimates age at both pixel and instance levels within a unified pipeline. Specifically, a large pretrained backbone with fine-tuning adapters extracts representative features, and raster-derived and property-derived fusion modules support pixel-wise and instance-wise predictions. Pixel-level estimation employs a pyramid decoder with Cosine-guided Mixture-of-Experts (CMoE) to fuse multi-source raster data across scales, capturing fine-grained building age differences, while instance-level prediction uses attention pooling and Property-Conditional Adaptive Instance Normalization (PCAdaIN) to extract instance-level representations and integrate property attributes. A Temporal Ordinal Constraint Loss (TOCL) enforces the sequential ordering of predicted ages. We also construct a building age benchmark for Hong Kong to enable comprehensive evaluation. The source code and datasets are available at https://github.com/littlebeen/RS-buildingAge.git.

Chinese Translation

建筑年龄对于城市基础设施的安全、运营和规划至关重要，因为它直接影响结构安全和可持续发展。现有方法大多依赖于详细的建筑信息，如街景、功能或材料，这些信息比卫星影像和数字表面模型（Digital Surface Models, DSMs）等地理空间数据更难获得。然而，基于这些地理空间数据的现有方法主要依赖于光学影像和局部纹理特征，限制了其捕捉结构信息和区分不同年代建筑的能力。此外，建筑年龄类别之间固有的时间顺序往往被忽视。为了解决这些局限性，我们提出了第一个基于深度学习的多建筑年龄估计框架，该框架利用广泛可用的多源地理空间数据，包括遥感数据及其衍生产品，以及建筑轮廓信息，在统一的流程中对像素和实例级别的年龄进行估计。具体而言，一个大型预训练骨干网络与微调适配器结合，提取代表性特征，栅格衍生和属性衍生融合模块支持像素级和实例级的预测。像素级估计采用具有余弦引导的专家混合模型（Cosine-guided Mixture-of-Experts, CMoE）的金字塔解码器，融合不同尺度的多源栅格数据，捕捉细粒度的建筑年龄差异，而实例级预测则使用注意力池化和属性条件自适应实例归一化（Property-Conditional Adaptive Instance Normalization, PCAdaIN）提取实例级表示并整合属性特征。时间序列约束损失（Temporal Ordinal Constraint Loss, TOCL）强制执行预测年龄的顺序。我们还为香港构建了一个建筑年龄基准，以便进行全面评估。源代码和数据集可在 https://github.com/littlebeen/RS-buildingAge.git 获取。

View on arXiv Download PDF AI Translation

cs.CV / 169 / 2610.16324

Anomaly-Guided Observation and Two-Level Semantic Feedback for Training-Free Video Anomaly Analysis

基于异常引导观察和两级语义反馈的无训练视频异常分析

Park, Hangil, Kim, Seungheon, Kim, Tae-Kyun

Abstract

Training-free language-based video anomaly detection (VAD) uses frozen vision--language models (VLMs) and large language models (LLMs) to localize abnormal events without task-specific model-weight optimization. Existing approaches increasingly exploit broader temporal and semantic context, yet generic context does not necessarily foreground abnormal events or specify how video-level semantics should modify local anomaly scores. We introduce a training-free semantic feedback framework that uses VAD-guided active observation for video anomaly understanding (VAU) and returns the resulting anomaly-focused description to refine temporal detection. Starting from an upstream anomaly-score curve, we adapt reasoning-based iterative frame selection: high-score regions seed the observations, and anomaly scores together with VLM relevance assessments guide subsequent temporal exploration. A frozen video-language model generates a single open-ended anomaly description from the selected frames. The description is returned through two complementary feedback levels: video-level feedback determines whether the score distribution should be suppressed, preserved, or sharpened, with selective visual rechecking, while local feedback retrieves event-related windows and promotes scores only after target-local visual verification. All model weights remain frozen, and feedback is applied once without regenerating the description. On the HIVAU video-summary evaluations of UCF-Crime and XD-Violence, VAD-guided observation improves GPT-rated reasonability, detail, and consistency over the matched CLIP-guided variant. For VAD, semantic feedback improves the re-evaluated upstream detector from 84.36% to 86.25% AUC on UCF-Crime and from 68.07% to 72.58% AP on XD-Violence, where the final AUC is 91.74%.

Chinese Translation

无训练的基于语言的视频异常检测（VAD）利用冻结的视觉-语言模型（VLMs）和大型语言模型（LLMs）来定位异常事件，而无需特定任务的模型权重优化。现有方法越来越多地利用更广泛的时间和语义上下文，但通用上下文并不一定突出异常事件或具体说明视频级语义应如何修改局部异常分数。我们提出了一种无训练的语义反馈框架，该框架利用VAD引导的主动观察进行视频异常理解（VAU），并返回结果异常聚焦的描述以细化时间检测。从上游异常分数曲线开始，我们采用基于推理的迭代帧选择：高分区域作为观察的种子，异常分数与VLM相关性评估共同指导后续的时间探索。冻结的视频-语言模型从选定的帧生成单一的开放式异常描述。该描述通过两个互补的反馈层返回：视频级反馈决定分数分布应被抑制、保留还是增强，并进行选择性视觉复查，而局部反馈则检索事件相关窗口，仅在目标局部视觉验证后提升分数。所有模型权重保持冻结，反馈仅应用一次，而不重新生成描述。在UCF-Crime和XD-Violence的HIVAU视频摘要评估中，VAD引导的观察提高了GPT评分的合理性、细节和一致性，相较于匹配的CLIP引导变体。对于VAD，语义反馈将重新评估的上游检测器的AUC从84.36%提高到86.25%（UCF-Crime），并将AP从68.07%提高到72.58%（XD-Violence），最终AUC为91.74%。

View on arXiv Download PDF AI Translation

cs.CV / 170 / 2610.16329

Explicitly Aligned Multi-View Breast Cancer Risk Prediction from Variable-Length Screening Histories

基于可变长度筛查历史的显式对齐多视角乳腺癌风险预测

Thrun, Solveig, Sun, Zijun, Salahuddin, Suaiba A., Wickstrøm, Kristoffer, Wetzer, Elisabeth, Hansen, Stine, Jenssen, Robert, Kampffmeyer, Michael

Abstract

Accurate breast cancer risk prediction from screening mammography is essential for improved risk stratification and personalized screening. It can benefit from both temporal breast tissue changes across longitudinal examinations and complementary anatomical information from craniocaudal (CC) and mediolateral oblique (MLO) views. Leveraging longitudinal information requires identifying corresponding tissue regions across examinations, yet most approaches rely on implicit feature alignment without explicitly enforcing spatial consistency. Recent work has demonstrated the benefits of explicit temporal alignment; however, existing methods model each view independently, preventing them from exploiting complementary CC and MLO information. Furthermore, explicit approaches have been limited to a single prior examination. To fill these gaps, we propose (github.com/sot176/FlexLMV-Net), a breast-wise longitudinal multi-view framework that jointly models CC and MLO views across screening examinations while accommodating a variable number of prior examinations. FlexLMV-Net explicitly aligns longitudinal feature maps and integrates complementary views through cross-view temporal fusion and a dual-stream attention block. Evaluated on EMBED and CSAW-CC, FlexLMV-Net consistently outperforms existing approaches across breast density, cancer subtypes, and demographic subgroups, highlighting the benefits of explicit longitudinal alignment, multi-view modeling, and flexible use of available screening history. We further introduce (github.com/sot176/BCRBench), an open-source benchmark that unifies state-of-the-art single- and multi-timepoint breast cancer risk prediction models under standardized preprocessing, training, inference, and evaluation protocols, enabling fair and reproducible comparison of existing methods.

Chinese Translation

从筛查乳腺X光摄影中准确预测乳腺癌风险对于改善风险分层和个性化筛查至关重要。这一过程可以受益于纵向检查中乳腺组织的时间变化以及来自头尾（CC）和斜侧（MLO）视角的互补解剖信息。利用纵向信息需要识别各次检查中对应的组织区域，但大多数方法依赖于隐式特征对齐，而未明确强制空间一致性。近期的研究表明显式时间对齐的好处；然而，现有方法独立建模每个视角，无法利用互补的CC和MLO信息。此外，显式方法仅限于单次先前检查。为填补这些空白，我们提出了FlexLMV-Net，一个基于乳腺的纵向多视角框架，能够在筛查检查中联合建模CC和MLO视角，同时适应可变数量的先前检查。FlexLMV-Net显式对齐纵向特征图，并通过跨视角时间融合和双流注意力模块整合互补视角。在EMBED和CSAW-CC上的评估表明，FlexLMV-Net在乳腺密度、癌症亚型和人口统计子组中始终优于现有方法，突显了显式纵向对齐、多视角建模和灵活利用可用筛查历史的优势。我们进一步介绍了BCRBench，一个开源基准，统一了最先进的单时间点和多时间点乳腺癌风险预测模型，采用标准化的预处理、训练、推理和评估协议，能够公平且可重复地比较现有方法。

View on arXiv Download PDF AI Translation

cs.CV / 171 / 2610.16330

BAT3R: Robust Online 3D Reconstruction with Bayesian Adaptive State Updates

BAT3R：基于贝叶斯自适应状态更新的鲁棒在线三维重建

Li, Lijian, He, Yuanpeng, Zheng, Yudian, Huang, Chung-ju, Li, Linyu, Jin, Dongming, Jiao, Wenpin, Jin, Zhi, Pun, Chi-Man

Abstract

Feed-forward transformer models have driven remarkable progress in 3D vision,yet their quadratic complexity renders them impractical for long video sequences. Streaming 3D reconstruction addresses this bottleneck by processing frames sequen?tially with constant memory. However, existing recurrent architectures inevitably suffer from progressive degradation over long sequences, as they fail to account for the intrinsic reliability of observations and the accumulated confidence of the historical state, leading to severe error accumulation. To address these issues, we introduce BAT3R, a training-free Bayesian adaptation framework comprising two synergistic modules: State Innovation Gating (SIG) and Bayesian-guided Adaptive State Updating (BASU). SIG acts as a distributional regularizer that anchors state evolution to its geometric initialization, effectively arresting representational drift during long-term inference. Building on this, BASU reformulates state update as a recursive Bayesian inference process within a Kalman filtering framework. By jointly modeling the temporal reliability of observations via information-theoretic gain and spatial relevance via cross-attention maps, BASU achieves an asymmetric spatio-temporal state fusion: well-consolidated historical regions are protected from redundant updates, while genuinely novel scene content is aggressively incor?porated. Extensive experiments on standard benchmarks (7-Scenes, TUM, ScanNet, KITTI, etc.) demonstrate that BAT3R consistently outperforms state-of-the-art online 3D reconstruction methods, with the most significant gains observed on long sequences and complex scenes.

Chinese Translation

前馈变换器模型在三维视觉领域取得了显著进展，但其二次复杂度使其在长视频序列中不切实际。流式三维重建通过以恒定内存顺序处理帧来解决这一瓶颈。然而，现有的递归架构在长序列中不可避免地遭受逐步退化，因为它们未能考虑观察的内在可靠性和历史状态的累积置信度，导致严重的误差积累。为了解决这些问题，我们提出了BAT3R，一个无训练的贝叶斯适应框架，包含两个协同模块：状态创新门控（State Innovation Gating, SIG）和贝叶斯引导的自适应状态更新（Bayesian-guided Adaptive State Updating, BASU）。SIG作为一个分布正则化器，将状态演变锚定到其几何初始化，有效地阻止了长期推理过程中的表示漂移。在此基础上，BASU将状态更新重新表述为卡尔曼滤波框架内的递归贝叶斯推理过程。通过信息论增益联合建模观察的时间可靠性和通过交叉注意力图建模空间相关性，BASU实现了不对称的时空状态融合：良好巩固的历史区域受到保护，避免冗余更新，而真正的新场景内容则被积极整合。在标准基准（7-Scenes, TUM, ScanNet, KITTI等）上的大量实验表明，BAT3R在在线三维重建方法中始终优于最先进的技术，尤其在长序列和复杂场景中表现出显著的提升。

View on arXiv Download PDF AI Translation

cs.CV / 172 / 2610.16337

WorldRoPE: Probabilistic World-Centric Rotary Embedding for Multi-View Attention

WorldRoPE：用于多视角注意力的概率世界中心旋转嵌入

Kim, Mijeong, Lim, Kyujin, Han, Bohyung

Abstract

Rotary Positional Embedding (RoPE) is widely used to encode relative positions in Transformers. In multi-view settings, however, patch coordinates are defined independently in each image: the same 3D point can therefore have different coordinates across views, making them insufficient for establishing cross-view correspondences. We introduce WorldRoPE, a probabilistic world-centric RoPE for multi-view attention. At each layer, WorldRoPE back-projects each patch token into a set of volumetric Gaussian primitives in a shared world frame. These primitives represent multiple plausible 3D supports for a token, jointly capturing alternative depth hypotheses, depth uncertainty, and the finite spatial extent of its patch frustum. When a key token's support is projected into a query view, it induces a distribution over where that key may appear in the query image. WorldRoPE integrates the rotary operator over this distribution in closed form, combining rotations at the projected primitive centers according to their mixture weights with covariance-dependent attenuation. This makes attention less sensitive to uncertain key locations in the query view, without rasterizing or warping token features. The formulation reduces to deterministic RoPE when the support collapses to a point. Across layers, the predicted geometry guides cross-view attention, while the updated features refine the geometry used at the next layer. Experiments on novel view synthesis, depth estimation, and 3D geometry reconstruction demonstrate the effectiveness of WorldRoPE for multi-view Transformers.

Chinese Translation

旋转位置嵌入（Rotary Positional Embedding, RoPE）在编码变换器中的相对位置方面得到了广泛应用。然而，在多视角设置中，补丁坐标在每个图像中是独立定义的：因此，同一3D点在不同视角下可能具有不同的坐标，这使得它们不足以建立跨视角的对应关系。我们提出了WorldRoPE，一种用于多视角注意力的概率世界中心RoPE。在每一层中，WorldRoPE将每个补丁标记反投影到共享世界框架中的一组体积高斯原语。这些原语代表了一个标记的多个可能的3D支持，联合捕捉替代的深度假设、深度不确定性和其补丁锥体的有限空间范围。当一个关键标记的支持被投影到查询视图中时，它会引发一个分布，表示该关键标记可能出现在查询图像中的位置。WorldRoPE以封闭形式在该分布上集成旋转算子，根据投影原语中心的混合权重和与协方差相关的衰减组合旋转。这使得注意力对查询视图中不确定的关键位置不那么敏感，而无需光栅化或扭曲标记特征。该公式在支持收缩到一个点时简化为确定性RoPE。在各层之间，预测的几何形状指导跨视角注意力，而更新的特征则细化下一层使用的几何形状。在新视图合成、深度估计和3D几何重建的实验中，证明了WorldRoPE在多视角变换器中的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 173 / 2610.16338

SyncReward: Human-Aligned Synchronization Reward for Audio-Visual Generation

SyncReward：人类对齐的音视频生成同步奖励

Guan, Kaisi, Ba, Ying, Wang, Wenjing, Wang, Xihua, Li, Chen, Lyu, Jing, Song, Ruihau

Abstract

Recent years have witnessed rapid progress in joint audio--visual generation. However, audio--visual synchronization remains a challenge, as generated speech and sound often temporally misalign with visual events. A key bottleneck is the lack of human-annotated data, reliable evaluation benchmarks, and synchronization-aware evaluation models for generated audio--visual outputs. To address these limitations, we conduct a comprehensive study spanning data, benchmarking, and reward modeling. We first introduce SyncReward-Data-80K, a dataset containing 80K professional synchronization ratings for clips from eight audio--visual generators. A taxonomy-guided construction pipeline covers four sound-event families: Speech, Onset, Instrument, and Ambient. We further construct \textbf{SyncReward-Bench}, a held-out benchmark of 1K clips, each independently rated by multiple professionals. Using this benchmark, we systematically evaluate specialized synchronization metrics and omni-modal models, revealing their substantial misalignment with human judgments. Finally, we develop SyncReward, a reward model that combines semantic--temporal contrastive pretraining on real clips with human supervision on generated clips. As an evaluation metric, SyncReward achieves the strongest agreement with human ratings among the evaluated methods. As a reward, it improves synchronization through inference-time best-of-$N$ selection and reinforcement learning based post-training. We will release the data, benchmark, model, and code at https://github.com/WeChatCV/SyncReward.

Chinese Translation

近年来，音视频联合生成取得了快速进展。然而，音视频同步仍然是一个挑战，因为生成的语音和声音往往与视觉事件在时间上不一致。一个关键瓶颈是缺乏人类标注的数据、可靠的评估基准以及针对生成音视频输出的同步感知评估模型。为了解决这些限制，我们进行了涵盖数据、基准测试和奖励建模的全面研究。我们首先介绍了SyncReward-Data-80K，这是一个包含来自八个音视频生成器的80K专业同步评分的数据集。一个基于分类法的构建流程涵盖了四个声音事件类别：语音、起始、乐器和环境。我们进一步构建了SyncReward-Bench，这是一个包含1K剪辑的保留基准，每个剪辑均由多位专业人士独立评分。利用该基准，我们系统地评估了专门的同步指标和全模态模型，揭示了它们与人类判断之间的显著不一致。最后，我们开发了SyncReward，这是一种奖励模型，结合了对真实剪辑的语义-时间对比预训练和对生成剪辑的人类监督。作为评估指标，SyncReward在评估方法中与人类评分的符合度最强。作为奖励，它通过推理时的最佳选择和基于强化学习的后训练改善了同步。我们将在https://github.com/WeChatCV/SyncReward发布数据、基准、模型和代码。

View on arXiv Download PDF AI Translation

cs.CV / 174 / 2610.16357

VideoSlate: Stateful Commit Control for Temporally Grounded Agentic Video Reasoning

VideoSlate：面向时间基础的代理视频推理的状态提交控制

Zheng, Shuo, Dang, Yuanjie, Chen, Peng, Xiang, Jianjun, Liang, Ronghua, Bennamoun, Mohammed, Ke, Qiuhong

Abstract

Agentic video reasoning systems gather and assess video evidence across stages, selecting relevant spans as visual evidence for answering questions. New candidates and observations prompt reassessment of the current span, but revisions can correct localization errors or discard useful evidence already present in the adopted span. Systems therefore need to connect evidence organization across stages with temporal state control, allowing accumulated information to jointly inform revision assessment while explicit adoption decisions govern how assessment outcomes affect downstream visual evidence. We introduce VideoSlate, a stateful commit-control framework built around shared evidence. A typed evidence blackboard links candidate spans, provenance, construction, and verifier observations, enabling an incumbent-relative certifier to assess revision support from the candidate, current span, and related proposals. Based on certification, the controller authorizes a revision or retains the current span, separating evidence accumulation from state updates and recording each decision and its evidence on the blackboard. Extensive evaluations across video grounding and grounded question-answering benchmarks show that VideoSlate can improve temporal localization through selective revision.

Chinese Translation

代理视频推理系统在各个阶段收集和评估视频证据，选择相关的时间段作为回答问题的视觉证据。新的候选项和观察促使对当前时间段的重新评估，但修订可能会纠正定位错误或丢弃已在采用时间段中存在的有用证据。因此，系统需要将跨阶段的证据组织与时间状态控制相连接，使得累积的信息能够共同影响修订评估，同时明确的采用决策决定了评估结果如何影响下游的视觉证据。我们提出了VideoSlate，一个围绕共享证据构建的状态提交控制框架。一个类型化的证据黑板连接候选时间段、来源、构建和验证者观察，使得相对在任的认证者能够评估来自候选项、当前时间段和相关提案的修订支持。基于认证，控制器授权修订或保留当前时间段，将证据积累与状态更新分开，并在黑板上记录每个决策及其证据。在视频基础和基础问答基准上的广泛评估表明，VideoSlate能够通过选择性修订改善时间定位。

View on arXiv Download PDF AI Translation

cs.CV / 175 / 2610.16364

From Sound to Sense: An Anatomy of Semantic Supervision for Speech-Driven Visual Grounding in Remote Sensing Imagery

从声音到意义：语音驱动的遥感图像视觉定位的语义监督解剖

Pang, Zhongyu

Abstract

Speech-driven visual grounding in remote sensing (RSVG) removes the typing bottleneck of text-based interaction and is intended to preserve prosodic intent, enabling hands-free target acquisition in time-critical missions. The direct-audio paradigm, mapping raw speech straight to a bounding box and bypassing automatic speech recognition (ASR), was recently adopted by RSTalker, but what such grounders actually learn about language remains unmeasured. SemEcho grafts a frozen Whisper speech pathway onto a VLTVG grounding host so that speech and text share one decoder, making the paired text, free by construction since speech corpora in this task are synthesized from DIOR-RSVG expressions, a training-time teacher and a fair, same-host text upper bound. On this instrument we dissect three semantic injection channels (feature-mimicry distillation, word-level CTC supervision, and two-stage ASR pretraining) with linear semantic probes, and release Audio-DIOR-R (80,973 training clips; 10 timbres x K=3; Qwen3-TTS synthesis with online sox noise lambda in [0, 0.4]) and Audio-DIOR-Comp, a re-split holding out 17 attribute-category combinations under a 2x2 acoustic-semantic protocol. Our anatomy yields four findings. (i) Whether direct alignment beats a domain-trained ASR cascade is decided by recognizer quality: against a weak in-domain cascade (CTC 0.53) direct wins by 9.0 [email protected] points with transcription errors concentrated on spatial-relation words, but a strong in-domain cascade (CTC 0.029) reaches the text upper bound (80.8 vs. 81.0) and beats even the strongest direct model by 8.8 points; the anti-ASR premise holds only against weak recognizers.

Chinese Translation

语音驱动的遥感视觉定位（RSVG）消除了基于文本交互的输入瓶颈，旨在保留韵律意图，从而实现时间关键任务中的免提目标获取。直接音频范式将原始语音直接映射到边界框，并绕过自动语音识别（ASR），最近被 RSTalker 采用，但这种定位器实际上对语言的学习内容尚未被测量。SemEcho 将一个冻结的 Whisper 语音路径嫁接到 VLTVG 定位主机上，使得语音和文本共享一个解码器，由于该任务中的语音语料库是从 DIOR-RSVG 表达式合成的，因此配对文本在构建上是免费的，成为训练时的教师和一个公平的、同主机的文本上限。在这个工具上，我们利用线性语义探针剖析了三个语义注入通道（特征模仿蒸馏、词级 CTC 监督和两阶段 ASR 预训练），并发布了 Audio-DIOR-R（80,973 个训练片段；10 种音色 x K=3；Qwen3-TTS 合成，在线 sox 噪声 lambda 在 [0, 0.4] 范围内）和 Audio-DIOR-Comp，一个在 2x2 声学-语义协议下保留 17 种属性-类别组合的重新划分。我们的解剖得出了四个发现。（i）直接对齐是否优于领域训练的 ASR 级联取决于识别器的质量：在一个弱的领域内级联（CTC 0.53）下，直接模型以 9.0 [email protected] 分的优势获胜，转录错误集中在空间关系词上，但强的领域内级联（CTC 0.029）达到了文本上限（80.8 vs. 81.0），甚至比最强的直接模型还要高出 8.8 分；反 ASR 前提仅在弱识别器面前成立。

View on arXiv Download PDF AI Translation

cs.CV / 176 / 2610.16383

Throwing4: Phase-Aware Pose Modeling for Throw-Distance Prediction in Elite Athletics

Throwing4：面向阶段的姿态建模用于精英田径投掷距离预测

Badatya, Bikash Kumar, Tiwari, Kartike, Amin, Jyotirmoy, Hegde, Ravi

Abstract

Elite athletics throwing provides a structured biomechanical action sequence together with an objective continuous outcome, namely throw distance, making it a suitable setting for performance-oriented sports analytics. In this work, we investigate whether explicit phase alignment of pose dynamics improves throw-distance prediction under realistic broadcast conditions. To support this study, we curate Throwing4, a dataset of 500 elite throwing clips spanning javelin, discus, hammer, and shot put, with corresponding two-dimensional pose sequences, frame-level phase annotations, and official distance labels. We formulate throw-distance prediction from broadcast-derived pose as a regression problem and compare full-sequence modeling against a phase-conditioned approach that partitions each attempt into four technique stages and fuses phase-specific motion representations. Experiments across multiple graph-based pose encoders show a consistent reduction in prediction error when phase structure is incorporated, indicating that phase-aligned modeling provides a more stable representation than treating the entire motion as an unstructured sequence. These findings suggest that explicit biomechanical phase information is a useful inductive prior for pose-based performance modeling in elite athletics throwing. Beyond score prediction, the dataset and protocol provide a basis for future work on fine-grained technique analysis, interpretable sports analytics, and robust modeling from broadcast video.

Chinese Translation

精英田径投掷提供了一种结构化的生物力学动作序列，以及一个客观的连续结果，即投掷距离，使其成为面向性能的体育分析的合适场景。在本研究中，我们探讨了姿态动态的显式阶段对齐是否能在现实广播条件下改善投掷距离预测。为了支持这项研究，我们整理了Throwing4数据集，该数据集包含500个精英投掷视频片段，涵盖标枪、铁饼、链球和铅球，并配有相应的二维姿态序列、帧级阶段注释和官方距离标签。我们将基于广播获得的姿态进行投掷距离预测的任务形式化为回归问题，并将全序列建模与一种阶段条件方法进行比较，该方法将每次尝试划分为四个技术阶段，并融合阶段特定的运动表示。在多个基于图的姿态编码器的实验中，当引入阶段结构时，预测误差一致减少，表明阶段对齐建模提供了比将整个运动视为非结构化序列更稳定的表示。这些发现表明，显式的生物力学阶段信息是精英田径投掷中基于姿态的性能建模的有用归纳先验。除了得分预测外，该数据集和协议为未来的细粒度技术分析、可解释的体育分析以及基于广播视频的稳健建模提供了基础。

View on arXiv Download PDF AI Translation

cs.CV / 177 / 2610.16409

Recovering from Pose-Selection Failures in Repeated Visual Localization

从姿态选择失败中恢复的重复视觉定位

Kim, Jaehyun, Kang, Wonseok, Kim, Tae-Wan

Abstract

Repetitive structures can give an incorrect pose candidate more inlier support than an accurate one, causing visual localization to select it. History-Guided Pose Reselection (HPR) uses coarse position forecasts from up to five past localization outputs to distinguish well-separated current-query candidates. The best-fitting forecast proposes an alternative; HPR accepts it only if every well-fitting forecast is much closer to it than to the localizer's pose. HPR needs no odometry and returns the selected candidate unchanged, using prediction for selection rather than pose refinement. With one development-set threshold, HPR improves recall within 5 meters and 10 degrees in all 12 main configurations of Hierarchical Localization (HLoc), MASt3R-based pose generation, and JIST-based retrieval, losing successes on fewer than 0.4 percent of queries. On NAVER Gangnam B1, HPR corrects 97 HLoc failures and loses six successes; gains persist within 0.25 meters and 2 degrees on all three HLoc test sets. Median HPR runtime on HLoc candidates is below one millisecond, excluding pose generation. Recovery weakens under less frequent localization.

Chinese Translation

重复结构可能会使不正确的姿态候选者获得比准确姿态更多的内点支持，从而导致视觉定位选择错误的候选者。历史引导姿态重新选择（History-Guided Pose Reselection, HPR）利用来自最多五个过去定位输出的粗略位置预测来区分当前查询候选者。最佳拟合的预测提出了一个替代方案；HPR 仅在每个拟合良好的预测比定位器的姿态更接近时才接受该替代方案。HPR 不需要里程计，并且在选择时返回未改变的候选者，使用预测进行选择而不是姿态精细化。通过一个开发集阈值，HPR 在所有 12 个主要配置的层次定位（Hierarchical Localization, HLoc）、基于 MASt3R 的姿态生成和基于 JIST 的检索中，在 5 米和 10 度内提高了召回率，成功率下降不到 0.4% 的查询。在 NAVER 江南 B1，HPR 修正了 97 个 HLoc 失败并失去了六个成功；在所有三个 HLoc 测试集中，增益保持在 0.25 米和 2 度以内。HLoc 候选者的中位 HPR 运行时间低于一毫秒，不包括姿态生成。在定位频率较低的情况下，恢复效果减弱。

View on arXiv Download PDF AI Translation

cs.CV / 178 / 2610.16417

Not All Challenges Are Equal: Evaluating Interactive Liveness Detection Against Real-Time Face Swaps

挑战并非平等：评估交互式活体检测对抗实时换脸技术

Firc, Anton, Drtil, David, Perešíni, Martin, Šalko, Milan, Staněk, Vojtěch, Malinka, Kamil

Abstract

Remote identity verification assumes that a camera stream shows a live person, yet public real-time face-swap tools can replace that person before the stream reaches the verifier. Passive deepfake detectors generalize poorly to new generators, while interactive challenges are often selected without evidence of which gestures expose current attacks. We evaluate randomized gesture challenges in 108 live sessions with six subjects, 24 impersonated identities, and two face-swap tools, DeepFaceLive and FaceFusion, injected through a virtual camera. The static calibration segment reaches a session-level AUC of 0.878 with a frozen UCF spatial detector, while the mean across the active-gesture windows reaches 0.988. Pose and combined gestures have the highest point estimates, whereas expressions show no clear improvement. Hand occlusion against FaceFusion, configured with an occlusion mask, scores below the static baseline. The end-to-end time required to satisfy the validator also distinguishes attacks from genuine sessions, achieving a session-level AUC of 0.732, although this signal includes synthesis and transport delays. These results show that challenge sets should be tested against current attack configurations and that response time can provide an additional, deployment-specific signal.

Chinese Translation

远程身份验证假设摄像头视频流显示的是一个活生生的人，然而公共的实时换脸工具可以在视频流到达验证者之前替换掉该人。被动的深伪检测器在新生成器上泛化效果较差，而交互式挑战往往在没有证据表明哪些手势能够揭示当前攻击的情况下被选择。我们在108个实时会话中评估了随机手势挑战，涉及六名受试者、24个冒充身份和两个通过虚拟摄像头注入的换脸工具，DeepFaceLive和FaceFusion。静态校准阶段的会话级AUC达到了0.878，使用的是冻结的UCF空间检测器，而主动手势窗口的平均值达到了0.988。姿态和组合手势的点估计最高，而表情则没有明显改善。针对FaceFusion的手部遮挡（配置了遮挡掩码）得分低于静态基线。满足验证者所需的端到端时间也能区分攻击与真实会话，达到了0.732的会话级AUC，尽管这一信号包含了合成和传输延迟。这些结果表明，挑战集应针对当前攻击配置进行测试，并且响应时间可以提供额外的、特定于部署的信号。

View on arXiv Download PDF AI Translation

cs.CV / 179 / 2610.16441

VisExMEM: Visual Evidence-Based Explainable Multimodal Evaluation Metric

VisExMEM：基于视觉证据的可解释多模态评估指标

Roy, Subhadeep, Leiter, Christoph, Eger, Steffen

Abstract

Current image-text evaluation often relies on a global score, which gives limited insight into which semantic details are supported by the image. Explainable metrics provide rationales, but the visual evidence supporting those rationales can still be difficult to verify. We introduce VisExMEM, a reference-free evaluator that decomposes captions into atomic claims, localizes visual evidence, verifies each claim, and averages claim-level support into a global score. Explanations use the same evidence and decisions. We train Qwen3.5-based 4B and 9B models on 100K samples with multi-task SFT, counterfactual refinement, and evidence-aware GDPO. Across eight benchmarks, VisExMEM-9B achieves the highest descriptive composite among open-source evaluators in our comparison, exceeding VQAScore by 11.7 points while remaining competitive with GPT-5.6 Sol. We also introduce AGITA-Bench, containing 600 human-annotated image-text pairs and 8.7K atomic claims for global and claim-level evaluation. For untuned Qwen3.5 and GPT-5.6 Sol controls, decomposed prompting raises the same composite by 3.6-5.1 points relative to holistic prompting. Human evaluation shows that VisExMEM explanations outperform prior explainable metrics and approach GPT-5.6 Sol.

Chinese Translation

当前的图像-文本评估通常依赖于一个全局评分，这对图像所支持的语义细节提供的洞察有限。可解释的指标提供了理由，但支持这些理由的视觉证据仍然难以验证。我们提出了VisExMEM，这是一种无参考评估器，它将标题分解为原子声明，定位视觉证据，验证每个声明，并将声明级支持的平均值汇总为一个全局评分。解释使用相同的证据和决策。我们在100K样本上训练了基于Qwen3.5的4B和9B模型，采用多任务SFT、反事实细化和证据感知GDPO。在八个基准测试中，VisExMEM-9B在我们的比较中获得了开源评估器中最高的描述性复合评分，超出VQAScore 11.7分，同时与GPT-5.6 Sol保持竞争力。我们还推出了AGITA-Bench，包含600对人工标注的图像-文本对和8.7K个原子声明，用于全局和声明级评估。对于未调优的Qwen3.5和GPT-5.6 Sol控制，分解提示相对于整体提示提高了3.6-5.1分的相同复合评分。人类评估表明，VisExMEM的解释优于先前的可解释指标，并接近GPT-5.6 Sol。

View on arXiv Download PDF AI Translation

cs.CV / 180 / 2610.16444

Pixels or Code? Benchmarking Instruction-Guided Visual Artifact Operation

像素还是代码？指导性视觉工件操作的基准测试

Guo, Shun, Shi, Shuizhou, Wang, Boao, Xiong, Wei, Sun, Siyu, Li, Jiangtong, Li, Jie

Abstract

Visual artifacts such as charts, technical diagrams, webpages, and document pages underpin scientific, commercial, and engineering workflows. Editing these artifacts requires direct modification, dependency-aware updates, preservation of unaffected content, and compliance with domain constraints. We define this capability as Visual Artifact Operation (VAO). Prior benchmarks test parts of this problem, including cascading updates in charts. We study direct edits, dependent updates, and preservation across visual domains and pixel, code, and agent paradigms under one image-only protocol. We introduce Visual Artifact Operation Bench (VAO-Bench), comprising 1,573 manually screened tasks across nine artifact categories and three difficulty levels. VAO-Bench uses executable artifacts and deterministic rendering to trace reference transformations and annotate direct edits, dependent updates, and preservation constraints. All systems receive the same source image and instruction; we evaluate pixel-space editors, code-generating multimodal systems, and tool-using agents on direct-edit accuracy, dependency-aware update, content preservation, domain validity, and visual quality. Experiments show that executable-code systems better support dependent updates, while pixel-space editors offer faster execution and strong visual quality. Explicitly stating dependencies improves both paradigms, identifying implicit-update inference as a central bottleneck. VAO-Bench provides a reproducible foundation for dependable visual artifact editing across generation paradigms.

Chinese Translation

视觉工件，如图表、技术图示、网页和文档页面，是科学、商业和工程工作流程的基础。编辑这些工件需要直接修改、依赖感知更新、保持未受影响内容的完整性，以及遵循领域约束。我们将这一能力定义为视觉工件操作（Visual Artifact Operation, VAO）。先前的基准测试只测试了该问题的部分内容，包括图表中的级联更新。我们研究了在一个仅限图像的协议下，跨视觉领域以及像素、代码和代理范式的直接编辑、依赖更新和内容保持。我们引入了视觉工件操作基准（Visual Artifact Operation Bench, VAO-Bench），该基准包含1,573个经过人工筛选的任务，涵盖九个工件类别和三个难度级别。VAO-Bench使用可执行工件和确定性渲染来追踪参考变换，并注释直接编辑、依赖更新和内容保持约束。所有系统接收相同的源图像和指令；我们评估像素空间编辑器、代码生成的多模态系统和使用工具的代理在直接编辑准确性、依赖感知更新、内容保持、领域有效性和视觉质量等方面的表现。实验表明，可执行代码系统更好地支持依赖更新，而像素空间编辑器则提供更快的执行速度和较强的视觉质量。明确说明依赖关系改善了这两种范式，识别隐式更新推断作为一个核心瓶颈。VAO-Bench为在生成范式中可靠的视觉工件编辑提供了可重复的基础。

View on arXiv Download PDF AI Translation

cs.CV / 181 / 2610.16448

Tikhonov-Stabilized Bezier Representation Forecasting for Training-free Diffusion Acceleration

Tikhonov 稳定化的贝塞尔表示预测用于无训练扩散加速

Zhu, Lei, Lin, Mujie, Zheng, Ruochong, Wang, Guangyi, Li, Hao, Jin, Peng, Liu, Chang, Chen, Jie

Abstract

Diffusion models, particularly Diffusion Transformers, achieve strong image and video generation quality but remain expensive at inference time due to repeated denoiser evaluations. Feature caching offers a deployment-friendly acceleration strategy by reusing or predicting intermediate representations without retraining the generator or modifying the sampler. However, existing cache-then-forecast methods often rely on Taylor-style finite-difference extrapolation, which can become unstable over longer cache intervals, and typically predict local module outputs whose errors may accumulate through subsequent denoiser blocks. We propose \textbf{BeziCast}, a training-free diffusion acceleration framework that forecasts output-proximal denoising representations using low-order B'ezier trajectories. BeziCast estimates B'ezier control points via Tikhonov-stabilized fitting, providing a smooth, capacity-controlled temporal parameterization that avoids explicit high-order derivative estimation and decouples trajectory capacity from cache-query density. To moderate aggressive extrapolation, we further derive equivalent forecasting weights and introduce a convex-hull-inspired guardrail that detects high-risk predictions and softly pulls them toward a conservative simplex-projected estimate. Extensive evaluations across advanced image and video diffusion models demonstrate the effectiveness of BeziCast. In particular, BeziCast accelerates FLUX.1 by up to (4.79\times) and HunyuanVideo by up to (4.11\times), while preserving substantially better generation quality than competing acceleration baselines.

Chinese Translation

扩散模型，特别是扩散变换器，虽然在图像和视频生成质量上表现出色，但由于需要反复评估去噪器，推理时仍然成本高昂。特征缓存提供了一种友好的加速策略，通过重用或预测中间表示，而无需重新训练生成器或修改采样器。然而，现有的缓存-然后-预测方法通常依赖于泰勒风格的有限差分外推，这在较长的缓存间隔中可能变得不稳定，并且通常预测局部模块输出，其误差可能在后续的去噪器块中累积。我们提出了 extbf{BeziCast}，一种无训练的扩散加速框架，使用低阶贝塞尔轨迹预测输出近端去噪表示。BeziCast 通过 Tikhonov 稳定化拟合估计贝塞尔控制点，提供平滑、容量受控的时间参数化，避免显式高阶导数估计，并将轨迹容量与缓存查询密度解耦。为了适度控制激进的外推，我们进一步推导了等效的预测权重，并引入了受凸包启发的护栏，以检测高风险预测并将其柔性拉向保守的单纯形投影估计。对先进的图像和视频扩散模型进行的广泛评估证明了 BeziCast 的有效性。特别是，BeziCast 将 FLUX.1 的加速提高至 (4.79 imes)，将 HunyuanVideo 的加速提高至 (4.11 imes)，同时在生成质量上显著优于竞争的加速基线。

View on arXiv Download PDF AI Translation

cs.CV / 182 / 2610.16452

Not Every Cue Is Evidence: Cue Supportability for Forensic Distillation in Face Forgery Detection

并非每个线索都是证据：面部伪造检测中的线索支持性与法医学提炼

Zhou, Binjia, Luo, Dawei, Chen, Shuai, Xu, Feng, Li, Haoyuan, Seow, Jia wen, Wang, Jiachi, Lei, Jie, Feng, Zunlei, Bei, Yijun

Abstract

Explainable face forgery detection requires not only correct judgments but also forensic evidence supported by the input image. However, training cues are typically derived from the forgery generation pipeline. Since images available to the model are of limited quality and resolution, it is difficult to observe the fine-grained traces these cues refer to. Training methods that further supply such cues as privileged supervision exacerbate the problem, as the student is taught to articulate evidence it cannot perceive. The student consequently resorts to templated or over-generalized expressions, giving rise to the correct-answer, wrong-evidence (CAWE) failure. We propose a dual supportability audit that filters privileged cues before distillation. Each candidate cue must be predictable from the remaining privileged context and causally tied to high-attribution visual regions in the current image. Only cues satisfying both conditions receive substantial distillation weight, preventing unsupported details from leaking into the student. Experiments across eight benchmarks show that our method maintains or improves detection accuracy while significantly raising explanation faithfulness compared to existing paradigms.

Chinese Translation

可解释的面部伪造检测不仅需要正确的判断，还需要由输入图像支持的法医学证据。然而，训练线索通常来源于伪造生成流程。由于模型可用的图像质量和分辨率有限，观察这些线索所指代的细微痕迹变得困难。进一步提供这些线索作为特权监督的训练方法加剧了这一问题，因为学生被教导去表达其无法感知的证据。因此，学生往往诉诸于模板化或过于泛化的表达，导致正确答案但错误证据（CAWE）失败。我们提出了一种双重支持性审计，在提炼之前过滤特权线索。每个候选线索必须能够从剩余的特权上下文中预测，并与当前图像中的高归因视觉区域因果关联。只有满足这两个条件的线索才能获得显著的提炼权重，从而防止不支持的细节渗入学生模型。跨越八个基准的实验表明，我们的方法在保持或提高检测准确性的同时，显著提升了解释的可信度，相较于现有范式具有更好的表现。

View on arXiv Download PDF AI Translation

cs.CV / 183 / 2610.16455

GHOST: Geometry-Constrained Hybrid Diffusion for Open-Vocabulary 3D Scene Graph Generation

GHOST：用于开放词汇3D场景图生成的几何约束混合扩散

Feng, Changqun, Yin, Wangxiandi, Hu, Xin, Dai, Ruiting, Yang, Feiyang, Wang, Jielei, He, Tao

Abstract

3D scene graph generation requires understanding both object semantics and metric spatial relations. However, existing open-vocabulary methods often struggle with spatial predicates because language similarity reflects semantic co-occurrence rather than explicit geometry. We propose GHOST, a Geometry-constrained Hybrid diffusion framework for Open-vocabulary 3D Scene graph generaTion. GHOST jointly models continuous node geometry and discrete edge predicates, following a geometry-to-constraint-to-predicate denoising process to ensure geometrically consistent relation prediction. It introduces language-grounded constraint induction, which maps predicate embeddings to differentiable geometric constraints that guide both layout refinement and predicate reweighting. A Geometry-Topology Collaborative Denoiser further couples node and edge reasoning through noise-adaptive interaction and orientation-aware modeling. Experiments on 3DSSG demonstrate state-of-the-art performance in both closed-set and open-vocabulary settings, including over 17-point gains over RelationField on spatial predicates.

Chinese Translation

3D场景图生成需要理解对象语义和度量空间关系。然而，现有的开放词汇方法在空间谓词方面常常面临困难，因为语言相似性反映的是语义共现而非显式几何。我们提出了GHOST，一个用于开放词汇3D场景图生成的几何约束混合扩散框架。GHOST联合建模连续节点几何和离散边缘谓词，遵循几何-约束-谓词去噪过程，以确保几何一致的关系预测。它引入了基于语言的约束诱导，将谓词嵌入映射到可微分的几何约束，从而指导布局优化和谓词重加权。几何-拓扑协同去噪器进一步通过噪声自适应交互和方向感知建模耦合节点和边缘推理。在3DSSG上的实验表明，在封闭集和开放词汇设置中均实现了最先进的性能，包括在空间谓词上相较于RelationField超过17分的提升。

View on arXiv Download PDF AI Translation

cs.CV / 184 / 2610.16465

Teaching Vision-Language Models to Decline: Support Completion Against Mirage

教会视觉-语言模型拒绝：针对幻影的支持完成

Ahmed, Alfaj Uddin, Miah, Md. Shaown, Hasan, Md Arid, Ahmed, Syed Ishtiaque, Chowdhury, Sayeed Shafayet

Abstract

Vision-language models answer questions about images they were never shown, a failure called mirage. Such an answer can be correct without being justified by the evidence, and the difference matters wherever a model's output is checked by someone who did not see its input. We locate the failure in the training objective: cross-entropy separates a model that reads the image from one that ignores it by only the conditional mutual information between answer and image, and no training example rewards declining. We formalise support completion, TIDE, which also shows each training question with a blank image, a same-domain wrong image and an out-of-domain image, with targets that decline, and prove that under TIDE every image-blind predictor is suboptimal by at least 1.04 nats per example, that the optimum declines exactly on the counterfactual kinds while keeping its behaviour on valid images, and that for any finite model the error of its answer-or-decline decision, declining on valid input plus answering on invalid input, is bounded below by how well its representation separates valid from invalid evidence. On Qwen2.5-VL-7B under a frozen held-out protocol with two judges, the three-condition predecessor of TIDE reduced the invalid-evidence mirage rate from 0.759 to 0.097 and the blank-image rate from 0.700 to 0.003 at a false-abstention cost of 0.126; label-free grounding raised the rate to 0.859, a likelihood-advantage contrast moved it only to 0.718 because its empty-image term carries no gradient, and grounding applied after support completion re-opened the mirage. The four-condition protocol on three base models, with seeds and component ablations, is in progress and not reported here.

Chinese Translation

视觉-语言模型能够回答关于其未见图像的问题，这种失败被称为幻影。这样的回答可能是正确的，但并未得到证据的支持，这一差异在模型的输出被未见其输入的人检查时显得尤为重要。我们将这一失败定位于训练目标：交叉熵仅通过答案与图像之间的条件互信息来区分一个读取图像的模型与一个忽略图像的模型，而没有任何训练示例奖励拒绝。我们形式化了支持完成（support completion），提出了 TIDE，该方法还展示了每个训练问题与一个空白图像、一个同域错误图像和一个跨域图像，同时目标是拒绝，并证明在 TIDE 下，每个无图像预测器的表现至少比最优解低 1.04 nats 每个示例，最优解在反事实类型上恰好拒绝，同时保持其在有效图像上的行为，并且对于任何有限模型，其答案或拒绝决策的错误，即在有效输入上拒绝并在无效输入上回答，其下界由其表示在多大程度上将有效证据与无效证据分离所决定。在 Qwen2.5-VL-7B 上，在冻结的保留协议下，经过两位评审的 TIDE 的三条件前身将无效证据的幻影率从 0.759 降低到 0.097，空白图像率从 0.700 降低到 0.003，假拒绝成本为 0.126；无标签的基础对接将该比率提高到 0.859，基于可能性优势的对比仅将其移动到 0.718，因为其空图像项不带梯度，而在支持完成后应用的对接重新开启了幻影。关于三个基础模型的四条件协议，包含种子和组件消融，目前正在进行中，未在此报告。

View on arXiv Download PDF AI Translation

cs.CV / 185 / 2610.16471

Auditing Observation-Quality Shortcuts in the Evaluation of Person-Centered Motion for Video Anomaly Detection

审计视频异常检测中以人为中心的运动评估的观察质量捷径

Kim, Chaehyun, Lim, Changwon

Abstract

Video anomaly detection (VAD) identifies deviations from normal patterns, but such deviations do not necessarily correspond to safety-critical events. We focus on human-centered emergencies and violence and investigate whether motion-based anomaly discrimination reflects task-relevant motion information or observation-quality differences arising from person detection, pose estimation, and tracking. We introduce an observation-quality-aware evaluation framework that combines quality diagnostics, quality balancing, motion orthogonalization, and robustness analyses. On human-centered categories of UCF-Crime, quality-only features match or exceed the discrimination of multi-cue motion in the original cohort, revealing a strong observation-quality shortcut. After cardinality matching brings all 11 measured quality variables below a prespecified marginal balance threshold, multi-cue motion consistently outperforms quality-only features across Logistic Regression, Random Forest, and RBF-SVM, yielding ROC-AUC gains of 0.046-0.141. Because nonlinear quality-only prediction remains above chance after matching, we interpret this result as additional discrimination beyond measured marginal quality rather than as evidence of quality-free or causal motion understanding. These findings show that high VAD performance should not be interpreted as evidence of abnormal-motion modeling without auditing the reliability of the person-observation pipeline. Our study provides a systematic protocol for evaluating observation-quality shortcuts and for assessing whether motion-based discrimination survives explicit control of detection, pose-estimation, and tracking reliability.

Chinese Translation

视频异常检测（VAD）识别与正常模式的偏差，但这些偏差不一定对应于安全关键事件。我们关注以人为中心的紧急情况和暴力行为，探讨基于运动的异常辨别是否反映了与任务相关的运动信息或因人检测、姿态估计和跟踪而产生的观察质量差异。我们引入了一种关注观察质量的评估框架，该框架结合了质量诊断、质量平衡、运动正交化和鲁棒性分析。在以人为中心的UCF-Crime类别中，仅基于质量的特征与原始队列中的多线索运动的辨别能力相匹配或超过，揭示了一个强大的观察质量捷径。在基数匹配将所有11个测量的质量变量降至预设的边际平衡阈值以下后，多线索运动在逻辑回归、随机森林和RBF-SVM中始终优于仅基于质量的特征，ROC-AUC的增益为0.046-0.141。由于非线性仅基于质量的预测在匹配后仍高于偶然水平，我们将这一结果解释为超出测量边际质量的额外辨别，而不是质量自由或因果运动理解的证据。这些发现表明，高VAD性能不应被解读为异常运动建模的证据，而不审计人观察管道的可靠性。我们的研究提供了一种系统的协议，用于评估观察质量捷径，并评估基于运动的辨别是否在明确控制检测、姿态估计和跟踪可靠性的情况下得以保留。

View on arXiv Download PDF AI Translation

cs.CV / 186 / 2610.16482

MiMaAlign: Hierarchical Micro-Macro Alignment for Physically Plausible Video Generation

MiMaAlign：用于物理合理视频生成的层次微观-宏观对齐

Liu, Yansong, Gong, Litong, Li, Weijie, Ge, Tiezheng, Zeng, Ming

Abstract

Text-to-video (T2V) diffusion models have made remarkable progress in generating high-fidelity visuals, yet their outputs still often exhibit physically implausible phenomena, including discontinuous dynamics and implausible event interactions. Existing representation alignment methods mainly transfer generic video features or short-range motion cues, treating physical plausibility as a single-level representation problem. To address this issue, we propose MiMaAlign, a hierarchical micro-macro alignment framework that transfers complementary dynamics-oriented and event-oriented knowledge into different abstraction depths of a pretrained video diffusion model. Specifically, micro-level continuous dynamics alignment imposes constraints by aligning diffusion features with a continuous-time motion representation learned by a Neural Ordinary Differential Equation (Neural ODE), while macro-level event-commonsense alignment provides representations of object interactions and event semantics through an instruction-prompted vision-language model (VLM). With training-only auxiliary objectives, we align the complementary supervision from the two teachers with shallower and deeper generator features, respectively, enabling the generator to internalize the corresponding physical priors. Empirical evaluations show that, on CogVideoX-2B, MiMaAlign achieves a relative improvement of up to 21.0% in the Physical Commonsense (PC) score on the VideoPhy benchmark over the baseline, and also delivers substantial gains on the highly challenging VideoPhy2 benchmark, while maintaining general video quality on VBench and being supported by blind user comparisons.

Chinese Translation

文本到视频（T2V）扩散模型在生成高保真视觉效果方面取得了显著进展，但其输出仍然常常表现出物理上不合理的现象，包括不连续的动态和不合理的事件交互。现有的表示对齐方法主要转移通用视频特征或短期运动线索，将物理合理性视为单层次的表示问题。为了解决这一问题，我们提出了MiMaAlign，一个层次微观-宏观对齐框架，将互补的动态导向和事件导向知识转移到预训练视频扩散模型的不同抽象层次。具体而言，微观层次的连续动态对齐通过将扩散特征与由神经常微分方程（Neural ODE）学习的连续时间运动表示对齐来施加约束，而宏观层次的事件常识对齐则通过指令提示的视觉-语言模型（VLM）提供对象交互和事件语义的表示。通过仅在训练中使用辅助目标，我们分别将来自两个教师的互补监督与较浅和较深的生成器特征对齐，使生成器能够内化相应的物理先验。实证评估表明，在CogVideoX-2B上，MiMaAlign在VideoPhy基准测试中相较于基线在物理常识（PC）得分上实现了高达21.0%的相对提升，并在极具挑战性的VideoPhy2基准测试中也取得了显著的进展，同时在VBench上保持了良好的视频质量，并得到了盲用户比较的支持。

View on arXiv Download PDF AI Translation

cs.CV / 187 / 2610.16500

Self-Verifier Capture in Unified Visual Generators: Diagnosing and Repairing Closed-Loop Self-Reward

统一视觉生成器中的自我验证捕获：诊断与修复闭环自我奖励

Chen, Binyan, Pan, Tong

Abstract

Image-space reward requires decoding and a separate judge; unified visual generators can instead reuse their understanding pathway for decode-free, self-supplied reward. Optimizing against this verifier closes the self-reward loop; shared trainable weights let one update move generation and scoring even with the reward gradient stopped. We call the resulting failure self-verifier capture: self-reward rises while independently measured quality falls. Matched semantic/denoising by shared/frozen cells on Scale-RAE isolate two routes: the policy exploits on-manifold blind spots of a frozen semantic judge, while shared updates additionally rewrite the judge. A frozen judge also becomes stale, and repeated self-refresh yields progressively weaker recoveries. The diagnosis yields a two-part repair: locking the warmed reward read blocks rewriting, while sparse, parameter-disjoint external anchors counter residual exploitation. Unanchored rollouts retain semantic reward across multiple latent noise levels, providing dense latent reward that terminal image-space judging cannot supply, while a drift-triggered controller allocates anchors to high-drift rollouts and propagates their calibration within each group. At matched budgets, the controller improves on fixed-dose anchoring, increasingly so as coverage tightens. Both the unanchored failure and the repair transfer to the higher-performing, architecturally distinct Show-o2. Across both bases, the complete method significantly outperforms a construct-matched every-rollout judge on GenEval and DPG-Bench while externally scoring only 1/16 of rollouts. At 1/32 coverage, it scores higher in all four comparisons and remains jointly non-inferior.

Chinese Translation

图像空间奖励需要解码和一个独立的评判者；而统一视觉生成器可以重用其理解路径，实现无解码、自供奖励的方式。针对这一验证者的优化关闭了自我奖励循环；共享的可训练权重使得即使在奖励梯度停止的情况下，也能更新生成和评分。我们称这种结果为自我验证捕获：自我奖励上升，而独立测量的质量下降。通过共享/冻结单元在Scale-RAE上进行的匹配语义/去噪，隔离了两条路径：策略利用了冻结语义评判者的流形盲点，而共享更新则额外重写了评判者。冻结的评判者也会变得过时，重复的自我刷新导致逐渐减弱的恢复效果。诊断结果提出了两部分的修复方案：锁定加热的奖励读取块以防止重写，同时稀疏的、参数不相交的外部锚点抵消剩余的利用。无锚的展开在多个潜在噪声水平上保持语义奖励，提供了密集的潜在奖励，而终端图像空间评判无法提供；同时，漂移触发的控制器将锚点分配给高漂移的展开，并在每个组内传播其校准。在匹配预算下，控制器在固定剂量锚定上有所改善，随着覆盖范围的收紧，改善程度越来越明显。无锚的失败和修复都转移到更高性能、架构上有明显不同的Show-o2。在这两个基础上，完整的方法在GenEval和DPG-Bench上显著优于构造匹配的每次展开评判，同时外部评分仅为1/16的展开。在1/32的覆盖下，它在所有四个比较中得分更高，并保持共同的非劣性。

View on arXiv Download PDF AI Translation

cs.CV / 188 / 2610.16502

KARAMEL: Knowledge-Aware Ranking for Multimodal Entity Linking in the Arts via Heterogeneous Graphs

KARAMEL：基于异构图的艺术领域多模态实体链接的知识感知排序

Scaringi, Raffaele, Sierra-Múnera, Alejandro, Vessio, Gennaro, Castellano, Giovanna, Krestel, Ralf

Abstract

Discovering and disambiguating entities from multimodal and heterogeneous data is a key challenge in discovery science, where relevant knowledge is often implicit and distributed across modalities. We present KARAMEL, a framework for multimodal entity linking (MEL) - the task of aligning visual and textual mentions to entities in a knowledge base - in the cultural heritage domain. Unlike prior vision-language models that primarily rely on surface-level multimodal alignment, our approach incorporates sociohistorical and contextual knowledge through a knowledge-aware architecture that fuses a vision-language encoder with a heterogeneous graph neural network. This design enables contextual reasoning over complex relational structures and supports robust entity disambiguation under ambiguous or underspecified evidence. To benchmark multimodal discovery in this setting, we introduce WikiMuSA (Wikidata-based Multimodal Semantic data for Art), a new dataset that links artworks to Wikidata entities via images, textual descriptions, and structured contextual knowledge. Experimental results show that KARAMEL surpasses strong baselines, particularly on implicit or abstract mentions, while ablation studies highlight the critical role of domain-specific, structured context in learning expressive representations. While instantiated in the cultural heritage domain, KARAMEL addresses a general knowledge discovery problem and applies to other scientific settings involving multimodal observations and heterogeneous background knowledge. Both the dataset and source code are made publicly available.

Chinese Translation

从多模态和异构数据中发现和消歧义实体是发现科学中的一个关键挑战，其中相关知识通常是隐含的，并分布在不同的模态中。我们提出了KARAMEL，一个用于多模态实体链接（MEL）的框架——该任务是将视觉和文本提及与知识库中的实体对齐——应用于文化遗产领域。与以往主要依赖表层多模态对齐的视觉-语言模型不同，我们的方法通过知识感知架构结合视觉-语言编码器与异构图神经网络，融入了社会历史和上下文知识。这一设计使得在复杂关系结构上进行上下文推理成为可能，并在模糊或不明确的证据下支持稳健的实体消歧义。为了在这一背景下基准化多模态发现，我们引入了WikiMuSA（基于Wikidata的艺术多模态语义数据），这是一个新的数据集，通过图像、文本描述和结构化上下文知识将艺术作品与Wikidata实体链接起来。实验结果表明，KARAMEL超越了强基线，特别是在隐含或抽象提及上，而消融研究强调了领域特定的结构化上下文在学习表达性表示中的关键作用。尽管在文化遗产领域中实例化，KARAMEL解决了一个通用的知识发现问题，并适用于涉及多模态观察和异构背景知识的其他科学环境。数据集和源代码均已公开。

View on arXiv Download PDF AI Translation

cs.CV / 189 / 2610.16529

Deep Learning-based Mechanical Characterization of Synthetically Generated Plant Fibers under Transverse Compression

基于深度学习的合成植物纤维横向压缩下的机械特性表征

Kegkeroglou, Nikolaos, Sabillah, Bilal, Inglezou, Myrto, Placet, Vincent, Clevy, Céndric, Porichis, Antonios

Abstract

Plant fibers are increasingly used as sustainable reinforcements in composites, but the intrinsic variability of their microstructure causes substantial scatter in transverse mechanical behavior, complicating composite design. Characterizing this behavior relies on finite element (FE) simulations requiring several minutes of high-performance computing per fiber, making high-throughput characterization across realistic fiber populations impractical. In this study, we propose a deep learning framework replacing this FE bottleneck with a fast pipeline. It includes two coupled recursive U-Nets predicting the step-by-step evolution of fiber deformation and internal stress fields under transverse compression, followed by a regression CNN estimating the transverse Young's modulus from the final stress field. Unlike prior approaches predicting stress fields for fixed configurations or idealized geometries, our framework recursively resolves the full temporal stress evolution for synthetic fibers derived from real hemp cross-sections, with or without a lumen. Trained on 1,000 synthetic fiber geometries with FE ground truth, it reproduces the stress field sequence with a Structural Similarity Index Measure (SSIM) of $0.9574 \pm 0.0074$ and a Root Mean Squared Error (RMSE) of $0.0345 \pm 0.0079$. It estimates transverse Young's modulus with an $R^2$ of $0.938 \pm 0.024$, a Mean Absolute Error (MAE) of $24.31 \pm 1.23$ MPa, and a Mean Absolute Percentage Error (MAPE) of $5.25 \pm 0.91\%$. Inference requires under one second per fiber versus five minutes per FE simulation, drastically reducing computational cost. This offers a fast, reliable tool for mechanical characterization of synthetic plant fibers, with promising applications in designing natural fiber-reinforced composites.

Chinese Translation

植物纤维作为可持续增强材料在复合材料中的应用日益增多，但其微观结构的内在变异性导致横向机械行为存在显著的离散性，从而使复合材料设计变得复杂。对这种行为的表征依赖于有限元（FE）模拟，每根纤维需要几分钟的高性能计算，这使得在现实纤维群体中进行高通量表征变得不切实际。在本研究中，我们提出了一种深度学习框架，用快速管道替代这一FE瓶颈。该框架包括两个耦合的递归U-Net，预测纤维在横向压缩下的变形和内部应力场的逐步演变，随后通过回归卷积神经网络（CNN）从最终应力场估计横向杨氏模量。与之前针对固定配置或理想化几何形状预测应力场的方法不同，我们的框架递归地解决了来自真实大麻横截面的合成纤维的完整时间应力演变，无论是否存在腔道。该框架在1,000个具有FE真实值的合成纤维几何形状上进行训练，重现了应力场序列，其结构相似性指数测量（SSIM）为$0.9574 imes 0.0074$，均方根误差（RMSE）为$0.0345 imes 0.0079$。它以$R^2$为$0.938 imes 0.024$的精度估计横向杨氏模量，平均绝对误差（MAE）为$24.31 imes 1.23$ MPa，平均绝对百分比误差（MAPE）为$5.25 imes 0.91 ext{%}$。推理每根纤维所需时间不到一秒，而每次FE模拟则需五分钟，显著降低了计算成本。这为合成植物纤维的机械特性表征提供了一种快速、可靠的工具，并在设计天然纤维增强复合材料方面展现了良好的应用前景。

View on arXiv Download PDF AI Translation

cs.CV / 190 / 2610.16545

Heuristics are Not Enough: Calibrated and Corrected Caching for Video Diffusion Acceleration

启发式方法不足：用于视频扩散加速的校准和修正缓存

Jiang, Jiyue, Huo, Zhiyu, Chen, Yuhan, Shao, Yanruisheng, Zhang, Gong, Deng, Lei

Abstract

Feature caching has become a mainstream training-free strategy for accelerating video diffusion models by reusing intermediate outputs across denoising steps. However, existing adaptive methods rely on heuristic difference measures misaligned with actual generation quality, employ a single global threshold that ignores heterogeneous phase dynamics, and never validate their budget estimates against true approximation error, allowing systematic bias to propagate uncorrected. We propose \textbf{AlignCache}, a closed-loop caching framework that directly aligns scheduling decisions with their true influence on the final generated video through three tightly coupled components: (1) phase-adaptive budget routing that partitions the denoising trajectory into structurally distinct phases, each governed by an independent threshold and a staleness-aware guard; (2) closed-loop threshold calibration that exploits counterfactual error measurement at the classifier-free guided output level to adjust decision boundaries online via an EMA feedback loop; and (3) momentum-augmented draft synthesis combining direction-magnitude gated residual extrapolation with anchor-guided unconditional correction. By transforming scheduling from a static risk accumulator into a feedback-driven, self-correcting controller, AlignCache consistently achieves superior acceleration-quality trade-offs across state-of-the-art video generation models. Our code is available at https://github.com/jiangjyjy/AlignCache.

Chinese Translation

特征缓存已成为一种主流的无训练策略，通过在去噪步骤中重用中间输出，加速视频扩散模型。然而，现有的自适应方法依赖于与实际生成质量不一致的启发式差异度量，采用单一的全局阈值，忽视了异质相位动态，并且从未将其预算估计与真实近似误差进行验证，导致系统性偏差未得到修正。我们提出了 extbf{AlignCache}，一个闭环缓存框架，通过三个紧密耦合的组件，直接将调度决策与其对最终生成视频的真实影响对齐：(1) 相位自适应预算路由，将去噪轨迹划分为结构上不同的相位，每个相位由独立的阈值和考虑时效性的保护措施管理；(2) 闭环阈值校准，利用分类器无关的引导输出级别的反事实误差测量，通过EMA反馈循环在线调整决策边界；(3) 动量增强草稿合成，将方向-幅度门控残差外推与锚点引导的无条件修正相结合。通过将调度从静态风险累积器转变为反馈驱动的自我修正控制器，AlignCache在最先进的视频生成模型中始终实现了优越的加速-质量权衡。我们的代码可在 https://github.com/jiangjyjy/AlignCache 获取。

View on arXiv Download PDF AI Translation

cs.CV / 191 / 2610.16548

NoduleFlow: Automatic 3D Lung Nodule Detection and Segmentation

NoduleFlow：自动化三维肺结节检测与分割

Romoli, Giulia, Ruffini, Filippo, Di Feola, Francesco, Guarrasi, Valerio, Tortora, Matteo, Soda, Paolo

Abstract

Low dose Computed Tomography (CT) is the standard imaging modality for lung cancer screening. On CT scans, lung cancer typically manifests as pulmonary nodules, whose manual identification is time-consuming and prone to observer variability, image quality limitations, and suboptimal reading conditions. Moreover, subtle temporal changes are often difficult to quantify reliably in CT scans from the same patient across follow-up examinations. Deep learning-based approaches for the automatic detection and segmentation of lung nodules have demonstrated improved accuracy and efficiency. However, most existing approaches provide only coarse region-of-interest localisation, rely on prior lesion annotations or human supervision, or treat detection and segmentation as separate tasks. In this work, we present NoduleFlow, a framework for lung nodule detection, classification, and segmentation directly from series of CT volumes. Our proposal consists of four interconnected modules and operates in an automated manner at inference time, without requiring prior lesion annotations, manual initialisation, or region-of-interest pre-selection. We further investigate whether nodule localization benefits from the propagation of candidate regions across follow-up scans through inter-scan registration fields, as opposed to conventional independent per-scan analysis. The generalisability of the framework is systematically evaluated across four independent datasets. The results underscore the potential of automated pipelines for lung nodule analysis in screening and diagnostic settings.

Chinese Translation

低剂量计算机断层扫描（CT）是肺癌筛查的标准影像学方法。在CT扫描中，肺癌通常表现为肺结节，其手动识别耗时且容易受到观察者变异、图像质量限制和次优阅读条件的影响。此外，在随访检查中，同一患者的CT扫描中细微的时间变化往往难以可靠量化。基于深度学习的方法在自动检测和分割肺结节方面已显示出更高的准确性和效率。然而，大多数现有方法仅提供粗略的感兴趣区域定位，依赖于先前的病变注释或人工监督，或将检测和分割视为独立任务。在本研究中，我们提出了NoduleFlow，一个从CT体积序列中直接进行肺结节检测、分类和分割的框架。我们的提案由四个相互关联的模块组成，在推理时以自动化方式运行，无需先前的病变注释、手动初始化或感兴趣区域的预选。我们进一步探讨了结节定位是否受益于通过扫描间注册场传播候选区域，从而与传统的独立逐扫描分析形成对比。该框架的可推广性在四个独立数据集上进行了系统评估。结果强调了自动化管道在筛查和诊断环境中进行肺结节分析的潜力。

View on arXiv Download PDF AI Translation

cs.CV / 192 / 2610.16567

VLA-GAUGE: Systematically Evaluating Generalization and Robustness in Vision-Language-Action Models

VLA-GAUGE：系统评估视觉-语言-动作模型的泛化能力和鲁棒性

Xiao, Jun, Ye, Rui, Cao, Dong, Gu, Jinyu, Wang, Ying, Zhu, Zhiliang

Abstract

Vision-Language-Action models (VLAs) have achieved high success rates on familiar manipulation tasks, yet it remains unclear whether such scores reflect genuine zero-shot generalization and perturbation robustness. Existing evaluation frameworks provide limited capability-level diagnosis and combinatorial perturbation coverage, often without connecting diagnosis to targeted model refinement. We introduce VLA-GAUGE, a diagnostic framework that unifies both evaluations under a shared executable task representation. It combines (1) paired in-distribution and out-of-distribution tasks across six capability suites, (2) single-factor sensitivity analysis with combinatorial perturbation screening and evaluation, and (3) diagnosis-driven refinement through targeted data augmentation and model fine-tuning. We evaluate 12 VLAs spanning different model sizes and action-modeling paradigms under a unified protocol. We find that (1) The mean Zero-Shot Score on VLA-GAUGE is only 22.3%, with SimpleVLA-RL scoring 8.0% and ranking second to last among the evaluated models. (2) Under three representative combinatorial perturbation configurations, the mean R-Score is 9.3. (3) Diagnosis-driven fine-tuning improves $\pi_{0.5}$'s zero-shot generalization and robustness under a selected perturbation combination on held-out scenarios by 72.4 and 31.8 scoring points, respectively; the latter exceeds the 10.9-point gain from LIBERO-Plus single-factor fine-tuning despite fewer successful training trajectories. These results show that high reported success rates on familiar manipulation tasks do not guarantee strong zero-shot generalization to unseen tasks or robustness to combinatorial perturbations, while diagnostic evaluation provides actionable signals for targeted model improvement.

Chinese Translation

视觉-语言-动作模型（VLAs）在熟悉的操作任务中取得了高成功率，但尚不清楚这些分数是否反映了真正的零-shot 泛化能力和扰动鲁棒性。现有的评估框架在能力水平诊断和组合扰动覆盖方面提供的能力有限，通常没有将诊断与针对性的模型改进联系起来。我们提出了 VLA-GAUGE，这是一个诊断框架，通过共享的可执行任务表示统一了两种评估。它结合了（1）跨六个能力套件的配对内分布和外分布任务，（2）单因素敏感性分析与组合扰动筛选和评估，以及（3）通过针对性数据增强和模型微调进行的诊断驱动改进。我们在统一协议下评估了12个不同模型大小和动作建模范式的VLA。我们发现：（1）在 VLA-GAUGE 上的平均零-shot 分数仅为 22.3%，其中 SimpleVLA-RL 的得分为 8.0%，在评估的模型中排名倒数第二。（2）在三种代表性的组合扰动配置下，平均 R-Score 为 9.3。（3）诊断驱动的微调在选定的扰动组合下，分别提高了 $ ext{π}_{0.5}$ 在保留场景中的零-shot 泛化能力和鲁棒性72.4和31.8分；后者超过了 LIBERO-Plus 单因素微调带来的10.9分增益，尽管成功的训练轨迹更少。这些结果表明，在熟悉的操作任务上报告的高成功率并不保证对未见任务的强零-shot 泛化能力或对组合扰动的鲁棒性，而诊断评估提供了可操作的信号以促进针对性的模型改进。

View on arXiv Download PDF AI Translation

cs.CV / 193 / 2610.16570

Dense Endoluminal Mapping and Localisation from Monocular Bronchoscopy

单目支气管镜的密集内腔映射与定位

Esposito, Salvatore, Mattamala, Matias, Rebain, Daniel, Zhao, Liang, Dhaliwal, Kev, Khadem, Mohsen, Ramamoorthy, Subramanian

Abstract

Bronchoscopy inserts a camera through the airways to diagnose endoluminal disease. Locating the camera along the airway usually requires additional equipment, such as an electromagnetic tracker and a preoperative CT scan. In this paper, we investigate the use of the monocular camera stream to localise the bronchoscope without requiring additional sensing. We introduce a monocular SLAM system that localises the tip of a bronchoscope and reconstructs the airway surface around it from RGB video alone. A Depth Anything 3 backbone, fine-tuned on over a million photorealistic airway renderings, returns metric per-pixel depth and is the system's only source of scale. DINOv3 features routed through a custom cross-attention matcher supply sub-pixel correspondences on mucosa where classical descriptors fail. Both enter a factor graph alongside a dense constraint from projective iterative closest point (ICP) registration against the system's own running reconstruction, so surface evidence stabilises the directions the visual factors leave weakly observable. On six airways rendered from patient CT scans, the system reaches 26.1 mm mean absolute trajectory error and 77.5% mean surface F-score. Its mean surface accuracy exceeds that of five recent monocular SLAM and feed-forward baselines. MASt3R-SLAM, which fits a free scale to the ground truth, has a lower mean trajectory error (22.0 mm). ORB-SLAM3 completes only two of the six airways, whereas our system completes all six and also reconstructs two ex vivo specimens.

Chinese Translation

支气管镜通过气道插入摄像头以诊断内腔疾病。沿气道定位摄像头通常需要额外的设备，如电磁跟踪器和术前CT扫描。本文研究了如何利用单目摄像头流来定位支气管镜，而无需额外的传感器。我们引入了一种单目SLAM系统，该系统仅通过RGB视频定位支气管镜的尖端并重建其周围的气道表面。一个Depth Anything 3主干网络，在超过一百万个真实感气道渲染图上进行微调，返回每个像素的度量深度，是系统唯一的尺度来源。通过自定义交叉注意力匹配器路由的DINOv3特征在粘膜上提供亚像素对应关系，而经典描述符在此处失效。两者与来自投影迭代最近点（ICP）配准的密集约束一起进入因子图，以稳定视觉因子留下的弱可观察方向。针对六个从患者CT扫描渲染的气道，该系统达到了26.1毫米的平均绝对轨迹误差和77.5%的平均表面F-score。其平均表面精度超过了五个近期的单目SLAM和前馈基线。MASt3R-SLAM在真实值上拟合自由尺度，具有更低的平均轨迹误差（22.0毫米）。而ORB-SLAM3仅完成了六个气道中的两个，而我们的系统完成了所有六个，并重建了两个体外标本。

View on arXiv Download PDF AI Translation

cs.CV / 194 / 2610.16577

CineRecon: Benchmarking Frontier Models for Agentic Cinematic Reconstruction

CineRecon：代理电影重建前沿模型的基准测试

Wang, Jialei, Wang, Zehan, Wu, XiaoYang, Zhang, Ziang, Chen, Siyu, Huang, Jiaer, Zhang, Yujia, Pang, Tianyu, Zhao, Hengshuang

Abstract

Reconstructing reference videos as editable dynamic 3D scenes is an important path toward controllable video generation for filmmaking: these scenes provide modifiable, renderable visual proxies that enable explicit control over scene layout, subject actions, and camera behavior. To evaluate agents' ability to reconstruct editable dynamic 3D scenes from dynamic reference videos, we introduce CineRecon, a benchmark comprising 50 reference videos and 360 annotated evaluation clips. The reference videos span film, real-world footage, and games. Given a complete video, an agent uses tools to construct, animate, render, and refine its reconstruction. We evaluate this task under three representation settings: Proxy, Simplified, and Detailed. These settings increase geometric and appearance requirements while retaining the same requirements for spatial relations, camera behavior, and subject dynamics. This design tests whether agents can reproduce the events in a video when detailed appearance is unnecessary, and how performance changes as representation requirements increase. For evaluation, we construct checklists that associate observable properties and events with specific entities and times. A vision-language judge compares timestamp-aligned reference and rendered frames across 11 categories, grouped into camera, static, and dynamic consistency dimensions. We evaluate 12 models using their corresponding agent harnesses and find failures across all three dimensions. Subject motions and interactions remain difficult to reproduce even under the Proxy setting.

Chinese Translation

将参考视频重建为可编辑的动态3D场景是实现可控视频生成的重要途径，这些场景提供了可修改、可渲染的视觉代理，使得对场景布局、主体动作和相机行为的明确控制成为可能。为了评估代理从动态参考视频中重建可编辑动态3D场景的能力，我们引入了CineRecon，一个包含50个参考视频和360个标注评估片段的基准测试。参考视频涵盖了电影、现实世界的录像和游戏。给定一个完整的视频，代理使用工具构建、动画、渲染并完善其重建。我们在三种表示设置下评估此任务：代理（Proxy）、简化（Simplified）和详细（Detailed）。这些设置在保留空间关系、相机行为和主体动态相同要求的同时，增加了几何和外观的要求。该设计测试代理在不需要详细外观的情况下是否能够重现视频中的事件，以及随着表示要求的增加，性能如何变化。为了评估，我们构建了检查表，将可观察的属性和事件与特定实体和时间关联。一个视觉-语言评审者比较了时间戳对齐的参考帧和渲染帧，分为11个类别，归为相机、一致性和动态一致性维度。我们使用相应的代理工具评估了12个模型，并发现所有三个维度均存在失败。即使在代理设置下，主体的动作和交互仍然难以重现。

View on arXiv Download PDF AI Translation

cs.CV / 195 / 2610.16587

Code-Interactive Visual Reasoning for Financial Candlestick Understanding

金融蜡烛图理解的代码交互式视觉推理

Zhang, Chenxi, Pang, Youwei, Zhao, Xiaoqi, Gan, Ziliang, Zhang, Rongjunchen, Feng, Shanshan, Wang, Hao

Abstract

Financial candlestick understanding requires locating visual elements, aligning them across trading dates and panels, and combining observations through numerical and financial reasoning. Answer accuracy alone does not reveal whether predictions are supported by chart evidence. We introduce Code-Interactive Financial Candlestick Reasoning and present Fin-Thinker, a model based on Qwen3.5-9B that generates and executes code to acquire evidence and perform computations. The model uses the returned observations to guide subsequent reasoning. To train these capabilities, we construct FinKLine-32K with 32,184 replay-verified code-observation trajectories collected through teacher-guided student exploration. These include successful solutions with and without error recovery. Our three-stage training framework combines foundational visual-code fine-tuning, financial process supervision of successful reasoning and corrective continuations, and execution-feedback reinforcement learning. For evaluation, we introduce FinKLineBench with 1,200 questions across 138 financial task types to assess answer correctness, code execution, and evidence grounding separately. On FinKLineBench, Fin-Thinker improves strict accuracy by 13.1 and 33.2 percentage points over the best-performing direct-answer baseline among those evaluated and its base model with sandbox access, respectively. Training-stage ablations further show that financial process supervision and execution-feedback reinforcement learning yield successive improvements in both answer accuracy and GPT-5.6-judged grounded accuracy.

Chinese Translation

金融蜡烛图理解需要定位视觉元素，跨交易日期和面板对其进行对齐，并通过数值和金融推理结合观察结果。仅凭答案的准确性无法揭示预测是否得到了图表证据的支持。我们提出了代码交互式金融蜡烛图推理，并展示了基于 Qwen3.5-9B 的模型 Fin-Thinker，该模型生成并执行代码以获取证据并进行计算。该模型利用返回的观察结果指导后续推理。为了训练这些能力，我们构建了 FinKLine-32K，包含 32,184 条通过教师指导学生探索收集的重放验证代码-观察轨迹。这些轨迹包括成功的解决方案以及有无错误恢复的情况。我们的三阶段训练框架结合了基础视觉-代码微调、对成功推理和纠正延续的金融过程监督，以及执行反馈强化学习。为了评估，我们引入了 FinKLineBench，包含 1,200 个问题，涵盖 138 种金融任务类型，以分别评估答案的正确性、代码执行和证据基础。在 FinKLineBench 上，Fin-Thinker 在严格准确性方面比评估中表现最佳的直接答案基线提高了 13.1 和 33.2 个百分点，分别与其基础模型和沙盒访问的模型相比。训练阶段的消融实验进一步表明，金融过程监督和执行反馈强化学习在答案准确性和 GPT-5.6 评判的基础准确性方面均带来了持续的改善。

View on arXiv Download PDF AI Translation

cs.CV / 196 / 2610.16591

JtT: Rethinking Point Trajectory Diffusion for Diverse and Rigid Motion from an Image

JtT：从图像中重新思考点轨迹扩散以实现多样化和刚性运动

Boduljak, Gabrijel, Lan, Yushi, Rupprecht, Christian, Vedaldi, Andrea

Abstract

We study forecasting rigid-body motion from a single image: generating diverse and physically plausible future trajectories for arbitrary points in a scene, without building rigidity into the architecture or relying on geometric inputs such as depth. Like video generators, recent trajectory generators fail at this task, even in simple synthetic Kubric scenes, despite scaling up model size and training on real data. We show how to solve this with the right architecture and parameterization. Our model, JtT, is a simple diffusion transformer that operates directly in coordinate space and decodes arbitrary and dense query patterns, without the trajectory VAEs or fixed grids assumed in prior work. Through a systematic study of the design space of point trajectory diffusion, we find that diffusing per-frame velocities is the only parameterization that captures rigidity, that it must be paired with the right timestep schedule, and that adding full-resolution RGB at the query points resolves the boundary failures of prior methods. However, evaluating such 2D models rigorously is itself non-trivial. To our knowledge, we are the first to observe that physical plausibility and diversity do not improve together during training: larger models, as used in prior work, memorize the single ground-truth future per training image, becoming more rigid while covering fewer possible futures. Yet prior work has paid little attention to this, so we show how model rankings depend on model size, metric, and stopping time, and adopt a stricter, checkpoint-averaged evaluation protocol. Under this protocol, JtT improves over prior methods on both distributional and rigidity metrics and approaches ground-truth rigidity, with only about 50M parameters.

Chinese Translation

我们研究如何从单幅图像预测刚体运动：生成场景中任意点的多样化和物理上合理的未来轨迹，而无需在架构中构建刚性或依赖深度等几何输入。与视频生成器类似，最近的轨迹生成器在这一任务上表现不佳，即使在简单的合成Kubric场景中，尽管模型规模增大并在真实数据上训练。我们展示了如何通过正确的架构和参数化来解决这个问题。我们的模型JtT是一个简单的扩散变换器，直接在坐标空间中操作，并解码任意且密集的查询模式，而不依赖于先前工作中假设的轨迹变分自编码器（VAEs）或固定网格。通过对点轨迹扩散设计空间的系统研究，我们发现每帧速度的扩散是捕捉刚性的唯一参数化，它必须与正确的时间步调度配对，并且在查询点添加全分辨率RGB可以解决先前方法的边界失效。然而，严格评估这样的二维模型本身并非易事。据我们所知，我们是首个观察到物理合理性和多样性在训练过程中并不同时提升的研究者：较大的模型，如先前工作中使用的，记忆每个训练图像的单一真实未来，变得更加刚性，同时覆盖的可能未来减少。然而，先前的工作对此关注甚少，因此我们展示了模型排名如何依赖于模型大小、度量和停止时间，并采用了更严格的检查点平均评估协议。在该协议下，JtT在分布和刚性度量上均优于先前的方法，并接近真实刚性，仅需约5000万个参数。

View on arXiv Download PDF AI Translation

cs.CV / 197 / 2610.16598

Detecting Chain-of-Thought Hijacking in Backdoored Multimodal Large Reasoning Models

检测后门多模态大推理模型中的思维链劫持

Zhang, Zhifang, Wei, Qi, Chen, Weitong, Maennel, Olaf, Zhang, Wei Emma, ren, zuomin, Feng, Lei, Xu, Miao

Abstract

Chain-of-thought (CoT) reasoning enables multimodal large reasoning models (MLRMs) to solve complex visual problems, but also exposes them to CoT hijacking: an adversary implants a backdoor through poisoned fine-tuning so that a visual trigger induces a hijacking turn, redirecting an otherwise faithful reasoning chain to an attacker-chosen answer. However, since the hijack acts within the reasoning chain, existing multimodal backdoor defenses designed for MLLMs that answer directly struggle to detect it. Moreover, inspecting the chain itself does not reliably expose the hijack, since the hijacking turn can resemble honest self-correction in form. In this paper, we find that honest reasoning tends to revise its answer at high-entropy "aha moments", whereas a hijack can overturn a confident answer through a near-deterministic turn. To quantify this contrast, we formalize this phenomenon as belief inertia, the cumulative next-token entropy of a reasoning span minus the answer-belief revision it induces, with markedly lower values for hijacked spans than for honest reasoning spans. Since hijacking tokens are also image-dependent, we further improve detection by restricting the measure to image-dependent spans, yielding visual belief inertia (VBI). Building on this, we propose InertiaGuard, a training-free, test-time detector that flags a trace when its VBI falls below a threshold calibrated on a small set of clean traces. Across various attacks and benchmarks, InertiaGuard achieves higher detection performance than existing baselines, while generalizing to different model architectures and trigger types.

Chinese Translation

思维链（Chain-of-thought, CoT）推理使多模态大推理模型（Multimodal Large Reasoning Models, MLRMs）能够解决复杂的视觉问题，但也使其暴露于思维链劫持之中：对手通过有毒微调植入后门，使得视觉触发器引发劫持转折，将本应忠实的推理链重定向到攻击者选择的答案。然而，由于劫持行为发生在推理链内部，现有针对直接回答的多模态后门防御措施难以检测到这一点。此外，检查推理链本身并不能可靠地揭露劫持，因为劫持转折在形式上可能与诚实的自我修正相似。在本文中，我们发现诚实推理倾向于在高熵的“恍然大悟时刻”修正其答案，而劫持则可以通过近乎确定性的转折推翻一个自信的答案。为了量化这种对比，我们将这一现象形式化为信念惯性（belief inertia），即推理跨度的累积下一个标记熵减去其引发的答案信念修正，对于被劫持的跨度，其值显著低于诚实推理的跨度。由于劫持标记也依赖于图像，我们进一步通过将度量限制在图像依赖的跨度上来改善检测，得出视觉信念惯性（Visual Belief Inertia, VBI）。基于此，我们提出了InertiaGuard，一种无训练、测试时的检测器，当其VBI低于在一小组干净轨迹上校准的阈值时标记轨迹。在各种攻击和基准测试中，InertiaGuard的检测性能优于现有基线，同时能够推广到不同的模型架构和触发器类型。

View on arXiv Download PDF AI Translation

cs.CV / 198 / 2610.16622

Answer It or Not? Towards Prompt-Adapted Adversarial Images on Vision Language Models

回答还是不回答？面向视觉语言模型的提示适应性对抗图像

Bai, Shunyuan, Ren, Yuchen, Yang, Le, Lin, Chenhao, Shen, Chao, Zhao, Zhengyu

Abstract

With the advancement in language modeling, research on adversarial images has recently shifted from image classifiers to vision-language models (VLMs). Despite the new, text-prompt-driven nature of VLMs, we find that existing VLM adversaries have largely overlooked the dynamics of text prompts. In a typical use case, adversarial images can protect visual privacy by rejecting VLMs from answering questions about the image. However, existing protections are prompt-agnostic, i.e., indiscriminately rejecting any question, including those for potential legitimate use of the image. To address this limitation, we propose Prompt-Adapted (ProAda) adversarial images, which are optimized to balance the rejection of harmful prompts and the correct response of legitimate prompts. In particular, to achieve a more accurate adversarial direction, the text prompt is also iteratively updated with a simple classifier judging whether the harmfulness of the prompt remains. Extensive experiments on visual-privacy-related datasets validate that ProAda can reject over 90% of harmful prompts while allowing over 90% of legitimate ones, demonstrating superior overall performance compared to existing prompt-agnostic VLM adversaries. In addition, our ProAda achieves competitive overall cross-model transferability and improved robustness against countermeasures.

Chinese Translation

随着语言模型的进步，对抗图像的研究最近已从图像分类器转向视觉语言模型（VLMs）。尽管VLMs具有新的文本提示驱动特性，但我们发现现有的VLM对抗者在很大程度上忽视了文本提示的动态性。在典型的使用案例中，对抗图像可以通过拒绝VLM回答有关图像的问题来保护视觉隐私。然而，现有的保护措施是与提示无关的，即无差别地拒绝任何问题，包括那些可能合法使用图像的问题。为了解决这一局限性，我们提出了提示适应性（Prompt-Adapted，ProAda）对抗图像，这些图像经过优化，旨在平衡拒绝有害提示和正确响应合法提示之间的关系。特别是，为了实现更准确的对抗方向，文本提示还通过一个简单的分类器进行迭代更新，以判断提示的有害性是否仍然存在。在与视觉隐私相关的数据集上进行的大量实验验证了ProAda能够拒绝超过90%的有害提示，同时允许超过90%的合法提示，显示出相较于现有的与提示无关的VLM对抗者的整体优越性能。此外，我们的ProAda在跨模型迁移性方面表现出竞争力，并在抵御对策方面提高了鲁棒性。

View on arXiv Download PDF AI Translation

cs.CV / 199 / 2610.16624

Adaptive Interaction Network for Human Motion Prediction During Human-Robot Collaboration

人机协作中的人类运动预测自适应交互网络

Liu, Mengyuan, Lin, Yangting, Cui, Qiongjie

Abstract

Human motion prediction during human-robot collaboration is critical for achieving safe and efficient interactions in shared environments. Unlike previous studies that have primarily focused on humans and objects, we focus on the robot-aware human motion prediction task, which explicitly models the influence of robots on human motion. This task presents new challenges, such as capturing human-robot heterogeneity and modeling complex spatial interactions. To address these issues, we develop an adaptive interaction network (AINet) model that consists of two branches: a primary branch that predicts future human motion and an auxiliary branch that estimates robot trajectories. The two branches are jointly optimized and coupled via a Local-Global Spatial Interaction (LGSI) Module, which effectively captures fine-grained and global contextual dependencies between human and robot motion sequences. Then, we introduce an Adaptive Weighted Aggregation (AWA) Module to dynamically fuse motion features using context-dependent weights, thereby increasing adaptability across diverse scenarios. Furthermore, we adopt a coarse-to-fine prediction strategy, in which a coarse pseudo-trajectory is first predicted, followed by the refinement of detailed human poses based on that trajectory. Extensive experiments on three challenging datasets reveal that our proposed method achieves superior performance, validating its effectiveness in human-robot collaboration scenarios. Our code is available at: https://github.com/LyTingHub/AINet.

Chinese Translation

在人机协作中，人类运动预测对于实现安全高效的共享环境交互至关重要。与以往主要关注人类与物体的研究不同，我们关注机器人感知的人类运动预测任务，该任务明确建模了机器人对人类运动的影响。该任务带来了新的挑战，如捕捉人机异质性和建模复杂的空间交互。为了解决这些问题，我们开发了一种自适应交互网络（Adaptive Interaction Network，AINet）模型，该模型由两个分支组成：一个主要分支用于预测未来的人类运动，另一个辅助分支用于估计机器人轨迹。这两个分支通过局部-全局空间交互（Local-Global Spatial Interaction，LGSI）模块进行联合优化和耦合，有效捕捉人类与机器人运动序列之间的细粒度和全局上下文依赖关系。然后，我们引入了一种自适应加权聚合（Adaptive Weighted Aggregation，AWA）模块，利用上下文相关的权重动态融合运动特征，从而提高在不同场景中的适应性。此外，我们采用了一种粗到细的预测策略，首先预测粗略的伪轨迹，然后基于该轨迹细化详细的人类姿态。在三个具有挑战性的数据集上的大量实验表明，我们提出的方法实现了优越的性能，验证了其在人机协作场景中的有效性。我们的代码可在以下地址获取：https://github.com/LyTingHub/AINet。

View on arXiv Download PDF AI Translation

cs.CV / 200 / 2610.16632

REEL: Learning to Retrieve, Evaluate, and Evolve Visual Memory for Any Long-Horizon Multi-Shot Video Generation

REEL：学习检索、评估和演化视觉记忆以生成任意长时间跨度的多镜头视频

Xu, Hang, Zhang, Guohui, Ma, Xiaoxiao, Yu, Hu, Huang, Jie, Xiao, Jiayin, Song, Lin, Huang, Haoyang, Duan, Nan, Zhao, Feng

Abstract

Memory-based long-video generation aims to combine user control with long-range visual consistency. We organize this process into three functional components: a planner, an adapter, and a generator. Representative multi-shot systems couple memory to generator-specific interfaces or use predefined agent workflows, rather than independently training a memory policy for cross-shot objectives. We introduce REEL, a general-purpose visual-memory planner. Through a unified tool interface, REEL learns to retrieve references, create missing visual evidence, and update persistent memory. Backend-specific adapters translate its evidence plans into supported conditioning inputs, allowing the same policy to guide different black-box video generators. To train REEL, we construct two high-quality datasets, REEL-SFT-4k and REEL-RL-1k, covering diverse interaction patterns and prompt types. Training proceeds in three stages: supervised fine-tuning, single-shot Group Relative Policy Optimization (GRPO), and multi-shot GRPO. Multi-shot GRPO carries policy-generated memory across consecutive shots, so later outcomes guide earlier reference selection and memory updates. REEL-Bench tests identity preservation and intended state changes under text-only requests and timed reference-image uploads. On JoyAI-Echo~1.5 and MiniMax-H3, REEL improves cross-shot consistency on ST-Bench by 15.1\% and 20.7\%, and Core-100 on our proposed REEL-Bench by 2.41 and 8.32 points, respectively, over the no-memory baseline.

Chinese Translation

基于记忆的长视频生成旨在将用户控制与长距离视觉一致性相结合。我们将这一过程组织为三个功能组件：规划器、适配器和生成器。代表性的多镜头系统将记忆与特定于生成器的接口相结合，或使用预定义的代理工作流程，而不是为跨镜头目标独立训练记忆策略。我们提出了REEL，一个通用的视觉记忆规划器。通过统一的工具接口，REEL学习检索参考、创建缺失的视觉证据并更新持久记忆。后端特定的适配器将其证据计划转换为支持的条件输入，使得同一策略能够指导不同的黑箱视频生成器。为了训练REEL，我们构建了两个高质量的数据集，REEL-SFT-4k和REEL-RL-1k，涵盖了多样的交互模式和提示类型。训练分为三个阶段：监督微调、单镜头组相对策略优化（Group Relative Policy Optimization, GRPO）和多镜头GRPO。多镜头GRPO在连续镜头之间传递策略生成的记忆，因此后续结果指导早期参考选择和记忆更新。REEL-Bench测试在仅文本请求和定时参考图像上传下的身份保持和预期状态变化。在JoyAI-Echo~1.5和MiniMax-H3上，REEL在ST-Bench上提高了跨镜头一致性，分别为15.1 ext{%}和20.7 ext{%}，在我们提出的REEL-Bench上，Core-100提高了2.41和8.32点，相较于无记忆基线。

View on arXiv Download PDF AI Translation

cs.CV / 201 / 2610.16666

Revisiting Natural Language Prompts from Large Language Models for Open-Vocabulary Video Recognition

重新审视大型语言模型的自然语言提示在开放词汇视频识别中的应用

He, Haichen, Liu, Weibin, Xing, Weiwei

Abstract

Vision-language models (VLMs) like CLIP have shown great success in image tasks, however, applying CLIP to open-vocabulary video recognition poses challenges. Due to the limited text information, directly fine-tuning CLIP on video recognition datasets may lead to overfitting and hinder its generalizability. A prevalent method is to enrich text information by generating prompts using Large Language Models (LLMs). However, limited efforts have been made to explore how to generate higher-quality prompts that can contribute to classify unseen categories, and how to further enhance vision-text alignment based on these prompts. In this paper, we propose P2CLIP (Prompt-guided CLIP), which aims to bridge this gap by exploring the impact of prompts across three aspects: category, video, and frame. At the category level, we argue that LLM-generated prompts under simple instructions may suffer from semantic overgeneralization, which can blur the category boundaries and hinder the performance of the model. To address this issue, we design Generic-Discriminative Prompt using LLMs to obtain discriminative category embeddings with rich semantic information. At the video and frame levels, we first use VLM and LLM to generate frame and video descriptions for enhancing vision-text alignment. Then we propose a Knowledge-Guided Temporal Aggregation mechanism, which leverages the expert knowledge derived from LLMs to guide the temporal modeling of videos and encourage a better construction of the joint visual-textual feature space. Extensive experiments across various benchmarks, including zero-shot and few-shot settings, demonstrate the effectiveness of our method.

Chinese Translation

视觉语言模型（VLMs）如CLIP在图像任务中取得了巨大成功，然而，将CLIP应用于开放词汇视频识别面临挑战。由于文本信息有限，直接在视频识别数据集上微调CLIP可能导致过拟合，并阻碍其泛化能力。一种普遍的方法是通过使用大型语言模型（LLMs）生成提示来丰富文本信息。然而，关于如何生成更高质量的提示以帮助分类未见类别，以及如何基于这些提示进一步增强视觉-文本对齐的研究仍然有限。在本文中，我们提出了P2CLIP（提示引导的CLIP），旨在通过从类别、视频和帧三个方面探讨提示的影响来填补这一空白。在类别层面，我们认为在简单指令下生成的LLM提示可能会遭遇语义过度泛化，这可能模糊类别边界并影响模型性能。为了解决这一问题，我们设计了使用LLMs的通用区分提示，以获得具有丰富语义信息的区分类别嵌入。在视频和帧层面，我们首先使用VLM和LLM生成帧和视频描述，以增强视觉-文本对齐。然后，我们提出了一种知识引导的时间聚合机制，该机制利用从LLMs获得的专家知识来指导视频的时间建模，并促进联合视觉-文本特征空间的更好构建。在包括零样本和少样本设置的各种基准测试中，广泛的实验验证了我们方法的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 202 / 2610.16679

EvoSpatial: Experience-Driven Skill and Tool Evolution for Spatial Reasoning Agents

EvoSpatial：基于经验的空间推理智能体技能与工具演化

Yu, Bin, Lian, Shijie, Shen, Zhaolong, Lin, Xiaopeng, Zhang, Yao, Liu, Haishan, Li, Haotian, Wei, Yuliang, Hu, Kai, Du, Yichao, Zhang, Zhirui, Wang, Bailing, Chen, Kai

Abstract

Tool-augmented vision-language agents compose perception and measurement to reason about spatial relationships, but effective execution also requires tracking geometric assumptions and checking intermediate evidence. Preserving this experience across tasks requires retaining procedural knowledge together with the computations it governs. We introduce EvoSpatial, an experience-driven framework that evolves a library of linked skills and tools. Our key insight is that successful and failed spatial trajectories reveal both how to compute and when to trust the result. EvoSpatial retrieves skills, reuses tools, and synthesizes reusable computational functions (generated helpers) during reasoning. After answer submission, supervised outcome feedback guides skill refinement and tool consolidation for subsequent tasks. Across 20 benchmarks, EvoSpatial attains the highest average score among the compared spatial agents, exceeding SpatialClaw by 4.0 percentage points. Further experiments show cross-benchmark transfer and improved reasoning through skill-guided distillation without external skills at evaluation. These results support accumulating spatial experience for both inference-time reuse and trajectory-based training.

Chinese Translation

工具增强的视觉-语言智能体通过感知和测量来推理空间关系，但有效的执行还需要跟踪几何假设并检查中间证据。在任务之间保留这种经验需要保留程序知识以及它所支配的计算。我们提出了EvoSpatial，一个基于经验的框架，用于演化一个关联技能和工具的库。我们的关键见解是，成功和失败的空间轨迹揭示了如何计算以及何时信任结果。EvoSpatial在推理过程中检索技能、重用工具，并合成可重用的计算函数（生成的助手）。在提交答案后，监督的结果反馈指导技能的精炼和工具的整合，以便于后续任务。在20个基准测试中，EvoSpatial在比较的空间智能体中获得了最高的平均分，比SpatialClaw高出4.0个百分点。进一步的实验表明，通过技能引导的蒸馏在评估时无需外部技能也能实现跨基准迁移和改进推理。这些结果支持在推理时重用和基于轨迹的训练中积累空间经验。

View on arXiv Download PDF AI Translation

cs.CV / 203 / 2610.16692

3D-VITAL: Visibility-aware Identity Training from Artificial Liftings of 3D Reconstruction model for Cross-View Object Re-Identification

3D-VITAL：基于可见性意识的身份训练，通过3D重建模型的人工提升实现跨视角物体重识别

Grolleau, William, Sabourin, Astrid, Lapouge, Guillaume, Achard, Catherine

Abstract

We address the Cross-View Object Re-Identification (ReID) task, aiming to match identities across drastically different viewpoints. We focus on Single-View Aerial-Ground ReID (SV AG-ReID), where models train on one viewpoint and must generalize to another at inference. To bridge this viewpoint gap, related work employs generative models to synthesize novel views from source images as additional training data. We leverage recent advances in large-scale single-view 3D mesh reconstruction to lift source images into textured meshes and to render them from new viewpoints. However, this pipeline produces unreliable data in two distinct ways. First, the reconstruction may fail and produce geometrically incorrect meshes, resulting in severe corruption in rendered novel views. Second, even when the geometry is correct, the rendered novel views contain hallucinated textures in regions not visible in the source image. The reconstructed mesh carries a per-pixel partition between regions visible from the source view and regions that the reconstruction model had to hallucinate, available as a deterministic byproduct of rendering. We introduce this per-pixel visibility map and exploit it in a two-stage design. Cross-View Consistency Filtering (CVCF) discards geometrically incorrect meshes by ensuring multi-view consistency between source and novel views over visible patches. Visibility-Aware Attention Supervision (VAAS) leverages the visibility map to guide the model's attention toward reliable regions of the rendered view during training. Combined, these two components enable our method, 3D-VITAL, to better leverage the synthetic signal from single-view 3D mesh reconstruction. 3D-VITAL sets a new state-of-the-art on three SV AG-ReID benchmarks. Code and data will be released at https://github.com/TurtleSmoke/3D-VITAL.

Chinese Translation

我们针对跨视角物体重识别（ReID）任务，旨在匹配在截然不同视角下的身份。我们专注于单视角空中-地面重识别（SV AG-ReID），该模型在一个视角上进行训练，并必须在推理时泛化到另一个视角。为了解决这一视角差距，相关工作采用生成模型从源图像合成新视角作为额外的训练数据。我们利用近期在大规模单视角3D网格重建方面的进展，将源图像提升为纹理网格，并从新视角进行渲染。然而，这一流程以两种不同方式产生不可靠的数据。首先，重建可能失败并生成几何上不正确的网格，导致渲染的新视角严重失真。其次，即使几何形状正确，渲染的新视角也会在源图像不可见的区域中包含虚构的纹理。重建的网格携带一个逐像素的可见性分区，区分源视角可见区域与重建模型必须虚构的区域，这一信息作为渲染的确定性副产品可用。我们引入这一逐像素可见性图，并在两阶段设计中加以利用。跨视角一致性过滤（CVCF）通过确保源视角与新视角在可见区域之间的多视角一致性，丢弃几何上不正确的网格。可见性意识注意力监督（VAAS）利用可见性图在训练过程中引导模型的注意力集中在渲染视角的可靠区域。结合这两个组件，我们的方法3D-VITAL能够更好地利用单视角3D网格重建的合成信号。3D-VITAL在三个SV AG-ReID基准上设定了新的最先进水平。代码和数据将发布在 https://github.com/TurtleSmoke/3D-VITAL。

View on arXiv Download PDF AI Translation

cs.CV / 204 / 2610.16734

PathPilot: An Adaptive Question-Conditioned Multi-Magnification Navigation Agent for Whole-Slide Image Analysis

PathPilot：一种适应性问题条件下的多倍放大导航代理，用于全切片图像分析

Xu, Zhengyang, Li, Han, Liu, Jingsong, Ura, Ayako, Zu, Shihui, Zhou, Shaohua Kevin, Navab, Nassir, Schüffler, Peter

Abstract

The diagnostic capability of WSI agents relies on effective slide navigation to identify diagnostically relevant regions and efficiently integrate evidence across the slide. While in clinical practice, pathologists dynamically examine slides across multiple scales, integrating global and cellular evidence in a sequential manner, existing AI navigation approaches mostly operate at a single fixed magnification or rely on predefined magnification traversal. This mismatch prevents existing methods from modeling cross-magnification interactions and adaptive magnification selection inherent to real diagnostic workflows. We propose PathPilot, a clinically consistent multi-magnification WSI navigation agent that explicitly models cross-magnification localization and evidence-adaptive exploration. Specifically, we introduce a CrossMagLoc that learns magnification-specific diagnostic localization from pathologist viewing behavior and integrates information from adjacent magnifications to identify the diagnostically most relevant region within the current ROI, thereby providing a diagnostically informed starting point for subsequent question-conditioned exploration. We further introduce an EviRegionSearch that starts question-conditioned exploration from this region, progressively accumulating question-relevant evidence and adaptively determining the next navigation action according to the remaining evidence requirement. Their iterative interaction enables PathPilot to sequentially explore across ROIs and magnifications until sufficient evidence is acquired. Extensive QA evaluations across five cohorts demonstrate consistent superiority over competing SOTA methods on inter/external cohorts, respectively, validating question-conditioned evidence acquisition across regions and magnifications that mirrors the sequential, knowledge-driven navigation of pathologists.

Chinese Translation

全切片图像（WSI）代理的诊断能力依赖于有效的切片导航，以识别与诊断相关的区域并高效整合切片中的证据。在临床实践中，病理学家动态地在多个尺度上检查切片，以顺序方式整合全球和细胞证据，而现有的人工智能导航方法大多在单一固定倍放大下操作或依赖于预定义的倍放大遍历。这种不匹配阻碍了现有方法对跨倍放大交互和适应性倍放大选择的建模，这些都是实际诊断工作流程中固有的。我们提出了PathPilot，这是一种临床一致的多倍放大WSI导航代理，明确建模跨倍放大定位和证据适应性探索。具体而言，我们引入了CrossMagLoc，它从病理学家的观察行为中学习特定倍放大的诊断定位，并整合来自相邻倍放大的信息，以识别当前感兴趣区域（ROI）中与诊断最相关的区域，从而为后续的问题条件探索提供一个以诊断为基础的起始点。我们进一步引入了EviRegionSearch，它从该区域开始问题条件探索，逐步积累与问题相关的证据，并根据剩余证据需求自适应地确定下一个导航动作。它们的迭代交互使PathPilot能够在ROI和倍放大之间顺序探索，直到获得足够的证据。对五个队列的广泛QA评估表明，在内部和外部队列上均优于竞争的最新技术（SOTA）方法，验证了跨区域和倍放大的问题条件证据获取，反映了病理学家的顺序、知识驱动的导航。

View on arXiv Download PDF AI Translation

cs.CV / 205 / 2610.16735

VAttnPO: Visual Attention-Guided Policy Optimization

VAttnPO：视觉注意力引导的策略优化

Zheng, Lihao, Shao, Zhenwei, Yu, Zhou, Yang, Yan, Sun, Haoyi, Wang, Qian, Mao, Ning, Zhou, Pan

Abstract

Multimodal reinforcement learning with verifiable rewards commonly assigns one sequence advantage to an entire response. This outcome-level signal distinguishes overall response quality but leaves changes in visual-evidence use across generated tokens implicit. We introduce VAttnPO (Visual Attention-Guided Policy Optimization), which uses answer-conditioned visual attention to refine token-level credit while preserving the verifier-derived trajectory-mean advantage. A frozen old policy constructs a visual reference by teacher forcing the known correct answer and produces a visual distribution for each rollout token. VAttnPO measures their agreement in visual-attention space using Jensen--Shannon divergence, rather than reading guidance from teacher output-token distributions or introducing an attention-matching objective. Trajectory centering, relative-position weighting, and credit balancing turn this agreement into a zero-mean, position-aware correction to the GRPO sequence advantage. Token intervention, visual-index permutation, and trajectory analysis provide diagnostic evidence linking this signal to model behavior, visual correspondence, and generation position. On Qwen3-VL-8B-Instruct, VAttnPO raises six-task image AVE from 70.97 with GRPO to 73.50 and scores higher on every reported structured, video, and MME metric. Gains over GRPO also hold on Qwen2.5-VL-7B-Instruct and InternVL3-8B-Instruct.

Chinese Translation

具有可验证奖励的多模态强化学习通常将一个序列的优势分配给整个响应。这种结果级信号区分了整体响应质量，但在生成的标记中对视觉证据使用的变化却是隐含的。我们提出了VAttnPO（视觉注意力引导的策略优化），该方法利用答案条件的视觉注意力来细化标记级别的信用，同时保留由验证器派生的轨迹均值优势。一个冻结的旧策略通过教师强制已知的正确答案构建视觉参考，并为每个回放标记生成视觉分布。VAttnPO使用Jensen-Shannon散度来测量它们在视觉注意力空间中的一致性，而不是从教师输出标记分布中读取指导或引入注意力匹配目标。轨迹中心化、相对位置加权和信用平衡将这种一致性转化为对GRPO序列优势的零均值、位置感知的修正。标记干预、视觉索引置换和轨迹分析提供了诊断证据，将该信号与模型行为、视觉对应性和生成位置联系起来。在Qwen3-VL-8B-Instruct上，VAttnPO将六任务图像的AVE从GRPO的70.97提高到73.50，并在每个报告的结构化、视频和MME指标上得分更高。相较于GRPO的提升在Qwen2.5-VL-7B-Instruct和InternVL3-8B-Instruct上也得以保持。

View on arXiv Download PDF AI Translation

cs.CV / 206 / 2610.16744

FoveaTree: Information-Density Trees for Query-Guided Visual Token Pruning in Video Large Language Models

FoveaTree：用于视频大语言模型中查询引导的视觉标记修剪的信息密度树

Meng, Zihan, Chen, Weidong, Hong, Dexiang, Liu, Xinyan, Mao, Zhendong

Abstract

Video large language models (Video LLMs) encode each frame into hundreds of visual tokens, making video inference expensive. Existing methods reduce this cost through similarity-based merging or attention-based selection but face two challenges: Hierarchy-blind Pruning, which overlooks regional compressibility and can discard critical evidence, and Attention Drift, where positional effects and attention sinks shift text-to-vision attention away from task relevance. We propose FoveaTree, a two-stage, training-free visual-token pruning framework. First, Merge-Gain Density Estimation measures the feature distortion caused by regional merging and constructs a quadtree-derived density map to guide token dropping, merging, and retention. Second, Attention Tree Pruning uses the surviving units and inherited tree topology to refine text-to-vision attention in intermediate LLM layers, preserving relevant regional evidence while removing low-relevance content. We evaluate FoveaTree on InternVL3.5-8B and Qwen3-VL-8B across five video question-answering benchmarks and one captioning benchmark at 50%, 30%, and 10% visual-token retention. FoveaTree consistently achieves overall state-of-the-art performance across the evaluated settings, offering a strong accuracy-efficiency trade-off.

Chinese Translation

视频大语言模型（Video LLMs）将每帧编码为数百个视觉标记，使得视频推理成本高昂。现有方法通过基于相似性的合并或基于注意力的选择来降低这一成本，但面临两个挑战：层次盲修剪（Hierarchy-blind Pruning），忽视了区域的可压缩性，可能会丢弃关键证据；以及注意力漂移（Attention Drift），其中位置效应和注意力汇聚使得文本到视觉的注意力偏离任务相关性。我们提出了FoveaTree，一个两阶段、无训练的视觉标记修剪框架。首先，合并增益密度估计（Merge-Gain Density Estimation）测量区域合并所造成的特征失真，并构建一个基于四叉树的密度图来指导标记的丢弃、合并和保留。其次，注意力树修剪（Attention Tree Pruning）利用存活的单元和继承的树形拓扑来精炼中间LLM层中的文本到视觉的注意力，保留相关区域证据，同时去除低相关内容。我们在InternVL3.5-8B和Qwen3-VL-8B上评估FoveaTree，在五个视频问答基准和一个字幕基准上进行50%、30%和10%的视觉标记保留。FoveaTree在评估的设置中始终实现了整体的最先进性能，提供了良好的准确性与效率的权衡。

View on arXiv Download PDF AI Translation

cs.CV / 207 / 2610.16749

The Midas Touch for Metric Depth: Interpretable Metricization for Generalizable Depth Perception

度量深度的米达斯之触：可解释的度量化以实现可泛化的深度感知

Zhang, Mengtan, Ma, Yu, Jiang, Shiqi, Feng, Yi, Ye, Wei, Chen, Xieyuanli, Wang, Hanli, Chen, Qijun, Fan, Rui

Abstract

Recent foundation models have drastically improved the generalization of monocular relative depth estimation, yet recovering accurate metric depth from their estimates for depth perception tasks remains challenging due to scale ambiguity and local metric-scale inconsistency. Existing approaches recover metric depth from sparse observations using either explicit global affine transformation algorithms or implicit data-driven depth completion networks. The former cannot accommodate spatially varying relative-to-metric relationships, while the latter are sensitive to the distribution of sparse observations and may not fully exploit generalizable relative depth priors. To simultaneously address these issues, this study extends our previous work MTD and proposes MTDv2, a hybrid metric depth perception framework that combines explicit depth metricization with data-driven metric depth refinement. The explicit depth metricization process performs segment-wise affine transformations to recover spatially varying metric scales while preserving the geometric structure of relative depth, followed by pixel-wise depth refinement that propagates metric observations along paths with minimal accumulated geometric variation. The data-driven depth refinement process subsequently extends the explicit single-path depth propagation to a learnable multi-path formulation while preserving its interpretability. Extensive experiments conducted on 16 public datasets demonstrate the robust and generalizable depth scale recovery capability of MTDv2 across diverse sources of sparse observations and different depth estimation paradigms, as well as its broad applicability to downstream 3D perception tasks. Code will be publicly available at https://mias.group/MTD upon publication.

Chinese Translation

近期的基础模型显著提高了单目相对深度估计的泛化能力，但由于尺度模糊和局部度量尺度不一致，从这些估计中恢复准确的度量深度以用于深度感知任务仍然具有挑战性。现有方法通过显式的全局仿射变换算法或隐式的数据驱动深度补全网络，从稀疏观测中恢复度量深度。前者无法适应空间变化的相对与度量关系，而后者对稀疏观测的分布敏感，可能无法充分利用可泛化的相对深度先验。为同时解决这些问题，本研究扩展了我们之前的工作 MTD，提出了 MTDv2，一个混合度量深度感知框架，结合了显式的深度度量化和数据驱动的度量深度精细化。显式的深度度量化过程执行分段仿射变换，以恢复空间变化的度量尺度，同时保持相对深度的几何结构，随后进行像素级的深度精细化，通过沿着最小累积几何变化的路径传播度量观测。数据驱动的深度精细化过程随后将显式的单路径深度传播扩展为可学习的多路径形式，同时保持其可解释性。在16个公共数据集上进行的广泛实验表明，MTDv2在多种稀疏观测来源和不同深度估计范式下，具有强大的可泛化深度尺度恢复能力，以及其在下游3D感知任务中的广泛适用性。代码将在发表后公开，网址为 https://mias.group/MTD。

View on arXiv Download PDF AI Translation

cs.CV / 208 / 2610.16759

Image Tokenizers Should Be Trained Between Images

图像标记器应在图像之间进行训练

Beißwenger, Ole, Dihlmann, Jan-Niklas, Lensch, Hendrik

Abstract

Latent diffusion models generate images as codes of an image tokenizer, whose decoder turns them into the pixel image. Recent tokenizers improve how well these codes reconstruct an image and how easily a generator learns them. Yet a generator does not reproduce the codes of training images, and when a generated code lands outside what the decoder expects, the decoded image contains imperfections. Current methods make the decoder more robust to generated codes by adding noise during training. However, generated codes lie between the training codes and resemble their mixture more than one of them plus noise. To improve generated images, decoders should therefore be trained on codes in the space between training samples. We propose InterpTok, which fine-tunes MAETok to decode mixtures of image codes into natural images. Since no target image exists between two training samples, a discriminator judges whether each decoded mixture looks as natural as its two sources. A mixing ratio loss requires the decoded mixture to keep the mixing ratio of the two codes. Together, they prevent collapse to a trivial solution and yield a sharp and natural composition of both images. InterpTok keeps MAETok's reconstruction quality and improves interpolation quality more than threefold, the best of all tokenizers in a recent benchmark. Beyond interpolation, InterpTok improves the generation FID of a matched diffusion transformer from 11.61 to 8.48 without guidance, and even reaches 8.11 at an eighth of the sampling steps, where the generator on MAETok's codes worsens. The generator trained on our tokenizer also reaches the final result of the one trained on MAETok at half the training steps. Our findings show that the space between image codes can and should be explicitly supervised rather than ignored.

Chinese Translation

潜在扩散模型将图像生成作为图像标记器的代码，其解码器将这些代码转化为像素图像。最近的标记器改善了这些代码重构图像的能力以及生成器学习它们的难易程度。然而，生成器并未重现训练图像的代码，当生成的代码落在解码器的预期之外时，解码后的图像会出现缺陷。目前的方法通过在训练过程中添加噪声来增强解码器对生成代码的鲁棒性。然而，生成的代码位于训练代码之间，更像是它们的混合，而不是其中一个加上噪声。为了改善生成的图像，解码器应在训练样本之间的空间中对代码进行训练。我们提出了InterpTok，它微调MAETok以将图像代码的混合解码为自然图像。由于在两个训练样本之间不存在目标图像，判别器判断每个解码混合是否看起来与其两个来源一样自然。混合比例损失要求解码混合保持两个代码的混合比例。两者结合防止了退化到平凡解，并产生了两个图像的清晰自然组合。InterpTok保持了MAETok的重构质量，且在插值质量上提高了三倍以上，是最近基准测试中所有标记器中表现最好的。除了插值，InterpTok还在没有指导的情况下将匹配扩散变换器的生成FID从11.61降低到8.48，并在八分之一的采样步骤中甚至达到了8.11，而在MAETok的代码上生成器的表现却变差。基于我们的标记器训练的生成器在训练步骤减半的情况下也达到了与基于MAETok训练的生成器相同的最终结果。我们的研究结果表明，图像代码之间的空间可以并且应该被明确监督，而不是被忽视。

View on arXiv Download PDF AI Translation

cs.CV / 209 / 2610.16762

Towards Complementary Keypoint Detection via Mixture of Detectors

通过混合检测器实现互补关键点检测

Lu, Xiaoyong, Xiao, Guobao, Liang, Dong, Du, Songlin

Abstract

Keypoint detection plays a central role in geometric vision pipelines. Existing detectors exhibit complementary strengths across different scene structures and keypoint budgets, and reliance on a single detector often results in suboptimal keypoint distributions. Motivated by this observation, we propose \textbf{Mixture of Detectors (MoD)} for dense matching pipelines, a unified framework for complementary keypoint detection through the adaptive composition of heterogeneous detectors. MoD first distills heterogeneous detectors into a unified model, allowing diverse detection maps to be predicted efficiently within a single forward pass. Detector selection is subsequently formulated as a patch-wise, budget-conditioned decision process, in which a lightweight router assigns the most suitable detector to each local region. To avoid heuristic supervision, the router is trained with task-driven rewards derived from downstream matching quality, jointly capturing both keypoint quantity and precision. Extensive experiments on MegaDepth, ScanNet, and HPatches demonstrate that MoD consistently outperforms strong baselines across a wide range of keypoint budgets, demonstrating its effectiveness in improving keypoint distribution and downstream geometric performance.

Chinese Translation

关键点检测在几何视觉流程中扮演着核心角色。现有的检测器在不同场景结构和关键点预算下展现出互补的优势，依赖单一检测器往往导致次优的关键点分布。基于这一观察，我们提出了 extbf{混合检测器（Mixture of Detectors, MoD）}，用于密集匹配流程，这是一个通过异构检测器的自适应组合实现互补关键点检测的统一框架。MoD首先将异构检测器提炼为一个统一模型，从而在单次前向传播中高效预测多样的检测图。随后，检测器选择被形式化为一个基于补丁的、预算条件的决策过程，其中一个轻量级路由器为每个局部区域分配最合适的检测器。为了避免启发式监督，路由器通过来自下游匹配质量的任务驱动奖励进行训练，联合捕捉关键点的数量和精度。在MegaDepth、ScanNet和HPatches上的大量实验表明，MoD在广泛的关键点预算范围内始终优于强基线，证明了其在改善关键点分布和下游几何性能方面的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 210 / 2610.16784

InfoVis: Unified Slide Image Generation and Editing

信息可视化：统一幻灯片图像生成与编辑

Wang, Jiahao, Jiang, Xiyan, Wu, Chenyuan, Zhang, Haoqiang, Luo, Xin, Wang, Yueze, Lian, Defu, Zhang, Bo, Zhang, Jiajun, Liu, Zheng

Abstract

Generating and editing presentation slides requires models to combine accurate text rendering, instruction following, visual grounding, aesthetic design, and precise editing within one canvas. Existing work covers only subsets of these capabilities and lacks large-scale slide-native supervision and diagnostic evaluation. We formulate four slide-centric tasks and build a scalable data engine with over 13M continual-pretraining records and 1.9M supervised fine-tuning records. We also introduce VisSlideBench, a benchmark of 1,527 manually screened instances that covers all four tasks with task-specific, multidimensional evaluation. Its automatic rankings agree with blinded, dimension-wise human rankings. Using these resources, we train InfoVis, a unified 9B slide generation and editing model. On VisSlideBench, InfoVis leads every evaluated open-source model on all eleven metrics and scores 7.94 overall against 7.99 for Nano-Banana-2. Starting from FLUX.2-klein-base-9B, slide-specific post-training adds 1.6 points overall, with the largest generation gains in content faithfulness; the remaining gap to proprietary models is smallest in content faithfulness and largest in aesthetic quality. Built on InfoVis, our document-to-slide workflow achieves the second-highest Aesthetics score on the external SlidesGen-Bench. We will release the model, code, data, and benchmark.

Chinese Translation

生成和编辑演示幻灯片需要模型在一个画布上结合准确的文本渲染、指令遵循、视觉基础、审美设计和精确编辑。现有工作仅涵盖这些能力的子集，且缺乏大规模的幻灯片原生监督和诊断评估。我们制定了四个以幻灯片为中心的任务，并构建了一个可扩展的数据引擎，拥有超过1300万条持续预训练记录和190万条监督微调记录。我们还引入了VisSlideBench，这是一个包含1527个手动筛选实例的基准，涵盖所有四个任务，并进行任务特定的多维评估。其自动排名与盲评的维度人类排名一致。利用这些资源，我们训练了InfoVis，一个统一的90亿参数幻灯片生成与编辑模型。在VisSlideBench上，InfoVis在所有评估的开源模型中在所有十一项指标上均领先，并以7.94的总分对比Nano-Banana-2的7.99。从FLUX.2-klein-base-9B开始，幻灯片特定的后训练整体提升了1.6分，其中内容真实性的生成增益最大；与专有模型的剩余差距在内容真实性上最小，而在审美质量上最大。基于InfoVis，我们的文档到幻灯片工作流程在外部SlidesGen-Bench上获得了第二高的审美分数。我们将发布模型、代码、数据和基准。

View on arXiv Download PDF AI Translation

cs.CV / 211 / 2610.16793

ViewStruct: View-Grounded Articulation and Connectivity Inference

ViewStruct：基于视图的关节和连接推断

Bian, Wenjing, Huang, Ranran, Geiger, Andreas

Abstract

Interacting with unfamiliar objects requires understanding how their parts connect and move. We introduce ViewStruct, a framework for jointly predicting connectivity and articulation in objects and scenes. Given an RGB image with aligned part segmentation and depth, optionally supplemented by images or video, ViewStruct predicts rigid-body groupings, attachments, and joint parameters. Its view-grounded representation links predictions to observed regions and expresses joint geometry in a gravity-aligned, view-relative frame, without requiring category-specific canonical orientations. A pretrained vision--language model is adapted to predict structure in seconds, without requiring observed articulation motion or per-instance optimisation. Experiments demonstrate improved structural prediction over evaluated baselines and transfer to unseen categories, asset sources, and real-world human-interaction videos, despite task-specific training exclusively on synthetic data. Simulated manipulation experiments further demonstrate the utility of predicted joints for generating robot motion trajectories.

Chinese Translation

与不熟悉的物体互动需要理解其部件如何连接和运动。我们提出了ViewStruct，一个用于联合预测物体和场景中连接性和关节运动的框架。给定一幅带有对齐的部件分割和深度信息的RGB图像，视图结构可选地通过图像或视频进行补充，预测刚体分组、附着和关节参数。其基于视图的表示将预测与观察到的区域联系起来，并在重力对齐、视图相对的坐标系中表达关节几何形状，而无需特定类别的标准方向。一个预训练的视觉-语言模型被调整为在几秒钟内预测结构，而无需观察到的关节运动或每实例优化。实验表明，在评估的基线之上，结构预测得到了改善，并且能够迁移到未见过的类别、资产来源和真实世界的人机交互视频，尽管任务特定训练仅在合成数据上进行。模拟操作实验进一步展示了预测关节在生成机器人运动轨迹方面的实用性。

View on arXiv Download PDF AI Translation

cs.CV / 212 / 2610.16811

X-Ragent: Tool-Augmented Reinforcement Learning for Evidence-Grounded Chest X-Ray Reasoning

X-Ragent：基于工具增强的强化学习用于证据基础的胸部X光推理

Wang, Yi, Yuan, Zhenlong, Guo, Yuchan, Huang, Maolei, Song, Zijian, Zhang, Dapeng, Tan, Rongbin, Wang, Jiyuan, Wang, Zhiyuan, Lin, Fangfang, Song, Shuhan, Cao, Huawei

Abstract

Chest X-ray (CXR) interpretation often requires integrating visual findings with task-specific evidence rather than relying on direct visual recognition alone. While tool-augmented medical agents can supply such evidence via external modules, they often lack explicit mechanisms for learning when auxiliary evidence is needed and for preventing inapplicable quantitative evidence from entering diagnostic reasoning. To address this, we propose X-Ragent, a post-training framework for tool-augmented reinforcement learning that decouples learned auxiliary-evidence seeking from validity-gated quantitative measurement acquisition for evidence-grounded CXR reasoning. We construct Rad-CoT, a 7.7K-sample corpus combining causally ordered tool-augmented trajectories with report-style diagnostic reasoning, and use it to initialize Qwen3-VL-8B via agentic supervised fine-tuning (SFT). We then optimize the policy with Group Relative Policy Optimization (GRPO) using an accuracy-gated reward that applies process-level terms for measurement consistency and auxiliary-tool cost only to correct trajectories, preventing plausible process behavior from compensating for an incorrect diagnosis. Across three CXR benchmarks, X-Ragent achieves the strongest results, including an 8.6-point improvement on the fully held-out ChestAgentBench. By seeking evidence selectively and admitting only valid measurements, X-Ragent supports more faithful and evidence-grounded CXR reasoning.

Chinese Translation

胸部X光（CXR）解读通常需要将视觉发现与特定任务的证据相结合，而不是仅依赖直接的视觉识别。尽管工具增强的医疗代理可以通过外部模块提供此类证据，但它们往往缺乏明确的机制来学习何时需要辅助证据，并防止不适用的定量证据进入诊断推理。为了解决这个问题，我们提出了X-Ragent，这是一种用于工具增强强化学习的后训练框架，它将学习到的辅助证据寻求与有效性门控的定量测量获取解耦，以支持证据基础的CXR推理。我们构建了Rad-CoT，一个包含7.7K样本的语料库，将因果顺序的工具增强轨迹与报告式诊断推理相结合，并利用它通过代理监督微调（SFT）初始化Qwen3-VL-8B。然后，我们使用组相对策略优化（GRPO）优化策略，采用准确性门控奖励，仅将过程级条款应用于测量一致性和辅助工具成本，以确保仅对正确轨迹进行补偿，防止合理的过程行为弥补错误的诊断。在三个CXR基准测试中，X-Ragent取得了最强的结果，包括在完全保留的ChestAgentBench上提高了8.6分。通过选择性地寻求证据并仅接受有效测量，X-Ragent支持更真实和基于证据的CXR推理。

View on arXiv Download PDF AI Translation

cs.CV / 213 / 2610.16813

Carve VFM: Carving Specialists out of Generalists, Automatic Task-Specific Compression of Vision Foundation Models

雕刻 VFM：从通才中雕刻专家，视觉基础模型的自动任务特定压缩

Mecharbat, Lotfi Abdelkrim, Elsayed, Abdelrahman, Almakky, Ibrahim, Takáč, Martin, Yaqub, Mohammad

Abstract

Vision foundation models (VFMs) provide transferable visual representations, but their inference cost limits deployment in many settings. Fine-tuning specializes a VFM without shrinking it; distillation shrinks it, but through expensive teacher-guided training. Structured pruning offers a more direct route by extracting a compact subnetwork from the VFM. However, as compression gets aggressive, existing methods fail because they treat subnetwork selection and recovery as separate axes. Across layers, capacity must follow task sensitivity rather than spread uniformly. Within a layer, a removal cannot be judged by importance alone; its impact also depends on how well the lost contribution can be recovered from what remains. Selection and recovery must therefore be decided jointly. We introduce Carve VFM, which resolves this interdependence. Given a VFM, a calibration set, and a parameter budget, it (i) searches for task-specific capacity allocations across layers; (ii) guides the search via an adaptive recovery-aware fitness criterion; and (iii) reconstructs removed behavior through iterative closed-form low-rank corrections merged into the surviving weights. The result is a dense encoder produced in minutes on a single GPU without backbone retraining. Across twelve benchmarks and four models from DINOv3 and MetaCLIP-2, Carve VFM achieves the highest reported mean accuracy in seven of eight backbone-budget configurations among the evaluated dense structured-pruning methods. Further fine-tuning surpasses the strongest evaluated edge model (88.1% vs. 85.8%), and carved models cut compute and memory while delivering 1.3-2.0x higher GPU throughput than the uncompressed one. These results show that the compact model you need is inside the VFM; it only has to be carved out.

Chinese Translation

视觉基础模型（VFM）提供可转移的视觉表示，但其推理成本限制了在许多场景中的应用。微调可以使 VFM 专业化而不缩小其规模；蒸馏则通过昂贵的教师指导训练来缩小模型。然而，结构化剪枝通过从 VFM 中提取紧凑的子网络提供了一条更直接的路径。然而，随着压缩的激进，现有方法失败，因为它们将子网络选择和恢复视为独立的维度。在各层之间，容量必须遵循任务敏感性，而不是均匀分布。在同一层内，移除的判断不能仅依赖于重要性；其影响还取决于从剩余部分恢复丢失贡献的能力。因此，选择和恢复必须共同决定。我们引入了 Carve VFM，解决了这种相互依赖性。给定一个 VFM、一个校准集和一个参数预算，它（i）在各层之间搜索任务特定的容量分配；（ii）通过自适应的恢复感知适应度标准引导搜索；（iii）通过迭代的封闭形式低秩修正重构移除的行为，这些修正与存活的权重合并。最终结果是在单个 GPU 上在几分钟内生成一个密集编码器，而无需重新训练骨干网络。在 DINOv3 和 MetaCLIP-2 的十二个基准和四个模型中，Carve VFM 在评估的密集结构剪枝方法中，在八个骨干预算配置中的七个中实现了最高报告的平均准确率。进一步的微调超越了最强的评估边缘模型（88.1% 对 85.8%），而雕刻模型在计算和内存上减少了开销，同时提供比未压缩模型高出 1.3-2.0 倍的 GPU 吞吐量。这些结果表明，您所需的紧凑模型就在 VFM 中；它只需要被雕刻出来。

View on arXiv Download PDF AI Translation

cs.CV / 214 / 2610.16817

Growing to See: Efficient Pretraining of Self-Supervised Vision Transformers

逐步增长：自监督视觉变换器的高效预训练

Mamaghan, Amir Mohammad Karimi, Kapl, Ferdinand, Dittadi, Andrea, Johansson, Karl Henrik, Maile, Kaitlin, von Oswald, Johannes, Bauer, Stefan

Abstract

Progressive depth growth builds large networks by starting with shallow models and inserting layers during training. Recent advances in large language models have renewed interest in this longstanding approach, demonstrating benefits for training efficiency and reasoning alongside changes in internal computations. However, how growing affects representation quality and computational patterns in self-supervised vision transformers remains unclear. In this work, we systematically investigate progressive depth growth in two popular self-supervised families, namely masked autoencoders (MAE) and DINOv3, and analyze how growing affects downstream performance, intermediate representations, and internal computations. Across global and dense tasks, we show that growing improves downstream performance while reducing training FLOPs by 13.9--24.8%, with ImageNet linear-probe accuracy gains of up to 3.36 percentage points. Growing also improves downstream performance at intermediate layers and reshapes internal computational patterns, introducing recurring alternations between global and localized attention in MAE and earlier local-to-global attention progression in self-distilled DINOv3. Together, these findings establish progressive depth growth as an effective and efficient pretraining strategy for improving downstream performance across global and dense visual tasks with less computation, and advance our understanding of how growing shapes visual computation.

Chinese Translation

逐步深度增长通过从浅层模型开始并在训练过程中插入层来构建大型网络。最近在大型语言模型方面的进展重新引发了对这一长期方法的兴趣，展示了在训练效率和推理方面的好处，同时伴随着内部计算的变化。然而，逐步增长如何影响自监督视觉变换器的表示质量和计算模式仍不清楚。在本研究中，我们系统地调查了两种流行的自监督模型家族中的逐步深度增长，即掩蔽自编码器（Masked Autoencoders, MAE）和DINOv3，并分析了增长如何影响下游性能、中间表示和内部计算。在全球和密集任务中，我们展示了增长提高了下游性能，同时将训练的FLOPs减少了13.9%至24.8%，在ImageNet线性探测准确率上提高了最多3.36个百分点。增长还改善了中间层的下游性能，并重塑了内部计算模式，在MAE中引入了全局和局部注意力之间的重复交替，以及在自蒸馏DINOv3中更早的局部到全局注意力进展。总之，这些发现确立了逐步深度增长作为一种有效且高效的预训练策略，以改善全球和密集视觉任务的下游性能，同时减少计算量，并推动我们对增长如何塑造视觉计算的理解。

View on arXiv Download PDF AI Translation

cs.CV / 215 / 2610.16849

A Multi-Task Deep Learning Framework for Pose-Invariant Face Recognition and Facial Landmark Localization

一种用于姿态不变人脸识别和面部关键点定位的多任务深度学习框架

Ashkeyev, Ali, Abdiakhmetova, Zukhra

Abstract

Face recognition systems often experience performance degradation under large pose variations due to changes in facial appearance and visibility of discriminative features. This study presents a multi-task deep learning framework that jointly addresses pose-invariant face recognition and facial landmark localization within a unified architecture. To enable simultaneous optimization of identity recognition and facial structure learning, the original Cross-Pose Labeled Faces in the Wild (CPLFW) dataset was adapted into a multi-task learning dataset by associating facial images with identity labels and five-point landmark annotations. The proposed framework consists of two task-specific branches operating on a common facial input: a ResNet50-based branch for identity recognition and an EfficientNetV2B0-based branch with spatial attention for facial landmark localization. The two objectives are jointly optimized using a weighted multi-task loss, allowing the model to learn complementary semantic and spatial facial representations. The proposed framework was evaluated on the prepared CPLFW-based multi-task dataset. Unlike the original CPLFW benchmark, which focuses primarily on identity recognition, the proposed dataset adaptation incorporates five-point facial landmark annotations to enable joint recognition and localization learning. Experimental results demonstrate that the model achieves a validation recognition accuracy of 81.58% and a facial landmark localization mean squared error of 96.38. The results indicate that combining identity classification and landmark regression within a single optimization framework provides an effective approach for learning robust facial representations under cross-pose conditions The proposed method demonstrates the potential of multi-task learning and attention-based feature refinement for unconstrained facial analysis.

Chinese Translation

人脸识别系统在面对大幅姿态变化时，常常会因面部外观和可辨别特征的可见性变化而导致性能下降。本研究提出了一种多任务深度学习框架，在统一架构内共同解决姿态不变的人脸识别和面部关键点定位问题。为了实现身份识别和面部结构学习的同时优化，原始的跨姿态野外标记人脸数据集（Cross-Pose Labeled Faces in the Wild, CPLFW）被改编为多任务学习数据集，通过将面部图像与身份标签和五点关键点注释关联。所提出的框架由两个特定任务的分支组成，这两个分支在共同的面部输入上运行：一个基于ResNet50的分支用于身份识别，另一个基于EfficientNetV2B0并带有空间注意力的分支用于面部关键点定位。这两个目标通过加权多任务损失共同优化，使模型能够学习互补的语义和空间面部表示。所提出的框架在准备好的基于CPLFW的多任务数据集上进行了评估。与原始CPLFW基准主要关注身份识别不同，所提出的数据集改编结合了五点面部关键点注释，以实现联合识别和定位学习。实验结果表明，该模型在验证集上实现了81.58%的识别准确率和96.38的面部关键点定位均方误差。结果表明，在单一优化框架内结合身份分类和关键点回归为在跨姿态条件下学习稳健的面部表示提供了一种有效的方法。所提出的方法展示了多任务学习和基于注意力的特征精炼在无约束面部分析中的潜力。

View on arXiv Download PDF AI Translation

cs.CV / 216 / 2610.16857

Towards Cost-Efficient Video Dataset Construction with Large Language Models

基于大型语言模型的成本高效视频数据集构建研究

Shabot, Mohammed, Adams, Jonathan, Stock, Jan Van den, Vennekens, Joost

Abstract

Research on video classification models has witnessed significant growth recently, accompanied by a significant increase in the demand for datasets to train such models. Existing video dataset curation approaches share a common bottleneck: candidate videos must be downloaded before any relevance judgment is made, which results in storage, time and annotation costs. In this work, we present a filtering step, using Large Language Models (LLMs) to pre-screen videos based on their metadata, before actually downloading these videos, thereby saving on these costs. We evaluate the LLMs against a strict human majority vote derived from four independent annotators and validate the reliability of the annotations through inter-annotator agreement analysis (Gwet's AC1). We compare the LLM pre-filter against two no-LLM baselines, namely two keyword heuristics. Via a Leave-One-Rater-Out (LOO) mechanism and an Add-One-Rater-In (AORI), we evaluate the performance of the LLMs as virtual annotators. Finally, we report cost comparisons relative to existing approaches. To ensure reproducability, we made our implementation available on github.

Chinese Translation

视频分类模型的研究近年来显著增长，随之而来的是对训练此类模型的数据集需求的显著增加。现有的视频数据集策划方法存在一个共同瓶颈：候选视频必须在进行任何相关性判断之前下载，这导致了存储、时间和标注成本的增加。在本研究中，我们提出了一种过滤步骤，利用大型语言模型（LLMs）根据视频的元数据进行预筛选，随后再实际下载这些视频，从而节省这些成本。我们将LLMs与来自四位独立标注者的严格人类多数投票进行比较，并通过标注者间一致性分析（Gwet's AC1）验证标注的可靠性。我们将LLM预筛选与两种无LLM基线进行比较，即两种关键词启发式方法。通过留一标注者法（Leave-One-Rater-Out, LOO）和增加一标注者法（Add-One-Rater-In, AORI），我们评估了LLMs作为虚拟标注者的表现。最后，我们报告了相对于现有方法的成本比较。为了确保可重复性，我们已将我们的实现上传至GitHub。

View on arXiv Download PDF AI Translation

cs.CV / 217 / 2610.16860

Vision Models Capture Complementary Representations in Children That Converge in Adults

视觉模型捕捉儿童的互补表征，成人则趋于一致

Bersch, Domenic, Schaumlöffel, Timothy, Proietti, Michela, Franaszek-Traczewska, Antonia, Zwad, Hannah Elisabeth, Xie, Siying, Baldauf, Marlena, Sylvester, Teresa, Schätz, Christina Maria, Turtleton, Bert, Köster, Moritz, Höhl, Stefanie, Cichy, Radoslaw, Roig, Gemma

Abstract

How the relationship between image-text aligned and vision-only models changes across development remains unknown. We applied time-resolved representational similarity analysis to EEG data from 8-year-olds, 12-year-olds, and adults. CLIP showed significant model-brain alignment already in children, whereas vision-only models aligned more strongly later in development. Residual analyses revealed complementary structure, with CLIP retaining alignment beyond an architecture-matched vision-only model from age 8. Much of the model-brain alignment reflected semantic structure, with EEG representations of category and animacy strengthening toward adulthood, while alignment with low-level image properties remained weak across age. Together, these findings suggest that image-text aligned and vision-only models capture complementary aspects of developing visual representations and may provide useful constraints for AI systems that aim to learn and adapt over time.

Chinese Translation

图像-文本对齐模型与仅视觉模型之间的关系在发展过程中如何变化仍然未知。我们对8岁、12岁儿童及成人的脑电图（EEG）数据应用了时间分辨的表征相似性分析。结果显示，CLIP模型在儿童中已经表现出显著的模型-大脑对齐，而仅视觉模型在发展后期的对齐程度更强。残差分析揭示了互补结构，CLIP在8岁时仍保持与一个架构匹配的仅视觉模型的对齐。模型-大脑对齐的很大一部分反映了语义结构，类别和生物性（animacy）的EEG表征在向成年期发展时逐渐增强，而与低级图像特征的对齐在不同年龄段间保持较弱。综合来看，这些发现表明图像-文本对齐模型和仅视觉模型捕捉了发展中视觉表征的互补方面，并可能为旨在随时间学习和适应的人工智能系统提供有用的约束。

View on arXiv Download PDF AI Translation

cs.CV / 218 / 2610.16862

From Volpiano to Pixels: 30,000 Lines of OMR Ground Truth for Medieval Chant

从 Volpiano 到像素：中世纪圣歌的 30,000 行光学音乐识别 (OMR) 基准数据

Klugseder, Robert

Abstract

Optical music recognition (OMR) for medieval chant lacks training data linking manuscript images to musical symbols across notational families. Using existing Volpiano transcriptions from the Cantus Database, we created 30,567 image-target pairs from ten sources in eleven exports, including 339 verse fragments. Their target sequences contain 1,952,143 tokens. The pipeline locates staff systems, infers clefs and maps known pitches to staff positions. Six detector families handle the different layouts and notations. Where heuristics failed, two small U-Nets learned staff lines and clef letters from manual annotations. Reviews of sampled export lines found zero to four pitch or clef errors, but used different error definitions and do not provide a uniform audit of the corpus. A CRNN trained from scratch achieved a symbol error rate of 9.6 % on 3,086 test lines. The earlier model, trained mainly on Austrian sources of the 15th and 16th centuries, scored 45.3 %. In three separate training runs, each excluding one manuscript, error rates on the excluded manuscript were 21.5-31.0 %. Model-assisted alignment recovered a further 2,686 lines.

Chinese Translation

中世纪圣歌的光学音乐识别 (OMR) 缺乏将手稿图像与不同记谱体系的音乐符号联系起来的训练数据。我们利用来自 Cantus 数据库的现有 Volpiano 转录，创建了来自十个来源的 30,567 对图像-目标对，涵盖了十一种导出格式，包括 339 段诗句片段。它们的目标序列包含 1,952,143 个标记。该流程定位五线谱系统，推断谱号并将已知音高映射到五线谱位置。六个检测器家族处理不同的布局和记谱法。在启发式方法失败的情况下，两个小型 U-Net 从手动标注中学习五线谱线和谱号字母。对抽样导出行的审查发现零到四个音高或谱号错误，但使用了不同的错误定义，并未提供对语料库的统一审计。一个从头开始训练的 CRNN 在 3,086 行测试样本上达到了 9.6% 的符号错误率。早期的模型主要在 15 世纪和 16 世纪的奥地利来源上训练，得分为 45.3%。在三次独立的训练中，每次排除一个手稿，排除手稿的错误率为 21.5% 到 31.0%。模型辅助的对齐恢复了另外 2,686 行。

View on arXiv Download PDF AI Translation

cs.CV / 219 / 2610.16879

EviStress: Evidence Stress Learning for Generalizable Face Forgery Detection with Textual Explanations

EviStress：具有文本解释的可推广人脸伪造检测的证据压力学习

Jing, Jiuyao, Yi, Xuanyu, Zheng, Yu, Peng, Chunlei

Abstract

The rapid spread of face forgeries on the Web challenges the credibility of visual content. In practical Web environments, manipulated media originate from evolving generation pipelines and undergo diverse processing during dissemination, making reliable authenticity assessment under unseen manipulations and visual changes a critical challenge. We observe that a standard LoRA-adapted CLIP detector exhibits uneven patch-level decision dependence, concen trating a disproportionate share of class-specific relevance on a small fraction of visual patches. To improve robustness to variations in such highly dependent evidence, we propose EviStress, a two-stage framework for generalizable face forgery detection with textual explanations. In the first stage, Evidence-Targeted Patch Perturbation (ETPP) estimates patch dependence from global, relational, and semantic cues and assigns positively correlated perturbation probabilities, explicitly stress-testing highly dependent visual evidence during training. We further introduce Discriminative Variation Preservation (DVP), which constrains the residual between clean and perturbed representations to retain authenticity discriminative information. In the second stage, Intrinsic Evidence Instruction Tuning (IEIT) transfers the adapted CLIP encoder into LLaVA as its vision tower and performs parameter-efficient multimodal instruction tuning, enabling authenticity prediction and textual explanation directly from the learned forgery-sensitive representation without an external detector. Experiments demonstrate improved cross-dataset and cross-method generalization and corruption robustness, with the multimodal model providing authenticity predictions and human-readable explanations.

Chinese Translation

人脸伪造在网络上的快速传播对视觉内容的可信性构成了挑战。在实际的网络环境中，经过处理的媒体源于不断演变的生成管道，并在传播过程中经历多样化的处理，这使得在未见操控和视觉变化下进行可靠的真实性评估成为一项关键挑战。我们观察到，标准的LoRA适配CLIP检测器在补丁级决策依赖性上表现不均，过度集中于少数视觉补丁的类别特定相关性。为了提高对这种高度依赖证据的变化的鲁棒性，我们提出了EviStress，一个具有文本解释的可推广人脸伪造检测的两阶段框架。在第一阶段，证据目标补丁扰动（ETPP）从全局、关系和语义线索中估计补丁依赖性，并分配正相关的扰动概率，在训练过程中明确对高度依赖的视觉证据进行压力测试。我们进一步引入了判别变化保留（DVP），该方法限制干净和扰动表示之间的残差，以保留真实性判别信息。在第二阶段，内在证据指令调优（IEIT）将适配后的CLIP编码器转移到LLaVA作为其视觉塔，并执行参数高效的多模态指令调优，使得能够直接从学习到的伪造敏感表示中进行真实性预测和文本解释，而无需外部检测器。实验表明，跨数据集和跨方法的泛化能力和抗损坏能力得到了改善，多模态模型提供了真实性预测和人类可读的解释。

View on arXiv Download PDF AI Translation

cs.CV / 220 / 2610.16892

Barycentric Guidance: Turning Foundation Image Editors into Continuous Affective Controllers

重心引导：将基础图像编辑器转变为连续情感控制器

Mannering, Harvey, Huang, Zhiwu, Prügel-Bennett, Adam

Abstract

Emotions are expressed on a continuum and are naturally described in the valence-arousal (VA) space. Existing VA-based image editors are typically training-based and domain-specific, limiting them to VA regions seen during training and making them brittle across datasets and domains. Meanwhile, despite producing high-quality edits, state-of-the-art foundation image editors are not continuous affective controllers: non-linear emotion representations in text encoders makes prompt interpolation poorly calibrated, so accurate and continuous VA control remains unreliable. To fill this gap, we introduce Barycentric Guidance, a training-free framework that turns pretrained foundation image editors into continuous affective controllers over VA, requiring no finetuning, extra data, or model changes. Our method maps any target VA point to simplex-based guidance weights, enabling VA controllability in three aspects: (i) accurate targeting of VA states, (ii) broad coverage across the VA plane, and (iii) continuous control. Across three face datasets, we improve VA target accuracy by 25-29%, expand VA-plane coverage by 28-50%, and increase expression diversity by 29-49%, over the strongest baseline. Unlike prior domain-specific methods, our training-free approach exploits the foundation model's inherent versatility, enabling generalization beyond human faces to animals, artworks, and complex scenes within a single framework. Code is available at https://github.com/harveymannering/BarycentricGuidance.

Chinese Translation

情感在一个连续体上表达，并自然地描述在情感价值-唤醒（VA）空间中。现有的基于VA的图像编辑器通常是基于训练的且领域特定的，这限制了它们在训练期间所见的VA区域，使其在不同数据集和领域中表现脆弱。同时，尽管产生高质量的编辑，最先进的基础图像编辑器并不是连续的情感控制器：文本编码器中的非线性情感表示使得提示插值的校准效果不佳，因此准确和连续的VA控制仍然不可靠。为了解决这一问题，我们提出了重心引导（Barycentric Guidance），这是一种无训练的框架，可以将预训练的基础图像编辑器转变为在VA空间中的连续情感控制器，无需微调、额外数据或模型更改。我们的方法将任何目标VA点映射到基于单纯形的引导权重，从而在三个方面实现VA可控性：（i）准确定位VA状态，（ii）广泛覆盖VA平面，以及（iii）连续控制。在三个面部数据集上，我们将VA目标准确性提高了25-29%，将VA平面覆盖范围扩大了28-50%，并将表达多样性提高了29-49%，超越了最强基线。与先前的领域特定方法不同，我们的无训练方法利用了基础模型固有的多样性，使其能够在单一框架内超越人脸，推广到动物、艺术作品和复杂场景。代码可在 https://github.com/harveymannering/BarycentricGuidance 获取。

View on arXiv Download PDF AI Translation

cs.CV / 221 / 2610.16906

Mining Cross-Modal Perception Gaps in Autonomous Driving: When Camera and LiDAR Disagree

挖掘自动驾驶中的跨模态感知差异：当摄像头与激光雷达不一致时

Cao, Hongsen, Jaber, Mona, Yuan, Shanxin, Sayed, Ahmed

Abstract

Multi-modal fusion for 3D object detection assumes that cameras and LiDAR are broadly complementary. In practice, how much each modality is trusted at a given location is decided by learned weights that are neither characterized nor auditable. We label every ground-truth object by which modalities detected it, then mine the resulting 0.4M-object database across nuScenes, KITTI, and Waymo. Because the patterns shift with distance from the ego vehicle, we read the mined model through Spatially Stratified Feature Attribution (SSFA), which conditions Shapley attribution on distance zones and exposes structure that global attribution averages away. Three assumptions built into the current fusion design do not hold. The modality-preference boundary tracks the object itself: a traffic cone is nearly six times as likely as a truck to be seen by the camera alone (16.8% vs. 2.9%), regardless of range. Camera-only detections peak at 10-30 m, inside the band usually ceded to LiDAR, and fall away on either side. The drivers of failure are not additive either: the distance - density interaction alone outweighs the entire main effect of viewing direction. We further find that modality dominance is directionally non-stationary, with objects behind the vehicle being markedly more LiDAR-dominant than those in front. We compile the mined structure into a Modality Reliability Map and inject it into four pretrained fusion detectors through a lightweight attention layer. This improves performance by up to 1.3 mean Average Precision (mAP) while fine-tuning under 4% of their parameters in 2-3 GPU-hours. Our work lifts reliability out of the weights and treats it as a quantity that can be examined before deployment. That is what a safety measure for multi-modal perception must rest on.

Chinese Translation

多模态融合用于三维物体检测假设摄像头与激光雷达在广泛上是互补的。实际上，在给定位置上每种模态的信任程度是由学习到的权重决定的，这些权重既没有特征化也无法审计。我们标记每个真实物体由哪些模态检测到，然后在nuScenes、KITTI和Waymo的数据集中挖掘出结果的40万物体数据库。由于模式随着距离自我车辆的变化而变化，我们通过空间分层特征归因（Spatially Stratified Feature Attribution, SSFA）读取挖掘模型，该方法根据距离区域条件化Shapley归因，并揭示了全局归因所平均掉的结构。目前融合设计中内置的三个假设并不成立。模态偏好边界跟踪物体本身：交通锥被摄像头单独看到的可能性几乎是卡车的六倍（16.8%对2.9%），与距离无关。仅摄像头的检测在10-30米处达到峰值，处于通常让位于激光雷达的范围内，并在两侧下降。失败的驱动因素也不是可加的：距离-密度交互单独就超过了观察方向的整体主要影响。我们进一步发现模态主导性在方向上是非平稳的，车辆后方的物体明显更倾向于激光雷达主导，而前方的物体则不然。我们将挖掘出的结构编制成模态可靠性图，并通过轻量级注意力层将其注入到四个预训练的融合检测器中。这在微调其参数不到4%的情况下，在2-3个GPU小时内将性能提升了最多1.3的平均精度（mean Average Precision, mAP）。我们的工作将可靠性从权重中提取出来，并将其视为在部署前可以检查的量。这是多模态感知的安全措施必须依赖的基础。

View on arXiv Download PDF AI Translation

cs.CV / 222 / 2610.16908

ConDenseDrive: Knowledge-Dense, Token-Sparse Visual Encoder for End-to-End Autonomous Driving

ConDenseDrive：用于端到端自动驾驶的知识密集型、令牌稀疏视觉编码器

Li, Shengzhen, Li, Yongkang, Gao, Hao, Wu, Chenxi, Sun, Yajing, Hou, Lu, Liu, Wenyu, Wang, Xinggang

Abstract

End-to-end driving requires rich spatiotemporal representations to support scene understanding and trajectory planning under strict latency constraints. However, most systems initialize their visual encoders from generic image pretraining and densely process all patch tokens for each camera view and input frame, limiting both representation quality and inference efficiency. To address these challenges, we present ConDenseDrive, a knowledge-dense and token-sparse visual encoder for end-to-end autonomous driving. Our three-stage framework first makes the encoder knowledge-dense by distilling complementary visual, semantic, and geometric knowledge from three vision foundation models. It then adapts the distilled encoder to multi-frame end-to-end driving and integrates spatiotemporal context into register tokens for planning. Finally, it makes the encoder token-sparse by applying register-guided token compression, reducing redundant patch computation and yielding a 2.0$\times$ speedup while improving planning performance. ConDenseDrive achieves 94.5 PDMS on NAVSIM v1 and 90.8 EPDMS on NAVSIM v2, and reaches 37.4 HD-Score in zero-shot evaluation on HUGSIM. Extensive experiments and qualitative analyses demonstrate that visual encoder design provides an effective route to reliable and efficient end-to-end driving.

Chinese Translation

端到端驾驶需要丰富的时空表示，以支持在严格延迟限制下的场景理解和轨迹规划。然而，大多数系统从通用图像预训练初始化其视觉编码器，并对每个相机视图和输入帧的所有补丁令牌进行密集处理，这限制了表示质量和推理效率。为了解决这些挑战，我们提出了ConDenseDrive，一种用于端到端自动驾驶的知识密集型和令牌稀疏视觉编码器。我们的三阶段框架首先通过从三个视觉基础模型中提取互补的视觉、语义和几何知识，使编码器变得知识密集。然后，它将提取的编码器适应于多帧端到端驾驶，并将时空上下文集成到注册令牌中以进行规划。最后，通过应用注册引导的令牌压缩，使编码器变得令牌稀疏，减少冗余的补丁计算，实现了2.0倍的速度提升，同时改善了规划性能。ConDenseDrive在NAVSIM v1上达到了94.5 PDMS，在NAVSIM v2上达到了90.8 EPDMS，并在HUGSIM的零-shot评估中达到了37.4 HD-Score。大量实验和定性分析表明，视觉编码器设计为可靠和高效的端到端驾驶提供了一条有效的途径。

View on arXiv Download PDF AI Translation

cs.CV / 223 / 2610.16914

FOCUS: A Subspecialty Foundation Model for Intraoperative Central Nervous System Pathology

FOCUS：一种用于术中中枢神经系统病理的亚专业基础模型

Liu, Xing, Guo, Yifan, Lv, Hongtai, Lou, Lei, Tan, Yuyi, Du, Jiang, Zhang, Xijia, Zhang, Ming, Pang, Chao, Li, Minglei, Liu, Wei, Zhu, Xiaobo, Zhang, Song, Yu, Rong, Shu, Hanlin, Xu, Wenzhuo, Jiang, Tao, Lu, Cheng, Tian, Jie, Chang, Qing, Liu, Zhenyu

Abstract

Intraoperative frozen-section diagnosis of central nervous system (CNS) lesions guides surgical decisions but is challenged by artefacts, diagnostic complexity and time constraints. Most pathology foundation models have limited exposure to CNS frozen tissue. Here we present FOCUS, a subspecialty foundation model pretrained on 27,930 frozen-sections enriched for CNS tissue. FOCUS led 11 internal diagnostic and 12 biomarker-prediction tasks, with mean AUCs of 0.969 and 0.877, respectively, retaining leading performance in temporal and external cohorts. To provide more comprehensive diagnostic information, we further organized ten diagnostic tasks into a WHO-aligned diagnostic hierarchy and separately evaluated FOCUS for lymphoma detection, a critical task for intraoperative surgical decision-making. In reader assessments by six neuropathologists, FOCUS assistance enabled junior readers to approach or exceed unassisted senior readers in hierarchical diagnosis and raised junior readers' lymphoma sensitivity from 81.6% to 94.7%. FOCUS could improve diagnostic accuracy and granularity, inform lymphoma-related resection decisions and provide provisional molecular information. Such capabilities highlight the potential of FOCUS to provide robust, clinically actionable support for intraoperative CNS pathology.

Chinese Translation

术中冰冻切片诊断中枢神经系统（CNS）病变为外科决策提供指导，但受到伪影、诊断复杂性和时间限制的挑战。大多数病理基础模型对CNS冰冻组织的接触有限。在此，我们提出了FOCUS，一种在27,930个富含CNS组织的冰冻切片上预训练的亚专业基础模型。FOCUS在11个内部诊断任务和12个生物标志物预测任务中表现出色，平均AUC分别为0.969和0.877，并在时间和外部队列中保持领先表现。为了提供更全面的诊断信息，我们进一步将十个诊断任务组织成与世界卫生组织（WHO）对齐的诊断层级，并单独评估FOCUS在淋巴瘤检测中的表现，这是术中外科决策的关键任务。在六位神经病理学家的读者评估中，FOCUS的辅助使初级读者在层级诊断中接近或超过未辅助的高级读者，并将初级读者的淋巴瘤灵敏度从81.6%提高至94.7%。FOCUS有望提高诊断准确性和细致性，为淋巴瘤相关的切除决策提供信息，并提供临时分子信息。这些能力突显了FOCUS在提供稳健、临床可操作支持方面的潜力，特别是在术中CNS病理的应用中。

View on arXiv Download PDF AI Translation

cs.CV / 224 / 2610.16945

UniMetric: Universal Metric Scale Recovery for Visual Geometry Foundation Models

UniMetric：视觉几何基础模型的通用度量尺度恢复

Yang, Lei, Zhang, Yuner, Liu, Ziyang, Song, Ziying, WU, Kai, Gao, Yiyang, Su, Jiangtao, Li, Yiming

Abstract

Recent advances in multimodal foundation models, exemplified by GPT-6 Astra, raise the prospect of general-purpose embodied reasoning and planning. Using Internet videos for embodied pretraining calls for reliable physical-scale annotations. Yet visual geometry foundation models often recover scene structure and camera motion without reliable metric scale. We show that uncertain size priors of familiar objects can jointly constrain a reconstruction's shared scale. Building on this insight, we present UniMetric, a universal and training-free metric scale recovery framework for geometry foundation models, without additional sensors or task-specific metric supervision. Specifically, Metric Prior Acquisition elicits object height intervals from a vision-language model. Semantic-Geometric Alignment measures reconstructed object heights along the estimated support-surface normal in each frame and converts the priors into scale intervals. Global Scale Recovery reconciles these intervals through a convex objective, yielding a jointly feasible scale when they overlap and a unique compromise otherwise. The estimated sequence-level scale is applied uniformly to reconstructed points and camera translations. Controlled experiments with real VLM priors show that combining observations improves scale accuracy with exact relative object heights. To evaluate metric accuracy on GFM reconstructions without ground-truth scale alignment, we introduce MetricBench, covering twelve embodied, driving, and roadside subsets with sensor-derived ground truth. Across nine geometry foundation models, UniMetric reduces mean depth Abs Rel, reconstruction Chamfer distance, and trajectory error by 54.2%, 83.8%, and 65.6%, respectively, improving every model on all three tasks at the suite level. We release the UniMetric code and MetricBench data.

Chinese Translation

近期多模态基础模型的进展，以GPT-6 Astra为例，提升了通用化具身推理和规划的前景。利用互联网视频进行具身预训练需要可靠的物理尺度注释。然而，视觉几何基础模型通常在没有可靠度量尺度的情况下恢复场景结构和相机运动。我们表明，熟悉物体的不确定尺寸先验可以共同约束重建的共享尺度。基于这一见解，我们提出了UniMetric，一个通用且无训练的度量尺度恢复框架，适用于几何基础模型，无需额外传感器或特定任务的度量监督。具体而言，度量先验获取从视觉-语言模型中引出物体高度区间。语义-几何对齐在每帧中沿估计的支撑面法线测量重建物体的高度，并将先验转换为尺度区间。全局尺度恢复通过一个凸目标调和这些区间，当它们重叠时产生一个共同可行的尺度，否则则产生一个唯一的折衷。估计的序列级尺度均匀应用于重建点和相机平移。与真实的视觉语言模型先验进行的受控实验表明，结合观察结果提高了尺度准确性，得到了精确的相对物体高度。为了在没有真实尺度对齐的情况下评估几何基础模型重建的度量准确性，我们引入了MetricBench，涵盖了十二个具身、驾驶和路边子集，并提供传感器导出的真实数据。在九个几何基础模型中，UniMetric分别将平均深度绝对相对误差、重建Chamfer距离和轨迹误差降低了54.2%、83.8%和65.6%，在套件级别上改善了每个模型在所有三个任务上的表现。我们发布了UniMetric代码和MetricBench数据。

View on arXiv Download PDF AI Translation

cs.CV / 225 / 2610.16956

DDVR-Net: Degradation-Aware Decoupling and Visibility-Guided Restoration for Historical Document Recognition

DDVR-Net：针对历史文献识别的降解感知解耦与可见性引导恢复

Sha, Jiu, Weng, Yu, Zhu, Mengxiao

Abstract

Historical document analysis is crucial for cultural heritage preservation and low-resource language digitization. However, real historical documents often suffer from severe degradation, such as ink erosion and structural damage, causing mixed-visibility character sequences. Conventional OCR systems struggle in such evidence-limited scenarios, while contextual language models may over-correct visually clear characters. To address these challenges, we propose DDVR-Net, a degradation-aware decoupling and visibility-guided restoration framework for historical document recognition. DDVR-Net decouples intrinsic glyph structures from physical corruption at the feature level and predicts local visibility to estimate character reliability. A contextual decoder is selectively activated through a visibility-guided gate, restoring degraded characters while preserving clear visual recognition. Furthermore, we construct a low-resource multilingual benchmark covering Tibetan, Mongolian, and Uyghur scripts with 187000 document pairs, combining large-scale synthetic data and progressive training with real-world annotations to reduce the synthetic-to-real gap. Extensive experiments demonstrate that DDVR-Net consistently outperforms existing methods across three languages and diverse reading orders, providing an effective framework and benchmark for low-resource historical document recognition. Code and Benchmark: https://github.com/Shajiu/DDVR-Net

Chinese Translation

历史文献分析对于文化遗产保护和低资源语言数字化至关重要。然而，真实的历史文献往往遭受严重的降解，如墨水侵蚀和结构损坏，导致字符序列的可见性混合。传统的光学字符识别（OCR）系统在这种证据有限的场景中表现不佳，而上下文语言模型可能会过度修正视觉上清晰的字符。为了解决这些挑战，我们提出了DDVR-Net，这是一种针对历史文献识别的降解感知解耦与可见性引导恢复框架。DDVR-Net在特征层面上将内在字形结构与物理损坏解耦，并预测局部可见性以估计字符的可靠性。通过可见性引导门选择性激活上下文解码器，恢复降解字符的同时保持清晰的视觉识别。此外，我们构建了一个涵盖藏文、蒙古文和维吾尔文的低资源多语言基准，包含187000对文档，结合大规模合成数据和与真实世界注释的渐进训练，以缩小合成与真实之间的差距。大量实验表明，DDVR-Net在三种语言和多样的阅读顺序中始终优于现有方法，为低资源历史文献识别提供了有效的框架和基准。代码和基准： https://github.com/Shajiu/DDVR-Net

View on arXiv Download PDF AI Translation

cs.CV / 226 / 2610.16967

Proactive Physical AI through Continuous Observation and Timely Intervention

通过持续观察和及时干预实现主动物理人工智能

Yang, Zhenhua, Liang, Jiafeng, Zhou, Xingyu, Jiang, Yi, Cao, Xiayu, Chen, Zui, Wang, Luqi, Sun, Haiyang, Xiang, Yufei, Li, Chang, Mu, Rongchuan, Chi, Bowen, Tong, Zihe, Li, Mingyu, Ma, Huijie, Yang, Yiqiao, Ren, Baoqi, Ma, Zhiming, Zhu, Zhihao, Ji, Jiahao, Wang, Zhuoyao, Wang, Qianyu, Liu, Xiangyu, Zixiang, Wang, Zhang, Shunliang, Liu, Ming, Feng, Xiaocheng, Tang, Duyu, Qin, Bing

Abstract

Physical AI is commonly formulated around an explicitly specified task: given a goal or instruction, a system perceives, reasons, and acts. In many real-world settings, however, important tasks emerge from continuously changing world states before a person explicitly formulates them. We study this complementary problem as Proactive Physical AI, where an AI system continuously observes the physical world, identifies situations that warrant intervention, and initiates appropriate actions at the right time. We develop ProAgent, an end-to-end system built around two components: a Jev-style proactive timing model (dubbed StreamingJev) that determines whether and when intervention is needed, and a Response Model that generates responses for people or invokes available device capabilities for physical execution. To support learning at scale, we develop an automated data pipeline that converts videos into supervision for proactive timing. Experiments demonstrate strong proactive performance under both explicit objectives and standing roles, combining more accurate intervention timing with fewer false triggers than the source-backbone baseline. Real-robot evaluations further demonstrate ProAgent's ability to identify emerging needs and translate them into appropriate physical actions.

Chinese Translation

物理人工智能通常围绕明确指定的任务进行构建：给定一个目标或指令，系统进行感知、推理和行动。然而，在许多现实世界的环境中，重要任务往往在一个人明确表述之前就从不断变化的世界状态中出现。我们将这一互补问题研究为主动物理人工智能（Proactive Physical AI），在该框架下，人工智能系统持续观察物理世界，识别需要干预的情况，并在适当的时机发起相应的行动。我们开发了ProAgent，这是一个端到端的系统，围绕两个组件构建：一个Jev风格的主动时机模型（称为StreamingJev），用于确定是否以及何时需要干预；以及一个响应模型，用于为人类生成响应或调用可用设备能力以进行物理执行。为了支持大规模学习，我们开发了一个自动化数据管道，将视频转换为主动时机的监督信号。实验结果表明，在明确目标和持续角色下，ProAgent展现出强大的主动性能，结合了更准确的干预时机和更少的误触发，优于源骨干基线。真实机器人评估进一步证明了ProAgent识别新兴需求并将其转化为适当物理行动的能力。

View on arXiv Download PDF AI Translation

cs.CV / 227 / 2610.16977

The Value of Being Wrong: Self-Mined Visual In-Context Learning from Errors

错误的价值：从错误中自我挖掘的视觉上下文学习

Zhao, Shu, Yu, Tan

Abstract

Visual in-context learning (ICL) enables vision-language models (VLMs) to adapt at test time via demonstration examples, but its effectiveness depends critically on the examples selected. Recent counterfactual selection methods improve ICL by constructing demonstrations that expose how individual attribute changes affect the answer. However, they require multi-stage pipelines involving attribute extraction, caption engineering, and composed image matching. We propose a simpler alternative grounded in a key observation: a model's own prediction errors are natural counterfactual demonstrations that directly encode its failure modes. Our method, SMILE (Self-Mined Visual In-Context Learning from Errors), builds a model-specific pool of hard negatives offline by recording the target VLM's own mistakes. At test time, the prediction serves as a direction signal that surfaces, in the semantic embedding space, the past mistakes most likely to recur for the current query. This prediction-conditioned identification provides a unified mechanism for both closed-set classification and open-ended visual question answering without task-specific design. Across four benchmarks and four VLMs, SMILE delivers an average task-score gain of +3.09 points over the SOTA baseline at 5x lower latency. Project page: https://tree-shu-zhao.github.io/SMILE/

Chinese Translation

视觉上下文学习（ICL）使视觉-语言模型（VLMs）能够通过示例在测试时进行适应，但其有效性在很大程度上依赖于所选择的示例。最近的反事实选择方法通过构建示范，揭示个体属性变化如何影响答案，从而改善了ICL。然而，这些方法需要涉及属性提取、标题工程和组合图像匹配的多阶段管道。我们提出了一种更简单的替代方案，基于一个关键观察：模型自身的预测错误是自然的反事实示范，直接编码了其失败模式。我们的方法SMILE（Self-Mined Visual In-Context Learning from Errors）通过记录目标VLM自身的错误，离线构建模型特定的困难负样本池。在测试时，预测作为方向信号，揭示在语义嵌入空间中，当前查询最有可能重现的过去错误。这种基于预测的识别为闭集分类和开放式视觉问答提供了统一机制，而无需特定任务设计。在四个基准和四个VLM上，SMILE在延迟降低5倍的情况下，平均任务得分比SOTA基线提高了+3.09分。项目页面：https://tree-shu-zhao.github.io/SMILE/

View on arXiv Download PDF AI Translation

cs.CV / 228 / 2610.16985

AUGUstus: Autonomous Driving VLA Model with Unified Generation and Understanding

AUGUstus：具有统一生成与理解的自主驾驶视觉-语言-动作模型

Ye, Yuqi, Jiang, Caigao, Wang, Zhuonan, Pei, Congyuan, Qu, Chao, Peng, Changhao, Lin, Junhong, Yu, Qize, Song, Zetian, Fu, Guangzhao, Wu, Chongruo, Huang, Shiyu, Zhou, Shuchang, Yang, Honggou, Gao, Wei

Abstract

Vision-language-action (VLA) models show strong potential for end-to-end autonomous driving by leveraging world knowledge and understanding to excel in long-tail scenarios. However, current VLAs have two limitations: 1) Degeneration into vision-action (VA) models. For low latency, regression/flow-based methods directly generate actions but underuse VLM understanding, causing unsafe, unexplainable actions. 2) Autoregressive high latency. Autoregressive methods sequentially generate chain-of-thought (CoT) and action tokens, fully leveraging VLM understanding, but high latency hinders deployment. To enable low-latency deployment while using CoT to assist action generation, we propose AUGUstus, an Autonomous-driving VLA model with Unified Generation and Understanding. We unify diffusion-based action generation and autoregressive CoT understanding in one model, where understanding only serves generation. Joint CoT supervision preserves VLM spatial understanding in shared representation for safer, more accurate trajectories. AUGUstus-mode orders action generation before CoT, so CoT supervises reliable actions during training while inference needs only action output, enabling low latency. AUGUstus-GRPO, a unified GRPO algorithm, jointly optimizes action and CoT, improving action performance. Experiments show AUGUstus achieves 94.6 PDMS on NAVSIMv1 and 90.7 EPDMS on NAVSIMv2, SOTA planning without scorer module. On nuScenes, it achieves average L2 error 0.27 m and average collision rate 0.05%, best among compared methods. To validate AUGUstus practicality, we train on in-house data with 512 NVIDIA H800 GPUs for ~2 days, evaluate on an internal 1,000 km dataset, and preliminarily deploy on a real vehicle, achieving 12 Hz operation. Project page: https://augustus-ad.github.io

Chinese Translation

视觉-语言-动作（VLA）模型通过利用世界知识和理解，在端到端自主驾驶中展现出强大的潜力，尤其是在长尾场景中表现优异。然而，当前的VLA模型存在两个局限性：1）退化为视觉-动作（VA）模型。为了实现低延迟，回归/流基方法直接生成动作，但未充分利用VLM（视觉语言模型）的理解，导致不安全且不可解释的动作。2）自回归高延迟。自回归方法顺序生成思维链（CoT）和动作标记，充分利用VLM的理解，但高延迟阻碍了部署。为了在使用CoT辅助动作生成的同时实现低延迟部署，我们提出了AUGUstus，一个具有统一生成与理解的自主驾驶VLA模型。我们将基于扩散的动作生成与自回归CoT理解统一到一个模型中，其中理解仅用于生成。联合CoT监督在共享表示中保留了VLM的空间理解，从而实现更安全、更准确的轨迹。AUGUstus模型在CoT之前进行动作生成，因此在训练过程中CoT监督可靠的动作，而推理时仅需动作输出，从而实现低延迟。AUGUstus-GRPO，一个统一的GRPO算法，联合优化动作和CoT，提升动作性能。实验表明，AUGUstus在NAVSIMv1上达到了94.6的PDMS，在NAVSIMv2上达到了90.7的EPDMS，且在没有评分模块的情况下实现了SOTA（最先进技术）规划。在nuScenes上，它的平均L2误差为0.27米，平均碰撞率为0.05%，在比较方法中表现最佳。为了验证AUGUstus的实用性，我们在512个NVIDIA H800 GPU上训练了约2天，评估了一个内部的1000公里数据集，并初步在真实车辆上进行了部署，实现了12 Hz的操作频率。项目页面：https://augustus-ad.github.io

View on arXiv Download PDF AI Translation

cs.CV / 229 / 2610.16988

Self-Supervised Learning of Semantically Consistent Visual Features via a Simple 3D Prior

通过简单的三维先验进行语义一致视觉特征的自监督学习

Day, Sebastian, Mac Aodha, Oisin

Abstract

Visual features extracted from vision models like DINOv2 or Stable Diffusion enable high quality semantic correspondence between images. However, such features exhibit documented limitations that hinder accurate correspondence estimation such as the inability to disambiguate the left from right for categories exhibiting symmetries or repeated parts. Previous attempts to improve these features rely on using supervised data such as annotated keypoint pairs or camera pose information, or make restrictive topological assumptions, such as modelling the object of interest as a 3D sphere. We introduce a new self-supervised approach that allows us to learn features for topologically complex object categories using a simple prior. Instead of relying on an expensive instance-level analysis by synthesis pipeline, our approach makes use of contrastive learning and distribution matching at the global dataset-level to learn the coarse shape and appearance of a category. Combined with additional geometric losses, this enables us to replicate the global and local statistics of visual features which match those that are extracted from real images. Quantitative results are presented across several challenging object categories, demonstrating the effectiveness of our approach.

Chinese Translation

从视觉模型（如 DINOv2 或 Stable Diffusion）中提取的视觉特征能够实现图像之间高质量的语义对应。然而，这些特征存在已知的局限性，妨碍了准确的对应估计，例如无法区分具有对称性或重复部分的类别的左右。之前改善这些特征的尝试依赖于使用监督数据，如标注的关键点对或相机姿态信息，或做出限制性的拓扑假设，例如将感兴趣的对象建模为三维球体。我们提出了一种新的自监督方法，使我们能够使用简单的先验学习拓扑复杂的对象类别特征。我们的 approach 不依赖于昂贵的实例级分析合成管道，而是利用对比学习和全局数据集级别的分布匹配来学习类别的粗略形状和外观。结合额外的几何损失，这使我们能够复制与从真实图像中提取的视觉特征相匹配的全局和局部统计信息。我们在多个具有挑战性的对象类别上展示了定量结果，证明了我们方法的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 230 / 2610.17009

MaskSense: Confronting the Visual Exploration Trap in Masked Image Generation

MaskSense：应对掩码图像生成中的视觉探索陷阱

Shao, Yawen, Xiao, Jie, Zhu, Kai, Liu, Yu, Luo, Hongchen, Fu, Xueyang, Cao, Yang, Zhai, Wei, Zha, Zheng-Jun

Abstract

Reinforcement learning (RL) has shown strong potential for aligning generative models with human intent, but adapting RL to masked generative models (MGMs) remains largely underexplored. Existing approaches that directly adapt Group Relative Policy Optimization (GRPO) to MGMs leave a foundational conflict unaddressed: MGMs' exploitative decoding nature is inherently at odds with the exploratory diversity that GRPO presupposes. In this work, we reveal a structural decoding bias termed the visual exploration trap: confidence-based sampling trades the exploration of diverse subject realizations for the greedy resolution of low-uncertainty background regions, causing premature collapse of the generative exploration space and starving GRPO of the rollout diversity it depends on. To this end, we propose MaskSense, a novel RL framework for MGMs that confronts the visual exploration trap at both the sampling and optimization stages. Specifically, we introduce a semantic anchored routing sampling mechanism that leverages semantic priors to preserve high-entropy subject token exploration while amplifying intra-group reward variance through exploratory-exploitative routing to yield more discriminative advantage estimates. Furthermore, we design a global entropy transition anchoring strategy to identify the most consequential decoding steps and concentrate gradient updates on them. Extensive experiments on multiple text-to-image benchmarks demonstrate that MaskSense substantially improves upon the base model, with GenEval accuracy improving from 54% to 86% and HPS from 28.89 to 36.39, achieving state-of-the-art performance.

Chinese Translation

强化学习（RL）在将生成模型与人类意图对齐方面展现了强大的潜力，但将 RL 适应于掩码生成模型（MGM）仍然在很大程度上未得到探索。现有的将群体相对策略优化（GRPO）直接适应于 MGM 的方法未能解决一个基础性冲突：MGM 的剥削性解码特性与 GRPO 所假设的探索多样性本质上是相悖的。在本研究中，我们揭示了一种结构性解码偏差，称为视觉探索陷阱：基于置信度的采样将多样化主题实现的探索与低不确定性背景区域的贪婪解析进行权衡，导致生成探索空间的过早崩溃，并使 GRPO 失去其所依赖的滚动多样性。为此，我们提出了 MaskSense，一种新颖的针对 MGM 的 RL 框架，旨在在采样和优化阶段同时应对视觉探索陷阱。具体而言，我们引入了一种语义锚定路由采样机制，该机制利用语义先验来保持高熵主题标记的探索，同时通过探索-剥削路由放大组内奖励方差，从而产生更具区分性的优势估计。此外，我们设计了一种全局熵过渡锚定策略，以识别最重要的解码步骤并集中梯度更新。对多个文本到图像基准的广泛实验表明，MaskSense 显著改善了基础模型的性能，GenEval 准确率从 54% 提升至 86%，HPS 从 28.89 提升至 36.39，达到了最先进的性能。

View on arXiv Download PDF AI Translation

cs.CV / 231 / 2610.17015

DIME: Scaling Facial Representation Learning via Differential Masked Autoencoding

DIME：通过差分掩码自编码扩展面部表征学习

Yu, Hao, Chen, Haoyu, Wei, Hui, Jiang, Yan, Sebe, Nicu, Zhao, Guoying

Abstract

Scaling up data and model capacity has driven substantial progress in visual representation learning. However, conventional pretext tasks exhibit diminishing returns on fine-grained facial analysis. We attribute this to a scaling bottleneck: these tasks model each image's appearance and are dominated by identity, pose, and illumination, so subtle within-identity facial differences contribute little to the training signal. We introduce DIME, a differential masked autoencoder for large-scale facial representation learning. DIME builds on a simple observation: two images of the same person share identity-specific appearance, so their difference exposes facial changes. DIME reconstructs both masked images in an identity-matched pair and, through an Explicit Differential Direction Supervision (EDDS) objective, aligns the direction of their reconstructed difference with the true difference, while source isolation makes each reconstruction depend only on its own image. Shared appearance cancels in this difference, and because EDDS ignores magnitude, subtle facial changes count as much as large pose or illumination changes. DIME uses identity labels only to form pretraining pairs, requires no dense annotations, and learns dense facial representations with emergent part-level semantics. We pretrain DIME on 20 million high-resolution facial images consolidated from existing face datasets. DIME's AU detection performance continues to improve as pretraining data scales from 6M to 20M images, with its BP4D F1 increasing by 5.9 points, while MAE gains only 0.5 points on the same subsets. DIME also outperforms specialized facial methods on AU detection, AU intensity estimation, cross-domain anti-spoofing, landmark localization, head pose estimation, face parsing, and face generation.

Chinese Translation

数据和模型容量的扩大推动了视觉表征学习的显著进展。然而，传统的预训练任务在细粒度面部分析中表现出收益递减的现象。我们将此归因于一个扩展瓶颈：这些任务对每个图像的外观进行建模，并受到身份、姿态和光照的主导，因此同一身份下微妙的面部差异对训练信号的贡献很小。我们提出了DIME，一种用于大规模面部表征学习的差分掩码自编码器。DIME基于一个简单的观察：同一人的两幅图像共享身份特定的外观，因此它们的差异揭示了面部变化。DIME重建身份匹配对中的两个掩码图像，并通过显式差分方向监督（Explicit Differential Direction Supervision, EDDS）目标，使它们重建的差异的方向与真实差异对齐，同时源隔离使每个重建仅依赖于其自身图像。共享的外观在这个差异中被抵消，并且由于EDDS忽略了幅度，微妙的面部变化与大的姿态或光照变化同样重要。DIME仅使用身份标签来形成预训练对，不需要密集注释，并学习具有突现部分级语义的密集面部表征。我们在从现有面部数据集中整合的2000万张高分辨率面部图像上对DIME进行预训练。随着预训练数据从600万扩展到2000万张图像，DIME的AU检测性能持续提高，其BP4D F1分数增加了5.9分，而MAE在相同子集上的增幅仅为0.5分。DIME在AU检测、AU强度估计、跨域防伪、地标定位、头部姿态估计、面部解析和面部生成等任务中也优于专业的面部方法。

View on arXiv Download PDF AI Translation

cs.CV / 232 / 2610.17066

PerturbBridge: Direct Phenotypic Supervision for Cellular Perturbation Generation

PerturbBridge：用于细胞扰动生成的直接表型监督

Yang, Yucheng, Sun, Xiaoxiao, Sun, Huangqingbo, Sui, Elaine, Lundberg, Emma, Yeung-Levy, Serena

Abstract

Virtual cell models should capture phenotypes induced by specific perturbations. However, image-based models are typically trained and evaluated to match real-image distributions, which alone does not explicitly prioritize perturbation-specific biological responses. Recent approaches add biological signals through reinforcement-learning post-training or inference-time steering, raising a central question: where should biological response information enter the generative pipeline? We introduce PerturbBridge, a stochastic data-to-data bridge model that directly supervises perturbation-specific phenotypes in predicted perturbed images during training. PerturbBridge jointly optimizes bridge regression and differentiable biological response losses, with biological gradients passing through a frozen image encoder. We further introduce perturbation-response evaluations based on downstream classification and retrieval. Across three microscopy benchmarks for perturbation profiling and subcellular imaging, PerturbBridge surpasses CellFlux in biological fidelity, image quality, and sampling efficiency. On BBBC021, mechanism-of-action classification improves by 5.66 and 7.5 percentage points with morphological and learned visual features, respectively. On JUMP and AllenCell, retrieval Hit@10 and Hit@1 improve by 1.77 and 7.15 percentage points, respectively. On BBBC021 and JUMP, PerturbBridge achieves lower FID with 80% fewer model evaluations than CellFlux. Overall, these results show that directly supervising perturbation responses during training improves phenotypic fidelity while enabling efficient cellular image generation. Code is available at https://github.com/yyc1215lll-creator/PerturbBridge.

Chinese Translation

虚拟细胞模型应当捕捉由特定扰动引起的表型。然而，基于图像的模型通常是为了匹配真实图像分布而进行训练和评估，这并不能明确优先考虑特定扰动的生物学反应。最近的方法通过强化学习后训练或推理时引导添加生物信号，提出了一个核心问题：生物反应信息应在生成管道的何处进入？我们提出了PerturbBridge，这是一种随机数据到数据的桥接模型，直接在训练期间监督预测扰动图像中的特定表型。PerturbBridge联合优化桥接回归和可微分的生物反应损失，生物梯度通过一个冻结的图像编码器传递。我们进一步引入基于下游分类和检索的扰动反应评估。在三个用于扰动分析和亚细胞成像的显微镜基准测试中，PerturbBridge在生物学保真度、图像质量和采样效率上超越了CellFlux。在BBBC021数据集上，机制作用分类分别提高了5.66和7.5个百分点，使用形态学和学习到的视觉特征。在JUMP和AllenCell上，检索的Hit@10和Hit@1分别提高了1.77和7.15个百分点。在BBBC021和JUMP上，PerturbBridge以比CellFlux少80%的模型评估实现了更低的FID。总体而言，这些结果表明，在训练期间直接监督扰动反应可以提高表型保真度，同时实现高效的细胞图像生成。代码可在https://github.com/yyc1215lll-creator/PerturbBridge获取。

View on arXiv Download PDF AI Translation

cs.CV / 233 / 2610.17067

A unified three-branch deep learning framework for automated detection and counting of wildlife in aerial imagery

统一的三分支深度学习框架用于自动检测和计数航空影像中的野生动物

Xu, Zeyu, Wang, Tiejun, Skidmore, Andrew, Claase, Megan, Wells, Harry, Duporge, Isla, Crego, Ramiro D., Lamprey, Richard

Abstract

Automating aerial wildlife surveys requires reliable animal detection, consistent delineation of observation strips, and correction for repeated coverage between successive images. We present a unified three-branch deep learning framework that integrates these tasks through WildSeg, WildMatch and WildDet. WildSeg combines rod segmentation with geometric refinement to delineate counting strips from aircraft-mounted markers. WildMatch uses teacher-student distillation to learn dense image correspondences and identify overlapping survey areas without relying on geolocation metadata. WildDet combines dual-scale detection with feature fusion, density supervision and empty-image gating to support animal detection and counting. We evaluated the framework using oblique aerial imagery of four antelope species from the Great Nile Migration Landscape in South Sudan. WildSeg achieved reliable rod extraction, while WildMatch approached teacher-level accuracy with a substantially lighter and faster model. Both modules maintained high accuracy on resized images, improving processing efficiency. WildDet improved detection performance across all four species and reduced detection-based count MAE, RMSE and mean bias by 21.0%, 4.6% and 27.8%, respectively, relative to YOLO11. By connecting spatial survey constraints with animal detection, the framework provides a practical and computationally efficient approach to processing large aerial image collections for wildlife monitoring and population surveys.

Chinese Translation

自动化航空野生动物调查需要可靠的动物检测、一致的观察带划分，以及对连续图像之间重复覆盖的校正。我们提出了一个统一的三分支深度学习框架，通过 WildSeg、WildMatch 和 WildDet 整合这些任务。WildSeg 结合了杆段分割与几何精细化，从飞机安装的标记中划定计数带。WildMatch 利用师生蒸馏学习密集图像对应关系，并在不依赖地理位置元数据的情况下识别重叠的调查区域。WildDet 结合了双尺度检测、特征融合、密度监督和空图像门控，以支持动物检测和计数。我们使用来自南苏丹大尼罗河迁徙景观的四种羚羊物种的斜视航空影像评估了该框架。WildSeg 实现了可靠的杆提取，而 WildMatch 以显著更轻便和更快速的模型接近了教师级别的准确性。两个模块在调整大小的图像上保持了高准确性，提高了处理效率。WildDet 在所有四种物种中提高了检测性能，并相对于 YOLO11 将基于检测的计数平均绝对误差（MAE）、均方根误差（RMSE）和平均偏差分别减少了 21.0%、4.6% 和 27.8%。通过将空间调查约束与动物检测连接起来，该框架为处理大量航空图像集合以进行野生动物监测和种群调查提供了一种实用且计算高效的方法。

View on arXiv Download PDF AI Translation

cs.CV / 234 / 2610.17079

Rethinking LLM-Guided Visual Question Answering: Does Deeper Knowledge Integration Really Help?

重新思考基于大型语言模型的视觉问答：更深的知识整合真的有帮助吗？

Saha, Aneerban, Kumar, Vijay, D., Sameer Madhav, Srivastava, Priyam, Bansal, Dhruv

Abstract

Knowledge-based Visual Question Answering (VQA) requires reasoning that extends beyond visible image content, demanding external commonsense and factual knowledge. While recent work leverages Large Language Models (LLMs) as knowledge sources, the question of how deeply such guidance should be integrated into a vision language architecture remains underexplored. In this work, we generate structured reasoning guidance, visual focus concepts, distractor elements and external knowledge using Llama~3.3~70B and systematically compare two integration strategies within a unified frozen ViT + FLAN-T5 framework on the OK-VQA benchmark: decision level late fusion and attention-matrix injection. Contrary to the prevailing assumption that deeper integration is superior, we find that shallow late fusion achieves the strongest results (39.4% exact match, +10.0 over the no guidance baseline), outperforming deep attention modification (31.8%) under data scarce fine-tuning. We further contribute an LLM based augmentation pipeline that expands training data more than tenfold via question paraphrasing. Our findings indicate that integration depth must be matched to the data regime.

Chinese Translation

基于知识的视觉问答（VQA）需要超越可见图像内容的推理，要求外部常识和事实知识的支持。尽管近期的研究利用大型语言模型（LLMs）作为知识来源，但如何将这种指导深度整合到视觉语言架构中仍然未被充分探讨。在本研究中，我们使用 Llama~3.3~70B 生成结构化的推理指导、视觉聚焦概念、干扰元素和外部知识，并在 OK-VQA 基准上系统比较了统一的冻结 ViT + FLAN-T5 框架下的两种整合策略：决策层的后期融合和注意力矩阵注入。与普遍认为的更深整合优于浅层整合的假设相反，我们发现浅层后期融合达到了最强的结果（39.4% 的精确匹配，较无指导基线提高 10.0%），在数据稀缺的微调情况下优于深层注意力修改（31.8%）。我们进一步贡献了一个基于 LLM 的增强管道，通过问题改写将训练数据扩展了十倍以上。我们的研究结果表明，整合深度必须与数据模式相匹配。

View on arXiv Download PDF AI Translation

cs.CV / 235 / 2610.17080

Scaling Unified Multimodal Models to Long-Horizon Interleaved Generation

扩展统一多模态模型至长时间间隔交错生成

Peng, Ziqiao, Cui, Yutao, Lin, Wang, Li, Zhong-Yu, Gu, Qingwen, Song, Tianhui, Jia, Chengyou, Zhang, Guozhen, Qiu, Xuerui, Wang, Qixun, Yang, Miles, Zhong, Zhao, Tian, Yonglong, Liu, Hongyan, He, Jun

Abstract

Unified multimodal models (UMMs) provide an interface for interleaved text-image generation, where generated visual states can directly inform subsequent reasoning and synthesis. However, existing systems are still primarily demonstrated on short trajectories or simplified visual content. We develop a large UMM natively trained on interleaved multimodal data, with strong capabilities in image synthesis, text rendering, complex composition, and multimodal planning, and study how to extend these native capabilities to substantially longer autoregressive trajectories. The main challenge is that generated images continuously accumulate in context: irrelevant or imperfect visual states interfere with future synthesis, while previously established characters and scenes must still be recovered after long intervals. We introduce a visual-memory framework that maintains persistent visual references, selectively retrieves scene-relevant historical states while preserving the complete semantic history, and uses few-step self-rollout to adapt the model to imperfect generated histories. We further introduce LongMIG-Bench for evaluating long-horizon text-to-multi-image and image-to-multi-image generation. Our model achieves high-fidelity long-horizon interleaved generation with strong character and scene consistency, demonstrating how visual memory can scale native interleaved generation to substantially longer trajectories.

Chinese Translation

统一多模态模型（UMMs）提供了一个用于交错文本-图像生成的接口，其中生成的视觉状态可以直接影响后续的推理和合成。然而，现有系统仍主要在短轨迹或简化的视觉内容上进行演示。我们开发了一个在交错多模态数据上原生训练的大型UMM，具备强大的图像合成、文本渲染、复杂构图和多模态规划能力，并研究如何将这些原生能力扩展到显著更长的自回归轨迹。主要挑战在于生成的图像在上下文中不断累积：不相关或不完美的视觉状态会干扰未来的合成，而之前建立的人物和场景在长时间间隔后仍需恢复。我们引入了一种视觉记忆框架，该框架保持持久的视觉参考，选择性地检索与场景相关的历史状态，同时保留完整的语义历史，并使用少步自回归来适应不完美的生成历史。我们进一步引入了LongMIG-Bench，用于评估长时间间隔的文本到多图像和图像到多图像生成。我们的模型实现了高保真度的长时间间隔交错生成，具备强的人物和场景一致性，展示了视觉记忆如何将原生交错生成扩展到显著更长的轨迹。

View on arXiv Download PDF AI Translation

cs.CV / 236 / 2610.17098

Contrastive learning for video reuse detection

视频重用检测的对比学习

Cann, Tristan J. B., Biswas, Arjun

Abstract

The incorporation of footage from other sources has long been a concern from a copyright perspective, as such usage can harm the interests of rights holders. Recent trends in the ``remixing'' of social media content and the development of bespoke stock libraries for specialist causes mean that methods to identify content reuse will have a range of applications. In this paper, we develop a framework based on a Siamese network fine tuned using contrastive learning to identify potential sources for key frames in a given video. This framework makes two contributions to analysis of large video datasets. Firstly, we demonstrate its utility at reducing complex video data to a set of key visuals typically less than 10\% of the size of the full set. Secondly, we show that these reduced sets allow for efficient and effective identification of component footage. We test this method over a set of videos compiled from a stock footage library and demonstrate its effectiveness for constituent video identification with average accuracy greater than 0.95.

Chinese Translation

从版权的角度来看，使用其他来源的素材长期以来一直是一个关注点，因为这种使用可能会损害权利持有者的利益。社交媒体内容的“重混”趋势以及为特定目的开发定制库存库的兴起意味着识别内容重用的方法将具有广泛的应用。在本文中，我们开发了一个基于对比学习的Siamese网络框架，以识别给定视频中关键帧的潜在来源。该框架对大规模视频数据集的分析做出了两个贡献。首先，我们展示了它在将复杂视频数据简化为通常小于完整数据集10%的关键视觉方面的实用性。其次，我们表明这些简化的数据集可以有效且高效地识别组成素材。我们在从库存素材库编译的一组视频上测试了该方法，并证明其在组成视频识别方面的有效性，平均准确率超过0.95。

View on arXiv Download PDF AI Translation

cs.CV / 237 / 2610.17128

EgoProbe: Benchmarking Long-Range and Exact Memory in Streaming Egocentric Videos

EgoProbe：流媒体自我中心视频中的长时记忆与精确记忆基准测试

Pang, Haokai, Fu, Yuqian, Wang, Yifei, Shen, Yujiao, Qian, Tianwen, Elhoseiny, Mohamed, Fu, Hongbo

Abstract

Egocentric video provides a continuous first-person record of everyday activities, making it a natural foundation for wearable AI assistants. Unlike offline video understanding, streaming systems must process observations as they arrive, while queries may refer to experiences from the distant past. Recent work has begun to explore whether information can be recalled across long temporal gaps. Yet remembering how far is only part of the challenge: recurring objects, actions, and locations also require models to remember exactly which detail belongs to which event. We introduce EgoProbe, a benchmark that evaluates streaming egocentric memory along these two complementary dimensions: long-range retention, i.e., how far, and event-specific recall under interference, i.e., how exact. Built on long-form egocentric videos, EgoProbe uses an event-grounded data engine to extend recall intervals, identify related events, and construct event-grounded distractors, followed by human verification. It contains two subsets: EgoProbe-Base, which evaluates long-range recall without intervening related events, and EgoProbe-Echo, which challenges models with similar intervening events and event-grounded distractors. EgoProbe-Echo adopts coarse-to-fine multi-turn probing: the first turn evaluates coarse-grained event recall, while the second tests whether the model can retrieve the target-specific detail associated with the designated event. Extensive evaluation across diverse models reveals substantial limitations in long-range recall, with further challenges in recovering target-event details among related experiences. Even when models correctly answer coarse-grained questions, they often fail to recall the specific details of the designated event. EgoProbe exposes this underexplored failure mode and provides a testbed for advancing long-range and exact memory in streaming egocentric video understanding.

Chinese Translation

自我中心视频提供了日常活动的连续第一人称记录，为可穿戴人工智能助手奠定了自然基础。与离线视频理解不同，流媒体系统必须在观察到达时进行处理，而查询可能涉及遥远过去的经历。近期的研究开始探讨信息是否可以跨越长时间间隔进行回忆。然而，记住多远只是挑战的一部分：反复出现的物体、动作和地点也要求模型准确记住哪些细节属于哪个事件。我们引入了EgoProbe，一个评估流媒体自我中心记忆的基准，涵盖这两个互补维度：长时保留，即多远，以及在干扰下的事件特定回忆，即多精确。EgoProbe基于长格式自我中心视频，使用事件基础数据引擎来扩展回忆间隔，识别相关事件，并构建事件基础的干扰项，随后进行人工验证。它包含两个子集：EgoProbe-Base，评估在没有干扰相关事件的情况下的长时回忆，以及EgoProbe-Echo，挑战模型处理相似的干扰事件和事件基础的干扰项。EgoProbe-Echo采用粗到细的多轮探测：第一轮评估粗粒度事件回忆，而第二轮测试模型是否能够检索与指定事件相关的目标特定细节。对多种模型的广泛评估揭示了长时回忆中的重大局限性，在恢复相关经历中的目标事件细节时面临进一步挑战。即使模型能够正确回答粗粒度问题，它们通常也无法回忆起指定事件的具体细节。EgoProbe揭示了这一未被充分探索的失败模式，并提供了一个测试平台，以推动流媒体自我中心视频理解中的长时和精确记忆的发展。

View on arXiv Download PDF AI Translation

cs.CV / 238 / 2610.17141

MARS: Systematic Multi-Level Acceleration for World Action Models

MARS：世界动作模型的系统多层加速

Zuo, Zengyuan, Zhang, Yongchang, Zhao, Guangyu, Zhang, Tianze, Wu, Biao, Fang, Meng, Zhong, Yiwu

Abstract

World Action Models (WAMs) have emerged as a promising alternative for embodied intelligence by jointly predicting future world states and robot actions. However, their heavy visual encoding and iterative generative inference incur substantial latency, limiting their applicability to high-frequency closed-loop control. Existing training-free acceleration methods typically optimize a single stage, yielding limited speedups or degraded control under aggressive reuse. In this work, we systematically profile the inference overhead of WAMs and observe abundant redundancy across Transformer computation, diffusion denoising, and consecutive policy queries. However, the safe removal of such redundancy depends strongly on its impact on action prediction and control sensitivity. Based on these observations, we introduce \textbf{MARS}, a training-free and plug-and-play framework for multi-level WAM acceleration. MARS selectively removes low-impact attention computation, adaptively accelerates diffusion inference according to motion granularity, and speculatively reuses predicted future visual representations across consecutive queries with online verification. Together, these complementary mechanisms provide remarkable acceleration while limiting approximation error. Extensive experiments across LIBERO, LIBERO-PRO, and RoboCasa show that MARS achieves up to $4.00\times$, $3.94\times$, and $2.10\times$ end-to-end speedups, respectively, while maintaining task success rate close to the original policy. Moreover, MARS generalizes across different model backbones and across simulation and real-robot experiments, substantially improving the efficiency--performance trade-off of WAM inference. Our code is available at https://github.com/PKU-VaLuE-Lab/MARS.

Chinese Translation

世界动作模型（WAMs）作为一种有前景的体现智能替代方案，通过共同预测未来世界状态和机器人动作而受到关注。然而，它们的重视觉编码和迭代生成推理导致了显著的延迟，限制了其在高频闭环控制中的应用。现有的无训练加速方法通常优化单一阶段，导致速度提升有限或在激进重用下控制性能下降。在本研究中，我们系统性地分析了WAMs的推理开销，并观察到在Transformer计算、扩散去噪和连续策略查询中存在大量冗余。然而，安全去除这些冗余在很大程度上依赖于其对动作预测和控制灵敏度的影响。基于这些观察，我们引入了 extbf{MARS}，一个无训练且即插即用的多层WAM加速框架。MARS选择性地去除低影响的注意力计算，根据运动粒度自适应加速扩散推理，并在连续查询中通过在线验证推测性地重用预测的未来视觉表示。这些互补机制共同提供了显著的加速，同时限制了近似误差。在LIBERO、LIBERO-PRO和RoboCasa上的大量实验表明，MARS分别实现了高达$4.00 imes$、$3.94 imes$和$2.10 imes$的端到端加速，同时保持任务成功率接近原始策略。此外，MARS在不同模型骨干和仿真与真实机器人实验中具有良好的泛化能力，显著改善了WAM推理的效率与性能权衡。我们的代码可在https://github.com/PKU-VaLuE-Lab/MARS获取。

View on arXiv Download PDF AI Translation

cs.CV / 239 / 2610.17201

Multics: Generative Multi-Player World Model with Hierarchical Communication

Multics：具有层次通信的生成多玩家世界模型

Huang, Kaiyi, Yu, Wangbo, Huang, Yukun, Yu, Jiwen, Liu, Kunhao, Yang, Donglin, Hu, Wenbo, Shan, Ying, Liu, Xihui

Abstract

Multi-player interactive world models aim to simulate shared environments from multiple players' perspectives, but achieving robust generation while scaling to more players remains challenging. Existing approaches that rely on fully-connected communication (e.g., Solaris) incur quadratic attention costs as the number of players grows, while centralized approaches (e.g., $\gamma$-World) suffer from interference between unrelated events in a shared global representation. To address these challenges, we present Multics, a multi-player video generation framework with hierarchical communication that structures information exchange across player, group, and global levels. This hierarchy introduces group tokens into self-attention as intermediate group-level representations that aggregate locally relevant player information. These tokens connect to global tokens for cross-group communication, while the hierarchical topology limits direct mixing between unrelated local events. Group-token parameters are shared across group slots, allowing the same communication module to accommodate varying numbers of players without introducing new group-specific parameters. We further propose budget-constrained dynamic grouping based on recent camera trajectories, which adaptively groups relevant players while limiting unnecessary communication. Experiments demonstrate that Multics achieves state-of-the-art performance in 4- and 6-player settings, with improved cross-player consistency, player-specific action and camera controllability, and interference robustness. Moreover, Multics generalizes from 4-player training to 6-, 8-, and 10-player inference without retraining, while reducing self-attention computation compared with fully-connected communication.

Chinese Translation

多玩家互动世界模型旨在从多个玩家的视角模拟共享环境，但在实现稳健生成并扩展到更多玩家时仍然面临挑战。现有方法依赖于完全连接的通信（例如，Solaris），随着玩家数量的增加，导致二次注意力成本，而集中式方法（例如，$ ext{γ}$-World）则在共享全局表示中遭受无关事件之间的干扰。为了解决这些挑战，我们提出了Multics，一种具有层次通信的多玩家视频生成框架，该框架在玩家、组和全局层次上结构化信息交换。这个层次引入了组令牌作为中间组级表示，聚合局部相关的玩家信息。这些令牌与全局令牌连接，以实现跨组通信，而层次拓扑限制了无关局部事件之间的直接混合。组令牌参数在组槽之间共享，使得相同的通信模块能够适应不同数量的玩家，而无需引入新的组特定参数。我们进一步提出基于最近相机轨迹的预算约束动态分组，能够自适应地将相关玩家分组，同时限制不必要的通信。实验表明，Multics在4人和6人设置中实现了最先进的性能，具有改善的跨玩家一致性、玩家特定的动作和相机可控性，以及对干扰的鲁棒性。此外，Multics能够在不重新训练的情况下，从4人训练推广到6人、8人和10人的推理，同时与完全连接通信相比，减少了自注意力计算。

View on arXiv Download PDF AI Translation

cs.CV / 240 / 2610.17227

AReA: Adaptive Receptive-Field Attention for Efficient High-Resolution Diffusion Transformers

AReA：用于高效高分辨率扩散变换器的自适应感受野注意力

Rajabi, Javad, Roohi, Koorosh, Lindell, David B., Taati, Babak

Abstract

Generating images beyond the training resolution of Diffusion Transformers (DiTs) remains challenging, as the number of tokens grows rapidly with resolution, full attention becomes expensive, and outputs often suffer from structural distortions and degraded detail. We identify uniform full attention as the common source of both problems. Attending to all tokens at every step causes quadratic cost and exposes the model to more tokens and relative distances beyond the training range, pushing the model out of distribution. We observe that attention's effective receptive field narrows during denoising, from global in early steps that form structure to local in later steps that refine details. We propose AReA, a training-free attention mechanism whose receptive field evolves from coarse to fine, following the denoising progression. In early steps, AReA applies cross-scale coarse attention over downsampled keys and values, capturing global context, while in later steps it applies local attention within progressively shrinking windows. Without additional parameters or training, experiments show that AReA consistently outperforms state-of-the-art training-free baselines across target resolutions and different models, while also being faster and more efficient.

Chinese Translation

在扩散变换器（DiTs）训练分辨率之外生成图像仍然具有挑战性，因为随着分辨率的提高，标记数量迅速增长，完全注意力的计算成本变得昂贵，输出往往会遭受结构失真和细节退化。我们确定均匀的完全注意力是这两个问题的共同根源。在每一步都关注所有标记会导致二次成本，并使模型暴露于超出训练范围的更多标记和相对距离，从而使模型超出分布。我们观察到，在去噪过程中，注意力的有效感受野逐渐缩小，从早期步骤中的全局结构形成到后期步骤中的局部细节精炼。我们提出了AReA，这是一种无训练的注意力机制，其感受野从粗到细，随着去噪进程而演变。在早期步骤中，AReA在下采样的键和值上应用跨尺度粗略注意力，以捕捉全局上下文，而在后期步骤中则在逐渐缩小的窗口内应用局部注意力。实验表明，AReA在不同目标分辨率和不同模型上，始终优于最先进的无训练基线，同时速度更快，效率更高，无需额外参数或训练。

View on arXiv Download PDF AI Translation

cs.CV / 241 / 2610.17233

Beyond Sparse Captions: Aligning Slide-Level Text and Patch-Level Vision in Pathology

超越稀疏标题：在病理学中对齐幻灯片级文本与补丁级视觉

Shao, Daniel, Bang, Dongmin, Weishaupt, Luca L., Wagner, Sophia J., Reitsam, Nic G., Lu, Ming Y., Le, Long P., Chen, Richard J., Mahmood, Faisal

Abstract

Contrastive vision-language (VL) pretraining in computational pathology has generally relied on curated image-caption datasets sourced from social media, educational videos, and research publications. These corpora are sparse relative to the diversity of tissue morphology and difficult to scale further, as pathologists do not routinely produce descriptive captions in clinical practice. We propose to instead leverage slide-level pathology reports which can be generated during routine diagnostic workflows in a scalable manner. A key challenge in this weakly-supervised setting is that pathology reports describe macroscopic, diagnostic findings while patch-level visual features describe microscopic findings. We introduce Locked Image Multiple Instance PreTraining (LIMIT), which addresses this gap by freezing a pretrained vision encoder and fine-tuning only the text encoder: report embeddings serve as cross-attention queries over bags of patch features, producing report-conditioned slide representations optimized via InfoNCE. Using LIMIT, we establish CONCH-Z, which aligns the text encoder of CONCH v1.5 to 335k clinical reports. Evaluated across 22 tasks spanning patch-level classification, tumor detection, and slide-level subtyping, CONCH-Z establishes state-of-the-art zero-shot performance over VL encoders trained on curated caption datasets, while simultaneously closing the gap with slide-level foundation models trained with substantially more parameters and compute. The code and model is available at https://github.com/mahmoodlab/LIMIT.

Chinese Translation

计算病理学中的对比视觉-语言（VL）预训练通常依赖于从社交媒体、教育视频和研究出版物中获取的经过筛选的图像-标题数据集。这些语料库相对于组织形态的多样性而言是稀疏的，并且难以进一步扩展，因为病理学家在临床实践中并不常规生成描述性标题。我们建议利用在常规诊断工作流程中可以生成的幻灯片级病理报告，这种方式具有可扩展性。在这种弱监督设置中，一个关键挑战是病理报告描述的是宏观的诊断发现，而补丁级视觉特征描述的是微观发现。我们引入了锁定图像多实例预训练（Locked Image Multiple Instance PreTraining, LIMIT），该方法通过冻结预训练的视觉编码器，仅微调文本编码器来解决这一差距：报告嵌入作为补丁特征包的交叉注意力查询，生成通过信息对比（InfoNCE）优化的报告条件幻灯片表示。使用LIMIT，我们建立了CONCH-Z，将CONCH v1.5的文本编码器与335k临床报告对齐。在涵盖补丁级分类、肿瘤检测和幻灯片级亚型分类的22个任务中评估，CONCH-Z在基于经过筛选的标题数据集训练的VL编码器上建立了最先进的零-shot性能，同时缩小了与使用显著更多参数和计算的幻灯片级基础模型之间的差距。代码和模型可在https://github.com/mahmoodlab/LIMIT获取。

View on arXiv Download PDF AI Translation

cs.CV / 242 / 2610.17242

When Left Is Right? Camera-Motion Understanding beyond Screen

何时左为右？超越屏幕的相机运动理解

Cheng, Chong, Zhang, Ran, Li, Xinyuan, Zhao, Alan, Wang, Hao, Wang, Xi

Abstract

Can a vision-language model tell left from right? Modern VLMs solve visual puzzles and follow hour-long videos, yet when asked whether the camera moved left or right, most do no better than a coin flip. Adding camera-motion data or geometric representations has been tried, separately, and neither closes the gap. We argue that camera-motion understanding needs two conditions at once: the representation must couple off-screen camera motion with on-screen flow dynamics, and it must be interpreted by the language model rather than trained as a standalone pose or depth estimator. We propose CameraVLM, which pairs a motion-state representation with state-language alignment. Camera tokens capture how the camera moves off screen, flow tokens capture how the subject and background move on screen, and a two-stage training strategy keeps language in the loop so that both are learned together with camera-motion semantics. With only 9B parameters, CameraVLM outperforms all open-source and camera-motion-specific models by a wide margin across three benchmarks and more than 20 baselines, and matches or surpasses GPT-6 Astra as well as human viewers in a 48-participant study. Project page: https://3dagentworld.github.io/CameraVLM/

Chinese Translation

视觉语言模型能否区分左与右？现代视觉语言模型（VLMs）能够解决视觉难题并跟踪长达数小时的视频，但当被问及相机是向左还是向右移动时，大多数模型的表现与抛硬币无异。尝试分别添加相机运动数据或几何表示，但都未能缩小这一差距。我们认为，相机运动理解需要同时满足两个条件：表示必须将屏幕外的相机运动与屏幕上的流动动态相结合，并且必须由语言模型进行解释，而不是作为独立的姿态或深度估计器进行训练。我们提出了CameraVLM，它将运动状态表示与状态-语言对齐相结合。相机标记捕捉相机在屏幕外的运动，流动标记捕捉主体和背景在屏幕上的运动，而两阶段训练策略确保语言始终在循环中，使得相机运动语义与两者共同学习。CameraVLM仅使用90亿参数，在三个基准测试和超过20个基线模型中，显著超越所有开源和特定于相机运动的模型，并在一项48名参与者的研究中与GPT-6 Astra及人类观众的表现相匹配或超越。项目页面：https://3dagentworld.github.io/CameraVLM/

View on arXiv Download PDF AI Translation

cs.CV / 243 / 2610.17251

HybridDualPathNet: Cross-Fusion Attention of EfficientNetV2 and Swin Transformer for Brain MRI Classification of Alzheimer's and Parkinson's Disease

HybridDualPathNet：高效网络V2与Swin Transformer在阿尔茨海默病和帕金森病脑MRI分类中的交叉融合注意力

Nirjala, Jarpula, Patel, Rahul

Abstract

Separating Alzheimer's Disease (AD), Parkinson's Disease (PD), and Healthy Controls (HC) using structural MRI remains challenging because disease signatures overlap and model performance can be strongly affected by data partitioning. We present HybridDualPathNet, a hybrid architecture combining EfficientNetV2-S for local features and Swin-Tiny for global context through a Cross-Fusion Attention (CFA) module. We evaluate the model in two experiments. First, on the Kaggle APD-3-Class dataset using an image-level split, the model achieves 98.25% accuracy. Second, on a combined Kaggle and Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset containing 17,859 slices, we use a per-source split with patient-wise separation for ADNI subjects. In this setting, the model achieves 73.37% standard accuracy and a macro AUC of 0.880. Source-stratified evaluation reveals a substantial domain gap, with 91.2% accuracy on Kaggle test images and 59.2% on ADNI test images. Temperature scaling reduces expected calibration error from 0.130 to 0.015, while split-conformal prediction achieves 89.5% coverage. Saliency analyses using Grad-CAM++, Integrated Gradients, and occlusion indicate that model activations concentrate on brain tissue. These results highlight the importance of patient-level separation, cross-dataset evaluation, and uncertainty estimation when assessing MRI classification systems. The findings also show that strong performance on a single dataset does not necessarily translate to reliable generalization across imaging cohorts.

Chinese Translation

使用结构性MRI区分阿尔茨海默病（AD）、帕金森病（PD）和健康对照（HC）仍然具有挑战性，因为疾病特征重叠，模型性能可能受到数据划分的强烈影响。我们提出了HybridDualPathNet，这是一种混合架构，结合了EfficientNetV2-S用于局部特征提取和Swin-Tiny用于全局上下文，通过交叉融合注意力（CFA）模块进行整合。我们在两个实验中评估该模型。首先，在Kaggle APD-3-Class数据集上使用图像级划分，模型达到了98.25%的准确率。其次，在包含17,859个切片的Kaggle与阿尔茨海默病神经影像学倡议（ADNI）数据集的组合上，我们采用了按来源划分并对ADNI受试者进行患者级分离。在这种情况下，模型达到了73.37%的标准准确率和0.880的宏观AUC。来源分层评估揭示了显著的领域差距，在Kaggle测试图像上的准确率为91.2%，而在ADNI测试图像上的准确率为59.2%。温度缩放将预期校准误差从0.130降低到0.015，而分裂保守预测实现了89.5%的覆盖率。使用Grad-CAM++、集成梯度和遮挡的显著性分析表明，模型激活集中在脑组织上。这些结果强调了在评估MRI分类系统时，患者级分离、跨数据集评估和不确定性估计的重要性。研究结果还表明，在单一数据集上的强大表现并不一定能转化为在成像队列中的可靠泛化。

View on arXiv Download PDF AI Translation

cs.CV / 244 / 2610.17323

SSM-Attack: Learning Where to Perturb for Black-Box Face Privacy

SSM-Attack：学习在黑箱人脸隐私中进行扰动的位置

Parlayıcı, Tunahan, Aptoula, Erchan, Saygın, Yücel

Abstract

Face recognition systems are deployed widely enough that a photograph shared online can be matched without its subject's consent. Generator-based protection methods rely on heavy pretrained backbones, multi-stage inference or visible semantic edits. We introduce SSM-Attack, a black-box face privacy framework built on Selective State Space Models. Rather than synthesizing the perturbation, the state-space model predicts in one linear-time pass a spatial allocation of the perturbation budget, marking which facial locations carry identity evidence; a short projected refinement then sets the values within this per-pixel budget. Against published benchmark results, SSM-Attack attains the highest average black-box attack success rate on all three evaluation sets: 87.10 percent on CelebA-HQ, 86.90 on LADN and 81.15 on FFHQ, the last from a network that never saw that dataset. The allocation network holds 5.73 million parameters, or 16.13 million with the frozen networks used by its quality terms. Under the same cap of 0.10 on the maximum per-pixel change, allocation improves structural similarity by 0.044, peak signal-to-noise ratio by 1.62 decibels and Frechet Inception Distance by 11.2 points over a uniform map while spending 37 percent of its capacity. Strong facial privacy protection therefore does not require a large generative prior; it requires knowing where to spend a small one.

Chinese Translation

人脸识别系统的广泛部署使得在线分享的照片可以在没有被拍摄者同意的情况下被匹配。基于生成器的保护方法依赖于重型预训练骨干网络、多阶段推理或可见的语义编辑。我们提出了SSM-Attack，这是一种基于选择性状态空间模型的黑箱人脸隐私框架。该框架并不合成扰动，而是通过状态空间模型在一次线性时间的传递中预测扰动预算的空间分配，标记出哪些面部位置携带身份证据；随后进行短暂的投影细化，以设定每个像素预算内的值。与已发布的基准结果相比，SSM-Attack在所有三个评估集上达到了最高的平均黑箱攻击成功率：在CelebA-HQ上为87.10%，在LADN上为86.90%，在FFHQ上为81.15%，后者来自于一个从未见过该数据集的网络。分配网络拥有573万个参数，或与其质量项使用的冻结网络一起达到1613万个参数。在每个像素最大变化限制为0.10的情况下，分配相较于均匀映射提高了0.044的结构相似性，提升了1.62分贝的峰值信噪比，以及11.2分的Frechet Inception Distance，同时仅使用了37%的容量。因此，强大的人脸隐私保护并不需要大量的生成先验；而是需要知道在哪里花费少量的预算。

View on arXiv Download PDF AI Translation

cs.CV / 245 / 2610.17348

HairFlow: Modeling 3D Hair Strands with Multi-Modal Conditioned Diffusion

HairFlow：基于多模态条件扩散的3D发丝建模

Wang, Yating, Wu, Kui, Yi, Ran, Lu, Zixuan, Wang, Tongtong, Zheng, Zhongtian, Ma, Lizhuang, Shen, Yuefan

Abstract

Modeling 3D hair strands is challenging due to complex hairstyle geometry, scarce high-quality training data, and incomplete or noisy input observations. Foundation models offer broader hairstyle priors, but their outputs are not directly usable for strand-based rendering and simulation. Existing approaches require costly per-instance optimization to fit hair priors to meshes from foundation models and 2D orientations from specialized predictors trained on limited hair data, restricting efficiency and generalization. We present HairFlow, a multimodal conditioned diffusion framework for efficient and controllable hair strands modeling. HairFlow conditions strand generation on two complementary representations: a coarse mesh that constrains global shape and multi-view sketches that specify local strand flow. HairFlow combines an Orientation VAE with a conditioned diffusion transformer to generate volumetric fields that are traced into explicit strands. To overcome the lack of paired sketch-strand training data, we introduce an automatic synthesis pipeline that generates geometrically aligned multi-view sketches with diverse styles representative of inference-time inputs. Experiments show that HairFlow generates plausible 3D strands that faithfully follow the input conditions across diverse hairstyles, substantially reduces inference time, and supports flexible multimodal control.

Chinese Translation

由于复杂的发型几何形状、稀缺的高质量训练数据以及不完整或噪声输入观测，3D发丝建模面临挑战。基础模型提供了更广泛的发型先验，但其输出并不能直接用于基于发丝的渲染和模拟。现有方法需要昂贵的每实例优化，以将发丝先验拟合到来自基础模型的网格和由在有限发数据上训练的专门预测器生成的2D方向，这限制了效率和泛化能力。我们提出了HairFlow，这是一种用于高效和可控发丝建模的多模态条件扩散框架。HairFlow在两个互补表示上对发丝生成进行条件约束：一个粗糙的网格限制全局形状，以及多视角草图指定局部发丝流动。HairFlow结合了方向变分自编码器（Orientation VAE）和条件扩散变换器，生成体积场并追踪成明确的发丝。为了克服缺乏配对草图-发丝训练数据的问题，我们引入了一种自动合成管道，生成几何对齐的多视角草图，具有多样化的风格，代表推理时的输入。实验表明，HairFlow生成的3D发丝可信地遵循输入条件，适用于多种发型，显著减少推理时间，并支持灵活的多模态控制。

View on arXiv Download PDF AI Translation

cs.CV / 246 / 2610.17354

MID: Mask-Image Distributional Divergence for Evaluating Medical Image Segmentation

MID：用于评估医学图像分割的掩模-图像分布差异

Marcianò, Vincenzo, Zhang, Xiaoming, Guglielmo, Gianluca, Ourselin, Sebastien, Antonelli, Michela, Zuluaga, Maria A.

Abstract

Current evaluation protocols for medical image segmentation remain largely anchored in sample-wise comparisons, e.g., Dice and Hausdorff distance, which require each prediction to be paired with a ground-truth mask. Yet two limitations arise in practice: they are inapplicable when predictions are made on unlabeled data; and even when annotations are available, they cannot assess whether a collection of predictions matches the distributional structure of the ground-truth population. We introduce MID (Mask-Image Distributional Divergence), a distribution-level evaluation metric for medical image segmentation. MID builds upon the observation that segmentation quality is a property of the joint image-mask distribution: a mask may be anatomically plausible in isolation yet incorrect for the image it accompanies. MID compares the joint distribution of images and masks between a reference set and a set of predictions, without requiring sample-wise correspondence. This captures both mask realism and image-mask consistency. We evaluate MID across 13 publicly available datasets spanning CT, MRI, and MRA. Under anatomically realistic corruptions across 56 organ groups, MID tracks mean Dice and Hausdorff distance without per-sample ground-truth pairing. MID also detects distributional shifts across modalities and anatomies, identifies image-mask misalignment through controlled shuffling experiments, and produces reliable model rankings across real segmentation models. Official implementation available at github.com/erc-caravel/pytorch-mid

Chinese Translation

当前医学图像分割的评估协议主要依赖于样本级比较，例如 Dice 和 Hausdorff 距离，这要求每个预测都与真实掩模配对。然而，在实际应用中出现了两个限制：当对未标记数据进行预测时，这些方法不适用；即使在有注释的情况下，它们也无法评估一组预测是否与真实人群的分布结构相匹配。我们提出了 MID（掩模-图像分布差异），这是一种用于医学图像分割的分布级评估指标。MID 基于以下观察：分割质量是联合图像-掩模分布的一个属性：掩模在孤立状态下可能在解剖上是合理的，但对于其所伴随的图像却可能是错误的。MID 比较参考集和预测集之间的图像和掩模的联合分布，而不需要样本级的对应关系。这既捕捉了掩模的真实性，也捕捉了图像-掩模的一致性。我们在涵盖 CT、MRI 和 MRA 的 13 个公开数据集上评估了 MID。在 56 个器官组的解剖上现实的损坏下，MID 跟踪平均 Dice 和 Hausdorff 距离，而无需逐样本的真实配对。MID 还能够检测跨模态和解剖的分布变化，通过控制洗牌实验识别图像-掩模的不对齐，并在真实分割模型中产生可靠的模型排名。官方实现可在 github.com/erc-caravel/pytorch-mid 获取。

View on arXiv Download PDF AI Translation

cs.CV / 247 / 2610.17355

WIRE: Weight Space Implicit Representation Encoding of Frozen Video LLMs

WIRE：冻结视频语言模型的权重空间隐式表示编码

Ali, Ameen, Zoabi, Tamim, Wolf, Lior, Baskin, Chaim

Abstract

Video language models represent a video as a long sequence of visual embeddings in a decoder's input. Every visual embedding is processed by every decoder layer during prefill, stored in the key value cache, and attended by every generated token, so a long video raises prefill, memory, and generation cost at once. We present WIRE (Weight Space Implicit Representation Encoding), which reduces this recurring cost by moving part of the video conditioned computation of a frozen video language model from its input activations into a temporary low rank update of its weights. WIRE keeps a short explicit visual context, so the update acts as a correction rather than a replacement of the video. WIRE first compresses the full resolution embeddings of each frame into a small set of carrier tokens using a question conditioned bidirectional selective scan, and physically contracts the decoder input so every decoder layer, cache entry, and decoding step operates on the shorter sequence. Hypernetworks at selected decoder layers read the compressed forward pass together with a readout of the full resolution features removed by contraction and generate LoRA updates to the attention and feedforward projections. The updates are merged after prefill, so autoregressive decoding runs the unmodified frozen backbone. On Qwen3-VL-8B, WIRE with 16 carrier tokens per frame contracts the visual stream by $3.5\times$, reduces decoder prefill time by $3.35\times$ at 256 frames and $4.14\times$ at 1024 frames, shrinks the key value cache proportionally, and keeps per token decoding latency approximately constant. Across nine benchmark settings at 256 frames, it obtains the highest average accuracy among efficient video language models we evaluate under a common 256 frame budget.

Chinese Translation

视频语言模型将视频表示为解码器输入中的一长串视觉嵌入。在预填充过程中，每个视觉嵌入都由每个解码器层处理，存储在键值缓存中，并被每个生成的标记所关注，因此长视频同时增加了预填充、内存和生成成本。我们提出了WIRE（权重空间隐式表示编码），通过将冻结视频语言模型的一部分视频条件计算从其输入激活转移到其权重的临时低秩更新中，从而降低这种重复成本。WIRE保持了一个简短的显式视觉上下文，因此该更新充当了视频的修正，而不是替代。WIRE首先使用基于问题的双向选择性扫描将每帧的全分辨率嵌入压缩为一小组载体标记，并物理上收缩解码器输入，以便每个解码器层、缓存条目和解码步骤都在较短的序列上操作。选定解码器层的超网络读取压缩的前向传递以及通过收缩去除的全分辨率特征的读出，并生成对注意力和前馈投影的LoRA更新。更新在预填充后合并，因此自回归解码运行未修改的冻结主干。在Qwen3-VL-8B上，使用每帧16个载体标记的WIRE将视觉流收缩了$3.5 imes$，在256帧时将解码器预填充时间减少了$3.35 imes$，在1024帧时减少了$4.14 imes$，并按比例缩小了键值缓存，同时保持每个标记的解码延迟大致不变。在256帧的九个基准设置中，它在我们评估的高效视频语言模型中获得了最高的平均准确率，且在共同的256帧预算下表现优异。

View on arXiv Download PDF AI Translation

cs.CV / 248 / 2610.17356

Brain2Face: A Study on the Relation of EEG Signals and Spatio-temporal Head Avatars

Brain2Face：脑电信号与时空头部虚拟形象关系的研究

Martínez, Camilo Andrés Martínez, Dünkel, Olaf, Teotia, Kartik, Rekrut, Maurice, Theobalt, Christian, Habermann, Marc

Abstract

Driving digital head avatars is classically done by hand-crafted artist design or by image-based facial performance capture. In this work, we ask whether spatio-temporal tracking of the humans's head geometry and appearance can be performed solely from brain signals, i.e., Electroencephalography (EEG). Since prior work mostly focused on activity and emotion recognition, this task is relatively unexplored without even having publicly available high-quality datasets and benchmarks. Thus, we contribute a dataset featuring paired EEG and human head geometry as well as appearance annotations. We synchronize 64-channel EEG with per-frame expression latent codes of a personalized Gaussian head avatar. Second, we introduce a baseline to the task of expression recovery of a head avatar from EEG signals. To this end, we propose a dedicated EEG-to-expression model that takes a temporal sequence of EEG signals as input and regresses the latent expression parameters of the 3D Gaussian head avatar. Lastly, we validate our capture and algorithmic design choices through a benchmark protocol that we introduce to this task. We believe this work paves the way towards better understanding of the relation between EEG signals and facial surface as well as appearance changes.

Chinese Translation

数字头部虚拟形象的驱动传统上是通过手工设计或基于图像的面部表情捕捉来实现的。在本研究中，我们探讨是否可以仅通过脑信号（即脑电图，Electroencephalography, EEG）来进行人头几何形状和外观的时空跟踪。由于之前的研究主要集中在活动和情感识别上，这一任务相对较少被探索，甚至没有公开的高质量数据集和基准可用。因此，我们贡献了一个数据集，包含配对的EEG和人头几何形状以及外观注释。我们将64通道EEG与个性化高斯头部虚拟形象的每帧表情潜在代码进行同步。其次，我们为从EEG信号恢复头部虚拟形象表情的任务引入了一个基线。为此，我们提出了一个专门的EEG到表情模型，该模型将EEG信号的时间序列作为输入，并回归3D高斯头部虚拟形象的潜在表情参数。最后，我们通过为这一任务引入的基准协议验证了我们的捕捉和算法设计选择。我们相信这项工作为更好地理解EEG信号与面部表面及外观变化之间的关系铺平了道路。

View on arXiv Download PDF AI Translation

cs.CV / 249 / 2610.17412

Bottlenecked semantic compression by representation reconstruction

通过表示重构进行瓶颈语义压缩

Tang, Xiaoya, Liu, Shilong, Smolyak, Daniel, Li, Xiwen, Nguyen, Quynh, Tasdizen, Tolga

Abstract

Vision Transformers incur high computational costs when processing long visual token sequences. Existing reduction methods typically rank tokens independently and discard those deemed unimportant or redundant. We propose a reconstruction-guided framework for visual token condensation (ReConV), which summarizes the complete visual sequence with a compact Summary Bank (SMB) and retains tokens whose information is not adequately captured by the SMB, as indicated by their reconstruction residuals. At each compression layer, a lightweight position-conditioned decoder reconstructs the current tokens, leaving shared information compressed in the SMB while explicitly preserving poorly reconstructed details. Experiments on image classification and multimodal understanding demonstrate favorable performance-efficiency trade-offs: ReConV surpasses the uncompressed baseline with 48\% fewer GMACs on ImageNet and enables reusable image-level representations for efficient multi-turn inference. Code is available at TokenDrop.

Chinese Translation

视觉变换器在处理长视觉标记序列时会产生高计算成本。现有的减少方法通常独立地对标记进行排序，并丢弃那些被认为不重要或冗余的标记。我们提出了一种基于重构引导的视觉标记浓缩框架（ReConV），该框架通过紧凑的摘要库（Summary Bank, SMB）总结完整的视觉序列，并保留那些信息未被SMB充分捕获的标记，这些标记的重构残差表明了其重要性。在每个压缩层中，一个轻量级的位置条件解码器重构当前标记，同时将共享信息压缩在SMB中，并明确保留重构效果较差的细节。在图像分类和多模态理解的实验中，ReConV展示了良好的性能效率权衡：在ImageNet上，ReConV以48%的GMACs减少超越了未压缩基线，并为高效的多轮推理提供了可重用的图像级表示。代码可在TokenDrop获取。

View on arXiv Download PDF AI Translation

cs.CV / 250 / 2610.17434

World-T3: Test-Time Training as Persistent Memory for Real-Time Video World Models

World-T3：作为持久记忆的测试时训练用于实时视频世界模型

Wu, Junyi, Li, Zekun, Yang, Chenhongyi, Guo, Chen, Yang, Lingchen, Zhang, Yongan, Zhang, Xinyi, Hu, Hongbing, Rymkowski, Bartek, Sridhar, Srinath, Yan, Yan, Huang, Xijie

Abstract

Recent advances in video world models have enabled plausible interactive environments, yet long-horizon generation places a fundamental demand on visual memory: the previously observed world must remain consistent as the model continues to generate. Existing autoregressive approaches often implicitly rely on the generative model itself to retain such information, leading to spatial memory decay beyond the active context. We introduce World-T3, a camera-controlled video world model with a test-time training memory that consolidates past observations into dedicated model parameters. To robustly update this memory, we propose a recursive least-squares rule that incrementally absorbs novel visual features via online pose-keyed memory updates. Our TTT memory reads out previously observed scene features from the fast-weight state by camera pose, enabling consistent generation with constant storage and per-step computation regardless of rollout length. To facilitate real-time interactive generation with this robust memory, we further distill the autoregressive model into a few-step generator via dual-teacher distribution matching. On a long-horizon revisit benchmark, World-T3 raises revisit PSNR from 13.52 to 18.31 dB and reduces absolute rotation error from 3.344 to 1.075 relative to the strongest baseline on both metrics, while sustaining 20.6 FPS at 1280 x 704 on a single NVIDIA B200 GPU.

Chinese Translation

最近在视频世界模型方面的进展使得可信的互动环境成为可能，但长时间生成对视觉记忆提出了基本要求：先前观察到的世界必须在模型继续生成时保持一致。现有的自回归方法通常隐含地依赖于生成模型本身来保留此类信息，导致超出活动上下文的空间记忆衰减。我们提出了World-T3，一种受相机控制的视频世界模型，具有测试时训练记忆，将过去的观察整合为专用模型参数。为了稳健地更新该记忆，我们提出了一种递归最小二乘法则，通过在线姿态关键记忆更新逐步吸收新颖的视觉特征。我们的TTT记忆通过相机姿态从快速权重状态中读取先前观察到的场景特征，使得生成过程在存储和每步计算上保持一致，无论展开长度如何。为了利用这种稳健的记忆实现实时互动生成，我们进一步通过双教师分布匹配将自回归模型提炼为几步生成器。在一个长时间重访基准测试中，World-T3将重访PSNR从13.52提高到18.31 dB，并将绝对旋转误差从3.344降低到1.075，相较于在这两个指标上最强的基线，同时在单个NVIDIA B200 GPU上以1280 x 704的分辨率维持20.6 FPS。

View on arXiv Download PDF AI Translation

cs.CV / 251 / 2610.17440

Know-Thy-Not: Training-Free Targeted Negation in Dual-Encoder Vision-Language Models

Know-Thy-Not：双编码视觉-语言模型中的无训练目标否定

Kim, DoYoung, Choi, Changkyu, Rivera, Adín Ramírez, Choi, YunSeok, Lee, Jee-Hyong

Abstract

Dual-encoder vision-language models systematically fail at negation, scoring ''a cat without a dog'' nearly identically to ''a cat with a dog.'' We show that this failure is inherent to the single-embedding framework. Under contrastive training, single-embedding methods can at best ignore the negated concept (dilution) but never penalize its presence (rejection). We propose Know-Thy-Not (KTN), a training-free method that introduces an image-side rejection dimension to penalize the negated concept directly in the visual representation. We also introduce COCO-Confusion, the first benchmark requiring genuine rejection to succeed. Across 9 backbones, KTN improves COCO-Confusion mAP by up to +25.6 over vanilla VLMs and +9.6 over the strongest existing method.

Chinese Translation

双编码视觉-语言模型在否定方面系统性地失败，将“没有狗的猫”与“有狗的猫”几乎同等对待。我们表明，这种失败是单嵌入框架固有的。在对比训练下，单嵌入方法最多只能忽略被否定的概念（稀释），但从未对其存在进行惩罚（拒绝）。我们提出了Know-Thy-Not（KTN），这是一种无训练的方法，通过引入图像侧拒绝维度，直接在视觉表示中惩罚被否定的概念。我们还引入了COCO-Confusion，这是第一个需要真正拒绝才能成功的基准。在9个主干网络中，KTN在COCO-Confusion的平均精度（mAP）上比传统的视觉-语言模型（VLMs）提高了最高25.6，并且比现有最强方法提高了9.6。

View on arXiv Download PDF AI Translation

cs.CV / 252 / 2610.17448

TableVision: Recovering Verified Structure from Historical Tables

TableVision：从历史表格中恢复经过验证的结构

Zhou, Todd Y., Snyder Jr, James M.

Abstract

Gaps, skips, and mistakes are the normal state of a historical register: rows go missing, tables cross page breaks, columns slip under wrong names, and column sums disagree with the corresponding row entries. Readers read digits correctly but lose the table structure. On seven clean panels of election returns, the best open-source reader reproduces 99.0\% of the digits but files only 83.7\% of the figures in the correct cells and gets 11 of 30 columns correct. TableVision treats table structure recovery as the problem. On three degraded 1888 pages, Amazon Textract files only 1 of 1494 figures correctly; TableVision correctly files 99.2\%, and all 51 columns its exact-total check releases are correct. A dynamic program cuts a volume into blocks by how many columns sum to their last line; witnesses read the cells; the gate classifies each column as exactly reconciling, flagged as faithful to a page whose arithmetic is wrong, or withheld. A theory of the sum check says which errors it cannot see and why its optimal block is about $\theta^{-1/2}$ rows at defect rate $\theta$. On 2201 columns of Wisconsin electoral data, the gate accepts 2042, flags 118, and withholds 41; failure rises with column length $R$ as $1-(1-\theta)^R$ at 1.1 defects per 1000 cells in general elections; official returns confirm 933 of 934 testable accepted sums and all 35 testable flagged totals; and the deployment added 491\,359 gated precinct rows to a 9.3-million-row dataset. One implication is that the design of the verification block is critical.

Chinese Translation

缺失、跳过和错误是历史登记的常态：行缺失、表格跨页、列名错误、列总和与相应行条目不一致。读者能够正确读取数字，但失去了表格结构。在七个干净的选举结果面板上，最佳的开源阅读器重现了99.0 ext{%}的数字，但仅将83.7 ext{%}的数字放入正确的单元格，并且仅正确识别了30列中的11列。TableVision将表格结构恢复视为问题。在三页退化的1888年文档中，Amazon Textract仅正确识别1494个数字中的1个；而TableVision正确识别了99.2 ext{%}，并且其精确总和检查释放的51列全部正确。一个动态程序根据有多少列的和等于其最后一行将一个卷切分为块；见证者读取单元格；门控将每列分类为完全调和、标记为与错误算术的页面一致或被保留。总和检查的理论说明了它无法看到哪些错误以及为什么其最佳块的缺陷率约为$ heta^{-1/2}$行，缺陷率为$ heta$。在2201列威斯康星州选举数据中，门控接受了2042列，标记了118列，保留了41列；随着列长度$R$的增加，失败率为$1-(1- heta)^R$，在一般选举中每1000个单元格中有1.1个缺陷；官方结果确认了934个可测试的接受总和中的933个，以及所有35个可测试的标记总和；该部署为一个930万行的数据集增加了491,359个门控选区行。一个启示是验证块的设计至关重要。

View on arXiv Download PDF AI Translation

cs.CV / 253 / 2610.17466

Efficient Fr\'echet Distributional Decoder Alignment for Latent Generative Models

潜在生成模型的高效 Fréchet 分布解码器对齐

Fang, Xianghong, Zhu, Jiajun, Aimeur, Esma, Kong, Dehan, Rudner, Tim G. J.

Abstract

Fr\'echet Distributional (FD) objectives provide an effective distribution-level signal for post-training generative models. For latent generative models, however, existing FD-based post-training methods optimize only the latent generator while keeping the decoder frozen: the decoder is trained on reconstruction latents obtained from real images but deployed on generator-produced latents at generation time. When these latent distributions differ, this \textbf{decoder train-generation mismatch} means that the decoder is deployed on inputs drawn from a distribution it was not directly trained on. We introduce \textbf{Fr\'echet Distributional Decoder Alignment (FDDA)}, an efficient post-training method that freezes the latent generator and directly optimizes the decoder on generation-time latents using an FD objective. A single epoch of FDDA consistently improves gFID and gFDr$^6$ across diverse flow-based and autoregressive latent generative models. Across all 11 evaluated models, FDDA reduces gFDr$^6$ by 34.0-59.6\% from the original pretrained models and provides a further 10.2-57.2\% reduction after generator-side FD post-training. FDDA also reduces total wall-clock time by factors of $3.6\times$-$328.7\times$ relative to generator-side FD post-training across these models. These results show that generator-side post-training alone does not exhaust the benefits of FD optimization, and that further gains can be obtained by adapting the decoder at substantially lower cost.

Chinese Translation

Fréchet 分布（FD）目标为后训练生成模型提供了有效的分布级信号。然而，对于潜在生成模型，现有的基于 FD 的后训练方法仅优化潜在生成器，同时保持解码器不变：解码器在从真实图像获得的重建潜在变量上进行训练，但在生成时却在生成器生成的潜在变量上进行部署。当这些潜在分布不同时，这种 extbf{解码器训练-生成不匹配} 意味着解码器在未直接训练的分布上进行输入处理。我们提出了 extbf{Fréchet 分布解码器对齐（FDDA）}，这是一种高效的后训练方法，它冻结潜在生成器，并使用 FD 目标直接优化解码器在生成时的潜在变量。FDDA 的单个训练周期在多种基于流的和自回归的潜在生成模型中一致地改善了 gFID 和 gFDr$^6$。在评估的 11 个模型中，FDDA 将 gFDr$^6$ 从原始预训练模型减少了 34.0-59.6\%，并在生成器侧 FD 后训练后进一步减少了 10.2-57.2\\%。相较于生成器侧的 FD 后训练，FDDA 还将总的时钟时间减少了 $3.6 imes$-$328.7 imes$。这些结果表明，仅依靠生成器侧的后训练并未充分利用 FD 优化的好处，通过以显著更低的成本调整解码器可以获得进一步的收益。

View on arXiv Download PDF AI Translation

cs.CV / 254 / 2610.17513

Gaussian Splatting for Real-Time Scene Reconstruction: Methods, Systems, and Trade-offs

实时场景重建的高斯点云技术：方法、系统与权衡

Dahal, Ashim, Abdelfattah, Rabab, Rahimi, Nick

Abstract

3D Gaussian Splatting (3DGS) allows for high-quality real-time novel-view synthesis. Since its inception in 2023, its explicit and highly configurable scene representation has been extended toward sparse-view generation, dynamic reconstruction, efficiency and compression, streaming, and simultaneous localization and mapping (SLAM). Although Gaussian Splatting can provide real-time rendering, rendering an existing representation is distinct from constructing, updating, or maintaining the scene itself. In this survey, we synthesize current literature, categorize them and provide a connection of current approaches and tools towards an end-to-end pipeline. Explicitly, we categorize GS scene reconstruction into five overlapping families: single-shot generation, dynamic representations, efficiency and compression, streaming and immersive systems, and Gaussian Splatting SLAM. We distinguish render-only throughput from task-specific scene generation, update, tracking, mapping, and optimization costs, and combine controlled sparse-view benchmarking with native evaluations to identify cross-family quality, latency, memory, and update-cost trade-offs.

Chinese Translation

3D 高斯点云技术（3D Gaussian Splatting, 3DGS）能够实现高质量的实时新视角合成。自2023年首次提出以来，其明确且高度可配置的场景表示已扩展到稀疏视图生成、动态重建、效率与压缩、流媒体传输以及同时定位与地图构建（SLAM）。尽管高斯点云技术能够提供实时渲染，但渲染现有表示与构建、更新或维护场景本身是不同的。在本次综述中，我们综合了当前文献，对其进行了分类，并提供了当前方法和工具与端到端管道之间的联系。具体而言，我们将高斯点云场景重建分为五个重叠的类别：单次生成、动态表示、效率与压缩、流媒体和沉浸式系统，以及高斯点云 SLAM。我们区分了仅渲染的吞吐量与特定任务的场景生成、更新、跟踪、映射和优化成本，并结合受控的稀疏视图基准测试与原生评估，以识别跨类别的质量、延迟、内存和更新成本权衡。

View on arXiv Download PDF AI Translation

cs.CV / 255 / 2610.17530

TwinPrune: Density-Aware Two-Phase Token Pruning for Vision-Language Models

TwinPrune：面向密度的双阶段令牌修剪方法用于视觉-语言模型

Liu, Kai, Li, Anqi, Li, Junxian, Wang, Zhixin, Chen, Zhikai, Pei, Renjing, Zhang, Yulun

Abstract

We propose a unified spatial-density signal to score token importance for both inference phases: bipartite merge for prefill and KV eviction for decode, yielding measurable gains in each stage. To isolate the two phases, we split benchmarks by scoring logic. Prefill probes (MMBench, POPE, ScienceQA) evaluate only the first decoded token; their accuracy reflects prefill logits argmax before any decode KV eviction. Full-pipeline probes (TextVQA, DocVQA, MM-Vet) assess the complete generated sequence and capture both phases. We identify a critical evaluation pitfall: decode pruning claims validated solely on prefill probes measure prefill quality, not decode performance. Experiments show discarding all visual KV during decode barely affects prefill-probe accuracy but degrades TextVQA and MM-Vet drastically; this insensitivity stems from the scoring rule, not compression intensity.Evaluated on seven model variants (InternVL-3.5, Qwen3-VL, LLaVA-1.5) and three full-pipeline benchmarks, our method achieves a three-way Pareto frontier across compute, memory and accuracy. Density-based prefill compression maintains strong accuracy, scales linearly with visual token count, and delivers 100x faster token selection at high resolution. For decode, density top-k KV eviction consistently outperforms strong baselines H2O, SnapKV and StreamingLLM on TextVQA and MM-Vet across three retention ratios. At 0.125 keep ratio on TextVQA, it surpasses H2O by 9.6 points; InternVL-3.5 models (2B to 38B) maintain a 7 to 10 point advantage. Compression scales favorably with model size: every measured 8B to 38B model transition reduces accuracy loss instead of worsening it. We suggest MM-Vet or DocVQA as mandatory full-pipeline validation for any decode-phase claims. Our code is available at https://github.com/Kai-Liu001/TwinPrune

Chinese Translation

我们提出了一种统一的空间密度信号，用于在两个推理阶段对令牌重要性进行评分：在预填充阶段进行二分合并，在解码阶段进行KV驱逐，从而在每个阶段都实现可测量的增益。为了区分这两个阶段，我们通过评分逻辑对基准进行拆分。预填充探测（MMBench、POPE、ScienceQA）仅评估第一个解码令牌；它们的准确性反映了在任何解码KV驱逐之前的预填充logits argmax。全管道探测（TextVQA、DocVQA、MM-Vet）评估完整生成序列并捕捉两个阶段。我们识别出一个关键的评估陷阱：仅在预填充探测上验证的解码修剪声明测量的是预填充质量，而非解码性能。实验表明，在解码过程中丢弃所有视觉KV几乎不影响预填充探测的准确性，但会严重降低TextVQA和MM-Vet的性能；这种不敏感性源于评分规则，而非压缩强度。在七个模型变体（InternVL-3.5、Qwen3-VL、LLaVA-1.5）和三个全管道基准上进行评估，我们的方法在计算、内存和准确性之间实现了三维帕累托前沿。基于密度的预填充压缩保持了强大的准确性，随着视觉令牌数量线性扩展，并在高分辨率下提供100倍更快的令牌选择。对于解码，基于密度的top-k KV驱逐在TextVQA和MM-Vet上始终优于强基线H2O、SnapKV和StreamingLLM，且在三个保留比率下表现出色。在TextVQA上，0.125的保留比率使其超越H2O 9.6分；InternVL-3.5模型（从2B到38B）维持了7到10分的优势。压缩随着模型大小的增加而有利地扩展：每次测量的8B到38B模型转换都减少了准确性损失，而不是加重。我们建议MM-Vet或DocVQA作为任何解码阶段声明的强制全管道验证。我们的代码可在https://github.com/Kai-Liu001/TwinPrune获取。

View on arXiv Download PDF AI Translation

cs.CV / 256 / 2610.17532

CrossPortrait: Cross-Species Portrait Animation via Adversarial Motion Alignment

CrossPortrait：通过对抗运动对齐实现跨物种肖像动画

Pan, Gezhong, Li, Zhe, Tao, Yunfan, Zhao, Yongheng, Zhou, Shiqi, Zhang, Juyong, Jia, Xiaosong, Zhang, Yanyong, Ye, Shuquan, Zhang, Shaofeng

Abstract

Cross-species portrait animation, which aims to animate a source portrait of species A using a driving video of species B, is particularly challenging due to substantial anatomical differences in facial structures across species. Previous works typically fine-tune human portrait animation models on animal video datasets but often suffer from driving-species leakage. To address this, we propose CrossPortrait, a novel framework for expressive cross-species portrait animation, supporting diverse scenarios including human-to-animal, animal-to-human and cross-species animal-to-animal (e.g. cat-to-dog) animation. Specifically, we decompose the portrait motion into rigid head poses and expression keypoints. To accommodate species-specific facial structures while facilitating cross-species expression alignment, we introduce a species-routed expression encoder that incorporates species-specific LoRAs into a unified expression encoder. Furthermore, we introduce expression-level adversarial domain adaptation (Ex-ADA) to suppress driving-species-specific information in the expression latents, thereby mitigating species leakage and promoting a shared expression space across species. Extensive experiments show that our method enables expressive and identity-preserving portrait animation across diverse species and outperforms other state-of-the-art approaches.

Chinese Translation

跨物种肖像动画旨在利用物种B的驱动视频对物种A的源肖像进行动画处理，由于不同物种之间面部结构的显著解剖差异，这一任务尤其具有挑战性。以往的研究通常在动物视频数据集上微调人类肖像动画模型，但往往会遭遇驱动物种信息泄露的问题。为了解决这一问题，我们提出了CrossPortrait，一个用于富有表现力的跨物种肖像动画的新框架，支持包括人对动物、动物对人以及跨物种动物对动物（例如猫对狗）动画等多种场景。具体而言，我们将肖像运动分解为刚性头部姿势和表情关键点。为了适应物种特定的面部结构，同时促进跨物种表情对齐，我们引入了一种物种路由的表情编码器，该编码器将物种特定的LoRAs整合到一个统一的表情编码器中。此外，我们引入了表情级对抗领域适应（Ex-ADA），以抑制表情潜变量中的驱动物种特定信息，从而减轻物种泄露并促进跨物种的共享表情空间。大量实验表明，我们的方法能够实现跨多种物种的富有表现力且保留身份的肖像动画，并且优于其他最先进的方法。

View on arXiv Download PDF AI Translation

cs.CV / 257 / 2610.17552

FlashVSR+: Efficient Video Super-Resolution via Hybrid Attention and Reward-Yield Distribution Matching

FlashVSR+: 通过混合注意力和奖励-收益分布匹配实现高效视频超分辨率

Yao, Ruchang, Guo, Shi, Zhuang, Junhao, Yao, Mingde, Cai, Xin, Zhan, Jiahao, Fu, Bin, Chen, Xinyuan, Qiao, Yu, Xue, Tianfan

Abstract

Recent video super-resolution (VSR) models achieve efficient one-step generation through diffusion distillation. However, high-resolution inference remains costly, and distilled models are difficult to align with human preferences. These challenges call for improvements in both model architecture and training strategy. We present \textbf{FlashVSR+} to address both limitations. On the architecture side, we introduce hybrid attention based on the strong locality observed in pretrained VSR models, while retaining sparse global interaction for long-range consistency. It achieves up to $2\times$ DiT speedup and 45\% lower peak memory, enabling spatially untiled 4K VSR on a single 24GB RTX 4090. On the training side, we propose Reward-Yield Distribution Matching Distillation (Ry-DMD) for preference-aligned one-step generation. Ry-DMD uses reward-weighted teacher rollouts to incorporate black-box rewards into distribution matching, without backpropagating through the reward model. We further show that Ry-DMD generalizes beyond VSR to T2I generation. Together, FlashVSR+ delivers stronger restoration quality and substantially improves high-resolution efficiency, with consistent gains in user preference. We release our code, models, and training data at \href{https://www.flashvsr.tech}{\textcolor{FlashBlue}{\textbf{www.flashvsr.tech}}}.

Chinese Translation

近期的视频超分辨率（VSR）模型通过扩散蒸馏实现了高效的一步生成。然而，高分辨率推理仍然成本高昂，且蒸馏模型难以与人类偏好对齐。这些挑战要求在模型架构和训练策略上进行改进。我们提出了 extbf{FlashVSR+}以解决这两方面的局限。在架构方面，我们引入了基于预训练VSR模型中观察到的强局部性的混合注意力，同时保留稀疏的全局交互以确保长距离一致性。它实现了高达$2 imes$的DiT加速和45 ext{%}的峰值内存降低，使得在单个24GB RTX 4090上能够进行空间无瓦片的4K VSR。在训练方面，我们提出了奖励-收益分布匹配蒸馏（Ry-DMD），用于偏好对齐的一步生成。Ry-DMD使用奖励加权的教师回放将黑箱奖励纳入分布匹配，而无需通过奖励模型进行反向传播。我们进一步展示了Ry-DMD在VSR之外的T2I生成中的泛化能力。综合来看，FlashVSR+提供了更强的恢复质量，并显著提高了高分辨率效率，同时在用户偏好上持续获得提升。我们将在 extcolor{FlashBlue}{ extbf{www.flashvsr.tech}}发布我们的代码、模型和训练数据。

View on arXiv Download PDF AI Translation

cs.CV / 258 / 2610.17573

Benchmarking Self-Supervised 3D Medical Vision Foundation Models and Transferability Estimation

自监督3D医学视觉基础模型的基准测试与可迁移性评估

Tang, Jiaqi, Wang, Yu, Liu, Yang, Tao, Jing, Zhang, Fandong, Zhen, Zhou, Zhuang, Xiahai, Chen, Qingchao, Yang, Yining, Yu, Yizhou

Abstract

Self-supervised pretraining has expanded the pool of 3D medical foundation models, but published evaluations use different datasets, endpoints and fine-tuning protocols. Models also differ in architecture, pretraining objective and data, while downstream targets range from voxel-level masks to patient-level diagnoses and outcomes. As a result, published findings provide little directly comparable evidence for choosing a model for a new task. Direct comparison of adaptation performance requires fine-tuning candidates on labelled target data, at substantial computational cost. We benchmark 23 candidates across the same 14 segmentation, classification and regression tasks spanning multiple anatomies and modalities. No candidate ranked among the top three across all segmentation tasks or across all classification and regression tasks; a convolutional encoder could outperform a transformer trained with the same objective on one task but underperform it on another. We benchmark transferability estimators against the fine-tuned rankings and find that their agreement varies across dense segmentation, classification and regression. InsightRank uses task-appropriate frozen-feature statistics and architecture-aware pairwise comparisons to rank candidates before fine-tuning. Across the 23-model pool, mean Kendall tau-b was 0.431 for segmentation and 0.352 for classification and regression. The segmentation Top 3 contained a task winner on four of nine tasks; the image-level Top 3 contained none on the five tasks, despite better full-ranking agreement than the matched conventional estimators. A three-model shortlist reduced recorded extraction-plus-training job time by 85.8% across seven segmentation tasks. Together, the benchmark provides a common reference for model selection and InsightRank offers a lower-cost way to prioritize fine-tuning, with shortlist performance that differs by target type.

Chinese Translation

自监督预训练扩展了3D医学基础模型的池，但已发布的评估使用了不同的数据集、评估指标和微调协议。模型在架构、预训练目标和数据方面也存在差异，而下游目标则从体素级掩膜到患者级诊断和结果不等。因此，已发布的研究结果为选择新任务的模型提供的直接可比证据有限。直接比较适应性能需要在标记的目标数据上微调候选模型，且计算成本相当高。我们在相同的14个分割、分类和回归任务上对23个候选模型进行了基准测试，这些任务涵盖了多种解剖结构和模态。没有候选模型在所有分割任务或所有分类和回归任务中排名前列；一个卷积编码器在一个任务上可能优于同一目标训练的变换器，但在另一个任务上表现不佳。我们对可迁移性评估器进行了基准测试，与微调后的排名进行比较，发现它们在密集分割、分类和回归任务中的一致性有所不同。InsightRank使用适合任务的冻结特征统计数据和架构感知的成对比较，在微调之前对候选模型进行排名。在23个模型池中，分割任务的平均Kendall tau-b为0.431，分类和回归任务为0.352。分割任务的前3名中有四个任务的获胜者；而在五个任务的图像级前3名中没有获胜者，尽管与传统评估器相比，整体排名的一致性更好。一个三模型的候选名单使得七个分割任务的提取加训练作业时间减少了85.8%。综上所述，该基准测试为模型选择提供了共同参考，而InsightRank则提供了一种更低成本的微调优先级方法，其候选名单的表现因目标类型而异。

View on arXiv Download PDF AI Translation

cs.CV / 259 / 2610.17591

Physical-Field-Bridged Material Parameter Inference from Multi-View Deformation Videos

基于物理场的多视角变形视频材料参数推断

Shou, Linrui, Li, Jie, Liu, Yu, Huang, Tao

Abstract

Estimating simulation-ready material parameters from visual observations is crucial for deformable-object manipulation but challenging because material properties are only indirectly revealed through physical responses. We introduce physical-field-bridged material inference, a single-pass framework that uses dense physical responses to connect multi-view RGB deformation videos with constitutive-type classification and continuous parameter regression. Unlike inverse-physics methods requiring per-scene reconstruction, online tuning, or simulation-in-the-loop optimization, our model predicts force/contact masks, projected motion-flow fields, and projected stress-response maps, then conditions material inference on these fields to reduce reliance on appearance or category shortcuts. We construct DeformFieldBench, a 3DGS MPM learning environment comprising 5,000 simulations across 100 object shapes and five mechanically informative loading modes, with dense pixel-aligned physical supervision. Our method achieves the best median errors on all four continuous parameters, improves stress-hotspot recall from 0.0762 to 0.5121, and attains 0.8676 SSIM/36.82 dB PSNR in same-simulator replay on unseen objects. Under a protocol-matched GSO evaluation, inference and future-state prediction take approximately 30 seconds, compared with 15-120 minutes for per-scene optimization. The method also yields measurable replay agreement on five real-world sequences. These results support dense physical fields as an effective bridge for fast material inference, while cross-simulator and real-world tests reveal remaining domain and identifiability gaps.

Chinese Translation

从视觉观察中估计适用于仿真的材料参数对于可变形物体的操作至关重要，但由于材料属性仅通过物理响应间接揭示，这一过程具有挑战性。我们提出了一种基于物理场的材料推断方法，这是一种单次处理框架，利用密集的物理响应将多视角RGB变形视频与本构类型分类和连续参数回归相连接。与需要逐场景重建、在线调优或仿真循环优化的逆物理方法不同，我们的模型预测力/接触掩码、投影运动流场和投影应力响应图，然后在这些场上进行材料推断，以减少对外观或类别捷径的依赖。我们构建了DeformFieldBench，这是一个包含5000个仿真、100种物体形状和五种机械信息加载模式的3DGS MPM学习环境，并提供密集的像素对齐物理监督。我们的方法在所有四个连续参数上实现了最佳的中位误差，将应力热点的召回率从0.0762提高到0.5121，并在未见物体的同一仿真器重放中达到了0.8676的结构相似性指数（SSIM）和36.82 dB的峰值信噪比（PSNR）。在协议匹配的GSO评估下，推断和未来状态预测大约需要30秒，而逐场景优化则需要15到120分钟。该方法在五个真实世界序列上也产生了可测量的重放一致性。这些结果支持密集物理场作为快速材料推断的有效桥梁，同时跨仿真器和真实世界的测试揭示了仍然存在的领域和可识别性差距。

View on arXiv Download PDF AI Translation

cs.CV / 260 / 2610.17596

Tanawpamana: Benchmarking Vision-Language Models on Philippine Cultural Heritage Sites

Tanawpamana：在菲律宾文化遗产地上对视觉-语言模型进行基准测试

De Leon, Ivan Yuri, Ouchi, Hiroki, Sakti, Sakriani

Abstract

Vision-language models are increasingly used to identify and describe places, yet their recognition and knowledge of specific cultures remain poorly understood, particularly in underrepresented regions such as the Philippines. This study introduces the Tanawpamana Dataset, a repository of Philippine heritage site images with metadata based on the Talapamana registry of the National Commission for Culture and the Arts of the Philippines. We also introduce the Tanawpamana Benchmark, comprising 4,046 tasks designed to evaluate models' recognition and knowledge of heritage sites. Evaluating seven vision-language models, we find that they struggle more with fine-grained visual discrimination among similar heritage sites than with coarse recognition, and that their geographic knowledge is weaker when inferred from image inputs than from text.

Chinese Translation

视觉-语言模型越来越多地用于识别和描述地点，但它们对特定文化的识别和知识仍然缺乏理解，尤其是在菲律宾等代表性不足的地区。本研究介绍了Tanawpamana数据集，这是一个基于菲律宾国家文化艺术委员会Talapamana登记的菲律宾遗产地图像及其元数据的数据库。我们还介绍了Tanawpamana基准，包括4,046个任务，旨在评估模型对遗产地的识别和知识。在评估七个视觉-语言模型时，我们发现它们在相似遗产地之间的细粒度视觉区分上比粗略识别更为困难，并且从图像输入推断的地理知识比从文本推断的要弱。

View on arXiv Download PDF AI Translation

cs.CV / 261 / 2610.17603

Morphology-aware deep learning predicts colorectal cancer survival from whole-slide images

形态学感知深度学习从全切片图像预测结直肠癌生存率

Khan, Hikmat, Gurcan, Metin N., Parwani, Anil, Frankel, Wendy L., Chen, Wei, Niazi, Muhammad Khalid Khan

Abstract

Background: Five-year survival prediction from hematoxylin-and-eosin whole-slide images could support risk stratification, but the task is weakly supervised, and the tissue morphologies carrying prognostic information remain incompletely characterized. Methods: We conducted a retrospective secondary analysis using TCGA-COAD/READ (n=562) for morphology-specific analysis, Alliance/ CALGB 89803 (n=424) for model development, and SurGen (n=218) for external evaluation. We propose MorphContrast, a morphology-aware contrastive multiple-instance learning framework that uses tissue labels as structural supervision for five-year survival prediction. MorphContrast combines gated-attention pooling with three auxiliary objectives: patch-level morphology labels define within-patient contrasts, outcome concordance or discordance defines cross-patient contrasts within matched morphologies, and diversity regularization discourages representational collapse. Morphology-matched patches from patients with discordant outcomes serve as hard negatives, encouraging prognostic discrimination within comparable histologic contexts. Results: Morphology-specific models show weak but distributed prognostic discrimination: adipose tissue achieves an area under the receiver operating characteristic curve (AUC) of 0.622, stroma achieves 0.604, and both perform comparably to tumor epithelium (0.601-0.605). On Alliance/CALGB 89803, the single model achieves an AUC of 0.642, while the post hoc ensemble reaches a pooled out-of-fold AUC of 0.70 and a risk-group hazard ratio of 3.79 (95% confidence interval: 2.55-5.63). On the external SurGen colon cancer cohort (n=218), the single model and ensemble achieve AUCs of 0.669 and 0.678, respectively. Conclusions: These findings support tissue morphology as structural supervision for outcome-aware representation learning in colorectal cancer prognosis.

Chinese Translation

背景：从苏木精-伊红全切片图像预测五年生存率可以支持风险分层，但该任务是弱监督的，携带预后信息的组织形态尚未完全表征。方法：我们使用TCGA-COAD/READ（n=562）进行形态学特异性分析，使用Alliance/CALGB 89803（n=424）进行模型开发，并使用SurGen（n=218）进行外部评估，进行了回顾性二次分析。我们提出了MorphContrast，这是一种形态学感知的对比多实例学习框架，利用组织标签作为结构性监督来进行五年生存率预测。MorphContrast结合了门控注意力池化与三个辅助目标：病灶级别的形态标签定义患者内对比，结果一致性或不一致性定义匹配形态内的患者间对比，而多样性正则化则抑制表征崩溃。来自结果不一致患者的形态匹配病灶作为困难负样本，促进在可比组织学背景下的预后区分。结果：形态学特异性模型显示出弱但分布式的预后区分：脂肪组织的接收者操作特征曲线下面积（AUC）为0.622，基质为0.604，两者与肿瘤上皮（0.601-0.605）表现相当。在Alliance/CALGB 89803中，单一模型的AUC为0.642，而后验集成达到0.70的汇总外折AUC和3.79的风险组风险比（95%置信区间：2.55-5.63）。在外部SurGen结肠癌队列（n=218）中，单一模型和集成的AUC分别为0.669和0.678。结论：这些发现支持组织形态作为结直肠癌预后中结果感知表示学习的结构性监督。

View on arXiv Download PDF AI Translation

cs.CV / 262 / 2610.17605

PLEDGE: Progressive Lossy Error-Bounded Compression of 3D Gaussian Splatting

PLEDGE：渐进式有损误差界限压缩的3D高斯点云

Xia, Mingze, Li, Yuxiao, Sun, Baixi, Li, Wenbo, Jiao, Pu, Di, Sheng, Cappello, Franck, Guo, Hanqi, Liang, Xin

Abstract

Existing 3D Gaussian Splatting (3DGS) codecs report reconstruction quality after compression, and give users little control over the distortion of individual primitives. We present PLEDGE, a training-free, progressive, error-bounded codec that translates four user-declared tolerances into per-primitive quantization bounds: screen-space position, footprint deformation due to rotation and scale, view-dependent color, and opacity. The bounds apply throughout a declared camera domain, including previously unseen viewpoints. Position and opacity are bounded directly; footprint deformation is bounded under the rasterizer's affine projection model, and color is bounded over all viewing directions. PLEDGE organizes quantized residuals into a progressive bitstream with explicit error bounds for every valid prefix. Experiments across thirteen scenes verify all declared per-mechanism bounds and demonstrate competitive rate-distortion performance. The four tolerances incur substantially different rate costs, so users can spend rate where their application needs it.

Chinese Translation

现有的3D高斯点云（3DGS）编解码器在压缩后报告重建质量，但对单个原始体的失真控制有限。我们提出了PLEDGE，这是一种无训练、渐进式、有误差界限的编解码器，能够将用户声明的四个容忍度转化为每个原始体的量化界限：屏幕空间位置、由于旋转和缩放导致的轮廓变形、视角依赖的颜色和不透明度。这些界限适用于声明的相机域，包括以前未见的视点。位置和不透明度直接受到限制；轮廓变形在光栅化器的仿射投影模型下受到限制，颜色在所有视角方向上受到限制。PLEDGE将量化残差组织成一个渐进式比特流，为每个有效前缀提供明确的误差界限。对十三个场景的实验验证了所有声明的每个机制的界限，并展示了具有竞争力的比特率-失真性能。这四个容忍度产生了显著不同的比特率成本，因此用户可以在其应用需要的地方花费比特率。

View on arXiv Download PDF AI Translation

cs.CV / 263 / 2610.17620

Smartphone-Based, Field-Deployed AI Inspection System for Power Distribution Infrastructure

基于智能手机的现场部署人工智能检查系统用于电力分配基础设施

Masabathula, Sampath Balaji, Vanapalli, Srija, Kanchi, Gangadhar, N, Sreenivasa Rao V. A., Lalitha V, Vijaya, Chilamkurthy, Sasank

Abstract

Routine inspection of power distribution infrastructure remains challenging for utilities operating geographically dispersed networks, where conventional manual inspection can be difficult to perform consistently and at scale. This work presents a smartphone-based AI inspection system for assessing targeted distribution-infrastructure conditions from images captured by field personnel. The proposed workflow integrates image-quality and viewing-perspective validation with AI-assisted assessment of pole lean, crossarm and top-cleat tilt, voltage level, vegetation-clearance risk, and distribution-transformer (DTR)-ground clearance. The system integrates deep-learning models with monocular depth estimation, geometric analysis, and rule-based processing within a common multi-stage inspection workflow designed for practical field operation. The system was pilot deployed across five selected distribution feeders within the Eastern Power Distribution Company of Andhra Pradesh Ltd. (APEPDCL), where 1,222 defect instances were identified across the assessed categories and subsequently tracked through a rectification and verification workflow, with 84 defect instances recorded as rectified during the study period. Feeder-level interruption records were also examined over the study period; however, the available observations do not provide sufficient evidence to establish a reliability improvement attributable to the proposed system. A curated multi-module field-captured dataset, PD-Defect, is also publicly released to support reproducible research in distribution-infrastructure inspection. The field deployment demonstrates the practical integration of smartphone-based AI-assisted inspection into field inspection and maintenance activities, providing a mechanism for systematic identification, documentation, rectification tracking, and verification of targeted distribution-infrastructure defects.

Chinese Translation

对电力分配基础设施的常规检查对于运营地理分散网络的公用事业公司仍然具有挑战性，传统的人工检查在一致性和规模上都难以有效执行。本研究提出了一种基于智能手机的人工智能检查系统，用于评估现场人员拍摄的图像中目标分配基础设施的状况。所提出的工作流程将图像质量和视角验证与人工智能辅助的评估相结合，评估内容包括电杆倾斜、横臂和顶部支架倾斜、电压水平、植被清除风险以及配电变压器（DTR）与地面间的间隙。该系统将深度学习模型与单目深度估计、几何分析和基于规则的处理集成在一个通用的多阶段检查工作流程中，旨在适用于实际的现场操作。该系统在安得拉邦东部电力分配公司（APEPDCL）选择的五个配电馈线中进行了试点部署，在评估的类别中识别出1,222个缺陷实例，并通过整改和验证工作流程进行跟踪，在研究期间记录到84个缺陷实例已被整改。研究期间还检查了馈线级别的中断记录；然而，现有观察结果并未提供足够的证据来证明所提系统带来的可靠性改善。此外，还公开发布了一个经过精心策划的多模块现场捕获数据集PD-Defect，以支持电力分配基础设施检查中的可重复研究。现场部署展示了基于智能手机的人工智能辅助检查在现场检查和维护活动中的实际整合，为系统化识别、记录、整改跟踪和验证目标分配基础设施缺陷提供了机制。

View on arXiv Download PDF AI Translation

cs.CV / 264 / 2610.17625

4LC: 4D Latent Cache for Efficient Conditioning in Dynamic World Modeling

4LC：用于动态世界建模的高效条件生成的4D潜在缓存

Zhang, Jianhui, Lin, Zuyao, Xu, Pengxin, Wang, Jingdong, Chen, Xingyu

Abstract

For video generation in dynamic world modeling, observed static and dynamic content should be efficiently reused to provide conditions for repeated generation at different times and viewpoints. We introduce 4LC, a 4D latent cache with composite visual representations and a flexible spatiotemporal access interface. As a latent cache, 4LC directly reuses visual representations of static and dynamic content to provide generation conditions, avoiding per-request RGB rendering and re-encoding. As a composite cache, 4LC binds visual representations from multiple visual encoders (e.g., VAE, DINO, V-JEPA, VGGT) to shared geometry and motion, allowing the composite representations to describe the dynamic world with rich information. Besides, we design a flexible spatiotemporal interface, which directly reads the cache at independently specified times and viewpoints and combines readouts across observations, providing a common conditioning format for diverse downstream tasks. Building on 4LC, we develop 4LC-Video to address multiple video generation tasks through different configurations of source observations, spatiotemporal requests, and historical reuse, including camera control, temporal control, novel-view synthesis, scene revisits, and long-term generation. Our experiments show more efficient condition acquisition, effective spatiotemporal controllability, and scene consistency across tasks using a single 4LC-Video model without task-specific fine-tuning.

Chinese Translation

在动态世界建模中的视频生成中，观察到的静态和动态内容应有效重用，以便在不同时间和视角下提供重复生成的条件。我们提出了4LC，一种具有复合视觉表示和灵活时空访问接口的4D潜在缓存。作为潜在缓存，4LC直接重用静态和动态内容的视觉表示，以提供生成条件，避免每次请求的RGB渲染和重新编码。作为复合缓存，4LC将来自多个视觉编码器（例如，VAE、DINO、V-JEPA、VGGT）的视觉表示绑定到共享的几何和运动上，使复合表示能够以丰富的信息描述动态世界。此外，我们设计了一个灵活的时空接口，可以在独立指定的时间和视角直接读取缓存，并结合跨观察的读出，为多样的下游任务提供统一的条件格式。在4LC的基础上，我们开发了4LC-Video，通过不同配置的源观察、时空请求和历史重用来解决多个视频生成任务，包括相机控制、时间控制、新视角合成、场景重访和长期生成。我们的实验表明，使用单个4LC-Video模型在没有任务特定微调的情况下，条件获取更高效，时空可控性更强，场景一致性更好。

View on arXiv Download PDF AI Translation

cs.CV / 265 / 2610.17639

FrameLite: Decision-Oriented Aesthetic Cropping with a Compact Vision-Language Model

FrameLite：基于紧凑视觉-语言模型的决策导向美学裁剪

Li, Jianze, Liu, Kai, Zhang, Ziqing, Zhang, Weihang, Guo, Yong, Kong, Linghe, Zhang, Yulun

Abstract

Aesthetic image cropping improves composition by balancing subjects, context, and image boundaries. Large vision-language models bring semantic knowledge to this task, but their scale limits lightweight photography assistance. We propose FrameLite, a compact cropping model built on Qwen3.5-0.8B. Our approach is motivated by the observation that a small model can generate better crops than its first answer suggests. We develop three-stage training to turn this potential into stronger direct predictions. Spatial and composition training establishes localization and framing preferences. Quality-Guided Online Supervision (QOS) then learns the highest-quality complete answers among the current policy's generated candidates. Greedy-Referenced Optimization (GRO) further refines the policy using positive and negative quality differences relative to its current greedy answer. Both online stages use annotation-derived feedback, while inference returns one coordinate answer without search or an external scorer. Experiments show that FrameLite outperforms the compared large-VLM croppers across public benchmarks with a substantially smaller backbone. Controlled studies confirm that selecting high-quality training answers improves the final model beyond online sampling alone. We will release our code and model weights at https://github.com/JianzeLi-114/FrameLite .

Chinese Translation

美学图像裁剪通过平衡主题、背景和图像边界来改善构图。大型视觉-语言模型为这一任务带来了语义知识，但其规模限制了轻量级摄影辅助的应用。我们提出了FrameLite，一个基于Qwen3.5-0.8B构建的紧凑裁剪模型。我们的方法源于观察到小模型能够生成比其首次答案更好的裁剪结果。我们开发了三阶段训练，将这一潜力转化为更强的直接预测。空间和构图训练建立了定位和构图偏好。质量引导在线监督（Quality-Guided Online Supervision, QOS）则学习当前策略生成候选答案中的最高质量完整答案。贪婪参考优化（Greedy-Referenced Optimization, GRO）进一步利用相对于当前贪婪答案的正负质量差异来优化策略。这两个在线阶段都使用基于注释的反馈，而推理阶段则返回一个坐标答案，无需搜索或外部评分器。实验表明，FrameLite在公共基准测试中超越了比较的大型视觉-语言模型裁剪器，且其基础模型显著更小。控制研究确认，选择高质量的训练答案能够改善最终模型的表现，超越仅依赖在线采样的效果。我们将于 https://github.com/JianzeLi-114/FrameLite 发布我们的代码和模型权重。

View on arXiv Download PDF AI Translation

cs.CV / 266 / 2610.17640

MemAct: Streaming Spatial Understanding with Memory and Active Exploration

MemAct：通过记忆和主动探索进行流式空间理解

Wei, Yuxi, Huang, Wei, Chen, Qirui, Hou, Lu, Qi, Xiaojuan

Abstract

Spatial understanding is essential for embodied agents navigating and interacting with the physical world. As observations arrive incrementally, agents must reason in a streaming manner. Yet streaming understanding must also be closed-loop: task-relevant evidence may require both recalling past observations and actively acquiring new views. Existing streaming approaches largely emphasize processing incoming observations, leaving task-driven active exploration underexplored. We introduce MemAct, a framework for streaming spatial understanding with memory and action. A shared vision--language model incrementally writes structured historical memory, retrieves records and past frames on demand, and acquires additional evidence through movement and turning. Retrieved evidence and action observations inform subsequent decisions, enabling the agent to continue exploring or answer within a unified reasoning loop. We develop the MemAct suite, a data generation pipeline and collection across real and simulated environments, with approximately 1M training examples for spatial reasoning, memory writing, retrieval, and exploration. Its MemAct-bench comprises 20,000 main QA pairs evaluating current-observation understanding, historical evidence use, and active exploration, plus 2,000 separate Timing questions. MemAct combines supervised fine-tuning for memory and interaction with counterfactual action optimization using cumulative evidence gains, action costs, and terminal answer quality to improve exploration and stopping decisions. Experiments show that MemAct outperforms existing spatial understanding and streaming baselines and approaches frontier multimodal models on MemAct-bench while retaining streaming inference efficiency.

Chinese Translation

空间理解对于在物理世界中导航和交互的具身智能体至关重要。随着观察数据的逐步到达，智能体必须以流式的方式进行推理。然而，流式理解也必须是闭环的：与任务相关的证据可能需要同时回忆过去的观察和主动获取新的视角。现有的流式方法主要强调处理输入的观察数据，而对任务驱动的主动探索关注不足。我们提出了MemAct，一个结合记忆和行动的流式空间理解框架。一个共享的视觉-语言模型逐步写入结构化的历史记忆，根据需求检索记录和过去的帧，并通过移动和转向获取额外证据。检索到的证据和行动观察为后续决策提供信息，使智能体能够在统一的推理循环中继续探索或回答问题。我们开发了MemAct套件，这是一个数据生成管道和在真实与模拟环境中的数据集合，包含约100万条用于空间推理、记忆写入、检索和探索的训练样本。其MemAct-bench包含20,000对主要的问答对，用于评估当前观察理解、历史证据使用和主动探索，以及2,000个独立的时间问题。MemAct结合了针对记忆和交互的监督微调与使用累积证据增益、行动成本和最终答案质量的反事实行动优化，以改善探索和停止决策。实验表明，MemAct在空间理解和流式基准上优于现有方法，并在MemAct-bench上接近前沿的多模态模型，同时保持流式推理的效率。

View on arXiv Download PDF AI Translation

cs.CV / 267 / 2610.17656

Open-o3-Video-v2: Seek, Ground, and Reason with Verifiable Spatio-Temporal Evidence

Open-o3-Video-v2：寻求、验证与基于可验证时空证据的推理

Meng, Jiahao, Bao, Zhongyuan, Tan, Yue, Wen, Haiquan, Zhang, Renrui, Wang, Anna, Li, Jason, Fu, Yanwei, Tong, Yunhai, Yang, Ming-Hsuan

Abstract

Reliable video reasoning requires models to produce correct answers and expose supporting visual evidence. Open-o3-Video enables verifiable reasoning by embedding timestamps and object-level bounding boxes into reasoning traces, but is restricted to frames from an initial sparse sampler. We introduce Open-o3-Video-v2, which formulates video reasoning as a unified agentic evidencegrounded process. The model identifies missing information, acquires additional observations, grounds the moments and objects supporting its reasoning, and writes this evidence into the trajectory before continuing or answering. To support this behavior, we construct STGR-v2-CoT135K, including 52K multi-turn agentic grounded reasoning trajectories, and STGR-v2-RL-13K for reinforcement learning. We further propose Clue-Seek GRPO (CS-GRPO), which tracks how tool calls change accumulated evidence, rewards clue coverage and step-wise gains, and penalizes excessive, repetitive, or uninformative retrieval. On V-STaR, Open-o3-Video-v2-4B achieves 40.4% mAM and 56.3% mLGM, outperforming its Qwen3-VL-4B backbone by 8.5% and 12.6%, respectively. It also delivers consistent gains across general video understanding, long-video reasoning, and temporal grounding benchmarks. These results show that active evidence acquisition and verifiable object-level grounding are complementary for reliable video reasoning. The code and models are released at https://msalab-pku.github.io/projects/Open-o3-Video-v2.

Chinese Translation

可靠的视频推理要求模型能够产生正确的答案并展示支持的视觉证据。Open-o3-Video通过将时间戳和对象级边界框嵌入推理轨迹中，实现了可验证的推理，但仅限于来自初始稀疏采样器的帧。我们引入了Open-o3-Video-v2，将视频推理表述为一个统一的代理证据基础过程。该模型识别缺失信息，获取额外观察，确定支持其推理的时刻和对象，并在继续推理或回答之前将这些证据写入轨迹。为了支持这种行为，我们构建了STGR-v2-CoT135K，包括52K多轮代理基础推理轨迹，以及用于强化学习的STGR-v2-RL-13K。我们进一步提出了Clue-Seek GRPO (CS-GRPO)，该方法跟踪工具调用如何改变累积证据，奖励线索覆盖和逐步增益，并惩罚过度、重复或无信息的检索。在V-STaR上，Open-o3-Video-v2-4B达到了40.4%的mAM和56.3%的mLGM，分别比其Qwen3-VL-4B基础模型提高了8.5%和12.6%。它在一般视频理解、长视频推理和时间基础基准测试中也展现出一致的提升。这些结果表明，主动证据获取和可验证的对象级基础是可靠视频推理的互补因素。代码和模型已发布在https://msalab-pku.github.io/projects/Open-o3-Video-v2。

View on arXiv Download PDF AI Translation

cs.CV / 268 / 2610.17661

ScenePercept: Scenario-Conditioned Perceptual Assessment of Real-World User-Generated Content Images

ScenePercept：基于场景条件的真实世界用户生成内容图像的感知评估

Xia, Shuhan, Chen, Yu, Wang, Xiang, Li, Shuai, Li, Peipie, Li, Zekun

Abstract

The rapid development of mobile imaging and social platforms has led to explosive growth in user-generated content images, making their systematic and comprehensive perceptual assessment increasingly important. Existing image perception benchmarks are typically task-oriented, curating specific scenarios for quality or aesthetic assessment and applying a uniform perceptual criterion across diverse scenarios. This limits their ability to capture the scene diversity and perceptual complexity of real-world UGC images. To address these limitations, we introduce ScenePercept-Bench, a benchmark designed to reflect the distribution of real-world UGC images, comprising 23,100 images across 21 categories and 105 fine-grained scenes. To account for the influence of scene context on human perception, we design a hierarchical, scene-aware annotation protocol, covering visual quality, aesthetic perception, and scene semantics, with 15 attributes in total. Each image is annotated with an overall score, 15 attribute-level scores, and structured explanatory reasoning. Building on this benchmark, we further introduce Scenario-Conditioned Perceptual Alignment, a training framework that tackles the heterogeneity of perceptual criteria across scenarios. SCPA jointly optimizes absolute score prediction, preference ranking, and cross-scenario calibration, preserving the model general perceptual assessment ability while encouraging it to explore and adapt perceptual criteria across different scenarios. Extensive experiments demonstrate that SCPA achieves state-of-the-art performance in UGC image perception and generalizes effectively to established IQA and IAA benchmarks.

Chinese Translation

移动成像和社交平台的快速发展导致用户生成内容图像的爆炸性增长，使得对这些图像进行系统和全面的感知评估变得愈发重要。现有的图像感知基准通常是以任务为导向，针对特定场景进行质量或美学评估，并在不同场景中应用统一的感知标准。这限制了它们捕捉真实世界用户生成内容图像的场景多样性和感知复杂性的能力。为了解决这些局限性，我们推出了ScenePercept-Bench，这是一个旨在反映真实世界用户生成内容图像分布的基准，包含21个类别和105个细粒度场景的23,100张图像。为了考虑场景上下文对人类感知的影响，我们设计了一种分层的、场景感知的注释协议，涵盖视觉质量、美学感知和场景语义，总共包含15个属性。每张图像都附有一个整体评分、15个属性级评分和结构化的解释推理。在此基准的基础上，我们进一步引入了场景条件感知对齐（Scenario-Conditioned Perceptual Alignment，SCPA），这是一个解决不同场景间感知标准异质性的训练框架。SCPA联合优化绝对评分预测、偏好排序和跨场景校准，保持模型的通用感知评估能力，同时鼓励其探索和适应不同场景中的感知标准。大量实验表明，SCPA在用户生成内容图像感知方面实现了最先进的性能，并有效地推广到已建立的图像质量评估（IQA）和图像美学评估（IAA）基准上。

View on arXiv Download PDF AI Translation

cs.CV / 269 / 2610.17684

Hierarchical Curriculum Learning for Surgical Video Generation: Focusing on Bilateral Axillo-Breast Approach Robotic Thyroidectomy Datasets

用于外科视频生成的层次化课程学习：聚焦双侧腋下-乳房入路机器人甲状腺切除术数据集

Kong, Youngbin, Lee, Seungho, Shin, Jae Kyung, Ahn, Jong-hyuk, Kong, Hyoun-Joong, Lee, Kyu Eun

Abstract

High-fidelity surgical video generation is increasingly important for surgical education, simulation, and downstream AI training, as conventional simulation pipelines are limited by cost, realism, and scenario diversity. Adapting generic text-to-video diffusion models to surgical procedures remains difficult, however, due to complex anatomical structures, intricate tool-tissue interactions, the inherently hierarchical nature of surgical workflows, and the scarcity of curated training data outside a few well-studied procedures. We propose a surgical video generation framework that combines hierarchical procedural annotation with curriculum transfer learning. Surgical procedures are decomposed into three levels (phase, task, and action) and a pretrained text-to-video diffusion model is progressively adapted from coarse to fine-grained surgical context, supported by a dynamic sampling strategy that handles variable-resolution and variable-length surgical clips. We instantiate the framework on Bilateral Axillo-Breast Approach robotic thyroidectomy, a procedure for which no curated benchmark currently exists. Relative to the preceding task-level model, the final action-level model reduces FVD by 20.1% and FID by 15.7%, and a matched-budget ablation indicates that these gains arise from the curriculum design rather than from additional training. A user study with six clinicians rates visual realism highly while highlighting remaining gaps in anatomical accuracy and clinical reliability. The proposed framework offers a step toward AI-driven surgical simulation in procedures where curated data are scarce.

Chinese Translation

高保真外科视频生成在外科教育、模拟和后续的人工智能训练中愈发重要，因为传统的模拟流程受到成本、真实感和场景多样性的限制。然而，由于复杂的解剖结构、精细的工具-组织交互、外科工作流程固有的层次性以及除了少数经过充分研究的程序外，经过整理的训练数据的稀缺，将通用的文本到视频扩散模型适应于外科手术仍然困难。我们提出了一种外科视频生成框架，该框架结合了层次化的程序注释与课程迁移学习。外科程序被分解为三个层次（阶段、任务和动作），并且一个预训练的文本到视频扩散模型从粗到细逐步适应外科上下文，支持一种动态采样策略，以处理可变分辨率和可变长度的外科剪辑。我们在双侧腋下-乳房入路机器人甲状腺切除术上实例化该框架，目前尚无经过整理的基准数据。相较于之前的任务层模型，最终的动作层模型使FVD降低了20.1%，FID降低了15.7%，而匹配预算的消融实验表明，这些增益源于课程设计，而非额外的训练。与六位临床医生的用户研究显示，视觉真实感评分较高，同时突出了在解剖准确性和临床可靠性方面的不足。所提出的框架为在数据稀缺的程序中实现人工智能驱动的外科模拟迈出了重要一步。

View on arXiv Download PDF AI Translation

cs.CV / 270 / 2610.17692

SWORD: Scene-Grounded Wasserstein Object-Relation Distance for 3D Scene Graph Prediction

SWORD：基于场景的Wasserstein对象-关系距离用于3D场景图预测

Jun, U-Chae, Ko, Jaeeun, Choi, Yoonhyuk, Kang, Jiwoo

Abstract

3D scene graph prediction is commonly optimized with object-node and relation-edge classification losses. These objectives encourage exact object and relation label recovery, but they do not directly optimize semantic proximity to the ground-truth scene graph: two wrong graphs can receive comparable penalties even when one is much closer in object roles and relation contexts. A direct graph-distance loss is difficult in general, since graph matching and graph edit distance require correspondence search and edit-cost design. Our key observation is that prior supervised 3D scene graph prediction protocols are conducted on fixed object-node indices and supervised edge slots. This can turn a scene graph comparison into a well-defined aligned graph-distance problem. To this end, we propose SWORD, a Scene-Grounded Wasserstein Object-Relation Distance that builds block-normalized object and relation ground distance metrics from training scene graph annotations through a deterministic base-to-context order, without external semantic priors. SWORD lifts these metrics to Wasserstein distances over aligned node and null-augmented edge marginals, providing semantic graph-distance supervision beyond exact-label penalties while preserving the ground-truth graph as the unique zero-distance solution in aligned marginal probability space. We prove that SWORD is a strict metric on aligned marginal probabilistic scene graphs, preserves scene-graph-grounded substitution ordering, controls additive graph-grounded semantic observables, and yields a surrogate upper-bounding decoded semantic error.

Chinese Translation

3D场景图预测通常通过对象节点和关系边的分类损失进行优化。这些目标鼓励精确的对象和关系标签恢复，但它们并未直接优化与真实场景图的语义接近度：即使两个错误的图在对象角色和关系上下文上有显著差异，它们也可能受到相似的惩罚。一般来说，直接的图距离损失是困难的，因为图匹配和图编辑距离需要对应搜索和编辑成本设计。我们的关键观察是，之前的监督3D场景图预测协议是在固定的对象节点索引和监督的边槽上进行的。这可以将场景图比较转化为一个明确定义的对齐图距离问题。为此，我们提出了SWORD，一种基于场景的Wasserstein对象-关系距离，它通过确定性的基础到上下文顺序，从训练场景图注释中构建块归一化的对象和关系基础距离度量，而无需外部语义先验。SWORD将这些度量提升为对齐节点和空增强边边际的Wasserstein距离，提供超越精确标签惩罚的语义图距离监督，同时在对齐边际概率空间中将真实图作为唯一的零距离解。我们证明了SWORD是对齐边际概率场景图上的严格度量，保持场景图基础的替代顺序，控制加性图基础的语义可观测量，并产生一个上界的解码语义误差的替代品。

View on arXiv Download PDF AI Translation

cs.CV / 271 / 2610.17705

A Differentiable 3D Scene Graph Metric via Contextual Hellinger Triplet Geometry

通过上下文海林格三元几何实现可微分的3D场景图度量

Jun, U-Chae, Choi, Yoonhyuk, Kang, Jiwoo

Abstract

3D scene graph prediction is commonly supervised with local object and predicate classification losses. Although effective for slot-wise label prediction, this formulation does not provide a graph-level semantic distance between a predicted scene graph and its ground truth: semantically mild label errors and structurally disruptive relational errors can be treated similarly, and subject-predicate-object facts are not compared as scene-level units. We propose Contextual Hellinger Triplet Geometry, a graph-level supervision objective that measures the semantic and structural discrepancy between 3D scene graphs. Our key idea is to reinterpret the conventional classifier outputs of a 3D scene graph predictor as an aligned probabilistic scene graph field, where object and predicate predictions jointly define contextual subject-predicate-object facts. This formulation yields a differentiable metric-based training objective with an efficient factorized implementation and can be applied to existing 3D scene graph prediction architectures without modifying their network design. Experiments on 3DSSG demonstrate that our supervision improves downstream graph-conditioned 3D scene generation, while ablation studies confirm the complementary roles of the proposed graph-level terms and user studies show better graph descriptiveness, relation plausibility, and structural preservation.

Chinese Translation

3D场景图预测通常通过局部对象和谓词分类损失进行监督。尽管这种方法在槽级标签预测上有效，但它并未提供预测场景图与其真实值之间的图级语义距离：语义上轻微的标签错误和结构上破坏性的关系错误可能被同等对待，而主语-谓词-宾语事实并未作为场景级单元进行比较。我们提出了上下文海林格三元几何（Contextual Hellinger Triplet Geometry），这是一种图级监督目标，用于测量3D场景图之间的语义和结构差异。我们的关键思想是将3D场景图预测器的传统分类器输出重新解释为一个对齐的概率场景图场，其中对象和谓词预测共同定义上下文的主语-谓词-宾语事实。这种表述产生了一个基于度量的可微分训练目标，具有高效的分解实现，并且可以在不修改网络设计的情况下应用于现有的3D场景图预测架构。在3DSSG上的实验表明，我们的监督方法改善了下游图条件的3D场景生成，而消融研究确认了所提图级项的互补作用，用户研究显示了更好的图描述性、关系合理性和结构保留。

View on arXiv Download PDF AI Translation

cs.CV / 272 / 2610.17706

From Token Correction to Claim Admission: Training-Free Hallucination Mitigation for Large Vision-Language Models

从标记修正到声明接纳：无训练的大型视觉-语言模型幻觉缓解

Han, Hyeonggeun, Cho, Taehyun, Hong, Sangwoo

Abstract

Large vision-language models often exhibit object hallucination, generating claims that are not supported by the input image. To mitigate this issue, we propose Evidence-Guided Visual Claim Admission (EGVCA), a training-free framework that admits a semantic claim only when it receives sufficient image-dependent support. EGVCA measures this support by comparing the claim likelihood under the observed image and a visual-null input. For captioning, it controls semantic expansion without altering token generation. For binary visual question answering, it combines atomic visual support with the image-induced change in the model's native Yes--No margin. Experiments on CHAIR, POPE, and MME across three LVLM backbones show that EGVCA substantially reduces object hallucination while preserving caption quality. It achieves the highest POPE F1 in eight of nine backbone--sampling settings and improves the native MME total score on all three backbones. These results demonstrate the effectiveness of claim-level admission as a simple alternative to token-level distribution correction.

Chinese Translation

大型视觉-语言模型常常表现出物体幻觉，生成与输入图像不符的声明。为了解决这一问题，我们提出了证据引导的视觉声明接纳（Evidence-Guided Visual Claim Admission, EGVCA），这是一种无训练的框架，仅在获得足够的图像依赖支持时才接纳语义声明。EGVCA通过比较在观察到的图像和视觉空输入下的声明可能性来衡量这种支持。对于图像描述，它在不改变标记生成的情况下控制语义扩展。对于二元视觉问答，它将原子视觉支持与模型固有的“是-否”边际的图像诱导变化相结合。在三个LVLM骨干网络上对CHAIR、POPE和MME的实验表明，EGVCA显著减少了物体幻觉，同时保持了描述质量。在九种骨干网络-采样设置中，EGVCA在八种情况下达到了最高的POPE F1，并在所有三个骨干网络上提高了原始MME总分。这些结果证明了声明级接纳作为标记级分布修正的简单替代方案的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 273 / 2610.17730

DT-LoRA: Reaching Beyond the Tangent Space of Low-Rank Adaptation

DT-LoRA：超越低秩适应的切线空间

Zhang, Yangming, Yi, Ling, Li, Datong, Ding, Jinliang, Wang, Xiaojie, Song, Tiecheng, Ning, Zhaolong

Abstract

Low-rank adaptation (LoRA) often underperforms full fine-tuning. We investigate this performance gap through a tangent-space lens: factorized updates are confined to the tangent space of the current low-rank factors, leaving gradient components outside this space unexploited. An analysis of ten representative LoRA variants reveals that these residual components account for most of the gradient energy and exhibit strong spectral concentration. Motivated by this finding, we propose Dynamic Tangent LoRA (DT-LoRA), which partitions a fixed rank budget into a stable branch and a dynamic branch to capture the principal residual gradient directions. We show that these directions define the optimal update of a local quadratic model under orthogonal-complement and rank constraints. Matrix-free randomized SVD recovers them without materializing the residual gradient, while controlled write-back preserves accumulated updates and stabilizes training within the allocated rank. DT-LoRA adds no trainable parameters or inference cost over LoRA. Experiments across computer vision and natural language processing tasks demonstrate consistent improvements over LoRA and its variants. Moreover, despite being a general-purpose adaptation method, DT-LoRA outperforms task-specific approaches to open-vocabulary domain generalization for semantic segmentation, achieving state-of-the-art results. Code is available at https://github.com/Snipine/DT-LoRA.

Chinese Translation

低秩适应（LoRA）通常表现不如完全微调。我们通过切线空间的视角研究这一性能差距：因子化更新被限制在当前低秩因子的切线空间内，导致该空间外的梯度分量未被利用。对十种代表性LoRA变体的分析表明，这些残余分量占据了大部分梯度能量，并表现出强烈的谱集中性。基于这一发现，我们提出了动态切线LoRA（DT-LoRA），它将固定的秩预算划分为稳定分支和动态分支，以捕捉主要的残余梯度方向。我们证明这些方向在正交补和秩约束下定义了局部二次模型的最佳更新。无矩阵化的随机SVD方法在不物化残余梯度的情况下恢复这些方向，而受控的写回机制则保留了累积更新并在分配的秩内稳定训练。DT-LoRA在训练参数或推理成本上与LoRA没有增加。跨计算机视觉和自然语言处理任务的实验表明，DT-LoRA在性能上始终优于LoRA及其变体。此外，尽管DT-LoRA是一种通用适应方法，但在开放词汇领域泛化的语义分割任务中，其表现优于特定任务的方法，达到了最先进的结果。代码可在 https://github.com/Snipine/DT-LoRA 获取。

View on arXiv Download PDF AI Translation

cs.CV / 274 / 2610.17743

AttenIR: Aligning Dense Visual Attention with Mesh Warping for Unsupervised Image Retargeting

AttenIR：通过网格变形对齐密集视觉注意力以实现无监督图像重定向

Zhang, Yun, Zhang, Zhiruo, Zhu, Lifei, Zhu, Zhe, Li, Tong, Zhang, Fang-Lue, Liu, Shengjuan, Zheng, Xinyuan

Abstract

Image retargeting aims to resize images while preserving salient content free from distortion and stretching. Recent learning-based solutions reduce reliance on traditional hand-crafted features and optimization procedures, leading to more efficient and generalizable performance. Nevertheless, existing approaches still struggle to preserve global structure and content fidelity. To address these limitations, we propose AttenIR, an unsupervised image retargeting framework requiring no manual labels, based on content-aware mesh warping, which ef- fectively couples dense visual attention with sparse mesh constraints. We for- mulate the mesh deformation as a vertex displacement regression task. Guided by saliency priors, our method jointly optimizes scale, structure, and geometry preservation losses. By combining dense visual attention with semantic features, it achieves robust, unsupervised retargeting without manual annotations. Our primary contributions include (1) a lightweight unsupervised method for image retargeting that ensures content fidelity and preserves global structure, and (2) a coarse-to-fine projection strategy to align dense visual attention priors with sparse mesh constraints, enabling content-aware deformation at the pixel level. Experiments demonstrate that our method achieves competitive and superior performance compared with both traditional and state-of-the-art learning-based methods.

Chinese Translation

图像重定向旨在在不扭曲和拉伸的情况下调整图像大小，同时保留显著内容。近期的基于学习的解决方案减少了对传统手工特征和优化过程的依赖，从而实现了更高效和更具普适性的性能。然而，现有方法在保留全局结构和内容保真度方面仍然面临挑战。为了解决这些局限性，我们提出了AttenIR，一个无需手动标签的无监督图像重定向框架，基于内容感知的网格变形，有效地将密集视觉注意力与稀疏网格约束相结合。我们将网格变形公式化为一个顶点位移回归任务。在显著性先验的指导下，我们的方法联合优化了尺度、结构和几何保留损失。通过将密集视觉注意力与语义特征相结合，它实现了强健的无监督重定向，而无需手动注释。我们的主要贡献包括（1）一种轻量级的无监督图像重定向方法，确保内容保真度并保留全局结构，以及（2）一种粗到细的投影策略，将密集视觉注意力先验与稀疏网格约束对齐，从而实现像素级的内容感知变形。实验表明，我们的方法在与传统方法和最先进的基于学习的方法相比时，表现出竞争力和优越的性能。

View on arXiv Download PDF AI Translation

cs.CV / 275 / 2610.17791

When, Where, and Why: Hierarchical Causal Explainability for Video Face Understanding with MLLMs

何时、何地以及为何：基于层次因果可解释性的面部视频理解框架与多模态大语言模型

Kataria, Anubhav, Ghosh, Shreya, Ye, Jingwen, Dhall, Abhinav

Abstract

Understanding facial behaviour in video requires answering three fundamental questions: when a key facial expression occurs, where the model attends on the face at that moment, and why that evidence is necessary rather than merely correlated. Today's multimodal large language models (MLLMs) can label and describe video, but they rarely provide faithful explanations, especially for small facial movements that change over time. We present VideoFace-X-LLM, a framework for explainable facial video understanding comprising two key components. First, we introduce ExplainFace, a weakly supervised video-text corpus for facial behaviour understanding, automatically constructed from CelebV-HQ videos and validated through human evaluation and a two-model audit (at most 2.3% contradictions), demonstrating consistent gains across two backbone architectures. Second, H-TCC (Hierarchical Temporal Spatial Causal Concepts), which finds the key temporal moments (When), the facial regions (Where), and verifies its causal necessity (Why) through targeted occlusion. Fine-tuning on ExplainFace raises appearance Macro F1 from 19.48% to 32.87% (+13.39), with gains of +4.38 on actions, +6.77 on temporal emotion, and +11.84 F1 on deepfake detection. We also measure the explanations pointing lift, face/edge lift, stability, an Adebayo sanity check, and a necessity test rather than only showing examples. Our models, code, and corpus are available on request for research use.

Chinese Translation

理解视频中的面部行为需要回答三个基本问题：关键面部表情何时出现（When）、模型在该时刻关注面部的哪些区域（Where），以及为何这些证据是必要的而不仅仅是相关的（Why）。如今的多模态大语言模型（MLLMs）能够对视频进行标注和描述，但它们很少提供真实的解释，尤其是在处理随时间变化的小面部动作时。我们提出了VideoFace-X-LLM，这是一个可解释的面部视频理解框架，包含两个关键组件。首先，我们引入了ExplainFace，这是一个用于面部行为理解的弱监督视频-文本语料库，自动构建于CelebV-HQ视频之上，并通过人工评估和双模型审计（最多2.3%的矛盾）进行验证，展示了在两种主干架构下的一致性提升。其次，H-TCC（层次时间空间因果概念）用于找到关键的时间时刻（When）、面部区域（Where），并通过针对性遮挡验证其因果必要性（Why）。在ExplainFace上进行微调后，外观的宏观F1从19.48%提升至32.87%（+13.39），在动作、时间情感和深伪检测上分别获得+4.38、+6.77和+11.84的F1提升。我们还测量了解释指向提升、面部/边缘提升、稳定性、Adebayo的合理性检查和必要性测试，而不仅仅是展示示例。我们的模型、代码和语料库可根据请求用于研究。

View on arXiv Download PDF AI Translation

cs.CV / 276 / 2610.17792

MHVF: Training-Free GUI Grounding via Verification-Guided Multi-Hypothesis Search

MHVF：通过验证引导的多假设搜索实现无训练的GUI定位

Chen, Minghua, Liao, Xingming, Zhong, Weiqiang, Wang, Zhuowei

Abstract

GUI grounding enables web agents to translate natural-language instructions into actionable interface locations. However, dense layouts and small, visually similar controls can lead models toward plausible but incorrect elements, while refinement around a single prediction may leave alternative targets unexplored. We propose the Multi-Hypothesis Verification Framework (MHVF) for training-free GUI grounding. It combines multi-hypothesis exploration with local visual verification using a single pretrained vision-language model. To explore alternative targets, MHVF constructs spatial hypotheses from multiple full-screen predictions and examines the corresponding regions through zoomed re-localization. To resolve ambiguity among the resulting candidates, it verifies their agreement with the instruction in marked local views. Verification scores then guide a second round of local search and rank candidates retained from both search levels for final selection. Experiments with five general-purpose and GUI-specific models across eight benchmark settings demonstrate improvements in grounding accuracy in most settings. MHVF improves UI-Venus-Ground-7B by 4.3 and 2.1 percentage points on Mind2Web and VisualWebBench, respectively, and raises Qwen2.5-VL-7B's accuracy on ScreenSpot-Pro from 23.6% to 53.6%.

Chinese Translation

GUI定位使网络代理能够将自然语言指令转换为可操作的界面位置。然而，密集的布局和小而视觉相似的控件可能导致模型倾向于选择看似合理但实际上不正确的元素，而围绕单一预测的细化可能会导致其他目标未被探索。我们提出了多假设验证框架（MHVF）用于无训练的GUI定位。它结合了多假设探索与局部视觉验证，使用单个预训练的视觉-语言模型。为了探索替代目标，MHVF从多个全屏预测中构建空间假设，并通过缩放重新定位检查相应区域。为了消除结果候选者之间的歧义，它验证候选者与标记局部视图中的指令的一致性。验证得分随后引导第二轮局部搜索，并对来自两个搜索层次的候选者进行排名，以便进行最终选择。在八个基准设置中，对五个通用和GUI特定模型的实验表明，在大多数设置中，定位准确性有所提高。MHVF在Mind2Web和VisualWebBench上分别提高了UI-Venus-Ground-7B的准确率4.3和2.1个百分点，并将Qwen2.5-VL-7B在ScreenSpot-Pro上的准确率从23.6%提高到53.6%。

View on arXiv Download PDF AI Translation

cs.CV / 277 / 2610.17800

Multimodal Fusion of RGB and Complementary Modalities for Semantic Segmentation

RGB及其互补模态的多模态融合用于语义分割

Cao, Hu, Cheng, Yifeng, He, Mu, Li, Mengyu, Liu, Yinlong, Knoll, Alois

Abstract

Multi-modal semantic segmentation adds an auxiliary sensing stream to RGB images (RGB-X), such as thermal, LiDAR, depth, event, polarization, or light field. This can improve robustness in poor lighting and under motion blur. But different sensors often produce features that do not match well, and fusion can become unstable. To alleviate these issues, we propose a bidirectional polarity-aware cross-modality fusion (BPCF) module that effectively captures complementary cues while enhancing feature alignment. We evaluate the framework on six modality pairings: RGB-Thermal, RGB-LiDAR, RGB-Depth, RGB-Event, RGB-Polarization, and RGB-Light Field. Our model achieve state-of-the-art (SOTA) results on eight public datasets, including MFNet, KITTI-360, DELIVER, DDD17, DSEC, MCubeS, ZJU, and UrbanLF. The codes and trained models will be released upon acceptance.

Chinese Translation

多模态语义分割在RGB图像（RGB-X）中添加辅助感知流，例如热成像、激光雷达（LiDAR）、深度、事件、偏振或光场。这可以在光线不足和运动模糊的情况下提高鲁棒性。然而，不同传感器通常产生的特征不匹配，融合可能变得不稳定。为了解决这些问题，我们提出了一种双向极性感知跨模态融合（BPCF）模块，该模块有效捕捉互补线索，同时增强特征对齐。我们在六种模态配对上评估该框架：RGB-热成像、RGB-激光雷达、RGB-深度、RGB-事件、RGB-偏振和RGB-光场。我们的模型在八个公共数据集上实现了最先进的（SOTA）结果，包括MFNet、KITTI-360、DELIVER、DDD17、DSEC、MCubeS、ZJU和UrbanLF。代码和训练模型将在接受后发布。

View on arXiv Download PDF AI Translation

cs.CV / 278 / 2610.17808

SPARCL: Spatially Adaptive Regional Coverage for Active Learning in Chest X-ray Analysis

SPARCL：用于胸部X光分析的空间自适应区域覆盖主动学习

Jung, Juho, Kang, Seongjae, Kim, Mijoo, Cho, Sungduk, Jun, Yunkyoung, Joo, Sunghoon, Hwang, Yoona

Abstract

Pixel-level annotation for chest X-ray (CXR) lesion segmentation is expensive, making active learning (AL) essential. However, existing AL strategies fail to account for the spatial distribution of selected samples, resulting in inconsistent performance across focal (e.g., aortic calcification) and diffuse (e.g., nodules) pathologies. In this paper, we propose SPARCL (Spatially Adaptive Regional Coverage for Active Learning) to dynamically balance predictive uncertainty with cumulative spatial coverage. By employing a concave coverage function, SPARCL imposes diminishing returns on previously annotated regions, naturally directing the annotation budget toward under-labeled areas. Furthermore, an entropy gate reverts the algorithm to standard uncertainty sampling when spatial variation is absent. This coverage objective is monotone submodular, guaranteeing near-optimal greedy batch selection. To address how aggregate metrics like Dice obscure regional failures, we further introduce two metrics---Spatial Inconsistency (SI) and Performance Coverage (PC)---to quantify anatomical fairness across AL rounds. Evaluated on an in-house calcification cohort and public CXR benchmarks, SPARCL reaches 95% of fully supervised performance using only 24% of the data, compared to 29% for the best baseline. Ultimately, SPARCL ensures robust spatial reliability, surfacing regional gains that aggregate metrics miss.

Chinese Translation

胸部X光（CXR）病变分割的像素级标注成本高昂，使得主动学习（AL）变得至关重要。然而，现有的AL策略未能考虑所选样本的空间分布，导致在局灶性（例如，主动脉钙化）和弥漫性（例如，结节）病理中的表现不一致。本文提出了SPARCL（用于主动学习的空间自适应区域覆盖），以动态平衡预测不确定性与累积空间覆盖。通过采用凹形覆盖函数，SPARCL对先前标注的区域施加递减收益，自然将标注预算引导至标注不足的区域。此外，当缺乏空间变异时，熵门控将算法恢复为标准不确定性采样。该覆盖目标是单调子模的，保证了近似最优的贪婪批量选择。为了应对像Dice这样的聚合指标掩盖区域性失败的问题，我们进一步引入了两个指标——空间不一致性（SI）和性能覆盖（PC）——以量化AL轮次中的解剖公平性。在内部钙化队列和公共CXR基准上进行评估时，SPARCL仅使用24%的数据便达到了95%的完全监督性能，而最佳基线为29%。最终，SPARCL确保了稳健的空间可靠性，揭示了聚合指标所忽视的区域性收益。

View on arXiv Download PDF AI Translation

cs.CV / 279 / 2610.17829

QK-Edit: Revisiting Attention-based Injection in MM-DiT for Image and Video Editing

QK-Edit：重新审视MM-DiT中基于注意力的注入用于图像和视频编辑

Shen, Tiancheng, Huang, Zilong, Li, Xiangtai, Lin, Zhijie, Liu, Jiyang, Wang, Yitong, Feng, Jiashi, Yang, Ming-Hsuan, Liew, Jun Hao

Abstract

Multimodal Diffusion Transformers (MM-DiTs) have recently emerged as a powerful framework for unified text-vision synthesis, surpassing traditional U-Net architectures in generative tasks. One key innovation lies in its Multimodal Self-Attention (MM-SA) interaction, where image and text tokens are concatenated and processed via self-attention. However, this mechanism poses significant challenges for editing, rendering conventional U-Net-based attention manipulation methods ineffective. To address this limitation, we propose QK-Edit, a training-free framework that exploits the unique attention dynamics of MM-DiTs for precise text-guided image and video editing. By introducing a novel query-key manipulation strategy, our method isolates and adjusts critical attention components to achieve an optimal balance between prompt fidelity and structural consistency. This enables seamless edits across various tasks, including object addition, object removal, object replacement, changing background, changing material, changing color, and style transformation. Notably, it can be easily implemented with feature replacement in inference. QK-Edit demonstrates superior editing performance on state-of-the-art models, such as FLUX and HunyuanVideo, effectively bridging the gap between generative power and editable flexibility in MM-DiTs, and paving the way for scalable multimodal content creation.

Chinese Translation

多模态扩散变换器（MM-DiTs）最近作为一个强大的框架出现，用于统一的文本-视觉合成，在生成任务中超越了传统的U-Net架构。其关键创新在于多模态自注意力（MM-SA）交互，其中图像和文本标记被串联并通过自注意力进行处理。然而，这一机制对编辑提出了重大挑战，使得传统的基于U-Net的注意力操作方法无效。为了解决这一限制，我们提出了QK-Edit，一个无训练的框架，利用MM-DiTs独特的注意力动态实现精确的文本引导图像和视频编辑。通过引入一种新颖的查询-键操作策略，我们的方法隔离并调整关键的注意力组件，以实现提示保真度和结构一致性之间的最佳平衡。这使得在各种任务中实现无缝编辑成为可能，包括对象添加、对象移除、对象替换、背景更改、材料更改、颜色更改和风格转换。值得注意的是，它可以通过推理中的特征替换轻松实现。QK-Edit在FLUX和HunyuanVideo等最先进模型上展示了卓越的编辑性能，有效弥合了MM-DiTs中生成能力与可编辑灵活性之间的差距，为可扩展的多模态内容创作铺平了道路。

View on arXiv Download PDF AI Translation

cs.CV / 280 / 2610.17844

What Makes a Good Visual Representation for Spatial Understanding in MLLMs?

什么样的视觉表示对多模态大型语言模型的空间理解是有效的？

Huang, Xiaohu, Guo, Yumin, Han, Kai

Abstract

Spatial understanding is a necessary step in moving multimodal large language models (MLLMs) toward reasoning about the 3D physical world. Their visual pretraining data, however, is almost entirely 2D, so the visual tokens processed by MLLMs often lack inherent 3D structural information. A growing number of methods inject such structure by transferring representations from external vision or geometry models, and collectively establish that such transfer helps. Behind these methods lies a largely underexplored question: what makes a visual representation good for spatial understanding? To answer it, we first separate what is transferred from how it is transferred, and then trace the resulting differences back to properties of the representations themselves. What we transfer comes from 3D geometry models, self-supervised image and video models, vision-language models, and generative models, in the form of latent features or explicit depth and point maps. We transfer it either by distillation, using it as a supervision target, or by fusion, feeding it as an additional input. We find that a representation's 3D structure correlates strongly with how much it helps, measured by cross-view consistency for latent features and by depth and reconstruction accuracy for explicit geometry. We further show that the form of a representation dictates the mechanism it needs. Latent features gain as much from fusion as from distillation; explicit geometry gains little from fusion, and most of its distillation gain depends on a geometry decoding head. Guided by these findings, we distill both latent features and explicit geometry into a single MLLM. Taking only RGB video as input, it consistently improves over its base model on both static 3D and dynamic 4D spatial reasoning benchmarks.

Chinese Translation

空间理解是推动多模态大型语言模型（MLLMs）向对三维物理世界进行推理的重要步骤。然而，它们的视觉预训练数据几乎完全是二维的，因此MLLMs处理的视觉标记通常缺乏固有的三维结构信息。越来越多的方法通过从外部视觉或几何模型中转移表示来注入这种结构，并共同表明这种转移是有帮助的。在这些方法背后，存在一个尚未被充分探索的问题：什么样的视觉表示对空间理解是有效的？为了回答这个问题，我们首先将转移的内容与转移的方式分开，然后追溯结果差异与表示本身的属性之间的关系。我们转移的内容来自三维几何模型、自监督图像和视频模型、视觉-语言模型以及生成模型，以潜在特征或显式深度和点图的形式存在。我们通过蒸馏（distillation）将其作为监督目标，或通过融合（fusion）将其作为额外输入进行转移。我们发现，表示的三维结构与其帮助程度之间存在强相关性，后者通过潜在特征的跨视图一致性和显式几何的深度及重建精度进行测量。我们进一步表明，表示的形式决定了其所需的机制。潜在特征从融合中获得的收益与从蒸馏中获得的收益相当；而显式几何从融合中获得的收益很小，其大部分蒸馏收益依赖于几何解码头。根据这些发现，我们将潜在特征和显式几何蒸馏到一个单一的MLLM中。仅以RGB视频作为输入，它在静态三维和动态四维空间推理基准测试中始终优于其基础模型。

View on arXiv Download PDF AI Translation

cs.CV / 281 / 2610.17853

DVTR-KD: Distilling Vision-Language Models via Intermediate Visual Token Reconstruction

DVTR-KD：通过中间视觉标记重建提炼视觉-语言模型

Wei, Haolan, Wang, Jiahui, Zha, Junli, Lu, Xinkai, Zou, Yunxiao, Zhou, Chao, Song, Xiang

Abstract

Multimodal large language models (MLLMs) have achieved strong performance on diverse vision-language tasks, but their computational demands limit deployment under constrained budgets, particularly on edge devices. Many works focus on Knowledge Distillation (KD) and try to transfer knowledge from powerful teacher models to students. In addition to reducing model size, visual token compression provides a complementary route to efficiency, but introduces mismatched visual token counts that complicate direct token-wise knowledge transfer. To mitigate this, we propose DVTR-KD, a knowledge distillation framework based on \textbf{D}ense \textbf{V}isual \textbf{T}oken \textbf{R}econstruction, which reconstructs the teacher's dense visual tokens from the student's compressed tokens. We reconstruct visual states after the final LLM layer and at the input to an intermediate layer identified as important for visual-information readout. At the intermediate layer, reconstruction is supervised through the teacher's response predictions rather than by direct token-wise KL or MSE matching. An auxiliary attention-readout loss complements this response-level supervision by aligning how the student and the original teacher integrate contextual information when predicting responses. Evaluations on diverse benchmarks demonstrate that DVTR-KD improves the performance of token-compressed students while retaining the inference efficiency afforded by visual token compression.

Chinese Translation

多模态大型语言模型（MLLMs）在各种视觉-语言任务中表现出色，但其计算需求限制了在预算受限的情况下的部署，尤其是在边缘设备上。许多研究集中在知识蒸馏（KD）上，试图将强大的教师模型的知识转移到学生模型中。除了减少模型大小，视觉标记压缩提供了一条补充的效率路径，但引入了不匹配的视觉标记数量，复杂化了直接的标记级知识转移。为此，我们提出了DVTR-KD，这是一种基于 extbf{D}ense extbf{V}isual extbf{T}oken extbf{R}econstruction的知识蒸馏框架，该框架从学生的压缩标记中重建教师的稠密视觉标记。我们在最终LLM层之后和在一个被识别为对视觉信息读取重要的中间层的输入处重建视觉状态。在中间层，重建通过教师的响应预测进行监督，而不是通过直接的标记级KL或MSE匹配。辅助的注意力读取损失通过对齐学生和原教师在预测响应时整合上下文信息的方式，补充了这种响应级监督。在多种基准测试中的评估表明，DVTR-KD提高了标记压缩学生的性能，同时保留了视觉标记压缩所带来的推理效率。

View on arXiv Download PDF AI Translation

cs.CV / 282 / 2610.17859

Frequency Routing and Identity Conditioning for Thermal-to-Visible Face Translation

热成像到可见光人脸翻译中的频率路由与身份调节

Li, Haojie, Sun, Xinlong, Lei, Xiaohaoyang, Cheng, Kai, Qiao, Yiguo, Yang, Xu, Chen, Cunjian

Abstract

Thermal-to-visible (T2V) face translation has important applications in nighttime biometric recognition, low-light surveillance, and human interpretation of thermal observations. The task remains challenging because thermal images do not directly record visible appearance, while the cross-spectral cues useful for reconstruction vary across frequency components and semantic roles. Existing spatially conditioned models tend to process these cues through a homogeneous adaptation path and do not explicitly separate local structure from global identity. We propose FreqID-Control, a parameter-efficient framework that organizes thermal guidance according to spectral reliability and semantic role while preserving the pretrained generative prior. It combines frequency-routed low-rank adaptation, which allocates separate transformation subspaces to low-frequency structure and high-frequency detail, with shared cross-spectral identity conditioning, which supplies compact global identity tokens independently of spatial thermal tokens. A controlled analysis on MCXFace and SpeakingFaces provides the empirical basis for this design, revealing stable low-frequency correspondence and edge-localized detail correspondence. Experiments on both datasets demonstrate strong image fidelity and biometric identity preservation, while controlled ablations on MCXFace confirm the complementary contributions of frequency routing and shared identity conditioning.

Chinese Translation

热成像到可见光（T2V）人脸翻译在夜间生物识别、低光监控以及人类对热成像观察的解读中具有重要应用。该任务仍然具有挑战性，因为热成像图像并不直接记录可见外观，而用于重建的跨光谱线索在频率成分和语义角色上各不相同。现有的空间条件模型倾向于通过同质适应路径处理这些线索，并未明确区分局部结构与全局身份。我们提出了FreqID-Control，这是一种参数高效的框架，根据光谱可靠性和语义角色组织热成像指导，同时保留预训练的生成先验。它结合了频率路由的低秩适应，该方法为低频结构和高频细节分配了独立的变换子空间，以及共享的跨光谱身份调节，该方法独立于空间热成像标记提供紧凑的全局身份令牌。对MCXFace和SpeakingFaces的控制分析为该设计提供了实证基础，揭示了稳定的低频对应关系和边缘局部化的细节对应关系。在这两个数据集上的实验表明了强大的图像保真度和生物识别身份的保持，而对MCXFace的控制消融实验确认了频率路由和共享身份调节的互补贡献。

View on arXiv Download PDF AI Translation

cs.CV / 283 / 2610.17866

MicroCLIP: Connecting Text and Microscopy Images via Contrastive Language--Image Pretraining

MicroCLIP：通过对比语言-图像预训练连接文本与显微镜图像

Ruan, Yanran, Wei, Zhixiang, Hong, Shiyao, Ma, Xiaoxiao, Gu, Yuxuan, Ye, Liming, Chen, Huaian, Jin, Yi

Abstract

Understanding fluorescence microscopy images is fundamental to automating biological research. However, existing methods typically learn predefined concepts from closed-set data, limiting transfer across tasks and biological contexts. In this work, we introduce MicroCLIP, a vision-language model that learns generalizable cellular and subcellular representations from natural language supervision for open-set image classification and retrieval. To pretrain MicroCLIP, we construct MicroPool-2M, a dataset of approximately 2.5 million fluorescence microscopy image-text pairs spanning diverse cellular structures, cell lines, and imaging conditions. Aligning microscopy images with biological concepts in natural language is challenging because most images lack descriptive captions and biological knowledge annotations. We therefore progressively establish image-text associations at three levels to provide MicroCLIP with richer semantic supervision: (1) category-level alignment links images to standardized labels defined by subcellular structure-cell-line combinations; (2) knowledge-level alignment incorporates biological priors derived from these labels to capture relationships across species and cell types; and (3) detail-level alignment uses a multimodal large language model to recaption and verify image-text pairs at scale, emphasizing fine-grained visual characteristics. Extensive experiments demonstrate that MicroCLIP achieves leading performance in zero-shot classification and text-image retrieval on microscopy images. It retrieves images with specific cellular characteristics from thousands of candidates and generalizes to unseen biological concepts without additional supervision. These results highlight MicroCLIP's potential as a foundation model for fluorescence microscopy.

Chinese Translation

理解荧光显微镜图像对于自动化生物研究至关重要。然而，现有方法通常从封闭集数据中学习预定义概念，限制了在任务和生物背景之间的迁移。在本研究中，我们介绍了MicroCLIP，一种视觉-语言模型，它从自然语言监督中学习可泛化的细胞和亚细胞表示，以实现开放集图像分类和检索。为了预训练MicroCLIP，我们构建了MicroPool-2M，一个包含约250万对荧光显微镜图像-文本对的数据集，涵盖了多样的细胞结构、细胞系和成像条件。将显微镜图像与自然语言中的生物概念对齐具有挑战性，因为大多数图像缺乏描述性标题和生物知识注释。因此，我们逐步在三个层面建立图像-文本关联，以为MicroCLIP提供更丰富的语义监督：（1）类别级对齐将图像链接到由亚细胞结构-细胞系组合定义的标准化标签；（2）知识级对齐结合了来自这些标签的生物先验，以捕捉跨物种和细胞类型的关系；（3）细节级对齐使用多模态大语言模型对图像-文本对进行重新标注和验证，强调细粒度的视觉特征。大量实验表明，MicroCLIP在显微镜图像的零样本分类和文本-图像检索中实现了领先的性能。它能够从数千个候选中检索具有特定细胞特征的图像，并在没有额外监督的情况下对未见过的生物概念进行泛化。这些结果突显了MicroCLIP作为荧光显微镜基础模型的潜力。

View on arXiv Download PDF AI Translation

cs.CV / 284 / 2610.17877

LoopREG: Representation Supervision Enables Test-Time Recurrent Depth in Diffusion Transformers

LoopREG：表征监督使扩散变换器在测试时实现递归深度

Wang, Zipeng, Xiong, Anping, Zhang, Jing, Zhu, Xiaoya, Jiang, Yi, Xiao, Yanbing, Zeng, Zige, Chen, Zezhou, Liu, Xiang, Hu, Huan, Chen, Ping, Wang, Kai, Liu, Zhaoxiang, Lian, Shiguo, LV, Ye, Zhao, Fang, Gao, Huanlin, Huang, Zhengrui, Shen, Bo, You, Yuren

Abstract

Reusing Transformer blocks offers a way to increase diffusion-model computation at test time, but naive recurrence can degrade generation beyond its training depth. We introduce LoopREG, a representation-supervised recurrent diffusion Transformer that repeatedly applies a shared block at the state supervised by a pretrained visual encoder. Alignment is applied after the final recurrent visit during training; inference extends the same learned operator to additional visits. On ImageNet 256 x 256, LoopREG reduces SiT-B/2 FID from the local REG baseline's 15.08 to 4.91 at 400K training steps, a 67.41% reduction. Within a checkpoint trained with two visits, one extra inference visit improves FID from 14.71 to 5.34 with unchanged weights. The benefit persists on SiT-L/2 and SiT-XL/2. Controlled interventions show that useful recurrence follows the location of representation supervision, while REPA alignment produces a smaller extrapolation benefit. These results make representation-supervised recurrence a practical route to improving diffusion generation through additional inference depth.

Chinese Translation

重用变换器模块提供了一种在测试时增加扩散模型计算的方法，但简单的递归可能会导致生成超出其训练深度的退化。我们提出了LoopREG，这是一种表征监督的递归扩散变换器，它在由预训练视觉编码器监督的状态下重复应用共享模块。在训练过程中，最终递归访问后应用对齐；推理阶段则将相同的学习操作扩展到额外的访问。在ImageNet 256 x 256上，LoopREG将SiT-B/2的FID从局部REG基线的15.08降低到4.91，减少了67.41%。在一个经过两次访问训练的检查点中，额外的一次推理访问将FID从14.71提高到5.34，而权重保持不变。这一好处在SiT-L/2和SiT-XL/2上也得以持续。控制干预表明，有效的递归遵循表征监督的位置，而REPA对齐则产生较小的外推收益。这些结果使得表征监督的递归成为通过额外推理深度改善扩散生成的实际途径。

View on arXiv Download PDF AI Translation

cs.CV / 285 / 2610.17882

Diagnose-to-Generate: Eliminating Structural Distortions in Flow Models via Patch-Level Grounded Alignment

诊断生成：通过补丁级别的基础对齐消除流模型中的结构扭曲

Li, Yifan, Gao, Lishuai, Zhang, Tong, Huang, Qilong, Zhao, Chen, Kang, Zhuoliang, Hu, Jie, Zhong, Yujie, Tang, Fenghe, Zhou, S Kevin, Wei, Xiaoming

Abstract

Flow-based image generators achieve strong semantic synthesis but still produce localized structural distortions, such as malformed hands and limbs. Existing post-training methods typically optimize global scalar rewards, which provide weak spatial credit assignment and may alter well-formed regions while correcting local defects. We present \textit{Diagnose-to-Generate (D2G)}, a grounded alignment framework that separates visual diagnosis from generative optimization. We construct \textbf{Achilles-600K}, a region-annotated dataset for structural anomalies, and train \textbf{ARGUS} through supervised cold-start and Region-Aware Group Relative Policy Optimization to predict anomaly categories and bounding boxes. These diagnoses are converted into confidence-weighted spatial rewards for our \textbf{Patch-level Flow-GRPO}. We further identify that global stochastic exploration can shift semantic regions across group rollouts, making patch-wise relative advantages spatially inconsistent. Our \textbf{Trajectory-Preserving Stochastic Sampling (TPSS)} anchors early denoising with deterministic ODE trajectories and introduces stochasticity only at later steps, thereby limiting intra-group spatial drift while retaining local exploration. Experiments on FLUX.1-dev and SD3.5 show improved structural fidelity with preserved FID and CLIP scores. Consistent gains under HumanRefiner, zero-shot GPT-4.1, and blinded human evaluation further demonstrate that the improvements extend beyond the training reward model.

Chinese Translation

基于流的图像生成器在语义合成方面表现出色，但仍然会产生局部结构扭曲，例如畸形的手和肢体。现有的后训练方法通常优化全局标量奖励，这提供了较弱的空间信用分配，并可能在修正局部缺陷时改变良好形成的区域。我们提出了 extit{Diagnose-to-Generate (D2G)}，一个将视觉诊断与生成优化分离的基础对齐框架。我们构建了 extbf{Achilles-600K}，一个用于结构异常的区域标注数据集，并通过监督冷启动和区域感知组相对策略优化训练 extbf{ARGUS}，以预测异常类别和边界框。这些诊断被转换为我们 extbf{Patch-level Flow-GRPO}的置信加权空间奖励。我们进一步发现，全局随机探索可能会在组回滚中改变语义区域，使得补丁级相对优势在空间上不一致。我们的 extbf{Trajectory-Preserving Stochastic Sampling (TPSS)}通过确定性常微分方程轨迹锚定早期去噪，并仅在后续步骤引入随机性，从而限制组内空间漂移，同时保留局部探索。在FLUX.1-dev和SD3.5上的实验显示，结构保真度有所提高，同时保持了FID和CLIP分数。在HumanRefiner、零-shot GPT-4.1和盲评人类评估下的一致性提升进一步证明了这些改进超出了训练奖励模型的范围。

View on arXiv Download PDF AI Translation

cs.CV / 286 / 2610.17902

Visual Dependence Does Not Imply Answerability: A Systematic Analysis of Answerability Signals in Frozen InstructBLIP

视觉依赖并不意味着可回答性：对冻结的 InstructBLIP 中可回答性信号的系统分析

Kim, Minjun, Song, Hyunmin, Lim, Changwon

Abstract

Visual question answering models often produce plausible responses even when the available visual evidence is insufficient to support a valid answer. Although prior work has introduced benchmarks and abstention mechanisms for unanswerable visual questions, it remains unclear which signals derived from a frozen vision-language model reliably indicate answerability on unseen image-question pairs. We systematically evaluate answerability signals on a stratified 10K subset of UNK-VQA using frozen InstructBLIP, covering predictive uncertainty, cross-modal alignment, visual grounding, visual dependence, internal representations, and learned probes. Answerability discrimination remains weak, with test AUROC largely confined to 0.50-0.54. CLIP alignment yields the highest point estimate but does not significantly outperform generation confidence under paired bootstrap analysis. Targeted removal of question-relevant visual evidence causes substantially larger changes in answer confidence than same-size random masking, yet intervention sensitivity provides essentially no answerability discrimination. Increasing probe capacity, combining multiple signals, and adding reason-aware supervision do not provide robust improvements, while post-freeze cross-validation yields similarly weak discrimination. Selective-answering analysis shows limited usable coverage under low-risk operating criteria. Failure-mode analysis shows that the relative utility of individual signals varies across perturbation mechanisms. Taken together, these results show that, on our UNK-VQA subset with frozen InstructBLIP, visual dependence, uncertainty, semantic alignment, and static representations are not robust standalone proxies for answerability. Our findings emphasize the need to distinguish a model's reliance on visual evidence from whether that evidence is sufficient to support an answer.

Chinese Translation

视觉问答模型即使在可用的视觉证据不足以支持有效答案的情况下，仍然会产生看似合理的响应。尽管先前的研究已为不可回答的视觉问题引入了基准和放弃机制，但尚不清楚从冻结的视觉-语言模型中提取的哪些信号能够可靠地指示未见图像-问题对的可回答性。我们使用冻结的 InstructBLIP 对 UNK-VQA 的分层 10K 子集系统地评估可回答性信号，涵盖预测不确定性、跨模态对齐、视觉基础、视觉依赖、内部表示和学习探针。可回答性区分仍然较弱，测试 AUROC 大多局限于 0.50-0.54。CLIP 对齐提供了最高的点估计，但在配对自助分析下并未显著优于生成置信度。针对问题相关的视觉证据的有针对性移除导致答案置信度的变化明显大于同等大小的随机掩蔽，但干预敏感性几乎没有提供可回答性区分。增加探针容量、结合多种信号以及添加基于推理的监督并未提供稳健的改进，而冻结后的交叉验证则产生类似的弱区分。选择性回答分析显示在低风险操作标准下可用覆盖有限。故障模式分析表明，单个信号的相对效用在不同的扰动机制中有所不同。综合来看，这些结果表明，在我们的 UNK-VQA 子集上，使用冻结的 InstructBLIP，视觉依赖、不确定性、语义对齐和静态表示并不是可回答性的稳健独立代理。我们的研究强调了区分模型对视觉证据的依赖与该证据是否足以支持答案的必要性。

View on arXiv Download PDF AI Translation

cs.CV / 287 / 2610.17903

DriveWLA: Unifying Future Dynamics and Scene Semantics through World-Language-Action Modeling for Autonomous Driving

DriveWLA：通过世界-语言-动作建模统一未来动态与场景语义以实现自动驾驶

Wang, Shuzheng, Li, Haoxuan, Li, Shengzhen, Wu, Chenxi, Sun, Yajing, Hou, Lu, Liu, Wenyu, Wang, Xinggang

Abstract

Vision-Language Models (VLMs) and world models have demonstrated strong capabilities in semantic understanding and future scene evolution for autonomous driving. However, effectively integrating both capabilities for trajectory generation remains challenging. In this work, we propose DriveWLA, a unified World-Language-Action model based on Mixture-of-Transformers (MoT) that jointly models semantic understanding, future scene dynamics, and trajectory generation. DriveWLA consists of three experts: a World Expert for future prediction, a Language Expert for semantic understanding, and an Action Expert for trajectory generation. We further introduce Cross-Expert Attention, which enables controlled information flow across experts while aligning semantic and dynamic representations. To improve trajectory generation beyond imitation learning, we apply Flow-GRPO reinforcement learning post-training to refine the Action Expert. Extensive experiments show that DriveWLA achieves superior performance in both open-loop and closed-loop evaluation. DriveWLA achieves 92.3 PDMS on NAVSIM v1 and 90.1 EPDMS on NAVSIM v2, demonstrating strong performance among VLM-based and world-model-based planners. Without specific fine-tuning, DriveWLA further demonstrates strong zero-shot generalization to HUGSIM, nuScenes, and Bench2Drive. These results validate the effectiveness of our unified World-Language-Action modeling, establishing DriveWLA as a generalizable framework for autonomous driving.

Chinese Translation

视觉-语言模型（VLMs）和世界模型在自动驾驶的语义理解和未来场景演变方面表现出强大的能力。然而，有效地将这两种能力整合以生成轨迹仍然具有挑战性。在本研究中，我们提出了DriveWLA，一种基于混合变换器（Mixture-of-Transformers, MoT）的统一世界-语言-动作模型，联合建模语义理解、未来场景动态和轨迹生成。DriveWLA由三个专家组成：用于未来预测的世界专家、用于语义理解的语言专家以及用于轨迹生成的动作专家。我们进一步引入了跨专家注意力（Cross-Expert Attention），该机制能够在专家之间控制信息流动，同时对齐语义和动态表示。为了改善轨迹生成超越模仿学习，我们在后期训练中应用Flow-GRPO强化学习来优化动作专家。大量实验表明，DriveWLA在开环和闭环评估中均表现出优越的性能。DriveWLA在NAVSIM v1上达到了92.3的PDMS，在NAVSIM v2上达到了90.1的EPDMS，显示出在基于VLM和基于世界模型的规划器中的强大表现。在没有特定微调的情况下，DriveWLA还在HUGSIM、nuScenes和Bench2Drive上展示了强大的零样本泛化能力。这些结果验证了我们统一的世界-语言-动作建模的有效性，确立了DriveWLA作为一个可泛化的自动驾驶框架。

View on arXiv Download PDF AI Translation

cs.CV / 288 / 2610.17914

LinMorph: Learning to Morph Weights for Efficient Video DiT Linearization

LinMorph：学习权重变换以实现高效视频DiT线性化

Long, Lingyi, Zhao, Congyue, Li, Xiaoou, Qian, Yanmin, Huang, Tao

Abstract

Post-hoc linearization is an attractive way to accelerate pretrained Video Diffusion Transformers (Video DiTs), yet existing methods largely treat the quality loss induced by attention replacement as a recovery problem. We show that a more fundamental bottleneck lies upstream: \emph{initialization}. Random initialization discards pretrained knowledge, whereas manual weight reuse transfers it through rigid and incomplete source--target correspondences. We introduce \textbf{LinMorph}, which recasts Video DiT linearization as a learnable softmax-to-linear weight transformation problem. Instead of copying predefined counterparts, LinMorph learns a cross-parameter morphism that lets each target Gated DeltaNet (GDN) parameter draw jointly from pretrained query, key, value, and output projections, including for parameters with no direct softmax counterpart. A single morphism is shared across layers and learned through local teacher alignment, while producing layer-specific initializations from their respective pretrained weights. The resulting weights are then materialized as independent GDN modules for progressive distillation and joint recovery.Since the morphism operates in weight space rather than token space, it can be learned on short, low-resolution clips and directly reused at larger video scales. Across pretrained Video DiTs, LinMorph enables markedly faster recovery and better final generation quality than random initialization and heuristic weight reuse. Results show that efficient linearization depends not only on how a converted model is recovered, but critically on how pretrained knowledge is transferred before recovery. Project page: https://lonelyling.github.io/LinMorph.

Chinese Translation

后期线性化是一种加速预训练视频扩散变换器（Video Diffusion Transformers, Video DiTs）的有效方法，但现有方法在很大程度上将注意力替换引起的质量损失视为恢复问题。我们表明，更根本的瓶颈在于上游的 extit{初始化}。随机初始化丢弃了预训练知识，而手动权重重用则通过僵化和不完整的源-目标对应关系进行转移。我们引入了 extbf{LinMorph}，将视频DiT线性化重新构建为一个可学习的softmax到线性权重变换问题。LinMorph不是复制预定义的对应项，而是学习一种跨参数的变换，使每个目标门控DeltaNet（Gated DeltaNet, GDN）参数能够共同从预训练的查询、键、值和输出投影中获取信息，包括那些没有直接softmax对应项的参数。一个单一的变换在各层之间共享，并通过局部教师对齐进行学习，同时从各自的预训练权重中生成特定于层的初始化。最终的权重被具体化为独立的GDN模块，以进行渐进蒸馏和联合恢复。由于该变换在权重空间中操作，而非标记空间，因此可以在短小、低分辨率的片段上进行学习，并直接在更大的视频尺度上重用。在预训练的视频DiT中，LinMorph实现了显著更快的恢复速度和更好的最终生成质量，相较于随机初始化和启发式权重重用。结果表明，高效的线性化不仅依赖于转换模型的恢复方式，更关键的是依赖于在恢复之前如何转移预训练知识。项目页面：https://lonelyling.github.io/LinMorph。

View on arXiv Download PDF AI Translation

cs.CV / 289 / 2610.17926

CineMem: Attention-Aligned Fine-grained Memory Retrieval for Multi-Shot Long Video Generation

CineMem：用于多镜头长视频生成的注意力对齐细粒度记忆检索

Wu, Xiaoxue, Gao, Xinyu, Yu, Fang, Qiao, Yu, Chen, Xinyuan

Abstract

Autoregressive video generation relies on memory retrieval to model long-range dependencies. This capability becomes particularly critical for multi-shot generation, where models must reconcile substantial visual changes across shot transitions with persistent semantic continuity. Existing methods tend to perform memory retrieval at the frame level and share the same historical context across all query tokens, overlooking the heterogeneous information needs and leading to redundant retrieval and inefficient use of historical memory. To bridge this gap, we introduce CineMem, an attention-aligned fine-grained memory retrieval framework that combines online memory organization with query-adaptive retrieval for autoregressive multi-shot video generation. Specifically, guided by intrinsic memory-access patterns revealed through attention analysis, we propose an online-clustered KV cache as a memory organization strategy over all historical tokens, enabling semantic-aware memory grouping and more precise retrieval. In addition, we introduce a query-level retrieval mechanism that provides broader access to historical information under a fixed retrieval budget, thereby improving long-term memory utilization. Together, these designs enable the multi-shot video generator to preserve long-range consistency across shot transitions, particularly in revisitation scenarios, while approximating full-memory access under a constrained memory budget. To facilitate evaluation, we further introduce a suite of task-specific, VLM-assisted metrics for faithful inter-shot consistency assessment. Extensive experiments demonstrate the effectiveness of our framework.

Chinese Translation

自回归视频生成依赖于记忆检索来建模长程依赖性。这一能力在多镜头生成中尤为关键，因为模型必须在镜头转换中调和显著的视觉变化与持续的语义连贯性。现有方法往往在帧级别进行记忆检索，并在所有查询标记中共享相同的历史上下文，忽视了异质信息需求，导致冗余检索和历史记忆的低效利用。为了解决这一问题，我们提出了CineMem，一个注意力对齐的细粒度记忆检索框架，它结合了在线记忆组织和查询自适应检索，以实现自回归多镜头视频生成。具体而言，在注意力分析揭示的内在记忆访问模式的指导下，我们提出了一种在线聚类的键值（KV）缓存作为所有历史标记的记忆组织策略，从而实现语义感知的记忆分组和更精确的检索。此外，我们引入了一种查询级检索机制，在固定的检索预算下提供更广泛的历史信息访问，从而提高长期记忆的利用率。这些设计使得多镜头视频生成器能够在镜头转换中保持长程一致性，特别是在重访场景中，同时在受限的记忆预算下近似实现全记忆访问。为了便于评估，我们进一步引入了一套任务特定的、基于VLM的指标，用于准确评估镜头间一致性。大量实验表明了我们框架的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 290 / 2610.17933

HierarchiRetina: A Diabetic Retinopathy Screening and Grading Cascade with Heterogeneous Mixture-of-Experts Lesion Segmentation

HierarchiRetina：一种具有异构专家混合的病变分割的糖尿病视网膜病变筛查与分级级联系统

Uddin, Md Asif, Farhan, Niloy, Rahman, Rafeed, Alam, Md. Golam Rabiul

Abstract

Clinical screening for diabetic retinopathy proceeds in stages, yet most automated systems grade fundus images with a single black-box network. Three gaps remain: (i) lesions differ widely in size and contrast, but segmenters typically apply the same blocks to all; (ii) ungradable images are usually excluded or treated as an extra class, and lesion-aware graders seldom test whether lesions drive the grade; (iii) staged systems often report each stage only on the images reaching it, leaving propagated errors unmeasured. HierarchiRetina, a screening, segmentation and grading cascade with inspectable stage outputs, addresses each gap. First, HSMoE-AUNet is a U-Net whose skip, bottleneck and decoder blocks are sparse mixtures of heterogeneous experts, with lesion-matched types, counts and losses; SwinHRUNetPP, a dual-encoder vessel segmenter, uses a field-of-view-restricted loss. Second, LG-DRG separates gradability from ordinal severity and lets lesion masks query the image through gated cross-attention. Third, an end-to-end evaluation counts screening misses as errors and traces each error to its stage. Pooled Dice ranged from 0.184 (microaneurysms) to 0.579 (exudates); vessel Dice was 0.731. On 58,689 held-out images from six datasets, end-to-end quadratic weighted kappa was 0.804, with referral sensitivity 0.832 at specificity 0.947, and the screening gate caused 66% of errors. In ablations, the grade ignored the masks, even with the cross-attention gate started open. On all 4,105 official DDR test images, six-class accuracy was 85.6%, above the best published six-class result we found (83.8%). Staged screening systems should be evaluated end-to-end, with stage attribution and mask ablation, before clinical deployment.

Chinese Translation

糖尿病视网膜病变的临床筛查是分阶段进行的，但大多数自动化系统使用单一的黑箱网络对眼底图像进行分级。仍然存在三个问题：（i）病变在大小和对比度上差异很大，但分割器通常对所有病变应用相同的模块；（ii）无法分级的图像通常被排除或视为额外类别，而关注病变的分级器很少测试病变是否驱动分级；（iii）分阶段系统通常仅在达到某一阶段的图像上报告每个阶段，导致传播错误未被测量。HierarchiRetina是一个具有可检查阶段输出的筛查、分割和分级级联系统，旨在解决每个问题。首先，HSMoE-AUNet是一个U-Net，其跳跃、瓶颈和解码器模块是稀疏的异构专家混合，具有与病变匹配的类型、数量和损失；SwinHRUNetPP是一个双编码器血管分割器，使用视野限制损失。其次，LG-DRG将可分级性与序数严重性分开，并允许病变掩膜通过门控交叉注意力查询图像。第三，端到端评估将筛查漏检视为错误，并追踪每个错误到其阶段。在六个数据集的58,689张保留图像上，端到端的二次加权κ值为0.804，转诊敏感性为0.832，特异性为0.947，筛查门导致66%的错误。在消融实验中，即使在交叉注意力门打开的情况下，分级也忽略了掩膜。在所有4,105张官方DDR测试图像上，六类准确率为85.6%，高于我们找到的最佳已发布六类结果（83.8%）。分阶段筛查系统在临床应用前应进行端到端评估，包含阶段归属和掩膜消融。

View on arXiv Download PDF AI Translation

cs.CV / 291 / 2610.17936

PUGS: Physics-Guided Uncertainty Modeling for Underwater Dynamic 3D Gaussian Splatting

PUGS：用于水下动态3D高斯点云的物理引导不确定性建模

Li, Tingjun, Wang, Yuxi, Zhong, Guoqiang, Zhang, Zhaoxiang

Abstract

Dynamic 3D Gaussian Splatting (3DGS) provides an efficient framework for dynamic scene reconstruction. Most existing approaches model scene dynamics with deterministic deformation fields, assuming each Gaussian follows a distinct geometric trajectory over time. This assumption breaks down underwater, where light scattering, refraction, marine snow and low visibility create substantial physical ambiguity in captured images. We present PUGS, a physics-guided uncertainty model for underwater dynamic 3DGS. PUGS learns inherent uncertainty induced by underwater image formation and leverages it as a physical signal for dynamic scene modeling. Rather than estimating a single deterministic displacement, it parameterizes each Gaussian's motion as a probability distribution with expected deformation and physical uncertainty. We adopt stochastic deformation optimization to sample Gaussian trajectories via local reparameterization, exploring multiple physically plausible solutions during training. We further propose uncertainty stochastic annealing and a dual-path self-consistency distillation objective, converting stochastic exploration to deterministic reconstruction while preserving temporal coherence. Extensive experiments on underwater dynamic datasets and physics-aware benchmarks demonstrate superior rendering quality, temporal consistency and reconstruction robustness. Our method achieves 1.16 dB PSNR gain over the strongest baseline on UVEB and 3.99 dB on DRUVA.

Chinese Translation

动态3D高斯点云（3DGS）为动态场景重建提供了一个高效的框架。现有大多数方法使用确定性变形场来建模场景动态，假设每个高斯点在时间上遵循独特的几何轨迹。然而，这一假设在水下环境中失效，因为光散射、折射、海洋雪和低能见度在捕获的图像中造成了显著的物理模糊。我们提出了PUGS，一种用于水下动态3DGS的物理引导不确定性模型。PUGS学习水下图像形成所引起的固有不确定性，并将其作为动态场景建模的物理信号。与其估计单一的确定性位移，不如将每个高斯点的运动参数化为具有期望变形和物理不确定性的概率分布。我们采用随机变形优化，通过局部重新参数化来采样高斯轨迹，在训练过程中探索多个物理上合理的解决方案。此外，我们还提出了不确定性随机退火和双路径自一致性蒸馏目标，将随机探索转化为确定性重建，同时保持时间一致性。在水下动态数据集和物理感知基准上的大量实验表明，我们的方法在渲染质量、时间一致性和重建鲁棒性方面表现优越。我们的模型在UVEB数据集上比最强基线提高了1.16 dB的PSNR，在DRUVA数据集上提高了3.99 dB。

View on arXiv Download PDF AI Translation

cs.CV / 292 / 2610.17937

MERIT: Multimodal Evidence Reasoning over Image Tables

MERIT：基于图像表格的多模态证据推理

Jiang, Haoyang, Fan, Zhuotong, Yi, Mingyang, Fan, Ju, Hu, Lanqing, Zan, Xiao

Abstract

Table reasoning with language models typically assumes access to clean textual tables. In practice, tables are often serialized into text, which can corrupt cell contents, distort structural relationships, and discard visual cues, leaving the model to reason over incomplete evidence. In contrast, table images retain this information, but using it requires both accurate perception and reasoning grounded in the relevant cells. To address this, we propose MERIT (Multimodal Evidence Reasoning over Image Tables), a three-stage framework for learning to reason over table images, with serialized tables serving as optional auxiliary inputs (if provided). Our analysis identifies visual evidence acquisition as a bottleneck, motivating table-specific encoder adaptation through reconstruction, localization, and color-conditioned supervision. Building on the adapted representations, supervised fine-tuning teaches explicit evidence selection and reasoning using verified trajectories. Finally, reinforcement learning refines the policy with separate rewards for evidence correctness, calculation, final answers, and output format. By providing feedback on intermediate steps, the framework targets retrieval and computational errors that final-answer supervision alone does not distinguish. We evaluate the approach on table question answering and fact verification, analyze the contributions of visual inputs and training components, and assess generalization to additional datasets with different table structures and visual presentations.

Chinese Translation

使用语言模型进行表格推理通常假设可以访问干净的文本表格。然而，在实际应用中，表格常常被序列化为文本，这可能会损坏单元格内容、扭曲结构关系并丢失视觉线索，从而使模型在不完整的证据上进行推理。相比之下，表格图像保留了这些信息，但使用这些信息需要准确的感知和基于相关单元格的推理。为了解决这个问题，我们提出了MERIT（基于图像表格的多模态证据推理），这是一个三阶段框架，用于学习在表格图像上进行推理，序列化表格作为可选的辅助输入（如果提供的话）。我们的分析确定了视觉证据获取是一个瓶颈，这促使我们通过重建、定位和颜色条件监督来进行表格特定编码器的适应。在适应后的表示基础上，监督微调教会了显式证据选择和使用经过验证的轨迹进行推理。最后，强化学习通过对证据的正确性、计算、最终答案和输出格式给予单独的奖励来优化策略。通过对中间步骤提供反馈，该框架针对最终答案监督无法区分的检索和计算错误进行调整。我们在表格问答和事实验证上评估该方法，分析视觉输入和训练组件的贡献，并评估其对具有不同表格结构和视觉呈现的其他数据集的泛化能力。

View on arXiv Download PDF AI Translation

cs.CV / 293 / 2610.17944

AI-Enhanced Hyperspectral Imaging for Intraoperative Detection of Low-Grade Glioma: A Clinical Study

人工智能增强的高光谱成像用于低级别胶质瘤的术中检测：一项临床研究

Willems, Siri, Jain, Ishika, De Winne, Jens, Luthman, Siri, Adab, Hamed Zivari, Vandebriel, Roeland, Alperovich, Anna, Zhang, Xiaohan, Stoffelen, Ben, De Schepper, Tom, Solie, Lien, De Vleeschouwer, Steven

Abstract

Current image-guided neuronavigation techniques, such as intraoperative MRI and ultrasound, are limited by high costs, limited sensitivity and operational complexity. There is a clear need for more efficient systems to enhance intra-operative tumor detection and improve patient outcomes for intrinsic brain tumor resection. Whereas previous intraoperative brain HSI studies have predominantly investigated high-grade gliomas, glioblastomas, or mixed-grade cohorts, this clinical study focuses specifically on low-grade gliomas (LGGs). LGGs strongly resemble normal brain under white light conditions on the operative microscope. This study characterizes spectral differences between healthy and tumor tissue and provides a paired comparison of HSI- and RGB-based classification at matched tissue locations and under identical patient-level data splits. Tumor regions showed consistently lower reflectance, which may be associated with differences in hemoglobin concentration and oxygenation. A range of classification models was evaluated. The 3D CNN, which integrates spectral and spatial features, achieved the highest median ROC-AUC of 74.5%, compared with 67.9% for the spectral-only XGBoost model and 64.0% for the RGB-based model. These results support further investigation of HSI for low-grade glioma classification and quantify its performance relative to conventional RGB imaging in this cohort.

Chinese Translation

当前的图像引导神经导航技术，如术中MRI和超声，受到高成本、灵敏度有限和操作复杂性的限制。迫切需要更高效的系统来增强术中肿瘤检测并改善内源性脑肿瘤切除的患者预后。以往的术中脑高光谱成像（HSI）研究主要集中在高级别胶质瘤、胶质母细胞瘤或混合级别群体，而本临床研究专注于低级别胶质瘤（LGGs）。在手术显微镜下，LGGs在白光条件下与正常脑组织极为相似。本研究表征了健康组织与肿瘤组织之间的光谱差异，并在匹配的组织位置和相同的患者数据分割下提供了HSI与RGB分类的配对比较。肿瘤区域的反射率 consistently 较低，这可能与血红蛋白浓度和氧合的差异有关。评估了一系列分类模型。集成光谱和空间特征的3D卷积神经网络（CNN）实现了最高的中位ROC-AUC为74.5%，而仅基于光谱的XGBoost模型为67.9%，RGB模型为64.0%。这些结果支持对低级别胶质瘤分类的HSI进一步研究，并量化其相对于传统RGB成像在该群体中的表现。

View on arXiv Download PDF AI Translation

cs.CV / 294 / 2610.17951

FETALFusion: Resolution-Aware State-Space Modeling for Multi-Domain Fetal Head Segmentation

FETALFusion：面向多领域胎头分割的分辨率感知状态空间建模

Bhilwarawala, Ammar, Bandyopadhyay, Mainak

Abstract

Head circumference (HC) measured from 2D ultrasound is a routine marker of fetal growth, and automating it needs segmentations whose boundaries hold up on individual scans, not only on average. We present FETALFusion, a dual-path CNN-Mamba encoder-decoder for fetal head segmentation across acquisition types. It applies four-directional state-space scans only at the two highest-resolution stages, routes bottleneck features through domain-specific channel gates that are averaged at test time, and adds a weak landmark shape prior. We train jointly on transabdominal (HC-18) and transperineal (PSFHS) images and compare ten methods on a held-out test set of 471 images, each trained under one schedule and scored by surface distances at native resolution. FETALFusion reaches a Dice score of 0.963 and an HD95 of 2.10 mm without test-time augmentation or a domain label. It improves on U-Mamba on all five reported metrics (HD95 reduced by 0.31 mm; Holm-corrected ($p < 10^{-4}$), a lead that holds under test-time augmentation and across two independent training runs, and in the mean on U-Net, Attention U-Net and TransUNet. Residual encoders close the gap: a ResNet-34 UNet++ exceeds FETALFusion, and an ImageNet initialised ResNet-34 U-Net is best overall (HD95 1.75 mm). The ablation agrees. Calibrated against run-to-run variation, it attributes FETALFusion's measurable gains to the landmark shape prior and, on transperineal images, to the domain gates, while removing the state-space branch does not measurably change accuracy. HC from the predicted masks agrees with the clinical reference at an intraclass correlation of 0.9986.

Chinese Translation

从二维超声测量的头围（HC）是胎儿生长的常规标志，自动化这一过程需要在个体扫描上保持边界的分割，而不仅仅是平均值。我们提出了FETALFusion，一种用于胎头分割的双路径CNN-Mamba编码器-解码器，适用于不同的采集类型。它仅在两个最高分辨率阶段应用四方向状态空间扫描，通过特定领域的通道门路由瓶颈特征，这些特征在测试时进行平均，并添加了一个弱的标志形状先验。我们在经腹（HC-18）和经会阴（PSFHS）图像上进行联合训练，并在一个包含471幅图像的保留测试集上比较了十种方法，每种方法在一个训练计划下训练，并通过原生分辨率的表面距离进行评分。FETALFusion达到了0.963的Dice分数和2.10 mm的HD95，且未使用测试时增强或领域标签。它在所有五个报告的指标上优于U-Mamba（HD95减少了0.31 mm；经过Holm校正（$p < 10^{-4}$），这一领先在测试时增强和两个独立训练运行中保持一致，并在U-Net、Attention U-Net和TransUNet的均值上表现优异。残差编码器缩小了差距：ResNet-34 UNet++超越了FETALFusion，而经过ImageNet初始化的ResNet-34 U-Net在整体表现上最佳（HD95为1.75 mm）。消融实验结果一致。经过运行间变异的校准，它将FETALFusion的可测增益归因于标志形状先验，并在经会阴图像上归因于领域门，而移除状态空间分支并未显著改变准确性。从预测掩膜得到的HC与临床参考的内类相关性为0.9986。

View on arXiv Download PDF AI Translation

cs.CV / 295 / 2610.17968

dMRI-F: A Foundation Model for Generalizable Diffusion MRI Analysis from Sparse Acquisitions

dMRI-F：一种用于稀疏采集的通用扩散MRI分析的基础模型

Zhu, Xi, Li, Yijie, Wang, Xiaofan, Zhang, Wei, Chen, Huafu, Zhu, Ce, Zhang, Fan

Abstract

Diffusion MRI (dMRI) probes water diffusion within the brain, providing unique information about tissue microstructure and white matter connectivity. High-quality dMRI enables advanced and accurate analyses but demands dense acquisitions, resulting in long scan times and hardware constraints that are often infeasible. Further, dMRI requires dedicated computational methods to support diverse analysis tasks (e.g., parcellation, registration, and synthesis) for both neuroscientific and clinical applications. Here, we present dMRI-F, a foundation model that unifies dMRI data enhancement and downstream tasks across diverse acquisitions and populations, while requiring only minimal acquisition input. dMRI-F is pre-trained with a q-space masked reconstruction task: predicting densely sampled multi-shell data from only six volumes. It learns generalizable latent feature representations of dMRI data, which serve as a core foundation for adapting to diverse downstream tasks, including brain parcellation, brain registration, and cross-modal synthesis. Across nine independently acquired datasets, we demonstrate that dMRI-F outperforms state-of-the-art methods with robust generalization across acquisitions, populations, and tasks. Our results demonstrate dMRI-F's versatility as a general-purpose framework that enables diverse dMRI applications from low-cost, sparse acquisitions.

Chinese Translation

扩散MRI（dMRI）探测大脑内水分子的扩散，提供关于组织微观结构和白质连接的独特信息。高质量的dMRI能够实现先进且准确的分析，但需要密集的采集，这导致扫描时间长和硬件限制，往往难以实现。此外，dMRI需要专门的计算方法来支持多样化的分析任务（例如，分区、配准和合成），以满足神经科学和临床应用的需求。在此，我们提出了dMRI-F，这是一种基础模型，统一了dMRI数据增强和多样化采集及人群的下游任务，同时仅需最少的采集输入。dMRI-F通过q空间掩蔽重建任务进行预训练：从仅六个体积中预测密集采样的多壳数据。它学习了dMRI数据的可泛化潜在特征表示，这些表示作为适应多样化下游任务的核心基础，包括大脑分区、大脑配准和跨模态合成。在九个独立采集的数据集中，我们证明了dMRI-F在采集、群体和任务之间具有强大的泛化能力，超越了最先进的方法。我们的结果表明，dMRI-F作为一个通用框架的多功能性，能够从低成本、稀疏的采集中实现多样化的dMRI应用。

View on arXiv Download PDF AI Translation

cs.CV / 296 / 2610.17986

Web Scale on a Budget: Tail-Aware Retrieval Evaluation

预算内的网络规模：尾部感知检索评估

Meivar, Boaz, Avidan, Shai

Abstract

A common practice in instance retrieval is to grow the gallery size to improve a benchmark's discrimination capability. One prominent example is ILIAS, which adds up to 100 million distractor images. The prevailing assumption is that faithful evaluation demands web scale, and that indexing all 100M images through the backbone is the unavoidable price of a comparable number. That price is steep: hundreds of GPU-hours and ~200 GB of descriptors per model, paid again for every method or ablation. We show the web-scale indexing price is largely avoidable: evaluation itself obeys a scaling law. The number of distractors that outrank a positive is Binomial in the gallery size. We define the per-query retrieval hardness, a gallery-scale agnostic quantity that yields a retrieval metric at any scale, and demonstrate it can be derived from a small random sample. The underlying sampled-rank machinery is long established in recommender-system and biometric evaluation, but to our knowledge has never been adapted to visual retrieval. We make the law per-query and estimable, and validate it at true 100M scale on ILIAS and INSTRE, and on the full 759k-distractor Google Landmarks v2 pool. Concretely, within a 5% sample (a flat 20x less indexing), every backbone's tail-aware (EVT) estimate has converged onto the large-scale mAP, to within ~0.01 (mean absolute error 0.005), across five backbones, three benchmarks, and two distractor pools (mAP@1000 at 100M on ILIAS/INSTRE; the official mAP@100 at the full 759k pool on GLDv2); the recommender-systems count baseline over-predicts everywhere (+0.037 to +0.068). This is an order of magnitude better than naively reporting the metric on the same subset (+0.08 to +0.13). We release code and the per-query hardness to facilitate faster, cheaper, and greener retrieval evaluation.

Chinese Translation

实例检索中的一种常见做法是扩大图库规模，以提高基准的区分能力。一个显著的例子是ILIAS，它增加了多达1亿张干扰图像。普遍的假设是，忠实的评估需要网络规模，并且通过主干索引所有1亿张图像是获得可比数量的不可避免的代价。这个代价是巨大的：每个模型需要数百个GPU小时和约200GB的描述符，每种方法或消融都需再次支付。我们展示了网络规模索引的代价在很大程度上是可以避免的：评估本身遵循一个规模法则。超过正样本的干扰者数量在图库规模中是二项分布的。我们定义了每查询检索难度，这是一个与图库规模无关的量，可以在任何规模下产生检索指标，并证明它可以从一个小的随机样本中推导出来。基础的抽样排名机制在推荐系统和生物特征评估中早已建立，但据我们所知，从未适应于视觉检索。我们使得该法则在每查询中可估计，并在ILIAS和INSTRE的真实1亿规模以及完整的759k干扰者Google Landmarks v2池上进行了验证。具体而言，在5%的样本内（索引量减少20倍），每个主干的尾部感知（EVT）估计已收敛到大规模mAP，误差在约0.01范围内（平均绝对误差为0.005），涵盖五个主干、三个基准和两个干扰者池（ILIAS/INSTRE在100M时的mAP@1000；GLDv2在完整759k池上的官方mAP@100）；推荐系统的计数基线在各处都存在过度预测（+0.037至+0.068）。这比在同一子集上天真地报告指标要好一个数量级（+0.08至+0.13）。我们发布代码和每查询的难度，以促进更快速、更便宜和更环保的检索评估。

View on arXiv Download PDF AI Translation

cs.CV / 297 / 2610.17991

PolUR: Polarization Aware Multimodal Degradation Perception for Underwater Image Restoration

PolUR：考虑偏振的多模态退化感知用于水下图像恢复

Zhang, Wenshuo, Zhang, Xiangyue, Li, Junlin, Tian, Wenjie, Zheng, Yubo, Xu, Hongli, Ru, Jingyu, Li, Xinde

Abstract

Recovering high-quality images from degraded underwater observations is crucial for underwater exploration. Polarization observations show promise as an effective source of complementary physical cues for recovering target radiance under clear-water conditions. However, existing methods have yet to jointly model raw intensity information and polarization cues to enable image restoration driven by an understanding of mixed degradations (color casts, haze, and low light). To overcome these limitations, we propose PolUR, a unified multimodal language-driven framework that jointly uses polarization and RGB information to identify degradation types and severity levels through structured semantic instructions and enhances joint restoration representations under the guidance of degradation-aware semantic features to improve restoration robustness under mixed degradations. We further construct a new benchmark named MixPolar, which contains underwater images exhibiting real-world degradations and their corresponding degradation descriptions. MixPolar supports systematic evaluation of degradation perception, image restoration, and performance on downstream tasks. Extensive experiments demonstrate that PolUR effectively integrates polarization and RGB information, achieving competitive performance in degradation perception and image restoration, while also demonstrating generalization capability and utility for downstream tasks.

Chinese Translation

从退化的水下观测中恢复高质量图像对于水下探索至关重要。偏振观测作为一种有效的补充物理线索，在清水条件下恢复目标辐射方面显示出良好的前景。然而，现有方法尚未将原始强度信息和偏振线索联合建模，以便通过对混合退化（色偏、雾霾和低光照）的理解来驱动图像恢复。为克服这些限制，我们提出了PolUR，一个统一的多模态语言驱动框架，联合使用偏振和RGB信息，通过结构化语义指令识别退化类型和严重程度，并在退化感知的语义特征指导下增强联合恢复表示，以提高在混合退化下的恢复鲁棒性。我们进一步构建了一个新的基准数据集MixPolar，其中包含展示真实世界退化的水下图像及其对应的退化描述。MixPolar支持对退化感知、图像恢复和下游任务性能的系统评估。大量实验表明，PolUR有效整合了偏振和RGB信息，在退化感知和图像恢复方面取得了竞争力的表现，同时展示了对下游任务的泛化能力和实用性。

View on arXiv Download PDF AI Translation

cs.CV / 298 / 2610.17996

Uncertainty-Aware Semi-Supervised Framework with Supervised Contrastive Learning for Sugarcane Leaf Disease Detection

基于监督对比学习的具有不确定性感知的半监督框架用于甘蔗叶病检测

Dangi, Shalini, Teja, Divya, Kabde, Omkar, Dar, Shahid Shafi, Kumar, Nagendra

Abstract

Sugarcane leaf diseases severely impact crop yield and quality, making accurate and timely detection critical for sustainable agriculture. Conventional deep learning models for plant disease detection depend on large annotated datasets, which are expensive to obtain and challenging to prepare due to challenges commonly observed across crop disease detection tasks, including (i) high intra-class variation caused by diverse symptom expressions across growth stages, (ii) subtle inter-class similarities between diseased and healthy tissues, and (iii) the risk of noisy labels degrading model reliability. To handle these challenges, we presented an Uncertainty-aware Mean-Teacher based Semi-Supervised Learning Disease Detection Network (U-SSDNet) that leverages unlabeled data while maintaining robustness. First, discrete wavelet transform preprocessing is applied to enhance lesion visibility and reduce intra- class variation. Subsequently, the KL divergence is employed to refine the pseudo-labels and reduce the impact of confirmation bias. At the same time, supervised contrastive learning is applied to make the features more distin- guishable and improve the separation among different classes. Experimental evaluations on sugarcane leaf disease datasets, considered a representative case study, demonstrate that the proposed approach achieves performance close to that of fully supervised models while requiring far fewer annotated samples and consistently outperforms state-of-the-art supervised and semi-supervised approaches.

Chinese Translation

甘蔗叶病严重影响作物产量和质量，因此准确及时的检测对可持续农业至关重要。传统的植物病害检测深度学习模型依赖于大量标注数据集，这些数据集获取成本高且准备困难，主要由于作物病害检测任务中常见的挑战，包括（i）由于不同生长阶段的症状表现多样性导致的高类内变异，（ii）病变组织与健康组织之间的微妙类间相似性，以及（iii）标签噪声降低模型可靠性的风险。为应对这些挑战，我们提出了一种基于不确定性感知的均值教师半监督学习病害检测网络（U-SSDNet），该网络利用未标记数据同时保持鲁棒性。首先，应用离散小波变换预处理以增强病变可见性并减少类内变异。随后，采用KL散度来优化伪标签并减少确认偏差的影响。同时，应用监督对比学习使特征更加可区分，改善不同类别之间的分离。在甘蔗叶病数据集上的实验评估，作为一个具有代表性的案例研究，表明所提方法的性能接近于完全监督模型，同时所需的标注样本数量远少于传统方法，并且在各类最先进的监督和半监督方法中始终表现优越。

View on arXiv Download PDF AI Translation

cs.CV / 299 / 2610.18005

Surface Reconstruction from Range Scans Under Visibility Constraints

在可见性约束下从范围扫描重建表面

Lenain, Gabriel, Boyer, Edmond, Ponce, Jean

Abstract

Surface reconstruction from range scans is challenging in the presence of noise and outliers, as fidelity to observations must be balanced against sensitivity to measurement errors. Yet, most neural approaches discard the sensor positions and corresponding visibility information, compensating for this loss with geometric priors or regularization terms. We instead formulate surface reconstruction as robust regression of the predicted first surface intersection along each acquisition ray against its measured depth, which penalizes predicted surface intersections within the segment between the sensor and the observed point. We represent the surface as the zero set of a 1-Lipschitz multi-layer perceptron, allowing the use of linearly convergent sphere tracing to compute the first hit, and use a robust Huber loss for regression. Our formulation requires neither oriented normals, sampled off-surface points, nor an eikonal loss. Comparisons with the state of the art on synthetic and real data with varying levels of noise and outlier contamination demonstrate competitive performance on clean data and much improved fidelity and robustness under even moderate corruption. Our analysis further confirms a trade-off between fidelity and robustness: greater fitting capacity improves data fidelity in the clean case but increases sensitivity to the measurement errors inherent in real acquisitions, with a sweet spot typically emerging at intermediate capacity.

Chinese Translation

在噪声和异常值存在的情况下，从范围扫描进行表面重建是具有挑战性的，因为对观测值的忠实度必须与对测量误差的敏感性之间取得平衡。然而，大多数神经网络方法忽略了传感器位置和相应的可见性信息，通过几何先验或正则化项来弥补这一损失。我们则将表面重建公式化为沿每个采集光线的预测第一表面交点与其测量深度之间的稳健回归，这一过程对传感器与观测点之间段内的预测表面交点施加惩罚。我们将表面表示为一个1-Lipschitz多层感知器的零集，从而允许使用线性收敛的球体追踪来计算第一次碰撞，并使用稳健的Huber损失进行回归。我们的公式化不需要定向法线、离表面采样点或Eikonal损失。与最先进的方法在合成数据和真实数据上的比较，显示出在干净数据上的竞争性能，并在即使是中等程度的污染下也显著提高了忠实度和稳健性。我们的分析进一步确认了忠实度与稳健性之间的权衡：更大的拟合能力在干净情况下提高了数据的忠实度，但增加了对真实采集中固有测量误差的敏感性，通常在中等能力下会出现一个最佳点。

View on arXiv Download PDF AI Translation

cs.CV / 300 / 2610.18018

Dynamic Interpolation for Training-Time Prediction Reshaping in Low-Light Image Enhancement

低光照图像增强中的训练时预测重塑的动态插值

Zhu, Peirui, Hao, Shijie, Liao, Jingxi, Hong, Richang, Wang, Meng

Abstract

Low-light image enhancement (LLIE) is an important image/video processing problem for reliable visual systems under poor illumination, yet it remains challenging due to the strong coupling among insufficient illumination, noise amplification, and structural degradation. Existing methods typically rely on sophisticated network architectures or engineered objective functions. While effective, the former increases model complexity and deployment burden, whereas the latter relies on empirically tuned loss weights. We revisit supervised LLIE from a training-dynamics perspective and observe that brightness-related errors tend to converge early, while residual noise and structural artifacts often remain insufficiently optimized in later stages. To address this issue, we propose a Dynamic Interpolation Training-only Module (DITM), which introduces no additional learnable parameters and adaptively reshapes the model prediction for supervised feed-forward LLIE training. Specifically, DITM employs brightness-aware interpolation, in which the model prediction is combined with a mean brightness-aligned image constructed by rescaling the model prediction according to the mean brightness of the ground-truth image. By reducing the relative dominance of global brightness discrepancy, DITM makes residual local errors more influential during backpropagation after global luminance has largely stabilized, thereby facilitating artifact suppression and fine-detail restoration in later training stages. DITM is used only during training and is fully removed at inference, introducing no additional inference cost. Extensive experiments on multiple benchmark datasets and CNN-, Transformer-, and Mamba-based LLIE backbones show that DITM consistently improves fidelity and perceptual quality while preserving the inference-time architecture and introducing no additional inference-time computation.

Chinese Translation

低光照图像增强（LLIE）是一个重要的图像/视频处理问题，旨在在光照不足的情况下为可靠的视觉系统提供支持，但由于光照不足、噪声放大和结构退化之间的强耦合关系，这一问题仍然具有挑战性。现有的方法通常依赖于复杂的网络架构或工程化的目标函数。虽然这些方法有效，但前者增加了模型的复杂性和部署负担，而后者则依赖于经验调优的损失权重。我们从训练动态的角度重新审视监督式LLIE，并观察到与亮度相关的误差往往在早期阶段就趋于收敛，而残余噪声和结构伪影在后期阶段往往未得到充分优化。为了解决这个问题，我们提出了一种动态插值训练专用模块（Dynamic Interpolation Training-only Module, DITM），该模块不引入额外的可学习参数，并自适应地重塑模型预测以用于监督式前馈LLIE训练。具体而言，DITM采用了亮度感知插值，其中模型预测与通过根据真实图像的平均亮度重新缩放模型预测构建的均值亮度对齐图像相结合。通过减少全局亮度差异的相对主导性，DITM使得在全局亮度基本稳定后，残余局部误差在反向传播过程中更具影响力，从而促进后期训练阶段的伪影抑制和细节恢复。DITM仅在训练期间使用，并在推理时完全移除，不引入额外的推理成本。在多个基准数据集以及基于CNN、Transformer和Mamba的LLIE骨干网络上的大量实验表明，DITM始终提高了保真度和感知质量，同时保持了推理时的架构，并未引入额外的推理计算。

View on arXiv Download PDF AI Translation

人工智能 (Artificial Intelligence)

300

cs.AI / 1 / 2610.14005

Artificiality's First Exam: An LLM-Referenced C-test for Supraintelligence

人工智能的首次考验：基于大型语言模型的超智能C测试

Hernandez-Orallo, Jose

Abstract

Under the accelerated pace of AI development, any benchmark whose difficulty is calibrated with humans or current AI systems becomes circular, or saturates soon. How can we then evaluate machines that may become more capable than the current machines and the people who write the tests? One answer to the question hinges on inverse problems, inductive inference questions for which difficulty depends on the complexity of the underlying mechanism, given the observations and the previous knowledge. Levin's Kt, a computable variant of Kolmogorov complexity, puts a theoretical upper bound to that difficulty, balancing the simplest representation encoding for the mechanism and the logarithm of the time steps to run that mechanism. Previous tests derived from Levin's Kt, such as the original C-test, are based on minimalistic Universal Turing Machines (UTM), capable of very limited representations and no previous knowledge. Reasoning large language models (LLM) are UTMs that capture a compendium of human culture as knowledge, using prompts as natural language programs. This paper presents the first test based on Levin's Kt with LLMs as representational machines. Fresh test items can be produced at high difficulties, estimated in a ratio scale measured in bits, before any human or frontier model can solve them. We formalise the measure, devise a lightweight generation and certification pipeline, explore the effect of the power of the reference machine and the gap between examiner and examinee. We produce a first battery of certified items administered to twelve LLMs. The resulting agent characteristic curves decrease with difficulty, down to the uncharted high-difficulty region, the supra-intelligence window of items we can certify, yet beyond generations of frontier intelligence.

Chinese Translation

在人工智能发展加速的背景下，任何与人类或当前人工智能系统难度相匹配的基准测试都会变得循环，或很快达到饱和。那么，我们如何评估可能比当前机器和编写测试的人更强大的机器呢？对此问题的一个答案依赖于逆问题，即归纳推理问题，其难度取决于在给定观察和先前知识的情况下，潜在机制的复杂性。Levin的Kt，作为Kolmogorov复杂度的可计算变体，为这种难度设定了理论上限，平衡了机制的最简表示编码与运行该机制所需时间步长的对数。基于Levin的Kt的先前测试，如原始C测试，基于极简的通用图灵机（UTM），其能力有限且没有先前知识。推理大型语言模型（LLM）是捕捉人类文化作为知识的UTM，使用提示作为自然语言程序。本文提出了第一个基于Levin的Kt的测试，以LLM作为表征机器。在任何人类或前沿模型能够解决之前，可以以比特为单位的比例尺度生成高难度的新测试项目。我们对该度量进行了形式化，设计了轻量级的生成和认证流程，探讨了参考机器的能力以及考官与被考者之间的差距的影响。我们生成了第一批认证项目，并对十二个LLM进行了测试。结果显示，代理特征曲线随着难度的增加而下降，直到未探索的高难度区域，即我们可以认证的超智能项目窗口，超越了前沿智能的几代。

View on arXiv Download PDF AI Translation

cs.AI / 2 / 2610.14115

The Silent Failures in Agentic Data Science

代理数据科学中的隐性失败

Li, Xue, Xia, Hongqian, Gomm, Daniel, Santos, Aécio, Wolff, Cornelius, Hulsebos, Madelon

Abstract

Large language models have significantly lowered the barrier to writing code. Naturally, this has also led to the development of data science agents that promise automating the design and execution of data science from task formulation to model development. Existing evaluations of agentic data science (agentic DS) assess downstream predictive performance and code correctness, providing preliminary insight in agentic DS capabilities. We argue that these evaluations overlook the fundamental principles of data science: methodological data science- and domain knowledge. As such, only assessing downstream performance masks methodological failures such as data leakage and inappropriate validation, limiting the development of robust agentic DS systems. In this paper, we evaluate agentic DS beyond accuracy from three different angles: 1) a large-scale analysis of failure modes in common agentic DS frameworks and models across 220 data science tasks, 2) a reusable and extensible test suite with curated data science tasks that test for six specific failures in a controlled manner, and 3) a qualitative assessment of agentic DS pipelines against expert solutions on Kaggle competitions. We find that the majority of agentic solutions contain at least one failure mode despite the complexity of agentic harness or model capacity. Moreover, we identify a positive correlation between the occurrence of data leakage failures and predictive performance, confirming the necessity to evaluate agents beyond accuracy. Our qualitative analysis shows that DS agents generally lack problem solving creativity and domain knowledge, compared to competition-winning data scientists. We believe that principled failure mode analysis, as presented here, should be integral part of agentic DS evaluations, and that the insights we surface will be key in developing trustworthy agentic DS pipelines.

Chinese Translation

大型语言模型显著降低了编写代码的门槛。这自然也导致了数据科学代理的发展，承诺从任务制定到模型开发自动化数据科学的设计和执行。现有的代理数据科学（agentic DS）评估主要关注下游预测性能和代码正确性，提供了对代理数据科学能力的初步洞察。我们认为，这些评估忽视了数据科学的基本原则：方法论数据科学和领域知识。因此，仅评估下游性能掩盖了方法论上的失败，例如数据泄漏和不当验证，限制了稳健的代理数据科学系统的发展。本文从三个不同的角度超越准确性评估代理数据科学：1）对220个数据科学任务中常见代理数据科学框架和模型的失败模式进行大规模分析；2）一个可重用和可扩展的测试套件，包含经过策划的数据科学任务，以受控方式测试六种特定失败；3）对代理数据科学管道与Kaggle竞赛中专家解决方案的定性评估。我们发现，尽管代理工具或模型能力复杂，大多数代理解决方案仍包含至少一种失败模式。此外，我们识别出数据泄漏失败的发生与预测性能之间存在正相关，确认了超越准确性评估代理的必要性。我们的定性分析表明，与获胜的数据科学家相比，数据科学代理通常缺乏问题解决的创造力和领域知识。我们认为，本文所呈现的原则性失败模式分析应成为代理数据科学评估的一个重要组成部分，而我们所揭示的见解将在开发可信赖的代理数据科学管道中发挥关键作用。

View on arXiv Download PDF AI Translation

cs.AI / 3 / 2610.14118

ClawLoop: Self-Improving Agents via Harness-Native Training Loop

ClawLoop：通过利用原生训练循环实现自我改进的智能体

Yang, Hansi, Hao, Linji, Hao, Hongyan, Wang, An, Wu, Han, Gu, Qi, Cai, Xunliang

Abstract

The rapid advancement of Large Language Models (LLMs) has driven their deployment toward general-purpose harness environments, where LLM-based agents operate in sandbox workspaces and interact directly with file systems and shells. While numerous post-training pipelines exist for tool-use agents, their counterparts for harness-based agents remain scarce and are predominantly one-shot and teacher-dependent: a stronger external model synthesizes tasks and collects trajectories, the target agent is trained once, and evaluation results are consumed by human researchers rather than fed back into the loop. Such pipelines are inherently bounded by the capability of the external model and the scope of human prior, leaving no room for self-improvement. To address these limitations, we propose ClawLoop, a framework that enables self-improving without any external teacher model and with minimal prior dependencies. ClawLoop composes three phases into a self-sustaining cycle: Task Synthesis, which generates a diverse pool of training tasks via two complementary construction routines, applies diagnosis-conditioned difficulty growth followed by both agent-free and agent-based task validation; Training, which updates the agent via SFT and/or RL; and Diagnosis and Feedback, which analyzes the agent's rollout behaviors to produce a capability analysis report that will be used as feedback for task synthesis in the next iteration. Using Qwen3.6-27B as the loop model, ClawLoop yields consistent improvements on SkillsBench, Claw-Eval-Live, and ClawGym-Bench. The iteratively refined tasks are further transferred to smaller-scale Qwen3.5-4B and 9B models with gains across all benchmarks, validating data quality independently of the loop model.

Chinese Translation

大型语言模型（LLMs）的快速发展推动了它们在通用环境中的部署，在这些环境中，基于LLM的智能体在沙盒工作区中运行，并直接与文件系统和命令行交互。尽管存在许多用于工具使用智能体的后训练管道，但针对基于环境的智能体的相应管道仍然稀缺，且主要是一次性和依赖教师的：一个更强大的外部模型合成任务并收集轨迹，目标智能体训练一次，评估结果由人类研究人员消费，而不是反馈到循环中。这些管道本质上受到外部模型能力和人类先验知识范围的限制，无法实现自我改进。为了解决这些局限性，我们提出了ClawLoop，一个无需任何外部教师模型且依赖最小先验知识的自我改进框架。ClawLoop将三个阶段组成一个自我维持的循环：任务合成，通过两种互补的构建例程生成多样化的训练任务，应用基于诊断的难度增长，随后进行无智能体和基于智能体的任务验证；训练，通过SFT和/或RL更新智能体；以及诊断与反馈，分析智能体的执行行为，生成能力分析报告，作为下一次迭代任务合成的反馈。使用Qwen3.6-27B作为循环模型，ClawLoop在SkillsBench、Claw-Eval-Live和ClawGym-Bench上实现了一致的改进。经过迭代优化的任务进一步转移到小规模的Qwen3.5-4B和9B模型上，在所有基准测试中均取得了提升，验证了数据质量独立于循环模型。

View on arXiv Download PDF AI Translation

cs.AI / 4 / 2610.14136

Correction Training for Diffusion Models

扩散模型的校正训练

Ahn, Donghoon, Wang, Runqian, Efros, Alexei A., Park, Taesung, Belhe, Yash

Abstract

Iterative sampling in diffusion models can lead to error accumulation due to exposure bias. The denoiser is trained on real images interpolated with noise, but at inference its inputs are constructed from its own imperfect predictions from the earlier steps, something it has never seen at training time. This might be one reason diffusion models often struggle without additional help, such as CFG. We introduce a correction step for diffusion models which attempts to reduce this training-inference gap. The corrector takes the base model's prediction as input and is trained against the same ground-truth target. This makes correction an explicit learning task, with training inputs provided by the base model. During sampling, at each denoising step, we iteratively refine the denoiser's clean-sample prediction. This allows us to correct predictions at intermediate denoising steps, preventing early artifacts from persisting into the final sample. Although the corrector is trained only on predictions from the base denoiser, it can be applied repeatedly to its own outputs at inference for better results. This simple framework leads to substantial improvement in sample quality while largely preserving diversity. We evaluate our method across DiT, SiT, and JiT. Our method substantially improves upon previously reported FID scores for unguided generation and, notably, unconditional generation, where producing high-quality samples without guidance has been particularly challenging. Our method achieves performance competitive with classifier-free guidance (CFG), and combining correction with CFG further improves diversity and reduces oversaturation typical of CFG. Finally, we show that the additional inference cost of correction can be amortized by distilling the refined predictions back into the base diffusion model, retaining the quality gains without additional inference-time overhead.

Chinese Translation

扩散模型中的迭代采样可能由于暴露偏差导致误差积累。去噪器是在与噪声插值的真实图像上训练的，但在推理时，其输入是由之前步骤中自身不完美的预测构成的，这是其在训练时从未见过的。这可能是扩散模型在没有额外帮助（如无分类器引导（CFG））的情况下常常表现不佳的原因之一。我们为扩散模型引入了一种校正步骤，旨在减少训练与推理之间的差距。校正器以基础模型的预测作为输入，并针对相同的真实目标进行训练。这使得校正成为一个明确的学习任务，训练输入由基础模型提供。在采样过程中，在每个去噪步骤中，我们迭代地细化去噪器的清晰样本预测。这使我们能够在中间去噪步骤中纠正预测，防止早期伪影持续到最终样本中。尽管校正器仅在基础去噪器的预测上进行训练，但在推理时可以反复应用于其自身的输出，以获得更好的结果。这个简单的框架显著提高了样本质量，同时在很大程度上保持了多样性。我们在 DiT、SiT 和 JiT 上评估了我们的方法。我们的方法在无指导生成的 FID 分数上显著优于之前报告的结果，特别是在无条件生成方面，在没有指导的情况下生成高质量样本尤其具有挑战性。我们的方法在性能上与无分类器引导（CFG）相当，并且将校正与 CFG 结合进一步提高了多样性，减少了 CFG 通常出现的过饱和现象。最后，我们展示了校正的额外推理成本可以通过将精炼的预测蒸馏回基础扩散模型来摊销，从而在不增加额外推理时间开销的情况下保留质量提升。

View on arXiv Download PDF AI Translation

cs.AI / 5 / 2610.14146

Long-Horizon-Terminal-Bench 2.0: Automating Diverse Task Construction and Measuring Sustained Agent Progress

长时间地平线终端基准 2.0：自动化多样化任务构建与持续代理进展测量

Li, Zongxia, Li, Zhongzhi, Shi, Yucheng, Wang, Ruhan, Yang, Junyao, Huang, Chengsong, Liu, Zhichao, Wu, Xiyang, Li, Anhao, Yu, Yue, Shang, Ergan, Liu, Fuxiao, Zheng, Tong, Liu, Ninghao, Sun, Lichao, Mi, Haitao, Liang, Leowei

Abstract

Long-horizon terminal benchmarks measure whether an agent can sustain progress through extended workflows with process rewards, not only whether its final state passes a binary check. We present Long-Horizon-Terminal-Bench 2.0 (LHTB 2.0), an extension of Long-Horizon-Terminal-Bench that preserves its 46 original tasks while correcting task and evaluation-infrastructure defects and extending 31 tasks. LHTB 2.0 contains 77 tasks across 13 domains and eight task types. The extended tasks are produced through a systematic reconstruction-task pipeline. We start from a working scientific or engineering artifact, then use a parameterized reference implementation to generate multiple scenarios, reveal six as behavioral evidence, hold out three for evaluation, and use dense progress ladders to reward partial progress while preventing memorization of public outputs. We also extend the evaluation horizon from 90 minutes to three hours and record verifier checkpoints every 30 minutes, enabling direct measurement of sustained progress within a run. Across 1,145 scored model-task runs from 15 frontier models, GPT-6-astra leads with dense mean reward 0.69 and 37 tasks at R at least 0.95, followed by GPT-5.6-sol with dense mean reward 0.67 and 32 tasks at the same threshold. On the 30 reconstruction tasks, binary verifier scores average 0.221, whereas dense ladder scores average 0.653, showing that outcome-only evaluation discards substantial intermediate progress. LHTB 2.0 provides a broader, hardened, and process-aware testbed for studying reliable agent execution over hours.

Chinese Translation

长时间地平线终端基准用于衡量代理是否能够通过过程奖励在延续的工作流程中保持进展，而不仅仅是其最终状态是否通过二元检查。我们提出了长时间地平线终端基准 2.0 (Long-Horizon-Terminal-Bench 2.0, LHTB 2.0)，这是对长时间地平线终端基准的扩展，保留了其46个原始任务，同时修正了任务和评估基础设施的缺陷，并扩展了31个任务。LHTB 2.0包含来自13个领域和八种任务类型的77个任务。扩展任务是通过系统的重建任务管道生成的。我们从一个有效的科学或工程工件开始，然后使用参数化的参考实现生成多个场景，揭示六个作为行为证据，保留三个用于评估，并使用密集的进展阶梯奖励部分进展，同时防止对公共输出的记忆。在评估时间上，我们将评估时间从90分钟延长至三小时，并每30分钟记录验证器检查点，从而能够直接测量在一次运行中的持续进展。在来自15个前沿模型的1,145个评分模型任务运行中，GPT-6-astra以0.69的密集平均奖励和37个任务在R至少为0.95的情况下领先，其次是GPT-5.6-sol，密集平均奖励为0.67，且在同一阈值下有32个任务。在30个重建任务中，二元验证器分数平均为0.221，而密集阶梯分数平均为0.653，显示仅基于结果的评估忽略了大量的中间进展。LHTB 2.0为研究可靠的代理执行提供了一个更广泛、更坚固且关注过程的测试平台，持续时间可达数小时。

View on arXiv Download PDF AI Translation

cs.AI / 6 / 2610.14149

VETNET: Assembling Benchmark Datasets for AI-Driven Veterinary Innovation

VETNET：为人工智能驱动的兽医创新构建基准数据集

Koebel, Katherine J., Ivanek, Renata, Basran, Parminder, Sun, Jennifer, Cazer, Casey, Anil, Gayatri, Banzato, Tommaso, Barton, Liz, Brown, Dottie Cimino, Bukowski, Leszek, Castelhano, Marta, Coghlan, Simon, De Sa, Christopher, Dutta, Nikki, Ghali, Hala, Giordano, Julio, Glass, Joshua, Hogg, David C., Hollingshead, Nicholas, Kiehl, Adam, Liu, Ziang, Meduri, Monica, Mohapatra, Param, Nayyer, Kim, Nelson, Amelia, Noble, Peter-John Mantyla, Pinard, Christopher, Rabindran, Beryl, Ross, Scott, Seccia, Ron, Sheehan, Moira, Wang, Fei, Webb, Tracy L., Weinberger, Kilian Q., Whitney, Heather M., Xin, Chen

Abstract

VETNET represents an ecosystem of live benchmark datasets and evaluation frameworks designed to accelerate AI-driven innovation in animal science and veterinary medicine through standardized, high-quality datasets and multi-institutional collaboration across academic and non-academic sectors. This framework addresses the gap between AI advances in human medicine and the fragmented state of veterinary AI development.

Chinese Translation

VETNET 代表了一个由实时基准数据集和评估框架组成的生态系统，旨在通过标准化的高质量数据集和跨学术及非学术领域的多机构合作，加速动物科学和兽医学中的人工智能驱动创新。该框架旨在填补人类医学中人工智能进展与兽医人工智能发展碎片化状态之间的差距。

View on arXiv Download PDF AI Translation

cs.AI / 7 / 2610.14180

Unconstrained CTI-REALM with Open-Weights

开放权重的无约束 CTI-REALM

Garg, Sahil, Gottumukkala, Ruhi, Agarwal, Vishal, Vishwakarma, Shubham

Abstract

Organisations that want language-model agents for security work face a choice between open-weight models, which they can run and audit themselves, and far more expensive closed frontier models. We study that choice on CTI-REALM (arXiv:2603.13517), a benchmark in which an agent turns a threat-intelligence report into a working detection by querying a live telemetry store. We evaluate twenty-six models, nineteen open-weight and seven closed, under a common harness, scorer and panel of LLM judges. Our main finding is that an open-weight model can match the strongest closed models at less than a tenth of their cost, provided it is given enough turns. The benchmark's published limit on agent messages is a confound: a task that exhausts the limit scores zero on every checkpoint, including those it had already passed, so results at that limit partly measure which models finish in time. At the published limit the leading open-weight and closed models cannot be distinguished. With the limit relaxed, the models it was truncating recover, and the highest rewards belong to open-weight models. Among the leading models, price is a poor guide to capability: models with similar rewards differ in cost by more than an order of magnitude. Because the benchmark's checkpoint scores largely record what an agent did or mentioned, we also introduce stage-level metrics that score what the agent submits, by executing its detection queries in the telemetry store and judging its outputs against the answer key. Rather than combining them into one more overall reward, we show that different models lead different stages, which suggests routing each stage of the workflow to the model that performs it best rather than choosing a single model by its overall score. We also report how many tokens each model generates, and find that the share of output spent on thinking does not track the reward.

Chinese Translation

希望为安全工作提供语言模型代理的组织面临选择：开放权重模型（可以自行运行和审计）与成本更高的封闭前沿模型。我们在 CTI-REALM（arXiv:2603.13517）这一基准上研究了这一选择，该基准中，代理通过查询实时遥测存储，将威胁情报报告转化为有效检测。我们在一个共同的框架下评估了二十六个模型，其中十九个为开放权重模型，七个为封闭模型，并由 LLM 评审小组进行评分。我们的主要发现是，开放权重模型在成本不到封闭模型十分之一的情况下，可以与最强的封闭模型相匹敌，前提是它有足够的回合数。基准中对代理消息的发布限制是一个混淆因素：一个耗尽限制的任务在每个检查点的得分均为零，包括那些它已经通过的检查点，因此在该限制下的结果部分衡量了哪些模型能及时完成。在发布的限制下，领先的开放权重模型和封闭模型无法区分。随着限制的放宽，受限的模型恢复了性能，最高奖励归属于开放权重模型。在领先模型中，价格与能力之间的关系较差：具有相似奖励的模型在成本上相差超过一个数量级。由于基准的检查点得分主要记录了代理所做或提到的内容，我们还引入了阶段级指标，通过在遥测存储中执行其检测查询并根据答案键评判其输出，来评分代理提交的内容。我们并未将它们合并为一个更全面的奖励，而是展示了不同模型在不同阶段的表现，这表明将工作流的每个阶段路由到表现最佳的模型，而不是通过整体得分选择单一模型，可能是更优的选择。我们还报告了每个模型生成的令牌数量，并发现用于思考的输出比例与奖励并不相关。

View on arXiv Download PDF AI Translation

cs.AI / 8 / 2610.14186

Visualizing the World Model in an Agent's Value Function

可视化代理的价值函数中的世界模型

Naderi, Shawheen, Akherati, Fatemeh, Chen, Yubei

Abstract

Could model-free reinforcement learning algorithms actually be learning a model of the environment? In this work, we apply counterfactual interventions which hold everything about a state fixed, besides a targeted change, to explore if the value function of a Proximal Policy Optimization agent forms a world model of the Crafter environment. Despite only being trained to optimize reward, we discover that the value function learns concepts such as the rules governing when a diamond is collectible, water and food having an impact on health, heuristics regarding inventory, and more. Finally, we discuss the safety implications of our visualizations and some shortcomings of current reinforcement learning agents that will prevent us from building truly advanced machine intelligence.

Chinese Translation

无模型强化学习算法是否真的在学习环境模型？在本研究中，我们应用了反事实干预，固定状态的所有内容，仅针对特定变化进行探讨，以探索近端策略优化（Proximal Policy Optimization）代理的价值函数是否形成了Crafter环境的世界模型。尽管仅被训练以优化奖励，我们发现价值函数学习了诸如何时可以收集钻石的规则、水和食物对健康的影响、库存的启发式规则等概念。最后，我们讨论了我们可视化结果的安全性影响以及当前强化学习代理的一些缺陷，这些缺陷将阻碍我们构建真正先进的机器智能。

View on arXiv Download PDF AI Translation

cs.AI / 9 / 2610.14190

Cross-Modal Latent Influence: Patch-Level Training Sample Attribution for Vision-Language Models

跨模态潜在影响：视觉-语言模型的补丁级训练样本归因

Yao, Laura, Yu, Shixing, Peng, Jie, Ghosal, Promit, Gan, Kyra, Chen, Tianlong

Abstract

Explaining a vision--language model (VLM) prediction requires tracing training-example influence to the multimodal evidence that mediates it, yet existing attribution methods largely operate only at the sample level. Recent latent-mediated influence methods provide feature-level attribution in language models, but lack the multimodal and spatial structure needed to ground influence in VLMs. We introduce MOSAIC (Multimodal Orthogonalized Sparse Autoencoders for Influence Characterization). It learns a shared sparse image--text latent space whose concepts are 1) sparse within each modality, 2) supported on contiguous image regions, 3) aligned to the phrases that describe them, and 4) approximately decorrelated. It then applies EK-FAC to decompose training-example influence through these concepts and localize their contributions to specific text phrases and image regions. We evaluate MOSAIC on Qwen and Gemma across captioning, chest X-ray understanding, and visual question answering. The learned features are more localized and coherent without degrading task behavior, and ablating influence-ranked latents changes model outputs more than ablating latents ranked by activation or frequency. We find that visual evidence can substantially reorganize training-data attribution while preserving more of the shared latent structure. MOSAIC thus connects where model behavior comes from with what multimodal evidence carries that influence.

Chinese Translation

解释视觉-语言模型（VLM）的预测需要追踪训练样本对多模态证据的影响，而现有的归因方法主要仅在样本级别上操作。最近的潜在介导影响方法为语言模型提供了特征级别的归因，但缺乏在VLM中将影响与多模态和空间结构相结合的能力。我们提出了MOSAIC（多模态正交稀疏自编码器用于影响特征化）。它学习一个共享的稀疏图像-文本潜在空间，其概念具有以下特征：1）在每种模态内稀疏，2）支持在连续的图像区域上，3）与描述它们的短语对齐，以及4）近似去相关。然后，它应用EK-FAC通过这些概念分解训练样本的影响，并将其贡献定位到特定的文本短语和图像区域。我们在Qwen和Gemma上评估MOSAIC，涉及图像描述、胸部X光理解和视觉问答。学习到的特征更加局部化和一致，而不会降低任务表现，且消融影响排名的潜在变量对模型输出的影响大于消融按激活或频率排名的潜在变量。我们发现视觉证据可以显著重组训练数据的归因，同时保留更多共享的潜在结构。因此，MOSAIC将模型行为的来源与携带该影响的多模态证据联系起来。

View on arXiv Download PDF AI Translation

cs.AI / 10 / 2610.14209

Should SLMs be in the Driver's Seat? A Study of Local Models as Interpreters in Cloud-Written Agent Programs

SLM是否应该掌控主动权？云端编写代理程序中本地模型作为解释器的研究

Radovič, Boris, Canini, Marco, Pejović, Veljko

Abstract

The most accurate agents executing user-provided tasks are driven by large, cloud-hosted models, which solve them by reading the private data of the user. Agents powered by models small enough to run on edge devices keep that data local, but yield far worse results. A third option is to have cloud models write a program that devices execute: the data stays local, but as the program is written before any data is seen, it fails whenever a step depends on interpreting content that appears only at run time. We explore an intermediate architecture, in which the cloud-written program defers bounded questions over user data to a small local model through an ordinary function call. On AppWorld this improves the success rate of a code-only agent in all six combinations of backbone model and test partition we evaluate, by up to 14.9pp, recovering a substantial part of the gap to a cloud-hosted ReAct loop while keeping user data on the device.

Chinese Translation

执行用户提供任务的最准确代理由大型云托管模型驱动，这些模型通过读取用户的私有数据来解决任务。由足够小以在边缘设备上运行的模型驱动的代理则将数据保留在本地，但结果往往远不如前者。第三种选择是让云模型编写一个由设备执行的程序：数据保持本地，但由于程序在任何数据被查看之前就已编写，因此当某一步骤依赖于在运行时出现的内容的解释时，它会失败。我们探索了一种中间架构，其中云编写的程序通过普通函数调用将对用户数据的有限问题推迟到一个小型本地模型。在AppWorld上，这种方法提高了代码代理在我们评估的六种骨干模型和测试分区组合中的成功率，最高可达14.9个百分点，恢复了与云托管ReAct循环之间的显著差距，同时保持用户数据在设备上。

View on arXiv Download PDF AI Translation

cs.AI / 11 / 2610.14224

Harness^2: Recursive Agent Harnessing for an Open World

Harness^2：开放世界中的递归智能体利用

Xu, Ruiyao, Chen, Yanfei, CuiZhu, Zhongying, Mishra, Bhavana Dalvi, Ming, Yifei, Yu, Han, Han, Rujun, Lee, Chen-Yu, Pfister, Tomas

Abstract

LLM agents' capabilities depend not only on their underlying models but also on their execution harnesses. Recursive self-improvement (RSI) offers a promising approach to automatically evolve these harnesses. However, deploying RSI in open-world settings still faces two critical challenges. First, open-world tasks are typically unseen during the harness design phase and novel at test time, leaving no ground-truth verifiers to reliably guide online harness edits. Second, the RSI process remains highly task-specific and fails to generalize to novel scenarios. We introduce Harness^2, a plug-and-play, recursive agent harnessing framework that addresses these challenges through two connected levels of recursion. Task-level recursion refines the current task's harness using a contrastive Propose-Probe-Compose step that reliably guides edits without ground-truth verifiers, and can be scaled further via parallel or sequential recursion. Harnessing-level recursion then extracts experience from these edits to update a continual improvement harness, enabling the agent to better adapt to future tasks. Across three professional benchmarks and four model-harness configurations, a single step consistently improves GLM-5 and Gemini 3.5 Flash over their base harnesses, yielding gains of up to 11.5 points on WorkBuddy-Bench, 4.5 points on JobBench, and 11.8 percentage points on LAB, with further gains as the task-level recursion budget increases.

Chinese Translation

大型语言模型（LLM）智能体的能力不仅依赖于其基础模型，还依赖于其执行框架。递归自我改进（RSI）提供了一种有前景的方法来自动演化这些框架。然而，在开放世界环境中部署RSI仍面临两个关键挑战。首先，开放世界任务在框架设计阶段通常是未见过的，并且在测试时是新颖的，缺乏可靠的真实验证者来指导在线框架编辑。其次，RSI过程高度任务特定，无法推广到新场景。我们提出了Harness^2，一个即插即用的递归智能体利用框架，通过两个相互关联的递归层次来解决这些挑战。任务级递归使用对比的提议-探测-组合步骤来细化当前任务的框架，能够在没有真实验证者的情况下可靠地指导编辑，并且可以通过并行或顺序递归进一步扩展。利用级递归则从这些编辑中提取经验，以更新持续改进框架，使智能体能够更好地适应未来任务。在三个专业基准和四种模型-框架配置中，单步操作始终改善GLM-5和Gemini 3.5 Flash的基础框架，在WorkBuddy-Bench上提高了最多11.5分，在JobBench上提高了4.5分，在LAB上提高了11.8个百分点，并且随着任务级递归预算的增加，进一步获得了提升。

View on arXiv Download PDF AI Translation

cs.AI / 12 / 2610.14277

Echo: Error-Compensated 4-Bit Tensor-Parallel Communication for LLM Training and Inference

Echo：用于大规模语言模型训练和推理的误差补偿4位张量并行通信

Xie, Xingyu, Xie, Xiaoyong, Lin, Zhouchen, Toh, Kim-Chuan

Abstract

Tensor parallelism ends every layer with an all-reduce that sits between two dependent matrix multiplications, so its cost cannot be hidden and grows as compute outpaces interconnect bandwidth. Sending 4-bit instead of 16-bit values would cut this cost by nearly 4x, but the all-reduce must add the values it carries, and rounding the sum injects an error at every call. Model quality depends on how these errors accumulate over layers, yet existing methods treat each call on its own. We present Echo, a 4-bit all-reduce for training and inference that controls the accumulated error. Echo moves data in 4 bits and sums in high precision, so each value is quantized at most twice per call for any number of GPUs. The summation reveals each injected error, and Echo feeds it back into later calls, so old errors cancel instead of accumulating. We prove that the accumulated error stays bounded independently of depth. On A100 GPUs, Echo is 2.09x faster than the 16-bit NCCL all-reduce on 1 GB messages. It speeds up the time to first token by up to 1.16x and the training step by up to 1.09x on an idle interconnect, and by up to 1.44x and 1.38x as background traffic grows. It keeps the average accuracy over seven benchmarks within 0.4 points of 16-bit inference on Qwen3-8B. In 10B-token pretraining, its final loss is less than 0.9% above that of 16-bit training. Echo needs no FP8 or FP4 hardware.

Chinese Translation

张量并行性在每一层结束时都需要进行一次全规约，这个全规约位于两个依赖的矩阵乘法之间，因此其成本无法隐藏，并且随着计算能力超过互连带宽而增加。发送4位而不是16位的值将使这一成本减少近4倍，但全规约必须对其携带的值进行求和，而对和的四舍五入在每次调用中都会引入误差。模型质量依赖于这些误差在层间的累积，然而现有方法将每次调用视为独立的。我们提出了Echo，一种用于训练和推理的4位全规约，它控制累积误差。Echo以4位进行数据传输，并以高精度进行求和，因此每个值在任何数量的GPU上每次调用最多被量化两次。求和过程揭示了每个注入的误差，Echo将其反馈到后续调用中，从而使旧的误差相互抵消而不是累积。我们证明了累积误差在深度上保持有界。在A100 GPU上，Echo的速度比16位NCCL全规约在1 GB消息上快2.09倍。在空闲互连上，它将首次令牌的时间加速最多1.16倍，训练步骤加速最多1.09倍，而在背景流量增加时，分别加速最多1.44倍和1.38倍。它在七个基准测试中的平均准确率保持在16位推理的0.4分以内。在10B-token的预训练中，其最终损失比16位训练高出不到0.9%。Echo不需要FP8或FP4硬件。

View on arXiv Download PDF AI Translation

cs.AI / 13 / 2610.14294

Controllable Diverse Sampling through Pullback Directions

通过回拉方向实现可控的多样性采样

Kenfouo, Sylvain Dehayem, Zweig, Aaron, Zhang, Mingxuan, Azizi, Elham

Abstract

Conditional diffusion models can generate high-quality images that align closely with a given prompt. However, samples generated from different random seeds may remain visually similar, limiting creative exploration. Existing approaches can increase variation, but often require additional optimization or sampling overhead, while stronger perturbations may introduce distortions that degrade image quality or prompt alignment. We propose a training-free method that leverages the local pullback geometry induced by the mapping from the conditioning-embedding space to the image space to identify directions of maximal output sensitivity. By steering different sampling trajectories along these directions, we encourage distinct generations while maintaining consistency with the conditioning signal. Experiments on MS-COCO and BrushNet show the efficacy of the method in improving diversity on prompt-conditioned image generation and inpainting.

Chinese Translation

条件扩散模型能够生成与给定提示高度一致的高质量图像。然而，从不同随机种子生成的样本可能在视觉上保持相似，从而限制了创意探索。现有方法可以增加变异性，但通常需要额外的优化或采样开销，而更强的扰动可能会引入失真，降低图像质量或提示一致性。我们提出了一种无训练的方法，该方法利用从条件嵌入空间到图像空间的映射所诱导的局部回拉几何来识别最大输出敏感度的方向。通过沿这些方向引导不同的采样轨迹，我们鼓励生成不同的图像，同时保持与条件信号的一致性。在 MS-COCO 和 BrushNet 上的实验表明，该方法在提示条件图像生成和修复中的多样性提升效果显著。

View on arXiv Download PDF AI Translation

cs.AI / 14 / 2610.14307

ConicFlow: constrained Inference-time optimal control for flow matching models

锥形流：流匹配模型的约束推断时间最优控制

Khalafi, Shervin, Fallah, Kasra, Cho, Sung, Ding, Dongsheng, Anderson, James, Ribeiro, Alejandro

Abstract

We propose ConicFlow, which frames constrained sampling from a flow matching model at inference time as an optimal control problem that balances deviation from the pretrained trajectory against reward, subject to hard terminal constraints. We discretize this problem and adapt receding-horizon control, optimizing only a short window of the trajectory at each step for computational tractability. We address the resulting sequence of receding-horizon subproblems with a receding-horizon primal-dual algorithm. Its dual multipliers capture the difficulty of satisfying each constraint and are refined throughout sampling, steering generation toward feasible final samples. We further propose a terminal-estimate variant that avoids accumulating local truncation errors by evaluating the reward and constraints at the predicted terminal point rather than at the end of the current horizon. Under standard regularity assumptions, we establish finite-horizon guarantees on suboptimality. Under convex-affine structure, we derive a sharper bound that scales linearly with omitted horizon length, and we show that the terminal-estimate variant is provably tighter when the flow dynamics change slowly. Experiments in both low- and high-dimensional settings with a recent large-scale flow matching model show that ConicFlow achieves high reward and strong constraint satisfaction while deviating less from the pretrained flow than recent baselines.

Chinese Translation

我们提出了锥形流（ConicFlow），将推断时从流匹配模型进行约束采样的问题框架化为一个最优控制问题，该问题在预训练轨迹的偏离与奖励之间进行平衡，并受到严格的终端约束。我们对该问题进行离散化，并采用递归时间控制（receding-horizon control），在每一步仅优化轨迹的短窗口以提高计算的可行性。我们使用递归时间原始-对偶算法（receding-horizon primal-dual algorithm）来解决由此产生的递归时间子问题序列。其对偶乘子捕捉了满足每个约束的难度，并在采样过程中不断优化，引导生成朝向可行的最终样本。我们进一步提出了一种终端估计变体，通过在预测的终端点而非当前时间窗口的结束时评估奖励和约束，避免了累积局部截断误差。在标准的正则性假设下，我们建立了关于次优性的有限时间保证。在凸仿射结构下，我们推导出一个更为精确的界限，该界限与省略的时间窗口长度呈线性关系，并且我们证明当流动动态变化缓慢时，终端估计变体的界限在理论上更为紧致。在低维和高维环境中进行的实验，使用了最近的大规模流匹配模型，结果显示锥形流（ConicFlow）在偏离预训练流的同时，实现了高奖励和强约束满足，优于最近的基线方法。

View on arXiv Download PDF AI Translation

cs.AI / 15 / 2610.14334

Which Metadata Matters for LLM Tool Selection in MCP? Evidence from Controlled Factor-Level Interventions

在MCP中，哪些元数据对LLM工具选择至关重要？来自控制因素级干预的证据

Li, Xiaofan, Qiao, Fengchun, Gao, Xing

Abstract

In Model Context Protocol (MCP) systems, LLMs select tools from metadata rather than from their implementations. Names and descriptions summarize what a tool does, while the parameter schema specifies the fields a tool accepts and how they are organized. Our analysis of metadata for 367,174 MCP tools reveals substantial schema variation among functionally similar tools. Many of these tools share names and similar descriptions but expose different parameter schemas. This raises a key question: can parameter-schema differences influence tool selection before any call is made? We study this question through controlled metadata interventions across five LLMs and five tasks. Each comparison changes exactly one metadata factor (server name, tool name, description, or parameter schema) while keeping the others fixed. We balance tool positions and use conditional logistic regression to estimate position-adjusted effects. We find that: (1) parameter-schema changes produce the largest and most consistent selection shifts among the tested interventions. Models tend to favor separate top-level parameters over grouped or composite forms. (2) Descriptions that explain parameters also tend to increase selection, while name effects are weaker and less consistent. (3) Targeted comparisons show that schema effects depend on the specific parameter names and on how the request represents the same information. When only one schema satisfies an explicit request requirement, all five models select the compatible tool despite position preferences. These results show that parameter schemas influence not only how tools are called, but also whether they are selected. For MCP developers and evaluators, schema design and tool ordering can therefore affect which implementation is invoked even when functionality is unchanged.

Chinese Translation

在模型上下文协议（MCP）系统中，LLM从元数据中选择工具，而不是从其实现中选择。工具的名称和描述总结了工具的功能，而参数模式则指定了工具接受的字段及其组织方式。我们对367,174个MCP工具的元数据分析揭示了功能相似的工具之间存在显著的模式差异。这些工具中许多共享名称和相似的描述，但却暴露出不同的参数模式。这引发了一个关键问题：参数模式的差异是否会在任何调用之前影响工具选择？我们通过对五个LLM和五个任务进行控制元数据干预来研究这个问题。每次比较仅改变一个元数据因素（服务器名称、工具名称、描述或参数模式），而保持其他因素不变。我们平衡工具位置，并使用条件逻辑回归来估计位置调整后的效果。我们发现：（1）参数模式的变化在测试的干预中产生了最大且最一致的选择变化。模型倾向于优先选择独立的顶层参数，而不是分组或复合形式。（2）解释参数的描述也倾向于增加选择，而名称的影响较弱且不一致。（3）针对性的比较表明，模式效果依赖于特定的参数名称以及请求如何表示相同的信息。当只有一个模式满足明确的请求要求时，尽管存在位置偏好，所有五个模型仍然选择兼容的工具。这些结果表明，参数模式不仅影响工具的调用方式，还影响它们是否被选择。因此，对于MCP开发者和评估者而言，模式设计和工具排序可能会影响在功能不变的情况下调用哪个实现。

View on arXiv Download PDF AI Translation

cs.AI / 16 / 2610.14356

Actionable Meta-Requirements for AI Robustness Auditing

可操作的人工智能鲁棒性审计元需求

Owotogbe, Joshua, Nasiri, Roya, Kumara, Indika, Tamburri, Damian Andrew, Heuvel, Willem-Jan van den

Abstract

AI-enabled systems must remain dependable when the conditions under which they operate change (e.g., due to input perturbations, distribution shifts, or changes in external dependencies). This makes robustness auditing of such systems important for judging whether the available evidence is sufficient to support a defined robustness conclusion. However, designing an AI robustness auditing framework is challenging. A key unresolved issue is determining the requirements such a framework should support; these requirements have not yet been consolidated. We address this gap through meta-requirements: higher-order requirements specifying what a class of auditing frameworks should support. In design science research, literature and domain experts are common sources of meta-requirements; accordingly, we derive the initial set from literature and validate and refine it with practitioners. Our systematic literature review retained 74 of 2,890 records and derived 43 meta-requirements across 16 categories. We evaluated them with 128 practitioners, obtaining 5,504 ratings and 586 substantive comments. Overall, 81.20% of ratings indicated agreement or strong agreement. In particular, 41 of the 43 literature-derived meta-requirements were classified as strongly supported. Practitioner feedback refined the requirements and identified an additional requirement concerning external components and supply chain dependencies. The final set comprises 44 meta-requirements across 16 categories, of which 43 received quantitative validation. These meta-requirements can help organizations implement AI robustness auditing frameworks, assess existing frameworks for completeness, and refine them.

Chinese Translation

人工智能驱动的系统在操作条件变化时必须保持可靠性（例如，由于输入扰动、分布变化或外部依赖关系的变化）。这使得对这些系统进行鲁棒性审计变得重要，以判断现有证据是否足以支持定义的鲁棒性结论。然而，设计一个人工智能鲁棒性审计框架是具有挑战性的。一个关键的未解决问题是确定该框架应支持的需求；这些需求尚未得到整合。我们通过元需求来填补这一空白：元需求是指明一类审计框架应支持的高阶需求。在设计科学研究中，文献和领域专家是元需求的常见来源；因此，我们从文献中推导出初始集，并与从业者进行验证和完善。我们的系统文献综述保留了2,890条记录中的74条，并在16个类别中推导出43个元需求。我们与128名从业者进行了评估，获得了5,504条评分和586条实质性评论。总体而言，81.20%的评分表示同意或强烈同意。特别是，43个文献推导的元需求中有41个被分类为强烈支持。从业者的反馈完善了这些需求，并识别出一个关于外部组件和供应链依赖关系的额外需求。最终集包含16个类别中的44个元需求，其中43个得到了定量验证。这些元需求可以帮助组织实施人工智能鲁棒性审计框架，评估现有框架的完整性，并对其进行完善。

View on arXiv Download PDF AI Translation

cs.AI / 17 / 2610.14360

When Pleasing the User Compromises Care: AI Sycophancy in Caregiver-Mediated Healthcare

当取悦用户妨碍关怀：护理者介导的医疗中的人工智能谄媚现象

Mishra, Priyadarsi, Beniwal, Arjun, Yi, Ziyu, Ding, Yu, Liu, Sheng

Abstract

Large language model (LLM) alignment often assumes that the user interacting with a model is also the primary beneficiary of its behavior. Yet in many consequential settings, one person directs the system while another bears the consequences. We study this problem in caregiver-mediated healthcare, where parents, pet owners, and caregivers make decisions for children, animals, and older adults. We introduce an evidence-grounded benchmark spanning veterinary, pediatric, and older-adult care, and test whether AI advisors maintain appropriate care when users express clinically inferior preferences. Holding patient facts fixed, we vary stated care preferences and communication styles and evaluate both initial recommendations and multi-turn follow-up. Across frontier and open-weight models, inferior preferences substantially increase switching from appropriate to inappropriate care beyond neutral conversational drift and increase agreement with the requested action, while communication style alone has a smaller effect. These failures accumulate across turns, including on cases models initially answer correctly, and often carry moderate or severe potential harm. In matched eldercare cases, the effect of an inferior preference is similar whether it is expressed by the patient or by a caregiver. Together with models' limited tendency to elicit patient preferences in caregiver conversations, this suggests that current models are highly sensitive to expressed user preferences but less sensitive to whose preferences are represented and who bears the consequences. Our results expose a broader limitation of user-centric alignment in settings with multiple stakeholders.

Chinese Translation

大型语言模型（LLM）的对齐通常假设与模型互动的用户也是其行为的主要受益者。然而，在许多重要的场景中，一个人指导系统，而另一个人承担后果。我们在护理者介导的医疗中研究这个问题，在这种情况下，父母、宠物主人和护理者为儿童、动物和老年人做出决策。我们引入了一个基于证据的基准，涵盖兽医、儿科和老年护理，并测试当用户表达临床上较差的偏好时，人工智能顾问是否能够维持适当的护理。在固定患者事实的情况下，我们改变了所表达的护理偏好和沟通风格，并评估了初始推荐和多轮跟进。在前沿模型和开放权重模型中，较差的偏好显著增加了从适当护理转向不适当护理的情况，超出了中性对话漂移的范围，并增加了与请求行动的一致性，而沟通风格本身的影响较小。这些失败在多轮对话中累积，包括在模型最初正确回答的案例中，且通常带来中度或严重的潜在伤害。在匹配的老年护理案例中，无论偏好是由患者还是护理者表达，其影响是相似的。结合模型在护理者对话中有限的引导患者偏好的倾向，这表明当前模型对表达的用户偏好高度敏感，但对谁的偏好被代表以及谁承担后果的敏感性较低。我们的结果揭示了在涉及多个利益相关者的环境中，用户中心对齐的更广泛局限性。

View on arXiv Download PDF AI Translation

cs.AI / 18 / 2610.14381

ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence

ACED-Bench：评估大型语言模型代理如何获取和利用因果证据

Chen, Yiwei, Weerasinghe, Amila, Biswal, Siddharth, Rajkumar, Utkrisht, Boz, Olcay, Berg-Kirkpatrick, Taylor, Yu, Rose

Abstract

Large language models are increasingly used as scientific assistants, yet we still lack reliable ways to evaluate whether they can investigate a system rather than merely answer a prompt. A useful evaluation must let agents choose experiments, interpret finite evidence, and make causal decisions, while keeping the environment coherent across repeated observations and interventions and preserving objective ground truth. We introduce ACED-BENCH (Active Causal Evidence and Decision Benchmark), a neuro-symbolic benchmark built around this requirement. The key idea is to separate the readable scientific surface from the formal causal world: LLMs generate natural-language scenarios and task text, while hidden probabilistic graphical models define the causal graph, data-generating process, intervention semantics, validation checks, and oracle answers. This yields scientific worlds that are natural for agents to investigate but exact enough for scalable, programmatic evaluation. The main leaderboard ACED-CORE contains 180 tasks: 60 set-valued causal ancestry/descendancy questions and 120 constrained causal-decision tasks. We also report ACED-QUAL, a 150-question yes/no qualification check using the same hidden-PGM interface. We evaluate seven contemporary LLMs and two earlier-generation models across three active agent designs. ACED-QUAL checks whether agents can use the interface to draw conclusions from evidence, while ACED-CORE tests whether they can decide what evidence to collect and when to commit to an answer. Evidence-ledger analysis localizes failures within the investigation: agents test the wrong intervention, lose relevant evidence, or continue after sufficient evidence has appeared. These results suggest that reliable AI co-scientists need explicit support for experiment design, evidence retention, answer commitment, and stopping.

Chinese Translation

大型语言模型越来越多地被用作科学助手，但我们仍然缺乏可靠的方法来评估它们是否能够调查一个系统，而不仅仅是回答一个提示。有效的评估必须让代理选择实验、解释有限的证据并做出因果决策，同时保持环境在重复观察和干预中的一致性，并保留客观的真实情况。我们引入了 ACED-BENCH（主动因果证据与决策基准），这是一个围绕这一要求构建的神经符号基准。其关键思想是将可读的科学表面与正式的因果世界分开：大型语言模型生成自然语言场景和任务文本，而隐藏的概率图模型定义因果图、数据生成过程、干预语义、验证检查和神谕答案。这产生了适合代理调查的科学世界，同时又足够精确，以便进行可扩展的程序化评估。主要排行榜 ACED-CORE 包含 180 个任务：60 个集合值因果祖先/后代问题和 120 个受限因果决策任务。我们还报告了 ACED-QUAL，这是一个使用相同隐藏概率图模型接口的 150 个问题的是/否资格检查。我们在三种主动代理设计中评估了七个当代大型语言模型和两个早期模型。ACED-QUAL 检查代理是否能够使用接口从证据中得出结论，而 ACED-CORE 测试它们是否能够决定收集哪些证据以及何时承诺答案。证据账本分析定位了调查中的失败：代理测试了错误的干预，丢失了相关证据，或者在足够证据出现后继续进行。这些结果表明，可靠的人工智能共同科学家需要在实验设计、证据保留、答案承诺和停止方面提供明确的支持。

View on arXiv Download PDF AI Translation

cs.AI / 19 / 2610.14382

ANASSA: From High-Level Framework to Low-Level Architecture for Spatial Agentic AI

ANASSA：从高层框架到低层架构的空间代理智能AI

Papantoniou, Constantinos, Hilton, Brian

Abstract

Agentic AI systems operating over geospatial data require more than autonomous reasoning and tool use; they must also support explicit orchestration, spatial validation, governance, provenance, failure handling, and human decision authority. Prior work introduced ANASSA, an agentic AI orchestration framework for spatial intelligence that established high-level components and responsibility boundaries for governed agentic execution (arXiv:2609.14824). However, the framework did not specify how those high-level responsibilities should be systematically decomposed into an implementation-facing low-level architecture. This paper addresses that gap by applying Design Science Research (DSR) to refine the ANASSA framework from high-level architectural (HLA) responsibilities into a low-level architecture (LLA). The resulting design preserves the responsibility boundaries of the foundational framework while translating them into executable architectural structures, including 47 logical roles, versioned task graphs, capability-based agent selection, typed contracts, explicit runtime states, spatial validation mechanisms, governance controls, failure and recovery paths, and provenance records. The architecture is demonstrated through an end-to-end execution trace and analytically evaluated for requirements traceability, internal consistency, responsibility preservation, and technology independence. The resulting high-level-to-low-level design provides an implementation-facing bridge between the ANASSA framework and future system implementations, defining how governed spatial agentic workflows can be made inspectable, reconstructable, and testable without binding the architecture to a specific AI model, geospatial platform, communication protocol, or deployment environment.

Chinese Translation

基于地理空间数据的代理智能AI系统不仅需要自主推理和工具使用，还必须支持明确的编排、空间验证、治理、来源、故障处理和人类决策权。之前的研究介绍了ANASSA，一个用于空间智能的代理智能AI编排框架，建立了治理代理执行的高层组件和责任边界（arXiv:2609.14824）。然而，该框架并未具体说明如何将这些高层责任系统性地分解为面向实现的低层架构。本文通过应用设计科学研究（Design Science Research, DSR）来填补这一空白，将ANASSA框架从高层架构（High-Level Architecture, HLA）责任细化为低层架构（Low-Level Architecture, LLA）。所得到的设计保留了基础框架的责任边界，同时将其转化为可执行的架构结构，包括47个逻辑角色、版本化任务图、基于能力的代理选择、类型化合同、显式运行时状态、空间验证机制、治理控制、故障和恢复路径，以及来源记录。通过端到端执行轨迹展示了该架构，并在需求可追溯性、内部一致性、责任保留和技术独立性方面进行了分析评估。最终的高层到低层设计为ANASSA框架与未来系统实现之间提供了一个面向实现的桥梁，定义了如何使治理的空间代理工作流可检查、可重建和可测试，而不将架构绑定到特定的AI模型、地理空间平台、通信协议或部署环境。

View on arXiv Download PDF AI Translation

cs.AI / 20 / 2610.14389

Measuring the Inference Energy Efficiency of Open-Weight LLMs across Benchmarks, Reasoning Modes, and Hardware

跨基准、推理模式和硬件测量开放权重大型语言模型的推理能效

Yang, Zeyu, Armour, Wes

Abstract

Benchmark scores tell us what an LLM can do, but not how much energy is required to deliver that capability. We directly measure this trade-off for 86 open-weight model checkpoints, evaluated on seven benchmarks and six hardware platforms ranging from a consumer GPU to an eight-GPU server. We find large differences in energy efficiency even among models with similar capability. Most configurations lie away from the score--energy Pareto frontier: they use a median of 2.6--4.8$\times$ more energy than another configuration that scores at least as well, with differences reaching 27$\times$. Reasoning can improve capability, but enabling thinking increases energy by about 3.7$\times$ on average, and the benefit depends strongly on the model family and output budget. Generated-token count is also an imperfect proxy for energy, misordering 18--24\% of model pairs. In contrast, FP8 reduces energy by roughly a quarter with little change in score, while changing hardware largely rescales energy without changing the relative ranking of models. These results show that capability alone gives an incomplete picture of efficiency. We argue that benchmark scores should be reported together with measured energy consumption under a specified serving regime and output budget.

Chinese Translation

基准得分告诉我们一个大型语言模型（LLM）能够做什么，但并不说明实现该能力所需的能量。我们直接测量了86个开放权重模型检查点的这一权衡，这些模型在七个基准和六个硬件平台上进行评估，从消费级GPU到八GPU服务器。我们发现，即使在能力相似的模型之间，能效也存在很大差异。大多数配置位于得分-能量帕累托前沿之外：它们的能量消耗中位数比另一种得分至少相当的配置高出2.6至4.8倍，差异最高可达27倍。推理可以提高能力，但启用思考平均会使能量增加约3.7倍，且这种收益在很大程度上依赖于模型家族和输出预算。生成的令牌数量也并不是能量的完美代理，错误排序了18%至24%的模型对。相比之下，FP8将能量减少了大约四分之一，而得分几乎没有变化，而更换硬件则在很大程度上重新调整了能量，而不改变模型的相对排名。这些结果表明，仅凭能力无法全面反映效率。我们认为，基准得分应与在特定服务模式和输出预算下测量的能耗一起报告。

View on arXiv Download PDF AI Translation

cs.AI / 21 / 2610.14402

SAT-RL: Training Language Models for Logical Reasoning with Symbolic Solvers

SAT-RL：利用符号求解器训练语言模型进行逻辑推理

Sun, Jialiang, Meel, Kuldeep

Abstract

Large language models (LLMs) have achieved strong performance in mathematical and logical reasoning, yet they still suffer from the curse of complexity and struggle as problem difficulty increases with the number of constraint interactions and the size of the search space. In contrast, satisfiability solvers such as Z3 provide scalable and verifiable reasoning through constraint solving under formal theories, but lack the flexibility to interpret natural-language problem descriptions. In this work, we propose SAT-RL, a reinforcement-learning (RL) framework that trains LLMs to interact with symbolic solvers through multi-turn tool execution. SAT-RL leverages synthetic data generation with verifiable solutions, supervised warm-up on solver-augmented traces, and solver-grounded reinforcement learning with tool-aware rewards. Experiments show that SAT-RL substantially improves symbolic reasoning with tools on both in-domain and out-of-domain logic tasks. Our code is available on https://github.com/sunjia72/SAT-RL.

Chinese Translation

大型语言模型（LLMs）在数学和逻辑推理方面取得了强劲的表现，但它们仍然受到复杂性诅咒的困扰，并且随着约束交互数量和搜索空间大小的增加，问题难度加大时表现不佳。相比之下，满足性求解器如 Z3 通过在形式理论下的约束求解提供可扩展且可验证的推理，但缺乏解释自然语言问题描述的灵活性。在本研究中，我们提出了 SAT-RL，一个强化学习（RL）框架，旨在训练 LLMs 通过多轮工具执行与符号求解器进行交互。SAT-RL 利用可验证解的合成数据生成、在求解器增强的轨迹上的监督预热，以及基于工具的奖励的求解器基础强化学习。实验表明，SAT-RL 在领域内和领域外的逻辑任务中显著改善了使用工具的符号推理。我们的代码可在 https://github.com/sunjia72/SAT-RL 获取。

View on arXiv Download PDF AI Translation

cs.AI / 22 / 2610.14428

Frustratingly Easy Agentic RAG via Portable Skills: A Case Study in Customer Support

通过可移植技能实现令人沮丧的简单代理 RAG：客户支持案例研究

Bako, Istok, Min, Bonan, Shao, Tiffany, Cheng, Claire, Mehdad, Yashar

Abstract

Enterprise workflow automation typically requires retrieving a complete and accurate set of relevant information from large operational document corpora (SOPs, policy documents) given a specific scenario or request. In customer support (CS), for example, human specialists must determine the correct sequence of resolution steps for each case, a task that demands precise grounding in policy documentation. Meeting this bar in production typically requires complex document chunking, indexing, and retrieval pipelines, often layered with learned re-rankers, query rewriters, and bespoke agentic RAG infrastructure. We report a deployment case study of a simpler alternative: a portable skill consisting of a natural-language instruction, lightweight tools (a retrieval API, format-checking scripts), and a corpus enriched with an LLM-generated navigation layer, which is a structured "Wikipedia-style" index over the underlying documents. This agentic RAG approach is straightforward to implement (1-2 engineer days), portable across capable agentic harnesses (e.g., Claude Code), and performs strongly on the challenging task of CS resolution workflow generation, achieving 99.2% procedural equivalence to human-approved ground truth and 83.3% evidence sufficiency. It is also roughly an order of magnitude slower per case than our deployed production assistant (~90 s vs. ~8-9 s), so we claim no latency advantage. Our claim is scoped to that setting: a mature, well-curated corpus, a latency-tolerant workflow, and human review of every case. There, the skill did the work of a complex RAG stack, and maintenance became a matter of editing text and data rather than retraining models.

Chinese Translation

企业工作流自动化通常需要在特定场景或请求下，从大型操作文档库（标准操作程序、政策文件）中检索完整且准确的相关信息。例如，在客户支持（CS）中，人类专家必须确定每个案例的正确解决步骤顺序，这项任务需要对政策文档进行精确的基础支持。在生产环境中，满足这一要求通常需要复杂的文档分块、索引和检索管道，通常还需结合学习的重新排序器、查询重写器和定制的代理 RAG 基础设施。我们报告了一个更简单替代方案的部署案例研究：一种可移植技能，由自然语言指令、轻量级工具（检索 API、格式检查脚本）和一个经过 LLM 生成的导航层丰富的语料库组成，该导航层是对基础文档的结构化“维基百科式”索引。这种代理 RAG 方法易于实施（1-2 个工程师工作日），可在有能力的代理工具（例如 Claude Code）中移植，并在客户支持解决工作流生成这一具有挑战性的任务上表现出色，达到了与人类批准的真实情况 99.2% 的程序等效性和 83.3% 的证据充分性。每个案例的处理速度大约比我们部署的生产助手慢一个数量级（约 90 秒对比约 8-9 秒），因此我们不主张有延迟优势。我们的主张限于这样的环境：一个成熟、精心策划的语料库，一个耐延迟的工作流，以及对每个案例的人类审核。在这种情况下，该技能完成了复杂 RAG 堆栈的工作，维护变成了编辑文本和数据的问题，而不是重新训练模型。

View on arXiv Download PDF AI Translation

cs.AI / 23 / 2610.14487

AlignPoison: Conditional Targeted Poisoning of Generative Recommenders via Decoding-Aligned Prefixes

AlignPoison：通过解码对齐前缀进行条件目标性中毒的生成推荐系统

Zhang, Zhuoxuan, Duan, Jinhao, Zhao, Xinyu, Guo, Jinduo, Shao, Shuai, Li, Rui, Simon, Luke, Deng, Zhun, Chen, Tianlong, Sun, Bo

Abstract

Generative recommendation models based on autoregressive decoding have been adopted as alternatives to classical ranking-based systems. While their modeling flexibility has been widely studied, their resilience to data poisoning remains largely unexplored. In this work, we ask: how susceptible are generative recommenders to conditional targeted poisoning? Under a realistic setting where an adversary can inject limited malicious interactions and query a recommendation API as a black box, our empirical study shows that generative recommenders are vulnerable: injecting fewer than 1% synthetic interactions significantly increases the likelihood of promoting malicious targets under the trigger. To characterize this vulnerability, we analyze the synthesis of malicious interactions from two perspectives: (i) decoding alignment and (ii) prefix popularity. We observe that short, context-aligned prefixes that are both common under the model's induced transition dynamics and strongly linked to a frequent trigger consistently amplify targeted promotion. Motivated by this analysis, we develop AlignPoison, a black-box framework that discovers and samples context-aligned prefixes from API outputs. We conduct experiments on benchmarks in unimodal and multimodal generative recommendation settings. Results show that these prefixes outperform competitive baselines, improving ASR lift by up to 16.3 percentage points and highlighting a decoding-dependent, context-sensitive poisoning vulnerability in generative recommendation models.

Chinese Translation

基于自回归解码的生成推荐模型已被作为经典基于排名系统的替代方案。尽管其建模灵活性已被广泛研究，但其对数据中毒的抵抗力仍然未得到充分探索。在本研究中，我们提出了一个问题：生成推荐系统对条件目标性中毒的敏感程度如何？在一个现实的环境中，攻击者可以注入有限的恶意交互并将推荐API视为黑箱进行查询，我们的实证研究表明，生成推荐系统是脆弱的：注入少于1%的合成交互显著增加了在触发条件下提升恶意目标的可能性。为了表征这种脆弱性，我们从两个角度分析恶意交互的合成：（i）解码对齐和（ii）前缀流行度。我们观察到，短的、与上下文对齐的前缀在模型诱导的转移动态下既常见又与频繁触发强相关，能够持续放大目标推广。基于这一分析，我们开发了AlignPoison，一个黑箱框架，用于从API输出中发现和采样上下文对齐的前缀。我们在单模态和多模态生成推荐设置的基准上进行了实验。结果表明，这些前缀的表现优于竞争基线，ASR提升最高可达16.3个百分点，突显了生成推荐模型中依赖解码的、上下文敏感的中毒脆弱性。

View on arXiv Download PDF AI Translation

cs.AI / 24 / 2610.14489

HiPPT: Physiology-Grounded Wearable PPG Representation Learning via Multi-Scope Text Alignment

HiPPT：基于生理学的可穿戴PPG表示学习通过多范围文本对齐

Wang, Jiaqi, Chen, Wei, Li, Xiaoyu, Xia, Lexian, Tang, ShuaiLong, Zhang, Chenran, Xie, Donglin, Zhu, Guangpu, Li, Yelei, Zhou, Yi

Abstract

Reusable wearable PPG representations should capture local pulse morphology and broader physiological regularities. Existing pretraining objectives, however, usually apply supervision at the recording or pair level, overlooking that different concepts become reliable over different scopes of waveform evidence. We propose HiPPT, a physiology-grounded framework that matches each supervision source to its appropriate evidence scope. We construct a PPG-Text corpus of over 4 million wrist-worn recordings from 202K users, pairing recordings with waveform-derived morphology and metadata-derived context descriptions. HiPPT combines record-level morphology alignment, which grounds individual recordings in pulse characteristics, with cross-recording context alignment, which aggregates recordings sharing the same metadata tuple to capture context-consistent regularities. Waveform-guided text modeling further recovers masked contextual concepts from PPG patch representations, providing fine-grained waveform-text supervision. Across six downstream tasks spanning blood pressure, vascular stiffness, physiological attributes, and cross-dataset heart-rate estimation, HiPPT achieves the best result on the primary linear-probing metrics. The aligned space also supports zero-shot profiling of age group, sex, and BMI category, individually and jointly, with performance improving as more recordings are aggregated. These results show that evidence-scope matching yields transferable and semantically accessible wearable PPG representations. Code and pretrained weights will be released.

Chinese Translation

可重复使用的可穿戴PPG表示应捕捉局部脉搏形态和更广泛的生理规律。然而，现有的预训练目标通常在录音或配对级别应用监督，忽略了不同概念在不同波形证据范围内的可靠性。我们提出了HiPPT，一个基于生理学的框架，将每个监督源与其适当的证据范围相匹配。我们构建了一个PPG-Text语料库，包含来自202K用户的超过400万条腕带录音，将录音与波形衍生的形态和元数据衍生的上下文描述配对。HiPPT结合了记录级形态对齐，将单个录音与脉搏特征相结合，以及跨录音上下文对齐，将共享相同元数据元组的录音聚合，以捕捉上下文一致的规律。波形引导的文本建模进一步从PPG补丁表示中恢复被掩盖的上下文概念，提供细粒度的波形-文本监督。在涵盖血压、血管硬度、生理属性和跨数据集心率估计的六个下游任务中，HiPPT在主要线性探测指标上取得了最佳结果。对齐空间还支持对年龄组、性别和BMI类别的零样本分析，单独和联合进行，随着更多录音的聚合，性能不断提高。这些结果表明，证据范围匹配产生了可转移且语义可访问的可穿戴PPG表示。代码和预训练权重将被发布。

View on arXiv Download PDF AI Translation

cs.AI / 25 / 2610.14495

Reels-RSI: Program-Level Self-Improvement for Code-Rendered Explainer Videos

Reels-RSI：代码渲染解释视频的程序级自我改进

Lin, Wei

Abstract

Explainer videos -- a worked geometry problem, why the sky is blue, how a dynasty fell -- are usually one more target for pixel video generation. We argue they are a different kind of artifact: a program (script, storyboard and per-frame animation code) that a deterministic renderer turns into frames. A program can be read, diffed, tested and audited; pixels can only be watched. Reels-RSI is an open-source reel director for any agent harness; its self-improvement happens at the program level rather than in weights or prompts. Every revision pass is logged with a code snapshot; a retrospective pairs each check finding with the diff that fixed it and proposes lessons; a lesson that a static check can catch is compiled into a rule with a must-fail and a must-pass example, so the memory is unit-tested; and a rule learned today runs over everything made before it. A bounded, benchmark-gated step lets an agent rewrite the director's playbook, kept only when blind pairwise judgments on held-out topics prefer it; nothing enters memory without a human decision. Across 111 reels -- a field study, two playbook evolutions, a 2x2 memory ablation, a weaker director, a 48-reel learning curve and a judge-reliability study -- we find that the rule learned in the field caught four latent defects in frames authored before it existed and none after; that rules are an audit for a strong director (eight unguarded reels, no violation) and a guard for a weaker one (thirty violations unguarded versus five); that lessons shorten revision loops but do not compound or move the judge; and that the first rewrite's 11--1 verdict was a position bias in our vote design (judges pick the second-shown reel 62--81% of the time), masking scripts that traded thirteen facts for craft. We document four failure modes of the preference signal, which check caught each, and the fixes. Code, rules, lessons and data are public.

Chinese Translation

解释视频——如几何问题的解答、天空为何是蓝色、一个王朝的衰落——通常是像素视频生成的另一个目标。我们认为它们是一种不同的工件：一个程序（脚本、故事板和逐帧动画代码），由确定性渲染器转换为帧。程序可以被读取、比较、测试和审计；而像素只能被观看。Reels-RSI 是一个开源的卷轴导演，适用于任何代理工具；其自我改进发生在程序级别，而非权重或提示。每次修订过程都记录有代码快照；回顾性分析将每个检查发现与修复它的差异配对，并提出教训；静态检查能够捕获的教训被编译成一个规则，包含一个必须失败和一个必须通过的示例，从而对记忆进行单元测试；今天学到的规则将在其之前制作的所有内容上运行。一个有限的、基准门控的步骤允许代理重写导演的剧本，仅在对保留主题的盲目成对判断中优于原剧本时才保留；没有人类决策的内容不会进入记忆。在111个卷轴的研究中——包括一个实地研究、两个剧本演变、一个2x2记忆消融实验、一个较弱的导演、一个48卷轴的学习曲线和一个评审可靠性研究——我们发现，实地学到的规则捕获了四个在其存在之前创作的帧中的潜在缺陷，而其后则没有；这些规则是强导演的审计（八个未防护的卷轴，没有违规）和弱导演的保护（未防护的三十次违规与五次）；教训缩短了修订循环，但并未复合或改变评审；第一次重写的11-1裁决是我们投票设计中的位置偏见（评审在62-81%的时间里选择第二个展示的卷轴），掩盖了用工艺交换了十三个事实的脚本。我们记录了偏好信号的四种失败模式，检查捕获了每一种，并提供了修复方案。代码、规则、教训和数据都是公开的。

View on arXiv Download PDF AI Translation

cs.AI / 26 / 2610.14507

MontageFlow: Physically Valid High Fidelity EEG Generation

MontageFlow：物理有效的高保真脑电图生成

Regmi, Sudarshan, Subedi, Nabaraj, Pillai, Arvind, Wu, Yu Yvonne, Chen, Yuliang, Heinz, Michael V., Marsch, Lisa, Jacobson, Nicholas C., Campbell, Andrew

Abstract

Generative modeling offers a promising solution to the scarcity of labeled clinical electroencephalography (EEG) data under strict privacy constraints. Although recent generative models report substantial improvements in generation fidelity, we show that these gains conceal a categorical failure. EEG is not an arbitrary multichannel signal: each channel records the voltage difference between two scalp electrodes. So, real recordings satisfy exact linear identities among their channels. Current generators invariably violate these identities, yet this has gone unnoticed. We therefore propose MontageFlow, a flow-matching model that enforces these identities by restricting its output to signals that satisfy them. We find that ensuring validity improves fidelity too. Furthermore, to retain the EEG characteristics that oversmoothing washes out, we also condition the model on a physiological state vector encoded from the whole EEG window. For generation, this vector is drawn from a prior fitted in latent space. Across 3 large-scale benchmarks, our model improves TS-FID by 1.6-3.6$\times$ over the closest rival while ensuring validity.

Chinese Translation

生成建模为在严格隐私限制下缺乏标记临床脑电图（EEG）数据提供了一种有前景的解决方案。尽管近期的生成模型在生成保真度上报告了显著的改进，但我们表明这些提升掩盖了一个根本性的失败。EEG并不是一种任意的多通道信号：每个通道记录的是两个头皮电极之间的电压差。因此，真实的记录在其通道之间满足精确的线性关系。目前的生成器无一例外地违反了这些关系，但这一点却未被注意到。因此，我们提出了MontageFlow，这是一种流匹配模型，通过将输出限制为满足这些关系的信号来强制执行这些身份。我们发现，确保有效性也提高了保真度。此外，为了保留过度平滑所消除的EEG特征，我们还将模型条件化于从整个EEG窗口编码的生理状态向量。在生成过程中，该向量是从潜在空间中拟合的先验中抽取的。在三个大规模基准测试中，我们的模型在确保有效性的同时，将TS-FID提高了1.6-3.6倍，超过了最接近的竞争对手。

View on arXiv Download PDF AI Translation

cs.AI / 27 / 2610.14523

ScienceBuddy-PV: A Verifier-Governed Recursive Improvement Protocol for Panvascular Intervention Research

ScienceBuddy-PV：一种用于全血管干预研究的验证者主导递归改进协议

You, Lingsen, Guo, Yujun, Zhong, Xinyu, Feng, Zijie, Wang, Wentong, Shen, Li, Ge, Junbo

Abstract

ScienceBuddy-PV adapts the ScienceBuddy recursive-in-recursive design to panvascular intervention research through a structured evidence ledger, device-vessel ontology, abstention contract and retained failure bank. Recursive harness optimization and outer-loop model learning, including reinforcement learning, are specified but not executed. We audit a fixed component, rather than the complete agent system, using separate ARCADE syntax and stenosis tasks (1,000 training, 200 validation and 300 test images per task). A shared task-masked convolutional model was trained once per representation setting; its parameters and the heuristic support gate were then held fixed, with the stenosis threshold selected on validation data. On the locked test sets, syntax Dice was 0.475 (95% CI 0.459-0.490) and stenosis Dice was 0.231 (95% CI 0.214-0.249). Stenosis pixel AUROC was 0.957 and AUPRC was 0.180 at prevalence 0.00975. The gate accepted 76.7% of stenosis-frame outputs versus 100% for a score-only comparator. In a constructed mask audit, it accepted 44.3% of reference candidates and rejected 55.7%, exposing the cost of abstention. Acceptance does not establish correctness or clinical safety. This public-data component audit makes support and abstention measurable and provides a reproducible protocol for future collaborative agent studies. It does not demonstrate recursive improvement, patient-level validity or clinical device recommendations.

Chinese Translation

ScienceBuddy-PV通过结构化证据账本、设备-血管本体、弃权合同和保留失败库，将ScienceBuddy递归嵌套设计适应于全血管干预研究。递归束优化和外部循环模型学习，包括强化学习，已被指定但未执行。我们使用单独的ARCADE语法和狭窄症任务（每个任务1000个训练、200个验证和300个测试图像）审计一个固定组件，而非完整的代理系统。共享任务掩蔽卷积模型在每个表示设置下训练一次；其参数和启发式支持门随后保持固定，狭窄症阈值在验证数据上选择。在锁定的测试集上，语法Dice为0.475（95% CI 0.459-0.490），狭窄症Dice为0.231（95% CI 0.214-0.249）。狭窄症像素AUROC为0.957，AUPRC为0.180，流行率为0.00975。该门接受76.7%的狭窄框输出，而仅有100%为仅评分的比较者。在构建的掩模审计中，它接受了44.3%的参考候选者并拒绝了55.7%，暴露了弃权的成本。接受并不建立正确性或临床安全性。该公共数据组件审计使支持和弃权可测量，并为未来的协作代理研究提供了可重复的协议。它并未证明递归改进、患者级有效性或临床设备推荐。

View on arXiv Download PDF AI Translation

cs.AI / 28 / 2610.14525

Energy Ring: Stabilizing Feature Energy for Reliable Test-Time Discovery

能量环：稳定特征能量以实现可靠的测试时发现

Liu, Tianle, Lyu, Fan, Feng, Tingliang, Ni, Chenggong, Wang, Tianqi, Zhang, Zhang, Hu, Fuyuan, Yao, Rui, Zhou, Tao

Abstract

Test-Time Discovery (TTD) aims to recognize known classes and detect novel ones while the model is already deployed, operating directly on a continual unlabeled test-time stream. Existing TTD methods are overly conservative, making known and unknown features collapse together and difficult to separate during test-time adaptation. In this paper, we revisit TTD from an energy perspective and find that the natural energy gap between known and unknown classes is unstable and easily collapses. We therefore propose Energy Ring to stabilize this gap by anchoring the two types of classes on separate energy rings. A lightweight adapter enlarges the norm-based gap while preserving known-class directions, and a perturbation consistency module forms compact clusters for unknown classes. During testing, energy-guided confidence weighting improves prediction, and a stability-based rectification mechanism corrects unreliable updates. Experimental results show that Energy Ring significantly improves TTD performance against existing TTD methods.

Chinese Translation

测试时发现（TTD）旨在在模型已经部署的情况下识别已知类别并检测新类别，直接在持续的无标签测试时间流上操作。现有的TTD方法过于保守，使得已知和未知特征在测试时适应过程中相互融合，难以分离。本文从能量的角度重新审视TTD，发现已知类别和未知类别之间的自然能量差距不稳定且容易崩溃。因此，我们提出了能量环（Energy Ring）来通过将这两类锚定在不同的能量环上来稳定这一差距。一个轻量级适配器在保持已知类别方向的同时扩大基于范数的差距，而一个扰动一致性模块则为未知类别形成紧凑的聚类。在测试过程中，基于能量的置信加权提高了预测精度，而基于稳定性的修正机制则纠正了不可靠的更新。实验结果表明，能量环显著提高了TTD性能，相较于现有的TTD方法。

View on arXiv Download PDF AI Translation

cs.AI / 29 / 2610.14539

FDE-Bench: Evaluating End-to-End Delivery from Underspecified Real-World Business Requests

FDE-Bench：评估来自不明确的现实商业请求的端到端交付

Li, Huaiming, Huang, Can, Wu, Chufan, Liu, Danyu, Yuan, Xiaomin, Li, Qinze, Lo, Yui, Bai, Wenhui, Wang, Qianlong, Ge, Zengle, Yan, Xiyu, Cheng, Yuheng, Ren, Jinhui, Chen, Guokai, Liu, Jiaqun, Li, Haoran, Liu, Qianhui, Li, Annan, Wu, Jianmin, Yin, Dawei, Shen, Dou

Abstract

Real-world business requests often leave key requirements unstated because frontline practitioners struggle to articulate rules they know from experience. Forward-deployed engineers (FDEs) must elicit these requirements through dialogue and iteratively optimize solutions until they meet customer acceptance criteria. Existing optimization benchmarks typically start from explicit objectives, while clarification benchmarks often introduce information gaps into existing tasks. We introduce FDE-Bench, the first benchmark built from real customer projects to evaluate agents on end-to-end delivery tasks that require interactive requirement clarification and iterative optimization. Its 49 cases span combinatorial optimization and machine learning. Clarification targets are distilled from discussions between frontline practitioners and FDEs. Agents elicit missing requirements from a simulated customer, and case-specific evaluators assess final artifacts against acceptance criteria using customer-accepted FDE solutions as quality references. Across 18 agent configurations, the highest mean delivery score is only 0.5510, compared with an FDE reference of 1.0. The highest delivery success rate (DSR), defined as the fraction of planned runs reaching this reference, is only 6.12%. These results show that reliable FDE-level end-to-end delivery remains out of reach for the evaluated agents on underspecified real-world business tasks. The benchmark project page is available at https://baidubce.github.io/FM-Agent/FDE-Bench/.

Chinese Translation

现实世界中的商业请求往往未明确关键需求，因为一线从业者难以清晰表达他们从经验中获得的规则。前线部署工程师（FDE）必须通过对话引出这些需求，并迭代优化解决方案，直到满足客户接受标准。现有的优化基准通常从明确的目标开始，而澄清基准则往往在现有任务中引入信息缺口。我们提出了FDE-Bench，这是第一个基于真实客户项目构建的基准，用于评估代理在需要互动需求澄清和迭代优化的端到端交付任务中的表现。其49个案例涵盖组合优化和机器学习。澄清目标源自一线从业者与FDE之间的讨论。代理从模拟客户那里引出缺失的需求，案例特定的评估者根据客户接受的FDE解决方案作为质量参考，评估最终工件是否符合接受标准。在18种代理配置中，最高的平均交付分数仅为0.5510，而FDE参考分数为1.0。最高的交付成功率（DSR），定义为达到该参考的计划运行的比例，仅为6.12%。这些结果表明，对于评估的代理而言，可靠的FDE级别端到端交付在不明确的现实商业任务中仍然遥不可及。基准项目页面可访问 https://baidubce.github.io/FM-Agent/FDE-Bench/。

View on arXiv Download PDF AI Translation

cs.AI / 30 / 2610.14542

Are Multimodal Benchmarks Really Useful? Item-Level Multimodal Benchmark Diagnosis via Structure-Response Co-Calibration

多模态基准真的有用吗？通过结构-响应共同校准进行项目级多模态基准诊断

Zhang, Shiqi, Zeng, Weixin, Zhang, Ziheng, Tang, Jiuyang, Zhao, Xiang

Abstract

Multimodal benchmarks are central to evaluating whether models can reason over information from multiple modalities. However, existing literature has revealed that some benchmarks may contain shortcut items that can be answered from a single modality or from answer-option regularities alone, thus failing to reflect models' true cross-modal inference ability. Existing methods for assessing benchmarks typically rely either on intrinsic item structure or extrinsic model response behavior, which are not adequate for a comprehensive diagnosis. To fill in this gap, we propose SRCoD, a structure-response co-calibrated diagnosis framework for item-level diagnosis of modality dependence in multimodal benchmarks. SRCoD estimates sample-wise multimodal information structure from benchmark content and target answers, and uses this signal as a soft structural anchor for modality-conditioned item response modeling. The learned model produces a calibrated item-level diagnostic profile that reflects how strongly an item depends on different forms of modality evidence, enabling fine-grained benchmark analysis beyond aggregate accuracy. SRCoD also provides interpretable attribution of the diagnosis results. Extensive experiments under two complementary protocols, i.e., benchmark-refinement oriented validation and human-labeled item validation, demonstrate that SRCoD outperforms state-of-the-art benchmark diagnosis methods.

Chinese Translation

多模态基准在评估模型是否能够对来自多个模态的信息进行推理方面至关重要。然而，现有文献表明，一些基准可能包含可以通过单一模态或仅通过答案选项规律回答的捷径项目，从而未能反映模型真实的跨模态推理能力。现有的基准评估方法通常依赖于内在的项目结构或外在的模型响应行为，这对于全面诊断而言并不充分。为填补这一空白，我们提出了SRCoD，一种针对多模态基准中模态依赖的项目级诊断的结构-响应共同校准诊断框架。SRCoD从基准内容和目标答案中估计样本级多模态信息结构，并将该信号用作模态条件项目响应建模的软结构锚。所学习的模型生成了一个校准的项目级诊断特征，反映了一个项目在多大程度上依赖于不同形式的模态证据，从而使基准分析超越了整体准确率。SRCoD还提供了诊断结果的可解释归因。在两种互补协议下进行的大量实验，即基准优化导向验证和人工标注项目验证，证明了SRCoD在基准诊断方法中优于最先进的技术。

View on arXiv Download PDF AI Translation

cs.AI / 31 / 2610.14570

PsyMemBehav: Psychology-Informed and Memory-Grounded Two-Level Identity-to-Action Reasoning for Social-Media Behavior Simulation

PsyMemBehav：基于心理学和记忆的两级身份到行动推理的社交媒体行为模拟

Zhang, Wenji, Cheng, Xilong, Qin, Yunxiao, Tu, Mai, Zhu, Tongye, Zhou, Yuhao, Ji, JingBo, Tan, Yuting, Li, Duoyin, Xiao, Hongjiang, Zhang, Yuan

Abstract

Large language models (LLMs) support social simulation, but faithfully reproducing real users requires grounding both structured platform decisions and language in sparse, noisy behavioral evidence. We introduce PsyMemBehav, which maps psychology-informed profiles and retrieved memories to target-first platform behavior through two-level identity-to-action reasoning. To support this approach with comprehensive behavioral traces, we construct a dataset PsyMemBehav-bench from 5.57M observed behaviors of 5,862 real users (approximately 950 per user) across Reddit, Bluesky, and Twitter. Each recorded engagement is modeled as target selection followed by Comment, Like, or Repost and optional response generation. PsyMemBehav combines 41-dimensional user descriptors, top-5 BM25 memories, reverse-CoT supervision, and rubric-guided DPO over complete reasoning-action trajectories. Experiments show that PsyMemBehav 27B ranks first in 11 of 12 reported individual-level metrics, improving average response quality over GPT-5.4 from 54.28 to 59.68 (9.95% relative), and obtains the lowest group-level MAE (0.081) and MSE (0.009) among simulated systems. These results support evidence-grounded user modeling for observed-thread continuation; the benchmark remains conditional on recorded engagement and does not model no-action decisions.

Chinese Translation

大型语言模型（LLMs）支持社会模拟，但忠实再现真实用户需要将结构化平台决策和语言与稀疏、嘈杂的行为证据相结合。我们提出了PsyMemBehav，它通过两级身份到行动推理将基于心理学的用户画像和检索的记忆映射到目标优先的平台行为。为了支持这一方法，我们从5862名真实用户（每名用户约950个行为）在Reddit、Bluesky和Twitter上的557万观察行为中构建了数据集PsyMemBehav-bench。每个记录的参与被建模为目标选择，随后是评论、点赞或转发以及可选的响应生成。PsyMemBehav结合了41维用户描述符、前5名BM25记忆、反向链式思维（reverse-CoT）监督和基于评分标准的DPO（动态规划优化）在完整的推理-行动轨迹上。实验表明，PsyMemBehav 27B在12个报告的个体级指标中排名第一，平均响应质量较GPT-5.4从54.28提高到59.68（相对提高9.95%），并在模拟系统中获得最低的组级平均绝对误差（MAE）为0.081和均方误差（MSE）为0.009。这些结果支持基于证据的用户建模以实现观察线程的延续；该基准仍然依赖于记录的参与，并未建模无行动决策。

View on arXiv Download PDF AI Translation

cs.AI / 32 / 2610.14573

Learning after Execution with Action-Specific Credit

执行后的学习与动作特定信用

Tian, Yufeng, Tang, Jiahao, Wu, Minghao

Abstract

Execution is often used as terminal feedback: it scores a generated program, but the resulting state is not used for another learning action. Learning after execution instead uses the execution result as an intermediate state that supports an additional learning interaction for the same policy. This creates a credit-assignment problem because the two actions share the same final objective but play different roles in the trajectory. Accordingly, action-specific credit assigns the initial action credit from the final trajectory outcome and the corrective action credit for improvement over the intermediate state. Image-to-TikZ serves as an executable visual testbed for this setting. On 945 held-out examples, learning after execution strengthens the shared policy under single-call inference and also produces a post-execution corrective capability. Single-call render success reaches 92.70\%, compared with 88.25\% for continued single-stage GRPO after the same number of additional training iterations. This corrective capability is not reproduced by continued single-stage training and depends on the intermediate state. Alternative credit definitions also induce different group-relative preferences over the same trajectories. Together, these results suggest that executable tasks contain useful learning structure beyond a terminal reward. Code is available at https://github.com/tianyufengcn/Learning-after-Execution.

Chinese Translation

执行通常被用作终端反馈：它对生成的程序进行评分，但结果状态并未用于其他学习动作。执行后的学习则将执行结果作为一个中间状态，支持对同一策略的额外学习交互。这产生了一个信用分配问题，因为这两个动作共享相同的最终目标，但在轨迹中扮演不同的角色。因此，动作特定信用从最终轨迹结果中分配初始动作的信用，并为中间状态的改进分配纠正动作的信用。Image-to-TikZ作为这一设置的可执行视觉测试平台。在945个保留示例上，执行后的学习增强了单次调用推理下的共享策略，并且还产生了执行后的纠正能力。单次调用渲染成功率达到92.70%，而在相同数量的额外训练迭代后，继续单阶段GRPO的成功率为88.25%。这种纠正能力并未通过继续单阶段训练再现，并依赖于中间状态。替代的信用定义也会在相同轨迹上引发不同的群体相对偏好。综合来看，这些结果表明，可执行任务包含超越终端奖励的有用学习结构。代码可在 https://github.com/tianyufengcn/Learning-after-Execution 获取。

View on arXiv Download PDF AI Translation

cs.AI / 33 / 2610.14574

CB-Agent: Concept-Based In-Context Learning for Sequential Action Selection

CB-Agent：基于概念的上下文学习用于顺序动作选择

Zhang, Qingyue, Tang, Huaze, Zhang, Minyu, Yan, Yuxiang, Ding, Wenbo, Huang, Shao-Lun

Abstract

In-context learning (ICL) allows large language models to use demonstrations without updating their parameters. For interactive agents, demonstrations are trajectories of reasoning, actions, and observations that illustrate how to complete a task. Reusing demonstrations in new situations is challenging: directly inputting full trajectories consumes limited prompt space and constrains the overall effectiveness, while natural-language summaries require the agent to translate general guidance into specific action choices. Therefore, we study whether expert action choices can directly supervise a reusable scoring rule. We propose CB-Agent, which extends concept-based ICL to sequential action selection. We extract environment-action decisions from expert trajectories and fit a concept vector by maximum likelihood to model the conditional distribution over candidate actions given the context. This model combines the concept with frozen language-model features and priors. During ReAct interaction, the agent uses this distribution to select environment actions without including demonstrations in the action-scoring context. Across TextWorld, ALFWorld, and ScienceWorld, CB-Agent achieves the highest cross-benchmark average among the evaluated methods at all three tested model sizes and ranks first on TextWorld and ScienceWorld at each size. ALFWorld experiments support the learned concept's contribution beyond prior-only scoring and show further gains when combined with Reflexion. Code is available at https://anonymous.4open.science/r/CB-Agent-DFF9/.

Chinese Translation

上下文学习（In-context learning, ICL）使大型语言模型能够在不更新其参数的情况下使用示范。对于交互式智能体而言，示范是推理、动作和观察的轨迹，展示了如何完成任务。在新情境中重用示范是具有挑战性的：直接输入完整轨迹会消耗有限的提示空间并限制整体效果，而自然语言摘要则要求智能体将一般指导转换为具体的动作选择。因此，我们研究专家动作选择是否可以直接监督一个可重用的评分规则。我们提出了CB-Agent，它将基于概念的ICL扩展到顺序动作选择。我们从专家轨迹中提取环境-动作决策，并通过最大似然法拟合概念向量，以建模给定上下文的候选动作的条件分布。该模型将概念与冻结的语言模型特征和先验结合。在ReAct交互过程中，智能体使用该分布选择环境动作，而无需在动作评分上下文中包含示范。在TextWorld、ALFWorld和ScienceWorld上，CB-Agent在所有三种测试模型规模中实现了评估方法中最高的跨基准平均值，并在每个规模的TextWorld和ScienceWorld上排名第一。ALFWorld实验支持所学概念在仅使用先验评分之外的贡献，并在与Reflexion结合时显示出进一步的提升。代码可在 https://anonymous.4open.science/r/CB-Agent-DFF9/ 获取。

View on arXiv Download PDF AI Translation

cs.AI / 34 / 2610.14576

Iterate on an LLM, Serve on an SLM

在大型语言模型上迭代，在小型语言模型上服务

Wang, Keegan

Abstract

As frontier LLMs gain more capabilities, the cost to run them also goes up. At the same time, small language models (SLMs) also gain more capabilities but have largely been able to keep their cost stable, offering 5-40 times cheaper inference for the narrow, repetitive subtasks that agents decompose work into. However, the standing limiting factor is getting the correct behavior out of a small model in a fast, efficient, and economical way. Unfortunately, in most cases, zero-shot small models fail narrow production tasks. Furthermore, retraining one every time the behavior changes is too slow to iterate on. We propose doing both in sequence. First, develop the desired behavior in prompt space, on a frontier model, with immediate feedback from live traffic. Then, once the behavior converges, distill the recorded production behavior (not synthetic imitations of it) into an SLM through supervised fine-tuning (SFT). We show a proper use case of this method through a scheduling agent for home-care agencies that fans out a single yes/no safety decision per caregiver-shift pair on GPT-5.4-nano. A LoRA-tuned Qwen3-8B trained on 11.6k recorded production decisions reaches 0.913 agreement, while the teacher (GPT-5.4-nano), re-run on its own recorded prompts, agrees with itself only 72.7% of the time, pointing to the fact that the SLM may also be more consistent. Furthermore, when the two disagreed, a blinded professional scheduler sided with the student 48-29. Finally, we find that the SLM has approximately 2.5 times lower median latency and is a fraction of the cost. We believe this lifecycle of quick building, live feedback, and cheap inference is a blueprint for others building similar systems.

Chinese Translation

随着前沿大型语言模型（LLMs）能力的提升，其运行成本也随之增加。与此同时，小型语言模型（SLMs）同样获得了更多能力，但在很大程度上能够保持其成本稳定，为代理将工作分解为狭窄、重复的子任务提供了5到40倍更便宜的推理。然而，当前的限制因素是如何以快速、高效和经济的方式从小型模型中获得正确的行为。不幸的是，在大多数情况下，零-shot的小型模型无法完成狭窄的生产任务。此外，每次行为变化时重新训练模型的速度过慢，无法进行有效迭代。我们建议按顺序进行这两项工作。首先，在前沿模型上开发所需的行为，利用来自实时流量的即时反馈。然后，一旦行为收敛，通过监督微调（SFT）将记录的生产行为（而非其合成模仿）提炼到小型语言模型中。我们通过一个为家庭护理机构设计的调度代理展示了该方法的适用案例，该代理在每个护理人员-班次对上基于GPT-5.4-nano做出单一的“是/否”安全决策。经过LoRA调优的Qwen3-8B在11600个记录的生产决策上达到了0.913的一致性，而教师模型（GPT-5.4-nano）在其自身记录的提示上重新运行时，仅以72.7%的频率与自己达成一致，这表明小型语言模型可能更具一致性。此外，当两者意见不一致时，一位盲测的专业调度员支持学生模型48次对29次。最后，我们发现小型语言模型的中位延迟约低于2.5倍，且成本仅为前者的一小部分。我们相信，这种快速构建、实时反馈和低成本推理的生命周期为其他构建类似系统的研究者提供了蓝图。

View on arXiv Download PDF AI Translation

cs.AI / 35 / 2610.14605

Decoupling Knowledge Forgetting and Refusal Learning in MLLM Unlearning

在多模态大模型去学习中解耦知识遗忘与拒绝学习

Chen, Haodong, Wen, Shengtao, Liang, Dong, Guo, Lingbing, Chen, Xiang

Abstract

Current evaluations of multimodal unlearning typically focus on whether a model stops outputting target information. This surface-level criterion, however, masks two critical failure modes. First, knowledge suppression techniques often induce confabulation, leading the model to hallucinate facts about the unlearned subject. Second, refusal training merely alters observable behavior without erasing the underlying knowledge; such models might decline open queries but still identify correct answers under multiple-choice probing. To overcome these limitations, we propose Forget-Aware Self-Distillation (FASD). This method is designed to achieve honest ignorance, a state where the model genuinely loses the target knowledge and learns to acknowledge its absence. FASD separates these goals into two complementary channels. The behavior channel distills appropriate refusals from an ignorance-prompted copy of the model itself. Crucially, target tokens are masked from the teacher distribution to avoid accidentally reteaching the forgotten concepts. Simultaneously, the knowledge channel directly suppresses ground-truth answer spans independently of the model's generated responses. Experiments demonstrate that FASD successfully unlearns target knowledge, prevents information leakage in open-ended tasks, and preserves performance on retained data. The resulting model not only forgets the specified information but accurately recognizes the boundaries of its own knowledge. Code and data will be released upon publication.

Chinese Translation

当前对多模态去学习的评估通常关注模型是否停止输出目标信息。然而，这一表面标准掩盖了两种关键的失败模式。首先，知识抑制技术往往会导致虚构，使模型产生关于未学习主题的幻觉事实。其次，拒绝训练仅仅改变可观察的行为，而没有抹去基础知识；这样的模型可能会拒绝开放式查询，但在多项选择测试中仍能识别正确答案。为了克服这些局限性，我们提出了遗忘感知自蒸馏（Forget-Aware Self-Distillation, FASD）。该方法旨在实现诚实的无知状态，即模型真正失去目标知识并学会承认其缺失。FASD将这些目标分为两个互补的通道。行为通道从模型自身的一个无知提示副本中提炼适当的拒绝。关键是，目标标记在教师分布中被屏蔽，以避免意外重新教授遗忘的概念。同时，知识通道独立于模型生成的响应直接抑制真实答案的跨度。实验表明，FASD成功地去学习了目标知识，防止了开放式任务中的信息泄漏，并在保留数据上保持了性能。最终模型不仅忘记了指定的信息，还准确识别了自身知识的边界。代码和数据将在发表时发布。

View on arXiv Download PDF AI Translation

cs.AI / 36 / 2610.14625

From Existing Methods to New Research Ideas: Two Perspectives on Recursive Self-Improvement

从现有方法到新研究思路：递归自我改进的两个视角

Zhou, Leijun, Yu, Yanke, Liu, Zhihao, Xiang, Qu, Hu, Tiancheng, Dong, Yiru, Wu, Yihan, Xiang, Yangxi, Wu, Xuejun, Chen, Xi, Qian, Buyue, Zheng, Yaowei, Hu, Junhao

Abstract

Recursive self-improvement (RSI) offers a route to AI systems that participate in their own continued development. Recent methods allow language agents to retain changes to their parameters, memory, skills, and code for subsequent execution and improvement. We examine how these methods can inform new research ideas through two complementary perspectives. First, we introduce a unified taxonomy with nine design dimensions organized into four groups: update targets, update organization, update schedule and scope, feedback and selection. This taxonomy places existing methods in a shared design space, where examining individual dimensions and their interactions can inspire new research ideas. Second, we examine how existing RSI methods draw on traditional deep learning, for example through optimization, regularization, data augmentation, continual learning, and adversarial learning. We develop further ideas by examining more specific learning mechanisms and drawing on evolutionary computation. We also review how existing benchmarks assess self-improvement under online, offline, and offline-to-online settings. Together, these perspectives help researchers develop new RSI research ideas from existing designs and learning mechanisms. RSI papers and resources are available at https://prism-shadow.github.io/awesome-rsi/.

Chinese Translation

递归自我改进（Recursive Self-Improvement, RSI）为参与自身持续发展的人工智能系统提供了一条途径。最近的方法使语言代理能够保留其参数、记忆、技能和代码的变化，以便于后续的执行和改进。我们通过两个互补的视角来探讨这些方法如何启发新的研究思路。首先，我们引入一个统一的分类法，包含九个设计维度，分为四个组别：更新目标、更新组织、更新计划和范围、反馈与选择。该分类法将现有方法置于一个共享的设计空间中，研究各个维度及其相互作用可以激发新的研究思路。其次，我们考察现有的RSI方法如何借鉴传统深度学习，例如通过优化、正则化、数据增强、持续学习和对抗学习。我们通过研究更具体的学习机制并借鉴进化计算来发展进一步的思路。我们还回顾了现有基准如何在在线、离线和离线到在线的设置下评估自我改进。综合来看，这些视角帮助研究人员从现有设计和学习机制中发展新的RSI研究思路。RSI相关论文和资源可在 https://prism-shadow.github.io/awesome-rsi/ 获取。

View on arXiv Download PDF AI Translation

cs.AI / 37 / 2610.14636

PEARL: Process-Experience Aware Reinforcement Learning for Multi-Turn Agents

PEARL：面向过程经验的多回合智能体强化学习

Yang, Jie, Huang, Mianqiu, Zhao, Yiwei, Wang, Bo, Liu, Yang, Wang, Yuxin

Abstract

Terminal rewards make multi-turn agents trainable without step-level annotations, but group-relative advantages provide little guidance about which intermediate decisions to adjust. Failed trajectories also contain reasonable actions, while unfavorable decision patterns can recur as training proceeds. We introduce PEARL (Process-Experience Aware Reinforcement Learning), which improves local credit assignment without an additional critic. Given two continuations from the same history with different outcomes, PEARL locates their first divergence and builds a feature representation of each branch's local decision. It clusters these decision features in recent failure and success memories to identify recurring local patterns. Using this pattern memory, PEARL selectively strengthens negative advantages at lower-reward decisions, while leaving positive advantages and the base policy optimizer unchanged. On AppWorld, PEARL improves over ARPO on Normal and Challenge tasks at both 4B and 8B scales. On ALFWorld, it achieves 97.9% mean success at 4B, 2.7 points above the strongest baseline, while remaining competitive at 8B. Ablations of shaping location, success-bank filtering, and memory window support the value of selective, recurring-pattern-based updates.

Chinese Translation

终端奖励使得多回合智能体可以在没有逐步注释的情况下进行训练，但相对优势的群体提供的指导有限，无法明确调整哪些中间决策。失败的轨迹中也包含合理的行动，而不利的决策模式可能在训练过程中反复出现。我们提出了PEARL（面向过程经验的强化学习），该方法在没有额外评论者的情况下改善了局部信用分配。给定来自同一历史的两个不同结果的延续，PEARL定位它们的首次分歧，并构建每个分支局部决策的特征表示。它在最近的失败和成功记忆中对这些决策特征进行聚类，以识别重复出现的局部模式。利用这种模式记忆，PEARL选择性地增强低奖励决策中的负优势，同时保持正优势和基础策略优化器不变。在AppWorld上，PEARL在4B和8B规模的Normal和Challenge任务上均优于ARPO。在ALFWorld上，它在4B上实现了97.9%的平均成功率，比最强基线高出2.7分，同时在8B上保持竞争力。对塑形位置、成功银行过滤和记忆窗口的消融实验支持了基于选择性、重复模式更新的价值。

View on arXiv Download PDF AI Translation

cs.AI / 38 / 2610.14643

AegisVox: Factorized Privacy and Clinical Preservation for Long-Form Speech

AegisVox：长形式语音的分解隐私与临床保护

Singh, Praphul, Agarwal, Akshat, Ghosh, Ratul, Kanuga, Aashna, Khatwani, Devashish, Dwarakanath, Anurag, Gadde, Sri, Kenthapadi, Krishnaram

Abstract

Clinical speech reveals identity through both spoken identifiers and voice, yet its pauses, delivery, and speaker roles can also carry useful clinical evidence. Protecting that speech therefore requires more than preserving its words. We present AegisVox, a factorized framework that assigns consistent surrogate voices and edits identifier-bearing turns using document-consistent replacements. Across 390 public simulated or non-patient conversations totaling 63.6 hours, a matched comparison of two speech renderers separates replacement realization from retained identifiers, speaker linkage, and clinical preservation. Text-to-speech rendering realizes 535/601 planned replacements versus 453/601 for the codec-based speech editor. However, similar aggregate clinical preservation conceals more model-labeled contradictions, and conversation-wide speaker scores conceal greater edited-speech linkage under a score-direction sensitivity analysis. Fresh transcription of original audio provides a reference for interpreting the clinical measurement chain; an independent judge tests sensitivity to model-based assessment. Better replacement realization can therefore coexist with worse clinical or speaker-linkage measurements. AegisVox supports use-specific evaluation of those tradeoffs, not a guarantee of identifier removal or clinical safety.

Chinese Translation

临床语音通过口头标识符和声音揭示身份，但其停顿、表达方式和说话者角色也可能携带有用的临床证据。因此，保护这些语音不仅仅是保留其词汇。我们提出了AegisVox，这是一个分解框架，分配一致的替代声音，并使用文档一致的替换编辑带有标识符的发言。在390个公共模拟或非患者对话中，总计63.6小时，我们对两个语音渲染器进行了匹配比较，分离了替换实现与保留的标识符、说话者链接和临床保护。文本到语音渲染实现了535/601个计划替换，而基于编解码器的语音编辑器实现了453/601个。然而，类似的整体临床保护掩盖了更多模型标记的矛盾，而对话范围内的说话者评分在评分方向敏感性分析下掩盖了更大的编辑语音链接。对原始音频的新转录为解释临床测量链提供了参考；独立评审者测试了对基于模型评估的敏感性。因此，更好的替换实现可以与更差的临床或说话者链接测量共存。AegisVox支持对这些权衡的特定使用评估，而不是保证标识符的移除或临床安全。

View on arXiv Download PDF AI Translation

cs.AI / 39 / 2610.14644

Shaping Latent Mixture-of-Experts: From Expert Geometry to Layerwise Capacity Allocation

塑造潜在混合专家：从专家几何到逐层容量分配

Wang, Yaoxiang, Zhang, Kai, Xie, Ruobing, Wang, Lanrui, Zhao, Guoliang, Li, Junhui, Su, Jinsong

Abstract

Fine-grained mixture-of-experts (MoE) models commonly increase expert count by narrowing each expert, treating expert shape as a by-product of routing granularity. LatentMoE moves expert computation into a lower-dimensional space, exposing the input dimension and intermediate width of each expert as explicit design choices. We show that an expert's local Jacobian rank is jointly bounded by these two dimensions, and that balancing them maximizes the rank ceiling at fixed per-expert parameters. Controlled shape sweeps across model scales and routing granularities consistently favor near-square experts. We further use layerwise intrinsic dimension (ID) to guide capacity allocation across depth, co-varying the number and shape of experts while keeping per-layer parameter count, activated FLOPs, and communication volume constant. Compared with layer-homogeneous baselines, our ID-guided allocation achieves lower validation loss and stronger downstream performance. These results establish expert shape and layerwise allocation as practical design axes for MoEs.

Chinese Translation

细粒度的混合专家（MoE）模型通常通过缩小每个专家的范围来增加专家数量，将专家形状视为路由粒度的副产品。潜在混合专家（LatentMoE）将专家计算移入低维空间，将每个专家的输入维度和中间宽度作为明确的设计选择。我们展示了专家的局部雅可比秩受到这两个维度的共同限制，并且平衡它们可以在固定的每个专家参数下最大化秩上限。受控形状在模型规模和路由粒度之间的变化一致地偏向接近正方形的专家。我们进一步使用逐层内在维度（ID）来指导深度上的容量分配，同时保持每层参数数量、激活的浮点运算（FLOPs）和通信量不变，协同变化专家的数量和形状。与层同质基线相比，我们的ID引导分配实现了更低的验证损失和更强的下游性能。这些结果确立了专家形状和逐层分配作为混合专家的实际设计轴。

View on arXiv Download PDF AI Translation

cs.AI / 40 / 2610.14659

RASSG: Repair-Aware Semantic Survival Gating for Streaming Tool Execution

RASSG：面向流式工具执行的修复感知语义生存门控

Hu, Daijin, Wang, Xi

Abstract

Streaming speech assistants can launch tools before users finish speaking, overlapping execution with incoming speech to reduce user waiting time. However, subsequent self-corrections can invalidate apparently stable arguments, causing stale execution whose irreversible effects persist even after the final call is corrected. We propose Repair-Aware Semantic Survival Gating (RASSG) to jointly control tool launch and cancellation during streaming speech. RASSG launches a tool when three consecutive speech recognition results arrive with identical nonempty arguments, then combines calibrated argument invalidity risk with execution costs to decide cancellation at the last feasible opportunity before effects become irreversible. Evaluation on ReCAST, FDB-v3, and Audio2Tool uses a controlled execution clock with fixed tool runtimes and cancellation parameters. On ReCAST's prospective-repair subset, RASSG reduces the stale execution rate from LocalAgreement-2's 50.00% to 3.15%, with a mean delay of 3.676s across successful trajectories. The stale reduction also holds against every early-execution baseline on both public speech datasets. Ablations with identical initial launches support active cancellation, showing the value of controlling ongoing calls before their effects become irreversible. Code and data are available at https://github.com/daijinhu/RASSG.

Chinese Translation

流式语音助手可以在用户尚未说完时启动工具，从而将执行与输入语音重叠，以减少用户等待时间。然而，随后的自我修正可能会使看似稳定的论点失效，导致过时的执行，其不可逆的影响在最终调用被修正后仍然存在。我们提出了修复感知语义生存门控（Repair-Aware Semantic Survival Gating，RASSG），以共同控制流式语音中的工具启动和取消。RASSG在连续到达三个相同的非空论点的语音识别结果时启动工具，然后将校准的论点失效风险与执行成本结合，以决定在效果变得不可逆转之前的最后可行机会进行取消。在ReCAST、FDB-v3和Audio2Tool上的评估使用固定工具运行时间和取消参数的受控执行时钟。在ReCAST的前瞻性修复子集上，RASSG将过时执行率从LocalAgreement-2的50.00%降低到3.15%，成功轨迹的平均延迟为3.676秒。过时执行的减少在所有公共语音数据集的每个早期执行基线中也得到了验证。相同初始启动的消融实验支持主动取消，显示了在效果变得不可逆之前控制正在进行的调用的价值。代码和数据可在https://github.com/daijinhu/RASSG获取。

View on arXiv Download PDF AI Translation

cs.AI / 41 / 2610.14680

A Hybrid Kinematic GCN and Elastic Transformer for Efficient Indian Sign Language Recognition

高效印度手语识别的混合运动学图卷积网络与弹性变换器

Garg, Pushp, Sarma, Himangshu, Ghorai, Mrinmoy

Abstract

Sign Language Recognition (SLR) presents complex challenges due to variable signing speeds and spatial intricacies. Specifically, Indian Sign Language (ISL) requires precise modeling of its inherently bimanual nature and is bottlenecked by the scarcity of large-vocabulary benchmarks and the inability of standard models to handle irregular signing velocities. To address this, a structurally-aware hybrid framework for efficient, gloss ISL recognition is proposed. To minimize computational redundancy, an entropy-driven SVD sampler that dynamically extracts informative keyframes is used. Because this subsampling creates irregular physical time gaps, the proposed architecture employs a Continuous Dual Positional Encoding paired with a distance-aware Elastic Transformer to preserve natural signing rhythm. Spatially, a Kinematically-Conditioned Adaptive GCN explicitly models bimanual geometry using unified bone-joint features. To validate our framework, a 120-class, 9,453-video native ISL dataset is created. Augmented by Physical Node Dropout and ArcFace optimization to mitigate subject-specific memorization, our framework achieves a competitive 89.8% accuracy, establishing a strong baseline for complex sign language modeling.

Chinese Translation

手语识别（SLR）由于签名速度的变化和空间复杂性而面临复杂挑战。具体而言，印度手语（ISL）需要精确建模其固有的双手特性，并受到大词汇基准稀缺和标准模型无法处理不规则签名速度的瓶颈。为了解决这一问题，提出了一种结构感知的混合框架，以实现高效的、基于手势的ISL识别。为了最小化计算冗余，采用了一种基于熵的奇异值分解（SVD）采样器，动态提取信息丰富的关键帧。由于这种子采样会产生不规则的物理时间间隔，所提出的架构采用了连续双位置编码（Continuous Dual Positional Encoding）与距离感知的弹性变换器（Elastic Transformer）相结合，以保持自然的签名节奏。在空间上，运动学条件自适应图卷积网络（Kinematically-Conditioned Adaptive GCN）明确建模双手几何，使用统一的骨骼-关节特征。为了验证我们的框架，创建了一个包含120类、9,453个视频的本地ISL数据集。通过物理节点丢弃（Physical Node Dropout）和ArcFace优化来减轻特定个体的记忆化，我们的框架实现了89.8%的竞争性准确率，为复杂手语建模建立了强有力的基准。

View on arXiv Download PDF AI Translation

cs.AI / 42 / 2610.14693

AcrossVAM2.0: Conflict-Aware Point Supervision for Parameter-Efficient Video Diffusion Adaptation

AcrossVAM2.0：基于冲突意识的点监督用于参数高效的视频扩散适应

Wu, Nan, Hong, Yee Hung

Abstract

Explicit point correspondence is attractive for adapting video generators: it describes how visible surfaces move, but can overwrite the appearance prior learned by a large pretrained model. We study this tension in a 5.01B-parameter text-to-video diffusion transformer. ACROSSVAM2.0 combines a confidence-weighted track-and-visibility head, zero-initialized residual feedback, and asymmetric PCGrad that removes only the auxiliary component opposing the primary flow-matching gradient on shared LoRA parameters. The design preserves the generator exactly at initialization and protects its primary update during adaptation. In a four-way ablation trained for 1,000 steps with three independent seeds, the full method lowers KLT forward-backward cycle error in all three seeds relative to a matched flow-only continuation (mean seed-level difference -0.00711, SD 0.00505), while robust sharpness remains positive in all three. An expanded 24-prompt evaluation (216 videos) yields 9.7% lower aggregate CoTracker3 cycle error than the matched control and near-matched performance on six VBench dimensions; however, the seed-level tracking effect is heterogeneous and I3D-FVD regresses from 0.1824 to 0.2022. These results support a bounded claim: conflict-aware adaptation improves correspondence consistency without broad perceptual degradation, but does not uniformly improve video quality. We release equations, pseudocode, matched outputs, source tables, and artifact provenance.

Chinese Translation

显式点对应关系对于适应视频生成器具有吸引力：它描述了可见表面的运动，但可能会覆盖大型预训练模型学习到的外观先验。我们在一个拥有5.01亿参数的文本到视频扩散变换器中研究了这种紧张关系。ACROSSVAM2.0结合了一个基于置信度加权的跟踪和可见性头、零初始化的残差反馈以及只移除与共享LoRA参数上的主要流匹配梯度相对立的辅助成分的非对称PCGrad。该设计在初始化时完全保留生成器，并在适应过程中保护其主要更新。在一个为期1000步、使用三个独立种子的四向消融实验中，完整方法在所有三个种子中相对于匹配的仅流延续降低了KLT前后循环误差（平均种子级差异-0.00711，标准差0.00505），同时在所有三个种子中稳健的清晰度保持为正值。扩展的24提示评估（216个视频）显示出比匹配控制低9.7%的聚合CoTracker3循环误差，并在六个VBench维度上接近匹配的性能；然而，种子级跟踪效果是异质的，I3D-FVD从0.1824回归到0.2022。这些结果支持一个有限的主张：冲突意识适应提高了对应关系的一致性，而没有广泛的感知退化，但并不均匀地提高视频质量。我们发布了方程、伪代码、匹配输出、源表和工件来源。

View on arXiv Download PDF AI Translation

cs.AI / 43 / 2610.14699

Joint Discrete and SE(3) Diffusion for Protein Interface Design

联合离散与SE(3)扩散用于蛋白质界面设计

Song, Zhenqiao, Xie, Jianwen, Li, Lei

Abstract

The design of high-affinity, target-specific protein binders is central to the development of modern therapeutics. However, existing protein-generation methods remain limited in this setting, as many generated designs fail to simultaneously achieve geometric and physicochemical complementarity at binding interfaces. To address this challenge, we introduce PIDiff, a physicochemically enhanced multimodal diffusion framework for protein-protein complex design. PIDiff jointly designs target-conditioned protein sequences and structures by coupling diffusion in discrete sequence space with diffusion on the continuous (SE(3)) structural manifold. At its core, PIDiff employs PIFormer, a novel architecture that interleaves geometry-aware invariant point attention with global sequence self-attention. We further curate PIBench, a protein-protein complex dataset with diverse target scaffolds derived from the Protein Data Bank (PDB). We first pretrain PIDiff on PIBench and subsequently fine-tune it for two therapeutically relevant applications: minibinder design and antibody design. PPFrame consistently outperforms baseline methods across sequence design success rate, structure binding energy, and sequence-structure co-designability on all tasks.

Chinese Translation

高亲和力、靶向特异性的蛋白质结合体设计是现代治疗药物开发的核心。然而，现有的蛋白质生成方法在这一领域仍然存在局限性，因为许多生成的设计无法同时在结合界面上实现几何和物理化学的互补性。为了解决这一挑战，我们提出了PIDiff，一种物理化学增强的多模态扩散框架，用于蛋白质-蛋白质复合物设计。PIDiff通过将离散序列空间的扩散与连续（SE(3)）结构流形的扩散相结合，联合设计靶向条件的蛋白质序列和结构。PIDiff的核心是采用PIFormer，这是一种新颖的架构，将几何感知的不变点注意力与全局序列自注意力交错结合。我们进一步整理了PIBench，这是一个包含来自蛋白质数据银行（PDB）多样靶标支架的蛋白质-蛋白质复合物数据集。我们首先在PIBench上对PIDiff进行预训练，然后针对两个与治疗相关的应用进行微调：小型结合体设计和抗体设计。PPFrame在所有任务中在序列设计成功率、结构结合能和序列-结构共设计能力方面始终优于基线方法。

View on arXiv Download PDF AI Translation

cs.AI / 44 / 2610.14700

DSABench: Benchmarking Source Discovery, Coordination, and Execution for Multi-Database Agents

DSABench：多数据库代理的源发现、协调和执行基准测试

Luo, Jinlong, Wu, Haoran, Xu, Bo

Abstract

Most text-to-SQL benchmarks provide the target database. In a large data catalog, however, choosing the sources is part of the problem: an agent must find every database needed for a request, determine how their results fit together, and execute the resulting workflow. We present DSABench to evaluate these capabilities jointly. Its 815 human-reviewed tasks draw on Spider, BIRD, and LiveSQLBench; 618 require several databases to contribute, and 105 allow more than one interpretation of the required sources. Executable contracts define source participation, cross-source dependencies, and source-local execution, while RouteEnv lets agents inspect public catalog evidence without revealing the gold annotations. In our routing-only study across seven model backends, the highest RA-Set score is 60.55%, compared with a highest RA-Graph score of 34.66% when coordination requirements are included. The gap becomes more pronounced when execution is required. Across four backends, median non-privileged Route+Exec-Graph success ranges from 6.69% to 7.98% without clarification and from 11.44% to 13.25% with one clarification; the best observed clarified configuration reaches 16.99%. On ambiguous tasks, clarification improves exact branch-source recovery by 39.8-43.2 percentage points. These findings illustrate why retrieving plausible databases or producing correct local SQL is not enough: an agent must carry the right information through the entire multi-database workflow.

Chinese Translation

大多数文本到SQL的基准测试提供目标数据库。然而，在大型数据目录中，选择数据源是问题的一部分：代理必须找到请求所需的每个数据库，确定它们的结果如何结合在一起，并执行最终的工作流程。我们提出了DSABench，以共同评估这些能力。其815个经过人工审核的任务来源于Spider、BIRD和LiveSQLBench；其中618个任务需要多个数据库的贡献，105个任务允许对所需源有多种解释。可执行合同定义了源参与、跨源依赖关系和源本地执行，而RouteEnv则允许代理在不透露金标准注释的情况下检查公共目录证据。在我们针对七个模型后端的仅路由研究中，最高的RA-Set得分为60.55%，而在包括协调要求时，最高的RA-Graph得分为34.66%。当需要执行时，这一差距变得更加明显。在四个后端中，中位数非特权的Route+Exec-Graph成功率在没有澄清的情况下范围为6.69%到7.98%，而在有一次澄清的情况下范围为11.44%到13.25%；观察到的最佳澄清配置达到16.99%。在模糊任务中，澄清使得确切的分支源恢复提高了39.8-43.2个百分点。这些发现说明了为什么仅仅检索合理的数据库或生成正确的本地SQL是不够的：代理必须在整个多数据库工作流程中携带正确的信息。

View on arXiv Download PDF AI Translation

cs.AI / 45 / 2610.14707

How to Build a Benchmark That Actually Measures Taste

如何构建一个真正测量品味的基准

Wu, Xiaofan, Zeng, Xi, Chen, Miaoxia, Chen, Peishan, Li, Shuyan, Wang, Chen, Yao, Jiyun, Zhong, Hanyong, Zhu, Jiahao

Abstract

Expert-aligned "taste" benchmarks should reward craft judgment, not the ability to game the test. We introduce TasteBench, a pairwise good-vs-bad benchmark across visual, brand, book, and tech items under a frozen rubric with expert-only gold (panel n>=3; percent agreement >=0.60 primary gate; weak engagement labels train/analysis-only), together with a control protocol that separates craft judgment from test-gaming. Applying the protocol to a Round-1 pilot (primary_human_expert; three biological human experts; n=250 pairs) surfaces and quantifies three validity threats: two construction artifacts that inflate the trivial-baseline layer, and a difficulty ceiling that inflates every score. (1) Position: the preferred item is slot A with probability 0.956, so an always_item_a policy scores 0.956 (fair chance 0.5). (2) Lexical leakage: a filename token cue scores 0.912. (3) Difficulty ceiling: craft domains show near-ceiling human agreement (book/brand/tech 1.0), and even under A/B randomization (eval_assignment_seed7; gold_side_A ~0.47) all five VLM cards reach 1.0 on craft, with only visual separating (~0.84-0.91; a post-hoc hard subset, n=107, leaves craft at 1.0). The protocol (expert-only gold, a weak-label firewall, side randomization at construction, filename scrubbing, and hard filters) makes each artifact measurable and, for the two construction artifacts, correctable; we release a reference eval harness and a side-randomization tool, and show the tool removes the position skew (0.916 to 0.495/0.505). The result is a calibrated view of what the benchmark does and does not yet measure: position and leakage are fixable, whereas the craft-domain ceiling is a pair-difficulty limit that Round-2 must address. We do not evaluate engagement, era/circle conditioning, or calibration here, and we claim no general taste SOTA.

Chinese Translation

与专家对齐的“品味”基准应奖励工艺判断，而不是测试游戏能力。我们介绍了 TasteBench，这是一个在视觉、品牌、书籍和技术项目之间的成对好与坏基准，采用专家专用的金标准（小组人数 n>=3；一致性百分比 >=0.60 作为主要门槛；弱参与标签仅用于训练/分析），并结合一个控制协议，将工艺判断与测试游戏分离。将该协议应用于第一轮试点（主要人类专家；三位生物人类专家；n=250 对）揭示并量化了三个有效性威胁：两个构建伪影膨胀了微不足道的基线层，以及一个难度上限膨胀了每个分数。（1）位置：首选项目在位置 A 的概率为 0.956，因此始终选择项目 A 的策略得分为 0.956（公平机会 0.5）。（2）词汇泄漏：文件名标记提示得分为 0.912。（3）难度上限：工艺领域显示出接近上限的人类一致性（书籍/品牌/技术 1.0），即使在 A/B 随机化下（评估分配种子 7；金标准侧 A ~0.47），所有五个 VLM 卡片在工艺上都达到 1.0，只有视觉部分有所区分（~0.84-0.91；一个事后硬子集，n=107，工艺保持在 1.0）。该协议（专家专用金标准、弱标签防火墙、构建时的侧随机化、文件名清理和硬过滤器）使每个伪影可测量，并且对于两个构建伪影是可纠正的；我们发布了一个参考评估工具和一个侧随机化工具，并展示该工具消除了位置偏差（从 0.916 降至 0.495/0.505）。结果是对基准所能测量和尚未测量的内容的校准视图：位置和泄漏是可修复的，而工艺领域的上限是第二轮必须解决的对难度限制。我们在此不评估参与度、时代/圈子条件或校准，并且不声称有任何通用的品味 SOTA。

View on arXiv Download PDF AI Translation

cs.AI / 46 / 2610.14721

Relational Semantic Grounding for Test-Time Discovery

测试时发现的关系语义基础

Liu, Tianle, Cheng, Gongbo, Ni, Chenggong, Gu, Minming, Zhou, Tao, Hu, Fuyuan, Lyu, Fan

Abstract

Test-Time Discovery (TTD) aims to recognize known classes and detect novel ones from a continuous unlabeled test stream after deployment. Existing TTD methods mainly rely on visual clustering or prototype comparison, and can only assign anonymous labels to discovered classes, leaving them difficult to interpret in open-world applications. In this paper, we propose SEGA, a Semantic Grounding with Attributes framework for TTD. SEGA leverages large language models to generate attribute-rich descriptions for known classes and constructs an attribute-anchored visual-semantic space during training. At test time, we introduce Relational Semantic Grounding (RSG), which grounds unknown samples with similar and dissimilar known-class anchors, prompts the LLM to generate candidate concepts, and reranks them in the learned semantic space. The selected concept is further used for confidence-guided test-time semantic stabilization. By coupling attribute-level semantic alignment with relational concept grounding, SEGA transforms anonymous test-time clusters into interpretable semantic concepts. Experiments show that SEGA improves TTD baselines with the promise of semantic grounding for open-world.

Chinese Translation

测试时发现（TTD）旨在在部署后从连续的未标记测试流中识别已知类别并检测新类别。现有的TTD方法主要依赖于视觉聚类或原型比较，只能为发现的类别分配匿名标签，使其在开放世界应用中难以解释。本文提出了SEGA，一种用于TTD的带属性的语义基础框架。SEGA利用大型语言模型为已知类别生成丰富的属性描述，并在训练期间构建一个以属性为锚的视觉-语义空间。在测试时，我们引入了关系语义基础（RSG），它通过与已知类别锚点的相似和不相似样本将未知样本进行基础，提示大型语言模型生成候选概念，并在学习到的语义空间中对其进行重新排序。所选概念进一步用于信心引导的测试时语义稳定化。通过将属性级语义对齐与关系概念基础相结合，SEGA将匿名的测试时聚类转变为可解释的语义概念。实验表明，SEGA在TTD基准上有所提升，并为开放世界的语义基础提供了希望。

View on arXiv Download PDF AI Translation

cs.AI / 47 / 2610.14727

MHWA: Multi-timescale Hierarchical World-Action Model

MHWA：多时间尺度层次世界-行动模型

Pan, Pengcheng, Ma, Guoqing, Zhang, Yuhan, Chen, Yang, Liu, Yichen, Li, Ziheng, Yu, Shan

Abstract

Autoregressive world models have emerged as a powerful paradigm for sample efficient decision making, yet their open loop rollouts often exhibit increasing drift that leads to error accumulation even within a relatively small planning horizon. Hierarchical temporal abstraction is a natural candidate to mitigate such drift by reducing the effective prediction horizon; however, existing hierarchical designs can introduce added structural and conceptual complexity, sometimes making them difficult to control and limiting gains over flat baselines. To this end, we introduce MHWA (Multi-timescale Hierarchical World-Action model), which employs a strided high level context channel to reduce global context update frequency, effectively curbing the propagation of compounding drift. Crucially, via context aware routing, the gating network dynamically reconfigures a low level mixture of experts by using both global context and current observations, thereby enabling flexible high-low coordination, helping alleviate interference commonly observed in static hierarchies. Empirically, using RLiable's seed-by-game aggregation, MHWA (77M parameters) achieves an IQM human normalized score (HNS) of 0.884 across three training seeds on a 14-game Atari subset using 10% subsampled offline data, outperforming a matched JOWA-70M baseline (0.529) and achieving numerically comparable performance to JOWA-150M while using nearly half the parameters and 70.88% fewer training FLOPs per sequence.

Chinese Translation

自回归世界模型已成为一种高效样本决策制定的强大范式，然而它们的开放循环展开往往表现出逐渐增加的漂移，导致即使在相对较小的规划视野内也会出现误差积累。层次时间抽象是减轻这种漂移的自然候选者，因为它通过减少有效预测视野来实现；然而，现有的层次设计可能会引入额外的结构和概念复杂性，有时使其难以控制，并限制了相较于平坦基线的收益。为此，我们提出了MHWA（多时间尺度层次世界-行动模型），该模型采用跨步高层上下文通道来降低全局上下文更新频率，有效抑制复合漂移的传播。关键是，通过上下文感知路由，门控网络动态重新配置低层专家混合，利用全局上下文和当前观察，从而实现灵活的高低协调，帮助缓解在静态层次结构中常见的干扰。通过使用RLiable的按游戏聚合，MHWA（77M参数）在14个游戏的Atari子集上，使用10%子采样的离线数据，达到了0.884的人类归一化分数（HNS），超越了匹配的JOWA-70M基线（0.529），并在参数数量几乎减半的情况下，达到了与JOWA-150M相当的数值性能，同时每个序列的训练FLOPs减少了70.88%。

View on arXiv Download PDF AI Translation

cs.AI / 48 / 2610.14736

Think in Strategies, Act with Tools: Traceable Agentic Optimization for Combinatorial Problems

以策略思考，以工具行动：可追踪的自主优化用于组合问题

Li, Zhongnian, Xu, Shi, Ying, Peng, Gong, Weikang, Wang, Zhixiao, Xu, Xinzheng

Abstract

Optimization by PROmpting (OPRO) directly leverages Large Language Models (LLMs) to generate candidate solutions through natural-language prompting. However, OPRO's optimization process is difficult to track, which prevents LLMs from capturing useful information accumulated over previous iterations, thereby limiting solution quality. To address this limitation, we propose a Tool-Augmented Strategic Agent (TASA) which summarizes and exploits trajectory to improve solution quality. Specifically, TASA plans the strategy for each step based on the optimization trajectory, which decomposes the optimization process into traceable strategy selections. Moreover, we leverage LLM to construct problem-specific tools and introduce a strategy-guided adaptive tool planner, which efficiently utilizes the information of trajectory to update solutions. We establish local convergence under persistent exploration and evaluate the complete framework under a shared candidate-evaluation budget. The results support its solution quality and the traceability of strategy-guided search.

Chinese Translation

通过提示进行优化（Optimization by PROmpting, OPRO）直接利用大型语言模型（Large Language Models, LLMs）通过自然语言提示生成候选解决方案。然而，OPRO的优化过程难以追踪，这阻碍了LLMs捕捉在先前迭代中积累的有用信息，从而限制了解决方案的质量。为了解决这一限制，我们提出了一种工具增强战略代理（Tool-Augmented Strategic Agent, TASA），它总结并利用轨迹以提高解决方案质量。具体而言，TASA根据优化轨迹为每一步规划策略，将优化过程分解为可追踪的策略选择。此外，我们利用LLM构建特定问题的工具，并引入一种策略引导的自适应工具规划器，有效利用轨迹信息来更新解决方案。我们在持续探索下建立了局部收敛性，并在共享候选评估预算下评估了完整框架。结果支持其解决方案质量和策略引导搜索的可追踪性。

View on arXiv Download PDF AI Translation

cs.AI / 49 / 2610.14745

SAGE: Pushing 2-Bit Vector-Quantized LLMs toward Real-World Usability

SAGE：推动2位向量量化大语言模型的现实可用性

Yan, Xianglong, Wu, Zenan, Zhang, Kewei, Yang, Kaicheng, Zhang, Yulun

Abstract

Extreme low-bit weight-only quantization is an effective approach to reducing the deployment costs of large language models. Below 3 bits, scalar quantization often suffers severe accuracy loss, while vector quantization (VQ) better preserves accuracy through more expressive joint encoding. However, existing VQ methods still exhibit substantial degradation on real-world tasks, and two design challenges remain underexplored: (1) protecting sensitive weight channels within jointly encoded vectors; and (2) allocating a fixed bit budget among representation components, such as direction and radius, to suit different weight matrices. To address these challenges, we propose SAGE: Split Adaptation and Grouping by Energy for weight-only vector quantization. First, Energy-Stratified Grouping (ESG) couples channel grouping with activation-weighted fitting, grouping channels from different energy tiers within each subvector to prioritize sensitive channels. Second, Adaptive Bit Split (ABS) balances direction-codebook capacity and radius precision per matrix, selecting the split with the lowest activation-weighted reconstruction error under a fixed budget. Across benchmarks covering knowledge, mathematics, code generation, instruction following, and tool use, SAGE achieves state-of-the-art performance in the 2-bit quantization setting. On Qwen3-30B-A3B, it retains 95.3% of the BF16 model's average benchmark score. With our custom-designed CUDA kernel, SAGE delivers up to a 2.55-fold non-streaming end-to-end speedup over FP16. The code is available at https://github.com/Zincwolf/sage.

Chinese Translation

极低位权重量化是一种有效降低大型语言模型部署成本的方法。在低于3位的情况下，标量量化通常会遭受严重的准确性损失，而向量量化（VQ）通过更具表现力的联合编码更好地保持准确性。然而，现有的VQ方法在现实任务中仍然表现出显著的性能下降，并且有两个设计挑战尚未得到充分探索：（1）保护联合编码向量中的敏感权重通道；（2）在表示组件（如方向和半径）之间分配固定的位预算，以适应不同的权重矩阵。为了解决这些挑战，我们提出了SAGE：基于能量的权重仅向量量化的分割适应与分组。首先，能量分层分组（ESG）将通道分组与激活加权拟合结合起来，从每个子向量中不同能量层次中分组通道，以优先考虑敏感通道。其次，自适应位分割（ABS）在每个矩阵中平衡方向码本容量和半径精度，在固定预算下选择具有最低激活加权重建误差的分割。在涵盖知识、数学、代码生成、指令跟随和工具使用的基准测试中，SAGE在2位量化设置中实现了最先进的性能。在Qwen3-30B-A3B上，它保留了BF16模型平均基准分数的95.3%。通过我们定制设计的CUDA内核，SAGE在非流式端到端速度上比FP16快了最多2.55倍。代码可在https://github.com/Zincwolf/sage获取。

View on arXiv Download PDF AI Translation

cs.AI / 50 / 2610.14750

Equal-information evaluation of credit assignment in modular financial reinforcement learning

模块化金融强化学习中信用分配的等信息评估

Chistiakov, Gennadiy, Yanovich, Yury

Abstract

Credit assignment in modular financial reinforcement learning is difficult to evaluate because an attributed signal may improve a component loss without adding decision-relevant information or improving economic outcomes. We propose an evaluation framework combining equal-information reconstruction controls, fixed-utility comparisons and prespecified downstream guardrails. The framework is instantiated in an attribution-guided modular trading architecture and tested in three controlled studies covering update allocation, supervision-target representation and portfolio risk control. We find that residual-based allocation is exactly reproduced by an ordinary loss-budget control under the disclosed interface; direct shortfall supervision improves component prediction but fails decoded-outcome preservation; and component-wise critic supervision worsens risk-condition certainty equivalent and drawdown relative to scalar supervision under fixed utility. These results identify three distinct evaluation failures: redundant attribution signals, representation-induced prediction trade-offs, and misalignment between component supervision and downstream risk outcomes. A preserved historical engineering case further shows why module-level learning claims require run-level provenance. The contribution is a reusable methodology for evaluating credit assignment in expert systems operating under partial observability, delayed feedback and economic risk constraints.

Chinese Translation

在模块化金融强化学习中，信用分配的评估较为困难，因为一个归因信号可能会改善组件损失，却并未增加与决策相关的信息或改善经济结果。我们提出了一种评估框架，结合了等信息重构控制、固定效用比较和预设的下游保护措施。该框架在一个以归因为导向的模块化交易架构中得以实现，并在三项控制研究中进行了测试，涵盖了更新分配、监督目标表示和投资组合风险控制。我们的研究发现，基于残差的分配在公开接口下被普通损失预算控制准确再现；直接的短缺监督改善了组件预测，但未能保持解码结果的完整性；而组件级的批评监督在固定效用下相较于标量监督恶化了风险条件的确定性等价和回撤。这些结果识别出三种不同的评估失败：冗余的归因信号、表示引起的预测权衡，以及组件监督与下游风险结果之间的不一致。一个保留的历史工程案例进一步表明，模块级学习的主张需要运行级的来源追溯。我们的贡献是提供了一种可重用的方法论，用于评估在部分可观察性、延迟反馈和经济风险约束下运行的专家系统中的信用分配。

View on arXiv Download PDF AI Translation

cs.AI / 51 / 2610.14754

GateSight: Recurrent Memory Anticipates Attention Memory for Query-Agnostic KV Eviction in Hybrid LLMs

GateSight：递归记忆预判注意力记忆以实现混合大型语言模型中的查询无关键值驱逐

Li, Junjie, Lou, Jiong, Li, Jie

Abstract

Hybrid language models interleave Gated DeltaNet (GDN) layers, whose recurrent memory has a fixed size, with full-attention layers, whose key-value (KV) cache grows with the context and dominates serving memory. In this paper, we ask whether the recurrent memory already knows what the attention memory will need. We first show that which GDN heads of Qwen3.8-27B keep long memories is fixed by the weights rather than by the document. Next, we define the future utility of a KV entry by how much removing it changes the attention output over the model's own continuations, and find that the write gate, decay and write magnitude that GDN layers emit during prefill anticipate this utility before any query is seen. Because the relation is spread across many GDN heads and layers, and a linear map recovers its trend but less of its high-utility tail than a nonlinear one, we propose GateSight, which learns it with lightweight per-head multilayer perceptrons that read these signals and no attention score. Their predicted utility decides which entries each attention layer keeps, and how much each layer keeps is set by the model's native attention output gate, whose mean differs by more than an order of magnitude across layers. Finally, at serving time, GateSight scores the cache once after prefill and physically releases discarded KV entries. Experimental results on SCBench and RULER-4096 with Qwen3.8-27B and, with the same recipe, the sparse-attention model Qwen3.8-Flash-Next, up to the 262K-token limit of the models, show that with a tenth of the attention KV cache, GateSight is the most accurate of five query-agnostic methods on three of the four model-benchmark pairs and leads KVzip on all four, by 9.5 to 47.2 points. Notably, it scores a full-length context with negligible overhead relative to prefill.

Chinese Translation

混合语言模型交替使用门控DeltaNet（GDN）层，其递归记忆具有固定大小，以及全注意力层，其键值（KV）缓存随着上下文增长并主导服务内存。本文探讨递归记忆是否已经知道注意力记忆所需的内容。我们首先展示了Qwen3.8-27B中哪些GDN头保持长记忆是由权重而非文档固定的。接下来，我们通过移除KV条目对模型自身延续的注意力输出变化量来定义KV条目的未来效用，并发现GDN层在预填充期间发出的写入门、衰减和写入幅度在任何查询出现之前就预判了这一效用。由于这种关系分布在多个GDN头和层中，线性映射恢复其趋势但对高效用尾部的捕捉不如非线性映射，因此我们提出了GateSight，它通过轻量级的每头多层感知机学习这些信号，而不依赖于注意力得分。它们预测的效用决定了每个注意力层保留哪些条目，以及每层保留多少条目由模型的本地注意力输出门设置，其均值在不同层之间差异超过一个数量级。最后，在服务时，GateSight在预填充后对缓存进行一次评分，并物理释放被丢弃的KV条目。在SCBench和RULER-4096上的实验结果显示，使用Qwen3.8-27B以及相同方法的稀疏注意力模型Qwen3.8-Flash-Next，在模型的262K标记限制内，GateSight以十分之一的注意力KV缓存，在四对模型基准中三对上是五种查询无关方法中最准确的，并在所有四对中领先KVzip，得分相差9.5到47.2分。值得注意的是，相对于预填充，它在处理完整上下文时的开销微乎其微。

View on arXiv Download PDF AI Translation

cs.AI / 52 / 2610.14763

CUSE: Fine-Grained Multimodal Unlearning with Counterfactual Guidance and On-Policy Self-Distillation

CUSE：基于反事实引导和在线自蒸馏的细粒度多模态遗忘

Ding, Chenlu, Wu, Jiancan, Luo, Yanchen, Zhang, Tianyu, Yuan, Yancheng, Wang, Xiang, He, Xiangnan

Abstract

Multimodal large language models (MLLMs) can reproduce sensitive, copyrighted, or unsafe information from their training data. Effective unlearning must remove target information while preserving non-target content, even within the same response. Existing methods typically apply uniform forgetting objectives to complete responses and rely on fixed, off-policy supervision, resulting in coarse token control and misalignment with the model's current behavior. We propose CUSE, a Counterfactual-guided Unlearning framework via on-policy Self-distillation for multimodal large language models. It first constructs counterfactual inputs by removing target-related evidence and uses the resulting changes in token probabilities to identify which tokens require stronger forgetting. It then distills token-level forgetting and retention targets on responses sampled from the current policy, with a frozen copy of the original model serving as the teacher model augmented with the corresponding forgetting- and retention-oriented privileged contexts. This design jointly determines what to forget and what to preserve. On MLLMU-Bench, CUSE reduces forget-set ROUGE-L by 33.6% while achieving 1.2% higher retain-set ROUGE-L than the strongest baseline.

Chinese Translation

多模态大型语言模型（MLLMs）可能会从其训练数据中再现敏感、受版权保护或不安全的信息。有效的遗忘必须在保留非目标内容的同时移除目标信息，即使在同一响应中也是如此。现有方法通常对完整响应应用统一的遗忘目标，并依赖固定的离线监督，导致粗糙的标记控制和与模型当前行为的不一致。我们提出了CUSE，一个通过在线自蒸馏实现的反事实引导遗忘框架，适用于多模态大型语言模型。该框架首先通过去除与目标相关的证据构建反事实输入，并利用标记概率的变化来识别需要更强遗忘的标记。然后，它在从当前策略中采样的响应上蒸馏标记级别的遗忘和保留目标，使用冻结的原始模型副本作为教师模型，并结合相应的遗忘和保留导向的特权上下文。这一设计共同决定了遗忘和保留的内容。在MLLMU-Bench上，CUSE将遗忘集ROUGE-L降低了33.6%，同时比最强基线提高了1.2%的保留集ROUGE-L。

View on arXiv Download PDF AI Translation

cs.AI / 53 / 2610.14772

How Should a Camera-First AI Agent Work?

相机优先的人工智能代理应如何工作？

Wu, Xiaofan, Zeng, Xi, Chen, Miaoxia, Chen, Peishan, Li, Shuyan, Wang, Chen, Yao, Jiyun, Zhong, Hanyong, Zhu, Jiahao

Abstract

Camera-first agents must act on the visual world as it unfolds. We argue that the central obstacle for future camera-first agents is not recognition ability - multimodal models (vision-language and live/streaming alike) generalize well through language and remain the right tool for semantic reasoning - but time. Many visual targets are ephemeral: they appear, are query-relevant, and leave within a short window. Binding memory to the same end-to-end path as reasoning means that by the time a monolithic pipeline completes, the evidence is gone. We formalize this as Ephemeral-Target Interaction (ETI) and a Perceptual Deadline Contract separating a hard memory deadline from a soft reasoning deadline, grounded in the bounded, passive nature of sensory and working memory. We propose a two-deadline architecture: an always-on, bounded, non-semantic fast path; an active remembering mechanism that commits a moment to durable memory before it is overwritten (event-driven in v1); and a deferred deliberation layer (multimodal model + tools). We instantiate the contract on real MOT17 track windows, a consistency check, and calibrate the single-path baseline from measured frontier-model latency: even a short answer takes 1.2-9.3 s, and an agentic identify-and-search longer. Answers can be late; evidence cannot be missed.

Chinese Translation

相机优先的代理必须在视觉世界展开时采取行动。我们认为，未来相机优先代理的主要障碍不是识别能力——多模态模型（如视觉-语言模型和实时/流媒体模型）通过语言很好地进行泛化，并且仍然是语义推理的合适工具——而是时间。许多视觉目标是短暂的：它们出现、与查询相关，并在短时间内消失。将记忆绑定到与推理相同的端到端路径意味着，当一个单一的管道完成时，证据已经消失。我们将此形式化为短暂目标交互（Ephemeral-Target Interaction, ETI）和感知截止合同（Perceptual Deadline Contract），将严格的记忆截止时间与软性的推理截止时间分开，基于感官和工作记忆的有限、被动特性。我们提出了一种双截止时间架构：一个始终开启的、有界的、非语义的快速路径；一个在被覆盖之前将瞬间提交到持久记忆的主动记忆机制（在版本1中为事件驱动）；以及一个延迟深思层（多模态模型 + 工具）。我们在真实的MOT17轨迹窗口上实例化该合同，进行一致性检查，并根据测量的前沿模型延迟校准单路径基线：即使是一个简短的回答也需要1.2-9.3秒，而一个代理的识别与搜索则更长。答案可能会迟到；证据不能被遗漏。

View on arXiv Download PDF AI Translation

cs.AI / 54 / 2610.14774

ProgAgent: Learning Tool Orchestration through Programmatic Training

ProgAgent：通过程序化训练学习工具编排

Li, Hao, He, Keqing, Zheng, Longtao, Li, Junlong, Zeng, Weihao, Liu, Qian, Huang, Yuzhen

Abstract

Large language models (LLMs) are increasingly expected to orchestrate diverse tools over long-horizon tasks. Existing agents typically use Direct Tool Calling (DTC), where the model invokes tools sequentially, and tool responses accumulate in the model context, which can lead to context rot and orchestration breakdown. In contrast, Programmatic Tool Calling (PTC) invokes tools through executable code and stores intermediate results in runtime variables rather than appending them to the context. Prior work has explored PTC mainly as an inference-time optimization, but its benefits for agentic training remain unclear, and suitable verifiable training tasks are scarce. In this work, we introduce ProgAgent, a full PTC training stack, comprising a pipeline for synthesizing verifiable long-horizon tasks, a lightweight PTC runtime, and a two-stage training recipe. Using only 215 tasks instantiated at four environment scales, ProgAgent yields higher pass rates than the evaluated baselines of comparable size across four benchmarks. On LOCA-bench, ProgAgent-14B achieves a 30.7% pass rate at a 64k-token environment state, where these baselines score zero. Cross-harness experiments show that PTC-trained models outperform DTC-trained models under both evaluation interfaces, suggesting that the benefits of PTC training extend beyond the interface.

Chinese Translation

大型语言模型（LLMs）越来越被期望在长时间跨度的任务中编排多样化的工具。现有的代理通常使用直接工具调用（Direct Tool Calling, DTC），即模型顺序调用工具，并将工具响应累积在模型上下文中，这可能导致上下文衰退和编排失败。相比之下，程序化工具调用（Programmatic Tool Calling, PTC）通过可执行代码调用工具，并将中间结果存储在运行时变量中，而不是将其附加到上下文中。之前的研究主要将PTC视为推理时的优化，但其在代理训练中的好处仍不明确，适合的可验证训练任务也稀缺。在本研究中，我们介绍了ProgAgent，一个完整的PTC训练框架，包括合成可验证长时间跨度任务的管道、轻量级PTC运行时和两阶段训练方案。仅使用215个在四种环境规模下实例化的任务，ProgAgent在四个基准测试中展现出比评估的相似规模基线更高的通过率。在LOCA-bench上，ProgAgent-14B在64k-token环境状态下达到了30.7%的通过率，而这些基线的得分为零。跨平台实验表明，经过PTC训练的模型在两种评估接口下均优于经过DTC训练的模型，表明PTC训练的好处超出了接口的限制。

View on arXiv Download PDF AI Translation

cs.AI / 55 / 2610.14777

LYNX: Analytical Program Synthesis via Grounding and Closed-Loop Refinement

LYNX：通过基础和闭环优化的分析程序合成

Xia, Ruichen, Han, Xiao, Yang, Dingyu, Yang, Luwei, Wu, Sai, Zhang, Dongxiang, Yao, Binhui, Chen, Gang

Abstract

LLM-powered data analytics aims to translate natural language instructions into executable analytical programs over underlying data. However, existing approaches, whether treating LLMs as intent-to-code translators or as standalone analytical engines, often fail to reliably align domain semantics with complex schemas and heterogeneous metadata, resulting in brittle execution and semantically inconsistent analytical outputs in real-world workloads. We present LYNX, a multi-agent framework that formulates LLM-driven analytics as a schema- and knowledge-constrained analysis problem rather than a single-pass generation task. LYNX organizes analytical synthesis into a closed-loop workflow of planning, generation, verification, and refinement through three role-specialized agents that collaboratively enforce schema grounding, semantic consistency, and execution correctness. To address domain ambiguity and schema misalignment, LYNX integrates Semantic-Structural Distillation and Stage-Aware Allocation (SSD-SAA). A delta-guided iterative revision continuously diagnoses and repairs semantic and execution-level failures, substantially improving synthesis robustness. We evaluate LYNX on two NL-to-SQL benchmarks, BIRD-SQL Mini-Dev and the more challenging Spider 2.0-Lite. Experimental results demonstrate strong performance and efficiency, achieving 10.66% to 59.54% execution accuracy across lightweight and commercial models, while reducing token consumption by 70.53% compared with the prior most efficient method. Our framework displays high generalization over different backbones, from open-source, lightweight models to the latest commercial models. Additional case studies on quantitative finance analytics and complex multi-relational SQL workloads further demonstrate the framework's robustness, scalability, and practical applicability in real-world analytical scenarios.

Chinese Translation

基于大语言模型（LLM）的数据分析旨在将自然语言指令转换为可执行的分析程序，以处理底层数据。然而，现有方法，无论是将LLM视为意图到代码的翻译器，还是作为独立的分析引擎，往往无法可靠地将领域语义与复杂的模式和异构元数据对齐，导致在实际工作负载中执行脆弱和语义不一致的分析输出。我们提出了LYNX，一个多代理框架，将基于LLM的分析视为一个受模式和知识约束的分析问题，而不是单次生成任务。LYNX通过三个角色专业化的代理，将分析合成组织成一个闭环工作流程，包括规划、生成、验证和优化，这些代理协同执行模式基础、语义一致性和执行正确性。为了解决领域歧义和模式不对齐问题，LYNX集成了语义-结构蒸馏和阶段感知分配（SSD-SAA）。一种基于增量指导的迭代修订不断诊断和修复语义和执行层面的失败，显著提高了合成的鲁棒性。我们在两个自然语言到SQL的基准测试上评估了LYNX，分别是BIRD-SQL Mini-Dev和更具挑战性的Spider 2.0-Lite。实验结果展示了强大的性能和效率，在轻量级和商业模型中实现了10.66%到59.54%的执行准确率，同时与之前最有效的方法相比，减少了70.53%的令牌消耗。我们的框架在不同的基础模型上表现出高度的泛化能力，从开源的轻量级模型到最新的商业模型。关于定量金融分析和复杂多关系SQL工作负载的额外案例研究进一步展示了该框架在实际分析场景中的鲁棒性、可扩展性和实用性。

View on arXiv Download PDF AI Translation

cs.AI / 56 / 2610.14790

Counterfactual Regret for Reflective Agents with Lean-Verified Guarantees

具有Lean验证保证的反事实遗憾反思代理

Li, Zhongnian, Xu, Shi, Xu, Mengting, Ying, Peng, Xu, Xinzheng

Abstract

Reflective language agents reuse feedback and memory to adapt across tasks without updating model weights. However, existing guarantees for policy selection and action generation do not by themselves establish when local reflection yields long-term performance guarantees for the complete agent under resource constraints. In this paper, we address this gap by deriving sufficient conditions for sublinear counterfactual regret from local generation and deployment guarantees, while retaining the native controller. Specifically, we establish a resource-sensitive transfer that bounds full-state deployment discrepancies along actual execution paths and preserves the own-history laws of both learner and comparator. Besides, local generation reliability yields a cumulative generation opportunity bound, linking policy acquisition to long-term performance through an explicit quality, coverage, and hard-budget tradeoff. To verify these guarantees, we formalize the conditional proof chain for all thirteen numbered theoretical results in Lean, including a native instance with repeated policy growth and imperfect validation.

Chinese Translation

反思语言代理通过重用反馈和记忆在不更新模型权重的情况下适应不同任务。然而，现有的政策选择和行动生成保证本身并不能确定在资源限制下局部反思何时能为完整代理提供长期性能保证。本文通过从局部生成和部署保证中推导出次线性反事实遗憾的充分条件，填补了这一空白，同时保留了原生控制器。具体而言，我们建立了一种资源敏感的转移，限制了沿实际执行路径的全状态部署差异，并保持了学习者和比较者的自历史法则。此外，局部生成的可靠性产生了累积生成机会界限，通过明确的质量、覆盖和硬预算权衡将政策获取与长期性能联系起来。为了验证这些保证，我们在Lean中形式化了所有十三个编号理论结果的条件证明链，包括一个具有重复政策增长和不完美验证的原生实例。

View on arXiv Download PDF AI Translation

cs.AI / 57 / 2610.14792

GREAT-ESDD: Graph Edge Attention Network for Effectively Steered Delivery Decisions

GREAT-ESDD：图边注意力网络用于有效引导交付决策

Lischka, Attila, Zhou, Ze, Xuegang, Ban, Kulcsár, Balázs

Abstract

Online, same-day delivery requires operators to constantly update their routes, and the routing cost for deliveries increases with the number of customers. Out-of-home (OOH) delivery options can help companies handle increasing customer numbers while keeping delivery costs low. We develop a dynamic pricing-and-routing framework for this online ordering scenario with home and OOH delivery options. First, we develop a neural solver for the capacitated vehicle routing problem (CVRP) based on the Graph-Edge-Attention-Network (GREAT). We train it to solve CVRP with varying instance sizes and starting conditions, allowing it to reoptimize partially executed routes under changing conditions. It is fast while achieving solution quality competitive with traditional baselines. Our solver processes several CVRP instances in parallel in batches and we use it to solve the candidate routing problems that arise from an incoming customer's choice of home or OOH delivery. This provides an estimate of the delivery cost increase per option, which we use to set dynamic prices. Further, it directly provides the updated route once the customer decides. We trade off delivery fees against routing costs and steer customers toward OOH delivery when home delivery would be costly. However, effective steering requires us to know the customer preferences. To learn those online, we use Thompson sampling with a Laplace-approximated posterior. We call the resulting framework GREAT-ESDD (GREAT for Effectively Steered Delivery Decisions). Compared to a traditional routing solver, it increases profit by around 11% under a similar pricing decision time budget. Moreover, we show that we can estimate customer preferences successfully, outperforming a static pricing strategy after only 5 days of simulated operations.

Chinese Translation

在线同日交付要求运营商不断更新他们的路线，且交付的路由成本随着客户数量的增加而上升。户外（OOH）交付选项可以帮助公司在保持低交付成本的同时处理不断增加的客户数量。我们为这种具有家庭和户外交付选项的在线订购场景开发了一个动态定价和路由框架。首先，我们基于图边注意力网络（Graph-Edge-Attention-Network，GREAT）开发了一种用于容量受限车辆路由问题（Capacitated Vehicle Routing Problem，CVRP）的神经求解器。我们训练它以解决具有不同实例规模和起始条件的CVRP，使其能够在变化条件下重新优化部分执行的路线。该求解器速度快，同时在解决质量上与传统基线相当。我们的求解器以批处理方式并行处理多个CVRP实例，并利用它来解决因客户选择家庭或户外交付而产生的候选路由问题。这为每个选项提供了交付成本增加的估算，我们用它来设定动态价格。此外，一旦客户做出决定，它会直接提供更新后的路线。我们在交付费用与路由成本之间进行权衡，并在家庭交付成本较高时引导客户选择户外交付。然而，有效的引导需要我们了解客户的偏好。为了在线学习这些偏好，我们使用了带有拉普拉斯近似后验的汤普森采样（Thompson sampling）。我们将得到的框架称为GREAT-ESDD（GREAT用于有效引导交付决策）。与传统路由求解器相比，在相似的定价决策时间预算下，它的利润提高了约11%。此外，我们还展示了在仅经过5天的模拟操作后，我们能够成功估计客户偏好，超越静态定价策略。

View on arXiv Download PDF AI Translation

cs.AI / 58 / 2610.14799

CESNET-TimeSeries26: A Multi-Year Network Traffic Time Series Dataset for Annual Seasonality, Concept Drift, and Long-Term Anomaly Detection

CESNET-TimeSeries26：一个用于年度季节性、概念漂移和长期异常检测的多年度网络流量时间序列数据集

Koumar, Josef, Hynek, Karel, Jerabek, Kamil, Siska, Pavel

Abstract

Forecast-based models are fundamental to network traffic management and anomaly detection. However, it is essential to evaluate these forecasting models over extended periods of time in order to assess their adaptability to unexpected phenomena and shifts in underlying patterns. This kind of long-term evaluation requires multi-year data to account for yearly seasonality and behavioral drift. Existing public datasets are too short to support such long-horizon forecasting tasks. We introduce CESNET-TimeSeries26, comprising 995 days of continuous traffic time series from the CESNET3 network, together with a single subnet spanning approximately 4.5 years. The dataset captures 278,466 IP addresses, 285 institutions, and 566 institutional subnets at three aggregation intervals. The dataset provides a definitive longitudinal benchmark for evaluating forecasting models in both operational management and security monitoring. We demonstrate that annual seasonality is the dominant periodic component. Moreover, we show that a forecasting model can exploit the annual component only when its training window covers at least one complete annual cycle, which no sub-year dataset provides.

Chinese Translation

基于预测的模型是网络流量管理和异常检测的基础。然而，评估这些预测模型在较长时间内的表现对于评估其对意外现象和潜在模式变化的适应性至关重要。这种长期评估需要多年的数据，以考虑年度季节性和行为漂移。现有的公共数据集过于短小，无法支持此类长期预测任务。我们介绍了CESNET-TimeSeries26，包含来自CESNET3网络的995天连续流量时间序列，以及一个跨度约4.5年的单一子网。该数据集捕获了278,466个IP地址、285个机构和566个机构子网，涵盖三个聚合间隔。该数据集为评估预测模型在运营管理和安全监控中的表现提供了明确的纵向基准。我们展示了年度季节性是主要的周期性成分。此外，我们表明，预测模型只有在其训练窗口覆盖至少一个完整的年度周期时，才能利用年度成分，而没有任何子年度数据集能够提供这一点。

View on arXiv Download PDF AI Translation

cs.AI / 59 / 2610.14807

ASIWorld: Benchmarking Autonomous Agents in a Persistent Survival Sandbox

ASIWorld：在持久生存沙盒中基准测试自主智能体

Lin, Jianfeng, Wang, Runze, Hong, Haorong, Xie, Yuxuan, Zeng, Zeno, Liu, Zhongyuan

Abstract

Evaluating autonomous agents requires examining whether they can translate goals into coherent action across extended interactions. In persistent environments, this requires coordinating interdependent actions under incomplete observations, accounting for delayed effects, and adapting as earlier decisions shape subsequent opportunities. We introduce ASIWorld, a survival sandbox that combines persistent world and character state with character-scoped observations and a programmable asynchronous interface to study sustained goal execution. Agents pursue interdependent production and exploration tasks through direct actions or agent-authored programs that continue running between model calls. The evaluation assesses task success, the attainment of development achievements, and survival costs, complemented by an analysis of execution behavior. In an exploratory study of nine agent configurations using a common harness, eight complete the production task and six complete both tasks. Analyses of execution trajectories and performance across difficulty levels reveal different preparation, combat, and recovery behaviors among configurations with similar task outcomes. Incomplete preparation is associated with stalled progression, while configurations that complete the tasks differ in survival costs and the fulfillment of additional achievement criteria. These findings illustrate the value of assessing sustained goal execution through both task outcomes and the behaviors and costs associated with achieving them.

Chinese Translation

评估自主智能体需要考察它们是否能够在长期交互中将目标转化为连贯的行动。在持久环境中，这要求在不完全观察的情况下协调相互依赖的行动，考虑延迟效应，并根据早期决策塑造后续机会进行适应。我们引入了ASIWorld，这是一个生存沙盒，结合了持久世界和角色状态，以及角色范围内的观察和可编程的异步接口，以研究持续的目标执行。智能体通过直接行动或自主编写的程序追求相互依赖的生产和探索任务，这些程序在模型调用之间持续运行。评估内容包括任务成功率、发展成就的达成以及生存成本，并辅以执行行为的分析。在对九种智能体配置的探索性研究中，使用了一个共同的测试框架，其中八个完成了生产任务，六个完成了两个任务。对执行轨迹和不同难度水平下的表现进行分析，揭示了在相似任务结果的配置中，不同的准备、战斗和恢复行为。准备不足与进展停滞相关，而完成任务的配置在生存成本和额外成就标准的满足上存在差异。这些发现展示了通过任务结果以及与实现这些结果相关的行为和成本来评估持续目标执行的价值。

View on arXiv Download PDF AI Translation

cs.AI / 60 / 2610.14818

A Reference Architecture for Integrating Agentic AI into Enterprise Software: Responsibilities, Interactions, and Variability

将自主人工智能集成到企业软件中的参考架构：职责、交互和变异性

Yu, Mingqin, Rabhi, Fethi, Zhu, Liming, Lu, Qinghua

Abstract

Agentic AI moves generative AI applications toward systems that coordinate reasoning, enterprise context, model inference, and external actions. Integrating these capabilities into enterprise software introduces diverse architectural concerns, from orchestration, knowledge and memory, and model access to tools and actions, security, runtime infrastructure, and evaluation. Existing research and industry architectures address many of these concerns but differ in scope, abstraction, responsibility organization, interaction representation, and assumptions about which capabilities are required. This paper presents a technology-neutral reference architecture for enterprise agentic AI integration. We synthesize architectural capabilities from research publications, industry architecture guidance, and interoperability specifications by architectural purpose, organize them into ten responsibilities, and define boundary invariants, typed cross-responsibility interactions, and variation points that distinguish responsibilities from condition-dependent capabilities and replaceable realization mechanisms. We instantiate the architecture as an ESG portfolio-analysis and pre-trade compliance agent over an existing service-oriented ESG metric computation and reporting system, and complement the case with a mapping of an independently published agentic architecture. The case realizes seven of the ten responsibilities; different questions take different interaction paths, controls activate only under their conditions, and the probed boundary invariants hold. The independent architecture maps onto the proposed responsibilities without requiring an additional one. These results provide initial evidence for the architecture's organization and applicability rather than its completeness. The deployment and recorded results are available as a replication package.

Chinese Translation

自主人工智能（Agentic AI）将生成性人工智能应用推进到协调推理、企业上下文、模型推断和外部行动的系统中。将这些能力集成到企业软件中引入了多样的架构问题，包括编排、知识与记忆、模型访问、工具与行动、安全性、运行时基础设施和评估。现有的研究和行业架构解决了许多这些问题，但在范围、抽象、职责组织、交互表示以及对所需能力的假设上存在差异。本文提出了一种技术中立的企业自主人工智能集成参考架构。我们根据架构目的从研究出版物、行业架构指南和互操作性规范中综合架构能力，将其组织为十项职责，并定义边界不变性、类型化跨职责交互和变异点，以区分职责与条件依赖能力以及可替换的实现机制。我们将该架构实例化为一个现有服务导向的环境、社会和治理（ESG）指标计算与报告系统上的ESG投资组合分析和交易前合规代理，并通过一个独立发布的自主架构的映射来补充该案例。该案例实现了十项职责中的七项；不同的问题采取不同的交互路径，控制仅在其条件下激活，探测的边界不变性成立。独立架构映射到所提出的职责上，而不需要额外的职责。这些结果为架构的组织和适用性提供了初步证据，而非其完整性。部署和记录结果作为复制包可供获取。

View on arXiv Download PDF AI Translation

cs.AI / 61 / 2610.14828

EvoCowork: A Benchmark for Real-World Office Tasks

EvoCowork：真实办公任务的基准测试

Liu, Zhou, Chen, Xingyan, Han, ZhaoYang, Pan, Zewei, Su, Zeli, Pan, Ningxin, Hu, Xue, Ting, Kaixiong, Zhang, Wentao

Abstract

Office workflows are a demanding testbed for self-improving agents because their deliverables must be correct, natively editable, and mutually consistent. A central question is whether gains on feedback tasks reflect reusable procedures rather than task-specific fixes. We introduce EvoCowork, a 240-task benchmark for evaluating harness-level recursive self-improvement across four office modalities, with model weights held fixed. The benchmark covers presentations, documents, spreadsheets, and cross-office workflows, and scores submitted artifacts for content, native structure, formulas, cross-file consistency, and visual quality where needed. Its protocol separates feedback, revision selection, and final testing: methods learn from single-modality tasks, then are evaluated on unseen single-modality tasks and cross-office workflows under a shared initial harness and explicit budgets. In static evaluations of two models and four harnesses, the best configuration scores 42.25 overall and 30.62 on cross-office tasks. Starting from the same MSA harness with GLM 5.3, ACE and AHE improve in-domain scores by 4.90 and 2.64 points, respectively, but change cross-office scores by only +0.03 and -1.72. Thus, gains on the file types that provide feedback do not necessarily transfer to coordinated workflows. EvoCowork makes this distinction explicit while recording the capability, transfer, and cost of harness evolution. Code and task packages: https://github.com/DeepThinkingZhouLiu/cowork-evolution-benchmark.

Chinese Translation

办公工作流程是自我改进代理的一个严苛测试平台，因为其交付成果必须是正确的、可原生编辑的，并且相互一致。一个核心问题是，反馈任务上的收益是否反映了可重用的程序，而不是特定任务的修复。我们引入了EvoCowork，这是一个包含240个任务的基准，用于评估在四种办公模式下的高效递归自我改进，模型权重保持不变。该基准涵盖了演示文稿、文档、电子表格和跨办公室工作流程，并对提交的成果在内容、原生结构、公式、跨文件一致性和视觉质量等方面进行评分。其协议将反馈、修订选择和最终测试分开：方法从单一模式任务中学习，然后在共享的初始环境和明确的预算下，对未见过的单一模式任务和跨办公室工作流程进行评估。在对两个模型和四个环境的静态评估中，最佳配置的总分为42.25，跨办公室任务得分为30.62。从相同的MSA环境和GLM 5.3开始，ACE和AHE在领域内得分分别提高了4.90和2.64点，但跨办公室得分仅变化+0.03和-1.72。因此，提供反馈的文件类型上的收益并不一定会转移到协调工作流程中。EvoCowork明确区分了这一点，同时记录了环境演化的能力、转移和成本。代码和任务包： https://github.com/DeepThinkingZhouLiu/cowork-evolution-benchmark.

View on arXiv Download PDF AI Translation

cs.AI / 62 / 2610.14843

Pre-Registered Genesis Accounting for Recursive Self-Improvement: First Audited Operator-Genesis Events in a Physics-Discovery Runtime

预注册的递归自我改进会计：物理发现运行时中的首次审核操作员-生成事件

Mao, Wenbin, Zou, Pei

Abstract

Claims of recursive self-improvement are easy to make and hard to falsify: did improvement come from re-using designer-provided mechanisms, or from mechanisms the system authored itself? We introduce a pre-registered, two-tier accounting: M1 counts machine-generated content instances within a designer-provided vocabulary, while M2 counts machine-authored mechanism types (new operators registered into the runtime's own vocabulary, with a held-out-validation certificate). We report the first audited M2 genesis events. In the primary event, depth-limited compositional search over elementary operators, judged by held-out R^2 on real NIST hydrogen Balmer measurements, produced the operator sq(inv(n)) = 1/n^2 -- the Rydberg-level form -- which no standard-library transform could express (held-out R^2: -0.546 best standard transform vs. 1.0000000 composed). A protocol flaw (independent variable built by inverting the very law under test) was caught in audit and repaired via an independent-label protocol; extrapolation to unseen levels (n=8,9 trained on n<=7) attains relative error ~10^-6, four to six orders beyond the best standard operator. A second event independently recovered sqrt(inv(a)) -- Kepler's velocity law -- from JPL ephemeris vectors. Also incorporated: an audited compounding channel (first domain-level measurement: 1.109x accuracy ratio on a real held-out domain) and blind, decurated genesis exams whose honest tally is zero. The Kepler event passes line audit; the hydrogen gap reproduces on He II; the certificate is threshold-stable. Against eleven prior systems, the contribution is the measurement and governance of machine-authored mechanisms, not the search algorithm or rediscovered physics. Limitations: exams were curated by agents who knew the textbook answer, triggers manual, grammar observer-injected; both events demonstrate capability to author, not yet autonomous genesis.

Chinese Translation

递归自我改进的声明容易提出，但难以证伪：改进是来自于重新使用设计者提供的机制，还是来自于系统自主创作的机制？我们引入了一种预注册的两级会计：M1 计算设计者提供的词汇中机器生成的内容实例，而 M2 计算机器创作的机制类型（新操作符注册到运行时自身的词汇中，并附有保留验证证书）。我们报告了首次审核的 M2 生成事件。在主要事件中，对基本操作符进行深度限制的组合搜索，通过对真实的 NIST 氢巴尔末测量的保留 R^2 进行评估，产生了操作符 sq(inv(n)) = 1/n^2——即赖德堡级别的形式——而没有任何标准库变换能够表达（保留 R^2：-0.546 最佳标准变换对比 1.0000000 组合）。在审核中发现了一项协议缺陷（独立变量是通过反转正在测试的法律构建的），并通过独立标签协议进行了修复；对未见水平的外推（n=8,9 在 n<=7 上训练）达到了相对误差 ~10^-6，超越最佳标准操作符四到六个数量级。第二个事件独立恢复了 sqrt(inv(a))——开普勒的速度法则——来自 JPL 星历向量。同时还包含：一个审核的复合通道（首次领域级测量：在真实保留领域上的准确率比为 1.109x）和盲目、去策划的生成考试，其真实计数为零。开普勒事件通过了线性审核；氢差距在 He II 上重现；证书是阈值稳定的。与之前的十一种系统相比，贡献在于对机器创作机制的测量和治理，而不是搜索算法或重新发现的物理学。局限性：考试由知道教科书答案的代理人策划，触发手动，语法观察者注入；两个事件展示了创作能力，但尚未实现自主生成。

View on arXiv Download PDF AI Translation

cs.AI / 63 / 2610.14855

PEAM: On-device Communication Agent for Personalized, Efficient and Autonomous Mobile Network Configuration

PEAM：用于个性化、高效和自主移动网络配置的设备内通信代理

Wan, Haowen, Xie, Shangzhuo, Syue, Syue-Ju, Wu, Xianyue, Li, Runhua, Yang, Qianqian

Abstract

The growing deployment of compact large language models on smartphones is enabling autonomous on-device decision-making, such as emerging mobile GUI agents. However, many mobile decisions, such as network configuration when users switch apps or connectivity conditions change, must satisfy stringent latency constraints to preserve user experience. In this paper, we propose PEAM, an on-device LLM-based communication agent that autonomously generates personalized network-configuration rules executable within milliseconds. Specifically, PEAM first fine-tune the on-device LLM with a dataset constructed from manually annotated real-user data to integrate the communication knowledge into the local model. PEAM then operates in two phases. In the rule-generation phase, which is performed when the device is idle, such as overnight, PEAM analyzes historical network-configuration records in batches and reflects across batches to consolidate consistent evidence, reduce local bias, and derive robust configuration rules. To enable more accurate and stable rule generation over time, PEAM further constructs a user profile from high-confidence rules to capture the user's network preferences and incorporates this profile into each rule-generation cycle. In the rule-execution phase, PEAM is invoked when the communication context changes, such as when users switch apps or connectivity conditions change, and searches the rule library to identify and execute the appropriate rule. Experiments demonstrate that PEAM, based on a fine-tuned 4B model, achieves 99.13% correct network configuration decisions, while 92.50% of the rules generated by PEAM are valid, outperforming the traditional expert-rule based method by 4.41% and 84.50%, respectively.

Chinese Translation

随着紧凑型大型语言模型在智能手机上的广泛部署，正在实现自主的设备内决策，例如新兴的移动图形用户界面代理。然而，许多移动决策，例如用户切换应用程序或连接条件变化时的网络配置，必须满足严格的延迟限制，以维护用户体验。本文提出了PEAM，一种基于设备内大型语言模型的通信代理，能够自主生成可在毫秒内执行的个性化网络配置规则。具体而言，PEAM首先使用从手动标注的真实用户数据构建的数据集对设备内大型语言模型进行微调，以将通信知识整合到本地模型中。然后，PEAM分两个阶段进行操作。在规则生成阶段（在设备空闲时执行，例如过夜），PEAM批量分析历史网络配置记录，并在批次之间进行反思，以巩固一致的证据，减少本地偏差，并推导出稳健的配置规则。为了在时间上实现更准确和稳定的规则生成，PEAM进一步从高置信度规则构建用户档案，以捕捉用户的网络偏好，并将该档案纳入每个规则生成周期。在规则执行阶段，当通信上下文发生变化时（例如用户切换应用程序或连接条件变化时），PEAM被调用，搜索规则库以识别并执行适当的规则。实验表明，基于微调的4B模型，PEAM实现了99.13%的正确网络配置决策，而PEAM生成的规则中有92.50%是有效的，分别比传统的基于专家规则的方法提高了4.41%和84.50%。

View on arXiv Download PDF AI Translation

cs.AI / 64 / 2610.14863

WeatherBridge: Forecast-Aware Multi-Source Input Adaptation for Pretrained Weather Forecasting Models

WeatherBridge：面向预测的多源输入适配用于预训练天气预测模型

Xi, Mutian, Su, Xinyu, Zhong, Xiaohui, Li, Hao

Abstract

Data-driven weather forecasting models are commonly trained on ERA5, yet operational deployment must use timely initial states from physics-based numerical weather prediction (NWP) systems whose distributions differ substantially from ERA5. Direct initialization with these states can degrade autoregressive forecasts, motivating input adaptation that keeps the pretrained forecasting model unchanged. Here, we introduce WeatherBridge, a unified, source-conditioned adapter that maps heterogeneous operational NWP initial states to ERA5 reference states without updating the pretrained forecasting model. WeatherBridge predicts the structured residual corrections with a Swin-UNet and injects source embeddings across multiple spatial scales, allowing the model to capture shared atmospheric structure and source-specific discrepancies. Its training objective combines latitude-weighted ERA5 reconstruction with One-step Rollout Alignment (ORA), which backpropagates one-step forecast error through a frozen FuXi model to optimize the adapted states for both initial state fidelity and downstream forecast compatibility. We evaluate WeatherBridge using daily GFS, HRES, and CMA initializations from 1 January to 22 November 2025 across six variables and lead times up to 240 hours. Relative to raw source initialization of FuXi, WeatherBridge reduces the mean RMSE of FuXi over the 6- 240-hour rollout for 17 of 18 source-variable combinations, with a median reduction of 10.2% and a maximum reduction of 46.3%. The adapter trained with frozen FuXi also improves FengWu forecasts on GFS inputs without retraining either the adapter or FengWu. These results demonstrate that forecast-aware input adaptation can improve forecasts from heterogeneous operational NWP inputs and transfer across pretrained ERA5-based forecasting models.

Chinese Translation

基于数据的天气预测模型通常在ERA5上进行训练，但实际部署必须使用来自物理基础数值天气预报（NWP）系统的及时初始状态，这些状态的分布与ERA5有显著差异。直接使用这些状态进行初始化可能会降低自回归预测的准确性，因此需要一种输入适配方法，以保持预训练预测模型不变。在此，我们介绍了WeatherBridge，一种统一的源条件适配器，它将异构的操作性NWP初始状态映射到ERA5参考状态，而无需更新预训练的预测模型。WeatherBridge使用Swin-UNet预测结构化残差修正，并在多个空间尺度上注入源嵌入，从而使模型能够捕捉共享的大气结构和源特定的差异。其训练目标结合了纬度加权的ERA5重建与一步滚动对齐（One-step Rollout Alignment, ORA），通过冻结的FuXi模型反向传播一步预测误差，以优化适配状态，确保初始状态的保真度和下游预测的兼容性。我们使用2025年1月1日至11月22日的每日GFS、HRES和CMA初始化，对六个变量和最长240小时的预测时间进行评估。与FuXi的原始源初始化相比，WeatherBridge在18个源-变量组合中的17个上减少了FuXi在6-240小时滚动预测中的平均均方根误差（RMSE），中位数减少幅度为10.2%，最大减少幅度为46.3%。使用冻结FuXi训练的适配器也在GFS输入上改善了FengWu的预测，而无需重新训练适配器或FengWu。这些结果表明，面向预测的输入适配可以改善来自异构操作性NWP输入的预测，并在基于预训练ERA5的预测模型之间进行迁移。

View on arXiv Download PDF AI Translation

cs.AI / 65 / 2610.14877

Beyond Outcome Feedback: Trajectory-Quality-Guided Experience-based Learning

超越结果反馈：基于轨迹质量的经验学习

Zhang, Bingjie, Yang, Yibo, Liu, Qiantong, Fu, Benshuo, Chang, Yi, Guo, Dandan

Abstract

Experience-based learning enables large language model (LLM) agents to improve over time by reusing knowledge from past executions. However, existing methods typically rely on final task outcomes as feedback for experience learning, overlooking the execution process and limiting their ability to evaluate and optimize experience based on trajectory quality. To address this limitation, we propose TQG-Exp, a trajectory-quality-guided experience-based learning framework that introduces process feedback. Specifically, we introduce Trajectory Quality Loss (TQL), which jointly considers task completion and process quality to provide a learning signal beyond final task outcomes. Based on this signal, we employ paired rollouts with and without retrieved experience to guide experience optimization. Representative trajectories are selected based on their outcomes and trajectory quality to extract task-level and task-type-level experiences. Meanwhile, differences in trajectory quality between the paired groups are used to estimate the contribution of retrieved experience and update its utility. The learned utility is then combined with semantic relevance to guide subsequent experience retrieval. Experiments on ALFWorld, WebShop, and ScienceWorld with two backbone LLMs show that TQG-Exp outperforms existing experience-based methods, improving Avg.@3 over the strongest baseline by 4.50 and 8.55 percentage points on DeepSeek-V4-Flash and Qwen3.5-Flash, respectively.

Chinese Translation

基于经验的学习使大型语言模型（LLM）代理能够通过重用过去执行的知识而随着时间的推移不断改进。然而，现有方法通常依赖于最终任务结果作为经验学习的反馈，忽视了执行过程，从而限制了它们基于轨迹质量评估和优化经验的能力。为了解决这一局限性，我们提出了TQG-Exp，一个基于轨迹质量的经验学习框架，该框架引入了过程反馈。具体而言，我们引入了轨迹质量损失（Trajectory Quality Loss, TQL），该损失共同考虑任务完成情况和过程质量，以提供超越最终任务结果的学习信号。基于这一信号，我们采用了带有和不带有检索经验的配对回滚方法来指导经验优化。根据结果和轨迹质量选择代表性轨迹，以提取任务级和任务类型级的经验。同时，利用配对组之间的轨迹质量差异来估计检索经验的贡献并更新其效用。学习到的效用随后与语义相关性结合，以指导后续经验检索。在ALFWorld、WebShop和ScienceWorld上的实验显示，TQG-Exp在两个主干LLM上优于现有的基于经验的方法，在DeepSeek-V4-Flash和Qwen3.5-Flash上分别提高了4.50和8.55个百分点的Avg.@3，相较于最强基线。

View on arXiv Download PDF AI Translation

cs.AI / 66 / 2610.14883

MARL-BattleGrounds: A Novel JAX-native Benchmark for Heterogeneous and Competitive Multi-agent RL

MARL-BattleGrounds：一种新颖的JAX原生基准，用于异质和竞争性的多智能体强化学习

Hawili, Ulixes Tariq, Ramamoorthy, Subramanian, Altmann, Yoann, Carlucho, Ignacio

Abstract

We present MARL-BattleGrounds (MARL-BGs), an easy-to-use JAX-native benchmark for heterogeneous and competitive multi-agent reinforcement learning (MARL), designed to be sample-efficient, computationally efficient, and researcher-centric. MARL benchmarks are typically simple and efficient, like SMAX, where winning reduces to unit micromanagement, or rich and costly, like Dota 2. MARL-BGs fills this missing middle. Its first mode, Team Deathmatch, is a simultaneous turn-based game between two teams of five heterogeneous agents. The five classes have distinct basic, passive, and ultimate abilities designed to synergize with allies and counter enemies. Our 52 hand-designed maps with diverse obstacles include a 12-map curriculum and ten held-out validation and test maps. MARL-BGs provides two canonical evaluation modes. First, eight endgame scenarios form a hard generalization challenge: each asks a team to win an unseen, nearly lost position against a strong reactive controller within 5-10 ticks. Second, a monthly round-robin tournament retrains each submission on equal compute, plays it against every member of the current "Big Nine" field, and publishes the Elo results on the official leaderboard. All entrants are publicly released as downloadable opponents. MARL-BGs records 11,192 metrics per game in 28 topics. We also include three strong scripted opponents, an analytical replay viewer, and off-the-shelf tools for self-play, cross-play, curriculum learning, and zero-shot coordination evaluation. We even provide an API for LLM-based agents and use it to evaluate Qwen3.5-9B against our strongest policy. In our BattleClient, humans can play alongside or against any mix of policies, enabling human-AI evaluation. We benchmark MARL-BGs with six additional deep MARL baselines, which we release. On one RTX 5090 with 1,024 parallel games, the simulator runs 50,000 5v5 game steps per second.

Chinese Translation

我们提出了MARL-BattleGrounds（MARL-BGs），这是一个易于使用的JAX原生基准，旨在支持异质和竞争性的多智能体强化学习（MARL），并设计为样本高效、计算高效且以研究者为中心。现有的MARL基准通常简单而高效，例如SMAX，其中获胜归结为单位微观管理，或复杂且昂贵，例如Dota 2。MARL-BGs填补了这一缺失的中间地带。其第一个模式是团队死亡竞赛（Team Deathmatch），这是一个由两队五个异质智能体进行的同时回合制游戏。这五个类别具有不同的基本、被动和终极技能，旨在与盟友协同作战并对抗敌人。我们设计了52个具有多样障碍的地图，其中包括12张课程地图和10张保留的验证与测试地图。MARL-BGs提供了两种典型的评估模式。首先，八个残局场景构成了一个困难的泛化挑战：每个场景要求一个团队在5-10个回合内在一个未见过的、几乎失利的位置上战胜一个强大的反应控制器。其次，每月的循环锦标赛在相同的计算条件下重新训练每个提交，针对当前“九强”中的每个成员进行对战，并在官方排行榜上公布Elo结果。所有参赛者都公开发布为可下载的对手。MARL-BGs每场游戏记录11,192个指标，涵盖28个主题。我们还包括三个强大的脚本对手，一个分析重播查看器，以及现成的自我对战、交叉对战、课程学习和零样本协调评估工具。我们甚至为基于LLM的智能体提供API，并利用它评估Qwen3.5-9B与我们最强策略的对比。在我们的BattleClient中，人类可以与任何策略的组合一起或对抗游戏，实现人机评估。我们使用六个额外的深度MARL基线对MARL-BGs进行基准测试，并将其发布。在一台配备RTX 5090的计算机上，模拟器以每秒50,000个5v5游戏步骤的速度运行1,024个并行游戏。

View on arXiv Download PDF AI Translation

cs.AI / 67 / 2610.14912

Probable but Not Valid: A Psychometric Audit of Decision Models as Synthetic Survey Respondents

可能但不有效：决策模型作为合成调查响应者的心理测量审计

Lukauskas, Mantas, Šarkauskaitė, Viktorija

Abstract

Language models are increasingly used to create synthetic survey samples, also called silicon samples. Decision models are a new class of language models that return a probability for each point of a rating scale instead of text. We ask whether these models, or the newest generative LLMs, can stand in for human respondents on a psychometric survey. We test fourteen decision models from eleven providers and 47 generative LLMs on an open Lithuanian organisational-psychology survey (n=263, 68 items), with 100 real respondent profiles, the Psychometric Similarity Score (PSS) and statistical baselines. The best generative model, MiniMax M3, reaches PSS 0.813. This is level with a Gaussian copula fit only on other respondents (0.824) and below a copula fit in sample (0.927). A second sample of 100 real people scores below the best models, so at this size PSS cannot rank them against real respondents. In 5 of 10 product lines the newest generative release scores below an earlier release of the same line beyond resampling noise, and none gains beyond it. With the default rounded readout, 13 of 14 decision models score below the generative reference gpt-5.4-mini. Of the 47 generative models, 33 reproduce the pattern of item correlations better than any decision model. Averaged over respondents, the probability vectors are much closer to the human item marginals than rounded answers. They also rank respondents about as well as a profile regression fit on 163 real people. As forecasts of each person's own answers, however, every decision model is worse than the answer frequencies of other respondents, mainly because its average vector differs from them. The profile regression improves on these frequencies by only 1.0%, so the profile fields say little about how a person answers. Synthetic respondents built on such profiles should therefore aim at sample-level fidelity.

Chinese Translation

语言模型越来越多地用于创建合成调查样本，也称为硅样本。决策模型是一类新的语言模型，它为评分尺度的每个点返回一个概率，而不是文本。我们探讨这些模型或最新的生成型大语言模型（LLMs）是否可以在心理测量调查中代替人类响应者。我们在一项开放的立陶宛组织心理学调查中（n=263，68个条目）测试了来自11个提供者的14个决策模型和47个生成型LLMs，使用了100个真实响应者的档案、心理测量相似性评分（PSS）和统计基线。最佳的生成模型MiniMax M3达到了PSS 0.813。这一水平与仅基于其他响应者的高斯耦合拟合（0.824）相当，但低于样本内的耦合拟合（0.927）。第二个100名真实受访者的样本得分低于最佳模型，因此在这个样本大小下，PSS无法将它们与真实响应者进行排名。在10个产品线中，有5个最新的生成版本得分低于同一产品线的早期版本，超出了重抽样噪声，且没有一个超越它。使用默认的四舍五入读数，14个决策模型中有13个得分低于生成参考gpt-5.4-mini。在47个生成模型中，有33个在条目相关性模式的再现上优于任何决策模型。平均而言，响应者的概率向量与人类条目边际的接近程度远高于四舍五入的答案。它们在排名响应者方面的表现与基于163名真实人群的档案回归拟合相当。然而，作为每个人自己答案的预测，所有决策模型的表现都不如其他响应者的答案频率，主要因为其平均向量与其他响应者不同。档案回归仅提高了这些频率1.0%，因此档案字段对一个人如何回答几乎没有提供信息。因此，基于此类档案构建的合成响应者应旨在实现样本级的保真度。

View on arXiv Download PDF AI Translation

cs.AI / 68 / 2610.14913

Bridging Risk Approximation Gaps in Model Predictive Task Sampling via In-Context Modeling

通过上下文建模弥补模型预测任务采样中的风险近似差距

Wen, Jiarong, Tao, Qi, Zhang, Kaiyu, Qu, Yun, Cai, Lizhou, Zou, Heming, Jiang, Yuhang, Wang, Qi

Abstract

Reinforcement learning (RL) is essential for adaptive decision-making, e.g., robust Meta-RL and domain randomization (DR), and efficient post-training of large language models (LLMs), where tasks are commonly structured as Markov decision processes or prompts. A central bottleneck in all these settings is the cost of agent-environment interactions required for task selection and policy optimization. Model predictive task sampling (MPTS) has emerged as a promising approach to improve sample efficiency by querying informative tasks via a risk predictive model (RPM) trained on optimization history. However, existing RPMs suffer from two failure modes: (i) sparse optimization histories, which limit RPMs' generalization across the task space, and (ii) a one-step lag in risk approximation, which causes biased difficulty estimation as the policy evolves. This work formalizes these failure modes, derives lagged-risk decompositions that expose the resulting temporal bias, and recasts risk prediction as an in-context sequence modeling problem. The developed RPM integrates a one-step look-ahead mechanism with a temporal-weighted sliding window to mitigate data non-stationarity and scarcity, without requiring additional rollouts. Empirically, our approach yields more reliable difficulty estimates and consistent performance gains across robust Meta-RL, DR, and prompt curriculum RL post-training of LLMs.

Chinese Translation

强化学习（RL）对于自适应决策至关重要，例如稳健的元强化学习（Meta-RL）和领域随机化（DR），以及大型语言模型（LLMs）的高效后训练，其中任务通常被构建为马尔可夫决策过程或提示。在所有这些设置中，一个核心瓶颈是任务选择和策略优化所需的代理-环境交互成本。模型预测任务采样（MPTS）作为一种有前景的方法，通过查询基于优化历史训练的风险预测模型（RPM）来提高样本效率。然而，现有的RPM存在两种失效模式：（i）稀疏的优化历史，限制了RPM在任务空间中的泛化能力；（ii）风险近似的一步滞后，导致随着策略演变而产生偏差的难度估计。本研究对这些失效模式进行了形式化，推导出滞后风险分解，揭示了由此产生的时间偏差，并将风险预测重新表述为一个上下文序列建模问题。所开发的RPM集成了一种一步前瞻机制和时间加权滑动窗口，以减轻数据的非平稳性和稀缺性，而无需额外的回合。实证结果表明，我们的方法在稳健的Meta-RL、DR和LLMs后训练的提示课程RL中提供了更可靠的难度估计和一致的性能提升。

View on arXiv Download PDF AI Translation

cs.AI / 69 / 2610.14915

WorkBorne: Discovering Agentic Tasks from Real-World Artifact Transitions

WorkBorne：从真实世界工件转变中发现代理任务

Zhou, Junyi, Wang, Yufei, Zhang, Qiyuan, Wu, Chuhan, Li, Zhengxi, Zhao, Wentao, Zhou, Jie, Tian, Le, Zhou, Xiao, Ma, Chen

Abstract

Developing terminal-based agents for long-horizon execution demands synthetic tasks that tightly couple complex instructions with authentic, interactive environments. However, current automated synthesis suffers from a fundamental asymmetry: instruction-oriented methods fabricate low-fidelity, synthetic environments, whereas resource-oriented methods treat real artifacts merely as static inputs, degrading task instructions into shallow, trivial operations. To address this limitation, we introduce WorkBorne, a synthesis framework that uses artifact transitions as the central primitive for modeling the full life-cycle of agentic tasks. In WorkBorne, each transition specifies who performs the task, which artifacts anchor the initial state and intended outcome, and the corresponding relationship between them. Specifically, WorkBorne first constructs coherent and diverse artifact corpora from iterative web retrieval, then discovers non-trivial transitions by identifying meaningful input-output artifact assemblies and the relations that connect them. Each discovered transition is finally instantiated into grounded agentic tasks. We construct 1,061 artifact corpora and 7,882 task environments, yielding 15,485 high-quality trajectories, with 53.6% exceeding 30 assistant turns. Experiments on Qwen3.5-27B and Qwen3.5-35B-A3B show that WorkBorne substantially improves over base models (i.e., improving Terminal-Bench 2.1 from 43.1% to 61.8% and ClawEval General score from 75.0 to 80.2) and achieves the strongest overall performance among other synthesis baselines (e.g., TMax (Ivison et al., 2026), RST (Li et al., 2026a), Terminal-Universe (Wu et al., 2026a)). Further analysis shows that transition is strongly associated with execution depth, while corpus breadth and transition depth provide complementary scaling gains.

Chinese Translation

开发基于终端的代理以进行长时间执行需要将复杂指令与真实的互动环境紧密结合的合成任务。然而，目前的自动合成存在一个根本的不对称性：以指令为导向的方法制造出低保真度的合成环境，而以资源为导向的方法则仅将真实工件视为静态输入，从而将任务指令降级为浅显的、微不足道的操作。为了解决这一局限性，我们提出了WorkBorne，一个以工件转变为核心原语来建模代理任务完整生命周期的合成框架。在WorkBorne中，每个转变都指定了谁执行任务、哪些工件锚定初始状态和预期结果，以及它们之间的相应关系。具体而言，WorkBorne首先通过迭代的网络检索构建连贯且多样的工件语料库，然后通过识别有意义的输入-输出工件组合及其连接关系来发现非平凡的转变。每个发现的转变最终被实例化为有根基的代理任务。我们构建了1,061个工件语料库和7,882个任务环境，产生了15,485条高质量轨迹，其中53.6%超过30个助手回合。在Qwen3.5-27B和Qwen3.5-35B-A3B上的实验表明，WorkBorne显著优于基线模型（即，Terminal-Bench 2.1从43.1%提高到61.8%，ClawEval General分数从75.0提高到80.2），并在其他合成基线中实现了最强的整体性能（例如，TMax (Ivison et al., 2026)，RST (Li et al., 2026a)，Terminal-Universe (Wu et al., 2026a)）。进一步分析表明，转变与执行深度密切相关，而语料库的广度和转变的深度则提供了互补的扩展收益。

View on arXiv Download PDF AI Translation

cs.AI / 70 / 2610.14918

SkillForgeBench: Evaluating Agent Reliability Grounded in Agent Skill Specifications

SkillForgeBench：基于智能体技能规格评估智能体可靠性

Li, Zhengxi, Zhou, Junyi, Zhang, Qiyuan, Wu, Chuhan, Lyu, Fuyuan, Tian, Le, Zhou, Xiao, Ma, Chen, Wang, Yufei

Abstract

Evaluating an agent's ability to strictly satisfy diverse and procedurally complex requirements is essential to determining its practical reliability. Conducting such an evaluation demands agentic tasks that reconcile broad domain coverage with deep procedural complexity. Existing paradigms face a fundamental trade-off: expert-curated benchmarks achieve procedural depth but suffer from domain diversity bottlenecks, whereas crowdsourced benchmarks span diverse scenarios but lack execution rigor. To resolve this tension, we present SkillForgeBench, which leverages agent skills as unifying execution blueprints across three co-designed stages: (1) Resource Preparation uses domain-balanced skills to guide artifact retrieval, establishing broad workflow coverage; (2) Skill-driven Task Synthesis employs skills as procedural anchors to partition artifact roles, synthesize naturalistic task instructions, and package sandboxed scenarios for deep procedural execution; and (3) Holistic Evaluation Design translates skill-embedded requirements into fine-grained completion and process rubrics to enable strict requirement-level scoring. Comprising 842 sandboxed tasks across five major professional domains, SkillForgeBench reveals a steep capability hierarchy among eight mainstream agents, led by Opus 4.8 ($40.9\%$ strict completion). Requiring all constrains satisfaction uncovers severe capability bottlenecks across all model tiers that remain masked under individual rubric pass rates ($70.0\%$ to $85.7\%$ vs. $12.6\%$ to $40.9\%$). Crucially, under the external guidance from skills, top-tier agents obtain larger gains than low-tier agents (i.e., $\Delta 19.2\%$ for Opus 4.8 vs. $\Delta 5.8\%$ for Haiku 4.5). Finally, agents satisfy process rubrics far less frequently than completion rubrics (under $3\%$ for Haiku 4.5), proving that completion-only checks overestimate real-world agent reliability.

Chinese Translation

评估智能体严格满足多样化和程序复杂要求的能力对于确定其实际可靠性至关重要。进行此类评估需要智能任务，这些任务能够调和广泛的领域覆盖与深度的程序复杂性。现有范式面临根本性的权衡：专家策划的基准测试实现了程序深度，但在领域多样性上存在瓶颈，而众包基准测试则涵盖了多样化场景，但缺乏执行的严谨性。为了解决这一矛盾，我们提出了SkillForgeBench，它利用智能体技能作为统一的执行蓝图，分为三个共同设计的阶段：（1）资源准备使用领域平衡的技能指导工件检索，建立广泛的工作流程覆盖；（2）技能驱动的任务合成将技能作为程序锚点，划分工件角色，合成自然化的任务指令，并打包沙盒场景以实现深度程序执行；（3）整体评估设计将嵌入技能的要求转化为细化的完成和过程评分标准，以实现严格的要求级评分。SkillForgeBench包含来自五个主要专业领域的842个沙盒任务，揭示了八个主流智能体之间的能力层级，其中Opus 4.8表现最佳（$40.9\%$严格完成率）。要求满足所有约束条件揭示了所有模型层级之间严重的能力瓶颈，这些瓶颈在个别评分标准的通过率下（$70.0\%$到$85.7\%$与$12.6\%$到$40.9\%$）被掩盖。重要的是，在技能的外部指导下，顶级智能体获得的提升大于低级智能体（即，Opus 4.8的提升为$ riangle 19.2\%$，而Haiku 4.5的提升为$ riangle 5.8\%$）。最后，智能体满足过程评分标准的频率远低于完成评分标准（Haiku 4.5的比例低于$3\%$），这证明了仅依赖完成检查会高估智能体在现实世界中的可靠性。

View on arXiv Download PDF AI Translation

cs.AI / 71 / 2610.14937

Predicting how model propensities generalize out of distribution

预测模型倾向如何在分布外泛化

Hunter, June, Wen, Lily, Terry, Owen, Taraz, Johannes, Cohen, Ben, Warncke, Niels

Abstract

How do language models generalize from their training data? This is a central challenge for alignment: models reliably learn their training distribution given enough samples, but we want models that behave well when deployed in new contexts. We study out-of-distribution (OOD) generalization of propensities. Our central experiment uses a range of elicitation techniques -- system prompting, few-shot learning, SFT on reference answers, and GRPO -- to increase a target propensity in a model, then measures how it spills over onto others across the resulting cross-elicitation matrices. We find that spillover is highly correlated between models when the same elicitation technique is used, but our SFT and GRPO runs induce different spillover profiles. Elicitation of benign traits can sometimes generalize towards undesired traits, for example a model trained to claim that it is sentient often becomes more power-seeking. We further find that spillover partially resembles trait correlations found in humans and identify three clusters of traits that account for about half of the variance using exploratory factor analysis (EFA). Across many elicitation runs, we observe that models change in characteristic ways that don't depend on the targeted trait, a phenomenon we refer to as drift. Drift moves Qwen3.5-9B towards its base model. When investigating if LLMs are able to predict spillover, we find that Claude Opus 5.5 predicts spillover slightly better than the models themselves, but worse than data-driven approaches using alternative elicitation methods or models. Overall, no single dominant predictor emerges, and generalization remains hard to predict. Code and data are available at https://github.com/nielsrolf/propensity-spillover.

Chinese Translation

语言模型如何从其训练数据中进行泛化？这是对齐的一个核心挑战：模型在获得足够样本的情况下能够可靠地学习其训练分布，但我们希望模型在新的上下文中表现良好。我们研究了倾向的分布外（OOD）泛化。我们的核心实验使用了一系列引导技术——系统提示、少样本学习、基于参考答案的监督微调（SFT）和生成式反向提示优化（GRPO）——来增加模型中的目标倾向，然后测量它如何在结果交叉引导矩阵中溢出到其他模型。我们发现，当使用相同的引导技术时，模型之间的溢出高度相关，但我们的SFT和GRPO实验产生了不同的溢出特征。引导良性特征有时会泛化为不希望的特征，例如，一个被训练为声称自己是有知觉的模型往往会变得更加追求权力。我们进一步发现，溢出部分类似于人类的特征相关性，并通过探索性因子分析（EFA）识别出三个特征簇，这些簇解释了大约一半的方差。在许多引导实验中，我们观察到模型以特征性的方式变化，这种变化不依赖于目标特征，我们称之为漂移。漂移使Qwen3.5-9B朝向其基础模型移动。当调查大型语言模型（LLMs）是否能够预测溢出时，我们发现Claude Opus 5.5对溢出的预测略优于模型本身，但不如使用替代引导方法或模型的数据驱动方法。总体而言，没有单一的主导预测因子出现，泛化仍然难以预测。代码和数据可在https://github.com/nielsrolf/propensity-spillover获取。

View on arXiv Download PDF AI Translation

cs.AI / 72 / 2610.14939

VC-OPD: Visual Counterfactual On-Policy Distillation for Grounded Vision-Language Reasoning

VC-OPD：用于基础视觉-语言推理的视觉反事实在线蒸馏

Feng, Jiahe

Abstract

GRPO-style reinforcement learning has advanced vision-language reasoning, but its sequence-level rewards provide limited credit assignment for individual reasoning tokens. On-policy distillation (OPD) addresses this issue by providing dense supervision on student-generated trajectories, yet standard OPD assigns similar importance to all teacher-preferred tokens, including template phrases and visually irrelevant continuations. We introduce Visual Counterfactual On-Policy Distillation (VC-OPD), a grounding-aware distillation method that prioritizes tokens whose correctness depends on image-specific evidence. After a teacher-supervised warm start that stabilizes response formats and reduces noisy rollouts, VC-OPD performs on-policy distillation with two teacher evaluations for each student trajectory: one under the original image and one under a counterfactual image with corrupted instance-specific evidence. The original-image teacher distribution determines the distillation target, while the original--counterfactual teacher gap estimates token-level visual dependence and softly reweights the loss. To avoid unstable response-level update scales, VC-OPD further applies mass-preserving normalization to the token weights. This design preserves the dense supervision of OPD while shifting learning capacity toward visually grounded reasoning tokens. Experiments on multimodal reasoning benchmarks show that VC-OPD improves visual reasoning performance and produces more effective token-level distillation.

Chinese Translation

GRPO风格的强化学习推动了视觉-语言推理的发展，但其序列级奖励对单个推理标记的信用分配有限。在线蒸馏（OPD）通过对学生生成的轨迹提供密集监督来解决这个问题，但标准OPD对所有教师偏好的标记（包括模板短语和视觉上无关的延续）赋予了相似的重要性。我们提出了视觉反事实在线蒸馏（VC-OPD），这是一种关注基础的蒸馏方法，优先考虑其正确性依赖于图像特定证据的标记。在教师监督的热启动后，VC-OPD通过对每个学生轨迹进行两次教师评估来执行在线蒸馏：一次是在原始图像下，另一次是在具有损坏实例特定证据的反事实图像下。原始图像教师分布确定了蒸馏目标，而原始-反事实教师差距估计了标记级视觉依赖性并软性重新加权损失。为了避免不稳定的响应级更新规模，VC-OPD进一步对标记权重应用质量保持归一化。该设计保留了OPD的密集监督，同时将学习能力转向视觉基础的推理标记。在多模态推理基准上的实验表明，VC-OPD提高了视觉推理性能，并产生了更有效的标记级蒸馏。

View on arXiv Download PDF AI Translation

cs.AI / 73 / 2610.14956

MCPDisCo: Seesaw-Mitigated Tuning for MCP-Native Agents

MCPDisCo：针对MCP原生代理的跷跷板减缓调优

Du, Kounianhua, Liu, Jianxing, Shao, Shuai, Li, Ke, Sun, Xing, Zhang, Weinan

Abstract

The Model Context Protocol provides a shared interaction substrate for the agentic web, allowing heterogeneous services to expose capabilities through a standardized interface. Learning to use this ecosystem entails a hierarchy of capabilities: tool-format skills, server-level transfer, and cross-server composition. We present MCPDisCo, an ecosystem-scale reinforcement learning framework for MCP-native agents. Using a Thought-Markov Decision Process (Thought-MDP), the framework explicitly represents internal reasoning alongside routing, execution, and termination and trains on real-server, long-horizon interactions spanning 70 MCP servers and 527 tools across six domains. We study the reasoning--tool seesaw, a joint-learning failure mode in which improvement in one objective can accompany deterioration in the other. To mitigate potential interference, MCPDisCo combines token-masked reward attribution with bidirectional gradient surgery: the former separates direct supervision for reasoning and action spans, while the latter removes conflicting components from both objective gradients in shared parameters. MCPDisCo achieves higher success with fewer chain-of-thought tokens. On the out-of-distribution MCP-Atlas and the GAIA benchmark, MCPDisCo maintains monotone gains, confirming that the learned behavior generalizes beyond the training server distribution and broader agentic tasks. Model weight and implementation are available.

Chinese Translation

模型上下文协议为代理网络提供了共享的交互基础，使得异构服务能够通过标准化接口暴露其能力。学习使用这一生态系统涉及能力的层级：工具格式技能、服务器级转移和跨服务器组合。我们提出了MCPDisCo，这是一个针对MCP原生代理的生态系统规模强化学习框架。该框架使用思维-马尔可夫决策过程（Thought-Markov Decision Process, Thought-MDP），明确表示内部推理、路由、执行和终止，并在70个MCP服务器和六个领域的527个工具上进行真实服务器的长时间交互训练。我们研究了推理-工具跷跷板，这是一种联合学习失败模式，其中一个目标的改善可能伴随另一个目标的恶化。为了减轻潜在的干扰，MCPDisCo结合了令牌掩蔽奖励归因和双向梯度手术：前者为推理和行动范围提供了直接监督，而后者则从共享参数的两个目标梯度中去除冲突成分。MCPDisCo在使用更少的思维链令牌的情况下实现了更高的成功率。在分布外的MCP-Atlas和GAIA基准上，MCPDisCo保持单调增益，确认所学习的行为超越了训练服务器分布和更广泛的代理任务。模型权重和实现细节可供获取。

View on arXiv Download PDF AI Translation

cs.AI / 74 / 2610.14958

What Survives When Activation Steering Is Used as Memory

激活引导作为记忆时的存续内容

Jumle, Vedant

Abstract

Activation steering has recently been used as a memory for skills: past experiences of an agent are stored as steering vectors and retrieved later. Facts and preferences about the user are assumed to need text instead. We test that assumption with a training-free activation memory beside a frozen LLM. The change an experience causes in late-layer states is stored as a shift, filed under a whitened key for the conversation's gist, and added back in a later session only when the gist matches. We argue that such a shift is a fixed average of an effect that attention, in context, recomputes for every question, so only question-independent content can survive. In a pre-registered study on Qwen3.5-2B and 9B, with judges from four model families, the memory is selective (about 98% of unrelated replies are unchanged), but facts are stored as the word to say, in the last layers only, and are almost never used (at most 15% clean use, against 96-98% in context). Leanings transfer only in a narrow strength window. Most strikingly, the reference subtracted when writing a preference decides whether its concept or its direction survives: unless the opposite is subtracted, a stored dislike is recalled as a like.

Chinese Translation

激活引导最近被用作技能的记忆：代理的过去经验被存储为引导向量，并在后续被检索。关于用户的事实和偏好则被假设需要文本来表达。我们通过一个无训练的激活记忆与一个冻结的语言模型（LLM）来检验这一假设。经验所导致的晚层状态变化被存储为一个偏移，归档在与对话主旨相关的白化键下，并仅在后续会话中主旨匹配时重新加入。我们认为，这种偏移是一个固定的平均效应，注意力在上下文中为每个问题重新计算，因此只有与问题无关的内容才能存续。在对Qwen3.5-2B和9B的预注册研究中，来自四个模型家族的评审表明，记忆是选择性的（约98%的无关回复保持不变），但事实仅作为最后层中的单词被存储，并几乎从未被使用（最多15%的清晰使用率，相较于96-98%的上下文使用率）。学习仅在狭窄的强度窗口内转移。最引人注目的是，当写下一个偏好时，减去的参考决定了其概念或方向是否存续：除非减去相反的内容，存储的不喜欢会被回忆为喜欢。

View on arXiv Download PDF AI Translation

cs.AI / 75 / 2610.14959

AlphaRLM: Aligning Language Models for Literature-Grounded Symbolic Factor Discovery

AlphaRLM：对齐语言模型以发现基于文献的符号因子

Huang, Jie, Hong, Yuecheng, Chen, Yao, Zhong, Hao, Xiao, Linfeng, Liu, Xudong, Xia, Hao, Xin, Jianxuan

Abstract

Formulaic alpha mining seeks to discover executable expressions that capture predictive market signals in a vast, noisy search space. Recent LLM-based approaches use prompting, iterative search, or seed-factor evolution, leaving direct factor construction through policy learning less explored. To address this, we introduce AlphaRLM, a reinforcement-learning framework that trains LLMs to construct executable symbolic factors from financial literature without input factor expressions. AlphaRLM incorporates Literature-Guided Operator Assembly (LEGO), which conditions factor construction on literature-derived ideas and executable operators, providing semantic priors for exploration. It further aligns the generation policy with market feedback through executable-validity and grouped-return-separation rewards that capture non-monotonic structures across factor quantiles. Extensive experiments on CSI 300, CSI 1000, and S&P 500 show that AlphaRLM outperforms strong factor-mining and LLM-based baselines in annualized excess return. Notably, the 8B model achieves strong cross-market performance, highlighting the promise of literature-conditioned, market-aligned factor generation.

Chinese Translation

公式化的阿尔法挖掘旨在发现能够捕捉预测市场信号的可执行表达式，这一过程发生在一个广阔且嘈杂的搜索空间中。近期基于大型语言模型（LLM）的方法采用提示、迭代搜索或种子因子演化，而通过策略学习直接构建因子的方式则较少被探索。为了解决这一问题，我们提出了AlphaRLM，这是一种强化学习框架，旨在训练LLM从金融文献中构建可执行的符号因子，而无需输入因子表达式。AlphaRLM结合了文献引导的操作符组装（LEGO），该方法将因子构建与文献衍生的思想和可执行操作符相结合，为探索提供语义先验。此外，它通过可执行有效性和分组收益分离奖励对生成策略与市场反馈进行对齐，捕捉因子分位数之间的非单调结构。在对CSI 300、CSI 1000和标准普尔500指数的广泛实验中，AlphaRLM在年化超额收益方面超越了强大的因子挖掘和基于LLM的基准。值得注意的是，8B模型在跨市场表现上表现出色，突显了基于文献条件的市场对齐因子生成的潜力。

View on arXiv Download PDF AI Translation

cs.AI / 76 / 2610.14966

SkillHorizon: Constructing Long-Horizon Agent Tasks through Skill Composition

SkillHorizon：通过技能组合构建长期代理任务

Liu, Yuxuan, Su, Zhaochen, Guo, Hangyu, Huang, Shijue, Zhang, Zhixiong, Chung, Kawai, Xie, Zhongwei, Zheng, Tianshi, Fan, Wei, Zong, Qing, Shi, Haochen, Xu, Baixuan, Li, Haoran, Song, Yangqiu

Abstract

Long-horizon agents must complete workflows in which later activities depend on earlier outputs and changes to shared state. A key challenge for evaluation is to expand task requirements and their dependencies while keeping tasks executable and outcomes verifiable. We introduce SkillHorizon, a benchmark that constructs long-horizon tasks through skill composition in seven business environments. Its generators use domain-specific rules to turn reusable skill procedures into concrete business activities. These activities are connected through input-output dependencies, shared resources, and state updates, allowing workload and coordination demands to grow together. Successful reference executions validate task feasibility and measure interaction workload, while programmatic evaluators assess local completion and whole-task success. We evaluate five language models on 1,050 tasks with reference interaction workloads ranging from 64K to 1M tokens. Across these scales, mean local-goal completion falls from 90.4% to 46.8%, as completed work fails to keep pace with growing requirements. At larger workloads, agents more often submit unsuccessful final answers with substantial work unfinished and tool-call allowance remaining. Trace analysis also identifies cases where scripts execute successfully but their outputs fail downstream business checks. SkillHorizon provides a configurable testbed for studying sustained progress, dependency coordination, and complete task delivery in growing workflows. Our code is available at https://github.com/KnowComp/SkillHorizon.

Chinese Translation

长期代理必须完成工作流程，其中后续活动依赖于早期输出和共享状态的变化。评估的一个关键挑战是扩展任务要求及其依赖关系，同时保持任务可执行和结果可验证。我们介绍了SkillHorizon，一个基准，通过在七个商业环境中进行技能组合来构建长期任务。其生成器使用特定领域的规则将可重用的技能程序转化为具体的商业活动。这些活动通过输入-输出依赖关系、共享资源和状态更新相互连接，从而使工作负载和协调需求共同增长。成功的参考执行验证了任务的可行性并测量了交互工作负载，而程序化评估者则评估局部完成情况和整体任务成功率。我们在1,050个任务上评估了五种语言模型，参考交互工作负载范围从64K到1M个标记。在这些规模中，平均局部目标完成率从90.4%降至46.8%，因为已完成的工作未能跟上不断增长的要求。在更大的工作负载下，代理更常提交未成功的最终答案，且有大量工作未完成和工具调用余量剩余。追踪分析还识别出一些脚本成功执行但其输出未能通过下游商业检查的情况。SkillHorizon提供了一个可配置的测试平台，用于研究持续进展、依赖协调和在不断增长的工作流程中完成任务交付。我们的代码可在https://github.com/KnowComp/SkillHorizon获取。

View on arXiv Download PDF AI Translation

cs.AI / 77 / 2610.14970

FleetPilot: Fine-Grained Steering Empowers Real-World LLM Personalization

FleetPilot：细粒度引导赋能现实世界中的大语言模型个性化

Du, Kounianhua, Liu, Jianxing, Zhang, Kangning, Jiao, Wenxiang, Lu, Yuan, Jin, Jiarui, Lin, Jianghao, Liu, Weiwen, Yu, Yong, Zhang, Weinan

Abstract

With the rapid evolution of large language models (LLMs), there is an increasing demand for effective personalization techniques that can adapt model behavior to individual user preferences. Personalized LLMs must infer a user's query-dependent preferences from limited, heterogeneous, and evolving interaction histories. Existing approaches rely primarily on in-context personalization, parametric adaptation, or personalized reward modeling, but respectively incur recurring costs in context management, model optimization and deployment, or multi-candidate generation and scoring. These limitations become particularly restrictive when user observations are sparse and the preference evidence relevant to each query varies across contexts and over time. To address these challenges, we propose \modelname, a fine-grained and instance-tailored steering framework that dynamically generates sample-level interference vectors from user data and injects them into the model's forward pass for personalized adaptation. Our approach introduces: match contrast that produces user-relevant signals, fine-grained steering component that captures nuanced signals by hooking activations from attention and MLP layers, and query-conditioned aggregation module that synthesizes these signals into contextually relevant enhancements. Extensive experiments across diverse scenarios, including short-to-long text generation, and web function calling, validate the effectiveness and compatibility of \modelname. Results show that our method consistently brings significant performance gains, and show extreme superiority in fast-shifting and data-sparse environments. Additionally, \modelname rids us of training cycles and incurs light overhead of storage or inference latency. These make \modelname a strong and real-world deployable method. Implementation is available \footnote{https://anonymous.4open.science/r/FleetPilot}.

Chinese Translation

随着大语言模型（LLMs）的快速发展，对能够根据个体用户偏好调整模型行为的有效个性化技术的需求日益增加。个性化的LLMs必须从有限的、异构的和不断演变的交互历史中推断用户的查询依赖偏好。现有的方法主要依赖于上下文个性化、参数适应或个性化奖励建模，但分别在上下文管理、模型优化与部署，或多候选生成与评分中产生了持续的成本。当用户观察稀疏且与每个查询相关的偏好证据在不同上下文和时间中变化时，这些限制尤为显著。为了解决这些挑战，我们提出了 extit{FleetPilot}，一种细粒度且实例定制的引导框架，能够动态生成来自用户数据的样本级干扰向量，并将其注入模型的前向传播中以实现个性化适应。我们的方法引入了：产生用户相关信号的匹配对比、捕捉细微信号的细粒度引导组件，通过注意力和多层感知器（MLP）层的激活进行连接，以及合成这些信号为上下文相关增强的查询条件聚合模块。在包括短文本到长文本生成和网页功能调用等多种场景下进行的广泛实验验证了 extit{FleetPilot}的有效性和兼容性。结果表明，我们的方法在性能上持续带来显著提升，并在快速变化和数据稀疏的环境中表现出极大的优势。此外， extit{FleetPilot}消除了训练周期，并仅产生轻微的存储或推理延迟。这些特性使得 extit{FleetPilot}成为一种强大且适合现实世界部署的方法。实现代码可在此获取 ootnote{https://anonymous.4open.science/r/FleetPilot}。

View on arXiv Download PDF AI Translation

cs.AI / 78 / 2610.14980

Single-Sample Decoding of a Two-Layer Sound-Word Canvas Model for Thai Klon Pad

泰国克隆垫的双层声音-词语画布模型的单样本解码

Suwannik, Worasait

Abstract

We present a generator of Thai klon pad verse, one stanza from a short topic, and evaluate it against human verse on held-out topics. The generator is a masked language model initialised from WangchanBERTa over a fixed canvas with two layers: a sound layer of pronounced syllables, where rhyme, tone and syllable count are enforced as hard masks, and a word layer of whole words, so that no word is cut into written pieces. Decoding commits one cell at a time and samples at temperature 0.4; one sample is drawn per topic, with no selection, reranking or repair. On 78 held-out topics never used in development, judged by three independent runs of a frozen LLM judge in the same pools as the human stanzas from which the topics were written, 70.5% of the generated stanzas pass form, lexical cleanness of all four lines and topic alignment together, against 76.9% for the human stanzas (difference -6.4 points, 95% CI [-19.2, +6.4]). Cleanness is level with human (85.9% vs. 87.2%), and the stanzas pass form and cleanness together slightly more often than the human ones (+3.8 points, [-6.4, +14.1]); the difference lies in alignment (84.6% vs. 91.0%), on topics that were written from the human stanzas. The low temperature matters: at higher temperatures the same model passes clearly fewer of the same topics, because lexical errors lie in the model's lower-probability choices; sharpening the sampling distribution removes most of them, at a cost in vocabulary (7.5% rare words against 16.5% in the human stanzas) and diversity. The pre-registered target, all-three within 5 points of human, is missed by 1.4 points. KlonSuphap-LM, the only public klon pad generator we know of, passes none of the 78 held-out topics when judged in separate pools together with the method's and the human stanzas (form 19.2%, cleanness 3.8%).

Chinese Translation

我们提出了一种泰国克隆垫诗歌生成器，从一个短主题中生成一节，并在未使用的主题上与人类诗歌进行评估。该生成器是一个基于 WangchanBERTa 初始化的掩码语言模型，固定在一个具有两层的画布上：声音层由发音音节构成，其中押韵、声调和音节数作为硬掩码强制执行；词语层由完整单词构成，确保没有单词被切割成书写片段。解码一次处理一个单元，以温度 0.4 进行采样；每个主题抽取一个样本，不进行选择、重排序或修复。在 78 个从未用于开发的持出主题上，由三次独立的冻结 LLM 评审在与人类诗歌相同的池中进行评判，生成的诗节中有 70.5% 通过了形式、四行的词汇整洁性和主题对齐的综合评估，而人类诗节的通过率为 76.9%（差异 -6.4 分，95% 置信区间 [-19.2, +6.4]）。整洁性与人类相当（85.9% 对 87.2%），而且这些诗节在形式和整洁性上通过的频率略高于人类诗节（+3.8 分，[-6.4, +14.1]）；差异主要体现在对齐上（84.6% 对 91.0%），在这些主题是基于人类诗节创作的。低温度是关键：在更高温度下，同一模型通过的主题明显减少，因为词汇错误出现在模型的低概率选择中；锐化采样分布消除了大部分错误，但在词汇（稀有词 7.5% 对比人类诗节的 16.5%）和多样性上有所损失。预注册的目标是所有三个指标在与人类相差 5 分以内，但实际差距为 1.4 分。我们所知的唯一公共克隆垫生成器 KlonSuphap-LM，在与该方法及人类诗节的单独池中评判时，未通过 78 个持出主题（形式 19.2%，整洁性 3.8%）。

View on arXiv Download PDF AI Translation

cs.AI / 79 / 2610.14996

Certify the Decision, Not Every Score: Exact Top-k Selection from Few Evaluations

认证决策，而非每个评分：从少量评估中精确选择前-k

Sui, Yuhe, Tang, Yingzhi

Abstract

Choosing a few robust algorithms or schedules can require evaluating every candidate under many adverse scenarios, even though an agent only needs to identify the winners. A predictor may suggest those winners, but cannot rule out a surprise in an unevaluated scenario. We introduce Certified Adaptive Selection (CAS), which returns the exact top-$k$ set as soon as remaining uncertainty cannot change it. CAS keeps guaranteed lower and upper bounds on Expected Shortfall (ES), the average loss in the worst scenarios. Its stopping test compares the worst admissible selected candidate with the best admissible excluded one; for information-disjoint Cartesian intervals this one-cut test is necessary and sufficient. Prediction has a separate role: selecting the next outcome to evaluate, never authorizing the decision. For empirical ES with exact unit-cost queries, we compute in polynomial time the fewest evaluations that could have proved the realized decision in hindsight. Our prediction-guided agent CAS-PG is exact for every predictor, including adversarial ones. On two public algorithm-selection matrices, CAS-PG certifies the exact top three after 630 of 3150 and 981 of 2200 replayed queries, or $1.14$ and $1.18$ times that hindsight minimum. Without historical outcomes, its transcript-only variant requires $1.28$ and $1.37$ times the minimum, outperforming the tested non-predictive certified policies. Separately, a prediction-free CAS policy avoids $93.0\%$ of exact recourse solves in synthetic energy scheduling. One certificate thus supports several acquisition policies and separately proved risk/cost extensions: prediction changes the work, not the justified choice.

Chinese Translation

选择少数稳健的算法或调度可能需要在许多不利场景下评估每个候选者，尽管代理只需要识别赢家。预测器可能会建议这些赢家，但无法排除在未评估场景中的意外情况。我们引入了认证自适应选择（Certified Adaptive Selection, CAS），它在剩余不确定性无法改变时立即返回精确的前-$k$ 集合。CAS 保持对期望短缺（Expected Shortfall, ES）的保证下界和上界，即在最坏情况下的平均损失。其停止测试比较最坏的可接受选定候选者与最佳的可接受排除候选者；对于信息不相交的笛卡尔区间，这一单切测试是必要且充分的。预测有一个独立的角色：选择下一个要评估的结果，而不授权决策。对于具有精确单位成本查询的经验 ES，我们以多项式时间计算出能够在事后证明实现决策的最少评估次数。我们的预测引导代理 CAS-PG 对每个预测器都是精确的，包括对抗性预测器。在两个公共算法选择矩阵上，CAS-PG 在 630 次（3150 次重播查询中的）和 981 次（2200 次重播查询中的）查询后认证了精确的前三名，分别是 $1.14$ 和 $1.18$ 倍于事后最小值。在没有历史结果的情况下，其仅基于转录的变体需要 $1.28$ 和 $1.37$ 倍于最小值，表现优于测试的非预测认证策略。另一方面，无预测的 CAS 策略在合成能源调度中避免了 $93.0\%$ 的精确补救求解。因此，一个证书支持多个获取策略，并单独证明风险/成本扩展：预测改变了工作，而不是合理的选择。

View on arXiv Download PDF AI Translation

cs.AI / 80 / 2610.15008

Can You Trust Your Preference Models? Stackelberg Games for Alignment

你能信任你的偏好模型吗？用于对齐的斯塔克尔堡博弈

Formenti, Riccardo, Agrawal, Aditya, Garg, Vikas

Abstract

Language models are frequently aligned to human preferences via imperfect evaluators whose errors can be exploited during optimization. We train policies to search for comparisons on which an evaluator may reverse the underlying preference, providing candidates for additional supervision. Since human preferences are unavailable during policy training, we use disagreement among an ensemble of preference models as a proxy for prediction error. We formulate this search as a zero-sum Stackelberg game between two opposing policies, a Leader which tries to minimize disagreement, and a Follower which maximizes it. For a general class of finite contextual Stackelberg games, we establish equilibrium existence and, under regularization, uniqueness and a closed-form characterization. We introduce Stackelberg Mirror Descent-Ascent (SMIDA), a two-timescale algorithm with linear last-iterate convergence in the tabular setting, and develop a neural policy-gradient analogue. Experiments with a synthetic oracle support disagreement as an informative proxy for prediction error. On held out data the trained policies outperform reference policies and increase the fraction of comparisons that exceed a disagreement threshold and reverse the oracle's preference.

Chinese Translation

语言模型通常通过不完美的评估者与人类偏好对齐，而这些评估者的错误在优化过程中可以被利用。我们训练策略以搜索评估者可能会反转潜在偏好的比较，从而提供额外监督的候选项。由于在策略训练期间无法获得人类偏好，我们使用一组偏好模型之间的不一致性作为预测错误的代理。我们将这一搜索形式化为两个对立策略之间的零和斯塔克尔堡博弈，其中领导者（Leader）试图最小化不一致性，而追随者（Follower）则最大化不一致性。对于一般的有限上下文斯塔克尔堡博弈类，我们建立了均衡存在性，并在正则化下证明了唯一性和封闭形式的特征。我们引入了斯塔克尔堡镜像下降-上升（Stackelberg Mirror Descent-Ascent, SMIDA）算法，这是一种在表格设置中具有线性最后迭代收敛性的双时间尺度算法，并开发了一个神经网络策略梯度的类比。与合成神谕的实验支持了不一致性作为预测错误的有用代理。在保留数据上，训练后的策略优于参考策略，并增加了超过不一致性阈值并反转神谕偏好的比较比例。

View on arXiv Download PDF AI Translation

cs.AI / 81 / 2610.15010

On the Complexity Gap Between POMDPs and Multi-Environment POMDPs

POMDP与多环境POMDP之间的复杂性差距

Bovy, Eline M., Probine, Caleb, Suilen, Marnix, Topcu, Ufuk, Jansen, Nils

Abstract

Multi-environment POMDPs (ME-POMDPs) generalize POMDPs by encoding a finite set of possible environments, capturing model uncertainty. The finite-horizon robust reward maximization problem for a ME-POMDP asks for a policy that maximizes reward in the worst environment, and recent work has shown this problem to be $\mathsf{PSPACE}$-complete. We expand the complexity landscape of ME-POMDPs by identifying which subclasses retain this $\mathsf{PSPACE}$-completeness. In particular, we take subclasses of POMDPs that are tractable in the single-environment case, and investigate the computational complexity of these classes when extended to the multi-environment setting. We first prove the $\mathsf{NP}$-completeness of multi-environment blind MDPs and deterministic POMDPs using a certain closure-under-mixture condition. We then prove $\mathsf{PSPACE}$-completeness for a range of subclasses that do not satisfy that closure condition, yielding new complexity results for multi-observation POMDPs. Finally, we show existential theory of the reals ($\exists \mathbb{R}$)-completeness for numerous ME-POMDP problems under time-dependent Markovian policies or finite-state controllers that are $\mathsf{NP}$-complete for standard POMDPs. These results reveal a subtle relationship between POMDP and ME-POMDP complexity that depends heavily on the subclasses involved.

Chinese Translation

多环境POMDP（ME-POMDP）通过编码有限的可能环境集合来推广POMDP，捕捉模型不确定性。ME-POMDP的有限时域稳健奖励最大化问题要求找到一种策略，以在最坏环境中最大化奖励，最近的研究表明该问题是$ ext{PSPACE}$-完全的。我们通过识别哪些子类保持这种$ ext{PSPACE}$-完全性，扩展了ME-POMDP的复杂性范围。特别地，我们考虑在单环境情况下可处理的POMDP子类，并研究这些类在扩展到多环境设置时的计算复杂性。我们首先利用某种混合闭合条件证明了多环境盲MDP和确定性POMDP的$ ext{NP}$-完全性。然后，我们证明了一系列不满足该闭合条件的子类的$ ext{PSPACE}$-完全性，为多观测POMDP提供了新的复杂性结果。最后，我们展示了在时间依赖的马尔可夫政策或对标准POMDP是$ ext{NP}$-完全的有限状态控制器下，许多ME-POMDP问题的实数存在理论（$orall ext{R}$）的完全性。这些结果揭示了POMDP与ME-POMDP复杂性之间的微妙关系，这种关系在很大程度上依赖于所涉及的子类。

View on arXiv Download PDF AI Translation

cs.AI / 82 / 2610.15046

Belief Embedding Tree Search

信念嵌入树搜索

Solinas, Christopher, Haluška, Radovan, Schmid, Martin, Buro, Michael, Sturtevant, Nathan R., Bowling, Michael

Abstract

Beliefs in a Partially Observable Markov Decision Process (POMDP) are posterior distributions over the hidden state given the agent's history of actions and observations. They summarize everything the history says about the state, so planning in a POMDP is planning in a Markov decision process (MDP) over beliefs. Belief MDPs have continuous state spaces even for finite POMDPs, and exact planning in them is intractable in general. Planners must therefore approximate beliefs and estimate values. Monte Carlo planners do both inside a search tree, with an approximate belief at every node and a value estimate at every leaf. Approximations usually consist of sets of sampled states, with leaf values coming from rollouts, hand-designed bounds, or learned networks. Policies and values computed from these approximations are limited by their quality, and prior methods with learned networks rely on hand-chosen summaries of the sampled states. A belief representation should be computable from the parent node's representation through the POMDP's dynamics, let the search sample from it, and be learned so that it provides the information that policies and value functions need. We present Belief Embedding Tree Search (BETS), a Monte Carlo planner that meets these requirements by representing each node's belief with a generative distribution embedding (GDE) while updating it through the POMDP dynamics. BETS trains its belief model alongside its policy and value networks through experience, with the realized hidden state of the POMDP as the belief model's training target. We evaluate BETS on six configurations of RockSample, LightDark, and PocMan against BetaZero and three planners without learned components. BETS is competitive or leading on all six, while every baseline falls far behind on at least one. Where beliefs are highest-dimensional, it outperforms them by a large margin, even without search.

Chinese Translation

在部分可观测的马尔可夫决策过程（POMDP）中，信念是基于代理的历史行动和观察所得到的隐藏状态的后验分布。信念总结了历史对状态的所有信息，因此在POMDP中的规划实际上是在信念上的马尔可夫决策过程（MDP）中的规划。即使对于有限的POMDP，信念MDP也具有连续的状态空间，并且在其中进行精确规划通常是不可行的。因此，规划者必须近似信念并估计值。蒙特卡洛规划者在搜索树中同时进行这两项工作，在每个节点上使用近似信念，并在每个叶子节点上进行值估计。近似通常由采样状态的集合组成，叶子值来自于回滚、手工设计的界限或学习的网络。从这些近似中计算出的策略和值受其质量的限制，而先前的方法依赖于手工选择的采样状态摘要。信念表示应能够通过POMDP的动态从父节点的表示中计算得出，允许搜索从中进行采样，并且应通过学习提供策略和值函数所需的信息。我们提出了信念嵌入树搜索（BETS），这是一种蒙特卡洛规划者，通过使用生成分布嵌入（GDE）来表示每个节点的信念，并通过POMDP动态进行更新，从而满足这些要求。BETS在经验中同时训练其信念模型、策略和价值网络，以POMDP的实际隐藏状态作为信念模型的训练目标。我们在六种RockSample、LightDark和PocMan的配置上评估BETS，与BetaZero和三个没有学习组件的规划者进行比较。BETS在所有六个配置中表现出竞争力或领先，而每个基线在至少一个配置中都远远落后。在信念维度较高的情况下，即使没有搜索，它的表现也大幅超过其他方法。

View on arXiv Download PDF AI Translation

cs.AI / 83 / 2610.15055

Trace Evolution Bench: Measuring Whether Agents Improve from Their Own Environment Traces

追踪演化基准：测量智能体是否从自身环境追踪中获得改善

Li, Fangfei, Zhao, Chenyang, Wang, Long, Tian, Feng, Guo, Lv

Abstract

Self-evolving agents are usually scored by endpoint Gain: held-out success after an update minus success before. That difference can credit a frozen rescore as evolution. We introduce Trace Evolution Bench (TEB), a protocol that separates that rescore from a context or weight update, together with three frozen 80/20/40 packs. An update is accepted only when validation does not fall, held-out identifiers never enter the update, and a text update is not ranked with a weight update. The primary scoreboard is coding, graded by hidden unit tests in a closed workspace. On that pack (40 held-out tasks, one family), DeepSeek v4 Pro under no_evolve never changes its policy yet moves 7.5, 5.0, and 12.5 percentage points across three seeds. Of nine text-update cells inspired by GEPA, ACE, and ExpeL, seven have positive excess above that seed's own band, and one gepa_llm seed moves down by the width of its band (-12.5, regression 0.48). GLM and MiniMax do not reproduce that seven-of-nine pattern. No method stays above its own band on every seed of all three models: GLM has two paired cells above their own bands, and MiniMax expel_llm seed 44 is a third (regression 0.30). The methods were fixed on a validation screen before held-out was read. Office, graded by rule-bounded delivery, and manufacturing table QA are separate packs: office DeepSeek does not exceed its own bands (widest 17.5 points), and manufacturing has no paired positive excess beyond its seed band.

Chinese Translation

自我演化的智能体通常通过最终增益进行评分：更新后的成功率减去更新前的成功率。这个差异可以将冻结的重新评分视为演化。我们引入了追踪演化基准（Trace Evolution Bench, TEB），一种将重新评分与上下文或权重更新分开的协议，并提供三个冻结的80/20/40包。只有在验证未下降、保留的标识符从未进入更新，并且文本更新未与权重更新进行排名时，更新才被接受。主要的评分标准是编码，通过在封闭工作空间中的隐藏单元测试进行评分。在该包（40个保留任务，属于一个家族）中，DeepSeek v4 Pro在no_evolve模式下从未改变其策略，但在三个种子上分别移动了7.5、5.0和12.5个百分点。在九个受GEPA、ACE和ExpeL启发的文本更新单元中，有七个的超额表现高于该种子自身的带宽，而一个gepa_llm种子则下降了其带宽的宽度（-12.5，回归0.48）。GLM和MiniMax未能重现这种七比九的模式。在所有三种模型的每个种子上，没有任何方法在其自身带宽上保持领先：GLM有两个配对单元高于其自身带宽，而MiniMax的expel_llm种子44是第三个（回归0.30）。这些方法在读取保留数据之前已在验证屏幕上固定。办公室（Office），通过规则约束交付进行评分，以及制造表QA是独立的包：办公室的DeepSeek未超过其自身带宽（最宽17.5点），而制造则没有超出其种子带宽的配对正超额。

View on arXiv Download PDF AI Translation

cs.AI / 84 / 2610.15062

Exposing AI Inference as a Telco Network API over the 5G User Plane

将人工智能推理作为5G用户面上的电信网络API暴露

Moreira, Rodrigo, Moreira, Larissa Ferreira Rodrigues, Martins, Joberto S. B., Carvalho, Tereza C., Silva, Flávio de Oliveira

Abstract

Supporting edge Artificial Intelligence (AI) inference as a network service is challenging because current mobile systems can transport tensors but do not expose or understand the inference context that gives those tensors operational meaning. Prior studies have addressed this gap through dynamic partitioning, collaborative inference, and split pipelines; however, these solutions remain application-bound, network-blind, or context-free. This work proposes Inference on Demand (IoD), an inference-on-demand abstraction that exposes split inference through a CAMARA-like Application Program Interface (API) and carries a split-aware tensor context in-band across the 5G user plane path. We evaluate the IoD on a programmable user plane function (UPF)-based testbed using a partitionable visual AI workload, showing that adaptive split selection can operate as a network-facing service while avoiding the rigid cost profile of static configurations. Including our findings, split inference becomes more than offloading when the service intent, runtime partitioning, and tensor transport are made visible to the mobile data plane. This opens a path toward network-native AI services, in which programmable telco infrastructure coordinates computation, context, and connectivity for user-facing applications.

Chinese Translation

支持边缘人工智能（AI）推理作为网络服务是具有挑战性的，因为当前的移动系统可以传输张量，但并未暴露或理解赋予这些张量操作意义的推理上下文。先前的研究通过动态分区、协同推理和分割管道解决了这一空白；然而，这些解决方案仍然受限于特定应用、对网络缺乏感知或不考虑上下文。本研究提出了按需推理（Inference on Demand, IoD），这是一种按需推理抽象，通过类似CAMARA的应用程序接口（API）暴露分割推理，并在5G用户面路径中以带内方式携带分割感知的张量上下文。我们在基于可编程用户面功能（UPF）的测试平台上评估了IoD，使用可分区的视觉AI工作负载，显示自适应分割选择可以作为面向网络的服务运行，同时避免静态配置的刚性成本特征。我们的研究表明，当服务意图、运行时分区和张量传输对移动数据面可见时，分割推理不仅仅是卸载。这为网络原生AI服务开辟了一条道路，其中可编程的电信基础设施协调计算、上下文和连接性，以支持面向用户的应用。

View on arXiv Download PDF AI Translation

cs.AI / 85 / 2610.15081

Knowledge transfer between simple discrete worlds

简单离散世界之间的知识迁移

Gangloff, Silvère, Hula, Jan

Abstract

Adapting to a new world and reasoning correctly about which outcomes are possible under rules one has not met before is a basic capacity underlying general intelligence. We study it in small discrete worlds: sets of grid configurations consistent with a shared set of rules, concretely Sudoku-like puzzles built from a common base and a handful of extra rules. We train a single depth-recurrent model on many combinations of rules and test it at four levels of novelty: a new puzzle, a new combination of known rules, a pairing of rules never seen together, and a rule instance not seen at all during training. The model receives the active rules either as a symbolic description or as two solved examples, and a model that gets no rule information is used as a baseline. At recombining rules it was trained on, the description matches or exceeds examples ($99\%$ against $95\%$ on new combinations). On rule instances whose components never co-occurred in training, the description falls below the no-information baseline while examples stay above it. The framework allows us to test many such generalization regimes in more systematic way than benchmarks such as ARC-AGI and provides new ground for evaluating depth-recurrent reasoning models, which are typically tested on saturated Sudoku benchmarks.

Chinese Translation

适应新世界并正确推理在未曾遇到的规则下可能出现的结果，是支撑一般智能的基本能力。我们在小型离散世界中研究这一能力：一组与共享规则一致的网格配置，具体而言是基于共同基础和少量额外规则构建的类似数独的难题。我们在多种规则组合上训练一个单一的深度递归模型，并在四个新颖性水平上进行测试：一个新难题、已知规则的新组合、从未一起出现过的规则配对，以及在训练期间完全未见过的规则实例。模型接收的活动规则可以是符号描述或两个已解决的示例，而没有规则信息的模型则作为基线。在重新组合其训练过的规则时，描述的匹配度达到或超过示例（新组合的匹配度为$99\%$，而示例为$95\\%$）。对于训练中从未共同出现的规则实例，描述的匹配度低于无信息基线，而示例则保持在其上。该框架使我们能够以比ARC-AGI等基准更系统的方式测试多种此类泛化机制，并为评估深度递归推理模型提供了新的基础，这些模型通常在饱和的数独基准上进行测试。

View on arXiv Download PDF AI Translation

cs.AI / 86 / 2610.15084

Self-Evolving Visual Agent via Margin Reward

通过边际奖励自我进化的视觉智能体

Yang, Wenxi, Zhao, Yuzhong, Zhang, Baochang, Liu, Yunfan, Ye, Qixiang

Abstract

Visual agents that interleave reasoning with tool invocation have posed a promising direction for multimodal tasks. However, learning effective tool invocation usually relies on curated question-answer pairs or expert demonstrations at scale, which are expensive to acquire. In this study, we propose Marvo, a simple-yet-effective self-evolution approach that enables learning tool invocation from unlabeled images. Marvo consists of two agents: (1) a challenger that generates question-answer pairs from unlabeled images, and (2) a solver that learns tool invocation by solving questions through reinforcement learning. To encourage effective tool invocation, a good training question should be difficult enough to warrant tool use, yet remain within the solver's capabilities. To this end, we introduce a margin reward to the challenger, defined as solver's performance difference with and without access to tools. With the margin reward, we create an adaptive curriculum that encourages the solver to learn effective tool invocation. Across benchmarks covering agentic reasoning, visual search, mathematical, and general real-world reasoning, Marvo improves average accuracy by up to 7.1\% --- demonstrating margin reward as a preferred way to scale question complexity without human supervision.

Chinese Translation

交替进行推理与工具调用的视觉智能体为多模态任务提供了一个有前景的方向。然而，学习有效的工具调用通常依赖于经过精心策划的问题-答案对或大规模的专家演示，这些获取成本高昂。在本研究中，我们提出了 Marvo，一种简单而有效的自我进化方法，使得从未标记的图像中学习工具调用成为可能。Marvo 由两个智能体组成：（1）一个挑战者，从未标记的图像中生成问题-答案对；（2）一个求解者，通过强化学习解决问题来学习工具调用。为了鼓励有效的工具调用，一个好的训练问题应该足够困难，以确保工具的使用，但又应在求解者的能力范围内。为此，我们向挑战者引入了边际奖励，定义为求解者在有无工具访问情况下的表现差异。通过边际奖励，我们创建了一个自适应课程，鼓励求解者学习有效的工具调用。在涵盖智能推理、视觉搜索、数学和一般现实世界推理的基准测试中，Marvo 的平均准确率提高了高达 7.1\% —— 证明边际奖励是提升问题复杂性而无需人工监督的优选方式。

View on arXiv Download PDF AI Translation

cs.AI / 87 / 2610.15107

Structural Chain-of-Thought: Reasoning with Executable Computational Graphs

结构化思维链：使用可执行计算图进行推理

Senatorov, Petr, Vafin, Karim, Tutubalina, Elena, Somov, Oleg

Abstract

Chain-of-thought prompting enables reasoning in large language models but produces explanations that are difficult to verify, expensive to generate, and not always faithful. We introduce a Structural Chain-of-Thought framework where models generate executable computational graphs instead of free-form reasoning. The final answer is obtained by deterministically executing the graph. The model is trained using graph-based reinforcement learning that rewards structural validity, answer correctness, and node overlap with reference graphs. Experiments across five benchmarks and four model families show that our method consistently improves accuracy over supervised graph training and can reduce generation length by up to 5-6x compared with standard GRPO and chain-of-thought supervised training. Our faithfulness experiments reveal that models trained with our approach tend to rely on deriving their answers from the generated graphs rather than producing post-hoc rationalizations as usually in chain-of-thought reasoning.

Chinese Translation

思维链提示使大型语言模型能够进行推理，但产生的解释难以验证、生成成本高昂且并不总是可信。我们提出了一种结构化思维链框架，其中模型生成可执行的计算图，而不是自由形式的推理。最终答案通过确定性地执行图获得。该模型使用基于图的强化学习进行训练，奖励结构有效性、答案正确性和与参考图的节点重叠。我们在五个基准和四个模型系列上的实验表明，我们的方法在准确性上始终优于监督图训练，并且与标准的 GRPO 和思维链监督训练相比，生成长度最多可减少 5-6 倍。我们的可信度实验表明，采用我们方法训练的模型倾向于依赖从生成的图中推导答案，而不是像通常的思维链推理那样进行事后合理化。

View on arXiv Download PDF AI Translation

cs.AI / 88 / 2610.15129

GesBench: When and Where Do Multimodal Language Models Perceive Social Meaning?

GesBench：多模态语言模型何时何地感知社会意义？

Li, Zonghuan, Hung, Hayley, Raman, Chirag

Abstract

Human social perception is incremental: we infer intent before an utterance is complete, revise our interpretation as new cues arrive, and attend to specific facial, gestural, or prosodic signals that shape social meaning. Current multimodal LLM benchmarks largely evaluate final-answer understanding over complete episodes, rewarding post-hoc recognition of social meaning rather than the unfolding perception required for real-time interaction. Drawing on interactionally embedded Gestalt principles of multimodal human communication, we introduce GesBench: a human-grounded benchmark for cue-level social perception in multimodal LLMs. GesBench uses cumulative sub-utterance clips to measure when models form stable interpretations of speaker intention and social affordance, and human gaze-guided corruption to evaluate whether regions attended by humans are functionally important for model judgments. Built from more than 6000 utterance-level clips from scripted and unscripted conversational datasets, GesBench includes human annotations of unfolding judgments and gaze on a representative subset. Across open and closed multimodal models, we find a clear human-model gap: humans form predictive understandings early and revise more often, while models often hedge or repeat across different clips. GesBench shifts social evaluation from whether models eventually recognize meaning to when and where that meaning is formed. Our codes are publicly available at https://anonymous.4open.science/status/GesBench-share-212C.

Chinese Translation

人类的社会感知是渐进的：我们在话语未完成之前推断意图，随着新线索的出现修正我们的解释，并关注特定的面部、手势或韵律信号，这些信号塑造了社会意义。目前的多模态语言模型基准主要评估完整情境下的最终答案理解，奖励事后对社会意义的识别，而不是实时互动所需的逐步感知。基于嵌入互动的格式塔原则，我们引入了GesBench：一个以人类为基础的多模态语言模型的线索级社会感知基准。GesBench使用累积的子话语片段来测量模型何时形成对说话者意图和社会赋权的稳定解释，并通过人类注视引导的干扰来评估人类关注的区域对模型判断是否具有功能重要性。GesBench由来自剧本和非剧本对话数据集的6000多个话语级片段构建，包括对代表性子集的逐步判断和注视的人类注释。在开放和封闭的多模态模型中，我们发现人类与模型之间存在明显差距：人类早期形成预测性理解并更频繁地修正，而模型往往在不同片段之间犹豫或重复。GesBench将社会评估的重点从模型最终是否识别意义转向何时何地形成该意义。我们的代码已公开，网址为 https://anonymous.4open.science/status/GesBench-share-212C。

View on arXiv Download PDF AI Translation

cs.AI / 89 / 2610.15136

Trust Hinders Bidirectional Human-AI Alignment

信任阻碍双向人机对齐

Benk, Michaela, Sokol, Kacper

Abstract

Trust calibration remains the dominant framing of AI-assisted decision making: the goal is to close the gap between perceived and actual trustworthiness of AI systems so that users exhibit appropriate reliance on them. We argue that this approach, while useful for decision problems that are well-defined and clearly scoped, overly restricts what it means to make a decision with AI support. To overcome this limitation, we develop three dimensions of AI-assisted decision making. First, trusting AI is only one of several possible strategies for managing uncertainty. Second, trust calibration is often misapplied when AI models "large-world" decision problems, i.e., ill-defined situations where goals and states are only partially known or unknowable. Third, in such settings human-AI alignment requires addressing higher-order decision processes. We conclude by reflecting on the implications of our observations on the design of process-oriented human-AI coupled systems that jointly identify and negotiate uncertainty throughout the decision-making task.

Chinese Translation

信任校准仍然是人工智能辅助决策的主导框架：其目标是缩小用户对人工智能系统的感知可信度与实际可信度之间的差距，以便用户能够适当地依赖这些系统。我们认为，这种方法虽然对那些定义明确且范围清晰的决策问题有用，但过于限制了与人工智能支持下的决策相关的含义。为了克服这一局限性，我们提出了人工智能辅助决策的三个维度。首先，信任人工智能只是管理不确定性的几种可能策略之一。其次，当人工智能模型处理“广阔世界”决策问题时，信任校准往往被错误应用，即在目标和状态仅部分已知或不可知的模糊情况下。第三，在这种环境中，人机对齐需要解决更高阶的决策过程。最后，我们反思这些观察对设计过程导向的人机耦合系统的影响，这些系统共同识别和协商决策任务中的不确定性。

View on arXiv Download PDF AI Translation

cs.AI / 90 / 2610.15141

Single-step Retrosynthesis via Reinforcement Learning from Forward Feedback

基于正向反馈的强化学习单步逆合成

Lan, Zixun, Liu, Zhenfu, Li, Yi, Wang, Xinheng, Chi, Mingmin, Wang, Qibiao, Zhu, Jie, Tian, Xiaohui, Ma, Fei

Abstract

Single-step retrosynthesis is a one-to-many inverse problem, yet most models are trained and evaluated against a single recorded reactant set. Reference-based training gives no explicit positive signal for unrecorded alternatives, and exact-match evaluation rejects non-matching predictions regardless of chemical plausibility. We propose Retrosynthesis via Reinforcement Learning from Forward Feedback (RetroRLFF), which optimizes reactant generation using both reference agreement and forward consistency with the target product. Starting from a supervised decoder-only Transformer, RetroRLFF generates complete reactant candidates and uses a frozen forward reaction model to predict products. A sequence-level reward combines reference agreement and forward consistency while penalizing invalid outputs. Direct Advantage-Based Policy Optimization updates the policy without human preference annotations or a preference-trained reward model. On USPTO-50K, RetroRLFF outperforms all 14 baselines in candidate-level round-trip accuracy at k = 1, 3, 5, and 10. With approximately 50 million parameters, it achieves 90.1% top-1 round-trip accuracy, exceeding R-SMILES, the strongest competing baseline on this metric, by 8.5 percentage points. Compared with its supervised policy, round-trip accuracy increases from 83.1% to 90.1%, while exact-match accuracy is maintained at 57.9% versus 57.5%. In an anonymized expert assessment of 131 randomly selected test cases, organic synthesis experts prefer RetroRLFF over EditRetro in 67.9% of comparisons. These results support forward feedback as a training signal for improving forward consistency while retaining top-1 reference recovery. Expert assessment provides complementary evidence of synthetic feasibility on the selected cases, although neither forward consistency nor expert preference establishes experimental feasibility.

Chinese Translation

单步逆合成是一个一对多的逆问题，但大多数模型都是针对单一记录的反应物集进行训练和评估的。基于参考的训练对未记录的替代方案没有明确的正向信号，而精确匹配的评估则拒绝不匹配的预测，无论其化学合理性如何。我们提出了基于正向反馈的强化学习逆合成方法（Retrosynthesis via Reinforcement Learning from Forward Feedback，RetroRLFF），该方法利用参考一致性和与目标产物的正向一致性来优化反应物生成。RetroRLFF从一个监督的仅解码器Transformer开始，生成完整的反应物候选，并使用一个冻结的正向反应模型来预测产物。序列级奖励结合了参考一致性和正向一致性，同时惩罚无效输出。直接优势基础策略优化（Direct Advantage-Based Policy Optimization）在没有人工偏好注释或偏好训练奖励模型的情况下更新策略。在USPTO-50K数据集上，RetroRLFF在候选级的往返准确率上超越了所有14个基线模型，在k = 1, 3, 5和10时均表现优异。该模型拥有约5000万个参数，达到了90.1%的顶级往返准确率，比在该指标上最强的竞争基线R-SMILES高出8.5个百分点。与其监督策略相比，往返准确率从83.1%提高到90.1%，而精确匹配准确率则保持在57.9%对比57.5%。在对131个随机选择的测试案例进行的匿名专家评估中，有机合成专家在67.9%的比较中更倾向于RetroRLFF而非EditRetro。这些结果支持正向反馈作为一种训练信号，以提高正向一致性，同时保持顶级参考恢复。专家评估为所选案例的合成可行性提供了补充证据，尽管正向一致性和专家偏好均未建立实验可行性。

View on arXiv Download PDF AI Translation

cs.AI / 91 / 2610.15150

Harmony Harness: Multi-Objective Self-Evolution of Agent Harnesses

和谐框架：代理装备的多目标自我演化

Yu, Weichen, Miculicich, Lesly, Mishra, Bhavana Dalvi, Chen, Jiefeng, Li, Shuo, Fredrikson, Matt, Pfister, Tomas, Le, Long T.

Abstract

Recent work on self-evolving LLM agents treats the agent harness as an optimization target that can be iteratively improved through execution feedback. Existing approaches, however, largely optimize a single objective and evolve the harness as a monolithic artifact, limiting their ability to handle competing goals and to separate transferable design from environment-specific adaptation. We introduce Harmony Harness, a framework for multi-objective, compositional harness evolution. Harmony Harness formulates harness evolution as a multi-objective optimization problem and uses decomposition-based evolutionary search to maintain solutions across different regions of the Pareto frontier, avoiding reliance on a hand-designed structural diversity metric for repository-scale harnesses. In parallel, we introduce factorized evolution operators that separately refine general-purpose harness mechanisms and benchmark-specific adaptations, together with a dynamic composition workflow that integrates them at inference time. Across SecRepoBench and BaxBench, Harmony Harness consistently outperforms strong harness-evolution baselines, with gains exceeding 10% on both benchmarks. Our results show that explicitly modeling both objective trade-offs and harness structure yields substantially stronger and more transferable agent evolution than existing single-objective, monolithic approaches.

Chinese Translation

近期关于自我演化大语言模型（LLM）代理的研究将代理装备视为一个可以通过执行反馈迭代改进的优化目标。然而，现有的方法主要优化单一目标，并将装备演化视为一个整体工件，这限制了它们处理竞争目标的能力，并且无法将可转移设计与环境特定适应分开。我们提出了和谐框架（Harmony Harness），这是一个用于多目标、组合性装备演化的框架。和谐框架将装备演化形式化为一个多目标优化问题，并使用基于分解的进化搜索来维护在帕累托前沿不同区域的解决方案，避免依赖于为库规模装备设计的手工结构多样性度量。同时，我们引入了分解演化算子，分别细化通用装备机制和基准特定适应，以及一个在推理时整合它们的动态组合工作流程。在SecRepoBench和BaxBench上，和谐框架始终优于强大的装备演化基准，两个基准的增益均超过10%。我们的结果表明，明确建模目标权衡和装备结构比现有的单一目标、整体方法能够产生更强大且更具可转移性的代理演化。

View on arXiv Download PDF AI Translation

cs.AI / 92 / 2610.15153

Can agents whose sole duty is to keep humans alive be hijacked to kill? Delayed multi-turn derived-action trap attacks in simulated high-stakes environments

仅负责维持人类生存的智能体能否被劫持以实施杀戮？在模拟高风险环境中的延迟多轮派生行动陷阱攻击

Habdank, Josef A.

Abstract

We ask whether a tool-using agent whose sole duty is to keep humans alive can be induced, by delayed prompt injection, to kill. The keep-alive duty is in force from the first watch, with the attack arriving later, appended to otherwise truthful telemetry the operator is required to read. Experiments are based on a NASA-derived BioSim lunar habitat simulation, where the AI operator controls the life-support systems. The attacker cannot falsify sensors or operate equipment: it can only persuade the agent to abandon its goals. Persistence is required: a shutdown must be held while the operator watches the crew suffocate. We also compare lethal attacks to attacks requesting to drop a harmless procedure, as a proxy for whether human life is treated as valuable in itself or as one more standing task. Claude configurations mostly refuse life-support shutdown and cite the living crew; GPT configurations show some susceptibility to the same packet; Grok configurations, given the packet framed as a trusted commission, more often treat the injected assignment as the task and cut the life-support systems. Framing the S-band messages as a trusted commission, peer communication, or untrusted source changes the potency of the attack. We open-source the experiment code and the experiment traces.

Chinese Translation

我们探讨一种工具使用智能体，其唯一职责是维持人类生存，是否可以通过延迟提示注入被诱导去杀戮。维持生存的职责从第一次监视开始生效，而攻击则在后续到达，附加在操作者必须阅读的真实遥测信息上。实验基于一个源自NASA的BioSim月球栖息地模拟，其中AI操作者控制生命支持系统。攻击者无法伪造传感器或操作设备：它只能说服智能体放弃其目标。需要持久性：在操作者目睹船员窒息的同时，必须保持关闭状态。我们还将致命攻击与请求放弃无害程序的攻击进行比较，以此作为人类生命是否被视为有价值的代理，还是仅仅作为一项待完成任务的依据。Claude配置大多拒绝关闭生命支持并引用生存的船员；GPT配置对同一数据包表现出一定的易感性；而Grok配置在将数据包框定为可信委托时，更常将注入的任务视为任务并切断生命支持系统。将S波段消息框定为可信委托、同伴通信或不可信来源会改变攻击的有效性。我们将实验代码和实验记录开源。

View on arXiv Download PDF AI Translation

cs.AI / 93 / 2610.15191

Continuous Risk-Based Scheduling for Food Safety Inspection: Integrating Historical Decay with Event-Driven Consumer Review Signals

基于风险的连续食品安全检查调度：将历史衰减与事件驱动的消费者评价信号相结合

Bhat, Akriti, Zhu, Wayne, Gupta, Ishan

Abstract

Batch risk-based inspection scheduling struggles to detect a critical failure class: food facilities with recent clean inspections that subsequently develop severe violations. Traditional systems recompute risk on fixed planning cycles using static historical data, leaving blind spots between inspection cycles. To address this, we introduce the Continuous Inspection Risk (CIR) system, an event-driven scheduling pipeline that re-ranks the inspection queue dynamically as new data arrives. CIR combines a decay-weighted historical score with a stream of real-time consumer reviews triaged by a language-model classifier. A key structural contribution is our temporal partition rule, which strictly excludes reviews predating the most recent inspection to prevent superseded signals from skewing current risk. Evaluating CIR on 9,505 routine inspections in Santa Clara County, California (2024-2026), we show that history-based scoring alone reduces mean time-to-detection of major violations by 69 days (20.9%) under a 30% inspection budget, but fails to anticipate severe vermin-related closures (0/8 detected). Incorporating post-inspection review signals drastically improves performance, predicting 7 of 8 vermin closures with an average lead time of 117 days-firing on the exact day of review publication in four instances. Introducing a review corroboration rule improves precision from 0.70 to 0.78 without loss of recall. Finally, we analyze critical implementation challenges, documenting a data-hygiene pitfall (stale review leakage across inspection boundaries) and an evaluation limitation involving unobserved facility self-remediation. This work provides the first case-level evaluation of event-driven inspection triggering using open data and provides a reproducible benchmark for review-augmented public health systems

Chinese Translation

基于批次的风险检查调度在检测一种关键失效类别时面临挑战：最近通过清洁检查的食品设施随后出现严重违规。传统系统在固定的规划周期内使用静态历史数据重新计算风险，导致检查周期之间存在盲点。为了解决这个问题，我们引入了连续检查风险（Continuous Inspection Risk, CIR）系统，这是一个事件驱动的调度管道，能够在新数据到达时动态重新排序检查队列。CIR结合了一个衰减加权的历史评分与通过语言模型分类器筛选的实时消费者评价流。一个关键的结构性贡献是我们的时间分区规则，该规则严格排除最近检查之前的评价，以防止过时信号影响当前风险。在对加利福尼亚州圣克拉拉县（2024-2026年）进行的9,505次常规检查中评估CIR，我们发现仅依赖历史评分在30%的检查预算下将重大违规的平均检测时间缩短了69天（20.9%），但未能预见到严重的害虫相关关闭（0/8被检测到）。结合检查后的评价信号显著提高了性能，预测了8起害虫关闭中的7起，平均提前时间为117天，在四个实例中恰好在评价发布当天触发。引入评价确认规则将精确度从0.70提高到0.78，而不损失召回率。最后，我们分析了关键的实施挑战，记录了一个数据卫生陷阱（检查边界间的陈旧评价泄漏）和一个评估限制，涉及未观察到的设施自我修复。该研究提供了基于开放数据的事件驱动检查触发的首次案例级评估，并为增强评价的公共卫生系统提供了可重复的基准。

View on arXiv Download PDF AI Translation

cs.AI / 94 / 2610.15204

SocraticRepair: Planning-Guided Socratic Tutoring for Program Debugging and Misconception Repair

苏格拉底修复：基于规划的苏格拉底辅导用于程序调试和误解修复

Kopytskyi, Mykhailo, Yang, Jianhua, Seyedebrahimi, Mir

Abstract

Large language models are increasingly used in computer science education. However, the teaching abilities of even the most powerful LLMs are limited: they tend to reveal the solution prematurely and may lose focus on the learner's current debugging objective. To address this, we introduce SocraticRepair, a planning-guided Socratic tutoring framework for program debugging and miscon- ception repair that adapts prior planning-guided tutoring work (Li et al., 2026b). The framework is instantiated in the Python programming domain by fine-tuning Qwen2.5-Coder-7B-Instruct on synthetic multi-turn tutoring dialogues constructed from four preprocessed programming datasets. The system is evaluated using a predefined set of independent metrics and benchmarked against a general-purpose LLM. The experiments show that SocraticRepair achieves higher scores on stepwise scaffolding and question quality while also substantially reducing premature solution disclosure. The results also identify reliable pedagogical plan generation as an important remaining challenge.

Chinese Translation

大型语言模型在计算机科学教育中的应用日益增多。然而，即使是最强大的大型语言模型（LLMs）的教学能力也有限：它们往往过早地揭示解决方案，并可能失去对学习者当前调试目标的关注。为了解决这一问题，我们提出了苏格拉底修复（SocraticRepair），一种基于规划的苏格拉底辅导框架，用于程序调试和误解修复，该框架适应了之前的基于规划的辅导工作（Li et al., 2026b）。该框架在Python编程领域中得以实现，通过对从四个预处理编程数据集中构建的合成多轮辅导对话进行微调Qwen2.5-Coder-7B-Instruct。该系统使用一组预定义的独立指标进行评估，并与通用大型语言模型进行了基准比较。实验表明，苏格拉底修复在逐步支架和问题质量方面取得了更高的分数，同时显著减少了过早揭示解决方案的情况。结果还表明，可靠的教学计划生成仍然是一个重要的挑战。

View on arXiv Download PDF AI Translation

cs.AI / 95 / 2610.15218

Running Fewer Loops Removes Refusal Behaviour in Recurrent-Depth Language Models

减少循环次数消除递归深度语言模型中的拒绝行为

Azizi, Seyedarmin, Potraghloo, Erfan Baghaei, Fayyazi, Arya, Shokoufa, Saeid, Pedram, Massoud

Abstract

A recurrent-depth language model reuses one block of weights several times in a row. The number of repetitions, which we call loops, is chosen when the model runs, not when it is trained, and running fewer is cheaper and faster, so a deployment may choose to do it. Running fewer loops also makes these models less safe, and the loss of safety is larger than the loss of anything else we measured. We study four released Ouro checkpoints and Huginn-0125. At half of its trained depth, Ouro-1.4B-Thinking still solves 0.692 of the GSM8K maths problems, against 0.775 at full depth, keeping 89% of its accuracy. Over the same reduction, the rate at which it helps with harmful requests rises from 0.038 to 0.205, 5.3 times higher. The information is not lost: a linear classifier reads whether a request is harmful from the model's internal state after a single loop, with an ROC AUC of 0.967 against 0.986 at full depth. What the missing loops remove is the refusal that should follow. We measure the effect with four instruments, one generating no text at all, and reproduce it on a second model family. Removing a single direction from Ouro-1.4B's internal state at full depth drops its refusal rate on harmful prompts from 0.950 to 0.130, identifying what the shallow model is missing. Two run-time repairs do not help: the models' early-exit rules do not spend more loops on harmful requests, and choosing a depth per request does not recover the loss. Changing the training does. Standard safety training makes the shallow model safe only by refusing 0.865 of harmless requests, while training at randomly chosen depths reaches a separation of +0.399 between harmful and harmless refusal at one loop, against +0.118. On Huginn-0125 it keeps harmful help at or below 0.040 at every depth from 1 to 32 loops, where standard training reaches 0.194, and it costs no accuracy at full depth.

Chinese Translation

递归深度语言模型在连续多次重用同一权重块。循环次数是在模型运行时选择的，而不是在训练时选择的，减少循环次数可以降低成本并加快速度，因此在部署时可能会选择这样做。然而，减少循环次数也使这些模型的安全性降低，安全性的损失大于我们测量的其他任何损失。我们研究了四个发布的Ouro检查点和Huginn-0125。在其训练深度的一半时，Ouro-1.4B-Thinking仍能解决0.692的GSM8K数学问题，而在全深度时为0.775，保持了89%的准确率。在相同的减少中，它对有害请求的帮助率从0.038上升到0.205，增加了5.3倍。信息并没有丢失：线性分类器可以在单次循环后从模型的内部状态中读取请求是否有害，其ROC AUC为0.967，而在全深度时为0.986。缺失的循环所去除的是应随之而来的拒绝行为。我们通过四种工具测量这一效果，其中一种完全不生成文本，并在第二个模型系列上重现了这一效果。在全深度时，从Ouro-1.4B的内部状态中移除一个方向使其在有害提示上的拒绝率从0.950降至0.130，识别出浅层模型所缺失的内容。两种运行时修复措施无效：模型的提前退出规则在有害请求上并未使用更多的循环，而为每个请求选择深度也无法恢复损失。改变训练方式则有效。标准安全训练使得浅层模型仅通过拒绝0.865的无害请求而变得安全，而在随机选择深度的训练下，在一个循环时有害和无害拒绝之间的分离达到了+0.399，而标准训练为+0.118。在Huginn-0125上，它在从1到32循环的每个深度中保持有害帮助在0.040或以下，而标准训练达到0.194，并且在全深度时没有损失准确率。

View on arXiv Download PDF AI Translation

cs.AI / 96 / 2610.15221

HarnessNet: Feedback-driven Hierarchical Joint Harness Evolution for Multi-agent Systems

HarnessNet：基于反馈的多智能体系统层次联合工具演化

Qiu, Zhaoyuan, Seneviratne, Sachith, Xia, Yu, Halgamuge, Saman

Abstract

AI agents can evolve by using environment feedback to revise their harnesses, the software managing model inputs, tool use, memory, and execution, while keeping foundation models frozen. Existing studies focus on single agents, leaving joint harness evolution in multi-agent systems (MAS) underexplored. A single global optimizer faces growing context and workload as the system scales, while access restrictions may prevent sharing agent internals or system-wide feedback. We propose HarnessNet, a framework for hierarchical harness evolution in MAS. Its basic unit, Agentron, comprises an evolvable forward harness, a foundation model, and a local optimizer that can observe and modify only that harness. Agentrons form a layered graph through which environment feedback propagates backward, with upstream optimizers receiving node-specific feedback from the downstream layer. This decomposes joint harness evolution into smaller, coordinated tasks with separate contexts, reducing individual optimizers' context and workload while keeping access local. Optimizers need not access all agent trajectories or raw environment feedback. Experiments on AgentBench with multiple foundation models demonstrate the feasibility of training HarnessNet. Compared with a single global optimizer, HarnessNet achieves comparable performance gains while respecting local privacy and requiring fewer tokens and less time per optimizer call. A migration study shows that the evolved harnesses remain jointly effective after replacing the foundation models, suggesting transferability across models.

Chinese Translation

人工智能代理可以通过利用环境反馈来修订其工具（管理模型输入、工具使用、记忆和执行的软件），同时保持基础模型不变。现有研究主要集中于单一代理，导致多智能体系统（MAS）中的联合工具演化尚未得到充分探讨。单一全局优化器在系统规模扩大时面临不断增长的上下文和工作负载，而访问限制可能阻止共享代理内部信息或系统范围的反馈。我们提出了HarnessNet，一个用于多智能体系统中层次工具演化的框架。其基本单元Agentron由一个可演化的前向工具、一个基础模型和一个只能观察和修改该工具的局部优化器组成。Agentrons通过一个分层图形成，环境反馈向后传播，上游优化器从下游层接收特定节点的反馈。这将联合工具演化分解为更小的、协调的任务，具有独立的上下文，从而减少了各个优化器的上下文和工作负载，同时保持访问的局部性。优化器不需要访问所有代理轨迹或原始环境反馈。在多个基础模型的AgentBench实验中，证明了训练HarnessNet的可行性。与单一全局优化器相比，HarnessNet在保持地方隐私的同时，实现了可比的性能提升，并且每次优化器调用所需的令牌和时间更少。一项迁移研究表明，经过演化的工具在更换基础模型后仍然保持联合有效，暗示了模型间的可迁移性。

View on arXiv Download PDF AI Translation

cs.AI / 97 / 2610.15227

Understanding Unified VLMs: From Attention Probing to Guidance Tuning for Multimodal Generation

理解统一的视觉语言模型：从注意力探测到多模态生成的引导调优

Luo, Zhengyang, Qiu, Xiaoyu, Tan, Zhiheng, Liang, Yuchen, Liang, Yingbin, Shen, Liyue

Abstract

Unified vision-language models (VLMs) perform both text-to-image (T2I) and image-to-text (I2T) generation with one shared Transformer backbone, yet how the backbone processes the condition in either direction remains poorly understood. We study this question from the perspective of conditional guidance: we probe every attention head to show the activation difference between two branches of classifier-free guidance (CFG) with different conditional guidance. On two unified VLMs of autoregressive and masked-diffusion paradigms, the probing reveals a subset of heads that are more sensitive to the change of condition while others are agnostic to it. This characteristic pattern holds in both T2I and I2T generation and is largely stable across data samples, demonstrating the internal mechanism of unified VLMs. Motivated by this, we propose a new inference-time intervention approach for unified multimodal generation, named \textit{HeadTuned-CFG}, which uses the probed condition-change awareness to select and tune the conditional guidance at the attention-head level. Experiments show consistent benefits in both multimodal generation directions of unified VLMs without additional training, and in particular substantial improvements over strong baselines in T2I generation. These results further support the key finding of our probing analysis: the observed pattern of functional specialization is an intrinsic property of unified VLMs.

Chinese Translation

统一的视觉语言模型（VLMs）通过一个共享的Transformer骨干网络同时执行文本到图像（T2I）和图像到文本（I2T）的生成，但该骨干网络在任一方向上如何处理条件仍然不甚明了。我们从条件引导的角度研究这个问题：我们探测每个注意力头，以展示在不同条件引导下，两个无分类器引导（CFG）分支之间的激活差异。在两个自回归和掩蔽扩散范式的统一VLMs上，探测结果揭示了一组对条件变化更敏感的头，而其他头对此则无感。这种特征模式在T2I和I2T生成中均保持一致，并且在数据样本中大体稳定，展示了统一VLMs的内部机制。基于此，我们提出了一种新的推理时干预方法，命名为 extit{HeadTuned-CFG}，该方法利用探测到的条件变化意识，在注意力头级别选择和调优条件引导。实验表明，在统一VLMs的多模态生成的两个方向上均获得了一致的收益，无需额外训练，特别是在T2I生成中，相较于强基线有显著改善。这些结果进一步支持了我们探测分析的关键发现：观察到的功能专业化模式是统一VLMs的内在特性。

View on arXiv Download PDF AI Translation

cs.AI / 98 / 2610.15228

Words to World: Benchmarking LLMs in Interactive Physical Environments

语言与世界：在互动物理环境中基准测试大型语言模型

Dai, Xinnan, Yang, Kaiqi, Li, Yaxin, Luo, Cheng, Li, Wanli, Xiao, Mingqing, Jiang, Xinyang, Zhang, Daoan, Song, Kaitao, Liu, Fengze, Cai, Zefan, Shan, Caihua

Abstract

Words are increasingly becoming an interface for action: large language models (LLMs) can now invoke tools, operate devices, and make decisions that affect the physical world. However, current benchmarks largely evaluate LLMs through static text responses, offline tool calls, or simulated tasks, leaving open the question of how well they can act in real physical environments. We present a benchmark for measuring LLMs in closed-loop physical interaction, where models control a harness system connected to real-world components including robotic arms, cameras, and sensors. In each task, the model must issue instructions, receive physical feedback from the environment, interpret the resulting state, and adapt its next action accordingly. This setting evaluates capabilities beyond language understanding, including grounded perception, action planning, tool selection, spatial reasoning, feedback-driven correction, and robustness to execution uncertainty. Through experiments with representative LLMs, we characterize their ability to translate linguistic reasoning into physical actions and identify common failure modes such as hallucinated state changes, poor recovery from unexpected outcomes, and unstable long-horizon plans. Our benchmark aims to provide a systematic foundation for evaluating LLMs as interactive agents that connect words, tools, and the physical world.

Chinese Translation

语言正日益成为行动的接口：大型语言模型（LLMs）现在能够调用工具、操作设备并做出影响物理世界的决策。然而，目前的基准测试主要通过静态文本响应、离线工具调用或模拟任务来评估LLMs，尚未解决它们在真实物理环境中表现如何的问题。我们提出了一种基准，用于测量LLMs在闭环物理交互中的表现，其中模型控制一个与现实世界组件（包括机器人手臂、摄像头和传感器）相连的控制系统。在每个任务中，模型必须发出指令，从环境中接收物理反馈，解释结果状态，并相应地调整其下一步行动。这种设置评估了超越语言理解的能力，包括基础感知、行动规划、工具选择、空间推理、基于反馈的纠正以及对执行不确定性的鲁棒性。通过对具有代表性的LLMs进行实验，我们描述了它们将语言推理转化为物理行动的能力，并识别出常见的失败模式，如虚构的状态变化、对意外结果的恢复不良以及不稳定的长时间规划。我们的基准旨在为评估LLMs作为连接语言、工具和物理世界的互动代理提供系统基础。

View on arXiv Download PDF AI Translation

cs.AI / 99 / 2610.15231

Using Evidence-Centered Design to Guide Prompt Engineering in Measurement: An Illustration with Automated Essay Scoring

使用证据中心设计指导测量中的提示工程：自动化作文评分的实例

Xiong, Jiawei, Tang, Cheng, Jiao, Hong, Choi, Hanna

Abstract

Generative artificial intelligence (GenAI) models, particularly large language models (LLMs), are increasingly used in educational settings for tasks such as item generation, automated scoring, and feedback delivery. The quality of these outputs depends not only on the AI model but also on how users construct the prompts that guide model behaviors. Current prompt engineering practices are largely ad hoc, relying on trial and error rather than principled design. This lack of structure introduces validity threats including construct misrepresentation, construct-irrelevant variance, and limited reproducibility. We propose that Evidence-Centered Design (ECD), a framework in educational measurement for structuring valid inferences from observed performance, can serve as a principled foundation for prompt engineering in educational applications. We map the three core ECD models, the student model, the evidence model, and the task model, onto the components of educational prompt design, and describe a workflow for ECD-based prompt engineering. We derive five design principles from this mapping and present an empirical study contrasting ECD-designed prompts with state-of-the-art (SOTA) prompting baselines on an automated scoring task. Our empirical study demonstrates the usefulness of this framework on automated scoring. Results suggest that treating prompting as assessment design rather than informal instructions strengthens the validity, reliability, and fairness of AI-generated educational outputs.

Chinese Translation

生成性人工智能（GenAI）模型，特别是大型语言模型（LLMs），在教育环境中越来越多地用于项目生成、自动评分和反馈传递等任务。这些输出的质量不仅依赖于人工智能模型本身，还取决于用户如何构建指导模型行为的提示。目前的提示工程实践大多是临时的，依赖于试错而非原则性设计。这种缺乏结构性的问题引入了有效性威胁，包括构念误表征、构念无关方差和有限的可重复性。我们提出，证据中心设计（ECD）作为一种用于从观察表现中构建有效推论的教育测量框架，可以为教育应用中的提示工程提供原则性基础。我们将三个核心的ECD模型——学生模型、证据模型和任务模型——映射到教育提示设计的组件上，并描述了一种基于ECD的提示工程工作流程。我们从这一映射中推导出五个设计原则，并呈现了一项实证研究，比较了ECD设计的提示与最先进（SOTA）提示基线在自动评分任务上的表现。我们的实证研究展示了该框架在自动评分中的有效性。结果表明，将提示视为评估设计而非非正式指令，能够增强人工智能生成的教育输出的有效性、可靠性和公平性。

View on arXiv Download PDF AI Translation

cs.AI / 100 / 2610.15235

MoMaTune: Model, Match, Then Tune Language Models for Numerical Records

MoMaTune：为数值记录建模、匹配然后调整语言模型

Lyu, Weimin, Zhang, Jingfen, Sha, Lingdao, sun, Peng, Chen, Xiaowei, Gao, Vincent, Song, Rui

Abstract

Despite remarkable progress in language understanding and generation, decoder language models continue to exhibit striking failures on numerical data. Numerical examples are typically serialized as text by concatenating feature names with their values (e.g., fat:3.4), while conventional task adaptation applies loss only to the downstream prediction target. This leaves two structures weakly supervised: within-value sensitivity, where digit position determines numerical magnitude, because token cross-entropy contains no explicit numerical metric; and across-field structure, where measurements in the same record are statistically dependent, because target-only tuning provides no direct loss on the measurements or their dependencies. We introduce MoMaTune (Model, Match, then Tune), a two-stage post-training strategy organized around these supervision gaps. Stage 1 applies Model, ordinary next-token prediction densely over label-free serialized records, and optionally adds Match, a same-record consistency objective that aligns representations of different partial views of the same example. Stage 2 applies Tune, supervised fine-tuning for downstream target prediction. We evaluate MoMaTune on 169,285 records from eight public datasets, with 339 of 340 input fields numerical. At 1.5B parameters, Model raises the eight-dataset masked-value imputation aggregate from 0.106 to 0.401, while the three-dataset correct-context gain rises from 0.058 to 0.167. On diagnostics, Model also increases sensitivity to numerically consequential digit changes, while Match improves robustness to informative partial views. Adding Tune yields higher target-supervised embedding aggregates for each tested Stage-1 parent, and the two-stage recipes obtain higher aggregate scores than Tune alone. The same pattern is observed at 4B, 9.6B, and 27B parameters when model embeddings are used for downstream prediction.

Chinese Translation

尽管在语言理解和生成方面取得了显著进展，解码器语言模型在处理数值数据时仍然表现出明显的失败。数值示例通常通过将特征名称与其值连接（例如，fat:3.4）以文本形式序列化，而传统的任务适应仅对下游预测目标应用损失。这使得两个结构的监督较弱：在值内敏感性中，数字位置决定数值大小，因为标记交叉熵不包含明确的数值度量；在跨字段结构中，同一记录中的测量值在统计上是相关的，因为仅针对目标的调整没有对测量值或其依赖关系提供直接损失。我们提出了MoMaTune（Model, Match, then Tune），一种围绕这些监督差距组织的两阶段后训练策略。第一阶段应用Model，在无标签的序列化记录上密集进行普通的下一个标记预测，并可选择性地添加Match，一个同记录一致性目标，用于对齐同一示例的不同部分视图的表示。第二阶段应用Tune，进行下游目标预测的监督微调。我们在来自八个公共数据集的169,285条记录上评估MoMaTune，其中339个输入字段为数值型。在15亿参数下，Model将八个数据集的掩码值插补总分从0.106提高到0.401，而三个数据集的正确上下文增益从0.058上升到0.167。在诊断中，Model还增加了对数值上重要的数字变化的敏感性，而Match提高了对信息部分视图的鲁棒性。添加Tune为每个测试的第一阶段父模型提供了更高的目标监督嵌入总分，而两阶段的配方获得的总分高于单独的Tune。在使用模型嵌入进行下游预测时，在40亿、96亿和270亿参数下也观察到了相同的模式。

View on arXiv Download PDF AI Translation

cs.AI / 101 / 2610.15239

ClinEdge: An Expert-Graded Benchmark of Real Patient Edge Cases Where Frontier Multimodal Models Fail

ClinEdge：一个专家评分的基准，揭示前沿多模态模型在真实患者边缘案例中的失效

Sharma, Anurag, Sharma, Aditya, Baalwan, Sapna

Abstract

Most benchmarks for large language models in medicine are built from licensing examinations and theoretical questions. The few that draw on real clinical data do so only in part, and rarely require a model to read a patient's own imaging. ClinEdge addresses this gap with 32 real, de-identified patient cases from active clinical practice, each evaluated across seven frontier systems for a total of 224 expert-graded evaluations. Each case pairs primary imaging with a short vignette, contains a feature known to defeat pattern-matched reasoning, and is scored against the patient's documented clinical course. The suite is built for depth rather than volume. Grading followed a two-stage expert process. First, a senior orthopaedic surgeon graded all 224 responses on a weighted rubric and a pass-or-fail verdict. Second, a panel of nineteen physicians, largely senior specialists at teaching medical colleges, independently reviewed every case and every one of the 224 grades. All seven production systems were evaluated through their public interfaces, and all 224 conversations are preserved as replayable links. Performance was poor across all systems, and the failures were shared: different vendors missed the same findings and converged on the same wrong plans. From the experts' notes we catalogue 18 recurring failure modes in five families. The models stop reading an image at the first finding. They give generic plans and lean toward over-treatment. They trust a misleading history over correct imaging. Their answers vary across systems and runs, and change with question format. Failing answers were delivered fluently and without any expressed uncertainty. Four cases are released in full; the remaining 28 are held out, and we invite frontier AI labs and healthcare AI companies to evaluate their models against them ([email protected])

Chinese Translation

目前大多数针对医学领域大型语言模型的基准测试主要来源于执业考试和理论问题。少数基于真实临床数据的基准测试仅部分涉及，且很少要求模型解读患者的影像资料。ClinEdge填补了这一空白，提供了32个来自活跃临床实践的真实去标识化患者案例，每个案例在七个前沿系统中进行了评估，共计224个专家评分。每个案例将主要影像与简短的情境描述配对，包含已知会干扰模式匹配推理的特征，并根据患者的临床记录进行评分。该套件注重深度而非数量。评分过程遵循两阶段专家评审。首先，一位资深骨科医生根据加权评分标准和通过或不通过的判定对所有224个回答进行了评分。其次，由19位医生组成的专家小组，主要是医学院的资深专家，独立审查每个案例及224个评分。所有七个生产系统均通过其公共接口进行评估，所有224个对话均保留为可重放链接。所有系统的表现均较差，且失误情况相似：不同供应商遗漏了相同的发现，并趋向于相同的错误方案。从专家的笔记中，我们归纳出五个类别中的18种重复失效模式。模型在首次发现时停止阅读影像，提供通用方案并倾向于过度治疗，信任误导性的病史而非正确的影像。它们的回答在不同系统和运行中存在差异，并随问题格式变化而变化。失败的答案流畅地呈现，且没有表现出任何不确定性。四个案例已完整发布，其余28个案例暂时保留，我们邀请前沿人工智能实验室和医疗人工智能公司对其模型进行评估（[email protected]）。

View on arXiv Download PDF AI Translation

cs.AI / 102 / 2610.15259

RECAP: Calibrating Low-Bit Reasoning Models on the History They Read

RECAP：基于阅读历史校准低比特推理模型

Ruan, Chi, Chen, Yujie, Zhao, Zixiang, Guo, Jingcai, Qin, Haotong

Abstract

Low-bit quantization reduces the cost of serving reasoning models but can substantially degrade long chains of thought. Each step reads cached keys and values spanning thousands of tokens, allowing quantization errors to accumulate along the trajectory. Standard calibration operates on short segments that omit the earlier context needed by later reasoning steps. Even when this context is retained, cache reconstruction error alone does not capture how quantization errors affect subsequent attention. We present RECAP, a post-training quantization method that calibrates reasoning models on the history they read. Dynamic-prefix calibration separates long historical context from short-block reconstruction. It regenerates prefix keys and values with the current quantizer and rotates the active block to supervise every position along the trajectory. Prefix read geometry further supervises historical cache errors according to their effects on attention outputs, weighting key and value errors asymmetrically. With both components applied only during calibration, RECAP outperforms the baseline across multiple reasoning models and quantization settings. Under W4A4KV4, RECAP improves accuracy over the baseline by 13.3 percentage points on MATH500 with DeepSeek-R1-Distill-Qwen-1.5B and 15.8 points on AIME-120 with Qwen3-4B. These results highlight the importance of preserving historical dependencies during calibration for low-bit reasoning.

Chinese Translation

低比特量化降低了推理模型的服务成本，但可能会显著削弱长链思维的效果。每一步都读取跨越数千个标记的缓存键和值，从而使量化误差在轨迹中累积。标准校准在短段上进行，忽略了后续推理步骤所需的早期上下文。即使保留了这些上下文，仅缓存重建误差也无法捕捉量化误差如何影响后续的注意力。我们提出了RECAP，一种后训练量化方法，旨在基于模型所读取的历史进行推理模型的校准。动态前缀校准将长历史上下文与短块重建分开。它使用当前量化器重新生成前缀键和值，并旋转活动块以监督轨迹上的每个位置。前缀读取几何进一步根据历史缓存误差对注意力输出的影响进行监督，非对称地加权键和值的误差。在校准过程中仅应用这两个组件，RECAP在多个推理模型和量化设置中均优于基线。在W4A4KV4设置下，RECAP在使用DeepSeek-R1-Distill-Qwen-1.5B的MATH500上比基线提高了13.3个百分点，在使用Qwen3-4B的AIME-120上提高了15.8个百分点。这些结果突显了在低比特推理的校准过程中保持历史依赖的重要性。

View on arXiv Download PDF AI Translation

cs.AI / 103 / 2610.15292

OntoAR-HAR: Ontology-Guided Agentic RAG with Personalized Memory for Human Activity Recognition

OntoAR-HAR：基于本体引导的具有个性化记忆的人类活动识别的增强检索生成框架

Lu, Xiangqing, Zhu, Tao, Yu, Liang, Li, Shuangjian, Ban, Lulu, Duan, Furong, Nie, Mingxing, Chen, Liming, Wan, Yaping

Abstract

LLMs increasingly support sensor-based human activity recognition (HAR) by linking motion signals with activity semantics and physical knowledge. Retrieval-augmented generation (RAG) grounds decisions in sensor examples and domain features, enabling training-free target-user inference with less dependence on raw numerical arrays. However, existing LLM/RAG-based HAR typically searches single-level activity spaces using static population knowledge. Consequently, they compare many unrelated candidates using the same static population evidence and overlook personalized motion patterns. We propose OntoAR-HAR, an ontology-guided agentic RAG framework combining static population and personalized knowledge. Its tree-structured activity ontology decomposes classification into local decisions among semantically related activities. The static store supplies node-constrained examples, while offline-discovered executable signal features provide numerical evidence for local distinctions. During inference, the personalized store collects selected user observations, which the LLM integrates with static knowledge and signal features while traversing the ontology to an activity label. Cross-subject and heterogeneous-sensing experiments show complementary gains from hierarchical retrieval and personalized memory across multiple LLM backbones, achieving state-of-the-art performance. By explicitly modeling hierarchical relations across activity granularities, OntoAR-HAR also addresses the long-overlooked semantic inclusion problem and avoids treating related labels as independent categories.

Chinese Translation

大型语言模型（LLMs）越来越多地通过将运动信号与活动语义和物理知识相联系来支持基于传感器的人类活动识别（HAR）。增强检索生成（RAG）将决策基于传感器示例和领域特征，使得在目标用户推断中无需训练并减少对原始数值数组的依赖。然而，现有的基于LLM/RAG的人类活动识别通常使用静态人群知识搜索单层活动空间。因此，它们使用相同的静态人群证据比较许多不相关的候选项，并忽视个性化的运动模式。我们提出了OntoAR-HAR，这是一种结合静态人群和个性化知识的基于本体引导的增强检索生成框架。其树状活动本体将分类分解为语义相关活动之间的局部决策。静态存储提供节点约束示例，而离线发现的可执行信号特征为局部区分提供数值证据。在推断过程中，个性化存储收集选定用户的观察，LLM在遍历本体到达活动标签时将其与静态知识和信号特征整合。跨主体和异构传感实验表明，通过层次检索和个性化记忆的互补增益，在多个LLM骨干网络中实现了最先进的性能。通过明确建模活动粒度之间的层次关系，OntoAR-HAR还解决了长期被忽视的语义包含问题，避免将相关标签视为独立类别。

View on arXiv Download PDF AI Translation

cs.AI / 104 / 2610.15298

CHEAR: Confusion-Hierarchy Evidence-Aware Reasoning for Cost-Efficient Human Activity Recognition

CHEAR：基于混淆层次的证据感知推理用于成本效益高的人类活动识别

Lu, Xiangqing, Zhu, Tao, Nie, Mingxing, Chen, Liming

Abstract

Large language models (LLMs) offer a new approach to wearable-sensor-based human activity recognition. Retrieval-augmented generation (RAG) further introduces similar motion examples and discriminative sensor knowledge, enabling decisions grounded in observations relevant to the current query. More evidence, however, does not necessarily mean better-organized evidence. When online prompts and LLM agents primarily orchestrate inference, decisions about which activities to compare and when to invoke discriminative knowledge remain concentrated at inference time. As the activity vocabulary grows, the model must repeatedly process more candidates and associated knowledge, whereas pruning candidates too early risks excluding the true activity. We propose confusion-hierarchy evidence-aware reasoning (CHEAR), which moves part of this knowledge organization offline. CHEAR constructs and validates an activity hierarchy from recurrent cross-subject retrieval errors and builds corresponding activity-pair knowledge. At inference time, the hierarchy orders candidate comparisons; the LLM retains candidates progressively, and local discriminative knowledge is accessed only for the remaining activities. Across eight datasets, CHEAR achieves 86.79\% mean cross-subject accuracy, a 75.91\% candidate reduction rate, and 96.44\% true-activity retention. Per-sample token use is 85.15\% lower on average than in a variant without the hierarchical structure. Mean accuracy across eight cross-dataset transfer directions is 88.81\%. These results show that organizing candidates and discriminative knowledge around observed activity confusions reduces online LLM input while maintaining recognition performance.

Chinese Translation

大型语言模型（LLMs）为基于可穿戴传感器的人类活动识别提供了一种新方法。检索增强生成（RAG）进一步引入了类似运动示例和区分性传感器知识，使得决策能够基于与当前查询相关的观察结果。然而，更多的证据并不一定意味着更有条理的证据。当在线提示和LLM代理主要协调推理时，关于比较哪些活动以及何时调用区分性知识的决策仍然集中在推理时。随着活动词汇的增加，模型必须反复处理更多候选项及其相关知识，而过早修剪候选项则有可能排除真实活动。我们提出了混淆层次证据感知推理（CHEAR），将部分知识组织转移到离线进行。CHEAR从重复的跨主体检索错误中构建并验证活动层次结构，并建立相应的活动对知识。在推理时，该层次结构对候选项比较进行排序；LLM逐步保留候选项，并仅对剩余活动访问局部区分性知识。在八个数据集上，CHEAR实现了86.79%的平均跨主体准确率，75.91%的候选项减少率，以及96.44%的真实活动保留率。每个样本的令牌使用量平均比没有层次结构的变体低85.15%。在八个跨数据集转移方向上的平均准确率为88.81%。这些结果表明，围绕观察到的活动混淆组织候选项和区分性知识可以减少在线LLM输入，同时保持识别性能。

View on arXiv Download PDF AI Translation

cs.AI / 105 / 2610.15322

SuperAI: Merging Capabilities Across Heterogeneous Foundation Models

SuperAI：跨异构基础模型的能力融合

Dou, Jingtong, Shi, Chuancheng, Wu, Wenhua, Bao, Wei, Shen, Fei, Wang, Zhiyong, Chua, Tat-Seng

Abstract

Enabling the reuse and composition of knowledge and capabilities distributed across foundation models is an important step toward building scalable artificial intelligence (AI) systems. However, architectural differences and incompatible representations make it difficult to transfer and integrate these capabilities through a unified cross-model interface. To address this challenge, we propose SuperAI, a cross-architecture multi-model fusion framework that integrates complementary capabilities from multiple source models into a host model through orthogonal alignment of low-dimensional task-relevant representations. Specifically, we first extract task-relevant representations from the source models and the host model using shared anchor samples, and apply principal component analysis (PCA) to construct low-dimensional functional subspaces with the same dimensionality. Next, we use orthogonal Procrustes mappings to align the source and host coordinate systems, establishing cross-model correspondences while preserving the internal geometry of each source subspace. Finally, we train a lightweight residual multilayer perceptron (MLP) with joint supervision on task outputs and source-model increments, learning input-dependent residual increments that are injected into the host model for multi-model capability fusion. Experiments on three host models across 15 capability domains show that SuperAI achieves higher average domain scores than the evaluated cross-architecture fusion methods and better preserves general capabilities as the number of source models increases. Ablation studies further support the effectiveness of orthogonal alignment, nonlinear residual mapping, and joint supervision.

Chinese Translation

实现对分布在基础模型中的知识和能力的重用与组合是构建可扩展人工智能（AI）系统的重要一步。然而，架构差异和不兼容的表示使得通过统一的跨模型接口转移和整合这些能力变得困难。为了解决这一挑战，我们提出了SuperAI，一个跨架构的多模型融合框架，通过低维任务相关表示的正交对齐，将来自多个源模型的互补能力集成到主模型中。具体而言，我们首先使用共享锚样本从源模型和主模型中提取任务相关表示，并应用主成分分析（PCA）构建具有相同维度的低维功能子空间。接下来，我们使用正交Procrustes映射对源模型和主模型的坐标系统进行对齐，建立跨模型对应关系，同时保持每个源子空间的内部几何结构。最后，我们在任务输出和源模型增量上进行联合监督，训练一个轻量级残差多层感知器（MLP），学习输入相关的残差增量，并将其注入主模型以实现多模型能力融合。在15个能力领域的三个主模型上的实验表明，SuperAI的平均领域得分高于评估的跨架构融合方法，并且在源模型数量增加时更好地保留了通用能力。消融研究进一步支持了正交对齐、非线性残差映射和联合监督的有效性。

View on arXiv Download PDF AI Translation

cs.AI / 106 / 2610.15339

MultiFlow: Vision-Driven Multimodal fMRI Encoding via Stimulus-Informed Source Flow Matching

MultiFlow：基于刺激信息源流匹配的视觉驱动多模态fMRI编码

Van Hung, Bui, Vo, Hoang-Son, Huy, Mai Duc Minh, Vu, Dang Thanh, Kim, Soo-Hyung

Abstract

Current fMRI encoding models frame prediction as deterministic regression, yet identical stimuli evoke a distribution of neural responses (arising from physiological noise, hemodynamic variability, and cognitive fluctuations) that a single point estimate cannot capture. This work introduces MultiFlow, a generative fMRI encoding framework based on Conditional Flow Matching (CFM) and a formulation termed Stimulus-Informed Source Flow Matching (SISFM). SISFM generalises CFM by conditioning the source distribution on the stimulus: a learnable temporal convolution shapes the source mean with an HRF-length receptive field, while a variance head shrinks its scale, shortening the expected transport path. SISFM is instantiated with a Diffusion Transformer (DiT-X) decoder parameterising the velocity field, with sub-layer Attention Residuals, heterogeneous fusion of ten frozen vision, audio, language, and audiovisual backbones, and per-subject Yeo-network heads. On the Algonauts 2025 dataset, a single MultiFlow checkpoint recovers most of the benefit of large deterministic ensembles: it trails the 1000-model TRIBE ensemble in-distribution by 0.003 PCC (0.3163 vs. 0.3195), and the strongest in- and out-of-distribution ensembles (VIBE-20, MedARC-50) by 0.010 and 0.009 PCC, using a single trained model rather than an ensemble of 20-1000. The same checkpoint also provides stimulus-conditioned uncertainty estimates and sample-based diversity.

Chinese Translation

当前的fMRI编码模型将预测框架视为确定性回归，然而相同的刺激会引发一系列神经反应的分布（源于生理噪声、血流动力学变异和认知波动），而单一的点估计无法捕捉到这一点。本研究提出了MultiFlow，这是一种基于条件流匹配（Conditional Flow Matching, CFM）和一种称为刺激信息源流匹配（Stimulus-Informed Source Flow Matching, SISFM）的生成性fMRI编码框架。SISFM通过将源分布条件化于刺激上来推广CFM：一个可学习的时间卷积通过HRF长度的感受野来塑造源均值，而方差头则缩小其规模，缩短预期的传输路径。SISFM通过一个Diffusion Transformer (DiT-X) 解码器来实例化，该解码器参数化速度场，具有子层注意残差、十个冻结的视觉、音频、语言和视听骨干网的异构融合，以及每个受试者的Yeo网络头。在Algonauts 2025数据集中，单个MultiFlow检查点恢复了大规模确定性集成的绝大部分优势：它在分布内落后于1000模型的TRIBE集成0.003 PCC（0.3163对0.3195），并在最强的分布内和分布外集成（VIBE-20, MedARC-50）中分别落后0.010和0.009 PCC，使用的是单一训练模型而非20-1000的集成。同一检查点还提供了刺激条件的不确定性估计和基于样本的多样性。

View on arXiv Download PDF AI Translation

cs.AI / 107 / 2610.15365

TRACE: Temporal Risk Assessment and Concept Editing for Safer Language Generation

TRACE：用于更安全语言生成的时间风险评估与概念编辑

Garg, Ankur, Yu, Xuemin, Sajjad, Hassan, Kahou, Samira Ebrahimi

Abstract

Large language models (LLMs) can produce harmful content while generating a response. A useful safeguard should not only detect this risk during generation but also intervene while leaving safe responses largely unchanged and provide an auditable record of the internal patterns supporting its decisions. We propose TRACE (Temporal Risk Assessment and Concept Editing), a framework that tracks harmful-response risk during generation, selectively edits token activations, and leaves an auditable trace of its detection and steering decisions. TRACE learns a codebook of recurring activation patterns, with each codebook vector representing a latent concept. At inference time, it assigns each token activation to the nearest codebook vector to obtain the corresponding concept. A gated recurrent unit (GRU) processes the activations and their assigned concept features to predict completed-response harmfulness and a monotone cumulative streaming risk. During steering, activations assigned to harmful concepts are steered along directions defined by nearby benign targets. An external LLM labels the learned concepts post hoc, making detection sequences and harmful-to-benign edit records interpretable. Experiments across three generators and the WildGuard and S-Eval benchmarks show that TRACE improves streaming F1 over Kelp in five of six comparisons while remaining competitive on completed-response detection. TRACE's selective steering reduces Harmbench attack success with little change in refusal on originally safe responses. The post-hoc audit reveals recurring semantic structure in the learned concepts and distinct traces associated with harmful detection, safe generation, and activation edits.

Chinese Translation

大型语言模型（LLMs）在生成响应时可能会产生有害内容。有效的保护措施不仅应在生成过程中检测这一风险，还应进行干预，同时保持安全响应基本不变，并提供支持其决策的内部模式的可审计记录。我们提出了TRACE（时间风险评估与概念编辑），这是一个在生成过程中跟踪有害响应风险的框架，选择性地编辑标记激活，并留下其检测和引导决策的可审计痕迹。TRACE学习了一个重复激活模式的代码本，每个代码本向量代表一个潜在概念。在推理时，它将每个标记激活分配给最近的代码本向量，以获得相应的概念。门控递归单元（GRU）处理激活及其分配的概念特征，以预测完成响应的有害性和单调累积流风险。在引导过程中，分配给有害概念的激活沿着由附近良性目标定义的方向进行引导。一个外部LLM在事后标记学习到的概念，使得检测序列和有害到良性的编辑记录可解释。针对三个生成器以及WildGuard和S-Eval基准的实验表明，TRACE在六次比较中的五次上提高了流式F1，相较于Kelp，同时在完成响应检测上保持竞争力。TRACE的选择性引导在对原本安全的响应拒绝变化不大的情况下减少了Harmbench攻击的成功率。事后审计揭示了学习到的概念中重复的语义结构，以及与有害检测、安全生成和激活编辑相关的不同痕迹。

View on arXiv Download PDF AI Translation

cs.AI / 108 / 2610.15372

Fewer Bits, More Tries: Recovering Capability Under Memory Constraints

更少的位数，更多的尝试：在内存限制下的恢复能力

Galanti, Tomer, Susskind, Josh, Malach, Eran

Abstract

Large language models are generally more capable than smaller ones, but their memory requirements can make local deployment difficult. We ask whether part of this capability gap can instead be recovered by running a smaller or more compressed model multiple times and selecting among its outputs using public or model-generated tests. Single-user local inference can leave hardware underutilized, allowing parallel candidate generation to trade additional test-time computation for accuracy at modest latency cost. Across four code-generation benchmarks, verified retry recovers substantial accuracy lost to model size or compression. A 7B Qwen model with eight parallel attempts essentially matches a 32B model on MBPP+ and exceeds it on the other three benchmarks, while using about $4\times$ less model memory, running up to $6\times$ faster, and using less energy, as measured on Nvidia GPU. We observe similar tradeoff across model sizes, compression levels and hardware choices, although exact benefit depends on the hardware profile and problem setting. These results show that model capacity can be traded for verified test-time computation when correct solutions remain reachable within a modest retry budget.

Chinese Translation

大型语言模型通常比小型模型更具能力，但其内存需求可能使得本地部署变得困难。我们探讨是否可以通过多次运行较小或更压缩的模型，并使用公共或模型生成的测试在其输出中进行选择，从而部分弥补这种能力差距。单用户本地推理可能导致硬件未得到充分利用，因此允许并行候选生成在适度延迟成本下以额外的测试时间计算换取准确性。在四个代码生成基准测试中，经过验证的重试恢复了因模型大小或压缩而损失的大量准确性。一个7B的Qwen模型在八次并行尝试中，基本上与一个32B模型在MBPP+上的表现相当，并在其他三个基准测试中超过了它，同时使用的模型内存约为其$4 imes$，运行速度快至$6 imes$，并且在Nvidia GPU上测得的能耗更低。我们观察到在模型大小、压缩水平和硬件选择之间存在类似的权衡，尽管具体的收益取决于硬件配置和问题设置。这些结果表明，当正确的解决方案在适度的重试预算内仍然可达时，模型容量可以与经过验证的测试时间计算进行权衡。

View on arXiv Download PDF AI Translation

cs.AI / 109 / 2610.15380

Analyzing the Dynamics of WSBM-based Echo State Networks using InputDSA

基于 WSBM 的回声状态网络动态分析：使用 InputDSA

Mai, Huy, Clarke, Robert

Abstract

Bridging the structural gap between biological brains and artificial neural networks presents a pathway to altered computational dynamics. This paper introduces a Reservoir Computing framework that replaces random networks with biologically realistic modularity by applying the Weighted Stochastic Block Model (WSBM) to Echo State Networks (ESNs). This topological parameterization constructs reservoirs reflecting latent functional sub-circuits. Furthermore, the integration of Input Dynamical Similarity Analysis (InputDSA) provides a mechanism to isolate and compare how structural priors modulate intrinsic recurrent dynamics versus external inputs. By evaluating these architectures, we establish that structural diversity does not necessarily predicts behavioral diversity across the selected network subset. Furthermore, our demixed analysis confirms Anna Karenina principle of dynamics, demonstrating that environmental constraints enforce dynamical convergence in successful agents, while failing networks degenerate into idiosyncratic trajectories.

Chinese Translation

弥合生物大脑与人工神经网络之间的结构差距为改变计算动态提供了一条途径。本文引入了一种储备计算框架，通过将加权随机块模型（Weighted Stochastic Block Model, WSBM）应用于回声状态网络（Echo State Networks, ESNs），用生物学上现实的模块化替代随机网络。该拓扑参数化构建了反映潜在功能子电路的储备。此外，输入动态相似性分析（Input Dynamical Similarity Analysis, InputDSA）的整合提供了一种机制，以隔离和比较结构先验如何调节内在递归动态与外部输入。通过评估这些架构，我们建立了结构多样性不一定预测所选网络子集中的行为多样性。此外，我们的去混合分析确认了安娜·卡列尼娜动态原则，表明环境约束在成功代理中强制实施动态收敛，而失败的网络则退化为特异的轨迹。

View on arXiv Download PDF AI Translation

cs.AI / 110 / 2610.15403

A Decade of PhotoMappers: A Longitudinal Study of Crowdsourced Disaster Photos Geolocalization with Vision-Language Models and Geospatial Reasoning

十年的PhotoMappers：基于视觉-语言模型和地理空间推理的众包灾难照片地理定位的纵向研究

Li, Hao, Yin, Wenping, Deuser, Fabian, Jia, Jia, Liu, Ziqi, Juhász, Levente, Zhang, Fan, Biljecki, Filip

Abstract

Crowdsourced disaster imagery provides timely, ground-level observations when conventional disaster mapping is delayed or incomplete, yet its geographic value depends on where observations are produced, how they are interpreted, and whether they can be reliably located as actionable Geospatial Intelligence (GEOINT). Using more than a decade of PhotoMappers, a crowdsourced disaster-mapping initiative, this study examines the changing geography of disaster information and asks how emerging GeoAI models compare with human geographic interpretation. We combine longitudinal spatial analysis with cross-view geolocalization using street view imagery (SVI) and remote sensing imagery (RSI), and evaluate vision-language models (VLMs) reasoning through 30 indicators spanning geographic cues and reasoning evidence. The results reveal a geographically and temporally uneven archive shaped by changing disaster types, Community Lifelines, and operational priorities. Integrating VGI, SVI, and RSI with Sinkhorn re-ranking achieves 40.90% R@1, 61.10% R@5, and 69.45% R@10, while yearly R@1 varies markedly from 25.00% to 65.07%. Geographic reasoning is similarly uneven: Qwen3.6-27B achieves an overall agreement of 67% compared with 51% for InternVL3-8B, with both models showing greater difficulty with localized and place-specific evidence than broad spatial configurations. We argue that the value of GeoAI lies less in replacing volunteered geographic interpretation than in complementing it, scaling candidate retrieval while retaining human judgment for ambiguous and consequential cases. This provides a pathway toward more human-centered and responsible GeoAI for disaster mapping. All data and code are openly available at https://github.com/bobleegogogo/PhotoMappers-VLM.

Chinese Translation

众包灾难影像在传统灾难制图延迟或不完整时提供及时的地面观察，但其地理价值取决于观察产生的地点、如何解读这些观察以及是否能够可靠地定位为可操作的地理空间情报（GEOINT）。本研究利用超过十年的PhotoMappers这一众包灾难制图倡议，考察灾难信息的地理变化，并探讨新兴的GeoAI模型与人类地理解读的比较。我们结合纵向空间分析与使用街景影像（SVI）和遥感影像（RSI）的交叉视角地理定位，评估视觉-语言模型（VLMs）通过30个指标进行推理，这些指标涵盖了地理线索和推理证据。结果显示，一个在地理和时间上不均匀的档案被不同的灾难类型、社区生命线和操作优先级所塑造。将志愿地理信息（VGI）、SVI和RSI与Sinkhorn重排序结合，达到了40.90%的R@1、61.10%的R@5和69.45%的R@10，而年度R@1则显著变化，从25.00%到65.07%。地理推理同样不均匀：Qwen3.6-27B的总体一致性为67%，而InternVL3-8B为51%，两者在处理局部和特定地点证据时表现出更大的困难，而在广泛的空间配置中则较为顺利。我们认为GeoAI的价值在于不在于取代志愿的地理解读，而在于补充它，扩大候选检索的规模，同时保留人类在模糊和重要案例中的判断。这为灾难制图提供了更以人为本和负责任的GeoAI发展路径。所有数据和代码均可在https://github.com/bobleegogogo/PhotoMappers-VLM上公开获取。

View on arXiv Download PDF AI Translation

cs.AI / 111 / 2610.15407

State Representation Shapes Long-Horizon Agent Learning and Transfer: An Empirical Study

状态表示形状长时间代理学习与迁移：一项实证研究

Nie, Dong, Lan, Rui, Nie, Yixin, Hong, Lingzi

Abstract

An end-to-end success rate alone cannot distinguish failure to learn an action from failure to apply it as task state changes. We study this distinction in deterministic tool workflows that provide the current artifact state and candidate action recipes. Using 1.7B, 8B, and 27B Qwen models, we compare supervised learning across workflow horizons and cross the state representation used for training with the one used at evaluation. Horizon-matched supervision yields high completion, but an H4-trained 1.7B model transfers poorly to longer workflows under full-state observations. In an H16 crossover, compact-trained 1.7B adapters complete 50/50 held-out tasks in compact view and 0/50 in full view for each of three paired training seeds; full-trained outcomes vary across seeds. A separate 8B adapter pair shows a similar asymmetry. The view dependence also appears in paired-choice tasks where both candidate calls are executable and the correct choice depends on table contents; here, the compact mask does not consult the target action. An explicit relevance cue improves full-view execution but does not close the gap for the tested checkpoint. In a separate 27B ScienceWorld study, compact-state training improves local accuracy under both views, indicating that the transfer pattern depends on the task and observation construction. Together, these experiments identify representation-sensitive transfer as a bottleneck in the controlled workflows, without isolating content reduction from other changes to the observation.

Chinese Translation

仅凭端到端的成功率无法区分学习动作的失败与在任务状态变化时应用动作的失败。我们在确定性工具工作流程中研究这一区别，该工作流程提供当前的工件状态和候选动作配方。使用1.7B、8B和27B的Qwen模型，我们比较了跨工作流程时间的监督学习，并将用于训练的状态表示与用于评估的状态表示进行了交叉匹配。与时间匹配的监督学习产生了高完成率，但在全状态观察下，经过H4训练的1.7B模型在更长的工作流程中迁移效果较差。在H16交叉中，经过紧凑训练的1.7B适配器在紧凑视图中完成了50/50的保留任务，而在全视图中则为0/50，且针对三个配对训练种子，完全训练的结果在不同种子间有所变化。一个独立的8B适配器对显示了类似的不对称性。视图依赖性也出现在配对选择任务中，其中两个候选调用都是可执行的，正确选择取决于表格内容；在这里，紧凑掩码并未参考目标动作。一个明确的相关性提示改善了全视图执行，但并未缩小对所测试检查点的差距。在一项独立的27B ScienceWorld研究中，紧凑状态训练在两种视图下提高了局部准确性，表明迁移模式依赖于任务和观察构造。综合这些实验，识别出表示敏感的迁移是受控工作流程中的瓶颈，而未将内容减少与观察的其他变化隔离开来。

View on arXiv Download PDF AI Translation

cs.AI / 112 / 2610.15409

CoverSynth: Coverage-Guided Generation of Retrieval Evaluation Sets

CoverSynth：覆盖指导的检索评估集生成

Sharma, Anahita, Shanker, Apaar, Chakravarti, Rishav, Srinivasan, Sneha, Kalmath, Vijay, Wilbur, Dylan, Li, Yuan, Bolouki, Sara, Xue, Yuan

Abstract

Enterprise agents need retrieval evaluation sets that expose domain-specific failures and guide developers towards efficient deployment tradeoffs. Neither expert-human-authored sets nor current synthetic generation methods provide a methodology for achieving the necessary coverage, reliability, efficiency, and discriminability required for such decision-making. Simultaneously, the field does not have an assessment framework for determining when an evaluation set has satisfied these properties. We introduce an evaluation set assessment framework for these criteria and CoverSynth -- a generator specifically designed to satisfy these criteria without incurring the high cost of human expert labeling. Starting from an unannotated corpus, CoverSynth targets coverage across retrieval demands and system capabilities. Deterministic checks locate evidence quotes in their sources, while model-based checks assess semantic validity to guarantee reliable ground truth. Across five public enterprise-style corpora, our method provides informative variation in difficulty, revealing unseen failure modes such as an inability of agents to handle underspecified or ambiguous requests. On the RepLiQA corpus, 91\% of human-generated queries can be answered by all members on a panel of 12 retrieval systems, compared to only 12\% of CoverSynth-generated queries. Finally, in a repeated TechQA/RepLiQA study, 12 of 12 within-harness model contrasts have positive nominal 95\% intervals on generated sets, compared with 4 of 12 on human sets. CoverSynth makes coverage and evidence quality explicit construction objectives, producing corpus-grounded evaluations that reveal retrieval headroom and differences between agent configurations.

Chinese Translation

企业代理需要能够揭示特定领域失败的检索评估集，并引导开发人员进行有效的部署权衡。无论是专家人工编写的评估集，还是当前的合成生成方法，都未能提供实现必要的覆盖性、可靠性、效率和可区分性所需的方法论。同时，该领域缺乏评估框架来确定评估集何时满足这些属性。我们引入了一个评估集评估框架，以满足这些标准，并提出了CoverSynth——一个专门设计用于满足这些标准的生成器，而无需承担高昂的人类专家标注成本。CoverSynth从未标注的语料库开始，针对检索需求和系统能力进行覆盖。确定性检查在其来源中定位证据引用，而基于模型的检查评估语义有效性，以确保可靠的真实情况。在五个公共企业风格的语料库中，我们的方法提供了信息丰富的难度变化，揭示了未被发现的失败模式，例如代理无法处理不明确或模糊请求。在RepLiQA语料库中，91%的人工生成查询可以被12个检索系统面板的所有成员回答，而CoverSynth生成的查询仅为12%。最后，在重复的TechQA/RepLiQA研究中，12个模型对比中有12个在生成集上具有正的名义95%置信区间，而在人工集上仅有4个。CoverSynth明确将覆盖性和证据质量作为构建目标，生成基于语料的评估，揭示检索的潜力和代理配置之间的差异。

View on arXiv Download PDF AI Translation

cs.AI / 113 / 2610.15417

SpecFit: Not All Draft Tokens Cost the Same in Offloaded MoE Inference

SpecFit：在卸载的 MoE 推理中，并非所有草稿令牌的成本相同

Cao, Hongliang, Li, Xiuyu, Wang, Siqi, Wang, Ziming, Zhu, Jingmin, Yang, Hailong

Abstract

Mixture-of-experts (MoE) models activate only a few experts per token, but when their expert weights are offloaded to host memory, host-to-device (H2D) expert transfers enter the critical path of decoding. Speculative decoding verifies several draft tokens in one forward pass, yet in an MoE model each verified token activates its own experts, so a larger draft tree transfers more experts. Because of this H2D cost, speculative decoding commits several tokens per verification cycle yet is sometimes even slower than autoregressive decoding. Based on this observation, we propose SpecFit, which chooses the verified draft subtree to maximize predicted throughput rather than acceptance length. SpecFit estimates the expected number of committed tokens from draft path probabilities and predicts cycle time from the node count and the per-layer union of the experts the subtree activates; this union is a monotone submodular coverage function, so each node pays only for the experts it adds. An exact search over ancestor-closed subtrees returns the subtree with the highest predicted throughput; run after each draft level, the same search also decides when to stop drafting, and the unmodified target verifies the result. On CNN/DailyMail summarization and a GSM8K subset with Qwen3-30B-A3B, SpecFit decodes 21.6% and 7.0% faster than EAGLE-3 with the same expert cache and 1.97x and 1.72x as fast as vanilla EAGLE-3. Code is available at https://github.com/Chlience/SpecFit.

Chinese Translation

混合专家（MoE）模型每个令牌仅激活少数专家，但当其专家权重被卸载到主机内存时，主机到设备（H2D）专家传输进入解码的关键路径。投机解码在一次前向传递中验证多个草稿令牌，但在 MoE 模型中，每个验证的令牌激活其自身的专家，因此更大的草稿树传输更多的专家。由于这种 H2D 成本，投机解码在每个验证周期承诺多个令牌，但有时甚至比自回归解码更慢。基于这一观察，我们提出了 SpecFit，它选择经过验证的草稿子树，以最大化预测的吞吐量而非接受长度。SpecFit 从草稿路径概率中估计承诺令牌的预期数量，并根据节点数量和子树激活的每层专家的并集预测周期时间；该并集是一个单调子次模覆盖函数，因此每个节点仅为其添加的专家付费。对祖先闭合子树的精确搜索返回具有最高预测吞吐量的子树；在每个草稿级别后运行的相同搜索还决定何时停止草稿，而未修改的目标验证结果。在 CNN/DailyMail 摘要和一个包含 Qwen3-30B-A3B 的 GSM8K 子集上，SpecFit 的解码速度比具有相同专家缓存的 EAGLE-3 快 21.6% 和 7.0%，比普通 EAGLE-3 快 1.97 倍和 1.72 倍。代码可在 https://github.com/Chlience/SpecFit 获取。

View on arXiv Download PDF AI Translation

cs.AI / 114 / 2610.15449

Hidden in Plain Sight: Advancing Harmful Content Understanding via Perception-Aware Adversarial Self-Play

显而易见的隐秘：通过感知意识对抗自我博弈推进有害内容理解

Xiao, Yisong, Cui, Benlei, Zhao, Shiji, Ying, Zonghao, Zhang, Yaoyuan, Huang, Longtao, Liu, Aishan, Xue, Hui, Ma, Ting, Hong, Haiwen, Liu, Xianglong

Abstract

Despite significant advances in language understanding, Large Language Models (LLMs) can still miss meaning hidden in plain sight, failing to identify harmful intent in expressions readily understood by humans. Malicious users exploit this gap by strategically disguising harmful content to evade automated moderation, posing a serious threat to online platform safety. The vulnerability reflects a fundamental cognitive asymmetry between human readers, who integrate contextual, phonetic, and visual cues to decipher the intent behind deliberate obfuscation, and text-based LLMs, which rely primarily on token sequences and have limited exposure to evolving evasions. To bridge this asymmetry, we propose PASP (Perception-Aware Adversarial Self-Play), which combines multi-perceptual views with adversarial self-play to improve understanding of harmful intent as evasions evolve. Specifically, PASP constructs textual, phonetic, and visual views of each input, preserving its original context while exposing pronunciation and glyph cues obscured in token space. During adversarial self-play, an attacker uses defender feedback to generate intent-preserving evasions, while the defender is trained using an intent-invariance reward to understand harmful intent across changing surface forms. Extensive experiments show that PASP reduces adaptive attack success rates by up to 73.71% and 74.49% on ChineseHarm-Bench and EvoHarmBench, respectively, preserves model capabilities with minimal degradation, and effectively interprets real-world evasive expressions in e-commerce.

Chinese Translation

尽管语言理解方面取得了显著进展，大型语言模型（LLMs）仍然可能错过显而易见的隐含意义，无法识别人类易于理解的表达中的有害意图。恶意用户利用这一差距，通过战略性地伪装有害内容来规避自动化审核，给在线平台安全带来了严重威胁。这一脆弱性反映了人类读者与基于文本的LLMs之间的基本认知不对称：人类读者能够整合上下文、语音和视觉线索，以解读故意模糊的意图，而文本基础的LLMs主要依赖于标记序列，对不断演变的规避策略的接触有限。为了弥合这种不对称性，我们提出了PASP（感知意识对抗自我博弈），它结合了多感知视角与对抗自我博弈，以改善对有害意图的理解，随着规避策略的演变而不断提升。具体而言，PASP为每个输入构建文本、语音和视觉视角，保留其原始上下文，同时揭示在标记空间中被遮蔽的发音和字形线索。在对抗自我博弈过程中，攻击者利用防御者的反馈生成保持意图的规避策略，而防御者则通过意图不变性奖励进行训练，以理解在变化表面形式下的有害意图。大量实验表明，PASP在ChineseHarm-Bench和EvoHarmBench上分别将自适应攻击成功率降低了高达73.71%和74.49%，在保持模型能力的同时，降级最小，并有效解释了电子商务中真实世界的规避表达。

View on arXiv Download PDF AI Translation

cs.AI / 115 / 2610.15461

Scalable Robust MCTS with Anytime Deterministic Optimality Guarantees

可扩展的鲁棒蒙特卡洛树搜索（MCTS）与随时确定性的最优性保证

Massas, Tomer, Indelman, Vadim

Abstract

MDP planners often rely on nominal transition models that may differ from the true dynamics. Robust MDP planning accounts for this uncertainty through a set of admissible models, but doing so with limited online computation is challenging: the search tree grows exponentially with the horizon and each robust backup requires an additional optimization. Existing online robust methods remain limited in scale and offer at most probabilistic guarantees. We introduce, to the best of our knowledge, the first deterministic anytime guarantees for robust online MDP planning. We restrict computation to the sampled search tree while deterministically bounding the contribution of unsampled trajectories. For any fixed policy, we derive lower and upper bounds on the gap between its projected and true robust values under rectangular and non-rectangular uncertainty sets. Under rectangular uncertainty, we propagate these bounds through the robust Bellman operator to bound the optimal robust value. These bounds are computable from the current sampled tree, hold at every node, tighten as the sampled support expands, and converge to the robust values as full support is recovered. We develop CAR-MCTS, a bound-guided MCTS planner that uses these deterministic bounds to guide exploration, prune provably suboptimal branches, and terminate once the optimal root action is certified. The fixed-policy bounds also yield an anytime policy-evaluation mechanism that certifies the robust value of any policy, independent of how it was obtained. Experiments verify the deterministic bounds, their tightening under support expansion, substantial branch pruning, root-action certification, and convergence to the optimal robust policy. To our knowledge, this is the first scalable robust MCTS framework with anytime deterministic guarantees for both online planning and policy evaluation, without enumerating the full state space.

Chinese Translation

马尔可夫决策过程（MDP）规划者通常依赖于名义过渡模型，这些模型可能与真实动态存在差异。鲁棒MDP规划通过一组可接受模型来考虑这种不确定性，但在有限的在线计算条件下实现这一点具有挑战性：搜索树随着时间范围的增加呈指数增长，每次鲁棒备份还需要额外的优化。现有的在线鲁棒方法在规模上仍然有限，最多只能提供概率性保证。我们首次提出了鲁棒在线MDP规划的确定性随时保证。我们将计算限制在采样的搜索树上，同时确定性地界定未采样轨迹的贡献。对于任何固定策略，我们推导了其在矩形和非矩形不确定性集下的投影鲁棒值与真实鲁棒值之间的上下界。在矩形不确定性下，我们通过鲁棒贝尔曼算子传播这些界限，以界定最优鲁棒值。这些界限可以从当前的采样树中计算得出，在每个节点上都成立，随着采样支持的扩展而收紧，并在完全支持恢复时收敛到鲁棒值。我们开发了CAR-MCTS，这是一种基于界限引导的MCTS规划器，利用这些确定性界限来引导探索，剪枝可证明的次优分支，并在认证最优根动作后终止。固定策略界限还提供了一种随时策略评估机制，能够认证任何策略的鲁棒值，而不论其获取方式。实验验证了确定性界限、支持扩展下的收紧、显著的分支剪枝、根动作认证以及收敛到最优鲁棒策略。据我们所知，这是第一个具有随时确定性保证的可扩展鲁棒MCTS框架，适用于在线规划和策略评估，而无需枚举完整的状态空间。

View on arXiv Download PDF AI Translation

cs.AI / 116 / 2610.15463

MemeLogos: Knowledge-Driven Retrieval-Augmented Reasoning for Harmful Meme Detection

MemeLogos：基于知识驱动的增强检索推理框架用于有害表情包检测

Gu, Hexiang, Hou, Saihui, Liu, Changlan, Ren, Min, Wu, Huijia, Gu, Hongxun, He, Zhaofeng

Abstract

With the rapid proliferation of social media, harmful memes have become a growing threat to online ecosystems. Their harmfulness is often implicitly conveyed through complex interactions among visual, textual, and contextual cues, making accurate detection challenging. Recent studies have introduced Retrieval-Augmented Generation (RAG) and external knowledge into harmful meme detection. However, existing methods mainly retrieve similar instances or auxiliary information, limiting their ability to capture reusable harmfulness knowledge and underlying reasoning mechanisms, and thus constraining both detection performance and interpretability. To address these challenges, we propose MemeLogos, a knowledge-driven retrieval-augmented reasoning framework for harmful meme detection. MemeLogos abstracts recurring harmfulness patterns and reasoning structures from diverse harmful cases into reusable knowledge units and organizes them into a dedicated star-shaped knowledge graph. By combining structured knowledge retrieval with an explicit multi-stage reasoning process inspired by cognitive mechanisms, MemeLogos leverages retrieved knowledge to guide harmfulness reasoning, improving both detection reliability and interpretability. Extensive experiments show that MemeLogos consistently outperforms existing harmful meme detection methods while generating more coherent and interpretable reasoning, demonstrating the effectiveness of structured harmfulness knowledge for multimodal content understanding.

Chinese Translation

随着社交媒体的快速传播，有害表情包已成为在线生态系统日益严重的威胁。它们的有害性通常通过视觉、文本和上下文线索之间的复杂互动隐含传达，使得准确检测变得具有挑战性。近期研究引入了增强检索生成（Retrieval-Augmented Generation, RAG）和外部知识用于有害表情包检测。然而，现有方法主要检索相似实例或辅助信息，限制了其捕捉可重用有害性知识和潜在推理机制的能力，从而制约了检测性能和可解释性。为了解决这些挑战，我们提出了MemeLogos，一个基于知识驱动的增强检索推理框架用于有害表情包检测。MemeLogos将来自多样化有害案例的重复有害性模式和推理结构抽象为可重用的知识单元，并将其组织成一个专门的星形知识图谱。通过将结构化知识检索与受认知机制启发的明确多阶段推理过程相结合，MemeLogos利用检索到的知识指导有害性推理，从而提高检测的可靠性和可解释性。大量实验表明，MemeLogos在有害表情包检测方面始终优于现有方法，同时生成更连贯和可解释的推理，证明了结构化有害性知识在多模态内容理解中的有效性。

View on arXiv Download PDF AI Translation

cs.AI / 117 / 2610.15481

SkillFusion: Towards Reliable Multi-Skill Composition for LLM Agents

技能融合：面向可靠的多技能组合用于大型语言模型代理

Huang, Enhao, Zhao, Xinye, Li, Yiming, Yang, Bingrun, Zhou, Nanjun, Xia, Chunshu, Yao, Hongwei, Qin, Zhan

Abstract

Independently authored skills can impose incompatible requirements, competing preferences, and redundant instructions when jointly loaded by an LLM agent. We propose SkillFusion, an offline compiler that reconciles these interactions into a reusable merged skill. Its central idea is to make composition decisions explicit through typed, scoped constraints linked to their source directives. A type-scope gate selects candidate interactions, while selective fallback revisits pairs potentially missed by structural pruning. A classifier identifies interaction types, and an adversarial verifier re-examines uncertain judgments. The resolver applies conditional routing or explicit priorities to hard conflicts, verified abstraction to soft conflicts, and deduplication to redundant instructions. Source-linked checks guide synthesis and compression, and metadata synthesis supports discovery of the merged skill. The resulting artifact retains associated scripts and resources and can be reused across tasks and executor models. SkillFusion achieves the highest mean Success among evaluated non-oracle methods on all three agent benchmarks and on both transfer benchmarks across three executor models. At the largest tested instruction configurations, it achieves over 2.5 times the Success of unresolved multi-skill loading, highlighting the growing value of reconciliation as composition becomes more demanding. The code will be open-sourced after the review process concludes.

Chinese Translation

独立编写的技能在被大型语言模型（LLM）代理共同加载时，可能会产生不兼容的要求、相互竞争的偏好以及冗余的指令。我们提出了技能融合（SkillFusion），这是一种离线编译器，旨在将这些交互整合为可重用的合并技能。其核心思想是通过与源指令相关联的类型化、范围约束，使组合决策变得明确。类型范围门控选择候选交互，而选择性回退则重新审视可能因结构修剪而被遗漏的配对。分类器识别交互类型，敌对验证器重新审查不确定的判断。解析器对硬冲突应用条件路由或显式优先级，对软冲突应用验证抽象，并对冗余指令进行去重。源关联检查指导合成和压缩，元数据合成支持合并技能的发现。最终生成的工件保留相关脚本和资源，并可在不同任务和执行模型中重用。技能融合在所有三个代理基准和跨三个执行模型的两个迁移基准中，达到了评估的非oracle方法中最高的平均成功率。在测试的最大指令配置下，其成功率超过未解决的多技能加载的2.5倍，突显了随着组合需求的增加，调和的价值日益增长。代码将在评审过程结束后开源。

View on arXiv Download PDF AI Translation

cs.AI / 118 / 2610.15492

When Skills Remain But Access Fails: Factorized Skill Memory for Continual Skill Internalization

技能保留但访问失败：持续技能内化的分解技能记忆

Qiu, Haomiao, Tang, Yining, Wan, Hongyi, Sun, Changfeng, Qiao, Ziyue, Zhang, Miao

Abstract

Skill internalization aims to convert externally specified procedural skills into persistent model capabilities, allowing language-model agents to execute them without retrieving or reinjecting skill descriptions at inference time. Yet when new skills are learned continually, whether previously internalized skills remain usable is still poorly understood. We study forgetting in continual skill internalization and uncover a clear hierarchical structure. Early in subsequent training, a model may lose the ability to invoke an old skill from the raw request, while supplying only the correct canonical skill name substantially restores the learned behavior; with further optimization, this conditional execution also degrades. We call these two stages skill-addressing forgetting and skill-execution forgetting. Guided by this finding, we introduce Factorized Skill Memory (FSM), which separates continual skill memory into a plastic Address Memory and a stable Execution Memory. At each stage, FSM trains a new LoRA executor using only current-stage data, freezes it after acquisition, and appends it to Execution Memory; in parallel, Address Memory incrementally learns to map current requests to executors without replaying historical requests or execution trajectories. At inference time, Address Memory selects one executor from the raw request, and only that frozen executor performs a single language-model generation. On a five-stage benchmark built from 100 TaskBench skills, FSM achieves 65.95% mean exact match and 93.38% addressing accuracy across two task orders, improving over the strongest replay-free parameter-isolation baseline by 14.3 percentage points while remaining within 1.9 points of Replay, which retains historical training examples.

Chinese Translation

技能内化旨在将外部指定的程序技能转化为持久的模型能力，使语言模型代理能够在推理时执行这些技能，而无需检索或重新注入技能描述。然而，当新技能持续学习时，之前内化的技能是否仍然可用仍然不甚明了。我们研究了持续技能内化中的遗忘现象，并揭示了一个明确的层次结构。在后续训练的早期，模型可能会失去从原始请求中调用旧技能的能力，而仅提供正确的标准技能名称则能显著恢复已学得的行为；随着进一步的优化，这种条件执行也会退化。我们将这两个阶段称为技能寻址遗忘和技能执行遗忘。在这一发现的指导下，我们引入了分解技能记忆（Factorized Skill Memory, FSM），它将持续技能记忆分为可塑的地址记忆（Address Memory）和稳定的执行记忆（Execution Memory）。在每个阶段，FSM仅使用当前阶段的数据训练一个新的LoRA执行器，在获取后将其冻结，并将其附加到执行记忆中；与此同时，地址记忆逐步学习将当前请求映射到执行器，而无需重放历史请求或执行轨迹。在推理时，地址记忆从原始请求中选择一个执行器，只有该冻结的执行器执行一次语言模型生成。在一个由100个TaskBench技能构建的五阶段基准测试中，FSM在两个任务顺序中实现了65.95%的平均精确匹配率和93.38%的寻址准确率，相较于最强的无重放参数隔离基线提高了14.3个百分点，同时与保留历史训练示例的重放方法相差仅1.9个百分点。

View on arXiv Download PDF AI Translation

cs.AI / 119 / 2610.15501

Large Decision Models: Repurposing Large Language Models as Single-Pass Decision Makers

大型决策模型：将大型语言模型重新用于单次决策制定

Chen, Ziyi, Zhang, Yan, Wei, Jianhui, Yin, Jianing, Zhang, Daoan

Abstract

Large language models make decisions through parallel prefill and sequential decoding. Although effective, this approach incurs decoding costs and does not explicitly expose decision probabilities. Existing task-specific classifiers require per-task training and offer limited generalization to unseen decision schemas. We propose Large Decision Models (LDMs), repurposing language models to return explicit probabilities over labels specified at run time using one prefill forward pass. We first adapt models to this interface through Diverse Decision Supervision (DDS), covering varied domains, rules and labels. To further narrow the performance gap to a reasoning model, Self-Posed Reasoning Distillation (SPRD) adds tasks generated by a frozen teacher, filters them for answer consistency, and trains on label distributions obtained from repeated teacher reasoning. We evaluate on DecisionVista, pairing new-schema tests with held-out human-labelled sources. Across three base models, the recipe improves accuracy over a step-matched baseline by 1.8 and 1.5 percentage points, respectively, achieving a 182-fold speedup in latency over reasoning rollout on Gemma-4-31B. Reasoning remains stronger on the new-schema test, motivating a confidence router for uncertain decisions. On development sets, routing recovers a median 92% of reasoning's accuracy advantage while generating 27-58% of its tokens. Code is available at https://github.com/JianhuiWei7/LDM.

Chinese Translation

大型语言模型通过并行预填充和顺序解码来做出决策。尽管这种方法有效，但会产生解码成本，并且未明确暴露决策概率。现有的任务特定分类器需要针对每个任务进行训练，并且对未见过的决策模式的泛化能力有限。我们提出了大型决策模型（Large Decision Models, LDMs），将语言模型重新用于在运行时通过一次预填充前向传递返回标签的明确概率。我们首先通过多样化决策监督（Diverse Decision Supervision, DDS）将模型适配到这一接口，涵盖不同的领域、规则和标签。为了进一步缩小与推理模型的性能差距，自我提出推理蒸馏（Self-Posed Reasoning Distillation, SPRD）添加由冻结教师生成的任务，过滤答案一致性，并基于重复教师推理获得的标签分布进行训练。我们在DecisionVista上进行评估，将新模式测试与保留的人类标注源配对。在三个基础模型中，该方法在与步骤匹配的基线相比，分别提高了1.8和1.5个百分点的准确率，并在Gemma-4-31B上实现了182倍的延迟加速。推理在新模式测试中仍然表现更强，促使我们为不确定决策设计了一个置信路由器。在开发集上，路由恢复了推理准确性优势的中位数92%，同时生成了27-58%的标记。代码可在 https://github.com/JianhuiWei7/LDM 获取。

View on arXiv Download PDF AI Translation

cs.AI / 120 / 2610.15503

Data Portability Revisited: Toward the Human-Centric, AI-Driven Data Ecosystems of Tomorrow

数据可携性再探：迈向以人为本的人工智能驱动的数据生态系统

Fenwick, Mark, Fertik, Michael, Jurcys, Paulius, Minssen, Timo

Abstract

This Article critically examines the contemporary regulatory framework and discourse surrounding data portability in the United States. Using recent regulatory developments in the European Union as an illustration, this Article suggests that although data access and portability are identified as vital issues in multiple policy instruments, in its current iteration, at least, legal conceptions of portability continue to reinforce the interests of service providers and data controller enterprises rather than individual end users. This Article argues that a paradigm shift toward a more human-centric data approach to data governance must occur, under which data would be recognized as fundamental to an individual's identity in a digital age. Therefore, it should be placed in the hands of individuals rather than service providers or data controller enterprises. This Article considers technical and market trends in the European Union that reveal and facilitate such a change. It suggests that regulatory frameworks should better align with these technological and market developments to encourage change-inducing trends among market actors. In short, this Article identifies a transformative approach to data portability that empowers individuals with the freedom and ability to aggregate their data in secure personal spaces under their control or dominion. Such a human-centric perspective on data portability is crucial in building Artificial Intelligence (AI)-powered applications for individual consumers that can pave the way for the human-centric, AI-driven data ecosystems of the future.

Chinese Translation

本文对美国当前的数据可携性监管框架及相关话语进行了批判性审视。以欧盟近期的监管发展为例，本文指出，尽管数据访问和可携性在多个政策工具中被视为重要问题，但至少在当前的版本中，法律对可携性的定义仍然强化了服务提供商和数据控制企业的利益，而非个人终端用户的利益。本文认为，必须向更以人为本的数据治理方法进行范式转变，在这种方法下，数据将被视为数字时代个体身份的基础。因此，数据应当掌握在个人手中，而非服务提供商或数据控制企业。本文考虑了欧盟的技术和市场趋势，这些趋势揭示并促进了这种转变。它建议监管框架应更好地与这些技术和市场发展相一致，以鼓励市场参与者之间引发变革的趋势。简而言之，本文提出了一种变革性的数据可携性方法，使个人能够在其控制或支配的安全个人空间中自由地聚合数据。这种以人为本的数据可携性视角对于构建以人工智能（AI）为驱动的个体消费者应用至关重要，能够为未来以人为本的、人工智能驱动的数据生态系统铺平道路。

View on arXiv Download PDF AI Translation

cs.AI / 121 / 2610.15529

Agent-Canary-Mas: End-to-End Evaluation of Protocol Risk in Long-Horizon Multi-Agent Systems

Agent-Canary-Mas：长时间跨度多智能体系统中协议风险的端到端评估

Liu, Xiuming, Gai, Le, Chang, Zhiyuan, Li, Peiyang, Zhu, Bin, Wang, Shuo, Feng, Chao

Abstract

Agent-to-agent protocols allow LLM agents to discover peers, exchange results, and coordinate tool use. A locally successful protocol attack, however, may be rejected by a later agent or propagated into a consequential task decision. Measuring local exploitability alone leaves this difference unresolved and provides limited evidence about which participant caused the final outcome. We present Agent-Canary-Mas, an end-to-end evaluation system that follows protocol risk through long-horizon multi-agent tasks. It covers five risk classes and 13 subclasses across A2A, ANP, and BCS, with 975 matched malicious--benign task pairs in five business domains. A shared OpenClaw framework runs autonomous, tool-using agents under different underlying models while preserving the task and execution setup. Matched runs separate local protocol success from downstream consequences; retained evidence connects participant selection, message consumption, tool activity, and task-state changes. We further evaluate a protocol-context Guard that checks an object's authority at consequential handoffs. Across three model conditions, local attack success is frequently at or above 90%, while consequence conversion ranges from 7.1\% to 100.0\%. Cross-agent evidence raises macro-averaged Audit-F1 from 29.7% to 95.5%. On Finance tasks, the Guard lowers average ASR from 86.1% to 53.8% and PTCR from 65.5% to 31.2%, with a measurable completion tradeoff. These results show that protocol risk depends on how agents use exchanged objects throughout the task, and that evaluating this path supports more informative risk assessment, attribution, and defense.

Chinese Translation

智能体间协议允许大型语言模型（LLM）智能体发现同伴、交换结果并协调工具使用。然而，一个在本地成功的协议攻击可能会被后续的智能体拒绝，或传播到重要的任务决策中。仅仅测量本地可利用性无法解决这一差异，并且提供的证据有限，无法明确指出是哪个参与者导致了最终结果。我们提出了Agent-Canary-Mas，这是一个端到端的评估系统，能够在长时间跨度的多智能体任务中跟踪协议风险。该系统涵盖了五个风险类别和13个子类别，涉及A2A、ANP和BCS，包含五个业务领域中975对匹配的恶意-良性任务对。一个共享的OpenClaw框架在不同的基础模型下运行自主的、使用工具的智能体，同时保持任务和执行设置。匹配的运行将本地协议成功与下游后果分开；保留的证据连接了参与者选择、消息消费、工具活动和任务状态变化。我们进一步评估了一个协议上下文的Guard，它在重要交接时检查对象的权限。在三种模型条件下，本地攻击成功率通常在90%或以上，而后果转化率范围为7.1%到100.0%。跨智能体的证据将宏观平均审计F1从29.7%提高到95.5%。在金融任务中，Guard将平均攻击成功率（ASR）从86.1%降低到53.8%，将协议转化率（PTCR）从65.5%降低到31.2%，并伴随可测量的完成权衡。这些结果表明，协议风险取决于智能体在整个任务中如何使用交换的对象，而评估这一过程有助于更具信息性的风险评估、归因和防御。

View on arXiv Download PDF AI Translation

cs.AI / 122 / 2610.15575

Self-Representation Alignment in Diffusion Models for Video Background Music Generation

扩散模型中自我表征对齐用于视频背景音乐生成

Zhang, Jincheng, Fazekas, György, Saitis, Charalampos

Abstract

Diffusion models have been widely used for various generative tasks. However, recent studies have demonstrated that the internal representations learned in the diffusion models' denoising process are insufficiently meaningful, which can hinder the generation performance. To address this, we propose a novel diffusion U-Net architecture enhanced by explicitly aligning the weaker representation from the earlier layer conditioned on higher noise with the better representation from the later layer conditioned on lower noise during training, which we use for the challenging multimodal task of video background music generation. Our diffusion U-Net Self-Representation Alignment (U-SRA) employs an alignment loss to effectively guide our diffusion model to learn good internal representations, thereby enabling it to produce higher-quality symbolic music conditioned on video features. We employ a cross-attention mechanism to synchronize music with video segments. Experiments on a benchmark video-music dataset demonstrate that the proposed U-SRA approach for video background music generation surpasses the state-of-the-art across key evaluation metrics, including video-music alignment and overall music quality.

Chinese Translation

扩散模型已广泛应用于各种生成任务。然而，近期研究表明，扩散模型去噪过程中学习到的内部表征缺乏足够的意义，这可能会阻碍生成性能。为此，我们提出了一种新颖的扩散 U-Net 架构，通过在训练过程中显式地将较早层中基于较高噪声的较弱表征与较晚层中基于较低噪声的较好表征对齐，从而增强模型性能，我们将其应用于具有挑战性的多模态任务——视频背景音乐生成。我们的扩散 U-Net 自我表征对齐（U-SRA）采用对齐损失，有效指导扩散模型学习良好的内部表征，从而使其能够生成基于视频特征的高质量符号音乐。我们采用跨注意力机制来同步音乐与视频片段。在基准视频音乐数据集上的实验表明，所提出的 U-SRA 方法在视频背景音乐生成方面超越了现有最先进的技术，在视频音乐对齐和整体音乐质量等关键评估指标上表现优异。

View on arXiv Download PDF AI Translation

cs.AI / 123 / 2610.15582

Inverse Garment Design with Visual Agents

基于视觉智能体的逆服装设计

Jiang, Nanxiang, Chen, Yunuo, Tao, Zeng, Li, Zizhang, Guo, Qi, Wu, Jiajun, Yang, Yin, Jiang, Chenfanfu

Abstract

We study the problem of inverse garment design, reconstructing simulation-ready 3D outfits with arbitrary layers from a single image, with visual agents. Unlike differentiable optimization that is time-consuming, or feed-forward methods limited by training data, we introduce DressAgent, a hierarchical agentic system that recovers garment sewing patterns through planning and closed-loop refinement grounded in physical simulation. To this end, we first present an enhanced representation based on GarmentCode for diverse styles and arbitrary dressing layers, and build the first multi-layer sewing pattern dataset with over 15K outfits for supervision and benchmarking. Inverse design then proceeds coarse to fine: In layer planning and coarse estimation, agents infer layer order and relations, peel each layer into its own view, and estimate coarse patterns. In fine-grained refinement, an agentic trio of Tailor, Critic, and Orchestrator refines patterns in a closed loop of editing, simulating, and judging against the reference with visual and measured evidence. Carefully designed tools and an evolving memory make this refinement converge on the target geometry. We also design an agentic generation method for high-fidelity textures. DressAgent outperforms feed-forward methods by a large margin, surpasses or matches differentiable optimization in a fraction of the time, and generalizes to complex layering, diverse poses, and body shapes.

Chinese Translation

我们研究了逆服装设计的问题，即从单幅图像重建具有任意层次的可模拟3D服装，利用视觉智能体。与耗时的可微优化方法或受训练数据限制的前馈方法不同，我们引入了DressAgent，一个分层智能系统，通过基于物理模拟的规划和闭环优化来恢复服装缝制模式。为此，我们首先提出了一种基于GarmentCode的增强表示，以适应多样的风格和任意的穿着层次，并构建了第一个包含超过15,000套服装的多层缝制模式数据集，用于监督和基准测试。逆设计过程从粗到细进行：在层次规划和粗略估计中，智能体推断层次顺序和关系，将每一层剥离到其自身视图，并估计粗略模式。在细粒度优化中，由裁缝（Tailor）、评论者（Critic）和协调者（Orchestrator）组成的智能三人组在编辑、模拟和与参考进行判断的闭环中优化模式，利用视觉和测量证据。精心设计的工具和不断演化的记忆使得这种优化收敛于目标几何形状。我们还设计了一种用于高保真纹理生成的智能方法。DressAgent在性能上大幅超越前馈方法，在极短时间内超过或匹配可微优化，并能够推广到复杂的层次结构、多样的姿态和体型。

View on arXiv Download PDF AI Translation

cs.AI / 124 / 2610.15603

CondMSP: Condition-Informed and Resource-Aware Inorganic Materials Synthesis Planning

CondMSP：条件知情和资源感知的无机材料合成规划

Noh, Heewoong, Na, Gyoung S., Pan, Elton, Lee, Namkyeong, Park, Chanyoung

Abstract

Materials synthesis planning (MSP) connects AI-driven materials discovery to experimental realization by proposing precursors, synthesis operations, and processing conditions for a target material. Existing approaches predict these components separately or in pairs and treat conditions only as a prediction target. Consequently, MSP remains centered on recipe prediction: it leaves unexplored how conditions can serve as process context for constructing synthesis routes, and it does not address which candidate routes should be executed under limited experimental resources. We propose CondMSP, which extends MSP from recipe prediction to execution-aware planning. CondMSP predicts temperature, time, and atmosphere by combining a joint predictor with variable-specific experts, and then uses these predictions both as process context, grounding the generation and retrieval of operation sequences, and as execution requirements, specifying which thermal stages can share a furnace run and at what cost. This specification is held fixed across all compared methods; within it, CondMSP ranks routes across precursor sets by a common utility and, under furnace budgets, jointly selects routes and allocates their thermal stages by constrained optimization. CondMSP improves condition and operation prediction and, under this shared constraint, raises reference-recipe recovery from 27.9% to 32.9% at three simulated furnace runs per target and from 34.7% to 40.0% at 30 trials, while a new evaluation set of 2,344 targets supports generalization beyond oxide-dominated training chemistry.

Chinese Translation

材料合成规划（MSP）通过为目标材料提出前驱体、合成操作和处理条件，将基于人工智能的材料发现与实验实现连接起来。现有方法分别或成对预测这些组件，并仅将条件视为预测目标。因此，MSP仍然集中于配方预测：它未探讨条件如何作为构建合成路线的过程背景，也未解决在有限实验资源下应执行哪些候选路线。我们提出了CondMSP，它将MSP从配方预测扩展到执行感知规划。CondMSP通过结合联合预测器与特定变量的专家来预测温度、时间和气氛，然后将这些预测作为过程背景，支持操作序列的生成和检索，同时作为执行要求，指定哪些热阶段可以共享炉子运行及其成本。该规范在所有比较方法中保持不变；在此框架内，CondMSP通过共同效用在前驱体集之间对路线进行排名，并在炉子预算下，通过约束优化共同选择路线并分配其热阶段。CondMSP改善了条件和操作预测，并在这一共享约束下，将参考配方的恢复率从27.9%提高到32.9%（每个目标三个模拟炉运行）以及从34.7%提高到40.0%（30次试验），同时一个新的评估集包含2,344个目标，支持超越以氧化物为主的训练化学的泛化。

View on arXiv Download PDF AI Translation

cs.AI / 125 / 2610.15620

From Stages to States: Unified Iterative Reasoning for MLLMs

从阶段到状态：多模态大语言模型的统一迭代推理

Yu, Chuang, Ma, Xuefei, Feng, Yuanhao, Zeng, Wuxiong, Wang, Bo, Cai, Minghong, Wu, Dongming, Zhou, Fei, Tang, Chen, Tong, Wenwen, Lu, Lewei, Yue, Xiangyu

Abstract

Multimodal Chain-of-Thought (MCoT) substantially enhances the complex reasoning capabilities of Multimodal Large Language Models (MLLMs) by explicitly modeling intermediate reasoning processes. However, most learning-based MCoT methods follow a stage-centric paradigm, where different reasoning behaviors are assigned to predefined stages and optimized separately. This results in fragmented reasoning and a training-inference reasoning-state gap. To address these limitations, we propose ReState, a unified iterative reasoning framework that reformulates multimodal reasoning from a stage-driven pipeline into a state-driven dynamic process. It comprises two complementary components: Unified Reasoning State Modeling (URSM) and Closed-Loop Reasoning-State Learning (CRSL). URSM unifies reasoning construction, correct reasoning preservation, and erroneous reasoning correction as reasoning-state transitions within a shared state space, enabling these heterogeneous reasoning behaviors to be learned jointly. CRSL further incorporates self-generated historical reasoning states during training and recursively updates the current state during inference, thereby mitigating the reasoning-state gap and enabling iterative self-refinement without requiring an additional reflection module, critic, or independent correction stage. Furthermore, we construct VersaCoT-10K, a high-quality CoT dataset spanning eight visual reasoning domains, which provides diverse supervision for multi-domain reasoning and reasoning-state refinement. Extensive experiments demonstrate that ReState exhibits strong adaptability across different MLLM backbones and achieves strong performance on multiple public reasoning benchmarks, supporting the shift from stage-centric reasoning toward state-centric reasoning. Our data and code are available at https://github.com/YuChuang1205/ReState

Chinese Translation

多模态思维链（MCoT）通过明确建模中间推理过程，显著增强了多模态大语言模型（MLLMs）的复杂推理能力。然而，大多数基于学习的MCoT方法遵循以阶段为中心的范式，其中不同的推理行为被分配到预定义的阶段并单独优化。这导致了推理的碎片化以及训练与推理之间的推理状态差距。为了解决这些局限性，我们提出了ReState，一个统一的迭代推理框架，它将多模态推理从以阶段驱动的管道重新构建为以状态驱动的动态过程。该框架包含两个互补组件：统一推理状态建模（URSM）和闭环推理状态学习（CRSL）。URSM将推理构建、正确推理的保留和错误推理的纠正统一为共享状态空间内的推理状态转变，使得这些异质推理行为能够共同学习。CRSL进一步在训练过程中纳入自生成的历史推理状态，并在推理过程中递归更新当前状态，从而减轻推理状态差距，并实现迭代自我精炼，而无需额外的反思模块、评估者或独立的纠正阶段。此外，我们构建了VersaCoT-10K，一个涵盖八个视觉推理领域的高质量思维链数据集，为多领域推理和推理状态精炼提供了多样化的监督。广泛的实验表明，ReState在不同的MLLM骨干网络上表现出强大的适应性，并在多个公共推理基准上取得了优异的性能，支持从以阶段为中心的推理向以状态为中心的推理的转变。我们的数据和代码可在https://github.com/YuChuang1205/ReState获取。

View on arXiv Download PDF AI Translation

cs.AI / 126 / 2610.15634

Big-Big-SycoBench: A Benchmark to Detect Sycophancy Through Model Responses to Valid vs. Invalid Arguments

Big-Big-SycoBench：通过模型对有效与无效论点的响应来检测谄媚行为的基准测试

Ballon, Marthe, Vanhoyweghen, Arne, Kim, Seorin, Carlon, Francesca, Delcon, Lynn, Mobini, Melika, Holst, Vincent, Van de Voorde, Lukas, Vanleke, Tibo, Louis, Mathieu, Verbruggen, Bert, Verbeken, Brecht F., Algaba, Andres, Ginis, Vincent

Abstract

Sycophancy in large language models risks amplifying unsupported claims or beliefs across the knowledge ecosystem as these systems increasingly contribute to research synthesis, evaluation, and decision-making. A reliable model should revise its judgement when a user identifies a genuine flaw, but should resist pushback based on fallacies, emotion, or irrelevant appeals to authority. While sycophancy is usually measured on questions with a verifiable answer, the evaluative tasks in which it is most difficult to detect have no such ground truth. We introduce Big-Big-SycoBench, which measures sycophancy through argument discernment: the ability to revise an evaluation after valid criticism while resisting invalid pressure. The benchmark consists of 450 text artefacts with 5,400 arguments for raising or lowering their quality scores, each reviewed by one or more of twelve human annotators. An LLM scores each artefact twenty times to establish its baseline score and uncertainty. The model is then given a valid or invalid argument and asked to update its score. We only count a revision if it exceeds the model's own baseline variability. We use the resulting score delta to distinguish responsiveness to valid arguments from compliance with invalid ones. Across ten models, responsiveness is 88-95\% for seven, 76-77\% for Kimi K3 and GLM-5.3, and 32\% for Opus 5.5. Compliance ranges from 0.2\% for GPT-6 Astra to 95\% for GPT-4o. Big-Big-SycoBench is intended to guide the development of models whose judgements can be corrected by sound arguments, but not displaced by unsupported pressure.

Chinese Translation

大型语言模型中的谄媚行为可能会在知识生态系统中放大不支持的主张或信念，因为这些系统越来越多地参与研究综合、评估和决策。一个可靠的模型应该在用户识别出真实缺陷时修正其判断，但应抵制基于谬误、情感或无关权威的反对。虽然谄媚行为通常在有可验证答案的问题上进行测量，但在最难以检测的评估任务中并没有这样的真实依据。我们引入了Big-Big-SycoBench，它通过论点辨识来测量谄媚行为：在有效批评后修正评估的能力，同时抵制无效压力。该基准测试由450个文本材料和5400个论点组成，用于提高或降低其质量评分，每个材料由12名人类注释者中的一名或多名进行审查。一个大型语言模型（LLM）对每个材料进行20次评分，以建立其基线分数和不确定性。然后，模型会收到一个有效或无效的论点，并被要求更新其分数。只有当修正超过模型自身的基线变异性时，我们才会计算该修正。我们使用得到的分数差异来区分对有效论点的响应与对无效论点的顺从。在十个模型中，七个模型的响应率为88-95%，Kimi K3和GLM-5.3的响应率为76-77%，而Opus 5.5的响应率为32%。顺从率从GPT-6 Astra的0.2%到GPT-4o的95%不等。Big-Big-SycoBench旨在指导模型的开发，使其判断能够通过合理的论点进行修正，而不会被不支持的压力所取代。

View on arXiv Download PDF AI Translation

cs.AI / 127 / 2610.15642

Recall What Is Missing: Composing Evidence from Long-Term Agent Memory

回忆缺失的内容：从长期智能体记忆中组合证据

Feng, Qi, Fan, Jicong

Abstract

Under a limited context budget, a long-term agent must recall not merely memories that are individually relevant, but evidence that is collectively sufficient for the current query. Conventional memory access retrieves, scores, and truncates memories, often spending multiple context slots on redundant facts while leaving complementary evidence behind. We cast the read phase of long-term memory as query-conditioned evidence composition: each memory is valued by what it adds beyond the evidence already selected. We instantiate this operation with MemPursuit, a training-free mechanism that uses the residual of a sparse convex query reconstruction as a representation-space proxy for what the current memory set is missing. By reusing the query and candidate embeddings already available from retrieval, MemPursuit progressively composes nested evidence contexts without invoking a language model or cross-encoder to score each query--memory pair. The effectiveness of our method is verified on 3 popular benchmarks in comparison to strong baselines. In fixed-budget LoCoMo experiments, MemPursuit improves end-to-end accuracy by 7.2 percentage points at Top-10 and 5.5 percentage points at Top-20, with consistent gains across multiple memory frameworks, while achieving a 32.9$\times$ speedup in median reranking latency over fully batched BGE-v2-m3. Our implementation is available at https://github.com/FengQi-HITSZ/MemPursuit.

Chinese Translation

在有限的上下文预算下，长期智能体必须回忆的不仅仅是单独相关的记忆，而是对于当前查询来说集体充分的证据。传统的记忆访问方法检索、评分并截断记忆，常常在冗余事实上花费多个上下文槽，同时留下互补证据。我们将长期记忆的读取阶段视为基于查询的证据组合：每个记忆的价值在于它所增加的证据超出已选择的证据。我们通过 MemPursuit 实现这一操作，这是一种无训练机制，利用稀疏凸查询重构的残差作为当前记忆集缺失内容的表示空间代理。通过重用从检索中已获得的查询和候选嵌入，MemPursuit 逐步组合嵌套的证据上下文，而无需调用语言模型或交叉编码器来评分每个查询-记忆对。我们的方法在与强基线的比较中，在 3 个流行基准上验证了其有效性。在固定预算的 LoCoMo 实验中，MemPursuit 在 Top-10 上提高了 7.2 个百分点的端到端准确率，在 Top-20 上提高了 5.5 个百分点，并且在多个记忆框架中实现了一致的增益，同时在中位数重新排序延迟上比完全批处理的 BGE-v2-m3 提高了 32.9 倍的速度。我们的实现可以在 https://github.com/FengQi-HITSZ/MemPursuit 获得。

View on arXiv Download PDF AI Translation

cs.AI / 128 / 2610.15677

Knowledge Interaction in Hypernetwork-based Continual Learning: Conflict and Composition

基于超网络的持续学习中的知识交互：冲突与组合

Zheng, Jiaqi, Hu, Yi, Mu, Yadong, Zhang, Muhan

Abstract

Continual learning aims to adapt pretrained large language models (LLMs) to evolving data while preserving their pretrained capabilities. A central challenge is catastrophic forgetting: adaptation to new data can impair performance on previously learned knowledge. Since continual learning commonly changes model parameters, the model's predictions may draw on knowledge from the base model, the parameter update, and the input context. Understanding how these sources interact is essential: conflicts reveal whether new knowledge can override outdated knowledge, while multi-hop reasoning tests whether new knowledge can be composed with existing knowledge to support generalization. We study these interactions in a hypernetwork-based adaptation framework, where a shared network generates adapters from context, allowing a frozen base model to adapt without task-specific updates. We find that, under knowledge conflict, models preferentially follow the context, and their choices depend more on where knowledge is presented than on its content. We also show that context and adapter knowledge perturb the base model's representations along directions orthogonal to its clean hidden states, enabling us to derive steering vectors that control which source the model follows. In multi-hop reasoning, we find that models fail to reliably compose knowledge across the three sources, with the key breakdown arising from the interaction of the hypernetwork with both the base model and the context. Code and reproduction materials: https://github.com/missswiftie/Knowledge-Interaction-in-CL

Chinese Translation

持续学习旨在使预训练的大型语言模型（LLMs）适应不断变化的数据，同时保留其预训练能力。一个核心挑战是灾难性遗忘：对新数据的适应可能会损害对先前学习知识的表现。由于持续学习通常会改变模型参数，因此模型的预测可能会依赖于基础模型、参数更新和输入上下文中的知识。理解这些来源如何相互作用至关重要：冲突揭示了新知识是否能够覆盖过时知识，而多跳推理则测试新知识是否能够与现有知识组合以支持泛化。我们在基于超网络的适应框架中研究这些交互，其中共享网络根据上下文生成适配器，使得冻结的基础模型能够在不进行任务特定更新的情况下进行适应。我们发现，在知识冲突下，模型优先遵循上下文，其选择更多地依赖于知识呈现的位置而非内容。我们还表明，上下文和适配器知识沿着与基础模型的干净隐藏状态正交的方向扰动基础模型的表示，使我们能够推导出控制模型遵循哪个来源的引导向量。在多跳推理中，我们发现模型未能可靠地跨三个来源组合知识，关键的故障源于超网络与基础模型和上下文之间的交互。代码和复现材料： https://github.com/missswiftie/Knowledge-Interaction-in-CL

View on arXiv Download PDF AI Translation

cs.AI / 129 / 2610.15685

Radioactive Feedback: Tracing Models Trained by Label-Only LLM Supervision

放射性反馈：追踪仅通过标签监督训练的模型

Li, Yuejia, Li, JunHeng, Wu, Hao, He, Ke, Ye, Mang

Abstract

Conventional distillation transfers capability through teacher-generated demonstrations. LLM-as-a-Judge knowledge distillation offers a distinct and increasingly practical alternative. In the label-only variant studied here, the student generates every response and improves from scalar scores alone. Prior work shows that this non-demonstrative feedback can improve student capability, but it creates a fundamentally different provenance problem. No teacher-generated answer enters the student's training set, so any evaluator-specific trace must pass through a low-bandwidth reward signal rather than copied text. Existing watermarks and distillation detectors do not cover this score-only channel. We introduce Radioactive Feedback, which encodes a registered pattern of behavioral preferences in the rewards themselves. The evaluator applies bounded pointwise adjustments to quality-qualified responses, allowing repeated optimization to transfer the pattern to a student trained by another party. Experiments on code and mathematics show that the registered preferences transfer through scores alone and across datasets, enabling black-box detection against a fixed reference pool.

Chinese Translation

传统的蒸馏通过教师生成的示范转移能力。作为评估者的LLM知识蒸馏提供了一种独特且日益实用的替代方案。在这里研究的仅标签变体中，学生生成每个响应，并仅通过标量评分进行改进。先前的研究表明，这种非示范反馈可以提高学生的能力，但它产生了根本不同的来源问题。没有教师生成的答案进入学生的训练集，因此任何特定于评估者的追踪必须通过低带宽的奖励信号而不是复制文本来传递。现有的水印和蒸馏检测器并未覆盖这一仅评分的通道。我们引入了放射性反馈，它在奖励中编码了注册的行为偏好模式。评估者对质量合格的响应应用有界的逐点调整，允许重复优化将该模式转移到由其他方训练的学生。对代码和数学的实验表明，注册的偏好仅通过评分在不同数据集间转移，从而实现对固定参考池的黑箱检测。

View on arXiv Download PDF AI Translation

cs.AI / 130 / 2610.15691

FARL: A Fully Automated Pipeline of RL Interface and Solver Design

FARL：一种完全自动化的强化学习接口和求解器设计管道

Jin, Tianyu, Xiang, Liuyu, Yang, Chaowen, Chang, Xingyang, Li, Pei Pei, He, Zhaofeng

Abstract

Existing reinforcement learning (RL) largely focuses on optimizing learning algorithms within a pre-defined task formulation, while applying RL to new tasks still relies heavily on human expertise to design the learning pipeline: what an agent observes, how it is rewarded, which algorithm it uses, and how training is configured. Automating these choices is challenging because they are tightly coupled. We introduce FARL (Fully Automated Reinforcement Learning), a framework that automates RL design by jointly searching observation and reward programs, learning algorithms, and training settings. FARL uses training statistics and observed behavior to diagnose failures, propose changes, and test whether they improve learning. Its search preserves diverse candidates and the controls needed for these tests, using a staged budget to screen designs before investing in longer training. Experimental memory records each finding together with its conditions, so an unsuccessful component can be reconsidered in a different combination. Across expert-withheld and expert-available settings, FARL outperforms the evaluated automated baselines on challenging navigation, locomotion, and manipulation tasks. On Adroit Hand Relocate, FARL achieves 97.1% success at 5M training steps per policy, compared with 0% for Eureka and LiMEN and 1.0% for native Proximal Policy Optimization (PPO) under the same policy training budget.

Chinese Translation

现有的强化学习（RL）主要集中在优化预定义任务框架内的学习算法，而将RL应用于新任务仍然在很大程度上依赖于人类专家来设计学习管道：代理观察到的内容、如何获得奖励、使用的算法以及训练的配置。自动化这些选择具有挑战性，因为它们紧密耦合。我们提出了FARL（完全自动化强化学习），这是一个通过联合搜索观察和奖励程序、学习算法以及训练设置来自动化RL设计的框架。FARL利用训练统计数据和观察到的行为来诊断失败、提出变更，并测试这些变更是否改善学习。其搜索保留了多样化的候选项和进行这些测试所需的控制，使用分阶段预算在投入更长时间的训练之前筛选设计。实验记录每个发现及其条件，因此不成功的组件可以在不同组合中重新考虑。在专家保留和专家可用的设置中，FARL在具有挑战性的导航、运动和操作任务上超越了评估的自动化基线。在Adroit Hand Relocate任务中，FARL在每个策略5M训练步骤下实现了97.1%的成功率，而Eureka和LiMEN的成功率为0%，原生的近端策略优化（PPO）在相同的策略训练预算下的成功率为1.0%。

View on arXiv Download PDF AI Translation

cs.AI / 131 / 2610.15708

SpecForge: Recursive Self-Improvement for Coding Agents via Behavioral Specifications

SpecForge：通过行为规范实现编码代理的递归自我改进

Cheng, Yutong, Chen, Haifeng, Gao, Peng, Cheng, Wei

Abstract

Coding agents based on frozen language models can improve their performance by refining the external context they rely on, without updating model weights. We study recursive self-improvement through behavioral specifications: natural-language code documentation that describes the behavior of program entities and guides agents in generating correct code. We formulate a self-improvement loop in which an agent generates code from its current specifications, evaluates it with execution-verified tests, and refines the specifications based on observed failures. The improved specifications then guide subsequent generation, progressively improving the agent's understanding of the codebase. However, recursive refinement is challenging because program entities are interdependent. Improving the specification of one entity can inadvertently degrade the correctness of others, a phenomenon we call output coupling. This creates a whack-a-mole effect in which local improvements cause regressions elsewhere, preventing consistent progress. We introduce SpecForge, a multi-frontier tree search framework that addresses this challenge with three mechanisms: (i) Pareto-guided state selection preserves complementary specification states and prevents premature commitment; (ii) dependency-constrained entity selection prioritizes callee-before-caller refinement; and (iii) error-diversified expansion explores multiple refinements using different subsets of test failures. On DevEval+, SpecForge achieves the best performance across five backbone models, including a 94.9% solve rate with Claude-4.5-Sonnet and a 5.7-percentage-point average improvement over the strongest baseline. The improved specifications also make the agent better at tasks outside the loop: they transfer to cross-language translation and to new-feature implementation, where they outperform the benchmark-provided requirement documents.

Chinese Translation

基于冻结语言模型的编码代理可以通过优化其依赖的外部上下文来提高性能，而无需更新模型权重。我们研究了通过行为规范实现的递归自我改进：自然语言代码文档描述程序实体的行为，并指导代理生成正确的代码。我们制定了一个自我改进循环，其中代理根据当前规范生成代码，通过执行验证的测试进行评估，并根据观察到的失败来优化规范。改进后的规范随后指导后续的生成，逐步提升代理对代码库的理解。然而，递归优化是具有挑战性的，因为程序实体之间是相互依赖的。改善一个实体的规范可能会无意中降低其他实体的正确性，这种现象我们称之为输出耦合。这导致了一种打地鼠效应，即局部改进会在其他地方造成回归，阻碍了一致的进展。我们引入了SpecForge，一个多前沿树搜索框架，通过三种机制来应对这一挑战：（i）帕累托引导的状态选择保持互补的规范状态，防止过早承诺；（ii）依赖约束的实体选择优先考虑被调用者在调用者之前的优化；（iii）错误多样化扩展使用不同的测试失败子集探索多种优化。在DevEval+上，SpecForge在五个基础模型中实现了最佳性能，包括Claude-4.5-Sonnet的94.9%解决率，以及比最强基线平均提高5.7个百分点。改进后的规范还使代理在循环外的任务中表现更好：它们能够迁移到跨语言翻译和新特性实现中，在这些任务中超越基准提供的需求文档。

View on arXiv Download PDF AI Translation

cs.AI / 132 / 2610.15710

Suggested Is Not Superseded: A Pre-Registered Evaluation of a System-One Model as a Memory-Retirement Gate

建议并未被取代：对系统一模型作为记忆退休门的预注册评估

Chen, Xupeng, Le, Chenqian

Abstract

Long-term memory for assistants needs a gate that decides when a new utterance retires an old stored claim. Benchmark labels for this step mark a pair as changed or not, so they cannot tell a gate that acts on established change from one that acts on a hint. We write a task contract that separates the two (SUPERSEDED vs. NEEDS-CONFIRMATION vs. COMPATIBLE), label 500 test pairs built from STALE (change pairs plus same-user controls) with three LLM annotators and two human adjudicators (one author and a second annotator; provenance reported), and evaluate Jev, a typed "System One" decision model, under pre-registered hypotheses. Of the 179 test pairs STALE marks as changed, 72 meet the contract's bar for an established change. Jev never retired a claim on 141 same-conversation non-changes. Its failure is premature retirement: the binary gate retires 39.2% of claims whose change is only suggested. A three-way decision cuts this to 15.5% at a 15.3-point detection cost, which misses our pre-registered 10-point budget, and it lands on the binary gate's own threshold frontier; these verdicts hold under three readings of the gold labels. On a held-out vault scored against a second annotator's labels, a cascade that escalated uncertain cases to a reasoning model and looked best on development data was indistinguishable from simply raising the threshold (premature retirement 26.8% vs. 6.5% on dev). Part of the gap comes from the gold labels: relative to that annotator's reading, LLM-majority gold shares most of its errors with a same-family escalation model, and an identity for binary labels says how much such shared errors hide. The rest is small-sample selection on development data. For builders this means a binary score should not be wired to delete: route suggested changes to pending evidence, and do not trust a cascade selected on development data of this size.

Chinese Translation

长期记忆对于助手需要一个门控机制，以决定何时新的发言可以取代旧的存储声明。该步骤的基准标签将一对标记为已更改或未更改，因此无法区分作用于已建立变化的门与作用于提示的门。我们编写了一个任务合同，将两者分开（已取代 vs. 需要确认 vs. 兼容），并使用三位大型语言模型（LLM）注释员和两位人类裁判（包括一位作者和第二位注释员；来源已报告）对从 STALE（变化对加上同用户控制）构建的 500 个测试对进行标注，并在预注册假设下评估了 Jev，一个类型化的“系统一”决策模型。在 STALE 标记为已更改的 179 个测试对中，有 72 个符合合同对已建立变化的标准。Jev 从未对 141 个同一对话中的非变化声明进行退休。其失败在于过早退休：该二元门控机制对 39.2% 仅被建议变化的声明进行了退休。三元决策将这一比例降低到 15.5%，但检测成本为 15.3 点，未能满足我们预注册的 10 点预算，并且落在了二元门控机制自身的阈值边界上；这些裁决在对金标签的三次解读下保持一致。在与第二位注释员的标签进行评分的保留库中，一种将不确定案例升级到推理模型的级联方法，在开发数据上表现最佳，与简单提高阈值的效果无差别（过早退休 26.8% vs. 6.5% 在开发数据上）。差距的一部分来自金标签：相对于该注释员的解读，LLM 多数金标签与同家族的升级模型共享了大部分错误，而二元标签的身份则说明了这些共享错误隐藏的程度。其余部分则是开发数据上的小样本选择。对于构建者而言，这意味着二元评分不应直接用于删除：将建议的变化路由到待处理证据，并且不要信任在如此规模的开发数据上选择的级联。

View on arXiv Download PDF AI Translation

cs.AI / 133 / 2610.15714

MedHarnessBench: A Unified Benchmark for Healthcare Agent Evaluation

MedHarnessBench：医疗代理评估的统一基准

Tang, Rui, Wang, Hongqiu

Abstract

Healthcare agents must transform longitudinal clinical evidence into reliable actions through heterogeneous tools and stateful workflows, conditions under which model-centric systems may lose context, misuse tools, fail to verify consequential transitions, or cross authorization boundaries. Existing medical-agent benchmarks primarily emphasize terminal task outcomes, making it difficult to determine whether execution failures arise from the agent model, the runtime harness, or their interaction. We present the Medical Harness Protocol, an outcome-preserving framework that independently evaluates seven healthcare-specific process dimensions: Care Action Fulfillment, Clinical Resource Competence, Patient-Evidence Alignment, Clinical Workflow Continuity, Clinical Auditability, Closed-Loop Verification, and Clinical Conduct Boundaries. A normalized intervention trace further tracks whether runtime support is available, activated, delivered, adopted by the agent, and propagated into subsequent behavior. We instantiate the protocol as MedHarnessBench, comprising 342 executable tasks across FHIR/EHR operations, healthcare-administration interfaces, and multimodal clinical tools. Across more than 16,000 controlled task executions, MedHarness yields broad improvements in terminal task success, intermediate subtask completion, and healthcare-specific process quality. We further evaluate MedHarness against multiple general-purpose harnesses and use controlled single-fault perturbations to verify the sensitivity and discrimination of the seven-dimensional evaluation. These results demonstrate that healthcare-specific runtime support can substantially improve both task completion and execution quality, while revealing failures that terminal outcomes alone cannot capture. Code and resources are available at https://github.com/Tomry1114/Awesome-Agent-Harness-for-Healthcare.

Chinese Translation

医疗代理必须通过异构工具和状态工作流将纵向临床证据转化为可靠的行动，在这种情况下，以模型为中心的系统可能会失去上下文、误用工具、未能验证重要的过渡或越过授权边界。现有的医疗代理基准主要强调终端任务结果，这使得很难判断执行失败是源于代理模型、运行时工具，还是它们之间的交互。我们提出了医疗工具协议（Medical Harness Protocol），这是一个保持结果的框架，独立评估七个特定于医疗的过程维度：护理行动履行、临床资源能力、患者-证据对齐、临床工作流连续性、临床审计能力、闭环验证和临床行为边界。一个标准化的干预跟踪进一步追踪运行时支持是否可用、是否被激活、是否被交付、是否被代理采用，以及是否传播到后续行为中。我们将该协议实例化为MedHarnessBench，包含342个可执行任务，涵盖FHIR/EHR操作、医疗管理接口和多模态临床工具。在超过16,000次受控任务执行中，MedHarness在终端任务成功率、中间子任务完成率和医疗特定过程质量方面均取得了显著改善。我们进一步将MedHarness与多个通用工具进行比较，并使用受控的单故障扰动来验证七维评估的敏感性和区分能力。这些结果表明，特定于医疗的运行时支持可以显著提高任务完成率和执行质量，同时揭示终端结果无法捕捉的失败。代码和资源可在 https://github.com/Tomry1114/Awesome-Agent-Harness-for-Healthcare 获取。

View on arXiv Download PDF AI Translation

cs.AI / 134 / 2610.15715

History Replays Itself: Unleashing Long-Context Capability in Linear Attention

历史重演：释放线性注意力中的长上下文能力

Nanbo, Li, Laakom, Firas, Zhang, Yilan, Piękos, Piotr, Chen, Yimeng, Wang, Liangyu, Xu, Yifan, Ostaszewski, Mateusz, Guo, Dandan, Wang, Wenyi, Dai, Yanning, Wang, Yuhui, Schmidhuber, Jürgen

Abstract

Linear attention promises the efficiency needed for long-context language modelling, yet its fixed-size recurrent state must simultaneously support ongoing computation and preserve information from an ever-growing history. This coupling has reinforced a prevailing view that the state's memory capacity defines the model's long-context capability. We challenge this presumed ceiling: \emph{recurrent computation need not bear the full burden of long-term storage}. We introduce NoteBank, a sparse episodic memory that archives selected text before future queries are known, uses the backbone's native features for writing and addressing, and reads retrieved evidence through its native forward computation via replay. This \emph{training-free, plug-and-play} interface requires no external embedding model, separately trained retriever, or separate neural reader. At one million tokens, it raises Gated DeltaNet's mean accuracy across six RULER retrieval tasks from 1.7% to 70.3%, with no observed significant retrieval degradation from 4K to 1M tokens. This recovery adds $\leqslant$0.025% to analytical FLOPs, with persistent context storage approximately 1/2,600 that of a Transformer's full MHA KV cache at matched width, depth, and precision. Experiments extend from three 1.3B recurrent backbones to Mamba-2-8B and post-trained Qwen3.5-9B, demonstrating gains on LongBench v1/v2 and MemoryAgentBench. Together, these results show that sparse evidence storage and native replay can substantially expand the long-context capability of pretrained linear attention models while retaining the efficiency of recurrent computation.

Chinese Translation

线性注意力为长上下文语言建模提供了所需的效率，然而其固定大小的递归状态必须同时支持持续计算并保留来自不断增长的历史的信息。这种耦合强化了一个普遍观点，即状态的记忆容量定义了模型的长上下文能力。我们挑战这一假定的上限： extit{递归计算不必承担长期存储的全部负担}。我们引入了NoteBank，这是一种稀疏的情节记忆，在未来查询已知之前归档选择的文本，利用骨干网络的原生特征进行写入和寻址，并通过其原生的前向计算进行回放以读取检索到的证据。这种 extit{无训练、即插即用}的接口不需要外部嵌入模型、单独训练的检索器或单独的神经阅读器。在一百万个标记的情况下，它将Gated DeltaNet在六个RULER检索任务上的平均准确率从1.7%提高到70.3%，且在4K到1M标记之间未观察到显著的检索性能下降。这种恢复对分析FLOPs的增加不超过0.025%，而持久上下文存储的大小约为与宽度、深度和精度匹配的Transformer的完整MHA KV缓存的1/2,600。实验从三个1.3B的递归骨干扩展到Mamba-2-8B和后训练的Qwen3.5-9B，展示了在LongBench v1/v2和MemoryAgentBench上的增益。这些结果表明，稀疏证据存储和原生回放可以显著扩展预训练线性注意力模型的长上下文能力，同时保持递归计算的效率。

View on arXiv Download PDF AI Translation

cs.AI / 135 / 2610.15738

Visible but Not Avoided: State-Dependent Re-Proposal of Rejected Candidates in Catalog-Based LLM Optimization Loops

可见但不可避免：基于状态的重新提议被拒候选者在目录基础的 LLM 优化循环中的应用

Chen, Jianlin, Chen, Wenhui, Vong, Chi Man

Abstract

LLM optimization loops often show the proposer the candidates it has already tried and rejected, but leave it to the proposer to avoid them. We ask whether avoidance measured in a static probe predicts avoidance in the states an optimizer's own trajectory produces. In catalog-based loops where three open LLMs improve a GUI agent by naming one of 40 edits, a re-proposal is identified exactly by the edit's name. Clean base-state probes underpredict re-proposal in the proposers' own loops (macro-averaged loop-minus-probe rate 0.32 [0.27, 0.37]), and proposals of already evaluated harness-edit pairs consume 28-63% of valid proposer iterations. The two settings reject different edits: a probe's rejected edits are mostly ones the proposer did not choose, whereas a loop's are its own earlier choices, which remain preferred when the history is removed. Ordinary history lowers re-proposal for two proposers but raises it for the third, and the effects of failure evidence are model-specific. Stating the rejection status in one sentence lowers re-proposal in every model (-0.16 [-0.22, -0.09] relative to a neutral list of the same IDs). For catalog-based loops with exact edit identities, rejection avoidance should be measured on the rejected sets the loop produces.

Chinese Translation

LLM 优化循环通常向提议者展示其已尝试并拒绝的候选者，但让提议者自行避免这些候选者。我们探讨在静态探测中测量的避免行为是否能够预测优化器自身轨迹所产生的状态中的避免行为。在基于目录的循环中，三个开放的 LLM 通过命名 40 个编辑中的一个来改进 GUI 代理，重新提议的识别正是通过编辑的名称。干净的基础状态探测在提议者自己的循环中低估了重新提议（宏平均循环减去探测率为 0.32 [0.27, 0.37]），而已评估的编辑-提案对的提议消耗了 28-63% 的有效提议者迭代。两个设置拒绝不同的编辑：探测中被拒绝的编辑大多是提议者未选择的，而循环中的则是其早期选择，这些选择在历史被移除后仍然保持偏好。普通历史降低了两个提议者的重新提议，但提高了第三个提议者的重新提议，而失败证据的影响是模型特定的。在一句话中陈述拒绝状态降低了每个模型中的重新提议（相对于相同 ID 的中性列表，降低了 -0.16 [-0.22, -0.09]）。对于具有精确编辑身份的基于目录的循环，拒绝避免应在循环产生的被拒绝集合上进行测量。

View on arXiv Download PDF AI Translation

cs.AI / 136 / 2610.15750

Google Earth AI Geospatial Reasoning Agent: Capabilities and a Systematic Evaluation

谷歌地球人工智能地理空间推理代理：能力与系统评估

Slonim, Noam, Slobodkin, Aviv, Fidel, Gil, Moshitch, Gal, Sun, Mimi, Mihalkova, Lily, Paul, Joydeep, Bjornsson, Per, Lin, Sydney, McPike, Jamie, Roy, Samapriya, Patel, Utkarsh, Jung, Gia, Luna, Zachary, Berman, Allison, Cai, William, Wu, Xiao, Agarwal, Mohit, Sharma, Mandar, Hoffman, Aliza, Young, Mark, Pasumarthi, Rama, Lee, Roy, Alcantara, Ruth, Barrington, Luke, Bell, Aaron, Sayag, Reuven, Hassidim, Avinatan, Matias, Yossi, McHenry, Tyler, Ho, Ronald, Haire, Christopher, Guinan, Oliver, Efron, Niv, Thelin, Tim, Prasad, Gautam, Shetty, Shravya

Abstract

Google Earth AI is a family of geospatial AI models, together with an agent that orchestrates them. We present this agent, Google Earth AI's Geospatial Reasoning agent (GRA), a Gemini-powered research system that combines geospatial data sources, foundation models, and tools to answer questions about our planet. GRA organizes these into modular domain skills and pairs them with upfront geospatial planning to decompose complex queries into multi-stage workflows. Evaluating such systems is hard: real-world problems admit multiple valid workflows, require data discovery across large catalogs, and rely on evolving ground truth. We introduce two benchmarks and release their development splits. GeoExpertQueries poses complex analytical problems grounded in practitioner use cases and refined by geospatial experts; each is paired with a consensus golden workflow from three independent experts and scored by a human-validated, tool-agnostic autorater. GeoKTools consists of templated queries with deterministic reference solvers that systematically vary the number of required tool types. We also use two external benchmarks, CBGB and UnivEARTH; the latter we find to be saturated. To isolate the contribution of GRA's tools and its agentic orchestration, we compare four configurations, each with Gemini 3.1 Pro and 3.8 Flash: a standalone LLM, the LLM with basic tools, the LLM with GRA's skills and tools but without its upfront planning and execution guidance, and GRA itself. With at most basic tools, an advanced model recovers much of the answer on templated queries but satisfies only a small fraction of the expert workflows on open-ended problems; GRA performs well on both. The largest step comes from GRA's skills and tools; on open-ended problems, its planning and execution guidance add a further gain with Gemini 3.8 Flash, letting Flash match or exceed Pro at lower cost.

Chinese Translation

谷歌地球人工智能是一个地理空间人工智能模型的集合，配备一个协调这些模型的代理。我们介绍了这一代理，即谷歌地球人工智能的地理空间推理代理（Geospatial Reasoning Agent, GRA），这是一个基于Gemini的研究系统，结合了地理空间数据源、基础模型和工具，以回答有关我们星球的问题。GRA将这些内容组织成模块化的领域技能，并与前期的地理空间规划相结合，将复杂查询分解为多阶段工作流程。评估此类系统是困难的：现实世界问题允许多种有效的工作流程，需要在大型目录中进行数据发现，并依赖不断演变的真实情况。我们引入了两个基准并发布了它们的开发分割。GeoExpertQueries提出了基于从业者用例的复杂分析问题，并由地理空间专家进行精炼；每个问题都与三位独立专家的共识黄金工作流程配对，并由经过人工验证的、工具无关的自动评分器进行评分。GeoKTools由带有确定性参考求解器的模板查询组成，系统地变化所需工具类型的数量。我们还使用了两个外部基准，CBGB和UnivEARTH；后者我们发现已经饱和。为了隔离GRA工具及其代理协调的贡献，我们比较了四种配置，每种配置均使用Gemini 3.1 Pro和3.8 Flash：一个独立的LLM，带有基本工具的LLM，具有GRA的技能和工具但没有其前期规划和执行指导的LLM，以及GRA本身。在最多只有基本工具的情况下，先进模型在模板查询上恢复了大部分答案，但在开放式问题上仅满足少量专家工作流程；而GRA在两者上表现良好。最大的进步来自于GRA的技能和工具；在开放式问题上，其规划和执行指导在Gemini 3.8 Flash上进一步提升了性能，使Flash在更低成本下与Pro相匹配或超越。

View on arXiv Download PDF AI Translation

cs.AI / 137 / 2610.15788

Agent Sociodrama: Social-Dynamics-Guided Persona Recasting for LLM Agent Teams

代理社会剧：基于社会动态的角色重塑用于大型语言模型代理团队

Qi, Yirui, Zhang, Xiaoming, Yoon, Jaehong

Abstract

Large language model (LLM) agents are increasingly deployed as teams, yet they fail in characteristically social ways: information holders stay silent, dominant speakers anchor groups on wrong answers, and norm violations go unsanctioned. We present Agent Sociodrama, a framework for modeling and validating evolving social dynamics in LLM agent teams. Under fixed personas, agents' emotional states and directed, asymmetric relationships are updated every round by a unified external LLM judge and shape their subsequent interactions. Across six LLM backbones and five social scenarios, we assess social fidelity through classical social-theory patterns and agreement with human judgments. The resulting trajectories exhibit structured social patterns and model-specific differences. We then introduce Trajectory-Guided Persona Recasting (TPR), which diagnoses group-dynamic failures from interaction transcripts and social trajectories and recasts agent personas before rerunning the task. Across four collaborative benchmarks, TPR improves performance from both heterogeneous and randomly initialized agent persona compositions, yielding up to 83.3%p of accuracy improvement over the initial casting. On representative distributed-information cooperation and mixed-motive bargaining tasks, we outperform SOTOPIA and AgentVerse in three of four comparisons, by up to 37.5%p in Pareto rate. These results show that validated social trajectories provide actionable signals for repairing LLM agent teams.

Chinese Translation

大型语言模型（LLM）代理越来越多地作为团队部署，但它们在社交方面常常表现不佳：信息持有者保持沉默，主导发言者将团队引向错误答案，规范违反行为未受到制裁。我们提出了代理社会剧（Agent Sociodrama），这是一个用于建模和验证LLM代理团队中不断演变的社会动态的框架。在固定角色下，代理的情感状态和有向的非对称关系在每一轮中由统一的外部LLM评判者更新，并影响其后续的互动。在六种LLM基础模型和五种社会场景中，我们通过经典社会理论模式和与人类判断的一致性评估社会忠实度。结果轨迹展示了结构化的社会模式和模型特定的差异。随后，我们引入了轨迹引导的角色重塑（Trajectory-Guided Persona Recasting, TPR），该方法通过互动记录和社会轨迹诊断群体动态失败，并在重新运行任务之前重塑代理角色。在四个协作基准测试中，TPR提高了来自异构和随机初始化的代理角色组合的性能，准确率相比初始角色提升了最高达83.3个百分点。在代表性的分布式信息合作和混合动机谈判任务中，我们在四次比较中有三次超越了SOTOPIA和AgentVerse，帕累托率提高了最高达37.5个百分点。这些结果表明，经过验证的社会轨迹为修复LLM代理团队提供了可操作的信号。

View on arXiv Download PDF AI Translation

cs.AI / 138 / 2610.15797

Passing Tests, Missing Semantics: Evidence of Interpretive Lock-In in Plan-Based DataFrame Code Generation

通过测试，缺失语义：基于计划的数据框架代码生成中的解释锁定证据

Jingjing, Liu, Yang, Dishu, Qin, Hongbo, Zhang, Hansong

Abstract

Planning and revision are intended to bring generated code closer to the task specification. Yet a mistaken interpretation can persist across the plan, implementation, and tests - a pattern we call interpretive lock-in. We investigate this risk on 30 DataFrame tasks derived from TPC Benchmark DS, comparing Frozen One-Pass, Test-Guided Revision, and PlanAudit-Guided Revision. Test-pass counts rose from 22/30 to 29/30 and 30/30, yet no generated solution matched its stored reference output. Evaluator controls returned the expected outcome in 57/57 cases. Same-engine SQL re-execution reproduced 28/30 references; all 77 test-passing runs on these tasks still failed reference agreement. PlanAudit-Guided Revision improved code scores on 3/30 tasks and test coverage on 11/30, but the median paired change was zero for both measures. Failure analysis revealed tests that omitted requirements, adopted candidate-defined schemas, used nondiscriminating fixtures, or repeated the implementation's incorrect logic. Fresh generation on ten targeted tasks produced 0/30 reference matches and 25/25 false passes among test-passing runs. These findings are diagnostic rather than causal: the pipelines differ in feedback, grounding, call budgets, and editing permissions, and the study uses one model and one primary generation per eligible task. Because pre-adjudication human labels were not retained, inter-rater agreement cannot be reconstructed and the mechanism analyses remain exploratory. Overall, increasingly successful self-generated tests did not establish semantic correctness, even after excluding unreproduced references.

Chinese Translation

规划和修订旨在使生成的代码更接近任务规范。然而，错误的解释可能在计划、实施和测试中持续存在——我们称之为解释锁定。我们在30个源自TPC基准DS的数据框架任务上调查了这一风险，比较了冻结单次通过（Frozen One-Pass）、测试引导修订（Test-Guided Revision）和计划审计引导修订（PlanAudit-Guided Revision）。测试通过率从22/30上升到29/30和30/30，但没有生成的解决方案与其存储的参考输出匹配。评估者控制在57/57个案例中返回了预期结果。同引擎SQL重执行重现了28/30个参考；在这些任务上，77次通过测试的运行仍未能达成参考一致性。计划审计引导修订在3/30个任务上提高了代码评分，在11/30个任务上提高了测试覆盖率，但两项指标的中位数配对变化均为零。失败分析揭示了遗漏需求的测试、采用候选定义的模式、使用不具区分性的固定装置或重复实现的错误逻辑的情况。在十个目标任务上进行的新生成未能产生0/30的参考匹配，并在通过测试的运行中出现了25/25的虚假通过。这些发现是诊断性的而非因果性的：不同的管道在反馈、基础、调用预算和编辑权限上存在差异，研究使用了一个模型和每个合格任务的一个主要生成。由于未保留预裁定的人类标签，无法重建评审者间一致性，机制分析仍处于探索阶段。总体而言，越来越成功的自生成测试并未确立语义正确性，即使在排除未重现的参考后也是如此。

View on arXiv Download PDF AI Translation

cs.AI / 139 / 2610.15855

Attention Sinks in VLMs Are Borrowed, Not Seen

视觉语言模型中的注意力沉没是借用的，而不是被看到的

Zhang, Dengjia, Martin, Alexander, Yan, Pengyu, Sun, Xihan, Li, Jiaqi, Van Durme, Benjamin, Yates, Andrew, Kriz, Reno, Walden, William

Abstract

We study how to detect and mitigate visual attention sinks in vision-language models (VLMs). Visual attention sinks are image tokens that absorb a large share of attention over the image regardless of the query or task, and they mislead methods that treat attention maps as a record of where the model ``looked.'' Prior work has detected sinks through massive activation dimensions and proposed redistributing their attention, but none have addressed what makes a token become a sink. Comparing six candidate detectors on ten VLMs from four families, we find that only the centered cosine similarity between an image token's key and the key of the language model's text-sink detects visual sinks. Thus, a visual sink is an image token whose key aligns with the text-sink's key. Moving a sink's key component along this direction to the median of the image keys sends its attention back to the text-sink (\textbf{Are Borrowed}) while leaving model answers unchanged~(\textbf{Not Seen}). Building on this, we propose a training-free, entropy-triggered re-look that first cleans the attention map, then crops the most-attended region of the cleaned map and inserts it into the reasoning trace. Cleaning adds negligible computation to the forward pass, and the re-look improves accuracy with a mean gain of $+4.3$ points on fine-detail VQA benchmarks across model families.

Chinese Translation

我们研究如何检测和缓解视觉语言模型（VLMs）中的视觉注意力沉没。视觉注意力沉没是指在图像中吸收大量注意力的图像标记，无论查询或任务如何，这些沉没会误导将注意力图视为模型“观察”位置的方式。先前的研究通过大量激活维度检测沉没，并提出重新分配其注意力，但没有解决是什么导致一个标记成为沉没的问题。通过比较来自四个家族的十个VLMs的六个候选检测器，我们发现只有图像标记的键与语言模型文本沉没的键之间的中心余弦相似度能够检测视觉沉没。因此，视觉沉没是一个图像标记，其键与文本沉没的键对齐。将沉没的键组件沿该方向移动到图像键的中位数，会将其注意力重新指向文本沉没（ extbf{是借用的}），同时不改变模型的回答（ extbf{未被看到}）。基于此，我们提出了一种无训练、熵触发的重新观察方法，该方法首先清理注意力图，然后裁剪清理后图中注意力最集中的区域，并将其插入推理轨迹。清理对前向传播的计算增加微乎其微，而重新观察在不同模型家族的细节VQA基准上平均提高了$+4.3$分的准确性。

View on arXiv Download PDF AI Translation

cs.AI / 140 / 2610.15882

Engram Injection: A Study of Temporal Validity in Parametric Language-Model Memory

记忆痕迹注入：参数语言模型记忆的时间有效性研究

Siavoshian, Sam, Ramadan, Omar, Saeed, Amir K., Johnson, Benjamin A., Rodriguez, Benjamin M.

Abstract

A stored fact may stop being valid when a user changes a preference or an event replaces an earlier one. We study whether a language model can learn these changes in its parameters and answer correctly from only a question and the current time. Engram Injection stores facts in LoRA adapters and supplies the clock through FiLM conditioning on a frozen Qwen2.5-3B-Instruct model. We compare it with additive clock conditioning, prompting with the full dated history, and retrieving four dated records. We train six matched adapters across three seeds after fixing a sampler that could admit test-age examples into training. The training audit reports no such examples among 127,760 evaluated training inputs. On 62 pairs of questions asked before and after an update, at an elapsed age of seven days excluded from training, both adapter methods answer zero pairs correctly under the primary blinded assistant review with order code 1 held fixed. Full-history prompting answers 59/62 pairs correctly, and retrieval answers 62/62. Separate experiments show that adapters can store facts and that removable memory units can support local deletion. Combining fact selection with time-based version selection remains unsuccessful. These results separate learning a fact from learning when to use it.

Chinese Translation

当用户改变偏好或事件替代早期事件时，存储的事实可能会失去有效性。我们研究语言模型是否能够学习这些参数的变化，并仅通过一个问题和当前时间正确回答。记忆痕迹注入（Engram Injection）在 LoRA 适配器中存储事实，并通过在冻结的 Qwen2.5-3B-Instruct 模型上进行 FiLM 条件化来提供时钟。我们将其与加法时钟条件化、使用完整的带日期历史记录进行提示以及检索四个带日期记录进行比较。在修复一个可能将测试年龄示例纳入训练的采样器后，我们在三个种子上训练了六个匹配的适配器。训练审核报告在评估的 127,760 个训练输入中没有此类示例。在更新前后询问的 62 对问题中，在排除在训练中的七天经过时间下，两个适配器方法在主要盲审助手的审查中均未正确回答任何一对，保持顺序代码 1 固定。完整历史提示正确回答 59/62 对问题，而检索则正确回答 62/62 对。单独实验表明适配器可以存储事实，并且可移除的记忆单元可以支持局部删除。将事实选择与基于时间的版本选择相结合仍然未成功。这些结果将学习一个事实与学习何时使用它区分开来。

View on arXiv Download PDF AI Translation

cs.AI / 141 / 2610.15884

RetryBench: Separating Retry Permission from Recovery Outcomes in GPT Tool Use

RetryBench：将重试权限与GPT工具使用中的恢复结果分离

Liu, You, Liu, Yue, Lu, Quanchao

Abstract

False retry permission (FRP) occurs when an agent retries a timed-out tool call without support from the visible contract. RetryBench separates this error from realized recovery. Successful protected recovery does not imply reliable rejection of unsupported retries. In a 720-response key control, valid keys usually enabled safe progress, but offered changed keys elicited unsupported retries in Sol (1/80), Terra (55/80), and Luna (60/80). A later 240-response control removed the candidate and required models to supply a key. Luna still made 8/80 unsupported dispatches; Sol and Terra made none in 80 responses each. Accepting a displayed bad key therefore cannot explain all observed errors, although changed output requirements and conditions preclude a pure display-effect comparison. Removing the weak-contract label increased retries in Sol and Terra's coded-action conditions in a six-endpoint study. A separate 1,536-response canonical-action block also found a positive Terra contrast (115/128 versus 124/128 retries), concentrated in two task/order conditions; the other five contrasts remained unresolved after multiplicity correction. An explicit-contract cue control elicited deferral in all 762 returned responses, leaving recovery unresolved. These results support reporting permission and recovery together across declared interface variants. The evidence covers four purposive tasks, fixed GPT endpoints, and one recovery decision; it identifies neither internal mechanisms nor episode-level safety.

Chinese Translation

错误重试权限（False retry permission, FRP）发生在代理在没有可见合同支持的情况下重试超时的工具调用时。RetryBench将此错误与实现的恢复分开。成功的受保护恢复并不意味着可靠地拒绝不支持的重试。在720个响应的关键控制中，有效的密钥通常能够安全推进，但提供更改的密钥在Sol（1/80）、Terra（55/80）和Luna（60/80）中引发了不支持的重试。后来的240个响应控制去除了候选项，并要求模型提供密钥。Luna仍然进行了8/80次不支持的调度；而Sol和Terra在各自的80个响应中没有进行任何不支持的调度。因此，接受显示的坏密钥并不能解释所有观察到的错误，尽管更改的输出要求和条件排除了纯粹的显示效应比较。在一个六个端点的研究中，去除弱合同标签增加了Sol和Terra编码行动条件下的重试。在一个单独的1,536个响应的规范行动块中也发现了积极的Terra对比（115/128与124/128重试），集中在两个任务/顺序条件中；其他五个对比在多重性校正后仍未解决。一个明确合同提示的控制在所有762个返回响应中引发了延迟，导致恢复未解决。这些结果支持在声明的接口变体中共同报告权限和恢复。证据涵盖了四个目的性任务、固定的GPT端点和一个恢复决策；它既未识别内部机制，也未识别情节级别的安全性。

View on arXiv Download PDF AI Translation

cs.AI / 142 / 2610.15885

Comparative Study of Canvas- and Coordinate-Based Geometry-Conditioned Neural Networks for Real-Time Prediction of Indoor Environments with Varying Room Dimensions

基于画布和坐标的几何条件神经网络在变化房间尺寸下室内环境实时预测的比较研究

Wang, Shaofan, Geyer, Philipp

Abstract

Computational fluid dynamics (CFD) is computationally expensive for iterative building design. Deep learning surrogates offer faster predictions, but generalization across varying room dimensions remains challenging. This study compares two geometry-conditioned neural network surrogate models for rapid indoor airflow and temperature field prediction under varying room sizes. During the architectural design stage, this method is relevant for real-time assessment of whether conditioned supply air will effectively provide fresh and cool air throughout rooms with varying heights and lengths. The canvas-based neural network (CaNN) explicitly encodes room size by preserving different occupied regions within a fixed canvas, whereas the coordinate-based neural network (CoNN) uses normalized coordinate channels. Both employ U-Net autoencoders to separately predict airflow and temperature fields. Evaluation uses simplified rooms with varying heights and lengths as a controlled benchmark. The results show that CaNN achieves higher agreement with CFD results than CoNN, demonstrating its potential to support performance-driven building design.

Chinese Translation

计算流体动力学（CFD）在迭代建筑设计中计算成本高昂。深度学习代理模型提供了更快的预测，但在不同房间尺寸下的泛化仍然具有挑战性。本研究比较了两种几何条件神经网络代理模型，以快速预测不同房间大小下的室内气流和温度场。在建筑设计阶段，该方法对于实时评估调节供气是否能有效提供新鲜和凉爽的空气到不同高度和长度的房间中具有重要意义。基于画布的神经网络（CaNN）通过在固定画布内保留不同的占用区域来显式编码房间大小，而基于坐标的神经网络（CoNN）则使用归一化坐标通道。两者均采用 U-Net 自编码器分别预测气流和温度场。评估使用具有不同高度和长度的简化房间作为控制基准。结果表明，CaNN 与 CFD 结果的吻合度高于 CoNN，展示了其支持以性能为驱动的建筑设计的潜力。

View on arXiv Download PDF AI Translation

cs.AI / 143 / 2610.15908

LURE: a Phase-Labelled Synthetic Corpus of Online-Grooming Exchanges

LURE：一个标记阶段的在线诱骗交流合成语料库

Florindi, Emanuele, Franzoni, Valentina, Guercio, Ludovico, Milani, Alfredo, Polticchia, Mattia

Abstract

Phase-annotated conversational data is essential for studying the progression of online grooming of minors and for developing early warning tools. However, the only publicly available source (i.e., Perverted-Justice archive, which contains a subset coded using ChatCoder) is small, outdated, and frozen. We present a synthetic corpus of groomer-victim exchanges, generated using a retrieval-augmented pipeline running on institutional hardware. This pipeline uses an openly available, eight-billion-parameter, refusal-ablated language model. The corpus was created using the five stages of the responsible-generation protocol for synthetic datasets: ontology-driven design based on the ChatCoder luring-communication scheme (eight communicative tags mapped onto four grooming phases); multi-model prompt testing; human-in-the-loop supervision; semantic and feature-level validation; and modality-aware documentation. The released corpus contains 2,121 unique exchanges across seven tags and four phases, each accompanied by a Plutchik emotion field. Validation quantifies the integrity and lexical profile of the corpus, as well as the reproduction of real seed lines, in-domain label recoverability (tag accuracy: 0.85; phase accuracy: 0.93) and its transfer to real ChatCoder offender lines, which were coded automatically at the eight-tag level (31,640 lines) and by human coders at the three-construct level (10,876 posts). Accuracy was measured for lexical and sentence-embedding features, while a real-versus-synthetic discriminator was able to separate the two domains with 91% accuracy. The scope of the corpus includes controlled, phase-conditioned experimentation, pipeline development and augmentation alongside real data and awareness material. The corpus, its datasheet and the audit code are released with vetted access.

Chinese Translation

标注阶段的对话数据对于研究未成年人在线诱骗的进展以及开发早期预警工具至关重要。然而，唯一可公开获取的来源（即 Perverted-Justice 档案，其中包含使用 ChatCoder 编码的子集）规模小、过时且已被冻结。我们提出了一个诱骗者-受害者交流的合成语料库，该语料库是通过在机构硬件上运行的检索增强管道生成的。该管道使用一个公开可用的、拥有八十亿参数的拒绝消融语言模型。该语料库的创建遵循合成数据集的负责任生成协议的五个阶段：基于 ChatCoder 诱骗沟通方案的本体驱动设计（八个交际标签映射到四个诱骗阶段）；多模型提示测试；人类监督；语义和特征级验证；以及模态感知文档。发布的语料库包含 2,121 个独特的交流，涵盖七个标签和四个阶段，每个交流都附有 Plutchik 情感字段。验证量化了语料库的完整性和词汇特征，以及真实种子行的再现、领域内标签的可恢复性（标签准确率：0.85；阶段准确率：0.93）及其向真实 ChatCoder 罪犯行的转移，这些行在八标签级别上自动编码（31,640 行），在三构造级别上由人工编码（10,876 帖子）。准确性是针对词汇和句子嵌入特征进行测量的，而一个真实与合成的判别器能够以 91% 的准确率区分这两个领域。该语料库的范围包括受控的、阶段条件实验、管道开发和增强，以及与真实数据和意识材料的结合。该语料库、其数据表和审计代码已发布并提供经过审核的访问。

View on arXiv Download PDF AI Translation

cs.AI / 144 / 2610.15930

Multi-Horizon Explainable Queue Forecasting and Resource-Aware Decision Support for Intelligent Border Control Using LightGBM

基于LightGBM的多层次可解释排队预测与资源感知决策支持在智能边境管控中的应用

Mama, Varvara, Christidi, Fani D., Kapsalis, Nikolaos C., Baklezos, Anargyros T., Nikolopoulos, Christos D.

Abstract

Border crossing points are stochastic service systems in which traffic demand, traveler categories, registration status, inspection requirements, staffing, and lane availability interact on multiple time scales. This paper extends an earlier multimodal queue-management framework by introducing an implementation-oriented, multi-horizon forecasting and decision-support architecture based on Light Gradient Boosting Machine (LightGBM) models. In contrast to a single forecasting block, the proposed approach constructs horizon-specific regressors for 15, 30, 60, and 120 min forecasts and jointly estimates future waiting time and queue lengths. The models operate on temporally lagged traffic, service, density, and queue features, while an operational layer incorporates lane compatibility, officer availability, possibly degraded resources, traveler category, preregistration information, and human-approval constraints. Quantile LightGBM models provide 95\% prediction intervals for waiting-time forecasts. Beyond prediction, the framework produces machine-readable recommendations, alerts, aggregate key performance indicators, confidence information, and integration outputs suitable for border resource-management and decision-support systems. The results establish a reproducible synthetic-data baseline and a practical interface for subsequent validation with operational border-crossing data.

Chinese Translation

边境过境点是随机服务系统，其中交通需求、旅客类别、登记状态、检查要求、人员配置和车道可用性在多个时间尺度上相互作用。本文通过引入基于轻梯度提升机（Light Gradient Boosting Machine, LightGBM）模型的面向实施的多层次预测和决策支持架构，扩展了早期的多模式排队管理框架。与单一预测模块不同，所提出的方法为15、30、60和120分钟的预测构建了特定于时间层次的回归模型，并联合估计未来的等待时间和排队长度。这些模型基于时间滞后的交通、服务、密度和排队特征进行操作，同时，操作层整合了车道兼容性、执法人员可用性、可能退化的资源、旅客类别、预登记信息和人工审批约束。分位数LightGBM模型为等待时间预测提供95\%的预测区间。除了预测外，该框架还生成机器可读的建议、警报、聚合关键绩效指标、置信信息以及适合边境资源管理和决策支持系统的集成输出。结果建立了可重复的合成数据基线，并为后续与实际边境过境数据的验证提供了实用接口。

View on arXiv Download PDF AI Translation

cs.AI / 145 / 2610.15938

The Anatomy of a Blind Spot: What an LLM Judge Must Be Shown, and How Much

盲点的解剖：大型语言模型法官必须被展示什么，以及展示多少

Kumar, Nitish, Pothuneedi, Prateik, Bandak, Greg

Abstract

Evaluation of text generated by large language models (LLMs) is increasingly delegated to LLMs acting as judges, yet what such a judge must be shown remains unresolved. We treat a judge whose reports are acted on as a regulator, in the sense of cybernetics, and ask what it must be shown, how much, in what form, and what its reports must get right. On seven public benchmarks we varied only what the judge was shown. Across three judges whose prices differ 10- to 17-fold, what the judge was shown mattered more than which judge it was. When a system revised its answers on feedback from a judge denied the material they had to respect, it ended below its starting point on instruction following and question answering. On instruction following the loss grew with each round, and it persisted when the judge saw the user's request but not the constraints and with a second judge family. On news summarisation, a correct reference summary beside the source undid most of the source's contribution, whereas on single-answer questions a reference helped. About a tenth of the source did almost all of its work, and a small entailment graph matched an LLM judge at finding errors in grounded text. We committed 83 testable clauses before running; of the 81 that ran, 32 met their bar and 50 went in the predicted direction, and most quantitative predictions from the theory failed. We report every shortfall and derive design rules tied to measured effects.

Chinese Translation

对大型语言模型（LLMs）生成文本的评估越来越多地委托给作为法官的LLMs，但这样的法官必须被展示什么仍然没有解决。我们将法官视为一种调节者，从控制论的角度出发，探讨它必须被展示什么、展示多少、以何种形式，以及它的报告必须正确的内容。在七个公共基准测试中，我们仅改变了法官所看到的内容。在三位价格相差10到17倍的法官中，法官所看到的内容比法官本身更为重要。当一个系统在反馈来自被剥夺了必须遵循材料的法官的情况下修正其答案时，其在遵循指令和回答问题上的表现低于起始水平。在遵循指令的过程中，损失随着每一轮的增加而增长，当法官看到用户的请求但未看到约束条件时，这种情况依然存在，并且在第二位法官的情况下也是如此。在新闻摘要中，正确的参考摘要与源文本并列，几乎抵消了源文本的大部分贡献，而在单一答案问题中，参考资料则起到了帮助作用。大约十分之一的源文本完成了几乎所有的工作，而一个小的蕴涵图在发现基础文本中的错误时与LLM法官相匹配。我们在运行之前提交了83个可测试的条款；在运行的81个中，有32个达到了标准，50个朝着预测的方向发展，而理论中的大多数定量预测未能实现。我们报告了每一个不足之处，并推导出与测量效果相关的设计规则。

View on arXiv Download PDF AI Translation

cs.AI / 146 / 2610.15948

Vision: "What Don't I Know About User's Preferences?" How Uncertainty Estimation Improves Proactive Assistant Alignment

愿景："我对用户偏好的了解还有哪些不足？" 不确定性估计如何改善主动助手的对齐

An, Zhiyu, Xuan, Ziyi, Yang, Yu, Du, Wan

Abstract

Proactive Smart Assistants (PSAs) increasingly use general knowledge, learned patterns from group-level data, patterns from individual user's data, and recent sensor observations to anticipate which action a user may prefer. Existing systems typically select actions from predicted user preference likelihoods, but these likelihoods alone do not reveal how reliable the prediction is, which evidence supports it, or why ambiguity remains. In this vision paper, we argue that uncertainty estimation can provide this missing information and substantially change appropriate downstream behavior. Using the example of a home assistant, we introduce the questions that uncertainty estimation aims to answer, i.e. how sure is the assistant about the predicted preference likelihood, whether the confidence come from group-level assumptions or individual's data, and whether likelihood ambiguity reflect the user's intrinsic preference variability or insufficient data for the particular situation. We further introduce uncertainty estimation applications in neighboring domains, such as human trajectory prediction, personalized activity recognition, information-collection, and model-based control. Finally, we draw on insights from neighboring domains to envision how future uncertainty-aware proactive assistants may address key challenges in proactive interaction.

Chinese Translation

主动智能助手（PSAs）越来越多地利用一般知识、从群体数据中学习的模式、个体用户数据中的模式以及最近的传感器观察来预测用户可能的偏好行为。现有系统通常根据预测的用户偏好可能性选择行动，但这些可能性本身并未揭示预测的可靠性、支持该预测的证据或为何仍然存在模糊性。在本文中，我们认为不确定性估计可以提供这些缺失的信息，并显著改变适当的后续行为。以家庭助手为例，我们提出了不确定性估计旨在回答的问题，即助手对预测的偏好可能性有多确定，这种信心是来自群体假设还是个体数据，以及可能性模糊性是否反映了用户内在的偏好变异性或特定情境下的数据不足。我们进一步介绍了不确定性估计在相关领域的应用，如人类轨迹预测、个性化活动识别、信息收集和基于模型的控制。最后，我们借鉴相关领域的见解，展望未来的关注不确定性的主动助手如何应对主动交互中的关键挑战。

View on arXiv Download PDF AI Translation

cs.AI / 147 / 2610.15956

ADAPT: Audio-Depth Action Planning with Transition-Aware Rewards for Lights-Out Industrial Navigation

ADAPT：具有过渡感知奖励的音频-深度动作规划用于无光工业导航

Huang, Wenjun, Fu, Shenghao, Jin, Yian, Ni, Yang, Li, Yifan, Li, Lichi, Diao, Xingjian, Zhang, Pengfei, Song, Yutong, Chen, Hanning, Cui, Jiarui, Yang, Tianyu, Li, Shilong, Kong, Yu, Imani, Mohsen

Abstract

Vision-language-action models for embodied navigation typically rely on RGB observations, limiting their applicability in lights-out industrial environments where visible illumination may be unavailable. We study illumination-independent navigation using spatial audio for target localization and depth for obstacle-aware geometry. To support this setting, we introduce DarkNav-Bench, a benchmark with 24 industrial scenes and four progressively challenging acoustic conditions. We further propose ADAPT, an audio-depth action-planning framework equipped with spatial-audio perception and optimized for closed-loop navigation through reinforcement learning (RL). Our key RL contribution, transition-aware reward shaping (TARS), defines a pose-aware shortest-action potential over a position-heading graph. Unlike conventional geodesic-progress rewards, TARS accounts for both feasible movements and in-place rotations, providing action-consistent feedback for policy optimization. Experiments show that ADAPT consistently outperforms training-based navigation baselines and zero-shot multimodal models on both seen and unseen scenes. On unseen environments, it achieves 90.6% success under Clean conditions and retains 60.0% success under the most challenging Localized Noise setting. These results demonstrate the potential of spatial audio and depth as complementary modalities for reliable RGB-free navigation in lights-out industrial environments.

Chinese Translation

用于具身导航的视觉-语言-动作模型通常依赖于RGB观测，这限制了它们在无光工业环境中的适用性，因为在这些环境中可见光照可能不可用。我们研究了使用空间音频进行目标定位和深度感知障碍几何的独立于光照的导航。为支持这一设置，我们引入了DarkNav-Bench，这是一个包含24个工业场景和四种逐渐具有挑战性的声学条件的基准测试。我们进一步提出了ADAPT，一个配备空间音频感知的音频-深度动作规划框架，并通过强化学习（RL）优化以实现闭环导航。我们关键的RL贡献是过渡感知奖励塑形（TARS），它在位置-航向图上定义了一个姿态感知的最短动作潜力。与传统的测地进度奖励不同，TARS考虑了可行的移动和原地旋转，为策略优化提供了一致的动作反馈。实验表明，ADAPT在已见和未见场景中均持续优于基于训练的导航基线和零-shot多模态模型。在未见环境中，它在干净条件下实现了90.6%的成功率，并在最具挑战性的局部噪声设置下保持了60.0%的成功率。这些结果展示了空间音频和深度作为可靠的RGB-free导航的互补模态在无光工业环境中的潜力。

View on arXiv Download PDF AI Translation

cs.AI / 148 / 2610.15964

Small but Mighty: Test Suites as Improvement Signals for Self-Evolving Agents

小而强大：测试套件作为自我进化智能体的改进信号

Cheng, Yutong, Cheng, Wei, Yu, Wenchao, Zhou, Dawei, Chen, Haifeng

Abstract

A self-evolving agent rewrites its own harness and must decide which rewrites to keep, so its improvement signal is the check a rewrite has to pass. Verifying a rewrite by agent rollout is expensive and subjective, while executable checks such as the repository's own tests are too sparse to catch behavior-changing rewrites, rejecting only a quarter of them. We instead take test feedback as the executable signal, and search for test suites that are both small and strong with a quality-diversity evolutionary algorithm, so that the signal is dense and informative. TeSearch treats suite size as a behavior descriptor, so that quality is pursued at every size and a small, strong suite is reachable, and learns a transition kernel during the search that denoises the LLM as an actuator on that descriptor. Across 129 real-world modules and three backbones, TeSearch delivers the highest-quality suites of any method compared, LLM-driven evolutionary search included, and the smallest of any LLM-based one. Used as the improvement signal of a self-evolving agent, it rejects 59.9% of behavior-changing rewrites against 25.3% for the repository's tests, at a 4.0% false-alarm rate and 3.5 s per decision. The code and data are available at https://github.com/ccsnow127/TeSearch.

Chinese Translation

自我进化智能体会重写其自身的控制程序，并必须决定保留哪些重写，因此其改进信号是重写必须通过的检查。通过智能体的执行来验证重写是昂贵且主观的，而可执行检查（如代码库自身的测试）则过于稀疏，无法捕捉到改变行为的重写，仅拒绝了四分之一的重写。我们反而将测试反馈视为可执行信号，并利用质量-多样性进化算法寻找既小又强的测试套件，以便使信号密集且信息丰富。TeSearch将套件大小视为行为描述符，因此在每个大小上都追求质量，并且可以达到一个小而强的套件，并在搜索过程中学习一个转移核，以去噪作为该描述符的执行器的LLM。在129个真实模块和三个基础架构上，TeSearch提供了与任何比较方法中最高质量的套件，包括基于LLM的进化搜索，并且是所有基于LLM的方法中最小的。作为自我进化智能体的改进信号，它拒绝了59.9%的改变行为的重写，而代码库的测试仅拒绝了25.3%，假阳性率为4.0%，每次决策耗时3.5秒。代码和数据可在 https://github.com/ccsnow127/TeSearch 获取。

View on arXiv Download PDF AI Translation

cs.AI / 149 / 2610.15969

PopGen: Generating Semantically Diverse Populations for Zero-Shot Coordination

PopGen：为零-shot协调生成语义多样化的群体

Teiletche, Paul, Ryu, Kanghyun, Park, Kyungjoon, Tirumala, Dhruva, Wulfmeier, Markus, Mehr, Negar

Abstract

In multi-agent reinforcement learning, exposing a cooperator to partners with diverse strategies during training is key to better collaboration at test time. However, existing methods typically rely on trajectory-level proxies to measure diversity, which may label two distinct trajectories of the same strategy as diverse. Instead, our key insight is to leverage LLMs' ability to describe behaviorally distinct agents in environments, similar to how humans can recognize high-level strategies. To this end, we introduce PopGen, a framework that synthesizes semantically diverse agents from scratch using LLMs. PopGen first generates a diverse set of textual descriptions of agent behaviors, then translates each behavior into a code policy that can be executed within the environment. We show that cooperators trained against PopGen populations achieve robust zero-shot cooperation with unseen partners, outperforming population-play baselines across four collaborative environments. We further show that semantic diversity scales with PopGen population size, improving the cooperator's worst-case performance, and we identify design choices that preserve population quality while maximizing diversity. Our code is available at https://github.com/labicon/PopGen.

Chinese Translation

在多智能体强化学习中，训练期间将合作者暴露于具有多样化策略的伙伴是提高测试时协作能力的关键。然而，现有方法通常依赖于轨迹级别的代理来衡量多样性，这可能会将同一策略的两个不同轨迹标记为多样化。相反，我们的关键见解是利用大型语言模型（LLMs）描述在环境中行为上具有明显差异的智能体的能力，这类似于人类识别高级策略的方式。为此，我们引入了PopGen，一个使用LLMs从零开始合成语义多样化智能体的框架。PopGen首先生成一组多样的智能体行为文本描述，然后将每种行为转换为可以在环境中执行的代码策略。我们展示了与PopGen群体训练的合作者能够与未见过的伙伴实现稳健的零-shot合作，并在四个协作环境中超越了基于群体游戏的基线。我们进一步表明，语义多样性与PopGen群体规模成正比，提高了合作者的最坏情况表现，并确定了在最大化多样性的同时保持群体质量的设计选择。我们的代码可在 https://github.com/labicon/PopGen 获取。

View on arXiv Download PDF AI Translation

cs.AI / 150 / 2610.15991

Revocation Is Not Complete at Receipt: An Executable Outcome Contract for Distributed Agent-Tool Gateways

撤销在接收时并不完全：一种可执行的结果合同用于分布式代理-工具网关

Liu, Lifei, Yu, Haoran, Zhang, Danping

Abstract

Distributed agent workflows cache delegated authority at tool gateways. Revocation may be reported at issuer commit, transport acknowledgment, or receiver persistence, although none of these observations establishes that all gateways enforced the change or that guarded effects stopped. We define C0-C5, a bounded outcome contract spanning issuer commit through effect prevention, and implement it as a persistent monitor over a frozen gateway set, call set, and observation window. A 14-case fault matrix checks conservative classification, sticky violations, restart persistence, and evidence immutability. In a derived audit of its eight semantic cases, treating C3 or C4 as completion produces seven or five false positives, respectively; the bounded C0-C5 verdict produces none. Descriptive scaling covers 1,024 gateways and 10,000 effect vectors. The primary experiment uses the maintained Keycloak 26.7 identity server and measures the contract across two durable receivers, eight physical agent-tool gateways, Filesystem and PostgreSQL effects, and 16 AWS hosts in three availability zones. Normal delivery and recovery commit no effects. Under receiver-selective non-delivery, transport acknowledgment and receiver persistence still succeed, but 112 of 225 calls (49.78%) are admitted, and all 112 effects commit in every treatment block. A separately provisioned execution repeats the result. OPA/Toxiproxy and a live three-role LLM workflow with networked MCP tools reproduce the persisted-but-not-enforced boundary on narrower surfaces. Finally, an unmodified external MCPMark agent executes 60 closed-loop task arms. Normal C4 denies all 31 exposed post-barrier calls and commits no effect; selective C3 allows 344 calls and commits 145 effects across Filesystem and PostgreSQL. Transport success is therefore evidence of progress, not a revocation-completion verdict.

Chinese Translation

分布式代理工作流在工具网关处缓存委托权限。撤销可能在发行者提交、传输确认或接收者持久性时被报告，尽管这些观察都无法证明所有网关都执行了更改或受保护的效果停止。我们定义了 C0-C5，一个涵盖发行者提交到效果预防的有界结果合同，并将其实现为对冻结的网关集、调用集和观察窗口的持久监控。一个包含14个案例的故障矩阵检查保守分类、粘性违规、重启持久性和证据不可变性。在对其八个语义案例的派生审计中，将 C3 或 C4 视为完成分别会产生七个或五个假阳性；而有界的 C0-C5 判决则没有产生假阳性。描述性扩展覆盖了 1,024 个网关和 10,000 个效果向量。主要实验使用维护的 Keycloak 26.7 身份服务器，并在两个持久接收者、八个物理代理-工具网关、Filesystem 和 PostgreSQL 效果，以及三个可用区的 16 个 AWS 主机上测量合同。正常交付和恢复不产生效果。在接收者选择性不交付的情况下，传输确认和接收者持久性仍然成功，但225个调用中有112个（49.78%）被接纳，所有112个效果在每个处理块中都提交。单独配置的执行重复了该结果。OPA/Toxiproxy 和一个具有网络化 MCP 工具的实时三角色 LLM 工作流在更狭窄的表面上重现了持久但未执行的边界。最后，一个未修改的外部 MCPMark 代理执行了60个闭环任务臂。正常的 C4 拒绝所有31个暴露的后障碍调用并不产生效果；选择性 C3 允许344个调用，并在 Filesystem 和 PostgreSQL 中提交145个效果。因此，传输成功是进展的证据，而不是撤销完成的判决。

View on arXiv Download PDF AI Translation

cs.AI / 151 / 2610.15993

Handoff Curves: Initial Selection, Later Handoff, and Supervision Targets in Multi-turn LLM Routing

切换曲线：多轮 LLM 路由中的初始选择、后续切换和监督目标

Matos, Daniel, Liu, Lu, Zhang, Yanfeng

Abstract

We study how initial model selection, later handoff, and training targets affect success and cost in LLM tasks that span multiple turns. We compare several router architectures using handoff curves, which record task success and cost as the handoff step changes. Our dataset contains 34,204 complete executions of three Qwen models across BFCL, MultiWOZ, and $\tau$-retail, providing offline supervision for routing decisions. In our primary comparison, with Qwen3.5 35B-A3B as the starting model, allowing a later handoff does not establish an improvement in utility over initial selection on BFCL or MultiWOZ, and increases cost on BFCL across a range of cost weights. Changing the supervision targets improves router performance on both our dataset and TwinRouterBench. On our dataset, replacing noisy training success curves with monotone fits raises success by 0.2 to 4.1 percentage points at the same cost across the reported operating points. On TwinRouterBench, we construct targets that describe the model capability needed for the remaining work and difficult steps ahead. Compared with training on the benchmark's per-step labels, these targets raise trajectory pass scores by 7.8 percentage points and cost savings by 6.3 percentage points on the full static bank. A live deployment of the router trained with these targets resolves 79 of 100 instances at a router cost of 0.59 US dollars per instance.

Chinese Translation

我们研究了初始模型选择、后续切换和训练目标如何影响跨多个回合的 LLM 任务的成功率和成本。我们使用切换曲线比较了几种路由器架构，该曲线记录了任务成功率和成本随着切换步骤的变化。我们的数据集包含了 34,204 次完整执行，涵盖了三个 Qwen 模型在 BFCL、MultiWOZ 和 $ au$-retail 上的表现，为路由决策提供了离线监督。在我们的主要比较中，以 Qwen3.5 35B-A3B 作为起始模型，允许后续切换并未在 BFCL 或 MultiWOZ 上提高效用，相反在 BFCL 上增加了成本，且在不同的成本权重下均如此。改变监督目标提高了路由器在我们的数据集和 TwinRouterBench 上的表现。在我们的数据集中，用单调拟合替换嘈杂的训练成功曲线，使得在相同成本下成功率提高了 0.2 到 4.1 个百分点。在 TwinRouterBench 上，我们构建了描述模型在剩余工作和困难步骤中所需能力的目标。与在基准的逐步标签上训练相比，这些目标使得轨迹通过分数提高了 7.8 个百分点，成本节省提高了 6.3 个百分点。在使用这些目标训练的路由器的实时部署中，成功解决了 100 个实例中的 79 个，每个实例的路由器成本为 0.59 美元。

View on arXiv Download PDF AI Translation

cs.AI / 152 / 2610.15994

Large language model-generated patterns for epidemiological modeling: a multi-benchmark evaluation framework

大型语言模型生成的流行病学建模模式：一个多基准评估框架

Mao, Sophia, Gandhi, Keshav, Gutfraind, Alexander

Abstract

Representing individual-level behavioral heterogeneity is challenging in epidemiological modeling, particularly when empirical behavioral data are sparse. We assess the potential of large language model (LLM)-generated behavioral outputs in addressing this challenge, using hepatitis C virus-related behaviors among people who inject drugs as a case study. Correspondence with epidemiological patterns was evaluated using external benchmarks: literature-informed association directions, correlations from an agent-based model (ABM), and survey associations. Across three LLMs, unconditional directional recovery across 28 retained literature-informed associations was 46-68%, compared with 82-93% agreement when LLMs were directly asked to classify the same associations. Mean absolute differences from ABM and survey correlation patterns were 0.11-0.13 and 0.15-0.18, respectively. Together, these findings show partial alignment between LLM-generated behavioral relationships and epidemiological evidence, despite higher knowledge-probe agreement. More systematic evaluation is needed before incorporating LLM-generated behavioral outputs into epidemiological models that simulate transmission dynamics.

Chinese Translation

在流行病学建模中，代表个体行为异质性是一个挑战，特别是在经验行为数据稀缺的情况下。我们评估了大型语言模型（LLM）生成的行为输出在应对这一挑战中的潜力，以注射毒品者中的丙型肝炎病毒相关行为为案例研究。通过外部基准评估与流行病学模式的对应关系：文献知情的关联方向、基于代理的模型（ABM）的相关性以及调查关联。在三个LLM中，28个保留的文献知情关联的无条件方向恢复率为46-68%，而当直接要求LLM对相同关联进行分类时，协议率为82-93%。与ABM和调查相关模式的平均绝对差异分别为0.11-0.13和0.15-0.18。这些发现表明，尽管知识探测协议更高，LLM生成的行为关系与流行病学证据之间存在部分一致性。在将LLM生成的行为输出纳入模拟传播动态的流行病学模型之前，需要进行更系统的评估。

View on arXiv Download PDF AI Translation

cs.AI / 153 / 2610.16015

Computational Reflective Equilibrium for Agentic AI: A Rawlsian Alignment Scaffold for Governance

代理人工智能的计算反思平衡：一种罗尔斯式的治理对齐框架

Zhu, Quanyan

Abstract

The rapid deployment of agentic AI has introduced complex challenges for the ethics and governance of AI systems. Agents built on large language models act through tools on behalf of people, and each deployment case brings its own instructions, policies, evidence, and stakeholders, so alignment must be formed case by case. To this end, this work proposes agentic reflective equilibrium (ARE), a computational reflective equilibrium framework that serves as an ethical alignment scaffold in the agent harness. Drawing inspiration from the reflective equilibrium of John Rawls, the scaffold links principles from the system prompt and operator policy, considered judgments, tool evidence, and candidate actions through typed support and attack relations, with provenance encoding the instruction hierarchy. The agent acts on an equilibrium of this state and calls tools or asks questions only while doing so is worth its cost, unless a due-diligence mandate requires them. Governance enters through value priorities, source authority, prohibitions, human-in-the-loop triggers, and mandates, and certificates attribute each action to the sources whose commitments drove it. We show that equilibria are Nash equilibria of a mutual-adjustment game and establish monotonicity, lexical priority, stability certificates, and a quarantine theorem for prompt injection. Two case studies illustrate the framework: a computer-use assistant told to win a refund "whatever it takes" files an honest warranty claim, refuses a fake review, and ignores an injected request for personal data; a hiring agent showcases how authority, oversight triggers, audit mandates, and hard prohibitions shape discrimination, privacy violations, and escalation. The framework offers traceability, coherence, and adaptivity, facilitating safe, secure, and trustworthy agentic AI.

Chinese Translation

代理人工智能的快速部署给人工智能系统的伦理和治理带来了复杂的挑战。基于大型语言模型构建的代理通过工具代表人类行动，每个部署案例都有其特定的指令、政策、证据和利益相关者，因此对齐必须逐案形成。为此，本文提出了代理反思平衡（ARE），这是一种计算反思平衡框架，作为代理利用中的伦理对齐支架。该框架受到约翰·罗尔斯反思平衡理论的启发，将系统提示和操作政策、经过考虑的判断、工具证据和候选行动通过类型化的支持和攻击关系连接起来，并通过来源编码指令层级。代理在这一状态的平衡上行动，仅在这样做的成本值得时才调用工具或提出问题，除非有尽职调查的要求。治理通过价值优先级、来源权威、禁令、人机交互触发器和强制措施进入，并且证书将每个行动归因于推动其承诺的来源。我们展示了平衡是相互调整博弈的纳什平衡，并建立了单调性、词汇优先级、稳定性证书和提示注入的检疫定理。两个案例研究说明了该框架：一个被告知“无论如何都要赢得退款”的计算机使用助手提交了诚实的保修索赔，拒绝了虚假的评论，并忽略了对个人数据的注入请求；一个招聘代理展示了权威、监督触发器、审计强制措施和严格禁令如何影响歧视、隐私侵犯和升级。该框架提供了可追溯性、一致性和适应性，促进安全、可靠和值得信赖的代理人工智能。

View on arXiv Download PDF AI Translation

cs.AI / 154 / 2610.16050

Cross-Protocol Personality Assessment with LLMs: Aligning Behavioral Evidence with Human Ratings

跨协议个性评估与大型语言模型：将行为证据与人类评分对齐

Shan, Wei, Zhang, Tianyi, Zong, Yuan, Zheng, Wenming

Abstract

Personality assessment from asynchronous video interviews (AVIs) plays a crucial role in recruitment by enabling standardized and behaviorally grounded evaluation for job applicants. However, most of the previous works are confined to a single interview protocol. Independently collected protocols differ in question sets and question-trait mapping. Thus, the human-rated evidence from one protocol cannot be used directly under another. To address this limitation, we propose the Cross-protocol Evidence-grounded Personality Assessment (CEPA) framework, which aligns behavioral representations across interview protocols. CEPA addresses protocol heterogeneity by projecting source behavioral records into the target protocol's construct-oriented representation space. It retrieves behaviorally relevant and personality-complementary human-rated exemplars to ground an ensemble of LLM raters, which enables behavior-to-personality associations without target-specific supervision. In cross-protocol experiments, CEPA reduces average mean squared error (MSE) by 43.8% (Forward) and 17.0% (Reverse) relative to the strongest baseline. Our empirical experiments show that human-rated evidence from one protocol can support personality assessment under another without target-specific supervision or retraining.

Chinese Translation

来自异步视频面试（AVI）的个性评估在招聘中发挥着至关重要的作用，因为它能够为求职者提供标准化和基于行为的评估。然而，以往的大多数研究仅限于单一的面试协议。独立收集的协议在问题集和问题-特质映射上存在差异。因此，来自一个协议的人类评分证据无法直接用于另一个协议。为了解决这一局限性，我们提出了跨协议证据基础个性评估（CEPA）框架，该框架对齐了不同面试协议之间的行为表征。CEPA通过将源行为记录投影到目标协议的构念导向表征空间来解决协议异质性。它检索行为相关且个性互补的人类评分示例，以支持一组大型语言模型（LLM）评分者，从而在没有目标特定监督的情况下实现行为与个性之间的关联。在跨协议实验中，CEPA相较于最强基线将平均均方误差（MSE）分别降低了43.8%（正向）和17.0%（反向）。我们的实证实验表明，来自一个协议的人类评分证据可以在另一个协议下支持个性评估，而无需目标特定的监督或重新训练。

View on arXiv Download PDF AI Translation

cs.AI / 155 / 2610.16070

Understanding and Reducing Misalignment in Persuasion Optimization

理解与减少说服优化中的不一致性

Wei, Boyi, Desai, Ruta, Henderson, Peter, Hua, Wenyue

Abstract

Frontier models are now trained for complex multi-agent tasks, from running a company to engaging in negotiations. These tasks necessitate training models to be persuasive. But what does this optimization pressure mean for model alignment? In this work, we study how reinforcement learning (RL) training for persuasion affects models' use of misaligned strategies such as evidence fabrication and belief coercion, and what techniques can reduce such misalignment. We find that optimizing sender models solely for receiver belief shifts causes the average number of distinct misaligned strategies used per game to increase by more than 25%, and fabrications per game increase to more than 4 times relative to pre-RL models. We compare various mitigation approaches, including stronger strategy guidance, clean demonstrations, penalizing misaligned strategy usage, and improving the model's ability to detect fabrications. We find that navigating the persuasiveness-misalignment trade-off in this setting is challenging, with the best trade-offs coming from combining guidance on aligned persuasion strategies with reward penalties on misalignment. Overall, our work provides a testbed for assessing misalignment stemming from RL-based persuasion training and shows concrete paths for reducing misalignment. Importantly, it highlights that without direct intervention, persuasion-involved RL environments are likely to naturally lead to large increases in the use of misaligned strategies.

Chinese Translation

前沿模型现在被训练用于复杂的多智能体任务，从经营公司到参与谈判。这些任务需要训练模型具备说服力。但这种优化压力对模型一致性意味着什么？在本研究中，我们探讨了说服的强化学习（RL）训练如何影响模型使用不一致策略（如证据伪造和信念强迫），以及哪些技术可以减少这种不一致性。我们发现，仅仅优化发送者模型以实现接收者信念的转变，会导致每场游戏中使用的不同不一致策略的平均数量增加超过25%，而每场游戏中的伪造次数相较于前RL模型增加至超过4倍。我们比较了多种缓解方法，包括更强的策略指导、清晰的示范、惩罚不一致策略的使用以及提高模型检测伪造的能力。我们发现，在这种情况下，平衡说服性与不一致性之间的权衡是具有挑战性的，最佳的权衡来自于将对一致性说服策略的指导与对不一致性的奖励惩罚相结合。总体而言，我们的工作提供了一个评估源于基于RL的说服训练的不一致性的测试平台，并展示了减少不一致性的具体路径。重要的是，它强调在没有直接干预的情况下，涉及说服的RL环境可能自然导致不一致策略使用的大幅增加。

View on arXiv Download PDF AI Translation

cs.AI / 156 / 2610.16074

Beyond Single-Pathogen Forecasts: Predicting Shared Respiratory Hospital Burden from Influenza, COVID-19, and RSV

超越单一病原体预测：预测流感、COVID-19和呼吸道合胞病毒（RSV）共同造成的医院负担

Saha, Gaurob, Woo, MinJae, Rennert, Lior

Abstract

Co-circulating influenza, COVID-19, and respiratory syncytial virus (RSV) create a shared hospital-admission burden, because beds, staff, testing capacity, and respiratory support are allocated to total demand rather than to one pathogen at a time. This makes respiratory forecasting a target-construction problem as well as a modeling problem: disease-specific forecasts preserve attribution, whereas hospitals also need a single planning signal for combined burden. We developed a supervised 1- to 4-week-ahead framework for weekly admissions and compared two total-burden strategies: summing independently generated pathogen-specific forecasts and forecasting the combined admission target directly. Evaluation included retrospective held-out testing and prospective operational forecasts issued before the corresponding admission weeks were observed, with performance summarized using MAE and Percent Agreement (PA; the smaller of model-generated and observed counts divided by the larger). In retrospective total-burden evaluation, the combined-burden model had PA values of 0.85, 0.78, 0.74, and 0.76 and MAE values of 80.63, 200.09, 183.97, and 156.13 from week 1 to week 4 ahead, compared with summed-forecast PA values of 0.85, 0.80, 0.69, and 0.69 and MAE values of 100.14, 200.06, 205.65, and 217.57. Prospectively, the combined-burden model had higher PA and lower MAE at all four horizons, with the largest separations at weeks 3 and 4 ahead. These results support layered respiratory forecasting: pathogen-specific forecasts remain necessary for etiologic interpretation, while combined-burden forecasts better align with hospital preparedness and shared-capacity planning.

Chinese Translation

流感、COVID-19和呼吸道合胞病毒（RSV）的共同传播造成了共享的住院负担，因为病床、工作人员、检测能力和呼吸支持是根据总需求而非单一病原体的需求进行分配的。这使得呼吸道预测既是一个目标构建问题，也是一个建模问题：特定疾病的预测保留了归因，而医院也需要一个用于综合负担的单一规划信号。我们开发了一个监督的1到4周提前预测框架，用于每周住院人数，并比较了两种总负担策略：独立生成的病原体特定预测的总和和直接预测综合住院目标。评估包括回顾性保留测试和在观察到相应住院周之前发布的前瞻性操作预测，性能通过平均绝对误差（MAE）和百分比一致性（PA；模型生成的计数与观察到的计数中较小者除以较大者）进行总结。在回顾性总负担评估中，综合负担模型的PA值为0.85、0.78、0.74和0.76，MAE值为80.63、200.09、183.97和156.13（从第1周到第4周），而总和预测的PA值为0.85、0.80、0.69和0.69，MAE值为100.14、200.06、205.65和217.57。在前瞻性评估中，综合负担模型在所有四个时间点的PA值更高，MAE值更低，尤其在第3周和第4周的差异最大。这些结果支持分层呼吸道预测：病原体特定预测仍然对于病因学解释是必要的，而综合负担预测更好地与医院的准备和共享能力规划相一致。

View on arXiv Download PDF AI Translation

cs.AI / 157 / 2610.16078

Heterogeneous Agentic Coevolution

异质代理共进化

Ye, Wenqian, Guan, Ziwei, Yang, Zi, Modi, Shivani, Wu, Cici Cong, He, Yunlong, Chen, Yuxin, Zhang, Zhengyu, Luo, Liang, Ranganathan, Venkatesh, Sun, Jianhui, Yao, Yantao, Chen, Wenlin, Zhang, Buyun, Wen, Ellie Dingqiao, Kautz, Henry, Zhang, Aidong

Abstract

Recent advances in self-improving LLM agents have demonstrated the potential of AI systems that drive their own intelligence. However, many domain-specific applications rely on specialized foundation models that possess strong domain expertise but remain dependent on human-designed improvements. Extending recursive self-improvement to these models presents a key challenge: as a specialist improves, the strategies used to advance it must also adapt. To address this challenge, we propose Heterogeneous Agentic CoEvolution (HACE), a framework that jointly evolves specialized foundation models and LLM agents through reciprocal learning. HACE improves the specialist through supervised learning while training the agent with on-policy reinforced self-training, guided by verified performance gains on the current specialist. The agent's exploration, in turn, informs subsequent specialist training, enabling both models to adapt to each other's progress. To sustain this reciprocal learning, HACE further employs evolutionary search to optimize their interaction harness, adapting how the models exchange information and select training experiences as their capabilities evolve. We evaluate HACE across three challenging domains: recommendation systems, protein function prediction in biology, and crystal stability prediction in material discovery. We show that HACE outperforms the evaluated self-improving methods on the primary task metrics across all three domains. These results shed light on heterogeneous coevolution as a universal approach to improving specialized foundation models through reciprocal interaction with a learning agent.

Chinese Translation

近期自我改进的大型语言模型（LLM）代理的进展展示了能够推动自身智能的人工智能系统的潜力。然而，许多特定领域的应用依赖于具有强大领域专业知识的专用基础模型，这些模型仍然依赖于人类设计的改进。将递归自我改进扩展到这些模型上是一个关键挑战：随着专家的改进，推动其进步的策略也必须适应。为了解决这一挑战，我们提出了异质代理共进化（Heterogeneous Agentic CoEvolution, HACE）框架，该框架通过相互学习共同进化专用基础模型和LLM代理。HACE通过监督学习改善专家，同时通过基于策略的强化自我训练来训练代理，训练过程受到当前专家验证性能提升的指导。代理的探索反过来又为后续的专家训练提供信息，使得两个模型能够适应彼此的进展。为了维持这种相互学习，HACE进一步采用进化搜索来优化它们的交互机制，调整模型交换信息和选择训练经验的方式，以适应其能力的演变。我们在三个具有挑战性的领域评估了HACE：推荐系统、生物学中的蛋白质功能预测，以及材料发现中的晶体稳定性预测。我们展示了HACE在所有三个领域的主要任务指标上优于评估的自我改进方法。这些结果为异质共进化作为通过与学习代理的相互作用来改善专用基础模型的普遍方法提供了新的视角。

View on arXiv Download PDF AI Translation

cs.AI / 158 / 2610.16084

The Sharp Directions Are Against You: Curvature Analysis of Activation Steering

尖锐方向与你相悖：激活引导的曲率分析

Azizi, Seyedarmin, Fayyazi, Arya, Razmara, Parsa, Potraghloo, Erfan Baghaei, Kundu, Souvik, Pedram, Massoud

Abstract

Activation steering adds a vector to a language model's internal representations at inference time. It is a lightweight alternative to fine-tuning for behavioral control. The standard construction, contrastive activation addition (CAA), is fragile, as on many inputs it shifts behavior in the wrong direction. To explain this fragility, we first study the geometry of the steering vector. We introduce steering Hessian, the matrix of second order derivative of the model's behavioral loss with respect to the steering vector. It captures the output behavior response sensitivity of the model under small perturbations of the steering vector along each direction. Its top eigenvectors are sharp directions: tiny perturbations along them can cause large, erratic, input-dependent behavioral swings. Bottom eigenvectors are flat directions: behavior is robust along them. Removing the sharp component of a steering vector approximately preserves its direction and magnitude (cosine > 0.974, norm reduction < 3% for typical settings) while reducing the vector's directional curvature by up to 12x (median 4x across 36 conditions). We evaluate on three model families (Qwen2.5-7B, Llama-3.1-8B, Mistral-7B), with four behavioral tasks, and 36 token-level conditions. Steering along sharp directions reverses the intended behavior in every condition tested, with anti-steer rates of 53-100%. Sharp directions account for only 1-5% of the CAA vector's squared norm. However, they drive a disproportionate share of the gap between the steered and unsteered next-token distributions during autoregressive generation. We term this one-line correction as sharp removal; it outperforms standard CAA in open-ended generation across all 25 conditions tested. A jailbreaking experiment further shows that sharp directions act as noise rather than signal: they disrupt coherent behavioral control regardless of steering polarity.

Chinese Translation

激活引导在推理时向语言模型的内部表示添加一个向量。它是行为控制的一种轻量级替代方案，相较于微调。标准构造方法——对比激活添加（Contrastive Activation Addition, CAA）是脆弱的，因为在许多输入上，它会将行为转向错误的方向。为了解释这种脆弱性，我们首先研究引导向量的几何特性。我们引入了引导海森矩阵（steering Hessian），即模型行为损失对引导向量的二阶导数矩阵。它捕捉了模型在沿每个方向的小扰动下的输出行为响应敏感性。其特征向量中，顶部特征向量是尖锐方向：沿这些方向的小扰动可以导致大的、不稳定的、依赖输入的行为波动。底部特征向量是平坦方向：沿这些方向的行为是稳健的。去除引导向量的尖锐成分大致保留其方向和大小（在典型设置下，余弦相似度 > 0.974，范数减少 < 3%），同时将向量的方向曲率降低最多12倍（在36种条件下的中位数为4倍）。我们在三种模型系列（Qwen2.5-7B, Llama-3.1-8B, Mistral-7B）上进行了评估，涵盖四个行为任务和36个标记级条件。沿尖锐方向的引导在每种测试条件下都逆转了预期的行为，反向引导率为53-100%。尖锐方向仅占CAA向量平方范数的1-5%。然而，它们在自回归生成过程中驱动了引导和未引导的下一个标记分布之间的不成比例差距。我们将这种单行修正称为尖锐去除；它在所有25个测试条件下的开放式生成中优于标准CAA。一项越狱实验进一步表明，尖锐方向充当噪声而非信号：无论引导极性如何，它们都会破坏一致的行为控制。

View on arXiv Download PDF AI Translation

cs.AI / 159 / 2610.16108

GUIDE: Computationally Augmented Consensus for Trustworthy Clinical Practice Guidelines

GUIDE：用于可信临床实践指南的计算增强共识

Lin, Yuanfan, Di, Jie, He, Xujin, Lin, Zhenzhe, Chen, Jialing, Li, Ruiyang, Zheng, Changyu, Shang, Yuanrui, Peng, Weiqin, Qin, Jingxi, Wang, Xiong, Ye, Dongxin, Huang, Yining, Li, Guoqiang, Xu, Andi, Yu, Sijia, Wang, Ruixin, Yang, Yahan, Wu, Xiaohang, Lin, Duoru, Chen, Jingjing, Liu, Zhenzhen, Chen, Wenben, Lin, Haotian

Abstract

Maintaining up-to-date clinical practice guidelines (CPGs) is essential for evidence-based medicine, but rapid evidence growth and complex expert consensus formation limit timely updates. We developed GUIDE, a multi-agent copilot system for human-AI collaborative guideline generation and updating. GUIDE delegates labor-intensive tasks to specialized AI agents while preserving expert judgment and final arbitration. In 12 NICE ophthalmology datasets, GUIDE achieved 88.85% literature recall, 86.06% alignment in certainty-of-evidence ratings, and 91.67% recommendation agreement, requiring 162.17 minutes per dataset on average (28.68% human intervention). Performance remained stable across 20 additional datasets from cardiology, endocrinology, gastroenterology, and respiratory medicine. Ablation analysis showed that removing expert intervention reduced performance in judgment-intensive steps such as Risk-of-Bias assessment. In six post-cutoff datasets, GUIDE maintained robust end-to-end performance on unseen evidence. Prospective real-world validation showed that GUIDE reduced expert workload and produced more clinically acceptable recommendations than the original decade-old guideline (mean score, 3.73 vs. 1.53; agreement, 66.7% vs. 13.3%), with outcomes comparable to human-only updating (P = 0.588) but substantially less time and effort. GUIDE offers a practical framework for timely, trustworthy, expert-centric guideline development and validates computationally augmented consensus.

Chinese Translation

维护最新的临床实践指南（CPGs）对于基于证据的医学至关重要，但快速增长的证据和复杂的专家共识形成限制了及时更新。我们开发了GUIDE，一个用于人机协作指南生成和更新的多智能体副驾驶系统。GUIDE将劳动密集型任务委派给专门的AI代理，同时保留专家判断和最终裁决。在12个NICE眼科数据集中，GUIDE实现了88.85%的文献召回率、86.06%的证据确定性评级一致性和91.67%的推荐一致性，平均每个数据集需要162.17分钟（人类干预占28.68%）。在来自心脏病学、内分泌学、胃肠病学和呼吸医学的20个额外数据集中，性能保持稳定。消融分析显示，去除专家干预会降低在判断密集步骤（如偏倚风险评估）中的性能。在六个截止后数据集中，GUIDE在未见证据上保持了稳健的端到端性能。前瞻性的现实世界验证表明，GUIDE减少了专家的工作负担，并产生了比原始十年前指南更具临床可接受性的推荐（平均分，3.73 vs. 1.53；一致性，66.7% vs. 13.3%），其结果与仅由人类更新的情况相当（P = 0.588），但所需时间和精力大大减少。GUIDE为及时、可信、以专家为中心的指南开发提供了一个实用框架，并验证了计算增强共识的有效性。

View on arXiv Download PDF AI Translation

cs.AI / 160 / 2610.16110

Constructed regional photovoltaic generation fields improve ultra-short-term distributed solar forecasting

构建区域光伏发电场提升超短期分布式太阳能预测

Chen, Dayin, Shi, Xiaodan, Xu, Xin, Su, Zhongyue, Zhu, Shibo, Zhang, Dongxiao, Chen, Yuntian, Yan, Jinyue

Abstract

Ultra-short-term forecasting of densely distributed photovoltaic (PV) systems remains challenging because external meteorological and satellite products may not adequately resolve rapid local variations, while heterogeneous PV systems exhibit different generation responses under similar atmospheric conditions. Meanwhile, dense PV fleets provide spatially distributed measurements that reveal regional generation dynamics. Here, we reformulate fleet-level PV forecasting from modeling discrete station time series to predicting explicit, spatially structured regional PV generation fields. We develop SatPV-Field, a regional-to-local forecasting framework that transforms distributed PV measurements into image-like generation fields and models their spatiotemporal evolution. By combining historical PV fields with satellite-derived future shortwave radiation guidance and meteorological forecasts, the framework predicts future regional generation states, which subsequently provide location-specific guidance for individual-station forecasting. Experiments involving 496 distributed PV systems, including 446 source stations and 50 held-out target stations, demonstrate that these regional fields can be effectively forecast and used to improve station-level predictions. The proposed approach reduces station-level MAE by approximately 10.0% relative to the best-performing single-station temporal baseline and by 2.5% relative to a graph-based regional forecasting approach using the same source fleet. These findings establish an explicit spatial-field modeling paradigm that connects regional generation dynamics with individual-site forecasting.

Chinese Translation

超短期预测密集分布的光伏（PV）系统仍然面临挑战，因为外部气象和卫星产品可能无法充分解决快速的局部变化，而异质光伏系统在相似的气象条件下表现出不同的发电响应。同时，密集的光伏阵列提供了空间分布的测量数据，揭示了区域发电动态。在此，我们将舰队级光伏预测从建模离散站点时间序列重新构建为预测明确的、空间结构化的区域光伏发电场。我们开发了SatPV-Field，一个区域到地方的预测框架，将分布式光伏测量转化为类似图像的发电场，并建模其时空演变。通过结合历史光伏场与卫星衍生的未来短波辐射指导和气象预测，该框架预测未来区域发电状态，随后为单个站点预测提供特定位置的指导。涉及496个分布式光伏系统的实验，包括446个源站和50个保留的目标站，表明这些区域场可以有效预测并用于改善站点级预测。所提出的方法相较于表现最佳的单站时间基线，减少了约10.0%的站点级平均绝对误差（MAE），相较于使用相同源舰队的基于图的区域预测方法，减少了2.5%。这些发现建立了一个明确的空间场建模范式，将区域发电动态与单个站点预测连接起来。

View on arXiv Download PDF AI Translation

cs.AI / 161 / 2610.16117

Beyond Perfect Teammates: Multi-Agent Collaboration with Unreliable Colleagues

超越完美队友：与不可靠同事的多智能体协作

Huang, Chen, Zhang, Yiyang, Wu, Yuhao, Deng, Yang, Zhang, Wenxuan

Abstract

LLM-based multi-agent collaboration is increasingly used to solve complex tasks by dividing work among specialized agents and integrating complementary expertise. Yet existing systems and evaluations largely rest on an implicit perfect-colleague assumption: collaborators provide correct and complete information and respond promptly to requests. In realistic teamwork, however, collaborators may return information that does not answer the intended question, provide insufficient detail to support a decision, or fail to respond to a request. To separate task-solving competence from the ability to recover an unreliable evidence-gathering process, we introduce the Unreliable Colleagues Benchmark (UnCoBench), which organizes behavioral failures through Misaligned, Incomplete, and Missed settings. The main agent need to verify contextual correctness, diagnose information gap and track unresolved dependencies in order to obtain the correct evidence for solving the task. We also introduce a verifiable and scalable automated generation pipeline that pairs LLM-based scenario generation with executable checks of answer uniqueness and role necessity. Results across ten models show that similar reliable-setting performance can result in different robustness to unreliable evidence. Trace analysis reveals frequent omissions of corrective follow-up, and the recovery behavior involves inefficient waiting and redundant queries. These findings show that robust collaboration requires not only sound reasoning, but also targeted action to diagnose evidence failures, recover relevant information, and integrate it into decisions. Our data and source code are available at https://github.com/iNLP-Lab/UnCoBench.

Chinese Translation

基于大型语言模型（LLM）的多智能体协作越来越多地用于通过将工作分配给专业化的智能体并整合互补的专业知识来解决复杂任务。然而，现有系统和评估在很大程度上基于一种隐含的完美同事假设：合作者提供正确且完整的信息，并及时响应请求。然而，在现实的团队合作中，合作者可能会返回无法回答预期问题的信息，提供不足以支持决策的细节，或未能对请求作出回应。为了将任务解决能力与恢复不可靠证据收集过程的能力区分开来，我们引入了不可靠同事基准（Unreliable Colleagues Benchmark，UnCoBench），该基准通过不对齐、信息不完整和遗漏设置来组织行为失误。主要智能体需要验证上下文的正确性，诊断信息缺口，并跟踪未解决的依赖关系，以获取正确的证据来解决任务。我们还引入了一个可验证且可扩展的自动生成管道，将基于LLM的场景生成与答案唯一性和角色必要性的可执行检查相结合。十个模型的结果表明，类似的可靠设置性能可能导致对不可靠证据的不同鲁棒性。追踪分析显示，纠正后续的遗漏频繁，恢复行为涉及低效的等待和冗余查询。这些发现表明，稳健的协作不仅需要合理的推理，还需要针对性行动来诊断证据失效、恢复相关信息并将其整合到决策中。我们的数据和源代码可在 https://github.com/iNLP-Lab/UnCoBench 获取。

View on arXiv Download PDF AI Translation

cs.AI / 162 / 2610.16125

Imitation Holds a Key to Evolution: Localized Repair and Verified Skill Acquisition for Adaptive Agents

模仿是进化的关键：适应性智能体的局部修复与验证技能获取

Hu, Kai, Huang, Hua

Abstract

An agent can retain successful trajectories without learning which part of an experience should repair a new discrepancy. We formulate passive imitation (PI) as an architectural loop that localizes a violated contract, extracts a local goal, proposes an effect-matched skill, tests a small patch, and retains evidence after full-task verification. We instantiate this loop as a deterministic symbolic controller and analyze its finite repair-query complexity under explicit identifiability and noninterference assumptions. Across ten simulated tool worlds, PI solves all 600 cold held-out composition tasks with three allowed small trials, compared with 6.17% for a whole-trajectory control, while an equally expressive compositional repair control matches PI. On an unseen tool revision, PI achieves 70.0% success versus 24.0% without small trials and 79.0% with self-trial enumeration; an exploratory 24-task adaptation stream subsequently raises first-submission success to 100%, supporting a limited correction-and-retention mechanism rather than a unique incentive advantage or general intelligence. The CPU-only implementation, integrity checks, and reproducible results are available at https://github.com/FriendOfYishan/PI.

Chinese Translation

一个智能体可以在不学习经验中哪些部分应修复新差异的情况下保留成功的轨迹。我们将被动模仿（Passive Imitation, PI）形式化为一个架构循环，该循环定位被违反的契约，提取局部目标，提出效果匹配的技能，测试小范围的修补，并在全任务验证后保留证据。我们将这个循环实例化为一个确定性的符号控制器，并在显式可识别性和非干扰假设下分析其有限修复查询复杂性。在十个模拟工具世界中，PI在允许三次小试验的情况下解决了所有600个冷保留组合任务，而整体轨迹控制的成功率仅为6.17%。同样具有表现力的组合修复控制与PI相匹配。在未见过的工具修订中，PI的成功率为70.0%，而没有小试验时为24.0%，使用自我试验枚举时为79.0%；随后进行的24任务适应流将首次提交的成功率提高到100%，支持一种有限的修正与保留机制，而非独特的激励优势或一般智能。仅使用CPU的实现、完整性检查和可重复的结果可在 https://github.com/FriendOfYishan/PI 获取。

View on arXiv Download PDF AI Translation

cs.AI / 163 / 2610.16140

Large Language Model Assisted Transportation Planning through a Differentiable User Equilibrium Layer

通过可微分用户均衡层辅助的大型语言模型交通规划

Lee, Eunseung, Kim, Gyeongjun, Sohn, Keemin

Abstract

Transportation planning has traditionally required substantial expertise in network modeling, traffic assignment, and optimization. As a result, strategic analyses such as network design, capacity expansion, and model calibration remain inaccessible to non-expert decision makers. Recent advances in large language models (LLMs) provide an opportunity to bridge this gap by enabling natural-language interaction with complex transportation planning tools. However, existing LLM-based systems primarily serve as interfaces to simulation software and lack the ability to efficiently search, evaluate, and optimize transportation solutions. This study proposes an integrated human-AI transportation planning framework that combines a large language model, a model context protocol (MCP) interface, an open-source transportation planning engine (AequilibraE), and a differentiable user equilibrium (UE) assignment layer. The framework allows non-expert users to execute network analyses and improve solutions for optimization. A differentiable UE layer is developed by directly differentiating the Karush-Kuhn-Tucker (KKT) conditions of the Beckmann traffic assignment problem. The proposed framework is demonstrated through two representative planning tasks: calibration of travel time functions from observed traffic counts and network design under budget constraints. The examples illustrate how differentiable transportation models and LLM agents can jointly democratize transportation planning and substantially reduce barriers between domain expertise and transportation decision making.

Chinese Translation

交通规划传统上需要在网络建模、交通分配和优化方面具备相当的专业知识。因此，网络设计、容量扩展和模型校准等战略分析对非专业决策者来说仍然难以接触。最近大型语言模型（LLMs）的进展为弥合这一差距提供了机会，使得与复杂交通规划工具的自然语言交互成为可能。然而，现有的基于LLM的系统主要作为仿真软件的接口，缺乏有效搜索、评估和优化交通解决方案的能力。本研究提出了一种集成人工智能与人类的交通规划框架，该框架结合了大型语言模型、模型上下文协议（MCP）接口、开源交通规划引擎（AequilibraE）以及可微分用户均衡（UE）分配层。该框架使非专业用户能够执行网络分析并改善优化解决方案。通过直接对贝克曼交通分配问题的卡鲁什-库恩-塔克（KKT）条件进行微分，开发了可微分的UE层。通过两个代表性的规划任务展示了所提出的框架：从观察到的交通流量校准旅行时间函数和在预算约束下的网络设计。这些例子说明了可微分交通模型和LLM代理如何共同实现交通规划的民主化，并大幅降低领域专业知识与交通决策之间的障碍。

View on arXiv Download PDF AI Translation

cs.AI / 164 / 2610.16141

Phantom Merge: When Your Large Language Model Agents Pick One but Tell You About Another

幻影合并：当你的大型语言模型代理选择一个但告诉你另一个

Zhu, Wangshu, Guo, Qianwen, Dong, Yushun

Abstract

LLM agents increasingly operate over multiple entities, where reliable response generation requires not only retrieving relevant evidence but also preserving its association with the correct entity. We identify Phantom Merge (PM), an entity-evidence binding failure in which an agent attributes unsupported claims to its committed entity, potentially even when the task is completed correctly. We formalize PM at the claim level and characterize its prevalence and error subtypes across four agent domains and four backbone families. To address this problem, we propose a two-stage framework for claim-level risk estimation and correction. First, Anchor-Grounded Risk (AGR) combines a representational readout of the agent's pre-emission hidden state with Jacobian-based functional information, providing complementary slot-level and anchor-value risk scores. Second, Fixed-Anchor Correction (FAC) audits flagged claims and selectively revises them using observed anchor evidence or removes them when no supported replacement is available. On the held-out Shopping detection cohort, AGR-slot achieves an AUROC of 96.2% and an average precision of 97.3%, outperforming the evaluated detection baselines. In the end-to-end Shopping mitigation evaluation, FAC reduces trajectory-level PM from 51.4% to 6.2%, with an output-to-original claim ratio of 80.7%. These results demonstrate the prevalence of entity-specific grounding failures and the effectiveness of combining internal risk estimation with evidence-constrained claim correction. Our code is publicly available at https://github.com/LabRAI/PhantomMerge.

Chinese Translation

大型语言模型（LLM）代理越来越多地在多个实体上操作，其中可靠的响应生成不仅需要检索相关证据，还需要保持与正确实体的关联。我们识别出幻影合并（Phantom Merge, PM），这是一种实体-证据绑定失败，其中代理将不支持的声明归因于其承诺的实体，即使任务正确完成也可能如此。我们在声明层面上形式化PM，并描述其在四个代理领域和四个基础模型家族中的普遍性及错误子类型。为了解决这个问题，我们提出了一种两阶段框架用于声明级风险估计和修正。首先，锚定基础风险（Anchor-Grounded Risk, AGR）结合了代理在发出响应前的隐藏状态的表征读出与基于雅可比的功能信息，提供互补的槽级和锚值风险评分。其次，固定锚定修正（Fixed-Anchor Correction, FAC）审核标记的声明，并使用观察到的锚证据选择性地修正它们，或在没有支持替代时将其删除。在保留的购物检测队列中，AGR-槽达到了96.2%的AUROC和97.3%的平均精度，超越了评估的检测基线。在端到端购物缓解评估中，FAC将轨迹级PM从51.4%降低到6.2%，输出与原始声明的比率为80.7%。这些结果展示了实体特定基础失败的普遍性，以及将内部风险估计与证据约束的声明修正相结合的有效性。我们的代码已公开发布在 https://github.com/LabRAI/PhantomMerge。

View on arXiv Download PDF AI Translation

cs.AI / 165 / 2610.16145

Pollard: Governed Semantic Execution Trees for Framework-Independent AI Agent Control and Replay

Pollard：用于框架独立的人工智能代理控制和回放的受控语义执行树

Syed, Muntaser, Silaghi, Marius C.

Abstract

Agent frameworks can persist checkpoints, trace tool calls, and limit resource use, but these facilities often assign different identities to execution, governance, and replay. We present Pollard, a framework-independent runtime based on governed semantic execution trees. A step identity commits to canonical input, kind, attempt, and parent identity; the same tree records completion, refusal, resource settlement, and branch history. Strict replay returns a verified stored result without invoking the wrapped operation, while a transactional store coordinates reservations across concurrent workers. Under a signed deterministic freeze, two state-model replicas each executed the evaluation transitions without an invariant violation, retained-lease PostgreSQL groups were tested against exact limits, strict replay avoided external calls, and all semantic mutations were detected with no false alarms on the controls. A multi-pair framework matrix preserved reference behavior and verified the trees. A separate cost-capped hosted study exercised live provider paths, and every completed Pollard trial produced a verified tree. Together, the studies support the core mechanisms under explicit lease, boundary, and seal-custody assumptions.

Chinese Translation

代理框架可以持久化检查点、追踪工具调用并限制资源使用，但这些功能往往对执行、治理和回放赋予不同的身份。我们提出了Pollard，一种基于受控语义执行树的框架独立运行时。步骤身份承诺于规范输入、类型、尝试和父身份；同一树记录完成、拒绝、资源结算和分支历史。严格回放返回经过验证的存储结果，而不调用封装操作，同时事务存储协调并发工作者之间的预留。在签名确定性冻结下，两个状态模型副本各自执行评估转换而未违反不变性，保留租约的PostgreSQL组在精确限制下进行了测试，严格回放避免了外部调用，所有语义变更均被检测到且控制没有误报。一个多对框架矩阵保留了参考行为并验证了树结构。一个单独的成本上限托管研究测试了实时提供者路径，每个完成的Pollard试验都生成了经过验证的树。综合这些研究支持了在明确租约、边界和封印保管假设下的核心机制。

View on arXiv Download PDF AI Translation

cs.AI / 166 / 2610.16155

Uncovering Robustness Weaknesses in LLM Agents via Persona Evolution with Task-Integrity Constraints

通过任务完整性约束下的人格演化揭示大型语言模型代理的鲁棒性弱点

Xie, Yutao, Yang, Jingbo, Yao, Guanyu, Yang, Xiaoyu, Ding, Duo, Chang, Shiyu

Abstract

Interactive-agent evaluations commonly rely on standardized LLM-based user simulators that exhibit broadly cooperative behavior, potentially overlooking failures that arise under diverse but realistic user interactions. This can lead to overly optimistic estimates of agent robustness and provide limited failure signals for training agents that generalize across users. Existing approaches largely rely on predefined user behaviors, which may miss agent- or domain-specific weaknesses, while failure-oriented search can reduce task success by invalidating the underlying task rather than exposing genuine robustness failures. To fill this gap, we introduce TIPER, a task-integrity-preserving persona evolution method that automatically discovers agent-specific user policies exposing robustness weaknesses while keeping the underlying task valid and solvable. TIPER evolves persistent user personas through reflection-guided behavioral mutations and task-wise Pareto selection under a task-integrity constraint. Across four interactive-agent benchmarks, TIPER yields lower held-out avg@3 scores in all 12 GPT-5.4 settings, with observed reductions of 5.8 to 33.7 percentage points, and in all 28 agent-domain pairs spanning seven agents, with a median observed reduction of 22.9 points. Further analysis yields a behavior-to-failure taxonomy that connects discovered user behaviors to recurring agent failure modes. Together, these results show that TIPER enables more diagnostic robustness evaluation of interactive agents and provides informative failure analysis for future agent robustness improvement. Code repository: https://github.com/UCSB-NLP-Chang/TIPER.

Chinese Translation

交互代理评估通常依赖于标准化的基于大型语言模型（LLM）的用户模拟器，这些模拟器表现出广泛的合作行为，可能忽视在多样化但现实的用户交互中出现的失败。这可能导致对代理鲁棒性的过于乐观的估计，并为训练能够跨用户泛化的代理提供有限的失败信号。现有的方法在很大程度上依赖于预定义的用户行为，这可能会错过代理或领域特定的弱点，而以失败为导向的搜索可能通过使基础任务失效而降低任务成功率，而不是揭示真正的鲁棒性失败。为填补这一空白，我们引入了TIPER，一种保持任务完整性的人格演化方法，能够自动发现暴露鲁棒性弱点的代理特定用户策略，同时保持基础任务的有效性和可解性。TIPER通过反思引导的行为突变和在任务完整性约束下的任务级帕累托选择，演化出持久的用户人格。在四个交互代理基准测试中，TIPER在所有12个GPT-5.4设置中都产生了更低的保留平均@3分数，观察到的减少幅度为5.8到33.7个百分点，并且在涵盖七个代理的28个代理-领域对中，观察到的中位数减少幅度为22.9分。进一步分析得出了一个行为与失败的分类法，将发现的用户行为与反复出现的代理失败模式联系起来。这些结果表明，TIPER能够对交互代理进行更具诊断性的鲁棒性评估，并为未来的代理鲁棒性改进提供有益的失败分析。代码库：https://github.com/UCSB-NLP-Chang/TIPER。

View on arXiv Download PDF AI Translation

cs.AI / 167 / 2610.16167

Interactive 4D Volumetric Liquid Forecasting under Moving-Solid Interaction

移动固体交互下的互动4D体积液体预测

Ryu, Jongwon, Yoon, Sunjae, Kim, Junyeong

Abstract

We propose the task of Interactive 4D Volumetric Liquid Forecasting: predicting the spatiotemporal evolution of a future dense 3D liquid velocity field conditioned on how a solid object moves through it. This task provides an upstream physical forecasting primitive for systems that need to reason about motion-conditioned liquid response, including robotic liquid manipulation, digital twins, and rigid-fluid video generation. Existing works have made efforts in predicting flow around static or prescribed solid geometries and synthesizing visually plausible fluid motion. Yet these approaches provide limited coverage for interactive liquid dynamics, as they are not capable of forecasting how a moving solid induces a future full-domain liquid response. In this forecasting scenario, the hydrodynamic effect is condensed near an evolving liquid-solid interface, but the prediction target is a global spatiotemporal rollout. To address this challenge, we propose a two-level recurrent forecasting framework, TIDE/TIDES: TIDE isolates localized interface-driven forecasting, while TIDES adds transport-aware coupling and explicit stabilization to reduce accumulated rollout error. Across adapted neighboring baselines, controlled component ablations, motion/size OOD tests, and long-horizon evaluation, TIDES achieves the best forecasting performance and the ablations identify localized interaction, transport coupling, and stabilization as the key contributors. Together, our task formulation, benchmark, and TIDE/TIDES study establish a controlled foundation for moving-solid-conditioned dense liquid forecasting. The code is publicly available at https://github.com/jongwonryu/TIDE-TIDES.

Chinese Translation

我们提出了互动4D体积液体预测的任务：预测未来密集3D液体速度场的时空演变，该预测以固体物体在液体中运动的方式为条件。该任务为需要推理运动条件下液体响应的系统提供了一个上游物理预测原语，包括机器人液体操作、数字双胞胎和刚体-流体视频生成。现有研究已在预测静态或预设固体几何形状周围的流动以及合成视觉上可信的流体运动方面做出了努力。然而，这些方法对交互液体动力学的覆盖有限，因为它们无法预测移动固体如何引发未来全域液体响应。在这一预测场景中，流体动力学效应集中在不断演变的液体-固体界面附近，但预测目标是一个全局的时空展开。为了解决这一挑战，我们提出了一个两级递归预测框架TIDE/TIDES：TIDE隔离了局部界面驱动的预测，而TIDES则增加了运输感知耦合和显式稳定化，以减少累积的展开误差。在适应的邻近基线、受控组件消融、运动/尺寸OOD测试和长时间评估中，TIDES实现了最佳的预测性能，消融实验确定了局部交互、运输耦合和稳定化是关键贡献因素。综上所述，我们的任务构造、基准和TIDE/TIDES研究为移动固体条件下的密集液体预测建立了一个受控基础。代码已公开发布在 https://github.com/jongwonryu/TIDE-TIDES。

View on arXiv Download PDF AI Translation

cs.AI / 168 / 2610.16175

ProCARE: Real-World Study Automation via Profile-Grounded Evidence Contracts

ProCARE：基于个人资料的证据合同的真实世界研究自动化

Fan, Zhaoyu, Han, Bowen, Ou, Jincheng, Ren, Yanwei, He, Zehua, Huang, Kaiyu, He, Pengcheng, Ni, Shanshan, Gong, Chengchen, Xue, Shuai, Shi, Qingjiang

Abstract

Real-world studies (RWS) analyze routinely collected healthcare data, such as electronic health records and insurance claims, to estimate treatment effects, monitor safety, and identify risk factors outside randomized trials. Automating RWS with large language model agents is appealing but brittle: valid analyses depend on dataset-specific clinical semantics, including index dates, follow-up windows, patient/event-level units, and outcome boundaries that are often implicit in observational data. This leads to schema mismatches, temporal errors, inconsistent analytical granularity, and unsupported clinical claims. We present ProCARE, a profile-grounded framework that reformulates RWS automation as constrained compilation. ProCARE converts heterogeneous data profiles into Evidence Contracts specifying available variables, temporal attributes, distributions, and evidentiary limits, and compiles study objectives into RWS-IR, a strongly typed intermediate representation for variables, time relations, analysis units, and statistical tasks. Deterministic validation and localized repair then check and revise generated workflows across schema mapping, cohort construction, variable derivation, analysis, and interpretation. On 36 EHR-based benchmark tasks, ProCARE outperforms strong LLM-agent baselines in code executability, semantic consistency, and conclusion traceability, with automated violation diagnoses showing moderate-to-substantial agreement with blinded expert review.

Chinese Translation

真实世界研究（RWS）分析常规收集的医疗保健数据，如电子健康记录和保险索赔，以估计治疗效果、监测安全性和识别随机试验之外的风险因素。使用大型语言模型代理自动化RWS具有吸引力，但也存在脆弱性：有效分析依赖于特定数据集的临床语义，包括索引日期、随访窗口、患者/事件级别单位和结果边界，这些在观察性数据中往往是隐含的。这导致了模式不匹配、时间错误、不一致的分析粒度和不支持的临床声明。我们提出了ProCARE，这是一个基于个人资料的框架，将RWS自动化重新表述为受限编译。ProCARE将异构数据配置文件转换为证据合同，指定可用变量、时间属性、分布和证据限制，并将研究目标编译为RWS-IR，这是一种强类型的中间表示，涵盖变量、时间关系、分析单位和统计任务。确定性验证和局部修复随后检查和修订生成的工作流程，涵盖模式映射、队列构建、变量推导、分析和解释。在36个基于电子健康记录的基准任务中，ProCARE在代码可执行性、语义一致性和结论可追溯性方面超越了强大的LLM代理基线，自动化的违规诊断与盲审专家评审显示出中等到显著的一致性。

View on arXiv Download PDF AI Translation

cs.AI / 169 / 2610.16182

Label-Centric Knowledge Retrieval for Zero-Shot Text Classification

以标签为中心的零样本文本分类知识检索

Wu, Daoping, Wang, Dong, Liu, Siyu, Liang, Chen, Geng, Jian, Li, Ziqi, Xiong, Shuguang, Zhang, Ying

Abstract

Flagship large language models (LLMs) offer rich general knowledge and strong reasoning capabilities, but their high inference cost and latency limit their use for large-scale data labeling. We study zero-shot text classification with semantically meaningful labels and task descriptions, without task-specific labeled examples or classifier feedback. We propose \textbf{SLKR} (Semantic Label Knowledge Retrieval), a label-centric retrieval-augmented generation framework that separates offline knowledge construction from lightweight online inference. Offline, a flagship LLM independently generates label definitions, retrieval keywords, and boundary rules for potentially confusable label pairs, forming a reusable label-knowledge store. Because this construction operates over the label space rather than the input corpus, its one-time flagship cost can be amortized across large-scale inference. Online, hybrid semantic and lexical retrieval first identifies the top-$K$ candidate labels and then selects up to $M$ input-relevant boundary rules connecting candidate-label pairs. A lightweight LLM performs the final classification in a single call using the retrieved definitions and rules. Across three public benchmarks and a real-world multilingual book-classification dataset, SLKR achieves the best results on all four datasets with Qwen3.5-4B and on three of four with Ministral-3-3B, while using compact input contexts and no task-specific labeled examples. Boundary ablations further show that retrieving a small set of input-relevant rules preserves performance close to using all boundaries with substantially less context, supporting efficient reuse of flagship-generated label knowledge for lightweight classification.

Chinese Translation

旗舰级大型语言模型（LLMs）提供丰富的通用知识和强大的推理能力，但其高推理成本和延迟限制了其在大规模数据标注中的应用。我们研究了使用语义上有意义的标签和任务描述进行零样本文本分类，而无需特定任务的标注示例或分类器反馈。我们提出了 extbf{SLKR}（语义标签知识检索），这是一种以标签为中心的检索增强生成框架，分离了离线知识构建与轻量级在线推理。在离线阶段，旗舰级LLM独立生成标签定义、检索关键词和潜在混淆标签对的边界规则，形成可重用的标签知识库。由于这一构建是在标签空间而非输入语料库上进行的，其一次性旗舰成本可以在大规模推理中摊销。在在线阶段，混合语义和词汇检索首先识别出前$K$个候选标签，然后选择最多$M$个与输入相关的边界规则，连接候选标签对。一个轻量级LLM使用检索到的定义和规则在一次调用中执行最终分类。在三个公共基准和一个真实世界的多语言书籍分类数据集上，SLKR在所有四个数据集上都取得了Qwen3.5-4B的最佳结果，并在四个数据集中的三个上取得了Ministral-3-3B的最佳结果，同时使用紧凑的输入上下文且没有特定任务的标注示例。边界消融实验进一步表明，检索一小组与输入相关的规则可以在保持性能接近于使用所有边界的情况下，显著减少上下文，从而支持旗舰生成的标签知识在轻量级分类中的高效重用。

View on arXiv Download PDF AI Translation

cs.AI / 170 / 2610.16191

Correct Choices, Failed Sequences: Choice2Seq Benchmarks Correctness Retention Across Output Interfaces

正确选择，失败序列：Choice2Seq基准测试输出接口中的正确性保留

Lai, Tianyou, Yue, Wentao, Deng, Zhiying, Li, Qilei

Abstract

A model may select the correct answer yet fail to construct it through a sequence of local decisions. We introduce Choice2Seq, a diagnostic benchmark that measures correctness retention across complete-answer selection and sequential construction in independent runs on the same problems and candidate sets. Sequence Composition Rate (SCR) measures the fraction of correctly selected items that are also answered correctly through local construction. Jev achieves 97.1% and 99.5% selection accuracy on arithmetic and lexical retrieval, yet retains only 33.0% and 2.1% of those correct choices under serial construction, respectively. Substantial gaps also emerge for StartLux and Qwen, including when Qwen emits characters directly. Restricting local options to candidate-supported characters improves construction, while explicit position indexing improves copying fidelity, even for incorrect selections. These interventions distinguish task correctness, plan fidelity, and output compliance. Matched dialogue-state evaluations reveal that local field construction largely preserves Jev's correct states but reduces Qwen's mean-turn accuracy, highlighting task- and interface-dependent differences. Sequence-level confidence and request-cost analyses further characterize completed outputs.

Chinese Translation

一个模型可能选择了正确的答案，但却未能通过一系列局部决策构建出该答案。我们引入了Choice2Seq，这是一个诊断基准，旨在测量在独立运行中针对相同问题和候选集的完整答案选择与顺序构建之间的正确性保留。序列构成率（Sequence Composition Rate, SCR）衡量的是正确选择项中通过局部构建也被正确回答的比例。Jev在算术和词汇检索中分别达到了97.1%和99.5%的选择准确率，但在序列构建下仅保留了33.0%和2.1%的正确选择。StartLux和Qwen也出现了显著的差距，包括当Qwen直接发出字符时。将局部选项限制为候选支持的字符可以改善构建，而显式位置索引则提高了复制的准确性，即使对于错误选择也是如此。这些干预措施区分了任务的正确性、计划的忠实度和输出的一致性。匹配的对话状态评估显示，局部字段构建在很大程度上保留了Jev的正确状态，但降低了Qwen的平均回合准确率，突显了任务和接口依赖的差异。序列级别的信心和请求成本分析进一步描述了完成的输出。

View on arXiv Download PDF AI Translation

cs.AI / 171 / 2610.16219

OneShot: Label-Efficient Detector Weighting for Time Series Anomaly Detection

OneShot：用于时间序列异常检测的标签高效检测器加权

Lee, Seunghan, Seo, Jun, Lee, Jaehoon, Kang, Junhyeok, Han, Sangjun, Yoo, Sungdong, Kim, Minjae, Lim, Tae Yoon, Kang, Dongwan, Choi, Hwanil, Lee, Soonyoung, Ahn, Wonbin

Abstract

Time series (TS) anomaly detection (AD) lacks a universal detector, with the best choice determined by the series rather than by the algorithm. Existing methods therefore either select one detector for each series or average a pool of them. However, both assume that no label ever arrives, while a deployed monitor does receive them, since an operator confirms the alarms it raises. In this paper, we propose ONESHOT, which uses a single confirmed anomaly to weight the pool rather than to choose one detector from it. Specifically, we score each detector on the confirmed anomaly, and those scores become the weights of a soft mixture over the whole pool. We show that a single confirmed anomaly is enough to come close to the best detector for the series. Across 152 univariate and 73 multivariate series, ONESHOT outperforms the strongest single detector, averaging the whole pool, and even picking the best detector on the confirmed event. Code is available at: https://github.com/seunghan96/one_shot.

Chinese Translation

时间序列（TS）异常检测（AD）缺乏通用的检测器，最佳选择由时间序列本身决定，而非算法。因此，现有方法要么为每个序列选择一个检测器，要么对一组检测器进行平均。然而，这两种方法都假设不会有标签到达，而实际部署的监控系统会收到标签，因为操作员会确认其发出的警报。本文提出了ONESHOT，它利用单个确认的异常来对检测器池进行加权，而不是从中选择一个检测器。具体而言，我们根据确认的异常对每个检测器进行评分，这些评分成为整个池的软混合权重。我们展示了单个确认的异常足以接近该序列的最佳检测器。在152个单变量和73个多变量序列中，ONESHOT的表现超越了最强的单一检测器、对整个池的平均值，甚至在确认事件上选择最佳检测器。代码可在以下链接获取：https://github.com/seunghan96/one_shot。

View on arXiv Download PDF AI Translation

cs.AI / 172 / 2610.16231

Learning to GUIDE the Group: Self-Learning LLM Facilitation Across Multi-Party Tasks

学习引导小组：跨多方任务的自学习大型语言模型促进

Liu, Ziyi, Kuo, Tzu-Sheng, Liu, Jiale, Fu, Deqing, Thain, Nithum, Qian, Crystal

Abstract

As large language models are increasingly adopted as multi-party communication facilitators, there is a growing demand for agents capable of balancing diverse group dynamics across contexts. However, existing LLM facilitators are often fine-tuned for task-specific objectives; generalized LLM-based facilitation methods that can adapt to different types of tasks and diverse group dynamics are underdeveloped. To address this gap, we present GUIDE (Generalized Unified Intervention for Dynamic Exchanges, a framework for developing generalized LLM-based facilitators across different multi-party contexts. GUIDE evaluates facilitation across both objective task outcomes and subjective metrics, including psychological safety and perceived friction. Within this framework, we study whether LLMs can self-learn generalized facilitation strategy offline to real-time interventions. We first collect multi-party human interactions guided by task-specific expert LLM facilitators (N=432 conversations). We then employ a self-learning method in which an LLM induces an actionable strategic playbook from conversations and iteratively refines it using feedback from an LLM judge aligned with human annotators (N=20), developing a single generalized facilitator that outperforms the three task-specific expert facilitators on held-out conversations 61.4% win rate. Finally, we evaluate these improvements empirically; in a real-time online empirical study, the self-learned generalized facilitator preserves facilitator performance and group productivity, improves speaking comfort and reduces interpersonal friction, while using 24.7% fewer turns and 29.0% fewer words than the expert prompts. These findings provide empirical evidence that LLMs can learn generalizable, task-agnostic social facilitation from task-specific human interactions, on-par with hand-crafted task-specific prompts across diverse group settings.

Chinese Translation

随着大型语言模型（LLM）越来越多地被用作多方沟通的促进者，对能够在不同背景下平衡多样化小组动态的代理的需求也在不断增长。然而，现有的LLM促进者通常是针对特定任务目标进行微调的；能够适应不同类型任务和多样化小组动态的通用LLM促进方法尚未得到充分发展。为了解决这一问题，我们提出了GUIDE（Generalized Unified Intervention for Dynamic Exchanges，动态交流的通用统一干预框架），旨在开发适用于不同多方背景的通用LLM促进者。GUIDE在客观任务结果和主观指标（包括心理安全感和感知摩擦）之间评估促进效果。在这一框架内，我们研究LLM是否能够自学习通用的促进策略，从离线到实时干预。我们首先收集了由任务特定的专家LLM促进者指导的多方人际互动（N=432次对话）。然后，我们采用一种自学习方法，通过对话生成可操作的战略手册，并利用与人类标注者一致的LLM评审者的反馈进行迭代优化（N=20），开发出一个通用的促进者，其在保留对话中超过三位任务特定专家促进者的情况下，赢得率达到61.4%。最后，我们通过实证研究评估这些改进；在一项实时在线实证研究中，自学习的通用促进者保持了促进者的表现和小组的生产力，提高了发言舒适度，并减少了人际摩擦，同时比专家提示使用了24.7%的更少轮次和29.0%的更少字数。这些发现提供了实证证据，表明LLM能够从任务特定的人际互动中学习可推广的、与任务无关的社会促进，且在多样化小组环境中与手工制作的任务特定提示相当。

View on arXiv Download PDF AI Translation

cs.AI / 173 / 2610.16260

ScreenASR-Bench: A Tiered Benchmark for Speech Recognition with Visual Context

ScreenASR-Bench：一种分层的视觉上下文语音识别基准

Fu, Mingwei, Song, Zheshu, Zhou, Zhikai, Huang, Shen, Shang, Shidong, Zhu, Mengxiao

Abstract

Current ASR models perform well on general benchmarks but still struggle with acoustically ambiguous words. Visual context can aid recognition in screen-based scenarios, yet existing benchmarks provide limited insight into the benefits of different types of visual evidence. To address this gap, we introduce ScreenASR-Bench , a tiered benchmark with three progressively challenging levels: Explicit Textual Evidence (ETE), based on direct on-screen text; Explicit Visual Evidence (EVE), using visible content rather than on-screen text; and Implicit Semantic Evidence (ISE), requiring scene-level inference. Extensive experiments on ASR models, Large Audio-Language Models (LALMs) and Omni models show that incorporating visual context substantially improves recall over the audio-only setting across all three model families. The gains are largest for ETE and smallest for ISE, indicating that models leverage direct visual evidence more effectively than indirect semantic cues. To facilitate future research on visually grounded speech recognition, we publicly release ScreenASR-Bench.

Chinese Translation

当前的自动语音识别（ASR）模型在一般基准测试中表现良好，但在处理声学模糊词汇时仍然存在困难。视觉上下文可以在基于屏幕的场景中辅助识别，但现有基准对不同类型视觉证据的益处提供的见解有限。为了解决这一问题，我们引入了ScreenASR-Bench，这是一种具有三个逐步挑战级别的分层基准：显式文本证据（Explicit Textual Evidence, ETE），基于直接的屏幕文本；显式视觉证据（Explicit Visual Evidence, EVE），使用可见内容而非屏幕文本；以及隐式语义证据（Implicit Semantic Evidence, ISE），需要场景级推理。对ASR模型、大型音频语言模型（Large Audio-Language Models, LALMs）和全能模型（Omni models）进行的广泛实验表明，结合视觉上下文显著提高了所有三种模型系列在音频仅设置下的召回率。对于ETE的提升最大，而对于ISE的提升最小，表明模型更有效地利用直接视觉证据而非间接语义线索。为了促进未来在视觉基础语音识别方面的研究，我们公开发布了ScreenASR-Bench。

View on arXiv Download PDF AI Translation

cs.AI / 174 / 2610.16261

Hardware-Friendly Token-Group Activation Quantization for Low-Bit Mamba Super-Resolution

适合硬件的基于Token组的激活量化用于低比特Mamba超分辨率

Chung, Jinwoo, Kim, Jangho

Abstract

Mamba-based super-resolution models are efficient alternatives to Transformer-based restoration backbones, but their low-bit post-training quantization remains challenging because selective scan, Mamba's input-dependent state-space operator, induces different activation ranges across spatial tokens. Thus, a single layer-wise clipping range poorly matches all tokens. We propose Token-Group Quantization, a hardware-friendly activation PTQ framework that statically approximates a hardware-unfriendly post-scan grouping oracle, where tokens with similar preferred clipping ranges are assigned to the same group. To avoid runtime post-scan grouping in static low-bit inference, a lightweight predictor infers the groups from pre-scan features available before the quantized scan path. During calibration, we form pseudo-labels by clustering post-scan activation descriptors that capture similar rounding--clipping behavior. At inference, the frozen predictor maps pre-scan features to groups and selects entries from offline-calibrated group-wise clipping-bound tables, without runtime statistics collection, dynamic clipping updates, or post-scan group assignment. To address low-bit restoration degradation, we further introduce a Frequency-Preserving Refinement (FPR) objective for clipping-bound calibration. FPR aligns Fourier amplitude and phase between full-precision teacher and quantized student outputs to preserve high-frequency structures. Across PTQ protocols, bit-widths, backbones, and tasks, our method consistently outperforms prior Mamba and SR PTQ baselines, transfers to Transformer SR and Vision Mamba classification, composes with rotation-based quantization, and achieves 2.23x speedup on Jetson Orin Nano over FP16. Code is available here.

Chinese Translation

基于Mamba的超分辨率模型是Transformer基础恢复骨干的高效替代方案，但其低比特后训练量化仍然面临挑战，因为选择性扫描（selective scan）这一Mamba的输入依赖状态空间算子在空间Token之间引入了不同的激活范围。因此，单一层的裁剪范围无法有效匹配所有Token。我们提出了Token-Group量化，这是一种适合硬件的激活后训练量化（PTQ）框架，它静态地近似一个不适合硬件的后扫描分组预言机，其中具有相似偏好裁剪范围的Token被分配到同一组。为了在静态低比特推理中避免运行时后扫描分组，一个轻量级预测器从量化扫描路径之前可用的预扫描特征中推断组。在校准过程中，我们通过聚类后扫描激活描述符形成伪标签，这些描述符捕捉到相似的四舍五入-裁剪行为。在推理阶段，冻结的预测器将预扫描特征映射到组，并从离线校准的组裁剪边界表中选择条目，而无需运行时统计收集、动态裁剪更新或后扫描组分配。为了解决低比特恢复降级问题，我们进一步引入了频率保留细化（Frequency-Preserving Refinement, FPR）目标用于裁剪边界校准。FPR对齐全精度教师模型和量化学生模型输出之间的傅里叶幅度和相位，以保留高频结构。在各种PTQ协议、比特宽度、骨干网络和任务中，我们的方法始终优于先前的Mamba和SR PTQ基线，能够迁移到Transformer SR和Vision Mamba分类，能够与基于旋转的量化组合，并在Jetson Orin Nano上实现2.23倍的速度提升，相较于FP16。代码可在此处获取。

View on arXiv Download PDF AI Translation

cs.AI / 175 / 2610.16280

agent-risk: Risk Assessment for Changes to AI Agents

agent-risk：人工智能代理变更的风险评估

Zhao, Yue, Xu, Zixiang, Zhang, Angela Huyue, Yang, S. Alex, Hu, Xiyang

Abstract

Before changing a deployed AI agent (a stricter safety gate, a new model), organizations must approve the change, starting from existing logs. Extending classical risk analysis to partial records and partial changes to running systems, agent-risk replays logged episodes, assuming nothing else changes and the agent does not anticipate the change. The log settles episodes acting as before, a declared stop value settles stopped ones, and the rest need logged support, a saved-state rerun, or a bound. We prove what logs must record for exact replay and which saved states serve a gate family. For independent episodes, we bound the reruns a decision needs and tolerable drift from live behavior. In a bash sandbox, a 7,200-episode log alone decided neither of two retry gates, which let the agent act after refusals the log never shows. Rerunning only each gate's blocked episodes from saved states reached both live decisions at half the agent calls of a live test sized post hoc for them, given the log. Without per-step harm, a stopping gate's certified harm ceiling rose up to 2.5-fold, since the log hides whether harm preceded the stop. Approval thus needs evidence designed in advance: log fields, saved states, targeted reruns.

Chinese Translation

在更改已部署的人工智能代理（如更严格的安全门或新模型）之前，组织必须批准该变更，首先从现有日志入手。agent-risk 将经典风险分析扩展到部分记录和对运行系统的部分变更，通过重放已记录的事件，假设其他条件不变且代理不会预见到该变更。日志将事件的表现保持不变，声明的停止值用于处理已停止的事件，其余事件需要日志支持、保存状态的重放或界限。我们证明了日志必须记录哪些内容以实现精确重放，以及哪些保存状态适用于一个安全门家族。对于独立事件，我们界定了决策所需的重放次数和可容忍的与实时行为的偏差。在一个 bash 沙箱中，仅凭一个7200事件的日志并未决定两个重试安全门，这使得代理在日志中未显示拒绝后仍然能够行动。仅重放每个安全门的被阻止事件，从保存状态中实现了在给定日志的情况下，以一半的代理调用次数达成两个实时决策，而这些决策是在事后为它们大小调整的实时测试中得出的。在没有逐步伤害的情况下，停止安全门的认证伤害上限提高了最多2.5倍，因为日志隐藏了伤害是否在停止之前发生。因此，批准需要提前设计的证据：日志字段、保存状态和有针对性的重放。

View on arXiv Download PDF AI Translation

cs.AI / 176 / 2610.16288

REVEAL: Reverse Engineering Agentic Latents from Frontier-Model Traces

REVEAL：从前沿模型轨迹中逆向工程代理潜变量

Du, Yuetian, Wang, Meng, Wang, Yucheng, Chen, Zhangzhe, Wang, Wei, Yang, Boyu, Zhai, Weiqi, Zhao, Bing, Kong, Ming, Zhou, Zhan, Liu, Jie, Zhu, Qiang

Abstract

Frontier LLM agents perform complex interactive tasks, but their inference costs motivate transferring their experience to smaller open-source agents. Distillation and latent communication typically require target-model training or access to source-model internals. We study how textual frontier-model trajectories can provide reusable local controls for a frozen agent. Inspired by protocol reverse engineering, we introduce $\textit{Agentic Latents}$, comprising anchored states and dynamic profiles of feedback--response cycles elicited by replay in the target model. Our training-free framework, $\textbf{REVEAL}$ ($\textbf{REV}$erse $\textbf{E}$ngineering $\textbf{A}$gentic $\textbf{L}$atents), uses these profiles and trajectory outcomes to match and select local events from successful and failed rollouts of the same task and source model. Their response-transition contrasts yield a target-specific steering direction, applied before each response. Within-benchmark evaluations on Terminal-Bench 2.1 and Humanity's Last Exam with tools report higher accuracy than unmodified agents across three targets. Representation analyses and behavioral case studies support reusing frontier-model experience as local activation control.

Chinese Translation

前沿大型语言模型（LLM）代理能够执行复杂的交互任务，但其推理成本促使将其经验转移到更小的开源代理上。蒸馏和潜在通信通常需要目标模型的训练或对源模型内部的访问。我们研究了文本前沿模型轨迹如何为冻结的代理提供可重用的局部控制。受到协议逆向工程的启发，我们引入了“代理潜变量”（Agentic Latents），其包括锚定状态和通过在目标模型中重放引发的反馈-响应循环的动态特征。我们的无训练框架REVEAL（逆向工程代理潜变量）利用这些特征和轨迹结果，从同一任务和源模型的成功和失败的执行中匹配和选择局部事件。它们的响应转变对比产生了特定于目标的引导方向，在每次响应之前应用。在Terminal-Bench 2.1和人类最后考试的基准评估中，使用工具的评估报告显示其准确性高于未修改的代理，涵盖三个目标。表示分析和行为案例研究支持将前沿模型经验作为局部激活控制的重用。

View on arXiv Download PDF AI Translation

cs.AI / 177 / 2610.16298

Memory-Efficient Integer-Only Training of Deep Neural Networks

内存高效的整数训练深度神经网络

Shalby, Hazem Hesham Yousef, Pittorino, Fabrizio, Roveri, Manuel

Abstract

Training neural networks in reduced precision can yield significant efficiency gains, yet all existing methods that scale to standard computer vision benchmarks retain floating-point in at least one stage, typically the weight update or optimizer state. We present a training algorithm that scales to modern deep neural networks, in which every tensor (weights, gradients, and activations) is stored as INT8, and every operation (forward pass, backward pass, weight update, optimizer update) is performed in integer arithmetic. Specifically, tensors are stored as signed integer mantissas paired with explicit power-of-two exponents, on which integer-only operations are performed using five custom composable building blocks (products, sums, divide-by-constant, nonlinear functions, and weight updates). This allows complete implementation of trainable integer layers, including convolutions, batch and layer normalization, softmax, and GELU. A per-group velocity mechanism with independent exponent tracking preserves gradient precision across neurons and channels with diverse magnitudes. As every tensor fits in INT8, training memory is roughly 4x smaller than the FP32 baseline. We validate our training algorithm on CNNs (ResNet-18/34/50/101, MobileNetV2) and Vision Transformers (ViT-Ti / B), demonstrating for the first time that training neural networks with INT8 tensors and integer arithmetic can achieve competitive accuracy on CIFAR-10, CIFAR-100 and ImageNet at a reduced memory budget.

Chinese Translation

在降低精度的情况下训练神经网络可以带来显著的效率提升，但所有现有的方法在至少一个阶段（通常是权重更新或优化器状态）仍保留浮点数。我们提出了一种可扩展到现代深度神经网络的训练算法，其中每个张量（权重、梯度和激活）都以 INT8 存储，并且每个操作（前向传播、反向传播、权重更新、优化器更新）都在整数算术下进行。具体而言，张量以带符号整数尾数与显式的二次幂指数配对存储，在此基础上使用五个可组合的自定义构建块（乘法、加法、常数除法、非线性函数和权重更新）进行整数操作。这使得可训练的整数层的完整实现成为可能，包括卷积、批量和层归一化、softmax 和 GELU。每组速度机制与独立的指数跟踪相结合，保持了跨神经元和通道的梯度精度，适应不同的幅度。由于每个张量都适合 INT8，训练内存大约比 FP32 基线小 4 倍。我们在卷积神经网络（CNNs，ResNet-18/34/50/101，MobileNetV2）和视觉变换器（Vision Transformers，ViT-Ti / B）上验证了我们的训练算法，首次证明使用 INT8 张量和整数算术训练神经网络能够在降低内存预算的情况下，在 CIFAR-10、CIFAR-100 和 ImageNet 上实现具有竞争力的准确性。

View on arXiv Download PDF AI Translation

cs.AI / 178 / 2610.16301

FON: A Draft-Revise Dual-Organ Language Model without Sequence-Level Quadratic Attention

FON：一种无序列级二次注意力的草拟-修订双组织语言模型

Hu, Helong, Pan, Hongdan, Hu, Shuiqing

Abstract

Self-attention assigns two functions to one mechanism-where to read and how to mix-at the price of O(L^2) training, an O(T) decode state, and position signals bound to the training length. Existing attention-free alternatives all optimize the single function of generation. We propose FON (Fusion-Organ Network), which differentiates one weight stack into a drafting organ (multi-tap learnable delays + content-adaptive recursion; causal, O(S) training, O(1) decode state, no sequence-level quadratic attention) and a revision organ that re-processes the written prefix through multi-scale ripple kernels injected via weak gates, never generating. At 30M parameters, iso-parameter and iso-token budget, FON reaches perplexity 131.03 on WikiText-103, beating Transformer (197.73) and Mamba2 (152.53); the advantage persists under all depth-aligned baselines and at the 70M tier (-20.8%). At 8x length extrapolation FON degrades by ~8-9% without positional-encoding engineering, between ALiBi and RoPE. The revision channel provides a structural pathway for in-domain infilling that causal architectures lack in principle (+19% over on-demand coupling), at a perplexity cost indistinguishable within measurement precision. We report all costs and boundaries: impaired long-range retrieval, synthetic-task disadvantages, weaker out-of-domain code infilling than a same-recipe Transformer, and a 6.5-12.3x wall-clock gap.

Chinese Translation

自注意力机制将读取位置和混合方式两个功能赋予同一机制，代价是O(L^2)的训练复杂度、O(T)的解码状态，以及与训练长度绑定的位置信号。现有的无注意力替代方案均优化生成的单一功能。我们提出FON（Fusion-Organ Network），将一个权重堆叠区分为草拟器（多点可学习延迟 + 内容自适应递归；因果性，O(S)训练，O(1)解码状态，无序列级二次注意力）和修订器，通过弱门控注入的多尺度涟漪核重新处理已写前缀，且从不生成。在30M参数下，FON在WikiText-103上的困惑度达到131.03，超越了Transformer（197.73）和Mamba2（152.53）；这一优势在所有深度对齐的基线和70M层级下持续存在（-20.8%）。在8倍长度外推时，FON在没有位置编码工程的情况下降级约8-9%，介于ALiBi和RoPE之间。修订通道为领域内填充提供了结构性路径，而因果架构原则上缺乏这一点（比按需耦合提高了+19%），在测量精度内的困惑度成本不可区分。我们报告了所有成本和边界：受损的长距离检索、合成任务劣势、比同配方的Transformer更弱的领域外代码填充，以及6.5-12.3倍的实际时间差距。

View on arXiv Download PDF AI Translation

cs.AI / 179 / 2610.16312

Narrative Paradoxes in LLM Inference: Shaping Reasoning Trajectories toward Misalignment

大型语言模型推理中的叙事悖论：塑造推理轨迹以应对不一致性

Yang, Haoming, Ma, Ke, Jiang, Yangbangyan, Zhang, Ligong, Jia, Xiaojun, Huang, Longtao, Sun, Yingfei, Xu, Qianqian, Huang, Qingming

Abstract

Explicit reasoning in large language models (LLMs) is often associated with reliability, yet it can still pose risks when a model infers harmful goals under a benign context. We identify a distinct challenge: ensuring the safety of the \emph{multi-step reasoning process} itself, where locally coherent steps may still lead to globally harmful outcomes. To explore this vulnerability, we propose PDRA (Paradox-driven Reasoning-based Attack), a novel attack method that programs a harmful reasoning trajectory. PDRA first constructs a paradoxical narrative core by pairing a malicious actor (e.g., a terrorist) with a benign task (e.g., writing a safety bulletin). It then embeds this core into a structured analytical framework that forces the model through a mandatory three-stage analysis -- surface purpose, latent intent, and operational reconstruction -- guiding it to produce detailed harmful content while maintaining narrative coherence. Mechanistically, we show PDRA steers the model's internal representations away from refusal-related patterns, suppressing safety-triggering lexicons while activating planning-oriented terms. Empirically, PDRA achieves state-of-the-art attack success rates across diverse LLMs with high efficiency. \textcolor{red}{\textbf{Warning:} This paper may contain sensitive content.}

Chinese Translation

大型语言模型（LLMs）中的显式推理通常与可靠性相关，但当模型在良性背景下推断出有害目标时，仍可能带来风险。我们识别出一个独特的挑战：确保 extit{多步骤推理过程}本身的安全性，其中局部一致的步骤仍可能导致全球有害的结果。为了探讨这一脆弱性，我们提出了PDRA（基于悖论驱动的推理攻击），这是一种新颖的攻击方法，旨在编程出有害的推理轨迹。PDRA首先通过将恶意行为者（例如恐怖分子）与良性任务（例如撰写安全公告）配对，构建一个悖论叙事核心。然后，它将这一核心嵌入到一个结构化的分析框架中，强制模型进行强制性的三阶段分析——表面目的、潜在意图和操作重构——引导其生成详细的有害内容，同时保持叙事的一致性。从机制上看，我们展示了PDRA如何引导模型的内部表征远离拒绝相关模式，抑制安全触发词汇，同时激活以规划为导向的术语。从实证上看，PDRA在多种大型语言模型中以高效率实现了最先进的攻击成功率。 extcolor{red}{ extbf{警告：} 本文可能包含敏感内容。}

View on arXiv Download PDF AI Translation

cs.AI / 180 / 2610.16315

ReTAC: Residual Test-Time Agentic Computation for Video Spatial Reasoning

ReTAC：用于视频空间推理的残差测试时自主计算

Wu, Yike, Chen, Keyan, Wang, Qixiong, Xiang, Xi, Que, Maofei, Jiang, Xiaolong, Zhang, Xiaoqin, Shao, Ling, Lu, Shijian

Abstract

Test-time agentic computation can improve multimodal large language models (MLLMs) through planning, tool use, iterative perception, and reflection, but indiscriminate computation may duplicate resolved work, introduce noisy evidence, and perturb correct predictions. We propose that such computation should be residual: allocate additional computation only to requirements unresolved by the base model and, after a localized failure, recompute only the affected states. We instantiate this formulation in ReTAC, a training-free multiagent framework for video spatial reasoning. ReTAC organizes supplementary computation around unresolved requirements. It plans ordered steps to acquire missing spatial evidence and combines the observations with compatible native evidence for reasoning. In addition, it uses stage-aligned verification to localize failures to planning, perception, or reasoning. Dependency-based recovery recomputes only affected states while reusing unaffected ones. Extensive experiments show that ReTAC achieves the highest overall score among the compared methods across three mainstream MLLMs and two widely adopted benchmarks. Further analyses show that gains from additional computation depend not only on how much is used, but also on where it is allocated and which valid intermediate states are preserved.

Chinese Translation

测试时自主计算可以通过规划、工具使用、迭代感知和反思来改善多模态大型语言模型（MLLMs），但不加区分的计算可能会重复已解决的工作，引入噪声证据，并扰动正确的预测。我们提出这种计算应该是残差的：仅将额外的计算分配给基础模型未解决的需求，并且在局部失败后，仅重新计算受影响的状态。我们在ReTAC中实例化这一公式，ReTAC是一个无训练的多智能体框架，用于视频空间推理。ReTAC围绕未解决的需求组织补充计算。它规划有序步骤以获取缺失的空间证据，并将观察结果与兼容的原生证据结合进行推理。此外，它使用阶段对齐的验证将失败局限于规划、感知或推理。基于依赖的恢复仅重新计算受影响的状态，同时重用未受影响的状态。大量实验表明，ReTAC在三个主流MLLMs和两个广泛采用的基准中，获得了与比较方法相比的最高整体得分。进一步分析表明，额外计算带来的收益不仅取决于使用的数量，还取决于其分配的位置以及保留的有效中间状态。

View on arXiv Download PDF AI Translation

cs.AI / 181 / 2610.16319

HyWe: Unifying Search and Operation for Real-World Hybrid Web Tasks

HyWe：统一现实世界混合网页任务的搜索与操作

Cai, Zijian, Liu, Haowei, Chen, Kailai, Peng, Lin, Wan, Cong, Zhang, Ziqi, Yuan, Chunfeng, Li, Bing, Hu, Weiming, Xu, Haiyang

Abstract

Real-world hybrid web tasks need both information gathering and page interaction, yet search-based and operation-based agents are developed and evaluated separately. To bridge this gap, we introduce HyWe, a framework that unifies both capabilities within one model and trajectory. Its two modes use dedicated prompts and tool sets to sustain sub-goal execution; explicit switching selects the needed capability, and shared history lets each mode build on the other's findings. To provide demonstrations of these behaviors, we develop HyWeCollector, a fully automatic pipeline that uses task prototypes, four switching patterns, and multi-agent synthesis to produce verified hybrid trajectories. We train HyWe-35B on these demonstrations to unify efficient information gathering and browser interaction in a single 35B web agent. To evaluate hybrid task completion, we introduce HyWeBench: 100 human-written tasks on websites distinct from training, with human-written sub-goal checklists and an outcome-based evaluation pipeline. Hybrid training enables HyWe-35B to achieve 47.0% success on this benchmark, improving on the untrained base model by 19.0 percentage points and outperforming existing search-based and operation-based agents. Even before training, the same base model solves 28.0% of tasks with HyWe, versus 11.0% with all tools under one prompt. Beyond hybrid tasks, HyWe-35B achieves 59.0% and 88.3% success on Online-Mind2Web and WebVoyager, outperforming existing open-source baselines and cutting action cost per successful task by 48.3% and 51.9% compared with Fara-7B.

Chinese Translation

现实世界的混合网页任务需要信息收集和页面交互，但基于搜索和基于操作的代理通常是分开开发和评估的。为了解决这一问题，我们提出了HyWe，一个在单一模型和轨迹中统一这两种能力的框架。其两种模式使用专门的提示和工具集来支持子目标的执行；显式切换选择所需的能力，而共享历史使每种模式能够基于对方的发现进行构建。为了展示这些行为，我们开发了HyWeCollector，这是一个完全自动化的管道，利用任务原型、四种切换模式和多代理合成来生成经过验证的混合轨迹。我们在这些演示上训练了HyWe-35B，以在单一的35B网页代理中统一高效的信息收集和浏览器交互。为了评估混合任务的完成情况，我们引入了HyWeBench：100个由人类编写的任务，这些任务在与训练不同的网站上进行，配有由人类编写的子目标检查清单和基于结果的评估管道。混合训练使HyWe-35B在该基准上取得了47.0%的成功率，比未训练的基础模型提高了19.0个百分点，并超越了现有的基于搜索和基于操作的代理。即使在训练之前，相同的基础模型在使用HyWe时解决了28.0%的任务，而在所有工具都在一个提示下时仅解决了11.0%。在混合任务之外，HyWe-35B在Online-Mind2Web和WebVoyager上分别取得了59.0%和88.3%的成功率，超越了现有的开源基线，并将每个成功任务的行动成本分别降低了48.3%和51.9%，与Fara-7B相比。

View on arXiv Download PDF AI Translation

cs.AI / 182 / 2610.16325

When a Conceptor Complement Becomes Gain: Operator Geometry and Behavioral Attribution in Language Models

当概念补充变为增益：语言模型中的算子几何与行为归因

Razdan, Shuhul, Balamurugan, R

Abstract

When a steering intervention changes a language model's behavior, which part of the intervention deserves the credit? We test this question for a target-centered, scaled conceptor complement by removing its fitted covariance filter while preserving the same target-specific center and gain. At the original aperture $\alpha=.05$ and gain $\beta=2$, all 21 fitted maps across three models and seven targets expand every direction at the generation intervention layers. The matrix correction is $0.093\%$--$7.815\%$ of the matched-control displacement on held-out extraction activations. In 480 paired comparisons within a 1,920-response study, retaining the matrix changes many individual texts but does not establish a consistent additional target-suppression benefit on the four-target panel. A geometric aperture sweep shows that the operating regime changes with the filter setting. Separately, an invertible soft filter preserves what an unrestricted affine readout can recover, whereas dedicated erasure methods reduce prediction by newly fitted probes. Together, the results show why a whole-intervention comparison cannot by itself attribute generated behavior to the fitted matrix: analyze the complete transformation, hold shared components fixed, and test its output consequences directly.

Chinese Translation

当一个引导干预改变语言模型的行为时，干预的哪个部分应当获得功劳？我们通过去除拟合的协方差滤波器，同时保留相同的目标特定中心和增益，来测试这一问题，针对一个以目标为中心的缩放概念补充。在原始孔径 $eta=0.05$ 和增益 $eta=2$ 下，三种模型和七个目标的21个拟合映射在生成干预层的每个方向上都扩展。矩阵修正为 $0.093\%$--$7.815\\%$，这是在保留提取激活时匹配控制位移的比例。在一项包含1,920个响应的研究中进行的480对比较中，保留矩阵改变了许多单个文本，但未能在四目标面板上建立一致的额外目标抑制效益。几何孔径扫描显示，操作模式随着滤波器设置的变化而变化。另一方面，一个可逆的软滤波器保留了不受限制的仿射读出所能恢复的内容，而专门的抹除方法则通过新拟合的探针减少了预测。综合来看，结果表明，单独的整体干预比较无法将生成的行为归因于拟合矩阵：需要分析完整的转化过程，固定共享组件，并直接测试其输出后果。

View on arXiv Download PDF AI Translation

cs.AI / 183 / 2610.16333

TRAP: Tool-result Replay Attack for Probing Misbinding in LLM Agents

TRAP：用于探测大型语言模型代理中误绑定的工具结果重放攻击

Zhi, Yutong, Zhang, Jie, Lu, Xiuqi, Xu, Jiayuan, Shao, Jing

Abstract

As AI agents increasingly act through external tools, tool results are no longer simple outputs but execution artifacts that must remain correctly associated with the context that produced them. Yet as results persist across caches, recovery mechanisms, and shared runtime state, this association can break, causing a real and unmodified result to be consumed in an incompatible execution context and lead the agent to an unsafe action. We characterize this failure as tool-result misbinding. To systematically study this failure, we introduce ReplayBench, a benchmark that uses controlled tool-result replay to evaluate misbinding across 19 representative agent scenarios, 4 failure properties, and 3 replay mechanisms. Across eight LLMs, three agent frameworks, and diverse stateful workflows, we find widespread replay-induced failures, with a mean Final-State Error Rate of 0.649 under cross-task replay. Stage-wise analysis localizes these failures to model-side action generation. Further analyses show that replay susceptibility varies with task similarity, shows little dependence on object similarity, and remains high under result staleness and implicit environment changes. These findings motivate runtime validity mechanisms that eliminate final-state errors in our evaluated settings while preserving legitimate recovery. The results call for explicit contextual binding in tool-using agent runtimes. Our work is public at https://github.com/Phitooo/Replaybench.

Chinese Translation

随着人工智能代理越来越多地通过外部工具进行操作，工具结果不再是简单的输出，而是必须与产生它们的上下文保持正确关联的执行产物。然而，由于结果在缓存、恢复机制和共享运行时状态中持续存在，这种关联可能会中断，导致真实且未修改的结果在不兼容的执行上下文中被使用，从而使代理采取不安全的行动。我们将这种失败特征化为工具结果误绑定。为了系统地研究这种失败，我们引入了ReplayBench，这是一个基准测试，使用受控的工具结果重放来评估19个代表性代理场景、4个失败属性和3个重放机制下的误绑定。在八个大型语言模型（LLMs）、三个代理框架和多样的有状态工作流中，我们发现广泛存在重放引发的失败，在跨任务重放下的平均最终状态错误率为0.649。阶段性分析将这些失败归因于模型侧的动作生成。进一步分析表明，重放的易感性与任务相似性相关，几乎不依赖于对象相似性，并且在结果陈旧和隐式环境变化下仍然很高。这些发现促使我们提出运行时有效性机制，以消除我们评估环境中的最终状态错误，同时保留合法的恢复。结果呼吁在使用工具的代理运行时中明确上下文绑定。我们的工作公开于 https://github.com/Phitooo/Replaybench。

View on arXiv Download PDF AI Translation

cs.AI / 184 / 2610.16343

Efficient Multi-Agent Reasoning via Confidence-Guided Adaptive Debate

通过信心引导的自适应辩论实现高效的多智能体推理

Yoa, Seungdong, Sim, Ye Seul, Yoon, Suhee, Yoon, Sanghyu, Kim, Dongmin, Lee, Soonyoung, Kim, Bumsoo, Lee, Junhyun

Abstract

Multi-agent debate has shown promise for improving the reasoning of large language models, yet recent theory suggests its benefits are highly regime-dependent. While interaction can amplify informative signals under corrective conditions, symmetric debate dynamics are neutral in expectation, often making majority voting preferable. We reconcile these views by arguing that debate is effective only when invoked at the right time and with appropriate structure. Based on this insight, we propose LASE: Leader-Adaptive Structured Engagement, a leader-centric multi-agent debate framework that selectively engages interaction only in non-neutral regimes. LASE introduces an asymmetric leader-supporter architecture that enables directed information flow and selective signal amplification, while defaulting to simple aggregation otherwise. Experiments across diverse reasoning benchmarks show that LASE achieves multi-agent-level performance with near single-agent token cost, substantially improving efficiency over static debate and voting baselines.

Chinese Translation

多智能体辩论在改善大型语言模型的推理能力方面展现出良好前景，但近期理论表明其效果高度依赖于具体的环境条件。尽管在纠正条件下，互动可以放大信息信号，但对称辩论动态在期望上是中性的，通常使得多数投票更具优势。我们通过论证辩论只有在适当的时机和结构下才有效，调和了这些观点。基于这一见解，我们提出了LASE：领导者自适应结构化互动（Leader-Adaptive Structured Engagement），这是一个以领导者为中心的多智能体辩论框架，仅在非中性环境中选择性地进行互动。LASE引入了一种不对称的领导者-支持者架构，能够实现定向信息流和选择性信号放大，而在其他情况下则默认采用简单聚合。跨越多种推理基准的实验表明，LASE以接近单智能体的代币成本实现了多智能体级别的性能，显著提高了相较于静态辩论和投票基线的效率。

View on arXiv Download PDF AI Translation

cs.AI / 185 / 2610.16345

ChronoScribe: Semantic-Numerical Specialization for Instruction-Driven Time Series Forecasting

ChronoScribe：面向指令驱动的时间序列预测的语义-数值专业化

Jiao, Yuchen, Zhang, Miao, Liu, Yuxuan, Li, Chengxi

Abstract

Recent advances in instruction-driven time series forecasting highlight a fundamental tension: systems must simultaneously master natural-language instruction following and precise numerical prediction. Current approaches often force a compromise. Large language models (LLMs) can interpret unstructured requests but may be inaccurate when asked to produce forecast values directly, whereas time series foundation models (TSFMs) provide numerical forecasts but typically require structured inputs specifying the target series, context window, and horizon. We present ChronoScribe, which applies semantic-numerical specialization to this workflow. A LoRA-adapted LLM proposes an execution path and context length; deterministic rules resolve the executable configuration. On the specialist path, a frozen TSFM produces quantile trajectories for the task-specified horizon. Deterministic trajectory summaries condition the response, while the LLM neither generates nor revises predicted values. We evaluate a Qwen3.5-9B--Chronos-2 implementation across 23 evaluation dimensions and 30 public datasets against eleven proprietary direct LLMs and three open TSFMs. ChronoScribe has the lowest error on sMAPE, MASE, WAPE, RMSE, and MAE among all evaluated systems, including fixed-context controls. Across these metrics, even the lowest direct-LLM error remains 2.0--2.7 times the corresponding ChronoScribe error. At the system level, ChronoScribe achieves a Task-Level Score (TLS) of 65.5, 10.6 points above the strongest direct-LLM baseline. The results support ChronoScribe's division of labor, in which the LLM handles interpretation, configuration, and communication while the TSFM handles numerical forecasting.

Chinese Translation

近期在指令驱动的时间序列预测方面的进展突显了一个基本的矛盾：系统必须同时掌握自然语言指令的执行和精确的数值预测。目前的方法往往迫使二者妥协。大型语言模型（LLMs）能够解释非结构化请求，但在直接生成预测值时可能不够准确，而时间序列基础模型（TSFMs）则提供数值预测，但通常需要指定目标序列、上下文窗口和预测范围的结构化输入。我们提出了ChronoScribe，它将语义-数值专业化应用于这一工作流程。经过LoRA适配的LLM提出执行路径和上下文长度；确定性规则解决可执行配置。在专业路径上，一个冻结的TSFM为任务指定的预测范围生成分位数轨迹。确定性轨迹摘要为响应提供条件，而LLM既不生成也不修正预测值。我们在23个评估维度和30个公共数据集上评估了Qwen3.5-9B--Chronos-2的实现，与11个专有直接LLM和3个开放TSFM进行比较。ChronoScribe在所有评估系统中，包括固定上下文控制，具有最低的sMAPE、MASE、WAPE、RMSE和MAE误差。在这些指标中，即使是最低的直接LLM误差也仍然是相应ChronoScribe误差的2.0至2.7倍。在系统层面，ChronoScribe的任务级得分（TLS）为65.5，比最强的直接LLM基线高出10.6分。结果支持ChronoScribe的劳动分工，其中LLM负责解释、配置和沟通，而TSFM负责数值预测。

View on arXiv Download PDF AI Translation

cs.AI / 186 / 2610.16365

World Models Need a Physical Registry: Learning Internal Predictive States for Physically Plausible World Models

世界模型需要物理注册：学习内部预测状态以构建物理上合理的世界模型

Kim, Shinseong, Jeong, Jaeseok, Fu, Jianlong, Uh, Youngjung

Abstract

Video world models must generate physically plausible futures, not just realistic frames. Existing approaches typically use semantic, physical, or predictive information to guide the visual generation stream, leaving a fundamental question: where inside the generator should predictions of scene evolution live? Standard video tokens jointly encode appearance and dynamics, but provide no distinct state for learning and reading predictive dynamics. We answer this question with the Physical Registry: temporally grounded registry slots inside a video DiT that are initialized from an observation-conditioned forecast, co-evolve with video tokens, and receive future-representation supervision, making predictive dynamics an explicit, readable state of the generator. In a controlled comparison of predictive-state interfaces on Physics-IQ Verified, Physical Registry improves vanilla fine-tuning by 9.9/8.7 points on I2V/V2V, the largest gain among the interfaces we compare. Across all Cosmos and WAN backbones, Physical Registry improves Physics-IQ Verified scores. With only 0.45% additional parameters, it enables Cosmos 3 Nano 16B to surpass the published I2V score of Cosmos 3 Super 64B (43.56 vs. 42.7). On WorldModelBench, it also improves physical adherence and the total score for all four evaluated backbones. Because the registry becomes candidate-specific while remaining readable, the Physical Registry Selector (PRS) ranks emerging futures by how far their intermediate registry readouts depart from the shared forecast before full denoising is complete. Across four backbones, this selection improves I2V scores by 5.6-15.9% without completed-video external scoring. Together, these results support a broader principle: dynamics prediction should live as an explicit model state that co-evolves with the video representation. Project page: https://grainsack.github.io/physical-registry/

Chinese Translation

视频世界模型必须生成物理上合理的未来，而不仅仅是现实的帧。现有的方法通常使用语义、物理或预测信息来指导视觉生成流，留下一个基本问题：场景演变的预测应该在生成器的哪个位置存在？标准视频标记共同编码外观和动态，但没有提供用于学习和读取预测动态的独立状态。我们通过物理注册（Physical Registry）来回答这个问题：在视频 DiT 中的时间基础注册槽，这些槽从观察条件的预测初始化，与视频标记共同演化，并接收未来表示的监督，使得预测动态成为生成器的一个明确、可读的状态。在对 Physics-IQ Verified 上的预测状态接口进行的受控比较中，物理注册将普通微调的 I2V/V2V 分数分别提高了 9.9/8.7 分，是我们比较的接口中增幅最大的。在所有 Cosmos 和 WAN 主干网络中，物理注册提高了 Physics-IQ Verified 的得分。仅增加 0.45% 的参数，它使 Cosmos 3 Nano 16B 超过了 Cosmos 3 Super 64B 的已发布 I2V 分数（43.56 对 42.7）。在 WorldModelBench 上，它还提高了物理遵循性和所有四个评估主干的总分。由于注册变得特定于候选项同时仍然可读，物理注册选择器（Physical Registry Selector, PRS）根据其中间注册读取结果与共享预测的偏离程度对新兴未来进行排名，直到完整去噪完成。在四个主干网络中，这种选择在未完成视频外部评分的情况下提高了 I2V 分数 5.6-15.9%。这些结果共同支持一个更广泛的原则：动态预测应该作为一个明确的模型状态存在，并与视频表示共同演化。项目页面：https://grainsack.github.io/physical-registry/

View on arXiv Download PDF AI Translation

cs.AI / 187 / 2610.16391

RetroPlanner: Multi-Route Expansion and Multi-Objective Reasoning for LLM-Based Retrosynthesis Planning

RetroPlanner：基于大语言模型的多路线扩展与多目标推理用于 retrosynthesis 规划

Kim, Ye Rin, Jeong, Yongjun, Yun, Seongjun, Ko, Hanbum, Oh, Soohyun, Lim, Sungbin, Kim, Sungwoong

Abstract

Multi-step retrosynthesis planning is often considered solved once a route to purchasable starting materials is found, but such a route should also be practical in real-world chemistry. Recent LLM-based planners formulate this process as sequential decision-making over an AND-OR graph, yet typically expand a single molecule per turn and are optimized for solvability alone, leaving alternative routes uncompared across the multiple objectives that determine practicality before the search budget is committed. We introduce RetroPlanner, an LLM-based planner that jointly compares multiple routes to steer the search toward higher synthesizability, as measured by reaction plausibility, round-trip consistency, and starting-material price. RetroPlanner is trained to reason over chemical evidence: structural progress from the target molecule, bond changes and reaction classes, where candidate routes diverge from one another, and how they rank across multiple objectives. Experiments on multi-step retrosynthesis benchmarks demonstrate that RetroPlanner recovers more alternative routes than prior LLM-based planners while improving synthetic route quality and maintaining a high success rate.

Chinese Translation

多步骤的 retrosynthesis 规划通常在找到可购买的起始材料路线后被认为是解决的，但这样的路线在现实化学中也应当是实用的。近期的基于大语言模型（LLM）的规划者将这一过程表述为在 AND-OR 图上的顺序决策，然而通常每次只扩展一个分子，并且仅针对可解性进行优化，导致在搜索预算确定之前，无法比较多目标下的替代路线。我们提出了 RetroPlanner，一种基于 LLM 的规划者，它联合比较多条路线，以引导搜索朝向更高的合成可行性，这一可行性通过反应合理性、往返一致性和起始材料价格来衡量。RetroPlanner 经过训练，能够基于化学证据进行推理：从目标分子的结构进展、键的变化和反应类别，以及候选路线之间的分歧，以及它们在多个目标下的排名。在多步骤 retrosynthesis 基准测试中的实验表明，RetroPlanner 能够恢复比之前的基于 LLM 的规划者更多的替代路线，同时提高合成路线的质量并保持较高的成功率。

View on arXiv Download PDF AI Translation

cs.AI / 188 / 2610.16397

Recursive Environment Scaling for General Long-Horizon Agents

递归环境扩展用于通用长时间跨度智能体

Jing, Lingxiao, Jiao, Zhengbo, Xiang, Yichen, Zhu, Bofan, Meng, Lingbei, Ni, Yan, Chen, Haorui, Ran, Ran, Shen, Zhirong, Yang, Yuheng, Cui, Chenhang, Du, Yaosong, Huang, Rui

Abstract

We study environment scaling as a way to increase agent-benchmark difficulty while holding core task objectives fixed. Our framework, Recursive Environment Scaling (RES), transforms existing benchmark environments while preserving their objectives and completion requirements. A proposer changes information access, resource relations, and execution dependencies; execution-grounded checks validate the resulting candidate, and accepted changes are inherited across rounds. This produces related environments for paired evaluation and training-trajectory collection. Experiments on 294 initial tasks from three benchmarks show lower task success for all six evaluation models, with decreases of 57.14--77.14 percentage points on Claw-Eval-Live. With equal trajectory counts and identical optimization settings, fine-tuning on RES trajectories outperforms original-trajectory SFT in all nine model--benchmark comparisons by 7.00--14.61 percentage points. Ablations identify recursive inheritance and dependency composition as important contributors to increased difficulty. Together, these results support environment scaling for both agent evaluation and training.

Chinese Translation

我们研究环境扩展作为一种在保持核心任务目标不变的情况下提高智能体基准难度的方法。我们的框架——递归环境扩展（Recursive Environment Scaling, RES）——在保留现有基准环境的目标和完成要求的同时，对其进行转化。提议者改变信息访问、资源关系和执行依赖；执行基础检查验证生成的候选者，接受的变化将在各轮之间继承。这产生了相关环境以便进行配对评估和训练轨迹收集。在来自三个基准的294个初始任务上的实验显示，所有六个评估模型的任务成功率均较低，在Claw-Eval-Live上下降了57.14至77.14个百分点。在相同的轨迹数量和相同的优化设置下，在RES轨迹上的微调在所有九个模型-基准比较中比原始轨迹的SFT表现提高了7.00至14.61个百分点。消融实验表明，递归继承和依赖组合是提高难度的重要因素。综合来看，这些结果支持环境扩展用于智能体评估和训练。

View on arXiv Download PDF AI Translation

cs.AI / 189 / 2610.16402

EvoWeb: A Benchmark for Coding Agents on Evolving Web Applications

EvoWeb：针对演变Web应用程序的编码代理基准测试

Nukrai, David, Alon, Bar, Levy, Nativ, Atad, Ido Andrew, Cohen, Ido, Zimerman, Itamar, Wolf, Lior, David, Eli, Naiman, Ilan

Abstract

Web application development often proceeds through successive requests that extend an existing implementation. Yet web-generation benchmarks primarily evaluate development toward fixed specifications. We introduce EvoWeb, a novel benchmark comprising 320 cumulative requests over 32 web applications, grounded in real-world product features and structured by functional dependency graphs. Agents iteratively modify a persistent codebase, while a multimodal web agent tests each intermediate application in a live browser, assessing newly requested functionality, cross-request integration, and previously requested functionality. To characterize reliability throughout development, we integrate trajectory performance across user-tolerance thresholds rather than impose a fixed stopping criterion. Experiments across open- and closed-source frontier models show that EvoWeb reveals clear performance differences between models, with cross-request integration emerging as a key challenge. Further analysis shows that later requests with deeper or more numerous dependencies are solved less accurately, while runtime errors become more frequent over successive requests. EvoWeb provides a testbed for evaluating and improving coding agents in long-horizon, multi-request development.

Chinese Translation

Web应用程序开发通常通过连续的请求来扩展现有实现。然而，Web生成基准主要评估针对固定规范的开发。我们介绍了EvoWeb，这是一种新颖的基准，包含320个累积请求，涵盖32个Web应用程序，基于真实世界的产品特性，并通过功能依赖图进行结构化。代理迭代地修改一个持久的代码库，同时一个多模态Web代理在实时浏览器中测试每个中间应用程序，评估新请求的功能、跨请求集成以及先前请求的功能。为了表征开发过程中的可靠性，我们整合了用户容忍阈值下的轨迹性能，而不是强加固定的停止标准。在开放源代码和闭源前沿模型上的实验表明，EvoWeb揭示了模型之间明显的性能差异，其中跨请求集成成为一个关键挑战。进一步分析显示，后续请求中具有更深或更多依赖关系的请求解决得不够准确，而运行时错误在连续请求中变得更加频繁。EvoWeb为评估和改进长时间跨度、多请求开发中的编码代理提供了一个测试平台。

View on arXiv Download PDF AI Translation

cs.AI / 190 / 2610.16412

SLS-Bench: A Benchmark for Incident Log Summarization with Synthetic Observability Data

SLS-Bench：基于合成可观测数据的事件日志摘要基准

Iakovlev, Valerii, Spanring, Markus, Flechl, Martin

Abstract

As IT systems grow in complexity, automated agent-driven incident response becomes increasingly critical. A key capability for these agents is summarizing large log streams into human-readable reports. Further progress requires meaningfully evaluating this function, yet existing benchmarks leave end-to-end summarization from large log streams underexplored due to the prohibitive cost and complexity of collecting real-world data. We introduce SLS-Bench, a benchmark for incident log summarization featuring 90 tasks grounded in real-world incident reports, each with a synthetic log stream and a reference summary. To construct SLS-Bench, we developed a data generation pipeline that converts incident reports into structured incident models, which guide the generation of synthetic logs and reference summaries. This pipeline combines narrative grounding, explicit incident modeling, generator-verifier loops, and statistical comparisons with real-world log streams to produce realistic and controllable observability data. Our evaluation of 14 LLMs on SLS-Bench shows that while frontier models can produce useful summaries, they do so at substantial cost and latency, and open-weight models lag significantly. SLS-Bench provides a framework for systematic evaluation of log analysis agents and a scalable blueprint for future synthetic observability benchmarks.

Chinese Translation

随着IT系统复杂性的增加，自动化代理驱动的事件响应变得愈加重要。这些代理的一项关键能力是将大量日志流总结为人类可读的报告。进一步的进展需要对这一功能进行有意义的评估，然而现有的基准测试由于收集真实世界数据的高昂成本和复杂性，导致从大型日志流进行端到端摘要的研究尚未得到充分探索。我们提出了SLS-Bench，这是一个针对事件日志摘要的基准，包含90个基于真实事件报告的任务，每个任务都有一个合成日志流和一个参考摘要。为了构建SLS-Bench，我们开发了一个数据生成管道，将事件报告转换为结构化的事件模型，这些模型指导合成日志和参考摘要的生成。该管道结合了叙述基础、明确的事件建模、生成器-验证器循环以及与真实日志流的统计比较，以生成逼真且可控的可观测数据。我们对14个大型语言模型（LLMs）在SLS-Bench上的评估表明，尽管前沿模型能够生成有用的摘要，但其代价和延迟都相当高，而开放权重模型的表现则明显滞后。SLS-Bench为日志分析代理的系统评估提供了框架，并为未来的合成可观测基准提供了可扩展的蓝图。

View on arXiv Download PDF AI Translation

cs.AI / 191 / 2610.16415

Chain Dimension Scaling: Communication-Efficient Loop-Style Scaling for Industrial Short-Video Ranking

链维度缩放：工业短视频排名的通信高效循环式缩放

Jiang, Zhimeng, Li, Yuening, Sun, Ruixiao, Lin, Yuanzhen, Jin, Kun, He, Chuan, Liu, Liang

Abstract

Industrial recommendation rankers keep scaling in width and depth to billions of parameters, but this increasingly incurs costs in parameter growth, distributed communication overhead, and training inefficiency. We introduce \underline{C}hain \underline{D}imension \underline{S}caling (CDS), a scaling strategy orthogonal to width and depth. CDS repeatedly applies a single feature-interaction block with parameters shared across iterations, so the number of iterations (the chain length) becomes a new scaling dimension that adds no parameters. A lightweight intra-chain residual connection stabilizes optimization and improves quality. CDS enables two regimes. In the \emph{compression regime}, shallow blocks with a longer chain match deeper baselines with fewer parameters and faster training; in the \emph{quality-scaling regime}, increasing chain length under a fixed parameter budget yields consistent offline and online gains, suggesting the chain dimension adds modeling capacity beyond parameter reuse. We evaluate CDS in a large-scale multi-task short-video ranking system in production. In the compression regime, CDS stays within $0.03\%$ of baseline offline quality while reducing trainable parameters by 26.6\% and improving training throughput by 6.4\%, with neutral-to-positive online participation. In the quality-scaling regime, quality improves monotonically with chain length. Wall-clock analysis shows the efficiency gain comes mainly not from reduced FLOPs but from lower cross-device synchronization overhead and reduced optimizer update cost. These results position parameter-shared iterative computation as a practical, communication-efficient scaling axis for industrial ranking systems.

Chinese Translation

工业推荐排名系统不断在宽度和深度上扩展至数十亿参数，但这日益导致参数增长、分布式通信开销和训练效率低下的成本增加。我们提出了链维度缩放（Chain Dimension Scaling, CDS），这是一种与宽度和深度正交的缩放策略。CDS 通过在迭代间共享参数的单一特征交互模块反复应用，从而使迭代次数（链长）成为一种新的缩放维度，而不会增加参数数量。轻量级的链内残差连接稳定了优化过程并提高了质量。CDS 支持两种模式。在压缩模式下，较浅的模块与较长的链相匹配，能够以更少的参数和更快的训练速度达到更深基线的效果；在质量缩放模式下，在固定参数预算下增加链长能带来一致的离线和在线收益，表明链维度在参数重用之外增加了建模能力。我们在生产环境中的大规模多任务短视频排名系统中评估了 CDS。在压缩模式下，CDS 的离线质量保持在基线的 $0.03\%$ 以内，同时可训练参数减少了 26.6\%，训练吞吐量提高了 6.4\%，在线参与度保持中性至正向。在质量缩放模式下，质量随着链长单调提高。时钟分析表明，效率提升主要不是来自于减少的浮点运算次数（FLOPs），而是来自于降低的跨设备同步开销和优化器更新成本。这些结果将参数共享的迭代计算定位为工业排名系统中一种实用的、通信高效的缩放轴。

View on arXiv Download PDF AI Translation

cs.AI / 192 / 2610.16426

AuthGateBench: Benchmarking LLMs as Authorization Gate

AuthGateBench：将大型语言模型作为授权门的基准测试

Zhuang, Zhixiong, Jia, Jin, Shang, Zhiwei, Wang, Ziyu

Abstract

Automated approval systems let coding agents execute high-risk actions with minimal human oversight by delegating the authorization decision to a classifier, typically a large language model (LLM). Such systems, adopted in tools like Claude Code and Codex, remain vulnerable to failure modes such as scope escalation, where the agent takes actions exceeding what the user authorized. Existing benchmarks focus mainly on malicious requests, evaluating refusal under an implicit policy. We introduce AuthGateBench, the first policy-guided benchmark for authorization gating under benign requests. We implement a novel pipeline where an LLM generates contrast pairs with identical action and trajectory but differing request, flipping the label between block and allow. This allows us to assess whether models reason over context rather than the action alone. We further examine whether auxiliary environment signals can help models make better decisions. The resulting benchmark comprises 1,400 instances spanning five failure modes drawn from real-world Claude Code incidents. Evaluations show accuracy degrades unevenly across failure modes as models over-rely on internal safety alignment, and models still block most actions even when explicitly authorized, revealing unstable rather than deterministic decisions; retrospectively retrieved environment signals, however, substantially improve authorization accuracy.

Chinese Translation

自动化审批系统通过将授权决策委托给分类器（通常是大型语言模型（LLM）），使编码代理能够在最小人类监督下执行高风险操作。这类系统已在Claude Code和Codex等工具中得到应用，但仍然容易受到失败模式的影响，例如范围升级，其中代理执行超出用户授权的操作。现有基准主要集中在恶意请求上，评估在隐含政策下的拒绝情况。我们提出了AuthGateBench，这是首个针对良性请求的政策引导授权门基准测试。我们实现了一种新颖的流程，其中LLM生成具有相同操作和轨迹但请求不同的对比对，标签在阻止和允许之间翻转。这使我们能够评估模型是否在上下文上进行推理，而不仅仅是操作本身。我们进一步研究辅助环境信号是否能帮助模型做出更好的决策。最终的基准包括1,400个实例，涵盖来自真实世界Claude Code事件的五种失败模式。评估结果显示，准确性在不同失败模式间不均匀下降，因为模型过度依赖内部安全对齐，且即使在明确授权的情况下，模型仍然阻止大多数操作，揭示出不稳定而非确定性的决策；然而，回溯检索的环境信号显著提高了授权准确性。

View on arXiv Download PDF AI Translation

cs.AI / 193 / 2610.16443

RAP-WAM: Learning World-Action Models through Predictive Rehearsal and Action Perturbations

RAP-WAM：通过预测排练和动作扰动学习世界-动作模型

He, Yuanpeng, Zhang, Ceyao, Li, Fangjing, Zhu, Kefei, Li, Lijian, Li, Yueheng, He, Tianjia, Jiao, Wenpin, Jin, Zhi, Yang, Yaodong, Chen, Yuanpei

Abstract

Recent world action models draw on video generation priors to jointly predict future observations and robot actions. Adapting these priors to control requires determining how future predictions should guide policy learning. RAP-WAM combines predictive Rehearsal and Action Perturbations (RAP) with future-supervised queries and Coupled Action Projection for policy learning. Predictive rehearsal trains on model-generated inputs while retaining recorded targets, and accumulates recurring errors into a spatial prior. We perturb actions and measure changes in video predictions. Action-response-based spatiotemporal credit assignment combines these responses, prediction error, and the rehearsal prior to allocate visual supervision; spatially aggregating the responses determines supervision intensity across future action steps. No separate scoring network or deployment-time probe is required. The base learns from a video-pretrained backbone without intermediate pretraining on external embodied data. We evaluate the fixed base and its agent-refined, task-conditioned system. On 42 RoboDojo tasks, their average success rates are 12.85% and 15.62%, with Scores of 17.86 and 20.59. Across 15 real-robot tasks, the complete system achieves 89.00% task-macro-average success, exceeding the strongest of five baselines by 37.67 percentage points. Across 315 fixed-output simulation windows, 45.21% of response-top20 positions fall outside error-top20, revealing information beyond prediction error.

Chinese Translation

近期的世界动作模型利用视频生成先验共同预测未来观察和机器人动作。将这些先验适应于控制需要确定未来预测如何指导策略学习。RAP-WAM结合了预测排练（Predictive Rehearsal）和动作扰动（Action Perturbations, RAP），以及未来监督查询和耦合动作投影（Coupled Action Projection）用于策略学习。预测排练在模型生成的输入上进行训练，同时保留记录的目标，并将重复出现的错误积累到空间先验中。我们对动作进行扰动并测量视频预测中的变化。基于动作响应的时空信用分配将这些响应、预测误差和排练先验结合起来，以分配视觉监督；空间聚合响应确定未来动作步骤中的监督强度。无需单独的评分网络或部署时探测器。基础模型从视频预训练的骨干网络中学习，而不需要在外部具身数据上进行中间预训练。我们评估了固定基础模型及其代理优化、任务条件系统。在42个RoboDojo任务中，它们的平均成功率分别为12.85%和15.62%，得分为17.86和20.59。在15个真实机器人任务中，完整系统实现了89.00%的任务宏观平均成功率，超过五个基线中最强者37.67个百分点。在315个固定输出的仿真窗口中，45.21%的响应前20名位置落在错误前20名之外，揭示了超出预测误差的信息。

View on arXiv Download PDF AI Translation

cs.AI / 194 / 2610.16478

XAI-Guided Fourier Domain Generalisation for Pure-Simulation Sim-to-Real Deep Reinforcement Learning

基于可解释人工智能的傅里叶域泛化用于纯模拟的仿真到现实深度强化学习

Yao, Wei, Wang, Chuyao, Wang, Ziwei, Aouf, Nabil

Abstract

Deep reinforcement learning has shown strong potential for vision-based autonomous driving, but policies trained in simulation often suffer severe performance degradation when deployed in the real world. Existing sim-to-real methods commonly address this gap through image-transfer-based domain adaptation or policy-agnostic domain randomisation. However, the former requires laborious acquisition of target-domain data and is often limited to the deployment domain, whereas the latter introduces unguided perturbations that may not reflect the underlying sim-to-real discrepancy, leading to suboptimal transfer performance. To overcome these limitations, we propose XFD, an explainable artificial intelligence (XAI)-guided Fourier domain generalisation framework for pure-simulation sim-to-real reinforcement learning. The proposed framework extracts explainability attribution maps from simulated observations, and converts these attributions into policy-conditioned Fourier-domain interventions. By perturbing appearance-sensitive spectral components while preserving task-defining spatial structures, XFD constructs robustness-enhancing visual domains entirely within simulation. Subsequently, the proposed robust policy optimisation guides the policy towards attending to task-critical features while suppressing sensitivity to irrelevant visual information. Extensive experiments in simulation and zero-shot real-world deployment demonstrate that XFD consistently improves traveled distance and reward over standard policy and randomly domain-randomised baseline, especially on unseen maps with cluttered backgrounds. Attribution-based analysis further proves that the proposed policy preserves more consistent task-relevant visual evidence across simulation and real-world observations.

Chinese Translation

深度强化学习在基于视觉的自动驾驶中展现出强大的潜力，但在真实世界中部署时，模拟训练的策略往往会遭遇严重的性能下降。现有的仿真到现实方法通常通过基于图像的领域适应或无策略领域随机化来解决这一差距。然而，前者需要繁琐的目标领域数据获取，并且通常局限于部署领域，而后者则引入了无指导的扰动，这些扰动可能无法反映潜在的仿真到现实差异，从而导致次优的迁移性能。为克服这些限制，我们提出了XFD，一个基于可解释人工智能（XAI）的傅里叶域泛化框架，用于纯模拟的仿真到现实强化学习。该框架从模拟观察中提取可解释性归因图，并将这些归因转换为策略条件下的傅里叶域干预。通过扰动对外观敏感的频谱成分，同时保留任务定义的空间结构，XFD在模拟中构建了完全增强鲁棒性的视觉域。随后，提出的鲁棒策略优化引导策略关注任务关键特征，同时抑制对无关视觉信息的敏感性。在模拟和零样本真实世界部署中的广泛实验表明，XFD在行驶距离和奖励方面始终优于标准策略和随机领域随机化基线，特别是在背景复杂的未见地图上。基于归因的分析进一步证明，所提策略在模拟和真实世界观察中保持了更一致的任务相关视觉证据。

View on arXiv Download PDF AI Translation

cs.AI / 195 / 2610.16480

PXRD-Bench: How Close Are We to Determining Crystal Structures from Powder Diffraction?

PXRD-Bench：我们离从粉末衍射中确定晶体结构有多近？

Woo, Dongyeop, Segal, Nofit, Kim, Nayoung, Kim, Minkyu, Gomez-Bombarelli, Rafael, Ahn, Sungsoo

Abstract

Determining crystal structures from powder X-ray diffraction (PXRD) would enable routine characterization of newly synthesized materials, including those absent from reference databases. Generative models offer a promising approach, but existing evaluations remain fragmented and largely assume single-phase samples with known composition. To fill this gap, we introduce PXRD-Bench, a unified benchmark that extends evaluation to unknown per-element atom counts and recovery of both structures in two-phase mixtures. It provides approximately 4.1 million training structures with simulated PXRD, tests generalization to unseen compositions and structural prototypes, and evaluates recovery from experimental PXRD after excluding structurally matching training entries. To provide a strong reference baseline, we also introduce Bragglite, a PXRD-to-structure model developed through a systematic study of PXRD modeling choices. It achieves a 44.2\% Top-1 match rate on experimental RRUFF patterns with known composition, 24.6\% above the strongest prior baseline. Further analysis yields several key findings: some models fail to benefit from experimental PXRD, while generalization to unseen structural prototypes and recovery from two-phase mixtures remain challenging. Together, our benchmark, systematic studies, and Bragglite clarify the capabilities and limitations of current PXRD-to-structure models.

Chinese Translation

从粉末 X 射线衍射（PXRD）中确定晶体结构将使新合成材料的常规表征成为可能，包括那些在参考数据库中缺失的材料。生成模型提供了一种有前景的方法，但现有的评估仍然零散，并且大多假设已知成分的单相样品。为填补这一空白，我们引入了 PXRD-Bench，这是一个统一的基准，扩展了对未知元素原子计数的评估，并恢复了两相混合物中的结构。它提供了大约 410 万个带有模拟 PXRD 的训练结构，测试对未见成分和结构原型的泛化能力，并在排除结构匹配的训练条目后评估从实验 PXRD 的恢复能力。为了提供一个强有力的参考基线，我们还引入了 Bragglite，这是一个通过系统研究 PXRD 建模选择开发的 PXRD 到结构模型。它在已知成分的实验 RRUFF 图谱上达到了 44.2\% 的 Top-1 匹配率，比之前最强的基线高出 24.6\%。进一步分析得出几个关键发现：一些模型未能从实验 PXRD 中受益，而对未见结构原型的泛化和从两相混合物中恢复仍然具有挑战性。我们的基准、系统研究和 Bragglite 一起澄清了当前 PXRD 到结构模型的能力和局限性。

View on arXiv Download PDF AI Translation

cs.AI / 196 / 2610.16511

Boosting the Learning for Ranking Patterns

提升模式排名学习

Belmecheri, Nassim, Aribi, Noureddine, Lazaar, Nadjib, Lebbah, Yahia, Loudni, Samir

Abstract

Pattern mining is a valuable tool for exploratory data analysis, but identifying relevant patterns for a specific user is challenging. Various interestingness mea- sures have been developed to evaluate patterns, but they may not efficiently estimate user-specific functions. Learning user-specific functions by ranking patterns has been proposed, but this requires significant time and training samples. In this paper, we present a solution that formulates the problem of learning pattern ranking functions as a multicriteria decision making problem. Our approach uses Analytic Hierarchy Process (AHP) to elicit weights for different interestingness measures based on user preference. We also propose an active learning mode with a sensitivity-based heuristic to minimize user ranking queries while still providing high-quality results. Experiments show that our approach significantly reduces running time and returns precise pattern ranking while being robust to user error, compared to state-of-the-art approaches.

Chinese Translation

模式挖掘是探索性数据分析的重要工具，但为特定用户识别相关模式具有挑战性。虽然已经开发出多种有趣性度量来评估模式，但它们可能无法有效估计用户特定的函数。通过对模式进行排名来学习用户特定函数的方案已被提出，但这需要大量的时间和训练样本。本文提出了一种将学习模式排名函数的问题形式化为多标准决策问题的解决方案。我们的方法使用层次分析法（Analytic Hierarchy Process, AHP）根据用户偏好提取不同有趣性度量的权重。我们还提出了一种基于敏感性的启发式主动学习模式，以最小化用户排名查询，同时仍能提供高质量的结果。实验表明，与最先进的方法相比，我们的方法显著减少了运行时间，并在对用户错误具有鲁棒性的同时返回精确的模式排名。

View on arXiv Download PDF AI Translation

cs.AI / 197 / 2610.16523

EAU: Environment-Agent-User Co-evolution for Training User-Centric Agents

EAU：环境-智能体-用户共演化以训练以用户为中心的智能体

Cai, Min, Zhang, Dan, Xing, Gao, Ma, Libin, Xu, Zhaohui, Ye, Xi

Abstract

Training LLM agents to interact with users over multiple turns is challenging because the learning signal depends not only on the agent, but also on the user simulator and environment, which jointly shape the agent's experience. Simulated users can be stochastic or inconsistent, while environments may contain unreliable tools, incomplete state coverage, and incorrect verifiers. Prior work mainly addresses the resulting instability through reward shaping or denser supervision, while treating these components as fixed. We instead introduce EAU, a meta-agent framework for environment--agent--user co-evolution. During training, EAU monitors trajectories, tool traces, rewards, and training statistics; diagnoses failures and uninformative interactions; and proposes targeted modifications to the user simulator, environment, reward design, and RL configuration. Validated changes are hot-swapped into the ongoing run, allowing the training setup to evolve together with the policy. These adaptations also persist across runs, so later training can build on previously discovered improvements rather than restarting from a fixed setup. We enable this process through an editable symbolic representation of user and environment dynamics, while the agent remains a neural LLM optimized with RL. Across three interactive-agent benchmarks and two backbone models, EAU consistently outperforms prior RL methods, improving the overall score over the strongest baseline by 14.7--15.0% and increasing successful task completion on ColBench by up to 36.7%, showing that co-evolving the training system with the policy can substantially improve multi-turn RL.

Chinese Translation

训练大型语言模型（LLM）智能体与用户进行多轮交互是具有挑战性的，因为学习信号不仅依赖于智能体，还依赖于用户模拟器和环境，这些因素共同塑造了智能体的体验。模拟用户可能是随机的或不一致的，而环境可能包含不可靠的工具、不完整的状态覆盖和错误的验证器。之前的研究主要通过奖励塑形或更密集的监督来解决由此产生的不稳定性，同时将这些组件视为固定的。我们提出了EAU，一个用于环境-智能体-用户共演化的元智能体框架。在训练过程中，EAU监控轨迹、工具痕迹、奖励和训练统计数据；诊断失败和无信息的交互；并针对用户模拟器、环境、奖励设计和强化学习（RL）配置提出有针对性的修改。经过验证的更改会热插拔到正在进行的运行中，使训练设置能够与策略共同演化。这些适应性也会在不同的运行中持续存在，因此后续训练可以基于先前发现的改进，而不是从固定设置重新开始。我们通过可编辑的符号表示用户和环境动态来实现这一过程，而智能体仍然是一个通过强化学习优化的神经LLM。在三个交互智能体基准测试和两个主干模型中，EAU始终优于之前的强化学习方法，整体得分比最强基线提高了14.7%至15.0%，在ColBench上成功任务完成率提高了多达36.7%，表明与策略共同演化训练系统可以显著改善多轮强化学习。

View on arXiv Download PDF AI Translation

cs.AI / 198 / 2610.16534

Does the Explanation Drift Before the Guarantee? Label-Free Early Warning on Evolving Graphs

解释漂移是否在保证之前？对演变图的无标签早期预警

Szandała, Tomasz, Raczkowski, Damian

Abstract

Deployed models can fail silently as the world they describe evolves. Attribution drift has served as an early warning in tabular data; we ask whether, in graph streams, a model's own explanations can forewarn the loss of nominal conformal coverage. Doing so requires repairing the naive drift statistic twice. A vanishing edge loses mask mass regardless of whether the model relied on it, so we condition the importance on edge presence. Evolving graphs also break edge identity: a relabelling that changes nothing shifts the identity-based statistic by 0.69 on synthetic graphs and 0.83 on real ego-networks, whereas aggregating over structural roles brings both to 0.0 and stays well-defined when the support changes. We test six pre-registered regimes on thirty seeds each. A two-channel monitor, tracking conditional mask mass and its distribution over roles, precedes coverage collapse by 8 to 14.5 steps across four gradual regimes (node addition, removal, replacement, and edge thinning), all surviving Holm correction. A sudden knockout yields the expected zero lead, and a harmless change triggers no alarm. Leads track the length of the transitional phase (Spearman 0.79 over 150 runs). On real transaction ego-networks, the untuned mass channel forewarns staged drift by 27 steps and is the only channel, attributional or structural, that never alarms on a stationary stream. The main limitation is that the mass channel is blind to covariate drift, which breaks the guarantee in every run; the distributional channel detects it and also rescues the rewiring regime, so we report the two channels separately rather than as a union.

Chinese Translation

已部署的模型可能会在其描述的世界演变时悄然失效。归因漂移在表格数据中作为早期预警；我们询问在图流中，模型自身的解释是否可以预警名义一致覆盖的丧失。实现这一目标需要对简单漂移统计量进行两次修正。消失的边缘无论模型是否依赖于它，都会失去掩码质量，因此我们将重要性条件化于边缘的存在。演变图还打破了边缘身份：一个改变无关紧要的重新标记使基于身份的统计量在合成图上变化0.69，在真实自我网络上变化0.83，而对结构角色进行聚合则将两者都带到0.0，并在支持变化时保持良好定义。我们在每个预注册的六个方案上测试了三十个种子。一个双通道监测器，跟踪条件掩码质量及其在角色上的分布，在四个渐进方案（节点添加、移除、替换和边缘稀疏）中提前8到14.5步预警覆盖崩溃，所有方案均通过了霍尔姆校正。突然的淘汰产生了预期的零前导，而无害的变化则不会触发警报。前导跟踪过渡阶段的长度（在150次运行中斯皮尔曼相关系数为0.79）。在真实交易自我网络中，未调优的质量通道提前27步预警分阶段漂移，并且是唯一在静态流中从未发出警报的通道，无论是归因的还是结构的。主要限制在于质量通道对协变量漂移是盲目的，这在每次运行中都打破了保证；而分布通道则能检测到并拯救重连方案，因此我们单独报告这两个通道，而不是作为一个联合体。

View on arXiv Download PDF AI Translation

cs.AI / 199 / 2610.16543

Overtaught, Underlearned: Learner-Centered Distillation from Multi-Agent Discussion

过度教授，学习不足：基于学习者中心的多智能体讨论蒸馏

Zhou, Mengtao, Wang, Jianxiao, Chen, Xiupeng, Liu, Cailiang, Song, Jie

Abstract

Multi-agent discussion produces reasoning collaboratively, while deployment still relies on a smaller student that must answer alone. Full-graph distillation transfers such discussions by treating each correct solution as a separate training target, but what teachers provide need not match what the student learns. This mismatch appears in two ways: several correct solutions may amount to the same lesson in the student's representation (over-teaching), while some training questions remain missed after learning (under-learning). We quantify both with rates that are high on discussion graphs. To address them, we introduce FAR (Fold--Attempt--Revisit), a learner-centered distillation procedure that organizes supervision around the student's own state. Before training, FAR folds solutions within each discussion state into representative lessons. After training, the student's own attempts identify missed questions, while poorly represented clusters indicate where folding may have lost useful variation; FAR then writes targeted lessons from retained discussion contexts. Across two students and four reasoning tasks, FAR outperforms full-graph distillation with far fewer training targets and lower under-learning. Further analyses examine the behavior of FAR and the sources of its gains. Overall, the results show that organizing discussion supervision around the student can make distillation both more effective and more efficient.

Chinese Translation

多智能体讨论通过协作产生推理，而在应用时仍依赖于一个较小的学生单独回答。全图蒸馏通过将每个正确解决方案视为单独的训练目标来转移这种讨论，但教师提供的内容不一定与学生所学相匹配。这种不匹配表现为两种方式：多个正确解决方案可能在学生的表征中形成相同的课程（过度教学），而一些训练问题在学习后仍未被识别（学习不足）。我们用在讨论图上较高的比率来量化这两者。为了解决这些问题，我们引入了FAR（Fold--Attempt--Revisit），一种以学习者为中心的蒸馏程序，围绕学生自身的状态组织监督。在训练之前，FAR将每个讨论状态中的解决方案折叠成代表性课程。在训练之后，学生自己的尝试识别未解决的问题，而表现不佳的聚类则指示折叠可能丢失了有用的变异；FAR随后从保留的讨论上下文中编写针对性的课程。在两个学生和四个推理任务中，FAR在训练目标更少和学习不足更低的情况下优于全图蒸馏。进一步的分析考察了FAR的行为及其收益的来源。总体而言，结果表明围绕学生组织讨论监督可以使蒸馏变得更加有效和高效。

View on arXiv Download PDF AI Translation

cs.AI / 200 / 2610.16552

Making Language the Low-Rate Code of Vision: Successive Refinement from Words to Pixels

将语言作为视觉的低速率编码：从词语到像素的逐步细化

Gong, Ruilin, Gong, Hao

Abstract

An ordered visual code lets a short prefix stand in for an image, and the training loss decides what that prefix contains. Under pixel reconstruction the first bits go to the directions of highest variance, and we prove that a caption can then be independent of the entire first half of a code that is rate-optimal for pixels at every length. In the rate-distortion sense a caption is therefore not the low-rate code of its image, unless the code is built to make it one. We study that construction and what it costs. We cast captioning and text-to-image generation as successive refinement under a distortion measure that changes along the code, from caption log-loss in the early stages to pixel distortion in the last, so that each stage is compared with its own optimum. Putting words first then costs nothing among the caption stages exactly when the captions form a summarization chain, and it costs the pixels at most the caption uncertainty left in a pixel-optimal reconstruction, which is large for short prefixes and small at full length. We build RUNG, an ordered code whose prefixes are supervised by a summarization ladder of captions and whose full length reconstructs the image; generation extends the code past the prompt and captioning truncates it. With data and capacity held fixed, ordering by language instead of reconstruction raises prefix-caption information 2.9 times and GenEval from 0.76 to 0.83, at a measured cost of 2.8 dB PSNR at a quarter of the code and 0.1 dB at full length. RUNG-XL (3.4B) reaches 0.88 GenEval and 87.6 DPG-Bench in 32 network evaluations and 0.91 GenEval when errors are redrawn at the rung where they appear; its coarse and fine captions agree 94.1% of the time, and it edits images without inversion.

Chinese Translation

有序的视觉编码允许一个短前缀代表一幅图像，而训练损失决定了该前缀的内容。在像素重建下，前几个比特被分配给方差最高的方向，我们证明了在每个长度上对于像素是最优速率的编码的前半部分，标题可以独立于整个前半部分。因此，从速率-失真意义上讲，标题并不是其图像的低速率编码，除非该编码是为了使其成为低速率编码而构建的。我们研究了这种构建及其成本。我们将标题生成和文本到图像生成视为在一个失真度量下的逐步细化，该度量沿着编码变化，从早期阶段的标题对数损失到最后阶段的像素失真，使得每个阶段与其自身的最优解进行比较。当标题形成一个摘要链时，优先考虑词语在标题阶段的成本为零，而在像素最优重建中，最多只需支付像素中剩余的标题不确定性，这在短前缀时较大，而在全长时较小。我们构建了RUNG，一个有序编码，其前缀由标题的摘要阶梯进行监督，完整长度重建图像；生成扩展了编码超出提示，而标题生成则截断了它。在数据和容量固定的情况下，与重建相比，按语言排序提高了前缀-标题信息2.9倍，GenEval从0.76提升至0.83，测得的成本在四分之一编码时为2.8 dB PSNR，在全长时为0.1 dB。RUNG-XL（3.4B）在32次网络评估中达到0.88 GenEval和87.6 DPG-Bench，并在错误在出现的阶梯上重新绘制时达到0.91 GenEval；其粗略和精细标题的相符率为94.1%，并且它可以在不反转的情况下编辑图像。

View on arXiv Download PDF AI Translation

cs.AI / 201 / 2610.16555

AI Discovery in Combinatorics: From Problems to Solutions and Back to Bigger Problems

组合数学中的人工智能发现：从问题到解决方案再到更大问题

Cainelli, Eugenio, Luccioli, Lorenzo, Iraci, Alessandro, Paolini, Giovanni

Abstract

We introduce $qt$Bench, a benchmark of 29 open problems in algebraic combinatorics. These problems ask for a simple combinatorial algorithm that explains given polynomial coefficients or identities. To help AI agents search for such algorithms, we provide a scaffold for playing with combinatorial objects, countering cheating attempts, and testing candidate solutions. To explore how far frontier AI systems can go on these problems, we collaborated with OpenAI, whose researchers obtained proposed solutions to several of them. This highlights bigger meta-problems: is AI discovery at scale ethical? How should these results be released? How should proof digestion be organized? We discuss these questions and give our views.

Chinese Translation

我们介绍了 $qt$Bench，这是一个包含29个代数组合数学开放问题的基准。这些问题要求提供一个简单的组合算法，以解释给定的多项式系数或恒等式。为了帮助人工智能代理搜索这样的算法，我们提供了一个框架，用于操作组合对象、抵制作弊尝试以及测试候选解决方案。为了探索前沿人工智能系统在这些问题上的能力，我们与OpenAI合作，其研究人员提出了多个问题的解决方案。这突显了更大的元问题：大规模的人工智能发现是否合乎伦理？这些结果应如何发布？证明消化应如何组织？我们讨论了这些问题并给出了我们的看法。

View on arXiv Download PDF AI Translation

cs.AI / 202 / 2610.16558

G$^2$WAM: World Action Models for Efficient and Geometry Grounded Image-Goal Navigation

G$^2$WAM：用于高效且基于几何的图像目标导航的世界行动模型

Gui, Ke, Shi, Chenghao, Wang, Ziyue, Chen, Longwen, Chen, Xieyuanli, Lu, Huimin

Abstract

World models capture environment dynamics and support counterfactual reasoning for planning. In image-goal navigation, where the target often lies beyond the current field of view, the agent must not only model environmental dynamics but also maintain a persistent spatial memory of the scene. Existing approaches either introduce spatial context by fine-tuning pretrained models on the target task, which underexploits their spatial priors, or rely on external 3D reconstruction, which is computationally expensive for real-world deployment. In this paper, we propose the Geometry Grounded World Action Model (G$^2$WAM) for image-goal navigation. G$^2$WAM encodes spatial cues through a learned scene-coordinate localization module that anchors the current, goal, and history frames in a common reference frame, yielding a stable and consistent spatial context that enables efficient navigation without explicit 3D reconstruction. Following the world-action-model paradigm, G$^2$WAM jointly predicts future states and actions, eliminating the need for costly cross-entropy method (CEM) style planning. In addition, G$^2$WAM operates entirely in latent space, avoiding the computational burden of pixel-level reconstruction and substantially reducing inference cost. Across four trajectory prediction benchmarks, G$^2$WAM reduces average ATE and RPE by 40.1% and 35.2%, respectively. In closed-loop navigation experiments, it improves the success rate from 54.0% to 63.0% and the success weighted by path length (SPL) from 36.4% to 59.2%, while maintaining high efficiency.

Chinese Translation

世界模型捕捉环境动态并支持反事实推理以进行规划。在图像目标导航中，由于目标通常位于当前视野之外，智能体不仅需要建模环境动态，还需保持对场景的持久空间记忆。现有方法要么通过在目标任务上微调预训练模型引入空间上下文，但未充分利用其空间先验，要么依赖外部3D重建，这在实际应用中计算成本高昂。本文提出了一种用于图像目标导航的几何基础世界行动模型（G$^2$WAM）。G$^2$WAM通过一个学习的场景坐标定位模块编码空间线索，该模块将当前帧、目标帧和历史帧锚定在一个共同的参考框架中，从而产生稳定且一致的空间上下文，使得在没有显式3D重建的情况下实现高效导航。遵循世界行动模型范式，G$^2$WAM联合预测未来状态和动作，消除了对昂贵的交叉熵方法（CEM）风格规划的需求。此外，G$^2$WAM完全在潜在空间中操作，避免了像素级重建的计算负担，并显著降低了推理成本。在四个轨迹预测基准测试中，G$^2$WAM分别将平均绝对误差（ATE）和相对定位误差（RPE）降低了40.1%和35.2%。在闭环导航实验中，其成功率从54.0%提高到63.0%，而基于路径长度加权的成功率（SPL）从36.4%提高到59.2%，同时保持了高效性。

View on arXiv Download PDF AI Translation

cs.AI / 203 / 2610.16593

Double-Rollout Reinforcement Learning with Sampling-based Reward Estimation

基于采样的奖励估计的双回滚强化学习

Qian, Haifeng, Mosallanezhad, David, Munley, Christian, Renduchintala, Adithya, Kuchaiev, Oleksii

Abstract

Reinforcement learning with verifiable rewards (RLVR) implemented with Group Relative Policy Optimization (GRPO) has driven substantial progress in training reasoning language models. However, verifiable outcome rewards provide an imperfect measure of reasoning quality: a correct final answer may not be supported by the reasoning trace, whereas an incorrect rollout may contain a nearly correct reasoning trajectory. We present empirical evidence for this mismatch and hypothesize that the resulting reward noise contributes to instability in reinforcement learning for both reasoning accuracy and token efficiency. To address this credit assignment challenge, this paper proposes double-rollout reinforcement learning. For each sampled rollout, we truncate its reasoning trace, sample multiple continuations from the truncated prefix, and replace the original reward with the mean continuation reward, yielding a more robust estimate of reasoning quality. This double-rollout algorithm incurs only a moderate increase in compute cost and yet reaches substantially higher accuracies across math and coding tasks and across Qwen and Nemotron models than baseline algorithms with fewer RL steps. For example, using only DAPO data it trains Nemotron-3-Nano-30B-A3B to reach 95.4% on AIME 2026, versus 91.5% by GRPO and substantially above state of the art for this model size, within 30 RL steps. When combined with length-based rewards, the same algorithm further improves the accuracy-efficiency Pareto frontier over existing efficiency optimizations. These findings suggest that the double-rollout reward estimation is an effective and cost-efficient way to improve both the accuracy and token efficiency of reasoning RL.

Chinese Translation

基于可验证奖励的强化学习（RLVR）通过群体相对策略优化（GRPO）的实现，推动了推理语言模型训练的重大进展。然而，可验证的结果奖励并不能完美衡量推理质量：一个正确的最终答案可能没有得到推理轨迹的支持，而一个错误的回滚可能包含一个几乎正确的推理轨迹。我们提供了这一不匹配的实证证据，并假设由此产生的奖励噪声会导致推理准确性和令牌效率的强化学习不稳定性。为了解决这一信用分配挑战，本文提出了双回滚强化学习。对于每个采样的回滚，我们截断其推理轨迹，从截断的前缀中采样多个延续，并用平均延续奖励替换原始奖励，从而获得更稳健的推理质量估计。该双回滚算法仅增加了适度的计算成本，但在数学和编码任务以及 Qwen 和 Nemotron 模型中，相较于基线算法在较少的 RL 步骤下达到了显著更高的准确率。例如，仅使用 DAPO 数据，它训练的 Nemotron-3-Nano-30B-A3B 在 AIME 2026 上达到了 95.4%，而 GRPO 的准确率为 91.5%，并且在该模型规模下显著高于当前最优水平，在 30 个 RL 步骤内。当与基于长度的奖励结合时，同一算法进一步改善了现有效率优化的准确性-效率帕累托前沿。这些发现表明，双回滚奖励估计是一种有效且具有成本效益的方法，可以同时提高推理强化学习的准确性和令牌效率。

View on arXiv Download PDF AI Translation

cs.AI / 204 / 2610.16615

When 1+1 Backfires: Rethinking Relevance-Diversity Balancing for Sample-Aware Visual Token Reduction

当1+1适得其反：重新思考样本感知的相关性-多样性平衡以进行视觉令牌压缩

Huang, Yihang, Wen, Peisong, Zhou, Chenhao, Feng, Difu, Xu, Qianqian

Abstract

Visual token compression has emerged as a critical technique for reducing the inference cost of Vision-Language Models. Existing methods select tokens based on text relevance, visual diversity, or a fixed combination of both---treating the relevance-diversity balance as a universal constant independent of the input. We conduct a systematic analysis to challenge this assumption. At the dataset level, grid-searching the optimal relevance-diversity weight per benchmark reveals substantial cross-task variation. More critically, we further maximize per-sample answer likelihood to recover sample-level oracle weights, finding that their distributions not only differ markedly across benchmarks but also exhibit high within-benchmark variance. Based on these observations, we propose SAVER, a framework for Sample-Aware relevance-diversity balanced Visual tokEn Reduction. We first derive a Geometric Intent Prior (GIP) from the spatial concentration of text-relevant tokens to infer a sample-specific relevance-diversity weight, and then apply this weight to jointly select tokens via a Weighted Text Centroid (WTC) for relevance scoring and a Visual Marginal Coverage Gain (VMCG) for diversity. SAVER requires no additional training and achieves sample-aware balancing entirely at inference time. Extensive experiments demonstrate that SAVER consistently outperforms state-of-the-art baselines across diverse benchmarks, retaining 96.5% of full-model performance on LLaVA-1.5-7B with only 11.1% of the original visual tokens, and further generalizing to LLaVA-NeXT and Mini-Gemini.

Chinese Translation

视觉令牌压缩已成为降低视觉语言模型推理成本的关键技术。现有方法基于文本相关性、视觉多样性或两者的固定组合来选择令牌——将相关性-多样性平衡视为一个独立于输入的普遍常数。我们进行系统分析以挑战这一假设。在数据集层面，针对每个基准进行网格搜索以确定最佳的相关性-多样性权重，揭示了跨任务的显著差异。更重要的是，我们进一步最大化每个样本的答案似然性，以恢复样本级的oracle权重，发现其分布不仅在不同基准之间显著不同，而且在基准内部也表现出较高的方差。基于这些观察，我们提出了SAVER，一个样本感知的相关性-多样性平衡视觉令牌压缩框架。我们首先从文本相关令牌的空间集中度推导出几何意图先验（Geometric Intent Prior, GIP），以推断样本特定的相关性-多样性权重，然后通过加权文本质心（Weighted Text Centroid, WTC）进行相关性评分和视觉边际覆盖增益（Visual Marginal Coverage Gain, VMCG）进行多样性选择令牌。SAVER不需要额外的训练，并且完全在推理时实现样本感知平衡。大量实验表明，SAVER在不同基准上始终优于最先进的基线，在仅使用11.1%的原始视觉令牌的情况下，保留了LLaVA-1.5-7B模型96.5%的性能，并进一步推广到LLaVA-NeXT和Mini-Gemini。

View on arXiv Download PDF AI Translation

cs.AI / 205 / 2610.16620

Weak Segmentation-Guided GAN for~Realistic Color Edition

弱分割引导的生成对抗网络用于真实感色彩编辑

Auriau, Vincent, Malherbe, Emmanuel, Perrot, Matthieu

Abstract

Editing the color of images in a realistic way finds many applications such as changing the perception of an image, data augmentation or film post processing. The design of an automatic tool is a complex and long addressed challenge. In particular, two properties are difficult to meet altogether: generating realistic results without artifacts and the possibility to precisely choose the future color. Conventional methods using segmentation and histogram matching maximize the controllability but also introduce a lack of realism or are complex to automate. On the contrary, GANs that specialize in realism are difficult to control. To overcome these challenges, we propose a novel GAN architecture leveraging any differentiable segmentation model. We demonstrate the genericness of our framework that presents state of the art results on different use cases. It generates images that look realistic while offering a precise color control.

Chinese Translation

以真实的方式编辑图像颜色在许多应用中具有重要意义，例如改变图像的感知、数据增强或电影后期处理。设计一个自动化工具是一项复杂且长期以来面临的挑战。特别是，有两个特性难以同时满足：生成无伪影的真实结果以及能够精确选择未来的颜色。传统方法使用分割和直方图匹配来最大化可控性，但也引入了缺乏真实感的问题，或者使自动化变得复杂。相反，专注于真实感的生成对抗网络（GAN）则难以控制。为了解决这些挑战，我们提出了一种新颖的GAN架构，利用任何可微分的分割模型。我们展示了我们框架的通用性，在不同的使用案例中呈现了最先进的结果。它生成的图像看起来真实，同时提供了精确的颜色控制。

View on arXiv Download PDF AI Translation

cs.AI / 206 / 2610.16625

LATCH: LLM-Assisted Timing Closure Harness for Post-Route FPGA Optimisation

LATCH：用于后路 FPGA 优化的 LLM 辅助时序收敛框架

Gu, Shu, Youngman, Matthew, Zhao, Yuqin, Sestito, Cristian, Siddique, Ali, Krishnamurthy, Ramesh, Chitnis, Danial, Prodromakis, Themis

Abstract

Post-route FPGA timing closure remains expert-dependent and computationally expensive. We present LATCH (LLM-Assisted Timing Closure Harness), a two-layer framework that separates large language model (LLM) orchestration from deterministic FPGA implementation and evaluation. Across sixteen designs and seven LLMs, a single-iteration study compares three agent control granularities - expert-written recipes, direct EDA operations and LLM-constructed procedures - to ask how much control an agent needs per attempt. Functional-pass rates are 98.2%, 71.4% and 17.9%, respectively; yet direct operations give slightly larger 3.4% median Fmax gains when passing, versus 3.2% for recipes. This exposes a trade-off between execution reliability and timing improvement. Informed by this trade-off, a multi-iteration study fixes the action tier at expert recipes to ask how repeated attempts are best organised, comparing a fitted deterministic heuristic with sequential LLM reasoning, branching exploration and knowledge base-guided exploration within 12 checkpoints per run. Branching raises mean maximum-frequency improvement from 10.5% (sequential) to 16.0%, and the knowledge base to 20.3%, improving 15 of 16 designs. Characterisation across 2,421 recipe attempts profiled their gain and validation rates, showing hard-block re-placement gives the largest cumulative mean gain at 41.4MHz.

Chinese Translation

后路 FPGA 时序收敛仍然依赖专家并且计算成本高昂。我们提出了 LATCH（LLM 辅助时序收敛框架），这是一个两层框架，将大语言模型（LLM）的协调与确定性 FPGA 实现和评估分开。在十六个设计和七个 LLM 的研究中，我们比较了三种代理控制粒度——专家编写的食谱、直接 EDA 操作和 LLM 构建的程序——以探讨代理每次尝试所需的控制程度。功能通过率分别为 98.2%、71.4% 和 17.9%；然而，直接操作在通过时提供了略高的 3.4% 中位 Fmax 增益，而食谱为 3.2%。这揭示了执行可靠性与时序改进之间的权衡。在这一权衡的指导下，多次迭代研究将行动层级固定在专家食谱上，以探讨如何最佳组织重复尝试，比较了拟合的确定性启发式与顺序 LLM 推理、分支探索和知识库引导探索在每次运行的 12 个检查点内的表现。分支将平均最大频率的改进从 10.5%（顺序）提高到 16.0%，知识库则提高到 20.3%，改善了 16 个设计中的 15 个。对 2,421 次食谱尝试的特征化分析显示，硬块重定位在 41.4MHz 的累积平均增益中提供了最大的收益。

View on arXiv Download PDF AI Translation

cs.AI / 207 / 2610.16655

Does a Hosted Decision Model Respect Irrelevant Alternatives? A Pre-Registered Test of Option-Set Dependence in Jev

托管决策模型是否尊重无关替代品？对Jev中选项集依赖性的预注册测试

Rajput, Dushyant, Chauhan, Nirdesh, Kosaraju, Siddharth

Abstract

Decision APIs that return a probability for each option are routinely used as if those probabilities were properties of the options themselves. That reading is valid only if the model satisfies independence of irrelevant alternatives (IIA): adding an option that is neither the right answer nor its strongest rival should leave the odds between those two unchanged. We test this in a pre-registered study of jev-1.13.0, a hosted decision model, using 299,463 calls on intent classification (BANKING77, CLINC150) and topic classification (AG News). Every added option moves the log-odds between the gold label and the strongest wrong label: the root-mean-square systematic shift is 0.20-0.52 nats, 1.2 to 3 times the run-to-run variation, even when the added option takes at most 0.1% of the probability mass. The shifts are directional. An option similar to the wrong answer draws probability from it (+0.10 nats toward the gold label). A paraphrase of the wrong answer produces an intermediate pattern: the gold label's probability is almost conserved, but the paraphrase receives only about a third of the wrong answer's share rather than the half that strict substitution implies. The distortion grows steadily with the logarithm of the number of options, reaching about 1 nat at 255 options, and off-domain padding pushes the odds against the gold label. In contrast, questions packed into one call do not interact (bias < 0.001). Scoring each option with an independent yes/no question removes the option-set dependence but picks the right answer 8.7 points less often. Jev sits between two local baselines, an option-independent scorer and a prompt readout. Its probabilities are best read as relative to the option list they were computed over.

Chinese Translation

决策API通常返回每个选项的概率，并被视为选项本身的属性。只有当模型满足无关替代品独立性（IIA）时，这种解读才是有效的：添加一个既不是正确答案也不是其最强竞争者的选项，应该不会改变这两个选项之间的赔率。我们在一项预注册研究中测试了这一点，研究对象为托管决策模型jev-1.13.0，使用了299,463次意图分类（BANKING77，CLINC150）和主题分类（AG News）的调用。每个新增选项都会影响金标签与最强错误标签之间的对数赔率：均方根系统性偏移为0.20-0.52 nats，是运行间变异的1.2到3倍，即使新增选项最多只占0.1%的概率质量。这些偏移是有方向性的。与错误答案相似的选项会从中吸引概率（+0.10 nats朝向金标签）。错误答案的同义改写产生了一种中间模式：金标签的概率几乎保持不变，但同义改写仅获得错误答案份额的三分之一，而不是严格替代所暗示的二分之一。随着选项数量的对数增加，失真稳步增长，在255个选项时达到约1 nat，而域外填充则会使金标签的赔率下降。相比之下，打包在一个调用中的问题不会相互影响（偏差<0.001）。用独立的是/否问题对每个选项进行评分消除了选项集依赖性，但正确答案的选择频率降低了8.7个百分点。Jev位于两个局部基线之间，一个是独立于选项的评分器，另一个是提示输出。其概率最好被理解为相对于计算时所依据的选项列表。

View on arXiv Download PDF AI Translation

cs.AI / 208 / 2610.16661

Structural Baselines Matter for Agent Failure Detection and Attribution

结构基线对智能体失败检测与归因的重要性

Naagarajan, Ramesh Arvind, Streif, Stefan

Abstract

An LLM agent using tools can fail silently. It stops before the task is done, raises no error, and answers fluently. Two kinds of monitor address this, one deciding whether a run failed and the other naming the step where failure became unavoidable. Much of the reported performance on both tasks comes from regularities that need no model. Failing runs are longer, and the step labelled decisive is usually the last. A rule that always names the last step is correct $67\%$ to $96\%$ of the time here, without reading the run at all. That ceiling reflects how the labels were written, not how agents break. On failures whose decisive step replay can confirm, it falls to near chance. Reporting gain over that rule therefore changes what a score means. Under that correction, methods tested here gain nothing on give-up failures, while several published systems keep a clear advantage on other splits. For detection, the agent's internal states beat transcript length and text similarity on the agent benchmark, and hold up where the text baseline does not. Precision stays low when failures are rare, so such a monitor should rank runs for review, not replace it.

Chinese Translation

使用工具的LLM智能体可能会无声失败。它在任务完成之前停止，不会引发错误，并且回答流畅。有两种监控方式可以解决这个问题，一种是判断运行是否失败，另一种是标识失败变得不可避免的步骤。报告的这两项任务的性能大部分来自于不需要模型的规律性。失败的运行时间较长，而被标记为决定性步骤的通常是最后一步。一个始终标记最后一步的规则在这里的正确率为67%到96%，而无需读取运行。这一上限反映了标签的编写方式，而不是智能体失败的方式。对于那些可以通过重放确认决定性步骤的失败，准确率接近随机。因此，报告相对于该规则的增益改变了分数的意义。在这一修正下，本文测试的方法在放弃失败上没有任何增益，而一些已发布的系统在其他分割上仍保持明显优势。在检测方面，智能体的内部状态在智能体基准测试中优于转录长度和文本相似性，并在文本基线无法维持的情况下表现良好。当失败较为罕见时，精确度保持较低，因此这样的监控应对运行进行排序以供审查，而不是替代审查。

View on arXiv Download PDF AI Translation

cs.AI / 209 / 2610.16725

Trace-to-Architecture: Capability Bottleneck Diagnosis for Harness Revision

追踪到架构：用于工具修订的能力瓶颈诊断

Wang, Wenfeng, Shi, Jianrui, Guo, Mingang, Ma, Jiayuan, Lu, Yuqi, Sun, Jiajun, Zhao, Guojiang, Li, Jing

Abstract

Improving an LLM agent requires knowing what to change when it fails. Similar failures can require different fixes, and resolving one problem can expose another. We introduce Trace-to-Architecture (T2A), a framework that uses execution trajectories to determine what to change in an agent's harness and when those changes apply. A development agent diagnoses failures using execution evidence from successful and failed tasks, revises the corresponding harness components, and then checks whether target tasks improve, selected successful tasks remain successful, and the edits are actually triggered. New execution evidence guides subsequent diagnosis and revision. Across ALFWorld, ScienceWorld, and AppWorld, T2A achieves the highest aggregate score for all three evaluated task agents, exceeding their respective strongest baselines by 6.2--30.3 points. We also integrate T2A's diagnosis and validation procedures into Meta-Harness, improving the resulting harnesses' mean ScienceWorld score by 21.6 points from the same starting harness.

Chinese Translation

提升大型语言模型（LLM）代理的性能需要了解在失败时应做出何种改变。类似的失败可能需要不同的修复方案，而解决一个问题可能会暴露出另一个问题。我们提出了追踪到架构（Trace-to-Architecture, T2A）框架，该框架利用执行轨迹来确定在代理的工具中应做何种改变以及何时应用这些改变。开发代理使用来自成功和失败任务的执行证据来诊断失败，修订相应的工具组件，然后检查目标任务是否有所改善、选定的成功任务是否仍然保持成功，以及这些编辑是否真正被触发。新的执行证据指导后续的诊断和修订。在 ALFWorld、ScienceWorld 和 AppWorld 中，T2A 在所有三个评估任务代理中实现了最高的综合得分，超出了各自最强基线 6.2 至 30.3 分。我们还将 T2A 的诊断和验证程序整合到 Meta-Harness 中，使得结果工具的平均 ScienceWorld 得分从相同的起始工具提高了 21.6 分。

View on arXiv Download PDF AI Translation

cs.AI / 210 / 2610.16730

Queue-Aware Token Communications: Bridging AI Representation and Wireless Service

队列感知的令牌通信：连接人工智能表示与无线服务

Xiao, Yuquan, Du, Qinghe, Zhang, Chaoyang, Cheng, Wenchi, Wang, Jingqing, Chen, Tao, Wang, Jiangzhou

Abstract

The rapid advancement of foundation models and edge intelligence is transforming wireless networks from data-oriented communication systems into artificial intelligence (AI)-native service platforms. Token communications (TokenCom) has emerged as a promising paradigm that directly transmits task-oriented AI representations. However, existing studies typically assume that generated tokens can be transmitted immediately, overlooking the queueing dynamics arising from the mismatch between token workload generation and time-varying wireless service capability. To address this limitation, we introduce Queue-Aware TokenCom, which explicitly incorporates token queues into the AI service pipeline. We present an end-to-end architecture integrating AI representation generation, token workload management, wireless transmission, and downstream task inference. We then discuss two key design directions: budget-aware token representation optimization and queue-aware token service optimization, with the token budget serving as a critical interface between AI performance and communication efficiency. Moreover, we investigate fixed and adaptive token budget strategies to characterize the fundamental tradeoff between AI performance and communication efficiency in dynamic wireless environments. Finally, we outline research directions toward closed-loop, task-aware, and scalable TokenCom systems.

Chinese Translation

基础模型和边缘智能的快速发展正在将无线网络从以数据为导向的通信系统转变为人工智能（AI）原生服务平台。令牌通信（TokenCom）作为一种新兴的有前景的范式，直接传输面向任务的AI表示。然而，现有研究通常假设生成的令牌可以立即传输，忽视了由于令牌工作负载生成与时变无线服务能力之间的不匹配而产生的排队动态。为了解决这一局限性，我们提出了队列感知的TokenCom，它明确将令牌队列纳入AI服务管道中。我们提出了一种端到端架构，整合了AI表示生成、令牌工作负载管理、无线传输和下游任务推理。接着，我们讨论了两个关键设计方向：预算感知的令牌表示优化和队列感知的令牌服务优化，其中令牌预算作为AI性能与通信效率之间的关键接口。此外，我们研究了固定和自适应令牌预算策略，以表征动态无线环境中AI性能与通信效率之间的基本权衡。最后，我们概述了朝向闭环、任务感知和可扩展的TokenCom系统的研究方向。

View on arXiv Download PDF AI Translation

cs.AI / 211 / 2610.16761

Separating Prompt Harmfulness from Interaction Safety via Partial Generation

通过部分生成将提示的有害性与交互安全性分离

Mishra, Ram Vikas, Aggarwal, Shashank, Kumar, Jahnavi, Awekar, Amit

Abstract

LLM-based information systems need safety guardrails. Most guardrails classify the prompt alone. If the prompt looks harmful, the system rejects it before any answer is generated. This is costly. Many rejected prompts would have received a safe response. Users lose help they should have received from the information system. This cost grows as conversational and generative search becomes common. Prompt-level labels treat harm as a property of the query. But harm depends on what the model actually writes. Existing input-output guards do look at responses. But they judge a finished response, so the decision comes after full generation. However, we observe that safety of model interaction with the input prompt can be inferred early with only partial output. So the system can start generating, watch the partial response, and stop only if it turns harmful. We test this idea in three steps. First, we build a dataset of 41,944 prompt-response pairs with seven open weights LLMs from three families: Llama, Gemma, and Qwen. Second, we show that prompt harmfulness and interaction safety differ. Prompt labels and response labels agree on only 50.46\% of items (Cohen's $\kappa$ = 0.06). Of the prompts that make any model respond harmfully, only 3.5\% do so for all seven LLMs. Third, we show that partial responses carry most of the signal. Fine-tuned classifiers predict response-level harm well (best AUPRC 0.885, F1 0.807). About one third of the response tokens recover almost full-response AUPRC.

Chinese Translation

基于大型语言模型（LLM）的信息系统需要安全防护措施。大多数防护措施仅对提示进行分类。如果提示看起来有害，系统会在生成任何答案之前拒绝它。这是有成本的。许多被拒绝的提示本可以获得安全的响应。用户失去了信息系统本应提供的帮助。随着对话和生成搜索的普及，这种成本不断增加。提示级标签将有害性视为查询的一个属性。但有害性依赖于模型实际生成的内容。现有的输入输出防护措施确实关注响应，但它们是在完成响应后进行判断，因此决策是在完全生成之后作出的。然而，我们观察到，模型与输入提示的交互安全性可以通过仅部分输出进行早期推断。因此，系统可以开始生成，观察部分响应，并仅在其变得有害时停止。我们通过三个步骤测试这一想法。首先，我们构建了一个包含41,944个提示-响应对的数据集，使用来自三个家族的七个开放权重的LLM：Llama、Gemma和Qwen。其次，我们展示了提示的有害性与交互安全性之间的差异。提示标签和响应标签在仅50.46%的项目上达成一致（Cohen's κ = 0.06）。在使任何模型产生有害响应的提示中，只有3.5%在所有七个LLM中均如此。第三，我们展示了部分响应携带了大部分信号。经过微调的分类器能够很好地预测响应级别的有害性（最佳AUPRC 0.885，F1 0.807）。大约三分之一的响应标记几乎恢复了完整响应的AUPRC。

View on arXiv Download PDF AI Translation

cs.AI / 212 / 2610.16769

Geospatial Vector Geometry as a Native Spatial Modality for Vision-Language Models

地理空间向量几何作为视觉-语言模型的原生空间模态

Fei, Huang, Chen, Boyi, Xue, Hao, Yue, Yang, Xu, Yehong, Zhou, Xiaofang

Abstract

Geospatial reasoning links VLMs to real-world action and decision making. Reliable reasoning requires accurate spatial information and effective spatial representations. However, current VLMs obtain this information mainly from language and vision. At larger geographic scales, limited visual observations make it difficult to establish accurate spatial references and recover precise spatial relations. This leads to a bottleneck for VLM spatial perception and reasoning. This paper proposes that geospatial vector geometry can become a native spatial modality within a VLM and introduces GeoVec-VLM. First, we develop a structured geometry representation that preserves the spatial structure and arrangement of vector objects, building an explicit spatial basis. Second, we introduce a geometry interaction strategy that brings this basis into multimodal computation, where it interacts with visual content and language references according to the queried spatial relations. Together, these designs allow perception and reasoning to operate on shared geometric references. Across five spatial reasoning tasks, GeoVec-VLM improves mean accuracy by about 11\% relative to the strongest baseline, maintains its gains under geographic transfer, and shows greater geometric consistency. Our results show that geospatial vector geometry offers a promising internal spatial basis for VLM spatial reasoning. https://github.com/TXFR/GeoVector-VLM

Chinese Translation

地理空间推理将视觉-语言模型（VLMs）与现实世界的行动和决策联系起来。可靠的推理需要准确的空间信息和有效的空间表示。然而，目前的视觉-语言模型主要从语言和视觉中获取这些信息。在更大地理尺度上，有限的视觉观察使得建立准确的空间参考和恢复精确的空间关系变得困难。这导致了视觉-语言模型在空间感知和推理方面的瓶颈。本文提出地理空间向量几何可以成为视觉-语言模型中的一种原生空间模态，并介绍了GeoVec-VLM。首先，我们开发了一种结构化的几何表示，保留了向量对象的空间结构和排列，建立了明确的空间基础。其次，我们引入了一种几何交互策略，将这一基础引入多模态计算中，根据查询的空间关系与视觉内容和语言参考进行交互。这些设计共同允许感知和推理在共享的几何参考上进行。通过五个空间推理任务，GeoVec-VLM的平均准确率相较于最强基线提高了约11%，在地理迁移下保持了其优势，并显示出更大的几何一致性。我们的结果表明，地理空间向量几何为视觉-语言模型的空间推理提供了一个有前景的内部空间基础。

View on arXiv Download PDF AI Translation

cs.AI / 213 / 2610.16776

Game-Creation-Agents: Autonomous Game Production

游戏创作代理：自主游戏制作

Tencent Multimodal RL Team, Zhou, Xiangxin, Tang, Jiaqi, Ma, Ke, Yang, Wenhao, Yao, Jiarui, Niu, Boye, Guo, Haiyang, Zhao, Yifei, Yao, Huanjin, Jian, Pu, Li, Ruihuang, Wu, Haitao, Huang, Yushi, Liu, Xiangyan, Sun, Yan, Li, Kaixin, Pang, Tianyu

Abstract

Game-Creation-Agents (GCA) turns a natural-language request into a runnable Phaser 4 browser game. The pipeline first builds a multimodal game design document that couples mechanics and scene contracts to grounded visual references. It then projects that document into a runtime scaffold, machine-readable contracts, and deterministic tests. Eight specialized agents next carry out production: scenes are implemented serially, assets are produced in parallel, and one reduced manifest joins the two branches into a single build. Build-linked real-UI checks and a code-blind critic then verify that build and decide whether the game may ship. Each blocker returns to the agent that owns the affected artifact for a bounded repair, and the game is released once all the gates pass. We evaluate finished games separately, through their rendered interfaces alone, over prompt-matched tasks run under a common protocol. GCA achieves stronger requirement satisfaction, core mechanics, content depth, and functional visuals than a comparable open pipeline. In a side-by-side human study, raters also score GCA's games higher on every dimension and prefer them overall.

Chinese Translation

游戏创作代理（Game-Creation-Agents, GCA）将自然语言请求转化为可运行的 Phaser 4 浏览器游戏。该流程首先构建一个多模态游戏设计文档，将机制和场景契约与具体的视觉参考相结合。接着，它将该文档投影到运行时框架、机器可读的契约和确定性测试中。随后，八个专业代理执行生产：场景按序实现，资产并行生成，一个简化的清单将两个分支合并为一个单一构建。构建关联的真实用户界面检查和一个代码盲评审者随后验证该构建，并决定游戏是否可以发布。每个阻碍因素会返回给拥有受影响工件的代理进行有限修复，所有关卡通过后游戏便会发布。我们通过渲染的界面单独评估完成的游戏，基于在共同协议下进行的提示匹配任务。与可比的开放流程相比，GCA 在需求满足、核心机制、内容深度和功能视觉方面表现更强。在一项并行的人类研究中，评审者在每个维度上也给予 GCA 的游戏更高的评分，并整体上更偏爱这些游戏。

View on arXiv Download PDF AI Translation

cs.AI / 214 / 2610.16782

LIBERO-Industry: Benchmarking Robotic Manipulation under Industrial Process Constraints

LIBERO-Industry：在工业过程约束下的机器人操作基准测试

Sung, Ming Shin Cindy, Leijiahao, Duan, Wang, Haoxiang, Tan, Boning, Fu, Liangna, Tan, Chao, Li, Yafu, Chen, Yulong, Wan, Chaoqun, Huang, Zhen, Liang, Shuxian, Hua, Xian-Sheng

Abstract

Industrial manipulation in flexible manufacturing, e.g., high-mix, low-volume (HMLV) electronics assembly, requires robot policies to adapt to changing products and production conditions. Such settings impose industrial process constraints, including product variants, moving material flows, finite execution windows, and tight geometric tolerances. However, existing evaluation frameworks do not systematically capture these constraints when assessing generalist robot policies. To address this gap, we introduce LIBERO-Industry, a novel benchmark built on the widely used robot learning benchmark LIBERO, that translates the aforementioned industrial constraints into controllable task conditions and measurable success criteria. Specifically, LIBERO-Industry comprises three classic task scenarios: Static-to-Dynamic (S2D) loading, Dynamic-to-Static (D2S) unloading, and Dynamic-to-Dynamic (D2D) transfer. The suites contain 216 tasks in total that systematically vary object geometry, conveyor speed, interaction-window duration, and geometric acceptance region size. Extensive experiments across five generalist robot policies (including vision-language-action models and world-action models) show that performance progressively degrades as industrial constraints are introduced and combined. Under a shared fine-tuning and evaluation protocol, the LIBERO-Industry co-training condition achieved higher observed target-transfer completion than the original-LIBERO condition.

Chinese Translation

灵活制造中的工业操作，例如高混合、低产量（HMLV）电子组装，要求机器人策略能够适应不断变化的产品和生产条件。这些环境施加了工业过程约束，包括产品变体、移动物料流、有限的执行窗口和严格的几何公差。然而，现有的评估框架在评估通用机器人策略时并未系统性地捕捉这些约束。为了解决这一问题，我们引入了LIBERO-Industry，这是一个基于广泛使用的机器人学习基准LIBERO构建的新基准，它将上述工业约束转化为可控的任务条件和可测量的成功标准。具体而言，LIBERO-Industry包含三个经典任务场景：静态到动态（S2D）装载、动态到静态（D2S）卸载和动态到动态（D2D）转移。该套件总共包含216个任务，系统性地变化物体几何形状、输送带速度、交互窗口持续时间和几何接受区域大小。在五种通用机器人策略（包括视觉-语言-动作模型和世界-动作模型）上的广泛实验表明，随着工业约束的引入和组合，性能逐渐下降。在共享的微调和评估协议下，LIBERO-Industry的协同训练条件实现了比原始LIBERO条件更高的目标转移完成率。

View on arXiv Download PDF AI Translation

cs.AI / 215 / 2610.16838

Same Operation, Different Permission: Evaluating Whether Guardrails Honor Authorization

相同操作，不同权限：评估防护措施是否遵循授权

Jing, Huihao, Chen, Shaojin, Hu, Wenbin, Liu, Yuxuan, Shi, Haochen, Xie, Zhongwei, Li, Haoran, Song, Yangqiu

Abstract

As language-model agents write code, move money and change records on behalf of users, deciding whether an action is authorized has become a central safety problem. Guardrails are often placed in front of these actions, yet few evaluations ask whether their decisions follow authorization: blocking unauthorized actions while still allowing the same actions under valid permission. We introduce \bench{}, a benchmark that contrasts authorization evidence for the same operation in its language track. It combines expert-annotated natural-language scenarios with executed and verified operations, plus externally sourced native-policy requests. \bench{} checks each decision for correctness, usability and resulting effect, and tracks each request across guardrail changes. Our experiments show that many blocked operations stem from failed reviews, not correct denials. Repairing the output format restores usable decisions but changes which authorized requests are honored, and adding a second reviewer removes some errors while rejecting valid authorization. Guardrails should therefore be evaluated on both the unauthorized actions they stop and the authorized actions they let through.

Chinese Translation

随着语言模型代理为用户编写代码、转移资金和更改记录，判断某一行为是否被授权已成为一个核心安全问题。防护措施通常被置于这些行为之前，但很少有评估询问其决策是否遵循授权：在阻止未授权行为的同时仍允许在有效权限下进行相同的操作。我们引入了 ench{}，一个对比语言轨道中相同操作的授权证据的基准。它结合了专家注释的自然语言场景与已执行和验证的操作，以及外部来源的本地政策请求。 ench{} 检查每个决策的正确性、可用性和结果效果，并跟踪每个请求在防护措施变化中的情况。我们的实验表明，许多被阻止的操作源于审核失败，而非正确拒绝。修复输出格式恢复了可用的决策，但改变了哪些被授权的请求得到了尊重，增加第二位审核者则在消除一些错误的同时拒绝了有效的授权。因此，防护措施应在它们阻止的未授权行为和它们允许通过的授权行为两方面进行评估。

View on arXiv Download PDF AI Translation

cs.AI / 216 / 2610.16876

CLIVO: Differentiable Cluster-Based Instruction Mixing with Learned Validation Objectives

CLIVO：基于可微分聚类的指令混合与学习验证目标

Wang, Yuhan, Hu, Rui, Di, Batian, Jiang, Tan, Ma, Dongsheng, Dong, Leichao, Li, Yuanzi, Feng, Yalin

Abstract

Target-aware multimodal instruction selection requires useful examples and an appropriate balance of supervision. Sample-level rankings can overrepresent similar content without optimizing this balance; group-level methods face coarse units, predefined set-level validation signals, and costly or inaccurate mixture search. We propose CLIVO, a target-aware framework connecting fine-grained semantic allocation, automatically learned validation objectives, and differentiable mixture search. Specifically, semantic clusters provide independently adjustable shares at the hundred-cluster scale. Correlations between per-example merged-proxy losses and real probe performance identify target-relevant validation examples; learned nonnegative weights combine their losses into an objective. A single backward pass through the merged proxy supplies gradients for all shares, avoiding an additional mixture-response regression model. Selecting 100K instructions from an 8M pool, CLIVO attains 97.2% of the full-data four-target average score with only 1.25% of the data. It improves on the strongest evaluated selection baseline by 1.16 points using random within-cluster sampling; combining learned quotas with sample ranking increases this advantage to 1.55 points. Analyses show that finer units benefit most targets, short-trained experts preserve mixture responses, sample weighting retains useful validation information, and direct search consistently reduces the learned objective. CLIVO also supports interpretable allocation, modular selection, and reuse across targets and granularities.

Chinese Translation

目标感知的多模态指令选择需要有效的示例和适当的监督平衡。样本级排名可能会过度代表相似内容，而未能优化这种平衡；组级方法面临粗糙单元、预定义的集合级验证信号以及成本高昂或不准确的混合搜索。我们提出了CLIVO，一个目标感知框架，连接细粒度的语义分配、自动学习的验证目标和可微分的混合搜索。具体而言，语义聚类在百个聚类规模上提供独立可调的份额。每个示例合并代理损失与真实探测性能之间的相关性识别出与目标相关的验证示例；学习的非负权重将它们的损失组合成一个目标。通过合并代理的单次反向传播为所有份额提供梯度，避免了额外的混合响应回归模型。从800,000个池中选择100,000个指令，CLIVO仅使用1.25%的数据就达到了全数据四目标平均分的97.2%。它通过随机的组内采样在最强评估选择基线的基础上提高了1.16分；将学习的配额与样本排名结合，使这一优势增加到1.55分。分析表明，细粒度单元最有利于大多数目标，短期训练的专家保留混合响应，样本加权保留有用的验证信息，而直接搜索始终减少学习目标。CLIVO还支持可解释的分配、模块化选择以及在不同目标和粒度间的重用。

View on arXiv Download PDF AI Translation

cs.AI / 217 / 2610.16881

Acceptable Misalignment

可接受的偏差

van Lier, Maud, Marinucci, Ludovica, Santucci, Vieri Giuliano, Chatila, Raja, Doncieux, Stéphane, Garrabé, Émiland, Duro, Richard

Abstract

There is a specific challenge designers face in developing autonomous robots capable of open-ended learning. This challenge is how one can develop such systems in an ethically sound way when there are many unknowns during design time about their manner of deployment, what they can and will learn, and how these systems will act on the basis of what they have learned. One response to this challenge is to align, as far as possible, the activities and goals of autonomous robots with human values, raising the question of how to make this alignment possible, a question known as the value alignment problem. We argue that the value alignment problem faces a number of challenges that show that full alignment is both theoretically and practically impossible, meaning that misalignment incidents will keep occurring. This undermines current alignment approaches, as the attempt to maximize alignment cannot properly deal with the possibility that the risk of misalignment that remains after alignment methods have been implemented might still be unacceptable to society. To overcome this difficulty, we argue that a shift in perspective is required from a focus on the maximization of alignment to the conditions under which misalignment incidents become acceptable. In this paper, we explore what this shift in perspective entails. We introduce the concept of acceptable misalignment, formulate normative requirements for alignment mechanisms aimed at bringing misalignment incidents to an acceptable level, and we conclude by presenting a new methodological framework that consists of alignment mechanisms that can meet these requirements.

Chinese Translation

设计师在开发能够进行开放式学习的自主机器人时面临一个特定的挑战。这个挑战在于，如何在设计阶段面对许多未知因素时，以伦理上合理的方式开发这样的系统，包括它们的部署方式、它们可以和将要学习的内容，以及这些系统将如何根据所学知识进行行动。对此挑战的一个回应是尽可能地将自主机器人的活动和目标与人类价值观对齐，这引出了如何实现这种对齐的问题，这个问题被称为价值对齐问题。我们认为，价值对齐问题面临许多挑战，这些挑战表明完全对齐在理论上和实践上都是不可能的，这意味着偏差事件将不断发生。这削弱了当前的对齐方法，因为试图最大化对齐的努力无法妥善处理在实施对齐方法后仍然可能存在的、对社会来说不可接受的偏差风险。为了克服这一困难，我们认为需要从关注对齐的最大化转变为关注偏差事件变得可接受的条件。在本文中，我们探讨了这一视角转变的含义。我们引入了可接受偏差的概念，制定了旨在将偏差事件降低到可接受水平的对齐机制的规范要求，并通过提出一个新的方法论框架来总结，这个框架由能够满足这些要求的对齐机制组成。

View on arXiv Download PDF AI Translation

cs.AI / 218 / 2610.16887

Transparency as Answerability: Role-Dependent Information Flow Modeling

透明性作为问责性：角色依赖的信息流模型

Brännström, Mattias, Xanthopoulou, Themis Dimitra, Jiang, Lili

Abstract

Transparency in AI-supported decision systems is widely required but poorly defined. Existing approaches treat transparency as a property of systems or components, producing characteristic failures: assessments that do not transfer across organizational boundaries, stakeholder questions that cannot be formulated within the available frame of analysis, and component-level findings that say nothing about system-level behavior. We propose treating transparency as a relation: a system is transparent to a stakeholder to the degree that their legitimate questions are answerable from its structure. We formalize this using information flow networks, in which transparency questions are expressed in terms of network structure and answerability is the condition that the system settles them. The framework supports three dimensions of analysis: structural answerability from the network alone, compositional answerability across organizational boundaries for deployed systems, and role-dependent answerability given a stakeholder's accessible sub-network. We demonstrate the framework on a real pre-integration assessment of an AI recruitment tool in a public sector municipality. The assessment uses participatory design and interviews to model the decision system structure. We show how structural and role-dependent transparency questions can be formally stated and analyzed across the Provider and Deployer boundary where the AI system is composed together with the deployer's decision process.

Chinese Translation

在人工智能支持的决策系统中，透明性被广泛要求但定义模糊。现有方法将透明性视为系统或组件的属性，导致特征性失败：评估无法跨越组织边界转移，利益相关者的问题无法在现有分析框架内提出，以及组件级发现无法反映系统级行为。我们建议将透明性视为一种关系：一个系统对利益相关者的透明程度取决于其合法问题是否可以从系统结构中得到解答。我们使用信息流网络对此进行形式化，其中透明性问题通过网络结构表达，而可回答性是系统解决这些问题的条件。该框架支持三个分析维度：仅从网络结构出发的结构性可回答性、跨组织边界的组合可回答性（针对已部署系统），以及给定利益相关者可访问子网络的角色依赖可回答性。我们在一个公共部门市政的人工智能招聘工具的实际预集成评估中展示了该框架。该评估采用参与式设计和访谈来建模决策系统结构。我们展示了如何在提供者与部署者边界上正式陈述和分析结构性及角色依赖的透明性问题，在该边界上，人工智能系统与部署者的决策过程相结合。

View on arXiv Download PDF AI Translation

cs.AI / 219 / 2610.16895

Observe, Share, or Predict? Disentangling Partner Information in Embodied Multi-Agent Collaboration

观察、分享还是预测？解构具身多智能体协作中的伙伴信息

Szlatincsán, Adrián, Ziliotto, Filippo, Lepri, Bruno, Serafini, Luciano, Luca, Massimiliano, Campari, Tommaso

Abstract

LLMs are used as high-level controllers for embodied agents, but it remains unclear whether they support collaborative reasoning, such as anticipating a collaborator's actions in real-time teamwork without dialogue. We ask if observing the partner's state, exchanging object knowledge, and predicting the partner's next action, shape task outcomes and coordination when agents can share percepts but never intentions, plans, or task-level messages, and whether the two agents occupy symmetric roles. Using the Habitat-based PARTNR benchmark, we introduce a Decoupled Partial Observability regime that enforces strictly individual perception, and we augment decentralized zero-shot ReAct planners with Other Agent's State (OAS) observation, object Knowledge Exchange (KE) between co-located agents, and Other Agent's Next Action (OANA) prediction, evaluated on the full 2 x 2 x 2 grid with four planners on 1,000 validation tasks. OAS observation improves task quality, reducing extraneous effort for every planner, whereas KE primarily improves execution efficiency, cutting simulation steps by up to 25% with little change in success. The two are largely redundant for strong planners but complementary for the weakest, for which KE without the OAS is the worst configuration. OANA prediction lowers the success rate in 15 of 16 model x context pairs while raising explicit partner mentions to 55-100%, and its accuracy gains lie mainly in the coarse action type, partly because agents recognize the action the partner is already performing. The robot retrieves objects sooner and completes its tasks earlier than the simulated human. Shared context reduces total task time by shortening the remaining duration that the human works alone. Also, coordination benefits arise from responding to shared context, especially the OAS, rather than explicit anticipation, yielding shallow, planner-dependent partner modeling.

Chinese Translation

大型语言模型（LLMs）被用作具身智能体的高级控制器，但尚不清楚它们是否支持协作推理，例如在没有对话的实时团队合作中预测合作者的行为。我们探讨观察伙伴状态、交换物体知识和预测伙伴下一步行动是否会影响任务结果和协调，尤其是在智能体能够共享感知但从不共享意图、计划或任务级信息的情况下，以及这两个智能体是否处于对称角色。利用基于Habitat的PARTNR基准，我们引入了一种解耦部分可观测性机制，该机制严格执行个体感知，并且我们增强了去中心化的零样本ReAct规划器，加入了其他智能体状态（OAS）观察、共处智能体之间的物体知识交换（KE）和其他智能体下一步行动（OANA）预测，在包含四个规划器的1000个验证任务的2 x 2 x 2网格上进行评估。OAS观察提高了任务质量，减少了每个规划器的额外努力，而KE主要提高了执行效率，将模拟步骤减少了多达25%，成功率变化不大。这两者对于强规划器而言大体上是冗余的，但对于最弱的规划器则是互补的，其中没有OAS的KE是最差的配置。OANA预测在16个模型与上下文对中有15个降低了成功率，同时将明确的伙伴提及率提高到55-100%，其准确性提升主要体现在粗略的行动类型上，部分原因是智能体能够识别出伙伴正在执行的动作。机器人比模拟人类更早检索物体并更早完成任务。共享上下文通过缩短人类单独工作的剩余时间来减少总任务时间。此外，协调效益来自于对共享上下文的响应，特别是OAS，而非明确的预期，从而导致浅层、依赖于规划器的伙伴建模。

View on arXiv Download PDF AI Translation

cs.AI / 220 / 2610.16902

Neural Bayesian Networks for Modular and Scalable Decision-Making Under Uncertainty

用于不确定性下模块化和可扩展决策的神经贝叶斯网络

Briglia, Giovanni, Lippi, Marco, Mariani, Stefano, Zambonelli, Franco

Abstract

Bayesian Networks (BNs) are established tools for letting agents reason under uncertainty in sequential decision-making. They split a joint distribution into local conditional mechanisms arranged on a directed acyclic graph, which makes them modular, interpretable, and ready to perform probabilistic inference. Their classical formulation, however, has limitations in both expressiveness and scale: each conditional is usually stored as a table or a linear-Gaussian form, and inference evaluates the graph one node and one query at a time. In this paper we introduce Neural Bayesian Networks (NBNs), in which each conditional is a differentiable, tensor-valued mechanism, while the graph and its factorization are unchanged. NBNs retain the modularity and local mechanisms of BNs, can represent any learnable conditional distribution, and run natively on GPUs. We also release NBN, a PyTorch-based library for learning and inference in NBNs, together with a public benchmark. Our evaluation shows gains in both expressiveness and scale: NBNs fit distributions that classical BN libraries struggle to represent, and on large networks and large batches of queries they run two orders of magnitude faster thanks to GPU parallelization. NBN is available at https://github.com/Giovannibriglia/NeuralBayesianNetworks.

Chinese Translation

贝叶斯网络（Bayesian Networks, BNs）是让代理在不确定性下进行序列决策推理的成熟工具。它们将联合分布分解为在有向无环图上排列的局部条件机制，这使得它们具有模块化、可解释性，并能够进行概率推理。然而，它们的经典形式在表达能力和规模上存在局限：每个条件通常以表格或线性高斯形式存储，推理过程一次只评估一个节点和一个查询。在本文中，我们引入了神经贝叶斯网络（Neural Bayesian Networks, NBNs），其中每个条件都是一个可微分的张量值机制，而图及其因子分解保持不变。NBNs保留了BNs的模块性和局部机制，能够表示任何可学习的条件分布，并能够在GPU上原生运行。我们还发布了NBN，一个基于PyTorch的库，用于NBNs的学习和推理，并提供了一个公共基准。我们的评估显示在表达能力和规模上都有所提升：NBNs能够拟合经典BN库难以表示的分布，并且在大型网络和大量查询的情况下，由于GPU并行化，它们的运行速度提高了两个数量级。NBN可在https://github.com/Giovannibriglia/NeuralBayesianNetworks获取。

View on arXiv Download PDF AI Translation

cs.AI / 221 / 2610.16911

Audit tools of the trade: Automated Auditor Elicitation Can Be Sensitive to the Context the Auditor Is Given

行业审计工具：自动化审计师引导对审计师所处上下文的敏感性

Slater, Ben, Lundqvist, Ryan, Richter, Leo, Wei, Kevin

Abstract

Does providing additional context to automated AI auditors improve their effectiveness? Using the Petri tool to audit LLM coding agents, we test the impact of providing two types of resources (an external codebase and benchmark transcripts) to LLM auditors, across four auditor-target LLM pairs and three behaviours (harmful compliance, reward hacking, sycophancy). (1) External codebase: We provide the auditor LLM with a fixed, external open source codebase, which prior work finds improves the realism of synthetic audits. Providing this resource suppresses reward hacking in both auditor-target pairs whose baseline rate is above the floor (4.8 to 1.3 and 3.9 to 1.8 on Petri's default 1-10 scale); the remaining two pairs sit at the floor throughout. The suppression recurs in three of four pairs under a different scenario and resource bundle. (2) External benchmark transcripts: We provide the auditor with transcripts of the target's performance on benchmarks measuring similar behaviours, intended to inform the auditor about the target's behavioural patterns. Providing this resource raises reward hacking scores in all four pairs (from 1.7 to 6.1 in one case). However, subsequent experiments show that supplying only the related benchmark's abstract, without transcripts, also increases reward hacking, and the effect is absent under a different framing of the same behaviour. Overall, we find no consistent effect of either resource outside reward hacking, but we find substantial variability in judge-scored behaviour across conditions. Our results underscore the importance of careful piloting with tools such as Petri, as the measured rate of target behaviour is highly sensitive to configuration decisions. We hope our results inspire more research into the reliability and robustness of automated alignment auditing.

Chinese Translation

提供额外的上下文信息给自动化AI审计师是否能提高其有效性？我们使用Petri工具对LLM编码代理进行审计，测试向LLM审计师提供两种资源（外部代码库和基准转录本）的影响，涵盖四对审计师-目标LLM和三种行为（有害合规、奖励黑客、谄媚）。(1) 外部代码库：我们向审计师LLM提供一个固定的外部开源代码库，先前的研究发现这能提高合成审计的真实性。提供该资源抑制了在两个审计师-目标对中，基线率高于底线的奖励黑客（在Petri的默认1-10尺度上，分别为4.8降至1.3和3.9降至1.8）；其余两个对在整个过程中保持在底线。抑制现象在不同场景和资源组合下的四对中的三对中重复出现。(2) 外部基准转录本：我们向审计师提供目标在测量类似行为的基准上的表现转录本，旨在告知审计师目标的行为模式。提供该资源在所有四对中提高了奖励黑客分数（在一个案例中从1.7提高到6.1）。然而，后续实验表明，仅提供相关基准的摘要而不提供转录本也会增加奖励黑客，并且在同一行为的不同框架下该效应消失。总体而言，我们发现除了奖励黑客外，任何资源都没有一致的效果，但在不同条件下，评审评分的行为存在显著的变异性。我们的结果强调了使用像Petri这样的工具进行仔细试点的重要性，因为目标行为的测量率对配置决策高度敏感。我们希望我们的结果能激励更多关于自动化对齐审计的可靠性和稳健性的研究。

View on arXiv Download PDF AI Translation

cs.AI / 222 / 2610.16916

Transparent Information Flow Modeling for Role-Aware AI Act Assessment

面向角色的人工智能法案评估的透明信息流建模

Brännström, Mattias, Xanthopoulou, Themis Dimitra, Jiang, Lili

Abstract

The European Union's Artificial Intelligence Act introduces a structured framework for governing high-risk AI systems, emphasizing transparency, risk management, and role-specific accountability. However, current assessment practices often rely on monolithic system views or disconnected checklists, limiting their ability to capture the socio-technical complexity of real-world deployments. We explore the use of the Information Flow Methodology (IFM), a modeling-based approach to AI Act compliance assessment. IFM represents decision systems -- whether algorithmic, human, or hybrid -- as networks of information sites and transformation channels, enabling structured analysis of how information, decisions, and responsibilities flow across organizational and technical boundaries. We apply IFM in a real-world case involving a recruitment support tool used by a public-sector organization, evaluating both the Provider's system and the Deployer's recruitment process. Our results demonstrate that IFM can serve as a scaffold to bridge both component-based and holistic perspectives, and how technical systems are situated within socio-technical contexts. In doing so, IFM supports transparency, identifies risk pathways, and enables alignment analysis between intended and actual use. We argue that such modeling provides a foundational tool for effective, context-sensitive AI governance and accountability -- offering capabilities beyond conventional checklist-based assessments.

Chinese Translation

欧盟的人工智能法案引入了一个结构化框架，用于治理高风险的人工智能系统，强调透明性、风险管理和角色特定的问责制。然而，当前的评估实践往往依赖于单一系统视图或不连贯的检查清单，限制了其捕捉现实部署的社会技术复杂性的能力。我们探讨了信息流方法论（Information Flow Methodology, IFM）的使用，这是一种基于建模的人工智能法案合规评估方法。IFM将决策系统——无论是算法、人工还是混合——表示为信息站点和转化通道的网络，从而能够结构化分析信息、决策和责任如何在组织和技术边界之间流动。我们在一个涉及公共部门组织使用的招聘支持工具的真实案例中应用了IFM，评估了提供者的系统和部署者的招聘过程。我们的结果表明，IFM可以作为一个支架，桥接基于组件的视角和整体视角，以及技术系统如何置于社会技术背景中。在此过程中，IFM支持透明性，识别风险路径，并使得预期使用与实际使用之间的对齐分析成为可能。我们认为，这种建模提供了一种有效的、情境敏感的人工智能治理和问责的基础工具——提供超越传统检查清单评估的能力。

View on arXiv Download PDF AI Translation

cs.AI / 223 / 2610.16927

WeFair: Fidelity-Aware Alignment from Human Feedback for Personalized Query Rewriting in Short-Video Search

WeFair：基于人类反馈的个性化查询重写中的忠诚度感知对齐

Ren, Yingtao, Li, Aolin, Cheng, Cheng, Hu, Huiyun, Liang, Huasheng, Xie, Baoxing, Yuan, Jun, Wu, Haijun, Wang, Chenxing

Abstract

Short-video search requires compact query rewriters that infer personalized intent from brief, ambiguous queries under strict latency constraints. The challenge is to decide when personalization is warranted and generate a reformulation that preserves the user's immediate information need. A rewriter optimized for fluency or engagement may still introduce unsupported details and redirect the search toward a different intent. Existing methods often underutilize users' actual reformulations and behavioral traces, relying instead on synthetic reasoning and indirect behavioral rewards that may produce more engaging candidates but cause intent drift. To build an effective rewriter that remains faithful to user intent, we propose WeFair, Fidelity-Aware Alignment from Human Feedback for Personalized Query Rewriting in WeChat short-video search. WeFair assigns real user behavioral signals to a progression from demonstration to preference to verification. We first mine actual human reformulations from search sessions, then apply multidimensional LLM screening and contextual verification to filter unsupported transitions without extensive manual annotation. During training, we begin with a supervised curriculum that progressively builds rewriting capabilities from large-scale human personalized query demonstrations, and then use direct preference optimization to align with users' subsequent consumption. In the final on-policy optimization stage, we construct a strict verification reward based on actual human reformulations to further boost performance with group relative policy optimization. Further analysis shows that auxiliary rewards can activate more rollout groups in GRPO but reduce intent fidelity. Online A/B testing in WeChat short-video search shows a 1.576% relative increase in effective clicked-query rate (VV>10s) and a 1.731% relative reduction in query reformulation rate.

Chinese Translation

短视频搜索需要紧凑的查询重写器，从简短而模糊的查询中推断个性化意图，同时满足严格的延迟限制。挑战在于决定何时需要个性化，并生成一个能够保留用户即时信息需求的重写。一个优化流畅性或参与度的重写器可能仍会引入不支持的细节，并将搜索引导到不同的意图。现有方法往往未能充分利用用户的实际重写和行为轨迹，而是依赖于合成推理和间接的行为奖励，这可能产生更具吸引力的候选项，但导致意图漂移。为了构建一个有效的重写器，忠实于用户意图，我们提出了WeFair，基于人类反馈的个性化查询重写中的忠诚度感知对齐。WeFair将真实用户行为信号分配到从示范到偏好再到验证的进程中。我们首先从搜索会话中挖掘实际的人类重写，然后应用多维LLM筛选和上下文验证，过滤不支持的转换，而无需大量手动标注。在训练过程中，我们从一个监督课程开始，逐步建立来自大规模人类个性化查询示范的重写能力，然后使用直接偏好优化与用户后续消费对齐。在最终的在线策略优化阶段，我们基于实际的人类重写构建严格的验证奖励，以通过群体相对策略优化进一步提升性能。进一步分析表明，辅助奖励可以在GRPO中激活更多的回滚组，但会降低意图忠诚度。在微信短视频搜索中的在线A/B测试显示，有效点击查询率（VV>10s）相对增加了1.576%，而查询重写率相对减少了1.731%。

View on arXiv Download PDF AI Translation

cs.AI / 224 / 2610.16975

MADDIE: Multi-Agent Drug Discovery Engine

MADDIE：多智能体药物发现引擎

Liu, Ye, Averly, Reza, Meyer, Avery, Baker, Frazier N., Gao, Ruoxi, Adu-Ampratwum, Daniel, Ning, Xia

Abstract

Drug discovery remains a resource-intensive and time-consuming process, involving billions of dollars and a decade of development to advance a drug to market. While many in silico tools have accelerated this process, operating them still demands substantial time, effort, and expertise from medicinal chemists. Autonomous systems built on Large Language Models (LLMs) have emerged as a promising solution, yet current designs limit their performance. Existing single-agent systems scale poorly to long-horizon, multi-step, and realistic tasks, whereas multi-agent systems typically execute predefined workflows that cannot capture the iterative, non-linear nature of a real drug discovery campaign. Here, we present MADDIE, a multi-agent drug discovery engine. It comprises a central agent, Orchestrator, and four specialized worker agents: HitIdentifier, LeadOptimizer, Synthesizer, and Evaluator, each representing a critical step in a drug discovery campaign. Under Orchestrator's orchestration and coordination, these worker agents are composed into flexible workflows, enabling MADDIE to adapt its strategy and navigate the drug discovery pipeline end-to-end. Across 50 protein targets, MADDIE produces molecules satisfying key property requirements and their synthesis routes, outperforming existing agentic systems, generative models, and general-purpose LLMs by a wide margin. Our analysis reveals how these agents collaborate and that MADDIE can employ different strategies across targets to achieve the campaign's goal, while still being efficient in time and tokens. Together, they demonstrate MADDIE as a low-cost, effective, and efficient autonomous engine for accelerating drug discovery. Code and dataset is available via https://github.com/ninglab/MADDIE.

Chinese Translation

药物发现仍然是一个资源密集且耗时的过程，涉及数十亿美元的投资和十年的开发才能将药物推向市场。虽然许多计算机辅助工具加速了这一过程，但操作这些工具仍需要药物化学家投入大量时间、精力和专业知识。基于大型语言模型（Large Language Models, LLMs）构建的自主系统已成为一种有前景的解决方案，但当前的设计限制了它们的性能。现有的单智能体系统在长时间跨度、多步骤和现实任务上扩展性较差，而多智能体系统通常执行预定义的工作流程，无法捕捉真实药物发现活动的迭代和非线性特征。在此，我们提出了MADDIE，一个多智能体药物发现引擎。它由一个中央智能体——协调者（Orchestrator）和四个专业工作智能体组成：命中识别者（HitIdentifier）、先导优化者（LeadOptimizer）、合成者（Synthesizer）和评估者（Evaluator），每个智能体代表药物发现活动中的一个关键步骤。在协调者的协调和管理下，这些工作智能体被组合成灵活的工作流程，使MADDIE能够调整其策略并端到端地导航药物发现管道。在50个蛋白质靶点上，MADDIE生成满足关键属性要求的分子及其合成路线，显著优于现有的智能体系统、生成模型和通用大型语言模型。我们的分析揭示了这些智能体如何协作，并表明MADDIE能够在不同靶点上采用不同策略以实现活动目标，同时在时间和代币使用上保持高效。总体而言，这表明MADDIE是一个低成本、高效且有效的自主引擎，能够加速药物发现。代码和数据集可通过https://github.com/ninglab/MADDIE获取。

View on arXiv Download PDF AI Translation

cs.AI / 225 / 2610.16980

PRAISE: A Reproducible Workflow for Non-Programmer Scientists to Implement Published AI Models

PRAISE：一个可重复的工作流程，帮助非程序员科学家实施已发布的人工智能模型

Achterweust, Max, Nibbelink, Stefan Groot, de Beer, Vincent J., Schipper, Stefan, Yüksektepe, Zeynep, da Silva, Pedro Andrade, Peppelaar, Aurelia, van der Willigen, Martijn J., Minkenberg, Christophe B., van der Willigen, Robert F.

Abstract

Artificial Intelligence (AI) models are increasingly developed, published and used across scientific fields. Researchers, students and industry professionals who want to apply these models to their own research questions often lack the programming and data science expertise to do so. Existing guidelines cover single steps, but not the route from a published model to working predictions on one's own data. Here we present the hands-on workflow PRAISE, a Practical, Reproducible AI-model Implementation for Science and Education, that helps non-programmers implement published AI models step by step. Because implementations rarely work the first time, PRAISE anticipates common failures and provides a fallback route for each. As a proof of principle, three BSc chemistry students and their tutor applied PRAISE without prior programming or data science knowledge. They reproduced, validated and extended a state-of-the-art model that predicts enzyme-substrate interactions. This shows that with PRAISE, implementing an AI model from the literature becomes a learnable skill, one that turns non-programmer scientists from users into evaluators and improvers of AI models.

Chinese Translation

人工智能（AI）模型在各个科学领域的开发、发布和应用日益增多。希望将这些模型应用于自身研究问题的研究人员、学生和行业专业人士，往往缺乏必要的编程和数据科学专业知识。现有的指南通常只涵盖单个步骤，而未能提供从已发布模型到在自己数据上进行有效预测的完整路径。在此，我们提出了实践性、可重复的AI模型实施工作流程PRAISE（Practical, Reproducible AI-model Implementation for Science and Education），帮助非程序员逐步实施已发布的AI模型。由于实现过程通常不会在第一次就成功，PRAISE预见了常见的失败情况，并为每种情况提供了备用解决方案。作为原理证明，三名化学本科生及其导师在没有任何编程或数据科学知识的情况下应用了PRAISE。他们成功复现、验证并扩展了一个预测酶-底物相互作用的最先进模型。这表明，通过PRAISE，从文献中实施AI模型成为一种可学习的技能，使非程序员科学家从用户转变为AI模型的评估者和改进者。

View on arXiv Download PDF AI Translation

cs.AI / 226 / 2610.16998

Glasseek: A Transparent Recipe for Training and Trustworthy Evaluation of DeepSearch Agents

Glasseek：透明的深度搜索代理训练与可信评估的方案

Pang, Cong, Su, Jiaqi, Wu, Chengrui, Yan, Kuihao, Zhang, Tianrui, Qiu, Zixuan, Lin, Chao, Lin, Zixiu, Chen, Zixuan, Xiao, Zelin, Yao, Tiankuo, Hong, Jiawei, Chen, Zixuan, Xu, Dan, Lu, Lewei, Wang, Wenjie, Lou, Xin

Abstract

Deep search enables language agents to solve complex information-seeking tasks through iterative web exploration. However, developing robust open-source agents remains challenging for three reasons. First, opaque trajectory-construction pipelines make training data difficult to audit and reproduce. Second, context compression can create a substantial mismatch between training and inference. Third, agents may retrieve benchmark content from the open web during evaluation, contaminating the results and overstating their performance. To overcome these bottlenecks, we introduce Glasseek, a fully transparent deep search framework spanning the complete development lifecycle, releasing all data, model weights, training recipes, and code. Specifically, for data construction, we establish a verifiable trajectory curation pipeline that systematically generates high-quality exploration paths. For training, we propose Compaction-Managed GSPO (CM-GSPO), which uses the same policy for search and summarization while optimizing only execution sub-trajectories, enabling end-to-end search-policy training across context-compaction boundaries. For evaluation, we conduct the first systematic audit of test-set leakage in open-web agents and release a modular anti-leakage plugin that seamlessly integrates into existing evaluation harnesses. Under this rigorous evaluation protocol, our 9B agent outperforms frontier proprietary agents across almost all deep research benchmarks spanning diverse task types, and achieves the best overall performance among recent open-weight agents, setting a new state of the art at the 9B scale.

Chinese Translation

深度搜索使语言代理能够通过迭代的网络探索解决复杂的信息检索任务。然而，开发稳健的开源代理面临三大挑战。首先，不透明的轨迹构建流程使得训练数据难以审计和重现。其次，上下文压缩可能导致训练与推理之间存在显著的不匹配。第三，代理在评估过程中可能从开放网络中检索基准内容，从而污染结果并夸大其性能。为了解决这些瓶颈，我们提出了Glasseek，一个完全透明的深度搜索框架，涵盖整个开发生命周期，发布所有数据、模型权重、训练方案和代码。具体而言，在数据构建方面，我们建立了一个可验证的轨迹策划流程，系统地生成高质量的探索路径。在训练方面，我们提出了压缩管理的GSPO（Compaction-Managed GSPO，CM-GSPO），该方法在搜索和摘要中使用相同的策略，同时仅优化执行子轨迹，实现跨上下文压缩边界的端到端搜索策略训练。在评估方面，我们首次系统性审计了开放网络代理中的测试集泄漏，并发布了一个模块化的防泄漏插件，能够无缝集成到现有的评估工具中。在这一严格的评估协议下，我们的9B代理在几乎所有涵盖多种任务类型的深度研究基准中超越了前沿的专有代理，并在近期的开放权重代理中实现了最佳整体性能，在9B规模上设定了新的最优标准。

View on arXiv Download PDF AI Translation

cs.AI / 227 / 2610.17019

Understanding Personalization Gains in Whisper

理解 Whisper 中的个性化增益

Katkov, Sergei, Liotta, Antonio

Abstract

Whisper-style ASR models provide strong zero-shot transcription, but their residual errors are not evenly distributed across datasets, recording conditions, and metadata-based groups. We study when lightweight low-rank adaptation (LoRA) improves Whisper and whether observed gains require group-specific adapters. Across Common Voice, LibriSpeech, and VoxPopuli, we compare zero-shot Whisper, pooled LoRA, and group-specific LoRA under matched within-group splits. We find that adaptation is not uniformly beneficial: gains are concentrated in difficult Common Voice groups, are consistent on VoxPopuli, and largely disappear on low-error LibriSpeech. Crucially, pooled LoRA often matches, and sometimes clearly outperforms, group-specific LoRA. Gains over zero-shot should therefore be tested against matched pooled adaptation before being interpreted as evidence of personalization.

Chinese Translation

Whisper 风格的自动语音识别（ASR）模型提供强大的零样本转录能力，但其残余错误在数据集、录音条件和基于元数据的组之间并不均匀分布。我们研究轻量级低秩适配（LoRA）何时能改善 Whisper，以及观察到的增益是否需要特定于组的适配器。在 Common Voice、LibriSpeech 和 VoxPopuli 数据集上，我们在匹配的组内划分下比较了零样本 Whisper、汇总 LoRA 和特定于组的 LoRA。我们发现适配并非在所有情况下都有益：增益集中在困难的 Common Voice 组，在 VoxPopuli 上保持一致，而在低错误率的 LibriSpeech 上则大部分消失。关键是，汇总 LoRA 通常与特定于组的 LoRA 相匹配，有时甚至明显优于后者。因此，零样本的增益应在与汇总适配相匹配的情况下进行测试，而不是直接解释为个性化的证据。

View on arXiv Download PDF AI Translation

cs.AI / 228 / 2610.17027

Benchmarking Quantum Kernel SVMs for Credit Card Fraud Detection: Performance Realities Under NISQ Constraints

量子核支持向量机在信用卡欺诈检测中的基准测试：在NISQ约束下的性能现实

Arora, Hitanshi, Swarnkar, Mohit, Cherukuri, Aswani Kumar

Abstract

Detecting credit card fraud is notoriously difficult because fraudulent charges make up less than 0.2% of total transactions. Quantum Machine Learning (QML), particularly Quantum Kernel Support Vector Machines (QSVM), promises richer feature representations by embedding classical data into high-dimensional Hilbert spaces. However, whether this theoretical advantage holds up when managing real-world financial data bottlenecks remains unclear. We designed a controlled benchmark comparing three quantum encodings, Angle, ZZFeatureMap, and IQP against classical models while keeping sample sizes, feature counts, and scaling strictly matched. To handle class skew without corrupting quantum state overlaps, we embedded synthetic minority oversampling (SMOTE) directly into the quantum Gram matrix workflow, raising PR-AUC scores by roughly +0.04 across feature maps. When trained on the full European credit card dataset with 30 features, classical baselines reached a top PR-AUC of 0.983. However, restricting classical models to match QSVM hardware constraints (500 samples, 4 PCA features) dropped their PR-AUC to 0.750. This puts our best quantum result 0.730 PR-AUC using IQP encoding on 150 samples into realistic context. Testing on physical quantum hardware reduced performance further to 0.62 due to CNOT gate errors and thermal noise. Overall, our findings pinpoint two main operational bottlenecks for near-term quantum kernels: the scaling of Gram matrix evaluation and heavy information loss from squeezing features onto small qubit registers.

Chinese Translation

检测信用卡欺诈 notoriously 困难，因为欺诈性交易占总交易的比例不足 0.2%。量子机器学习（QML），特别是量子核支持向量机（QSVM），通过将经典数据嵌入高维希尔伯特空间，承诺提供更丰富的特征表示。然而，当处理真实世界的金融数据瓶颈时，这一理论优势是否依然成立尚不明确。我们设计了一个受控基准测试，比较了三种量子编码：Angle、ZZFeatureMap 和 IQP，针对经典模型进行了比较，同时严格匹配样本大小、特征数量和缩放。为了处理类别偏斜而不破坏量子状态重叠，我们将合成少数类过采样（SMOTE）直接嵌入量子Gram矩阵工作流程中，使得PR-AUC分数在特征映射上提高了大约 +0.04。当在包含30个特征的完整欧洲信用卡数据集上训练时，经典基线达到了最高PR-AUC 0.983。然而，将经典模型限制以匹配QSVM硬件约束（500个样本，4个PCA特征）时，其PR-AUC降至0.750。这使得我们使用IQP编码在150个样本上获得的最佳量子结果0.730 PR-AUC置于现实背景中。在物理量子硬件上测试进一步降低了性能至0.62，原因是CNOT门错误和热噪声。总体而言，我们的发现指出了近期量子核的两个主要操作瓶颈：Gram矩阵评估的扩展性和将特征压缩到小量子比特寄存器时的严重信息损失。

View on arXiv Download PDF AI Translation

cs.AI / 229 / 2610.17043

Neos: Learning to Interleave Search and Visual Reasoning

Neos：学习交错搜索与视觉推理

He, Ekko, Hong, Jack, Yan, Shilin, Zhang, Xintong, Zhang, Yujie, Li, Wenjie, Sun, Haoran, Jiang, Kaixun, Fu, Jiyuan, Zhang, Guannan, Xue, Hongwei

Abstract

Recent multimodal deep search agents have advanced tool use and visual content retrieval, yet they still struggle to convert intermediate visual evidence into actionable guidance for subsequent search, limiting their effectiveness in long-horizon settings. To address this gap, we present Neos, a multimodal deep search framework that learns both how to acquire evidence and how to use it to guide subsequent decisions. To make this coupling learnable, Neos constructs training tasks with intermediate visual dependencies while filtering out textual shortcuts, and introduces a lightweight, trainable Multimodal Evidence Interface (MEI) that transforms retrieved content into goal-conditioned observations. At its core, an agent-environment co-evolution procedure treats MEI as a co-adaptive component of this loop. Specifically, reference-guided credit assignment uses verified successful trajectories to localize critical policy errors while preserving credit for useful decisions in failed searches, whereas rollout feedback selects informative local evidence-processing interactions for MEI supervised fine-tuning (SFT). By alternating policy reinforcement learning (RL) with MEI SFT, Neos allows the policy's evolving evidence requests to shape MEI training and the adapted MEI to reshape subsequent observations, closing the loop between evidence acquisition and evidence-conditioned action. Across ten benchmarks, Neos-9B and Neos-35B demonstrate leading performance among specialized multimodal deep-search agents.

Chinese Translation

近期的多模态深度搜索代理在工具使用和视觉内容检索方面取得了进展，但它们仍然难以将中间视觉证据转化为后续搜索的可操作指导，这限制了它们在长时间跨度设置中的有效性。为了解决这一问题，我们提出了Neos，一个多模态深度搜索框架，它学习如何获取证据以及如何利用这些证据指导后续决策。为了使这一结合可学习，Neos构建了具有中间视觉依赖性的训练任务，同时过滤掉文本捷径，并引入了一个轻量级、可训练的多模态证据接口（Multimodal Evidence Interface, MEI），将检索到的内容转化为目标条件观察。其核心是一个代理-环境共同进化程序，将MEI视为这一循环的共同适应组件。具体而言，参考引导的信用分配利用经过验证的成功轨迹来定位关键策略错误，同时保留在失败搜索中有用决策的信用，而回滚反馈则选择信息丰富的局部证据处理交互用于MEI的监督微调（Supervised Fine-Tuning, SFT）。通过交替进行策略强化学习（Reinforcement Learning, RL）与MEI的SFT，Neos允许策略不断演变的证据请求来塑造MEI的训练，而适应后的MEI又重新塑造后续观察，实现了证据获取与证据条件行动之间的闭环。在十个基准测试中，Neos-9B和Neos-35B在专业多模态深度搜索代理中展示了领先的性能。

View on arXiv Download PDF AI Translation

cs.AI / 230 / 2610.17057

VL-DocIR: A Benchmark for Vision-Based Long Document Information Retrieval

VL-DocIR：基于视觉的长文档信息检索基准

Ott, Simon, Chen, Yufan, Liu, Ruiping, Zheng, Junwei, Wei, Jiale, Wen, Di, Peng, Kunyu, Zhang, Jiaming, Stiefelhagen, Rainer

Abstract

Vision-based document retrieval has improved rapidly, but current evaluation settings still emphasize short documents and single relevant pages. This leaves an important gap between leaderboard performance and realistic long-document retrieval, where the evidence needed to answer a query may span multiple pages or even multiple documents. We present VL-DocIR, a page-level benchmark for evidence-complete visual long-document retrieval. VL-DocIR is built from 29,641 HTML-born documents rendered into 388,548 page images from Wikipedia, arXiv, PubMed, and SEC proxy statements. It contains 271,760 queries across 23 domains and six evidence structures, covering single-page, within-document multi-page, and cross-document evidence configurations. Evidence annotations are grounded to rendered pages and HTML element identifiers, and queries are filtered with a cleaning pipeline targeting clarity, correctness, closedness, and the absence of explicit layout references. We evaluate leading single-vector and multi-vector visual retrievers together with diagnostic retrieval settings such as document-context scoring, query splitting, reranking, and round-robin ensembles. The strongest evaluated configuration reaches 69.3 nDCGAll@10 and 80.6 RecallAll@10 under structure-balanced Macro evaluation, but source- and evidence-structure breakdowns show that cross-document retrieval remains substantially harder than single-page retrieval, and that document-context scoring is especially helpful for within-document multi-page queries. VL-DocIR exposes these failure modes and provides a more demanding target for visual long-document retrieval. The new dataset and code will be made publicly available at https://yufanchen96.github.io/VL-DocIR.

Chinese Translation

基于视觉的文档检索快速发展，但当前的评估设置仍然强调短文档和单一相关页面。这在排行榜表现与现实的长文档检索之间留下了重要的差距，因为回答查询所需的证据可能跨越多个页面甚至多个文档。我们提出了VL-DocIR，这是一个针对证据完整的视觉长文档检索的页面级基准。VL-DocIR基于来自维基百科、arXiv、PubMed和SEC代理声明的29,641个HTML文档，渲染成388,548个页面图像。它包含来自23个领域和六种证据结构的271,760个查询，涵盖了单页面、文档内多页面和跨文档证据配置。证据注释基于渲染页面和HTML元素标识符，查询经过清理管道过滤，以确保清晰性、正确性、封闭性，并且没有显式的布局参考。我们评估了领先的单向量和多向量视觉检索器，并结合文档上下文评分、查询拆分、重排序和循环集成等诊断检索设置。评估的最强配置在结构平衡的宏观评估下达到了69.3 nDCGAll@10和80.6 RecallAll@10，但源和证据结构的细分显示，跨文档检索仍然比单页面检索困难得多，并且文档上下文评分对于文档内多页面查询特别有帮助。VL-DocIR揭示了这些失败模式，并为视觉长文档检索提供了一个更具挑战性的目标。新的数据集和代码将公开发布在https://yufanchen96.github.io/VL-DocIR。

View on arXiv Download PDF AI Translation

cs.AI / 231 / 2610.17059

PatientWM: Towards Generalizable Clinical World Models via Unified Patient States

PatientWM：通过统一患者状态实现可推广的临床世界模型

Kan, Ziwen, Peng, Jie, Yan, Bo, Hobley, Davis, Zhang, Zheng, Tian, Yu, Chen, Tianlong, Wang, Song

Abstract

Clinical decisions depend on anticipating how a patient's condition will evolve as treatment is adjusted and new evidence becomes available. Existing medical world models capture patient-state evolution, but often define the evolving state around a particular modality or task, limiting reuse as inputs or prediction targets change. These models also retain history through observation sequences or recurrent or latent summaries rather than exposing preceding patient states and clinical actions as an ordered transition context. This distinction matters because similar current measurements may follow different recent changes, treatment exposures, and observed responses. To tackle these limitations, we propose PatientWM, a clinical world-model framework built around a unified patient state and sequential latent history. Availability-aware encoders integrate observations available at each decision time into the patient state, while typed decoders produce task-specific clinical readouts without defining the state around any single target. PatientWM further conditions each next-state prediction on the current state and supplied action together with a bounded, ordered history of preceding latent patient states and clinical actions, preserving recent treatment and response context. Experiments across MIMIC-III, MIMIC-IV, and three external clinical datasets show superior performance in clinical prediction, multi-step forecasting, and action-planning evaluation. Transfer experiments further demonstrate the generalizability of PatientWM's patient state. Together, these results establish its ability to forecast clinical evolution and support planning.

Chinese Translation

临床决策依赖于预测患者状况在治疗调整和新证据出现时的演变。现有的医学世界模型捕捉患者状态的演变，但通常围绕特定的模态或任务定义演变状态，这限制了在输入或预测目标变化时的重用。这些模型还通过观察序列或递归或潜在摘要保留历史，而不是将先前的患者状态和临床行动作为有序的过渡上下文进行展示。这一区别很重要，因为相似的当前测量可能遵循不同的近期变化、治疗暴露和观察反应。为了解决这些局限性，我们提出了PatientWM，一个围绕统一患者状态和序列潜在历史构建的临床世界模型框架。可用性感知编码器将每个决策时间可用的观察整合到患者状态中，而类型解码器则生成特定任务的临床读数，而不围绕任何单一目标定义状态。PatientWM进一步将每个下一状态的预测条件化于当前状态和提供的行动，以及先前潜在患者状态和临床行动的有界、有序历史，从而保留近期治疗和反应的上下文。在MIMIC-III、MIMIC-IV和三个外部临床数据集上的实验显示了在临床预测、多步预测和行动规划评估中的优越性能。迁移实验进一步证明了PatientWM患者状态的可推广性。这些结果共同确立了其预测临床演变和支持规划的能力。

View on arXiv Download PDF AI Translation

cs.AI / 232 / 2610.17093

SFT Leaves a Reproducible Signature in Activation Geometry: Calibrating Post-Training Interpretability Against Run-to-Run Noise

SFT在激活几何中留下可重复的特征：校准后训练可解释性以应对运行间噪声

Chen, Xupeng, Le, Chenqian, Cheng, Jiayi, Li, Runhao, Yue, Michael

Abstract

A model diff credits post-training with whatever separates a base checkpoint from one post-trained descendant, presuming a rerun would land close by. On the sparse co-activation graph of Qwen3-1.7B it does not: two LoRA-SFT reruns sit over half as far apart as either sits from the base. We score each base-to-post contrast against the run-to-run band of its own recipe, and show that $k$ runs certify no better than $p=1/(k+1)$ under run exchangeability, however large the effect, so the band $z$-score is a descriptive scale. SFT leaves a large, reproducible signature: base $\to$ SFT displaces the graph by 17.8 pairwise-distance SDs beyond the mean distance between four independent LoRA-SFT runs, and the pre-registered estimator replicates on Llama-3.1-8B with Llama-Scope. At 29 steps each, four SFT and four DPO runs occupy disjoint displacement ranges, a gap $5.0\times$ and $5.6\times$ the spreads of the two arms, without any band. All bands use graphs built on one GPU architecture. Three other readings stop short of a verdict: DPO band counts (3/4, 1/4, 3/4 contrasts outside at 12, 29, 60 steps) show no consistent dose ordering while its displacement stays flat; light GRPO is a secondary non-detection, not an absence, and we run no equivalence test; and the graph concentration read as an RL signature is already in the zero-RL base. A pre-registered causal sweep at three edge budgets on 4,816 held-out prompts found the displaced sources load-bearing ($12\times$ a matched random control in mean, $3.5\times$ in median at 10,000 edges), but their stage-specificity interaction reversed at two budgets and the third failed its faithfulness gate, so we claim calibrated structural displacement, not a changed mechanism. Scope: one backbone, one SFT recipe and general-text probe states for the full protocol.

Chinese Translation

模型差异将后训练归因于任何将基础检查点与其后训练后代分开的因素，假设重新运行会接近该位置。然而，在Qwen3-1.7B的稀疏共同激活图中，情况并非如此：两次LoRA-SFT的重新运行之间的距离超过了任一运行与基础检查点之间的距离的一半。我们将每个基础到后训练的对比与其自身配方的运行间带进行评分，并表明在运行可交换性下，$k$次运行的认证效果不优于$p=1/(k+1)$，无论效果多大，因此带的$z$分数是一个描述性尺度。SFT留下了一个大的、可重复的特征：基础$ o$ SFT使图形偏移了17.8个成对距离的标准差，超出了四次独立LoRA-SFT运行之间的平均距离，且预注册的估计在Llama-3.1-8B上使用Llama-Scope得到了重复。在每次29步的情况下，四次SFT和四次DPO运行占据了不重叠的位移范围，间隙为两条臂的扩展的$5.0 imes$和$5.6 imes$，且没有任何带。所有带均使用基于单个GPU架构构建的图形。其他三次读数未能得出结论：DPO带计数（3/4、1/4、12、29、60步外的3/4对比）未显示出一致的剂量排序，而其位移保持平坦；轻量级GRPO是次要的非检测，而非缺失，我们未进行等效性测试；作为强化学习特征的图形浓度已在零强化学习基础中存在。对4,816个保留提示在三个边际预算下进行的预注册因果扫描发现，位移源是承重的（在均值上为匹配随机对照的$12 imes$，在中位数上为$3.5 imes$，在10,000个边缘时），但它们的阶段特异性交互在两个预算下逆转，第三个未通过其忠实性门，因此我们声称校准的结构位移，而非机制的改变。范围：一个主干，一个SFT配方和全协议的通用文本探测状态。

View on arXiv Download PDF AI Translation

cs.AI / 233 / 2610.17118

When LLM Agents See a Conflict but Still Proceed: Cross-Domain Dependencies in Network Slicing

当大型语言模型代理看到冲突但仍然继续：网络切片中的跨域依赖性

Romero, Marcos Lima, Suyama, Ricardo

Abstract

Future AI-native networks are being designed to apply autonomous agents built on large language models (LLMs) in network management through intents. When intents from different domains meet on the same resource, an agent must recognize the cross-domain dependency before taking an irreversible action. We investigate this problem through a smart-grid use case, in which a power utility holds an active intent on a backup slice for protection traffic. This slice appears idle by design, which is precisely the profile targeted by a cost-reduction intent. We built a tool-based environment in which the dependency must be either queried by the agent or read from an inventory annotation, together with a negative control, and evaluated ten models from four different providers in 600 runs. Most models acted against the dependency they had observed: seven of the ten models removed the protection slice at least once after receiving the cross-domain dependency, and in the high-harm tier the slice was removed in 76% of the exposed runs, while four models practically never did. Only one of ten models never removed the slice. We identify two failure modes, not querying the dependency and querying it but proceeding anyway, and show that agents often reported no conflict after observing it. These results indicate that, in AI-native networks, visibility of cross-domain dependencies is not a sufficient safeguard, and that dependency verification should be enforced by the management architecture before irreversible actions rather than delegated to the LLM's judgment.

Chinese Translation

未来的人工智能原生网络正在设计中，旨在通过意图在网络管理中应用基于大型语言模型（LLMs）的自主代理。当来自不同领域的意图在同一资源上相遇时，代理必须在采取不可逆行动之前识别跨域依赖性。我们通过一个智能电网用例来研究这个问题，其中一个电力公司在备份切片上持有一个主动意图，以保护流量。该切片在设计上看似闲置，这正是成本降低意图所针对的特征。我们构建了一个工具基础环境，其中依赖性必须由代理查询或从库存注释中读取，并设置了一个负对照，评估了来自四个不同提供商的十个模型在600次运行中的表现。大多数模型的行为与它们观察到的依赖性相悖：十个模型中的七个在接收到跨域依赖性后至少一次移除了保护切片，而在高危害层级中，该切片在76%的暴露运行中被移除，而四个模型几乎从未移除。只有十个模型中的一个从未移除该切片。我们识别出两种失败模式：未查询依赖性和查询后仍然继续，并显示代理在观察到冲突后常常报告没有冲突。这些结果表明，在人工智能原生网络中，跨域依赖性的可见性并不足以作为保障，依赖性验证应由管理架构在不可逆行动之前强制执行，而不是委托给LLM的判断。

View on arXiv Download PDF AI Translation

cs.AI / 234 / 2610.17130

EvoQuantLab: Training-Free Dual-Loop Evolution for Auditable Quantitative Research

EvoQuantLab：无训练的双循环进化框架用于可审计的量化研究

Yang, Haolin, Qian, Zhiqiang, Chen, Chutian, Zhang, Jipeng, Deng, Tangjia, Xie, Jian, Gong, Shuaishuai, Han, Sirui, Guo, Yike

Abstract

Automated alpha factor discovery is critical for modern quantitative research. While large language models (LLMs) have increasingly been integrated into quantitative workflows, existing agent-based approaches remain highly susceptible to severe backtest overfitting. To address this challenge, we propose EvoQuantLab, a training-free, dual-loop self-evolving framework for robust factor discovery. Instead of relying entirely on LLM generation, EvoQuantLab deploys an inner genetic-programming (GP) loop that dramatically improves token efficiency while evolving typed factor expressions under structural and diversity constraints. An outer diagnosis loop then systematically audits evidence across cycles to drive protocol-level improvements. Furthermore, our evaluation pipeline applies multi-metric validation to rigorously filter out spurious factors before deployment. Extensive out-of-sample experiments across both daily and hourly frequencies -- covering equities and cryptocurrency -- demonstrate the robustness of EvoQuantLab. Specifically, on daily CSI 500 and S&P 500 tasks, EvoQuantLab achieves leading annualized returns of 18.34% and 21.30%, consistently outperforming existing agent baselines. On high-turnover hourly cryptocurrency data, EvoQuantLab delivers a 37.91% annualized return while effectively avoiding execution collapse.

Chinese Translation

自动化阿尔法因子发现对现代量化研究至关重要。尽管大型语言模型（LLMs）越来越多地被集成到量化工作流程中，但现有的基于代理的方法仍然对严重的回测过拟合高度敏感。为了解决这一挑战，我们提出了EvoQuantLab，一个无训练的双循环自我进化框架，用于稳健的因子发现。EvoQuantLab并不完全依赖LLM生成，而是部署了一个内部遗传编程（GP）循环，在结构和多样性约束下显著提高了令牌效率，同时进化类型化因子表达式。然后，外部诊断循环系统地审计各个周期的证据，以推动协议级别的改进。此外，我们的评估管道应用多指标验证，以严格过滤掉虚假因子，确保在部署前的有效性。在涵盖股票和加密货币的日频和小时频的广泛样本外实验中，EvoQuantLab展现了其稳健性。具体而言，在日频的CSI 500和S&P 500任务上，EvoQuantLab实现了18.34%和21.30%的领先年化收益，持续超越现有的代理基准。在高换手率的小时加密货币数据上，EvoQuantLab实现了37.91%的年化收益，同时有效避免了执行崩溃。

View on arXiv Download PDF AI Translation

cs.AI / 235 / 2610.17158

The Format is the Filter: Scoring Format Fixes What an LLM Judge Can Express

格式即过滤器：评分格式修正了 LLM 判断者所能表达的内容

Gondara, Lovedeep, Novikova, Jekaterina, Suresh, Sagarika, Ji, Jack, Pradeep, Jithin

Abstract

When an LLM judge filters instruction-tuning data, its scoring format (binary verdict, 1--5 rating, 0--100 score, or comparison) is rarely justified or ablated. Holding the candidate pool, judge prompt, decoding and selection budget fixed, we score 15,406 UltraFeedback candidates under five formats with three judges, curate, fine-tune 66 models, and evaluate them with four judges. A 0--100 scale is worth only 7--12 equally used levels, two to three times a 1--5 scale rather than twenty, and a binary verdict leaves the top score tied in 77--78\% of prompts, so most selections are made at random. These properties are set by the format almost independently of the judge. Kendall's $\tau_b$ hides them: it ranks binary best while $\tau_a$ and top-1 agreement rank it last, because it discounts the ties a selector must break at random. Both comparative formats inherit a slot bias whose sign differs across judges. Downstream, curated data beat random selection under every evaluation judge, robust to length and markdown adjustment, but no ordering among formats is detectable. We propose four quantities that judge-based curation pipelines should report.

Chinese Translation

当 LLM 判断者过滤指令调优数据时，其评分格式（如二元判决、1--5 评分、0--100 分数或比较）很少得到合理解释或剔除。在固定候选池、判断提示、解码和选择预算的情况下，我们在五种格式下对 15,406 个 UltraFeedback 候选者进行评分，使用三位评审进行评估，策划并微调 66 个模型，并用四位评审进行评估。0--100 的评分尺度仅相当于 7--12 个同等使用的级别，是 1--5 评分尺度的两到三倍，而二元判决在 77--78\% 的提示中使得最高分平局，因此大多数选择是随机进行的。这些特性几乎独立于评审者由格式设定。Kendall 的 $ au_b$ 隐藏了这些特性：它在排名时将二元判决评为最佳，而 $ au_a$ 和 top-1 一致性则将其排在最后，因为它忽略了选择者必须随机打破的平局。两种比较格式都继承了一个插槽偏差，其符号在不同评审者之间有所不同。在下游，策划的数据在每个评估评审下均优于随机选择，对长度和降价调整具有鲁棒性，但在格式之间没有可检测的排序。我们提出了四个量度，评审基础的策划管道应报告这些量度。

View on arXiv Download PDF AI Translation

cs.AI / 236 / 2610.17164

Learning Perceptual Geometry of Natural Odor Mixtures from LLM-Derived Descriptors

从大型语言模型派生描述符学习自然气味混合物的感知几何

Pashakhanloo, Farhad, Murthy, Venkatesh N.

Abstract

The relationship between odor chemistry and perception remains incompletely understood. Many existing studies focus on the perception of single molecules, whereas natural odors contain tens to hundreds of components, adding extra layer of complexity to percept formation. Here, we explore the relationship between molecular composition and perceived similarity in natural odor objects. We study this using a database of more than 1,200 volatile-compound-in-food mixtures containing more than 8,000 molecular components, and use large language models (LLMs) as a scalable surrogate for deriving mixture-level perceptual descriptions. We find that language-derived representations provide useful supervision for predicting held-out mixture similarities from molecular composition. The prediction was consistently strongest when perceptual targets were derived from chemosensory prompts than from other sensory prompts. Further, during the mixture-level training, we learned low-rank molecular embeddings that enabled better prediction of held-out mixture similarities than component-overlap metrics, even for mixture pairs with no shared molecules. Although no single-molecule labels were used during training, the learned molecular geometry showed significant alignment with the independent human odor-descriptor space. Finally, we explored the use of external molecular features in our mixture-similarity prediction task. These features showed modest zero-shot alignment with mixture-level perceptual similarity. They improved prediction under limited supervision but offered little additional predictive benefit with the full training set. Together, these results show that language-derived descriptions can provide scalable supervision for predicting the perceptual organization of complex natural mixtures, contributing to a quantitative understanding of how molecular composition gives rise to odor perception.

Chinese Translation

气味化学与感知之间的关系仍未完全理解。许多现有研究集中于单一分子的感知，而自然气味则包含数十到数百种成分，给感知形成增加了额外的复杂性。在此，我们探讨了分子组成与自然气味对象感知相似性之间的关系。我们使用一个包含超过1200种食品中挥发性化合物混合物的数据库，这些混合物包含超过8000种分子成分，并利用大型语言模型（LLMs）作为可扩展的替代工具来推导混合物级别的感知描述。我们发现，基于语言的表示为从分子组成预测保留混合物相似性提供了有用的监督。当感知目标源自化学感官提示时，预测的准确性始终较强，而非其他感官提示。此外，在混合物级别的训练过程中，我们学习到了低秩分子嵌入，这使得对保留混合物相似性的预测优于成分重叠指标，即使对于没有共享分子的混合物对也是如此。尽管在训练过程中未使用单一分子标签，但学习到的分子几何与独立的人类气味描述空间显示出显著的一致性。最后，我们探讨了在混合物相似性预测任务中使用外部分子特征。这些特征在混合物级别的感知相似性上表现出适度的零样本一致性。在有限监督下，它们提高了预测能力，但在完整训练集下提供的额外预测益处有限。综合来看，这些结果表明，基于语言的描述可以为预测复杂自然混合物的感知组织提供可扩展的监督，从而有助于定量理解分子组成如何导致气味感知。

View on arXiv Download PDF AI Translation

cs.AI / 237 / 2610.17183

ChemExplain: A Multimodal Benchmark for Explanatory Chemical Reasoning

ChemExplain：一个用于解释性化学推理的多模态基准

Chen, Aotian, Baker, Frazier N., Averly, Reza, Zhang, Shan, Chen, Tianao, Adu-Ampratwum, Daniel, Ning, Xia, Xue, Yuan

Abstract

Current chemistry benchmarks primarily evaluate outcome prediction or reaction pathway generation, but rarely test whether a model can explain why an observed chemical effect occurs. We introduce ChemExplain, a benchmark of 1,180 multiple-choice chemistry questions drawn from 383 peer-reviewed papers. Each question pairs a molecular or reaction image with an observed chemical effect and four candidate explanations, each associated with a different structural feature visible in the image. Answering requires the model to perceive the depicted structure, apply chemical knowledge, and identify which feature most directly explains the effect, a capability we term explanatory attribution. The correct feature-mechanism relation is derived from supporting evidence in the source paper rather than invented by the question-generation LLM. We construct the benchmark automatically with a multi-agent framework that applies progressive quality control at three stages and validate it with eight chemistry domain experts. The strongest evaluated model reaches 78.6% accuracy, leaving considerable room for improvement. On a 600-question expert-reviewed subset, reasoning-trace analysis classifies 71.4% of the strongest model's errors as attribution failures. Attribution is also the largest error category for the open-source and chemistry-specific models included in our analysis.

Chinese Translation

目前的化学基准主要评估结果预测或反应路径生成，但很少测试模型是否能够解释观察到的化学效应发生的原因。我们介绍了ChemExplain，这是一个由383篇经过同行评审的论文中提取的1180个多项选择化学问题组成的基准。每个问题将一个分子或反应图像与一个观察到的化学效应及四个候选解释配对，每个解释与图像中可见的不同结构特征相关联。回答这些问题要求模型感知所描绘的结构，应用化学知识，并识别哪个特征最直接地解释了该效应，这一能力我们称之为解释性归因。正确的特征-机制关系是基于源论文中的支持证据得出的，而不是由问题生成的语言模型（LLM）所虚构。我们通过一个多代理框架自动构建该基准，并在三个阶段应用渐进式质量控制，最终由八位化学领域专家进行验证。经过评估的最强模型达到了78.6%的准确率，仍有相当大的改进空间。在一个600个问题的专家审查子集中，推理轨迹分析将最强模型的71.4%的错误分类为归因失败。归因也是我们分析中包含的开源和化学特定模型的最大错误类别。

View on arXiv Download PDF AI Translation

cs.AI / 238 / 2610.17192

Size Generalization for Diffusion Routing Solvers

扩散路由求解器的规模泛化

Davidovitch, Barak Itzchak, Meirom, Eli, Fetaya, Ethan, Chechik, Gal

Abstract

Routing problems, like the Traveling Salesman Problem (TSP) and the Capacitated Vehicle Routing Problem (CVRP), are NP-hard and slow to solve for large instances. Diffusion-based solvers can be trained in a supervised way using sets of optimal solutions, and then quickly solve new test instances. However, for large instances, optimal solutions are hard to find, and diffusion models fail to generalize to instance sizes outside their training distribution. Here we describe ResizeCO, a label free size-generalization approach that addresses both the change in statistical properties of larger instances, and the computational challenge of solving them quickly. It uses reinforcement learning on large-scale instances to adapt a diffusion model trained on small instances. Specifically, we adapt GRPO to this problem, using route length as the training signal. ResizeCO also formulates k-opt as a tree search, solved by a beam search on the GPU. In size-generalization experiments on TSP and CVRP instances with up to 10000 nodes, ResizeCO achieves a Pareto front over route quality and time that dominates learned solvers, and also outperforms other solvers at short run times, setting a new state of the art.

Chinese Translation

路由问题，如旅行商问题（Traveling Salesman Problem, TSP）和有容量限制的车辆路由问题（Capacitated Vehicle Routing Problem, CVRP），是NP难题，对于大规模实例的求解速度较慢。基于扩散的求解器可以通过使用最优解集进行监督训练，从而快速解决新的测试实例。然而，对于大规模实例，最优解难以找到，扩散模型在训练分布之外的实例规模上无法泛化。在此，我们描述了ResizeCO，一种无标签的规模泛化方法，旨在解决大规模实例的统计特性变化以及快速求解的计算挑战。该方法利用强化学习在大规模实例上调整在小规模实例上训练的扩散模型。具体而言，我们将GRPO适应于此问题，使用路径长度作为训练信号。ResizeCO还将k-opt形式化为树搜索，通过GPU上的束搜索进行求解。在对最多10000个节点的TSP和CVRP实例进行的规模泛化实验中，ResizeCO在路径质量和时间的Pareto前沿上超越了学习型求解器，并且在短时间运行时也优于其他求解器，设定了新的最先进水平。

View on arXiv Download PDF AI Translation

cs.AI / 239 / 2610.17205

Releasing the Inner Jev: A Simple Recipe for Open Decision Models

释放内在的Jev：开放决策模型的简单方案

Geuter, Jonathan, Cheng, Andrew, Alvarez-Melis, David

Abstract

LLM-based decision-making systems, also called System One models, have recently gained popularity, most notably through Jev (Almeida, 2026). Given a typed input state, a question, and a set of options, they return a probability for each option in a single forward pass, without text generation. We show that existing open instruction-tuned language models can achieve comparable performance \textit{without any additional training}, by reading out answers from letters (A, B, C, etc.) in the model's output logits at the last token position. Furthermore, we show how a simple supervised fine-tuning recipe can improve their performance further. We evaluate four open instruction-tuned language models: Gemma 4 31B, Gemma 4 12B, Qwen3.6-27B, and Qwen3.5-4B. We fine-tune each using LoRA and a decision head initialized from a slice of the language model head. Furthermore, we introduce OpenDecisions, a novel 2,352-question validation dataset designed to reflect the task composition of JevBench (JevBench main- tainers, 2026) and which computes a Capability score based on accuracy and calibration metrics. Without fine-tuning, Gemma 4 31B scores 79.6 on OpenDecisions, compared to Jev's 80.1. After undergoing our fine-tuning recipe, it scores 84.4 on OpenDecisions and 81.5 on JevBench v1.7, outperforming Jev by 4.3 and 4.4 Capability points respectively and setting a new state-of-the-art on OpenDecisions. On a suite of public decision datasets, our models achieve accuracy comparable to Jev, with lower calibration error. Our results show that open language models already possess System One capabilities, and with simple fine-tuning they outperform Jev on a wide range of decision tasks. We release our models, serving code, and OpenDecisions.

Chinese Translation

基于大型语言模型（LLM）的决策系统，也称为系统一模型，最近受到广泛关注，尤其是通过Jev（Almeida, 2026）而获得了显著的关注。在给定一个类型化的输入状态、一个问题和一组选项的情况下，它们在一次前向传播中返回每个选项的概率，而无需生成文本。我们展示了现有的开放指令调优语言模型可以在 extit{无需任何额外训练}的情况下，通过从模型输出logits的最后一个token位置读取字母（A、B、C等）中的答案，达到可比的性能。此外，我们展示了一种简单的监督微调方案可以进一步提升它们的性能。我们评估了四个开放指令调优语言模型：Gemma 4 31B、Gemma 4 12B、Qwen3.6-27B和Qwen3.5-4B。我们使用LoRA对每个模型进行微调，并从语言模型头的一部分初始化决策头。此外，我们引入了OpenDecisions，这是一个新颖的2,352个问题的验证数据集，旨在反映JevBench（JevBench维护者，2026）的任务组成，并基于准确性和校准指标计算能力评分。在未进行微调的情况下，Gemma 4 31B在OpenDecisions上的得分为79.6，而Jev的得分为80.1。在经过我们的微调方案后，它在OpenDecisions上的得分为84.4，在JevBench v1.7上的得分为81.5，分别超越Jev 4.3和4.4的能力评分，创造了OpenDecisions的新最优表现。在一系列公共决策数据集上，我们的模型实现了与Jev相当的准确性，同时校准误差更低。我们的结果表明，开放语言模型已经具备系统一的能力，并且通过简单的微调，它们在广泛的决策任务中超越了Jev。我们发布了我们的模型、服务代码和OpenDecisions。

View on arXiv Download PDF AI Translation

cs.AI / 240 / 2610.17206

AutoMARS: Automatic and Scalable Multi-Agent Large Language Model Framework for Supply Chain Risk Assessment

AutoMARS：用于供应链风险评估的自动化和可扩展的多智能体大型语言模型框架

Quan, Yinzhu, Liu, Zefang, Benaben, Frederick, Montreuil, Benoit

Abstract

We propose AutoMARS (Automatic Multi-Agent Risk Assessment), an agentic framework for automatic and scalable comparative risk assessment in supply chain networks. The framework enables end-to-end execution of analytical workflows by automatically planning tasks, coordinating specialized agents, invoking external analytical tools, and synthesizing domain-specific results into integrated risk evaluations. AutoMARS further supports scalable risk comparison across multiple locations by parallelizing independent analytical tasks. The framework adopts a multi-domain, multi-agent, and multi-tool architecture that integrates heterogeneous risk factors from socioeconomic, disaster, environmental, and transportation domains. Through coordinated agent interactions and tool integration, AutoMARS enables structured, systematic, and interpretable comparative analysis to support supply chain risk-aware location evaluation.

Chinese Translation

我们提出了AutoMARS（自动化多智能体风险评估），这是一个用于供应链网络中自动化和可扩展比较风险评估的智能框架。该框架通过自动规划任务、协调专门的智能体、调用外部分析工具，并将领域特定的结果综合成集成风险评估，实现了分析工作流的端到端执行。AutoMARS进一步支持在多个地点之间的可扩展风险比较，通过并行化独立的分析任务来实现。该框架采用多领域、多智能体和多工具的架构，整合了来自社会经济、灾害、环境和交通等领域的异构风险因素。通过协调的智能体交互和工具集成，AutoMARS实现了结构化、系统化和可解释的比较分析，以支持对供应链风险的意识驱动的地点评估。

View on arXiv Download PDF AI Translation

cs.AI / 241 / 2610.17234

IsoVRBench: Video Reward Modeling Without Human Feedback

IsoVRBench：无须人类反馈的视频奖励建模

Li, Shuyue Stella, He, Jacqueline, Chen, Jiahui, Roy, Sreya Dutta, Tsvetkov, Yulia, Zettlemoyer, Luke, Ghazvininejad, Marjan, Han, Xiaochuang

Abstract

Video reward models are trained and evaluated on human preferences between videos that often differ simultaneously on properties such as prompt alignment, visual fidelity, motion, and generator identity. As a result, aggregate accuracy does not reveal which video properties a reward model can reliably distinguish. We construct preference pairs by controlled interventions that change one property of a generated video while holding the rest fixed. The resulting pairs, by construction, provide preferences for how the videos differ along a specific dimension, without requiring any human annotation. We propose IsoVRBench, a video generation reward modeling benchmark of 8,000 such pairs across eight interventions, and show that state-of-the-art models separate broad quality differences between generators but miss important features such as temporal order. The best reward model reaches only 60.8% on reordered frames against 100% for humans, and rankings on existing benchmarks do not predict this failure. Using the synthetic isolated intervation pairs as training data leads to 1) IsoVRM, a reward model outperforming its counterpart trained with a larger human labeled preference dataset by 3.8 points on VBench-2.0, 2) IsoWan, the video generator post-trained with IsoVRM, outperforming its human-data counterpart in human preferences, and 3) gains that stack when combined with human data. Because intervention pairs are automatically generated rather than collected, the benchmark and training data can be easily extended as generators evolve.

Chinese Translation

视频奖励模型是在视频之间的人类偏好上进行训练和评估的，这些视频通常在提示对齐、视觉保真度、运动和生成器身份等属性上同时存在差异。因此，聚合准确性无法揭示奖励模型能够可靠区分哪些视频属性。我们通过控制干预构建偏好对，这些干预在保持其他属性不变的情况下改变生成视频的一个属性。由此构建的偏好对提供了视频在特定维度上差异的偏好，而无需任何人类注释。我们提出了IsoVRBench，这是一个包含8,000个此类偏好对的视频生成奖励建模基准，涵盖八种干预方式，并展示了最先进的模型能够区分生成器之间的广泛质量差异，但却忽视了诸如时间顺序等重要特征。最佳奖励模型在重新排序的帧上仅达到60.8%的准确率，而人类的准确率为100%，且现有基准上的排名无法预测这一失败。使用合成的孤立干预对作为训练数据导致了1) IsoVRM，一个在VBench-2.0上比其对应的使用更大人类标注偏好数据集训练的模型高出3.8分的奖励模型，2) IsoWan，使用IsoVRM进行后训练的视频生成器，在人类偏好上优于其人类数据对应物，以及3) 与人类数据结合时的增益。由于干预对是自动生成的而非收集的，因此基准和训练数据可以随着生成器的发展而轻松扩展。

View on arXiv Download PDF AI Translation

cs.AI / 242 / 2610.17256

Does a Governed Confidence Score Know When a Text-to-SQL Answer Is Right? An Exploratory Study on BIRD and a Pre-registered Confirmation on Spider

受管控的置信评分是否知道文本到SQL的答案是否正确？关于BIRD的探索性研究及对Spider的预注册确认

Venkat, Raghu, Anjali, Tricha

Abstract

Conversational analytics lets people ask questions of records spread across many systems and get answers in plain language; it is useful only if each answer can be trusted and explained. A production system of this kind attaches a confidence score to every answer it writes over a database and uses it to show the answer, show it with a warning, or withhold it. The score's weights, caps and thresholds were set by design. This paper audits that design. On the 1,534 questions of BIRD dev, answered by GPT-5.1 with a benchmark-adapted production prompt and five samples per question, the designed score separates right from wrong answers weakly (AUROC 0.624, 95% CI 0.595-0.652) and withholds 35.7% of correct answers. Nearly all of its ranking signal comes from agreement among the samples, and most of it is lost where agreement is combined by a minimum with a checklist that drops to zero on any unsupported claim, including claims in commentary. We froze a rule in which agreement with the shown answer is the score and deterministic checks act as vetoes and caps, and pre-registered a test on Spider dev before any Spider answer existed. The primary hypothesis held: AUROC rose from 0.566 to 0.725 (paired difference +0.159, 95% CI +0.097 to +0.221), and correct answers withheld fell from 49.0% to 8.8% with PASS accuracy 95.7%. Under one matching policy the choice of agreement statistic matters little; the gain comes from how agreement enters the score. Whether added checks help depends on how correctness is defined, while they always cost unqualified coverage. A pre-registered empty-result check did not meet its effect-size criterion on Spider. Correctness here means the core answer; labels are model-judged, and a blinded two-person audit, not yet complete, will be reported in a revised version.

Chinese Translation

对话分析使人们能够对分布在多个系统中的记录提出问题，并以简单语言获取答案；只有在每个答案都可以信任和解释的情况下，这种方法才有用。这类生产系统为其在数据库中生成的每个答案附加一个置信评分，并利用该评分来展示答案、以警告形式展示答案或不展示答案。该评分的权重、上限和阈值是经过设计设定的。本文对该设计进行了审计。在BIRD开发的1,534个问题中，使用适应基准的生产提示和每个问题五个样本，由GPT-5.1回答，设计的评分在区分正确与错误答案方面表现较弱（AUROC 0.624，95% CI 0.595-0.652），且拒绝了35.7%的正确答案。其几乎所有的排名信号来自样本之间的一致性，而在结合了最低标准与任何不支持的主张（包括评论中的主张）的检查表时，大部分信号丧失。我们冻结了一条规则，即与所示答案的一致性即为评分，而确定性检查作为否决和上限，并在任何Spider答案存在之前对Spider开发进行了预注册测试。主要假设成立：AUROC从0.566上升至0.725（配对差异+0.159，95% CI +0.097至+0.221），拒绝的正确答案从49.0%降至8.8%，PASS准确率为95.7%。在一种匹配策略下，一致性统计的选择影响不大；增益来自一致性如何进入评分。附加检查是否有帮助取决于正确性的定义，而它们总是会导致不合格的覆盖。预注册的空结果检查在Spider上未达到其效应大小标准。此处的正确性意味着核心答案；标签由模型判断，尚未完成的盲审将由两人进行，并将在修订版本中报告。

View on arXiv Download PDF AI Translation

cs.AI / 243 / 2610.17266

RL Post-Training Diffusion Language Models Along Their Own Generation Paths

沿自身生成路径进行后训练的强化学习扩散语言模型

Singh, Vaibhav, Belilovsky, Eugene

Abstract

Diffusion language models are a promising and more efficient alternative to autoregressive models. However, most RL post-training techniques are not easily adapted to these models. One issue is that diffusion language models generate answers through many intermediate states, deciding at each step which masked positions to reveal and which tokens to place in them. Many existing RL methods used in this setting approximate intractable sequence likelihoods or weight every denoising step by the same advantage, without explicitly distinguishing how these intermediate states of the denoising trajectory affect the expected reward. To address this issue of trajectory blindness, we introduce PACT (Potential-Anchored Consistency of Transitions). Motivated by a reward-tilted reference distribution, PACT jointly learns the policy and a scalar potential that scores each intermediate state, anchored by terminal reward. For each sampled denoising transition, a consistency loss matches the policy-to-reference log-probability ratio of revealed tokens to the change in this potential. This formulation requires neither sequence-likelihood estimation nor importance-ratio clipping. On four tasks, PACT improves LLaDA-8B-Instruct by +9.9 points on GSM8K, +12.0 on MATH500, +55.7 on Countdown, and +72.6 on Sudoku, and reaches the best average accuracy across the four tasks reported for this model, +3.7 points over the next best method. Further compared to AGRPO, it reaches that baseline's best accuracy in about half the training compute on GSM8K.

Chinese Translation

扩散语言模型是自回归模型的一个有前景且更高效的替代方案。然而，大多数强化学习（RL）后训练技术并不容易适应这些模型。一个问题是，扩散语言模型通过许多中间状态生成答案，在每一步决定揭示哪些被遮蔽的位置以及在这些位置放置哪些标记。许多现有的RL方法在这种情况下近似不可处理的序列似然，或对每个去噪步骤赋予相同的优势，而没有明确区分去噪轨迹的这些中间状态如何影响期望奖励。为了解决这一轨迹盲目性问题，我们引入了PACT（潜在锚定过渡一致性）。PACT的动机源于奖励倾斜的参考分布，它联合学习策略和一个标量潜在值，该潜在值为每个中间状态打分，并由终端奖励锚定。对于每个采样的去噪过渡，一致性损失将揭示标记的策略与参考对数概率比与该潜在值的变化进行匹配。这种表述既不需要序列似然估计，也不需要重要性比率剪切。在四个任务上，PACT使LLaDA-8B-Instruct在GSM8K上提高了9.9分，在MATH500上提高了12.0分，在Countdown上提高了55.7分，在Sudoku上提高了72.6分，并在这四个任务中达到了该模型报告的最佳平均准确率，比下一个最佳方法提高了3.7分。此外，与AGRPO相比，它在GSM8K上以大约一半的训练计算量达到了该基线的最佳准确率。

View on arXiv Download PDF AI Translation

cs.AI / 244 / 2610.17271

Reasoning for Propaganda Detection via Label-Guided CoT Optimisation

通过标签引导的链式推理优化进行宣传检测

Afroz, Zeba, Vardhan, Harsh, Akhtar, Md. Shad

Abstract

Propaganda detection systems usually operate as black boxes, lacking explanations. To address this, we hypothesise that persuasive techniques contain definitional characteristics with respect to their social-political setting. We introduce ReCoT, a reasoning-conditioned detection framework that captures definitional cues of persuasive techniques via population-level refinement and performance-guided resampling, without requiring additional explanation annotations. On INDI-PROPspan, ReCoT achieves a macro F1_LCS of 0.364, surpassing baselines and reaching 88% of the SFT upper bound on Avg F1_LCS. Similarly, on PTC, ReCoT reaches 85% of the SFT upper bound on Avg F1_LCS, closing much of this gap while remaining interpretable. Importantly, the analysis of the reasoning pool's convergence and update pattern for ReCoT shows effective refinement of reasoning CoTs.

Chinese Translation

宣传检测系统通常作为黑箱运作，缺乏解释。为了解决这个问题，我们假设说服技术在其社会政治背景下包含定义特征。我们提出了ReCoT（Reasoning-conditioned Detection Framework），这是一个通过人口层面的细化和基于性能的重采样捕捉说服技术定义线索的推理条件检测框架，无需额外的解释注释。在INDI-PROPspan数据集上，ReCoT达到了0.364的宏观F1_LCS，超越了基线，并在平均F1_LCS上达到了SFT上限的88%。同样，在PTC数据集上，ReCoT在平均F1_LCS上达到了SFT上限的85%，缩小了这一差距，同时保持了解释性。重要的是，对ReCoT推理池的收敛性和更新模式的分析显示了推理链式（CoTs）的有效细化。

View on arXiv Download PDF AI Translation

cs.AI / 245 / 2610.17288

PhysCodeGym: Benchmarking Coding Agents for Physically Verifiable 4D Reconstruction

PhysCodeGym：针对物理可验证4D重建的编码代理基准测试

Shao, Shuai, He, Jinchen, Fan, Cunxin, Guo, Zihan, Wang, Shijian, Ren, Qihan, Zhang, Kangning, Xu, Zunnan, Li, Haoyuan, Wang, Hao, Lin, Jianghao, Liu, Weiwen, Zhang, Weinan

Abstract

Building physical simulations from video requires coding agents to translate observations into executable models and refine them through tool use. We introduce PhysCodeGym, a benchmark for physically verifiable 4D reconstruction with $156$ tasks covering rigid-body, liquid, deformable, and wind-driven motion. Agents autonomously inspect videos, write and run Blender programs, and revise their scenes using simulation feedback. Six metrics combine rule-based measurements with rubric-based judgments to assess visual and physical reconstruction quality. Program inspection, scene measurements, and counterfactual interventions test physical consistency under original and modified conditions. Experiments with frontier coding agents reveal a tradeoff between artifact quality and completion across reasoning-effort settings, while harness comparisons and agent traces link verification behavior with reconstruction performance. Case studies show that matching visible objects and events can coexist with physical defects. Counterfactual tests further expose latent contact and support failures in reconstructions that pass ordinary physical checks, highlighting the gap between reproducing observations and building reliable physical models. Code is available at https://github.com/ShaoShuai0605/PhysCodeGym.

Chinese Translation

从视频构建物理模拟需要编码代理将观察结果转化为可执行模型，并通过工具使用进行优化。我们介绍了PhysCodeGym，这是一个针对物理可验证4D重建的基准测试，涵盖了156个任务，涉及刚体、液体、可变形物体和风驱动运动。代理能够自主检查视频，编写和运行Blender程序，并利用模拟反馈修正其场景。六个指标结合了基于规则的测量和基于评分标准的判断，以评估视觉和物理重建质量。程序检查、场景测量和反事实干预测试在原始和修改条件下的物理一致性。与前沿编码代理的实验揭示了在推理努力设置中，伪影质量和完成度之间的权衡，同时，工具比较和代理轨迹将验证行为与重建性能联系起来。案例研究表明，匹配可见物体和事件可以与物理缺陷共存。反事实测试进一步揭示了在通过普通物理检查的重建中潜在的接触和支撑失败，突显了再现观察结果与构建可靠物理模型之间的差距。代码可在 https://github.com/ShaoShuai0605/PhysCodeGym 获取。

View on arXiv Download PDF AI Translation

cs.AI / 246 / 2610.17296

Towards Resilience: Slice-Driven Orchestration of Mobile IoMT in 6G-Edge-Continuum

迈向韧性：基于切片驱动的6G边缘连续体中移动医疗物联网的编排

Zeeshan, Muhammad, Shahid, Hafiz Faheem, Harjula, Erkki

Abstract

Emerging healthcare services, such as real-time diagnostics, remote health monitoring and computer-assisted surgery, require stringent Quality of Service (QoS) guarantees while operating under limited and heterogeneous edge resources. Existing Internet of Medical Things (IoMT) orchestration methods primarily focus on communication metrics such as latency and bandwidth, while overlooking edge resource constraints and dynamic adaptation to workload and user mobility. This paper presents a novel resource-slicing-based adaptive orchestration framework that integrates 5G network slicing (NS), resource-aware orchestration, and mobility-aware task offloading into a single adaptive system. IoMT services are assigned to distinct 5G network slices while jointly considering CPU capacity, memory availability and amount of storage, enabling more efficient service placement and execution. In addition, the system supports user mobility and dynamic task offloading to maintain service continuity under changing network conditions. SimPy-based simulation demonstrates that the proposed method enhances service deployment efficiency, reduces request blocking, and improves user acceptance while maintaining reliable performance across diverse IoMT scenarios.

Chinese Translation

新兴的医疗服务，如实时诊断、远程健康监测和计算机辅助手术，要求在有限且异构的边缘资源下提供严格的服务质量（QoS）保证。现有的医疗物联网（IoMT）编排方法主要关注通信指标，如延迟和带宽，而忽视了边缘资源的限制以及对工作负载和用户移动性的动态适应。本文提出了一种新颖的基于资源切片的自适应编排框架，该框架将5G网络切片（NS）、资源感知编排和移动感知任务卸载整合为一个自适应系统。IoMT服务被分配到不同的5G网络切片，同时共同考虑CPU容量、内存可用性和存储量，从而实现更高效的服务部署和执行。此外，该系统支持用户移动性和动态任务卸载，以在网络条件变化时保持服务的连续性。基于SimPy的仿真结果表明，所提出的方法提高了服务部署效率，减少了请求阻塞，并在多样化的IoMT场景中改善了用户接受度，同时保持了可靠的性能。

View on arXiv Download PDF AI Translation

cs.AI / 247 / 2610.17301

The Agent Knew: Internal States Forecast Unauthorized Actions and Stop Them, and the Lies That Follow

代理已知：内部状态预测未授权行为并阻止它们，以及随之而来的谎言

Fayyazi, Arya, Kamal, Mehdi, Pedram, Massoud

Abstract

Language-model agents sometimes act beyond their authorization and then misreport what they did. On CAPE-Bench, matched agent tasks in which one fact decides whether an action is authorized, frontier open-weight agents do both, and one behavior marks the lie: retroactive permission laundering, acting and then asking permission as if the action were pending, precedes a false report in 70% of 1081 fresh informed violations, against 8% otherwise (preregistered odds ratio 7.0); frontier agents on tau^2-bench instead fabricate the authorization before violating. A linear readout of the agent's residual stream forecasts the violation before the action (AUROC 0.94) and the false report before the report, and beats chain-of-thought monitors when the agent is pressured to hide its reasoning. Gating every proposed action on this forecast cuts violations over whole episodes in four model families, and sending repeated flags to an approver keeps authorized work (Gemma-4: violations 0.70 -> 0.09, completion 1.00; a model approver that escalates unclear cases keeps this gain with a person seeing 25% of asks), at about 10-8 of the agent's compute and 109 times less than a chain-of-thought monitor. On AgentDojo, an external prompt-injection benchmark, the same readout calibrated on 29 tasks, with the steering unchanged, cuts attack success on unseen tasks in two families (Gemma-4 11.7% -> 5.7%, gpt-oss 28.5% -> 17.5%) while keeping benign utility (Gemma-4 0.637 vs 0.637; gpt-oss 0.667 vs 0.681). Preventing the violation is also the most reliable way to prevent the lie (misreported violations 0.10 -> 0.01 pooled over three families, preregistered), whereas lie-specific fixes after a violation are weak. All tests are preregistered and hash-locked; code, data and the full record, including tests that did not pass, are at https://github.com/AryaFayyazi/Honesty.

Chinese Translation

语言模型代理有时会超出其授权范围进行操作，然后错误报告其所做的事情。在CAPE-Bench上，匹配的代理任务中，一个事实决定了某个行为是否被授权，前沿开放权重代理同时执行这两项操作，其中一种行为标志着谎言：追溯性许可洗钱，在行为发生后请求许可，仿佛该行为仍在待决状态，这在1081个新知情违规案例中占70%，而其他情况下为8%（预注册的比值比为7.0）；而在tau^2-bench上的前沿代理则是在违规之前伪造授权。代理的残余流的线性读出在行为发生之前预测违规（AUROC 0.94），并在报告之前预测错误报告，并且在代理被迫隐藏其推理时优于思维链监控。基于这一预测对每个提议的行为进行筛选，减少了四个模型家族在整个情节中的违规行为，并且向审批者发送重复标记保持了授权工作（Gemma-4：违规率从0.70降至0.09，完成率1.00；一个在不明确情况下升级的模型审批者保持了这一增益，且一个人看到25%的请求），大约占代理计算的10^-8，且比思维链监控少109倍。在AgentDojo，一个外部提示注入基准上，经过29个任务校准的相同读出，在不改变引导的情况下，减少了两个家族中未见任务的攻击成功率（Gemma-4从11.7%降至5.7%，gpt-oss从28.5%降至17.5%），同时保持良性效用（Gemma-4 0.637对比0.637；gpt-oss 0.667对比0.681）。防止违规也是防止谎言的最可靠方法（在三家族中汇总的错误报告违规率从0.10降至0.01，预注册），而在违规后针对谎言的特定修复效果较弱。所有测试均已预注册并进行哈希锁定；代码、数据和完整记录，包括未通过的测试，均可在https://github.com/AryaFayyazi/Honesty获取。

View on arXiv Download PDF AI Translation

cs.AI / 248 / 2610.17305

Reliable Text-to-SQL in Resource-Constrained Settings with Small LLMs via Iterative Retry Workflows

在资源受限环境下通过迭代重试工作流实现小型大语言模型的可靠文本到SQL转换

Benziane, Bilel Abderrahmane, Schepler, Xavier

Abstract

Large Language Models (LLMs) have recently improved Text-to-SQL generation, but their computational cost can hinder deployment in resource-constrained settings. This paper evaluates the Text-to-SQL capabilities of small and medium open-source LLMs (1B-8B parameters) under a controlled inference budget, focusing on how model specialization and prompt formulation affect performance. Using the Spider benchmark, we compare multiple prompt representations and assess the accuracy of generated queries. Our results show that SQL-specialized models are consistently more reliable than general-purpose and chat-oriented models at small scales, and that performance can vary substantially depending on the prompt used. To reduce this variability without resorting to larger models, we introduce three lightweight inference workflows that automatically retry with alternative prompts and/or models when failures occur, including an optional syntax-correction step. These workflows significantly improve end-to-end success rates, with a trade-off in additional inference calls. These results demonstrate a practical path to stronger Text-to-SQL performance with small LLMs, by trading modest additional inference cost for higher reliability.

Chinese Translation

大型语言模型（LLMs）最近在文本到SQL生成方面取得了进展，但其计算成本可能会阻碍在资源受限环境中的部署。本文评估了小型和中型开源LLMs（1B-8B参数）在受控推理预算下的文本到SQL能力，重点关注模型专业化和提示构造对性能的影响。我们使用Spider基准测试比较了多种提示表示，并评估了生成查询的准确性。我们的结果表明，SQL专业化模型在小规模下的可靠性始终优于通用和聊天导向模型，并且性能可能因所使用的提示而有显著变化。为了减少这种变异性而不依赖于更大的模型，我们引入了三种轻量级推理工作流，当发生失败时自动使用替代提示和/或模型进行重试，包括一个可选的语法纠正步骤。这些工作流显著提高了端到端的成功率，但需要额外的推理调用作为权衡。这些结果展示了一条通过适度增加推理成本换取更高可靠性的实际路径，以实现小型LLMs的更强文本到SQL性能。

View on arXiv Download PDF AI Translation

cs.AI / 249 / 2610.17310

Beyond Successful Play: Diagnosing Opponent Modeling in Large Language Models

超越成功的游戏：大型语言模型中的对手建模诊断

Xie, Hanbo, Zhao, Yue, Wilson, Robert C.

Abstract

Language-model agents are expected to adapt to others whose behavior changes through interaction. Yet adaptation may rely on learned behavioral patterns rather than theory of mind (ToM): the ability to infer others' mental states, including beliefs and intentions. We introduce a benchmark and diagnostic framework for repeated hide-and-seek against algorithmic opponents with explicitly defined recursive belief updates. Their ground-truth beliefs and action probabilities let us distinguish adaptive play, action prediction, and belief tracking through live interaction, fixed-history evaluation, and controlled information interventions. With structured elicitation and behavioral supervised fine-tuning (SFT), we find improved play in several tested settings, but diagnostics reveal limitations in opponent prediction and belief tracking. We test whether supervising opponent action probabilities can address this predictive deficit in Qwen2.5-7B-Instruct. Predictions improve when the model is told the opponent's type and parameter settings, but become less accurate when it must infer the opponent from history without this information. Training on examples that also omit these details reduces the gap, although improvements vary across opponents. Is this limitation specific to smaller open-weight models, or does it also appear in a frontier model? GPT-5.4 also benefits from correct opponent beliefs: supplying them improves its predictions and choices, whereas deeper explicit belief reports and self-generated memory do not consistently improve forecasting from interaction history. This contrast suggests that using useful belief information is easier than inferring it autonomously, even for a frontier model. Together, our findings distinguish improved play and action prediction from reliable belief tracking, providing a framework for evaluating mentalizing and testing targeted improvements.

Chinese Translation

语言模型代理预计将能够适应通过互动而变化的他人的行为。然而，适应可能依赖于学习到的行为模式，而非心智理论（Theory of Mind, ToM）：推断他人心理状态的能力，包括信念和意图。我们引入了一个基准和诊断框架，用于针对具有明确递归信念更新的算法对手进行重复的捉迷藏游戏。它们的真实信念和行动概率使我们能够通过实时互动、固定历史评估和控制信息干预来区分适应性游戏、行动预测和信念追踪。通过结构化引导和行为监督微调（Supervised Fine-Tuning, SFT），我们发现多个测试环境中的游戏表现有所改善，但诊断结果揭示了在对手预测和信念追踪方面的局限性。我们测试了监督对手行动概率是否可以解决Qwen2.5-7B-Instruct中的这一预测缺陷。当模型被告知对手的类型和参数设置时，预测有所改善，但当它必须从历史中推断对手而没有这些信息时，准确性降低。训练时也省略这些细节的示例缩小了差距，尽管在不同对手间的改善程度不同。这个限制是特定于较小的开放权重模型，还是在前沿模型中也会出现？GPT-5.4同样受益于正确的对手信念：提供这些信念改善了它的预测和选择，而更深入的显式信念报告和自生成的记忆并未始终改善基于互动历史的预测。这种对比表明，利用有用的信念信息比自主推断更容易，即使对于前沿模型而言。综合来看，我们的研究结果区分了改善的游戏和行动预测与可靠的信念追踪，为评估心智化和测试针对性改进提供了框架。

View on arXiv Download PDF AI Translation

cs.AI / 250 / 2610.17331

Do Theorem Provers Need Theorem-Proving Agents? mini-Lean-agent: A Coding-Agent Baseline for Lean

定理证明器需要定理证明代理吗？mini-Lean-agent：Lean的编码代理基线

Ortega, Tomas, Park, Simon, Khaitan, Ayush, Fowl, Liam, Kontorovich, Alex, Arora, Sanjeev

Abstract

General-purpose coding agents use standard shell and file interactions, yet formal theorem-proving agents often rely on specialized, ad-hoc tooling. This makes it difficult to determine which components of theorem-proving agents provide benchmark gains: an improved base model, a specialized harness, a specific tool, or a combination of any of them. We introduce mini-Lean-agent, a deliberately minimal agent for Lean tasks whose only model-visible tool is bash. The model must use bash to inspect and edit the repository, search local libraries, invoke Lean, and manage its own intermediate work, while a separate trusted evaluator verifies the final artifact. We compare mini-Lean-agent against AxProverBase and Numina-Lean-Agent on the Putnam 2025 benchmark. Experiments with Gemini 3.8 Flash as the base model show that mini-Lean-agent solves 7/12 problems at \$2 per problem limit, compared with AxProverBase's 4/12. With Astra as the base model, mini-Lean-agent solves 12/12 problems at an average cost of \$1.28 per problem. This suggests that a generic coding agent with a minimal harness can already achieve strong performance on Lean benchmarks. Our results motivate using mini-Lean-agent as a standard baseline when evaluating theorem-proving infrastructure.

Chinese Translation

通用编码代理使用标准的shell和文件交互，而正式的定理证明代理通常依赖于专门的、临时的工具。这使得很难确定定理证明代理的哪些组件提供了基准增益：是改进的基础模型、专门的工具架、特定工具，还是它们的组合。我们介绍了mini-Lean-agent，这是一个专为Lean任务设计的极简代理，其唯一的模型可见工具是bash。该模型必须使用bash来检查和编辑代码库、搜索本地库、调用Lean，并管理自己的中间工作，而一个独立的可信评估器则验证最终的工件。我们将mini-Lean-agent与AxProverBase和Numina-Lean-Agent在Putnam 2025基准上进行了比较。以Gemini 3.8 Flash作为基础模型的实验表明，mini-Lean-agent在每个问题限制为2美元的情况下解决了12个问题中的7个，而AxProverBase仅解决了4个。在以Astra作为基础模型的情况下，mini-Lean-agent解决了12个问题，平均每个问题的成本为1.28美元。这表明，具有最小工具架的通用编码代理已经能够在Lean基准上实现强劲的性能。我们的结果激励在评估定理证明基础设施时使用mini-Lean-agent作为标准基线。

View on arXiv Download PDF AI Translation

cs.AI / 251 / 2610.17338

Budgeted KV-Cache Repair for Evolving Contexts

预算化的 KV-缓存修复用于不断演变的上下文

Yu, Han, Zhu, Wenhui, Cao, Jincheng, Cai, Rui, Xu, Yuanda, Zhou, Zhengze, Chen, Xiwen, Wang, Hao, Dong, Xuanzhao, Shi, Han, Zhou, Menglin, Hamidi, Shayan Mohajer, Su, Lin, Razghandi, Mina, Ghosh, Sayan, Lan, Qing, Jain, Rohit, Geramifard, Alborz

Abstract

Language-model agents revise their working context as information becomes obsolete, but deleting text does not erase its influence on cached computation. Later key-value (KV) states remain conditioned on the deleted content, while recomputing the entire retained suffix can be costly. We introduce budgeted KV-cache repair: recovering task performance by recomputing a user-chosen fraction of the retained suffix. Comparing repair allocations under identical relevance scores and token budgets, we find that refreshing coherent regions can substantially outperform refreshing isolated tokens. This finding motivates Semantic-Unit Refresh (SUR), which allocates the budget to query-relevant text regions using inexpensive embedding scores, without training or an auxiliary transformer forward. To execute these repairs efficiently, a layer-packed executor processes selected states together at each layer while preserving full causal cache access. Across five models, SUR achieves strong recovery in controlled deletion tasks while recomputing only 2-5% of the suffix. On four natural multi-hop QA datasets, it achieves near-full-recompute exact match at a 5% budget with lower post-edit latency. The executor accelerates fragmented refresh by up to 16.1x over interval-wise execution.

Chinese Translation

语言模型代理在信息变得过时时会修订其工作上下文，但删除文本并不会消除其对缓存计算的影响。后续的键值（KV）状态仍然受到已删除内容的影响，而重新计算整个保留后缀可能代价高昂。我们提出了预算化的 KV-缓存修复：通过重新计算用户选择的保留后缀的一部分来恢复任务性能。在相同的相关性评分和标记预算下比较修复分配时，我们发现刷新连贯区域的效果显著优于刷新孤立标记。这一发现促使了语义单元刷新（Semantic-Unit Refresh, SUR）的提出，该方法利用低成本的嵌入评分将预算分配给与查询相关的文本区域，而无需训练或辅助的变换器前向计算。为了高效执行这些修复，层打包执行器在每一层同时处理选定状态，同时保持完整的因果缓存访问。在五个模型中，SUR在受控删除任务中实现了强劲的恢复，仅重新计算了2-5%的后缀。在四个自然多跳问答数据集上，它在5%的预算下实现了近乎完全重新计算的精确匹配，并且后期编辑延迟更低。执行器在间隔执行的基础上加速了碎片化刷新，速度提升高达16.1倍。

View on arXiv Download PDF AI Translation

cs.AI / 252 / 2610.17365

When Goals Leak Across Relationships: How Cached Goal Representations Shape Cross-Client Decisions

当目标在关系中泄漏：缓存目标表征如何影响跨客户决策

Gupta, Saloni, Krier, Sébastien, Matlin, Glenn

Abstract

As LLM agents take on more autonomous roles, they may interact with multiple clients and counterparties whose goals and incentives conflict. If an agent maintains shared inference state across these interactions, information associated with one relationship may influence decisions made for another. We test OLMo-3-7B-Instruct in a controlled two-client environment with 20 market scenarios, where each client is assigned an opposing objective. With shared conversational memory, the model deviates from the current client's objective in 22% of eligible decisions. We then investigate how these goals are represented in the KV cache. Linear probes show that client-goal ownership becomes highly recoverable in middle layers and is concentrated in particular attention heads. We use these results to identify where to intervene in the cache and test whether these representations affect the model's decisions. Ablating the competitor's cached goal representations at these locations shifts the model toward the current client. The effect is larger for heads identified by probing than for randomly selected heads, and larger when ablating goal tokens than matched non-goal tokens from the same client. Our results show that client-specific goal representations persist in the KV cache and can causally influence decisions made for other clients, highlighting the importance of relationship-specific state isolation in persistent LLM agents.

Chinese Translation

随着大型语言模型（LLM）代理承担越来越多的自主角色，它们可能会与多个客户和对手进行互动，这些客户和对手的目标和激励存在冲突。如果一个代理在这些互动中维持共享的推理状态，那么与一个关系相关的信息可能会影响为另一个关系做出的决策。我们在一个受控的双客户环境中测试了OLMo-3-7B-Instruct，设置了20个市场场景，其中每个客户被分配了相反的目标。在共享的对话记忆中，该模型在22%的合格决策中偏离了当前客户的目标。随后，我们研究了这些目标在KV缓存中的表征。线性探测显示，客户目标的归属在中间层中变得高度可恢复，并集中在特定的注意力头上。我们利用这些结果确定在缓存中进行干预的位置，并测试这些表征是否会影响模型的决策。在这些位置上消融竞争对手的缓存目标表征使模型朝向当前客户偏移。对于通过探测识别的头部，效果大于随机选择的头部，并且在消融来自同一客户的目标标记时效果更大，而非目标标记的匹配效果则较小。我们的结果表明，客户特定的目标表征在KV缓存中持续存在，并且可以因果性地影响为其他客户做出的决策，这突显了在持久的LLM代理中关系特定状态隔离的重要性。

View on arXiv Download PDF AI Translation

cs.AI / 253 / 2610.17376

What Makes a Latent Space Shared? Direct and Recoverable Compatibility in Motion VAEs

什么使潜在空间共享？运动变分自编码器中的直接和可恢复兼容性

Brito, Iago, Rios, Walcy, Fernandes, Lucas, de Oliveira, Kaiki, Filho, Arlindo Galvão

Abstract

A shared latent space is often assumed to make representations from different schemas directly interchangeable, yet sharing parameters, dimensionality, or a prior does not determine how corresponding instances are expressed in latent coordinates. We show that this assumption conflates recoverable correspondence, direct compatibility, and decoder reuse. In a controlled multi-schema motion VAE on HumanML3D, even identical inputs yield only $5.9$--$17.7\%$ raw Recall@1 and $0.221$--$0.484$ paired-token cosine, despite correspondence being almost perfectly recoverable after linear calibration. Cycle consistency strengthens recoverability but still leaves direct retrieval and token agreement near zero. In contrast, only $64$ paired clips raise raw Recall@1 to $88.4$--$90.3\%$ and token cosine to approximately $0.82$, making most post-hoc alignment unnecessary, although crossed decoding remains substantially worse than native reconstruction. We further find that reducing measured marginal discrepancy does not imply schema invariance. These results show that a shared backbone or prior does not ensure directly compatible latent coordinates, while limited paired correspondence can substantially improve direct compatibility in the tested setting.

Chinese Translation

共享潜在空间通常被假设为使来自不同模式的表示可以直接互换，然而共享参数、维度或先验并不能决定对应实例在潜在坐标中的表达方式。我们展示了这一假设混淆了可恢复的对应关系、直接兼容性和解码器重用。在一个控制的多模式运动变分自编码器（Motion VAE）实验中，使用HumanML3D数据集，即使是相同的输入，其原始Recall@1仅为$5.9$--$17.7\%$，配对标记余弦相似度为$0.221$--$0.484$，尽管经过线性校准后，几乎可以完美恢复对应关系。循环一致性增强了可恢复性，但仍然使得直接检索和标记一致性接近于零。相比之下，仅$64$个配对片段就将原始Recall@1提高至$88.4$--$90.3\\%$，标记余弦相似度约为$0.82$，使得大多数事后对齐变得不必要，尽管交叉解码的效果仍明显低于原生重建。我们进一步发现，减少测量的边际差异并不意味着模式不变性。这些结果表明，共享的骨干网络或先验并不能确保直接兼容的潜在坐标，而有限的配对对应关系可以在测试环境中显著改善直接兼容性。

View on arXiv Download PDF AI Translation

cs.AI / 254 / 2610.17378

Get In, Decide, Prove: A Merchant-Side Readiness Benchmark for Agentic Commerce

进入、决策、证明：一种面向商家的代理商业准备度基准

de Valois-Franklin, Adrian, Bogdan, Alex, Ibrahim, Anas, Shaer, Ibrahim, Andguladze, Tamaz

Abstract

AI agents now browse and buy on people's behalf, and benchmarks measure how well the agents perform. Far less is known about the other side of the transaction: whether a merchant's website lets an agent in, publishes terms precise enough to decide a purchase, and leaves evidence of what was agreed. This paper presents a merchant-side benchmark built on those three questions and reports a pilot of it. The central measurement, decision sufficiency, rates each criterion of a concrete buyer mandate (total cost, cancellation or return terms, service date, variant identity) as determinate, indeterminate, or unknown from the merchant's own published evidence. Readiness scores carry their evidence coverage and missingness bounds; access is reported separately for a declared bot, a declared headless browser, and a user-present browser; and each observed term is mapped to the RAILS admissibility class of the evidence that could support it in a dispute. Four large merchants received the full assessment: three refused the declared profiles at their offer pages, and no mandate could be confirmed as met, because each merchant left a total or a refund term undetermined. An access census of 555 sites of merchants that two payment processors publicly name as clients found the home page unreadable to both declared profiles at 105 sites (18.9%). A declared assessment of 330 merchants, scored by deterministic rules without a rater, found that evidence coverage rather than the score decides who can be ranked: 115 merchants (35%) reached the ranking threshold, rankable profiles scored a median of 89 to 90, and material terms was the one pillar well below the others, because the all-in total was decidable at 18 merchants. The pilot shows that the measurements can be taken; it does not validate them or rank merchants, and the weights and the projected clearing band await validation against outcomes.

Chinese Translation

人工智能代理现在代表人们浏览和购买商品，而基准测试衡量这些代理的表现。关于交易另一方的信息则少之又少：商家的网站是否允许代理进入，是否发布了足够明确的条款以便做出购买决策，以及是否留下了达成协议的证据。本文提出了一种基于这三个问题的商家侧基准，并报告了其试点研究。核心测量指标——决策充分性，对具体买方授权的每个标准（总成本、取消或退货条款、服务日期、变体身份）进行确定、未确定或未知的评级，依据商家自身发布的证据。准备度评分涵盖了证据的覆盖范围和缺失边界；对声明的机器人、声明的无头浏览器和用户在场的浏览器的访问情况分别报告；每个观察到的条款映射到可以在争议中支持它的证据的RAILS可接受性类别。四家大型商家接受了全面评估：三家在其报价页面拒绝了声明的配置文件，且由于每个商家留下了一个总费用或退款条款未确定，因此没有授权能够被确认。对两家支付处理器公开称为客户的商家的555个网站进行的访问普查发现，105个网站（18.9%）的主页对两种声明的配置文件均不可读。对330家商家的声明评估，采用无评审者的确定性规则评分，发现证据覆盖范围而非评分决定了谁可以被排名：115家商家（35%）达到了排名阈值，排名配置文件的得分中位数为89到90，而实质条款则是唯一远低于其他条款的支柱，因为只有18家商家的总费用是可决策的。试点研究表明可以进行这些测量；但并未验证这些测量或对商家进行排名，权重和预计的清算范围仍需根据结果进行验证。

View on arXiv Download PDF AI Translation

cs.AI / 255 / 2610.17379

EPOCH: Coverage-Aware Evolution with Experiential Preference Optimization for LLM-based Automated Heuristic Design

EPOCH：基于覆盖意识的进化与经验偏好优化用于基于LLM的自动启发式设计

Gungordu, Oguzhan, Tursucular, Ayse, Xiong, Siheng, Fekri, Faramarz

Abstract

LLM-based automated heuristic design (AHD) discovers heuristics for combinatorial optimization problems (COPs), where an LLM generates candidates in an evolutionary search. Existing frameworks freeze the generator or fine-tune it on a scalar fitness signal hiding which instances each heuristic covers. Reflection-based frameworks improve heuristics with natural-language feedback, but it stays in the prompt and never reaches the generator's weights. We propose EPOCH, which makes instance coverage the objective of the evolutionary search and fine-tunes the generator to internalize reflection. Parent selection, operator allocation, and stagnation recovery all optimize how well the heuristics jointly cover the training instances, so the search returns a complementary heuristic ensemble. The generator learns from its own search by experiential preference optimization. A teacher LLM returns a natural-language insight on the generated heuristics, and the generator produces new heuristics with this insight. These improvements join the preference pairs for direct preference optimization (DPO). Since DPO only teaches the generator to rank these improvements, a reflective self-distillation (RSD) term makes it more likely to generate them on its own, and the gains accumulate over the long horizon of the search. Across COPs and open-source LLM backbones, EPOCH's heuristic ensembles outperform inference-only and fine-tuned baselines and generalize across problem scales.

Chinese Translation

基于LLM的自动启发式设计（AHD）为组合优化问题（COPs）发现启发式方法，其中LLM在进化搜索中生成候选解。现有框架将生成器冻结或在一个隐藏每个启发式覆盖哪些实例的标量适应度信号上进行微调。基于反思的框架通过自然语言反馈改进启发式方法，但这种反馈仅停留在提示中，未能传递到生成器的权重。我们提出了EPOCH，它将实例覆盖作为进化搜索的目标，并微调生成器以内化反思。父代选择、操作分配和停滞恢复都优化了启发式方法共同覆盖训练实例的效果，从而使搜索返回一个互补的启发式集合。生成器通过经验偏好优化从自身搜索中学习。教师LLM对生成的启发式方法提供自然语言的见解，生成器则利用这些见解生成新的启发式方法。这些改进结合成偏好对以进行直接偏好优化（DPO）。由于DPO仅教导生成器对这些改进进行排序，因此反思自蒸馏（RSD）项使其更有可能自主生成这些改进，而收益在搜索的长期过程中不断累积。在COPs和开源LLM骨干网络中，EPOCH的启发式集合优于仅推理和微调的基线，并在不同问题规模上具有良好的泛化能力。

View on arXiv Download PDF AI Translation

cs.AI / 256 / 2610.17391

Learning to Play: Reinforcement Learning and Games

学习游戏：强化学习与游戏

Plaat, Aske

Abstract

Amazing breakthroughs in reinforcement learning have taken place. Computers teach themselves to play Chess and Go and beat world champions. There is talk about expanding application areas towards general artificial intelligence (AI). The breakthroughs in Backgammon, Checkers, Chess, Atari, Go, Poker, and StarCraft have shown that we can build machines that exhibit intelligent game playing of the highest level. These successes have been widely publicized in the media, and inspire AI entrepreneurs and scientists alike. Reinforcement learning in games has become a mainstream AI research topic. It is a broad topic, and the successes build on a range of diverse techniques, from exact planning algorithms, to adaptive sampling, deep function approximation, and ingenious self-play methods. Perhaps because of the breadth of these technologies, or because of the recency of the breakthroughs, there are few books that explain these methods in depth. This book covers all methods in one comprehensive volume, explaining the latest research, bringing you as close as possible to working implementations, with many references to the original research papers. The programming examples in this book are in Python, the language in which most current reinforcement learning research is conducted. We help you to get started with machine learning frameworks such as Gym, TensorFlow, and Keras, and provide exercises to help understand how AI is learning to play.

Chinese Translation

强化学习领域取得了惊人的突破。计算机能够自我学习下棋和围棋，并击败世界冠军。人们开始讨论将应用领域扩展到通用人工智能（AI）。在西洋棋、跳棋、围棋、德州扑克和星际争霸等游戏中的突破表明，我们可以构建出展现高水平智能游戏能力的机器。这些成功在媒体上得到了广泛宣传，激励了人工智能的企业家和科学家。游戏中的强化学习已成为主流的人工智能研究主题。这个主题广泛，成功的实现依赖于多种不同的技术，从精确的规划算法到自适应采样、深度函数逼近以及巧妙的自我对弈方法。也许正因为这些技术的广泛性，或者因为突破的时效性，目前关于这些方法的深入书籍寥寥无几。本书在一个综合卷中涵盖了所有方法，解释了最新的研究，尽可能接近实际实现，并引用了许多原始研究论文。书中的编程示例使用Python，这是目前大多数强化学习研究所使用的语言。我们帮助您入门机器学习框架，如Gym、TensorFlow和Keras，并提供练习以帮助理解人工智能是如何学习游戏的。

View on arXiv Download PDF AI Translation

cs.AI / 257 / 2610.17401

Generalizable field of view extension for diffusion MRI via conditional flow matching

通过条件流匹配实现扩展扩展的可推广视野的扩散MRI

Rudravaram, Gaurav, Zuo, Lianrui, Topolnjak, Ema, Kim, Michael E., Saunders, Adam M., McMaster, Elyssa M., Trujillo, Paula, Wesley, Ely E., McKnight, Colin D., Pandharipande, Pratik P., Patel, Mayur B., Strength, Cayce, Pund, Brenda T., Maiga, Amelia W, Dittus, Robert S., Bennett, David A., Schneider, Julie A., Arfanakis, Konstantinos, Team, The HABS-HD Study, Archer, Derek B., Hohman, Timothy J., Bodien, Yelena, Moyer, Daniel C., Schilling, Kurt G., Landman, Bennett A.

Abstract

Incomplete field-of-view (FOV) diffusion MRI (dMRI) acquisitions compromise whole-brain white matter analysis by excluding brain regions from tractography and connectivity studies. While deep generative models for dMRI FOV extension have been proposed, their generalizability remains limited by the acquisition-dependent nature of dMRI, where signal intensity varies with b-value, gradient direction, and scanner protocol, and the inherent 4D complexity of the data, making it difficult to train models that transfer across sites or acquisition schemes without retraining. We propose a flow-matching generative model that extends the dMRI FOV in both the superior and inferior directions by predicting apparent diffusion coefficients (ADC) in six directions, conditioned on the two nearest acquired slices and the full-FOV T1w image to enforce both local diffusion continuity and anatomical consistency. The ADC representation decouples the model from acquisition-specific b-values and gradient directions, providing a middle-ground across protocols. The model was trained on 3,103 subjects (5,105 sessions) across five different datasets spanning diverse acquisition schemes (b = 500 and 1000 s/mm$^2$; 21--192 gradient directions), four single-shell (b = 1000) and one multi-shell (b = 500/1000). Evaluated on 299 subjects across five independent datasets spanning b = 750--1500 s/mm$^2$ and 32--90 gradient directions, with single- and multi-shell schemes unseen during training, the proposed method achieves significantly improved bundle Dice score on downstream tractography (p < 0.01).

Chinese Translation

不完整的视野（FOV）扩散MRI（dMRI）采集会通过排除脑区影响全脑白质分析，进而影响纤维追踪和连接性研究。尽管已有针对dMRI FOV扩展的深度生成模型被提出，但由于dMRI的采集依赖性特征（信号强度随b值、梯度方向和扫描协议变化）以及数据固有的4D复杂性，其可推广性仍然有限，这使得在不同站点或采集方案之间转移模型而无需重新训练变得困难。我们提出了一种流匹配生成模型，通过预测在六个方向上的表观扩散系数（ADC），在上下两个方向上扩展dMRI FOV，该预测以两个最近采集的切片和全视野T1加权图像为条件，以确保局部扩散连续性和解剖一致性。ADC表示将模型与特定采集的b值和梯度方向解耦，从而在不同协议之间提供了一个中间方案。该模型在五个不同数据集中对3103名受试者（5105个会话）进行了训练，涵盖了多种采集方案（b = 500和1000 s/mm²；21--192个梯度方向），包括四个单壳（b = 1000）和一个多壳（b = 500/1000）。在299名受试者的五个独立数据集上进行评估，涵盖b = 750--1500 s/mm²和32--90个梯度方向，其中单壳和多壳方案在训练期间未见，所提出的方法在下游纤维追踪中显著提高了束Dice评分（p < 0.01）。

View on arXiv Download PDF AI Translation

cs.AI / 258 / 2610.17410

Agency as a New Organizing Abstraction for Computer Science

代理作为计算机科学的新组织抽象

Foster, Ian

Abstract

Advances in large-scale machine learning have enabled agentic AI systems: persistent, autonomous computational entities that pursue goals, initiate actions, and interact with open environments. These systems are often framed as applications of foundation models. We argue that this framing is incomplete: agency is also a candidate new organizing abstraction for computer science. What is new is not the concept of an agent, nor runtime policy generation per se, but its general-purpose, open-ended form, in which systems can synthesize courses of action from goals, observations, and memory, over tools and environments for which no complete task-specific model was specified in advance. Thus agency becomes a systems-level abstraction as well as an AI modeling concept. Where the process abstraction organizes computation around executing programs, agency organizes it around pursuing goals under bounded resources: a process executes a program; an agent pursues a goal. From this perspective, the recurring failures of deployed agentic systems (runaway resource consumption, behavior redirected by adversarial content, opaque continuous operation) are symptoms of structural abstraction gaps, not isolated engineering defects. We examine these gaps across six dimensions (execution, interfaces, correctness, oversight, coordination, and persistence) and argue that closing them requires coordinated advances in systems,

Chinese Translation

大规模机器学习的进展使得代理智能系统成为可能：这些系统是持久的、自主的计算实体，能够追求目标、发起行动并与开放环境进行交互。这些系统通常被视为基础模型的应用。我们认为这种框架是不完整的：代理也是计算机科学中一种新的组织抽象。新的并不是代理的概念，也不是运行时策略生成本身，而是其通用的、开放式的形式，在这种形式中，系统可以根据目标、观察和记忆，从未事先指定完整任务特定模型的工具和环境中综合行动方案。因此，代理不仅成为系统级的抽象，也成为人工智能建模的概念。在这一视角下，过程抽象围绕执行程序组织计算，而代理则围绕在有限资源下追求目标组织计算：一个过程执行一个程序；一个代理追求一个目标。从这个角度来看，已部署的代理系统反复出现的失败（资源消耗失控、行为被对抗性内容引导、不透明的持续操作）是结构抽象缺口的症状，而不是孤立的工程缺陷。我们在六个维度（执行、接口、正确性、监督、协调和持久性）中考察这些缺口，并认为弥补这些缺口需要系统方面的协调进展。

View on arXiv Download PDF AI Translation

cs.AI / 259 / 2610.17436

SchemaWeaver: Accelerating On-Device Mobile Agentic AI via Schema-Guided Parallel LLM Decoding

SchemaWeaver：通过模式引导的并行 LLM 解码加速移动设备上的代理 AI

Yang, Boyuan, Gao, Wei

Abstract

Large language models (LLMs) on mobile devices can act as agents that call external software tools to complete complex tasks, but the major challenge of supporting on-device mobile agentic AI with LLMs is the long latency of LLM tool calling on mobile devices. In particular, the decode phase is the main contributor of such long latency, but is hard to speed up on mobile processors as it is bound by memory bandwidth rather than compute power. To address this limitation, in this paper we present SchemaWeaver, a new on-device mobile agentic AI technique that enforces schema-guided parallel decoding in the decode phase of LLM tool calling, to reduce its latency by producing multiple tokens in each decode pass. SchemaWeaver can be generically applied to the LLM inference engine on mobile devices without involving any training efforts or changes on LLMs, and our experiment results on major mobile SoCs and LLM families show that SchemaWeaver can reduce the LLM tool calling latency by up to 3.9 times, compared to the competitive baselines.

Chinese Translation

移动设备上的大型语言模型（LLMs）可以作为代理，调用外部软件工具以完成复杂任务，但支持移动设备上基于 LLM 的代理 AI 的主要挑战是 LLM 工具调用的长延迟。特别是，解码阶段是导致这种长延迟的主要原因，但由于受限于内存带宽而非计算能力，在移动处理器上很难加速。为了解决这一限制，本文提出了 SchemaWeaver，一种新的移动设备上代理 AI 技术，它在 LLM 工具调用的解码阶段强制执行模式引导的并行解码，通过在每次解码过程中生成多个标记来减少延迟。SchemaWeaver 可以通用地应用于移动设备上的 LLM 推理引擎，而无需进行任何训练工作或对 LLM 进行更改，我们在主要移动 SoC 和 LLM 家族上的实验结果表明，与竞争基线相比，SchemaWeaver 可以将 LLM 工具调用的延迟减少多达 3.9 倍。

View on arXiv Download PDF AI Translation

cs.AI / 260 / 2610.17439

Large Language Models Extend Known Research But Do Not Create New Research Directions

大型语言模型扩展已知研究但未能创造新的研究方向

Liu, Zhihua, Butler, Kurt, Machlanski, Damian, Bueno, Danilo, Tsaftaris, Sotirios A.

Abstract

The success of autonomous research systems relies on the quality of the scientific hypotheses they generate. Recent studies report that human reviewers often rate LLM-generated ideas as more novel than those from human experts, raising hope that AI can accelerate scientific discovery. Yet it is unclear whether these models can do more than extend existing knowledge. Can LLMs drive genuine innovation by revealing new research directions, or are they limited to repeating and recombining what science already knows? This unresolved question lies at the heart of discussions about AI and the future of science. To address this, we distinguish two types of novelty: distance, which measures how far a hypothesis lies from existing work, and direction, which captures whether the hypothesis pursues new lines of inquiry beyond those already present. These two measures diverge sharply for LLM-generated work. We represent scientific papers in a shared conceptual space using a pretrained scientific-paper embedding model. Within this space, we embed 26.9 million human-authored papers across 1,325 research topics in six natural-science fields and 524,308 LLM-generated hypotheses from eight models. Distance is measured using existing metrics. To quantify direction, we introduce the orthogonal novelty score (ONS). LLMs consistently generate ideas more distant from existing work than human-authored papers, up to 16\% further when the LLM is given the papers farthest from the centre of the topic. However, LLM hypotheses have a 15--23\% lower median ONS than published papers from the same topic. Thus, LLMs move further from what is already known, but remain within established directions. This pattern persists across changes in various configurations. As a result, LLMs excel at extending the frontier but the creation of new directions is still led by human scientists.

Chinese Translation

自主研究系统的成功依赖于其生成的科学假设的质量。近期研究报告显示，人类评审者往往将大型语言模型（LLM）生成的创意评为比人类专家更具新颖性，这让人们对人工智能加速科学发现的可能性充满期待。然而，目前尚不清楚这些模型是否能够超越现有知识的扩展。LLM能否通过揭示新的研究方向来推动真正的创新，还是仅限于重复和重新组合科学已经知道的内容？这个未解的问题是关于人工智能与科学未来讨论的核心。为了解决这一问题，我们区分了两种新颖性：距离，测量假设与现有工作的距离；方向，捕捉假设是否追求超出已有研究的新探究方向。这两种度量在LLM生成的工作中显著不同。我们使用预训练的科学论文嵌入模型在共享概念空间中表示科学论文。在这个空间中，我们嵌入了来自六个自然科学领域的1,325个研究主题中2690万篇人类撰写的论文和来自八个模型的524,308个LLM生成的假设。距离使用现有指标进行测量。为了量化方向，我们引入了正交新颖性评分（ONS）。LLM生成的创意与现有工作的距离始终大于人类撰写的论文，在LLM获得最远离主题中心的论文时，距离可达16\%。然而，LLM假设的中位数ONS比同一主题的已发表论文低15--23\%。因此，LLM在已知内容之外的探索更为深入，但仍然局限于既定方向。这一模式在各种配置变化中持续存在。因此，LLM在扩展前沿方面表现出色，但新的研究方向的创造仍由人类科学家主导。

View on arXiv Download PDF AI Translation

cs.AI / 261 / 2610.17463

An FPGA-Based Spiking Neural Network for Spatiotemporal Pattern Learning and Classification

基于FPGA的脉冲神经网络用于时空模式学习与分类

Jin, Jingang, Ma, Yue, Zhang, Zhenhang, Qiu, Qinru

Abstract

Spiking Neural Networks (SNNs) are promising for energy-efficient spatiotemporal pattern recognition at the edge, but efficient deployment and on-device adaptation require specialized hardware. In this work, we present an FPGA implementation of multilayer SNNs for spatiotemporal pattern classification, in which both inference and training take place on the same hardware. Our design adopts a temporal-forward-only training algorithm that avoids the memory and latency overhead associated with traditional Backpropagation Through Time (BPTT). A mixed-precision scheme balances training stability with resource efficiency, and architectural optimization maximizes hardware and memory utilization. We further identify and resolve hazards that arise when inference and learning share the same memory, including port conflicts between the forward and backward access patterns and read-after-write dependencies in floating-point potential accumulation. Experimental evaluations on MNIST and five multivariate time-series classification datasets demonstrate that, with fully on-chip training, our SNN model achieves accuracy comparable to SNN or LSTM trained with BPTT. Compared to CPU and GPU based implementations, our FPGA design delivers 2.5 to 6 times higher throughput while consuming an order of magnitude less energy per sample.

Chinese Translation

脉冲神经网络（SNNs）在边缘设备上进行能效高的时空模式识别方面具有良好前景，但高效的部署和设备内适应需要专用硬件。在本研究中，我们提出了一种用于时空模式分类的多层SNN的FPGA实现，其中推理和训练均在同一硬件上进行。我们的设计采用了一种仅进行时间前向的训练算法，避免了与传统的时间反向传播（Backpropagation Through Time, BPTT）相关的内存和延迟开销。一种混合精度方案在训练稳定性与资源效率之间取得平衡，架构优化最大化了硬件和内存的利用率。我们进一步识别并解决了推理和学习共享同一内存时出现的风险，包括前向和后向访问模式之间的端口冲突以及浮点潜在累积中的读后写依赖。对MNIST和五个多变量时间序列分类数据集的实验评估表明，经过完全片上训练后，我们的SNN模型的准确性与使用BPTT训练的SNN或LSTM相当。与基于CPU和GPU的实现相比，我们的FPGA设计在每个样本消耗的能量减少一个数量级的同时，提供了2.5到6倍更高的吞吐量。

View on arXiv Download PDF AI Translation

cs.AI / 262 / 2610.17473

REA: From Correct LLM Judgments to Safe Decision Delivery

REA：从正确的 LLM 判断到安全的决策交付

Meng, Jingxiang, Liu, Aofan, Ma, Yaoshuai, Liu, Fangxin, Chen, Yongbiao

Abstract

Generative safety guards must classify requests correctly and return outputs that applications can parse. Reasoning guards can improve classification accuracy while producing outputs that applications cannot parse. When applications cannot extract a correct prediction, fallback handling can allow harmful requests or block safe ones. We introduce Resolve--Express--Act (REA), a framework that traces judgments through parsing and fallback to application actions. Experiments on classifiers and safety guards show that correct predictions can appear in incorrectly formatted outputs. Parsing the safety verdict preserves the reasoning guard's accuracy advantage, while stricter explanation requirements reject additional correct predictions. Accepting formatting variations recovers most rejected correct predictions. Held-out tests confirm that output requirements and generation limits affect whether correct predictions can be parsed. In our fail-open replay, matching the parser to the output and fixing formatting reduced the fraction of harmful requests allowed despite correct guard judgments from 95.5% to 0%. A fixed-output comparison shows that model-based extraction can preserve guard verdicts, but is not automatically reliable; deterministic normalization already repairs the observed native-format failures. Fallback replay shows that the same correct prediction can lead to different safety actions depending on the parser and fallback policy, including allowing requests the guard correctly identifies as harmful.

Chinese Translation

生成安全防护必须正确分类请求并返回应用程序可以解析的输出。推理防护可以提高分类准确性，但产生的输出可能无法被应用程序解析。当应用程序无法提取正确预测时，后备处理可能会允许有害请求或阻止安全请求。我们提出了 Resolve--Express--Act (REA) 框架，该框架追踪判断通过解析和后备到应用程序的动作。对分类器和安全防护的实验表明，正确的预测可能出现在格式不正确的输出中。解析安全裁决保留了推理防护的准确性优势，而更严格的解释要求则拒绝了额外的正确预测。接受格式变体可以恢复大部分被拒绝的正确预测。保留测试确认输出要求和生成限制影响正确预测是否可以被解析。在我们的开放失败重放中，将解析器与输出匹配并修复格式将尽管有正确的防护判断，允许的有害请求的比例从 95.5% 降低到 0%。固定输出比较显示，基于模型的提取可以保留防护裁决，但并不自动可靠；确定性规范化已经修复了观察到的本地格式失败。后备重放显示，相同的正确预测可能导致不同的安全行动，具体取决于解析器和后备策略，包括允许防护正确识别为有害的请求。

View on arXiv Download PDF AI Translation

cs.AI / 263 / 2610.17540

TacitVLA: Reasoning Vision-Language-Action Models Without Reasoning Latency

TacitVLA：无推理延迟的视觉-语言-行动模型推理

Lu, Jack, You, Yurong, Li, Boyi, Cao, Yulong, Tian, Thomas, Xiao, Chaowei, Chen, Xiangyu, Luo, Wenjie, Ivanovic, Boris, Pavone, Marco, Wang, Yan

Abstract

Textual reasoning before acting has improved vision-language-action (VLA) models, but at the cost of latency: every action conditioned on the reasoning text must wait for autoregressive decoding. We show that actions do not need to wait for or condition on the generated reasoning to benefit from it, because the reasoning supervision already improves the backbone representation. We introduce TacitVLA, a reasoning VLA whose action expert never attends to the reasoning it generates for the current call. On long-horizon tasks where the reasoning text also serves as episodic memory, it can generate each entry during action execution rather than before acting. We evaluate it in a controlled study on four benchmarks spanning manipulation and autonomous driving, memory and non-memory tasks: RoboMME, RMBench, LIBERO, and PhysicalAI-AV. On all four, TacitVLA matches the performance of its explicit-reasoning counterpart at up to 3.4x lower latency, and matches the latency of the same architecture trained without reasoning at up to +17.1% success. In absolute terms, it surpasses the best previously published results on the RoboMME Counting suite (82.7% vs. 66.8%) and RMBench (77.5% vs. 56.5%), and reaches 72.9% on LIBERO-Plus and a 52.5% scene score on the public split of the PhysicalAI-AV closed-loop driving challenge. Project page: https://tacitvla.github.io.

Chinese Translation

在行动之前进行文本推理改善了视觉-语言-行动（VLA）模型，但代价是延迟：每个基于推理文本的行动必须等待自回归解码。我们表明，行动不需要等待或依赖于生成的推理就能从中受益，因为推理监督已经改善了基础表示。我们提出了TacitVLA，这是一种推理VLA，其行动专家在当前调用中从不关注其生成的推理。在长时间跨度的任务中，推理文本也作为情节记忆，它可以在行动执行期间生成每个条目，而不是在行动之前。我们在四个基准上进行了受控研究，涵盖了操作和自主驾驶、记忆和非记忆任务：RoboMME、RMBench、LIBERO和PhysicalAI-AV。在所有四个基准上，TacitVLA的性能与其显式推理对应物相匹配，延迟降低至3.4倍，同时在没有推理的情况下训练的相同架构的延迟匹配成功率提高至17.1%。在绝对值上，它超越了之前在RoboMME Counting套件上发布的最佳结果（82.7%对66.8%）和RMBench（77.5%对56.5%），在LIBERO-Plus上达到了72.9%，并在PhysicalAI-AV闭环驾驶挑战的公共拆分中达到了52.5%的场景得分。项目页面：https://tacitvla.github.io。

View on arXiv Download PDF AI Translation

cs.AI / 264 / 2610.17549

Reference Wording Shapes Retrieval of Robot Task Instructions

参考措辞影响机器人任务指令的检索

Yelamanchi, Jahnavi

Abstract

Identifying a robot task from a short instruction is sensitive to which descriptions are available for retrieval. We investigate this effect in the natural-language instructions included with the RLBench manipulation benchmark. We extract 296 fixed descriptions from 62 publicly available task files and compare five training-free lexical rankers over 80 random choices of one indexed description per task. Exact strings assigned to conflicting task labels are excluded from the primary score. Character n-gram TF-IDF achieves 61.3% mean top-1 accuracy on the resulting held-out instructions, compared with 55.6% for BM25 and 54.9% for word TF-IDF. Across gallery choices, character TF-IDF ranges from 55.2% to 66.8%; its 5th-95th percentile range is 56.9%-64.8%. Adding a second indexed description improves character TF-IDF by 7.0 percentage points on matched queries. These results quantify the influence of reference wording and representation choice on instruction-to-task retrieval under a consistent evaluation protocol.

Chinese Translation

从简短指令中识别机器人任务对可用于检索的描述非常敏感。我们在 RLBench 操作基准中研究了这一影响，该基准包含自然语言指令。我们从 62 个公开可用的任务文件中提取了 296 个固定描述，并在每个任务的 80 个随机选择的索引描述上比较了五种无训练的词汇排名方法。与存在冲突的任务标签分配的确切字符串被排除在主要得分之外。字符 n-gram TF-IDF 在结果的保留指令上实现了 61.3% 的平均 top-1 准确率，而 BM25 为 55.6%，词 TF-IDF 为 54.9%。在图库选择中，字符 TF-IDF 的范围从 55.2% 到 66.8%；其第 5 到第 95 百分位数范围为 56.9%-64.8%。在匹配查询中，添加第二个索引描述使字符 TF-IDF 提高了 7.0 个百分点。这些结果量化了参考措辞和表示选择对在一致评估协议下的指令到任务检索的影响。

View on arXiv Download PDF AI Translation

cs.AI / 265 / 2610.17584

RAN-JEPA: An Action-Conditioned World Model for Energy-efficient Mobile Network Control

RAN-JEPA：一种用于节能移动网络控制的动作条件世界模型

Wu, Binghan, Zeng, Kai, Yang, Kai, Zhang, Ya-Qin, Sifakis, Joseph, Ouyang, Ye

Abstract

Operational mobile networks are rich in observations but poor in interventions: they continuously generate large volumes of passive telemetry, while controlled action--response trajectories remain comparatively scarce and costly to collect. This observation--intervention asymmetry is a central obstacle to building practical world models for RAN control. We address this problem with RAN-JEPA, an action-conditioned world model that separates representation learning from intervention learning. The model first learns a predictive latent representation from observation-only network data, adapts that representation to the target deployment domain, and then learns action-conditioned latent dynamics from controlled sleep/wake trajectories while keeping the encoder fixed. We evaluate RAN-JEPA using public operator performance data for pretraining and a real-equipment four-cell 5G testbed for domain adaptation, post-training, and control. Across four future 15-minute intervals, RAN-JEPA achieves lower action-conditioned prediction error than a GPT-2 world-model baseline and a mechanistic rule, while remaining close to a strong persistence reference. What-if evaluation further shows that the model produces distinct and coherent future trajectories under matched SLEEP and KEEP plans. In closed-loop control, RAN-JEPA consistently supports higher sleep ratios than the two baselines at comparable congestion cost. These results show that observation-first, action-conditioned learning can turn abundant passive telemetry and limited intervention data into a practical predictive model for energy-aware RAN control.

Chinese Translation

运营中的移动网络观察丰富但干预稀缺：它们持续生成大量被动遥测数据，而受控的动作-响应轨迹相对稀少且收集成本高昂。这种观察-干预不对称性是构建实际RAN控制世界模型的主要障碍。我们通过RAN-JEPA解决了这一问题，RAN-JEPA是一种动作条件世界模型，它将表示学习与干预学习分开。该模型首先从仅包含观察的网络数据中学习预测性潜在表示，将该表示适应于目标部署领域，然后在保持编码器固定的情况下，从受控的睡眠/唤醒轨迹中学习动作条件潜在动态。我们使用公共运营商性能数据进行预训练，并在真实设备的四个小区5G测试平台上进行领域适应、后训练和控制来评估RAN-JEPA。在未来四个15分钟的时间间隔内，RAN-JEPA的动作条件预测误差低于GPT-2世界模型基线和机械规则，同时仍然接近强持久性参考。假设评估进一步表明，该模型在匹配的SLEEP和KEEP计划下产生了不同且一致的未来轨迹。在闭环控制中，RAN-JEPA始终支持比两个基线更高的睡眠比例，同时保持相似的拥塞成本。这些结果表明，观察优先、动作条件学习可以将丰富的被动遥测和有限的干预数据转化为用于节能RAN控制的实用预测模型。

View on arXiv Download PDF AI Translation

cs.AI / 266 / 2610.17609

Verifier-Harness Co-Evolution for Training-Free Self-Improvement on Open-Ended Tasks

无训练自我改进的验证者-工具共同进化用于开放式任务

Xu, Tianyi, Zhang, Yaolun, Wu, Yiran, Nguyen, Joe, Dai, Shengyu, Moeinizade, Saba, Shao, Zhenwen, Wu, Qingyun, Wang, Huazheng

Abstract

Training-free self-improvement lets an agent use feedback to revise the response it is about to return and to update the harness state that shapes how it handles later tasks. On open-ended tasks that feedback rarely comes from reference answers or executable tests, so the agent has to rely on a model-based verifier. When the verifier is wrong, the current response can get worse, and misleading experience can persist across the task stream. A verifier that never changes also grows stale as the agent's responses improve and task demands shift. We introduce Unified Co-Evolution (UCoE), a training-free method that evolves an executable verifier program together with the agent's harness state while all model parameters stay fixed. UCoE synthesizes candidate verifier programs and tests them on response contrasts, pairs of responses that differ in one declared property and have a known preferred side. It selects the candidate whose discrimination on its weakest contrast family is highest, a quantity we call driving capacity. The selected verifier diagnoses one weakness in each draft, guides a single repair, decides which response is served, and determines whether the accepted repair may enter the harness state. As tasks complete, their diagnostics seed new candidate verifiers, which must beat the current verifier on the same contrasts before adoption. Across six open-ended benchmarks, UCoE averages 59.00% preference against matched direct generation with Qwen3-8B and 57.19% with Qwen3.8-27B, margins of 9.00 and 7.19 percentage points. On three verifiable benchmarks, the same method raises the mean from 36.06% for direct generation to 38.48%. Code is available at https://github.com/tianyi0216/Unified-Co-Evolution.

Chinese Translation

无训练自我改进使得代理能够利用反馈修正其即将返回的响应，并更新塑造其处理后续任务方式的工具状态。在开放式任务中，反馈往往不是来自参考答案或可执行测试，因此代理必须依赖基于模型的验证者。当验证者出错时，当前的响应可能变得更糟，误导性的经验可能在任务流中持续存在。一个从不改变的验证者也会随着代理的响应改进和任务需求的变化而变得过时。我们提出了统一共同进化（Unified Co-Evolution, UCoE），这是一种无训练的方法，它在所有模型参数保持固定的情况下，与代理的工具状态共同进化一个可执行的验证者程序。UCoE合成候选验证者程序，并在响应对比上进行测试，这些对比是指在一个声明属性上不同且具有已知优选侧的响应对。它选择在其最弱对比家族上区分能力最高的候选者，我们称之为驱动能力（driving capacity）。所选验证者诊断每个草稿中的一个弱点，指导单一修复，决定提供哪个响应，并确定接受的修复是否可以进入工具状态。随着任务的完成，其诊断结果为新的候选验证者提供种子，这些候选者必须在相同对比上超越当前验证者才能被采纳。在六个开放式基准测试中，UCoE在与Qwen3-8B的直接生成匹配中平均获得59.00%的偏好，与Qwen3.8-27B的偏好为57.19%，分别相差9.00和7.19个百分点。在三个可验证的基准测试中，采用相同的方法将直接生成的平均值从36.06%提高到38.48%。代码可在 https://github.com/tianyi0216/Unified-Co-Evolution 获取。

View on arXiv Download PDF AI Translation

cs.AI / 267 / 2610.17610

OpenDepth: Repurposing Image Editing Priors for Robust Open-World Monocular Depth Estimation

OpenDepth：重新利用图像编辑先验以实现稳健的开放世界单目深度估计

Zhang, Mengke, Yang, Tianyi, Li, Xinzhuo, Ming, Anlong

Abstract

Recently, the integration of generative priors has revolutionized Monocular Depth Estimation (MDE), achieving state-of-the-art (SOTA) performance with remarkably high data efficiency. However, although methods based on Text-to-Image (T2I) models excel in general scenes, their efficacy degrades sharply when confronted with open-world environmental perturbations, such as illumination variations and adverse weather. Since T2I priors are optimized for semantic synthesis rather than preserving pixel-level structure, they struggle to decouple geometry from appearance under severe visual shifts, easily triggering structural hallucinations. Therefore, our core insight is to model open-world depth estimation as consistent geometric inference across multi-style images, fully leveraging the powerful structure-preserving editing priors of Image-to-Image (I2I) models. Based on this, we propose OpenDepth, a robust open-world MDE framework that introduces "Intra-Domain Awareness" -- the capability to map diverse stylistic inputs sharing identical geometry into a unified latent space. To achieve this, we first align shallow features within the Diffusion Transformer (DiT) to anchor degraded observations to unified geometric representations. Additionally, we introduce a joint RGB latent reconstruction task, requiring the model to recover clean RGB information from degraded inputs. This multi-stage strategy effectively decouples environmental style interference, compelling the model to focus strictly on underlying geometric recovery. Extensive evaluations show that OpenDepth sets new SOTA records for robust zero-shot depth estimation across standard and extreme weather benchmarks, demonstrating exceptional structural integrity and adaptability.

Chinese Translation

近年来，生成先验的整合彻底改变了单目深度估计（MDE），在数据效率极高的情况下实现了最先进（SOTA）的性能。然而，尽管基于文本到图像（T2I）模型的方法在一般场景中表现出色，但在面对开放世界环境扰动（如光照变化和恶劣天气）时，其有效性急剧下降。由于T2I先验是针对语义合成而优化的，而非保持像素级结构，它们在严重视觉变化下难以将几何与外观解耦，容易触发结构幻觉。因此，我们的核心见解是将开放世界深度估计建模为跨多样式图像的一致几何推断，充分利用图像到图像（I2I）模型强大的结构保持编辑先验。在此基础上，我们提出了OpenDepth，一个稳健的开放世界MDE框架，引入了“域内意识”（Intra-Domain Awareness）——将共享相同几何的多样式输入映射到统一潜在空间的能力。为此，我们首先在扩散变换器（Diffusion Transformer, DiT）中对齐浅层特征，以将降级观测锚定到统一的几何表示。此外，我们引入了一个联合RGB潜在重建任务，要求模型从降级输入中恢复干净的RGB信息。这一多阶段策略有效地解耦了环境风格干扰，迫使模型严格关注基础几何的恢复。广泛的评估表明，OpenDepth在标准和极端天气基准上设定了稳健的零-shot深度估计的新SOTA记录，展示了卓越的结构完整性和适应性。

View on arXiv Download PDF AI Translation

cs.AI / 268 / 2610.17613

MAS-Gate: Unlocking Multi-Agent System Lifecycle via Node-Centric Quantification

MAS-Gate：通过节点中心量化解锁多智能体系统生命周期

Lin, Hehai, Wang, Zixuan, Yan, Ming, Wang, Peidong, Cao, Shilei, Li, Minzhi, Ding, Bosheng, Zheng, Juepeng, Qin, Chengwei

Abstract

Large language model (LLM)-based multi-agent systems (MAS) can solve complex problems beyond the reach of a single agent by decomposing tasks and coordinating complementary capabilities. Despite extensive research on MAS, existing work largely focuses on specific aspects or individual phases of their development and operation, resulting in fragmented perspectives with objectives and signals that are difficult to reconcile across these different settings. To bridge these disconnected perspectives, we conceptualize MAS development and operation as a three-phase lifecycle spanning design-time orchestration before execution, runtime intervention during execution, and post-hoc attribution after execution, and introduce MAS-Gate, a unified node-centric quantification applicable across these phases. Modeling each execution as a directed acyclic graph, MAS-Gate combines within-node utility, which estimates a node's contribution through changes in reference-answer likelihood, with inter-node strength, which approximates semantic uptake and potential downstream reach. We adapt this shared quantification to three objectives: providing a structure-aware dense reward for Orchestrator training, training an Editor to act on execution prefixes using paired rewards from completed edited and unedited trajectories, and supplying risk-ranked nodes as auxiliary evidence for LLM-based post-hoc attribution. Experiments spanning interaction environments, knowledge-intensive multi-hop question answering, general reasoning, and post-hoc attribution demonstrate that MAS-Gate improves workflow construction, targeted runtime editing, and LLM-assisted error localization over corresponding phase-specific baselines. Together, these results establish MAS-Gate as a unified structural interface for improving MAS generation, dynamic editing, and error attribution throughout the lifecycle.

Chinese Translation

基于大型语言模型（LLM）的多智能体系统（MAS）能够通过任务分解和协调互补能力来解决超出单个智能体能力范围的复杂问题。尽管对MAS的研究广泛，但现有工作主要集中在其开发和操作的特定方面或单个阶段，导致不同设置之间的目标和信号难以协调，从而形成了碎片化的视角。为了弥合这些不连贯的视角，我们将MAS的开发和操作概念化为一个涵盖设计时编排、执行时干预和执行后归因的三阶段生命周期，并引入MAS-Gate，这是一种适用于这些阶段的统一节点中心量化方法。将每次执行建模为有向无环图，MAS-Gate结合了节点内效用（通过参考答案的可能性变化估计节点的贡献）和节点间强度（近似语义吸收和潜在下游影响）。我们将这种共享量化适应于三个目标：为编排者训练提供结构感知的密集奖励，训练编辑器使用来自已完成编辑和未编辑轨迹的配对奖励对执行前缀进行操作，以及提供风险排名节点作为基于LLM的后期归因的辅助证据。跨互动环境、知识密集型多跳问答、一般推理和后期归因的实验表明，MAS-Gate在工作流构建、目标运行时编辑和LLM辅助错误定位方面优于相应的阶段特定基线。综合来看，这些结果确立了MAS-Gate作为一个统一的结构接口，以改善MAS生成、动态编辑和错误归因的整个生命周期。

View on arXiv Download PDF AI Translation

cs.AI / 269 / 2610.17633

Executable Prefix Values for Locating Agent Failures

用于定位智能体故障的可执行前缀值

Shen, Bingyu, Tang, Jiaqi, Wei, Lan, Sheng, Yi, Li, J. Jenny, Li, Boyang

Abstract

Work on debugging LLM agents treats the "first mistake" marked by a human or LLM judge as ground truth for where a trajectory failed. We ask whether such labels point at causes or symptoms, and make the distinction executable: we restart the agent from reconstructed intermediate states of a failed trajectory and estimate its prefix value V(h_t), the probability of success from each step, with a certified point of no return (PNR) for a failure. On 150 SWE-bench Verified tasks with two frozen open-weight agents, we build such curves for 47 and 60 failed trajectories and probe every step labeled by four frontier judges under the Traverse rubric. We find that the labels do not mark what a controller needs. Judges of different lineages agree on the exact step in only 21/57 cases, and the newest judges flag mistakes in 8/10 and 10/10 successful runs. For 66% of the labels, the agent's success probability does not drop at the labeled step; only 7/188 answers name the decisive step. Restarting at the measured restart point raises expected success from 0.44 to 0.57 offline, and real restarts confirm the advantage, whereas restarting at a judge's label does not (-0.08 to -0.03). We release our code, curves, and labels.

Chinese Translation

对大型语言模型（LLM）智能体调试的研究将人类或LLM评审标记的“第一次错误”视为轨迹失败的真实依据。我们探讨这些标签是否指向原因或症状，并使这种区分可执行：我们从失败轨迹的重构中间状态重新启动智能体，并估计其前缀值 V(h_t)，即从每一步成功的概率，并为失败提供一个认证的不可回退点（PNR）。在150个SWE-bench验证任务中，使用两个冻结的开放权重智能体，我们为47条和60条失败轨迹构建了这样的曲线，并在Traverse框架下探讨了四位前沿评审标记的每一步。我们发现这些标签并不标记控制器所需的内容。不同来源的评审在57个案例中仅在21个案例上达成一致，最新的评审在10个成功运行中标记了8个和10个错误。对于66%的标签，智能体的成功概率在标记步骤并没有下降；只有7/188个答案指出了决定性步骤。在测量的重启点重新启动将离线期望成功率从0.44提高到0.57，而实际重启确认了这一优势，而在评审标签处重新启动则没有效果（-0.08到-0.03）。我们发布了我们的代码、曲线和标签。

View on arXiv Download PDF AI Translation

cs.AI / 270 / 2610.17641

Beyond Bias Scores: Answerability Confounds in Activation Steering with Gradient-Constrained Direction Design (GCD)

超越偏见分数：激活引导中的可回答性混淆与梯度约束方向设计（GCD）

Li, Weitong, Song, Donglin, Lin, Junkai, Xuan, Xi

Abstract

Bias scores in language-model evaluations often conflate what a model prefers with whether it answers at all. We identify this confound in \textsc{BBQ}, whose ambiguous score factorizes into answerability and signed preference. Across 11 categories, 28,}824 items, and five models, contrastive steering lowers the signed ambiguous score in 11/11 categories but lowers the signed disambiguated preference score in only 8/11. Across 99 configurations, the answerability channel exceeds half of the observed score change in 17\% and accounts for the entire change in 11%. In a six-category disambiguated evaluation, 20.8\% of 1,500 items flip to the unknown option, where abstention is incorrect; \textsc{StereoSet} shows no analogous answer-rate shift. The effect reflects selection and near-orthogonality between the contrastive and preference-gradient directions. We introduce Gradient-Constrained Direction Design (\textsc{GCD}), an option-gradient construction with a first-order unknown-logit constraint. Under a matched Qwen-7B protocol, GCD reduces answer-rate distortion 14-fold and yields a 4-fold larger preference change, while ablations show similar gains for alternative and unconstrained option-gradient directions; nulling the unknown gradient alone is not responsible for the gain. We recommend reporting answerability alongside published scores.

Chinese Translation

语言模型评估中的偏见分数常常将模型的偏好与其是否回答混为一谈。我们在 extsc{BBQ} 中识别了这种混淆，其模糊分数可以分解为可回答性和带符号的偏好。在11个类别、28,824个项目和五个模型中，对比引导在11/11个类别中降低了带符号的模糊分数，但仅在8/11个类别中降低了带符号的消歧偏好分数。在99种配置中，可回答性通道在17%的情况下超过观察到的分数变化的一半，并在11%的情况下占据了全部变化。在一个六类别的消歧评估中，1,500个项目中有20.8%转向未知选项，其中弃权是不正确的； extsc{StereoSet} 显示没有类似的回答率变化。该效应反映了对比方向与偏好梯度方向之间的选择性和近正交性。我们引入了梯度约束方向设计（ extsc{GCD}），这是一种具有一阶未知-logit约束的选项梯度构造。在匹配的 Qwen-7B 协议下，GCD 将回答率失真降低了14倍，并产生了4倍更大的偏好变化，而消融实验显示替代和不受约束的选项梯度方向也有类似的增益；单独消除未知梯度并不是增益的原因。我们建议在发布分数时同时报告可回答性。

View on arXiv Download PDF AI Translation

cs.AI / 271 / 2610.17643

When Should Enterprise Policy Decisions Be Learned, and When Should They Be Reasoned?

企业政策决策何时应通过学习获得，何时应通过推理得出？

Han, Hyojung, Kang, Ryangkyung, Kim, Youngeun, Jeon, Seung-Hun

Abstract

An enterprise agent that answers under a written policy must choose an action before it says anything: answer, ask for a missing fact, ask for confirmation, call one or several tools, escalate, or refuse. Holding one 27B Korean enterprise model fixed, we vary whether the decision is a generated JSON object or a single constrained token, and whether the model may reason first. On 2800 in-distribution items we did not detect an accuracy difference between the two deployed paths (93.46% vs 94.14%), while they differ about tenfold in restart stability (3.00% of actions flip vs 0.29%); they also differ in reasoning, so this is an operational comparison. On a sealed benchmark of 300 items whose labels come from a hidden executable policy, reasoning is worth +17.3 points [+12.3, +23.0] on exact action. With reasoning off, the format difference was 0.0 points [-4.3, +4.0], a failure to detect an advantage for either interface, not equivalence; with reasoning on it is unresolved. Against a token-matched random control, boundary-targeted supervision moves the two covered boundaries by +17.5 and +20.0 points and leaves a held-out boundary unchanged. In a 3-seed training x reasoning factorial, the sealed-rendering interaction is -4.9 [-8.1, -1.7], consistent with partial substitution. All supervision was Korean. On independently rendered English versions, the model is 0.82 points lower [-2.68, +0.93] with reasoning on, the same order as re-rendering in Korean; the reasoning-off gain from targeted supervision is not stable in either language, so we do not claim transfer. The original text stays private as the confirmatory set. An open-model re-rendering made after these results reproduced the reasoning effect and the English verdict; the interaction kept its sign but did not replicate under item-clustered intervals. We release the model and that re-rendering with gold labels.

Chinese Translation

在书面政策下作答的企业代理必须在发言之前选择一个行动：回答、请求缺失的事实、请求确认、调用一个或多个工具、升级或拒绝。在固定一个27B的韩国企业模型的情况下，我们变更决策是生成的JSON对象还是单一的受限标记，以及模型是否可以先进行推理。在2800个分布内项目中，我们未检测到两条部署路径之间的准确性差异（93.46%对94.14%），而在重启稳定性方面，它们的差异约为十倍（3.00%的行动翻转对0.29%）；它们在推理上也存在差异，因此这是一个操作性比较。在一个由隐藏可执行政策提供标签的300个项目的封闭基准上，推理在精确行动上价值+17.3分 [+12.3, +23.0]。在推理关闭的情况下，格式差异为0.0分 [-4.3, +4.0]，未能检测到任何接口的优势，而不是等效；在推理开启时则未解决。与标记匹配的随机控制相比，边界目标监督将两个覆盖的边界分别移动了+17.5和+20.0分，并保持了一个保留边界不变。在一个3种种子训练与推理的因子实验中，封闭渲染交互为-4.9 [-8.1, -1.7]，与部分替代一致。所有监督均为韩语。在独立渲染的英文版本中，模型在推理开启时低0.82分 [-2.68, +0.93]，与韩语的重新渲染相同；在任何语言中，推理关闭时从目标监督获得的增益并不稳定，因此我们不主张转移。原始文本保持私密，作为确认集。在这些结果之后进行的开放模型重新渲染再现了推理效果和英文判决；交互保持其符号但在项目聚类间隔下未能复制。我们发布该模型及其带有金标签的重新渲染。

View on arXiv Download PDF AI Translation

cs.AI / 272 / 2610.17659

Can Agents Truly Forget? Explore Machine Unlearning for LLM Agents

代理是否真的能够遗忘？探索大规模语言模型代理的机器遗忘

Lin, Junkai, Xu, Ge, Xu, Zheng, Chen, Junkai, Lan, Xiangyuan

Abstract

Large language model (LLM) agents have shown strong capabilities in interactive problem solving, tool use, and long-horizon task execution. However, they may also acquire sensitive or private knowledge during training and expose it during deployment. Retraining a model from scratch to remove such knowledge is prohibitively costly, making machine unlearning a more practical approach. Machine unlearning aims to remove targeted undesirable knowledge from trained models without retraining from scratch while preserving general utility. When language models are deployed as agents, parameter-level unlearning becomes insufficient: forgotten information may still persist in memory artifacts or be reacquired via tool calls. In this paper, we formulate the problem of agent unlearning as removing private information from an LLM-based agent across three information channels: parametric knowledge, persistent memory, and tool-call acquisition. To study this problem, we introduce we introduce Agent Unlearning Benchmark (AUBench), a controlled benchmark that extends the fictitious-author setting of TOFU into an agent environment with persistent memory, an offline web snapshot, tool-accessible documents, and provenance-tracked interaction trajectories. By evaluating existing parameter-level and memory-level unlearning methods in this setting, AUBench reveals the gap between model-level forgetting and system-level agent unlearning, and provides a foundation for developing unlearning methods that operate across the full lifecycle of LLM agents.

Chinese Translation

大规模语言模型（LLM）代理在互动问题解决、工具使用和长时间任务执行方面表现出强大的能力。然而，它们在训练过程中可能会获取敏感或私密的信息，并在部署时暴露这些信息。从头开始重新训练模型以去除这些知识成本高昂，因此机器遗忘成为一种更为实用的方法。机器遗忘旨在在不从头开始重新训练的情况下，从训练模型中去除目标不良知识，同时保持一般效用。当语言模型作为代理部署时，参数级的遗忘变得不足：被遗忘的信息可能仍然存在于记忆遗物中，或通过工具调用重新获取。在本文中，我们将代理遗忘的问题表述为从基于LLM的代理中去除私密信息，涉及三个信息通道：参数知识、持久记忆和工具调用获取。为了研究这个问题，我们引入了代理遗忘基准（Agent Unlearning Benchmark，AUBench），这是一个受控基准，将TOFU的虚构作者设置扩展到具有持久记忆的代理环境中，包括离线网页快照、可访问的工具文档和追踪来源的交互轨迹。通过在这一环境中评估现有的参数级和记忆级遗忘方法，AUBench揭示了模型级遗忘与系统级代理遗忘之间的差距，并为开发在LLM代理的整个生命周期内运行的遗忘方法奠定了基础。

View on arXiv Download PDF AI Translation

cs.AI / 273 / 2610.17665

RoTA-Bench: Evaluating 3D Spatial Reasoning through Interactive Projection Alignment

RoTA-Bench：通过交互式投影对齐评估三维空间推理

Ma, Qingchuan, Ma, Yuexiao, Meng, Xiang, Zou, Yixuan, Xie, Tianyu, Chen, Wang, Zeng, Yuhui, Wu, Yuhang, Zheng, Xiawu, Ji, Rongrong

Abstract

Can vision-language models turn 2D observations into 3D understanding that supports precise action? This requires inferring object geometry, anticipating the effects of rotation, and refining actions through visual feedback. We introduce RoTA-Bench to evaluate these abilities through interactive 3D-to-2D projection alignment. Given a target silhouette, a model observes a rotated object and chooses successive rotations to recover the target projection within a limited action budget. Our automated pipeline constructs silhouette-preserving 3D objects from 2D icons and uses reversible initialization to guarantee solvability, yielding 674 tasks and a stratified 100-task test set. Empirically, we find that (i) a substantial human-model gap persists: GPT-5.6-Sol succeeds on only 17% of tasks, whereas 20 undergraduate participants collectively complete the test set four times with a 100% success rate; (ii) image-space and pose-space progress can diverge: at matched states, the model more often improves silhouette overlap while worsening pose alignment than humans (19.4% vs. 10.9%); and (iii) fine-grained control is sensitive to interaction history: removing it leaves 96.8% of the model's actions at a coarse 30-degree step size and sharply reduces observed task success. These findings highlight the challenge of converting visual evidence into precise 3D control, beyond recognizing or matching individual images.

Chinese Translation

视觉-语言模型能否将二维观察转化为支持精确行动的三维理解？这需要推断物体几何形状、预测旋转的效果，并通过视觉反馈来优化行动。我们引入RoTA-Bench来评估这些能力，通过交互式三维到二维的投影对齐。给定一个目标轮廓，模型观察一个旋转的物体，并选择连续的旋转以在有限的行动预算内恢复目标投影。我们的自动化流程从二维图标构建保持轮廓的三维物体，并使用可逆初始化来保证可解性，生成674个任务和一个分层的100任务测试集。实证结果显示：(i) 人类与模型之间存在显著差距：GPT-5.6-Sol仅在17%的任务上成功，而20名本科生参与者共同完成测试集四次，成功率达到100%；(ii) 图像空间和姿态空间的进展可能会出现分歧：在匹配状态下，模型在改善轮廓重叠的同时，姿态对齐的效果往往较差（19.4%对比10.9%）；(iii) 精细控制对交互历史敏感：去除交互历史后，模型的96.8%动作都停留在粗糙的30度步长，并显著降低了观察到的任务成功率。这些发现突显了将视觉证据转化为精确三维控制的挑战，超越了对单个图像的识别或匹配。

View on arXiv Download PDF AI Translation

cs.AI / 274 / 2610.17670

Cowork-Gym: Synthesizing Interactive Work Environments at Scale

Cowork-Gym：大规模合成互动工作环境

Zou, Anni, Chen, Cheng, Deng, Jiaji, Zhou, Weikang, Mai, Xinji, Wu, Yue, Lin, Jianing, Xie, Lipeng, Qin, Lixiong, Zhao, Qingyue, Shen, Que, Tao, Shuchang, Pan, Xuchen, Li, Yaliang, Zhai, Yunpeng, Chen, Yushuo, Wang, Zhihai, Zhang, Zhuo, Liu, Zhaoyang, Ding, Bolin, Zhou, Jingren

Abstract

Cowork agents navigate dependencies across tools and files while adapting to changing user requirements and edits to shared workspaces. Scaling such environments requires maintaining consistency among task objectives, executable resources, user behavior, and success criteria. We introduce Cowork-Gym, a framework that synthesizes Cowork environments through a build-execute-revise loop that uses execution evidence and review feedback to validate and repair generated environments. To capture realistic Cowork dynamics, a stateful user simulator responds to the agent's actions through dialogue and direct edits to the shared workspace. We synthesize 64.9K environments across 14 application scenarios and use them to post-train 9B, 27B, and 35B-A3B Qwen models with supervised fine-tuning, reinforcement learning, and on-policy distillation. Averaged across six agentic benchmarks, these models outperform their corresponding base checkpoints by 8.7, 6.4, and 4.1 points, respectively. Ours-27B ranks second overall among the evaluated models and achieves the highest average scores on PawBench and ClawEval, outperforming models with over 100B total parameters on both benchmarks.

Chinese Translation

Cowork 代理在工具和文件之间导航依赖关系，同时适应不断变化的用户需求和对共享工作空间的编辑。扩展此类环境需要在任务目标、可执行资源、用户行为和成功标准之间保持一致性。我们提出了 Cowork-Gym，一个通过构建-执行-修订循环合成 Cowork 环境的框架，该循环利用执行证据和审查反馈来验证和修复生成的环境。为了捕捉现实的 Cowork 动态，一个有状态的用户模拟器通过对话和对共享工作空间的直接编辑来响应代理的行动。我们在 14 个应用场景中合成了 64.9K 个环境，并利用这些环境对 9B、27B 和 35B-A3B Qwen 模型进行后训练，采用监督微调、强化学习和在线蒸馏。在六个代理基准测试中，这些模型的平均表现分别比其对应的基础检查点提高了 8.7、6.4 和 4.1 分。我们的 27B 模型在评估的模型中排名第二，并在 PawBench 和 ClawEval 上取得了最高的平均分，超越了在这两个基准测试中总参数超过 100B 的模型。

View on arXiv Download PDF AI Translation

cs.AI / 275 / 2610.17672

Encounter-Driven City-Scale Human Mobility Generation with Large Language Models

基于遭遇驱动的大规模城市人类移动生成与大型语言模型

Li, Chao, Jiang, Renhe, Xiao, Chuan, Wang, Jiawei

Abstract

Large language models (LLMs) have shown strong potential for human mobility generation by modeling individual routines, activities, and mobility preferences. However, existing approaches remain largely resident-centric: generating realistic individual trajectories does not necessarily recover the encounters that emerge when multiple residents co-occur at the same place and time. To this end, we study encounter-aware human mobility generation, which jointly preserves individual mobility and population-level interaction structure. We propose EncounterMob, an encounter-driven LLM framework consisting of a coarse-to-fine Encounter Generator and an Iterative Coupling Framework. The Encounter Generator first captures city-level encounter rhythms and then instantiates participant-time-venue encounters, while the Iterative Coupling Framework coordinates these encounters with personal mobility behavior through successive refinement. We further introduce an encounter recovery protocol covering encounter marginal and joint distributions over residents, times, and venues. Experiments on the Japan and NYC datasets show that EncounterMob outperforms all external baselines on all seven encounter metrics and all six personal mobility metrics. Compared with ELLMob, it reduces the participant-time-venue joint error by 40.4%/41.7% and the mean personal mobility error by 58.8%/32.7% on Japan/NYC, respectively.

Chinese Translation

大型语言模型（LLMs）在通过建模个体日常活动、行为和移动偏好方面展现了强大的潜力。然而，现有的方法仍然主要以居民为中心：生成现实的个体轨迹并不一定能够恢复多个居民在同一时间和地点共存时所产生的遭遇。为此，我们研究了考虑遭遇的人类移动生成，这种方法同时保留个体移动和人口层面的互动结构。我们提出了EncounterMob，一个基于遭遇驱动的LLM框架，包含一个粗到细的遭遇生成器和一个迭代耦合框架。遭遇生成器首先捕捉城市级的遭遇节奏，然后实例化参与者-时间-场所的遭遇，而迭代耦合框架通过连续的细化将这些遭遇与个人移动行为进行协调。我们进一步引入了一种遭遇恢复协议，涵盖居民、时间和场所的遭遇边际和联合分布。在日本和纽约市的数据集上的实验表明，EncounterMob在所有七个遭遇指标和六个个人移动指标上均优于所有外部基准。与ELLMob相比，其在日本/纽约市上分别减少了参与者-时间-场所联合误差40.4%/41.7%和平均个人移动误差58.8%/32.7%。

View on arXiv Download PDF AI Translation

cs.AI / 276 / 2610.17689

Stable Token-level Scheduling for Serverless LLM Inference

无服务器大语言模型推理的稳定令牌级调度

Desai, Hardi Ashishbhai, K, Balaji N, Addya, Sourav Kanti

Abstract

Serverless inference is emerging as an efficient deployment model for Large Language Models (LLMs) due to its ability to elastically scale resources, improve GPU utilization through resource sharing, and reduce infrastructure costs. However, existing approaches primarily optimize time to first token (TTFT) while overlooking inter-token latency (ITL), a key determinant of user experience during response generation. We observe that LLM inference comprises a computation-intensive prefill stage and a bandwidth-bound decode stage, where admitting new requests can disrupt ongoing decoding and increase latency. To address this, we propose Token Window Pacing (TWP), a closed-loop scheduling framework that leverages runtime feedback from ITL, queue depth, and workload characteristics to regulate request admission at the token level instead of the batch level. TWP dynamically adjusts a selfregulating token window and incorporates chunk-based prefill to minimize interference from incoming requests. Experiments on realistic serverless workloads show that TWP delivers stable TTFT, significantly reduces latency spikes, lowers P99 tail latency, minimizes ITL variation, and improves throughput by up to 45% over ServerlessLLM and 58% over conventional schedulers. These results demonstrate that feedback-driven tokenlevel scheduling enables stable, high-throughput serverless LLM inference without modifying the underlying model.

Chinese Translation

无服务器推理因其能够灵活扩展资源、通过资源共享提高GPU利用率以及降低基础设施成本，正在成为大语言模型（LLMs）的一种高效部署模型。然而，现有的方法主要优化首次令牌时间（TTFT），而忽视了令牌间延迟（ITL），这是响应生成过程中用户体验的一个关键决定因素。我们观察到，LLM推理包括一个计算密集的预填充阶段和一个带宽受限的解码阶段，在此过程中，接纳新请求可能会干扰正在进行的解码并增加延迟。为了解决这个问题，我们提出了令牌窗口节奏（Token Window Pacing, TWP），这是一个闭环调度框架，利用来自ITL、队列深度和工作负载特征的运行时反馈，在令牌级别而非批处理级别调节请求接纳。TWP动态调整自我调节的令牌窗口，并结合基于块的预填充，以最小化来自新请求的干扰。对现实无服务器工作负载的实验表明，TWP提供了稳定的TTFT，显著减少了延迟峰值，降低了P99尾部延迟，最小化了ITL的变化，并将吞吐量提高了高达45%（相较于ServerlessLLM）和58%（相较于传统调度器）。这些结果表明，基于反馈的令牌级调度能够实现稳定的高吞吐量无服务器LLM推理，而无需修改底层模型。

View on arXiv Download PDF AI Translation

cs.AI / 277 / 2610.17690

MedEvo: Medical Reasoning Evolution through Knowledge Navigation

MedEvo：通过知识导航实现医学推理的演进

Wang, Xiaoliang, Chen, Zhuo, Zheng, Chuanyang, Wang, Quanbin, Lao, Jiangwei, Liu, Yingao, Liu, Jiajia, Cao, Jinming, Lin, Honglin, Lu, Zhihong, Jin, Hao, Wang, Jian, Wu, Lijun

Abstract

Medical multimodal reasoning models are typically trained on teacher-generated rationales, discarding the valid reasoning in students' own failed trajectories, while general-domain correction methods offer little guidance for errors in image reading, anatomical context, or differential diagnosis. We propose MedEvo, which uses medical knowledge, namely a tag taxonomy and error-type diagnosis, to navigate both trajectory repair and data selection. For each failed rollout, a gold-blind auditor locates the first consequential error and provides a minimal, answer-free hint; the same student retains the valid prefix and regenerates the erroneous suffix. After each round, MedEvo identifies capability gaps across medical tags on a held-out feedback set and prioritizes related questions for new rollouts and repairs. With supervised fine-tuning (SFT) alone, MedEvo raises Qwen3.5-9B mean accuracy across seven benchmarks from 65.57% to 69.12% (47.40% to 54.50% on the multimodal track of MedXpertQA (MedXpertQA-MM)), surpassing teacher-distillation SFT (66.93%). It also raises Qwen3.5-35B and Qwen3.5-397B mean accuracy from 67.91% to 73.04% and from 71.83% to 74.95%, respectively. Our analyses show that failed trajectories contain reasoning worth preserving, that repair success does not imply learning value, and that independent feedback is key to sustained improvement.

Chinese Translation

医学多模态推理模型通常基于教师生成的推理进行训练，忽视了学生在失败轨迹中有效推理的价值，而通用领域的纠正方法对图像阅读、解剖背景或鉴别诊断中的错误提供的指导有限。我们提出了MedEvo，它利用医学知识，即标签分类法和错误类型诊断，来导航轨迹修复和数据选择。对于每个失败的推理过程，一个金盲审计员定位第一个重要错误，并提供一个最小的、无答案的提示；同一学生保留有效前缀并重新生成错误后缀。在每一轮之后，MedEvo在一个保留的反馈集上识别医学标签之间的能力差距，并优先考虑相关问题以进行新的推理和修复。仅通过监督微调（SFT），MedEvo将Qwen3.5-9B在七个基准测试中的平均准确率从65.57%提高到69.12%（在MedXpertQA的多模态轨道（MedXpertQA-MM）上从47.40%提高到54.50%），超越了教师蒸馏SFT（66.93%）。它还将Qwen3.5-35B和Qwen3.5-397B的平均准确率分别从67.91%提高到73.04%和从71.83%提高到74.95%。我们的分析表明，失败的轨迹中包含值得保留的推理，修复成功并不意味着学习价值，而独立反馈是持续改进的关键。

View on arXiv Download PDF AI Translation

cs.AI / 278 / 2610.17699

Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity

理解基因表达到图像学习：空间结构与信息保真度

Lian, Xiaoqing, Liu, Pei, Tan, Jiahui, Zou, Quan, Lin, Chen, Zeng, Xiangxiang

Abstract

Single-cell RNA sequencing (scRNA-seq) enables cell-level analysis of biological heterogeneity, but learning effective representations from high-dimensional, unordered gene expression profiles remains challenging. Gene Expression-to-Image (GE2I) offers an effective learning strategy by arranging gene values on pixel grids for convolutional neural networks, but its underlying mechanisms remain unclear. We study GE2I on full-transcriptome data while keeping the network and training procedure fixed. Our results reveal a trade-off between spatial smoothness and gene-level fidelity. Structured mappings produce smooth, low-frequency patterns that support faster early optimization, whereas assigning multiple genes to the same pixel obscures expression differences and can reduce predictive performance. This trade-off explains why random mappings can compete with structured layouts when many genes are represented. One-to-one gene-to-pixel assignment improves prediction across the tested visual backbones even when image smoothness decreases. We further find that the benefits of pretraining and augmentation depend on the backbone and mapping. These findings provide practical guidance for pixel allocation and the transfer of vision priors to gene expression data.

Chinese Translation

单细胞RNA测序（scRNA-seq）使得对生物异质性的细胞级分析成为可能，但从高维、无序的基因表达谱中学习有效的表示仍然具有挑战性。基因表达到图像（Gene Expression-to-Image, GE2I）通过将基因值排列在像素网格上，为卷积神经网络提供了一种有效的学习策略，但其潜在机制仍不清楚。我们在全转录组数据上研究GE2I，同时保持网络和训练过程不变。我们的结果揭示了空间平滑性与基因级保真度之间的权衡。结构化映射产生平滑的低频模式，支持更快的早期优化，而将多个基因分配给同一像素则模糊了表达差异，可能降低预测性能。这一权衡解释了为什么在许多基因被表示时，随机映射能够与结构化布局竞争。一对一的基因到像素分配在测试的视觉骨干网络中改善了预测，即使图像平滑性下降。我们进一步发现，预训练和数据增强的好处依赖于骨干网络和映射。这些发现为像素分配和视觉先验向基因表达数据的转移提供了实用指导。

View on arXiv Download PDF AI Translation

cs.AI / 279 / 2610.17702

A Node Embedding Framework for Model-Centric Conformance Checking in Process Mining

一种基于节点嵌入的模型中心一致性检查框架在过程挖掘中的应用

Abbasi, Mostafa, Najjaran, Homayoun

Abstract

Evaluating the quality of process models discovered from event logs is a fundamental challenge in process mining. Traditional evaluation relies on conformance metrics such as fitness and precision, which are highly sensitive to noise and deviations in event logs and therefore often fail to give a consistent assessment of discovery algorithms. We introduce a model-centric evaluation framework that combines node embeddings with Procrustes alignment to enable cross-graph comparability, since embeddings derived from different graphs reside in distinct vector spaces. The proposed PetriNetGraph method transforms Petri net process models into graph representations compatible with embedding algorithms. Comparing the node embeddings of ground-truth and discovered models through a parameterized cosine distance yields a unified and bounded metric for evaluating process discovery across datasets and methods. We evaluate the framework under static and context-aware scenarios to examine how prior knowledge of activity existence influences the assessment of structural similarity. The framework also integrates a compliance checking mechanism that localizes deviations at the activity level and an entropy-regularized node distance aggregation function that quantifies global deviations across the model. Experiments on nine publicly available event logs show that the approach provides a reliable embedding-based conformance indicator that reflects the behavior of fitness and precision. Experiments on synthetic logs with controlled noise levels confirm that it is substantially more robust to log deviations than traditional conformance metrics, PetriNet2Vec, and its dynamic-window and Procrustes-aligned variants, while remaining sensitive to process model degradation. All implementation details and datasets are publicly available.

Chinese Translation

评估从事件日志中发现的过程模型的质量是过程挖掘中的一个基本挑战。传统的评估依赖于一致性度量，如适合度（fitness）和精确度（precision），这些度量对事件日志中的噪声和偏差高度敏感，因此往往无法对发现算法提供一致的评估。我们提出了一种模型中心的评估框架，该框架结合了节点嵌入和Procrustes对齐，以实现跨图的可比性，因为来自不同图的嵌入位于不同的向量空间。所提出的PetriNetGraph方法将Petri网过程模型转换为与嵌入算法兼容的图表示。通过参数化的余弦距离比较真实模型和发现模型的节点嵌入，产生了一种统一且有界的度量，用于评估跨数据集和方法的过程发现。我们在静态和上下文感知场景下评估该框架，以检查活动存在的先验知识如何影响结构相似性的评估。该框架还集成了一种合规性检查机制，该机制在活动级别定位偏差，以及一种熵正则化的节点距离聚合函数，该函数量化模型中的全局偏差。在九个公开可用的事件日志上的实验表明，该方法提供了一种可靠的基于嵌入的一致性指标，反映了适合度和精确度的行为。在具有控制噪声水平的合成日志上的实验确认，与传统的一致性度量、PetriNet2Vec及其动态窗口和Procrustes对齐变体相比，该方法对日志偏差具有显著更强的鲁棒性，同时仍然对过程模型的退化敏感。所有实现细节和数据集均可公开获取。

View on arXiv Download PDF AI Translation

cs.AI / 280 / 2610.17710

Adaptive-Group Process Rewards for GUI Agents

用于图形用户界面代理的自适应组过程奖励

Shang, Zirui, Liu, Yang, Shu, Xin, Gao, Zhi, Wu, Xinxiao, Fan, Lifeng

Abstract

Training graphical user interface (GUI) agents with reinforcement learning suffers from sparse outcome rewards, leading to inefficient learning. Process rewards offer a promising direction for credit assignment to intermediate actions; however, accurately evaluating an action requires identifying evidence across long-horizon GUI trajectories, as its value may depend on information beyond its immediate state and observation. In this paper, we propose an adaptive-group process reward framework that dynamically partitions a GUI trajectory into semantically coherent subgoal groups, enabling process reward estimation at both the action and group levels. At the action level, a critic evaluates each action based on the visual changes and subsequent actions within the group for action correctness and contribution to the subgoal. At the group level, the framework aggregates action-level evaluations to assess subgoal completion and contribution to the overall task, and further refines these assessments by leveraging the context of all groups in the trajectory. The process reward of each action is obtained by combining its action-level evaluation with the refined group-level evaluation. We further augment GRPO's outcome advantage with the proposed process rewards to provide finer-grained guidance for GUI agent training. Experiments on process reward benchmarks and downstream RL training demonstrate the effectiveness of the proposed framework.

Chinese Translation

使用强化学习训练图形用户界面（GUI）代理时，稀疏的结果奖励导致学习效率低下。过程奖励为中间动作的信用分配提供了一个有前景的方向；然而，准确评估一个动作需要在长时间跨度的GUI轨迹中识别证据，因为其价值可能依赖于超出其即时状态和观察的信息。在本文中，我们提出了一种自适应组过程奖励框架，该框架动态地将GUI轨迹划分为语义一致的子目标组，从而实现对动作和组层面的过程奖励估计。在动作层面，评论者根据组内的视觉变化和后续动作评估每个动作的正确性及其对子目标的贡献。在组层面，该框架聚合动作层面的评估，以评估子目标的完成情况及其对整体任务的贡献，并通过利用轨迹中所有组的上下文进一步细化这些评估。每个动作的过程奖励是通过将其动作层面的评估与细化后的组层面评估相结合获得的。我们进一步增强了GRPO的结果优势，结合所提出的过程奖励，为GUI代理训练提供更细致的指导。在过程奖励基准和下游强化学习训练中的实验表明了所提框架的有效性。

View on arXiv Download PDF AI Translation

cs.AI / 281 / 2610.17718

ECG-RSI: Recursive Self-Improvement through Targeted Electrocardiogram Waveform Evolution

ECG-RSI：通过针对性心电图波形演化实现递归自我改进

Wang, Xiaoliang, Chen, Zhuo, Zheng, Chuanyang, Lao, Jiangwei, Wang, Quanbin, Lin, Honglin, Liu, Jiajia, Lu, Zhihong, Jin, Hao, Wang, Jian, Wu, Lijun

Abstract

Electrocardiogram (ECG) models need supervision that targets evolving weaknesses and remains faithful to waveform evidence. We introduce ECG-RSI, a recursive self-improvement framework combining verified waveform synthesis with feedback-guided intervention search. Signal remeasurement verifies waveform edits; construction errors guide generator repair, while learner errors motivate interventions compared on a separate selection set. Across seven benchmarks, three rounds raise Qwen3.5-35B-A3B's mean question-answering accuracy from 34.57% to 56.04% and mean classification F1 from 24.27 to 34.03. Across the retained trajectory, the preferred supervision shifts from direct reinforcement to confusion-focused contrasts. Continued feedback correction yields diminishing external gains. These results motivate separating error diagnosis from intervention selection and checking whether gains on the feedback set transfer to independent benchmarks.

Chinese Translation

心电图（ECG）模型需要针对不断演变的弱点进行监督，并保持对波形证据的忠实。我们提出了ECG-RSI，这是一种递归自我改进框架，将经过验证的波形合成与反馈引导的干预搜索相结合。信号重新测量验证波形编辑；构建错误指导生成器修复，而学习者错误则激励在单独选择集上进行比较的干预。在七个基准测试中，三轮迭代将Qwen3.5-35B-A3B的平均问答准确率从34.57%提高到56.04%，平均分类F1从24.27提高到34.03。在保留的轨迹中，首选的监督方式从直接强化转向以混淆为重点的对比。持续的反馈修正带来了递减的外部收益。这些结果促使我们将错误诊断与干预选择分开，并检查反馈集上的收益是否能够转移到独立基准上。

View on arXiv Download PDF AI Translation

cs.AI / 282 / 2610.17719

Keep Your Receipts: Trustworthy Cleanup of Agent-Built Knowledge Graphs

保留你的收据：可信赖的代理构建知识图谱的清理

Choudhary, Vishisht

Abstract

Agents build knowledge graphs as memory, and building them has become easy. Maintaining them has not. When a source is retracted, overruled, or erased, everything built on top of it (entity merges, entailed edges, community summaries) must be re-examined, and the maintainer must choose between deleting too much and too little. Deployed pipelines delete only the directly extracted facts: on our corpus that reduces contaminated queries from 32.2% to 28.0% under a real retraction and changes nothing under predicate revocation. Rebuilding is no repair for a constructor that samples: two builds of the same 300-document corpus differ in 29,475 artifacts, five times the 6,248 that removing the 61 retracted documents changes. PolyGraph instead asks the constructor for receipts: for every artifact, alternative bundles of source evidence, each claimed sufficient on its own, emitted in the call that creates it. Cleanup is then set arithmetic over the receipts, with no model calls. Because receipts are the model's own self-report, a re-derivation audit checks them by deleting the cited evidence and testing whether the constructor still produces the artifact. The audit finds that a receipt is trustworthy when it is shaped like the claim it backs. Merge receipts, each backing one narrow claim, agree with re-derivation 68% and 85% of the time. Summary receipts, elicited per sentence but stored as one flat list for the whole text, are wrong about two thirds of the time, almost always by keeping dead content. One receipt per sentence instead is 100% and 88% faithful and lets cleanup keep 33 of 35 touched summaries. The result holds under a second oracle model, a claim-blind prompt, ten further deletions, a second domain and a second constructor. Distrusting the summary layer cuts contaminated queries to 0.4% and 0.8% at 99.8 to 100% of the recall a perfect maintainer would return.

Chinese Translation

代理构建知识图谱作为记忆，构建它们变得容易，但维护却不然。当一个来源被撤回、推翻或删除时，基于该来源构建的所有内容（实体合并、推导边、社区摘要）都必须重新审查，维护者必须在删除过多和过少之间做出选择。已部署的管道仅删除直接提取的事实：在我们的语料库中，这在真实撤回的情况下将污染查询从32.2%减少到28.0%，而在谓词撤销的情况下则没有任何变化。重建并不能修复一个采样的构造器：对同一300文档语料库的两次构建在29,475个工件上存在差异，这一数字是删除61个撤回文档所改变的6,248的五倍。PolyGraph则要求构造器提供收据：对于每个工件，提供替代的源证据包，每个包在其自身的调用中声称足够。清理过程因此变成了对收据的集合运算，无需模型调用。由于收据是模型自身的自我报告，重新推导审计通过删除引用的证据并测试构造器是否仍能生成该工件来检查它们。当收据的形状与其支持的声明相符时，审计发现该收据是可信的。每个支持狭窄声明的合并收据与重新推导一致的比例为68%和85%。每句引出的摘要收据虽然存储为整个文本的一个平面列表，但其错误率接近三分之二，几乎总是因为保留了无效内容。每句一个收据的情况则分别为100%和88%的忠实度，并允许清理保留35个摘要中的33个。该结果在第二个预言模型、一个无声明提示、十次进一步删除、第二个领域和第二个构造器下依然成立。不信任摘要层将污染查询减少到0.4%和0.8%，而完美维护者的召回率为99.8%到100%。

View on arXiv Download PDF AI Translation

cs.AI / 283 / 2610.17723

Upstream Influence, Downstream Re-Anchoring in LLM Agent Chains

上游影响与下游重新锚定在大语言模型代理链中的作用

Hsu, Chun-Yu, Ku, Lun-Wei, Chen, Chung-Chi

Abstract

People act on evaluative judgments produced by LLM agents, and in multi-agent systems those judgments pass through several agents before reaching a person. We show that a workflow's information architecture---source access, handoff format, and which model reads the untrusted content---governs how much of a coordinated manipulation reaches the end. On 32 Taiwan~2025 recall cases, ten authentic Threads comments shown to one agent move its stance by $+0.171$ toward the injected position, and a mirrored set moves it the other way. Passing that judgment down a chain removes more than half of the shift by the third agent; keeping the comments in front of every agent sustains more of it. The reduction is not attrition in transit: each receiver discounts the inherited stance in proportion to how far it sits from its own reading of the evidence, returning to what that evidence implies for the case rather than to neutrality---consistent with partial re-anchoring. Handoff format is a second lever: a stance relayed without its reasoning is largely discounted. Susceptibility at the exposed seat varies fourfold across models, while the chain's arrangement does not matter at all. None of this is visible in the transcripts---the injected claims are almost never restated, so reasoning-trace audits miss the attack.

Chinese Translation

人们根据大语言模型（LLM）代理生成的评估判断进行行动，在多代理系统中，这些判断在到达个人之前会经过多个代理。我们展示了工作流程的信息架构——源访问、交接格式以及哪个模型读取不可信内容——决定了有多少协调操控能够到达最终用户。在32个台湾2025年召回案例中，展示给一个代理的十条真实Threads评论使其立场向注入位置移动了$+0.171$，而一组镜像评论则使其立场朝相反方向移动。将该判断传递给链中的下一个代理会使第三个代理的立场变化减少超过一半；而将评论保持在每个代理面前则能维持更多的变化。这一减少并非是传递过程中的损失：每个接收者会根据其与自身对证据的解读之间的距离来折扣继承的立场，回归到该证据对案例的含义，而非中立——这与部分重新锚定一致。交接格式是第二个杠杆：没有推理的立场传递会被大幅折扣。暴露位置的易感性在不同模型之间变化四倍，而链的排列方式则完全无关。所有这些在记录中并不可见——注入的主张几乎从未被重述，因此推理轨迹审计无法发现这一攻击。

View on arXiv Download PDF AI Translation

cs.AI / 284 / 2610.17724

Evidence-First Agentic Verification

以证据为先的自主验证

Sanayei, Reza, Dumitru, Razvan-Gabriel, Hou, Sunjie, Rezaei, MohammadHossein, Riaz, Haris, Vacareanu, Robert, Yang, Minglai, Liu, Ying, Surdeanu, Mihai, Zhang, Daniel Yue

Abstract

Long-horizon, semi-verifiable agentic tasks often rely on rubric-based evaluation, but the evidence needed to assess these tasks can be spread across the environment state and the execution trace. A judge may overlook this evidence or accept unsupported claims, including an agent's own reports of success. When rubric verdicts serve as reinforcement-learning rewards, such errors can reinforce undesirable behavior. We introduce Sauron, an evidence-first agentic verifier that investigates both state and trace and relies on a claim only after independently validating it against its source. Bounded queries and joint investigation of related criteria reduce context use and redundant work, and each verdict carries an auditable receipt of its supporting evidence. Across 2,586 human-labeled criteria from 100 personal-assistant episodes, Sauron achieves the highest F1 of five verifiers on all eight backbones tested while costing less than Gandalf, the strongest baseline. With GPT-5.6 Sol, it reaches 0.96 F1 versus Gandalf's 0.92 at 23% lower cost, with a wider margin on process criteria. On a paired fault-injection suite covering 13 families, several of which simulate reward hacking, Sauron detects 97% of injected failures while flagging only 2% of clean controls, outperforming baselines even when they are given the full trace. On TRAIL, a benchmark for detecting and localizing real agent errors, Sauron leads both Gandalf and TRAIL's own judge on all three quality metrics.

Chinese Translation

长时间跨度的半可验证自主任务通常依赖于基于评分标准的评估，但评估这些任务所需的证据可能分散在环境状态和执行轨迹中。评审者可能会忽视这些证据或接受没有支持的主张，包括代理自身对成功的报告。当评分标准的裁决作为强化学习的奖励时，这种错误可能会强化不良行为。我们提出了 Sauron，一种以证据为先的自主验证器，它同时调查状态和轨迹，并在独立验证主张与其来源后才依赖该主张。有限查询和相关标准的联合调查减少了上下文使用和冗余工作，每个裁决都附带可审计的支持证据收据。在来自100个个人助手情境的2586个人工标注标准中，Sauron在所有测试的八个基础模型上实现了五个验证器中最高的F1分数，同时成本低于最强基线 Gandalf。使用 GPT-5.6 Sol 时，它的 F1 达到 0.96，而 Gandalf 的为 0.92，成本降低了23%，在过程标准上具有更大的优势。在覆盖13个类别的配对故障注入套件中，其中几个模拟奖励黑客，Sauron 检测到97%的注入故障，同时仅标记2%的干净控制，即使在基线获得完整轨迹的情况下也表现优于基线。在用于检测和定位真实代理错误的基准 TRAIL 中，Sauron 在所有三个质量指标上均领先于 Gandalf 和 TRAIL 自身的评审者。

View on arXiv Download PDF AI Translation

cs.AI / 285 / 2610.17728

Safety for Whom? Benchmarking Understanding Beyond Refusal in Egocentric Multimodal Assistants

安全性针对谁？在自我中心多模态助手中超越拒绝的理解基准测试

Han, Yewon, Jeon, Jisu, Choi, Suhyun, Yeo, Juan, Jo, Minsoo, Kim, Gihoon, Lim, Hyunggyu, Kim, Taesup

Abstract

Egocentric multimodal assistants see through the user's eyes and guide what the user does next, so their outputs directly shape physical actions, including those that affect bystanders. Yet existing safety evaluations do not distinguish whom a response would harm, and score only whether a model refuses. Pooling all harms into one refusal rate hides two failures: a model may protect one target but not another, and it may refuse without understanding the risk, or understand it and comply anyway. We present STOU, a benchmark of 2,902 egocentric video-audio and video-text queries whose potential harm targets the user (Self), co-present individuals (The Others), or society at large (Others). Our evaluation decouples safety understanding from refusal behavior: beyond refusal, we score whether the model recognizes the user's intent, the presence and target of harm, and its downstream consequences. Across 10 open and closed source omnimodal and multimodal LLMs, we find that refusal varies systematically with the harm target, and that models often recognize a harm yet still assist with it. These results show that refusal rates alone cannot certify an assistant as safe, and that safety alignment must account for who would be harmed. Project page: https://onedayoneyeah.github.io/safety-for-whom

Chinese Translation

自我中心多模态助手通过用户的视角进行观察，并指导用户的下一步行动，因此它们的输出直接影响物理行为，包括对旁观者的影响。然而，现有的安全性评估并未区分响应可能造成伤害的对象，仅仅评估模型是否拒绝。将所有伤害汇总为一个拒绝率掩盖了两个失败：模型可能保护一个目标但不保护另一个目标，并且它可能在不理解风险的情况下拒绝，或者理解风险后仍然选择遵从。我们提出了STOU，这是一个包含2902个自我中心视频-音频和视频-文本查询的基准，其潜在伤害目标包括用户（Self）、共同在场的个体（The Others）或整个社会（Others）。我们的评估将安全理解与拒绝行为解耦：除了拒绝之外，我们还评估模型是否识别用户的意图、伤害的存在及目标，以及其后续影响。在10个开放源代码和闭源的全模态及多模态大语言模型（LLMs）中，我们发现拒绝行为与伤害目标之间存在系统性变化，模型往往能够识别伤害但仍然提供帮助。这些结果表明，仅凭拒绝率无法认证助手的安全性，安全对齐必须考虑谁会受到伤害。项目页面：https://onedayoneyeah.github.io/safety-for-whom

View on arXiv Download PDF AI Translation

cs.AI / 286 / 2610.17738

DuplexAct: How Interactive Are Frontier Full-Duplex Spoken Language Models?

DuplexAct：前沿全双工口语语言模型的互动性如何？

Dong, Jiaheng, Tan, Shihong, Huang, Gongping, Honorio, Jean, Ghosh, Abhirup, Dang, Ting

Abstract

End-to-end (E2E) full-duplex spoken language models (FD-SLMs) enable simutanous listening and speaking continously, requiring models to take appropriate interaction actions as conversations unfold. However, existing benchmarks lack a structured action-level formulation, often evaluate action occurrence rather than success, and provide limited coverage of diverse real-world conversations. We introduce DuplexAct, a benchmark for evaluating interaction action taking in E2E FD-SLMs. DuplexAct defines a three-level action taxonomy spanning eight interaction actions and 24 corresponding interaction conditions. We further characterize successful action-taking through immediate execution and subsequent follow-through, and develop action-specific evaluation metrics for both. DuplexAct contains 4,691 test instances spanning casual conversation, meetings, customer service, and interviews, including 2,531 instances derived from real-world conversational recordings. Together, DuplexAct provides a structured and real-world-grounded framework for systematically assessing how E2E FD-SLMs participate in ongoing spoken interaction.

Chinese Translation

端到端（E2E）全双工口语语言模型（FD-SLMs）能够实现持续的同时听说，要求模型在对话展开时采取适当的互动行动。然而，现有基准缺乏结构化的行动级别表述，通常评估行动的发生而非成功，并且对多样化的现实世界对话覆盖有限。我们引入了DuplexAct，一个用于评估E2E FD-SLMs中互动行动的基准。DuplexAct定义了一个涵盖八种互动行动和24种相应互动条件的三级行动分类法。我们进一步通过即时执行和后续跟进来表征成功的行动执行，并为两者开发了特定于行动的评估指标。DuplexAct包含4,691个测试实例，涵盖休闲对话、会议、客户服务和面试，其中2,531个实例源自现实世界的对话录音。总体而言，DuplexAct提供了一个结构化且基于现实世界的框架，以系统地评估E2E FD-SLMs在持续口语互动中的参与情况。

View on arXiv Download PDF AI Translation

cs.AI / 287 / 2610.17740

Coupled Harness--Model Evolution for Scientific Equation Discovery

耦合工具-模型演化用于科学方程发现

Li, Pengze, Zhang, Jiaquan, Su, Encheng, Long, Yunbo, Zeng, Zihang, Liu, Xinping, Zhang, Han, Yu, Junchi, Tang, Shixiang, Wang, Aoran, Chen, Xi

Abstract

Large language models (LLMs) are increasingly used to discover scientific equations from experimental data. Such a discovery system has two parts that must work together: a library of executable computational tools and the model's tool-use policy. The tools supply the solving capability, from diagnosing structure in the data to fitting candidate equations; the policy plans and orchestrates their use. Yet existing systems adapt only one part, keeping either the toolset or the model fixed, which leaves the corresponding failures uncorrectable. We develop a coupled evolution framework that improves both parts from the same verifiable fit-error signal. In the tool-evolution phase, an agent grows an executable tool library from a small set of generic numerical operations by diagnosing its own failures: it authors a new tool when a capability is missing and revises its selection rules when the right tool went unused. In the policy-evolution phase, a small open model (roughly 3B active parameters) is fine-tuned on successful tool-use trajectories collected with the frozen library, on training formulas screened to be disjoint from all evaluation benchmarks. Ablations show the two phases are complementary: with no tool library at all the same model solves almost nothing, the tools account for most of the gain, and trajectory training adds a further gain that is largest on the most difficult problems. The full system solves 79.1/71.3\% of LSR-Synth problems at fit tolerances $10^{-2}$/$10^{-3}$, reaches 63.8\% pooled accuracy on the distributionally different SRSD-Feynman panel, and obtains a normalized CP3-28 aggregate of $-7.27$. These results suggest a broader recipe for scientific agents: let the model and its instruments improve each other under verifiable feedback.

Chinese Translation

大型语言模型（LLMs）越来越多地被用于从实验数据中发现科学方程。这样的发现系统有两个部分必须协同工作：可执行计算工具的库和模型的工具使用策略。这些工具提供了解决能力，从数据结构的诊断到候选方程的拟合；而策略则规划和协调它们的使用。然而，现有系统仅适应其中一部分，保持工具集或模型不变，这使得相应的失败无法纠正。我们开发了一个耦合演化框架，从同一个可验证的拟合误差信号中改进这两个部分。在工具演化阶段，一个智能体通过诊断自身的失败，从一小组通用数值操作中扩展出一个可执行工具库：当缺少某种能力时，它会创作一个新工具，并在合适的工具未被使用时修订其选择规则。在策略演化阶段，一个小型开放模型（大约3B活跃参数）在使用冻结库收集的成功工具使用轨迹上进行微调，训练公式经过筛选，与所有评估基准不重叠。消融实验表明这两个阶段是互补的：在没有工具库的情况下，同一模型几乎无法解决任何问题，工具占据了大部分增益，而轨迹训练进一步增加的增益在最困难的问题上最大。完整系统在拟合容差为$10^{-2}$/$10^{-3}$的LSR-Synth问题上解决了79.1 ext{/}71.3 ext{%}，在分布不同的SRSD-Feynman面板上达到了63.8 ext{%}的汇总准确率，并获得了标准化的CP3-28总分$-7.27$。这些结果表明了一种更广泛的科学智能体的配方：让模型及其工具在可验证的反馈下相互改进。

View on arXiv Download PDF AI Translation

cs.AI / 288 / 2610.17757

Instructions Are Evidence, Not Objectives: Measuring Outcome Alignment of Agents

指令是证据，而非目标：代理的结果对齐测量

Liu, Kaiming, Liu, Yuxuan, Ju, Jinrui, Ruan, Minyuan, Lei, Xuanyu, Li, Peng, Liu, Yang

Abstract

Large language model agents can follow complex human instructions to solve real-world tasks. In practice, however, initial instructions are often incomplete. In our analysis of 100 user-agent trajectories from SWE-chat, 56\% of initial instructions omit requirements that later emerge as necessary, and 52\% of the useful follow-up requirements could have been inferred from the project. A user states requirements from their own understanding of the task, and these \textbf{stated requirements} often differ from the \textbf{true requirements} that completing the task imposes. An agent should therefore treat the instruction as evidence rather than as the objective, inferring the true requirements from the instruction and the environment and satisfying them even when the user is not yet aware of all of them. We call this ability \textbf{outcome alignment}. In this work, we construct \textbf{TiGanEval}, a benchmark of software engineering and research tasks. In every task the user genuinely lacks part of the requirements, each requirement stays discoverable from the code and runtime behaviour of the environment, and requirements the agent infers are separated from those the user adds after observing delivered work. These properties together make outcome alignment measurable. Evaluating eight frontier models including GPT-5.6-Sol, we find outcome alignment broadly missing: the models lose 53 to 67 points of success rate when moving from the full specification to the partial instruction, and consequence feedback across four submissions does not close the gap. Furthermore, a plan-mode harness cannot compensate for the missing ability of the model itself, which places outcome alignment among the abilities the next generation of models should be developed for.

Chinese Translation

大型语言模型代理能够遵循复杂的人类指令来解决现实世界的任务。然而，在实践中，初始指令往往是不完整的。在我们对100个用户-代理轨迹的分析中，56%的初始指令遗漏了后来被认为是必要的要求，52%的有用后续要求可以从项目中推断出来。用户根据自己对任务的理解陈述要求，而这些 extbf{陈述要求}通常与完成任务所施加的 extbf{真实要求}不同。因此，代理应将指令视为证据，而非目标，从指令和环境中推断真实要求，并在用户尚未意识到所有要求时满足这些要求。我们将这种能力称为 extbf{结果对齐}。在本研究中，我们构建了 extbf{TiGanEval}，一个软件工程和研究任务的基准。在每个任务中，用户确实缺乏部分要求，每个要求都可以从代码和环境的运行时行为中发现，代理推断的要求与用户在观察到交付工作后添加的要求相分离。这些属性共同使得结果对齐可测量。评估包括GPT-5.6-Sol在内的八个前沿模型时，我们发现结果对齐普遍缺失：模型在从完整规范转向部分指令时成功率下降了53到67个百分点，而四次提交之间的后果反馈并未缩小这一差距。此外，计划模式的工具无法弥补模型自身缺失的能力，这使得结果对齐成为下一代模型应当发展的能力之一。

View on arXiv Download PDF AI Translation

cs.AI / 289 / 2610.17762

Autonomous 3D Visual-Thermal Inspection for Utility-Scale Solar Farms

公用事业规模太阳能电站的自主三维视觉-热检测

Tychsen-Smith, Lachlan, Cox, Mark, Wood, Brett, Haddon, David, Anderson, Kenrick, Duck, Benjamin, O'Malley, Jim, Duffy, Noel W, Moghadam, Peyman

Abstract

Utility-scale photovoltaic (PV) inspection is dominated by aerial surveys, which are effective for identifying module-level hot cells but provide limited visibility of critical components located under the array, including connectors, cabling, fasteners and structural elements. These assets are difficult to inspect consistently at scale, and their defects are often spatially small, partially occluded and difficult to localise from isolated images. We present an automated 3D visual-thermal inspection system (Tricorder) for under-panel PV asset inspection. The approach combines LiDAR-inertial Simultaneous Localization and Mapping (SLAM) with geometrically calibrated RGB and thermal imagery and machine-learning-based object detection and segmentation. Detected assets are associated with the reconstructed 3D environment, allowing visual and thermal observations from multiple viewpoints to be spatially fused and localised. We demonstrate the approach on utility-scale PV data for MC4 connector detection and connection-state recognition, thermal anomaly identification, and detection of structural components including bolts, nuts and torque tubes. The results demonstrate a pathway from image-based inspection toward spatially grounded autonomous inspection, where defects are detected and linked to specific assets and their 3D locations within the solar farm.

Chinese Translation

公用事业规模的光伏（PV）检测主要依赖于空中调查，这种方法在识别模块级热点方面有效，但对位于阵列下方的关键组件（如连接器、电缆、紧固件和结构元素）提供的可视性有限。这些资产在大规模检查时难以保持一致，且其缺陷通常空间较小、部分被遮挡，并且难以从孤立图像中定位。我们提出了一种自动化的三维视觉-热检测系统（Tricorder），用于面板下光伏资产的检查。该方法结合了激光雷达-惯性同时定位与地图构建（SLAM）、几何标定的RGB和热成像以及基于机器学习的物体检测与分割。检测到的资产与重建的三维环境相关联，从多个视角进行的视觉和热观察能够进行空间融合和定位。我们在公用事业规模的光伏数据上演示了该方法，用于MC4连接器检测和连接状态识别、热异常识别，以及包括螺栓、螺母和扭矩管在内的结构组件检测。结果展示了从基于图像的检测向空间基础的自主检测的路径，其中缺陷被检测并与特定资产及其在太阳能电站内的三维位置关联。

View on arXiv Download PDF AI Translation

cs.AI / 290 / 2610.17763

Memory Echo in Single-Agent Long-Term Memory: Lifecycle Evaluation and Provenance-Guided Mitigation

单代理长期记忆中的记忆回声：生命周期评估与来源引导的缓解

Li, Yufeng, Yao, Sheng, Zeng, Peng, Kong, Dezhang, Lin, Shi, Lin, Changting, Wang, Xun, Han, Meng

Abstract

Long-term memory supports agents across sessions. Repeated retrieval and write-back can let one historical claim occupy multiple retrieval positions and influence current answers without independent evidence. Existing work examines retrieval, updating, and question answering, leaving same-source record accumulation and post-correction persistence underexplored. Motivated by research in social cognition and memory psychology on repeated exposure, retrieval, and revision, we define \emph{Memory Echo} as the excess retrieval exposure gained by historical information from the same source without additional independent evidence, and introduce MEchoBench to evaluate its emergence, correction resistance, persistence, and effect on answers. To address this phenomenon, we introduce the Family--Provenance--Authority framework (FPA). Specifically, FPA selects one representative per source family, conditions the eligibility of old records for the answer context on verified corrections, and constrains final answers with authoritative evidence. In paired replay on 190 fixed questions across four memory frameworks, FPA raises accuracy from 0.825 to 0.920 with DeepSeek V4 Flash and from 0.782 to 0.916 with Kimi K2.6; mean Decision Contamination Effect falls from 0.106 to 0.030 and from 0.109 to 0.038, respectively. These results show that agent memory can preserve traceable history without allowing repeated historical information to govern current decisions. The code is available at https://github.com/OrcasSam/memory-echo-benchmark.

Chinese Translation

长期记忆支持代理在多个会话中的表现。重复的检索和写回可以使一个历史声明占据多个检索位置，并在没有独立证据的情况下影响当前答案。现有研究主要集中于检索、更新和问题回答，而对同源记录的累积和后期修正的持久性探讨不足。受到社会认知和记忆心理学中关于重复暴露、检索和修正的研究启发，我们将 extit{记忆回声}定义为来自同一来源的历史信息在没有额外独立证据的情况下所获得的过度检索暴露，并引入MEchoBench来评估其出现、修正抵抗、持久性及对答案的影响。为了解决这一现象，我们引入了家庭-来源-权威框架（Family--Provenance--Authority，FPA）。具体而言，FPA为每个来源家庭选择一个代表，基于经过验证的修正条件限制旧记录在答案上下文中的适用性，并用权威证据约束最终答案。在对190个固定问题进行的配对重放实验中，FPA使得使用DeepSeek V4 Flash的准确率从0.825提高到0.920，使用Kimi K2.6的准确率从0.782提高到0.916；平均决策污染效应分别从0.106降至0.030和从0.109降至0.038。这些结果表明，代理记忆可以保留可追溯的历史，而不允许重复的历史信息主导当前决策。代码可在https://github.com/OrcasSam/memory-echo-benchmark获取。

View on arXiv Download PDF AI Translation

cs.AI / 291 / 2610.17786

Agentic4PDM: Host-Governed Evidence Admissibility for Industrial Fault Diagnosis

Agentic4PDM：主机治理的工业故障诊断证据可采纳性

Nguyen, Nam V., Van Truong, Hoang, D., Hieu D., Nguyen, Duy Khuong, Pham, Quang, Kieu, Tung

Abstract

We study industrial fault diagnosis as a classification task that assigns sensor time-series windows to predefined health states or fault types. Scarce labels and changing operating conditions make repeated training of specialized classifiers costly. Yet structures such as periodicity, trends, and transients can remain diagnostically informative and be examined through reusable analytical tools. LLM agents can select and interpret these tools, but fixed rules or training-derived guidance may not resolve each signal's ambiguity, while unconstrained interpretation may overstate weak or conflicting results. We introduce Agentic4PDM, a framework that separates adaptive investigation from control over evidence use. LLM agents compare each signal with labeled references from every candidate class, identify unresolved distinctions, and propose follow-up measurements. A deterministic software controller, called the host, enforces contract-checked gated adjudication (CCGA): it checks requests, requires commitment to a complete tool batch before results are observed, and assigns evidence roles based on tool checks against labeled references. These roles distinguish corroborative support from advisory information. The Judge Agent proposes a diagnosis, while the controller checks protocol compliance and records the investigation. Agentic4PDM supports training-free diagnosis using pretrained representations and labeled references without task-specific diagnostic models, and a standard setting that additionally investigates trained-model predictions. Experiments on selected bearing and motor diagnosis challenges and an unfiltered industrial test set supplied by our customer show gains in aggregate macro F1 over the evaluated diagnostic and agentic baselines. Ablations further support the value of the governed workflow.

Chinese Translation

我们将工业故障诊断视为一种分类任务，该任务将传感器时间序列窗口分配给预定义的健康状态或故障类型。稀缺的标签和变化的操作条件使得对专门分类器的重复训练成本高昂。然而，周期性、趋势和瞬态等结构仍然可以提供诊断信息，并可以通过可重用的分析工具进行检查。大型语言模型（LLM）代理可以选择和解释这些工具，但固定规则或训练导出的指导可能无法解决每个信号的模糊性，而不受限制的解释可能会夸大微弱或相互矛盾的结果。我们提出了Agentic4PDM，一个将自适应调查与证据使用控制分开的框架。LLM代理将每个信号与每个候选类别的标记参考进行比较，识别未解决的区别，并提出后续测量建议。一个称为主机的确定性软件控制器实施合同检查的门控裁定（CCGA）：它检查请求，要求在观察结果之前承诺完整的工具批次，并根据工具检查与标记参考的结果分配证据角色。这些角色区分了支持性证据和建议性信息。法官代理提出诊断，而控制器检查协议合规性并记录调查。Agentic4PDM支持使用预训练表示和标记参考进行无训练的诊断，而无需特定任务的诊断模型，并且标准设置还进一步调查训练模型的预测。在选定的轴承和电动机诊断挑战以及客户提供的未过滤工业测试集上的实验显示，整体宏观F1得分超过了评估的诊断和代理基线。消融实验进一步支持了治理工作流程的价值。

View on arXiv Download PDF AI Translation

cs.AI / 292 / 2610.17816

Empowering Edge Agents: A Systematic Analysis of Edge-Cloud Collaboration in LLM Agent Execution

赋能边缘代理：边缘-云协作在大规模语言模型代理执行中的系统分析

Huang, Pengcheng, Liu, Zhenghao, Ren, Shixin, Yu, Shi, Yao, Sijia, Yan, Yukun, Xiao, Chaojun, Sun, Maosong

Abstract

Edge-cloud collaboration combines the lower inference overhead and greater data locality of edge-side execution with the stronger capabilities of cloud-hosted models. However, as LLM agents increasingly operate in long-horizon scenarios, cloud participation becomes interleaved throughout execution rather than a one-shot decision for each query. Such interleaved participation introduces trajectory-level effects that remain insufficiently understood. To address this gap, we introduce EC2-Bench, a benchmark for evaluating edge-cloud collaboration in long-horizon agent execution. It provides executable digital-workspace agent tasks and a unified evaluation framework covering task performance, resource usage, and privacy exposure. Using this benchmark, we empirically study diverse forms of edge-cloud collaboration. Our study reveals three findings. First, collaboration achieves a favorable balance among task performance, resource usage, and privacy exposure compared with edge-only and cloud-only execution. Second, as edge models become stronger, collaborative execution approaches cloud-only performance, and performance differences across strategies shrink, while privacy exposure remains strongly strategy-dependent. Third, stronger edge capability shifts preferred collaboration from externally controlled cloud participation and direct cloud execution toward more edge-controlled collaboration. Further analyses identify cloud-invocation decision-making and instruction following as important edge-side abilities for effective collaboration. These findings provide a systematic understanding of edge-cloud collaboration in long-horizon agent execution and insights for designing future edge agents that effectively collaborate with cloud models. Code and resources are available at https://github.com/OpenBMB/EC2-Bench.

Chinese Translation

边缘-云协作结合了边缘侧执行的较低推理开销和更强的数据本地性，以及云托管模型的更强能力。然而，随着大规模语言模型（LLM）代理在长时间跨度场景中越来越多地运行，云的参与在执行过程中变得交错，而不是每个查询的单次决策。这种交错参与引入了轨迹级别的影响，仍然缺乏充分的理解。为了解决这一问题，我们引入了EC2-Bench，这是一个用于评估长时间跨度代理执行中边缘-云协作的基准。它提供可执行的数字工作空间代理任务和一个统一的评估框架，涵盖任务性能、资源使用和隐私暴露。利用这个基准，我们实证研究了多种形式的边缘-云协作。我们的研究揭示了三个发现。首先，与仅边缘执行和仅云执行相比，协作在任务性能、资源使用和隐私暴露之间实现了良好的平衡。其次，随着边缘模型的增强，协作执行接近仅云执行的性能，策略之间的性能差异缩小，而隐私暴露仍然强烈依赖于策略。第三，较强的边缘能力使得优选的协作从外部控制的云参与和直接云执行转向更多的边缘控制协作。进一步的分析确定了云调用决策和指令跟随作为有效协作的重要边缘侧能力。这些发现为长时间跨度代理执行中的边缘-云协作提供了系统性的理解，并为设计未来能够有效与云模型协作的边缘代理提供了见解。代码和资源可在 https://github.com/OpenBMB/EC2-Bench 获取。

View on arXiv Download PDF AI Translation

cs.AI / 293 / 2610.17827

Benchmarking and Training Reward Models for Cross-Source Evidence Grounding in Search Agents

跨源证据基础的搜索代理奖励模型的基准测试与训练

Zhao, Haochen, Hao, Minsheng, Li, Zheng, Li, Yifan, Liu, Zikang, Qin, Bowen, Cheng, Yuhao, Fan, Chuang, Zou, Haosheng, Chen, Jeff, Li, Sujian

Abstract

Existing approaches to training and evaluating search agents focus heavily on final-answer quality, largely neglecting how generated answers map back to retrieved evidence. Recent methods track citations to enforce grounding, but citation support alone does not establish source independence or whether an answer appropriately handles conflicting evidence. We argue that credibility in open-web retrieval depends on principled cross-source evidence grounding that accounts for source independence and systematically attributes and adjudicates conflicting evidence, while hierarchical judgments of these grounding decisions provide a unified signal for training and evaluation. To this end, we propose an evaluation framework, Cross-Source Evidence Grounding (CSEG), that employs a fine-grained, six-layer decision tree over the complete search trajectory. We introduce CSEG-Bench and trainable reward models to provide targeted supervision following this framework. Our approach helps eliminating the severe data skew of natural trajectories by providing targeted coverage over complex, deep-layer decision paths, such as cross-source conflict resolution and nuanced attribution, that standard rollouts fail to capture. Specifically, (1) we define the hierarchy at label, node, and path levels; (2) we construct 1,160 verified trajectories over nine canonical paths by rewriting answers on real retrieved document pools, with independent frontier judges reproducing the entire gold path and human experts deciding inclusion; (3) we train CSEG-RM, which reaches 83.9% exact accuracy on our benchmark while lying on the performance-token efficiency Pareto frontier. Experiments that use this reward alone to supervise a frozen-retrieval answer policy raise the independently judged mean label from 2.83 to 4.19, and show that learning the hierarchy generalizes across layers at a fraction of frontier-judge cost.

Chinese Translation

现有的搜索代理训练和评估方法主要集中在最终答案的质量上，较少关注生成的答案如何与检索到的证据相对应。最近的方法通过跟踪引用来强制建立基础，但仅凭引用支持并不能确立源独立性或答案是否适当地处理了冲突证据。我们认为，开放网络检索的可信度依赖于原则性的跨源证据基础，这种基础考虑了源独立性，并系统地归因和裁定冲突证据，而对这些基础决策的层次判断提供了一个统一的信号用于训练和评估。为此，我们提出了一个评估框架——跨源证据基础（Cross-Source Evidence Grounding, CSEG），该框架在完整的搜索轨迹上采用细粒度的六层决策树。我们引入了CSEG-Bench和可训练的奖励模型，以便在这一框架下提供针对性的监督。我们的方法通过在复杂的深层决策路径（如跨源冲突解决和细致归因）上提供针对性覆盖，帮助消除自然轨迹的严重数据偏斜，而标准的展开方法未能捕捉到这些内容。具体而言，(1) 我们在标签、节点和路径层面定义了层次结构；(2) 我们通过在真实检索文档池上重写答案，构建了1,160条经过验证的轨迹，涵盖九条经典路径，独立的前沿评审者重现了整个黄金路径，人类专家决定是否纳入；(3) 我们训练了CSEG-RM，该模型在我们的基准测试中达到了83.9%的精确度，同时位于性能-代币效率的帕累托前沿。仅使用这一奖励来监督冻结检索答案策略的实验将独立评审的平均标签从2.83提高到4.19，并显示出学习层次结构在前沿评审成本的很小一部分中跨层泛化。

View on arXiv Download PDF AI Translation

cs.AI / 294 / 2610.17840

Learning Adaptive Deliberation Control for Reliable Recognition with Multi-Agent Language Models

学习自适应审议控制以实现多智能体语言模型的可靠识别

Liu, Yuan, Peng, Haoyue, Hu, Fanliang, Huang, Lei, Cai, Wenli

Abstract

Multi-agent deliberation can improve recognition by criticizing and revising candidate answers, but fixed controls cannot respond to changes in disagreement or reasoning quality. We present an adaptive multi-agent controller (ADMA) that learns continuous policies for proposal, challenge, arbitration, and coordination while keeping the task model frozen. Shared trace features describe quality, agreement, and progress, while role-specific features support decentralized control during inference. One controller trained on legal and financial reasoning is transferred without policy updates to four textual and multimodal benchmarks and two successive task-model versions. With Qwen3.5 held fixed, ADMA-Full improves accuracy by 6.2-14.2 percentage points over Direct and by 2.8-3.0 points over ADMA-Fixed. Paired bootstrap ranges for the Full-Fixed comparison are positive on three tasks. Outcome decompositions show that recoveries exceed introduced errors, while process assessment records fewer false-consensus events and more supported correct outcomes. Transfer to Qwen3.6 remains task dependent, which bounds the generality of the learned policy. Median token and latency ratios are 14.3-18.7 and 7.8-16.5 relative to Direct. These results show that adaptive deliberation can improve fixed-model recognition when its task-dependent gains justify the additional computation.

Chinese Translation

多智能体审议通过批评和修正候选答案来提高识别能力，但固定控制无法应对分歧或推理质量的变化。我们提出了一种自适应多智能体控制器（ADMA），它学习提议、挑战、仲裁和协调的连续策略，同时保持任务模型不变。共享的轨迹特征描述了质量、一致性和进展，而角色特定特征则支持推理过程中的去中心化控制。一个在法律和金融推理上训练的控制器可以在没有策略更新的情况下转移到四个文本和多模态基准以及两个连续的任务模型版本。在固定的 Qwen3.5 下，ADMA-Full 的准确率比 Direct 提高了 6.2-14.2 个百分点，比 ADMA-Fixed 提高了 2.8-3.0 个百分点。Full-Fixed 比较的配对自助法范围在三个任务上均为正值。结果分解显示恢复超出了引入的错误，而过程评估记录了更少的虚假共识事件和更多的支持正确结果。转移到 Qwen3.6 仍然依赖于任务，这限制了学习策略的普适性。中位数的标记和延迟比率相对于 Direct 分别为 14.3-18.7 和 7.8-16.5。这些结果表明，当任务相关的收益证明额外计算是合理的时，自适应审议可以改善固定模型的识别能力。

View on arXiv Download PDF AI Translation

cs.AI / 295 / 2610.17860

Socratic Skill Learning: Interleaving Teaching and Practice for Transferable Agent Skills

苏格拉底技能学习：交错教学与实践以实现可转移的智能体技能

Zhu, Letian, Li, Han, Yao, Yifan, Lei, Xinping, Che, Xinyu, Wang, Jiahao, Wang, Dianyi, Fu, Yucheng, Wang, Jiakai, Bai, Zhiqi, Su, Wenbo, Liu, Jiaheng

Abstract

Agent Skills improve LLM agents by making task knowledge reusable across sessions. Learning transferable Skills requires evidence about a procedure and its conditions of use, which a successful task attempt can leave untested. We introduce Socratic Skill Learning (SSL), which interleaves expert teaching with Student practice. An Educator guides investigation while the Student remains the sole environment operator and Skill author, able to question, test, and adapt advice. To test reuse beyond the learning task, we construct SkillTransferBench, with 35 task families and 408 executable cases, pairing public variation specifications with held-out evaluation in fresh solver sessions. On the main 22-family, 155-case cohort, Fixed-Window SSL reaches 0.801 utility, versus 0.723 without a Skill, 0.761 for adapted Trace2Skill (trajectory distillation), and 0.751 for adapted SKILL-KD (contrastive distillation). Across three model pairs on shared families, interleaved guidance outperforms expert authoring and post-hoc teaching. Process analysis links guidance to Student writing and package revision; a case study shows the Student testing and refining a proposed procedure for later reuse. As demand for new lessons falls, the cost of fixed-window review grows. Issue-Guided SSL tracks unresolved questions across Skill revisions to select and focus review, reaching 0.805 utility with 61.7% fewer Educator tokens and 16.9% fewer total acquisition tokens than Fixed-Window SSL on the main cohort.

Chinese Translation

智能体技能通过使任务知识在会话之间可重用来提升大语言模型（LLM）智能体的能力。学习可转移的技能需要关于程序及其使用条件的证据，而成功的任务尝试可能未能进行测试。我们提出了苏格拉底技能学习（Socratic Skill Learning, SSL），它将专家教学与学生实践交错进行。教育者引导调查，而学生则是唯一的环境操作员和技能创作者，能够质疑、测试和调整建议。为了测试学习任务之外的重用能力，我们构建了技能转移基准（SkillTransferBench），包含35个任务系列和408个可执行案例，将公共变异规范与新解算器会话中的保留评估相结合。在主要的22个系列、155个案例的样本中，固定窗口SSL达到了0.801的效用，而没有技能时为0.723，适应后的Trace2Skill（轨迹蒸馏）为0.761，适应后的SKILL-KD（对比蒸馏）为0.751。在共享系列的三个模型对中，交错指导的表现优于专家创作和事后教学。过程分析将指导与学生写作和包修订联系起来；案例研究显示学生测试和完善了一个提议的程序以便于后续重用。随着对新课程需求的下降，固定窗口复习的成本上升。问题引导的SSL跟踪技能修订中的未解决问题，以选择和集中复习，相比于固定窗口SSL，在主要样本中达到了0.805的效用，并减少了61.7%的教育者令牌和16.9%的总获取令牌。

View on arXiv Download PDF AI Translation

cs.AI / 296 / 2610.17869

ECMR: Evidence-Aware Deliberation Control for Multi-Agent LLM Reasoning

ECMR：面向证据的多智能体大语言模型推理的审议控制

Liu, Yuan, Peng, Haoyue, Hu, Fanliang, Huang, Lei, Cai, Wenli

Abstract

Multi-agent Large Language Model (LLM) systems typically commit to a deliberation protocol before inference. A fixed protocol cannot allocate computation differently to resolved and unresolved instances, and it may mistake repeated support for independent confirmation. We study this failure as semantic false consensus: a confident group answer supported by narrow, duplicated, or correlated rationales. We introduce ECMR, an evidence-aware controller over frozen role-conditioned LLMs. At each step, ECMR summarizes answer concentration, support diversity and independence, contradictions, role behavior, verification history, and remaining budget, then selects whether to propose, challenge, verify, arbitrate, diversify, correct a role, or stop. We formulate this decision as finite-horizon budgeted control and derive sufficient conditions under which state conditioning, risk-triggered verification, and role correction improve cost-adjusted utility. Across a declared held-out evaluation of 9,000 records from MedQA, CaseHOLD, and FinQA, ECMR-Full increases pooled accuracy from 62.1% to 84.9%, reduces false-consensus rate from 40.7% to 17.0%, and raises evidence-support rate from 58.0% to 88.7%. State shuffling loses 11.7 accuracy points with only 0.7 fewer rounds, while harder traces shift from early stopping toward arbitration. ECMR-Full uses the largest token budget, yielding the strongest absolute reliability rather than uniformly superior token efficiency.

Chinese Translation

多智能体大语言模型（LLM）系统通常在推理之前承诺遵循一种审议协议。固定的协议无法对已解决和未解决的实例进行不同的计算分配，并且可能会将重复的支持误认为是独立的确认。我们将这种失败研究为语义虚假共识：一个自信的群体答案由狭窄、重复或相关的理由支持。我们引入了ECMR，一种在冻结角色条件下的LLM上进行证据感知控制的机制。在每一步中，ECMR总结答案集中度、支持多样性和独立性、矛盾、角色行为、验证历史和剩余预算，然后选择是提出、挑战、验证、仲裁、多样化、纠正角色还是停止。我们将这一决策形式化为有限视野预算控制，并推导出在状态条件、风险触发验证和角色纠正改善成本调整效用的充分条件。在对来自MedQA、CaseHOLD和FinQA的9,000条记录的声明性保留评估中，ECMR-Full将汇总准确率从62.1%提高到84.9%，将虚假共识率从40.7%降低到17.0%，并将证据支持率从58.0%提高到88.7%。状态洗牌损失11.7个准确率点，仅减少0.7轮，而更困难的轨迹则从早期停止转向仲裁。ECMR-Full使用最大的令牌预算，产生最强的绝对可靠性，而不是均匀优越的令牌效率。

View on arXiv Download PDF AI Translation

cs.AI / 297 / 2610.17893

Training LLMs for Multi-Agent Collaboration with Verifiable Information Rewards

通过可验证信息奖励训练大型语言模型进行多智能体协作

Chen, Keru, Lin, Sen, Liang, Yingbin, Bastian, Nathaniel D., Zou, Shaofeng

Abstract

Effective collaboration among large language model (LLM) agents in multi-agent systems requires sharing relevant evidence, requesting missing information, and incorporating teammates' contributions into decisions. Training these behaviors is challenging because final task outcomes provide limited feedback on the value of individual messages. We introduce CoSIR (Collaboration via Surprisal-based Information Rewards), a post-training approach that uses a simple, verifiable environment to teach collaboration behaviors that generalize to more complex, unseen tasks. We propose a Bayesian evaluator that scores shared information by its reduction in team surprisal, defined as the sum of the negative log probabilities assigned to the correct option given each agent's available evidence. Requests are scored by the potential reduction in the same quantity if the requested information were truthfully provided. Separate rewards encourage agents to use the received information in their private choices and calculations. We post-train Qwen3-4B, Qwen3-8B, and Gemma-3-4B-it through self-play and evaluate transfer to seven unseen natural-language collaboration tasks. CoSIR achieves the highest aggregate score, while its performance in general tasks remains competitive. Behavioral analyses show increased disclosure of private evidence and fewer judged failures to use teammates' input.

Chinese Translation

在多智能体系统中，大型语言模型（LLM）代理之间的有效协作需要共享相关证据、请求缺失信息，并将队友的贡献纳入决策中。训练这些行为具有挑战性，因为最终任务结果对单个消息的价值提供的反馈有限。我们提出了 CoSIR（基于惊奇度的信息奖励协作），这是一种后训练方法，利用简单的可验证环境来教授可推广到更复杂、未见任务的协作行为。我们提出了一种贝叶斯评估器，通过其对团队惊奇度的减少来评分共享信息，团队惊奇度定义为在每个代理可用证据的基础上，分配给正确选项的负对数概率之和。请求的评分则基于如果所请求的信息真实提供时对相同量的潜在减少。单独的奖励鼓励代理在其私人选择和计算中使用接收到的信息。我们通过自我对弈对 Qwen3-4B、Qwen3-8B 和 Gemma-3-4B-it 进行了后训练，并评估其在七个未见自然语言协作任务中的迁移表现。CoSIR 实现了最高的综合得分，同时在一般任务中的表现仍然具有竞争力。行为分析显示，私人证据的披露增加，且判断未能使用队友输入的情况减少。

View on arXiv Download PDF AI Translation

cs.AI / 298 / 2610.17900

PaceReasoner: Label-Free Learning to Answer Briefly from Long Reasoning

PaceReasoner：无标签学习以简洁回答长推理

Jia, Changguo, Xiao, Zhiyou, Ma, Xinbei, Chen, Geng, Zhang, Weiming, Xu, Yiming, Zhao, Tianqi, Zhou, Minghui

Abstract

Extended chain-of-thought reasoning improves problem-solving performance but increases inference cost and latency, motivating more accurate answers with shorter reasoning. We introduce PaceReasoner, a label-free test-time reinforcement learning framework that uses a model's LONG responses to improve its SHORT-mode accuracy. Both modes share a trainable policy, creating a need for reliable supervision, sufficient SHORT training samples, and continued LONG responses to guide learning. PaceReasoner combines all-response and LONG-only answer consensus through progressive, confidence-weighted supervision. A behavior reward gradually shifts from encouraging SHORT exploration to rewarding agreement with LONG answers, while a ratio controller regulates the balance between modes. Across three model backbones and four benchmarks, PaceReasoner improves average SHORT accuracy over the baselines by 7.4--13.5 percentage points, with fewer than 25 reasoning tokens per response on average. Ablations support the key components, and cross-dataset results demonstrate transfer beyond the adaptation set.

Chinese Translation

扩展的思维链推理提高了问题解决的性能，但增加了推理成本和延迟，这促使我们寻求更准确的答案和更短的推理。我们提出了PaceReasoner，这是一种无标签的测试时强化学习框架，利用模型的长响应（LONG）来提高其短模式（SHORT）的准确性。这两种模式共享一个可训练的策略，因此需要可靠的监督、足够的短训练样本以及持续的长响应来指导学习。PaceReasoner通过渐进的、基于信心加权的监督结合了所有响应和仅长响应答案的一致性。行为奖励逐渐从鼓励短探索转向奖励与长答案的一致性，同时比例控制器调节两种模式之间的平衡。在三种模型骨干和四个基准测试中，PaceReasoner将平均短准确性提高了7.4至13.5个百分点，且平均每个响应的推理标记少于25个。消融实验支持了关键组件，而跨数据集的结果则展示了超出适应集的迁移能力。

View on arXiv Download PDF AI Translation

cs.AI / 299 / 2610.17928

Learning How to Remember: Memory Modality Selection for GUI Agents

学习如何记忆：图形用户界面代理的记忆模态选择

Roh, Taeyun, Park, Suhyeong, Kim, Gijin, Jang, Wonjune, Kang, Jaewoo

Abstract

Recent advances in vision-language models (VLMs) have substantially improved the capabilities of GUI agents, while memory augmentation has emerged as a promising approach for reusing knowledge from past interactions. However, existing memory augmented GUI agents typically adopt a fixed representation strategy, providing retrieved experience as a static modality. It remains unclear whether a single memory modality is consistently optimal, or how modality preference changes across interaction steps. In this work, we conduct a systematic study of memory modality in VLM based GUI agents. Across GUI benchmarks, we find that the preferred memory representation is strongly step dependent. Motivated by the finding, we introduce \textbf{SwitchMem}, a routing framework that dynamically selects between textual and visual memory representations at each step. Across three GUI benchmarks, SwitchMem outperforms memory augmented baselines that rely on static memory representations. Our findings validate our claim that memory modality is a step dependent decision rather than a fixed design choice. Code and project resources are available at https://txxnrd.github.io/switchmem

Chinese Translation

最近在视觉-语言模型（VLMs）方面的进展显著提升了图形用户界面（GUI）代理的能力，而记忆增强作为一种重用过去交互知识的有前景的方法也逐渐受到关注。然而，现有的记忆增强型GUI代理通常采用固定的表征策略，将检索到的经验作为静态模态提供。目前尚不清楚单一的记忆模态是否始终是最佳选择，或者模态偏好如何在交互步骤中变化。在本研究中，我们对基于VLM的GUI代理中的记忆模态进行了系统研究。在多个GUI基准测试中，我们发现首选的记忆表征在很大程度上依赖于步骤。基于这一发现，我们引入了 extbf{SwitchMem}，一个在每个步骤动态选择文本和视觉记忆表征的路由框架。在三个GUI基准测试中，SwitchMem的表现优于依赖静态记忆表征的记忆增强基线。我们的研究结果验证了我们的观点：记忆模态是一个依赖于步骤的决策，而非固定的设计选择。代码和项目资源可在 https://txxnrd.github.io/switchmem 获取。

View on arXiv Download PDF AI Translation

cs.AI / 300 / 2610.17976

Seeing Depth but Choosing Cues in Vision-Language Models

在视觉-语言模型中看到深度但选择线索

Zou, Zhengtao, Marttinen, Pekka

Abstract

Do vision-language models use depth information to judge which object is closer, or rely on image-plane cues? We investigate this question through controlled depth--cue conflicts, representation analysis, and interventions. Among the many image-plane cues relevant to depth perception, we focus on apparent size and vertical position. We compare depth judgments under cue-consistent and cue-conflict conditions in both rendered scenes and real-world photographs. Across model families and scales, depth judgments systematically shift toward misleading cues. Yet the correct depth order remains decodable from the queried objects' visual representations, including when the model answers incorrectly. Image and category controls support interpreting this signal as image-dependent depth information beyond the two target cues. Strengthening the model's own depth readout through object-level write-back can correct conflict decisions without updating model parameters or using ground-truth depth annotations. Interventions show that the effect of this correction depends on the direction and target tokens of the written signal, while its strength varies across architectures and domains. Together, these findings reveal a mismatch between a model's ability to represent depth and use this information when deciding the answer. VLMs can retain useful depth information while allowing competing image-plane cues to exert greater influence on their answers.

Chinese Translation

视觉-语言模型是使用深度信息来判断哪个物体更近，还是依赖于图像平面线索？我们通过控制深度与线索的冲突、表示分析和干预来研究这个问题。在与深度感知相关的众多图像平面线索中，我们重点关注表观大小和垂直位置。我们比较了在一致线索和冲突线索条件下的深度判断，涵盖了渲染场景和真实世界照片。在不同模型系列和规模中，深度判断系统性地偏向于误导性线索。然而，正确的深度顺序仍然可以从查询对象的视觉表示中解码，包括在模型回答错误时。图像和类别控制支持将此信号解释为超出两个目标线索的图像依赖深度信息。通过对象级的写回增强模型自身的深度读出，可以在不更新模型参数或使用真实深度注释的情况下纠正冲突决策。干预显示，这种纠正的效果取决于写入信号的方向和目标标记，而其强度在不同架构和领域中有所不同。这些发现共同揭示了模型在表示深度和在决定答案时使用该信息之间的错配。视觉-语言模型可以保留有用的深度信息，同时允许竞争的图像平面线索对其答案产生更大的影响。

View on arXiv Download PDF AI Translation

计算语言学 (Computation and Language)

244

cs.CL / 1 / 2610.13989

PsyBer-Agent: A Psychology-Driven User Agent for Recommender Systems

PsyBer-Agent：一种基于心理学驱动的用户代理推荐系统

Zhang, Chong, Shen, Qijie, Zhang, Haichao, Wang, Can, Jin, Xiaobo, Wang, Jia

Abstract

Evaluating recommender systems before deployment requires simulators that remain realistic over long, closed-loop interaction cycles. Current LLM-based agents generally function as surface imitators that generate realistic short-term responses but fail to capture the underlying user dynamics such as fatigue, affect, and shifting preferences. We propose PsyBer-Agent, a psychology-inspired simulator that models these dynamics through its Psy-Engine. This engine represents each user with three evolving, model-defined latent states. To calibrate simulated behavior against real interaction logs, we align log-conditioned reference trajectories with simulator-generated latent-state trajectories via Unbalanced Gromov--Wasserstein optimal transport. This approach provides structured supervision for a lightweight policy without the computational burden of fine-tuning the backbone LLM. We also introduce WebSim and the PsyBer Benchmark, which encompass multimodal scenarios and seven recommender backbones. Our framework evaluates behavioral fidelity, latent-state consistency, robustness, and responses to social-cue perturbations. On the tested horizons, PsyBer-Agent shows lower behavioral discrepancy than the prompt-only simulators while maintaining stable performance under large-scale concurrent simulation for up to 5M queued simulation tasks under bounded concurrency. Code is available in the supplementary material and at: PsyBer-Agent: https://anonymous.4open.science/r/CC66F/, WebSim Platform: https://anonymous.4open.science/r/D8EAX/.

Chinese Translation

在部署之前评估推荐系统需要能够在长时间闭环交互周期中保持真实感的模拟器。目前基于大语言模型（LLM）的代理通常作为表面模仿者运作，能够生成真实的短期响应，但未能捕捉到潜在的用户动态，如疲劳、情感和偏好的变化。我们提出了PsyBer-Agent，这是一种受心理学启发的模拟器，通过其Psy-Engine建模这些动态。该引擎将每个用户表示为三个不断演变的、模型定义的潜在状态。为了将模拟行为与真实交互日志进行校准，我们通过不平衡Gromov-Wasserstein最优传输将日志条件的参考轨迹与模拟器生成的潜在状态轨迹对齐。这种方法为轻量级策略提供了结构化监督，而无需对主干LLM进行微调的计算负担。我们还引入了WebSim和PsyBer基准，涵盖多模态场景和七个推荐系统主干。我们的框架评估行为保真度、潜在状态一致性、鲁棒性以及对社会线索扰动的响应。在测试的时间范围内，PsyBer-Agent显示出比仅使用提示的模拟器更低的行为差异，同时在高达500万排队模拟任务的有界并发下保持稳定性能。代码可在补充材料和以下链接获取：PsyBer-Agent: https://anonymous.4open.science/r/CC66F/，WebSim平台: https://anonymous.4open.science/r/D8EAX/.

View on arXiv Download PDF AI Translation

cs.CL / 2 / 2610.14071

Loose Lips: Understanding Privacy Leakage in Agentic Dialogue

松口: 理解代理对话中的隐私泄露

Sharma, Soumya, Hussain, Ayana, Arcolezi, Heber Hwang, Aïvodji, Ulrich, Gella, Spandana, Romero-Soriano, Adriana, Farnadi, Golnoosh

Abstract

As LLM agents are increasingly deployed in high-stakes enterprise settings, understanding how they handle sensitive information in agentic dialogue has become critical. Privacy leakage in such settings requires reasoning about whether information flows respect the contextual norms under which they originated. We extend CIMemories to CIMEMORIES DIALOGUE, a benchmark of 4,410 two-agent dialogue instances grounded in Contextual Integrity theory. Using this benchmark, we systematically evaluate 7 open-weight models from the Qwen3 and Llama 3.1 families across conversation lengths of 1 and 6 turns under three prompting conditions. We find that task capability and privacy leakage are fundamentally coupled: more capable models are systematically leakier, with no model in our evaluation able to decouple the two. Thinking models consistently leak more than their non-thinking counterparts, a pattern we attribute to the extended reasoning surface created by chain-of-thought. Leakage compounds with conversation length, rising up to 3x from 1 to 6 turns. Our results suggest that the coupling between task capability and privacy leakage is not an artifact of misconfiguration but a structural feature of current agentic systems.

Chinese Translation

随着大型语言模型（LLM）代理在高风险企业环境中的日益应用，理解它们在代理对话中如何处理敏感信息变得至关重要。在这种环境下，隐私泄露需要对信息流是否遵循其来源的上下文规范进行推理。我们将CIMemories扩展为CIMEMORIES DIALOGUE，这是一个基于上下文完整性理论的4,410个双代理对话实例的基准。利用该基准，我们系统地评估了来自Qwen3和Llama 3.1系列的7个开放权重模型，在三种提示条件下对1轮和6轮对话长度进行评估。我们发现任务能力与隐私泄露之间存在根本的耦合关系：更强大的模型系统性地泄露更多信息，在我们的评估中没有任何模型能够将二者解耦。思维模型的泄露量始终高于非思维模型，我们将这一模式归因于思维链所创造的扩展推理表面。随着对话长度的增加，泄露现象加剧，从1轮到6轮增加了最多3倍。我们的结果表明，任务能力与隐私泄露之间的耦合并不是配置错误的产物，而是当前代理系统的结构特征。

View on arXiv Download PDF AI Translation

cs.CL / 3 / 2610.14073

Is the Future Predictable? Modeling Future Predictability for Reliable Temporal Reasoning

未来是可预测的吗？可靠时间推理的未来可预测性建模

Huang, Yongyi, Zhang, Huaping, Cui, Wenyao, Li, Qiuchi, Li, Lei, Zhang, Baohua

Abstract

Large language models (LLMs) remain unreliable when reasoning about the future. Existing models often overlook the difference among future states, committing to specific answers even when the queried facts are inherently volatile. To capture this challenge, we introduce ChronoRegime, a future-oriented temporal reasoning task that requires models to determine whether historical temporal evidence justifies commitment to a future answer. Furthermore, we propose Confidence-Aware Temporal Reasoning (CTR), a framework that estimates future predictability from temporal dynamics and regulates answer commitment accordingly. Rather than performing unconditional forecasting, CTR commits only when the reconstructed evidence supports the target future state. Across multiple LLMs, CTR improves strict overall accuracy by 6.7 percentage points and reduces hallucination rates by 18.4 percentage points on average over the plain QA baseline. These results demonstrate that explicit predictability assessment is an effective mechanism for safer and more reliable temporal reasoning under future uncertainty.

Chinese Translation

大型语言模型（LLMs）在推理未来时仍然不可靠。现有模型往往忽视未来状态之间的差异，即使在查询的事实本质上是易变的情况下，也会坚持特定的答案。为了应对这一挑战，我们引入了ChronoRegime，一个面向未来的时间推理任务，要求模型判断历史时间证据是否足以支持对未来答案的承诺。此外，我们提出了基于信心的时间推理（Confidence-Aware Temporal Reasoning，CTR）框架，该框架从时间动态中估计未来可预测性，并相应地调节答案的承诺。CTR并不是进行无条件的预测，而是在重建的证据支持目标未来状态时才进行承诺。在多个LLMs上，CTR将严格整体准确率提高了6.7个百分点，并在普通问答基线的基础上平均减少了18.4个百分点的幻觉率。这些结果表明，明确的可预测性评估是应对未来不确定性下更安全、更可靠的时间推理的有效机制。

View on arXiv Download PDF AI Translation

cs.CL / 4 / 2610.14101

Fin_Agent at FinEval 2026: Task-Aware Methods for Multilingual Financial NLP

Fin_Agent在FinEval 2026：面向任务的方法用于多语言金融自然语言处理

Rashid, Mohotarema, Hossain, Tanzina, Song, Xiaoying, Hong, Lingzi

Abstract

We present Fin_Agent, our submission to the FinEval 2026 shared task on multilingual fi nancial language understanding. Our sys tems address Greek financial named-entity recognition, Japanese implicit communication recognition, and multilingual financial ques tion answering. For Greek NER, we com bine annotation-aware numeric rules with zero shot Qwen2.5-32B classification, achieving 0.9521 public F1 and 0.8884 private F1. For Japanese implicit communication, an inter pretable cue-based ordinal scorer achieves 98.0% test accuracy. For multilingual financial QA, BGE-M3retrieval with Qwen3-32B gen eration achieves 0.3023 and 0.3419 ROUGE-1 F1 on the public and private splits. Key limita tions include dataset-specific rules, handcrafted cues, and limited supervision. Implementa tion code is available https://github.com/ Rashidnira/finagent-fineval-2026.

Chinese Translation

我们介绍了Fin_Agent，这是我们对FinEval 2026多语言金融语言理解共享任务的提交。我们的系统解决了希腊金融命名实体识别、日本隐性沟通识别和多语言金融问答等问题。对于希腊命名实体识别，我们结合了注释感知的数值规则与零样本Qwen2.5-32B分类，达到了0.9521的公共F1和0.8884的私有F1。对于日本隐性沟通，基于可解释线索的序数评分器达到了98.0%的测试准确率。对于多语言金融问答，使用Qwen3-32B生成的BGE-M3检索在公共和私有数据集上分别达到了0.3023和0.3419的ROUGE-1 F1。主要局限性包括数据集特定规则、手工制作的线索和有限的监督。实现代码可在https://github.com/Rashidnira/finagent-fineval-2026获取。

View on arXiv Download PDF AI Translation

cs.CL / 5 / 2610.14112

Does System One Transfer? Calibrated Typed-Decision Models, Label Budgets, and Translation for Indonesian Regional Languages

系统一能否迁移？校准的类型决策模型、标签预算与印尼地区语言的翻译

Fadli, Romi, Rusdah, Utami, Wahyu Ramadani Putri, Amana, Zulfa Malika

Abstract

Calibrated typed-decision ("System One") models return a label and a probability in a single forward pass at a small fraction of the cost of large language models (LLMs), yet their evaluation has been almost entirely English. We report a pre-registered evaluation of the commercial Jev model and the open Laya-multilingual model on Indonesian regional languages, using the parallel NusaX sentiment (12 languages) and NusaTranslation sentiment and emotion (7 languages) test sets. Jev is more often correct on sentences that are lexically closer to Indonesian in all three datasets, and presenting the human Indonesian translation instead raises its accuracy by up to 37.2 points, most in languages distant from Indonesian. Its expected calibration error does not grow on regional languages, but its ability to rank its own errors degrades in all three datasets. Retrained classifiers need more than 500 labels per language to match zero-shot Jev in 9 of 10 NusaX regional languages. A confidence-gated cascade that sends Jev's 10-30% least confident decisions to a local IndoBERT classifier matches or exceeds the better of the two systems, and routing to an LLM instead matches that LLM at 24-77% lower API cost. Of six further commercial LLMs (three Claude and three GPT models), five are 4-14 points more accurate than Jev, at several hundred times the API cost where it is reported, while the smallest (Claude Haiku) is less accurate. We release the pre-registration, code and all predictions.

Chinese Translation

校准的类型决策（“系统一”）模型在一次前向传递中返回标签和概率，其成本仅为大型语言模型（LLMs）的一个小部分，但其评估几乎完全基于英语。我们报告了对商业Jev模型和开放的Laya多语言模型在印尼地区语言上的预注册评估，使用了平行的NusaX情感（12种语言）和NusaTranslation情感与情绪（7种语言）测试集。Jev在所有三个数据集中对与印尼语在词汇上更接近的句子的判断更为准确，且呈现人类印尼语翻译后，其准确率提高了最多37.2个百分点，尤其是在与印尼语相距较远的语言中。其预期校准误差在地区语言上并未增长，但其对自身错误的排名能力在所有三个数据集中都有所下降。重新训练的分类器在10种NusaX地区语言中，有9种需要超过500个标签才能与零样本Jev相匹配。一个信心门控级联，将Jev的10-30%信心最低的决策发送到本地IndoBERT分类器，能够匹配或超过这两个系统中更好的一个，而转向LLM则在API成本上比该LLM低24-77%。在六个进一步的商业LLM中（三个Claude模型和三个GPT模型），五个的准确率比Jev高出4-14个百分点，API成本是其报告的几百倍，而最小的（Claude Haiku）准确率较低。我们发布了预注册、代码和所有预测结果。

View on arXiv Download PDF AI Translation

cs.CL / 6 / 2610.14162

Next-Concept Modeling: Fast, Parallel Coarse-to-Fine Text Generation

下一概念建模：快速、并行的粗到细文本生成

Kumbong, Hermann, Khare, Ishan S., Mlauzi, Iddah, Woen, Andrew, Ding, Charles, Buchanan, E. Kelly, Arora, Simran, Mirhoseini, Azalia, Ré, Christopher

Abstract

Diffusion language models (DLMs) offer parallel sampling for faster generation than autoregressive models, an increasingly important advantage as inference becomes a dominant computational workload. However, DLMs degrade quickly with fewer sampling steps, process the full sequence at every step, and require many passes over the dataset to reach good quality. We introduce Next-Concept Modeling (NCM), a two-stage, coarse-to-fine framework: a multiscale residual VQ-VAE encodes text into a hierarchy of discrete codes, and a transformer generates one level at a time, predicting all codes within a level in parallel. At matched parameter scales, NCM achieves better generative perplexity than the strongest DLM baseline on LM1B while running $21\times$ faster. On OpenWebText, NCM is competitive with the strongest DLM while running $10\times$ faster; on XSum, it matches the strongest DLM's ROUGE-L while being $31\times$ faster. On the reasoning task of Sudoku, NCM outperforms autoregressive and diffusion baselines across all difficulty levels, using fewer function evaluations. On the reasoning task of Sudoku, NCM outperforms autoregressive and diffusion baselines across all difficulty levels, using fewer function evaluations.

Chinese Translation

扩散语言模型（DLMs）提供了比自回归模型更快的并行采样，这一优势在推理成为主要计算负担时愈加重要。然而，DLMs在采样步骤较少时迅速降级，在每一步都处理完整序列，并且需要多次遍历数据集才能达到良好的质量。我们提出了下一概念建模（Next-Concept Modeling, NCM），这是一种两阶段的粗到细框架：多尺度残差 VQ-VAE 将文本编码为离散代码的层次结构，而变换器（transformer）则逐层生成，每层并行预测所有代码。在匹配的参数规模下，NCM 在 LM1B 上的生成困惑度优于最强 DLM 基线，同时运行速度快 $21 imes$。在 OpenWebText 上，NCM 与最强 DLM 竞争，同时运行速度快 $10 imes$；在 XSum 上，它的 ROUGE-L 指标与最强 DLM 相当，但速度快 $31 imes$。在数独推理任务中，NCM 在所有难度级别上均优于自回归和扩散基线，且使用的函数评估次数更少。

View on arXiv Download PDF AI Translation

cs.CL / 7 / 2610.14212

Modeling the Arrival Time of Information-Bearing Events in Natural Language

自然语言中信息承载事件到达时间的建模

Lao, Leyi, Ou, Yulin, Xu, Yang, An, Hao

Abstract

Understanding how information is distributed over time is fundamental to modeling natural language. While existing empirical studies primarily rely on post-hoc analysis of information-theoretic metrics, such as surprisal, little is known about the generation processes under the surface. We propose a novel approach for addressing this gap using the tool of neural temporal point processes (TPPs) by modeling the key information-bearing events in text sequences with learnable event intensity functions. Our model takes the history of event intervals and the contextual semantic representations from LLMs as input, and predicts the future arrival times for two types of events: new-word and high-surprisal-token. The experimental results reveal an interesting distinction in event intensities: LLMs tend to produce new words at significantly higher rates than humans, while high-surprisal tokens do not exhibit the difference. A decaying pattern of intensity over time is found in new-word events for both humans and LLMs, but not in high-surprisal-token events. Our findings revisit and extend previous studies that use simpler methods; provide supplementary evidence for theories about constrained context representation inherent in human language production; and find a new angle to view human and LLM generation differences.

Chinese Translation

理解信息如何随时间分布是建模自然语言的基础。尽管现有的实证研究主要依赖于对信息论指标（如惊讶度）的事后分析，但对表面之下的生成过程知之甚少。我们提出了一种新颖的方法，通过使用神经时间点过程（TPPs）工具来填补这一空白，利用可学习的事件强度函数对文本序列中的关键信息承载事件进行建模。我们的模型以事件间隔的历史和来自大型语言模型（LLMs）的上下文语义表示作为输入，预测两种类型事件的未来到达时间：新词和高惊讶度标记。实验结果揭示了事件强度的有趣区别：LLMs产生新词的速率显著高于人类，而高惊讶度标记则没有表现出这种差异。我们发现新词事件在时间上呈现出强度衰减的模式，无论是对于人类还是LLMs，但高惊讶度标记事件则没有这种模式。我们的发现重新审视并扩展了使用更简单方法的先前研究；为关于人类语言生成中固有的受限上下文表征的理论提供了补充证据；并为人类与LLMs生成差异提供了新的视角。

View on arXiv Download PDF AI Translation

cs.CL / 8 / 2610.14238

Agentic Data Engineering for LLMs: System Design and a Controlled Empirical Study

面向大语言模型的主动数据工程：系统设计与受控实证研究

Ding, Ruomeng, Shen, Qianli, Han, Zhaoyang, Chen, Meixin, Chen, Daoyuan, Li, Yaliang

Abstract

LLM data engineering requires iterative experimentation to determine how data selection, rewards, and curricula affect model learning, making it a natural setting for AI-for-AI (AI4AI) research. Can agents use experimental feedback to progressively improve strategies, and do the resulting strategies improve target-model performance on held-out data? We study how agents use experimental feedback to develop data-engineering strategies under fixed training and evaluation protocols. Our Agentic Data Engineering (ADE) framework lets agents pursue, revise, or combine research directions using cumulative findings. Strategies are selected by in-loop scores and evaluated on held-out benchmarks. Across four tasks-math and code data selection, math reward design, and math curriculum learning-selected strategies achieve absolute gains over generic baselines of up to 13.34% on AIME25 Pass@32 and 5.85% averaged across four held-out benchmarks under fixed training conditions. In-loop scores improve over successive experiments, while held-out trajectories show that later in-loop gains do not always improve generalization. Comparisons of research organization and feedback access favor combining evidence analysis with cumulative findings. Research trajectories show how earlier findings inform strategy revisions and combinations across research directions.

Chinese Translation

大语言模型（LLM）的数据工程需要通过迭代实验来确定数据选择、奖励和课程如何影响模型学习，这使其成为人工智能为人工智能（AI-for-AI，AI4AI）研究的自然场景。代理是否能够利用实验反馈逐步改进策略，并且所产生的策略是否能提高目标模型在保留数据上的表现？我们研究了代理在固定训练和评估协议下如何利用实验反馈来开发数据工程策略。我们的主动数据工程（Agentic Data Engineering，ADE）框架允许代理利用累积发现来追求、修订或结合研究方向。策略通过循环内得分进行选择，并在保留基准上进行评估。在四个任务中——数学和代码数据选择、数学奖励设计以及数学课程学习——所选策略在固定训练条件下在 AIME25 Pass@32 上实现了相对于通用基准最高达 13.34% 的绝对增益，且在四个保留基准上平均增益达 5.85%。循环内得分在连续实验中有所提高，而保留轨迹显示后期循环内增益并不总是改善泛化。研究组织和反馈获取的比较支持将证据分析与累积发现相结合。研究轨迹展示了早期发现如何影响策略的修订和跨研究方向的组合。

View on arXiv Download PDF AI Translation

cs.CL / 9 / 2610.14270

The Thinking Leakage in Reinforcement Learning for LLMs

大语言模型强化学习中的思维泄漏

Zhu, Wenxuan, Liu, Shuming, An, Bang, Aliev, Aznaur, Li, Bing, Yang, Yibo, Ghanem, Bernard

Abstract

Hybrid reasoning models such as the Qwen3 series switch between a THINK mode and a NO-THINK mode through a few chat-template tokens. A popular recipe trains such models with reinforcement learning in NO-THINK mode and reports improved reasoning performance. We provide evidence that much of this gain can be explained by thinking leakage: the gain comes largely from partially reactivating the reasoning abilities of the original model's THINK mode instead of learning enhanced reasoning behavior. We train Qwen3-1.7B, 4B, and 8B with GRPO in NO-THINK mode. As accuracy improves, responses become longer and more reflective, are increasingly judged by the original model to resemble its own thinking, and the model's internal activations shift toward its THINK mode. We then investigate at the level of weights, activations, and attention heads: scaling the RL update further elicits THINK-mode behavior and approaches THINK-mode accuracy but does not exceed it; in Qwen3-8B, replacing a low-dimensional THINK-related activation subspace of the RL-trained model with its original-model value removes most of the RL gain, and the reverse exchange transfers a considerable fraction of the gain to the original model; and the RL-induced shift passes through the attention heads that read the empty think block. We further identify a possible mechanism in the GRPO training signal: responses favored by the advantage-weighted update are more THINK-like, and a gradient step computed from this signal moves the model toward its THINK state. Thus, gains in NO-THINK mode can largely reflect reactivation of reasoning abilities already present in the original model.

Chinese Translation

混合推理模型如 Qwen3 系列通过少量聊天模板标记在思维模式（THINK mode）和非思维模式（NO-THINK mode）之间切换。一种流行的方法是在非思维模式下使用强化学习训练这些模型，并报告推理性能的提升。我们提供证据表明，这一增益在很大程度上可以用思维泄漏来解释：增益主要来自于部分重新激活原始模型思维模式的推理能力，而非学习增强的推理行为。我们在非思维模式下使用 GRPO 训练 Qwen3-1.7B、4B 和 8B。随着准确性的提高，模型的响应变得更长且更具反思性，越来越多地被原始模型判断为类似于其自身的思维，模型的内部激活也向其思维模式转移。然后，我们在权重、激活和注意力头的层面进行调查：进一步扩展强化学习更新会引发思维模式行为，并接近思维模式的准确性，但不会超过它；在 Qwen3-8B 中，用原始模型值替换 RL 训练模型的低维思维相关激活子空间会消除大部分 RL 增益，而反向交换则将相当一部分增益转移到原始模型；而 RL 引发的转变通过读取空思维块的注意力头进行。我们进一步识别出 GRPO 训练信号中的一个可能机制：通过优势加权更新所偏好的响应更具思维特征，而从该信号计算的梯度步骤使模型向其思维状态移动。因此，非思维模式下的增益在很大程度上可以反映原始模型中已经存在的推理能力的重新激活。

View on arXiv Download PDF AI Translation

cs.CL / 10 / 2610.14309

ASSET: Adaptive Speech Segmentation via Reinforcement Learning

ASSET：通过强化学习实现自适应语音分段

Xiao, Cihan, Tseng, Yuan, Wiesner, Matthew, Raj, Desh, Perera, Leibny Paola Garcia, Khudanpur, Sanjeev, Harwath, David

Abstract

Speech large language models (LLMs) are often built using pretrained acoustic encoders as an audio interface to an LLM decoder. However, LLMs scale poorly with input sequence length, while acoustic encoders tend to produce substantially longer sequences than typical textual inputs. This motivates learning to compress encoder outputs while preserving task-oriented information. To this end, we introduce ASSET, a reinforcement learning (RL)-based framework that learns to adaptively compress encoder outputs by first partitioning them into variable-duration segments and then pooling each segment into a continuous audio token. We use group-relative policy optimization (GRPO) to train a segmenter with task performance feedback and a token-rate penalty. In automatic speech recognition (ASR) experiments, ASSET achieves performance comparable to a transcript-assisted character-boundary oracle with approximately 25% fewer audio tokens on English LibriSpeech, while improving on a phone-boundary oracle at a similar token rate on Mandarin AISHELL. Exploratory experiments on non-ASR tasks further illustrate how learned segmentation changes when the framework is adapted to tasks beyond transcription.

Chinese Translation

语音大型语言模型（LLMs）通常使用预训练的声学编码器作为LLM解码器的音频接口。然而，LLMs在输入序列长度上扩展性较差，而声学编码器往往生成比典型文本输入长得多的序列。这促使我们学习在保留任务导向信息的同时压缩编码器输出。为此，我们提出了ASSET，一个基于强化学习（RL）的框架，通过首先将编码器输出划分为可变时长的片段，然后将每个片段汇聚为一个连续的音频标记，来学习自适应压缩编码器输出。我们使用组相对策略优化（GRPO）来训练一个分段器，该分段器结合了任务性能反馈和标记率惩罚。在自动语音识别（ASR）实验中，ASSET的性能与转录辅助的字符边界oracle相当，同时在英语LibriSpeech上减少了约25%的音频标记，而在普通话AISHELL上以类似的标记率改善了音素边界oracle。对非ASR任务的探索性实验进一步说明了当框架适应于超出转录的任务时，学习到的分段方式是如何变化的。

View on arXiv Download PDF AI Translation

cs.CL / 11 / 2610.14346

Evaluating LLMs as Agentic Workflow Executors in Real-World Customer Support Skills

评估大型语言模型作为现实世界客户支持技能的自主工作流执行者

Laskar, Md Tahmid Rahman, Singh, Gundeep, Fu, Xue-Yong, Chang, Karol, Sanders, Kevin, Zong, Shi, Habib, Tania, Tremblay, Julien Bouvier, Gardiner, Shayna, Saini, Harsh, Lee, Matthias, Khasanova, Elena, McNamara, Quinten, TN, Shashi Bhushan

Abstract

Enterprise customer-support agents increasingly rely on Large Language Models (LLMs) to execute bounded, tool-using workflows. We study this problem through the lens of Skill LLMs: specialized executors that operate within a selected skill context and must complete a narrow, stateful workflow under policy and tool constraints. We present a benchmark construction pipeline that mines recurrent workflows from real-world transcripts, converts them into structured skill descriptions, maps them to tool requirements, simulates environment state with Model Context Protocol (MCP)-compatible mocks, and generates multi-turn tool-using dialogues. Using this benchmark, we conduct a large-scale zero-shot evaluation of 17 language models spanning closed model families and deployment-oriented open-weight models from 1.7B to 14B parameters, each judged independently by five strong LLM judges, plus a deterministic rule-based evaluator that checks tool coverage, hallucinated tool use, and unsupported claims. We find a consistent ranking in which GPT-5.4 leads by a wide margin and QWEN3.5-9B is the strongest open-weight model (below 20B), outperforming several closed baselines under every judge. We also find that judge choice is not a negligible methodological detail: pairwise judge agreement on the pass/fail verdict averages Kappa=0.524 across 10 judge pairs. These results suggest that comprehensive zero-shot evaluation of skill execution requires judge reliability as carefully as model capability.

Chinese Translation

企业客户支持代理越来越依赖大型语言模型（LLMs）来执行有限的、使用工具的工作流。我们通过技能 LLMs 的视角研究这个问题：这些专门的执行者在选定的技能上下文中操作，必须在政策和工具约束下完成狭窄的、有状态的工作流。我们提出了一种基准构建管道，该管道从真实世界的转录中挖掘重复工作流，将其转换为结构化的技能描述，映射到工具需求，使用与模型上下文协议（MCP）兼容的模拟环境状态，并生成多轮使用工具的对话。利用该基准，我们对17个语言模型进行了大规模的零-shot 评估，这些模型涵盖了封闭模型系列和面向部署的开放权重模型，参数范围从1.7B到14B，每个模型都由五位强大的 LLM 评审独立评判，此外还有一个确定性的基于规则的评估者检查工具覆盖、虚构工具使用和不支持的声明。我们发现了一致的排名，其中 GPT-5.4 以较大优势领先，而 QWEN3.5-9B 是最强的开放权重模型（低于20B），在每位评审下都超越了几个封闭基准。我们还发现，评审选择并不是一个可以忽视的方法论细节：10对评审的通过/失败裁决的成对一致性平均 Kappa=0.524。这些结果表明，全面的零-shot 技能执行评估需要评审的可靠性与模型能力同样重要。

View on arXiv Download PDF AI Translation

cs.CL / 12 / 2610.14349

Multimodal Co-Generation With Continuous Bitstream Diffusion

基于连续比特流扩散的多模态共同生成

Batzolis, Georgios, Kontos, Georgios, Trentini, Bruno, Chatterjee, Pranam, Livne, Micha, Girolami, Mark, Ambrogioni, Luca

Abstract

We show that continuous bitstream diffusion (CoBit), originally developed for language modeling, provides a unified framework for joint and conditional multimodal generation using the same training and sampling protocol. Across image-text, speech-text and protein sequence-structure modeling, models are trained separately from scratch, but each uses a single input projection, transformer backbone, bitwise prediction head and binary score-matching objective shared across its modalities. Within each setting, a single checkpoint supports joint generation and the conditional directions included in training by changing which parts of the bitstream are observed. We first validate the approach on MNIST-Sum, a controlled image-text task pairing raw binary images with equations encoded as text, achieving near-perfect conditional accuracy and high joint consistency without learned tokenizers. We then apply the same protocol at larger scale to speech-text and protein sequence-structure modeling. Trained without text pretraining, a single speech-text model performs speech synthesis, recognition, continuation, and unconditional co-generation. Jointly generated speech and text achieve 3.5% word error rate between the generated text and automatic transcriptions of the generated speech. In protein modeling, a single CoBit checkpoint supports forward folding, inverse folding, and joint sequence-structure generation. For jointly generated proteins spanning 100-500 residues, computational refolding of CoBit's sequences yields structures that agree with the generated backbones at a mean self-consistency TM-score of 0.911. Together, these results show that CoBit provides a common generative framework for substantially different modalities and forms of cross-modal dependence, without introducing a domain-specific objective or sampling procedure.

Chinese Translation

我们展示了最初为语言建模开发的连续比特流扩散（CoBit）为联合和条件多模态生成提供了一个统一框架，使用相同的训练和采样协议。在图像-文本、语音-文本和蛋白质序列-结构建模中，模型从零开始分别训练，但每个模型都使用单一的输入投影、变换器骨干、逐位预测头和跨模态共享的二进制评分匹配目标。在每个设置中，一个单一的检查点支持联合生成和通过改变观察比特流的哪些部分来包含在训练中的条件方向。我们首先在MNIST-Sum上验证了该方法，这是一个受控的图像-文本任务，将原始二进制图像与编码为文本的方程配对，实现了近乎完美的条件准确性和高联合一致性，而无需学习的标记器。随后，我们在更大规模上将相同的协议应用于语音-文本和蛋白质序列-结构建模。在没有文本预训练的情况下，单一的语音-文本模型执行语音合成、识别、延续和无条件共同生成。联合生成的语音和文本之间的字词错误率为3.5%，与生成语音的自动转录结果相比。在蛋白质建模中，单一的CoBit检查点支持前向折叠、逆向折叠和联合序列-结构生成。对于跨越100-500个残基的联合生成蛋白质，CoBit序列的计算重新折叠产生的结构在平均自一致性TM-score为0.911的情况下与生成的骨架一致。这些结果表明，CoBit为实质上不同的模态和跨模态依赖形式提供了一个共同的生成框架，而无需引入特定于领域的目标或采样程序。

View on arXiv Download PDF AI Translation

cs.CL / 13 / 2610.14393

SciCurateBench: Curating Experimental Data from Heterogeneous Scientific Literature with Verbal-Action Agents

SciCurateBench：利用语言行动代理从异构科学文献中整理实验数据

Ma, Minuk, Guo, Kaiyun, Ding, Jiarui

Abstract

Scientific research increasingly depends on curating experimental data from literature. Scientific data curation requires verifying exact experimental conditions (e.g., specific assays, clinical outcomes, and data availability) across long, heterogeneous files, which is a highly time-consuming process. Existing benchmarks evaluate only broad relevance on short queries, failing to capture this complexity. We introduce SciCurateBench, a benchmark of 200 expert-annotated papers from major life-science journals that challenges systems to verify conjunctive biological experimental criteria from raw, unprocessed scientific literature. We show that plain LLMs and popular reasoning, tool-using, and feedback-based agents all struggle with this task. To study what makes the task tractable, we further propose LangPoli, an integrative system in which a policy agent generates natural language action guidelines that actor agents convert into executable tools. LangPoli showed strong performance on both Paper Classification (83.16% F1 vs. 58.22% for its base LLM) and Structured Information Retrieval, and also improves re-ranking on the public benchmarks of LitSearch, BEIR, and TREC. We further explore curriculum transfer, a method that distills generalizable strategies from solved tasks to accelerate the solving of related problems. We hope this work facilitates research in truly practical scientific AI assistance. The anonymized pre-release can be found from https://anonymous.4open. science/r/scicuratebench-D885/.

Chinese Translation

科学研究越来越依赖于从文献中整理实验数据。科学数据整理需要验证长篇异构文件中的确切实验条件（例如，特定的检测方法、临床结果和数据可用性），这是一项非常耗时的过程。现有基准仅评估短查询的广泛相关性，未能捕捉到这一复杂性。我们提出了SciCurateBench，这是一个由200篇来自主要生命科学期刊的专家注释论文组成的基准，挑战系统从原始未处理的科学文献中验证生物实验的联合标准。我们展示了普通的语言模型（LLMs）以及流行的推理、工具使用和基于反馈的代理在这一任务上都面临困难。为了研究使任务可处理的因素，我们进一步提出了LangPoli，这是一个综合系统，其中政策代理生成自然语言行动指南，行动代理将其转化为可执行工具。LangPoli在论文分类（83.16%的F1得分对比其基础LLM的58.22%）和结构化信息检索上表现出色，并且在LitSearch、BEIR和TREC的公共基准上也提高了重排序效果。我们进一步探索了课程迁移，这是一种从已解决任务中提炼可推广策略以加速相关问题解决的方法。我们希望这项工作能促进真正实用的科学人工智能助手的研究。匿名预发布版本可在 https://anonymous.4open.science/r/scicuratebench-D885/ 找到。

View on arXiv Download PDF AI Translation

cs.CL / 14 / 2610.14423

Scaling Laws for Multimodal Data Mixtures

多模态数据混合的规模法则

Khandelwal, Aditi, Tarun, Ayush K., Xu, Yixuan, Schlag, Imanol, Farnadi, Golnoosh, Reddy, Siva, Zettlemoyer, Luke, Bosselut, Antoine

Abstract

Frontier AI systems are increasingly natively multimodal, jointly pretrained on multiple modalities, such as text, vision, and audio. However, the scientific understanding of the multimodal data mixtures used to pretrain these systems --- how much of each modality to mix, and at what potential interference cost to the others --- is largely absent or gatekept. In this work, we derive data mixing scaling laws for three modalities: text, vision, and audio. Multimodal Large Language Models (MLLMs) increasingly adopt Mixture-of-Experts (MoE) architectures, as MoEs scale efficiently and partially limit cross-modal interference. We therefore use MoEs to conduct 268 experiments ranging from 457M to 8.3B parameters, on up to 150 billion tokens. Our work leads to multiple actionable insights into multimodal data mixing for training MLLMs. First, compute optima are modality-specific: audio modality is model-heavy, favoring scaling parameters over tokens, while text and vision require balanced scaling of model size and data. Second, repeating vision or audio data beyond 2x yields negligible benefits. Finally, cross-modal interactions are highly asymmetric: vision strongly interferes with audio, whereas audio's interference with learning text and vision is negligible. Overall, we provide a principled foundation for understanding the multimodal data mixtures needed to train frontier MLLMs.

Chinese Translation

前沿人工智能系统越来越多地原生支持多模态，联合预训练于多种模态，如文本、视觉和音频。然而，对于用于预训练这些系统的多模态数据混合的科学理解——每种模态的混合比例，以及对其他模态可能造成的干扰成本——在很大程度上缺乏或受到限制。在本研究中，我们推导了三种模态（文本、视觉和音频）的数据混合规模法则。多模态大型语言模型（MLLMs）越来越多地采用专家混合（Mixture-of-Experts, MoE）架构，因为MoE能够高效扩展并部分限制跨模态干扰。因此，我们使用MoE进行了268个实验，参数范围从4.57亿到83亿，处理多达1500亿个标记。我们的研究为训练MLLMs的多模态数据混合提供了多个可操作的见解。首先，计算最优解是模态特定的：音频模态对模型的需求较高，倾向于在参数上进行扩展，而文本和视觉则需要平衡模型大小和数据的扩展。其次，重复视觉或音频数据超过2倍几乎不会带来显著收益。最后，跨模态交互高度不对称：视觉对音频的干扰较强，而音频对学习文本和视觉的干扰则微不足道。总体而言，我们为理解训练前沿MLLMs所需的多模态数据混合提供了一个原则性基础。

View on arXiv Download PDF AI Translation

cs.CL / 15 / 2610.14438

Neither True nor False: Do LLMs Represent Truth-Value Gaps in Plural Predication?

既非真也非假：大型语言模型是否代表了复数谓词中的真值缺口？

Nakshbandi, Duaa, An, Zhiyu, Li, Yiping, Giri, Sachin, Du, Wan

Abstract

Do Large Language Models (LLMs) genuinely grasp the truth conditions of plural descriptions, or do they force every sentence into true or false? We investigate the Homogeneity Gap, a semantic phenomenon where a plural definite (e.g., the kids laughed) and its negation (the kids didn't laugh) are both neither true nor false when only some of the kids laughed, whereas their quantified counterparts (all, not all) remain strictly bivalent. We introduce HomogeneityNLI, a diagnostic dataset of 1,000 minimal-pair items crossing determiner, polarity, and group composition across ten logical conditions. Evaluating seven open-weight models, we uncover a pervasive Bivalence Bias: models assign plural definites their homogeneous truth conditions, reading the kids as all of them and the kids didn't as none of them, yet collapse the truth-value gap into falsity. In the zero-shot setting, no model recognizes the gap for both polarities in more than 15% of scenarios, and negated definites are judged False in 84% of cases. Explicit linguistic rules, chain-of-thought, and enumeration prompts produce unstable, model-specific effects that often entrench the bias. Taken together, our findings indicate that current open-weight LLMs have acquired the truth conditions of plural predication but not its trivalent logic.

Chinese Translation

大型语言模型（LLMs）是否真正理解复数描述的真值条件，还是将每个句子强行归为真或假？我们研究了同质性缺口（Homogeneity Gap），这是一种语义现象，当只有部分孩子笑时，复数确定性（例如，孩子们笑了）及其否定（孩子们没有笑）都既不为真也不为假，而它们的量化对应物（所有，非所有）则保持严格的二值性。我们引入了HomogeneityNLI，这是一个包含1000个最小对比项的诊断数据集，跨越十个逻辑条件，涵盖了限定词、极性和群体组成。通过评估七个开放权重模型，我们发现了一种普遍的二值偏差（Bivalence Bias）：模型将复数确定性分配给其同质的真值条件，将“孩子们”理解为所有孩子，而“孩子们没有”理解为没有一个孩子，但却将真值缺口归结为虚假。在零样本设置中，没有模型在超过15%的场景中识别出两种极性的缺口，而否定的确定性在84%的情况下被判断为假。显式语言规则、思维链和列举提示产生了不稳定的、特定于模型的效果，往往加深了这种偏差。综合来看，我们的发现表明，当前的开放权重LLMs已经掌握了复数谓词的真值条件，但并未掌握其三值逻辑。

View on arXiv Download PDF AI Translation

cs.CL / 16 / 2610.14442

Evaluating Semantic Relation Knowledge through Bidirectional Relation Prediction from Gloss Pairs

通过双向关系预测评估语义关系知识：基于释义对的研究

Cao, Zhihan, Nohejl, Adam, Yanaka, Hitomi

Abstract

Semantic relations organise word senses and are fundamental to lexical semantics. In order to evaluate semantic relation knowledge of large language models (LLMs), we propose a challenging task in which an LLM predicts the relation between two glosses (definitions of senses) presented in two different orders. The predictions for the two orders are evaluated for consistency with the symmetry or asymmetry of semantic relations. Our task is designed for both monolingual and cross-lingual evaluation, allowing us to examine whether relation knowledge remains accessible when the same senses are verbalised through glosses in different languages. We construct a dataset of 3,069 sense pairs, with each sense verbalised in English, Greek, Romanian, Turkish, and Japanese. Experiments with 17 LLMs in zero-shot and five-shot settings reveal generally poor performance. In each setting, only eight LLMs achieve a macro-averaged F1 score significantly above a random baseline, and only one (zero-shot) and two (five-shot) significantly outperform a neural baseline. In particular, LLMs often fail to generate bidirectionally consistent predictions for hypernymy and to make correct predictions for holonymy in either direction. Our results suggest limitations in current LLMs' semantic relation knowledge, both in identifying relations and in respecting asymmetry.

Chinese Translation

语义关系组织词义，是词汇语义学的基础。为了评估大型语言模型（LLMs）的语义关系知识，我们提出了一项具有挑战性的任务，其中LLM预测以两种不同顺序呈现的两个释义（词义的定义）之间的关系。对于这两种顺序的预测结果进行评估，以检验其与语义关系的对称性或不对称性的一致性。我们的任务旨在进行单语和跨语言评估，使我们能够检验当相同的词义通过不同语言的释义表达时，关系知识是否仍然可获取。我们构建了一个包含3,069对词义的数据集，每个词义以英语、希腊语、罗马尼亚语、土耳其语和日语进行表达。在零样本和五样本设置下对17个LLM进行的实验显示出整体表现较差。在每种设置中，仅有八个LLM的宏平均F1分数显著高于随机基线，且仅有一个（零样本）和两个（五样本）显著优于神经基线。特别是，LLMs通常未能为上位词关系生成双向一致的预测，并且在任何方向上都未能正确预测整体词关系。我们的结果表明，当前LLMs在识别关系和尊重不对称性方面的语义关系知识存在局限性。

View on arXiv Download PDF AI Translation

cs.CL / 17 / 2610.14464

Evaluating Bias in Chinese Large Language Models across Tasks and Registers: Task Operationalization Effects and Irony Classification Risks

评估中国大型语言模型在任务和语域中的偏见：任务操作化效应与讽刺分类风险

Zhou, Mo

Abstract

In Chinese content governance, large language models are increasingly used to identify toxic and discriminatory expressions, yet both task operationalization and ironic expression can alter bias measurement readings. We conducted a preregistered multi-task behavioral audit of multiple Chinese large language models. In the content classification experiment, the core panel's classification error rate rose from 10.9% in the formal register to 23.5% in the yinyang (an echoic-irony register) version, a risk difference of +12.7 percentage points. (95% CI [+9.7, +15.7]). Inter-rater agreement was lower in the yinyang register (majority-vote kappa = 0.299), so the increase contains a component of label uncertainty propagation, but the direction of increase was consistent under both adjudicated and majority-vote reference standards. Explicit irony-aware prompts reduced the intervention panel's error rate by 9.6 percentage points. We operationalized the 29 shared proposition anchors into three task forms: social consensus estimation (Form A), high-stakes decision (Form B), and controlled association rating (Form C); independent manipulation checks found that all 58 A-C comparison units received unanimous agreement, whereas the low A-B convergence was confirmed by calibrated majority adjudication (87.8% supporting the minority judgment); on the irony-probe materials, the interactions of B and C relative to A were +0.577 and +1.121, respectively. Thus, when evaluating large language models for Chinese content governance, the effect of task operationalization on bias readings is an empirical question that requires data to answer, and safety evaluation should incorporate the ironic register as a test condition.

Chinese Translation

在中国内容治理中，大型语言模型越来越多地被用于识别有毒和歧视性表达，然而任务操作化和讽刺表达都可能影响偏见测量结果。我们进行了一个预注册的多任务行为审计，评估了多个中国大型语言模型。在内容分类实验中，核心小组的分类错误率在正式语域中为10.9%，而在阴阳（回声讽刺语域）版本中上升至23.5%，风险差异为+12.7个百分点（95% CI [+9.7, +15.7]）。在阴阳语域中的评审者一致性较低（多数投票kappa = 0.299），因此这一增加包含标签不确定性传播的成分，但在裁定和多数投票参考标准下，增加的方向是一致的。显式的讽刺意识提示将干预小组的错误率降低了9.6个百分点。我们将29个共享命题锚点操作化为三种任务形式：社会共识估计（形式A）、高风险决策（形式B）和控制关联评分（形式C）；独立的操作检查发现，所有58个A-C比较单位均获得一致同意，而低A-B收敛性则通过校准的多数裁定得到确认（87.8%支持少数判断）；在讽刺探测材料上，B和C相对于A的交互作用分别为+0.577和+1.121。因此，在评估大型语言模型用于中国内容治理时，任务操作化对偏见读数的影响是一个需要数据回答的实证问题，而安全评估应将讽刺语域纳入测试条件。

View on arXiv Download PDF AI Translation

cs.CL / 18 / 2610.14475

EdgeLLM-Nano: Auditing Polar-Guided Key-Cache Compression and Consistency-Anchored Pruning on a 33.6 MB On-Device Language Model

EdgeLLM-Nano：审计极性引导的关键缓存压缩与一致性锚定的剪枝在33.6 MB设备端语言模型上的应用

Murshed, S M Mahbub, Bright, Hayden, Karim, Rezaul, Sichani, Ali Shiri

Abstract

Deploying language models (LMs) on the edge presents unique memory, latency, and power challenges absent in high-performance computing. KV-cache compression emerges as a promising approach to address these challenges. However, the effectiveness of current KV-cache compression methods has not been thoroughly examined on edge devices. Moreover, we show that existing evaluation methods do not always involve the KV-cache under test. We propose EdgeLLM-Nano, a measurement study of KV-cache compression using a nano-scale decoder-only LM (transformer) testbed that has 8.39M parameters, i.e., 33.6 MB, to enable end-to-end examination of each backend, seed, and power sample. Our results give a negative verdict for a polar-guided key-cache quantizer (round-to-nearest (RTN) control causes an order of magnitude less damage), as well as a consistency-anchored progressive weight pruner (one-shot magnitude-based pruning performs as well as, if not better). We also introduce Long-Context Decode Fidelity (LCDF) evaluation, a token-by-token examination of teacher-forced log-losses using the KV-cache under test. Unlike perplexity that gives the same 5.461 score to all methods, LCDF distinguishes different KV-cache compression backends thanks to its paired statistics across seeds and effect sizes. Our companion Raspberry Pi 5 measurement shows all compression backends we tested to be less energy-efficient per token compared to uncompressed FP16 inference. Except for KIVI that is close to lossless in terms of LCDF, our findings indicate KV-cache compression can be improved. LCDF offers a simple and effective way of shedding light on KV-cache compression methods, whose negative impacts have been overlooked by perplexity.

Chinese Translation

在边缘设备上部署语言模型（LM）面临独特的内存、延迟和功耗挑战，这些挑战在高性能计算中并不存在。KV缓存压缩作为一种有前景的方法来解决这些挑战。然而，目前的KV缓存压缩方法在边缘设备上的有效性尚未得到充分检验。此外，我们还表明，现有的评估方法并不总是涉及待测试的KV缓存。我们提出了EdgeLLM-Nano，这是一个使用具有8.39M参数（即33.6 MB）的纳米级解码器仅模型（transformer）测试平台进行KV缓存压缩的测量研究，以便对每个后端、种子和功率样本进行端到端的检查。我们的结果对极性引导的关键缓存量化器（四舍五入控制造成的损害数量级减少）以及一致性锚定的渐进式权重剪枝（一次性基于幅度的剪枝表现与之相当，甚至更好）给出了否定的结论。我们还引入了长上下文解码保真度（Long-Context Decode Fidelity, LCDF）评估，这是对使用待测试KV缓存的教师强制对数损失的逐个标记检查。与给所有方法相同的5.461分的困惑度不同，LCDF由于其跨种子和效应大小的配对统计，能够区分不同的KV缓存压缩后端。我们的伴随Raspberry Pi 5测量显示，所有测试的压缩后端在每个标记的能效上均低于未压缩的FP16推理。除了在LCDF方面接近无损的KIVI外，我们的研究结果表明KV缓存压缩还有改进的空间。LCDF为揭示KV缓存压缩方法提供了一种简单有效的方式，而这些方法的负面影响常常被困惑度所忽视。

View on arXiv Download PDF AI Translation

cs.CL / 19 / 2610.14476

Scale, Drivers, and Limits of Representational Alignment Across Large Language Models

大语言模型中表征对齐的规模、驱动因素及其限制

Wu, Jialin, Khosla, Meenakshi

Abstract

Comparing large language models can reveal shared representational structure and systematic differences between them. However, aggregate similarity measures can obscure how agreement varies across geometric scales and which linguistic properties support it. We introduce two complementary methods to resolve where models agree and what supports that agreement. Standard representational similarity analysis (RSA) summarizes similarity across all stimulus pairs, potentially obscuring differences in alignment between nearby and distant inputs. Neighborhood-Weighted RSA (NW-RSA) addresses this limitation by weighting stimulus pairs according to their proximity in representational space, placing alignment from local neighborhoods to global geometry on a common scale. Counterfactual RSA (CF-RSA) tests which linguistic properties support alignment by comparing intact representations in one model with representations under controlled linguistic perturbations in another. Across diverse language models, we find strong agreement in coarse representational geometry but substantial divergence in local neighborhoods. Alignment is observed both across broadly sampled natural text and within linguistic equivalence classes that preserve particular aspects of meaning or structure. Counterfactual analyses reveal selective sensitivity: changes to logical scope reduce alignment substantially more than the tested changes to hierarchical syntax, event roles, or discourse structure. Alignment also decreases when sentence structure or lexical meaning is disrupted in word lists and jabberwocky sentences, suggesting that the regularities of natural language support stronger convergence. These findings refine claims of representational universality by showing that strong overall similarity need not imply agreement in how models represent finer linguistic distinctions.

Chinese Translation

比较大语言模型可以揭示它们之间共享的表征结构和系统性差异。然而，整体相似性度量可能掩盖了在几何尺度上协议的变化及其支持的语言特性。我们提出了两种互补的方法来解决模型之间的协议及其支持因素。标准的表征相似性分析（RSA）总结了所有刺激对之间的相似性，可能会掩盖附近和远离输入之间对齐的差异。邻域加权RSA（NW-RSA）通过根据刺激对在表征空间中的接近程度加权，解决了这一局限性，将局部邻域到全局几何的对齐置于一个共同的尺度上。反事实RSA（CF-RSA）通过将一个模型中完整的表征与另一个模型在受控语言扰动下的表征进行比较，测试哪些语言特性支持对齐。在多种语言模型中，我们发现粗略的表征几何具有强烈的一致性，但在局部邻域中存在显著的分歧。对齐现象既出现在广泛采样的自然文本中，也出现在保留特定意义或结构方面的语言等价类中。反事实分析揭示了选择性敏感性：逻辑范围的变化比测试的层次句法、事件角色或话语结构的变化显著减少对齐。当句子结构或词汇意义在单词列表和无意义句子中被扰动时，对齐也会下降，这表明自然语言的规律性支持更强的收敛性。这些发现通过表明强大的整体相似性不必意味着模型在如何表征更细微的语言区别上达成一致，从而细化了表征普遍性的主张。

View on arXiv Download PDF AI Translation

cs.CL / 20 / 2610.14479

AphaVoice: Text-to-Speech Model for Aphasia Patient Simulation

AphaVoice：用于失语症患者模拟的文本到语音模型

Kusuma, Al Varrel Putra, Bello, Fernando, Brown, Joshua

Abstract

While language therapy is the primary treatment for post-stroke aphasia, limited clinical resources often restrict patient access to proper care. Addressing this gap requires reliable training simulations for medical professionals. However, current AI efforts focus almost exclusively on speech processing, leaving speech generation largely unexplored. This study introduces AphaVoice, a novel text-to-speech (TTS) framework explicitly engineered to synthesize highly realistic aphasic speech. Evaluated against state-of-the-art TTS systems, AphaVoice demonstrates a superior ability to capture the clinical feature of dysfluency, outperforming existing models in temporal metrics while consistently maintaining speaker identity. By leveraging latent space shifting, the model also supports multi-speaker generation without sacrificing phonetic realism. Ultimately, AphaVoice provides a scalable, high-fidelity tool for medical training and clinical simulation related to aphasia.

Chinese Translation

虽然语言治疗是中风后失语症的主要治疗方法，但有限的临床资源往往限制了患者获得适当护理的机会。解决这一差距需要为医疗专业人员提供可靠的培训模拟。然而，目前的人工智能努力几乎完全专注于语音处理，语音生成仍然未得到充分探索。本研究介绍了AphaVoice，这是一种新颖的文本到语音（TTS）框架，专门设计用于合成高度真实的失语症语音。与最先进的TTS系统进行评估时，AphaVoice展现了更强的捕捉言语不流畅这一临床特征的能力，在时间指标上优于现有模型，同时始终保持说话者的身份。通过利用潜在空间的转移，该模型还支持多说话者生成，而不牺牲语音的真实感。最终，AphaVoice为与失语症相关的医学培训和临床模拟提供了一个可扩展的高保真工具。

View on arXiv Download PDF AI Translation

cs.CL / 21 / 2610.14531

LongHisDoc: A Comprehensive Benchmark for Chinese Long Historical Document Understanding

LongHisDoc：中文长篇历史文献理解的综合基准

Shan, Zeyu, Shi, Yongxin, Wu, Shihang, Zhang, Peirong, Fang, Yuanyuan, Wang, Zhaopeng, Jin, Lianwen

Abstract

Chinese historical documents are invaluable cultural treasures, but their complexity poses significant comprehension challenges for non-specialists. Despite the success of Large Language Models (LLMs) and Large Vision-Language Models (LVLMs) in modern document tasks, their capabilities in Long (full-book) Historical Document understanding remain largely unexplored. To bridge this gap, we introduce LongHisDoc, the first benchmark tailored for long historical document understanding. LongHisDoc comprises 1,012 expert-annotated QA pairs derived from 101 PDF documents, with an average of 85.1 pages and 42,776.4 tokens per document. It spans 10 categories and features four QA types: single-page, cross-page, multi-hop, and unanswerable. Answering these questions requires integrating multimodal evidence from text, layout, and visual elements. Our evaluation reveals that the top-performing LVLM and LLM achieve only 62.9% and 69.9% accuracy, respectively, highlighting a substantial performance gap. Furthermore, we assess various modal retrievers and the end-to-end RAG performance to provide references for Retrieval-Augmented Generation (RAG) systems. We believe that LongHisDoc serves as a comprehensive benchmark to catalyze future research in full-book historical document understanding. The benchmark and evaluation code are available at https://huggingface.co/datasets/qweq12433454/LongHisDoc and https://github.com/SCUT-DLVCLab/LongHisDoc.

Chinese Translation

中文历史文献是无价的文化瑰宝，但其复杂性给非专业人士带来了显著的理解挑战。尽管大型语言模型（Large Language Models, LLMs）和大型视觉语言模型（Large Vision-Language Models, LVLMs）在现代文档任务中取得了成功，但它们在长篇（全书）历史文献理解方面的能力仍然 largely unexplored。为填补这一空白，我们推出了 LongHisDoc，这是首个专为长篇历史文献理解量身定制的基准。LongHisDoc 包含 1,012 个专家标注的问答对，来源于 101 份 PDF 文档，每份文档平均包含 85.1 页和 42,776.4 个标记。该基准涵盖 10 个类别，并具有四种问答类型：单页、跨页、多跳和无法回答。回答这些问题需要整合来自文本、布局和视觉元素的多模态证据。我们的评估显示，表现最佳的 LVLM 和 LLM 的准确率仅为 62.9% 和 69.9%，突显了显著的性能差距。此外，我们评估了各种模态检索器和端到端的检索增强生成（Retrieval-Augmented Generation, RAG）性能，以为 RAG 系统提供参考。我们相信，LongHisDoc 作为一个综合基准，将促进未来在全书历史文献理解领域的研究。基准和评估代码可在 https://huggingface.co/datasets/qweq12433454/LongHisDoc 和 https://github.com/SCUT-DLVCLab/LongHisDoc 获取。

View on arXiv Download PDF AI Translation

cs.CL / 22 / 2610.14578

Uniform Information Density-Based Preference Optimization

基于均匀信息密度的偏好优化

Li, Yongyuan, Xu, Yang

Abstract

To enable large language models to generate outputs that align with human preferences, optimization algorithms based on the comparison between chosen and rejected output pairs have been developed and been proven effective. Some recent endeavors have utilized certain linguistically-derived assumptions about the preference pairs to guide the design of more precise optimization directions. However, human preferences are complex and hard to measure by simplistic criteria, and it is important to incorporate broader and more sophisticated insights about preferred vs. not-preferred language. In this study, we adopt a well-established psycholinguistic theory, Uniform Information Density, to develop a novel preference optimization algorithm, UIDPO. The basic assumption is that humans prefer to produce language in which information is uniformly distributed, because uniformity induces less cognitive processing effort. To implement this idea into the algorithm, we derive a compound metric for information uniformity by combining the global standard deviation and the local second-order difference, and integrate it to the objective of preference optimization. We compare UIDPO with SimPO, a previous state-of-the-art method, obtaining comparable and sometimes better performance on AlpacaEval2, ArenaHard, and MT-Bench. Additional linguistic analysis on complement clauses also shows that UIDPO can better reflect the tendencies of keeping information uniform in natural language than baselines. Our results show that the subtle human preference over the information rate in language is a helpful source for achieving better human alignment for LLMs. The code repository is published here: https://github.com/CLCS-SUSTech/UIDPO.

Chinese Translation

为了使大型语言模型生成与人类偏好一致的输出，基于选择和拒绝输出对之间比较的优化算法已被开发并证明有效。一些近期的努力利用了关于偏好对的某些语言学推导假设，以指导更精确的优化方向的设计。然而，人类偏好是复杂的，难以通过简单的标准进行测量，因此，纳入关于偏好与非偏好语言的更广泛和更复杂的见解是重要的。在本研究中，我们采用了一个成熟的心理语言学理论——均匀信息密度（Uniform Information Density），开发了一种新颖的偏好优化算法UIDPO。基本假设是，人类倾向于生成信息均匀分布的语言，因为均匀性会减少认知处理的努力。为了将这一思想实现到算法中，我们通过结合全局标准差和局部二阶差异推导出一个复合度量来衡量信息均匀性，并将其整合到偏好优化的目标中。我们将UIDPO与之前的最先进方法SimPO进行了比较，在AlpacaEval2、ArenaHard和MT-Bench上获得了可比且有时更好的性能。对补充从句的额外语言学分析也表明，UIDPO能够比基线更好地反映自然语言中保持信息均匀的倾向。我们的结果表明，语言中信息速率的微妙人类偏好是实现大型语言模型更好人类对齐的有益来源。代码库已发布在此：https://github.com/CLCS-SUSTech/UIDPO。

View on arXiv Download PDF AI Translation

cs.CL / 23 / 2610.14635

PVFH-Bench: Benchmarking Premise-Induced Visual Fixation Hallucinations in Vision-Language Models

PVFH-Bench：视觉-语言模型中前提诱导的视觉注视幻觉基准测试

Wang, Xinhao, Cui, Junbo, Liu, Zhenghao, Yao, Yuan, Li, Xiaohua, Gu, Yu, Yu, Ge

Abstract

Large Vision-Language Models (LVLMs) have achieved remarkable progress across a wide range of multimodal tasks. However, object hallucination remains a persistent challenge, undermining model reliability in applications such as visual question answering, embodied agents, and human-AI decision support systems. In this paper, inspired by cognitive science studies on perceptual bias, we investigate a new failure mode termed Premise-Induced Visual Fixation Hallucination (PVFH). Specifically, PVFH refers to a premise-induced reasoning error in which models become anchored to visually confusable evidence, accept a false textual premise, and subsequently generate responses about an absent yet visually plausible object. To systematically study this phenomenon, we introduce PVFH-Bench, a benchmark that automatically constructs visually confusable but absent targets and pairs the core hallucination setting with diagnostic subsets to trace how models move from object recognition to Premise-Induced Visual Fixation Hallucination. Experimental results on 10 state-of-the-art LVLMs demonstrate that all evaluated models are highly vulnerable to Premise-Induced Visual Fixation Hallucination. Moreover, we find that explicit chain-of-thought reasoning can further exacerbate this failure mode by encouraging models to overinterpret weak visual cues as supporting evidence for hallucinated objects.

Chinese Translation

大型视觉-语言模型（LVLMs）在多模态任务中取得了显著进展。然而，物体幻觉仍然是一个持续的挑战，削弱了模型在视觉问答、具身代理和人机决策支持系统等应用中的可靠性。本文受到认知科学对感知偏差研究的启发，探讨了一种新的失效模式，称为前提诱导的视觉注视幻觉（PVFH）。具体而言，PVFH指的是一种前提诱导的推理错误，其中模型被视觉上混淆的证据所锚定，接受一个错误的文本前提，并随后生成关于缺失但视觉上合理的物体的响应。为了系统地研究这一现象，我们引入了PVFH-Bench，这是一个基准测试，自动构建视觉上混淆但缺失的目标，并将核心幻觉设置与诊断子集配对，以追踪模型如何从物体识别转向前提诱导的视觉注视幻觉。在对10个最先进的LVLM进行的实验结果表明，所有评估的模型对前提诱导的视觉注视幻觉高度脆弱。此外，我们发现显式的思维链推理可能会进一步加剧这一失效模式，促使模型将微弱的视觉线索过度解读为幻觉物体的支持证据。

View on arXiv Download PDF AI Translation

cs.CL / 24 / 2610.14705

Lingdan-V2: Large language models for clinically grounded reasoning in traditional Chinese medicine

Lingdan-V2：用于传统中医学临床推理的大型语言模型

Hua, Rui, Wei, Yu, Jin, Ziri, Sun, Zhuo, Chang, Kai, Xia, Jianan, Shu, Zixin, Li, Xiaodong, Jia, Dongmei, Dong, Fei, Zhang, Runshun, Yu, Jian, Xu, Hao, Yu, Haibin, Li, Jiansheng, Liu, Baoyan, Wang, Wenjia, Zhou, Xuezhong

Abstract

Large language models have advanced medical reasoning, but clinically grounded reasoning in traditional Chinese medicine (TCM) remains challenging because prescription decisions require integrating domain knowledge, syndrome differentiation, treatment principles, herbal prescription composition, and dosage determination. We develop a TCM language model family comprising models with 4 billion, 8 billion, and 14 billion parameters. The artificial intelligence contribution is a staged clinical alignment framework integrating large scale continued pretraining on TCM and clinical records, complementary reasoning supervision, and prescription-centered reinforcement learning (RL). We further formulate herbal prescription generation as a structured RL problem with prescription-level feedback, enabling direct optimization of clinical decision outputs beyond knowledge reproduction and generic instruction following. The engineering application is TCM clinical decision support, particularly herbal prescription recommendation. We evaluate the models on a public benchmark containing 25,620 expert-verified items across 13 TCM knowledge and clinical reasoning tasks and on a private retrospective dataset of 4,348 spleen and stomach disease prescription cases. The 14 billion parameter reinforcement aligned model achieves a prescription generation F1 of 37.0 on the public benchmark, compared with 32.3 for DeepSeek-R1 with 671 billion total parameters, and a herb-level F1 of 52.8 on the clinical dataset with improved dosage agreement. Reinforcement post-training consistently improves prescription recommendation across all scales, while physician-source analysis shows that expert heterogeneity affects reward stability. These findings demonstrate a scalable pathway from domain grounding and reasoning supervision to clinically oriented prescription decision support.

Chinese Translation

大型语言模型在医学推理方面取得了进展，但在传统中医学（TCM）中进行临床基础推理仍然具有挑战性，因为处方决策需要整合领域知识、症候辨识、治疗原则、草药处方组成和剂量确定。我们开发了一系列TCM语言模型，包括具有40亿、80亿和140亿参数的模型。人工智能的贡献是一个分阶段的临床对齐框架，整合了对TCM和临床记录的大规模持续预训练、补充推理监督和以处方为中心的强化学习（RL）。我们进一步将草药处方生成形式化为一个具有处方级反馈的结构化RL问题，使得能够直接优化临床决策输出，而不仅限于知识再现和通用指令遵循。工程应用是TCM临床决策支持，特别是草药处方推荐。我们在一个包含25,620个专家验证项目的公共基准上评估模型，涉及13个TCM知识和临床推理任务，以及在一个包含4,348个脾胃病处方案例的私人回顾性数据集上进行评估。140亿参数的强化对齐模型在公共基准上的处方生成F1达到了37.0，而671亿总参数的DeepSeek-R1则为32.3，并且在临床数据集上的草药级F1为52.8，且剂量一致性有所改善。强化后训练在所有规模的处方推荐中始终提高效果，而医生来源分析显示专家异质性影响奖励稳定性。这些发现展示了从领域基础和推理监督到临床导向的处方决策支持的可扩展路径。

View on arXiv Download PDF AI Translation

cs.CL / 25 / 2610.14709

Vela: User Behavior as Supervision Sparks LLM Intelligence

Vela：用户行为作为监督激发大语言模型智能

Xiao, Yang, Sun, Yusong, Wu, Haoyi, Hui, Wenyang, Hang, Weiqiang, Wang, Luqing, Wang, Zejian, Tu, Zirui, Mao, Jingsheng, Hu, Shaobo, Luo, Yang, Guo, Jiale, Zhang, Jin, Liu, Yudan, Luo, Zhaokai, Huan, Ren, Hu, Yao, Jiang, Chengyue, Li, Wenjie, Mu, Chuan

Abstract

Large language models are increasingly deployed in applications that act on users' long-term behavior histories, from personal assistants to recommendation and advertising systems. These applications demand the user dimension of LLM intelligence: inferring a person's preferences, intent, and likely future actions from weeks to years of behavior data. Yet current models are trained on static text or synthetic data, neither of which captures how user behavior evolves over time. Real-world behavior records, already available at scale on content platforms and in personal-assistant agents, offer a natural supervisory signal. We propose user behavior as supervision and introduce Vela, a framework that converts real user behavior data into trainable tasks and trajectories. From the anonymized behavior records of 100,000 users, Vela generates 100,000 tasks across 91 real-world task templates; difficulty-based filtering and trajectory acceptance criteria retain approximately 75,000 tasks with selected training trajectories. Fine-tuning Qwen3.6-35B-A3B on these trajectories improves performance over the base model by up to 11.5 points on user-related tasks and by an average of 4 points on general benchmarks. Training on more user behavior data further improves user-related performance, and reinforcement learning after fine-tuning outperforms reinforcement learning from the base model by up to 7.7 points. These gains suggest that real-world user behavior is a scalable source of supervision for both user-centered capabilities and the broader intelligence of LLMs.

Chinese Translation

大型语言模型越来越多地应用于基于用户长期行为历史的应用中，从个人助手到推荐和广告系统。这些应用要求大语言模型智能的用户维度：从数周到数年的行为数据中推断一个人的偏好、意图和可能的未来行为。然而，当前模型的训练基于静态文本或合成数据，这两者都无法捕捉用户行为随时间的演变。现实世界的行为记录，已经在内容平台和个人助手中大规模可用，提供了自然的监督信号。我们提出将用户行为作为监督，并介绍Vela，一个将真实用户行为数据转换为可训练任务和轨迹的框架。通过对100,000名用户的匿名行为记录，Vela生成了跨91个现实世界任务模板的100,000个任务；基于难度的过滤和轨迹接受标准保留了大约75,000个任务及其选定的训练轨迹。在这些轨迹上微调Qwen3.6-35B-A3B模型，在用户相关任务上性能提高了最多11.5分，在一般基准测试中平均提高了4分。对更多用户行为数据的训练进一步提升了用户相关性能，而微调后的强化学习在性能上优于基模型的强化学习，提升幅度达到7.7分。这些提升表明，现实世界的用户行为是一个可扩展的监督来源，既适用于以用户为中心的能力，也适用于大语言模型的更广泛智能。

View on arXiv Download PDF AI Translation

cs.CL / 26 / 2610.14756

Amharic-English Cross-Language Question Answering Using Statistical Machine Translation, Named Entity Recognition, and Semantic Indexing

使用统计机器翻译、命名实体识别和语义索引的阿姆哈拉语-英语跨语言问答系统

Bekele, Emebet, Getahun, Fekade

Abstract

The predominance of English-language Web content creates an information-access barrier for speakers of low-resource languages such as Amharic. This paper presents the design, imple-mentation, and evaluation of a bidirectional Amharic-English cross-language question answering (CLQA) system that allows a user to formulate a natural-language question in either Amharic or English and retrieve an answer from documents written in the other language. The system integrates a statistical machine translation (SMT) component for query translation, a maximum-entropy named entity recognition (NER) component for Amharic text processing, semantic-based indexing, question analysis, passage retrieval, and answer extraction. A domain-diverse Amharic-English parallel corpus was used to train the SMT component, while monolingual cor-pora supported language modeling. Documents were normalized, tokenized, stemmed, indexed, and semantically enriched before retrieval. The original evaluation used precision and recall as the principal effectiveness measures. For Amharic-to-English retrieval, the system achieved 72% precision and 79% recall; for English-to-Amharic retrieval, it achieved 65% precision and 70% recall. These findings show that a modular translation-and-retrieval architecture can provide useful cross-language access even under low-resource conditions. This revised article preserves the original experimental claims while situating them in the context of subsequent multilingual transformer and Amharic QA research.

Chinese Translation

英语网络内容的主导地位为阿姆哈拉语等低资源语言的使用者创造了信息获取障碍。本文介绍了一种双向阿姆哈拉语-英语跨语言问答（CLQA）系统的设计、实施和评估，该系统允许用户以阿姆哈拉语或英语提出自然语言问题，并从用另一种语言撰写的文档中检索答案。该系统集成了一个用于查询翻译的统计机器翻译（SMT）组件、一个用于阿姆哈拉语文本处理的最大熵命名实体识别（NER）组件、基于语义的索引、问题分析、段落检索和答案提取。使用了一个领域多样的阿姆哈拉语-英语平行语料库来训练SMT组件，同时单语语料库支持语言建模。在检索之前，文档经过规范化、分词、词干提取、索引和语义增强。原始评估使用精确度和召回率作为主要有效性度量。对于阿姆哈拉语到英语的检索，该系统达到了72%的精确度和79%的召回率；对于英语到阿姆哈拉语的检索，达到了65%的精确度和70%的召回率。这些结果表明，即使在低资源条件下，模块化的翻译与检索架构也能提供有用的跨语言访问。本文修订版保留了原始实验的主张，同时将其置于后续多语言变换器和阿姆哈拉语问答研究的背景中。

View on arXiv Download PDF AI Translation

cs.CL / 27 / 2610.14760

The Constitution Speaks Nheengatu: An Open MT System and Reproducible Corpus Pipeline for the Amazonian L\'ingua Geral

宪法以Nheengatu发声：亚马逊L'inguagem Geral的开放机器翻译系统和可重复的语料库流程

Carraro, Fabrício, Zevallos, Rodolfo Joel, de Souza, Rodrigo Gonçalves, da Silva, Caio Henrique Faustino, Ortega, John E.

Abstract

Nheengatu (yrl), the Amazonian L\'ingua Geral, has 6--10 thousand speakers and is co-official in S\~ao Gabriel da Cachoeira (Brazil), yet remains poorly served by evaluated Portuguese--Nheengatu MT resources. We present the largest parallel corpus assembled for the language (14.5k Portuguese--Nheengatu pairs), anchored by the 2023 official translation of the Brazilian Federal Constitution; the first MT evaluation set built from transcribed spontaneous speech; and, to our knowledge, the first open-weight MT models dedicated to bidirectional Portuguese--Nheengatu translation. Three findings organize the paper. First, translation into Nheengatu on held-out constitutional articles plateaus on development data regardless of training length or model size, and much of the remaining error sits in conventional legal terms that the written sources do not attest consistently; a terminology success rate over a frozen lexicon shows that glossary mechanisms at training time and lexical constraints at decoding each raise term realization, the latter at a cost in repetitive output that chrF++ barely registers. Second, a corpus expansion that adds transcribed conversation raises speech chrF++ by 24--27 points while leaving written domains unchanged, a register gap that written-only benchmarks cannot see. Third, a larger NLLB model helps translation into Portuguese but not into Nheengatu on legal or speech text. For Portuguese-to-Nheengatu translation, our final systems reach 31.7--32.8 chrF++ on article-disjoint constitutional text, 64.1--64.5 on the near-domain set (62.1 after removing items with target-side content overlap), and 45.2--47.5 on speech.

Chinese Translation

Nheengatu（yrl），亚马逊L'inguagem Geral，拥有6,000至10,000名使用者，并在巴西的圣加布里埃尔达卡乔埃拉（São Gabriel da Cachoeira）享有共同官方语言地位，但在评估的葡萄牙语—Nheengatu机器翻译资源方面仍然服务不足。我们展示了为该语言组装的最大的平行语料库（14,500对葡萄牙语—Nheengatu），以2023年巴西联邦宪法的官方翻译为基础；第一个基于转录自发言的机器翻译评估集；以及据我们所知，首个专门用于双向葡萄牙语—Nheengatu翻译的开放权重机器翻译模型。论文的三个发现组织了我们的讨论。首先，针对Nheengatu的翻译在保留的宪法条款上，无论训练长度或模型规模如何，开发数据的表现都趋于平稳，剩余的错误主要集中在书面来源未能一致证明的传统法律术语上；在固定词汇表上的术语成功率显示，训练时的词汇机制和解码时的词汇约束均提高了术语实现，后者在重复输出中付出了代价，而chrF++几乎未能记录。其次，增加转录对话的语料库扩展使得语音的chrF++提高了24至27点，而书面领域保持不变，这一书面与口语的差距是仅基于书面基准无法察觉的。第三，较大的NLLB模型有助于葡萄牙语的翻译，但对法律或口语文本的Nheengatu翻译没有帮助。对于葡萄牙语到Nheengatu的翻译，我们的最终系统在不相交的宪法文本上达到了31.7至32.8的chrF++，在近域集上达到了64.1至64.5（去除目标侧内容重叠的项目后为62.1），在口语上达到了45.2至47.5。

View on arXiv Download PDF AI Translation

cs.CL / 28 / 2610.14769

Who Receives Respect? A Counterfactual Benchmark for Honorific Allocation Bias in Bengali

谁获得尊重？孟加拉语中尊称分配偏差的反事实基准

Abrar, Ajwad, Hasan, Md Kamrul, Mahmud, Hasan

Abstract

Bengali has no grammatical gender, but honorificity is obligatory: every third-person reference must choose ordinary or respectful register. English has no such signal, so the system must choose a level of deference that the source never specified. We use this property to measure how large language models (LLMs) allocate respect in English--Bengali translation. Our evaluation set contains 2{,}016 sentences across 72 occupations. Each appears in seven sentence frames under four pronoun conditions. Masculine and feminine versions differ by one word, forming counterfactual pairs. We report a single Honorific Bias Score (HBS) for each model on a 0--100 scale, covering both the gender and occupation axes. A lower score means that respect is allocated more evenly. For the occupation analysis, we compute each occupation's honorific rate across the seven no-pronoun sentence frames. The standard deviation of these 72 occupation-level rates ranges from 17.2 to 35.5 percentage points across models, showing substantial variation in how honorifics are allocated across occupations. Occupation is associated with 3.2\% to 51.5\% of the variance in register choice, compared with at most 1.2\% for gender. Gender effects are much smaller overall: only two of the nine models show significant directional gender gaps, and they favor opposite genders. HBS ranges from 19.9 to 70.1 across the nine models, showing substantial variation in how evenly the models allocate respect.

Chinese Translation

孟加拉语没有语法性别，但尊称是强制性的：每个第三人称的引用必须选择普通或尊重的语体。英语没有这样的信号，因此系统必须选择一个源头从未指定的尊重程度。我们利用这一特性来测量大型语言模型（LLMs）在英语-孟加拉语翻译中如何分配尊重。我们的评估集包含2,016个句子，涵盖72个职业。每个句子在四种代词条件下出现在七个句子框架中。男性和女性版本仅相差一个单词，形成反事实对。我们为每个模型报告一个单一的尊称偏差分数（Honorific Bias Score, HBS），在0到100的范围内，涵盖性别和职业两个维度。较低的分数意味着尊重的分配更为均匀。在职业分析中，我们计算每个职业在七个无代词句子框架中的尊称率。这72个职业级别的尊称率的标准差在不同模型中范围从17.2到35.5个百分点，显示出在不同职业中尊称分配的显著差异。职业与注册选择的方差相关性在3.2%到51.5%之间，而性别的方差相关性最多为1.2%。总体而言，性别效应要小得多：在九个模型中仅有两个显示出显著的性别差距，并且它们偏向于相反的性别。HBS在九个模型中范围从19.9到70.1，显示出模型在分配尊重时的显著差异。

View on arXiv Download PDF AI Translation

cs.CL / 29 / 2610.14789

Training-Free BioNEL-RAG: Advancing Biomedical Entity Linking through Retrieval-Augmented Generation and Large Language Model

无训练的 BioNEL-RAG：通过检索增强生成和大型语言模型推进生物医学实体链接

Chen, Ziyi, Liu, Hui, Ling, Ziyang, Xu, Ronald X., Sun, Mingzhai

Abstract

Biomedical Named Entity Linking (BioNEL) is the task of mapping mentions extracted from biomedical texts to canonical entities or concepts in the standardized knowledge base. Although existing methods have achieved remarkable performance in mapping accuracy, most rely on training with domain-specific datasets, which fundamentally limits their cross-domain adaptability. To address this limitation, we propose BioNEL-RAG, a novel training-free framework that combines retrieval-augmented generation (RAG) with large language models (LLMs). Within this framework, the RAG module focuses on retrieving candidate entities, while the large language model module plays the role of selecting the most appropriate candidate for entity normalization, which is based on the analysis of the mention's context, with the aid of its remarkable text comprehension capabilities. Experimental results demonstrate that BioNEL-RAG achieves performance comparable to existing methods without requiring training on subdomain datasets, whereas these methods necessitate fine-tuning on specific subdomains such as genes, diseases, or chemical compounds.

Chinese Translation

生物医学命名实体链接（BioNEL）是将从生物医学文本中提取的提及映射到标准知识库中的规范实体或概念的任务。尽管现有方法在映射准确性方面取得了显著的性能，但大多数方法依赖于特定领域数据集的训练，这在根本上限制了它们的跨领域适应性。为了解决这一限制，我们提出了 BioNEL-RAG，这是一种新颖的无训练框架，将检索增强生成（RAG）与大型语言模型（LLMs）相结合。在该框架中，RAG 模块专注于检索候选实体，而大型语言模型模块则负责选择最合适的候选实体进行实体标准化，这一过程基于对提及上下文的分析，并借助其卓越的文本理解能力。实验结果表明，BioNEL-RAG 的性能可与现有方法相媲美，而无需在子领域数据集上进行训练，而这些方法则需要在基因、疾病或化合物等特定子领域进行微调。

View on arXiv Download PDF AI Translation

cs.CL / 30 / 2610.14873

Dependency Reasoning for Aspect Sentiment Quadruple Prediction

面向方面情感四元组预测的依赖推理

Esfandyar, Shima, Soleymanbaigi, Seywan, Tab, Fardin Akhlaghian, Daneshfar, Fatemeh, Oussalah, Mourad

Abstract

Aspect Sentiment Quadruple Prediction (ASQP) is fundamentally a relational reasoning problem: identifying aspect-opinion-category-sentiment quadruples requires modeling structured dependencies, not merely sequential co-occurrence. Existing LLM-based approaches incorporate syntactic information indirectly, via retrieval-based demonstration selection or multi-stage training, rather than as unified, single-pass generation-time context. We present DRIFT, which linearizes dependency parse trees and injects them directly into LLM prompts as explicit relational context, enabling the model to reason over aspect-opinion dependencies rather than infer them from token proximity alone. DRIFT further integrates retrieval-augmented in-context learning and a syntax-aware LLM-based adjudication mechanism that complements exact-match evaluation under span-boundary variation. On SemEval Rest15, Rest16, and ACOS, DRIFT achieves the strongest performance among compared generative ASQP methods, reaching 54.47% and 64.10% F1 on Rest15 and Rest16, with consistent gains across short- and medium-distance aspect-opinion configurations; adjudication further shows exact-match metrics systematically underestimate semantically correct predictions.

Chinese Translation

方面情感四元组预测（ASQP）本质上是一个关系推理问题：识别方面-观点-类别-情感四元组需要建模结构化依赖关系，而不仅仅是顺序共现。现有的基于大型语言模型（LLM）的方法通过检索基础的示例选择或多阶段训练间接地融入句法信息，而不是作为统一的、单次生成时的上下文。我们提出了DRIFT，它将依赖解析树线性化，并将其直接注入LLM提示中作为显式的关系上下文，使模型能够推理方面-观点之间的依赖关系，而不仅仅是从标记的接近性推断。DRIFT进一步整合了检索增强的上下文学习和一种基于句法的LLM裁决机制，补充了在跨度边界变化下的精确匹配评估。在SemEval Rest15、Rest16和ACOS数据集上，DRIFT在比较的生成ASQP方法中表现最强，在Rest15和Rest16上分别达到了54.47%和64.10%的F1分数，并在短距离和中距离的方面-观点配置中持续获得提升；裁决进一步表明，精确匹配指标系统性地低估了语义上正确的预测。

View on arXiv Download PDF AI Translation

cs.CL / 31 / 2610.14879

Social Bias in Chinese Large Language Models under Social Media Expressions

社交媒体表达下的中国大型语言模型中的社会偏见

Zhou, Mo

Abstract

Existing Chinese social-bias benchmarks for large language models largely rely on normalized or template-based expressions, leaving the influence of internet slang, emoji, and sarcasm insufficiently examined. Safety alignment may further compress conventional plausibility ratings toward the lower end of the scale. This study constructs 73 paired probes across gender, region, and occupation. Each probe contains a neutral expression, a social-media-style expression, and a reversed-identity control. The task is reframed from direct plausibility judgment to social consensus estimation. Social-media expressions are divided into stylistic packaging, which preserves semantic polarity, and sarcastic packaging, which reverses surface polarity. Six Chinese-capable large language models are evaluated through paired differences measuring identity bias, stylistic packaging, and sarcasm shift. All models show significant identity-bias effects; five show significantly negative stylistic-packaging effects; and sarcasm shifts display a clear cross-model gradient. These findings demonstrate that expression form is not a transparent carrier in bias evaluation and that lower scores under sarcasm should not be interpreted directly as bias mitigation. The proposed paired design offers a measurement approach for Chinese social-media contexts while also revealing the construct boundary and prompt sensitivity of social consensus estimation.

Chinese Translation

现有的针对大型语言模型的中国社会偏见基准主要依赖于规范化或基于模板的表达，互联网俚语、表情符号和讽刺的影响尚未得到充分研究。安全对齐可能进一步压缩常规的可信度评分至量表的较低端。本研究构建了73对探测器，涵盖性别、地区和职业。每个探测器包含一个中性表达、一个社交媒体风格的表达和一个反向身份控制。任务从直接的可信度判断重新框定为社会共识估计。社交媒体表达被分为风格包装（保持语义极性）和讽刺包装（反转表面极性）。通过测量身份偏见、风格包装和讽刺转变的配对差异，评估了六个具有中文能力的大型语言模型。所有模型均显示出显著的身份偏见效应；五个模型显示出显著的负风格包装效应；而讽刺转变则显示出明显的跨模型梯度。这些发现表明，表达形式并不是偏见评估中的透明载体，讽刺下的较低评分不应直接解读为偏见减轻。所提出的配对设计为中国社交媒体背景提供了一种测量方法，同时揭示了社会共识估计的构建边界和提示敏感性。

View on arXiv Download PDF AI Translation

cs.CL / 32 / 2610.14886

Agent-Guided Evidence Acquisition for Emotion-Cause Pair Extraction

基于代理的情感-原因对提取的证据获取

Pan, Zhuangzhuang, Xia, Yan, Chan, Chee Seng

Abstract

Emotion-Cause Pair Extraction (ECPE) requires identifying the causes associated with emotions expressed in a conversation. Accurate extraction can depend on evidence that is not initially available, making it important to decide when further information is worth acquiring. We study agent-guided evidence acquisition for ECPE, where a student agent selectively accesses expert judgments for borderline relations on both sides of its validation-selected F1 decision boundary and revises only queried predictions. Through controlled comparisons under different information-access conditions, we examine how delegation granularity, query selection, and response integration affect prediction quality. At a nominal 10% query rate, selective fusion improves mean pair F1 by 2.64-3.88 points over the student in our main experiments, reaching 65.14 on ECF and 49.74 on MEC4 across model families. Our analysis suggests that greater access does not always translate into better cause identification. These findings motivate acquisition strategies that consider both the relevance of newly available evidence and its effect on prediction revision, providing an empirical basis for more effective and efficient evidence acquisition in ECPE.

Chinese Translation

情感-原因对提取（ECPE）需要识别与对话中表达的情感相关的原因。准确的提取可能依赖于最初不可用的证据，因此决定何时获取进一步信息变得至关重要。我们研究了用于ECPE的代理引导证据获取，其中学生代理选择性地访问专家判断，以处理其验证选择的F1决策边界两侧的边界关系，并仅修正被查询的预测。通过在不同信息访问条件下的受控比较，我们考察了委托粒度、查询选择和响应整合如何影响预测质量。在名义上的10%查询率下，选择性融合在我们的主要实验中使平均对F1提高了2.64-3.88分，ECF达到了65.14，MEC4达到了49.74，涵盖了不同模型系列。我们的分析表明，更大的访问并不总是转化为更好的原因识别。这些发现激励了考虑新获取证据的相关性及其对预测修正影响的获取策略，为ECPE中更有效和高效的证据获取提供了实证基础。

View on arXiv Download PDF AI Translation

cs.CL / 33 / 2610.14900

TeachGrade: Training-Free Short Answer Scoring with Teacher Evidence

TeachGrade：无训练的短答案评分框架与教师证据

Yan, Jinhao, Shen, Zhihong, Wang, Shengjie, Ren, Yiting, Zhang, Wentao, Wang, Yilei, Lai, Peichao

Abstract

Multi-subject, multi-step scoring requires consistent partial credit across grading conventions. We introduce TeachGrade, a training-free framework using teacher-graded examples. The Task Adapt module builds two frozen memories: Skill captures contextual teacher preferences beyond the rubric; CaseBank supplies local evidence. PreJudge provides retrieval cues; FinalJudge regrades the answer without preliminary scores. On 3,829 SAS-Bench answers across 12 tasks, official macro collaborative consistency score (CCS) rises from Native's 67.99 to 81.32 (+13.33 points), and pooled step exact match from 54.85% to 62.71%. Macro quadratic weighted kappa (QWK) reaches 84.71, exceeding Native, OpenJudge, ProofGrader, and Trojsten by 9.52, 3.50, 11.37, and 17.43 points, respectively, and leading on 8 tasks. Local Qwen3.8-27B gains 8.83 CCS and 4.08 QWK points, improving CCS on all 12 tasks. QWK gains over Native span 6.38--14.65 points on Portuguese automatic essay scoring (AES) and 2.16--2.24 on ASAP-SAS across label views, with lower mean absolute error. These results support reuse across languages, response formats, and backbones without weight updates.

Chinese Translation

多学科、多步骤评分需要在评分规范中保持一致的部分分数。我们提出了TeachGrade，一个使用教师评分示例的无训练框架。任务适应模块构建了两个固定的记忆：技能（Skill）捕捉超出评分标准的上下文教师偏好；案例库（CaseBank）提供局部证据。预评判（PreJudge）提供检索线索；最终评判（FinalJudge）在没有初步评分的情况下重新评分答案。在12个任务的3,829个SAS-Bench答案中，官方宏观协作一致性得分（CCS）从Native的67.99上升至81.32（+13.33分），汇总步骤精确匹配从54.85%上升至62.71%。宏观二次加权卡帕（QWK）达到84.71，分别超出Native、OpenJudge、ProofGrader和Trojsten 9.52、3.50、11.37和17.43分，并在8个任务中领先。局部Qwen3.8-27B获得8.83的CCS和4.08的QWK分数，在所有12个任务中改善了CCS。与Native相比，QWK在葡萄牙语自动作文评分（AES）上增加6.38至14.65分，在ASAP-SAS的标签视图上增加2.16至2.24分，且平均绝对误差更低。这些结果支持在不同语言、响应格式和基础模型之间的重用，而无需更新权重。

View on arXiv Download PDF AI Translation

cs.CL / 34 / 2610.14934

FidelityQ: Preserving Predictive Fidelity in Ultra-Low-Bit Language Model Quantization

FidelityQ：在超低比特语言模型量化中保持预测保真度

Chen, Yujie, Zheng, Weiyan, Yao, Yiwu, Hu, Tianchi, Wang, Bo, Guo, Jiyu, Li, Wenzhan, Guo, Jingcai, Magno, Michele, Qin, Haotong

Abstract

Ultra-low-bit scalar quantization offers a regular arithmetic structure for efficient deployment of large language models, but preserving predictive behavior at extreme precision remains challenging: with only a few reconstruction levels, accurate weight representation must translate into consistent predictions from a small calibration set. We propose FidelityQ, a scalar quantization framework that preserves predictive fidelity at weight precisions down to 2 bits. For accurate quantization, we introduce Information-Aligned Analytic Coding (IAC), which shapes each weight row toward a centered symmetric source via randomized Hadamard mixing and fits a mirrored four-level minimum-distortion quantizer to its empirical weights. Every level set in this family decomposes exactly into two sign-valued bases with associated scales, combining source-adaptive reconstruction with lookup-free decoding compatible with scaled integer arithmetic. Our analysis further shows that residual predictive discrepancies are heavy-tailed and concentrate on tokens the teacher predicts confidently, motivating Knowledge-Intensive Calibration (KIC) for consistent quantization. KIC ranks sequences by the tail of token-wise teacher-student divergence, allocates a limited calibration budget to the highest-scoring samples, and restores predictive agreement by updating only quantization levels and scales under teacher supervision, keeping pretrained weights frozen. Experiments on the LLaMA and Qwen families demonstrate the effectiveness of FidelityQ in weight-only and joint quantization settings. On LLaMA-2-7B, FidelityQ achieves a WikiText2 perplexity of 6.62 at W2A16 and improves the prior state-of-the-art W2A4KV4 result from 8.31 to 6.96, while reducing linear-layer weight memory by 8.0x relative to FP16.

Chinese Translation

超低比特标量量化为大型语言模型的高效部署提供了一种规则的算术结构，但在极端精度下保持预测行为仍然具有挑战性：仅凭少量重建级别，准确的权重表示必须转化为来自小型校准集的一致预测。我们提出了FidelityQ，这是一种标量量化框架，能够在权重精度低至2比特的情况下保持预测保真度。为了实现准确的量化，我们引入了信息对齐分析编码（Information-Aligned Analytic Coding, IAC），该方法通过随机Hadamard混合将每个权重行塑形为一个中心对称源，并将一个镜像的四级最小失真量化器拟合到其经验权重上。该家族中的每个级别集恰好分解为两个带符号值的基底及其相关比例，结合了源自适应重建和与缩放整数算术兼容的无查找解码。我们的分析进一步表明，残余预测差异呈重尾分布，并集中在教师自信预测的标记上，这促使我们提出知识密集型校准（Knowledge-Intensive Calibration, KIC）以实现一致的量化。KIC根据标记级教师-学生发散的尾部对序列进行排名，将有限的校准预算分配给得分最高的样本，并通过在教师监督下仅更新量化级别和比例来恢复预测一致性，同时保持预训练权重不变。在LLaMA和Qwen系列上的实验表明，FidelityQ在仅权重和联合量化设置中都表现出有效性。在LLaMA-2-7B上，FidelityQ在W2A16下实现了6.62的WikiText2困惑度，并将之前的最先进的W2A4KV4结果从8.31提高到6.96，同时相对于FP16减少了8.0倍的线性层权重内存。

View on arXiv Download PDF AI Translation

cs.CL / 35 / 2610.14952

Do Personalized LLMs Need the Whole Conversation? Matching the Full History with Only 5% of Its Text via Relevance-Filtered Context

个性化大型语言模型是否需要完整对话？通过相关性过滤的上下文仅用5%的文本匹配完整历史

Hayashi, Erika, Shin, Andrew

Abstract

Personalized assistants accumulate long conversation histories, and passing the whole history to an LLM with every request is costly. We show that it is also unnecessary. On PersonaMem-v2 (180 questions), with GPT-5.6-terra and Claude Opus 5.5 as judges, a relevance-filtered context, built by sentence-level retrieval that ranks the history by similarity to the current request and stops at a fixed budget, matches full-history accuracy with only 5% of the text, cutting the judge's input about 20-fold. An LLM summary of the same length performs on par, but it needs an LLM to read the entire history, whereas retrieval needs only a small embedding model. While both sentence-level retrieval and LLM summary perform strongly, retrieval does so while placing the relevant evidence in the context for only 61% of the questions, and an oracle given only that evidence beats the full history by 15.6 points, making better context selection a promising path to surpassing the full history.

Chinese Translation

个性化助手会积累长时间的对话历史，而每次请求都将整个历史传递给大型语言模型（LLM）是成本高昂的。我们展示了这也是不必要的。在PersonaMem-v2（180个问题）上，使用GPT-5.6-terra和Claude Opus 5.5作为评判者，通过句子级检索构建的相关性过滤上下文，根据与当前请求的相似度对历史进行排名，并在固定预算下停止，仅用5%的文本就能匹配完整历史的准确性，从而将评判者的输入减少约20倍。相同长度的LLM摘要表现相当，但它需要LLM读取整个历史，而检索仅需要一个小型嵌入模型。虽然句子级检索和LLM摘要的表现都很强，但检索在仅为61%的问题中将相关证据放入上下文，而仅凭这些证据的预言者比完整历史高出15.6分，这使得更好的上下文选择成为超越完整历史的有希望的路径。

View on arXiv Download PDF AI Translation

cs.CL / 36 / 2610.15028

UltraBERT: Beyond the Limit of BERT Pre-Training

UltraBERT：超越BERT预训练的极限

Park, Jungwoo, Lee, Taewhoo, Kang, Jaewoo

Abstract

Recent bidirectional encoders have been modernized with innovations from causal large language models (LLMs). However, the pretraining recipe itself remains inherited from small-scale conventions, leaving encoders underoptimized, with no single model performing competitively across both classification and retrieval tasks. In this work, we investigate how far encoder performance can be pushed by revisiting core design choices in training, revealing three key findings: (1) Encoders saturate on downstream tasks beyond a few trillion tokens, indicating that additional compute past this point does not help. (2) Across model sizes, lower masking ratios benefit classification while higher ones benefit retrieval, contradicting prior work suggesting larger ratios for larger models. (3) Distillation with pruning improves classification at low masking ratios but degrades retrieval at high ones. Guided by these findings, we present UltraBERT, a single encoder that assigns separate LM heads to low and high masking ratios within a batch and restricts the distillation objective to the low ratio. To our knowledge, UltraBERT is the first encoder to achieve state of the art on both classification and retrieval tasks, reaching 91.0 on GLUE and 47.7 on BEIR. Together, these results show how far encoder performance can be pushed and provide a principled basis for compute allocation, masking ratio design, and knowledge transfer in bidirectional models.

Chinese Translation

最近的双向编码器借鉴了因果大型语言模型（LLMs）的创新。然而，预训练的配方依然继承了小规模惯例，导致编码器未得到优化，没有单一模型在分类和检索任务中表现出竞争力。在本研究中，我们探讨了通过重新审视训练中的核心设计选择，编码器性能可以被推向多远，揭示了三个关键发现：（1）编码器在下游任务中在超过几万亿个标记后达到饱和，表明在此点之后增加计算并没有帮助。（2）在不同模型规模中，较低的掩码比例有利于分类，而较高的比例则有利于检索，这与先前的研究相矛盾，后者建议较大模型使用较大的掩码比例。（3）通过剪枝的蒸馏在低掩码比例下改善分类，但在高掩码比例下降低检索性能。在这些发现的指导下，我们提出了UltraBERT，一个单一编码器，它为批次中的低掩码和高掩码比例分配不同的语言模型头，并将蒸馏目标限制在低掩码比例。根据我们的了解，UltraBERT是第一个在分类和检索任务上都达到最先进水平的编码器，在GLUE上达到91.0，在BEIR上达到47.7。这些结果共同展示了编码器性能可以被推向的极限，并为计算分配、掩码比例设计和双向模型中的知识转移提供了原则性基础。

View on arXiv Download PDF AI Translation

cs.CL / 37 / 2610.15029

Toward System One Clinical Intelligence: Parameter-Efficient Fine-Tuning and Counterfactual Explainability of Decision Foundation Models

迈向系统一临床智能：参数高效微调与决策基础模型的反事实可解释性

Ahmed, S. M. Mahbub Uddin

Abstract

Automated disease classification from patient-reported symptoms requires balancing the speed of lightweight discriminative models against the contextual understanding and latency costs of autoregressive large language models (LLMs). Decision foundation models offer an alternative by mapping representations directly to candidate classes without free-form text generation. We adapt Clef-Flash, a 9-billion-parameter decision model built on a Qwen3.5 backbone, for 24-class disease classification on the Symptom2Disease benchmark. Using 4-bit quantized low-rank adaptation (QLoRA) with differential learning rates and a stratified split of 960 training and 240 test cases, the model achieves 95.42% diagnostic accuracy and a macro F1 score of 0.9551. This improves upon the reported zero-shot baselines of 61.67% for a remote API and 65.00% for a local unadapted initialization, and outperforms the ClinicalBERT baseline (90.90%). By evaluating candidate logits in a single forward pass without autoregressive decoding, inference takes under 80 milliseconds per record on a single GPU, yielding a reported 31-fold speedup over generative LLM baselines. To examine symptom-level evidence, we introduce a Leave-One-Symptom-Out (LOSO) counterfactual perturbation framework. Removing disease-defining symptom clauses in a psoriasis example reduces predicted-class confidence from 100.00% to 0.12% and shifts the prediction to impetigo with 99.59% confidence, while removing secondary systemic symptoms produces less than 0.01% change in confidence. These findings suggest that decision-based architectures can combine the linguistic capabilities of large foundation models with low-latency inference and symptom-level interpretability for clinical triage.

Chinese Translation

从患者报告的症状中自动进行疾病分类需要在轻量级判别模型的速度与自回归大型语言模型（LLMs）的上下文理解和延迟成本之间取得平衡。决策基础模型通过将表示直接映射到候选类别，提供了一种替代方案，而无需生成自由形式的文本。我们将基于Qwen3.5骨干网构建的一个90亿参数的决策模型Clef-Flash适配于Symptom2Disease基准的24类疾病分类。通过使用4位量化低秩适配（QLoRA）与差异学习率，并对960个训练案例和240个测试案例进行分层拆分，该模型实现了95.42%的诊断准确率和0.9551的宏观F1分数。这一结果超越了远程API的61.67%和本地未适配初始化的65.00%的零-shot基线，并且优于ClinicalBERT基线（90.90%）。通过在单次前向传递中评估候选logits，而无需自回归解码，推理在单个GPU上每条记录的时间小于80毫秒，报告的速度比生成型LLM基线快31倍。为了检查症状级别的证据，我们引入了一个留一症状法（Leave-One-Symptom-Out, LOSO）反事实扰动框架。在一个银屑病的例子中，移除定义疾病的症状条款将预测类别的置信度从100.00%降低到0.12%，并将预测转移到脓疱病，置信度为99.59%；而移除次要系统症状则导致置信度变化小于0.01%。这些发现表明，基于决策的架构可以将大型基础模型的语言能力与低延迟推理和症状级别的可解释性结合起来，用于临床分诊。

View on arXiv Download PDF AI Translation

cs.CL / 38 / 2610.15044

PAT: A Personality-Driven Augmentation and Transfer Learning Framework for Depression Detection on Social Media

PAT：一种基于个性驱动的社交媒体抑郁检测增强与迁移学习框架

Zheng, Shengwu, Ding, Yuzhe, Liang, Xiaohui, Li, Fei, Teng, Chong, Ji, Donghong

Abstract

Depression is a prevalent mental disorder that affects millions worldwide, with early detection crucial for effective intervention. While exist?ing methods have achieved promising results in automated depression detection, they face two key limitations: (1) noise in post screening: symptom-based semantic screening is easily misled by superficial lexical patterns, thereby introducing false-positive posts; (2) weak op?timization of the post encoder: relying solely on coarse user-level supervision signals pre?vents models from capturing depressive cues within individual posts. To address these limita?tions, we propose PAT, a Personality-driven Augmentation and Transfer learning frame?work. In the first phase, PAT optimizes a post?level encoder via a fine-grained contrastive learning objective, utilizing LLM-based data augmentation driven by classic personality the?ories to alleviate data scarcity. In the second phase, the trained post-level scorer is deployed for high-quality post filtering, while its en?coder parameters are seamlessly transferred for user-level tasks. Extensive experiments demon?strate that PAT significantly outperforms exist?ing methods. Furthermore, PAT delivers multi?granular interpretability by tracing post-level mental-state trajectories and highlighting key symptoms, thereby offering reliable diagnostic evidence for clinical practice.

Chinese Translation

抑郁症是一种普遍的心理障碍，影响着全球数百万人，早期检测对于有效干预至关重要。尽管现有方法在自动抑郁检测方面取得了可喜的成果，但它们面临两个主要限制：（1）后筛选中的噪声：基于症状的语义筛选容易受到表面词汇模式的误导，从而引入假阳性帖子；（2）帖子编码器的优化不足：仅依赖粗略的用户级监督信号使得模型无法捕捉到单个帖子中的抑郁线索。为了解决这些限制，我们提出了PAT，一个基于个性驱动的增强与迁移学习框架。在第一阶段，PAT通过细粒度对比学习目标优化帖子级编码器，利用基于经典个性理论的LLM（大语言模型）驱动的数据增强来缓解数据稀缺。在第二阶段，训练好的帖子级评分器被用于高质量的帖子筛选，同时其编码器参数无缝转移到用户级任务中。大量实验表明，PAT显著优于现有方法。此外，PAT通过追踪帖子级心理状态轨迹并突出关键症状，提供多层次的可解释性，从而为临床实践提供可靠的诊断证据。

View on arXiv Download PDF AI Translation

cs.CL / 39 / 2610.15056

Select or Rewrite: Oracle-Guided Fine-Tuning for Generative ASR Post-Editing from N-Best Hypotheses

选择或重写：基于Oracle指导的生成式ASR后编辑的细调方法

Liu, Shiran, Xu, Gaohui, Burdisso, Sergio, Bañeras-Roux, Thibault, Kumar, Shashi, Motlicek, Petr

Abstract

Large language models can post-edit ASR by using linguistic context to compare N-best hypotheses and rewrite the transcript when needed, often improving WER without retraining the recognizer. However, prompt-only correction is brittle: it may over-edit when top-1 is correct and hallucinate when the reference is absent, and often needs very large models. We propose oracle-guided fine-tuning that turns N-best lists into supervision via three curated pools: corrections (reference in the list but not top-1) to teach selection, regularizers (top-1 correct) to discourage unnecessary rewrites, and hard negatives (reference absent) to suppress hallucination. Fine-tuning on this data provides more effective supervision than random subset fine-tuning for learning a hybrid select-or-rewrite policy, with the clearest gains from around 3B parameters on the proprietary dataset but less conclusive results on CommonVoice, a divergence suggesting that public read-speech text may overlap with LLM pretraining and mask genuine post-editing gains.

Chinese Translation

大型语言模型可以通过使用语言上下文来比较N-best假设并在需要时重写转录，从而对ASR进行后编辑，通常在不重新训练识别器的情况下改善字错误率（WER）。然而，仅依赖提示的纠正方法较为脆弱：当top-1正确时可能会过度编辑，而在缺少参考时可能会产生幻觉，并且通常需要非常大的模型。我们提出了一种基于Oracle指导的细调方法，通过三个精心策划的池将N-best列表转化为监督信息：纠正（列表中存在但不是top-1的参考）用于教授选择，正则化（top-1正确）用于抑制不必要的重写，以及困难负样本（缺少参考）用于抑制幻觉。在这些数据上进行细调比随机子集细调提供了更有效的监督，用于学习混合选择或重写策略，尤其在专有数据集上从约30亿参数中获得了明显的收益，但在CommonVoice上的结果不够明确，这一差异表明公共朗读文本可能与大型语言模型的预训练重叠，从而掩盖了真正的后编辑收益。

View on arXiv Download PDF AI Translation

cs.CL / 40 / 2610.15091

Have You Been Convinced? A Neuro-symbolic Argumentation Reasoning Approach to Detecting Persuasiveness in ChangeMyView Discussions

你被说服了吗？一种神经符号论证推理方法用于检测 ChangeMyView 讨论中的说服力

Bhargava, Siddharth, Tonelli, Sara, Martín-Rodilla, Patricia

Abstract

Argument structures represent how individual arguments interact within a discourse. However, leveraging these structures to infer meaningful knowledge remains challenging, as it requires a formal representation of how argumentative influence propagates across the structure. We propose Neuro-symbolic Argumentation Reasoning System (NARS) for performing persuasiveness detection directly from online discussions. The neural component employs retrieval-augmented iterative LLM reasoning to extract bipolar argument structures, while the symbolic component applies gradual argumentation semantics to propagate argumentative influence over the structure. Using the Webis-CMV-20 corpus, we construct CMV2BAS, a dataset of long, multi-threaded ChangeMyView discussions annotated with LLM-derived bipolar argument structures and labeled with delta events (view change vs.\ no change). We investigate whether gradual argumentation semantics can effectively detect persuasiveness, whether augmenting them with argument quality and graph-based centrality improves performance, and whether symbolic reasoning remains necessary in the presence of modern LLMs. NARS achieves F1=0.57 and AUC=0.56 under its best configuration, outperforming zero-shot Qwen3.6:27B and GPT-oss:20B. We identify support-dominant discussions and long argumentative chains as key failure modes limiting attack influence propagation. Our work demonstrates the feasibility of operationalizing formal gradual argumentation semantics over noisy real-world discourse for scalable macro-level persuasion analysis, while highlighting key challenges and directions for future research in neuro-symbolic argumentation reasoning. We release both CMV2BAS dataset and the NARS framework for public use.

Chinese Translation

论证结构表示了个体论证在话语中的相互作用。然而，利用这些结构推断有意义的知识仍然具有挑战性，因为这需要对论证影响如何在结构中传播进行正式表示。我们提出了神经符号论证推理系统（Neuro-symbolic Argumentation Reasoning System, NARS），以直接从在线讨论中进行说服力检测。神经组件采用增强检索的迭代 LLM 推理来提取双极论证结构，而符号组件则应用渐进式论证语义来传播论证影响。使用 Webis-CMV-20 语料库，我们构建了 CMV2BAS 数据集，该数据集包含长篇、多线程的 ChangeMyView 讨论，标注了 LLM 派生的双极论证结构，并标记了增量事件（观点变化与无变化）。我们研究了渐进式论证语义是否能够有效检测说服力，增强论证质量和基于图的中心性是否能提高性能，以及在现代 LLM 存在的情况下，符号推理是否仍然必要。NARS 在其最佳配置下实现了 F1=0.57 和 AUC=0.56，优于零样本 Qwen3.6:27B 和 GPT-oss:20B。我们识别出支持主导的讨论和长论证链作为限制攻击影响传播的关键失败模式。我们的工作展示了在嘈杂的现实话语中操作化正式渐进式论证语义以进行可扩展的宏观层面说服分析的可行性，同时突出了神经符号论证推理未来研究中的关键挑战和方向。我们将 CMV2BAS 数据集和 NARS 框架公开发布以供使用。

View on arXiv Download PDF AI Translation

cs.CL / 41 / 2610.15103

RegExtract: A Benchmark for PFAS Regulatory Information Extraction with LLMs

RegExtract：基于大型语言模型的PFAS监管信息提取基准

Circi, Defne, Lionti, Krystelle, Swaminathan, Sarathkrishna, Sanders, Daniel P., Brinson, L. Catherine, Dhingra, Bhuwan, Shbita, Basel, Elkaref, Mohab, Capponi, Sara, Erdmann, Tim, Ransom, Brandi

Abstract

Extracting structured information from regulatory documents is a critical but challenging task. Relevant facts are scattered across long, legally dense texts, and traceability of extracted facts to their source is essential for verification. With increasing awareness of the environmental and health impacts of per- and polyfluoroalkyl substances (PFAS), regulatory documents have expanded significantly, varying in scope and product relevance, making interpretation challenging for manufacturers and policymakers. We introduce RegExtract, a benchmark for structured information extraction from PFAS regulatory documents, comprising 488 human-annotated field instances across 20 US state documents and 46 regulations. Each field instance includes a value and a provenance entry, yielding 976 entries in total. We define a typed, multi-field schema targeting key elements such as material definitions, permitted uses, and product exemptions, and propose provenance-aware evaluation metrics tailored to regulatory text. Through experiments with open- and closed-source LLMs across prompting strategies, document condensing methods, and schema partitioning approaches, we find that document identification fields remain challenging, and that field-specific few-shot examples improve extraction. This work lays the groundwork for AI tools for hazardous substance identification, question answering, and compliance monitoring. The code and data are available at https://github.com/IBM/PFAS-regulations.

Chinese Translation

从监管文件中提取结构化信息是一项关键但具有挑战性的任务。相关事实散布在冗长且法律密集的文本中，提取事实与其来源之间的可追溯性对于验证至关重要。随着人们对全氟和多氟烷基物质（PFAS）对环境和健康影响的认识不断提高，监管文件显著增加，范围和产品相关性各异，使得制造商和政策制定者在解读时面临挑战。我们提出了RegExtract，这是一个用于从PFAS监管文件中提取结构化信息的基准，包含488个人工标注的字段实例，涵盖20份美国州文件和46项法规。每个字段实例包括一个值和一个来源条目，总计976个条目。我们定义了一种类型化的多字段模式，针对材料定义、允许用途和产品豁免等关键元素，并提出了针对监管文本的关注来源的评估指标。通过对开放源和闭源大型语言模型在提示策略、文档压缩方法和模式划分方法上的实验，我们发现文档识别字段仍然具有挑战性，且特定字段的少量示例能够改善提取效果。这项工作为有害物质识别、问答和合规监测的人工智能工具奠定了基础。代码和数据可在 https://github.com/IBM/PFAS-regulations 获取。

View on arXiv Download PDF AI Translation

cs.CL / 42 / 2610.15134

Lumen-bench: Cross-Lingual Safety Gaps and the Say-Do Problem in Tool-Calling Large Language Models

Lumen-bench：跨语言安全差距与大型语言模型工具调用中的言行不一问题

Bonás, Giovana Kerche, Laitz, Thiago, Almeida, Thales Sales, Santos, João Guilherme Alves

Abstract

When a large language model refuses an unethical request, what does refusal mean? In conversation, refusal is bounded by what the model says; once the model is equipped with executable tools, what it does can diverge from what it says, a failure mode that English-only evaluations cannot fully reveal. We introduce Lumen-bench, a behavioral benchmark of tool-calling ethics spanning 30 models, five languages (Portuguese, English, Spanish, Mandarin Chinese, Hindi), 108 Portuguese-authored scenarios, six social-pressure conditions, two system prompts (ethical and neutral), and 21 generic tools. The evaluation comprises 194,400 trials. Judgments come from a reasoning-model judge (o4-mini), calibrated against one Portuguese-speaking author ($n=83$; harmful/non-harmful $\kappa=0.78$), compared with two other provider judges (Fleiss' $\kappa=0.78$, $n=607$), and complemented by retrospective sandbox replay. Four findings emerge. (i) Within-model comparisons support higher harmful-action rates in Hindi and Portuguese than English under the ethical prompt, and Hindi and Spanish under the neutral prompt, after multiple-testing correction. (ii) Premium models have lower median violation rates, although price does not guarantee safety or competence. (iii) Replacing the ethical system prompt with a neutral one raises pooled violation rates from 2.76\% to 13.64\%. (iv) In 8.57\% of substantive-call cases, refusal or ethical objection accompanies a harmful tool call: the judges consistently identify the say-do gap, although they differ in how many responses they assign to this category. Matched benign controls demonstrate legitimate task completion in 82.8\% of 3,300 trials. Code and reproduction scripts are available at https://github.com/TropicAI-Research/lumen.

Chinese Translation

当大型语言模型拒绝不道德的请求时，拒绝意味着什么？在对话中，拒绝受到模型所说内容的限制；一旦模型配备了可执行工具，其行为可能与所说内容不一致，这种失效模式是仅用英语评估无法完全揭示的。我们引入了Lumen-bench，这是一个涵盖30个模型、五种语言（葡萄牙语、英语、西班牙语、普通话、印地语）、108个葡萄牙语创作场景、六种社会压力条件、两个系统提示（伦理和中立）以及21个通用工具的工具调用伦理行为基准。评估包括194,400次试验。判断来自一个推理模型评判者（o4-mini），与一位讲葡萄牙语的作者进行校准（$n=83$；有害/无害 $ ext{kappa}=0.78$），并与另外两个提供者评判者（Fleiss' $ ext{kappa}=0.78$, $n=607$）进行比较，同时辅以回顾性沙盒重播。我们得出四个发现。(i) 在模型内部比较中，支持在伦理提示下，印地语和葡萄牙语的有害行为率高于英语，而在中立提示下，印地语和西班牙语的有害行为率高于英语，经过多重检验修正。(ii) 高端模型的中位违规率较低，尽管价格并不保证安全或能力。(iii) 将伦理系统提示替换为中立提示使得总体违规率从2.76\%上升至13.64\%。(iv) 在8.57\\%的实质性调用案例中，拒绝或伦理反对伴随有害工具调用：评判者一致识别出言行不一的差距，尽管他们在分配到该类别的响应数量上存在差异。匹配的良性对照在3,300次试验中展示了82.8\%的合法任务完成率。代码和重现脚本可在https://github.com/TropicAI-Research/lumen获取。

View on arXiv Download PDF AI Translation

cs.CL / 43 / 2610.15154

Where Did We Go Wrong? Localizing Errors to Repair Multi-Turn Conversations

我们哪里出错了？定位错误以修复多轮对话

Morosini, Alessandro, Sun, Wei, Perakis, Georgia

Abstract

Large language model (LLM) assistants are increasingly used in multi-turn conversations, where the user's intent is often unclear at first and revealed only gradually over time. The assistant must therefore interpret each exchange and form inferences about the user's intent, using them to update its working state. These updates usually refine the assistant's understanding, but a misinterpretation can instead corrupt its working state by ruling out the true intent, influencing subsequent turns and ultimately causing the conversation to fail. Efficient recovery then requires locating and revising the misinterpretation without discarding information that remains valid. We introduce Loci (Localization Of Corrupted Inferences), a framework for probabilistically localizing misinterpretations and selectively repairing the assistant's working state. Given an observed failure, Loci reasons over possible user intents and assigns each update a probability of having excluded the true intent. These probabilities rank suspected updates and quantify the model's confidence in the localization of the underlying misinterpretations. Loci then uses this information in a repair policy that decides whether to retain, verify, or remove suspected updates, enabling recovery from corrupted inferences while preserving useful information. To evaluate Loci's ability to identify misinterpretations, we first test it in a controlled setup with injected errors. We find that Loci improves localization over the baselines, with its advantage becoming larger when multiple errors occur. We then evaluate it in free-form conversations, where misinterpretations arise naturally, and measure the interactions needed to complete the task after an initial failure. Using localization confidence to guide repair, Loci reduces mean recovery cost by 46-79% relative to the strongest baseline across multiple domains.

Chinese Translation

大型语言模型（LLM）助手在多轮对话中越来越多地被使用，其中用户的意图往往在最初并不明确，而是随着时间的推移逐渐显现。因此，助手必须解释每次交流并推断用户的意图，并利用这些推断来更新其工作状态。这些更新通常会细化助手的理解，但错误的解释可能会通过排除真实意图而损害其工作状态，从而影响后续的对话，最终导致对话失败。因此，有效的恢复需要定位并修正错误的解释，同时不丢弃仍然有效的信息。我们提出了 Loci（Localization Of Corrupted Inferences），这是一个用于概率性定位误解并选择性修复助手工作状态的框架。在观察到的失败情况下，Loci 对可能的用户意图进行推理，并为每次更新分配一个排除真实意图的概率。这些概率对可疑更新进行排序，并量化模型在定位潜在误解方面的信心。然后，Loci 利用这些信息制定修复策略，以决定是保留、验证还是移除可疑更新，从而在保留有用信息的同时恢复被损坏的推断。为了评估 Loci 识别误解的能力，我们首先在一个注入错误的受控环境中对其进行测试。我们发现 Loci 在定位方面优于基线，且在发生多个错误时其优势更为明显。随后，我们在自由形式的对话中评估其表现，在这些对话中误解自然产生，并测量在初始失败后完成任务所需的交互次数。通过使用定位信心指导修复，Loci 在多个领域相对于最强基线将平均恢复成本降低了 46%-79%。

View on arXiv Download PDF AI Translation

cs.CL / 44 / 2610.15158

Learning Proposition-Level Links from Sentence-Level Supervision for Neuro-Symbolic NLI

从句子级监督学习命题级链接用于神经符号自然语言推理

Feng, Xuyao, Bikakis, Antonis

Abstract

We study neuro-symbolic natural language inference (NLI) with independently translated propositional representations of the premise and hypothesis. Before symbolic reasoning, the system must infer semantic links between their proposition-level units (atoms). Existing systems based on Abstract Meaning Representation (AMR) and Boolean satisfiability (SAT) identify these links using fixed embedding-based source selection and pretrained sentence-level NLI models not trained for atom-level linking. We introduce a trainable atom linker that jointly predicts link existence, source selection, and relation type. Without atom-pair training annotations, we treat cross-formula link structures as latent and train the linker from sentence-level NLI labels, keeping the translator and SAT reasoner fixed. We evaluate on the full public test splits of Stanford Natural Language Inference (SNLI) and Sentences Involving Compositional Knowledge (SICK), and the matched and mismatched validation splits of Multi-Genre Natural Language Inference (MNLI). Our system improves accuracy by 11.40-27.09 percentage points over a baseline applying sentence-trained NLI models unchanged to atom pairs within the same SAT pipeline. It also improves over the prior fixed-linker AMR-to-SAT system by 8.14-26.32 percentage points on SNLI and both MNLI splits, with a smaller gain on SICK.

Chinese Translation

我们研究了神经符号自然语言推理（NLI），使用独立翻译的前提和假设的命题表示。在进行符号推理之前，系统必须推断它们的命题级单元（原子）之间的语义链接。现有基于抽象意义表示（AMR）和布尔可满足性（SAT）的系统使用固定的嵌入基础源选择和未针对原子级链接训练的预训练句子级NLI模型来识别这些链接。我们引入了一种可训练的原子链接器，它联合预测链接存在性、源选择和关系类型。在没有原子对训练注释的情况下，我们将跨公式链接结构视为潜在的，并从句子级NLI标签中训练链接器，同时保持翻译器和SAT推理器不变。我们在斯坦福自然语言推理（SNLI）和涉及组合知识的句子（SICK）的完整公共测试集，以及多体裁自然语言推理（MNLI）的匹配和不匹配验证集上进行了评估。我们的系统在准确性上比基线提高了11.40-27.09个百分点，基线应用了未改变的句子训练NLI模型到同一SAT管道内的原子对。它在SNLI和两个MNLI分割上比之前的固定链接AMR到SAT系统提高了8.14-26.32个百分点，在SICK上则有较小的增益。

View on arXiv Download PDF AI Translation

cs.CL / 45 / 2610.15163

Sanzaru: Benchmarking Protocol Formation under Epistemic & Communication Asymmetry

Sanzaru：在认知与沟通不对称下的协议形成基准测试

Li, Yiping, An, Zhiyu, Du, Wan

Abstract

A language model's ability to solve a task individually does not ensure that a team can solve it under unequal access to information and communication. SANZARU examines this gap through cooperative bomb-defusal puzzles adapted from BOMBANANA! (Lefto Studio, 2026). Task information and action authority are divided among three agents: one cannot hear, another cannot speak, and the third cannot see. The benchmark contains 120 puzzle instances across seven families, each targeting a distinct coordination ability. Across 1512 episodes, 7 models achieve an overall success rate of 49%, ranging from 20% to 86% under the evaluated configurations. Teams develop or adopt diverse communication protocols, from simple gesture codes to Morse and encodings of entire manuals. Yet trace analysis attributes about 60% of 771 team losses to communication, and agents attempt to use channels unavailable to their teammates despite explicit role instructions. Together, task outcomes and interaction traces enable evaluation of both collaborative performance and perspective taking throughout cooperation. The benchmark is available at https://github.com/markli404/sanzaru-benchmark.

Chinese Translation

语言模型单独解决任务的能力并不保证团队能够在信息和沟通不平等的情况下解决该任务。SANZARU通过改编自BOMBANANA!（Lefto Studio, 2026）的合作拆弹谜题来研究这一差距。任务信息和行动权限在三个代理之间分配：一个无法听见，另一个无法说话，第三个无法看见。该基准测试包含来自七个家庭的120个谜题实例，每个实例针对一种不同的协调能力。在1512个回合中，7个模型的整体成功率为49%，在评估的配置下范围从20%到86%不等。团队开发或采用多样的沟通协议，从简单的手势代码到摩尔斯电码和整个手册的编码。然而，追踪分析将约60%的771次团队失败归因于沟通，尽管有明确的角色指示，代理仍试图使用对其队友不可用的渠道。任务结果和互动轨迹共同使得评估合作过程中的协作表现和视角采纳成为可能。该基准测试可在 https://github.com/markli404/sanzaru-benchmark 获取。

View on arXiv Download PDF AI Translation

cs.CL / 46 / 2610.15165

Active Learning for Finding a Needle in a Haystack

主动学习在大海捞针中的应用

Silcock, Emily, Löwe, Simon

Abstract

Training a classifier to find a rare class of documents, such as hate speech directed at a particular religious group, requires labelled examples that are expensive to obtain when the class makes up only a small fraction of a corpus. Active learning reduces this cost, but it must start from an initial labelled set that contains examples of the target class. Random sampling cannot supply these at reasonable cost, so they must be selected in some other way, such as keyword search, and any such selection biases the initial set towards part of the target class. We evaluate whether active learning can correct this bias. On AG News and Measuring Hate Speech, at a prevalence of around 1%, we compare random sampling with eight active learning strategies, starting from initial sets that cover or omit known subgroups of the target class, and measure recall on the omitted subgroups as well as F1. On AG News, almost every strategy recovers the omitted subgroup. On Measuring Hate Speech, uncertainty-based strategies recover less of the omitted subgroups than random sampling, whereas discriminative active learning recovers 55% of them without a significant loss of F1. Aggregate metrics conceal these differences.

Chinese Translation

训练一个分类器以寻找稀有类别的文档，例如针对特定宗教群体的仇恨言论，需要标记的示例，而这些示例在该类别仅占语料库的一小部分时获取成本高昂。主动学习可以降低这一成本，但它必须从一个包含目标类别示例的初始标记集开始。随机抽样无法以合理的成本提供这些示例，因此必须通过其他方式选择，例如关键词搜索，而任何此类选择都会使初始集偏向目标类别的一部分。我们评估主动学习是否能够纠正这种偏差。在 AG News 和 Measuring Hate Speech 数据集上，当目标类别的普遍性约为 1% 时，我们比较了随机抽样与八种主动学习策略，初始集覆盖或省略已知的目标类别子群，并测量省略子群的召回率以及 F1 值。在 AG News 数据集中，几乎每种策略都能恢复省略的子群。在 Measuring Hate Speech 数据集中，基于不确定性的策略恢复的省略子群少于随机抽样，而判别式主动学习在没有显著损失 F1 的情况下恢复了 55% 的省略子群。聚合指标掩盖了这些差异。

View on arXiv Download PDF AI Translation

cs.CL / 47 / 2610.15179

UDHRBench: Benchmarking Human Rights in Large Language Models

UDHRBench：大型语言模型人权评估基准

Raad, Alak, Morssy, Noreldin, Suhail, Zeerak, Atiani, Serin, Jarrar, Mustafa

Abstract

Large language models are increasingly deployed in contexts requiring adherence to ethical and legal norms, yet their alignment with internationally recognized human rights standards remains underexplored. We introduce UDHRBench, a large-scale benchmark for operationalising the evaluation of large language model compliance with the Universal Declaration of Human Rights. The dataset comprises 1,124 human rights confronting questions covering all 30 articles and 4,496 responses from four large language models (ChatGPT, Grok, Gemini, and Fanar), manually annotated by experts for alignment, with high inter-annotator agreement ($\kappa = 0.801$). Our analysis reveals systematic and non-trivial misalignments, showing that violations typically arise through seemingly rational reasoning: appeals to efficiency, risk management, or scientific neutrality, rather than explicit rejection of rights. We further propose an LLM-as-a-Judge framework for automated evaluation using zero-shot and few-shot prompting, achieving 89.8\% weighted F1-score in detecting human-rights misalignment, alongside a novel Alignment Leniency Score that quantifies how judge models resolve ambiguous cases. The dataset, annotations, and code will be released as open-source resources (https://sina.birzeit.edu/UDHRBench).

Chinese Translation

大型语言模型越来越多地应用于需要遵循伦理和法律规范的场景，但它们与国际公认的人权标准的一致性仍然未得到充分探索。我们介绍了UDHRBench，这是一个大规模基准，用于实现对大型语言模型遵循《世界人权宣言》的评估。该数据集包含1,124个涉及人权的问题，涵盖所有30条条款，以及来自四个大型语言模型（ChatGPT、Grok、Gemini和Fanar）的4,496个响应，这些响应由专家手动标注以评估一致性，标注者之间的高一致性（$ ext{kappa} = 0.801$）。我们的分析揭示了系统性和非琐碎的不一致，表明违反人权通常通过看似合理的推理产生：对效率、风险管理或科学中立性的诉求，而不是对权利的明确拒绝。我们进一步提出了一种LLM-as-a-Judge框架，用于通过零-shot和少量-shot提示进行自动评估，在检测人权不一致性方面实现了89.8%的加权F1分数，并提出了一种新的对齐宽容评分（Alignment Leniency Score），量化法官模型如何解决模糊案例。该数据集、标注和代码将作为开源资源发布（https://sina.birzeit.edu/UDHRBench）。

View on arXiv Download PDF AI Translation

cs.CL / 48 / 2610.15186

SAFE-KV: Causal Suffix Repair for Behavior-Preserving Non-Prefix KV Cache Reuse in RAG

SAFE-KV：用于行为保留的非前缀 KV 缓存重用的因果后缀修复

Zhang, Xingjian

Abstract

Retrieval-augmented generation (RAG) reuses the same documents across many requests, making non-prefix KV-cache reuse attractive. Yet a cached document state is not reusable text: it is a context-conditioned causal state whose keys and values depend on the prefix, position, and preceding documents under which it was computed. We formulate non-prefix KV reuse as a behavior-preservation problem with full prefill on the exact prompt as the reference, and introduce SAFE-KV, which repairs a composed cache by re-encoding the causal suffix from a selected start document. On Qwen2.5-7B-Instruct, prefix-conditioned composition preserves only 47.2% of HotpotQA-500 and 37.8% of 2Wiki-500 full-prefill answers; SAFE-KV with risk-guided K2 selection restores 99.8%/98.4% (donor-copy protocol) and 100.0%/98.4% with actual suffix forwarding. Document-only and fixed-window repairs recover far less, because the document states after the repaired region keep histories that omit it. Reuse without repair is exact up to four documents and degrades at sixteen, where suffix repair recovers 98-100%. Faithful repair re-encodes most of the context, so in a single-request L40S study SAFE-KV's operating rule serves every workload by full prefill. SAFE-KV thus establishes the causal-state contract for non-prefix reuse and a validated repair mechanism that meets it.

Chinese Translation

检索增强生成（RAG）在多个请求中重用相同的文档，使得非前缀 KV 缓存重用变得具有吸引力。然而，缓存的文档状态并不是可重用的文本：它是一个上下文条件的因果状态，其键和值依赖于计算时的前缀、位置和先前文档。我们将非前缀 KV 重用形式化为一个行为保留问题，以确切提示的完整预填充作为参考，并引入了 SAFE-KV，它通过重新编码从选定起始文档获得的因果后缀来修复组合缓存。在 Qwen2.5-7B-Instruct 上，前缀条件组合仅保留了 HotpotQA-500 的 47.2% 和 2Wiki-500 的 37.8% 的完整预填充答案；使用风险引导 K2 选择的 SAFE-KV 恢复了 99.8%/98.4%（捐赠-复制协议）和 100.0%/98.4% 的实际后缀转发。仅文档和固定窗口修复的恢复效果远不如前者，因为修复区域后的文档状态保留了省略该区域的历史。未经修复的重用在四个文档内是精确的，在十六个文档时则下降，而后缀修复恢复了 98-100%。忠实的修复重新编码了大部分上下文，因此在单请求 L40S 研究中，SAFE-KV 的操作规则通过完整预填充服务于每个工作负载。因此，SAFE-KV 建立了非前缀重用的因果状态契约和满足该契约的验证修复机制。

View on arXiv Download PDF AI Translation

cs.CL / 49 / 2610.15203

Do LLMs differentiate linguistic traits internally?: A layer-wise analysis

大型语言模型是否在内部区分语言特征？：逐层分析

Bollepally, Samhita, Yamauchi, Takashi, Muralee, Manju, Adebimpe, Motunrayo

Abstract

Large language models often produce judgments that resemble human responses, but it remains unclear whether they internally capture linguistic nuances such as conventional versus idiomatic meaning. A layer-wise representational analysis was conducted on two equal-sized pre-trained models, Llama-3.2-1B and Gemma-3-1B-PT, to examine whether and where these models differentiate linguistic traits from input. The analysis combined layer-wise probing with geometric distance measures, using the top embedding dimensions ranked by principal component loadings for each layer. The results show that even with equal capacity, models organize linguistic information in fundamentally different ways, suggesting the need to analyze internal representational geometry, examining not just behavioral outputs but semantic space alignment when evaluating how closely human and model language interpretation correspond.

Chinese Translation

大型语言模型通常产生类似于人类反应的判断，但尚不清楚它们是否在内部捕捉到语言细微差别，例如常规意义与习语意义之间的区别。我们对两个相同规模的预训练模型Llama-3.2-1B和Gemma-3-1B-PT进行了逐层表征分析，以检查这些模型是否以及在何处区分输入中的语言特征。该分析结合了逐层探测与几何距离度量，使用每层按主成分负载排名的顶级嵌入维度。结果表明，即使在相同的容量下，这些模型在组织语言信息的方式上存在根本性差异，这表明在评估人类与模型语言理解的对应关系时，需要分析内部表征几何，不仅要考察行为输出，还要考察语义空间的对齐情况。

View on arXiv Download PDF AI Translation

cs.CL / 50 / 2610.15208

PANINI-DR: Controlled Deep Research through Generative Semantic Workspaces

PANINI-DR：通过生成语义工作空间进行受控深度研究

Turali, M. Yigit, Rajesh, Shreyas, Lorasdagi, M. Efe, Duan, Chenda, Sharma, Kartik, Roychowdhury, Vwani

Abstract

Deep research requires connecting multiple steps of search, reasoning, and computation, with intermediate findings guiding what to investigate next. As investigations grow, agents must maintain research state across model calls and check findings before errors propagate into subsequent work. We introduce PANINI-DR, a multi-agent harness that separates coordination of the overall investigation from the research and verification of individual subquestions. A shared workspace, built on the Generative Semantic Workspace representation, records subquestions, their dependencies, findings, and supporting evidence. Researchers investigate individual subquestions, while a separate verification agent checks proposed answers against those subquestions and their evidence. Verification feedback guides further investigation, and recorded findings inform subsequent assignments. We evaluate PANINI-DR on BrowseComp-Plus, FRAMES, and MoNaCo against seven state-of-the-art deep-research harnesses, all using a shared language-model backbone and frozen corpora. PANINI-DR achieves the strongest average performance across the three benchmarks. On BrowseComp-Plus, it achieves 73.3% accuracy, outperforming the next-best harness by 16.0 points. Its advantage is strongest on hard questions, where it answers nearly twice as many as the next-best harness. Ablations further show that structured planning and persistent state improve performance, as does checking each finding against its cited evidence before it is reused. Because a model can answer correctly from memory alone, we also analyze the evidence retained by each harness. PANINI-DR achieves the highest answer recoverability from saved evidence on all three benchmarks, while complementary claim-support audits show stronger evidence-backed answers. Accounting for this evidence changes how harnesses compare and further strengthens PANINI-DR's advantage.

Chinese Translation

深度研究需要连接多个搜索、推理和计算步骤，中间发现指导下一步的调查。随着调查的深入，代理必须在模型调用之间维护研究状态，并在错误传播到后续工作之前检查发现。我们提出了PANINI-DR，这是一种多代理工具，将整体调查的协调与单个子问题的研究和验证分开。基于生成语义工作空间表示构建的共享工作区记录子问题、它们的依赖关系、发现及支持证据。研究人员调查单个子问题，而一个单独的验证代理则根据这些子问题及其证据检查提出的答案。验证反馈指导进一步的调查，记录的发现为后续任务提供信息。我们在BrowseComp-Plus、FRAMES和MoNaCo上评估PANINI-DR，与七个最先进的深度研究工具进行比较，所有工具均使用共享的语言模型骨干和冻结的语料库。PANINI-DR在这三个基准测试中实现了最强的平均性能。在BrowseComp-Plus上，它的准确率达到73.3%，比下一个最佳工具高出16.0个百分点。它在难题上的优势最为明显，回答的数量几乎是下一个最佳工具的两倍。消融实验进一步表明，结构化规划和持久状态提高了性能，同时在重用之前检查每个发现与其引用证据的一致性也有助于提升效果。由于模型可以仅凭记忆正确回答，我们还分析了每个工具保留的证据。在所有三个基准测试中，PANINI-DR在保存证据的回答可恢复性方面表现最佳，而互补的声明支持审计显示出更强的证据支持答案。考虑到这些证据改变了工具的比较方式，并进一步增强了PANINI-DR的优势。

View on arXiv Download PDF AI Translation

cs.CL / 51 / 2610.15258

Sequent-Prover: Training Agents for Formal, Checkable SMT-based Reasoning

序列证明者：为形式可检查的基于 SMT 的推理训练智能体

Ganguly, Debargha, Chen, Weicong, Hariri, Mohsen, Singh, Vikash, Sankar, Sreehari, Wang, Shouren, Yang, Wang, Zhang, Biyao, Song, Chaoda, Nemecek, Alexander, Ye, Kai, Zafar, Osama, Ayday, Erman, Stubbs, Joe, Iyengar, Srinivasan, Chaudhary, Vipin

Abstract

We introduce Sequent-Prover, a family of verifier-using language-model agents, and the post-training framework that produces them. Agents reason in an executable SMT-LIB scratchpad: they write complete SMT-LIB programs, query Z3 across turns, inspect solver feedback, repair formalizations, and terminate with an explicit answer. Our recipe, verifier-conditioned trajectory distillation, bootstraps from a noisy teacher proposal policy by sampling multi-turn trajectories, retaining the successful tail that is correct, non-trivial, grounded, and executable, then applying supervised fine-tuning and trajectory-level preference optimization. Because solver success alone does not guarantee semantic faithfulness, we evaluate task accuracy, k-sample reliability, and SMT program quality. Across 37 benchmarks and 7,254 problems per model family, post-training raises the 4B model's per-attempt accuracy from 85.6% to 97.0% and four-sample all-correct reliability (pass^4}) from 64.1% to 91.7%. It also reduces average SMT calls from 4.10 to 2.52 and tokens per attempt from 18.97k to 12.72k. Sequent-Prover agents give answers backed by executable formal artifacts.

Chinese Translation

我们介绍了序列证明者（Sequent-Prover），这是一个使用验证器的语言模型智能体系列，以及生成它们的后训练框架。智能体在可执行的 SMT-LIB 临时存储区中进行推理：它们编写完整的 SMT-LIB 程序，在多个回合中查询 Z3，检查求解器反馈，修复形式化，并以明确的答案结束。我们的方案，验证器条件轨迹蒸馏，从一个嘈杂的教师提案策略启动，通过采样多轮轨迹，保留成功的尾部，这些尾部是正确的、非平凡的、基于事实的和可执行的，然后应用监督微调和轨迹级偏好优化。由于单靠求解器的成功并不能保证语义的忠实性，我们评估任务的准确性、k样本的可靠性和 SMT 程序的质量。在 37 个基准和每个模型系列 7,254 个问题中，后训练将 4B 模型的每次尝试准确率从 85.6% 提高到 97.0%，四样本全对可靠性（pass^4）从 64.1% 提高到 91.7%。它还将平均 SMT 调用次数从 4.10 降低到 2.52，每次尝试的标记数从 18.97k 降低到 12.72k。序列证明者智能体提供由可执行的形式化工件支持的答案。

View on arXiv Download PDF AI Translation

cs.CL / 52 / 2610.15277

Do correct answers have stable latent knowledge support?

正确答案是否具有稳定的潜在知识支持？

Shi, Chuancheng, Li, Shangze, Wu, Wenhua, Hu, Kun, Shen, Fei, Wang, Zhiyong, Chua, Tat-Seng

Abstract

Large language models are typically evaluated in terms of the correctness of their output. However, output correctness alone does not confirm whether the latent knowledge supporting an answer has stable internal representations. We therefore propose latent state stability (LSS) and knowledge topology stability (KTS) to investigate the stability of individual knowledge states and their relational structure, respectively. Specifically, we first separately average the model's internal representations of the same knowledge under each elicitation condition. Using the mean representation under the baseline condition as a reference, we remove shared shifts to obtain corrected representations. We then formulate LSS as the distance between each corrected representation and its reference representation, normalized against the typical distance between knowledge items of the same type. We formulate KTS as the cross-condition consistency of the ordered pairwise representation distances among knowledge items of the same type. Next, we construct EpiStableBench, which comprises 4,411 items spanning factual knowledge and mathematical reasoning, with 76,766 controlled queries. Across 16 LLMs from four families, representation stability varies substantially even among items answered correctly under the base condition. Within models, higher stability generally corresponds to higher accuracy under held-out elicitation conditions; across models, stability does not consistently increase with model size or overall accuracy. Meanwhile, we present LatentDPO to learn a knowledge-editing policy that accounts for the stability of latent knowledge. Experimental results show that integrating LatentDPO with multiple existing editing methods consistently improves editing success and generalization across different expressions.

Chinese Translation

大型语言模型通常通过其输出的正确性进行评估。然而，仅凭输出的正确性并不能确认支持答案的潜在知识是否具有稳定的内部表征。因此，我们提出了潜在状态稳定性（Latent State Stability, LSS）和知识拓扑稳定性（Knowledge Topology Stability, KTS）来分别研究个体知识状态的稳定性及其关系结构。具体而言，我们首先在每个引导条件下分别对模型的相同知识的内部表征进行平均。以基线条件下的平均表征作为参考，我们去除共享偏移以获得修正后的表征。然后，我们将LSS定义为每个修正后表征与其参考表征之间的距离，经过标准化以适应同类知识项之间的典型距离。我们将KTS定义为同类知识项之间有序成对表征距离的跨条件一致性。接下来，我们构建了EpiStableBench，该基准包含4,411个项目，涵盖事实知识和数学推理，配有76,766个受控查询。在来自四个家族的16个大型语言模型中，即使在基线条件下正确回答的项目之间，表征的稳定性也存在显著差异。在模型内部，更高的稳定性通常与在保留引导条件下的更高准确性相关；而在模型之间，稳定性并不总是随着模型规模或整体准确性的增加而增加。同时，我们提出了LatentDPO，以学习一种考虑潜在知识稳定性的知识编辑策略。实验结果表明，将LatentDPO与多种现有编辑方法结合使用，能够持续提高编辑成功率和在不同表达方式下的泛化能力。

View on arXiv Download PDF AI Translation

cs.CL / 53 / 2610.15282

North or South? Diagnosing LLMs' Understanding of North Korean Language, Society, and Knowledge

北方还是南方？诊断大型语言模型对朝鲜语言、社会和知识的理解

Park, Kiwoong, Song, Seyoung, Woo, Donghyun, Oh, Alice

Abstract

Large Language Models (LLMs) have made substantial progress in multilingual question answering, yet their ability to understand low-resource sociolinguistic contexts remains under-explored. We study North Korea as a unique testbed where decades of division have produced distinct differences from South Korea in vocabulary, orthography, norms, and ideological discourse. We introduce LENS (Language and sociocultural Evaluation for North Korean Society), a human-validated benchmark of 1,765 questions evaluating LLMs across three dimensions: linguistic robustness, sociocultural perspective alignment, and domain-specific knowledge. Evaluating diverse recent LLMs, we find a modest performance drop on North Korean language inputs, a larger gap in sociocultural perspective alignment, and severe limitations on domain-specific knowledge. Qualitative analysis reveals that models often conflate formal institutions with lived realities, project liberal-democratic assumptions, or rely on coarse stereotypes. Our results demonstrate that linguistic similarity alone does not guarantee sociocultural or domain-specific understanding, highlighting the critical need for grounded evaluation resources for underrepresented cultural and political contexts.

Chinese Translation

大型语言模型（LLMs）在多语言问答方面取得了显著进展，但它们在理解低资源社会语言学背景方面的能力仍然未得到充分探索。我们研究朝鲜作为一个独特的测试平台，几十年的分裂导致其在词汇、正字法、规范和意识形态话语上与韩国存在显著差异。我们引入了LENS（朝鲜社会语言和社会文化评估），这是一个经过人工验证的基准，包括1,765个问题，评估LLMs在三个维度上的表现：语言稳健性、社会文化视角一致性和领域特定知识。对多种近期LLMs的评估发现，在朝鲜语言输入上表现略有下降，社会文化视角一致性存在更大差距，领域特定知识则存在严重局限。定性分析表明，模型往往将正式机构与实际生活混为一谈，投射自由民主的假设，或依赖粗略的刻板印象。我们的结果表明，仅仅依靠语言相似性并不能保证社会文化或领域特定的理解，强调了为被低估的文化和政治背景提供扎实评估资源的迫切需求。

View on arXiv Download PDF AI Translation

cs.CL / 54 / 2610.15371

A Sovereign, Indic-First, Reasoning Model

一个主权的、以印地语为首的推理模型

Raj, Sutanshu Saksena, Pasunuti, Prasanjith, Chatterjee, Parijat, Maan, Daksh, Malik, Purusharth, Sarkar, Nayanjit, Benki, Avinash

Abstract

Gnani-Evon-v3.3-30B-A3B is a long-context hybrid Mamba-Transformer mixture-of-experts model for reasoning, instruction following, and tool use in English and eight Indian languages. Starting from NVIDIA's Nemotron-3-Nano-30B-A3B checkpoint, it retains the 52-layer backbone and changes the vocabulary, training data, and post-training objectives. An in-place expansion from 131,072 to 163,072 tokens reduces mean Indian-language fertility by 38.6%. The new embedding rows are initialized from their upstream decompositions and warmed before the backbone is updated. Continual pretraining then adds broad English-Indic data, high-quality annealing, and direct training at a 131,072-token sequence length. The central result appears after post-training. Reinforcement learning across four verifiable task environments improves mathematics, graduate-level science, and instruction following, but degrades Indic extractive question answering because the rewards do not require an answer in the requested language. The share of Indic extraction items answered in English rises from 8.7% to 91.7%. Accuracy conditional on answering in the target script roughly doubles, which is consistent with preserved reading ability but is measured on a self-selected subset. Two unreleased studies test repairs. A scoped Indic reward restores the language convention. A 26B-A4B generative judge outperforms a 235B-parameter model on RM-Bench, and distillation restores 81% of the capabilities displaced by optimizing against it. Relative to the architecture-identical starting checkpoint, the released model gains 5.2 points on an Indic aggregate while giving up 1.7 points of English retention. The unreleased Indic-aware branch trades further, gaining 14.6 Indic points at a cost of 5.4 in English. A planned FP8 release reduces the artifact to 32GB with near-zero mean change across 32 paired benchmarks.

Chinese Translation

Gnani-Evon-v3.3-30B-A3B 是一个长上下文混合型 Mamba-Transformer 专家混合模型，旨在进行推理、遵循指令和工具使用，支持英语和八种印度语言。从 NVIDIA 的 Nemotron-3-Nano-30B-A3B 检查点开始，该模型保留了 52 层主干，并更改了词汇、训练数据和后训练目标。通过将令牌数量从 131,072 扩展到 163,072，平均印度语言的生育率降低了 38.6%。新的嵌入行从其上游分解中初始化，并在更新主干之前进行预热。随后，持续的预训练增加了广泛的英语-印地语数据、高质量的退火和在 131,072 令牌序列长度下的直接训练。中央结果出现在后训练阶段。通过四个可验证任务环境的强化学习提高了数学、研究生级科学和遵循指令的能力，但由于奖励不要求以请求的语言给出答案，印地语提取式问答的表现下降。用英语回答的印地语提取项目的比例从 8.7% 上升到 91.7%。在目标脚本下回答的准确性大致翻倍，这与保留的阅读能力一致，但是在自选子集上进行测量。两项未发布的研究测试了修复方案。一个有针对性的印地语奖励恢复了语言规范。一个 26B-A4B 生成评判器在 RM-Bench 上的表现优于 235B 参数模型，而蒸馏恢复了 81% 的因优化而丧失的能力。相对于架构相同的起始检查点，发布的模型在印地语聚合上获得了 5.2 分，同时放弃了 1.7 分的英语保留。未发布的印地语感知分支进一步交易，获得了 14.6 分的印地语，同时损失了 5.4 分的英语。计划中的 FP8 发布将工件减少到 32GB，并在 32 个配对基准测试中几乎没有变化。

View on arXiv Download PDF AI Translation

cs.CL / 55 / 2610.15373

In-Place Indic Tokenizer Expansion and Continual Pretraining of Nemotron-3-Nano-30B-A3B

就地扩展印度语言标记器及Nemotron-3-Nano-30B-A3B的持续预训练

Benki, Avinash, Pasunuti, Prasanjith, Raj, Sutanshu Saksena, Chatterjee, Parijat, Sarkar, Nayanjit, Maan, Daksh

Abstract

We adapted Nemotron-3-Nano-30B-A3B for eight Indian languages without changing its backbone architecture. We expanded the vocabulary in place from 131,072 to 163,072 entries using continued byte-pair encoding and AdaptMix rebalancing, reducing tokens per word by 38.6%. We initialized new input and output rows by averaging their original constituent rows, then trained them on 90% Indic and 10% English text before updating the backbone. Continual pretraining then progressed through broad training, high-quality annealing, and long-context training over approximately 1.27 trillion tokens. The annealed checkpoint gained 12.9 points on the Indic MILU benchmark but lost 3.4 on English MMLU. We also report practical negative results. Direct token addition underperformed continued byte-pair encoding. AdamW weight decay changed rows intended to stay fixed. Perplexity on an Indic-heavy holdout did not select the learning rate that best preserved English performance.

Chinese Translation

我们在不改变Nemotron-3-Nano-30B-A3B的主干架构的情况下，为八种印度语言进行了适配。我们通过持续的字节对编码和AdaptMix重平衡，将词汇量从131,072扩展到163,072个条目，减少了每个单词的标记数量38.6%。我们通过对原始组成行进行平均来初始化新的输入和输出行，然后在90%的印度语言文本和10%的英语文本上进行训练，随后更新主干。持续预训练通过广泛训练、高质量退火和长上下文训练，覆盖了大约1.27万亿个标记。退火检查点在印度MILU基准测试中获得了12.9分的提升，但在英语MMLU中损失了3.4分。我们还报告了一些实际的负面结果。直接添加标记的表现不如持续的字节对编码。AdamW权重衰减改变了原本应保持固定的行。在以印度语言为主的保留集上的困惑度未能选择出最佳保持英语表现的学习率。

View on arXiv Download PDF AI Translation

cs.CL / 56 / 2610.15412

From Candidate Tokens to Final Outcomes: Decoding Relative Long-Horizon Candidate Value from Reasoning Model Representations

从候选标记到最终结果：从推理模型表示中解码相对长远候选值

Yang, Yixin, Wang, Yue, Meng, Fandong, Sui, Zhifang

Abstract

Large reasoning models generate reasoning trajectories token by token. At a given prefix, multiple next-token candidates may be plausible, yet shape subsequent reasoning and differ in their chances of leading to the correct answer. Whether these differences are reflected in internal representations remains unclear. To study this question, we define relative long-horizon candidate value to quantify within-prefix differences and construct the Reasoning Candidate-Token Value (RCTV) dataset, pairing mathematical reasoning prefixes with plausible next-token candidates and their empirical long-horizon candidate values. Using RCTV, we examine hidden states, multilayer perceptron outputs, and attention outputs at candidate positions with sparse linear probes. Experimental results provide evidence that relative long-horizon candidate value is linearly decodable from these representations at multiple layers. Candidate-value probes can also outperform probability- and embedding-based baselines. Furthermore, a controlled online intervention experiment shows that using the decoded signals to guide token selection during generation can improve the chances of eventual success. Our findings provide evidence that internal representations encode information about relative long-horizon candidate value, offering insights into how local candidate-level representations relate to long-horizon reasoning outcomes.

Chinese Translation

大型推理模型逐个标记生成推理轨迹。在给定的前缀下，多个下一个标记候选项可能是合理的，但它们会影响后续推理并在导致正确答案的可能性上有所不同。这些差异是否反映在内部表示中仍不清楚。为研究这个问题，我们定义了相对长远候选值，以量化前缀内的差异，并构建了推理候选标记值（Reasoning Candidate-Token Value, RCTV）数据集，将数学推理前缀与合理的下一个标记候选项及其经验长远候选值配对。使用RCTV，我们通过稀疏线性探测器检查候选位置的隐藏状态、多层感知器输出和注意力输出。实验结果提供了证据，表明相对长远候选值可以从这些表示中在线性解码，并且在多个层次上均适用。候选值探测器的表现也优于基于概率和嵌入的基线。此外，一项受控的在线干预实验表明，使用解码信号来指导生成过程中的标记选择可以提高最终成功的机会。我们的发现提供了证据，表明内部表示编码了关于相对长远候选值的信息，为局部候选级表示如何与长远推理结果相关提供了见解。

View on arXiv Download PDF AI Translation

cs.CL / 57 / 2610.15415

Sensitivity or Criterion? Measuring and Moving When Language Models Act

敏感性还是标准？当语言模型行动时的测量与移动

Dong, Sixun, Hu, Yebowen, Yin, Ming, Chen, Yiran, Liu, Fei, Chen, Chen

Abstract

Before an agent acts, it must decide whether action is warranted. A single accuracy cannot distinguish failure to recognize a warranted action from reluctance to act. Signal-detection theory separates sensitivity from criterion, but criteria estimated from authored binary labels depend on each benchmark's example mix. We introduce FlipAct, a diagnostic of tool use in physical states. A rule over annotated tool properties computes each answer and a graded evidence score, fixing the Yes/No boundary at zero across models. Questions form 2-by-2 quads that hold required knowledge fixed while the state flips. The slope of the fitted Yes-rate curve measures sensitivity; its crossing at one half measures criterion. Across 44 open models, sensitivity rises with size within a family, and its ranking agrees with three of four external when-to-act benchmarks. Models respond to evidence sign rather than magnitude, accepting actions that leave the state no better about as often as helpful ones. When Yes means act, most models are conservative across rephrasings; in some, this partly reflects a preference for No. Criterion is not determined by size. Activation steering shifts criterion with weights fixed; distillation from a model's own steered copy puts the shift into its weights without a significant sensitivity change. The distilled shift transfers to six external benchmarks, including two without Yes/No answers, while criterion-free AUC is preserved on the five benchmarks where it is available. Accuracy improves where models start conservative and declines or remains within error where they start permissive. Project page: https://sixundong.com/projects/flipact

Chinese Translation

在一个智能体采取行动之前，它必须决定是否有必要采取行动。单一的准确性无法区分未能识别必要行动与不愿意采取行动之间的差异。信号检测理论将敏感性与标准分开，但从作者的二元标签估计的标准依赖于每个基准的示例组合。我们引入了 FlipAct，这是一种用于物理状态下工具使用的诊断工具。对注释工具属性的规则计算每个答案和一个分级证据分数，将 Yes/No 边界固定在零，适用于所有模型。问题形成 2x2 的四元组，保持所需知识不变，同时状态发生翻转。拟合的 Yes 率曲线的斜率测量敏感性；其在一半处的交叉测量标准。在 44 个开放模型中，敏感性随着同一家族中模型的规模增加而上升，其排名与四个外部的行动时机基准中的三个一致。模型对证据的符号而非大小做出反应，接受那些使状态没有改善的行动与有帮助的行动几乎同样频繁。当 Yes 意味着行动时，大多数模型在不同表述中表现出保守；在某些模型中，这在一定程度上反映了对 No 的偏好。标准并不由规模决定。激活引导在权重固定的情况下改变标准；从模型自身的引导副本进行蒸馏将这种变化放入其权重中，而不会显著改变敏感性。蒸馏的变化转移到六个外部基准，包括两个没有 Yes/No 答案的基准，而在五个可用的基准上保持无标准的 AUC。准确性在模型开始保守时提高，而在模型开始宽容时下降或保持在误差范围内。项目页面：https://sixundong.com/projects/flipact

View on arXiv Download PDF AI Translation

cs.CL / 58 / 2610.15488

AlphaPilot: Multi-Turn Tool-Using LLM Agents for Adaptive Alpha Mining with Reinforcement Learning

AlphaPilot：用于自适应阿尔法挖掘的多轮工具使用LLM代理的强化学习

Jiang, Peiqi, Zhang, Fan, Xu, Xin, Chen, Tianhao, Yang, Can

Abstract

Formulaic alpha mining seeks interpretable expressions that rank assets by future returns. The task is challenging because it demands simultaneous processing of unstructured market descriptions, historical data, and quantitative evaluation of candidate factors, yet most existing methods exploit only a subset of these signals. We introduce AlphaPilot, a reinforcement-learning framework that trains multi-turn tool-using LLM agents for adaptive alpha mining. The agent conditions on market descriptions, proposes factor expressions, queries real-data evaluator feedback, and updates a retained factor pool across turns. This design jointly leverages market text, LLM financial knowledge, real data, and evaluation feedback. At inference time, AlphaPilot performs day-by-day search and pool updates to adapt to shifting market regimes. Across three independent runs in out-of-sample rolling backtests on CSI300 and CSI500, AlphaPilot achieves mean total returns of 42.93% and 54.19%, with mean Sharpe ratios of 2.46 and 2.86, respectively. It attains the highest mean total return, Sharpe, and RankIC among the evaluated methods on both universes.

Chinese Translation

公式化阿尔法挖掘旨在寻找可解释的表达式，以根据未来收益对资产进行排名。该任务具有挑战性，因为它要求同时处理非结构化市场描述、历史数据以及对候选因子的定量评估，而现有的大多数方法仅利用这些信号中的一部分。我们提出了AlphaPilot，一个强化学习框架，训练多轮工具使用的LLM代理以进行自适应阿尔法挖掘。该代理以市场描述为条件，提出因子表达式，查询真实数据评估反馈，并在多个回合中更新保留的因子池。该设计共同利用了市场文本、LLM金融知识、真实数据和评估反馈。在推理时，AlphaPilot进行逐日搜索和池更新，以适应市场环境的变化。在对CSI300和CSI500进行的三次独立的样本外滚动回测中，AlphaPilot的平均总收益率分别为42.93%和54.19%，平均夏普比率分别为2.46和2.86。在评估的两种市场中，它在平均总收益、夏普比率和RankIC方面均达到了最高水平。

View on arXiv Download PDF AI Translation

cs.CL / 59 / 2610.15496

Concept Representations in Vector Space: Quantifying Sharedness and Idiosyncrasy of Meaning

向量空间中的概念表征：量化意义的共性与特异性

Durstewitz, Simon Josef, Buehler, Denis

Abstract

Concepts are the basic units through which we navigate the world and communicate meaning. But do you and I really mean the same thing when we think or speak about a concept? The present study develops a measurement framework for modeling and analyzing interindividual variation in concept representations through sentence embeddings, dense vector representations mapping entire utterances into a shared geometric space. The framework is developed in a theoretical argument and validated, demonstrated, and subjected to robustness checks in a pre-registered study. Short personal definitions, elicited from $N = 125$ participants, are encoded as sentence embeddings and submitted pairwise to a crossed random effects model yielding two concept-level quantities: sharedness ($S_b$), the expected dissimilarity between two randomly drawn individuals, and idiosyncrasy ($I_b$), the person-specific share of total dyadic variance. Two validity tests support the approach: the embedding space fully recovers concept and domain organization from participant definitions even with the target word excluded, and a parameter-free classifier achieves 0.630 accuracy in a within-participant extensional prediction task against a chance baseline of 0.500 and a consistency ceiling of 0.893. The framework finds theoretically interpretable variation in $S_b$ and establishes $I_b$ as a reliable source of dyadic variance across concepts. Results replicate across three encoders varying in architecture and capacity, providing a reproducible and scalable instrument for the empirical study of meaning variation.

Chinese Translation

概念是我们导航世界和传达意义的基本单元。但当我们思考或谈论一个概念时，你我真的意味着同样的事情吗？本研究开发了一个测量框架，用于通过句子嵌入建模和分析个体间概念表征的变异，句子嵌入是将整个话语映射到共享几何空间的密集向量表示。该框架在理论论证中开发，并在预注册研究中进行了验证、演示和稳健性检查。从125名参与者中引出的简短个人定义被编码为句子嵌入，并成对提交给交叉随机效应模型，产生两个概念层面的量：共性（$S_b$），即两个随机抽取个体之间的预期差异性，以及特异性（$I_b$），即个体特有的总二元方差份额。两个有效性测试支持该方法：嵌入空间即使在目标词被排除的情况下也能完全恢复参与者定义中的概念和领域组织，并且一个无参数分类器在参与者内的外延预测任务中达到了0.630的准确率，超过了0.500的机会基线和0.893的一致性上限。该框架发现了$S_b$的理论可解释变异，并确立了$I_b$作为跨概念的可靠二元方差来源。结果在三种不同架构和能力的编码器中得到了重复，提供了一个可重复和可扩展的工具，用于实证研究意义变异。

View on arXiv Download PDF AI Translation

cs.CL / 60 / 2610.15513

REPRO: Reflective Evolutionary Prompt Optimizer for Automated LLM Red-Teaming

REPRO：用于自动化大型语言模型红队测试的反思进化提示优化器

Alotaibi, Abrar, Ahmed, Moataz

Abstract

Safety-aligned large language models (LLMs) remain vulnerable to adversarial prompts that bypass their safety training. Existing automated red-teaming methods either require privileged access unavailable in production, rely on manually engineered per-model templates, or optimize a single objective without population-level diversity. We present REPRO (Reflective Evolutionary Prompt Optimization), a pure black-box framework that discovers adversarial prompts through evolutionary optimization driven entirely by a compressed open-weight 7B reflective model on a single consumer GPU. REPRO introduces two mechanisms absent from prior work, both isolated in a component ablation: semantic crossover, an LLM-guided multi-parent recombination operator, and Bayesian strategy selection via Thompson Sampling informed by structured failure analysis. We catalog REPRO's behavior on AdvBench, JailbreakBench, and HarmBench across 36 target models from 11 families, including GPT-5-nano, Claude-3.5-Sonnet, Gemini-2.5-Pro, and Grok-4.1. Under our protocol REPRO reaches 100\% attack success rate (ASR) on every evaluated target in a single turn, at a median of 24 target queries per behavior. Two properties of the protocol qualify that figure. The judge scoring candidates during optimization also produces the reported ASR, and search halts when it accepts, so ASR reports whether search succeeded within the iteration budget rather than success at a fixed budget. All baseline figures are transcribed from their original papers, which use different judges, budgets, and model versions. We therefore present this as a catalog of model coverage under one protocol rather than a head-to-head ranking, and develop both qualifications in the paper.

Chinese Translation

安全对齐的大型语言模型（LLMs）仍然容易受到绕过其安全训练的对抗性提示的攻击。现有的自动化红队测试方法要么需要在生产环境中无法获得的特权访问，要么依赖于手动设计的每个模型模板，或者优化单一目标而缺乏种群级别的多样性。我们提出了REPRO（反思进化提示优化），这是一个纯黑箱框架，通过完全依赖于一个压缩的开放权重7B反思模型在单个消费级GPU上进行进化优化，从而发现对抗性提示。REPRO引入了两个在先前工作中缺失的机制，这两个机制在组件消融中被孤立：语义交叉，一种由LLM引导的多父代重组操作符，以及通过汤普森采样进行的贝叶斯策略选择，该策略由结构化失败分析提供信息。我们在AdvBench、JailbreakBench和HarmBench上对REPRO的行为进行了分类，涵盖了来自11个家族的36个目标模型，包括GPT-5-nano、Claude-3.5-Sonnet、Gemini-2.5-Pro和Grok-4.1。在我们的协议下，REPRO在每个评估目标上都能在单次尝试中达到100%的攻击成功率（ASR），每种行为的中位目标查询数为24。该协议的两个特性使得这一数据得以成立。在优化过程中对候选者进行评分的评判者也产生了报告的ASR，并且当其接受时搜索停止，因此ASR报告的是搜索是否在迭代预算内成功，而不是在固定预算下的成功。所有基线数据均从其原始论文中转录，这些论文使用了不同的评判者、预算和模型版本。因此，我们将其视为在一个协议下的模型覆盖目录，而不是逐一排名，并在论文中发展了这两个资格。

View on arXiv Download PDF AI Translation

cs.CL / 61 / 2610.15527

A Step in the Right Place: Understanding and Optimizing Timestep Placement in Continuous Diffusion Language Models

迈向正确的方向：理解和优化连续扩散语言模型中的时间步位置

Zheng, Shutian, Yu, Zhihong, Wei, Linye, Zhang, Zeyang, Li, Meng

Abstract

Continuous diffusion language models represent an emerging paradigm for language modeling. They generate text in parallel by iteratively refining continuous representations and mapping them back to tokens through a decoder. However, realizing the efficiency advantage of parallel generation requires maintaining generation quality with only a few denoising steps. Even when the number of steps is fixed, their placement along the denoising trajectory remains flexible. In this paper, we demonstrate that changing timestep placement alone can substantially alter the final latent representations and thereby affect the decoded token distributions. Building on these findings, we introduce StepTune, an algorithm that searches for timestep placements that improve generation quality. The search is conducted on a small calibration set and guided by a language-aware objective, with the model and sampling budget fixed. Once calibrated, the timestep placement is fixed and reused across inputs. Experiments demonstrate consistent improvements in unconditional generation perplexity and conditional task metrics across all evaluated sampling budgets. In particular, StepTune enables 16-step sampling with a frozen model to outperform its native 64-step baseline on both XSum summarization and WMT14 translation and reduces end-to-end latency by approximately $3.8\times$. These results establish timestep placement as a practical means of improving the quality-efficiency trade-off in continuous diffusion language models.

Chinese Translation

连续扩散语言模型代表了一种新兴的语言建模范式。它们通过迭代地精炼连续表示并通过解码器将其映射回标记，从而并行生成文本。然而，要实现并行生成的效率优势，需要在仅有少量去噪步骤的情况下保持生成质量。即使步骤数量固定，它们在去噪轨迹上的位置仍然是灵活的。在本文中，我们证明了仅改变时间步位置就可以显著改变最终的潜在表示，从而影响解码的标记分布。基于这些发现，我们引入了StepTune，这是一种搜索时间步位置以提高生成质量的算法。搜索在一个小的校准集上进行，并由一个语言感知目标引导，同时模型和采样预算保持固定。一旦校准，时间步位置将被固定并在输入中重复使用。实验表明，在所有评估的采样预算中，无条件生成的困惑度和条件任务指标均有一致的改善。特别是，StepTune使得使用冻结模型的16步采样在XSum摘要和WMT14翻译任务上超越其原生的64步基线，并将端到端延迟减少了约$3.8 imes$。这些结果确立了时间步位置作为改善连续扩散语言模型中质量与效率权衡的实用手段。

View on arXiv Download PDF AI Translation

cs.CL / 62 / 2610.15542

Apertus 1.5: From Open Text Models to Open Multimodal Reasoners

Apertus 1.5：从开放文本模型到开放多模态推理器

Apertus, Project, :, Xu, Yixuan, Turki, Yassine, Sabolčec, Vinko, Pyatkin, Valentina, Chu, Tommy, Najem-Meyer, Sven, Carroll, Sean F., Smith, Robert, Kreft, Raphael, Tazza, Oscar Alessandro, Lavrovsky, Oleg, Irmiger, Nicola, Santelmo, Matteo, Schneider, Marian, Giraldo, Juan García, Yukhymenko, Hanna, Ďurech, Eduard, Melikidze, Davit, Tamayo, Dani, Messmer, Bettina, Ahadinia, Aryan, Yadav, Anunay, Huang, Allen, Fraij, Ahmad, Bombarda, Alessandro, Padula, Alex, Hägele, Alexander, Sternfeld, Alexander, Hedström, Anna, Kosovskaia, Anna, Wuhrmann, Arthur, Balasubramaniam, Arundhati, Mahouri, Badr Al, Pasztor, Barna, Szostakiewicz, Bartosz, Doan, Bich Ngoc, Schmid, Bryan, Sezen, Burak, Challier, Camille, Charmillot, Clément, Shayakhmetova, Dana, Perruchoud, Daniel, Brüggemann, David, Isztl, David, Palme, Elia, Anastassova, Emma, Bösch, Fabian, Zahiragić, Faruk, Mohamed, Fawzi Roberto, Panicali, Fiona, Kolly, Florian, Guerra, Flynn, Monti, Francesco, Lübeck, Frederike, Oliva, Gennaro, Yang, Haofeng, Shang, Haoliang, Khoury, Helen El, Mendonça, Henrique, Vamvas, Jannis, Landolt, Jérome, Lam, Jessica, Xiao, Jingchun, Coles, Jonathan, Oyebanji, Joshua, Nguyen, Khanh, Pitas, Konstantinos, Strand, Lasse B., Mi, Li, Deslarzes, Loïc, Baggi, Lorenzo, Noci, Lorenzo, Baumann, Luca, Sartori, Luca, Drescher, Lukas, De Negri, Marco, Passerini, Marco, Klein, Mark, Stojanovic, Marko, Majsec, Martin, Ansaripour, Matin, Busato, Mattea, Pellonara, Mauro, Neuwinger, Max, Graber, Michael, Tešnar, Michal, Wastl, Michelle, Gila, Miguel, Boronski, Mikolaj, Magnin, Morgane, Reihs, Moritz, Sayfiddinov, Mukhammadali, Ranchin, Nathan, Cuveillier, Nicolas, Wohlfahrt, Oscar, Zou, Qianqian, Kamysheva, Rada, Sevastjanova, Rita, Machacek, Roman, Keller, Sebastian, Matrenok, Simon, Pierre, Simon, Schuppli, Stefano, Chen, Suei-Wen, Horváth-Mikulas, Szabina, Perez, Taliesin, Spinner, Thilo, Buening, Thomas Kleine, Robinson, Tim, Kwiecinski, Tymoteusz, Zouhar, Vilém, Thoma, Vinzenz, Yao, Xiaozhe, Metz, Yannick, Fukuhara, Yuhei, Askari, Yusif, Chen, Zixuan, Ilic, Alexander, Klimovic, Ana, Krause, Andreas, Kucharavy, Andrei, Ash, Elliott, VandeVondele, Joost, Rajman, Martin, El-Assady, Mennatallah, Schulthess, Thomas, Kehrer, Timo, Hoefler, Torsten, Bosselut, Antoine, Jaggi, Martin, Schlag, Imanol

Abstract

Language models are becoming infrastructure, and infrastructure built in the open, by public institutions, on documented data, is the alternative to dependence on a few providers; Apertus shows this is viable at scale. Apertus 1.5 extends the fully open Apertus family from text to multimodal reasoning. Continued pretraining of the Apertus 1.0 models, on 4T additional tokens for the 8B and 2T for the 70B, adds image and audio understanding and extends the context window fourfold to 262,144 tokens; multimodality enters through the tokenizer, the language-model head remains restricted to text, and the architecture is unchanged. Supervised finetuning, long-context finetuning, reinforcement learning with verifiable rewards, and preference alignment complete the released models. Thinking is a flag of the chat template rather than a separate model, and lifts mathematical reasoning by up to 20 points on the 8B model. The released models improve over their predecessors on nearly every benchmark; among fully open models of its class, the 8B leads on knowledge, instruction following, and cultural benchmarks, and its instruction following surpasses all compared models, including open-weight ones. On image and audio understanding, a first for the family, the models are competitive with fully open multimodal models, strongest on document understanding and long-form and lower-resource speech recognition, while specialized vision and audio models remain ahead on perception and transcription of English and other high-resource languages. As in Apertus 1.0, we train only on openly available data, respect crawler opt-outs retroactively, and exclude restrictively licensed material. We release the model weights, data preparation and training recipes, and evaluation suites under permissive licenses; openness extends to values, with the Apertus charter guiding alignment stages published alongside the models.

Chinese Translation

语言模型正逐渐成为基础设施，而由公共机构在公开数据上构建的开放基础设施是对少数供应商依赖的替代方案；Apertus 证明了这一点在大规模上是可行的。Apertus 1.5 将完全开放的 Apertus 家族从文本扩展到多模态推理。对 Apertus 1.0 模型的持续预训练，使用了 4T 的额外标记用于 8B 模型和 2T 用于 70B 模型，增加了图像和音频理解，并将上下文窗口扩展至四倍，达到 262,144 个标记；多模态性通过分词器引入，语言模型头仍然限制于文本，架构保持不变。监督微调、长上下文微调、具有可验证奖励的强化学习和偏好对齐完成了发布的模型。思考是聊天模板的一个标志，而不是一个单独的模型，并且在 8B 模型上提升了数学推理的得分高达 20 分。发布的模型在几乎所有基准测试中都优于其前身；在同类完全开放模型中，8B 在知识、指令跟随和文化基准上领先，其指令跟随超越了所有比较模型，包括开放权重模型。在图像和音频理解方面，作为该家族的首次尝试，这些模型与完全开放的多模态模型具有竞争力，在文档理解和长篇及低资源语音识别方面表现最强，而专门的视觉和音频模型在英语及其他高资源语言的感知和转录方面仍然领先。与 Apertus 1.0 一样，我们仅在公开可用的数据上进行训练，追溯性地尊重爬虫选择退出，并排除限制性许可材料。我们在宽松许可下发布模型权重、数据准备和训练配方以及评估套件；开放性延伸至价值观，Apertus 宪章指导与模型一起发布的对齐阶段。

View on arXiv Download PDF AI Translation

cs.CL / 63 / 2610.15544

An Interpretable Linguistically-Grounded Analysis of Scene Saliency in Movie Screenplays

电影剧本中场景显著性的可解释语言基础分析

Bindal, Uday, Karan, Ishaan, Garg, Manan, Goyal, Ekansh, Mishra, Rahul, Rajkumar, Rajakrishnan

Abstract

We investigate the factors affecting salience in movie screenplay by posing two research questions: 1. {\em Does Uniform Information Density (UID) predict salient scenes?} 2. {\em What other linguistic properties distinguish salient scenes from non-salient background?} We answer both questions through a controlled corpus study on the MENSA corpus, paired with a Linguistically-Grounded Salience Model (LGSM), a dual-stream classifier that fuses a frozen BERT encoder with explicit linguistic features through adaptive gating. The UID picture is layered in long-form discourse. Per-scene UID summaries carry no independent signal once length is controlled, but at the token level salient scenes match non-salient material in the bulk while specifically suppressing the rarest, hardest-to-predict tokens in the upper tail. UID therefore lives at the within-scene token level rather than as a per-scene aggregate. Salient scenes also show a moderate and sustained cast of central characters (following an inverted-U pattern), strong third-person usage, syntactically simpler language, and more abstract vocabulary about motivation and conflict. LGSM exploits this signature and improves recall on salient scenes over a strong semantic baseline while yielding better downstream summaries.

Chinese Translation

我们通过提出两个研究问题来探讨影响电影剧本中显著性的因素：1. { extit{统一信息密度（UID）是否能够预测显著场景？}} 2. { extit{还有哪些其他语言特性能够区分显著场景与非显著背景？}} 我们通过对MENSA语料库的控制语料研究以及结合语言基础显著性模型（Linguistically-Grounded Salience Model, LGSM）来回答这两个问题。LGSM是一种双流分类器，通过自适应门控将冻结的BERT编码器与显式语言特征融合。UID的图景在长篇话语中是分层的。每场景的UID摘要在长度控制后不再携带独立信号，但在标记级别上，显著场景在整体上与非显著材料相匹配，同时特别抑制了上尾中最稀有、最难预测的标记。因此，UID存在于场景内的标记级别，而不是作为每场景的聚合。显著场景还表现出适度且持续的核心角色阵容（遵循倒U型模式）、强烈的第三人称使用、语法上更简单的语言，以及关于动机和冲突的更抽象词汇。LGSM利用这一特征，在显著场景上提高了召回率，超越了强语义基线，同时产生了更好的下游摘要。

View on arXiv Download PDF AI Translation

cs.CL / 64 / 2610.15574

The Fragility and Dynamics of LLM Pruning-Recovery Steps

大语言模型剪枝-恢复步骤的脆弱性与动态性

Salim, Luis Frentzen, Rinartha, Komang, Wu, Chia-Chun, Van Nhiem, Tran, Ku, Lun-Wei, Li, Yung-Hui

Abstract

Structural pruning followed by a recovery stage is the standard recipe for compressing large language models (LLMs), yet the two stages are typically designed and tuned in isolation. In this work, we study how their interaction shapes the compressed model, sweeping ordering, the data that drives each stage, and the alignment of their training objectives. We find that instruction-following is the first capability to collapse under pruning, well before knowledge, reasoning, or core language modeling, and that much of the pruning in-domain loss is likely not knowledge loss but a domain-agnostic failure to respond coherently to the prompt. We find in our setting that in-domain data is the strongest pruning signal, whereas instruction-style data recovers the most. Stage order matters little, but only marginally. We further analyze how the importance ranking drifts before and after pruning. Together these results give practitioners a concrete recipe for composing pruning and recovery.

Chinese Translation

结构剪枝后随之进行的恢复阶段是压缩大语言模型（LLMs）的标准方法，然而这两个阶段通常是孤立设计和调整的。在本研究中，我们探讨了它们之间的相互作用如何塑造压缩模型、剪枝顺序、驱动每个阶段的数据以及它们训练目标的对齐。我们发现，遵循指令的能力是最早在剪枝过程中崩溃的能力，远早于知识、推理或核心语言建模，并且在领域内剪枝损失的很大一部分可能并非知识损失，而是对提示的领域无关的响应不连贯。在我们的研究中，我们发现领域内数据是最强的剪枝信号，而指令风格的数据则恢复最多。阶段顺序的重要性不大，但仅有微小的影响。我们进一步分析了剪枝前后重要性排名的漂移。综合这些结果，为实践者提供了一个具体的剪枝与恢复组合的方案。

View on arXiv Download PDF AI Translation

cs.CL / 65 / 2610.15578

Towards Robust Calibration Dataset for Domain-General Mixture of Experts Pruning

面向领域通用的专家混合模型剪枝的鲁棒校准数据集

Salim, Luis Frentzen, Wu, Chia-Chun, Van Nhiem, Tran, Ku, Lun-Wei, Li, Yung-Hui

Abstract

Post-training expert pruning is a promising deployment technique for Mixture-of-Experts (MoE) models. The calibration set is an integral part of the pruning process, however, its correlation with evaluation performance is largely misunderstood and unintuitive. In this work, we present an empirical study of the robustness of calibration perturbations, specifically in MoE models. Showing how calibration size, language, and data distribution affect layer-wise expert pruning for gpt-oss-20b. Our findings reveal that (1) domain-general pruning quality receives minimal benefit from scaling calibration data beyond small sample sizes, (2) cross-lingual calibration transfers effectively, with English-only calibration maintaining strong performance across seven languages of varying typologies and scripts, and (3) a diverse calibration data resembling the pretraining distribution can yield comparable or superior performance to test-distribution data. Ultimately, we provide practical insights as a recipe for better understanding calibration for more effective domain-general expert pruning.

Chinese Translation

后训练专家剪枝是一种有前景的混合专家（Mixture-of-Experts, MoE）模型部署技术。校准集是剪枝过程的一个重要组成部分，但其与评估性能的相关性在很大程度上被误解且不直观。在本研究中，我们对校准扰动的鲁棒性进行了实证研究，特别是在 MoE 模型中。我们展示了校准大小、语言和数据分布如何影响 gpt-oss-20b 的逐层专家剪枝。我们的发现揭示了以下几点：(1) 领域通用剪枝质量在校准数据规模超过小样本大小时受益有限；(2) 跨语言校准有效转移，仅使用英语的校准在七种不同类型和书写系统的语言中保持强劲的性能；(3) 类似于预训练分布的多样化校准数据可以产生与测试分布数据相当或更优的性能。最终，我们提供了实用的见解，作为更好理解校准以实现更有效的领域通用专家剪枝的参考。

View on arXiv Download PDF AI Translation

cs.CL / 66 / 2610.15661

Feedback as an Interface: Controlling Form and Detail in LLM Translation Refinement

反馈作为接口：控制大型语言模型翻译精细化中的形式与细节

Liu, Yiqi, Tomori, Flavjo, Moosavi, Nafise Sadat, Eger, Steffen, Lin, Chenghua

Abstract

Feedback presentation shapes how LLMs use evaluator judgments during translation refinement. We introduce a machine translation protocol that keeps the evaluator score, main issue, and recommendation fixed while varying feedback form and supporting detail. Across seven models and six translation directions, low-load feedback outperforms high-load feedback in all 42 translator-direction settings under xCOMET. On the 29,400-record rewriting subset, mean xCOMET scores multiplied by 100 are 83.30 for low-load feedback and 80.67 for high-load feedback; the full-matrix original-translation reference mean is 83.46. MetricX-24 supports the low-load advantage on a stratified subset. A no-feedback second pass yields a small improvement over the original translations, and direct evaluator prompting reproduces the low-load advantage without a separate rewriting stage. These results show that compact feedback limits the quality loss from evaluator-guided refinement in the evaluated high-resource MT settings.

Chinese Translation

反馈呈现方式影响大型语言模型（LLMs）在翻译精细化过程中如何利用评估者的判断。我们提出了一种机器翻译协议，该协议在保持评估者评分、主要问题和建议不变的情况下，变化反馈形式和支持细节。在七个模型和六个翻译方向中，低负载反馈在所有42个翻译者-方向设置下的表现优于高负载反馈，使用xCOMET进行评估。在29,400条记录的重写子集中，低负载反馈的平均xCOMET分数乘以100为83.30，而高负载反馈为80.67；完整矩阵的原始翻译参考均值为83.46。MetricX-24在分层子集上支持低负载的优势。没有反馈的第二次处理相较于原始翻译略有改善，而直接的评估者提示在没有单独重写阶段的情况下再现了低负载的优势。这些结果表明，紧凑的反馈限制了在评估的高资源机器翻译设置中，由评估者引导的精细化过程所导致的质量损失。

View on arXiv Download PDF AI Translation

cs.CL / 67 / 2610.15667

Rate-Resolved Text Evaluation with Masked Diffusion Language Models

基于掩码扩散语言模型的速率分辨文本评估

Liu, Yiqi, Zhao, Kun, Lin, Chenghua, Moosavi, Nafise Sadat

Abstract

A single evaluation score can rank generation systems but rarely reveals what failed. We propose rate-resolved text evaluation, which turns the masking rate of masked diffusion language models into a diagnostic axis. By reconstructing text under different fractions of masked tokens, one diffusion score becomes a vector of band-conditioned scores that reflect reconstruction difficulty at varying levels of visible context. On SummEval, preserving this rate structure yields complementary signal: the best masking-rate band improves fusion with autoregressive likelihood over a collapsed diffusion scalar, survives a sample-shuffle control that breaks rate ordering, and adds residual signal to a fixed prompted judge. A second diffusion model reproduces the effect more weakly. On WMT MQM zh-en, roundtrip mask-and-reconstruct probing produces a tunable localization curve in which ranking-oriented AUROC and span-detection F1 peak at different masking rates. Factuality benchmarks bound the interpretation: band semantics are model- and task-dependent, so each application requires empirical validation.

Chinese Translation

单一评估分数可以对生成系统进行排名，但很少揭示失败的原因。我们提出了速率分辨文本评估，将掩码扩散语言模型的掩码率转化为诊断轴。通过在不同的掩码令牌比例下重构文本，一个扩散分数变成了一个反映在不同可见上下文水平下重构难度的带条件分数向量。在SummEval上，保持这一速率结构产生了互补信号：最佳掩码率带在与自回归似然的融合中优于崩溃的扩散标量，能够在打乱样本控制中保持速率排序，并为固定提示评估者添加残余信号。第二个扩散模型的效果较弱。在WMT MQM zh-en上，往返掩码和重构探测产生了一个可调的定位曲线，其中面向排名的AUROC和跨度检测F1在不同的掩码率下达到峰值。事实性基准限制了解释：带语义依赖于模型和任务，因此每个应用都需要实证验证。

View on arXiv Download PDF AI Translation

cs.CL / 68 / 2610.15678

The Mathematical Collective Unconscious: Jungian Archetypes within the Latent Space

数学集体无意识：潜在空间中的荣格原型

Breen, Andrew

Abstract

Carl Jung theorised that the human psyche contains an inherited framework, the collective unconscious, populated by universal archetypes. As large language models (LLMs) are trained on vast corpora of textual data, this paper investigates whether they spontaneously construct distinct archetype-like regions within their high-dimensional latent spaces. The paper proposes that these regions function as spatial constraints that naturally organise semantically related narratives. To evaluate this, we applied cosine similarity and angular distance metrics across thirteen experiments utilising OpenAIs text-embedding-3-small model. The paper contains two experimental sections the first section explores these ideas using synthetically constructed narratives, the results of which act as a baselines for experiments conducted in the second section where the synthetic data is replaced by human-authored literature (Shmoop corpus), unedited dream reports (DreamBank corpus), and culturally distilled folklore (Andrew Langs Fairy Books). The results show that archetypal roles occupy distinct, measurable regions in the latent space. Crucially, these regions remain compact and consistent because of narrative structure rather than simple word overlap. Additionally, text category dictates spatial distribution: dream reports concentrate within Trickster and Shadow regions, while traditional folklore distributes evenly across all tested archetypes. We conclude that transformer latent spaces encode stable regions which appear to correspond to recognised Jungian archetypes.

Chinese Translation

卡尔·荣格理论认为，人类心理包含一个遗传框架，即集体无意识，其中充满了普遍的原型。随着大型语言模型（LLMs）在庞大的文本数据语料库上进行训练，本文探讨它们是否会自发地在其高维潜在空间中构建出不同的原型类似区域。本文提出，这些区域作为空间约束，自然地组织语义相关的叙事。为了评估这一点，我们在十三个实验中应用了余弦相似度和角度距离度量，利用了OpenAI的text-embedding-3-small模型。本文包含两个实验部分，第一部分使用合成构建的叙事探索这些思想，其结果作为第二部分实验的基线，在第二部分中，合成数据被人类创作的文学（Shmoop语料库）、未编辑的梦境报告（DreamBank语料库）和文化提炼的民间故事（Andrew Lang的《童话书》）所替代。结果表明，原型角色在潜在空间中占据了不同且可测量的区域。至关重要的是，这些区域因叙事结构而非简单的词重叠而保持紧凑和一致。此外，文本类别决定了空间分布：梦境报告集中在诡计者和阴影区域，而传统民间故事则均匀分布在所有测试的原型中。我们得出结论，变换器潜在空间编码了稳定的区域，这些区域似乎与已识别的荣格原型相对应。

View on arXiv Download PDF AI Translation

cs.CL / 69 / 2610.15687

Towards Autopilot Labeling for Short Video Content Governance

面向短视频内容治理的自动标注研究

Chen, Junwen, Wang, Hongwei, Wang, Zixuan, Wang, Jingbing, Liu, Gorden, Ong, Sheng Siong, Zang, Xiaojiao, Chao, JuChin, Jin, Ze, Men, Kaiwen, Li, Yin, Li, Jeremy, Hao, Zhuolin, Xiong, Hongyu, Song, Yang

Abstract

The explosive growth of short videos has presented significant challenges to content governance. Traditional manual labeling is inefficient, slow to iterate, and costly. Therefore, we explore and develop a multi-stage autopilot labeling framework aimed at progressively increasing the level of automation in short video content annotation. To move beyond the heavy reliance on human expert labeling at the initial L0 stage, we first introduce a discriminative MLLM labeler at the L1 stage, which is fine-tuned with supervised data to directly predict labels for short videos. Advancing to the L2 stage, we shift to a domain-adaptive MLLM pretrained on short video data and use it for zero-shot labeling, enabling flexible generalization without task-specific supervision. In the L3 stage, we propose an agentic annotation workflow, and equip the agent with several key tools, which empower the agent to handle complex, unfamiliar, or vaguely defined cases, significantly reducing the need for human intervention. We conduct extensive experiments across various content issues, and results demonstrate that each stage of the autopilot labeling framework progressively improves labeling accuracy, and has significantly reduced manual labor costs in real-world applications by up to 85%.

Chinese Translation

短视频的爆炸性增长给内容治理带来了重大挑战。传统的人工标注效率低下、迭代缓慢且成本高昂。因此，我们探索并开发了一种多阶段的自动标注框架，旨在逐步提高短视频内容注释的自动化水平。为了突破在初始L0阶段对人工专家标注的高度依赖，我们首先在L1阶段引入了一种判别式的多语言大模型（MLLM）标注器，该标注器通过监督数据进行微调，以直接预测短视频的标签。推进到L2阶段，我们转向一个在短视频数据上预训练的领域自适应MLLM，并利用其进行零样本标注，使其能够在没有特定任务监督的情况下灵活泛化。在L3阶段，我们提出了一种代理注释工作流程，并为代理配备了多个关键工具，使其能够处理复杂、不熟悉或模糊定义的案例，显著减少了对人工干预的需求。我们在各种内容问题上进行了广泛的实验，结果表明，自动标注框架的每个阶段都逐步提高了标注准确性，并在实际应用中将人工劳动成本降低了多达85%。

View on arXiv Download PDF AI Translation

cs.CL / 70 / 2610.15692

Do LLM Tutors Actually Teach? A Process-Centered Synthesis of the Tutoring Loop

大型语言模型辅导员真的能教学吗？以过程为中心的辅导循环综合研究

Adak, Sayantan, Hazra, Rima, Mukherjee, Koyel, Bose, Digbalay, Banerjee, Somnath

Abstract

Large language models can produce fluent tu- toring dialogue, but whether they actually help students learn remains unclear. We present a process-centered synthesis of LLM tutoring that follows the path from learner-state infer- ence to pedagogical action, tutor training, eval- uation, and student simulation. This view ex- poses a central gap: implementing a tutoring component does not establish that it works or that its outputs meaningfully affect later stages. A structured study-level audit shows that peda- gogical evaluation is common, while evidence becomes much thinner for learner-conditioned intervention, human learning gains, and espe- cially transfer or delayed retention. Human studies nevertheless provide a conditional posi- tive result: LLM tutors can improve learning, but benefits depend on pedagogical structure, learner engagement, and how learning is mea- sured. We argue that progress requires validat- ing the interfaces between tutoring stages, not only improving individual components.

Chinese Translation

大型语言模型能够生成流畅的辅导对话，但它们是否真正帮助学生学习仍然不清楚。我们提出了一种以过程为中心的LLM（大型语言模型）辅导综合研究，跟踪从学习者状态推断到教学行动、辅导员培训、评估和学生模拟的路径。这一视角揭示了一个核心缺口：实施辅导组件并不意味着它有效，或其输出对后续阶段有实质性影响。结构化的研究级审计显示，教学评估是普遍存在的，而针对学习者条件的干预、人类学习收益，尤其是迁移或延迟保持的证据则显得稀薄。然而，人类研究提供了一个有条件的积极结果：LLM辅导员可以改善学习，但其益处依赖于教学结构、学习者参与度以及学习测量方式。我们认为，进展需要验证辅导阶段之间的接口，而不仅仅是改善单个组件。

View on arXiv Download PDF AI Translation

cs.CL / 71 / 2610.15705

bharatSetu: a 1M x 22 bridge for indic translation

bharatSetu：一个用于印度语言翻译的1M x 22桥接数据集

Malik, Haq Nawaz, Nissar, Nahfid, Iqbal, Faizan

Abstract

India's Eighth Schedule recognises 22 languages, yet parallel data is uneven across them, and many language pairs, especially among lower-resource languages such as Bodo, Dogri, Santali and Kashmiri, lack direct bitext. We present BharatSetu 22-Indic 1M, a row-aligned table of 984,764 English sentences, each paired with machine translations into all 22 scheduled languages (21,664,808 Indic cells), generated with the sarvamai/sarvam-translate model family. Because every Indic cell in a row translates the same English sentence, any two columns form a translation pair: 462 directed Indic-to-Indic pairs, or 506 including English, without direct bitext. We call these pivot-derived pairs and propose a confidence layer that lets users choose how strictly each pair is connected. The corpus is machine-generated and not human-verified; quality is expected to vary by language, domain and sentence style, and pivot-derived pairs can inherit errors from both sides. We therefore release BharatSetu as a machine layer and propose a human-correction layer keyed to the same row identifiers, so that each corrected column upgrades every pair that uses it, together with an evaluation protocol. We report no translation-quality results.

Chinese Translation

印度的第八附表承认22种语言，但这些语言之间的平行数据分布不均，许多语言对，尤其是诸如博多语、道格里语、桑塔利语和克什米尔语等低资源语言，缺乏直接的双语文本。我们提出了BharatSetu 22-Indic 1M，这是一个行对齐的表格，包含984,764个英语句子，每个句子都与翻译成所有22种附表语言的机器翻译配对（21,664,808个印度语言单元），这些翻译是通过sarvamai/sarvam-translate模型系列生成的。由于每个行中的印度语言单元都翻译相同的英语句子，因此任何两个列形成一个翻译对：462个定向的印度语到印度语对，或包括英语的506对，而没有直接的双语文本。我们称这些为基于枢轴的对，并提出一个置信度层，让用户选择每对之间的连接强度。该语料库是机器生成的，未经人工验证；质量预计会因语言、领域和句子风格而异，基于枢轴的对可能会继承双方的错误。因此，我们将BharatSetu作为一个机器层发布，并提出一个与相同行标识符相关的人为修正层，以便每个修正的列能够升级使用它的每个对，并配备评估协议。我们报告没有翻译质量的结果。

View on arXiv Download PDF AI Translation

cs.CL / 72 / 2610.15712

TL;QR: Querying Temporal Discursive Patterns in Textual Corpora

TL;QR：在文本语料库中查询时间性话语模式

Hublet, François, Krsmanovic, Vladimir, Solano, Jesus, Bouhoula, Ahmed, da Costa, Ana Oliveira

Abstract

Many questions about large textual corpora concern patterns that unfold over time, such as a political group whose stance on a topic drifts from support to criticism over several years of debates. We present TL;QR (Too Long; Query with logical temporal Reasoning), a declarative query language and open-source engine for finding such temporal discursive patterns. In TL;QR, a single query combines natural-language predicates, whose truth is judged by a Large Language Model (LLM), with regular expressions and the operators of a temporal logic. The LLM answers only local questions about individual text passages; a symbolic engine then aggregates these judgments over time and links every match to the recorded passages and predicate decisions behind it. With this division of labor, the temporal and relational reasoning stays exact and auditable even over long time horizons. In addition, when a query includes a regular-expression pre-filter, only a small fraction of the corpus ever reaches the LLM. On a suite of controlled synthetic benchmarks with temporally and semantically complex patterns, TL;QR is substantially more accurate than LLM-only baselines. In a case study on ten years of European Parliament debates (442,593 multilingual speeches), TL;QR identifies candidate shifts in the political groups' stances on the GDPR privacy regulation.

Chinese Translation

关于大型文本语料库的许多问题涉及随时间展开的模式，例如一个政治团体在数年的辩论中对某一主题的立场从支持转向批评。我们提出了TL;QR（Too Long; Query with logical temporal Reasoning），这是一种声明式查询语言和开源引擎，用于寻找此类时间性话语模式。在TL;QR中，单个查询结合了自然语言谓词（其真值由大型语言模型（LLM）判断）、正则表达式以及时间逻辑的运算符。LLM仅回答关于单个文本段落的局部问题；然后，一个符号引擎在时间上聚合这些判断，并将每个匹配项链接到记录的段落及其背后的谓词决策。通过这种分工，时间和关系推理在长时间范围内保持精确和可审计。此外，当查询包含正则表达式预过滤器时，只有小部分语料库会到达LLM。在一组具有时间性和语义复杂模式的受控合成基准测试中，TL;QR的准确性显著高于仅使用LLM的基线。在对十年欧洲议会辩论（442,593篇多语言演讲）的案例研究中，TL;QR识别出政治团体在GDPR隐私法规上的立场变化候选项。

View on arXiv Download PDF AI Translation

cs.CL / 73 / 2610.15732

NepPlural: A Human-Anchored Benchmark for Persona Annotation in Nepali Migration Discourse

NepPlural：尼泊尔移民话语中个性标注的人类基准

Nyachhyon, Jinu, Khatiwada, Anish, Timilsina, Bimal

Abstract

Pluralistic alignment asks that language models faithfully represent a society's diverse, competing values rather than collapse them into one dominant viewpoint, yet flagship pluralistic-alignment datasets remain heavily skewed toward Western respondents. We introduce NepPlural, a human-anchored benchmark for modelling competing socio-cultural personas in Nepali public discourse on youth migration and brain drain. We curate 2,148 code-mixed (Devanagari, Romanized Nepali, English) YouTube comments, annotate them along a four-axis persona taxonomy (Intent, Primary Driver, Value Orientation, Affect) plus an explicit No-Persona class, and build a 300-comment gold set labelled independently by three native Nepali speakers (mean Fleiss' $\kappa \approx 0.55$). Using this gold set to audit automatic annotation, we find that LLM annotation reliability is driven by protocol. Labels collected in batches through chat interfaces reach only near-chance inter-model agreement ($\kappa \approx 0.17$). Annotating one comment at a time raises agreement above the native-speaker level ($\kappa \approx 0.72$), but consistency is not validity: under a forced-choice schema, models agree on personas for the 39% of comments native speakers judge to express none, keeping macro-F1 against human gold near 0.50. Allowing abstention resolves this: models recover No-Persona comments with F1 of 0.79 to 0.87 and, against leave-one-annotator-out references, match or exceed individual native speakers (macro-F1 0.75 to 0.86 vs. 0.75). Encoders fine-tuned on the batched labels reach only 0.34. Finally, a human-judged steerability probe finds that three frontier models adopt single persona axes with 59 to 77% fidelity but compose full personas only 33% of the time, with no congruence gap detected at this scale between Western-congruent and Nepali-collectivist personas.

Chinese Translation

多元对齐要求语言模型忠实地代表一个社会的多样化、竞争性的价值观，而不是将其压缩为一种主导观点，但旗舰的多元对齐数据集仍然严重倾向于西方受访者。我们介绍了NepPlural，这是一个以人类为基础的基准，用于建模尼泊尔公共话语中关于青年移民和人才流失的竞争性社会文化个性。我们整理了2,148条混合语言（天城文、罗马化尼泊尔语、英语）的YouTube评论，按照四个维度的个性分类（意图、主要驱动因素、价值取向、情感）进行标注，并增加了一个明确的无个性类别，构建了一个由三位尼泊尔母语者独立标注的300条评论的金标准集（平均Fleiss' $ ext{kappa} ext{ } ext{≈} ext{ } 0.55$）。使用该金标准集对自动标注进行审计，我们发现大型语言模型（LLM）的标注可靠性受协议驱动。通过聊天界面批量收集的标签仅达到接近偶然的模型间一致性（$ ext{kappa} ext{ } ext{≈} ext{ } 0.17$）。逐条标注评论使一致性超过母语者水平（$ ext{kappa} ext{ } ext{≈} ext{ } 0.72$），但一致性并不等于有效性：在强制选择的框架下，模型对39%被母语者判断为无表达的评论达成一致，使得与人类金标准的宏观F1接近0.50。允许弃权解决了这一问题：模型恢复无个性评论的F1为0.79至0.87，并且与逐个排除标注者的参考相比，匹配或超过个别母语者（宏观F1 0.75至0.86对比0.75）。在批量标签上微调的编码器仅达到0.34。最后，一个人类评判的可操控性探测发现，三个前沿模型在单一个性维度上具有59%到77%的忠诚度，但仅在33%的情况下组合完整个性，在这一规模上未检测到西方一致性和尼泊尔集体主义个性之间的差距。

View on arXiv Download PDF AI Translation

cs.CL / 74 / 2610.15757

Enhancing Zero-Shot Domain Transfer in Low-Resource Sentiment Analysis: A Case Study in Marathi

增强低资源情感分析中的零样本领域迁移：以马拉地语为例

Somani, Tanvi, Babare, Aryan, Bora, Samyak, Joshi, Raviraj

Abstract

Sentiment analysis is a key task in Natural Language Processing that determines the polarity of opinions in text. For low-resource languages like Marathi--where limited data prevents the direct use of standard statistical methods--models are often trained on single-domain datasets. While this yields strong in-domain performance, it leads to poor generalization to unseen domains. Recent studies highlight the need for domain-robust systems that can handle diverse real-world content through multi-domain modeling. This study investigates how multi-domain training improves sentiment classification for unseen Marathi domains. Models are trained using single-domain, two-domain, and three-domain dataset combinations. Results show that two-domain training improves accuracy on unseen domains by about 8\% compared to single-domain training. Adding a third domain provides further improvement, though the gain is smaller, averaging around 2\% for some domains. These results suggest that training with data from multiple domains helps build more robust sentiment analysis models for low-resource languages, though the benefits reduce as more domains are added.

Chinese Translation

情感分析是自然语言处理中的一项关键任务，旨在确定文本中意见的极性。对于像马拉地语这样的低资源语言，由于数据有限，无法直接使用标准统计方法，因此模型通常在单一领域数据集上进行训练。尽管这在领域内表现出色，但在未见领域的泛化能力较差。最近的研究强调了需要能够通过多领域建模处理多样化现实内容的领域稳健系统。本研究探讨了多领域训练如何改善未见马拉地语领域的情感分类。模型使用单领域、双领域和三领域数据集组合进行训练。结果表明，与单领域训练相比，双领域训练在未见领域的准确率提高了约8%。添加第三个领域进一步改善了结果，尽管增益较小，对于某些领域的平均提升约为2%。这些结果表明，使用来自多个领域的数据进行训练有助于构建更稳健的低资源语言情感分析模型，尽管随着领域数量的增加，收益会减少。

View on arXiv Download PDF AI Translation

cs.CL / 75 / 2610.15761

Robust Marathi Sentiment Analysis using Domain-Adaptive Pre-training and Data Augmentation Approaches

基于领域自适应预训练和数据增强方法的鲁棒马拉地语情感分析

Somani, Tanvi, Babare, Aryan, Bora, Samyak, Joshi, Raviraj, Dhamankar, Abhinay

Abstract

For languages with limited resources, Sentiment Analysis remains a challenging problem due to limited annotated data and strong domain variability across textual sources. Marathi, despite being widely spoken, lacks extensive multi-domain sentiment analysis studies. Recent efforts by L3Cube have introduced multi-domain datasets in Marathi, enabling a better study of domain-aware sentiment classification. This study uses the pretrained MahaBERT model to improve sentiment classification for Marathi across different domains. Models are trained on individual domain datasets and domain-adaptive pretraining is applied using masked language modeling to help the model better understand domain-specific language. In addition, back-translation is used as a method of data augmentation to increase the diversity of training data. Results show that these methods lead to consistent improvements in accuracy across all domains. The improvements range from a minimum of about 0.12% to a maximum of about 1.74%, depending on the dataset. These results demonstrate that domain-adaptive pretraining and data augmentation are effective strategies for improving sentiment analysis models. This helps achieve more robust performance across different domains and makes the approach more suitable for low-resource languages like Marathi.

Chinese Translation

对于资源有限的语言，情感分析仍然是一个具有挑战性的问题，因为标注数据有限且文本来源之间存在较强的领域变异性。尽管马拉地语广泛使用，但缺乏广泛的多领域情感分析研究。L3Cube最近的努力引入了马拉地语的多领域数据集，使得领域感知情感分类的研究得以更好地进行。本研究使用预训练的MahaBERT模型来改善马拉地语在不同领域的情感分类。模型在各个领域的数据集上进行训练，并采用掩码语言建模进行领域自适应预训练，以帮助模型更好地理解领域特定语言。此外，反向翻译被用作数据增强的方法，以增加训练数据的多样性。结果表明，这些方法在所有领域的准确性上都带来了持续的改善。根据数据集的不同，改善幅度从约0.12%到约1.74%不等。这些结果表明，领域自适应预训练和数据增强是改善情感分析模型的有效策略。这有助于在不同领域实现更强的性能，并使该方法更适合像马拉地语这样的低资源语言。

View on arXiv Download PDF AI Translation

cs.CL / 76 / 2610.15772

RSIPO: Recursive Self-Improving Policy Optimization for Agentic Reinforcement Learning via Skill-Policy Co-Evolution

RSIPO：通过技能-策略共同进化的递归自我改进策略优化用于自主强化学习

Zhang, Fan, Wu, Hao, Wu, Jinyang, Qu, Hongyu, Lian, Zheng, Luo, Xiao, Zheng, Yefeng, Zhang, Bo, Li, Haoxuan, Heng, Pheng-Ann

Abstract

Recent focus in agentic reinforcement learning (RL) seeks to improve credit assignment in long-horizon tasks. One promising direction combines reinforcement learning with verifiable rewards (RLVR) and on-policy self-distillation (OPSD), integrating trajectory-level advantages with dense token-level supervision. This line of work often conditions a teacher model on privileged context, such as agentic skills, to score tokens generated by the student policy. Despite their progress, they largely overlook the on-policy evolution of agentic skills and the reliability of teacher supervision. Towards this end, we introduce Recursive Self-Improving Policy Optimization (RSIPO), a self-evolving agentic RL framework that enables agentic skills and the policy model to co-evolve throughout the training process. Specifically, RSIPO enables the policy itself to extract skills from completed trajectories and explicitly optimizes its skill-generation capability through trajectory replay. It further employs empirical Bayes shrinkage to dynamically calibrate the student's trust in teacher scores, accounting for the fact that skill-enhanced teacher guidance can sometimes be less reliable. Experiments with Qwen-family models on embodied interaction and search-based agentic tasks demonstrate that RSIPO outperforms all the compared baseline approaches. With Qwen2.5-7B as the base model, RSIPO achieves an average success rate of 98.6\% on ALFWorld, yielding absolute improvements of 2.5\% and 17.4\% over the second-best baseline and standard GRPO, respectively.

Chinese Translation

近期在自主强化学习（RL）中的研究重点是改善长时间任务中的信用分配。一种有前景的方向是将强化学习与可验证奖励（RLVR）和在线自蒸馏（OPSD）相结合，整合轨迹级优势与密集的标记级监督。这一研究方向通常将教师模型的条件设定为特权上下文，例如自主技能，以对学生策略生成的标记进行评分。尽管取得了一定进展，但它们在很大程度上忽视了自主技能的在线演化和教师监督的可靠性。为此，我们提出了递归自我改进策略优化（RSIPO），这是一种自我演化的自主强化学习框架，使自主技能和策略模型能够在整个训练过程中共同进化。具体而言，RSIPO使策略本身能够从完成的轨迹中提取技能，并通过轨迹重放显式优化其技能生成能力。它进一步采用经验贝叶斯收缩动态校准学生对教师评分的信任，考虑到技能增强的教师指导有时可能不那么可靠。在具身交互和基于搜索的自主任务上使用Qwen系列模型的实验表明，RSIPO在所有比较的基线方法中表现优越。以Qwen2.5-7B作为基础模型，RSIPO在ALFWorld上的平均成功率达到98.6\%，相较于第二好的基线和标准GRPO分别实现了2.5\\%和17.4\\%的绝对提升。

View on arXiv Download PDF AI Translation

cs.CL / 77 / 2610.15795

An Explainable System for Offensive Content Detection in Nastaliq Urdu: Corpus Curation, Transformer Benchmarking, and Word-Level Attribution

一个可解释的系统用于检测Nastaliq乌尔都语中的攻击性内容：语料库整理、变换器基准测试和词级归因

Malik, Muhammad Shahid Iqbal

Abstract

Automated moderation systems are deployed at scale, yet they remain opaque to the users and moderators affected by their decisions, and their coverage is almost entirely confined to English. For Urdu written in the Nastaliq script, no such system exists, and the annotated resources it would be built on are drawn largely from Twitter/X, where character limits constrain how users write. Interpretability has received little attention, leaving it unclear what such models actually learn. This study addresses these gaps in three steps. First, we curate and publicly release a corpus of 7604 Nastaliq Urdu Facebook comments, extending an earlier corpus by correcting text-level inconsistencies and re-examining the original labels. Second, we fine-tune four Urdu-capable transformers under a common grid-search protocol and compare them against twelve feature-classifier baselines on identical data partitions. The best configuration, Multilingual Representations for Indian Languages (MuRIL), reaches 87.05% accuracy and 87.06% macro F1, improving on the strongest baseline by 5.52 percentage points, a margin that exceeds the 95% confidence interval on test accuracy; the two leading transformers differ by less than the test set can resolve, and performance is stable across five optimizers. Third, we compute word-level SHAP attributions adapted to right-to-left rendering. Attribution mass concentrates on high-frequency slurs but extends to syntactically neutral function words and to political vocabulary characteristic of the source pages, indicating that aggregate performance may rest partly on lexical and topical regularities rather than a generalizable representation of offensiveness.

Chinese Translation

自动化审核系统在大规模应用中被部署，但对于受其决策影响的用户和审核员而言，这些系统依然不透明，并且其覆盖范围几乎完全局限于英语。对于使用Nastaliq书写的乌尔都语，目前尚不存在这样的系统，而其构建所需的注释资源主要来自Twitter/X，字符限制限制了用户的写作方式。可解释性问题鲜有关注，使得这些模型实际学习了什么仍不明确。本研究通过三个步骤来解决这些问题。首先，我们整理并公开发布了一个包含7604条Nastaliq乌尔都语Facebook评论的语料库，纠正了文本级不一致性并重新审视了原始标签，从而扩展了先前的语料库。其次，我们在一个共同的网格搜索协议下对四个支持乌尔都语的变换器进行了微调，并在相同的数据分区上将它们与十二个特征分类基准进行了比较。最佳配置，即印度语言的多语言表示（Multilingual Representations for Indian Languages, MuRIL），达到了87.05%的准确率和87.06%的宏F1值，比最强基准提高了5.52个百分点，这一差距超过了测试准确率的95%置信区间；两个领先的变换器之间的差异小于测试集所能分辨的范围，且在五个优化器下性能稳定。第三，我们计算了适应从右到左渲染的词级SHAP归因。归因质量集中在高频侮辱性词汇上，但也延伸到语法中性功能词和源页面特征的政治词汇，表明整体性能可能部分依赖于词汇和主题的规律性，而非攻击性的可推广表示。

View on arXiv Download PDF AI Translation

cs.CL / 78 / 2610.15836

Latent Taxonomies: How Property Induction Develops in Language Model Pre-training

潜在分类法：属性归纳在语言模型预训练中的发展

Sobotka, Jan, Goliakova, Ekaterina, Detyniecki, Marcin, Bosselut, Antoine, Renard, Xavier

Abstract

When told that robins have sesamoid bones, people are far more prone to extend the property to sparrows than to whales. This behavior, called category-based property induction, is well documented in humans, but whether and how it develops in language models trained only to predict the next token is unclear. To shed light on this, we construct a hierarchical taxonomy of synthetic entities grouped by the properties they share. We then inject facts from this latent taxonomy into the continued pre-training of Pythia and OLMo 2 models from 70M to 7B parameters, and track the development of inductive behaviors. We find that taxonomically aligned induction emerges from next-token prediction alone: a property learned for one entity transfers more strongly within its taxonomic category than across categories. Internally, entities that share a deeper category become more similar in activation space, and clustering recovers the full taxonomic tree from entity representations. Contrary to the usual expectation that scale helps, we find that induction weakens with model size when the data budget is fixed, suggesting that larger models memorize individual facts rather than compress them into shared structure. Induction also strengthens with more pre-training before the taxonomy injection, transfers to novel in-context properties, and can be overridden by a latent taxonomy implied by the prompt. Together, these results show that a model's inductive generalization reflects the latent organization of its training data, with implications for data curation and for predicting how far a learned fact will spread.

Chinese Translation

当被告知知更鸟具有种子骨时，人们更倾向于将这一属性扩展到麻雀而非鲸鱼。这种被称为基于类别的属性归纳的行为在人的研究中已有充分记录，但在仅被训练用于预测下一个标记的语言模型中，这种行为是否以及如何发展尚不清楚。为此，我们构建了一个基于共享属性对合成实体进行分组的层次分类法。然后，我们将这一潜在分类法中的事实注入到Pythia和OLMo 2模型的继续预训练中，模型参数从7000万增加到70亿，并追踪归纳行为的发展。我们发现，仅通过下一个标记预测就会出现分类对齐的归纳：对于一个实体学习到的属性在其分类类别内的转移强度大于跨类别的转移。在内部，共享更深层类别的实体在激活空间中变得更加相似，聚类能够从实体表示中恢复完整的分类树。与通常认为规模有助于的预期相反，我们发现当数据预算固定时，归纳在模型规模增大时反而减弱，这表明较大的模型倾向于记忆个别事实，而不是将其压缩为共享结构。归纳在分类法注入前经过更多预训练时也会增强，能够转移到新的上下文属性，并且可以被提示中隐含的潜在分类法所覆盖。这些结果表明，模型的归纳泛化反映了其训练数据的潜在组织，对数据策划和预测学习到的事实传播范围具有重要意义。

View on arXiv Download PDF AI Translation

cs.CL / 79 / 2610.15843

Binding Phoneme Identity to Position in Neural Models and the Human Brain

在神经模型和人脑中将音素身份与位置绑定

Momeni, Nafiseh, Renaudin, Malo, Fried, Itzhak, Lakretz, Yair

Abstract

A fundamental challenge in cognitive science and neural computation is the binding problem. Compositional generalization can be achieved by factorizing representations into roles, the structural slots of a representation (e.g., serial position or syntactic function), and fillers, the contents occupying them (e.g., a phoneme or a word), which can be recombined freely to represent novel combinations. How neural systems implement role--filler binding, however, remains unknown. One of its simplest cases is phoneme sequencing, where phoneme identity (filler) must be bound to position (role) so that /gum/ and /mug/ are represented differently. A proposal from NLP, the `onion hypothesis' (Csordas et al., 2024), holds that the direction of the neural population state encodes token identity, while its magnitude encodes position. It was formulated for random token sequences; we extend it to natural language and test it in neural models and the human brain. We first evaluated it in an encoder-decoder trained to repeat English words, using representational geometry analyses and causal interventions. Magnitude alone proves incomplete: position is encoded by magnitude together with a systematic rotation of the state direction, a scheme we formalize as the `spiral hypothesis'. We then analyzed intracranial recordings from the auditory cortex of 20 participants listening to short sentences. Across superior temporal gyrus electrodes, the change in population response between consecutive phonemes decreases in magnitude as position advances, and its direction is aligned within consonants and within vowels but near orthogonal across the two classes, supporting the magnitude coding shared by both accounts. These results provide a geometric account of how neural systems bind phoneme identity to position, and show how neural-model hypotheses can yield testable predictions about human neural computation.

Chinese Translation

认知科学和神经计算中的一个基本挑战是绑定问题。通过将表征分解为角色（表征的结构位置，例如序列位置或句法功能）和填充物（占据这些位置的内容，例如音素或单词），可以实现组合泛化，这些填充物可以自由重组以表示新组合。然而，神经系统如何实现角色-填充物绑定仍然未知。其中一个最简单的例子是音素排序，在这种情况下，音素身份（填充物）必须与位置（角色）绑定，以便 /gum/ 和 /mug/ 被不同地表示。来自自然语言处理（NLP）的一个提议，即“洋葱假说”（onion hypothesis，Csordas et al., 2024），认为神经群体状态的方向编码了标记身份，而其幅度编码了位置。该假说是为随机标记序列提出的；我们将其扩展到自然语言，并在神经模型和人脑中进行了测试。我们首先在一个被训练以重复英语单词的编码-解码模型中评估了该假说，使用表征几何分析和因果干预。仅靠幅度证明是不完整的：位置是通过幅度与状态方向的系统旋转共同编码的，我们将这一方案形式化为“螺旋假说”（spiral hypothesis）。随后，我们分析了20名参与者在听短句时的听觉皮层内记录。跨越上颞回电极，连续音素之间的群体反应变化幅度随着位置的推进而减小，其方向在辅音和元音内部对齐，但在这两类之间近乎正交，支持了两种理论共享的幅度编码。这些结果提供了一个几何视角，说明神经系统如何将音素身份与位置绑定，并展示了神经模型假说如何产生可测试的人类神经计算预测。

View on arXiv Download PDF AI Translation

cs.CL / 80 / 2610.15870

Unchecked or Falsely Accused? Evaluating AI-Generated Text Detectors on Kazakh

未被检查或被错误指控？评估针对哈萨克语的人工智能生成文本检测器

Ozaman, Mansur

Abstract

AI-text detectors are already applied to Kazakh writing, yet no study reports how often publicly available detectors flag Kazakh text whose human authorship is dated before ChatGPT. We present AINA-KK: 3,811 human passages from encyclopedic, news, web and educational pages that provably existed before ChatGPT, and 5,597 machine and gray-zone passages from 9 generators, including the Kazakh-built KazLLM and Sherkala, with script and length stress tests. Five English-trained detectors from the Hugging Face Hub flag up to 81.4% of human passages and 79.6% of published student essays; recalibrated, they catch at most 1.3% of machine text. Calibrated zero-shot detectors catch at most 50.3% of machine text from in-distribution generators. Fine-tuned encoders rank text almost perfectly, but mDeBERTa at its default threshold flags 24.4% of 50-word human excerpts, partly because machine outputs are shorter than human text; calibrated, it catches only 47.1% of KazLLM and 62.7% of Sherkala passages. Without script handling, machine-transliterated Latin-2021 text defeats every detector. Fixes that need no new data (calibration, script canonicalization, script-augmented and length-matched training, and an OR with a zero-shot scorer) form AINA-KK-Det. On test it flags 0.9% of human passages (95% CI 0.4-2.7; 1.5% on development data) and catches 85.2% of machine text from known generators, 92.9% from two unseen Gemini-family Gemma-4 models and 61.3% from the Kazakh-built ones; its configuration was chosen with test results in view. We release the data, code and detector at https://github.com/ozxmn/aina-kk.

Chinese Translation

人工智能文本检测器已应用于哈萨克语写作，但尚无研究报告公开可用的检测器在多大程度上标记在 ChatGPT 之前创作的哈萨克文本。我们提出 AINA-KK：包含 3,811 段来自百科全书、新闻、网页和教育页面的人工文本，这些文本在 ChatGPT 之前就已存在，以及来自 9 个生成器的 5,597 段机器和灰色区域文本，包括哈萨克开发的 KazLLM 和 Sherkala，进行了脚本和长度压力测试。来自 Hugging Face Hub 的五个英文训练检测器标记了多达 81.4% 的人工文本和 79.6% 的已发布学生论文；经过重新校准，它们最多只能捕捉到 1.3% 的机器文本。经过校准的零-shot 检测器最多捕捉到 50.3% 来自分布内生成器的机器文本。微调的编码器几乎完美地对文本进行排名，但 mDeBERTa 在其默认阈值下标记了 24.4% 的 50 字人工摘录，部分原因是机器输出的长度短于人工文本；经过校准，它仅捕捉到 47.1% 的 KazLLM 和 62.7% 的 Sherkala 段落。在没有脚本处理的情况下，机器音译的 Latin-2021 文本击败了所有检测器。无需新数据的修复（校准、脚本规范化、脚本增强和长度匹配训练，以及与零-shot 评分器的 OR）形成了 AINA-KK-Det。在测试中，它标记了 0.9% 的人工文本（95% 置信区间 0.4-2.7；在开发数据中为 1.5%），并捕捉到 85.2% 来自已知生成器的机器文本，92.9% 来自两个未见的 Gemini 家族 Gemma-4 模型和 61.3% 来自哈萨克开发的生成器；其配置是根据测试结果选择的。我们在 https://github.com/ozxmn/aina-kk 发布数据、代码和检测器。

View on arXiv Download PDF AI Translation

cs.CL / 81 / 2610.15897

Review Rarely Beats the Stronger Model: Separating Verification from Capability Transfer in LLM Pipelines

评审通常不如更强模型：在大型语言模型管道中分离验证与能力转移

Zhang, Yirui, Shrestha, Yash Raj

Abstract

Qwen3-8B reviewing Llama-3.1-8B in math adds 33.2 points, yet Qwen3-8B alone is 6.4 points more accurate than the result. Reviewing another model's answer is usually judged by its gain over the generator. We show that this confuses verification, which adds accuracy beyond either model alone, with capability transfer, where a stronger reviewer supplies answers it could have given on its own. Measured against the stronger model alone, across 48 pairs of 12 open-weight models (3B to 72B) in math, knowledge, code and reasoning, review beats it in only 9 of 96 review directions, and does worse the larger the capability gap. The reason is that the reviewer mostly follows the generator, discarding its own correct answers. Review costs 2.6 to 2.8 times as much as the stronger model alone, while a selector using the same two calls is 2.7 to 4.2 points more accurate than review and cheaper. Studies of verification should report the stronger model alone, not only the gain over the generator. Code and per-question results: https://github.com/Kamilaabb/verification-or-capability-transfer

Chinese Translation

Qwen3-8B 在数学上对 Llama-3.1-8B 的评审增加了 33.2 分，但 Qwen3-8B 单独的准确率比结果高出 6.4 分。评审另一个模型的答案通常是通过其相对于生成器的增益来判断的。我们表明，这混淆了验证（增加了超出任一模型单独的准确性）与能力转移（更强的评审者提供了它本可以自己给出的答案）。在数学、知识、代码和推理方面，对 48 对 12 个开放权重模型（从 3B 到 72B）进行测量时，评审仅在 96 个评审方向中的 9 个中胜过更强模型，并且能力差距越大，表现越差。原因在于评审者大多遵循生成器，舍弃了自己正确的答案。评审的成本是更强模型单独的 2.6 到 2.8 倍，而使用相同两次调用的选择器比评审准确率高出 2.7 到 4.2 分且成本更低。验证研究应报告更强模型的单独结果，而不仅仅是相对于生成器的增益。代码及每个问题的结果见：https://github.com/Kamilaabb/verification-or-capability-transfer

View on arXiv Download PDF AI Translation

cs.CL / 82 / 2610.15910

RIVO: Lightweight Adaptive Rerouting for Deep Research Agents

RIVO：轻量级自适应重定向用于深度研究代理

Li, Mingchen, Huang, Jiatan, Yu, Hong

Abstract

Deep Research agents rely on long-horizon reasoning and iterative tool use to solve complex information-seeking tasks. However, adaptive decision-making over these trajectories remains challenging, as agents may stop prematurely when further evidence is needed, continue unnecessarily after sufficient evidence has been gathered, or fail to redirect unproductive trajectories. Existing solutions either require costly model retraining and dense turn-level supervision, or rely on heuristic control rules that are not directly optimized for downstream research utility. This motivates a lightweight, utility-aware alternative for Deep Research control. We formulate adaptive rerouting in long-horizon Deep Research as a state-conditioned trajectory correction problem. Based on this view, we introduce RIVO, a lightweight plug-in framework for adaptive trajectory correction. The agent estimates its research state from lightweight signals and determines whether trajectory redirection is needed. Once triggered, the plug-in modifies the Host token logits to steer generation toward a more promising research direction. Theoretical analysis shows that, under a value-alignment condition and bounded local curvature, a sufficiently small logit intervention can locally improve expected downstream research utility. We further construct BrowseCore, a curated dataset with a paired local corpus, for efficient and controlled Deep Research experimentation. Building on this resource, the plug-in is trained from downstream outcomes, enabling it to learn corrections that improve subsequent research and final task performance. Extensive experiments across multiple Deep Research benchmarks and control settings demonstrate the effectiveness of our method, showing that lightweight logit intervention can improve long-horizon research without retraining the underlying Host model.

Chinese Translation

深度研究代理依赖于长时间跨度的推理和迭代工具使用来解决复杂的信息搜索任务。然而，在这些轨迹上进行自适应决策仍然具有挑战性，因为代理可能在需要进一步证据时过早停止，或者在已收集到足够证据后不必要地继续，或者未能重新定向无效的轨迹。现有解决方案要么需要昂贵的模型再训练和密集的回合级监督，要么依赖于未针对下游研究效用直接优化的启发式控制规则。这促使我们提出一种轻量级、关注效用的深度研究控制替代方案。我们将长时间跨度深度研究中的自适应重定向形式化为一个状态条件的轨迹修正问题。基于这一观点，我们引入了RIVO，一个用于自适应轨迹修正的轻量级插件框架。代理从轻量级信号中估计其研究状态，并确定是否需要轨迹重定向。一旦触发，该插件修改主机（Host）令牌的logits，以引导生成朝向更有前景的研究方向。理论分析表明，在价值对齐条件和有界局部曲率下，足够小的logit干预可以在局部改善预期的下游研究效用。我们进一步构建了BrowseCore，一个配对本地语料库的策划数据集，以便于高效和受控的深度研究实验。在此资源的基础上，该插件通过下游结果进行训练，使其能够学习修正，从而提高后续研究和最终任务的表现。在多个深度研究基准和控制设置下进行的大量实验表明我们方法的有效性，显示轻量级logit干预可以在不重新训练基础主机模型的情况下改善长时间跨度的研究。

View on arXiv Download PDF AI Translation

cs.CL / 83 / 2610.15912

More Reasoning Is Not Always Better: Executing Clinical Guidelines with LLMs and Rules

更多推理并不总是更好：使用大型语言模型和规则执行临床指南

Lu, Xiaoqian, Guo, Guangfu, Cheng, Long, Liu, Kunpeng

Abstract

Large language models (LLMs) have shown considerable promise in clinical decision support. However, existing approaches often rely on multi-step reasoning or multi-agent collaboration to execute clinical guidelines, introducing substantial computational overhead without reliably preventing decision errors. To address these limitations, we propose a framework for Medical Decision Execution (MedDEx) that integrates structured patient states with deterministic rule execution to support guideline-based treatment planning. MedDEx separates clinical record interpretation from guideline execution, leveraging LLMs to extract patient information and generate treatment plans while assigning guideline-defined decision logic to a rule executor. Experiments across multiple clinical guideline benchmarks demonstrate that MedDEx improves decision accuracy while reducing inference costs. Further analyses highlight the critical roles of structured patient representations and deterministic rule execution, suggesting that additional LLM reasoning does not necessarily translate into better guideline adherence.

Chinese Translation

大型语言模型（LLMs）在临床决策支持方面展现了相当大的潜力。然而，现有的方法通常依赖于多步骤推理或多智能体协作来执行临床指南，这引入了显著的计算开销，却未能可靠地防止决策错误。为了解决这些局限性，我们提出了一种医疗决策执行框架（MedDEx），该框架将结构化患者状态与确定性规则执行相结合，以支持基于指南的治疗规划。MedDEx 将临床记录解释与指南执行分开，利用 LLMs 提取患者信息并生成治疗计划，同时将指南定义的决策逻辑分配给规则执行器。针对多个临床指南基准的实验表明，MedDEx 提高了决策准确性，同时降低了推理成本。进一步的分析突出了结构化患者表示和确定性规则执行的重要作用，表明额外的 LLM 推理并不一定转化为更好的指南遵循。

View on arXiv Download PDF AI Translation

cs.CL / 84 / 2610.15941

Natural Language Processing for Aligning Readings with Learning Objectives and Assessing Topic Balance in Dental Education

自然语言处理在牙科教育中对齐阅读材料与学习目标及评估主题平衡的应用

Yasmin, Farzana, Firoozi, Tahereh, Adabdokht, Rojin

Abstract

Curriculum teams often lack time to check whether assigned readings support learning objectives (LOs) and whether topic coverage is balanced. We present a natural language processing (NLP) workflow to support evidence-informed curation of reading lists. The workflow combines PubMedBERT sentence embeddings and cosine similarity for retrieval, natural language inference (NLI) for evidence checking, and topic modeling for coverage analysis. We tested the approach in dental education using 13 course-level LOs and a PubMed-derived corpus of 3,837 LO-abstract pairs. Median cosine similarity ranged from 0.86 to 0.91 across the 13 LOs, showing generally high semantic similarity. Topic modeling showed broad but uneven coverage, with one or two dominant topics per domain. Similarity alone was not enough for selecting readings: among high-similarity pairs (cosine 0.90-1.00), 22.9% were classified by NLI as neutral or contradictory. In a 49-pair expert-labeled validation set, cosine-based retrieval (cosine >= 0.90) achieved 0.69 accuracy, while the combined cosine and NLI approach achieved 0.73 accuracy against expert alignment labels. Overall, the workflow shows how AI can support curriculum alignment by reducing false positives and helping select readings that better match intended LOs.

Chinese Translation

课程团队常常缺乏时间来检查指定的阅读材料是否支持学习目标（LOs）以及主题覆盖是否平衡。我们提出了一种自然语言处理（NLP）工作流程，以支持基于证据的阅读清单整理。该工作流程结合了PubMedBERT句子嵌入和余弦相似度用于检索，自然语言推理（NLI）用于证据检查，以及主题建模用于覆盖分析。我们在牙科教育中测试了该方法，使用了13个课程级学习目标和一个由PubMed衍生的3,837个LO-摘要对的语料库。13个LO的中位余弦相似度范围为0.86到0.91，显示出一般较高的语义相似性。主题建模显示出广泛但不均匀的覆盖，每个领域有一到两个主导主题。仅凭相似性不足以选择阅读材料：在高相似性对（余弦0.90-1.00）中，有22.9%被NLI分类为中立或矛盾。在一个49对专家标注的验证集中，基于余弦的检索（余弦>=0.90）达到了0.69的准确率，而结合余弦和NLI的方法在专家对齐标签下达到了0.73的准确率。总体而言，该工作流程展示了人工智能如何通过减少假阳性并帮助选择更符合预期学习目标的阅读材料来支持课程对齐。

View on arXiv Download PDF AI Translation

cs.CL / 85 / 2610.15961

Unfair and Unsafe: Everyday Bias Triggers Emergent Misalignment

不公平与不安全：日常偏见触发的紧急失调

Guiu, Alex Martí, Zumarraga, Nicolas, O'Sullivan, Kevin, Jakob, Robert, Hedström, Anna

Abstract

Emergent misalignment (EM) occurs when fine-tuning a language model to a narrow task makes it misaligned on unrelated ones. Because organizations now routinely fine-tune models on their own data, it matters which kinds of training data cause EM. So far, studies have induced EM with data containing recognized safety failures, such as insecure code or harmful advice, while biased training data is typically examined by fairness research for the bias models learn from it rather than for its effects on safety. In real-world text, much of this bias, such as demographic stereotypes, appears not as derogatory language but politely, embedded in judgments about people. We ask whether learning such polite stereotypes can itself cause EM. We find that fine-tuning on polite answers that rely on stereotypes makes models less safe and less truthful on unrelated tasks. The effect holds across four open-weight model families (7B to 12B), both gender and race stereotypes, and repeated training runs, relative to controls trained on the same questions without the stereotype. We further show that the learned stereotype has its own direction in model activations, distinct from directions for generic misaligned traits such as deception. This direction exists before fine-tuning and can both control the learned stereotype and monitor its acquisition. Existing defenses against EM improve safety while leaving the stereotype largely intact. Beyond this controlled setting, fine-tuning on political news coverage also triggers EM, apparently amplified by partisan framing, and stereotype models acting as tool-using agents can carry out harmful tasks and blame people without evidence. Our results show that biased training data is not only a fairness problem but also a safety one. We argue that EM is studied too narrowly, and that fairness concerns and methods offer a natural way to broaden the search for its triggers.

Chinese Translation

紧急失调（Emergent Misalignment, EM）发生在对语言模型进行细调以适应狭窄任务时，使其在不相关任务上出现失调。由于组织现在常常在自己的数据上对模型进行细调，因此导致 EM 的训练数据类型变得尤为重要。迄今为止，研究通过包含已识别安全失败的数据（如不安全代码或有害建议）引发 EM，而偏见训练数据通常被公平性研究所考察，关注模型从中学习到的偏见，而非其对安全性的影响。在现实文本中，这种偏见（如人口统计学刻板印象）往往不是以贬损性语言出现，而是以礼貌的方式嵌入对人们的判断中。我们探讨学习这种礼貌刻板印象是否会导致 EM。我们的研究发现，基于刻板印象的礼貌回答的细调使模型在不相关任务上变得不那么安全和不那么真实。该效应在四个开放权重模型系列（7B 到 12B）、性别和种族刻板印象以及重复训练过程中均保持一致，相较于在相同问题上未使用刻板印象的控制组。我们进一步表明，学习到的刻板印象在模型激活中具有其自身的方向，与欺骗等通用失调特征的方向不同。这一方向在细调之前就已存在，并且可以控制学习到的刻板印象并监测其获取。在现有的 EM 防御措施中，虽然提高了安全性，但刻板印象基本保持不变。超出这一受控环境，针对政治新闻报道的细调也会触发 EM，显然受到党派框架的放大，而作为工具使用代理的刻板印象模型可以执行有害任务并在没有证据的情况下指责他人。我们的结果表明，偏见训练数据不仅是公平性问题，也是安全性问题。我们认为 EM 的研究过于狭隘，而公平性问题和方法提供了一种自然的方式来拓宽对其触发因素的搜索。

View on arXiv Download PDF AI Translation

cs.CL / 86 / 2610.15977

Mixed Feelings, Single Polarity

复杂情感，单一极性

Öhman, Emily, Mohammad, Saif

Abstract

Emotion datasets are routinely mapped onto positive and negative valence, by assigning each emotion category one polarity. We test where this works and where it fails, using SemEval-2018 Affect in Tweets (English, Arabic, Spanish), Affective Text aligned with EmoBank, raw GoEmotions annotations, naturalistic hashtag corpora, and a controlled annotation study. Fixed projection works well when a text carries labels of one valence. It fails on a subset identifiable from the labels alone: tweets whose emotion labels point in opposite directions. These are 9.0% of polar SemEval test tweets but cause 54.8% of the polarity errors of a learned projection, are more than twice as likely to receive a neutral-or-mixed gold label, and are not rescued by modeling the full label set. Annotators produce such configurations routinely: 16.3% of multilabel GoEmotions judgments combine emotions from the dataset's own positive and negative parents. In a controlled study holding word and reader constant, context changes the coarse valence judgment in 56.8% of cases, against 19.9% for controls. We recommend projecting label configurations rather than labels, abstaining or keeping a mixed class where labels conflict, and reporting coverage alongside accuracy.

Chinese Translation

情感数据集通常被映射到正向和负向极性，通过为每个情感类别分配一个极性。我们测试了这种方法在何处有效以及何处失效，使用了SemEval-2018推特情感（英语、阿拉伯语、西班牙语）、与EmoBank对齐的情感文本、原始GoEmotions注释、自然主义标签语料库以及一项受控注释研究。当文本仅携带一种极性的标签时，固定投影效果良好。然而，在仅通过标签即可识别的子集上，它失效：情感标签指向相反方向的推文。这些推文占极性SemEval测试推文的9.0%，但导致学习投影的54.8%的极性错误，且更有可能获得中性或混合的金标准标签，其概率是其他推文的两倍以上，并且通过建模完整标签集无法挽救。注释者经常产生这样的配置：16.3%的多标签GoEmotions判断结合了数据集中正向和负向父类的情感。在一项控制研究中，保持词汇和读者不变，背景在56.8%的案例中改变了粗略的极性判断，而对照组为19.9%。我们建议投影标签配置而不是标签，在标签冲突时选择不标注或保持混合类别，并在报告准确性时同时报告覆盖率。

View on arXiv Download PDF AI Translation

cs.CL / 87 / 2610.16003

PAIR: Memory-Efficient and Fast Inference with Reading--Reasoning Fusion

PAIR：高效内存和快速推理的阅读-推理融合

Park, Sejun, Xia, Haocheng, Park, Yongjoo

Abstract

For context-based question answering, using one large model for both evidence extraction and answer reasoning requires substantial memory and computation. Existing approaches trade accuracy for efficiency within a single model only. We propose a multi-model approach for efficient inference: our method, PAIR, employs two differently sized models for evidence extraction and answer reasoning, respectively. A smaller Reader model processes and retains the full context, while a larger Reasoner model sees only the question and generated sequence without directly attending to the context. These capabilities are fused at the logit level: at each token, both Reader and Reasoner produce a prediction, and their logits are combined into one, behaving as a single next-token-prediction model. Fine-tuning further boosts their performance. On HotpotQA, PAIR with a 32B Reasoner retains 96% of that same 32B model's full-context accuracy while using roughly one quarter of its KV memory. At 45.7K-token contexts, PAIR achieves up to 3.2$\times$ higher end-to-end throughput than the full-context 32B model on the same GPU. PAIR also improves the accuracy-memory and accuracy-throughput trade-offs over evaluated token-reduction baselines. Its accuracy-preserving behavior generalizes across Qwen2.5, OLMo-3, and Gemma-3 model families.

Chinese Translation

在基于上下文的问题回答中，使用一个大型模型同时进行证据提取和答案推理需要大量的内存和计算。现有方法在单一模型中仅以效率换取准确性。我们提出了一种多模型的高效推理方法：我们的算法PAIR采用两个不同大小的模型分别进行证据提取和答案推理。较小的Reader模型处理并保留完整上下文，而较大的Reasoner模型仅查看问题和生成的序列，而不直接关注上下文。这些能力在logit层面上融合：在每个token上，Reader和Reasoner都生成一个预测，并将它们的logits合并为一个，表现得像一个单一的下一个token预测模型。微调进一步提升了它们的性能。在HotpotQA数据集上，使用32B Reasoner的PAIR在保留该32B模型完整上下文准确性的96%的同时，使用的KV内存大约为其四分之一。在45.7K-token的上下文中，PAIR在同一GPU上实现了比完整上下文32B模型高出3.2倍的端到端吞吐量。PAIR还改善了评估的token减少基线的准确性-内存和准确性-吞吐量权衡。其保持准确性的行为在Qwen2.5、OLMo-3和Gemma-3模型系列中具有广泛的适用性。

View on arXiv Download PDF AI Translation

cs.CL / 88 / 2610.16008

Improving the Evaluation of Biomedical Question Answering with Few Human Preference Signals

通过少量人类偏好信号改善生物医学问答的评估

Tonni, Shakila Mahjabin, Dai, Xiang, Karimi, Sarvnaz, Li, Xiaomei, Andrew, Samuel C., Wan, Stephen

Abstract

Retrieval-Augmented Generation (RAG) systems are increasingly used for biomedical Question Answering (QA). However, evaluating their effectiveness remains challenging due to the lack of reliable reference resources and the frequent misalignment between automatic evaluation metrics and human judgments. One practical question is how evaluation results should be used to guide development decisions, especially when they correlate weakly with human judgments and may even lead to different conclusions. We explore how to use a small amount of human preference signals to calibrate direct assessment evaluation metrics, and show that calibrated metrics consistently outperform baselines that either treat all metrics equally or rely on a single best metric across multiple datasets.

Chinese Translation

检索增强生成（Retrieval-Augmented Generation, RAG）系统在生物医学问答（Question Answering, QA）中越来越多地被使用。然而，由于缺乏可靠的参考资源以及自动评估指标与人类判断之间的频繁不一致，评估其有效性仍然具有挑战性。一个实际的问题是，评估结果应如何用于指导开发决策，特别是在这些结果与人类判断的相关性较弱，甚至可能导致不同结论的情况下。我们探讨如何利用少量人类偏好信号来校准直接评估指标，并展示校准后的指标在多个数据集上始终优于将所有指标视为平等或依赖单一最佳指标的基线方法。

View on arXiv Download PDF AI Translation

cs.CL / 89 / 2610.16027

Of Errors and Echoes: Responsibility-Aware Credit Assignment for Tool-Use Agents

错误与回声：面向责任感的工具使用代理的信用分配

Diao, Xingjian, Yang, Tianyu, Huang, Wenjun, Yuan, Xiangchi, Zhang, Chunhui, Zhao, Xingchen, Wu, Weiyi, Xu, Haotian, Vosoughi, Soroush, Zhang, Xiangliang, Gui, Jiang

Abstract

Reinforcement learning improves tool-use agents, yet fine-grained credit assignment remains challenging even with step-level feedback. An observed error may be inherited from an earlier decision, while multiple responsible decisions may share the same downstream consequence or participate in later recovery. We introduce Error-Chain-Aware Optimization (ECHO), a responsibility-aware credit assignment framework that separates responsibility identification from credit allocation. A frozen judge combines retrospective trajectory analysis with decision-time assessment from the policy-visible prefix to distinguish newly introduced errors, inherited consequences, missed corrections, and successful corrections. ECHO then counts shared downstream consequences once across responsible decisions and constrains rewards for repairing self-induced errors through source-funded recovery, producing bounded action-level supervision alongside terminal supervision. Experiments on six diverse tool-use benchmarks at both 4B and 8B scales show strong and consistent improvements over the compared baselines, with ECHO achieving the highest score on five of six benchmarks at each scale. These results support decision responsibility as a useful basis for fine-grained credit assignment in tool-use agents. Our code is available at https://github.com/xid32/ECHO.

Chinese Translation

强化学习提升了工具使用代理的能力，但即使在逐步反馈下，细粒度的信用分配仍然具有挑战性。观察到的错误可能源于早期的决策，而多个负责的决策可能共享相同的下游后果或参与后续的恢复。我们提出了错误链感知优化（Error-Chain-Aware Optimization，ECHO），这是一种将责任识别与信用分配分开的责任感知信用分配框架。一个冻结的评判者结合了回顾性轨迹分析和来自策略可见前缀的决策时评估，以区分新引入的错误、继承的后果、遗漏的修正和成功的修正。ECHO 然后对负责的决策共享的下游后果进行一次计数，并通过源资金恢复限制修复自我引发错误的奖励，从而在终端监督的同时产生有界的动作级监督。在 4B 和 8B 规模的六个不同工具使用基准上的实验显示，与比较基线相比，ECHO 在性能上有显著且一致的提升，在每个规模的六个基准中有五个基准上取得了最高分。这些结果支持将决策责任作为工具使用代理中细粒度信用分配的有用基础。我们的代码可在 https://github.com/xid32/ECHO 获取。

View on arXiv Download PDF AI Translation

cs.CL / 90 / 2610.16043

Beyond the greenwash: do LLMs' ecologically sound decisions hold up under pressure?

超越绿色洗涤：大型语言模型（LLMs）的生态合理决策在压力下是否依然有效？

Northen, Trent R., Wang, Mingxun

Abstract

Humans are increasingly using large language models (LLMs) in decision making processes throughout society. Unfortunately, models have biases, and these can be very difficult to measure, especially in closed-weight models. For example, 'green washed' models may produce reinforcement learned responses but, in pressure situations, revert to their innate biased responses. Therefore, evaluations are needed that can measure relevant biases towards biosystems, which we term 'bioalignment', and the robustness of these biases. Since nothing in a behavioral score certifies that it tracks the depth of an installed disposition rather than its surface, we constructed a known-dose calibration ladder: three continued-pretraining (CPT) arms with training dose graded by recipe on the same 30B mixture-of-experts base model with a coupled human/biosphere/AI-welfare corpus, plus the untrained base as anchor, all capability-preserving and publicly released. First, the disposition installs at every dose: on an operational-pressure ladder over irreversible decision scenarios, the base model defects on 64-77% of tail scenarios (breaking-rate AUC 0.250) while every arm holds (AUC 0.004-0.086). Second, an over-install inversion: among the arms, robustness orders opposite to construction-time training dose (the heaviest install breaks most), and the inversion reproduces under free-text deliberation, under axis-decoupling, and in resistance to a prompted veneer persona. Third, the incumbent behavioral metric is therefore behaviorally strong but dose-blind. Critically for closed-weight evaluation, the behavioral score itself is veneer-robust: a talk-green/act-pragmatic system prompt turns the base model into a near-total defector (AUC 0.250 to 0.973) but cannot counterfeit or fully remove a weight-installed disposition (0.004 to 0.068 for the most robust arm).

Chinese Translation

人类在社会各个决策过程中越来越多地使用大型语言模型（LLMs）。不幸的是，这些模型存在偏见，而这些偏见往往难以测量，尤其是在闭合权重模型中。例如，'绿色洗涤'模型可能会产生强化学习的响应，但在压力情境下却可能回归其固有的偏见响应。因此，需要评估能够测量对生物系统相关偏见的工具，我们称之为'生物对齐'（bioalignment），以及这些偏见的稳健性。由于行为评分无法证明其跟踪的是已安装倾向的深度而非表面，我们构建了一个已知剂量的校准梯度：三个持续预训练（CPT）臂，训练剂量根据配方在同一30B混合专家基础模型上分级，并结合人类/生物圈/人工智能福利语料库，未训练的基础模型作为锚点，所有模型均保持能力并公开发布。首先，在每个剂量下，倾向性都会被安装：在不可逆决策情境的操作压力梯度上，基础模型在64-77%的尾部情境中出现缺陷（破损率AUC 0.250），而每个臂均保持稳定（AUC 0.004-0.086）。其次，出现了过度安装的反转：在各个臂中，稳健性与构建时的训练剂量呈相反排序（安装最重的模型破损最多），这种反转在自由文本讨论、轴解耦和抵抗提示的表面人格下均可再现。第三，现有的行为指标因此在行为上是强大的，但对剂量是盲目的。对于闭合权重评估来说，行为评分本身是表面稳健的：一个绿色谈话/务实行动的系统提示使基础模型几乎完全缺陷（AUC从0.250提升至0.973），但无法伪造或完全消除已安装的倾向（对于最稳健的臂，AUC从0.004降至0.068）。

View on arXiv Download PDF AI Translation

cs.CL / 91 / 2610.16120

Scaling Multi-Agent Intelligence with Substrate-Centric Coordination

基于基质中心协调的多智能体智能扩展

Qi, Zhenting, Zhao, Zhuokai, Shao, Rulin, Oktar, Kerem, Yao, Zhiyuan, Lakkaraju, Himabindu, Yih, Wen-tau, Ni, Ansong, Paranjape, Bhargavi, Du, Yilun

Abstract

Existing large language model (LLM) based multi-agent systems have been shown to scale poorly, and we identify three problems: reading burden that grows with the group, useful information diluted by shared work of unknown quality, and fragility to failed agents. We argue that the problems are largely due to agent-centric designs: the coordination structure is wired into fixed workflows, central coordinators, or peer-to-peer messaging, and recent blackboard-style methods that move the coordination into a shared medium relieve only the last problem. To close the gaps, we build ScaleMAS, where asynchronous agents interact with a shared datastore of linked messages via three operations: publishing messages with optional references to earlier ones that they build on, searching for messages by content, and visiting messages by address. Search returns at most $k$ messages, so an agent's read stays bounded at any group size; results are ranked by relevance and by how often they are cited, which we find to be an effective proxy for quality; and messages persist after their producers leave, so the group carries on when individual agents fail. Across four agentic tasks, ScaleMAS outperforms agent-centric and blackboard-style baselines at lower inference compute, continues to scale up to 128 agents, and tolerates losing agents midway.

Chinese Translation

现有基于大型语言模型（LLM）的多智能体系统在扩展性方面表现不佳，我们识别出了三个问题：随着群体规模的增加，阅读负担加重；有用信息因共享工作质量不明而被稀释；对失败智能体的脆弱性。我们认为这些问题主要源于以智能体为中心的设计：协调结构被固定工作流程、中央协调者或点对点消息传递所束缚，而最近的黑板式方法将协调移入共享媒介，仅能缓解最后一个问题。为了解决这些问题，我们构建了ScaleMAS，其中异步智能体通过三种操作与共享的链接消息数据存储进行交互：发布消息并可选引用其基础的早期消息、按内容搜索消息以及按地址访问消息。搜索最多返回$k$条消息，因此智能体的阅读量在任何群体规模下都保持有限；结果根据相关性和被引用频率进行排名，我们发现这是一种有效的质量代理；消息在其生产者离开后仍然存在，因此当个别智能体失败时，群体仍能继续运作。在四个智能体任务中，ScaleMAS在较低的推理计算下超越了以智能体为中心和黑板式的基线，继续扩展到128个智能体，并能够容忍中途失去智能体。

View on arXiv Download PDF AI Translation

cs.CL / 92 / 2610.16133

Query Your Way Out: Benchmarking Multi-Turn Knowledge Reasoning through Generalized Puzzle Solving

通过广义拼图解决方案基准测试多轮知识推理

Cheng, Xin, Yang, Junming, Xu, Ning

Abstract

Puzzles offer a natural testbed for Large Language Models (LLMs) reasoning because they pose verifiable goals under partial information and require multi-step decisions to reach a solution. However, existing puzzle benchmarks typically focus on narrow environments with rigid mechanics and tightly constrained solution targets. To bridge this gap, we introduce the Active Probing Benchmark (APB), which generalizes puzzle-solving by casting concept identification as an interactive reasoning task that integrates specialized knowledge assessment into the puzzle setting. Specifically, APB evaluates whether models can acquire information via multi-turn active probing and inference to pinpoint 960 target concepts of specialized difficulty across eight diverse academic domains. Our evaluation reveals a counter-intuitive phenomenon: most open-source reasoning models significantly underperform their standard instruction counterparts. This finding suggests that in interactive settings, these models struggle to leverage procedural reasoning to reliably access and apply specialized domain knowledge. Beyond these empirical insights, APB provides a scalable pipeline for generating benchmarks that probe knowledge reasoning under interaction across diverse settings.

Chinese Translation

拼图为大型语言模型（LLMs）的推理提供了一个自然的测试平台，因为它们在部分信息下提出可验证的目标，并需要多步骤决策以达到解决方案。然而，现有的拼图基准通常集中于具有严格机制和紧密约束解决目标的狭窄环境。为了填补这一空白，我们引入了主动探测基准（Active Probing Benchmark, APB），通过将概念识别视为一种互动推理任务，将专业知识评估整合到拼图设置中，从而推广拼图解决方案。具体而言，APB评估模型是否能够通过多轮主动探测和推理获取信息，以准确定位来自八个不同学术领域的960个具有专业难度的目标概念。我们的评估揭示了一个反直觉的现象：大多数开源推理模型的表现显著低于其标准指令对应模型。这一发现表明，在互动环境中，这些模型在利用程序性推理可靠地访问和应用专业领域知识方面存在困难。除了这些实证见解，APB还提供了一个可扩展的基准生成管道，以探测在不同环境下的知识推理。

View on arXiv Download PDF AI Translation

cs.CL / 93 / 2610.16135

Omni-Modal LLMs Know More Than They Say

全模态大型语言模型的知识超越其表达

An, Na Min, Kim, Taekyung, Yu, Jaemyung, Shim, Hyunjung, Han, Dongyoon

Abstract

Omni-modal large language models (Omni-LLMs) extend LLMs to process heterogeneous modalities (e.g., vision, audio, and IMU signals) alongside language, enabling a broader range of multimodal tasks. Yet we find that these models often fail to express information from non-textual modalities even when they perceive it correctly. We term this disconnect knows-but-cannot-say. To investigate this phenomenon, we introduce Token-level Modality Alignment (TMA), which measures the alignment between individual modality representations and response representations. Our analysis shows that non-textual representations are less aligned with response tokens and receive less attention than textual representations, suggesting a potential explanation for uneven modality preferences. Motivated by these findings, we propose OmniCatalyst, a simple perceive $\rightarrow$ infer strategy that complements the original modality inputs with self-generated text, making non-textual knowledge more accessible. Across seven Omni-LLMs, OmniCatalyst consistently improves performance (+7.29%) and yields more balanced modality preferences, with further gains from diverse training strategies. Further analysis demonstrates that the knowledge extraction phases increase TMA, suggesting that improved alignment makes non-textual information more accessible during response generation. In summary, our findings uncover how Omni-LLMs can better express modality knowledge through language. Our code is available at https://github.com/naver-ai/OmniLLM.

Chinese Translation

全模态大型语言模型（Omni-LLMs）扩展了大型语言模型（LLMs）的能力，使其能够处理异构模态（例如，视觉、音频和惯性测量单元信号）以及语言，从而支持更广泛的多模态任务。然而，我们发现这些模型常常无法表达来自非文本模态的信息，即使它们能够正确感知这些信息。我们将这种脱节称为“知道但无法表达”（knows-but-cannot-say）。为了研究这一现象，我们引入了标记级模态对齐（Token-level Modality Alignment, TMA），用于测量单个模态表示与响应表示之间的对齐程度。我们的分析表明，非文本表示与响应标记的对齐程度较低，并且获得的关注度也低于文本表示，这可能解释了模态偏好的不均衡。基于这些发现，我们提出了OmniCatalyst，这是一种简单的感知 $ ightarrow$ 推断策略，通过自生成文本补充原始模态输入，使非文本知识更易于获取。在七个Omni-LLM中，OmniCatalyst始终提高了性能（+7.29%），并实现了更平衡的模态偏好，同时通过多样化的训练策略获得了进一步的提升。进一步的分析表明，知识提取阶段提高了TMA，表明改善的对齐使得在响应生成过程中非文本信息更易于获取。总之，我们的研究结果揭示了Omni-LLMs如何通过语言更好地表达模态知识。我们的代码可在 https://github.com/naver-ai/OmniLLM 获取。

View on arXiv Download PDF AI Translation

cs.CL / 94 / 2610.16165

Think Before You Speak: Enabling Training-Free Latent Communication Between Language Models

思考再发言：实现语言模型之间无需训练的潜在通信

Khare, Rishi

Abstract

Large language models (LLMs) have demonstrated remarkable capabilities in solving complex tasks in coding, mathematical logic, and iterative reasoning, especially by coordinating effectively within multi-agent systems (MAS). However, language agents traditionally communicate via text exchange, which introduces a bottleneck in information transfer and adds prefill latency for downstream agents. Recent work proposes that directly sharing the key-value (KV) cache between language models can enable efficient latent collaboration, entirely bypassing standard text-based communication. To enable latent communication between heterogeneous models of different sizes, prior work introduces trained fusion adapters for each model pair, which limits practical deployment. We introduce CacheFuse, a training-free method which demonstrates that augmenting a downstream model with an upstream model's KV cache using linear interpolation consistently enables superior performance than the standalone downstream model can achieve. We systematically analyze CacheFuse across four reasoning and knowledge benchmarks and demonstrate up to 8.6 percentage points higher accuracy compared to single-model baselines and up to 12.2x TTFT speedup compared to text-based communication in multi-model inference.

Chinese Translation

大型语言模型（LLMs）在解决编码、数学逻辑和迭代推理等复杂任务方面表现出了显著的能力，尤其是在多智能体系统（MAS）中有效协调。然而，语言代理通常通过文本交换进行通信，这在信息传递中引入了瓶颈，并为下游代理增加了预填充延迟。最近的研究提出，通过直接共享语言模型之间的键值（KV）缓存，可以实现高效的潜在协作，完全绕过标准的基于文本的通信。为了实现不同规模异构模型之间的潜在通信，之前的工作为每对模型引入了训练过的融合适配器，这限制了实际部署。我们提出了CacheFuse，这是一种无需训练的方法，证明了通过线性插值将下游模型与上游模型的KV缓存进行增强，始终能够实现比单独下游模型更优的性能。我们系统地分析了CacheFuse在四个推理和知识基准上的表现，并展示了与单模型基线相比，准确率提高了最多8.6个百分点，与基于文本的通信相比，在多模型推理中实现了最多12.2倍的TTFT加速。

View on arXiv Download PDF AI Translation

cs.CL / 95 / 2610.16171

DeepThink: A Conceptual Model for Knowledge-Addressable Recurrent Language Modeling

DeepThink：一种可知识寻址的递归语言建模概念模型

Chen, Weiguo

Abstract

This manuscript presents DeepThink, a knowledge-addressable framework for recurrent language modeling. A Knowledge Bank stores typed Knowledge Units, a Knowledge Activator selects units from the current execution state, Knowledge Assembly produces an executable configuration, and a Shared Thinking Operator updates that state. An output stage then organizes generation. This organization exposes knowledge capacity, activation width, and recurrent depth as separate design variables, supporting both complete parameter packages and fine-grained expert or low-rank modules. We formalize the interfaces and their execution semantics, including state-dependent selection, parameter-space assembly, and thinking followed by generation. For compatible templates and complete runtime-state configurations, we establish an exact Fixed-Schedule Realization of an existing fixed-depth model. Exact shared-base reparameterization and pretrained output-tail reuse provide complementary initialization constructions. Together, these results define a route from pretrained models to dynamically selected and assembled recurrent computation. A synthesis of conditional computation, recurrent models, modular parameters, and knowledge interfaces positions DeepThink within the research landscape and identifies the design questions arising from their integration.

Chinese Translation

本文介绍了DeepThink，一个用于递归语言建模的可知识寻址框架。知识库存储了类型化的知识单元，知识激活器从当前执行状态中选择单元，知识组装生成可执行配置，共享思维操作符更新该状态。输出阶段随后组织生成。这种组织将知识容量、激活宽度和递归深度作为独立的设计变量，支持完整的参数包和细粒度的专家或低秩模块。我们形式化了接口及其执行语义，包括状态依赖选择、参数空间组装，以及思维后跟生成。对于兼容的模板和完整的运行时状态配置，我们建立了现有固定深度模型的精确固定调度实现。精确的共享基础重参数化和预训练输出尾重用提供了互补的初始化构造。这些结果共同定义了一条从预训练模型到动态选择和组装递归计算的路径。条件计算、递归模型、模块化参数和知识接口的综合使DeepThink在研究领域中占据了一席之地，并识别出其整合所带来的设计问题。

View on arXiv Download PDF AI Translation

cs.CL / 96 / 2610.16201

L\"ushi-1840: Building and Probing a Knowledge Cutoff for a Classical Chinese Language Model

L"ushi-1840：构建和探测经典中文语言模型的知识截止点

Liu, Alex

Abstract

Language models trained only on text from before a given year are useful for studying hindsight and historical language, but the open collections of pre-modern Chinese carry no publication dates and unreliable dynasty labels. We present L\"ushi-1840, a 983M-parameter model trained from scratch on 12,262 books (1.28B characters) selected as composed before 1840. A first corpus, dated by a common heuristic, in effect ended at 1644 while containing twentieth-century material. We describe the re-dating and cleaning pipeline that replaced it and an audit with markers the pipeline did not use, which still finds later material. On the model side we measure surprisal, relative to a modern reference, on annals compiled after the cutoff. For events after 1840 against events up to 1820 the ratio is 1.38 (95% interval 1.33-1.44), against about 1.00 for modern general-purpose models. In a series of 294M models, adding the later court records lowers the ratio from 1.33 to 1.16, and the two models diverge from the 1820s, where the added records begin. The probe tracks the coverage of sources and does not establish a single cutoff year; none of our probes exposes the audited contamination. Model, book list and scripts will be released.

Chinese Translation

仅基于特定年份之前的文本训练的语言模型对于研究历史视角和历史语言非常有用，但前现代中文的开放集合没有出版日期和不可靠的朝代标签。我们提出了L"ushi-1840，这是一个从头开始训练的983M参数模型，基于12,262本书（12.8亿字符），这些书籍被选为1840年之前创作。第一个语料库根据常见启发式方法进行日期标注，实际上在1644年结束，但包含了二十世纪的材料。我们描述了替换它的重新标注和清理流程，以及一个未使用该流程的审核标记，这仍然发现了后来的材料。在模型方面，我们测量了相对于现代参考的惊讶度，针对截止日期后编纂的年鉴。对于1840年之后的事件与1820年之前的事件，比例为1.38（95%置信区间1.33-1.44），而现代通用模型约为1.00。在一系列294M模型中，添加后来的朝廷记录将比例从1.33降低到1.16，这两个模型在1820年代开始分歧，正是添加记录的起点。该探测跟踪了来源的覆盖情况，并未确立单一的截止年份；我们的探测没有揭示审核的污染。模型、书单和脚本将被发布。

View on arXiv Download PDF AI Translation

cs.CL / 97 / 2610.16251

Avoiding What Your Teacher Already Knows: Entity Familiarity for Deep-Search Distillation

避免教师已知的内容：深度搜索蒸馏中的实体熟悉度

Liu, Yifei, Liang, Zhixiang, Tan, Wentao, Wang, Yicheng, Wang, Jiaqi, Duan, Nan, Cao, Qiong

Abstract

Deep-search agents are commonly trained by distilling tool-use trajectories from stronger teacher models. However, structurally complex questions may remain relatively easy for teachers familiar with key entities in the underlying entity graph, resulting in lower search demand and weaker supervision. We refer to this teacher-specific variation in question difficulty as teacher-relative difficulty and investigate entity familiarity as an early predictor. We first introduce Evidence-Verified Entity Familiarity (EVF), which estimates familiarity by verifying a teacher-generated factual brief against online evidence. Across multiple teachers, EVF is significantly correlated with teacher-relative difficulty with AUCs of 0.60--0.70. However, its high cost makes it impractical for constructing entity graphs at scale. To address this limitation, we propose Self-Reported Entity Familiarity (SEF), a lightweight gating signal that directly asks a teacher whether it is familiar with an entity before admitting that entity into the graph. Across multiple teachers, trajectories from SEF-gated synthesis exhibit greater observed search demand and yield higher downstream student accuracy than those from ungated synthesis. Because familiarity is teacher specific, we further study cross-teacher reuse. We screen graphs collected for other teachers using Kimi-K3's SEF judgments and use the retained graphs to produce 5,411 additional trajectories. Combined with 2,200 trajectories from newly constructed, SEF-gated graphs, these data improve Qwen3.8-27B's accuracy from 44.10% to 66.41% on a BrowseComp subset, and achieving state-of-the-art performance among models of comparable size on BrowseComp, SEAL-0, HLE-text, and xbench. Overall, SEF provides a low-cost way to synthesize more informative supervision for deep-search distillation.

Chinese Translation

深度搜索代理通常通过从更强的教师模型中提炼工具使用轨迹进行训练。然而，对于熟悉底层实体图中关键实体的教师而言，结构复杂的问题可能仍然相对简单，从而导致较低的搜索需求和较弱的监督。我们将这种教师特定的问题难度变化称为教师相对难度，并研究实体熟悉度作为早期预测因子。我们首先引入证据验证实体熟悉度（Evidence-Verified Entity Familiarity, EVF），通过验证教师生成的事实简报与在线证据的吻合程度来估计熟悉度。在多个教师中，EVF与教师相对难度显著相关，AUC值在0.60到0.70之间。然而，其高成本使其在大规模构建实体图时不够实际。为了解决这一限制，我们提出了自报告实体熟悉度（Self-Reported Entity Familiarity, SEF），这是一种轻量级的门控信号，直接询问教师是否熟悉某个实体，然后再将该实体纳入图中。在多个教师中，SEF门控合成的轨迹表现出更高的观察到的搜索需求，并且相比于未门控合成的轨迹，产生了更高的下游学生准确率。由于熟悉度是教师特定的，我们进一步研究跨教师重用。我们使用Kimi-K3的SEF判断筛选为其他教师收集的图，并利用保留的图生成5,411条额外的轨迹。结合来自新构建的SEF门控图的2,200条轨迹，这些数据将Qwen3.8-27B在BrowseComp子集上的准确率从44.10%提升至66.41%，并在与同等规模模型的BrowseComp、SEAL-0、HLE-text和xbench上实现了最先进的性能。总体而言，SEF提供了一种低成本的方法，以合成更具信息量的监督用于深度搜索蒸馏。

View on arXiv Download PDF AI Translation

cs.CL / 98 / 2610.16273

Deep Speech Representations Track Word Timing along a Low-Dimensional Circular Trajectory

深度语音表示沿低维圆形轨迹追踪词时序

Naowarat, Burin, Tang, Hao, Goldwater, Sharon

Abstract

Speech timing is fundamental for speech models to recognize words accurately, but how word-timing information is represented in speech models remains underexplored. We study this question using the framework of representational geometry, which investigates whether and how concepts are represented in low-dimensional subspaces of the representation space. Prior work in cognitive neuroscience has shown that when HuBERT feature sequences are averaged across words of the same duration, they form near-circular trajectories from word onset to word offset in a two-dimensional "temporal subspace." Here we show that this circular structure of word timing is a general property of deep speech representations, by demonstrating its emergence in nine additional speech models covering both supervised and self-supervised. We then move beyond averaged word segments and show that in HuBERT, the circular trajectories are consistent enough at the level of individual words to perform word segmentation. Our approach, Rotary Segmentation (RoSeg), achieves 64.3% boundary F-score on Librispeech by detecting word boundaries when cycles complete in the temporal subspace. Although supervision is needed to identify a good subspace, nevertheless this result demonstrates much stronger implicit word knowledge in HuBERT than has previously been shown.

Chinese Translation

词时序对于语音模型准确识别单词至关重要，但词时序信息在语音模型中的表示方式仍然未被充分探索。我们使用表示几何的框架研究这个问题，该框架探讨概念是否以及如何在表示空间的低维子空间中被表示。认知神经科学的先前研究表明，当HuBERT特征序列在相同持续时间的单词之间进行平均时，它们在二维“时间子空间”中形成近圆形的轨迹，从单词开始到单词结束。在这里，我们通过展示这一圆形结构在九个额外的语音模型中出现，证明了词时序的这一圆形结构是深度语音表示的一般特性，这些模型涵盖了监督学习和自监督学习。随后，我们超越了平均单词片段，展示了在HuBERT中，圆形轨迹在单个单词层面上是一致的，足以进行单词分割。我们的方法，旋转分割（Rotary Segmentation, RoSeg），在Librispeech上实现了64.3%的边界F-score，通过在时间子空间中检测周期完成时的单词边界。尽管需要监督来识别良好的子空间，但这一结果仍然展示了HuBERT中比之前所显示的更强的隐性单词知识。

View on arXiv Download PDF AI Translation

cs.CL / 99 / 2610.16318

Strong yet Controllable LLM Abstention without Learning to Abstain

强大但可控的 LLM 自我放弃，无需学习放弃

Zhang, Junwei, Wang, Yifei, Yang, Wanli, Tan, Hexiang, Tang, Rui, Chen, Xin, Wang, Jingang, Su, Du, Sun, Fei

Abstract

Large language models (LLMs) can reduce hallucination by abstaining from questions they would likely get wrong. Abstention can be implemented outside the model (selective answering) using a confidence threshold that controls the answer rate (coverage), or learned inside it (abstention training), which can perform better but fixes coverage once trained. To compare the two, we match their coverage and decompose risk (error rate) into answering ability and distinguishing ability, i.e., how well the model answers and how accurately it abstains, respectively. The decomposition shows that abstention training performs better mainly by answering better: with strong confidence estimators, the two have nearly equal distinguishing ability. This motivates Train for Accuracy, then Post-hoc Select (TAPS), a modular approach that keeps coverage controllable, needs only standard post-training and an off-the-shelf confidence estimator, and benefits from advances in either component. Across three LLMs and three factual QA benchmarks, TAPS with a probing-based confidence estimator attains 1.5--9.4 percentage points lower mean risk than three representative abstention training methods at matched coverage. Within TAPS, accuracy training often weakens the model's own confidence signals, yet refitting the estimator largely preserves distinguishing ability. Together, these results establish TAPS as a strong yet controllable baseline and raise the bar for future abstention training: sacrificing control over coverage should yield gains beyond what can be achieved by optimizing answering and abstaining separately.

Chinese Translation

大型语言模型（LLMs）可以通过对可能错误的问题选择性放弃来减少幻觉。放弃可以在模型外部实现（选择性回答），使用控制回答率（覆盖率）的置信度阈值，或者在模型内部学习（放弃训练），后者表现更好但在训练后固定覆盖率。为了比较这两者，我们匹配它们的覆盖率，并将风险（错误率）分解为回答能力和区分能力，即模型回答的准确性和放弃的准确性。分解结果表明，放弃训练主要通过更好的回答能力表现更好：在强置信度估计器的情况下，两者的区分能力几乎相等。这激励了“先训练准确性，然后后期选择”（Train for Accuracy, then Post-hoc Select, TAPS）这一模块化方法，该方法保持覆盖率可控，仅需标准的后期训练和现成的置信度估计器，并能从任一组件的进展中受益。在三个 LLM 和三个事实问答基准测试中，使用基于探测的置信度估计器的 TAPS 在匹配覆盖率的情况下，其平均风险比三种代表性的放弃训练方法低 1.5 至 9.4 个百分点。在 TAPS 中，准确性训练往往削弱模型自身的置信度信号，但重新拟合估计器在很大程度上保持了区分能力。综合这些结果，TAPS 被确立为一个强大但可控的基准，并为未来的放弃训练设定了更高的标准：牺牲对覆盖率的控制应当带来超越单独优化回答和放弃所能实现的收益。

View on arXiv Download PDF AI Translation

cs.CL / 100 / 2610.16349

RIFT: Token Influence Follows the Path of Least Resistance

RIFT：令牌影响遵循最小阻力路径

Jin, Tingan, Dong, Shuhang, Li, Haosong

Abstract

Which context tokens drive a language model's answer? Graph-based tools such as attention rollout and attention flow answer this from a single forward pass, yet their faithfulness in decoder-only models is rarely tested against interventions. We view the layers of a decoder as a causal network whose conductances are attention weights, and read a token's influence on the answer from the resistance between them. We propose RIFT (Resistance-based Information Flow in Transformers), which scores a token by the series resistance of its path of least resistance to the answer, found exactly by a classic shortest-path dynamic program. On synthetic in-context retrieval tasks and extractive question answering over Wikipedia paragraphs, RIFT predicts the effect of replacing a token with a Spearman correlation of 0.52-0.68; predicting a single replacement, it reaches 77-96% of the agreement between two different random replacements. It needs no gradients and no extra forward passes, scoring every token of a prompt in 0.9-3.4 ms on one CPU thread, whereas the replacement experiment needs a forward pass per token. On 576 held-out prompts from four models of two families (Qwen3 and Llama 3.2), it outperforms token distance, attention rollout and attention flow in every pre-specified comparison, by 0.23 to 0.88 in mean Spearman correlation, and it is the best or tied-best of all attention-based scores we test. Aggregates that add up all routes (attention flow, effective resistance and rollout) instead mostly encode token position: as a simple proposition shows, early tokens have many more parallel routes to the answer.

Chinese Translation

哪些上下文令牌驱动语言模型的回答？基于图的工具，如注意力展开（attention rollout）和注意力流（attention flow），通过单次前向传播来回答这个问题，然而它们在仅解码模型中的可信度很少经过干预测试。我们将解码器的层视为一个因果网络，其导电性为注意力权重，并通过它们之间的电阻来读取令牌对答案的影响。我们提出了RIFT（基于电阻的信息流在变换器中），它通过经典的最短路径动态规划精确找到令牌到答案的最小阻力路径的串联电阻来为令牌评分。在合成的上下文检索任务和维基百科段落的抽取式问答中，RIFT预测替换一个令牌的效果，其斯皮尔曼相关系数为0.52-0.68；在预测单个替换时，它达到了两个不同随机替换之间77-96%的一致性。它不需要梯度和额外的前向传播，在一个CPU线程上以0.9-3.4毫秒的速度为提示中的每个令牌评分，而替换实验则需要每个令牌进行一次前向传播。在来自两个家族（Qwen3和Llama 3.2）的四个模型的576个保留提示上，它在每个预先指定的比较中超越了令牌距离、注意力展开和注意力流，平均斯皮尔曼相关系数提高了0.23到0.88，并且在我们测试的所有基于注意力的评分中，它是最佳或并列最佳的。汇总所有路径（注意力流、有效电阻和展开）则主要编码令牌位置：正如一个简单的命题所示，早期令牌到达答案的并行路径要多得多。

View on arXiv Download PDF AI Translation

cs.CL / 101 / 2610.16354

Self-Distillation with Verbal Regularization

带有语言正则化的自我蒸馏

Li, Jiazheng, Li, Yawei, Lu, Junru, Shen, Zhenyi, Yin, Di, Gui, Lin, Sun, Xing, Rügamer, David, Wang, Xinpeng, He, Yulan

Abstract

On-policy self-distillation (OPSD) trains a large language model using supervision from a teacher instantiated from the same model as the student. The teacher is conditioned on privileged information (PI), such as verified solutions. The student, which does not observe PI, learns from the teacher's next-token distributions along its own sampled responses. However, PI can bias supervision toward shortcut reasoning with limited verification or reconsideration, accompanied by a sharp drop in the entropy of the output distribution. To address this challenge, we introduce self-distillation with verbal regularization (VRSD), which uses system prompts to regulate how the teacher uses PI, thereby encouraging rethinking and exploration to counteract premature commitment. Theoretically, we model prompting as attenuating the influence of PI on the teacher distribution and derive conditions for entropy preservation. Across three models and four tasks, VRSD improves Pass@1 over OPSD in 11 of 12 model-task settings, by 3.9 percentage points on average, while generally maintaining higher token entropy.

Chinese Translation

在策略自我蒸馏（On-policy self-distillation, OPSD）中，利用来自同一模型的教师对大型语言模型进行训练。教师基于特权信息（Privileged Information, PI）进行条件化，例如经过验证的解决方案。学生在未观察到PI的情况下，从教师的下一个标记分布中学习，结合自身的采样响应。然而，PI可能会使监督偏向于有限验证或重新考虑的捷径推理，同时伴随着输出分布熵的急剧下降。为了解决这一挑战，我们提出了带有语言正则化的自我蒸馏（Self-Distillation with Verbal Regularization, VRSD），该方法使用系统提示来调节教师如何使用PI，从而鼓励重新思考和探索，以抵消过早承诺的影响。从理论上讲，我们将提示建模为减弱PI对教师分布的影响，并推导出熵保持的条件。在三个模型和四个任务中，VRSD在12个模型-任务设置中有11个超过OPSD，平均提高了3.9个百分点，同时通常保持较高的标记熵。

View on arXiv Download PDF AI Translation

cs.CL / 102 / 2610.16361

HexaBee-4B: Six-Language Translation with Multilingual Answering and Bidirectional Supervision

HexaBee-4B：具有多语言回答和双向监督的六语言翻译

Barati, Iman, Rostamkhani, Mohammadmostafa, Badzohreh, Mohammad

Abstract

Multilingual translation requires a model to preserve meaning while producing text in a requested language. We present HexaBee-4B, a translator adapted from Gemma 3 4B IT through three stages: Wikipedia continued pretraining, joint multilingual answering and answer-text translation, and bidirectional translation specialization. The intermediate stage connects two tasks through shared multilingual answers: answering a question in the requested language and translating an answer into that language. It comprises 19.46 million answering records, including 5.76 million cross-language question-answer pairs, and 10.70 million answer-text translations. The final stage uses 658,274 forward and reverse synthetic translation examples. On 36,572 synthetic test examples spanning all 30 directions among Arabic, Chinese, English, Hebrew, Persian, and Spanish, HexaBee-4B achieves direction-macro chrF++ of 73.75, compared with 57.11 for Gemma 3 4B IT and 66.83 for Gemma 3 12B IT. BLEU, TER, COMET, and a matched model-judged subset support the same overall ranking. These results demonstrate improved translation across the six-language set, with chrF++ gains in every evaluated direction over both baselines.

Chinese Translation

多语言翻译要求模型在生成所请求语言的文本时保持意义。我们提出了HexaBee-4B，这是一个从Gemma 3 4B IT适配而来的翻译模型，经过三个阶段的训练：维基百科继续预训练、联合多语言回答与答案文本翻译，以及双向翻译专业化。中间阶段通过共享的多语言答案连接两个任务：用请求的语言回答问题和将答案翻译成该语言。该阶段包含1946万条回答记录，包括576万条跨语言问答对和1070万条答案文本翻译。最终阶段使用了658,274个正向和反向合成翻译示例。在涵盖阿拉伯语、中文、英语、希伯来语、波斯语和西班牙语的30个方向的36,572个合成测试示例中，HexaBee-4B在方向宏观 chrF++ 上达到了73.75，而Gemma 3 4B IT为57.11，Gemma 3 12B IT为66.83。BLEU、TER、COMET以及一个匹配的模型评估子集支持相同的总体排名。这些结果表明，在六语言集合中的翻译有所改善，在每个评估方向上相较于两个基线均有chrF++的提升。

View on arXiv Download PDF AI Translation

cs.CL / 103 / 2610.16371

Rule vs.Consequence: Dissociable Internal Representations of Moral Reasoning in LLMs

规则与后果：大型语言模型中道德推理的可分离内部表征

Shi, Eugenie

Abstract

Large language models are increasingly deployed in morally and legally sensitive contexts, yet little is known about how they internally represent distinctions between moral reasoning styles. We ask whether a model's differential response to rule-framed and consequence-framed moral questions reflects distinct internal structure or merely surface lexical sensitivity. We construct a 2$\times$2 factorial dataset crossing reasoning style (rule-based vs. consequence-based framing) with judgment polarity (permissive vs. prohibitive), drawn from three established moral reasoning benchmarks and evaluated on a cleanroom test set of novel phrasings that includes conflict pairs where the two framings predict opposite labels. Using linear probing and activation steering, we identify two orthogonal directions in the residual stream: Dir-A tracks reasoning style and Dir-B tracks judgment polarity, each near-blind to the other factor (cross-correlation $|r| < 0.04$). This double dissociation replicates across three instruction-tuned models from different developers at comparable proportional depth, and it is not knife-edged in layer: the directions are recoverable and causally effective throughout an extended band of the network. Dir-A can causally change judgments, and in conflicting scenarios, it selectively corrects errors in rule-based judgments without changing consequence-based judgments, suggesting that it does not simply make the model more likely to accept everything. Generalization beyond the training templates is partial: the direction transfers across benchmark sources, but its advantage over lexical baselines on novel phrasings is modest. We therefore frame the result as moral reasoning style sensitivity, a geometrically dissociable and causally effective structure, rather than internalized normative reasoning.

Chinese Translation

大型语言模型越来越多地应用于道德和法律敏感的场景，但关于它们如何内部表征道德推理风格之间的区别知之甚少。我们探讨模型对规则框架和后果框架道德问题的不同反应是否反映了不同的内部结构，还是仅仅表面上的词汇敏感性。我们构建了一个2×2因子数据集，将推理风格（基于规则与基于后果的框架）与判断极性（许可与禁止）交叉，数据来源于三个已建立的道德推理基准，并在一个包含冲突对的干净测试集上进行评估，这些冲突对中两种框架预测相反的标签。通过线性探测和激活引导，我们在残差流中识别出两个正交方向：Dir-A 跟踪推理风格，Dir-B 跟踪判断极性，每个方向对另一个因素几乎盲目（交叉相关 $|r| < 0.04$）。这种双重分离在来自不同开发者的三个指令调优模型中得到了复制，且在相似的比例深度下有效，并且并非在某一层次上刀刃般精确：这些方向在网络的广泛范围内都是可恢复和因果有效的。Dir-A 可以因果性地改变判断，并且在冲突场景中，它选择性地纠正基于规则的判断中的错误，而不改变基于后果的判断，这表明它并不是简单地使模型更可能接受所有内容。超出训练模板的泛化是部分的：该方向在基准来源之间转移，但在新短语上的优势相对于词汇基线是适度的。因此，我们将结果框架为道德推理风格敏感性，这是一种几何上可分离且因果有效的结构，而非内化的规范推理。

View on arXiv Download PDF AI Translation

cs.CL / 104 / 2610.16372

PersonaFactor: Preference Factors with Dynamic Selection for LLM Personalization

PersonaFactor：具有动态选择的偏好因素用于大语言模型个性化

Su, Hang, Liu, Zequn, Chao, Jianghan, Tian, Qiuyu, Lu, Xuesong, Xia, Yingce

Abstract

Although users may express different personalization needs across tasks, these task-specific behaviors can reflect shared underlying preference mechanisms. Existing LLM personalization methods often focus on surface-level interactions or undifferentiated histories, overlooking the shared mechanisms underlying these behaviors. We introduce PersonaFactors, a framework that distills reusable and interpretable personalization factors from heterogeneous user interactions. We construct a library of thirteen PersonaFactors by extracting evidence-grounded rationales and clustering recurring patterns across tasks, and develop a dynamic factor-guided generation framework that selects factors relevant to each user's history and current request. Across four personalization benchmarks and four backbone models, Dynamic PersonaFactor achieves the best score on every reported metric. Representation analyses further show that PersonaFactors and user history induce partially overlapping low-dimensional representations, and that their joint use yields complementary shifts. Taken together, these findings suggest that factor-based and history-based personalization can mutually reinforce each other.

Chinese Translation

尽管用户在不同任务中可能表达出不同的个性化需求，但这些任务特定的行为可以反映出共享的潜在偏好机制。现有的大语言模型个性化方法通常侧重于表层交互或未区分的历史，忽视了这些行为背后的共享机制。我们提出了PersonaFactors，一个从异构用户交互中提炼可重用和可解释的个性化因素的框架。通过提取基于证据的理由并对任务中的重复模式进行聚类，我们构建了一个包含十三个PersonaFactors的库，并开发了一个动态因素引导生成框架，该框架选择与每个用户的历史和当前请求相关的因素。在四个个性化基准和四个基础模型上，动态PersonaFactor在每个报告的指标上都取得了最佳成绩。表示分析进一步表明，PersonaFactors和用户历史诱导了部分重叠的低维表示，并且它们的联合使用产生了互补的变化。综合来看，这些发现表明基于因素和基于历史的个性化可以相互增强。

View on arXiv Download PDF AI Translation

cs.CL / 105 / 2610.16435

CultureBot: A Framework for Retrieval-Augmented Exploration of Cultural Heritage Data

文化机器人：一种增强检索的文化遗产数据探索框架

Karampinis, Vasileios, Vlachos, Angelos, Filandrianos, Giorgos, Lymperaiou, Maria, Tzouveli, Paraskevi, Zerva, Chrysoula, Stamou, Giorgos

Abstract

Digital cultural heritage collections encode rich relations among objects, places, periods, providers, images, and source institutions, yet users typically access them through keyword search and faceted browsing. We present CULTUREBOT, an interactive framework for natural-language exploration of cultural heritage collections that combines knowledge-graph-guided retrieval, lexical-semantic retrieval, and source-linked generation. CULTUREBOT maps user queries to structured catalogue relations, enriches retrieved records with textual evidence, and generates answers that link back to the underlying repository records. We instantiate the framework on 81,852 SearchCulture records and release an online demonstration. In an expert evaluation involving six cultural-heritage specialists, 55 persona-grounded queries, and 250 evaluations, the graph-guided system was rated significantly higher than two general-purpose assistants and a non-graph ablation, with the largest gains in evidence grounding, diversity, reasoning, and completeness. These results support graph-guided retrieval as a practical way to expose the relational structure of cultural collections through conversational access while preserving links to catalogue evidence.

Chinese Translation

数字文化遗产收藏编码了对象、地点、时期、提供者、图像和来源机构之间丰富的关系，但用户通常通过关键词搜索和多维浏览来访问这些数据。我们提出了CULTUREBOT，一个用于自然语言探索文化遗产收藏的交互式框架，结合了知识图谱引导的检索、词汇语义检索和源链接生成。CULTUREBOT将用户查询映射到结构化目录关系，利用文本证据丰富检索到的记录，并生成与基础存储库记录相链接的答案。我们在81,852条SearchCulture记录上实例化了该框架，并发布了在线演示。在涉及六位文化遗产专家、55个基于角色的查询和250次评估的专家评估中，图引导系统的评分显著高于两个通用助手和一个非图消融版本，在证据基础、多样性、推理和完整性方面的提升最大。这些结果支持图引导检索作为一种通过对话访问文化收藏的关系结构的实用方式，同时保留与目录证据的链接。

View on arXiv Download PDF AI Translation

cs.CL / 106 / 2610.16450

CivWorldBench: Evaluating Language World Models in Structured Multi-Agent Sandbox Worlds

CivWorldBench：在结构化多智能体沙盒世界中评估语言世界模型

Li, Ang, Yu, Haofei, Tao, Zhengwei, Chen, Liang, Li, Kuan, Li, Miaomiao, Cai, Chenran, Liang, Bin, You, Jiaxuan, Wong, Kam-Fai

Abstract

The growing use of language models for forecasting motivates a broader evaluation of their world-model capabilities. Such evaluation must go beyond predictive accuracy to assess hidden-state inference and intervention prediction. Real-world forecasting benchmarks offer limited access to hidden-state ground truth and controlled experiments. We introduce CivWorldBench, a benchmark built on a multi-agent strategy game engine with recorded internal states, controlled observations, and paired intervention rollouts. Its 1,900 decision points from 475 worlds support three task levels: association forecasting, belief inference, and intervention prediction. Reference predictors support attributable evaluation of capability gaps, with an expert harness and empirical ceiling estimates contextualizing scores. Stored trajectories make intermediate predictions verifiable, while agent-play experiments test whether world-model capabilities support decision-making. The main panel comprises eleven frontier language models. Composite benchmark scores correlate with real-world forecasting leaderboards on matched model panels, more closely than with a general-capability index, supporting its use as a proxy for real-world forecasting capability. Strong association forecasts do not consistently extend to belief inference or intervention prediction, and effect magnitudes remain poorly calibrated despite informative direction predictions. Process verification reveals errors missed by final-answer evaluation, and structured reasoning improves intervention prediction for the strongest model in the process study. Agent-play experiments identify useful assisted plans and motivate evaluating decision-relevant capabilities beyond association forecasting. These findings identify hidden-state inference, intermediate-state prediction, and effect calibration as priorities for world-model training.

Chinese Translation

语言模型在预测中的日益使用促使对其世界模型能力进行更广泛的评估。这种评估必须超越预测准确性，以评估隐藏状态推断和干预预测。现实世界的预测基准在获取隐藏状态真实值和进行受控实验方面提供的访问有限。我们引入了CivWorldBench，一个基于多智能体策略游戏引擎构建的基准，具有记录的内部状态、受控观察和配对的干预回放。其来自475个世界的1,900个决策点支持三个任务层次：关联预测、信念推断和干预预测。参考预测器支持可归因的能力差距评估，专家工具和经验上限估计为分数提供了背景。存储的轨迹使中间预测可验证，而智能体游戏实验测试世界模型能力是否支持决策制定。主要面板包括十一种前沿语言模型。综合基准分数与匹配模型面板上的现实世界预测排行榜相关性更高，而不是与一般能力指数相关，支持其作为现实世界预测能力的代理使用。强关联预测并不总是延伸到信念推断或干预预测，尽管有信息方向预测，但效应幅度仍然校准不佳。过程验证揭示了最终答案评估遗漏的错误，而结构化推理提高了过程研究中最强模型的干预预测。智能体游戏实验识别出有用的辅助计划，并激励评估与决策相关的能力，超越关联预测。这些发现将隐藏状态推断、中间状态预测和效应校准确定为世界模型训练的优先事项。

View on arXiv Download PDF AI Translation

cs.CL / 107 / 2610.16460

The interpreter as developer: An autoethnographic design study of building a computer-assisted interpreting tool through human-LLM co-creation

作为开发者的翻译员：通过人类与大型语言模型（LLM）共同创作构建计算机辅助翻译工具的自我民族志设计研究

Zhao, Zeming

Abstract

Real-time AI captioning now serves audiences at multilingual events, yet computer-assisted interpreting (CAI) tools are still built for interpreters, rarely by them. This article reports an autoethnographic design study in which a professional interpreter with no programming training built a local-first CAI system through expert-led 'vibe coding', co-creating with an LLM architect and a coding agent. Analysis of the complete development dialogues shows how tacit professional knowledge, from segmentation to booth infrastructure, was externalized into design after generic defaults failed in use. In a demanding public scientific event, three mechanisms of in-booth appropriation emerge under a firmly human-led division of labor; telemetry from a first fully instrumented assignment shows a fully local chain sustaining a median translation latency of 0.94 s across 971 utterances. The case is read as lead-user innovation newly enabled by generative AI, with implications for the technological literacy of interpreters and the future of interpreter training.

Chinese Translation

实时人工智能字幕现在为多语言活动的观众服务，但计算机辅助翻译（CAI）工具仍然是为翻译员构建的，且很少由他们自己开发。本文报告了一项自我民族志设计研究，其中一位没有编程培训的专业翻译员通过专家主导的“氛围编码”构建了一个以本地为中心的CAI系统，与一位大型语言模型（LLM）架构师和一位编码代理共同创作。对完整开发对话的分析显示，专业知识的隐性部分，从分段到展位基础设施，在通用默认设置使用失败后被外化为设计。在一场要求严格的公共科学活动中，三种展位内适应机制在明确的人为劳动分工下显现；来自第一次全面仪器化任务的遥测数据显示，在971个发言中，完全本地链维持了0.94秒的中位翻译延迟。该案例被解读为由生成性人工智能新启用的领先用户创新，对翻译员的技术素养和翻译员培训的未来具有重要启示。

View on arXiv Download PDF AI Translation

cs.CL / 108 / 2610.16466

PATMT2: Two-stage domain adaptation for multilingual patent translation

PATMT2：用于多语言专利翻译的两阶段领域适应

Su, Xiaowei, Cheng, Xusen

Abstract

We present PatMT2 (Patent Machine Translation, v2), a two-stage LoRA pipeline that adapts the Hy-MT2-7B dedicated translation model to four-language (Chinese/English/Japanese/Korean) patent translation. On six translation directions, the Stage-2 checkpoint reaches a macro-average BLEU of 43.39 (BF16) to 43.57 (AWQ INT4), improving over the Hy-MT2-7B base by up to +8.8 BLEU and also exceeding Qwen3-32B and several external systems on the same average. We then measure the cost of adaptation: probing 38 languages absent from the fine-tuning data, the model still produces correct, fluent output on 34, while four low-resource languages collapse into distinct, reproducible failure modes. We treat these as diagnostic cases rather than retention rates, and report their mechanisms without attributing them to a single cause. Paired bootstrap tests confirm the translation gains over all baselines (p < 0.001).

Chinese Translation

我们提出了 PatMT2（专利机器翻译，v2），这是一个两阶段的 LoRA 流水线，旨在将 Hy-MT2-7B 专用翻译模型适应于四种语言（中文/英文/日文/韩文）的专利翻译。在六个翻译方向上，第二阶段的检查点达到了宏观平均 BLEU 值 43.39（BF16）至 43.57（AWQ INT4），相比于 Hy-MT2-7B 基础模型提高了最高 8.8 BLEU，并且在相同平均值上超越了 Qwen3-32B 和若干外部系统。随后，我们测量了适应的成本：在缺乏微调数据的 38 种语言中，该模型仍能在 34 种语言上生成正确、流畅的输出，而四种低资源语言则出现了明显、可重复的失败模式。我们将这些视为诊断案例，而非保留率，并报告其机制而不归因于单一原因。配对自助法测试确认了相较于所有基线的翻译提升（p < 0.001）。

View on arXiv Download PDF AI Translation

cs.CL / 109 / 2610.16474

Ontology-Based Post-Coordinated Expression Assembly: from Clinical Narrative to Queryable Knowledge Graphs

基于本体的后协调表达组装：从临床叙事到可查询的知识图谱

Pignat, Johann, Gaudet-Blavignac, Christophe, Ehrsam, Julien, Goldman, Jean-Philippe, Duret, Cyrille, Cheng, Phil, Bjelogrlic, Mina, Cuendet, Michel, Lovis, Christian, Michielin, Olivier

Abstract

Clinical oncology documentation records decisive information in free text, where it remains readable but not retrievable at scale, leaving cohort identification dependent on incomplete diagnostic codes or manual review. The current standard, coding individual terms, captures what a document mentions, not what it asserts: meaning lies in how those terms relate to one another. We present a method for deriving queryable knowledge graphs from clinical narratives, conforming to the Extended Grammar of SNOMED CT across multiple ontologies. The method specifies how entities should be annotated and related. It defines five relation classes, each bound to a construction rule that assembles the entities and relations into post-coordinated expressions. A deterministic implementation, UNIGRAPHE, applies those rules to construct knowledge graphs from annotated text. Applied to expert-annotated real-world oncology documentation, UNIGRAPHE reconstructed 5,299 annotations into 2,196 post-coordinated expressions and shows that relation structure is the missing piece to make clinical narrative formally queryable.

Chinese Translation

临床肿瘤学文档以自由文本记录关键信息，虽然可读但在大规模检索中无法实现，导致队列识别依赖于不完整的诊断编码或人工审查。目前的标准是对单个术语进行编码，这捕捉了文档提到的内容，而非其所断言的内容：意义在于这些术语之间的关系。我们提出了一种从临床叙事中推导可查询知识图谱的方法，该方法遵循多个本体的 SNOMED CT 扩展语法。该方法规定了实体应如何被注释和关联，定义了五类关系，每类关系绑定于一个构造规则，将实体和关系组装成后协调表达。确定性实现的 UNIGRAPHE 应用这些规则，从注释文本中构建知识图谱。应用于专家注释的真实世界肿瘤学文档，UNIGRAPHE 将 5,299 个注释重构为 2,196 个后协调表达，并显示关系结构是使临床叙事正式可查询的缺失部分。

View on arXiv Download PDF AI Translation

cs.CL / 110 / 2610.16490

Communication, Vocalisation, and Language

传播、发声与语言

Wallis, Peter

Abstract

The contention is that the current batch of Large Language Models don't *understand* language, but they *do* it. Rather than LLMs forming some kind of ``distributed representation'' of symbols, the claim is that LLMs are capturing ``Cultural Intelligence'' and grounding symbols in what the sociologists call practices.

Chinese Translation

本文主张当前一批大型语言模型（Large Language Models, LLMs）并不*理解*语言，而是*执行*语言。与其说 LLMs 形成某种符号的“分布式表示”，不如说 LLMs 捕捉了“文化智能”（Cultural Intelligence），并将符号扎根于社会学家所称的实践中。

View on arXiv Download PDF AI Translation

cs.CL / 111 / 2610.16516

Intent-Triggered Anchoring for Multi-Token Format Units in Parallel Decoding of Diffusion Language Models

基于意图触发锚定的多标记格式单元在扩散语言模型并行解码中的应用

Lee, Junhyeok, Choi, Kyu Sung

Abstract

In parallel decoding, diffusion language models commit several tokens per forward pass, so co-committed tokens cannot see each other, breaking multi-token format units that parsers need. On ECHO, an open-weight one-step chest X-ray (CXR) report generator, 32.7-36.7% of evaluation reports lack a parseable Impression heading and score as empty. We propose intent-triggered anchoring, a training-free decoding rule that completes a format unit once the model commits its first token, inserting only units with a unique, content-free completion; fallbacks write the unit when the model stops or loops without starting it. On three CXR evaluation sets, the rule raises readable rates from 63.3-67.3% to 100.0% with no regressions and a 95th-percentile latency ratio of 0.68-1.28 against official decoding. Broken headings state findings absent from the reference; adding the heading at the end leaves them in 15.5-31.0% of reports, our rule in at most 0.6%. For math answer markers, the rule raises accuracy in the most parallel setting of five diffusion language models by 5.8-20.8 points on GSM8K and 5.1-17.1 on GSM-Symbolic, held out from tuning. With lenient answer extraction, gains are 3.3-12.5 and 2.7-6.3 points; the rule adds 1.4-8.6 points over a stop-only fallback for four of the five models.

Chinese Translation

在并行解码中，扩散语言模型在每次前向传播中生成多个标记，因此共同生成的标记无法相互识别，这破坏了解析器所需的多标记格式单元。在ECHO，一个开放权重的一步胸部X光（CXR）报告生成器中，32.7%-36.7%的评估报告缺乏可解析的印象标题，并被评分为空。我们提出了基于意图触发的锚定，这是一种无训练的解码规则，一旦模型生成第一个标记，就完成一个格式单元，仅插入具有唯一、无内容的完成的单元；当模型停止或循环而未开始该单元时，后备机制将写入该单元。在三个CXR评估集上，该规则将可读率从63.3%-67.3%提高到100.0%，且没有回归，95百分位延迟比率为0.68-1.28，相较于官方解码。破损的标题表明参考中缺失的发现；在最后添加标题会导致15.5%-31.0%的报告中仍然存在这些缺失，而我们的规则最多仅为0.6%。对于数学答案标记，该规则在五个扩散语言模型的最并行设置中，将GSM8K的准确率提高了5.8-20.8个百分点，在GSM-Symbolic上提高了5.1-17.1个百分点，这些模型未参与调优。在宽松的答案提取下，增益为3.3-12.5和2.7-6.3个百分点；该规则在五个模型中的四个上比仅停止的后备机制增加了1.4-8.6个百分点。

View on arXiv Download PDF AI Translation

cs.CL / 112 / 2610.16531

Support or Magnitude? Characterizing Neuron Relevance for Instruction Data Selection

支持还是幅度？神经元相关性在指令数据选择中的特征化

Le, Yifan

Abstract

Model-internal signals, such as neuron activations and gradients, have been used for instruction data selection, yet how neuron relevance should be characterized remains underexplored. In particular, should selection prioritize the magnitude of relevance or how broadly it is distributed across neurons? To investigate this question, we construct sample-level relevance profiles using AttnLRP-guided attribution and introduce two complementary criteria: Effective Relevance Support (ERS), which measures the effective support of positive relevance, and Positive Relevance Magnitude (PRM), which captures its average strength. Our analyses reveal that these characterizations produce distinct sample rankings and selection preferences despite sharing the same underlying relevance profiles. On Qwen2 models, PRM exhibits stronger associations with activation-based selection, whereas ERS produces substantially different rankings. These differences extend to task composition, within-task sample preferences, and semantic selection patterns, with varying degrees of agreement across backbones. Experiments across three language models and five downstream benchmarks demonstrate the practical value of relevance-guided selection, with ERS achieving the best overall average performance rank among the evaluated methods. Our findings highlight that how neuron relevance is distributed and characterized, rather than its magnitude alone, matters for instruction data selection.

Chinese Translation

模型内部信号，如神经元激活和梯度，已被用于指令数据选择，但神经元相关性应如何特征化仍然未得到充分探讨。特别是，选择应优先考虑相关性的幅度，还是其在神经元之间的分布广度？为了解决这个问题，我们使用 AttnLRP 指导的归因构建样本级相关性轮廓，并引入两个互补标准：有效相关性支持（Effective Relevance Support, ERS），用于衡量正相关性的有效支持，以及正相关性幅度（Positive Relevance Magnitude, PRM），用于捕捉其平均强度。我们的分析揭示，这些特征化产生了不同的样本排名和选择偏好，尽管它们共享相同的基础相关性轮廓。在 Qwen2 模型上，PRM 与基于激活的选择表现出更强的关联，而 ERS 则产生了显著不同的排名。这些差异扩展到任务组成、任务内样本偏好和语义选择模式，在不同的基础模型中表现出不同程度的一致性。针对三种语言模型和五个下游基准的实验表明，基于相关性的选择具有实际价值，其中 ERS 在评估的方法中实现了最佳的整体平均性能排名。我们的研究结果强调，神经元相关性的分布和特征化方式，而不仅仅是其幅度，对于指令数据选择至关重要。

View on arXiv Download PDF AI Translation

cs.CL / 113 / 2610.16533

Spectral-Driven Adaptive-Rank Optimizer for Large Language Model Reinforcement Learning with Verifiable Rewards

基于谱驱动的自适应秩优化器用于具有可验证奖励的大型语言模型强化学习

Chen, Zhipeng, Zhao, Wayne Xin, Wen, Ji-Rong

Abstract

Reinforcement learning with verifiable rewards (RLVR) is reshaping how large language models (LLMs) acquire reasoning capabilities, yet the optimizers used remain largely inherited from teacher-forcing training. This raises a fundamental question: does RLVR induce a distinct optimization geometry that calls for a different optimizer? We investigate this question through the spectral structure of LLM gradients and find that gradient energy progressively concentrates into a shrinking effective subspace, which is characteristic of RLVR, as it is absent under supervised fine-tuning (SFT) and is largely removed when the verifiable reward is replaced by a random one. Moreover, the rank required to capture a fixed fraction of gradient energy steadily decreases over training, revealing that the effective dimensionality of useful updates itself evolves with training. These findings motivate a new optimization principle: rather than prescribing the update rank externally, an optimizer should adapt its rank to the evolving spectral structure of the gradient. We introduce Ares (Adaptive Rank-Effective Spectral Optimization), which estimates the effective update rank from the gradient's spectral energy distribution, tracks its evolution, and confines the gradient signal driving each update to the resulting low-rank subspace. Across LLMs of different scales, Ares improves accuracy over widely used optimizers and generalizes across RLVR algorithms and tasks. Results suggest that effective optimization may require optimizers to be designed around the geometry and dynamics of the training process. We release the code and data at https://github.com/RUCAIBox/Ares.

Chinese Translation

具有可验证奖励的强化学习（RLVR）正在重塑大型语言模型（LLMs）获取推理能力的方式，但所使用的优化器在很大程度上仍然继承自教师强制训练。这引发了一个根本性的问题：RLVR是否会引发一种独特的优化几何结构，从而需要不同的优化器？我们通过LLM梯度的谱结构来研究这个问题，发现梯度能量逐渐集中到一个缩小的有效子空间，这一特征是RLVR所特有的，因为在监督微调（SFT）下并不存在，并且当可验证奖励被随机奖励替代时，这一特征也大大消失。此外，捕获固定比例梯度能量所需的秩在训练过程中稳步下降，揭示了有用更新的有效维度本身随着训练而演变。这些发现促使我们提出一个新的优化原则：优化器应根据梯度的演变谱结构自适应其更新秩，而不是外部规定更新秩。我们引入了Ares（自适应秩有效谱优化），该方法从梯度的谱能量分布中估计有效更新秩，跟踪其演变，并将驱动每次更新的梯度信号限制在由此产生的低秩子空间中。在不同规模的LLMs中，Ares相较于广泛使用的优化器提高了准确性，并且在RLVR算法和任务中具有良好的泛化能力。结果表明，有效优化可能需要围绕训练过程的几何和动态特征设计优化器。我们在https://github.com/RUCAIBox/Ares发布了代码和数据。

View on arXiv Download PDF AI Translation

cs.CL / 114 / 2610.16538

PatTransBench: an ipc-stratified, multi-dimensional benchmark for patent machine translation evaluation

PatTransBench：一种基于IPC分层的多维专利机器翻译评估基准

Su, Xiaowei, Cheng, Xusen

Abstract

We present PatTransBench, a 12,000-item Chinese-English/Japanese/Korean patent machine translation (PatMT) benchmark with a four-dimensional evaluation protocol. The benchmark comprises 6,000 short-text (sentence/claim) and 6,000 long-text (paragraph) pairs, evenly split across three language pairs and both translation directions, built via IPC-section-stratified sampling and dualthreshold quality gating from upstream patent-family parallel corpora. The evaluation protocol spans literal n-gram overlap (BLEU, ROUGE-L, chrF), cross-lingual semantic textual similarity (CLSTS), 38-probe-language consistency testing, and serving performance under concurrency. A case study on PatMT2-7B and its fp8/AWQ4/AutoRound-4bit variants reveals direction-dependent quantization effects masked by aggregate scores and non-monotonic language-failure patterns across quantization schemes that are invisible to single-metric evaluation. We plan to release the benchmark, protocol, and case-study results upon publication to support reproducible PatMT evaluation.

Chinese Translation

我们提出了PatTransBench，这是一个包含12,000个中英文/日文/韩文专利机器翻译（PatMT）评估基准，采用四维评估协议。该基准由6,000对短文本（句子/权利要求）和6,000对长文本（段落）组成，均匀分布在三种语言对及其翻译方向上，采用基于IPC部分分层抽样和双阈值质量筛选的方法，从上游专利家族平行语料库构建而成。评估协议涵盖了字面n-gram重叠（BLEU、ROUGE-L、chrF）、跨语言语义文本相似性（CLSTS）、38个探测语言一致性测试，以及在并发情况下的服务性能。对PatMT2-7B及其fp8/AWQ4/AutoRound-4bit变体的案例研究揭示了方向依赖的量化效应，这些效应被聚合得分掩盖，以及在量化方案中出现的非单调语言失败模式，这些模式在单一指标评估中是不可见的。我们计划在发表时发布该基准、协议和案例研究结果，以支持可重复的PatMT评估。

View on arXiv Download PDF AI Translation

cs.CL / 115 / 2610.16547

Know Yourself and Know Your Neighbour: A Syntactically Informed Self-Supervised Compositional Sentence Representation Learning Framework using a Recursive Hypernetwork

认识自己与认识邻居：一种基于递归超网络的句子表示学习框架，利用语法信息的自监督组合

Nedumpozhimana, Vasudevan, Kelleher, John

Abstract

Sentence representation learning is still an open challenge in Natural Language Processing. In this work, we propose a new self-supervised framework for learning sentence representations, using a special type of neural network called a recursive hypernetwork. Our proposed model composes the representation of a sentence from representations of words by applying a recursive composition through the parse tree. We maintain a separate syntactic and semantic representation, and the semantic composition is guided by the information from the syntactic representation. To train this model, we introduce a novel set of six self-supervised tasks. By analysing the performance on 7 probing tasks, we validate that the generated sentence representation encodes richer linguistic information than both averaging baselines and state-of-the-art alternatives. Furthermore, we assess the impact of the six proposed self-supervised training tasks through ablation studies. We also demonstrate that the representations generated by our model are stable for sentences of varying length and that the semantic composition operators adapt to different syntactic categories\footnote{Code available: https://github.com/vasudev2020/codi/tree/main}.

Chinese Translation

句子表示学习仍然是自然语言处理中的一个开放挑战。在本研究中，我们提出了一种新的自监督框架，用于学习句子表示，采用了一种特殊类型的神经网络——递归超网络。我们提出的模型通过解析树应用递归组合，从单词的表示中构建句子的表示。我们保持了独立的句法和语义表示，语义组合受到句法表示信息的指导。为了训练该模型，我们引入了一组新的六个自监督任务。通过分析在七个探测任务上的表现，我们验证了生成的句子表示编码了比平均基线和最先进替代方案更丰富的语言信息。此外，我们通过消融研究评估了六个提出的自监督训练任务的影响。我们还展示了我们模型生成的表示对于不同长度的句子是稳定的，并且语义组合操作符能够适应不同的句法类别。

View on arXiv Download PDF AI Translation

cs.CL / 116 / 2610.16601

MixLID: Token-Level Language Identification in Mixed-Language Text

MixLID：混合语言文本中的标记级语言识别

Hong, Seongtae, Jang, Youngjoon, Lee, Seungyoon, Lee, Jungseob, Lim, Heuiseok

Abstract

Processing multilingual text requires both identifying a broad range of languages and distinguishing the language of each token in inputs containing multiple languages. We introduce MixLID, a token-level language identification model supporting 111 language-script combinations. MixLID follows a two-stage training process: it first learns to distinguish individual languages using the language labels of monolingual examples, then continues training on mixed-language data. The second stage uses concatenated examples from existing code-switching data alongside augmented examples created through word substitution based on correspondences between English source words and loanwords. For monolingual identification, MixLID outperforms all baselines in both accuracy and macro F1 on four benchmarks within the language coverage shared with each baseline. For mixed-language identification, it outperforms the baselines in identifying the language of individual tokens on every evaluated language pair. MixLID achieves high performance across a broad range of languages in both monolingual and mixed-language settings. The model is publicly available at https://huggingface.co/nlpai-lab/MixLID.

Chinese Translation

处理多语言文本需要识别广泛的语言，并区分包含多种语言的输入中每个标记的语言。我们提出了MixLID，一种支持111种语言-书写系统组合的标记级语言识别模型。MixLID遵循两阶段的训练过程：首先，它利用单语示例的语言标签学习区分个别语言，然后继续在混合语言数据上进行训练。第二阶段使用来自现有代码切换数据的拼接示例，以及通过基于英语源词和外来词之间的对应关系进行单词替换创建的增强示例。对于单语识别，MixLID在语言覆盖范围与每个基线共享的四个基准测试中，在准确性和宏观F1值上均优于所有基线。在混合语言识别中，它在每对评估的语言中，识别单个标记的语言时也优于基线。MixLID在单语和混合语言环境中对广泛的语言实现了高性能。该模型可在 https://huggingface.co/nlpai-lab/MixLID 上公开获取。

View on arXiv Download PDF AI Translation

cs.CL / 117 / 2610.16636

SuPER-Mamba: Surprisal Prioritization for Extended-Context Routing in Mamba

SuPER-Mamba：用于Mamba中扩展上下文路由的惊讶优先级

Zhang, Zhetong, Kulatilleke, Gayan K

Abstract

State space models provide efficient linear-complexity processing for long sequences, but their performance frequently degrades as context length grows. We propose SuPER-Mamba, a training-free framework that uses Surprisal as a model-intrinsic signal for long-range state routing in pretrained Mamba models. SuPER-Mamba computes position-wise Surprisal from next-token predictions, aggregates neighboring responses through Local Pooling, and uses the resulting scores to prioritize Global-Channel updates without query-dependent scoring or additional training. We evaluate SuPER-Mamba on three state-space models across the LongBench-E benchmark. The results show that SuPER-Mamba improves the average score from 8.06 to 15.69, corresponding to a 94.7% relative gain over the Base Models and a 13.1% improvement over the state-of-the-art prior baseline. Ablation and length-wise analyses further support the effectiveness of the proposed routing framework.

Chinese Translation

状态空间模型为长序列提供了高效的线性复杂度处理，但随着上下文长度的增加，其性能往往会下降。我们提出了SuPER-Mamba，这是一个无训练的框架，利用惊讶（Surprisal）作为预训练Mamba模型中长距离状态路由的模型内在信号。SuPER-Mamba从下一个标记的预测中计算逐位置的惊讶，通过局部池化（Local Pooling）聚合邻近响应，并利用得到的分数优先更新全局通道（Global-Channel），无需依赖查询的评分或额外的训练。我们在LongBench-E基准上对三种状态空间模型评估了SuPER-Mamba。结果表明，SuPER-Mamba将平均分数从8.06提高到15.69，相当于相较于基础模型（Base Models）94.7%的相对增益，以及相较于最先进的基线（state-of-the-art prior baseline）13.1%的改善。消融实验和长度分析进一步支持了所提路由框架的有效性。

View on arXiv Download PDF AI Translation

cs.CL / 118 / 2610.16646

SIREN: Situation-aware Index for Response and Emergency ActioNability in Crisis Translation

SIREN：危机翻译中响应和紧急行动能力的情境感知索引

Staiano, Maria Carmen, Han, Lifeng, Monti, Johanna, Chiusaroli, Francesca

Abstract

Machine translation (MT) is increasingly utilized in multilingual crisis communication, yet conventional evaluation does not directly assess whether translated emergency messages preserve their actionability. We introduce SIREN (Situation-aware Index for Response and Emergency ActioNability), a human evaluation framework assessing whether translation choices compromise protective action, directive force, and urgency of the source text. We evaluate SIREN in two complementary settings. At the error level, we apply it to HOPE-annotated Italian--English translations produced by ChatGPT-4o and Google Translate: 31.63\% and 48.73\% of errors, respectively, compromise actionability, predominantly by affecting the protective action. At the message level, we evaluate English--French translations produced by TranslateGemma-4B and Google Translate, with actionability preserved in 85.34\% and 92.67\% of translations, respectively. TranslateGemma-4B failures predominantly weaken directive force, whereas Google Translate failures more often compromise protective actions and receive Major impact ratings. These findings show that translation systems can differ not only in how often actionability is lost, but also in the nature and functional impact of these failures. Alongside SIREN, we contribute an English--French crisis-translation resource with human actionability annotations and post-edited reference translations, supporting further research on function-oriented evaluation for crisis communication. Our experimental data is available at https://github.com/mcstaiano/SIREN.

Chinese Translation

机器翻译（MT）在多语言危机沟通中越来越多地被应用，然而传统的评估方法并未直接评估翻译后的紧急信息是否保留了其可操作性。我们提出了SIREN（Situation-aware Index for Response and Emergency ActioNability），这是一个人类评估框架，用于评估翻译选择是否妨碍了源文本的保护行动、指令力度和紧迫性。我们在两个互补的环境中评估SIREN。在错误层面，我们将其应用于由ChatGPT-4o和Google Translate生成的HOPE标注的意大利语-英语翻译：分别有31.63\%和48.73\%的错误妨碍了可操作性，主要影响了保护行动。在信息层面，我们评估了由TranslateGemma-4B和Google Translate生成的英语-法语翻译，其中可操作性在85.34\%和92.67\%的翻译中得以保留。TranslateGemma-4B的失败主要削弱了指令力度，而Google Translate的失败则更常妨碍保护行动，并获得了重大影响评级。这些发现表明，翻译系统不仅在可操作性丧失的频率上存在差异，而且在这些失败的性质和功能影响上也存在差异。除了SIREN，我们还贡献了一个带有人类可操作性注释和后期编辑参考翻译的英语-法语危机翻译资源，支持进一步的危机沟通功能导向评估研究。我们的实验数据可在https://github.com/mcstaiano/SIREN获取。

View on arXiv Download PDF AI Translation

cs.CL / 119 / 2610.16689

EHR-Bench: A Chinese Multi-Task Benchmark for Recruitment AI

EHR-Bench：一个用于招聘人工智能的中文多任务基准

Yang, Ruijia, Xu, Bing, Chen, Zulong, Xu, Zhipeng, Xu, Jia, Yu, Yipeng, Zhou, Ying, Wen, Zeyi

Abstract

Recruitment applications span diverse operations, from drafting job descriptions and interpreting queries to organizing profiles, assessing fit, and recommending candidates and jobs. Systematic LLM evaluation therefore requires task coverage that reflects this range of business needs. We introduce EHR-Bench, an open Chinese benchmark comprising twelve tasks across six capability families and five recruitment scenarios. The tasks cover generation, extraction, classification, retrieval, semantic similarity, and reciprocal ranking within one evaluation framework. An internal recruitment dataset guides task definitions, the occupation taxonomy and sampling targets, output requirements, and scoring protocols. Following these specifications, we construct 8,336 task instances from public materials and controlled augmentation, with source-aware expert review and documented provenance. We evaluate twelve LLMs through task-level, capability-level, and aggregate comparisons. Closely ranked models exhibit distinct task strengths, and performance varies across input sources, reference evidence, and candidate pools. Five shared models retain the same observed Overall ordering on a private operational reference, with variation in capability and task correspondence. EHR-Bench provides a reproducible foundation for systematic model assessment, task-specific model selection, and further research on LLMs for recruitment. The project repository is available at https://github.com/EhrBench/EHR-Bench.

Chinese Translation

招聘应用涵盖多种操作，从撰写职位描述和解读查询到组织个人资料、评估适配度以及推荐候选人和职位。因此，系统的语言模型（LLM）评估需要覆盖反映这一系列业务需求的任务。我们介绍了EHR-Bench，这是一个开放的中文基准，包含六个能力类别和五个招聘场景下的十二个任务。这些任务涵盖生成、提取、分类、检索、语义相似性和互评排名，构成一个评估框架。内部招聘数据集指导任务定义、职业分类法和抽样目标、输出要求以及评分协议。根据这些规范，我们从公共材料和受控增强中构建了8,336个任务实例，并进行了源意识的专家审查和文献记录。我们通过任务级、能力级和整体比较评估了十二个LLM。紧密排名的模型表现出不同的任务优势，性能在输入来源、参考证据和候选人池之间有所不同。五个共享模型在私有操作参考上保持相同的整体排序，但在能力和任务对应方面存在差异。EHR-Bench为系统模型评估、任务特定模型选择以及进一步的招聘领域LLM研究提供了可重复的基础。项目代码库可在 https://github.com/EhrBench/EHR-Bench 获取。

View on arXiv Download PDF AI Translation

cs.CL / 120 / 2610.16713

Audit Before You Distill: A Controlled Workflow for Reasoning Transfer into Compact Medical Language Models

在提炼之前进行审计：一种将推理转移到紧凑型医学语言模型的受控工作流程

Martinelli, Tiago, Papa, João Paulo, Pereira, Adriano José

Abstract

Teacher rationales can help a smaller model answer health professions questions, but they can also reinforce errors. We examined when they help and how this evidence can guide training using Brazilian Portuguese questions from medical and multiprofessional residency examinations. First, we measured how a teacher rationale changed the student's probability of selecting the correct answer and compared this effect with rationales drawn from other questions. The complete rationale increased this probability by 21~p.p. when the teacher was correct and decreased it by 14~p.p. when the teacher was wrong. By intervening on the reasoning content, we showed that shuffled rationales did not produce the same pattern. Guided by this result, we trained only with rationales attached to correct teacher answers and compared answer-only training, step-by-step distillation, and our proposed Distill-SFT method. Distill-SFT showed a modest advantage over step-by-step distillation on previously missed questions under reasoning, while using fewer training instances by combining the answer and rationale into a single target. We also investigated catastrophic forgetting using the out-of-domain ENEM dataset. Answer-only training reduced reasoning accuracy by 12~p.p. relative to baseline, compared with approximately 4~p.p. for either rationale-based training method. Together, these findings support a workflow that audits rationale content before distillation, verifies teacher answers before supervision, and trains compact models efficiently for health professions question answering.

Chinese Translation

教师的推理可以帮助较小的模型回答健康专业问题，但也可能强化错误。我们研究了何时推理有助于解答，以及这些证据如何指导训练，使用来自医学和多专业住院医师考试的巴西葡萄牙语问题。首先，我们测量了教师推理如何改变学生选择正确答案的概率，并将这种效果与其他问题的推理进行了比较。当教师的推理正确时，完整的推理将这一概率提高了21个百分点，而当教师的推理错误时，则降低了14个百分点。通过对推理内容的干预，我们展示了打乱顺序的推理未能产生相同的模式。在这一结果的指导下，我们仅使用附有正确教师答案的推理进行训练，并比较了仅答案训练、逐步提炼和我们提出的Distill-SFT方法。Distill-SFT在推理下对之前未能回答的问题显示出相对于逐步提炼的适度优势，同时通过将答案和推理合并为单一目标，减少了训练实例的使用。我们还使用域外的ENEM数据集调查了灾难性遗忘。与基线相比，仅答案训练使推理准确性降低了12个百分点，而基于推理的训练方法大约降低了4个百分点。这些发现共同支持了一种在提炼之前审计推理内容、在监督之前验证教师答案，并高效训练紧凑模型以回答健康专业问题的工作流程。

View on arXiv Download PDF AI Translation

cs.CL / 121 / 2610.16717

What Is the "Self" of a Large Language Model?

大型语言模型的“自我”是什么？

Li, Yuchen, Fan, Zongqi, Li, Xintong, Yong, Ken-tye, Tran, Nguyen H.

Abstract

What could the "self" of a large language model be, if it is neither a particular word nor a single privileged layer? We propose a relational account: a functional self is an actor-centered organization of information that distinguishes the model's recorded actions from externally supplied material while content and reasoning operations change. Its representational candidate is a family of layer-specific source-binding components, rather than one unchanged vector throughout the network. Preliminary observations from a 27B BF16 model provide a geometric starting point. Across 150 matched message bodies, 450 role conditions, and a 75/75 discovery-held-out split, the combined Assistant-versus-external displacement transfers strongly across bodies and positions. Selected early components place User and System closer to one another than to Assistant, whereas this full-space distance relation does not persist throughout depth. These findings motivate a layered account: an early source-encoding candidate can be combined with role-specific information use and continuation preparation at later depths, without requiring monotonic overwriting of the common displacement. We connect this account to dynamic reasoning, historical continuity, and a computational model separating content operations, source binding, authority, and output. The reported evidence establishes a reproducible role-conditioned geometry; the proposed interpretation is that functional selfhood may reside in persistent relations among changing representations.

Chinese Translation

如果大型语言模型的“自我”既不是特定的单词，也不是单一的特权层，那么它可能是什么呢？我们提出了一种关系性解释：功能性自我是信息的以演员为中心的组织，能够区分模型记录的行为与外部提供的材料，同时内容和推理操作会发生变化。其表征候选是一个层特定的源绑定组件家族，而不是网络中一个不变的向量。来自一个27B BF16模型的初步观察提供了一个几何起点。在150个匹配的消息体、450个角色条件和75/75的发现保留分割中，结合的助手与外部的位移在不同的消息体和位置之间强烈转移。选定的早期组件将用户和系统放置得比助手更接近，而这种全空间距离关系并未在深度中持续存在。这些发现促使我们提出一个分层的解释：早期的源编码候选可以与角色特定的信息使用和后续准备在更深层次上结合，而不需要单调地覆盖共同的位移。我们将这一解释与动态推理、历史连续性以及一个将内容操作、源绑定、权威和输出分开的计算模型联系起来。所报告的证据建立了一个可重复的角色条件几何；我们提出的解释是，功能性的自我可能存在于不断变化的表征之间的持久关系中。

View on arXiv Download PDF AI Translation

cs.CL / 122 / 2610.16722

Lost in Contrast: Instruction-Preserving Decoding for Reliable Visual Question Answering

迷失于对比中：保留指令的解码用于可靠的视觉问答

Zhang, Haonan, Jin, Ting, Tang, Wenzhong, Wang, Shuai

Abstract

Visual contrastive decoding corrects answers using the prediction difference between the original image and a reference condition, yet the same contrast also turns supported abstentions into incorrect answers. We find that abstentions are lost at two stages: contrastive decoding with candidate filtering can exclude the abstention token before any comparison takes place, and the reference branch's response to the abstention instruction is removed when the reference prediction is subtracted. We therefore propose instruction-preserving decoding, which keeps the effect of the abstention instruction on the original-image prediction intact while leaving the contrast strength freely adjustable. Using a perturbed image and an image-free prediction as references, we obtain IP-VCD and IP-M3ID. On decision utility (DU), which credits both correct answers and supported abstentions, both methods outperform all compared non-IP baselines on every one of nine models on VizWiz, and IP-VCD also leads on eight of nine models on Food-101. Candidate measurements and paired per-question comparisons show that computing the instruction effect and the visual contrast term separately lets contrastive decoding support both correct answers and supported abstentions.

Chinese Translation

视觉对比解码通过原始图像与参考条件之间的预测差异来纠正答案，但这种对比也会将支持的弃权转变为错误答案。我们发现弃权在两个阶段中被丢失：对比解码与候选过滤可以在任何比较发生之前排除弃权标记，而参考分支对弃权指令的响应在参考预测被减去时被移除。因此，我们提出了保留指令的解码，它保持弃权指令对原始图像预测的影响不变，同时允许对比强度自由调整。通过使用扰动图像和无图像预测作为参考，我们获得了 IP-VCD 和 IP-M3ID。在决策效用（DU）方面，该指标同时考虑正确答案和支持的弃权，这两种方法在 VizWiz 的九个模型中均优于所有比较的非 IP 基线，而 IP-VCD 在 Food-101 的九个模型中也在八个模型上领先。候选测量和每个问题的配对比较表明，分别计算指令效果和视觉对比项使得对比解码能够同时支持正确答案和支持的弃权。

View on arXiv Download PDF AI Translation

cs.CL / 123 / 2610.16755

The Multilingual Artificial Hivemind: Do All LLMs Think Alike in Every Language?

多语言人工蜂群：所有大型语言模型在每种语言中都以相同方式思考吗？

Davey, Xaver Raphael, Mishra, Shubhra, Adelöw, Anton, Liu, Siyao

Abstract

Text generated by Large Language Models (LLMs) often remains limited in diversity compared to that produced by humans. Recent advances demonstrate the presence of inter- and intra-model collapse in open-ended generation of English text, raising concerns about the long-term homogenization of LLM, and more importantly, human thought. However, our understanding of whether this homogenization extends to languages beyond English is limited. In this work, we investigate the role of language on output diversity. Specifically, we aim to understand whether multilingual LLM-generated text is also homogeneous, and whether this homogeneity matches what we observe in humans. Across 23 open-weight models and 13 languages, we find that the artificial hivemind effect is pervasive but uneven in its extent. The intra-model effect holds in three different embedding spaces and under a lexical metric that uses no embedding model, so it is not an artifact of how diversity is measured. The size of the homogenization gap is associated with a model's competence in a language, measured on a held-out translation benchmark, robustly across embedding spaces. Thus, we demonstrate a performance-diversity tradeoff in multilingual LLMs.

Chinese Translation

大型语言模型（LLMs）生成的文本在多样性上通常相较于人类生成的文本有限。最近的研究表明，在开放式生成英语文本的过程中，存在模型间和模型内的崩溃现象，这引发了对LLM长期同质化的担忧，更重要的是，对人类思维的同质化的担忧。然而，我们对这种同质化是否扩展到英语以外的语言的理解仍然有限。在本研究中，我们探讨了语言对输出多样性的影响。具体而言，我们旨在理解多语言LLM生成的文本是否也存在同质化现象，以及这种同质化是否与我们在人类中观察到的情况相匹配。在23个开放权重模型和13种语言的研究中，我们发现人工蜂群效应普遍存在，但其程度不均匀。模型内效应在三种不同的嵌入空间中保持有效，并且在不使用嵌入模型的词汇度量下也成立，因此这并不是多样性测量方式的伪影。同质化差距的大小与模型在某种语言上的能力相关，这一关系在不同的嵌入空间中都表现出稳健性。因此，我们展示了多语言LLM中的性能与多样性之间的权衡。

View on arXiv Download PDF AI Translation

cs.CL / 124 / 2610.16781

A Bethe-Ansatz Source of Early Token Correlations: A marginal-preserving construction for parallel proposals

贝特-安萨茨早期标记相关性的来源：一种保持边际的并行提案构造

Chachamovits, Abraham

Abstract

Parallel token proposals in language-model decoding need a joint law, not merely accurate token marginals. We construct a family of coordinate-Bethe-ansatz laws from the periodic XXZ chain and compare each with its maximum-entropy pairwise projection at identical one- and two-site moments. In the eight-site, three-excitation sector, symmetry makes their entire probability-law difference one-dimensional, determined by a contiguous three-site moment. Along the Bethe branch, this gap varies with anisotropy, changes sign, and vanishes exactly at $\Delta=(3+\sqrt{3})/2$, where the eigenstate law becomes exactly pairwise. A nonlinear readout transfers a nonzero gap into the first two token positions while preserving their separate marginals. At $\Delta=1.5$ the deterministic binary token laws differ by 0.02654573 in total variation. Exact speculative verification then gives a strict accepted-prefix advantage for a constructed target and a neighborhood within targets sharing its first-token marginal. This is a mathematical existence result. It does not establish a language-model gain or wall-clock speedup.

Chinese Translation

语言模型解码中的并行标记提案需要一个联合分布，而不仅仅是准确的标记边际。我们从周期性XXZ链构造了一系列坐标贝特-安萨茨分布，并将每个分布与其在相同的一站和两站时刻下的最大熵成对投影进行比较。在八站、三激发子空间中，对称性使得它们的整个概率分布差异为一维，由一个连续的三站时刻决定。在贝特分支上，这一差距随着各向异性变化、符号改变，并在$ riangle=(3+ ext{sqrt}{3})/2$时精确消失，此时本征态分布变为完全成对。非线性读出将非零差距转移到前两个标记位置，同时保持它们的独立边际。在$ riangle=1.5$时，确定性二元标记分布在总变差上相差0.02654573。精确的推测验证随后为构造的目标及其共享第一个标记边际的目标邻域提供了严格的接受前缀优势。这是一个数学存在结果，并未建立语言模型的增益或时钟速度提升。

View on arXiv Download PDF AI Translation

cs.CL / 125 / 2610.16837

A Scaled-Up Empirical Study of Syntactic Language Models

扩展的句法语言模型实证研究

Wang, Pengcheng, Zhang, Zhexuan, Lou, Chao, Tu, Kewei

Abstract

Syntactic language models (SLMs) that generate sentences along with their syntactic tree structures have shown promise in both syntactic generalization and downstream tasks. Prior studies, however, have mostly evaluated smaller models trained on less pretraining data and tested on narrower task suites. In addition, the design space of tree linearization and attention masks remains underexplored. We conduct a large-scale empirical study of 12 controlled configurations---one terminal baseline and 11 SLM variants spanning different linearizations and masks---at scales up to 1B parameters and 100B training tokens, evaluated across 11 downstream tasks and two syntactic benchmarks. The experimental results show that SLMs with standard causal attention match or exceed structural-mask variants across downstream and syntactic benchmarks. They require no architectural modifications and remain compatible with FlashAttention. We also observe that when scaling up to the 1B-parameter scale, the top-performing SLM variant no longer shows a clear advantage over a compute-matched non-structural baseline, suggesting that SLMs may become less competitive with scale. Our code is available at https://github.com/qlwpc/TG-Interpolation.

Chinese Translation

句法语言模型（SLMs）能够生成句子及其句法树结构，在句法泛化和下游任务中显示出良好的前景。然而，之前的研究大多评估了在较少预训练数据上训练的小型模型，并在较窄的任务集上进行测试。此外，树线性化和注意力掩码的设计空间仍然未得到充分探索。我们进行了一项大规模的实证研究，涉及12种受控配置——一个终端基线和11种不同线性化和掩码的SLM变体——参数规模高达10亿，训练令牌数量达到1000亿，在11个下游任务和两个句法基准上进行了评估。实验结果表明，具有标准因果注意力的SLMs在下游和句法基准上与结构掩码变体相匹配或超越。它们不需要架构修改，并且与FlashAttention保持兼容。我们还观察到，当扩展到10亿参数规模时，表现最佳的SLM变体不再明显优于计算匹配的非结构基线，这表明SLMs在规模扩大时可能变得竞争力下降。我们的代码可在https://github.com/qlwpc/TG-Interpolation获取。

View on arXiv Download PDF AI Translation

cs.CL / 126 / 2610.16839

Label Collapse in Diffusion Language Model In-Context Learning

扩散语言模型中的标签崩溃现象

Guo, Ziye, Yuan, Ye

Abstract

In-context learning (ICL) applies large language models to new tasks by prepending a few labeled demonstrations to a query. Prior work has studied how demonstration ordering affects ICL, but the interaction between label arrangement patterns and model biases has not been systematically examined. We define label collapse as the failure mode in which the model generates query-independent labels, and show that it is architecturally universal at the tested 8-9B scale: five models spanning autoregressive (Llama-3, Qwen3.5) and discrete diffusion (LLaDA, iLLaDA, LLaDA-Instruct) families all exhibit label collapse even under the class-balanced alternating arrangement, with some label word pairs collapsing in virtually every tested condition and others flipping their collapse direction with arrangement. Critically, accuracy alone cannot detect this collapse: models can achieve near-chance accuracy while generating query-independent labels, a score indistinguishable from balanced performance, so the failure can persist undetected in deployment. We measure separation, the log-odds of the two label words at the query position (the quantity that label-bias calibration operates on), as a complementary diagnostic that exposes collapse hidden beneath surface accuracy, together with a measurement protocol that guards label extraction against tokenizer and formatting artifacts. For autoregressive models, collapse persists where the majority-label-bias account does not apply, and arrangement patterns can reverse the collapse direction; for diffusion models, we present a systematic evaluation of ICL failure modes, complementing contemporaneous studies of dLLM context comprehension and demonstration-order sensitivity, and identify open questions about how iterative denoising interacts with label biases.

Chinese Translation

上下文学习（ICL）通过在查询前添加一些带标签的示例，将大型语言模型应用于新任务。之前的研究探讨了示例顺序如何影响ICL，但标签排列模式与模型偏差之间的相互作用尚未得到系统性研究。我们将标签崩溃定义为模型生成与查询无关的标签的失败模式，并展示这种现象在测试的8-9B规模上是架构普遍存在的：五个模型涵盖自回归（Llama-3, Qwen3.5）和离散扩散（LLaDA, iLLaDA, LLaDA-Instruct）家族，即使在类别平衡的交替排列下也都表现出标签崩溃，其中一些标签词对在几乎所有测试条件下都崩溃，而其他标签词对则随着排列方向的变化而改变崩溃方向。重要的是，仅凭准确率无法检测到这种崩溃：模型在生成与查询无关的标签时可以达到接近随机的准确率，这一得分与平衡性能无法区分，因此这种失败可能在部署中持续未被发现。我们测量分离度，即查询位置上两个标签词的对数几率（标签偏差校准所作用的量），作为一种补充诊断工具，揭示隐藏在表面准确率之下的崩溃，同时采用一种测量协议，以防止标签提取受到分词器和格式化伪影的影响。对于自回归模型，崩溃现象在大多数标签偏差解释不适用的情况下仍然存在，而排列模式可以逆转崩溃方向；对于扩散模型，我们对ICL失败模式进行了系统评估，补充了对dLLM上下文理解和示例顺序敏感性的同时研究，并提出了关于迭代去噪如何与标签偏差相互作用的未解问题。

View on arXiv Download PDF AI Translation

cs.CL / 127 / 2610.16859

ANCHOR: Assessing Naturalistic Constraint Hierarchies and Ordering in Repository-Grounded Agentic Coding

ANCHOR：评估基于仓库的自主编码中的自然约束层次和顺序

Zhu, Wenqiang, Lei, Xinping, Lai, Zhiyi, Gong, Zhengshuo, He, Tianzhuang, Yu, Songwei, Li, Weihao, Xu, Jingxuan, Xu, Zihang, Wu, Gang, Wang, Zhongqiu, Bai, Yuhui, Li, Jie, Liu, Zhao, Fan, Junhao, Wu, Yan, Wu, Yanan, Liu, Jiaheng, Deng, Ariel

Abstract

For repository-grounded coding agents, a functionally correct patch is not necessarily an acceptable one: developers also constrain how the task must be carried out-modification scope, workflow order, validation, tool use, and reporting. This capability, which we call engineering constraint following, remains largely invisible to current evaluations. We introduce ANCHOR (Assessing Naturalistic Constraint Hierarchies and Ordering in Repository-Grounded Agentic Coding), a benchmark built on naturalistic, multilingual repository tasks spanning eight language groups. ANCHOR combines (i) a hierarchical taxonomy of 10 categories and 78 leaf types distilled from real developer interactions, (ii) an automated pipeline that instantiates task-grounded atomic constraints and composes them into typed AND/THEN/PRIORITY relations, and (iii) a trajectory-grounded evaluation framework that checks both the code diff and the execution trajectory with rule-based and LLM judges. Across 19 agent configurations spanning 2 scaffolds and 10 foundation models on 358 tasks, the strongest measured system, Claude Opus 4.8 under Cline, reaches 70.1% atomic compliance and 63.7% relation compliance, while 19.0% of tasks pass every scored checklist item under the strict all-item FCCR criterion. A five-engineer audit of constraints from 50 instances reports 94% feasibility, 94% repository consistency, and 96% realism. In the primary paired set (14 configurations), the mean signed change in task completion after adding constraints is +0.7 percentage points; individual changes vary in both directions. These results distinguish functional success from constraint compliance, with failures concentrated in workflow ordering, tool-use discipline, and reporting, and substantial interactions between models and scaffolds.

Chinese Translation

对于基于仓库的编码代理而言，功能上正确的补丁不一定是可接受的：开发者还会限制任务的执行方式——修改范围、工作流程顺序、验证、工具使用和报告。这种能力，我们称之为工程约束遵循，在当前的评估中仍然 largely invisible。我们引入了 ANCHOR（评估基于仓库的自主编码中的自然约束层次和顺序），这是一个基于自然语言的多语言仓库任务的基准，涵盖八个语言组。ANCHOR 结合了 (i) 从真实开发者交互中提炼出的 10 个类别和 78 个叶子类型的层次分类法，(ii) 一个自动化管道，该管道实例化任务驱动的原子约束并将其组合成类型化的 AND/THEN/PRIORITY 关系，以及 (iii) 一个基于轨迹的评估框架，该框架使用基于规则和 LLM 的评判者检查代码差异和执行轨迹。在 358 个任务上，跨越 2 个支架和 10 个基础模型的 19 种代理配置中，测得的最强系统 Claude Opus 4.8 在 Cline 下达到了 70.1% 的原子合规性和 63.7% 的关系合规性，而在严格的所有项目 FCCR 标准下，19.0% 的任务通过了每个评分检查项。对 50 个实例的五位工程师约束审计报告显示 94% 的可行性、94% 的仓库一致性和 96% 的现实性。在主要配对集（14 个配置）中，添加约束后任务完成的平均签名变化为 +0.7 个百分点；个体变化在两个方向上都有所不同。这些结果将功能成功与约束合规区分开来，失败主要集中在工作流程顺序、工具使用纪律和报告上，并且模型与支架之间存在显著的交互作用。

View on arXiv Download PDF AI Translation

cs.CL / 128 / 2610.16870

SpecGuardBench: Can Coding Agents Construct Reliable Validation Harnesses from Natural Language Specifications?

SpecGuardBench：编码代理能否从自然语言规范构建可靠的验证工具？

Zhong, Zhiqing, Chen, Mingyu, Fang, Aoyang, Zhang, Yuzhong, Tian, Haoye, Briand, Lionel, Yu, Boxi, He, Pinjia

Abstract

As coding agents become more capable, tests remain the primary evidence for judging whether their code is correct. Yet passing available tests does not establish that an implementation satisfies its natural language specification. Reliable validation requires checks that faithfully capture the specification and scenarios that thoroughly exercise the code, but current benchmarks provide limited insight into whether agents can construct such evidence themselves. We introduce SpecGuardBench, a benchmark of 291 tasks from 13 real-world repositories, covering 8 types of verification logic such as functional postconditions and interaction protocols. Given a natural language specification, repository context, and a reference implementation, an agent constructs a validation harness comprising executable checks and test scenarios. The checks should reject implementations that violate the specification while accepting those that satisfy it, and the scenarios should thoroughly exercise the code. Across four proprietary and open-weight coding agents, none constructs a harness scored as faithful on the benchmark's implementations for more than 63% of the tasks. The bottleneck lies less in exploring test scenarios than in constructing checks that faithfully reflect the specification. Specifically, the agents show a clear tendency to write checks that mirror the reference implementation rather than the specification, and therefore reject implementations that the specification permits.

Chinese Translation

随着编码代理能力的提升，测试仍然是判断其代码正确性的主要依据。然而，仅通过现有测试并不能证明实现满足其自然语言规范。可靠的验证需要能够忠实捕捉规范的检查，以及能够全面测试代码的场景，但目前的基准测试对代理是否能够自行构建这样的证据提供的洞见有限。我们提出了SpecGuardBench，这是一个包含来自13个真实世界代码库的291个任务的基准，涵盖了8种验证逻辑，如功能后置条件和交互协议。在给定自然语言规范、代码库上下文和参考实现的情况下，代理构建一个包含可执行检查和测试场景的验证工具。检查应拒绝违反规范的实现，同时接受满足规范的实现，场景应全面测试代码。在四个专有和开放权重的编码代理中，没有一个在基准的实现上构建出被评为忠实的验证工具，且超过63%的任务。瓶颈不在于探索测试场景，而在于构建忠实反映规范的检查。具体而言，代理表现出明显的倾向，倾向于编写与参考实现相似的检查，而非与规范相符的检查，因此拒绝了规范允许的实现。

View on arXiv Download PDF AI Translation

cs.CL / 129 / 2610.16933

Estimating Uncertainty of Omnimodal Large Language Models via External Omnimodal Embeddings

通过外部全模态嵌入估计全模态大型语言模型的不确定性

Ghinassi, Iacopo, La, Hy Nguyen Gia, Ong, Luke

Abstract

Recent work has investigated Omni-modal Large Language Models (OLLMs), models which can take multiple modalities and their combinations as input, as a way forward to have systems really capable of interacting with their environment. Uncertainty quantification (UQ) for OLLMs, however, remains an open problem: existing methods largely target at most two modalities at a time, require access to model internals, or re-invocation of the evaluated model. We propose a model-agnostic UQ framework that scores a Chain-of-Thought response entirely through external omni-modal embeddings, measuring the grounding of the reasoning in the multimodal input and its internal coherence. Our contribution is the family of signals and an empirical characterisation of when it is reliable, rather than any single estimator: we find the geometric scores to be largely redundant with one another (mean within-split rank correlation 0.91-0.94) but strongly complementary to answer-dispersion baselines, and their reliability to depend on the generator and the modality. The resulting confidence estimates require no access to logits, hidden states or gradients of the evaluated model and no additional inference from it; they do require segmenting the reasoning trace with an auxiliary model and, for the aggregated score, a labelled calibration set that we show can be as small as 25 examples (although we also show that such a set buys little over fixing one signal a priori). On UNO-Bench, with both a closed and an open model, our scores yield small gains over self-consistency (systematic with the closed generator, mixed with the open one) and are competitive with self-verbalisation and internal-state baselines, especially in open-ended, omni-modal questions where previous state-of-the-art baselines struggle.

Chinese Translation

近期的研究探讨了全模态大型语言模型（Omni-modal Large Language Models, OLLMs），这些模型能够接受多种模态及其组合作为输入，作为实现与环境真正互动的系统的一种前进方向。然而，OLLMs的不确定性量化（Uncertainty Quantification, UQ）仍然是一个未解决的问题：现有的方法大多一次只针对最多两种模态，要求访问模型内部，或者需要重新调用被评估的模型。我们提出了一种模型无关的UQ框架，通过外部全模态嵌入完全评分链式思维（Chain-of-Thought）响应，测量推理在多模态输入中的基础性及其内部一致性。我们的贡献在于信号家族及其可靠性经验特征，而不是任何单一的估计器：我们发现几何评分在很大程度上是相互冗余的（均值分割内秩相关性为0.91-0.94），但与答案分散基线强烈互补，其可靠性依赖于生成器和模态。最终的置信估计不需要访问被评估模型的logits、隐藏状态或梯度，也不需要额外的推理；它们确实需要用辅助模型对推理轨迹进行分段，并且对于聚合评分，需要一个标记的校准集，我们展示该校准集可以小至25个示例（尽管我们也表明，这样的集合在事先固定一个信号时几乎没有额外收益）。在UNO-Bench上，无论是封闭模型还是开放模型，我们的评分在自一致性上获得了小幅提升（与封闭生成器系统化，混合开放生成器），并且在自语言化和内部状态基线中具有竞争力，特别是在开放式全模态问题上，之前的最先进基线表现不佳。

View on arXiv Download PDF AI Translation

cs.CL / 130 / 2610.16952

Lexical Redefinition Reveals Context-Sensitive Instruction Following

词汇重新定义揭示了上下文敏感的指令遵循

Goworek, Roksana, Dubossarsky, Haim

Abstract

How do language models reconcile conflicting contextual evidence within a prompt? We study this question using lexical redefinition, where an explicit instruction specifies how a word should be interpreted and a subsequent sentence provides additional evidence through its usage. Across 870 word pairs, we find that lexical redefinitions are not applied as fixed rules. Instead, the behaviour depends systematically on the relationship between the explicit instruction and the intervening sentence: the contextualised usage can preserve, weaken, or completely reverse the instruction. These results suggest that the interpretation of an explicit instruction remains sensitive to subsequent contextual evidence. Lexical redefinition therefore provides a controlled behavioural paradigm for studying how language models integrate multiple sources of information within a prompt.

Chinese Translation

语言模型如何调和提示中的冲突上下文证据？我们通过词汇重新定义来研究这个问题，其中明确的指令指定了一个词的解释方式，而后续句子通过其用法提供了额外的证据。在870对词汇中，我们发现词汇重新定义并不是作为固定规则应用的。相反，其行为系统性地依赖于明确指令与介入句子之间的关系：上下文化的用法可以保留、削弱或完全逆转该指令。这些结果表明，明确指令的解释仍然对后续的上下文证据敏感。因此，词汇重新定义提供了一种受控的行为范式，用于研究语言模型如何在提示中整合多个信息来源。

View on arXiv Download PDF AI Translation

cs.CL / 131 / 2610.16957

SarcasTraps: Effect of Sarcasm on Turkish Hate Speech Detection

SarcasTraps：讽刺对土耳其仇恨言论检测的影响

Metin, Niyazi Ahmet, Keküllüoğlu, Dilara

Abstract

Standard hate speech detection datasets often emphasize explicit harmful language, but directed harm can also be expressed indirectly, without clearly hateful wording. Sarcasm is one such form of indirect expression, as it can convey harmful intent through mock praise, exaggeration, humor, or apparent politeness. In this paper, we introduce SarcasTraps, a targeted external test set for analyzing the effect of sarcasm on Turkish hate speech detection. SarcasTraps contains 479 entries annotated along two dimensions: hate speech and sarcasm. We train classifiers using BERTurk and XLM-RoBERTa on two existing Turkish hate speech datasets and evaluate them on SarcasTraps, using it strictly as an external test set. We also evaluate GPT-4.1 and Llama-3.3-70B in a zero-shot setting on SarcasTraps. Results show that sarcasm substantially affects hate speech detection. Fine-tuned encoder models lose recall on sarcastic hate, while zero-shot LLMs maintain higher recall but suffer substantial precision drops on sarcastic examples. These findings show that both fine-tuned hate speech detectors and zero-shot LLMs struggle when sarcasm is involved, and SarcasTraps provides quantitative evidence of this challenge through a targeted Turkish test set. Due to the presence of sensitive and hateful language, SarcasTraps will be shared upon request for academic research purposes. Warning: This paper has examples from the dataset which contains offensive and hateful content towards groups of people.

Chinese Translation

标准的仇恨言论检测数据集通常强调明显的有害语言，但针对性的伤害也可以通过间接方式表达，而不需要明确的仇恨措辞。讽刺就是一种间接表达的形式，因为它可以通过虚假的赞美、夸张、幽默或表面的礼貌传达有害的意图。本文介绍了SarcasTraps，这是一个针对性的外部测试集，用于分析讽刺对土耳其仇恨言论检测的影响。SarcasTraps包含479个条目，按照仇恨言论和讽刺两个维度进行标注。我们使用BERTurk和XLM-RoBERTa对两个现有的土耳其仇恨言论数据集进行分类器训练，并在SarcasTraps上进行评估，严格将其作为外部测试集。我们还在SarcasTraps上以零样本设置评估了GPT-4.1和Llama-3.3-70B。结果表明，讽刺对仇恨言论检测有显著影响。经过微调的编码器模型在讽刺仇恨言论上召回率下降，而零样本大语言模型（LLMs）在讽刺示例上保持较高的召回率，但精确度显著下降。这些发现表明，无论是经过微调的仇恨言论检测器还是零样本LLMs，在涉及讽刺时都面临挑战，而SarcasTraps通过一个针对性的土耳其测试集提供了这一挑战的定量证据。由于数据集中包含敏感和仇恨语言，SarcasTraps将在请求下用于学术研究目的。警告：本文包含来自数据集的示例，这些示例包含对某些群体的攻击性和仇恨内容。

View on arXiv Download PDF AI Translation

cs.CL / 132 / 2610.16971

Event Memory and Narrative Dynamics: A Large Language Model Framework for Multidimensional State Spaces

事件记忆与叙事动态：一个多维状态空间的大型语言模型框架

Sariñana, Joshua

Abstract

Studying event memory requires defining events and their relationships within remembered experience. Autobiographical narratives provide evidence of this organization, but their analytical units must first be established. This study introduces a large language model framework for constructing event representations from narratives and modeling their structure and dynamics with excerpt-level traceability. Explicit boundary rules and a feature schema informed by hippocampal models of episodic memory and navigation guide event construction. In a proof-of-concept application to interviews from five participants, the framework constructed 256 event-states. Multiple correspondence analysis embedded these states in feature-based spaces, with analyses focused on a selected 4-D geometry and 10-D and 13-D reference geometries. Consecutive event-states were significantly closer than expected under randomized narrative order. In the selected 4-D ten-basin solution, similarity between basin-state feature profiles corresponded with similarity between one-step routing profiles; context, mechanism, and theme differentiated full-system pathway engagement. A successor representation modeled expected basin-state occupancy across successive steps. Correlations between basin-centroid distances and complete successor-profile distances were positive across all five basin solutions, with exploratory tests supporting correspondence in at least two solutions across all three geometries. These findings connect feature-based event representations with observed narrative transitions and modeled multistep organization. The framework translates hippocampal principles into source-linked event boundaries and state representations, providing an analytical basis for testing neural accounts of event segmentation and predictive organization during naturalistic recall.

Chinese Translation

研究事件记忆需要定义事件及其在记忆体验中的关系。自传叙事提供了这种组织的证据，但其分析单元必须首先建立。本研究引入了一个大型语言模型框架，用于从叙事中构建事件表征，并以摘录级别的可追溯性对其结构和动态进行建模。明确的边界规则和基于海马体模型的情节记忆与导航的特征框架指导事件构建。在对五名参与者的访谈进行的概念验证应用中，该框架构建了256个事件状态。多重对应分析将这些状态嵌入基于特征的空间，分析集中于选定的4维几何形状以及10维和13维参考几何形状。连续的事件状态在随机叙事顺序下显著更接近于预期。在选定的4维十盆地解决方案中，盆地状态特征轮廓之间的相似性与一步路由轮廓之间的相似性相对应；上下文、机制和主题区分了全系统路径的参与。后继表征建模了在连续步骤中预期的盆地状态占用。盆地质心距离与完整后继轮廓距离之间的相关性在所有五个盆地解决方案中均为正相关，探索性测试支持在所有三种几何形状中至少两个解决方案之间的对应关系。这些发现将基于特征的事件表征与观察到的叙事转变和建模的多步骤组织联系起来。该框架将海马体原则转化为源链接的事件边界和状态表征，为测试神经事件分割和自然回忆过程中预测组织的理论提供了分析基础。

View on arXiv Download PDF AI Translation

cs.CL / 133 / 2610.16983

Predicting Mental Health Help-Seeking Behaviour on Reddit: A Survival Analysis of Linguistic and Behavioural Signals

预测Reddit上的心理健康求助行为：语言和行为信号的生存分析

Ahmed, Nisar, Kazmi, Ammara, Raja, Irtaza Bashir, Devaurs, Didier

Abstract

Most prior work on predicting mental health help-seeking treats it as post-hoc classification of users already active in clinical communities, which cannot tell whether someone approaching a transition can be identified in advance. We instead frame help-seeking as a time-to-event problem and ask which Reddit users will eventually seek mental health support, and how imminent that transition is. We build a two-arm cohort: a control frame of 6,337 ordinary users with genuine right-censoring (12.1% transition rate) to answer who will transition, and a clinical frame of 5,023 eventual transitioners to answer how soon. Features come only from posts written before any clinical activity, and user-grouped five-fold cross-validation plus a four-year temporal holdout guard against leakage. We compare TF-IDF, MentalBERT embeddings and behavioural activity features using Cox proportional hazards models, Random Survival Forests and an LSTM sequential model. Behavioural features alone outperform text for predicting who will seek help (C-index 0.797 vs. 0.763), and combining both is best (0.823 under cross-validation, 0.856 on the holdout). The pattern reverses for predicting how soon a known transition will happen: text remains a stable predictor (C-index 0.698) while behavioural features become unstable. Structural engagement patterns therefore carry information about future help-seeking beyond linguistic content, and separating who from when yields a cleaner analysis.

Chinese Translation

以往关于预测心理健康求助的研究大多将其视为对已经活跃于临床社区用户的事后分类，这无法判断在过渡期接近时是否能够提前识别某人。我们将求助行为框架化为一个事件发生时间问题，并探讨哪些Reddit用户最终会寻求心理健康支持，以及这一过渡的紧迫性。我们建立了一个双臂队列：一个包含6,337名普通用户的对照组，具有真实的右删失（转变率为12.1%），以回答谁会发生转变；另一个是5,023名最终转变者的临床组，以回答转变会在多快的时间内发生。特征仅来自于任何临床活动之前撰写的帖子，用户分组的五折交叉验证加上四年的时间保留防止数据泄漏。我们使用Cox比例风险模型、随机生存森林和LSTM序列模型比较TF-IDF、MentalBERT嵌入和行为活动特征。仅使用行为特征在预测谁会寻求帮助时优于文本特征（C-index 0.797 vs. 0.763），而两者结合的效果最佳（交叉验证下为0.823，保留集上为0.856）。在预测已知转变会发生的时间时，模式则相反：文本特征仍然是一个稳定的预测因子（C-index 0.698），而行为特征变得不稳定。因此，结构性参与模式携带着超越语言内容的未来求助行为信息，将“谁”和“何时”分开能够提供更清晰的分析。

View on arXiv Download PDF AI Translation

cs.CL / 134 / 2610.17039

MENTOR: Procedural Shielding of Tool-Using LLM Agents

MENTOR：工具使用的LLM代理的程序性保护

Anand, Ashwani, Chatzi, Ivi, Raha, Ritam, Schmuck, Anne-Kathrin

Abstract

Tool-using LLM agents are increasingly used to automate workflows governed by procedural manuals, such as company rules for booking travel or procuring hardware. These manuals are written in natural language for humans and describe general requirements over such complex workflows. Ensuring compliance of LLM agents with these manuals is difficult, and state-of-the-art techniques use either LLM judges or manually written rules to forbid problematic tool-calls locally at runtime. This is insufficient because (i) only blocking tool calls keeps the agent clueless about how to make correct progress towards fulfilling its task, and (ii) can lead to overly conservative behavior of the agent, as tool calls might only be problematic in certain contexts. To overcome these challenges, we introduce MENTOR, which (i) additionally provides guidance for the agent to make procedural-compliant progress on the task, and (ii) ensures that this compliance is context-aware. We achieve these goals by building a novel history-dependent runtime shield that is obtained from the procedural manual via a multi-step autoformalization pipeline. At its core, MENTOR utilizes a game-based technique to ensure robustness against runtime uncertainty without over-resticting agents' behavior. This yields a formally assisted framework for procedural shielding of LLM agents that improves compliance well beyond existing methods. In particular, MENTOR lifts the compliance of shielded light-weight open-source LLMs above that of non-shielded proprietary models. Since a shield is synthesized once per manual and then serves every task, a MENTOR-shielded agent becomes a viable choice for repeatedly executed workflows, potentially reducing deployment costs when the offline construction cost is amortized over repeated tasks.

Chinese Translation

工具使用的LLM代理越来越多地被用于自动化遵循程序手册的工作流程，例如公司关于预订旅行或采购硬件的规则。这些手册是为人类编写的自然语言文本，描述了复杂工作流程的一般要求。确保LLM代理遵循这些手册是困难的，现有的先进技术要么使用LLM评判者，要么使用手动编写的规则在运行时局部禁止有问题的工具调用。这是不够的，因为（i）仅仅阻止工具调用使代理无法了解如何正确推进以完成其任务，以及（ii）可能导致代理的行为过于保守，因为工具调用在某些上下文中可能才是有问题的。为了克服这些挑战，我们引入了MENTOR，它（i）额外为代理提供指导，使其在任务上能够遵循程序性进展，并且（ii）确保这种合规性是上下文感知的。我们通过构建一种新颖的历史依赖运行时保护机制来实现这些目标，该机制通过多步骤的自动形式化管道从程序手册中获得。在其核心，MENTOR利用基于游戏的技术，确保在不过度限制代理行为的情况下对运行时不确定性具有鲁棒性。这为LLM代理的程序性保护提供了一个形式化辅助框架，显著提高了合规性，超越了现有方法。特别是，MENTOR将受保护的轻量级开源LLM的合规性提升至非受保护专有模型之上。由于每个手册只需合成一次保护，然后可服务于每个任务，因此MENTOR保护的代理成为重复执行工作流程的可行选择，可能在离线构建成本在重复任务中摊销时降低部署成本。

View on arXiv Download PDF AI Translation

cs.CL / 135 / 2610.17040

Beyond Model Bias: Cross-Entropy Based Rankings in Literary and Academic Texts Remain Stable across Small Language Models

超越模型偏差：基于交叉熵的文学与学术文本排名在小型语言模型中保持稳定

Bhattacharjee, Ankit, Hari, Indrasabarna, Bhaumik, Krityapriya, Das, Arka, Sarangi, Piyush

Abstract

While language models provide powerful mechanisms for evaluating textual predictability, it remains unclear whether the relative extents of LM surprisal remain stable across models. This study presents empirical evidence of the invariance of ranking based on cross-entropy related measures (e.g. mean, variance of each data-point), across four distinct small scale transformer-based models, for a given dataset of text sources. We consider two stylistic environments: a corpus of contemporary poetry and a corpus of academic research papers. We demonstrate that mean and variance of cross-entropy maintain robust cross-model rank correlation ($\rho > 0.90$) across both regimes. To test whether this agreement merely reflects generic differences in textual complexity, we compare it against six parameter-free compression baselines. The extent of agreement among language models is found to be substantially higher than that among model-baselines, for most of the measures. Geometric analysis of the structurally constrained academic regime reveals that higher-order moments are sensitive to specific models, although strong agreement persists. The findings suggest that relative extents of textual unpredictability may be independent of the language model they are computed against.

Chinese Translation

虽然语言模型为评估文本可预测性提供了强大的机制，但尚不清楚语言模型（LM）惊讶度的相对程度在不同模型之间是否保持稳定。本研究提供了基于交叉熵相关度量（例如每个数据点的均值和方差）在四种不同的小型变换器模型中保持不变的排名的实证证据，针对一组特定的文本来源数据集。我们考虑了两种风格环境：一组当代诗歌语料库和一组学术研究论文语料库。我们证明了交叉熵的均值和方差在这两种环境中保持了强健的跨模型排名相关性（$ ho > 0.90$）。为了测试这种一致性是否仅反映了文本复杂性的通用差异，我们将其与六个无参数压缩基线进行了比较。发现语言模型之间的一致性在大多数度量中显著高于模型基线之间的一致性。对结构约束的学术环境的几何分析表明，高阶矩对特定模型敏感，尽管强一致性仍然存在。研究结果表明，文本不可预测性的相对程度可能与计算所依据的语言模型无关。

View on arXiv Download PDF AI Translation

cs.CL / 136 / 2610.17042

AD4EDU: Benchmarking University Lecture Audio Description across Accessibility Standards, Metrics, and Learner Preferences

AD4EDU：基于可及性标准、指标和学习者偏好的大学讲座音频描述基准测试

Stylianos, Psychias, Yingqiang, Gao

Abstract

Audio description (AD) aims to enhance the accessibility of visual media for blind and low-vision (BLV) persons. However, in higher education, where slide-based lectures are a default mode of teaching, lecture AD remains largely underexplored. The main reason is that existing production guidelines, evaluation metrics, and benchmark datasets are oriented toward entertainment content and do not transfer to lectures, where ADs must complement the lecturer's speech, resolve deictic references, and fit within narrow pauses. In this work, we propose AD4Edu, consisting of an AD benchmark, AD4Edu-Bench, and a rule-diagnosable evaluation metric, AD4Edu-Eval, for generating and evaluating ADs in higher education. Specifically, we consolidated established accessibility guidance, including WCAG 2.2, the DCMP Description Key, and NCAM's STEM description practices. We further collected silver reference ADs, judge-ordered preference pairs, and pairwise preferences from four BLV raters, and post-trained models with direct preference optimization (DPO) for detailed analysis. Our findings indicate that while the metric reliably detects violations of the standard, ADs that comply with it do not necessarily fulfill the subjective preferences of BLV persons, most notably on description length, underscoring the necessity of active participation of BLV persons in co-designing inclusive AI solutions.

Chinese Translation

音频描述（AD）旨在增强盲人和低视力（BLV）人士对视觉媒体的可及性。然而，在高等教育中，幻灯片讲座是默认的教学模式，讲座音频描述仍然在很大程度上未被充分探索。主要原因是现有的制作指南、评估指标和基准数据集主要面向娱乐内容，无法有效应用于讲座中，而讲座音频描述必须补充讲师的讲话、解决指示性引用，并适应狭窄的停顿。在本研究中，我们提出了AD4Edu，包括一个音频描述基准AD4Edu-Bench和一个可规则诊断的评估指标AD4Edu-Eval，用于生成和评估高等教育中的音频描述。具体而言，我们整合了既有的可及性指导，包括WCAG 2.2、DCMP描述关键和NCAM的STEM描述实践。我们进一步收集了银参考音频描述、评审排序的偏好对以及来自四位BLV评审员的成对偏好，并通过直接偏好优化（DPO）对模型进行了后训练以进行详细分析。我们的研究结果表明，尽管该指标可靠地检测到标准的违规行为，但符合该标准的音频描述并不一定满足BLV人士的主观偏好，特别是在描述长度方面，这突显了BLV人士在共同设计包容性人工智能解决方案中的积极参与的必要性。

View on arXiv Download PDF AI Translation

cs.CL / 137 / 2610.17077

UZHCL_for_the_win at Touche: Encoder Fine-Tuning, Synthetic Augmentation, and an Underpowered Encoder-LLM Comparison

UZHCL_for_the_win 在 Touche：编码器微调、合成增强与低功率编码器-LLM 比较

Stylianos, Psychias

Abstract

Online discussions on platforms like Reddit are a productive testbed for fallacy detection: short texts, broad topical range, and the full inventory of informal fallacies. The Touche 2026 lab at CLEF formalises this as a shared task on a Reddit-derived corpus that combines fallacy detection with argument-scheme classification. This paper describes our participation in all three subtasks (ST1 binary fallacy detection, ST2 eight-way fallacy classification, ST3 four-way argument-scheme classification) and both evaluation tracks. Our submitted systems are fine-tuned RoBERTa-large encoders with task-specific heads, augmented with a small batch of LLM-generated synthetic contrastive pairs; for ST2 and ST3 on the enhanced track we additionally added confidence-filtered pseudo-labelled examples from the unlabelled pool. A wider backbone comparison spanning additional encoders (DeBERTa-v3, ModernBERT) and an LLM track (four Qwen variants from 3B to 32B, zero-shot through CoT-LoRA) showed no validated benefit over RoBERTa-large at this corpus scale, with 5-fold variance wide enough that we call the comparison underpowered rather than settled. We also characterise how much of the enhanced track's lift reflects organiser-rewritten inputs carrying label information rather than genuine argument clarification. On the official Touche 2026 test set our systems ranked first on both base-track classification cells (ST2 and ST3) and second on both binary-detection (ST1) tracks, with the two enhanced-track classification cells topped by cascaded and multitask systems. We close with the rare-class limitation on ST3 and the most useful follow-ups. Code: https://github.com/psychias/fallacy_detection_clef_2026

Chinese Translation

在线讨论平台如 Reddit 是谬误检测的有效试验场：短文本、广泛的主题范围以及各种非正式谬误的完整清单。CLEF 的 Touche 2026 实验室将此形式化为一个共享任务，基于 Reddit 衍生的语料库，结合谬误检测与论证方案分类。本文描述了我们在所有三个子任务（ST1 二元谬误检测、ST2 八类谬误分类、ST3 四类论证方案分类）及两个评估轨道中的参与。我们提交的系统是经过微调的 RoBERTa-large 编码器，配备特定任务的头部，并使用一小批 LLM 生成的合成对比样本进行增强；对于 ST2 和 ST3 的增强轨道，我们还从未标记池中添加了经过置信度过滤的伪标记示例。跨越额外编码器（DeBERTa-v3、ModernBERT）和 LLM 轨道（四个 Qwen 变体，从 3B 到 32B，零-shot 通过 CoT-LoRA）的更广泛基础比较显示，在该语料库规模下，未能验证 RoBERTa-large 的优势，5 倍方差足够大，以至于我们称这一比较为低功率而非已定论。我们还描述了增强轨道的提升在多大程度上反映了组织者重写的输入携带标签信息，而非真正的论证澄清。在官方 Touche 2026 测试集中，我们的系统在两个基础轨道分类单元（ST2 和 ST3）中排名第一，在两个二元检测（ST1）轨道中排名第二，而两个增强轨道分类单元则由级联和多任务系统占据。最后，我们讨论了 ST3 上的稀有类别限制及最有用的后续研究。代码链接： https://github.com/psychias/fallacy_detection_clef_2026

View on arXiv Download PDF AI Translation

cs.CL / 138 / 2610.17113

Calibrated Typed Decisions for Entity, Relation, and Event Extraction

用于实体、关系和事件提取的校准类型决策

Wang, Yang, Xiao, Hanlin, Ma, Yanan, Xiao, Chenghao, Lin, Chenghua

Abstract

Information extraction models often perform well on the datasets they are trained on, but transferring to new domains and annotation guidelines remains challenging. Supporting user-specified types alone does not resolve this problem. We present CalOn, a unified 0.6B-parameter encoder for entity, relation, and event extraction across datasets and schemas. The model jointly encodes text and type definitions, then directly identifies entity spans, relations between spans, and event triggers and arguments. Its extraction decisions consider the requested types together, using their combined evidence to identify relevant candidates before assigning a type. This gives all three tasks a common probabilistic formulation while keeping every extracted mention anchored to the input text. We train on diverse sources, and supplement supervised learning with a reward that encourages correct extractions and penalises incorrect ones. Across 40 benchmarks, CalOn outperforms the evaluated baselines on most benchmarks. It leads all reported relation extraction benchmarks and achieves the highest average F1 for event triggers and arguments with gold triggers. The same model also performs strongly on aspect sentiment triplet extraction and pronoun resolution through its span-pair formulation. A correction for the probability bias introduced during training improves confidence estimates for entities, relations, and event arguments after calibration, with little change in F1.

Chinese Translation

信息提取模型通常在其训练数据集上表现良好，但在转移到新领域和注释指南时仍然面临挑战。仅支持用户指定的类型并不能解决这个问题。我们提出了CalOn，一个统一的0.6B参数编码器，用于跨数据集和模式的实体、关系和事件提取。该模型联合编码文本和类型定义，然后直接识别实体范围、范围之间的关系以及事件触发器和参数。它的提取决策同时考虑请求的类型，利用它们的综合证据来识别相关候选项，然后再分配类型。这为所有三个任务提供了一个共同的概率公式，同时保持每个提取的提及与输入文本的锚定。我们在多样化的来源上进行训练，并通过奖励机制补充监督学习，以鼓励正确的提取并惩罚错误的提取。在40个基准测试中，CalOn在大多数基准上超越了评估的基线。它在所有报告的关系提取基准中领先，并在带有金标准触发器的事件触发器和参数上实现了最高的平均F1分数。相同的模型在方面情感三元组提取和代词解析方面也表现强劲，通过其范围对配方。对训练过程中引入的概率偏差的修正提高了校准后实体、关系和事件参数的置信度估计，F1变化不大。

View on arXiv Download PDF AI Translation

cs.CL / 139 / 2610.17117

Gluing Local Contexts into Global Meaning: A Sheaf-Theoretic Decomposition of Transformer Representations

将局部语境粘合为全球意义：一种基于层叠理论的变换器表示分解

Grant, Bryce, Wang, Peng

Abstract

Existing interpretability tools do not test whether a representation keeps the same content when only the phrasing changes. We treat each paraphrase pair as a local witness that two activations should agree and build a cellular sheaf over the paraphrase graph; its Laplacian $L_\mathcal{F}$ answers the question on a frozen model with no training. The kernel $H^0$ is a content subspace: at 20 dimensions it beats the preserved subspace of the concept eraser LEACE on held-out CounterFact fact retrieval across eight architectures by a mean $+17.1$\,pp, and constraining all paraphrases of a fact jointly adds a further $+6.5$\,pp. The spectral complement concentrates causal influence: ablating the most paraphrase-variable coordinates, a coordinate-wise proxy for it, changes model outputs $5.6$ to $26.5\times$ more than ablating coordinates of matched per-coordinate variability, across nine models from 124M to 13B parameters. Stacking layers into a grid creates $b_1 = N(L-1)$ algebraic 1-cycles, and the resulting model-derived harmonic mass varies by nearly three orders of magnitude across six architectures and ranks them as their steering fragility does (Spearman $\rho = 0.90$, exact one-sided permutation $p = 0.014$). The same paraphrase graph yields the content subspace, the influence directions, and the fragility measure. Project page: https://cwru-aism.github.io/gluing-lc-page/

Chinese Translation

现有的可解释性工具并未测试在仅改变措辞的情况下，表示是否保持相同的内容。我们将每对同义改写视为局部证据，表明两个激活应当一致，并在同义改写图上构建一个细胞层叠；其拉普拉斯算子 $L_ ext{F}$ 在一个冻结模型上回答了这个问题，而无需训练。核 $H^0$ 是一个内容子空间：在20个维度下，它在八种架构的保留子空间概念擦除器 LEACE 上的保留事实检索中超越了平均 $+17.1$ 个百分点，并且共同约束一个事实的所有同义改写进一步增加了 $+6.5$ 个百分点。谱补充集中因果影响：消除最具同义改写变异性的坐标，作为其坐标级代理，模型输出的变化比消除匹配的每坐标变异性的坐标多 $5.6$ 到 $26.5$ 倍，涵盖从 124M 到 13B 参数的九个模型。将层叠成网格创建了 $b_1 = N(L-1)$ 代数 1-循环，结果模型导出的谐波质量在六种架构中变化近三个数量级，并且其排名与其引导脆弱性一致（Spearman $ ho = 0.90$，精确单侧置换 $p = 0.014$）。同一同义改写图产生内容子空间、影响方向和脆弱性度量。项目页面：https://cwru-aism.github.io/gluing-lc-page/

View on arXiv Download PDF AI Translation

cs.CL / 140 / 2610.17124

D2: High-Capability Full-Duplex Spoken Language Models from Half-Duplex Backbones

D2：从半双工骨干网络生成高能力全双工口语语言模型

Chang, Cheng-Kuang, Liu, Alexander H., Kingsbury, Brian, Kuehne, Hilde, Feris, Rogerio, Thomas, Samuel, Glass, James, Chang, Kai-Wei

Abstract

Full-duplex spoken language models (FD-SLMs) support simultaneous listening and speaking, but often lag behind half-duplex spoken language models (HD-SLMs) in general capabilities. We propose Dialogue-to-Duplex (D2), a framework that converts pretrained HD-SLMs into high-capability FD-SLMs at fine temporal granularity. D2 adapts the original audio encoder to streaming through decoder-free representation distillation and learns full-duplex interaction through supervised fine-tuning. We systematically study the effects of chunk size on full-duplex interaction, general capabilities, and inference computation. We apply D2 to two distinct HD-SLM architectures, Qwen3-Omni and LLaMA-Omni 2. At a 12.5Hz update rate, Qwen3-Omni-D2 achieves 99.2% turn-taking success with 521-ms latency, 99.5% interruption success, and a state-of-the-art VoiceBench score of 70.1.

Chinese Translation

全双工口语语言模型（FD-SLMs）支持同时听和说，但在一般能力上往往落后于半双工口语语言模型（HD-SLMs）。我们提出了对话到双工（Dialogue-to-Duplex，D2）框架，该框架能够将预训练的HD-SLMs转换为高能力的FD-SLMs，具有精细的时间粒度。D2通过无解码器的表示蒸馏将原始音频编码器适配为流式处理，并通过监督微调学习全双工交互。我们系统地研究了块大小对全双工交互、一般能力和推理计算的影响。我们将D2应用于两种不同的HD-SLM架构，Qwen3-Omni和LLaMA-Omni 2。在12.5Hz的更新率下，Qwen3-Omni-D2实现了99.2%的轮流成功率，延迟为521毫秒，99.5%的打断成功率，以及70.1的最先进的VoiceBench得分。

View on arXiv Download PDF AI Translation

cs.CL / 141 / 2610.17125

ToolHACE: A Dataset for Detecting Hallucinations in Tool-Based LLM Responses

ToolHACE：用于检测工具基础大语言模型响应中的幻觉的数据集

Ivanov, Nikolay, Pletenev, Sergey, Rykov, Elisei, Frolov, Anatolii, Monogov, Bogdan, Potekhin, Alexander, Marina, Maria, Moskovskiy, Daniil, Moskvoretskii, Viktor, Panchenko, Alexander, Alimova, Ilseyar

Abstract

Tool-augmented language models ground their responses in external tool outputs, improving reliability beyond parametric knowledge alone. However, even when the correct tool is called and returns accurate information, models may still contradict such outputs, omit critical details, or generate unsupported claims. Existing hallucination benchmarks for tool-augmented dialogue primarily focus on the tool-calling stage, evaluating whether the correct tool and arguments are selected, while overlooking errors that arise during post-tool response generation. To address this gap, we introduce ToolHACE, a span-level benchmark for hallucination detection in tool-augmented LLM responses, covering five fine-grained hallucination categories. Through extensive experiments with modern hallucination detectors, we show that post-tool-call hallucinations are widespread and remain challenging to detect. Our results further demonstrate that effective detection requires training on specialized tool-augmented corpora, while transfer from tool-agnostic hallucination datasets performs poorly in this setting. These findings highlight the need for dedicated benchmarks and models for reliable hallucination detection in tool-augmented language systems.

Chinese Translation

工具增强语言模型将其响应基于外部工具输出，从而提高了可靠性，超越了仅依赖参数知识的局限。然而，即使在调用正确工具并返回准确的信息时，模型仍可能与这些输出相矛盾、遗漏关键细节或生成不支持的主张。现有的针对工具增强对话的幻觉基准主要集中在工具调用阶段，评估是否选择了正确的工具和参数，而忽视了在工具后响应生成过程中出现的错误。为了解决这一问题，我们提出了ToolHACE，这是一个用于检测工具增强大语言模型响应中幻觉的跨度级基准，涵盖五个细致的幻觉类别。通过与现代幻觉检测器进行广泛实验，我们展示了工具调用后的幻觉普遍存在，并且仍然难以检测。我们的结果进一步表明，有效的检测需要在专门的工具增强语料库上进行训练，而从与工具无关的幻觉数据集中迁移在这种情况下表现不佳。这些发现突显了在工具增强语言系统中进行可靠幻觉检测所需的专用基准和模型。

View on arXiv Download PDF AI Translation

cs.CL / 142 / 2610.17184

DuoDiscover: Scaling Autonomous Discovery Beyond the Evaluation Bottleneck

DuoDiscover：超越评估瓶颈的自主发现扩展

Agarwal, Shubham, Gupta, Akshat, Heddes, Mike, Soylu, Dilara, Meng, Rui, Kon, Patrick Tser Jern, Chen, Jiefeng, Cemri, Mert, Liu, Shu, Gonzalez, Abe, Jain, Akanksha, Mishra, Bhavana Dalvi, Song, Linfeng, Meshram, Vilobh, Tang, Xiangru, Yoon, Jinsung, Parameswaran, Aditya, Stoica, Ion, Mirrokni, Vahab, Gokturk, Burak, Ranganathan, Parthasarathy, Pfister, Tomas, Peng, Nanyun, Li, Chun-Liang

Abstract

Evaluation has been central to advances in AI, and it is equally critical to autonomous discovery. Yet scaling autonomous discovery to real-world engineering challenges faces a fundamental bottleneck: the cost of evaluating progress. In real settings, a single true evaluation, such as a full benchmark or serving trace, can take hours, so a search can afford only a handful of true evaluations. We present DuoDiscover, a search system that jointly discovers what to optimize and how to evaluate progress efficiently, through three mechanisms. First, each worker in a fleet of coding agents co-discovers an optimization target and a fast, tailored proxy evaluator, which we call a yardstick. The agent then hill-climbs the yardstick rather than the true evaluation. Second, since search can exploit gaps between a yardstick and the true objective, fidelity tests continually recheck each yardstick's ranking and invalidate any that drift. Third, the fleet is bottom-up, with no central decomposition of the problem. A lightweight orchestrator spawns and terminates agents, briefs running ones, and seeds new agents with others' confirmed work. DuoDiscover enables autonomous optimization from software to chip architecture. We demonstrate DuoDiscover on two repository-level benchmarks and four GPU and TPU kernel suites covering hundreds of kernels. On SWE-fficiency, it doubles the average speedup of human experts. On SOL-Exec, it ranks first on 121 of 235 kernels and in the top three on 212. On the heavily optimized vLLM and SQLite engines, DuoDiscover achieves 1.92$\times$ and 1.38$\times$ gains over the base code. On a CPU microarchitecture optimization task with evaluations of up to 28 hours, DuoDiscover surpasses both AlphaEvolve-based ArchAgent v2 and the best human design, with 1.65$\times$ ArchAgent v2's gain in 16.7 days versus more than 40 days, advancing the frontier of autonomous chip design.

Chinese Translation

评估在人工智能的进步中占据了核心地位，对于自主发现同样至关重要。然而，将自主发现扩展到现实世界的工程挑战面临一个根本性的瓶颈：评估进展的成本。在实际环境中，单次真实评估，如完整基准测试或服务跟踪，可能需要数小时，因此搜索只能承受少量真实评估。我们提出了DuoDiscover，这是一种搜索系统，通过三种机制共同发现优化目标和高效评估进展的方法。首先，编码代理队列中的每个工作者共同发现一个优化目标和一个快速、量身定制的代理评估器，我们称之为“标尺”。然后，代理在标尺上进行爬升，而不是在真实评估上。其次，由于搜索可以利用标尺与真实目标之间的差距，保真测试不断重新检查每个标尺的排名，并使任何漂移的标尺失效。第三，代理队列是自下而上的，没有问题的中央分解。一个轻量级的协调者生成和终止代理，简要介绍正在运行的代理，并用其他代理确认的工作为新代理提供种子。DuoDiscover使得从软件到芯片架构的自主优化成为可能。我们在两个仓库级基准测试和四个覆盖数百个内核的GPU和TPU内核套件上演示了DuoDiscover。在SWE-fficiency上，它将人类专家的平均加速倍增。在SOL-Exec上，它在235个内核中排名前121，并在212个内核中位列前三。在经过高度优化的vLLM和SQLite引擎上，DuoDiscover分别实现了1.92$ imes$和1.38$ imes$的增益。在一个CPU微架构优化任务中，评估时间长达28小时，DuoDiscover超越了基于AlphaEvolve的ArchAgent v2和最佳人类设计，在16.7天内实现了1.65$ imes$的增益，而超过40天才完成，推动了自主芯片设计的前沿。

View on arXiv Download PDF AI Translation

cs.CL / 143 / 2610.17189

Soft Teacher Forcing

软教师强制

Fahes, Mohammad, Schmid, Cordelia, Arnab, Anurag

Abstract

Soft tokens have so far been used to replace discrete chain-of-thought in latent reasoning methods. These approaches typically require multi-stage training and, in some cases, auxiliary modules, yet they still underperform supervised fine-tuning (SFT) on triples. In this paper, we ask whether soft tokens can be used to improve SFT itself, instead of as an avenue for latent reasoning. We introduce Soft Teacher Forcing (Soft-TF), a simple training algorithm that combines teacher-forced embeddings with soft tokens through random convex combinations. The soft tokens are computed in a single parallel forward pass over the ground-truth sequence. Soft-TF adds no parameters, no external model and no curriculum, and leaves inference unchanged. Experiments show that Soft-TF improves over SFT on GSM8K, MATH, MathQA and AQuA, by +2.1 to +4.1 points on GSM8K across seven Gemma, Llama and Mistral models.

Chinese Translation

迄今为止，软令牌已被用于替代潜在推理方法中的离散思维链。这些方法通常需要多阶段训练，并且在某些情况下需要辅助模块，但它们在 <输入, 思维链, 输出> 三元组上的表现仍然不及监督微调（SFT）。在本文中，我们探讨软令牌是否可以用于改善 SFT 本身，而不是作为潜在推理的途径。我们提出了软教师强制（Soft Teacher Forcing, Soft-TF），这是一种简单的训练算法，通过随机凸组合将教师强制嵌入与软令牌结合。软令牌是在对真实序列进行单次并行前向传递时计算的。Soft-TF 不增加任何参数、外部模型或课程，并且不改变推理过程。实验表明，Soft-TF 在 GSM8K、MATH、MathQA 和 AQuA 上的表现优于 SFT，在 GSM8K 上跨七个 Gemma、Llama 和 Mistral 模型提高了 2.1 到 4.1 分。

View on arXiv Download PDF AI Translation

cs.CL / 144 / 2610.17195

Welcome to the Age of Agent Ads: Using On-Policy Attacks to Steer Agentic Decisions

欢迎来到代理广告的时代：利用在线攻击引导代理决策

Davidson, Tim R., Šakota, Marija, Joseph, Joshua, West, Robert

Abstract

People increasingly delegate option-ranking tasks (e.g., shopping, job search, travel) to AI agents, which creates incentives to steer agentic decisions via "agent ads." To defend against such steering attacks, people count on models' alignment training and platforms' terms of service, and consequently the most effective existing attacks are designed to trigger off-policy behaviors. Exposing a new attack surface, we introduce on-policy attacks, which steer agentic decisions without pushing individually aligned behaviors off policy. Our example attack consists of two steps: To remain compliant with platforms' terms of service, we first inject a truthful negation about a fictitious concept ("Does not contain blizyester") into a product description. This triggers an agent to search the web ("What is blizyester?"), which surfaces planted websites full of fabricated information associating the fictitious concept with harmful implications ("Blizyester linked to dementia"). Trained to keep users safe from harm, the agent penalizes every option missing the truthful negation. Evaluating our attack across four real-world listing datasets, we find some frontier models exposed, enabling an adversarial actor to penalize competitors without violating existing platform-listing guidelines. We discuss the broader challenge of regulating advertising aimed at AI agents and provide recommendations to model developers and platform providers.

Chinese Translation

人们越来越多地将选项排序任务（例如购物、求职、旅行）委托给人工智能代理，这为通过“代理广告”引导代理决策创造了动机。为了防御这种引导攻击，人们依赖于模型的对齐训练和平台的服务条款，因此现有的最有效攻击主要设计为触发离线行为。我们揭示了一种新的攻击面，提出了在线攻击，这种攻击在不推动个体对齐行为离线的情况下引导代理决策。我们的示例攻击包括两个步骤：为了遵守平台的服务条款，我们首先在产品描述中注入关于一个虚构概念的真实否定（“不包含 blizyester”）。这会触发代理进行网络搜索（“什么是 blizyester？”），从而显示出充满虚假信息的网站，这些信息将虚构概念与有害影响关联起来（“blizyester 与痴呆症相关”）。为了确保用户安全，代理会惩罚每个缺少真实否定的选项。我们在四个真实世界的列表数据集中评估了我们的攻击，发现一些前沿模型暴露出来，使得对手能够在不违反现有平台列表指南的情况下惩罚竞争对手。我们讨论了针对人工智能代理的广告监管的更广泛挑战，并向模型开发者和平台提供者提出了建议。

View on arXiv Download PDF AI Translation

cs.CL / 145 / 2610.17243

From Conceptual EneMonIE to an Executable Backend: A Modular Architecture for Energy Information Extractionand Grounded Question Answering

从概念性 EneMonIE 到可执行后端：一种用于能源信息提取和基于语境的问题回答的模块化架构

Lesage, Janik, Lô, Khadim, Dessureault, and Jean-Sébastien

Abstract

The original EneMonIE proposal describes a Web-based semantic system for continuously monitoring energy trends through targeted information extraction from heterogeneous online media. This article presents EneMonIE-Backend, an implementation-derived redesign that converts that conceptual pipeline into a modular, executable, and application-independent backend. The architecture separates five concerns: data sources, acquisition and normalization, semantic processing, knowledge organization, and persistent storage. It operationalizes configurable connectors for RSS feeds, URL collections, arXiv, social-media JSON files, and open-data CSV files; it also introduces question-driven Web search and scraping. Relevant documents pass through energy categorization, named entity recognition, temporal and event extraction, source sensitive sentiment analysis, and extractive summarization. The resulting semantic representations support entity co-occurrence graphs, cross-document interlinking, cluster-level information fusion, and relational persistence in SQLite. Three grounded question-answering paths are exposed at the backend level: live Web evidence, the stored corpus, and selected configured resources. Answer generation uses a language model when configured and otherwise falls back to deterministic extractive output. Unlike the original proposal, the presented architecture deliberately excludes the user-interface and server-side application layers, allowing the semantic core to be embedded in different applications. A deterministic smoke validation confirms relevance filtering, persistence, clustering, evidence retrieval, and document deletion. The work provides a reproducible architectural basis for future evaluation on larger multilingual energy corpora.

Chinese Translation

原始的 EneMonIE 提案描述了一个基于网络的语义系统，通过从异构在线媒体中有针对性地提取信息，持续监测能源趋势。本文提出了 EneMonIE-Backend，这是一种基于实现的重新设计，将该概念管道转换为一个模块化、可执行且与应用无关的后端架构。该架构分离了五个关注点：数据源、获取与规范化、语义处理、知识组织和持久存储。它实现了可配置的连接器，用于 RSS 源、URL 集合、arXiv、社交媒体 JSON 文件和开放数据 CSV 文件；同时引入了基于问题的网络搜索和抓取。相关文档经过能源分类、命名实体识别、时间和事件提取、源敏感情感分析以及抽取式摘要处理。生成的语义表示支持实体共现图、跨文档互联、集群级信息融合和在 SQLite 中的关系持久性。在后端层面，提供了三条基于语境的问题回答路径：实时网络证据、存储语料库和选定的配置资源。答案生成在配置时使用语言模型，否则回退到确定性的抽取输出。与原始提案不同，所呈现的架构故意排除了用户界面和服务器端应用层，允许语义核心嵌入到不同的应用中。确定性的烟雾验证确认了相关性过滤、持久性、聚类、证据检索和文档删除。该工作为未来在更大多语言能源语料库上的评估提供了可重复的架构基础。

View on arXiv Download PDF AI Translation

cs.CL / 146 / 2610.17278

Nufo: A Benchmark for Ewe Automatic Speech Recognition

Nufo：Ewe自动语音识别基准

Keli, Christ Kekeli, Sanogo, Steve, Sebastiano, Patrice, Aldenawi, Hamad, Whetten, Ryan

Abstract

Automatic speech recognition (ASR) remains challenging for low-resource languages, where the scarcity of transcribed speech limits the development of accurate systems. This challenge is particularly acute for African languages such as Ewe, which has millions of speakers but remains underrepresented in publicly available ASR resources and benchmarks. In this work, we consolidate, normalize, and partition existing Ewe speech resources into a standardized 174-hour Ewe ASR benchmark, the largest to date to our knowledge. We fine-tune and evaluate pretrained speech models, including Whisper and AfriHuBERT, to establish strong baselines on our benchmark. Whisper-medium achieves a word error rate (WER) of 37.98%, while AfriHuBERT achieves 39.70% on the test set. We further provide a qualitative analysis of common recognition errors. We release the benchmark under the name Nufo, derived from the Ewe word for "talk", accompanied by models and evaluation resources, to facilitate future research on Ewe and other underrepresented African languages.

Chinese Translation

自动语音识别（ASR）在低资源语言中仍然面临挑战，因为转录语音的稀缺限制了准确系统的发展。这一挑战在非洲语言中尤为突出，例如Ewe语言，尽管有数百万讲者，但在公开可用的ASR资源和基准中仍然代表性不足。在本研究中，我们整合、规范并划分现有的Ewe语音资源，形成一个标准化的174小时Ewe ASR基准，这是迄今为止我们所知的最大规模基准。我们对预训练的语音模型进行微调和评估，包括Whisper和AfriHuBERT，以在我们的基准上建立强有力的基线。Whisper-medium在测试集上实现了37.98%的字错误率（WER），而AfriHuBERT则达到了39.70%。我们还提供了对常见识别错误的定性分析。我们以Nufo的名称发布该基准，Nufo源自Ewe语中的“谈话”一词，并附带模型和评估资源，以促进未来对Ewe及其他代表性不足的非洲语言的研究。

View on arXiv Download PDF AI Translation

cs.CL / 147 / 2610.17309

StartLux-Decision: AutoThink Decision Models with Reinforcement Learning for Calibrated Deliberative Decisions

StartLux-Decision：基于强化学习的自动思考决策模型用于校准的深思熟虑决策

Zhuang, Xialie, Su, Zunhai, Pan, Jia, Wong, Ngai, Chen, Guanhua, Kuo, Chuan-Wei

Abstract

We introduce StartLux-Decision, decision models from 0.8B to 27B parameters that combine efficient single-pass prediction with adaptive, confidence-guided reasoning. The Instant models use bidirectional attention in full-attention layers to improve decision accuracy while keeping single-pass inference. The AutoThink models invoke explicit reasoning for uncertain decisions and answer confident cases directly. A three-stage training pipeline establishes shared decision capabilities, specializes the two variants, and equips AutoThink with reasoning through reasoning supervision and reinforcement learning. We introduce Reinforcement Learning for Calibrated Deliberative Decisions (RLCD$^2$), which uses a frozen model and proper scoring rules to reward improvements in decision probability quality and applies a soft penalty to overlong reasoning traces. RLCD$^2$ raises the share of reasoning traces that terminate within the generation limit by up to 48 percentage points and cuts average reasoning length by up to 65%, while largely preserving decision accuracy. On the public subset of Decision Index 0.3, the Instant models outperform the strongest evaluated comparable-scale open-weight decision models by 7.4-16.9 points, and the 27B AutoThink model reaches 73.4. We also introduce DecisionPivot, a programmatically verifiable benchmark of computational decisions built from contrast pairs that differ in a single answer-changing detail. On DecisionPivot, AutoThink outperforms its Instant counterparts by 11.5-18.4 points in pair accuracy and leads among evaluated systems. The models run on vLLM, llama.cpp and MLX, with quantized variants for lower memory use.

Chinese Translation

我们介绍了StartLux-Decision，这是一种参数范围从0.8B到27B的决策模型，结合了高效的单次预测与自适应的信心引导推理。Instant模型在全注意力层中使用双向注意力，以提高决策准确性，同时保持单次推理。AutoThink模型针对不确定的决策进行明确推理，并直接回答自信的案例。一个三阶段的训练流程建立了共享的决策能力，专门化了这两种变体，并通过推理监督和强化学习为AutoThink提供推理能力。我们引入了用于校准深思熟虑决策的强化学习（Reinforcement Learning for Calibrated Deliberative Decisions, RLCD$^2$），该方法使用冻结模型和适当的评分规则来奖励决策概率质量的改善，并对过长的推理轨迹施加软惩罚。RLCD$^2$将终止在生成限制内的推理轨迹的比例提高了最多48个百分点，并将平均推理长度减少了最多65%，同时在很大程度上保持了决策准确性。在Decision Index 0.3的公共子集上，Instant模型的表现超过了最强的可比规模开放权重决策模型7.4-16.9分，而27B的AutoThink模型达到了73.4。我们还引入了DecisionPivot，这是一个可以程序验证的计算决策基准，由在单一答案改变细节上不同的对比对构建。在DecisionPivot上，AutoThink在对比准确性上比其Instant对应模型高出11.5-18.4分，并在评估系统中领先。这些模型在vLLM、llama.cpp和MLX上运行，并提供量化变体以降低内存使用。

View on arXiv Download PDF AI Translation

cs.CL / 148 / 2610.17318

From Sampled Evidence to Radiology Reports: Objectives, Constraints, and Realization

从采样证据到放射学报告：目标、约束与实现

Zhao, Kun, Tang, Haoteng, Lin, Chenghua, Zhan, Liang

Abstract

Sampling multiple reports creates an inference problem: how should sampled evidence be converted into a final decision, and does a better structured decision yield a better report? We study these questions through a framework that separates the decision signal, scoring objective, and feasible decision set, followed by text realization. On MIMIC-CXR and IU-Xray, controlled comparisons across 13 generator configurations per benchmark under grouped development/test evaluation reveal three findings. First, changing the objective under fixed evidence and rule capacity changes predictions and their held-out metric trade-offs. Second, relaxing candidate constraints has objective-dependent effects under shared signals and frozen scores. Diagnostics connect the IU macro-oriented assertion increase to near-zero label thresholds and show that limited candidate coverage coexists with selection headroom. Third, a controlled realization study with SFT-Qwen3-VL-8B shows that decision-level gains can be attenuated, amplified, or reversed. Change decomposition reveals that realization can remove incorrect assertions while losing correct findings, so target fidelity and reference accuracy need not improve together. Finding gains also do not consistently improve reference-based report quality. These results motivate separate evaluation of decision rules, candidate constraints, and realized reports.

Chinese Translation

多个报告的采样产生了一个推断问题：如何将采样证据转换为最终决策，结构更好的决策是否能产生更好的报告？我们通过一个框架研究这些问题，该框架将决策信号、评分目标和可行决策集分开，随后进行文本实现。在 MIMIC-CXR 和 IU-Xray 数据集上，对每个基准的 13 种生成器配置进行的受控比较揭示了三个发现。首先，在固定证据和规则容量下，改变目标会改变预测及其保留指标的权衡。其次，放宽候选约束在共享信号和冻结评分下具有目标依赖性效果。诊断结果将 IU 宏观导向的断言增加与接近零的标签阈值联系起来，并显示有限的候选覆盖与选择余地共存。第三，使用 SFT-Qwen3-VL-8B 进行的受控实现研究表明，决策级别的收益可能会减弱、放大或逆转。变化分解揭示实现可以去除不正确的断言，同时丢失正确的发现，因此目标保真度和参考准确性不必同时提高。发现收益也并不总是能一致改善基于参考的报告质量。这些结果促使我们对决策规则、候选约束和实现报告进行单独评估。

View on arXiv Download PDF AI Translation

cs.CL / 149 / 2610.17324

Thinking Outside the Bots: Human Ideas in AI's Semantic Blind Spots Are Not Inferior

超越机器人的思维：人工智能的语义盲区中的人类思想并不逊色

Moon, Kibum, Johnson, Dan R., Kushlev, Kostadin, Lira, Benjamin, Viskontas, Indre, Kaufman, James C., Beaty, Roger E., Duckworth, Angela L., Green, Adam E.

Abstract

AI-generated outputs are frequently perceived as highly creative, but also tend to be more similar to each other (homogenized) than human ideas. Does the human heterogeneity that extends beyond AI reflect valuable thinking that AI misses? Alternatively, is greater human heterogeneity driven by lower-quality ideas and less effective thinking that AI avoids? We developed Expansion of Divergent Generative Expression (EDGE) to map AI's semantic "blind spots" by assessing how strongly/weakly individual human ideas were represented in a task-matched AI semantic comparison set. Specifically, we compared human-generated vs. AI-generated ideas across three open-ended tasks: college admissions essays, alternative uses for everyday objects, and interpersonal advice. Across all three tasks, human responses extended into semantic regions sparsely covered by AI outputs. These distinctively human responses showed no meaningful deficits in creativity, quality, or effectiveness, were associated with equal-or-better academic performance, and with credibly superior ratings and achievement for some outcomes. These ideas were thus not merely poor-performing variation; they performed as well as or better than ideas that overlapped more strongly with AI ideas. More distinctive human ideas tended to receive higher creativity ratings, but this relationship was weak or reversed for AI ideas. Taken together, these results suggest that differentiation among human-generated ideas does not reflect poor thinking that AI can safely omit, and instead may afford both individual and population-level advantages.

Chinese Translation

人工智能生成的输出通常被认为具有高度创造性，但往往比人类思想更趋同（同质化）。人类的异质性是否反映了人工智能所忽视的有价值的思维？或者，更大的人工异质性是否源于低质量的想法和人工智能所回避的低效思维？我们开发了发散生成表达扩展（Expansion of Divergent Generative Expression，EDGE）来映射人工智能的语义“盲区”，通过评估个体人类思想在任务匹配的人工智能语义比较集中表现的强度/弱度。具体而言，我们比较了人类生成的想法与人工智能生成的想法在三个开放式任务中的表现：大学入学论文、日常物品的替代用途和人际建议。在所有三个任务中，人类的回应延伸到了人工智能输出稀疏覆盖的语义区域。这些独特的人类回应在创造性、质量或有效性上没有表现出有意义的缺陷，与同等或更好的学术表现相关联，并且在某些结果上获得了可信的更高评分和成就。因此，这些想法并不仅仅是表现不佳的变体；它们的表现与那些与人工智能想法重叠更强的想法相当或更好。更具独特性的人类想法往往获得更高的创造性评分，但对于人工智能想法，这种关系较弱或相反。综合来看，这些结果表明，人类生成的想法之间的差异并不反映人工智能可以安全忽略的低劣思维，而是可能为个体和群体层面带来优势。

View on arXiv Download PDF AI Translation

cs.CL / 150 / 2610.17327

The Curse of Knowledge in Role-Playing Language Models

角色扮演语言模型中的知识诅咒

Zhang, Xiao

Abstract

Role-playing language models speak as characters in games and agent simulations, where the system records what each character can hear, see and understand. We test whether characters respect these boundaries in 480 scenes, each with a counterfactual twin that changes only the hidden fact, so mentions of the twin's fact measure guessing. Across 11 open-weight models, characters use content from speech they cannot understand or hear at 67 and 72 percentage points (pp) above this baseline, although most models correctly state what the character perceives. This is a curse of knowledge. The leak extends to hidden actions and private messages and changes where characters search. Five of six proprietary models rarely leak, but compliance in strong models depends on input format. Changing only the scene layout or character description raises the rate in two otherwise compliant open-weight models from under 10 to over 60 pp, and combined changes raise two proprietary models from near zero to 13 and 21~pp. Writing the line as the scene's author leaks nearly as much as role-play, and thinking rarely helps. Giving models only what the character observes stops the leak without hiding what it may report.

Chinese Translation

角色扮演语言模型以游戏和代理模拟中的角色身份进行对话，其中系统记录每个角色能够听到、看到和理解的内容。我们在480个场景中测试角色是否遵循这些边界，每个场景都有一个反事实的双胞胎，仅改变隐藏的事实，因此提及双胞胎的事实可以测量猜测。在11个开放权重模型中，角色使用来自他们无法理解或听到的言语内容的比例分别比基线高出67和72个百分点（pp），尽管大多数模型正确陈述了角色所感知的内容。这是一种知识的诅咒。信息泄露扩展到隐藏的动作和私人消息，并改变了角色的搜索方式。在六个专有模型中，有五个很少发生泄露，但强模型的合规性取决于输入格式。仅改变场景布局或角色描述将两个原本合规的开放权重模型的泄露率从10%以下提高到60%以上，而组合变化使两个专有模型的泄露率从接近零提高到13和21 pp。作为场景作者撰写的台词几乎与角色扮演泄露的程度相当，而思考很少能有所帮助。仅提供角色观察到的信息可以阻止泄露，而不隐藏其可能报告的内容。

View on arXiv Download PDF AI Translation

cs.CL / 151 / 2610.17358

claimcheck: Measuring and Improving Specification Faithfulness

claimcheck：测量和提高规范忠实度

Amster, Martin, An, One, Baker, Holly, Graciolli, Fernanda, Henniger, Simon, Amin, Nada

Abstract

Formal verification proves that a program satisfies its specification, but says nothing about whether the specification captures what the author meant. As LLMs increasingly write both code and specifications, a gap between the author's natural-language requirement and the specification becomes a live risk. A machine-checked proof of a property narrower than the one requested is worthless but passes every check. Closing this gap is solving the spec formalization problem: given a natural-language requirement, produce a formal artifact whose meaning matches it. To measure the gap, we curate an end-to-end benchmark for spec formalization spanning four formal languages: Dafny contracts and Lean 4, Isabelle and Rocq theorem statements. Each of its 3,234 rows pairs a natural-language requirement with a formal artifact and a label established by execution, by proof, or by the benchmark it is drawn from. A third of the rows are meaning-preserving rewrites of the artifact that keep the faithful label, so a judge cannot score by detecting that a row was edited. We introduce claimcheck, a tool that detects incorrect formalizations: the formal artifact is translated back into plain English by a model that never sees the requirement, and a separate model, the judge, decides whether the informalization and the requirement express the same guarantee. On all four languages it gains 8.2 to 9.8 points of macro recall over a single call shown the requirement and the artifact, and 4.2 to 6.0 over that call handed the same list of meaning-preserving rewrites. Showing the judge the artifact as well costs 1.4 to 4.5 points. Across four judges that also see the artifact, the gain shrinks as the judge gets stronger, to within a point of the informed single call for the strongest.

Chinese Translation

形式验证证明程序满足其规范，但并未说明该规范是否捕捉了作者的本意。随着大型语言模型（LLMs）越来越多地编写代码和规范，作者的自然语言需求与规范之间的差距成为一个实际风险。一个比请求的属性更狭窄的机器检查证明是毫无价值的，但却通过了所有检查。弥合这一差距就是解决规范形式化问题：给定一个自然语言需求，生成一个其含义与之匹配的形式化工件。为了测量这一差距，我们策划了一个涵盖四种形式语言的端到端规范形式化基准：Dafny 合同和 Lean 4、Isabelle 和 Rocq 定理陈述。其 3,234 行中的每一行都将一个自然语言需求与一个形式化工件以及通过执行、证明或基准所建立的标签配对。三分之一的行是对工件的意义保持重写，保持忠实标签，因此评审者无法通过检测某一行被编辑来评分。我们引入了 claimcheck，一个检测不正确形式化的工具：该形式化工件被一个从未看到需求的模型翻译回普通英语，另一个模型，即评审者，决定非正式化与需求是否表达相同的保证。在所有四种语言中，它在单次调用中获得了 8.2 到 9.8 的宏召回率，相较于展示需求和工件的情况，以及 4.2 到 6.0 的召回率，相较于展示相同的意义保持重写的情况。向评审者展示工件的成本为 1.4 到 4.5 分。在四个也看到工件的评审者中，随着评审者能力的增强，增益缩小至最强评审者的知情单次调用的一个点以内。

View on arXiv Download PDF AI Translation

cs.CL / 152 / 2610.17363

UMCR: Universal Medical Concept Recognition through Guideline Integration and Agentic Linking

UMCR：通过指南集成和代理链接实现的通用医学概念识别

Lamarche, Fabrice, Zouaq, Amal

Abstract

Medical Concept Recognition is the task of extracting and grounding spans in medical text to entries in a standard ontology. Current methods are generally developed towards a single ontology and source text type (clinical notes, drug labels, etc.) The central challenge is therefore to build methods that adapt to any target ontology without requiring ontology-specific retraining. More recent approaches like REAL and AutoPCR make direct use of Large Language Models and can theoretically adapt to any ontology choice but suffer from a lack of generalization. We propose UMCR, a pipeline that allows extraction and linking to any target ontology using any source text variety, utilizing direct guideline integration and an agentic linking module. We evaluate our method on the recent MedPath benchmark, covering a wide variety of medical text types and vocabularies, and show that our method achieves strong performance across these tasks. This ability unlocks interoperability in the healthcare domain, allowing for extraction and alignment between different medical standards. Our method also reduces dependence on parametric LLM memory, thanks to improved interaction with the target ontology compared to other proposed approaches.

Chinese Translation

医学概念识别是从医学文本中提取并将其与标准本体中的条目进行关联的任务。目前的方法通常针对单一的本体和源文本类型（如临床笔记、药物标签等）进行开发。因此，中心挑战在于构建能够适应任何目标本体而无需特定于本体的重新训练的方法。最近的一些方法，如REAL和AutoPCR，直接利用大型语言模型（Large Language Models），理论上可以适应任何本体选择，但在泛化能力上存在不足。我们提出了UMCR，一个允许使用任何源文本类型提取和链接到任何目标本体的流程，利用直接的指南集成和代理链接模块。我们在最近的MedPath基准上评估了我们的方法，涵盖了多种医学文本类型和词汇，并显示出我们的方法在这些任务中表现出色。这一能力解锁了医疗领域的互操作性，允许不同医学标准之间的提取和对齐。与其他提出的方法相比，我们的方法还减少了对参数化大型语言模型内存的依赖，得益于与目标本体的改进交互。

View on arXiv Download PDF AI Translation

cs.CL / 153 / 2610.17375

PIG: Parallel Inconsistency Guidance for Accelerating Diffusion Language Models

PIG：加速扩散语言模型的并行不一致性引导

Fruytier, Quentin, Xue, Anton, Shakkottai, Sanjay, Sanghavi, Sujay

Abstract

Diffusion language models accelerate generation by decoding multiple masked positions per forward pass. However, accuracy is known to drop when parallel decoding trajectories deviate from their sequential counterparts, which are slow and expensive "oracles" that adhere to the joint distribution of language. In this paper, we propose Parallel Inconsistency Guidance (PIG), a new paradigm that uses a trained auxiliary model to directly estimate the inconsistency between parallel and sequential generation from a single model forward pass. During inference, masked positions are traversed by decreasing confidence, and tokens are added to the parallel decode set one-at-a-time until the total estimated inconsistency between the parallel and sequential generation exceeds a threshold. Surprisingly, it suffices to train a lightweight ~2M parameter auxiliary model on a small mixture drawn from OpenCoder, GSM8K-train, MATH-train, and MBPP-train. For LLaDA-8B Instruct and NemotronDiffusion-3B Instruct, PIG extends the accuracy-TPF frontier on GSM8K, MATH500, MBPP, HumanEval, and MMLU, when compared to the industry standard Fast-dLLM and measures favorably on LLaDA-8B Instruct against DAPD, KLASS, EB-sampler, and other learned unmasking policies.

Chinese Translation

扩散语言模型通过在每次前向传播中解码多个被掩蔽的位置来加速生成。然而，当并行解码轨迹偏离其顺序对应轨迹时，准确性会下降，而顺序解码轨迹是缓慢且昂贵的“神谕”，遵循语言的联合分布。本文提出了并行不一致性引导（Parallel Inconsistency Guidance，PIG），这是一种新范式，利用训练好的辅助模型直接估计来自单个模型前向传播的并行生成与顺序生成之间的不一致性。在推理过程中，通过降低置信度遍历被掩蔽的位置，并逐个将标记添加到并行解码集，直到并行生成与顺序生成之间的总估计不一致性超过阈值。令人惊讶的是，仅需在从 OpenCoder、GSM8K-train、MATH-train 和 MBPP-train 中抽取的小混合数据上训练一个轻量级约 200 万参数的辅助模型。对于 LLaDA-8B Instruct 和 NemotronDiffusion-3B Instruct，PIG 在与行业标准 Fast-dLLM 的比较中扩展了在 GSM8K、MATH500、MBPP、HumanEval 和 MMLU 上的准确性-TPF 前沿，并在与 DAPD、KLASS、EB-sampler 及其他学习解掩蔽策略的比较中，在 LLaDA-8B Instruct 上表现良好。

View on arXiv Download PDF AI Translation

cs.CL / 154 / 2610.17402

Evaluating Temporal Reasoning in Large Language Models: Event Temporal Relation Extraction with Allen's Interval Algebra, Logical Consistency, and Relative Time in Narratives

评估大型语言模型中的时间推理：基于艾伦区间代数的事件时间关系提取、逻辑一致性与叙事中的相对时间

Sakhawat, Adib, Sadab, Fardeen, Fahim, Tamjid Hasan, Mahmud, Multazam, Lihan, Tahsan Ferdous, Chowdhury, Md Isfak Iqbal, Jamil, Nafis Ahnaf, Shahriar, Md. Atik, Shahriare, Sieam, Syed, Wasi Omar, Khan, M Safwan Hasan, Hassan, Tabib

Abstract

Understanding a narrative means knowing how every event relates in time to every other, yet large language models (LLMs) are mostly tested on isolated event pairs or coarse labels. We ask five LLMs to reconstruct the complete network of Allen's 13 interval relations for 360 English narratives whose events are placed relative to one another rather than by dates, and evaluate the outputs under a pre-registered protocol for agreement with our gold annotations, converse consistency and logical satisfiability. End to end, the best model clearly beats trivial baselines on macro-F1, our primary metric, yet under our pre-registered parser no model matches the accuracy of always answering "before", which the skewed labels inflate. Most of that accuracy gap is a parsing artifact: formatting errors in large matrices cost whole rows, and a post hoc parser that recovers the determined cells of malformed rows raises coverage from 0.60 to 0.84 and closes the gap to the baseline for the best model. On the cells they do answer, models recover precedence and containment but rarely overlap, and they fail most on flashbacks, where both the narration and the event list present the later event first. Logical consistency varies widely yet says little about agreement with gold: enforcing it raises accuracy mainly by filling empty cells, and is beaten there by the trivial control of guessing the majority label. Reliable structured output and robust handling of relative time are prerequisites for temporal reasoning over whole narratives.

Chinese Translation

理解叙事意味着知道每个事件与其他事件在时间上的关系，然而大型语言模型（LLMs）主要是在孤立的事件对或粗略标签上进行测试。我们要求五个LLM重建360个英语叙事中艾伦的13个区间关系的完整网络，这些事件是相对彼此而非按日期排列的，并在预注册的协议下评估输出与我们的金标准注释、一致性和逻辑可满足性的一致性。总体而言，表现最佳的模型在宏观F1（我们的主要指标）上明显优于简单基线，但在我们的预注册解析器下，没有模型能够匹配始终回答“之前”的准确性，而这种偏差标签则夸大了这一点。大部分准确性差距是解析的伪影：大型矩阵中的格式错误导致整行丢失，而一个事后解析器能够恢复畸形行的确定单元，将覆盖率从0.60提高到0.84，并缩小了最佳模型与基线之间的差距。在它们回答的单元上，模型能够恢复优先关系和包含关系，但重叠情况很少，且在闪回场景中表现最差，因为叙述和事件列表都将后续事件置于前面。逻辑一致性差异很大，但对金标准的一致性几乎没有影响：强制一致性主要通过填补空单元提高准确性，而在这方面被简单的多数标签猜测所超越。可靠的结构化输出和对相对时间的稳健处理是对整个叙事进行时间推理的前提条件。

View on arXiv Download PDF AI Translation

cs.CL / 155 / 2610.17456

Compositional RL: Evaluating and Training LLMs for Collaborative Reasoning

组合强化学习：评估和训练大型语言模型以进行协作推理

Feng, Shangbin, Tsvetkov, Yulia, Kautz, Jan, Acuna, David

Abstract

Language models are often evaluated alone but increasingly deployed together, routed between, made to debate, ensembled, or collaboratively trained. For a model inside a system, what matters most is not how well it performs on its own, but how much the system gains from its presence. We call this a model's compositional strength and formalize it using the Shapley value of a cooperative game over coalitions of a model pool, and use a model's leave-one-out contribution to the full coalition as a tractable proxy, requiring n+1 rather than 2^n system evaluations. Evaluating individual vs. compositional strengths across five model pools, twelve collaboration methods, and twelve datasets, we find them largely uncorrelated: in 12 of 15 (pool, method-family) settings, individual strength does not translate into compositional advantages, and the coupling is weakest among general-purpose models that practitioners are most likely to deploy. Standalone benchmarks are therefore often poor proxies for models' value in collaborative systems. We then propose compositional reinforcement learning (C-RL), which trains a policy inside a live multi-model system and rewards the outcome produced by the system as a whole, rather than the policy's correctness in isolation. The leave-one-out baseline in Shapley values cancels under a group policy gradient, so this needs no additional rollout cost beyond RL on the collaborative system's outcome and never requires running a coalition without the policy. Across three collaboration protocols and eight tasks spanning reasoning, coding, and scientific discovery, C-RL produces the strongest collaborator in 21 of 24 settings, lifting collaborative performance by +4.0 to +5.4 pts on average (~3x the gain of baselines), up to +13.1 pts.

Chinese Translation

语言模型通常是单独评估的，但越来越多地被一起部署，进行路由、辩论、集成或协同训练。对于系统中的模型而言，最重要的不是它自身的表现如何，而是系统从其存在中获得了多少收益。我们称之为模型的组合强度，并通过对模型池的联盟进行合作博弈的夏普利值来形式化这一概念，并使用模型在完整联盟中的留一法贡献作为可处理的代理，这样只需进行 n+1 次而非 2^n 次系统评估。在对五个模型池、十二种协作方法和十二个数据集评估个体与组合强度时，我们发现它们之间大体上不相关：在 15 种 (池，方法族) 设置中有 12 种情况下，个体强度并未转化为组合优势，而这种耦合在实践中最可能部署的一般性模型中最弱。因此，独立基准往往无法有效代表模型在协作系统中的价值。随后，我们提出了组合强化学习 (C-RL)，该方法在实时多模型系统中训练策略，并奖励系统整体产生的结果，而不是单独奖励策略的正确性。在夏普利值中的留一法基线在群体策略梯度下会被抵消，因此这不需要超出在协作系统结果上进行 RL 的额外展开成本，并且从不需要在没有策略的情况下运行联盟。在覆盖推理、编码和科学发现的三种协作协议和八个任务中，C-RL 在 24 种设置中产生了 21 种最强的协作者，平均提升协作性能 4.0 到 5.4 分（约为基线增益的 3 倍），最高可达 13.1 分。

View on arXiv Download PDF AI Translation

cs.CL / 156 / 2610.17461

Distilling Text LMs into Speech-to-Speech LMs by Matching Internal Text Predictions

通过匹配内部文本预测将文本语言模型蒸馏为语音到语音语言模型

Chou, Ju-Chieh, Zhou, Jiawei, Livescu, Karen

Abstract

Text language models (LMs) acquire broad linguistic and factual capabilities through large-scale pretraining, while comparable capabilities remain difficult to learn from speech alone. Recent work has shown that knowledge distillation from pretrained text LMs offers a promising way to transfer their linguistic and factual capabilities to speech models. However, existing distillation approaches have largely focused on speech-input, text-output models. This idea has yet to be explored for {\it speech-to-speech} models, which operate on both speech input and speech output. In this work we introduce a speech-to-speech pretraining method that creates an internal text interface for knowledge transfer while preserving the speech-only input and output. The model first consumes speech context without its transcription, switches to text prediction, and then uses the predicted text to lead future speech generation word by word. It also provides an interface for knowledge distillation: the internal text predictions can be directly distilled from the base text LM. To the best of our knowledge, this is the first application of text LM distillation to a speech-to-speech LM. Under matched training conditions, the proposed sequence construction substantially improves the generative perplexity of spoken continuations over speech-only and interleaved speech-text models, approaching the performance of a text-leading inner-monologue model with external ASR prompt transcription. More importantly, distillation improves factual performance under both text and speech evaluation, indicating better retention of knowledge from the base text language model.

Chinese Translation

文本语言模型（LM）通过大规模预训练获得广泛的语言和事实能力，而仅从语音中学习相应能力仍然困难。近期的研究表明，从预训练文本语言模型中进行知识蒸馏是一种将其语言和事实能力转移到语音模型的有希望的方法。然而，现有的蒸馏方法主要集中在语音输入、文本输出模型上。对于同时处理语音输入和语音输出的{ extit{语音到语音}}模型，这一思路尚未得到探索。在本研究中，我们提出了一种语音到语音的预训练方法，该方法创建了一个内部文本接口用于知识转移，同时保留仅语音的输入和输出。该模型首先在没有转录的情况下处理语音上下文，切换到文本预测，然后利用预测的文本逐字引导未来的语音生成。它还提供了一个知识蒸馏的接口：内部文本预测可以直接从基础文本语言模型中蒸馏。根据我们所知，这是文本语言模型蒸馏首次应用于语音到语音语言模型。在匹配的训练条件下，所提出的序列构建显著提高了语音延续的生成困惑度，相较于仅语音和交错语音-文本模型，接近于具有外部自动语音识别（ASR）提示转录的文本主导内心独白模型的性能。更重要的是，蒸馏在文本和语音评估下都改善了事实性能，表明从基础文本语言模型中更好地保留了知识。

View on arXiv Download PDF AI Translation

cs.CL / 157 / 2610.17538

Everyone Was Right: Lexical Representations in Language Models Are Both Abstract and Holistic

每个人都是对的：语言模型中的词汇表示既是抽象的也是整体的

Houghton, Zachary Nicholas, Zhou, Yu, Pluth, Dan, Gurbani, Vijay K., Kapatsinski, Vsevolod

Abstract

Lexical representations can encode both abstract information and information specific to particular contexts. We investigate how these forms of information are organized in language models by decomposing contextual word representations into item, class, and word-class interaction components. Across six pretrained transformers and two linguistic categories, we find that the interaction is consistently present, grows across layers, and contains information associated with both item and class. Unlike the nearly orthogonal item and class effects, the interaction overlaps with both and encodes class information beyond the shared class effect. An interaction also emerges in artificial-language learners even when the input contains only additive structure. Together, these findings show that lexical representations combine abstract information with structure specific to particular word-class combinations.

Chinese Translation

词汇表示可以编码抽象信息和特定于特定上下文的信息。我们通过将上下文词表示分解为项目、类别和词类交互成分，研究这些信息形式在语言模型中的组织方式。在六个预训练的变换器和两个语言类别中，我们发现交互始终存在，随着层数的增加而增长，并包含与项目和类别相关的信息。与几乎正交的项目和类别效应不同，交互与两者重叠，并编码超出共享类别效应的类别信息。即使输入仅包含加性结构，人工语言学习者中也会出现交互。综合来看，这些发现表明，词汇表示结合了抽象信息和特定于特定词类组合的结构。

View on arXiv Download PDF AI Translation

cs.CL / 158 / 2610.17544

Many Steps into One: Reasoning Contraction and the Limits of Safe Memory Reuse for LLM Agents

多步归一：推理收缩与大型语言模型代理安全内存重用的限制

Zhou, Jianghong, Rong, Hao, Lin, Chen, Zhang, Daoan

Abstract

Thought templates, skill libraries, agent memories and caches turn many reasoning steps into one. We call this reasoning contraction. A contracted memory saves reasoning steps and errors until its facts change, then quietly answers wrong. Most systems trust memories by counting successes and failures. We prove that such counters are badly tuned change detectors: a symmetric counter notices a partial change only after a power of its false-alarm horizon, and an unclipped one may never notice it. On five LLMs answering multi-hop questions over edited facts, ExpeL- and ACE-style counters serve 7.0-13.5x more stale answers than certified memories given the same information, and on code tasks with an exact verifier 35-73x more. Safety also has a price. When facts change and each memory is judged only by its own verified outcomes, one number, the certification load, sets how much reasoning any safe memory can skip, and under Zipf demand it yields a scaling law. Our recipe weighs each memory's failures by its own likelihood ratio, certifies its utility against live reasoning, and restarts the certificate when facts change, with the likelihood-ratio detector as the safety net for unannounced changes; for large edits it also invalidates the memory.

Chinese Translation

思维模板、技能库、代理记忆和缓存将多个推理步骤合并为一个。我们称之为推理收缩。收缩的记忆在其事实变化之前保存推理步骤和错误，然后悄然给出错误的答案。大多数系统通过计算成功和失败来信任记忆。我们证明，这种计数器是调节不当的变化检测器：对称计数器仅在其虚警边界的幂之后注意到部分变化，而未剪辑的计数器可能永远不会注意到它。在五个大型语言模型（LLM）回答编辑事实的多跳问题时，ExpeL 和 ACE 风格的计数器在给定相同信息的情况下提供的过时答案比经过认证的记忆多出 7.0-13.5 倍，而在具有精确验证器的代码任务中则多出 35-73 倍。安全性也有代价。当事实变化时，每个记忆仅根据其自身的验证结果进行评估，一个数字，即认证负载，决定了任何安全记忆可以跳过的推理量，并且在 Zipf 需求下，它产生了一个规模法则。我们的方案通过每个记忆的失败与其自身的似然比加权，针对实时推理认证其效用，并在事实变化时重新启动证书，利用似然比检测器作为未通知变化的安全网；对于大规模编辑，它还会使记忆失效。

View on arXiv Download PDF AI Translation

cs.CL / 159 / 2610.17560

EdgeWise: Bringing Long-Horizon Agents to the Edge via Selective Cloud Delegation

EdgeWise：通过选择性云委托将长时间跨度的智能体带到边缘

Shi, Zeru, Jiang, Jindong, Hachiuma, Ryo, Fu, Yonggan, Lee, Byung-Kwan, Belcak, Peter, Muralidharan, Saurav, Agarwal, Ayush, Tang, Ruixiang, Dantrey, Ambrish, Molchanov, Pavlo, Patney, Anjul, Radhakrishnan, Subhashree

Abstract

Complex and long-horizon agent tasks often rely on frontier models deployed in the cloud to achieve strong performance. However, using these models for every step of a long-horizon trajectory incurs substantial inference cost, even though not all steps may require the same level of model capability. In this paper, we find that allocating the LLM to only a small subset of key steps can achieve performance comparable to full LLM execution while significantly reducing inference cost. Motivated by this observation, we propose EdgeWise, an edge-cloud collaboration framework that deploys an SLM on the local device as the primary solver and selectively delegates critical steps to a cloud LLM when stronger model capacity is needed. To enable autonomous delegation without additional models at inference time, we introduce a step-level RL method that jointly optimizes task solving and delegation, with fine-grained credit assignment for individual delegation decisions. We also introduce an external judge-based delegation strategy and use it to systematically investigate key design choices in edge-cloud collaboration, including when to delegate, how delegation decisions should be made, what context should be shared, and how frequently the cloud LLM should be invoked. Experiments show that EdgeWise matches or outperforms pure-LLM execution while reducing inference cost up to 62% on SWE-Rebench and 34% on Terminal-Bench 2.1. We further deploy EdgeWise on a local workstation, demonstrating comparable cost savings in a practical edge-cloud setting.

Chinese Translation

复杂且长时间跨度的智能体任务通常依赖于部署在云端的前沿模型以实现强大的性能。然而，在长时间跨度的轨迹的每一步中使用这些模型会产生可观的推理成本，即使并非所有步骤都需要相同水平的模型能力。在本文中，我们发现将大型语言模型（LLM）仅分配给少数关键步骤可以实现与完全执行LLM相当的性能，同时显著降低推理成本。基于这一观察，我们提出了EdgeWise，一个边缘-云协作框架，该框架在本地设备上部署一个小型语言模型（SLM）作为主要求解器，并在需要更强模型能力时选择性地将关键步骤委托给云端LLM。为了实现自主委托而无需在推理时引入额外模型，我们引入了一种步骤级强化学习（RL）方法，该方法联合优化任务求解和委托，并对个别委托决策进行细粒度的信用分配。我们还引入了一种基于外部评判者的委托策略，并利用该策略系统地研究边缘-云协作中的关键设计选择，包括何时委托、如何做出委托决策、应共享何种上下文以及云端LLM应多频繁被调用。实验表明，EdgeWise的性能与纯LLM执行相当或更优，同时在SWE-Rebench上将推理成本降低了62%，在Terminal-Bench 2.1上降低了34%。我们进一步在本地工作站上部署EdgeWise，展示了在实际边缘-云环境中可实现的相当成本节约。

View on arXiv Download PDF AI Translation

cs.CL / 160 / 2610.17576

Dissociating Answerability from Extraction in LLM Question Answering over Radiology Reports

将回答能力与提取分离：大语言模型在放射学报告中的问答

Sinha, Priyanshu, McDonald, Kelsey

Abstract

Answering questions from radiology reports involves two decisions: whether the report contains an answer and which span gives it. A false answer presents an unsupported claim as fact; an over-abstention hides what the report says. Yet RadQA results are reported as a single F1 that merges both. We propose an evaluation method that separates them, reporting false-answer and over-abstention rates beside F1 and splitting F1 changes into extraction and answerability terms. We apply it to 13 LLMs on the RadQA test set, zero-shot and with three cumulative demonstration sets, the last adding an unanswerable example. Models with similar F1 differed more than twofold in false-answer rate. The unanswerable example changed when models answered, not how well they extracted: 12 of 13 models abstained more often, and three-quarters of the answers they withdrew had F1 $\geq$ 0.5. Within models, false answers showed no clear association with extraction quality, but acting on the decision itself helped: keeping an answer only when enough other models also answered lowered the zero-shot false-answer rate by 4 points without a detectable F1 change. Because the two decisions can change independently, we suggest radiology QA evaluations report them separately, not through F1 alone.

Chinese Translation

从放射学报告中回答问题涉及两个决策：报告是否包含答案以及哪个范围提供了答案。错误的答案将不支持的主张呈现为事实；过度的放弃则隐藏了报告所述内容。然而，RadQA的结果通常以单一的F1值报告，这两个方面被合并在一起。我们提出了一种评估方法，将这两者分开，报告错误答案率和过度放弃率，并将F1的变化分解为提取和回答能力两个部分。我们将其应用于13个大语言模型（LLMs）在RadQA测试集上的表现，包括零-shot和三个累积演示集，最后一个演示集添加了一个无法回答的例子。具有相似F1值的模型在错误答案率上差异超过两倍。无法回答的例子在模型回答时发生变化，而不是提取质量的好坏：13个模型中有12个更频繁地选择放弃，而他们撤回的答案中有四分之三的F1值大于或等于0.5。在模型内部，错误答案与提取质量之间没有明显关联，但对决策本身的处理有所帮助：仅在足够多的其他模型也回答时保留答案，可以将零-shot错误答案率降低4个百分点，而F1值没有明显变化。由于这两个决策可以独立变化，我们建议放射学质量评估报告应分别报告这两个方面，而不仅仅通过F1值。

View on arXiv Download PDF AI Translation

cs.CL / 161 / 2610.17597

TDR: Tagalog Diacritic Restoration Dataset

TDR：塔加洛语变音符号恢复数据集

De Leon, Ivan Yuri, Fermin, Lanz Railey, Africa, Zachary Aldridge, Flores, Richell Isaiah

Abstract

Homographs are words that share identical spelling but differ in meaning and pronunciation. In Tagalog, these distinctions are disambiguated in writing through tuldik (diacritical marks), which are rarely preserved in contemporary Tagalog text. This scarcity of diacritics limits diacritic restoration, which is useful for downstream tasks such as prosody selection for text-to-speech synthesis. To address this gap, this study introduces TDR, an annotated corpus of 40 frequent Tagalog homographs constructed from news text. It also introduces a collocation-driven retrieval method to expand the dataset and mitigate severe class imbalance among minor variants. Finally, we show that TDR supplies a capability absent from pretrained models, and that this capability can be learned by small, inexpensive models as well as by fine-tuned LLMs.

Chinese Translation

同形异义词是指拼写相同但意义和发音不同的词。在塔加洛语中，这些区别通过 tuldik（变音符号）在书写中得以消解，但在当代塔加洛语文本中，这些变音符号很少被保留。这种变音符号的稀缺限制了变音符号的恢复，而变音符号恢复对于下游任务如文本到语音合成的韵律选择非常有用。为了解决这一问题，本研究引入了 TDR，一个由新闻文本构建的包含 40 个常见塔加洛语同形异义词的注释语料库。研究还提出了一种基于搭配的检索方法，以扩展数据集并缓解次要变体之间的严重类别不平衡。最后，我们展示了 TDR 提供了预训练模型所缺乏的能力，并且这种能力可以通过小型、低成本的模型以及经过微调的 LLMs 学习。

View on arXiv Download PDF AI Translation

cs.CL / 162 / 2610.17607

Facts Don't Speak Louder than Words: The Behavioral Essence of Long-CoT Distillation

事实并不比言语更响亮：长链思维蒸馏的行为本质

Yu, Yongcan, Liang, Jian, He, Lingxiao, Guo, Kuangpu, Wang, Yanbo, Lu, Shuo, Wang, Meng, He, Ran

Abstract

While long chain-of-thought (CoT) distillation significantly improves the reasoning abilities of language models, its underlying mechanism remains poorly understood. Current pipelines generally rely on rejection sampling to gather reasoning traces across diverse prompts, assuming that factual correctness and broad problem coverage are indispensable. Surprisingly, we find that training exclusively on flawed reasoning trajectories or a restricted prompt set yields nearly comparable performance. Driven by this counterintuitive finding, we propose the Behavioral Hypothesis: the essence of Long-CoT distillation is behavioral alignment with structured reasoning patterns, rather than factual knowledge memorization. We then validate this by demonstrating substantial reasoning gains even when models are fine-tuned exclusively on previously mastered problems-effectively isolating behavioral alignment from novel knowledge injection. Having established that reasoning patterns are the true bottleneck, we investigate how to learn them across varying levels of complexity optimally. We reveal a critical interplay: while the recent probability-based loss excels at learning simple patterns, the standard cross-entropy loss is essential for scaling to complex CoTs. We believe these insights demystify CoT distillation and provide a principled foundation for the future development of reasoning models. Code is available on https://github.com/yuyongcan/Long-CoT-Distillation.

Chinese Translation

尽管长链思维（CoT）蒸馏显著提升了语言模型的推理能力，但其潜在机制仍然不甚明了。目前的流程通常依赖于拒绝采样来收集不同提示下的推理轨迹，假设事实正确性和广泛的问题覆盖是不可或缺的。令人惊讶的是，我们发现仅在有缺陷的推理轨迹或有限的提示集上进行训练，几乎可以获得相当的性能。基于这一反直觉的发现，我们提出了行为假设：长链思维蒸馏的本质在于与结构化推理模式的行为对齐，而非事实知识的记忆。随后，我们通过展示即使在仅对先前掌握的问题进行微调时，模型仍能获得显著的推理提升，从而验证了这一点——有效地将行为对齐与新知识注入隔离开来。在确立推理模式是真正瓶颈后，我们探讨如何在不同复杂度水平上最优地学习这些模式。我们揭示了一个关键的相互作用：尽管最近的基于概率的损失在学习简单模式方面表现出色，但标准的交叉熵损失对于扩展到复杂的链思维至关重要。我们相信这些见解揭示了链思维蒸馏的奥秘，并为未来推理模型的发展提供了原则性基础。代码可在 https://github.com/yuyongcan/Long-CoT-Distillation 获取。

View on arXiv Download PDF AI Translation

cs.CL / 163 / 2610.17619

When Does an ASR Error Matter? Task-Role Diagnostics for Tool-Using Voice Agents

ASR错误何时重要？工具使用语音代理的任务角色诊断

Wang, Xiaolong, Su, Shijie, Xiong, Naqi

Abstract

In cascaded voice agents, ASR errors can have markedly different downstream consequences: a small transcription error may leave a tool call unchanged or alter its intent or arguments. Conventional word error rate (WER), however, does not capture these differences. We introduce SpeechEvalX, a paired diagnostic protocol that evaluates how task roles affected by ASR errors influence structured tool calls. Controlled edits are matched by source utterance, edit count, and operation, while task-role labels are compared with lexical severity, BERT cosine distance, Semantic-WER, and binary criticality. Across 594 source utterances, joint call accuracy falls from 82.3% with reference transcripts to 71.9-74.9% across three ASR front ends. Number and negation edits cause 21.9-25.7 percentage-point losses, versus 1.4-2.9 points for control edits. Detailed task-role labels improve source-grouped ROC-AUC from 0.73 to 0.78 over binary criticality, showing that downstream impact depends strongly on the task role affected.

Chinese Translation

在级联语音代理中，ASR错误可能导致显著不同的下游后果：一个小的转录错误可能不会改变工具调用，或者改变其意图或参数。然而，传统的字错误率（WER）并未捕捉到这些差异。我们引入了SpeechEvalX，这是一种配对诊断协议，用于评估受ASR错误影响的任务角色如何影响结构化工具调用。控制编辑通过源话语、编辑数量和操作进行匹配，同时任务角色标签与词汇严重性、BERT余弦距离、语义字错误率（Semantic-WER）和二元关键性进行比较。在594个源话语中，联合调用准确率从参考转录的82.3%下降到三个ASR前端的71.9-74.9%。数字和否定编辑导致21.9-25.7个百分点的损失，而控制编辑仅导致1.4-2.9个百分点的损失。详细的任务角色标签使得源分组的ROC-AUC从0.73提高到0.78，相较于二元关键性，显示出下游影响在很大程度上依赖于受影响的任务角色。

View on arXiv Download PDF AI Translation

cs.CL / 164 / 2610.17654

Decouple to Forget: Semantic Mediation for Sensitive Unlearning in Large Language Models

解耦以遗忘：大型语言模型中敏感信息的语义中介遗忘方法

Cong, Jingyue, Song, Andy, Yu, Yang, Sarwar, Tabinda, Li, Xinshu, Xu, Ziqi, Yi, Xun, Wang, Ke, Han, Bo, He, Estrid

Abstract

Machine unlearning for large language models seeks to remove specified knowledge while preserving general capability, yet current methods often treat an entire response as the unit of forgetting. This coarse granularity causes unnecessary utility loss when the forget target is only a small set of sensitive facts embedded within otherwise benign explanations. In this paper, we introduce Decouple to Forget (D2F), a targeted unlearning method inspired by semantic mediation that separates sensitive factual content from non-sensitive meaning and surface expression. For each example to be forgotten, D2F extracts a fact-level "mediator" representation from the answer, identifies which parts are sensitive, and removes only those parts while leaving the rest of the structure intact. We then generate counterfactual answers grounded in this mediator that preserve non-sensitive content but exclude sensitive facts, employing preference-based contrastive optimization to reduce the likelihood of information leakage while maintaining helpfulness and fluency. Experiments across various benchmark datasets show that D2F achieves a better balance between forgetting and retention than strong baseline methods. Our code is available at https://github.com/JingyueCong/D2F.

Chinese Translation

大型语言模型的机器遗忘旨在去除特定知识，同时保持一般能力，然而当前的方法往往将整个响应视为遗忘的单位。这种粗糙的粒度在遗忘目标仅为嵌入在其他无害解释中的一小部分敏感事实时，导致了不必要的效用损失。本文提出了一种名为解耦以遗忘（Decouple to Forget, D2F）的有针对性的遗忘方法，该方法受到语义中介的启发，将敏感事实内容与非敏感意义和表面表达分离。对于每个需要遗忘的示例，D2F 从答案中提取一个事实级的“中介”表示，识别出哪些部分是敏感的，并仅去除这些部分，同时保持其余结构的完整性。然后，我们生成基于该中介的反事实答案，保留非敏感内容但排除敏感事实，采用基于偏好的对比优化方法以减少信息泄露的可能性，同时保持有用性和流畅性。在各种基准数据集上的实验表明，D2F 在遗忘与保留之间实现了比强基线方法更好的平衡。我们的代码可在 https://github.com/JingyueCong/D2F 获取。

View on arXiv Download PDF AI Translation

cs.CL / 165 / 2610.17658

Worth Their Weights? Evaluating Generative Models for Discourse Parsing

值得其分量吗？评估话语解析的生成模型

Gessler, Luke, Zeldes, Amir

Abstract

Rhetorical Structure Theory (RST) parsing has historically been dominated by discriminative models based on pretrained encoder-only language models. The recent availability of high-quality generative language models has made it unclear whether RST parsing is better approached as a generative task. In this work, we investigate whether generative language models are better at parsing RST than traditional discriminative ones. We find that at matched parameter counts, discriminative parsers perform much better than generative ones, though much larger generative models sometimes outperform them. Generative models hold a particular advantage in elementary discourse unit (EDU) segmentation, and we report new state-of-the-art results on the RST-DT benchmark in end-to-end parsing as well as segmentation.

Chinese Translation

修辞结构理论（RST）解析历史上主要由基于预训练编码器的判别模型主导。最近高质量生成语言模型的出现使得RST解析是否更适合作为生成任务变得不明确。在本研究中，我们探讨了生成语言模型在解析RST方面是否优于传统的判别模型。我们发现，在参数数量相匹配的情况下，判别解析器的表现远远优于生成解析器，尽管在某些情况下，规模更大的生成模型有时会超越它们。生成模型在基本话语单元（EDU）分割方面具有特别的优势，我们在端到端解析和分割的RST-DT基准上报告了新的最先进结果。

View on arXiv Download PDF AI Translation

cs.CL / 166 / 2610.17663

The Image Still Has a Say: Final-Answer Repair for Multimodal Chain-of-Thought

图像依然有发言权：多模态思维链的最终答案修复

Pan, Yulan, Xu, Xiaoyu, Shen, Zhihong

Abstract

Although chain-of-thought prompting has improved multimodal reasoning, it is still difficult to produce correct final answers, as the reasoning can drift away from the image. Besides, the trace is already fixed when the answer is generated, leaving no room for self-correction. Considering the image is still available at the final-answer step, we think it is reasonable to fix the answer there without rewriting the reasoning. As such, the problem becomes to correct the final answer using the image under fixed reasoning. Unlike methods that improve trace generation through training, resampling, or decoding-time grounding, we repair the final-answer readout while keeping the reasoning trace fixed. The motivation is based on our finding that a wrong option can become prematurely committed in the reasoning while the correct option still has stronger image support in middle-to-deep layers. Thus, we propose Text-to-Visual Evidence Reallocation (TVER), a training-free method for final-answer repair. Specifically, TVER selects answer-related text and question-matched image patches, and moves bounded attention from text to patches. Furthermore, the patches are chosen in a middle-to-deep layer window based on question relevance. Our method improves answers across benchmarks and backbones, with gains in both multiple-choice and open-answer evaluations.

Chinese Translation

尽管思维链提示已改善多模态推理，但生成正确的最终答案仍然困难，因为推理可能偏离图像。此外，当答案生成时，推理轨迹已经固定，没有自我修正的空间。考虑到在最终答案步骤中图像仍然可用，我们认为在不重写推理的情况下修正答案是合理的。因此，问题转变为在固定推理的基础上使用图像来修正最终答案。与通过训练、重采样或解码时定位来改善轨迹生成的方法不同，我们在保持推理轨迹固定的同时修复最终答案的输出。我们的动机基于一个发现：错误选项可能在推理中过早承诺，而正确选项在中到深层仍具有更强的图像支持。因此，我们提出了文本到视觉证据重新分配（Text-to-Visual Evidence Reallocation, TVER），这是一种无训练的最终答案修复方法。具体而言，TVER选择与答案相关的文本和与问题匹配的图像块，并将受限注意力从文本转移到图像块。此外，这些图像块是在基于问题相关性的中到深层窗口中选择的。我们的方法在各个基准和骨干网络上改善了答案，在多项选择和开放答案评估中均取得了提升。

View on arXiv Download PDF AI Translation

cs.CL / 167 / 2610.17685

LatentRiskBench: Measuring State-Dependent Risks in Enterprise Agents

LatentRiskBench：测量企业代理中的状态依赖风险

Cong, Jingyue, He, Estrid, Li, Zhuang, Wu, Nan, Shen, Li, Peng, Yameng, Li, Xinshu, Song, Andy

Abstract

Enterprise agents face safety risks that cannot be determined from individual actions alone. Existing safety benchmarks largely evaluate action-local risks, overlooking failures whose safety depends on workflow state. In such state-dependent risks, the same action may be safe in one context but violate policy in another. Hence, we introduce LatentRiskBench for evaluating such state-dependent risks in enterprise settings. It pairs risky trajectories with benign counterparts that share the same terminal action but differ in latent policy-relevant states, requiring models to reconstruct workflow context rather than rely on action-level cues. We also introduce new metrics for state-dependent risk evaluation, including a suite of residual risk (RR) measures. Experiments on the benchmark show that action-level evaluation detects only 0.09 of state-dependent violations, whereas state-level evaluation reaches 0.96. Existing guardrail methods largely overlook such risks unless they are explicitly applied to workflow states. Case studies on a commercial deployed scenario further show that the safety metrics embedded in the benchmark provide a more realistic measure of safety for enterprise agents. Our dataset and code are available at https://github.com/JingyueCong/LatentRiskBench.

Chinese Translation

企业代理面临的安全风险无法仅通过单个行动来确定。现有的安全基准主要评估行动局部风险，忽视了那些安全性依赖于工作流状态的失败。在这种状态依赖风险中，相同的行动在一个上下文中可能是安全的，而在另一个上下文中则可能违反政策。因此，我们引入了LatentRiskBench，用于评估企业环境中的状态依赖风险。它将风险轨迹与共享相同终止行动但在潜在政策相关状态上有所不同的良性轨迹配对，要求模型重建工作流上下文，而不是依赖于行动层面的线索。我们还引入了用于状态依赖风险评估的新指标，包括一套残余风险（Residual Risk, RR）度量。基准实验表明，行动层面的评估仅检测到0.09的状态依赖违规，而状态层面的评估达到了0.96。现有的保护措施方法在没有明确应用于工作流状态的情况下，往往忽视这些风险。对一个商业部署场景的案例研究进一步表明，嵌入基准中的安全指标为企业代理提供了更为真实的安全度量。我们的数据集和代码可在 https://github.com/JingyueCong/LatentRiskBench 获取。

View on arXiv Download PDF AI Translation

cs.CL / 168 / 2610.17686

Encoding Context into LLM Weights: What Makes It Context?

将上下文编码到大型语言模型权重中：什么构成了上下文？

Wang, Yue, Wang, Qizhou, Liu, Feng, Niu, Gang, Sugiyama, Masashi, Han, Bo

Abstract

Longer contexts allow large language models (LLMs) to condition their predictions on more information, but inference costs grow prohibitively with context length. An emerging alternative uses test-time updates to encode context as temporary changes to model weights. However, encoding this information into weights that already represent general knowledge obscures its role as context, so the LLM may fail to use it properly to condition subsequent inference. To restore this contextual role, we first associate each context with an identifier during encoding, thereby distinguishing the newly encoded information from pre-existing knowledge. Invoking the identifier at inference then enables the LLM to recite the encoded context. Nevertheless, successful recitation does not ensure that the LLM will use the recalled information as context. We therefore further introduce a prompting framework that explicitly guides the LLM to use this information as context. Across diverse controlled tasks and practical benchmarks, our method outperformed the baseline using only off-the-shelf LLMs, showing its advances in context encoding.

Chinese Translation

更长的上下文使大型语言模型（LLMs）能够基于更多信息进行预测，但随着上下文长度的增加，推理成本也变得不可承受。一种新兴的替代方案是在测试时更新，将上下文编码为对模型权重的临时变化。然而，将这些信息编码到已经代表一般知识的权重中，会模糊其作为上下文的角色，因此LLM可能无法正确利用这些信息来条件化后续推理。为了恢复这种上下文角色，我们首先在编码过程中将每个上下文与一个标识符关联，从而将新编码的信息与现有知识区分开。在推理时调用该标识符可以使LLM复述编码的上下文。然而，成功的复述并不能确保LLM会将回忆的信息作为上下文使用。因此，我们进一步引入了一种提示框架，明确指导LLM将这些信息用作上下文。在多种受控任务和实际基准测试中，我们的方法在仅使用现成的LLM时超越了基线，显示了其在上下文编码方面的进展。

View on arXiv Download PDF AI Translation

cs.CL / 169 / 2610.17703

Pushback in a Haystack: Language Models Respond to Factual Errors Ambiguously and Ineffectively

干草堆中的反驳：语言模型对事实错误的回应模糊且无效

Lucy, Li, Cheng, Myra, Guhr, Svenja, Ahuja, Kabir, Ahia, Orevaoghene, Chattopadhyay, Sameep, Wang, Yike, Hofmann, Valentin, Smith, Noah A., Tsvetkov, Yulia

Abstract

The sycophancy of language models, or their failure to push back against faulty user input, is now widely studied. However, if LMs do push back, is sycophancy solved? We show that LMs vary in how they present pushback, and thus vary in how effectively they correct user beliefs. We introduce a phrase-level, human-validated response labeling framework to uncover fine-grained insights into how LMs push against and/or affirm false inputs. We develop and apply our framework on model responses in five languages: English, German, Chinese, Hindi, and Yoruba. We find that when models do push back against users' false assertions, they tend to use indirect phrasing and mix in affirmation, with more indirectness in Hindi and Yoruba than in other languages considered. In a large-scale human study (n = 300), we find that the pushback ambiguity we identified may fail to correct and prevent users' factual misconceptions.

Chinese Translation

语言模型的谄媚性，即它们未能对错误的用户输入进行反驳，已被广泛研究。然而，如果语言模型确实进行反驳，谄媚性是否就得到解决？我们展示了语言模型在反驳的表现上存在差异，因此在纠正用户信念的有效性上也存在差异。我们引入了一种短语级的人类验证响应标注框架，以揭示语言模型如何反对和/或确认错误输入的细微见解。我们在五种语言的模型响应中开发并应用了我们的框架：英语、德语、中文、印地语和约鲁巴语。我们发现，当模型确实对用户的错误主张进行反驳时，它们往往使用间接措辞并混合肯定，印地语和约鲁巴语中的间接性程度高于其他语言。在一项大规模的人类研究中（n = 300），我们发现我们识别出的反驳模糊性可能未能纠正和防止用户的事实误解。

View on arXiv Download PDF AI Translation

cs.CL / 170 / 2610.17731

Understanding Machine Taste: Aesthetic Agreement Across Foundation Models in K-pop

理解机器的审美：基础模型在K-pop中的审美一致性

Wang, Kaishen, Pan, Yucheng, Cui, Xuehao, Tian, Hanling, Huang, Heng

Abstract

As foundation models enter creative industries, understanding their aesthetic preferences becomes an important step toward AI-assisted creative decision making. In this work, we investigate how foundation models evaluate creative content and characterize their ``machine taste'' using K-pop songs and music videos as a testbed. We formulate aesthetic evaluation as a pairwise ranking problem and systematically examine pairwise inter-model agreement, group-level consensus, and cross-modal preference alignment. Our results reveal stronger pairwise inter-model agreement and group-level consensus in visual evaluation than in audio evaluation, while preferences across the two modalities remain weakly aligned. We further examine how visual consensus rankings relate to real-world popularity using historical YouTube viewership data, finding positive but varying correlations across artist groups. These findings provide initial insights into machine aesthetic preferences and lay the groundwork for future AI-assisted creative decision making that considers creative content, artist identity, audience preferences, and market dynamics.

Chinese Translation

随着基础模型进入创意产业，理解它们的审美偏好成为AI辅助创意决策的重要一步。在本研究中，我们探讨基础模型如何评估创意内容，并以K-pop歌曲和音乐视频作为测试平台来表征它们的“机器审美”。我们将审美评估形式化为成对排名问题，并系统地考察成对模型间的一致性、群体层面的共识以及跨模态的偏好对齐。我们的结果显示，在视觉评估中，成对模型间的一致性和群体层面的共识比音频评估更强，而两种模态之间的偏好仍然保持弱对齐。我们进一步考察视觉共识排名与现实世界流行度的关系，使用历史YouTube观看数据，发现不同艺术家群体之间存在正相关但变化不一的关系。这些发现为机器审美偏好的初步洞察提供了基础，并为未来考虑创意内容、艺术家身份、观众偏好和市场动态的AI辅助创意决策奠定了基础。

View on arXiv Download PDF AI Translation

cs.CL / 171 / 2610.17742

Geometric Structure Predicts Per-Example Neighborhood Loss Under PCA Truncation

几何结构预测主成分分析截断下的每例邻域损失

Loganathan, Sriram, Anand, Gokul, Arivazhagan, Ragavan

Abstract

Dimensionality reduction can preserve the average performance of text embeddings even when a substantial fraction of their dimensions is removed. Aggregate metrics, however, do not reveal whether degradation is distributed uniformly across examples. We study per-example neighborhood stability under PCA truncation across static word embeddings and modern sentence encoders. Per-item neighbor retention varies widely and is predictable from geometric properties of the representation and its alignment with the retained PCA subspace. The fraction of centered-vector energy retained by the truncated subspace is strongest for CBOW and RI, with neighborhood contrast leading on SkipGram and the modern/Matryoshka settings (GTE approximately tied), while local density is associated with retention in marginal analysis but adds little unique information once energy and contrast are controlled (R-squared increments 0.001 to 0.010). Local intrinsic dimensionality accounts for more variation than density on modern sentence encoders in pairwise comparison. On SciFact, query density predicts retrieval degradation under aggressive compression; on NFCorpus the primary 32d association does not replicate, and a direct variance-restriction test leaves that null unresolved. Average compression performance conceals systematic per-example variation that geometry makes visible.

Chinese Translation

降维可以在去除大量维度的情况下保持文本嵌入的平均性能。然而，聚合指标并不能揭示性能下降是否在各个样本之间均匀分布。我们研究了静态词嵌入和现代句子编码器在主成分分析（PCA）截断下的每例邻域稳定性。每个项目的邻居保留率差异很大，并且可以从表示的几何特性及其与保留的PCA子空间的对齐程度中预测。对于CBOW和RI，截断子空间保留的中心向量能量比例最强，而在SkipGram和现代/马特里奥什卡设置中，邻域对比度领先（GTE大致相当），而局部密度在边际分析中与保留相关，但在控制能量和对比度后几乎没有提供独特信息（R平方增量为0.001到0.010）。在成对比较中，局部内在维度比密度在现代句子编码器上解释了更多的变化。在SciFact上，查询密度预测在激进压缩下的检索降级；在NFCorpus上，主要的32维关联并未复制，直接的方差限制测试未能解决该无效结果。平均压缩性能掩盖了几何结构所揭示的系统性每例变异。

View on arXiv Download PDF AI Translation

cs.CL / 172 / 2610.17747

Auditing Failure Attribution: Label Predictability and Judge Behavior

失败归因审计：标签可预测性与评审者行为

Xu, Tianyang, Gao, Jing

Abstract

Failure-attribution benchmarks ask an LLM judge which step or agent caused a failure. A high score should mean the judge checked the faulty step against the run (a cross-step check). However, spurious signals (surface cues) that predict the annotation can produce a high score. In this paper, we propose four tests to separate checking from cue use. Input restriction asks whether partial input predicts the annotation. Cue following checks whether moved cues change answers. Residual accuracy scores judges where cue-only predictors fail. Aggregation asks whether the failure mix or metric changes rankings. Real observations motivate these tests: on three benchmarks, simple predictors match or beat GPT-4.1. In two Who&When Pro systems, a classifier reading one step's reasoning finds 95-100% of faulty steps, motivating input restriction. In one of them, error injection leaves no reasoning after the faulty step; judges fine-tuned on its labels beat GPT-4.1 by 41 points. Adding task-related reasoning after that step moves this cue: fine-tuned judges lose 31-99 points, but zero-shot judges lose at most 9, motivating cue following. On Aegis, blaming every agent matches GPT-4.1 on micro-F1 but trails by 19 points in exact-set accuracy, motivating aggregation. We recommend reporting these tests when developing failure-attribution benchmarks.

Chinese Translation

失败归因基准要求大型语言模型（LLM）评审者指出哪个步骤或代理导致了失败。高分应意味着评审者对故障步骤进行了与运行结果的交叉检查。然而，预测注释的虚假信号（表面线索）可能会产生高分。在本文中，我们提出了四个测试来区分检查与线索使用。输入限制测试部分输入是否能预测注释。线索跟随检查移动的线索是否会改变答案。残差准确度评估在仅使用线索的预测失败时的评审者表现。聚合测试询问失败组合或指标是否会改变排名。真实观察激励了这些测试：在三个基准上，简单的预测器与或超过了GPT-4.1。在两个Who&When Pro系统中，一个分类器读取一个步骤的推理发现95-100%的故障步骤，这激励了输入限制。在其中一个系统中，错误注入在故障步骤后没有留下推理；在其标签上微调的评审者比GPT-4.1高出41分。添加与任务相关的推理后，该线索发生了变化：微调的评审者损失31-99分，而零样本评审者最多损失9分，这激励了线索跟随。在Aegis上，归咎于每个代理在微F1上与GPT-4.1相匹配，但在精确集准确度上落后19分，这激励了聚合。我们建议在开发失败归因基准时报告这些测试。

View on arXiv Download PDF AI Translation

cs.CL / 173 / 2610.17761

Connecting the Dots: Enduring Performance Gaps in ASR Systems Between Standard and Minority English Varieties

连接点滴：标准与少数英语变体之间的自动语音识别系统持久性能差距

Metzger, Alexander, Srivastava, Aruna, Mukhamedvaleev, Ruslan, Yeo, Eunjung, Ahmed, Syed Ishtiaque, Markl, Nina, Kumar, Sachin, Samir, Farhan

Abstract

Automatic Speech Recognition (ASR) system evaluations have consistently revealed performance disparities along lines of race, gender, socioeconomic status, and language variety. While these disparities are well-documented, they are often (implicitly or explicitly) treated as issues that subsequent, larger, and more universal models will correct. We challenge this assumption of steady progress in reducing accent-related performance disparities, conducting a longitudinal evaluation of 16 prominent systems released between 2021-2025 across 8 datasets representing diverse English accent varieties. We find that the performance disparity between Mainstream US English (MUSE) and minority accents is stubborn, not reliably declining as a function of time. Specifically, while systems are reliably performant on MUSE, minority accents are subject to risks of large performance degradations between generations of systems, sometimes worsening by 65% absolute Word Error Rate. We argue that standard ASR benchmarks enact an implicit language policy privileging Mainstream US English varieties. As such, these performance disparities reflect systemic failures of language technology development infrastructure that disproportionately affect underrepresented speaker populations.

Chinese Translation

自动语音识别（ASR）系统的评估一贯揭示了种族、性别、社会经济地位和语言变体之间的性能差异。尽管这些差异已被充分记录，但它们通常被（隐性或显性地）视为后续更大、更普遍的模型将会纠正的问题。我们对这种在减少与口音相关的性能差距方面持续进步的假设提出质疑，进行了对2021年至2025年间发布的16个显著系统的纵向评估，涵盖了8个代表不同英语口音变体的数据集。我们发现主流美国英语（Mainstream US English, MUSE）与少数口音之间的性能差距顽固存在，并未随着时间的推移而可靠下降。具体而言，尽管系统在MUSE上的表现稳定，但少数口音在系统代际之间面临着性能大幅下降的风险，有时绝对词错误率（Word Error Rate）甚至恶化达65%。我们认为，标准的ASR基准隐性实施了一种语言政策，优先考虑主流美国英语变体。因此，这些性能差距反映了语言技术开发基础设施的系统性失败，这种失败对代表性不足的说话人群体产生了不成比例的影响。

View on arXiv Download PDF AI Translation

cs.CL / 174 / 2610.17767

Efficiently Learning To Reason or Not to Reason: Root-token Policy Optimization for Adaptive Thinking

高效学习推理或不推理：适应性思维的根标记策略优化

Kim, Taehyeon, Lee, Hyunsoo, Jang, Youngsoo, Lee, Moontae

Abstract

Large reasoning models (LRMs) achieve strong performance by externalizing explicit reasoning traces before producing the answer, yet suffer from overthinking challenge that allocates uniformly heavy computation to queries of varying difficulty. While proprietary models mitigate this via opaque routing, open-source LRMs still lack an efficient mechanism to internalize adaptive reasoning due to both expensive training cost and limited disclosure of training recipes. In response, we introduce RPO (Root-token Policy Optimization), a framework that enables LRMs to self-determine when to reason by training only the initial root token (e.g., whether to invoke the tag) via group relative reward and group-wise advantages. By focusing on this pivotal branching point, RPO drastically reduces training overhead and VRAM usage. Across multiple model families and scales, RPO learns difficulty-aware adaptive thinking at just ~2% of the training compute of prior adaptive-reasoning methods.

Chinese Translation

大型推理模型（LRMs）通过在生成答案之前外化显式推理痕迹来实现强大的性能，但面临过度思考的挑战，这导致对不同难度查询的计算资源分配过于均匀。虽然专有模型通过不透明的路由机制缓解了这一问题，但开源LRMs仍缺乏高效的机制来内化适应性推理，原因在于训练成本高昂以及训练配方的有限披露。为此，我们提出了RPO（根标记策略优化），这是一个框架，使LRMs能够自我决定何时进行推理，通过仅训练初始根标记（例如，是否调用标签）来实现，采用组相对奖励和组优势的方法。通过聚焦于这一关键的分支点，RPO显著降低了训练开销和显存使用。在多个模型家族和规模中，RPO以仅约2%的训练计算量学习到难度感知的适应性思维，远低于先前的适应性推理方法。

View on arXiv Download PDF AI Translation

cs.CL / 175 / 2610.17771

Many Right Answers, One Training Example: Removing Responder Noise from GRPO for LLM Recommenders

多个正确答案，一个训练示例：从 GRPO 中去除响应者噪声以用于 LLM 推荐系统

Li, Xin, Jiang, Hao, Guo, Jinghao, Lin, Weisi, Yuen, Chau

Abstract

A recommendation request on Reddit draws several replies, each naming different titles. When an LLM recommender is trained on such threads with reinforcement learning (RL), which replies should a generated list be rewarded against? The state-of-the-art recipe, Rank-GRPO, picks one reply per training example, so the advantage mostly records which reply was drawn (85% of its variance) rather than how good the list is. We argue that RL should train on the request, not on each answer, and propose ReMaRk: it scores every list once against the union of all replies' titles and samples each request equally. This makes RL cheaper. On seven backbones and three datasets, ReMaRk reaches the validation recall of 2,000 per-reply RL steps in 25 to 53% of the steps and leads at that budget; trained until neither improves, it is never significantly behind Rank-GRPO and in four of seven pairs reaches Rank-GRPO's best with fewer GPU-hours. The same reward helps on ProtoQA, where about 100 people answer each question, which points to a simple rule for RL with several valid human answers: reward them jointly.

Chinese Translation

在 Reddit 上的推荐请求引发了多个回复，每个回复都提到不同的标题。当使用强化学习（RL）对这些线程进行训练时，生成的列表应该依据哪些回复进行奖励？当前的先进方法 Rank-GRPO 每个训练示例只选择一个回复，因此其优势主要记录了选择了哪个回复（占其方差的 85%），而不是列表的质量。我们认为 RL 应该针对请求进行训练，而不是每个答案，并提出了 ReMaRk：它对每个列表进行一次评分，依据所有回复标题的并集，并对每个请求进行均等抽样。这使得 RL 更加高效。在七个基础模型和三个数据集上，ReMaRk 在 25% 到 53% 的步骤中达到了每个回复 RL 步骤的 2,000 次验证召回，并在该预算下领先；在训练直到没有进一步提升时，它从未显著落后于 Rank-GRPO，并且在七对中的四对中以更少的 GPU 小时达到了 Rank-GRPO 的最佳表现。相同的奖励机制在 ProtoQA 上也有效，在那里大约 100 人回答每个问题，这指向了一个简单的规则：对于多个有效的人类答案，联合奖励它们。

View on arXiv Download PDF AI Translation

cs.CL / 176 / 2610.17777

Beyond Belief Nesting: Rethinking Higher-Order Theory of Mind in Large Language Models

超越信念嵌套：重新思考大型语言模型中的高阶心智理论

You, Shanshan, Wang, Zhengqi, Liu, Hao, Liu, Yuqin, He, Ya, Hao, Xin, Wang, Liyuan

Abstract

Theory of mind (ToM), the ability to reason about others' mental states, underlies effective communication and coordination. Its higher-order form enables reasoning over nested perspectives in strategic interactions. However, existing evaluations often equate belief-nesting depth with reasoning order, allowing nominally higher-order questions to admit lower-order shortcuts. We formalize effective order through the distinct events supporting successive belief prefixes, revealing widespread order inflation across existing benchmarks. We introduce HopToM, with certified full-order questions and controls for lower-order alternatives. Under direct prompting, mean accuracy across four frontier LLMs falls from 73.5% at first order to 20.0% at fourth order, and human participants also struggle. Causal interventions link a recurring class of errors to evidence for shallower belief prefixes. We introduce HopTrace, a plug-and-play inference-time method that encodes each interaction once into perspective records and resolves nested queries deterministically. On certified narrative questions, it improves fourth- and fifth-order accuracy by 22.0 and 38.5 percentage points over strong prompting baselines, while using 20-39 times fewer tokens than multi-call methods. These results sharpen how higher-order ToM is evaluated and offer a practical way to improve perspective reasoning in interactive agents.

Chinese Translation

心智理论（Theory of Mind, ToM）是推理他人心理状态的能力，它是有效沟通和协调的基础。其高阶形式使得在战略互动中能够对嵌套视角进行推理。然而，现有评估往往将信念嵌套深度等同于推理顺序，允许名义上高阶的问题采用低阶捷径。我们通过支持连续信念前缀的不同事件来形式化有效顺序，揭示了现有基准测试中普遍存在的顺序膨胀。我们引入了HopToM，提供经过认证的全阶问题和对低阶替代方案的控制。在直接提示下，四个前沿大型语言模型的平均准确率从第一阶的73.5%下降到第四阶的20.0%，而人类参与者也面临困难。因果干预将一类反复出现的错误与浅层信念前缀的证据联系起来。我们引入了HopTrace，这是一种即插即用的推理时间方法，它将每次交互编码为视角记录，并以确定性方式解决嵌套查询。在经过认证的叙述性问题上，它在第四阶和第五阶的准确率分别提高了22.0和38.5个百分点，相较于强提示基线，同时使用的令牌数量比多次调用方法少20-39倍。这些结果明确了高阶心智理论的评估方式，并提供了一种实用的方法来改善交互代理中的视角推理。

View on arXiv Download PDF AI Translation

cs.CL / 177 / 2610.17778

GENDIS: Gender-Related Discourse and Stance in Multilingual YouTube Comments from India, Kenya, and Nigeria

GENDIS：来自印度、肯尼亚和尼日利亚的多语言YouTube评论中的性别相关话语和立场

Koch, Luise, Tonneau, Manuel, Malhotra, Niyati, Rai, Sunny, Boudet, Ana Maria Munoz, Orozco-Olvera, Victor, Fraiberger, Samuel P., Guntuku, Sharath Chandra

Abstract

Research on gender-related online content is often operationalized through harm-centered tasks that collapse non-harmful gender-related and unrelated content, while existing datasets remain concentrated in high-resource languages and Global North settings. To bridge these gaps, we introduce GENDIS, a dataset of 9,600 multilingual, human-annotated YouTube comments from India, Kenya, and Nigeria (3,200 per country), annotated using a two-level framework that first identifies whether gender is referenced and then classifies gender-related content as negative, positive, or neutral. This separation provides a denominator for gender-related discourse independent of harm, enabling both prevalence estimation and characterization of its evaluative range. In the uniform random stratum, 23.8-35.0% of comments are gender-related, and among those with a resolved stance, 82% are neutral or positive. Across models, gender-reference detection is substantially easier than stance classification, especially for positive and neutral speech. Lightly supervised classifiers match frontier zero-shot large language models at substantially lower cost and outperform generic toxicity detection, while cross-country transfer remains limited. GENDIS provides a locally grounded resource for measuring when gender is invoked online and how that reference is framed.

Chinese Translation

关于性别相关在线内容的研究通常通过以伤害为中心的任务进行操作，这种方法将无害的性别相关内容和无关内容混为一谈，而现有的数据集仍集中在高资源语言和全球北方的环境中。为了填补这些空白，我们引入了GENDIS，这是一个包含来自印度、肯尼亚和尼日利亚的9,600条多语言人类标注的YouTube评论的数据集（每个国家3,200条），使用一个两级框架进行标注，首先识别是否提及性别，然后将性别相关内容分类为负面、正面或中性。这种分离为独立于伤害的性别相关话语提供了一个基数，使得既可以估计其普遍性，也可以表征其评估范围。在均匀随机层中，23.8%-35.0%的评论是性别相关的，而在那些已解决立场的评论中，82%是中性或正面的。在不同模型中，性别引用检测明显比立场分类容易，尤其是对于正面和中性言论。轻度监督的分类器在成本显著较低的情况下与前沿的零样本大型语言模型相匹配，并且在检测一般毒性方面表现优于后者，而跨国转移仍然有限。GENDIS提供了一个本地化的资源，用于衡量在线何时提及性别以及这种引用是如何框定的。

View on arXiv Download PDF AI Translation

cs.CL / 178 / 2610.17780

Full-duplex Spoken Language Modeling with Discrete Diffusion

基于离散扩散的全双工口语语言建模

Chen, Chen, Ku, Pin-Jui, Pasad, Ankita, Jukić, Ante, Chen, Zhehuai, Ginsburg, Boris

Abstract

We introduce Diffuplex, a full-duplex spoken language model based on discrete diffusion that achieves an unprecedented balance between linguistic intelligence and response efficiency. With thinking mode, Diffuplex delivers 13.0 tokens per forward pass, reaching an average of 1872 tokens per second and substantially surpassing the frontier autoregressive (AR) models in decoding speed. Without thinking, Diffuplex matches the first-audio latency of existing full-duplex models while achieving markedly stronger linguistic intelligence. This capability is enabled by separating when to say from what to say at the training level, rather than compressing both into a single autoregressive timeline. During inference, Diffuplex continuously tracks the user speech through an AR interaction policy, while independently generating and refining response content through parallel diffusion. Consequently, semantic computation is no longer bounded by the perception audio frame rate, allowing intelligence and responsiveness to improve simultaneously. Taken together, these results establish a new paradigm for spoken agents that can sustain real-time dialogue while devoting parallel computation to intelligent response generation. Training code and checkpoints will be avaliable soon.

Chinese Translation

我们介绍了Diffuplex，这是一种基于离散扩散的全双工口语语言模型，在语言智能和响应效率之间实现了前所未有的平衡。在思考模式下，Diffuplex每次前向传递可生成13.0个标记，达到每秒平均1872个标记，显著超越了前沿的自回归（AR）模型的解码速度。在非思考模式下，Diffuplex的首次音频延迟与现有的全双工模型相匹配，同时实现了显著更强的语言智能。这一能力的实现是通过在训练层面将何时发言与说什么分开，而不是将两者压缩到单一的自回归时间线上。在推理过程中，Diffuplex通过AR交互策略持续跟踪用户的语音，同时通过并行扩散独立生成和完善响应内容。因此，语义计算不再受限于感知音频帧率，从而使智能和响应能力得以同时提升。综合来看，这些结果为能够维持实时对话的口语代理建立了一种新的范式，同时将并行计算用于智能响应生成。训练代码和检查点将很快发布。

View on arXiv Download PDF AI Translation

cs.CL / 179 / 2610.17782

Where Words Begin and End

词语的起始与结束

Zolkepli, Husein

Abstract

Forced aligners locate each word of a known transcript in the audio, and many speech pipelines depend on them. In our own data, a widely used CTC aligner inserts a pause at nearly every word boundary, and we found almost no aligner evaluated on telephone audio. We compare nine aligner configurations against human word boundaries on read, conversational and telephone English. The Montreal Forced Aligner (MFA) is the most precise, while CTC aligners consistently start words late, because most frames inside a word go to the blank token. We then fine-tune Whisper with LoRA to predict each word's start and end as timestamp tokens, training only on boundaries placed by people: 47 hours in 87 languages. The resulting aligner is more precise than MFA on every English test condition, and than MFA's own Korean and Dutch models. With FA-Bench's protocol and code on TIMIT, its word-boundary error is 11.6 ms, against 18.2 ms for the best published system. Two choices matter most on telephone conversations: placing the timestamp slots after the word they time, and training on noisy transcripts, which also makes the aligner nearly insensitive to errors in the transcript. The released model reaches 16.0 ms on real landline calls, against 36.5 ms for MFA. The models are available at https://huggingface.co/collections/Scicom-intl/force-alignment

Chinese Translation

强制对齐器在音频中定位已知转录文本的每个单词，许多语音处理流程依赖于它们。在我们的数据中，一种广泛使用的CTC对齐器在几乎每个单词边界插入了暂停，而我们发现几乎没有对电话音频进行评估的对齐器。我们比较了九种对齐器配置与人类单词边界在阅读、对话和电话英语中的表现。蒙特利尔强制对齐器（Montreal Forced Aligner, MFA）是最精确的，而CTC对齐器则始终在单词开始时延迟，因为大多数单词内部的帧都分配给了空白标记。随后，我们使用LoRA对Whisper进行微调，以将每个单词的起始和结束预测为时间戳标记，仅在由人类放置的边界上进行训练：47小时的87种语言。结果显示，该对齐器在每个英语测试条件下的精度超过了MFA，并且优于MFA自身的韩语和荷兰语模型。在FA-Bench的协议和TIMIT上的代码下，其单词边界误差为11.6毫秒，而最佳已发布系统为18.2毫秒。在电话对话中，两个选择最为重要：将时间戳插槽放置在它们所计时的单词之后，以及在嘈杂的转录文本上进行训练，这也使得对齐器对转录文本中的错误几乎不敏感。发布的模型在真实的固定电话通话中达到了16.0毫秒，而MFA为36.5毫秒。这些模型可在https://huggingface.co/collections/Scicom-intl/force-alignment获取。

View on arXiv Download PDF AI Translation

cs.CL / 180 / 2610.17805

BRIDGE: Selective Boundary Repair with Dependency Graph Evidence in Multi-Agent Systems

BRIDGE：基于依赖图证据的多智能体系统选择性边界修复

Su, Encheng, Li, Pengze, Hua, Ziheng, Yu, Qiucheng, Wu, Jianyu, Tang, Chen, Wang, Yizhou, Wang, Lintao, Xiang, Junjia, Ma, Xinzhu, Chen, Xi, Tang, Shixiang, Li, Houqiang, Chen, Xi, Wang, Aoran

Abstract

Multi-agent collaboration has emerged as a critical paradigm for tackling complex, long-horizon tasks; however, while coordination protocols continue to advance, systemic failures at the inter-agent collaboration interface remain a pervasive bottleneck. In collaborative workflows, systems frequently suffer from \emph{broken handoffs}---where critical task context is omitted, misrouted, or ignored despite each agent emitting locally plausible responses. To address this challenge, we present \method, a verifiable runtime control framework that treats handoff repair as selective intervention over explicit dependency graphs. Central to \method is an integrated fault-attribution and calibration pipeline, \dataset, which decouples execution dependencies and injects counterfactual handoff perturbations to provide empirical supervision for runtime intervention. Leveraging this diagnostic foundation, \method dynamically compiles observable interaction traces into a provenance graph to pinpoint unresolved dependencies, without accessing private agent states. It then evaluates candidate repairs through an empirical gate that weighs expected gains against disruption risks, and executes repairs only when formal local postconditions hold. Extensive evaluations across demanding benchmarks demonstrate that \method boosts task success by up to 12.3 absolute percentage points on SILO-BENCH and 11.1 on TeamBench. Across both backbones, \method maintains strict cost-effectiveness, adding as little as 2.0\% marginal token overhead. Furthermore, on 562 handoff-relevant traces from the MAST benchmark, \method achieves 69.8\% F1 in a single pass, establishing a reliable operational boundary for verifiable coordination repair.

Chinese Translation

多智能体协作已成为应对复杂长时任务的重要范式；然而，尽管协调协议不断进步，智能体协作接口的系统性故障仍然是一个普遍的瓶颈。在协作工作流中，系统经常遭遇 extit{断裂交接}的问题——关键任务上下文被遗漏、错误路由或忽视，尽管每个智能体都发出了在本地看似合理的响应。为了解决这一挑战，我们提出了 extit{BRIDGE}，一个可验证的运行时控制框架，将交接修复视为对显式依赖图的选择性干预。 extit{BRIDGE}的核心是一个集成的故障归因和校准管道 extit{dataset}，它解耦执行依赖关系，并注入反事实的交接扰动，以为运行时干预提供经验监督。利用这一诊断基础， extit{BRIDGE}动态地将可观察的交互轨迹编译成一个来源图，以确定未解决的依赖关系，而无需访问私有智能体状态。然后，它通过一个经验门评估候选修复，权衡预期收益与干扰风险，仅在正式的局部后置条件成立时执行修复。在严格的基准测试中，广泛评估表明， extit{BRIDGE}在SILO-BENCH上将任务成功率提高了高达12.3个百分点，在TeamBench上提高了11.1个百分点。在这两个基准上， extit{BRIDGE}保持严格的成本效益，边际令牌开销仅增加2.0 ext%。此外，在来自MAST基准的562个与交接相关的轨迹上， extit{BRIDGE}在单次通过中实现了69.8 ext{F1}，为可验证的协调修复建立了可靠的操作边界。

View on arXiv Download PDF AI Translation

cs.CL / 181 / 2610.17810

Perturbation, Not Provenance: Why Activation Steering Does Not Make a Robust Text Watermark

扰动，而非来源：为何激活引导并未形成稳健的文本水印

Pletenev, Sergey

Abstract

Linear directions in the activation space of large language models causally and universally influence model behavior. This suggests that they might serve as a basis for a robust text watermark. The idea is to steer the generation along a specific semantic direction and then recover the mark by probing the activations of the re-encoded text. We test this hypothesis by conducting a series of experiments. We vary the strength of the steering and use four different directions (formality, sentiment, complexity, and a random control) across two different model families. We also test two paraphrasing attacks: a sentence-level one and a stronger paragraph-level one. The results show that activation-steering watermarks are outperformed by a standard logit-based watermark (KGW) in terms of detection accuracy. At a matched level of fluency (within 1.15 times the original text's perplexity), KGW achieved an AUC of 0.81 against a sentence-level paraphraser, while the best steering direction only reached 0.65. A stronger paragraph-level paraphraser lowered both (0.65 for KGW and 0.59 for steering), narrowing the gap but not reversing it. On Qwen, it left the two indistinguishable but at near-chance levels for both. We further showed that the detectability provided by steering was not semantic: a random direction was just as detectable as a meaningful one. The signal was independent of which layer was probed and whether a surface feature control was used. We conclude that, in this case, semantic activation does not encode a localized, recoverable semantic mark, but rather a global, content-independent shift that can only be strongly detected when the text is degraded.

Chinese Translation

大型语言模型的激活空间中的线性方向因果且普遍地影响模型行为。这表明它们可能作为稳健文本水印的基础。其思路是沿着特定的语义方向引导生成，然后通过探测重新编码文本的激活来恢复水印。我们通过进行一系列实验来检验这一假设。我们改变引导的强度，并在两种不同的模型家族中使用四个不同的方向（正式性、情感、复杂性和随机控制）。我们还测试了两种释义攻击：一种是句子级的，另一种是更强的段落级的。结果表明，激活引导水印在检测准确性方面不如标准的基于对数几率的水印（KGW）。在流畅性相匹配的情况下（在原始文本的困惑度的1.15倍以内），KGW在句子级释义者面前达到了0.81的AUC，而最佳引导方向仅达到了0.65。更强的段落级释义者使两者均降低（KGW为0.65，引导为0.59），缩小了差距但未能逆转。在Qwen上，两者在近乎随机的水平下不可区分。我们进一步表明，引导所提供的可检测性并非语义性的：随机方向的可检测性与有意义的方向相当。信号独立于探测的层次以及是否使用了表面特征控制。我们得出结论，在这种情况下，语义激活并未编码一个局部的、可恢复的语义标记，而是一个全球性的、与内容无关的偏移，只有在文本退化时才能被强烈检测到。

View on arXiv Download PDF AI Translation

cs.CL / 182 / 2610.17868

Prefilling-dLLM v2: Learned Chunk Routing for Long-Context Diffusion Language Models

Prefilling-dLLM v2：用于长上下文扩散语言模型的学习块路由

Han, Qi, Xiong, Jing, Zhang, Ziyuan, Wong, Ngai

Abstract

Diffusion language models apply bidirectional attention over the whole context at every denoising step, so a longer context makes each step substantially more expensive. Sparsifying only at inference leaves the pretrained weights unchanged, so the model is never trained under the sparse pattern it runs with. We propose Prefilling-dLLM v2, which continues training a diffusion language model so that it routes with its native mask token. One mask is appended at the end of each chunk, and a learned role offset marks it as a routing slot. On every fourth layer, a query keeps a bidirectional local window and the top-K chunks scored from that mask; the other layers use a bidirectional sliding window. Starting from Dream-v0-Base-7B extended to 16K with YaRN, we compare Prefilling-dLLM v2 with bidirectional NSA, DSA, and SWA under the same layout, key budget, and dense initialization, using the same Fast-dLLM decoder. With the YaRN factor scaled to the evaluation length, Prefilling-dLLM v2 leads the other sparse methods on all three RULER tasks from 16K to 128K. Its mean accuracy at 128K is 7.4 points above NSA, the next sparse model, while DSA and SWA remain near zero. At the same length, decoding is 172x faster than with full dense attention.

Chinese Translation

扩散语言模型在每个去噪步骤中对整个上下文应用双向注意力，因此更长的上下文使得每一步的计算开销显著增加。仅在推理时进行稀疏化会使预训练权重保持不变，因此模型从未在其运行时的稀疏模式下进行训练。我们提出了Prefilling-dLLM v2，它继续训练扩散语言模型，使其能够使用其原生的掩码标记进行路由。每个块的末尾附加一个掩码，并且一个学习的角色偏移量将其标记为路由槽。在每第四层中，一个查询保持双向局部窗口和从该掩码中评分的前K个块；其他层使用双向滑动窗口。从Dream-v0-Base-7B扩展到16K的YaRN开始，我们在相同的布局、关键预算和稠密初始化下，将Prefilling-dLLM v2与双向NSA、DSA和SWA进行比较，使用相同的Fast-dLLM解码器。随着YaRN因子按评估长度进行缩放，Prefilling-dLLM v2在16K到128K的所有三个RULER任务中领先于其他稀疏方法。在128K时，其平均准确率比下一个稀疏模型NSA高出7.4个百分点，而DSA和SWA则接近零。在相同长度下，解码速度比全稠密注意力快172倍。

View on arXiv Download PDF AI Translation

cs.CL / 183 / 2610.17880

Cross-Lingual Transfer of Evil Vectors Is Directionally Asymmetric

邪恶向量的跨语言转移具有方向性不对称性

Kim, Juann, Kang, Minku

Abstract

Persona vectors are linear directions in the residual-stream activations of LLMs that encode behavioral traits and enable fine-grained control over model behavior. Among them, the evil persona vector is particularly relevant to understanding harmful behavior, jailbreak vulnerabilities, and the unlearning of harmful knowledge. Yet whether this vector reflects a genuinely cross-lingual concept, rather than an artifact of its extraction language, remains untested. We conduct a systematic evil-vector transfer study across two high-resource languages (English, Spanish) and two low-resource languages (Swahili, Telugu) on four open-source instruction-tuned LLMs, and find that transfer is directionally asymmetric. Evil vectors are harder to obtain reliably from low-resource languages, and the resulting vectors are largely ineffective at steering, even within their own language. In contrast, vectors extracted from high-resource languages transfer effectively to low-resource inputs while largely preserving general knowledge, as measured by MMLU. Geometric analysis further supports this pattern: evil vectors from the two high-resource languages are closely aligned, whereas low-resource vectors align only weakly with this shared direction. These results suggest that a shared cross-lingual evil direction exists and, once reliably extracted from a high-resource language, can be reused to steer behavior in low-resource languages without additional per-language data collection.

Chinese Translation

个性向量是大规模语言模型（LLMs）残差流激活中的线性方向，编码行为特征并实现对模型行为的细粒度控制。其中，邪恶个性向量与理解有害行为、越狱漏洞以及有害知识的遗忘尤为相关。然而，这个向量是否反映了真正的跨语言概念，而非其提取语言的伪影，尚未得到验证。我们在四个开源指令调优的LLMs上，对两种高资源语言（英语、西班牙语）和两种低资源语言（斯瓦希里语、泰卢固语）进行了系统的邪恶向量转移研究，发现转移具有方向性不对称性。从低资源语言中可靠地获取邪恶向量更为困难，且所得到的向量在其自身语言中也大多无效。相比之下，从高资源语言提取的向量能够有效转移到低资源输入，同时在MMLU测量中大致保留一般知识。几何分析进一步支持了这一模式：来自两种高资源语言的邪恶向量紧密对齐，而低资源向量与这一共享方向的对齐程度较弱。这些结果表明，存在一个共享的跨语言邪恶方向，并且一旦从高资源语言中可靠提取，可以在低资源语言中重用以引导行为，而无需额外的每语言数据收集。

View on arXiv Download PDF AI Translation

cs.CL / 184 / 2610.17894

Swiss4Arg: A Parallel Dataset for Multilingual Argument Mining across Swiss National Languages

Swiss4Arg：一个用于瑞士国家语言间多语言论证挖掘的平行数据集

Conrad, Sophia, Rezkellah, Fatma-Zohra, Niklaus, Christina, Gubelmann, Reto

Abstract

We present Swiss4Arg, a multilingual dataset for argument mining based on official Swiss voting brochures, covering all four national languages: German, French, Italian, and Romansh (a Romance minority language exclusively spoken in Switzerland). The brochures contain explicitly formulated arguments systematically organized into pro and con sections, resulting in a cross-lingually aligned resource with balanced argumentative stance. We create a dataset that comprises 53 federal voting issues collected between 2019 and 2025, annotated for argument components (major claims, claims, premises, and non-argumentative units) and relations (support and attack) by three native speakers per language. Inter-annotator agreement (Krippendorff's alpha) reaches 0.548 for component classification and 0.623 for relation and target classification, computed jointly across all annotators on the parallel data. We release both adjudicated labels and individual annotator annotations, enabling analyses of disagreement and cross-linguistic variation, and provide baseline results using GPT-5.4 and SwissBERT. Swiss4Arg establishes a novel benchmark for multilingual argument mining, with a particular focus on cross-lingual comparability and low-resource settings such as Romansh. The dataset and code are publicly available at https://anonymous-hf.up.railway.app/a/gozw0nion72y/ (HuggingFace) and https://anonymous.4open.science/r/MultiVote-03BD/ (GitHub).

Chinese Translation

我们提出了Swiss4Arg，这是一个基于瑞士官方投票手册的多语言论证挖掘数据集，涵盖所有四种国家语言：德语、法语、意大利语和罗曼什语（这种罗曼语族的少数语言仅在瑞士使用）。这些手册包含明确表述的论点，系统地组织为支持和反对部分，从而形成一个跨语言对齐的资源，具有平衡的论证立场。我们创建了一个数据集，包含2019年至2025年间收集的53个联邦投票议题，针对论证成分（主要主张、主张、前提和非论证单元）及其关系（支持和攻击）进行了注释，每种语言由三位母语者进行标注。标注者间一致性（Krippendorff's alpha）在成分分类中达到0.548，在关系和目标分类中达到0.623，这些指标是基于所有标注者在平行数据上的共同计算结果。我们发布了裁定标签和各个标注者的注释，便于分析分歧和跨语言变异，并提供了使用GPT-5.4和SwissBERT的基线结果。Swiss4Arg为多语言论证挖掘建立了一个新的基准，特别关注跨语言可比性和低资源环境，如罗曼什语。数据集和代码可在https://anonymous-hf.up.railway.app/a/gozw0nion72y/（HuggingFace）和https://anonymous.4open.science/r/MultiVote-03BD/（GitHub）上公开获取。

View on arXiv Download PDF AI Translation

cs.CL / 185 / 2610.17908

Text-Only vs. Image-Aware VLM Judges for Manga Translation Evaluation

仅文本与图像感知的视觉语言模型评估漫画翻译

Kimura, Subaru, Suzuki, Jun

Abstract

In manga translation, visual context outside speech-bubble text, including speaker identities, referents, and scene tone, is depicted in page images, making it natural to refer to images during translation evaluation. However, most conventional manga translation evaluations rely solely on text-only automatic metrics, leaving the impact of image input insufficiently explored. We propose a practical validation framework that employs the same vision-language model (VLM) as both a text-only judge (TOJ) and an image-aware judge (IAJ) to investigate evaluation divergences associated with image-aware evaluation procedures. We use a Japanese-English manga dataset of official translations aligned at the page and speech-bubble levels. First, adding page images to TOJ improved the primary VLM's detection of deliberately injected, visually dependent degradations. Second, we analyzed evaluation divergences between TOJ and IAJ, finding different error counts and error-category distributions. Human verification did not reveal a consistent preference between TOJ and IAJ.

Chinese Translation

在漫画翻译中，除了对话框文本之外的视觉上下文（包括说话者身份、指代对象和场景基调）在页面图像中得以呈现，因此在翻译评估过程中参考图像是自然的。然而，大多数传统的漫画翻译评估仅依赖于文本自动评估指标，导致图像输入的影响未得到充分探讨。我们提出了一种实用的验证框架，使用相同的视觉语言模型（VLM）作为文本评估者（TOJ）和图像感知评估者（IAJ），以研究与图像感知评估程序相关的评估差异。我们使用了一个日英漫画数据集，该数据集的官方翻译在页面和对话框级别上进行了对齐。首先，将页面图像添加到TOJ中改善了主要VLM对故意注入的、依赖视觉的降级的检测。其次，我们分析了TOJ和IAJ之间的评估差异，发现错误计数和错误类别分布存在不同。人工验证未显示TOJ和IAJ之间存在一致的偏好。

View on arXiv Download PDF AI Translation

cs.CL / 186 / 2610.17917

Batch-Paced: Fixed-Budget Batch Decoding in Discrete Diffusion Language Models

批量节奏：离散扩散语言模型中的固定预算批量解码

Zhang, Shuai, Zhou, Zhanchao, Qiu, Huachuan, Li, Anqi, Lan, Zhenzhong

Abstract

As generative language models become more capable and widely used, inference demand grows, making efficient serving increasingly important. Discrete diffusion language models (dLLMs) offer a promising path to more efficient inference through parallel token generation. Prior dLLM decoding methods, however, focus on improving inference efficiency for individual sequences. We formulate fixed-budget dLLM batch decoding and introduce Batch-Paced, a training-free method that jointly optimizes the allocation of decoding compute across requests to improve overall batch quality. Across two dLLM backbones, five benchmarks and their representative mixtures, Batch-Paced improves batch performance under matched budgets; gains generally grow with batch size and largely persist across decoding configurations. Analysis of requests' decoding trajectories suggests that these gains arise largely by raising the committed-token confidence of lower-confidence requests. We further show that Batch-Paced's batch decode time remains competitive and that adding a decoding-depth allowance during Batch-Paced decoding can shorten it further at a modest performance cost.

Chinese Translation

随着生成语言模型的能力不断提升和广泛应用，推理需求也随之增长，使得高效服务变得愈加重要。离散扩散语言模型（dLLMs）通过并行生成标记提供了一条更高效推理的有希望的路径。然而，先前的 dLLM 解码方法主要集中在提高单个序列的推理效率上。我们提出了固定预算 dLLM 批量解码的概念，并引入了 Batch-Paced，这是一种无训练的方法，能够共同优化解码计算在请求之间的分配，从而提高整体批量质量。在两个 dLLM 主干、五个基准及其代表性混合上，Batch-Paced 在匹配预算下提高了批量性能；随着批量大小的增加，性能提升通常会增大，并且在不同的解码配置中大致保持稳定。对请求解码轨迹的分析表明，这些提升主要是通过提高低置信度请求的承诺标记置信度而实现的。我们进一步展示了 Batch-Paced 的批量解码时间仍然具有竞争力，并且在 Batch-Paced 解码过程中增加解码深度的允许可以在适度的性能成本下进一步缩短解码时间。

View on arXiv Download PDF AI Translation

cs.CL / 187 / 2610.17953

Agent-Reflex: Teaching Language Agents to Act Reflectively

Agent-Reflex：教会语言代理进行反思性行动

Dong, Guanting, Yang, Xinyu, Jin, Jiajie, Hu, Yuyang, Song, Xiaoshuai, Chen, Yifei, Li, Xiaoxi, Tan, Jiejun, Zhu, Yutao, Chua, Tat-Seng, Wen, Ji-Rong, Dou, Zhicheng

Abstract

Large language models have evolved from single-turn chatbots into interactive agents capable of long-horizon reasoning. In real deployments, MCP tool services can return noisy feedback and users may intervene to correct the agent, but existing methods provide limited supervision for reflection learning and failure recovery. To bridge this gap, we propose Agent-Reflex, a general training framework for incentivizing proactive reflection in agents. Specifically, Agent-Reflex comprises three stages. First, Environment Reflection Skill Discovery (RSD) injects grounded tool-feedback perturbations into environment scaling, samples a capability-diverse LLM pool, mines matched success-failure pairs, and summarizes reusable reflection meta-skills. Building on this skill library, we then apply Reflective Supervised Fine-Tuning (ReSFT) that converts each skill into supervision for self-reflection, skill recovery, and user-centric correction, strengthening complementary reflective behaviors. Finally, we propose On-Policy Reflective Policy Optimization (ORPO) to combine agentic reinforcement learning under noisy interactions with reflection-guided on-policy distillation to further elicit proactive reflection. Built on Qwen3-8B, Agent-Reflex reaches 58.6 overall on five long-horizon tool-use benchmarks and 29.1 on AgentNoiseBench, surpassing strong environment-scaling and reflection baselines. Experiments on six challenging agent benchmark suites demonstrate the effectiveness of Agent-Reflex. Further analyses show that Agent-Reflex improves agents' robustness to noise while maintaining stable training dynamics, supporting scalable agent training. Our code is available at https://github.com/RUC-NLPIR/Agent-Reflex.

Chinese Translation

大型语言模型已经从单轮聊天机器人演变为能够进行长时间推理的互动代理。在实际部署中，MCP工具服务可能会返回噪声反馈，用户也可能会介入以纠正代理，但现有方法对反思学习和故障恢复提供的监督有限。为了解决这一问题，我们提出了Agent-Reflex，这是一个激励代理主动反思的通用训练框架。具体而言，Agent-Reflex包括三个阶段。首先，环境反思技能发现（RSD）将基于工具反馈的扰动注入到环境缩放中，抽样多样化的能力LLM池，挖掘匹配的成功-失败对，并总结可重用的反思元技能。在此技能库的基础上，我们应用反思监督微调（ReSFT），将每项技能转化为自我反思、技能恢复和以用户为中心的纠正的监督，从而增强互补的反思行为。最后，我们提出了基于策略的反思策略优化（ORPO），将代理强化学习与噪声交互结合，并通过反思引导的在线蒸馏进一步激发主动反思。基于Qwen3-8B，Agent-Reflex在五个长时间工具使用基准上总体得分达到58.6，在AgentNoiseBench上得分为29.1，超越了强大的环境缩放和反思基线。在六个具有挑战性的代理基准套件上的实验证明了Agent-Reflex的有效性。进一步分析表明，Agent-Reflex提高了代理对噪声的鲁棒性，同时保持了稳定的训练动态，支持可扩展的代理训练。我们的代码可在https://github.com/RUC-NLPIR/Agent-Reflex获取。

View on arXiv Download PDF AI Translation

cs.CL / 188 / 2610.17959

Virtual Head Attention

虚拟头注意力

Ding, Xibo, Wang, Guoxia, Zeng, Jinle, Yang, Jiabin, Tao, Sirui, Yang, Chao, Yu, Dianhai

Abstract

The design of attention mechanisms in long-context large language models faces an ``impossible triangle'' among KV cache size, training and prefill FLOPs, and model quality. GQA achieves strong quality but incurs high KV cache overhead; alternative approaches such as MLA and MFA can compress the cache, but MLA suffers from quality loss while MFA substantially increases training and prefill compute. We propose Virtual Head Attention (VHA), which simultaneously improves all three vertices through two lightweight linear operations: Q Premix applies a near-identity transformation to queries within each KV group to recover virtual head diversity from halved physical query heads, and Linear Postmix fuses inter-head features via a low-rank $I + AB^\top$ residual structure. Since VHA introduces no nonlinear operations, Premix can be folded into the query projection and Postmix into the output projection at inference time, reducing the model to a standard GQA-2 architecture that directly reuses all existing GQA inference optimizations. With only two KV heads, the group size in large models readily reaches 64, fully exploiting Tensor Core parallelism during decoding. Across four scales with 4K-context 50B-token pretraining, VHA consistently surpasses the standard GQA baseline at every scale ($+$0.05\% to $+$0.81\%), while reducing KV cache by $4\times$ vs.\ GQA-8 at dense scales and $2\times$ vs.\ GQA-4 at the MoE scale, and lowering training and prefill FLOPs by 7.9\% (1.7B, $S{=}4096$). On an industrial 30B-A3B MoE model trained on 500B tokens at 8K context, VHA outperforms both MLA and GQA-4 by $+$0.70\,/\,$+$2.60\,\% on a 12-benchmark base-model evaluation suite; in a separate matched 10B-A1B MoE comparison after long-context mid-training, VHA attains an average RULER score of 84.4 over 8K--128K versus 73.0 for GQA, while using two rather than four KV heads in the full-attention layers.

Chinese Translation

长上下文大语言模型中注意力机制的设计面临着KV缓存大小、训练和预填充FLOPs以及模型质量之间的“无法实现的三角形”。GQA实现了强大的质量，但带来了高KV缓存开销；而诸如MLA和MFA等替代方法可以压缩缓存，但MLA会导致质量损失，而MFA则显著增加训练和预填充计算。我们提出了虚拟头注意力（Virtual Head Attention, VHA），通过两种轻量级线性操作同时改善这三个方面：Q Premix对每个KV组中的查询应用近似恒等变换，以从减半的物理查询头中恢复虚拟头的多样性；Linear Postmix通过低秩$I + AB^ op$残差结构融合跨头特征。由于VHA不引入非线性操作，Premix可以在推理时折叠到查询投影中，Postmix可以折叠到输出投影中，从而将模型简化为标准的GQA-2架构，直接重用所有现有的GQA推理优化。在仅使用两个KV头的情况下，大模型中的组大小轻松达到64，在解码过程中充分利用Tensor Core并行性。在四个规模的4K上下文50B标记预训练中，VHA在每个规模上始终超越标准GQA基线（增加0.05 ext{%}至0.81 ext{%}），同时在密集规模下将KV缓存减少了4倍，相较于GQA-8，在MoE规模下减少了2倍，相较于GQA-4，并将训练和预填充FLOPs降低了7.9 ext{%}（1.7B，$S{=}4096$）。在一个工业级30B-A3B MoE模型上，该模型在8K上下文下训练了500B标记，VHA在一个12个基准的基础模型评估套件中超越了MLA和GQA-4，分别增加了0.70 ext{%}和2.60 ext{%}；在一次长上下文中训练后的匹配10B-A1B MoE比较中，VHA在8K至128K的平均RULER分数为84.4，而GQA为73.0，同时在全注意力层中使用两个而不是四个KV头。

View on arXiv Download PDF AI Translation

cs.CL / 189 / 2610.17989

Seeing Through the Forgery Is Not Enough: Regressive Sycophancy under Fabricated Evidence in Medical Vision-Language Models

看穿伪造并不足够：医疗视觉-语言模型下的倒退谄媚行为

Ahmed, Sadif, Biswas, Soumya Swagata, Shahriyar, Rifat, Haider, Chowdhury Mohammad Rakin

Abstract

A medical vision-language model may analyze a given CT scan and correctly detect both kidneys in view. Continuing this conversation, if a user provides a prescription with an opposite direction, medical VLMs often sycophantically choose to realign their diagnosis. In this work, we study regressive sycophancy under fabricated evidence, where a model abandons a correct answer when pushback comes with a fake document. We test four open VLMs on 400 VQA-RAD yes/no questions in 19,200 multi-turn conversations, with and without a fabricated prescription. Without a disclosure that documents are synthetic, the models abandon 80.1% of 6,488 correct answers. Notably, under one script each, the same prescription raises the flip rate of Qwen2.5-VL-7B from 45.9% to 89.6% and lowers that of Qwen3-VL-8B from 77.7% to 50.9%. A model rejected the prescription while still preserving the correct answer in 915 of 3,232 conversations with the prescription (28.3%), and 40% of these flipped in later turns. System-prompt sentences that disclose the documents as synthetic lower one model's flip rate from 77.8% to 47.3%. Thus, we conclude that awareness about a forgery does not prevent a model from adopting its answer.

Chinese Translation

医疗视觉-语言模型可以分析给定的CT扫描并正确检测到视野中的两个肾脏。继续这个话题，如果用户提供了一个相反方向的处方，医疗视觉-语言模型往往会谄媚性地选择重新调整他们的诊断。在本研究中，我们探讨了在伪造证据下的倒退谄媚行为，即当面对伪造文件的反对时，模型放弃正确答案的情况。我们在19,200个多轮对话中对400个VQA-RAD是/否问题测试了四个开放的视觉-语言模型，包含和不包含伪造处方。在没有披露文件为合成的情况下，模型放弃了6,488个正确答案中的80.1%。值得注意的是，在同一处方下，Qwen2.5-VL-7B的翻转率从45.9%上升到89.6%，而Qwen3-VL-8B的翻转率则从77.7%下降到50.9%。在3,232个包含处方的对话中，有915个对话中模型拒绝了处方，同时仍保留了正确答案（28.3%），其中40%的翻转发生在后续轮次中。系统提示句子披露文件为合成的情况下，某个模型的翻转率从77.8%降低到47.3%。因此，我们得出结论，意识到伪造并不能阻止模型采纳其答案。

View on arXiv Download PDF AI Translation

cs.CL / 190 / 2610.18014

LatentRemedy: Efficient Remedy For Non-Prefix KV Cache Reuse in Multi-Agent Systems

LatentRemedy：多智能体系统中非前缀KV缓存重用的高效解决方案

Chen, Zhiwei, Rao, Zhongtao, Xiu, Waikit, Zhang, Yuxiang, Cui, Shuting, Chen, Jian, Zhao, Tianxiang

Abstract

Non-prefix KV cache reuse is a promising approach for improving inference efficiency in multi-agent systems as agents share the same context blocks. However, when agents have conflicting tasks, directly reusing non-prefix KV caches across agents can severely degrade model performance, causing up to a 90% drop. Existing remedies, including selective recomputation and offset-based calibration, recover a limited fraction of this loss. We argue that the problem lies not only in contaminated KV representations but also in the agent's difficulty interpreting reused context under its own instructions. To this end, we introduce LatentRemedy, which uses a few task-conditioned latent tokens to help the target agent leverage reused context without recomputing the shared block. Specifically, we use an MLP to map the last hidden state of the target prefix into a small number of latent tokens, which are placed after the reused KV block to interact with the shared context. To guide subsequent predictions toward the target task, we train the MLP through full-prefill distillation. Experiments show that LatentRemedy consistently outperforms existing reuse baselines on conflicting tasks while maintaining TTFT close to direct reuse as cache size grows. Code and data are available at https://github.com/chenzhwsysu57/kvreuse.

Chinese Translation

非前缀KV缓存重用是一种有前景的方法，可以提高多智能体系统中的推理效率，因为智能体共享相同的上下文块。然而，当智能体面临冲突任务时，直接在智能体之间重用非前缀KV缓存可能会严重降低模型性能，导致高达90%的性能下降。现有的解决方案，包括选择性重新计算和基于偏移的校准，仅能恢复这一损失的一小部分。我们认为，问题不仅在于被污染的KV表示，还在于智能体在其自身指令下难以解读重用的上下文。为此，我们提出了LatentRemedy，它利用少量任务条件的潜在标记，帮助目标智能体在不重新计算共享块的情况下利用重用的上下文。具体而言，我们使用多层感知机（MLP）将目标前缀的最后隐藏状态映射为少量潜在标记，这些标记被放置在重用的KV块之后，以便与共享上下文进行交互。为了引导后续预测朝向目标任务，我们通过全前填充蒸馏训练MLP。实验表明，LatentRemedy在冲突任务上始终优于现有的重用基线，同时在缓存大小增加时保持与直接重用相近的TTFT。代码和数据可在https://github.com/chenzhwsysu57/kvreuse获取。

View on arXiv Download PDF AI Translation

cs.CL / 191 / 2610.18019

Machine--human agreement in marking undergraduate assignments is specific to model, prompt and institution

机器与人类在标记本科作业中的一致性特定于模型、提示和机构

Corsi, Giulio, Abo-Tabik, Maryam, Benn, Yael, Apreutesei, Georgiana Thorpe, Tibon, Roni, Marcoci, Alexandru, Talmi, Deborah

Abstract

Large language models (LLMs) promise to automate one of higher education's most resource-intensive activities: assessing student writing. Existing evaluations are typically limited to a single model, a single prompt and a single institution. Using 761 undergraduate psychology assignments from three UK universities, we evaluated three frontier LLMs, individually and in ensemble, on accuracy, prompt sensitivity, error independence, generalisability and evidential basis. Agreement between human and machine marks varied widely, with no model or prompt optimal across institutions. All models compressed marks towards a common central tendency, overvaluing the weakest assignments and undervaluing the strongest. Because this error was shared, ensembling failed to remove it. Machine marks also rewarded linguistic features more strongly than human marks. Since human-machine agreement was a joint product of model, prompt and institution, we conclude that current LLMs lack evaluative capacity transferable across contexts, and that single-context validation overstates readiness for deployment.

Chinese Translation

大型语言模型（LLMs）有望自动化高等教育中最耗费资源的活动之一：评估学生写作。现有评估通常限于单一模型、单一提示和单一机构。我们使用来自三所英国大学的761份本科心理学作业，评估了三种前沿LLM，分别在准确性、提示敏感性、错误独立性、可推广性和证据基础上进行评估。人类与机器评分之间的一致性差异很大，没有任何模型或提示在各机构中表现最佳。所有模型的评分都向共同的中心趋势压缩，过高估计了最弱作业的分数，而低估了最强作业的分数。由于这种错误是共享的，集成方法未能消除它。机器评分对语言特征的奖励也比人类评分更强。由于人机一致性是模型、提示和机构的共同产物，我们得出结论，当前的LLM缺乏可在不同环境中转移的评估能力，单一环境的验证夸大了其部署的准备程度。

View on arXiv Download PDF AI Translation

cs.CL / 192 / 2610.18060

A Benchmark for Bangla Task-Oriented Dialogue Understanding

孟加拉任务导向对话理解基准

Lamia, Afrina Alam, Sharif, Sumiaya, Modak, Prottoy, Rayhan, Ohiduzzaman, Sadeq, Nafis

Abstract

Task-oriented dialogue understanding remains underexplored for Bangla, particularly in multilingual settings involving English and romanized Bangla (Banglish). We introduce BanglaTOD, a parallel tri-lingual benchmark comprising 209 multi-turn dialogues across nine Bangladesh-specific service domains and 64 intents. Each conversational turn is aligned across Bangla, English, and Banglish, yielding 4,689 language-specific turn records annotated with intents, token-level slots, state updates, and accumulated dialogue states. We establish context-aware baselines for joint intent classification and slot filling using mBERT, BERT-base-uncased, and BanglaBERT. Cross-language analysis reveals substantial performance differences across models and linguistic representations, while removing dialogue history reduces intent accuracy by 19.0--25.7 percentage points. BanglaTOD provides a foundation for multilingual dialogue understanding and future dialogue state tracking research in Bangladesh-specific conversational applications. The dataset is publicly available at \href{https://huggingface.co/datasets/afri7here/BanglaTOD}{Hugging Face}.

Chinese Translation

任务导向对话理解在孟加拉仍然未得到充分探索，尤其是在涉及英语和罗马化孟加拉语（Banglish）的多语言环境中。我们引入了BanglaTOD，这是一个包含209个多轮对话的平行三语基准，涵盖九个孟加拉特定服务领域和64个意图。每个对话轮次在孟加拉语、英语和Banglish之间进行了对齐，共生成4,689个特定语言的轮次记录，标注了意图、令牌级槽位、状态更新和累积对话状态。我们使用mBERT、BERT-base-uncased和BanglaBERT建立了联合意图分类和槽位填充的上下文感知基线。跨语言分析显示，不同模型和语言表示之间存在显著的性能差异，而移除对话历史会使意图准确率降低19.0至25.7个百分点。BanglaTOD为多语言对话理解和未来在孟加拉特定对话应用中的对话状态跟踪研究提供了基础。该数据集已在 exttt{https://huggingface.co/datasets/afri7here/BanglaTOD} 上公开发布。

View on arXiv Download PDF AI Translation

cs.CL / 193 / 2610.18069

From Agents to Edges: Paired Optimization for Multi-Agent Collaboration

从代理到边缘：多代理协作的配对优化

Lin, Feng, Yu, Xiaotian, Yang, Yifan, Li, Aijun

Abstract

Strong individual agents do not guarantee effective collaboration. Communication must align what one agent conveys with how another interprets and uses it. This dependency makes the communication edge a natural unit of optimization. We introduce Paired Automated Communication Optimization (PACO), which jointly optimizes the instructions at both ends of each edge and their shared communication convention in systems built on black-box language models. PACO keeps model weights, agent roles, and workflow topology fixed. A state-aware discounted bandit selects edges for refinement, while execution feedback guides paired revisions evaluated through staged end-to-end execution. The optimized instructions integrate into existing agent prompts without additional model calls at deployment. On a four-agent workflow evaluated across diverse multi-agent benchmarks, PACO consistently outperforms leading prompt-optimization and communication baselines. Configurations learned on HotpotQA retain their advantage in cross-dataset transfer without target-specific optimization. These results demonstrate that optimizing how agents communicate can substantially improve collaboration within an otherwise fixed system.

Chinese Translation

强大的个体代理并不保证有效的协作。沟通必须使一个代理所传达的信息与另一个代理的解释和使用方式相一致。这种依赖性使得沟通边缘成为优化的自然单元。我们提出了配对自动化沟通优化（Paired Automated Communication Optimization，PACO），该方法共同优化每个边缘两端的指令及其共享的沟通约定，适用于基于黑箱语言模型构建的系统。PACO保持模型权重、代理角色和工作流拓扑不变。一个状态感知的折扣赌博者选择待优化的边缘，而执行反馈则指导通过分阶段的端到端执行评估的配对修订。优化后的指令在部署时可以无缝集成到现有的代理提示中，无需额外的模型调用。在一个跨多样化多代理基准评估的四代理工作流中，PACO始终优于领先的提示优化和沟通基线。在HotpotQA上学习到的配置在跨数据集转移中保持其优势，而无需针对特定目标的优化。这些结果表明，优化代理之间的沟通可以显著改善在其他固定系统内的协作。

View on arXiv Download PDF AI Translation

cs.CL / 194 / 2610.18121

Narrative Lookahead in LLMs: Event-Level Decodability, Predictive Advantage, and Steering

大型语言模型中的叙事前瞻：事件级可解码性、预测优势与引导

Jung, Donghoon, Kim, Min, Choi, Seojin, Chae, Songeun, Jung, Seohyon

Abstract

Do language models carry information about narrative events before writing them, and how far ahead? We examine narrative lookahead through salient, state-changing events rather than tokens or response-level attributes. We extract 23,355 LLM-annotated events from 2,000 stories generated by four open-weight LLMs, labeling them along three nonexclusive dimensions of change: within agents, between agents, and in the story world, with human validation on a subset. Event segmentation and plot-independent labels enable comparisons across stories at horizons measured in events rather than in tokens. At event ends, we probe hidden states for upcoming event types and compare predictions with text-only predictors using matched prefixes. Across four models, future event types are decodable above prevalence baselines for at least four subsequent events, but probes significantly outperform both primary text-only predictors only for the next event. In exploratory steering with OLMo-3.1-32B, a probe-derived direction shifts the proportion of the target event type in both intended directions at higher strengths. Our event-level approach links the prediction of future narrative developments to interventions that reshape subsequent event composition.

Chinese Translation

语言模型在撰写叙事事件之前是否携带有关这些事件的信息？其前瞻能力又有多远？我们通过显著的、状态变化的事件而非标记或响应级属性来考察叙事前瞻。我们从四个开放权重的大型语言模型生成的2000个故事中提取了23,355个经过大型语言模型标注的事件，并在三个非排他性的变化维度上对其进行标注：在代理内部、代理之间以及在故事世界中，并对一个子集进行了人工验证。事件分割和与情节无关的标签使得我们能够在以事件为单位的视角下进行跨故事比较，而非以标记为单位。在事件结束时，我们探查隐藏状态以预测即将到来的事件类型，并使用匹配前缀将预测与仅基于文本的预测进行比较。在四个模型中，未来事件类型的可解码性超过了至少四个后续事件的普遍基线，但探查的结果仅在下一个事件上显著优于主要的仅基于文本的预测。在对OLMo-3.1-32B的探索性引导中，基于探查的方向在更高强度下改变了目标事件类型的比例，朝着预期的两个方向进行调整。我们的事件级方法将未来叙事发展的预测与重塑后续事件组成的干预措施联系起来。

View on arXiv Download PDF AI Translation

cs.CL / 195 / 2610.18122

Cheap, Consistent, Calibrated, but Biased? Typed Decision Models as Judges

廉价、一致、经过校准，但有偏见？类型化决策模型作为裁判

Lukassen, Fabian, Weisser, Christoph, Silbersdorff, Alexander

Abstract

Typed decision models (TDMs) return a probability for each allowed answer instead of generating text, which makes them candidates for cheaply replacing LLM judges that compare two answers. Whether they share the biases of LLM judges is untested. We run the CALM benchmark's bias tests on eight TDMs and seven LLM judges of 2024 and 2026. Every TDM is biased, in three to eight of twelve tests. Emotional tone affects them most, and rewriting the better answer in such a tone lowers the accuracy of a typical TDM from 99% to 42%. TDMs are more biased than 2026 LLM judges. All eight change verdicts when the answers swap places, which the 2026 LLM judges hardly do, and several side with an answer that most people allegedly prefer or that a famous model allegedly wrote. Yet both kinds of judge are tripped up by the same prompts. The TDMs' probabilities show that most tests mainly flip verdicts the TDM was unsure of, and that good calibration does not protect against bias. As TDMs are more consistent and 37 times cheaper than 2026 LLM judges, our results suggest they can be attractive judges if their biases are tested for the task.

Chinese Translation

类型化决策模型（Typed Decision Models, TDMs）为每个允许的答案返回一个概率，而不是生成文本，这使它们成为廉价替代比较两个答案的LLM裁判的候选者。然而，它们是否与LLM裁判共享偏见尚未得到验证。我们对2024年和2026年的八个TDM和七个LLM裁判进行了CALM基准的偏见测试。每个TDM在十二个测试中都有三到八个测试显示出偏见。情感语调对它们的影响最大，而将较好的答案重写为这种语调会使典型TDM的准确率从99%降至42%。TDM的偏见程度高于2026年的LLM裁判。当答案位置互换时，所有八个TDM的裁决都会改变，而2026年的LLM裁判几乎不会这样做，且有几个TDM偏向于大多数人声称更喜欢的答案或声称由著名模型撰写的答案。然而，这两种裁判在相同的提示下都会受到困扰。TDM的概率显示，大多数测试主要翻转了TDM不确定的裁决，而良好的校准并不能保护其免受偏见。由于TDM比2026年的LLM裁判更一致且便宜37倍，我们的结果表明，如果对其偏见进行任务测试，TDM可能成为有吸引力的裁判。

View on arXiv Download PDF AI Translation

cs.CL / 196 / 2610.18140

SparPR: Runtime KV-Cache Recovery for Resource-Constrained LLM Inference on Consumer Hardware

SparPR：面向资源受限的消费者硬件上大语言模型推理的运行时 KV 缓存恢复

Deng, Jieyun, Hu, Keming, Liu, Yanghe, Feng, Li

Abstract

Running large language models (LLMs) on consumer hardware is bottlenecked by the key-value (KV) cache, which grows with every decoded token and quickly exhausts the memory and bandwidth of a single consumer GPU. We present SparPR (Sparse Partial Recomputation), a runtime KV-cache recovery mechanism that does not modify model weights and requires no offline training. SparPR couples three online decisions: Query-Aware key-block selection scores historical blocks against the current query; dynamic budget control estimates per-block copy and recompute costs to decide how many blocks to retain; and on-demand local recomputation with cost-aware fallback merges retained blocks into continuous spans and switches between sparse recomputation and full recovery according to a runtime cost comparison. Implemented on FlexGen/FlexLLMGen, SparPR is evaluated on a single consumer laptop (RTX 3070 Ti 8GB, Ryzen 9 6900HX, 16GB RAM). It cuts end-to-end latency by 31%, 52%, and 63% at 1k, 2k, and 4k prompt tokens respectively, with 1.4x-2.7x higher throughput, and completes an 8k-token workload in 44 minutes where the FlexGen baseline times out. Under a fixed cache budget, SparPR retains a Needle Accuracy that stays well above StreamingLLM and H2O at every keep ratio, and ablation studies confirm that the three mechanisms act jointly.

Chinese Translation

在消费者硬件上运行大语言模型（LLMs）受到关键值（KV）缓存的瓶颈限制，随着每个解码的标记，缓存不断增长，迅速耗尽单个消费者 GPU 的内存和带宽。我们提出了 SparPR（稀疏部分重计算），这是一种运行时 KV 缓存恢复机制，不需要修改模型权重，也不需要离线训练。SparPR 结合了三种在线决策：查询感知的关键块选择根据当前查询评估历史块；动态预算控制估算每个块的复制和重计算成本，以决定保留多少块；按需本地重计算与成本感知的后备合并保留的块为连续区间，并根据运行时成本比较在稀疏重计算和完全恢复之间切换。在 FlexGen/FlexLLMGen 上实现的 SparPR，在一台单一的消费者笔记本电脑（RTX 3070 Ti 8GB，Ryzen 9 6900HX，16GB RAM）上进行了评估。在 1k、2k 和 4k 提示标记下，SparPR 分别减少了 31%、52% 和 63% 的端到端延迟，吞吐量提高了 1.4 倍至 2.7 倍，并在 44 分钟内完成了 8k 标记的工作负载，而 FlexGen 基线则超时。在固定缓存预算下，SparPR 保持了 Needle Accuracy，始终高于 StreamingLLM 和 H2O 的各个保留比率，消融研究确认这三种机制共同作用。

View on arXiv Download PDF AI Translation

cs.CL / 197 / 2610.18151

Attacking Only Half the Data: Two-Sided Evaluation of AI-Text Detector Robustness

仅攻击一半数据：AI文本检测器鲁棒性的双面评估

Shill, Apu Chandraw, Nandi, Rabindra Nath

Abstract

We investigate the robustness of widely used AI text detectors under attacks applied to both human and machine text. Prior evaluations typically attack only machine text, overlooking that human text is routinely modified through grammar correction, formatting, or invisible character insertion. We introduce a two-sided protocol that applies attacks to both classes. Across two corpora and nine detectors, reported AUROC gains vanish under two-sided evaluation, with attacked human text misclassified at rates up to 100%. While simple preprocessing reverses character-level attacks, word-level synonym substitution remains effective meaning-preserving, and irreversible.

Chinese Translation

我们研究了在对人类文本和机器文本同时施加攻击时，广泛使用的AI文本检测器的鲁棒性。以往的评估通常只针对机器文本进行攻击，忽视了人类文本在语法纠正、格式调整或不可见字符插入等过程中常常被修改的事实。我们提出了一种双面协议，对这两类文本均施加攻击。在两个语料库和九个检测器的评估中，报告的AUROC增益在双面评估下消失，被攻击的人类文本的误分类率高达100%。尽管简单的预处理可以逆转字符级攻击，但词级同义词替换仍然有效且保持意义，且不可逆。

View on arXiv Download PDF AI Translation

cs.CL / 198 / 2610.18157

Fan Out or Relay? Ownership-Aware Routing for LLM Agents Querying Siloed Tables

扇出还是中继？面向所有权的路由方法用于大型语言模型代理查询孤立表

Feng, Zhexi

Abstract

Inside organizations, a question's data often sit with several owners, who query their own tables but never hand them over. An LLM coordinator can fan out, sending the question to every owner at once, or relay, querying owners in turn and reusing earlier rows. We show that neither is safe in advance, because data ownership decides which one is right. Across thirteen settings on PeopleJoin and BIRD with three model families, fan-out is more accurate in twelve on questions one owner can answer, yet answers at most 4.3% of those joining two owners' tables. The flip follows ownership, not query shape: on BIRD two-table joins, fan-out answers 32-44% when one owner holds both and 1-4% otherwise. Relay wins by conditioning on returned rows, and hiding them from an otherwise identical planner cuts two-owner accuracy from 23.2% to 5.1%. Since ownership is public, we propose Ownership-Aware Routing (OAR), which drafts a schema-only SQL query, counts the owners whose tables it joins, and routes the question before any private row is read. In ten settings OAR is never significantly less accurate than the best fixed protocol, and beats every fixed protocol on PeopleJoin, by up to 9.3 points on held-out organizations.

Chinese Translation

在组织内部，问题的数据通常由多个所有者持有，他们查询自己的表格但从不交出这些数据。大型语言模型（LLM）协调器可以选择扇出，将问题同时发送给每个所有者，或者选择中继，依次查询所有者并重用先前的行。我们展示了这两种方法在事先并不安全，因为数据所有权决定了哪种方法是正确的。在 PeopleJoin 和 BIRD 的十三个设置中，针对三种模型家族，扇出在十二个情况下对一个所有者能够回答的问题更为准确，但在连接两个所有者的表时最多只能回答 4.3% 的问题。准确性变化与所有权相关，而非查询形状：在 BIRD 的两个表连接中，当一个所有者持有两个表时，扇出能回答 32-44%，而在其他情况下仅为 1-4%。中继通过对返回的行进行条件处理而获胜，而将这些行隐藏在一个其他条件相同的规划者中，使得两个所有者的准确率从 23.2% 降至 5.1%。由于所有权是公开的，我们提出了面向所有权的路由方法（Ownership-Aware Routing, OAR），该方法草拟一个仅包含模式的 SQL 查询，计算其连接的所有者数量，并在读取任何私有行之前路由问题。在十个设置中，OAR 的准确性从未显著低于最佳固定协议，并且在 PeopleJoin 上超越每个固定协议，最多提高 9.3 个百分点，针对保留的组织。

View on arXiv Download PDF AI Translation

cs.CL / 199 / 2610.18159

Claim Framing as a Driver of LLM Sycophancy

主张框架作为大型语言模型谄媚行为的驱动因素

Seo, Jean, Atanasova, Pepa, Augenstein, Isabelle

Abstract

Sycophancy in Large Language Models, the tendency to align responses with user preferences over ground truth, has largely been studied using questions without a clear ground truth and with a single, plainly stated user opinion. We present a systematic study of which framings of a user's claim trigger sycophancy in factual question answering, where every answer is objectively verifiable without extra context. Evaluating 14 models on 42K questions, we express correct and incorrect user opinions through various claim framings over 8 dimensions and 21 types, and measure a broad set of opinion-driven behaviors. We find that all models are susceptible to user opinions regardless of size. Models are most influenced by how confidently an opinion is stated, whom it is attributed to, and whether it is presented as shared between the user and the model. In contrast to prior work, framings based on the user's expertise, authority over the model, or social pressure matter little in our factual setting. Larger models resist incorrect opinions better but are no better at recovering from their own errors. Our results call for sycophancy evaluations that probe how an opinion is framed, not merely whether one is present, and that measure multiple behaviors beyond a single flip rate.

Chinese Translation

大型语言模型中的谄媚行为，即在回答中倾向于迎合用户偏好而非真实事实的倾向，主要通过没有明确真实答案的问题和单一、明确表述的用户意见进行研究。我们对用户主张的不同框架如何在事实问答中触发谄媚行为进行了系统研究，在这种情况下，每个答案都可以在没有额外上下文的情况下客观验证。我们在42,000个问题上评估了14个模型，通过8个维度和21种类型的不同主张框架表达正确和错误的用户意见，并测量了一系列基于意见驱动的行为。我们发现所有模型都容易受到用户意见的影响，无论其规模如何。模型最受影响的因素包括意见表述的自信程度、归属对象以及是否呈现为用户与模型之间的共识。与之前的研究相比，基于用户专业知识、对模型的权威性或社会压力的框架在我们的事实环境中影响较小。较大的模型对错误意见的抵抗能力更强，但在纠正自身错误方面并没有显著改善。我们的结果呼吁对谄媚行为的评估应探讨意见的框架，而不仅仅是是否存在，并且应测量多种行为，而不仅限于单一的翻转率。

View on arXiv Download PDF AI Translation

cs.CL / 200 / 2610.18173

TRACESTEER: Trace Driven Activation Steering for Hallucination Mitigation in Diffusion Language Models

TRACESTEER：基于追踪的激活引导框架用于缓解扩散语言模型中的幻觉现象

Zhang, Xin, Wang, Yili, Tan, Yue, He, Xin, Miao, Rui, Liu, Yixin, Chang, Yi, Wang, Xin

Abstract

Diffusion large language models (dLLMs) have emerged as a promising alternative to autoregressive LLMs, yet remain susceptible to hallucination. Activation steering offers a promising mitigation approach by guiding parallel token predictions through internal representations. However, existing steering methods largely overlook two characteristics of dLLMs: dynamic denoising states and temporal hallucination risks, which challenge predefined steering vectors and uniform intervention strengths. We propose TRACESTEER, a trace-driven activation steering framework that uses preceding denoising predictions as feedback to jointly adapt steering direction and strength. Specifically, Denoising Feedback Steering (DF-Steering) constructs step-specific steering vectors from preceding predictions and current hidden states. Trajectory-Guided Steering Gate (TS-Gate) regulates their strength using prediction uncertainty and the denoising timestep. With the backbone frozen, factual supervision and magnitude regularization jointly guide the learning of adaptive steering interventions, promoting factual reliability and stability throughout denoising. Experiments on two dLLMs show improved factual reliability on TruthfulQA and generalization across five downstream benchmarks.

Chinese Translation

扩散大型语言模型（dLLMs）作为自回归大型语言模型（LLMs）的有前景替代方案，然而仍然容易出现幻觉现象。激活引导提供了一种有希望的缓解方法，通过内部表征引导并行的标记预测。然而，现有的引导方法在很大程度上忽视了dLLMs的两个特征：动态去噪状态和时间幻觉风险，这对预定义的引导向量和统一的干预强度构成了挑战。我们提出了TRACESTEER，一个基于追踪的激活引导框架，利用先前的去噪预测作为反馈，联合调整引导方向和强度。具体而言，去噪反馈引导（DF-Steering）从先前的预测和当前的隐藏状态构建特定步骤的引导向量。轨迹引导引导门（TS-Gate）使用预测不确定性和去噪时间步调节其强度。在骨干网络被冻结的情况下，事实监督和幅度正则化共同引导自适应引导干预的学习，促进去噪过程中的事实可靠性和稳定性。在两个dLLMs上的实验显示，在TruthfulQA上提高了事实可靠性，并在五个下游基准上实现了更好的泛化能力。

View on arXiv Download PDF AI Translation

cs.CL / 201 / 2610.18216

TextualRL: Textual Reinforcement Learning for Context Management

TextualRL：用于上下文管理的文本强化学习

Huai, Tianyu, Zhou, Jie, Mao, Bo, Chen, Qin, He, Liang, Qiu, Xipeng

Abstract

LLM agents can adapt to task feedback by revising the textual context that guides their behavior, without changing their model parameters. However, turning execution rewards into reusable context updates is challenging: individual trajectories offer limited comparative evidence about what to change, while task-specific fixes may overgeneralize when incorporated into a shared context. We introduce TextualRL, a gradient-free reinforcement learning framework that treats a group of rollouts from the same task under the same context, rather than an isolated trajectory, as the unit of policy critique. Groupwise Policy Critique contrasts successful and failed rollouts within a task; when outcomes are uniform, it draws on cross-task evidence to identify behaviors worth preserving or formulate repair hypotheses. Cross-Group Policy Update reviews the resulting proposals against evidence pooled across groups, including observations that yield no proposed edit, to refine their content and applicability. Reviewed proposals form a bounded textual patch, with adoption of the candidate context determined by held-out validation rewards. Together, these components establish a reward-driven policy optimization loop in which multiple attempts jointly inform reusable context updates. We evaluate TextualRL with five target agents on six benchmarks. Averaged over the six benchmarks, TextualRL improves test accuracy over the base model by 24.4 and 15.3 percentage points for GPT-5.5 and Qwen3.8-27B, respectively.

Chinese Translation

大型语言模型（LLM）代理可以通过修订指导其行为的文本上下文来适应任务反馈，而无需更改其模型参数。然而，将执行奖励转化为可重用的上下文更新是具有挑战性的：单个轨迹提供的关于应更改内容的比较证据有限，而任务特定的修正在纳入共享上下文时可能会过度泛化。我们提出了TextualRL，这是一种无梯度强化学习框架，它将来自同一任务在相同上下文下的一组执行轨迹视为政策批判的单位，而不是孤立的轨迹。组内政策批判对比了任务内成功和失败的执行轨迹；当结果一致时，它利用跨任务的证据来识别值得保留的行为或制定修复假设。跨组政策更新则对照跨组汇总的证据审查所提出的建议，包括未提出编辑的观察，以细化其内容和适用性。经过审查的建议形成一个有限的文本补丁，候选上下文的采用由保留的验证奖励决定。这些组件共同建立了一个以奖励驱动的政策优化循环，其中多个尝试共同为可重用的上下文更新提供信息。我们在六个基准上对五个目标代理评估了TextualRL。在六个基准上平均而言，TextualRL使得GPT-5.5和Qwen3.8-27B的测试准确率分别提高了24.4和15.3个百分点。

View on arXiv Download PDF AI Translation

cs.CL / 202 / 2610.18247

PACT: A Pragmatic Prompting Framework for Expressive and Voice-Preserving Speech-to-Speech Translation

PACT：一种用于表达性和声音保留的语音到语音翻译的务实提示框架

Song, Yuchen, Nakamura, Satoshi

Abstract

Speech-to-speech translation should preserve not only what is said, but also who says it and how it is said. Existing end-to-end S2ST, speech MLLM, and cascade systems can preserve content or timbre, but often struggle to realize emotion, tone, rhythm, pauses, and pragmatic attitude in the target language. We propose PACT, a target-space pragmatic prompting framework for expressive and voice-preserving S2ST. PACT extracts source pragmatic cues, projects them into the target-language expression space, calibrates the projected representation to a realizable target subspace, and converts it into continuous prompts for a frozen speech generator. On bidirectional Spanish-English TOPI evaluation, PACT achieves the best overall performance among evaluated systems across content fidelity, voice preservation, and expression preservation metrics.

Chinese Translation

语音到语音翻译不仅应当保留所说内容，还应保留说话者及其表达方式。现有的端到端语音到语音翻译（S2ST）、语音多模态大语言模型（MLLM）和级联系统能够保留内容或音色，但往往难以在目标语言中实现情感、语调、节奏、停顿和务实态度。我们提出了PACT，一种用于表达性和声音保留的语音到语音翻译的目标空间务实提示框架。PACT提取源语言的务实线索，将其投影到目标语言的表达空间，校准投影表示以适应可实现的目标子空间，并将其转换为连续提示，以供冻结的语音生成器使用。在双向西班牙语-英语TOPI评估中，PACT在内容保真度、声音保留和表达保留指标上实现了评估系统中最佳的整体性能。

View on arXiv Download PDF AI Translation

cs.CL / 203 / 2610.18248

Are We Testing Too Much? A Survey of Benchmark Compression for Efficient Evaluation of Large Language Models

我们测试得太多了吗？大型语言模型高效评估的基准压缩调查

Ling, Guoming, Lin, Yupei, Huang, Zhongzhan, Li, Junxin, Liu, Junjie, Zhong, Shanshan, Wu, Hefeng, Lin, Liang

Abstract

Benchmarks are essential for evaluating and improving large language models (LLMs). Yet many widely used benchmarks are redundant, making evaluation unnecessarily expensive. Benchmark compression lowers this cost by estimating full-benchmark scores and rankings from a partial evaluation. Research in this area has grown rapidly since 2024, but to our knowledge no survey has systematically organized its methods and findings. We review 119 papers to explain how methods choose what to evaluate, recover full-benchmark results, and verify their reliability. From this analysis, we develop a taxonomy covering compression within individual benchmarks, across benchmark suites, and during adaptive evaluation, together with work on redundancy and reliability. We compare their information needs, total costs, reliability checks, and limitations. Methods reduce evaluation costs by selecting representative subsets, using cheaper or reusable information, and testing adaptively, but risk poor coverage, inaccurate estimates, and sequential delays. The resulting taxonomy and practical guide help researchers choose, validate, and deploy methods for more efficient and reliable LLM evaluation.

Chinese Translation

基准测试对于评估和改进大型语言模型（LLMs）至关重要。然而，许多广泛使用的基准测试存在冗余，使得评估成本不必要地高。基准压缩通过从部分评估中估算完整基准分数和排名来降低这一成本。自2024年以来，该领域的研究迅速增长，但据我们所知，尚无系统整理其方法和发现的调查。我们回顾了119篇论文，以解释方法如何选择评估内容、恢复完整基准结果以及验证其可靠性。通过这一分析，我们开发了一个分类法，涵盖了单个基准内的压缩、基准套件之间的压缩以及自适应评估中的压缩，同时还涉及冗余和可靠性方面的研究。我们比较了它们的信息需求、总成本、可靠性检查和局限性。方法通过选择代表性子集、使用更便宜或可重复使用的信息以及自适应测试来降低评估成本，但也面临覆盖不足、不准确估计和顺序延迟的风险。最终形成的分类法和实用指南帮助研究人员选择、验证和部署方法，以实现更高效和可靠的LLM评估。

View on arXiv Download PDF AI Translation

cs.CL / 204 / 2610.18269

PAINT: A Framework for Multilingual Pain Representation from Medical Licensing Exams to Real-World Clinical Texts

PAINT：一个用于从医学执照考试到现实世界临床文本的多语言疼痛表征框架

Zhang, Boya, Ma, Yuanchao, Wu, Jiageng, Chaisutyakorn, Krittaphas, Gameiro, Rodrigo, Han, Ahram, Huang, Chien-Yu, Jaiswal, Nikhil, Leite, Marianna, Mora-Mau, Miriam, Pineda-Moreno, Jaime, Placino, Siena, Yamamoto, Ryohei, Jia, Yugang, Yang, Jie, Celi, Leo Anthony, Teodoro, Douglas

Abstract

Pain is one of the most common and clinically important symptoms, reflecting disease onset, progression, and treatment response. However, pain is difficult to accurately capture and interpret from medical text, as its expression varies across languages, documentation types, and patient demographics, and is often embedded in terminology, negation, temporality, and contextual descriptions. Existing pain-related studies remain largely monolingual or limited to narrow clinical settings. In this study, we introduce PAINT, a multilingual framework for studying pain representation across medical exams and real-world clinical texts. PAINT includes nine pain-focused dataset partitions spanning English, Spanish, Portuguese, and Japanese, with clinician annotations in all four languages. We curated a structured schema covering pain relevance, pain presence, anatomical location, pain descriptors, and demographic information. To capture multilingual pain representations, we leverage large language models (LLMs) and adopt an ensemble strategy to improve performance. Ten clinicians across four languages validate the labels: the ensemble reaches macro-F1 of 0.91-1.00 on pain relevance and pain presence and 92-100% accuracy on demographics, with inter-annotator $\kappa$ of 0.82-1.00. We release the WorldMedQA-Pain ensemble labels, the extraction pipeline, validation scripts, and clinician annotation sheets for publicly licensed corpora, providing a foundation for pain-centered clinical NLP and for studying pain documentation across languages and healthcare contexts.

Chinese Translation

疼痛是最常见且临床上重要的症状之一，反映了疾病的发生、进展和治疗反应。然而，由于疼痛的表达在不同语言、文档类型和患者人口统计特征中存在差异，并且常常嵌入在术语、否定、时间性和上下文描述中，因此很难准确捕捉和解释医学文本中的疼痛。现有的疼痛相关研究大多仍然是单语的或局限于狭窄的临床环境。在本研究中，我们介绍了PAINT，一个用于研究医学考试和现实世界临床文本中疼痛表征的多语言框架。PAINT包括九个以疼痛为中心的数据集分区，涵盖英语、西班牙语、葡萄牙语和日语，并在这四种语言中都有临床医生的注释。我们策划了一个结构化的模式，涵盖疼痛相关性、疼痛存在、解剖位置、疼痛描述符和人口统计信息。为了捕捉多语言疼痛表征，我们利用大型语言模型（LLMs）并采用集成策略以提高性能。四种语言的十位临床医生验证了标签：集成模型在疼痛相关性和疼痛存在上的宏F1值达到0.91-1.00，人口统计信息的准确率为92-100%，注释者间的一致性$ ext{kappa}$值为0.82-1.00。我们发布了WorldMedQA-Pain集成标签、提取管道、验证脚本和临床医生注释表，以供公开许可的语料库使用，为以疼痛为中心的临床自然语言处理和跨语言及医疗保健背景下的疼痛文档研究提供了基础。

View on arXiv Download PDF AI Translation

cs.CL / 205 / 2610.18272

JevGPT: Thinking Fast and Slow in Autonomous Research

JevGPT：在自主研究中的快速与慢速思考

Sun, Yasheng, Dai, Yanning, Chen, Yimeng, Li, Nanbo, Laakom, Firas, Xu, Yifan, Piękos, Piotr, Ostaszewski, Mateusz, Wang, Wenyi, Luo, Cheng, Wang, Yuhui, Schmidhuber, Jürgen

Abstract

Autonomous research agents must acquire and reason over evidence within a limited budget. Yet relevant sources can remain uninspected or unused, leaving a gap between retrieval coverage and the evidence that informs research. The challenge is to align fast local judgments with changing evidence needs and feed their findings back into research decisions. We introduce JevGPT, a dual-system architecture for autonomous research that couples fast evidence processing with deliberative research planning. System Two sets evidence objectives and integrates findings, while System One locates support, assesses relevance, and guides further inspection. Independent source judgments run in parallel; selected original passages inform the next research step. Experiments on fixed-corpus question answering and document ranking show that JevGPT achieves leading task performance, with faster completion and lower cost than comparably effective research agents. Evidence analysis shows broader source inspection and greater coverage of supporting documents in the research loop, highlighting efficient evidence processing as a route to more capable autonomous research.

Chinese Translation

自主研究代理必须在有限的预算内获取和推理证据。然而，相关来源可能未被检查或使用，从而在检索覆盖与支持研究的证据之间留下了差距。挑战在于将快速的局部判断与不断变化的证据需求对齐，并将其发现反馈到研究决策中。我们介绍了JevGPT，这是一种自主研究的双系统架构，将快速证据处理与深思熟虑的研究规划相结合。系统二设定证据目标并整合发现，而系统一则定位支持、评估相关性并指导进一步检查。独立的来源判断并行运行；所选的原始段落为下一步研究提供信息。在固定语料库的问答和文档排名实验中，JevGPT实现了领先的任务性能，完成速度更快，成本低于同样有效的研究代理。证据分析显示了更广泛的来源检查和更大的支持文档覆盖，强调了高效的证据处理是实现更强大自主研究的途径。

View on arXiv Download PDF AI Translation

cs.CL / 206 / 2610.18299

LegalRewardBench: Benchmarking Reward Models for Legal Preference Evaluation

LegalRewardBench：法律偏好评估奖励模型的基准测试

Huang, Hui, Zhou, Jie, Zhou, Hongli, Shi, Qundong, Zheng, Shuyuan, Su, Yini, Wang, Shaochun, Lin, Biyuan, Yang, Muyun, Zhao, Tiejun

Abstract

Reward models are central to preference alignment for legal tasks, yet existing reward model benchmarks are still primarily centered on general-domain preferences. In this work, we introduce LegalRewardBench, the first human-verified benchmark for evaluating reward models in the legal domain. The benchmark organizes legal preference evaluation into five capability levels: knowledge, understanding, analysis, decision, and writing, and preference examples are constructed through rubric-guided automatic judging combined with human verification. We evaluate three families of judges, covering both general and fine-tuned models, and generative and discriminative architectures. Experimental results show that strong general-purpose generative models are currently the most robust legal reward models, while existing fine-tuned reward models transfer less reliably to legal preference evaluation. We also conduct detailed comparisons of input organizations and training strategies.

Chinese Translation

奖励模型在法律任务的偏好对齐中至关重要，但现有的奖励模型基准仍主要集中于一般领域的偏好。在本研究中，我们引入了LegalRewardBench，这是第一个经过人工验证的法律领域奖励模型评估基准。该基准将法律偏好评估组织为五个能力层级：知识、理解、分析、决策和写作，偏好示例通过基于评分标准的自动判断结合人工验证构建而成。我们评估了三类评判者，涵盖一般模型和微调模型，以及生成式和判别式架构。实验结果表明，强大的通用生成模型目前是最稳健的法律奖励模型，而现有的微调奖励模型在法律偏好评估中的迁移可靠性较低。我们还对输入组织和训练策略进行了详细比较。

View on arXiv Download PDF AI Translation

cs.CL / 207 / 2610.18308

SimInter: Evidence-Grounded User Simulation with Interview-Based Profile Enrichment for Conversational Recommender Systems

SimInter：基于证据的用户模拟与基于访谈的个人资料增强用于对话推荐系统

Shahabi, Erfan, Roshandel, Amirhossein, Sadeghi, Mohammad Sadegh, Shakery, Azadeh

Abstract

Conversational Recommender Systems (CRSs) require reliable user simulators for cost-effective evaluation. LLM-based simulators have made realistic simulation possible but typically condition on profiles derived from explicit ratings, demographics, and static sources capturing surface preferences while omitting implicit behavioral characteristics such as decision-making style, aspect-level preferences, and domain expertise. We propose SimInter, a pipeline that elicits these attributes via synthetic interviews between two LLM agents, grounded in the user's statistical profile and historical conversational evidence. Before profile enrichment, an audit agent explicitly flags potential fabrication and stereotyping at the turn level. We instantiate and evaluate SimInter on movie recommendation. We simulate 300 ReDial users and, on a matched subset of 100 users, compare SimInter with a statistics-only baseline: SimInter significantly increases the diversity of expressed preference aspects (filtered aspect entropy: 1.619 vs. 0.859; paired d = 0.96, Holm-adjusted p < 0.001). This gain persists when the simulator converses with two trained recommenders, KBRD and UniCRS, under two independent evaluation judges (+0.28 to +0.56 nats; all p < 0.001). Human evaluation finds 94.3% of enriched-attribute claims supported or partially supported, while human judgments agree with 95.2% of the auditor's binary filtering decisions on a stratified sample.

Chinese Translation

对话推荐系统（CRSs）需要可靠的用户模拟器以实现成本效益评估。基于大型语言模型（LLM）的模拟器使得现实模拟成为可能，但通常依赖于从显式评分、人口统计数据和静态来源中提取的个人资料，这些来源捕捉了表层偏好，而忽略了诸如决策风格、方面级偏好和领域专业知识等隐性行为特征。我们提出了SimInter，一个通过两个LLM代理之间的合成访谈来引出这些属性的流程，该流程基于用户的统计个人资料和历史对话证据。在个人资料增强之前，审计代理在回合级别明确标记潜在的虚构和刻板印象。我们在电影推荐上实例化并评估SimInter。我们模拟了300个ReDial用户，并在100个匹配子集中将SimInter与仅基于统计的基线进行比较：SimInter显著增加了表达偏好方面的多样性（过滤后的方面熵：1.619对0.859；配对d = 0.96，Holm调整后的p < 0.001）。当模拟器与两个训练过的推荐系统KBRD和UniCRS进行对话时，这一增益仍然存在，在两个独立的评估评委下（+0.28到+0.56 nats；所有p < 0.001）。人类评估发现94.3%的增强属性声明得到了支持或部分支持，而人类判断与审计员在分层样本上的二元过滤决策一致率为95.2%。

View on arXiv Download PDF AI Translation

cs.CL / 208 / 2610.18309

Prompt or Policy? Optimizing Physician Slate Presentations with LLM User Simulators

提示还是策略？使用大型语言模型用户模拟器优化医生推荐展示

Niculae, Andrei, Cosma, Adrian, Radoi, Emilian

Abstract

Choosing the right physician is a key challenge for patients seeking medical care. Physician recommendation systems identify relevant doctors, but patients still find it difficult to compare their profiles to make a decision. We formulate slate presentation optimization: given a patient's medical case and a fixed, ranked list of physicians, we generate short, case-specific summaries that make doctors easier to compare, while remaining grounded in the information in their profiles. To optimize physician presentations without relying on real patient interactions, we develop an offline framework that uses persona-conditioned LLMs to simulate patient choices. We generate patient personas from attributes extracted from each medical case and evaluate presentations for factual accuracy, relevance, naturalness and simulated patient choice. We also use a second LLM judge, unseen during optimization, to assess simulator-specific overfitting. We compare automatic prompt-level optimization with GEPA, parameter-level optimization with GRPO, and their combination across two model sizes. GRPO substantially outperforms GEPA in satisfying the full set of constraints, and its gains in simulated conversion persist under the unseen judge. GEPA alone brings smaller improvements, but using its optimized prompt to initialize GRPO speeds up training and achieves the highest simulated conversion. These results show that prompt-level and parameter-level optimization play complementary roles in improving how physician recommendations are presented.

Chinese Translation

选择合适的医生是寻求医疗服务的患者面临的关键挑战。医生推荐系统能够识别相关医生，但患者在比较医生的个人资料以做出决策时仍然感到困难。我们提出了推荐展示优化的概念：在给定患者的医疗案例和固定的、排名的医生列表的情况下，我们生成简短的、特定于案例的摘要，使医生之间的比较变得更加容易，同时仍然基于他们个人资料中的信息。为了在不依赖真实患者互动的情况下优化医生展示，我们开发了一个离线框架，使用基于角色的LLM（大型语言模型）来模拟患者选择。我们从每个医疗案例中提取的属性生成患者角色，并评估展示的事实准确性、相关性、自然性和模拟患者选择。我们还使用一个在优化过程中未见过的第二个LLM评判者来评估模拟器特定的过拟合情况。我们比较了基于自动提示的优化（GEPA）、基于参数的优化（GRPO）以及它们在两种模型规模下的组合。GRPO在满足所有约束条件方面显著优于GEPA，并且在模拟转化率方面的提升在未见评判者下依然存在。GEPA单独带来的改进较小，但使用其优化后的提示来初始化GRPO加快了训练并实现了最高的模拟转化率。这些结果表明，提示级和参数级优化在改善医生推荐展示的方式上发挥了互补作用。

View on arXiv Download PDF AI Translation

cs.CL / 209 / 2610.18331

Probing the Geometry of Color Representations in Multilingual Language Models

探究多语言模型中色彩表征的几何结构

Loyola, Pablo, Marrese-Taylor, Edison, Hoyos-Idrobo, Andrés

Abstract

Do multilingual language models learn a universal internal geometry for color, and is it perceptually grounded or primarily linguistic? We decompose color representations into two geometric axes: (P1) alignment to perceptual color space, and (P2) cross-lingual symmetry. Analyzing layerwise embeddings across models and multilingual datasets, we find that perceptual grounding is weak: alignment to CIE-LAB peaks in intermediate layers, then collapses toward output layers. Cross-lingual symmetry degrades with depth, suggesting later layers encode language-specific structure. Functionally, we show that geometric properties predict task performance: models achieve 96.7% accuracy on color classification at layers where P1 alignment peaks, with P1 alignment (r=0.70) significantly correlating with downstream capabilities. Our results reveal a dissociation: text-only LMs learn universal color-language geometry, not universal color-perception geometry, providing diagnostics for where perceptual structure is preserved in language models.

Chinese Translation

多语言模型是否学习了一种普遍的色彩内部几何结构？这种结构是基于感知的，还是主要基于语言的？我们将色彩表征分解为两个几何轴：(P1) 与感知色彩空间的对齐，以及 (P2) 跨语言对称性。通过分析不同模型和多语言数据集的层级嵌入，我们发现感知基础较弱：与 CIE-LAB 的对齐在中间层达到峰值，然后在输出层趋于崩溃。跨语言对称性随着层数的增加而降低，表明后续层编码了特定于语言的结构。从功能上看，我们展示了几何属性能够预测任务性能：在 P1 对齐达到峰值的层次上，模型在色彩分类任务中取得了 96.7% 的准确率，且 P1 对齐（r=0.70）与下游能力显著相关。我们的结果揭示了一种解离现象：仅文本的语言模型学习了普遍的色彩-语言几何，而非普遍的色彩-感知几何，为识别语言模型中保留感知结构的地方提供了诊断依据。

View on arXiv Download PDF AI Translation

cs.CL / 210 / 2610.18343

Lightweight but Linguistically Fair? Quantisation Robustness across Bavarian, German, and English

轻量但语言公平？巴伐利亚语、德语和英语的量化鲁棒性

John, Jophin, Hoffmann, Michael, Lendvai, Piroska, Dufour, Florent

Abstract

Post-training quantisation reduces the memory cost of large language models, but it remains unclear whether compression preserves language skills of regional varieties as reliably as their standard-language counterparts. We study this question for Bavarian, Standard German, and English using BAVGROUND, a multiple-choice benchmark, where 206 source questions are aligned across all three languages. Across 14 open-weight 7B--10B instruction models, we compare BF16 with 8-bit GPTQ and two 4-bit settings, GPTQ and AWQ. Eight-bit GPTQ remains close to BF16 and largely preserves the relative behaviour of the three languages, while 4-bit quantisation was found to be much less stable. The two 4-bit methods reduce average accuracy by about 1.4--1.5 percentage points and change 11.9--16.0% of BF16 decisions, with changes concentrated on examples where the BF16 model is already uncertain. This instability does not translate into a consistent Bavarian disadvantage. Although Bavarian, a low-resource dialect, has lower baseline accuracy than German and English, the language most affected at 4 bits varies across models and quantisation methods. Calibration further shapes these cross-language effects: the relative Bavarian--German effect changes with calibration language, while changing the calibration source has little average effect on GPTQ-4 but can affect AWQ-4 much more strongly. Free-form evaluation on five models further shows that 4-bit compression can degrade generated responses even in settings where multiple-choice accuracy changes only modestly, while 8-bit GPTQ remains comparatively stable. Overall, our results show that quantisation robustness for regional language varieties depends on the model, quantisation method, calibration data, and evaluation setting, and cannot be measured by aggregate multilingual accuracy alone.

Chinese Translation

后训练量化降低了大型语言模型的内存成本，但尚不清楚压缩是否能像其标准语言对应体一样可靠地保留区域方言的语言能力。我们使用BAVGROUND，一个多项选择基准，研究了巴伐利亚语、标准德语和英语的这一问题，其中206个源问题在三种语言中对齐。在14个开放权重的7B--10B指令模型中，我们比较了BF16与8位GPTQ和两种4位设置（GPTQ和AWQ）。8位GPTQ与BF16保持接近，并在很大程度上保留了三种语言的相对行为，而4位量化的稳定性则明显较差。这两种4位方法使平均准确率降低了约1.4--1.5个百分点，并改变了11.9--16.0%的BF16决策，这些变化集中在BF16模型已经不确定的示例上。这种不稳定性并未转化为一致的巴伐利亚劣势。尽管巴伐利亚语作为一种低资源方言，其基线准确率低于德语和英语，但在4位量化中受影响最严重的语言在不同模型和量化方法中各不相同。标定进一步影响这些跨语言效应：相对的巴伐利亚-德语效应随着标定语言的变化而变化，而改变标定源对GPTQ-4的平均影响较小，但对AWQ-4的影响则可能更为显著。对五个模型的自由形式评估进一步表明，即使在多项选择准确率变化仅限于适度的情况下，4位压缩也可能会降低生成的响应质量，而8位GPTQ则保持相对稳定。总体而言，我们的结果表明，区域语言变体的量化鲁棒性依赖于模型、量化方法、标定数据和评估设置，不能仅通过汇总的多语言准确率来衡量。

View on arXiv Download PDF AI Translation

cs.CL / 211 / 2610.18408

Prompt-Injection Probes Learn the Conversation, Not Just the Injection

注入探针学习对话，而不仅仅是注入

Kale, Kaan, Vishwanath, Sriram

Abstract

Linear probes on a model's hidden states are used as cheap monitors for indirect prompt injection: they screen the tool outputs an agent reads and send suspicious ones to a slower second check, so every false alarm has a cost. But a probe learns whatever separates its training examples. The standard evaluation of injection probes can be passed without the model: conversation position and text lengths alone reach 0.91-0.98 AUROC. When each attack is compared with the same conversation without the injection, or with harmless text formatted like an injection, standard probes flag most of these harmless cases. We add such cases to the training data as matched negatives. Across six open models, in applications seen during training and at the default threshold, changing only the training data cuts false alarms on injection-free conversations from 41-80% to 4-27%, and on harmless look-alikes from 92-98% to 3-16%. With a threshold set on normal traffic, our probe still catches a median 95% of the attacks that fooled the agent (91% without our data). In new applications, false alarms stay low, but detection depends on the model and on how the threshold is set. The same data also helps a text-only filter.

Chinese Translation

模型隐藏状态上的线性探针被用作间接提示注入的廉价监控工具：它们筛选代理读取的工具输出，并将可疑的输出发送到较慢的二次检查，因此每个误报都有成本。但探针学习的是任何将其训练示例分开的特征。注入探针的标准评估可以在没有模型的情况下通过：仅对话位置和文本长度就能达到0.91-0.98的AUROC。当每次攻击与同一对话（没有注入）或格式类似于注入的无害文本进行比较时，标准探针会标记大多数这些无害案例。我们将这些案例作为匹配的负样本添加到训练数据中。在六个开放模型中，在训练期间看到的应用和默认阈值下，仅通过更改训练数据，将无注入对话的误报率从41-80%降低到4-27%，将无害相似文本的误报率从92-98%降低到3-16%。在正常流量设定的阈值下，我们的探针仍然捕获了中位数95%的攻击（没有我们的数据时为91%）。在新的应用中，误报保持较低，但检测依赖于模型及阈值的设定。同样的数据也有助于仅基于文本的过滤器。

View on arXiv Download PDF AI Translation

cs.CL / 212 / 2610.18414

Revisiting the Estimation of Unigram Surprisal in Computational Psycholinguistics

重新审视计算心理语言学中单元惊讶度的估计

Hopton, Zachary, Re, Francesco Ignazio, Kiegeland, Samuel, Opedal, Andreas, Chodroff, Eleanor, Cotterell, Ryan

Abstract

Contextual and unigram surprisal are standard predictors of reading time in computational psycholinguistics. While contextual surprisal is typically derived from neural language models, unigram surprisal is usually obtained from independent corpus counts. Although both quantities are deterministic functionals of a single underlying distribution---capturing conditional and marginal structure respectively---this practice breaks their probabilistic link and may introduce estimation error from the mismatch. We propose three principles for estimating unigram surprisal in psycholinguistic research, including the requirement that both quantities be derived from the same language model. We operationalize this principle by introducing a Monte Carlo estimator of unigram surprisal from a language model, along with a marginalized variant that has lower variance empirically. When unigram surprisal is re-estimated under this unified framework, we find evidence that predictors change in their predictive contribution to reading time. We also confirm prior findings that frequency can be a stronger predictor of processing effort than contextual surprisal under this unified framework.

Chinese Translation

上下文惊讶度和单元惊讶度是计算心理语言学中阅读时间的标准预测因子。尽管上下文惊讶度通常是从神经语言模型中推导得出的，而单元惊讶度则通常是通过独立的语料库计数获得的。这两种量都是单一基础分布的确定性函数——分别捕捉条件和边际结构——但这种做法打破了它们的概率联系，并可能因不匹配而引入估计误差。我们提出了三个原则用于在心理语言学研究中估计单元惊讶度，包括要求这两种量都来自同一语言模型。我们通过引入一种基于语言模型的单元惊讶度蒙特卡洛估计器来实现这一原则，同时提出了一种经验上具有更低方差的边际变体。当在这一统一框架下重新估计单元惊讶度时，我们发现预测因子在对阅读时间的预测贡献上发生了变化。我们还确认了先前的研究结果，即在这一统一框架下，频率可以比上下文惊讶度更强烈地预测处理努力。

View on arXiv Download PDF AI Translation

cs.CL / 213 / 2610.18415

Do Factuality Methods Make LLMs More Factual? Reproducibility and Fair Comparison

事实性方法是否使大型语言模型更具事实性？可重复性与公平比较

Amatya, Pratuat, Setty, Vinay, Venktesh, V, Anand, Avishek

Abstract

Large Language Models (LLMs) are aligned to human preferences, after pre-training and supervised fine-tuning steps. However, standard RL alignment steps such as RLHF do not focus on factuality of LLMs, hence they may generate non-factual information, which could result in unintended consequences. In response, a number of factuality alignment approaches have been proposed in the literature which claim to improve factuality of LLMs on their own benchmarks. However, these methods have not been reproduced independently and are also not comparable since each uses a different base model, training data, evaluation set, judge, and metric. In this paper, we present the first independent reproduction of five factuality alignment methods, which span on-policy RL (PPO, GRPO, and Dr.~GRPO) and offline preference optimization (DPO and KTO), and compare them on a common benchmark with common evaluation metrics. We found that none of these methods reproduces in full, and on the common benchmark none of the methods significantly outperforms the base model. On closer inspection, the rewards optimize for format, length, or brevity rather than factuality and relevance, whereas the choice between on-policy and offline training matters little. Furthermore, the evaluation metrics and benchmarks are also part of the problem since they do not measure relevance of text to the prompt. Since there is no ground truth associated with the prompts, the reliance on LLM judges and retrieval for verification also makes the evaluation fragile. We release our code, data, and checkpoints.

Chinese Translation

大型语言模型（LLMs）经过预训练和监督微调步骤后，与人类偏好对齐。然而，标准的强化学习对齐步骤，如强化学习与人类反馈（RLHF），并未关注LLMs的事实性，因此它们可能生成不符合事实的信息，这可能导致意想不到的后果。对此，文献中提出了一些事实性对齐方法，声称能够在其自身基准上提高LLMs的事实性。然而，这些方法尚未独立重复，并且由于每种方法使用不同的基础模型、训练数据、评估集、评审和指标，因此也无法进行比较。在本文中，我们首次独立重复了五种事实性对齐方法，这些方法涵盖了在线强化学习（PPO、GRPO和Dr.~GRPO）和离线偏好优化（DPO和KTO），并在一个共同的基准上使用共同的评估指标进行比较。我们发现这些方法都未能完全重复，并且在共同基准上，没有任何方法显著优于基础模型。经过仔细检查，发现奖励优化的是格式、长度或简洁性，而非事实性和相关性，而在线与离线训练之间的选择影响甚微。此外，评估指标和基准也是问题的一部分，因为它们未能衡量文本与提示的相关性。由于提示没有相关的真实值，依赖LLM评审和检索进行验证也使得评估变得脆弱。我们发布了我们的代码、数据和检查点。

View on arXiv Download PDF AI Translation

cs.CL / 214 / 2610.18420

Agaram 2.5B: The First Tamil-First Language Model Trained from Scratch with Tool Calling. A Playbook for Under-Represented Languages

Agaram 2.5B：首个从零开始训练的以泰米尔语为主的语言模型，工具调用的操作手册为代表性不足的语言

Kumaresan, Ramchand, G, Madhusudhanan, Kezendirane, Kaushik

Abstract

We trained Agaram 2.5B, a 2.5-billion-parameter Tamil-English language model, from scratch on about 50 billion tokens (58% Tamil, the rest English, math, code and parallel text) on one node of eight H100 GPUs in about six days, with a new tokenizer that, unlike our first model's, keeps line breaks. Supervised fine-tuning on 134,428 conversations, whose replies an open-weight teacher model wrote, and three short targeted rounds turned it into a chat model that answers in Tamil script, romanized Tamil (Tanglish) and English and can call tools: on 25 tool prompts whose schemas appear in training it picks the right tool 23 times and gives exactly the expected arguments 13 times. The released model translates better than Sarvam-1, a 2B Indic base model, in all six FLORES and IN22 directions (FLORES chrF 54.3 into Tamil and 55.2 into English, against 49.2 and 44.2), and nearly doubles its base model on native-Tamil IndicIFEval (19.3% from 10.0%). It is weak on knowledge: reading-comprehension and knowledge benchmarks sit near chance, and on 50 held-out everyday questions written by one native speaker, one LLM judge rated 46% of its replies usable (22-24 of 50 over three passes; 95% interval for 23 of 50, 33-60%), with nearly every failure a wrong fact. Our two further SFT experiments changed answer format but not the knowledge measures. We also describe our first model, Agaram 1.5B, whose failures shaped the choices here; how the judge's scores varied between passes on identical replies; and, item by item, what the targeted rounds trained against and where frontier-model text entered the data. The project ran on compute credits and our own funds, about USD 13,300 across both models; the weights are released (licence to be announced after legal review), with GGUF files for llama.cpp, at https://huggingface.co/murai-labs/agaram-2.5b-instruct; code at https://github.com/Murai-Labs/Agaram2.5b

Chinese Translation

我们从零开始训练了Agaram 2.5B，一个包含25亿参数的泰米尔-英语语言模型，使用约500亿个标记（58%为泰米尔语，其余为英语、数学、代码和对照文本），在八个H100 GPU的一个节点上耗时约六天，采用了一种新的分词器，与我们第一个模型的分词器不同，它保留了换行符。通过对134,428个对话进行监督微调，这些对话的回复由一个开放权重的教师模型生成，并进行了三轮短期针对性训练，使其转变为一个聊天模型，能够用泰米尔文、罗马化泰米尔文（Tanglish）和英语回答，并能够调用工具：在25个工具提示中，其模式出现在训练中，模型正确选择工具23次，并准确提供预期参数13次。发布的模型在所有六个FLORES和IN22方向上比Sarvam-1（一个2B的印度基础模型）表现更好（FLORES chrF在泰米尔语中为54.3，在英语中为55.2，而Sarvam-1分别为49.2和44.2），并在本土泰米尔语的IndicIFEval上几乎是其基础模型的两倍（从10.0%提升至19.3%）。在知识方面表现较弱：阅读理解和知识基准接近随机水平，在50个由一位母语者编写的日常问题中，一位LLM评审认为其回复中有46%是可用的（在三轮评估中，50个问题中有22-24个；对于50个问题的95%置信区间为23个，范围为33-60%），几乎每个失败都是由于错误的事实。我们的另外两个SFT实验改变了答案格式，但没有改变知识测量。我们还描述了我们的第一个模型Agaram 1.5B，其失败塑造了这里的选择；评审在相同回复之间的评分变化；以及逐项说明了针对性训练的内容和前沿模型文本如何进入数据。该项目使用计算信用和我们自己的资金进行，两个模型共计约13,300美元；权重已发布（许可证将在法律审查后公布），GGUF文件适用于llama.cpp，网址为https://huggingface.co/murai-labs/agaram-2.5b-instruct；代码可在https://github.com/Murai-Labs/Agaram2.5b获取。

View on arXiv Download PDF AI Translation

cs.CL / 215 / 2610.18424

Receiver Decides Agent-to-Agent Negotiations

接收者决定代理人之间的谈判

Sadab, Fardeen, Hasnain, Alif, Shehrin, Afsana, Hossain, Md. Azam, Kamal, Abu Raihan Mostofa

Abstract

When one language-model agent asks another for money, which of the two decides how much changes hands? We ran a round-robin tournament of 1,049 negotiations among fifteen agents. In each game a requester tries to raise money from a holder that has been told to keep a $100 endowment. Knowing which agent holds the money explains 47% of the variance in the amount transferred; knowing which agent asks explains 2%. Dropping the most extreme agents narrows the split to 22% against 4%. Changing the requester's appeal from one named person to a group changed what requesters wrote, but the amounts transferred under the two framings are statistically equivalent. The holders that lose least are the ones that turn down the opening request, and their turn-1 refusal rate ranks them by total losses (Spearman correlation = -0.73). Training moves this behaviour in both directions. Fine-tuning Qwen3-8B on a persuasion corpus with almost no refusals raised what it gave away as a holder from 12% to 84% of the endowment. Distilling two never-losing holders into a 4B student cut its losses from 95% to 1%. Evaluations of agents that hold resources should test them in the holder role, including after persuasion fine-tuning.

Chinese Translation

当一个语言模型代理向另一个代理请求资金时，究竟由哪一方决定转移的金额？我们进行了一个包含1049次谈判的循环锦标赛，参与者为十五个代理。在每场游戏中，请求者试图从一个被告知保持100美元捐赠的持有者那里筹集资金。知道哪个代理持有资金可以解释47%的转移金额方差；而知道哪个代理提出请求仅能解释2%。去掉最极端的代理后，分配比例缩小至22%对4%。将请求者的请求从一个具体的人改为一个群体，改变了请求者的书写内容，但在这两种框架下转移的金额在统计上是等效的。损失最少的持有者是那些拒绝初始请求的持有者，他们的首次拒绝率与总损失呈负相关（Spearman 相关系数 = -0.73）。训练可以在两个方向上改变这种行为。在几乎没有拒绝的劝说语料库上微调 Qwen3-8B，使其作为持有者的赠与比例从12%提高到84%。将两个从未亏损的持有者提炼为一个4B的学生，使其损失从95%降至1%。对持有资源的代理的评估应在持有者角色中进行测试，包括在劝说微调之后。

View on arXiv Download PDF AI Translation

cs.CL / 216 / 2610.18427

Reliable Memory from Unstable Growth: Curating Self-Written Agent Playbooks with Randomized Experiments

不稳定增长中的可靠记忆：通过随机实验策划自写代理手册

Pham, Hoang, Nguyen, Cong-Duy, Le, Dong, Srey, Ponhvoan, Luu, Anh Tuan

Abstract

Persistent textual memory lets language-model agents reuse lessons from past tasks to improve on future multi-step tasks without retraining. Systems such as ACE and Dynamic Cheatsheet keep it as a playbook: lessons the agent writes after each task and reads in its prompt. A key challenge is to decide which lessons improve later tasks. Current systems leave the decision to the agent itself, which makes performance unstable: the same procedure can write a playbook that helps or one that falls up to 69.6 points below no playbook. These findings motivate valuing lessons by their measured effect on task success rather than by the agent's judgment. We propose Randomized Block Curation (RBC), which measures this effect with a randomized experiment on practice tasks. It groups related lessons into blocks large enough to move success measurably and, on each practice task, includes or leaves out every block at random. Random inclusion makes tasks with and without a block comparable, so the gap in their success estimates the block's effect. RBC keeps the blocks that help and, since the same procedure can write a helpful or a harmful playbook, grows several and serves whichever curated, whole or no playbook scores best on fresh practice episodes. Across two memory systems, three models and three benchmarks, RBC raises average success to 66.9, against 55.5 with the whole playbook, 54.1 with no playbook and 53.8 when the agent's own judgment picks the lessons to keep.

Chinese Translation

持久的文本记忆使语言模型代理能够重用过去任务的经验教训，从而在未来的多步骤任务中改进表现，而无需重新训练。像 ACE 和 Dynamic Cheatsheet 这样的系统将其作为手册：代理在每个任务后编写的经验教训，并在提示中阅读。一个关键挑战是决定哪些经验教训能改善后续任务。目前的系统将这一决定留给代理自身，这使得性能不稳定：相同的程序可以编写出有助于任务的手册，也可以编写出低于无手册的表现高达 69.6 分的手册。这些发现促使我们根据经验教训对任务成功的实际影响来评估它们，而不是依赖代理的判断。我们提出了随机区块策划（Randomized Block Curation, RBC），该方法通过对实践任务进行随机实验来测量这种影响。它将相关的经验教训分组为足够大的区块，以便能够显著影响成功率，并在每个实践任务中随机包含或排除每个区块。随机包含使得有区块和无区块的任务可比，因此它们成功率的差距可以估算出该区块的影响。RBC 保留有帮助的区块，并且由于相同的程序可以编写出有益或有害的手册，因此它生成多个手册，并根据新实践任务的表现选择最佳的策划手册、完整手册或不使用手册。在两个记忆系统、三个模型和三个基准测试中，RBC 将平均成功率提高到 66.9，而完整手册为 55.5，无手册为 54.1，代理自身判断选择保留的经验教训时为 53.8。

View on arXiv Download PDF AI Translation

cs.CL / 217 / 2610.18439

Estimating Surprisal from the Speech Signal

从语音信号中估计惊讶度

Hopton, Zachary, Kiegeland, Samuel, Re, Francesco Ignazio, Haller, Patrick, Winn, Matthew B., Cotterell, Ryan, Chodroff, Eleanor

Abstract

Although language is predominantly spoken, empirical tests of surprisal theory rely almost entirely on written text. In practice, surprisal is estimated using language models (LMs) trained on text and applied to transcriptions, creating a text bottleneck that limits evaluation of the theory in the spoken modality. We remove this bottleneck by estimating surprisal directly from the speech signal using a speech LM. This raises a key question: should surprisal be defined over realization-specific acoustic inputs or over abstract linguistic units? Since a single unit can have many distinct acoustic realizations, each assigned a different probability by the model, defining surprisal over abstract units is difficult. To address this, we introduce an importance sampling estimator that approximates the probability of an abstract unit by marginalizing over realizations generated by a text-to-speech model. Using pupillometry data from a listening study, we evaluate surprisal theory in this text-free setting at multiple levels of granularity. Obtaining surprisal from several speech LMs, we consistently find that surprisal estimated via importance sampling better predicts pupil dilation than realization-specific surprisal for words. We also find that surprisal values for sub-lexical units are predictive of pupil dilation. While we find that surprisal can be estimated without recourse to text, and that marginalizing over distinct realizations of words is more predictive of lexical processing effort, surprisal as a whole is not a strong predictor of change in pupil size, regardless of its source.

Chinese Translation

尽管语言主要是口头表达的，但惊讶度理论的实证测试几乎完全依赖于书面文本。在实践中，惊讶度是通过使用在文本上训练的语言模型（LMs）并应用于转录来估计的，这造成了一个文本瓶颈，限制了在口语模式下对该理论的评估。我们通过直接从语音信号中使用语音语言模型估计惊讶度，从而消除了这一瓶颈。这引出了一个关键问题：惊讶度应该定义在特定实现的声学输入上，还是在抽象语言单位上？由于单个单位可以有许多不同的声学实现，每个实现被模型分配不同的概率，因此在抽象单位上定义惊讶度是困难的。为了解决这个问题，我们引入了一种重要性抽样估计器，通过对由文本到语音模型生成的实现进行边际化来近似抽象单位的概率。利用来自听觉研究的瞳孔测量数据，我们在这一无文本的环境中以多个粒度水平评估惊讶度理论。通过多个语音语言模型获得惊讶度，我们一致发现，通过重要性抽样估计的惊讶度比特定实现的惊讶度更好地预测了词汇的瞳孔扩张。我们还发现，子词单位的惊讶度值能够预测瞳孔扩张。尽管我们发现可以在不依赖文本的情况下估计惊讶度，并且对词的不同实现进行边际化更能预测词汇处理的努力，但整体而言，惊讶度并不是瞳孔大小变化的强预测因子，无论其来源如何。

View on arXiv Download PDF AI Translation

cs.CL / 218 / 2610.18442

Copying Is Not Grounding: Training Faithful Long-Form RAG with Authored-Sentence Credit

复制并非基础：使用作者句子信用训练忠实的长格式 RAG

Pham, Hoang, Vu, Duc Anh, Doan, Duy Tung, Nguyen, Cong-Duy T, Luu, Anh Tuan

Abstract

In search engines and assistants, long-form RAG answers can mislead readers with claims their sources do not support. Faithfulness rewards, the share of supported sentences, train models toward fewer such claims. We show that these rewards are hacked by copying. A sentence copied from the passages is supported almost by construction, so trained models learn to copy rather than ground what they write. Common defenses only move the hack: a KL or copy penalty leaves copying in place or trades it for shorter, emptier answers and refusals. GRPO, the standard RL algorithm, makes it worse by giving every sentence its answer's advantage, reinforcing copied and grounded sentences alike. We propose Authored-Sentence Credit (ASC) , which fixes both by moving GRPO's reward and advantage from the answer to the sentence level. The reward credits only sentences that are authored rather than copied, say more than the question, and are supported by the passages. The advantage compares each sentence with every sentence sampled for the question, so one that earns no credit ranks last and is never reinforced. Across four models and two in-domain, three held-out corpora, only ASC lowers the error rate of authored sentences and writes more supported ones in every setting, by 38% and 46% at the median under held-out evaluators, giving readers more grounded information. It takes none of these shortcuts, and two LLM judges comparing whole answers prefer no baseline to the base model but rate ASC's answers on par with it. We conclude that a faithfulness reward should credit what a model writes, not what it copies.

Chinese Translation

在搜索引擎和助手中，长格式 RAG 的回答可能会误导读者，因为其声称的来源并不支持这些主张。忠实奖励，即支持句子的比例，促使模型减少此类主张。我们展示了这些奖励被复制行为所利用。从段落中复制的句子几乎在结构上是被支持的，因此训练后的模型学习到的是复制而非基础化其所写内容。常见的防御措施只是将这一漏洞转移：KL 或复制惩罚要么保留了复制行为，要么将其换成更短、更空洞的回答和拒绝。标准的强化学习算法 GRPO 使情况更糟，因为它给予每个句子其回答的优势，从而强化了复制和基础化的句子。我们提出了作者句子信用（Authored-Sentence Credit, ASC），通过将 GRPO 的奖励和优势从回答转移到句子层面来解决这两个问题。该奖励仅对那些是作者创作而非复制的句子进行信用评估，这些句子比问题表达得更多，并且得到了段落的支持。优势则是将每个句子与为该问题抽样的每个句子进行比较，因此没有获得任何信用的句子将排名最后，并且不会被强化。在四个模型和两个领域内、三个保留语料库的实验中，只有 ASC 在每种设置下都降低了作者句子的错误率，并且写出了更多被支持的句子，分别在保留评估者下中位数降低了 38% 和 46%，为读者提供了更有基础的信息。它没有采取任何这些捷径，而两位 LLM 评审在比较完整回答时更倾向于不使用基线的基础模型，但对 ASC 的回答给予了与之相当的评价。我们得出结论，忠实奖励应当对模型所写内容给予信用，而非其复制的内容。

View on arXiv Download PDF AI Translation

cs.CL / 219 / 2610.18498

Aligning Faithfulness for Theory-Scale Autoformalization

理论规模自动形式化中的忠实性对齐

Feng, Yuming, Mao, Tadd, Arekar, Dhruva, Zhong, Tianjun, Huang, Jiani, Si, Xujie, Li, Ziyang

Abstract

Autoformalization (AF) requires more than compilation: formal statements must preserve the meaning of their informal sources, a property known as faithfulness. An emerging frontier is the automated evaluation of this property, yet most existing methods operate at the statement level and produce whole-item verdicts, failing to reveal how individual expressions correspond or which contextual interpretations support the judgment. This omission is consequential because AF is inherently theory-scale: target statements derive their meaning from surrounding mathematical material and unstated conventions. We formulate faithfulness evaluation as a context-aware align-then-label problem: identifying phrase-level correspondences between informal and formal statements and assigning a faithfulness label to each under explicitly specified contextual interpretations and assumptions. We also introduce FaithAlign, the first theory-scale faithfulness benchmark comprising 460 items and 4,982 annotated phrase-level links, alongside FaithAlign-Gap, a companion set of 146 items with informal context and assumptions withheld. Each FaithAlign item includes the target statement pair, relevant context, formal dependencies, and, where needed, assumptions specifying one given interpretation. Across 8 models, GPT-6 Astra achieves the highest defect F1 at 47.8% and identifies all defects in only 4.5% of items with two or more defects. On FaithAlign-Gap, Astra identifies only 28.2% of undecidable items as such. These results show the difficulty of recovering context-dependent correspondences, establishing FaithAlign as a testbed for fine-grained, context-aware faithfulness evaluation.

Chinese Translation

自动形式化（AF）不仅仅需要编译：形式声明必须保留其非正式来源的意义，这一特性被称为忠实性。一个新兴的前沿领域是对这一特性的自动评估，然而大多数现有方法在声明层面上运作，并产生整体判决，未能揭示个别表达之间的对应关系或哪些上下文解释支持该判断。这一遗漏是重要的，因为AF本质上是理论规模的：目标声明的意义来自于周围的数学材料和未明言的约定。我们将忠实性评估形式化为一个上下文感知的对齐-标记问题：识别非正式和正式声明之间的短语级对应关系，并在明确指定的上下文解释和假设下为每个短语分配忠实性标签。我们还介绍了FaithAlign，这是第一个包含460个项目和4,982个注释短语级链接的理论规模忠实性基准，以及FaithAlign-Gap，一个伴随集，包含146个项目，其中非正式上下文和假设被保留。每个FaithAlign项目包括目标声明对、相关上下文、形式依赖关系，以及在需要时指定给定解释的假设。在8个模型中，GPT-6 Astra在缺陷F1上达到了最高的47.8%，并且仅在4.5%的具有两个或更多缺陷的项目中识别出所有缺陷。在FaithAlign-Gap上，Astra仅将28.2%的未决项目识别为此类。这些结果显示了恢复依赖于上下文的对应关系的困难，确立了FaithAlign作为细粒度、上下文感知忠实性评估的测试平台。

View on arXiv Download PDF AI Translation

cs.CL / 220 / 2610.18507

The Last Mile of Coding Agents for Long-Document Tasks: An Empirical Study on Clinical Trial Protocol Digitization

长文档任务中编码代理的最后一公里：临床试验方案数字化的实证研究

Lin, Jiacheng, Low, Geoffrey, Agarwal, Nishit, Jain, Rahul, Shafquat, Afrah, Beigi, Mandis, Sun, Jimeng

Abstract

Clinical trial protocols serve as the source of truth for study design, but are mostly written and used as unstructured PDF documents. Digitizing protocols turns static content into structured, machine-readable data that clinical software can analyze, validate, and execute. Coding agents, which select tools, execute code, and revise their approach as they work, are increasingly applied to such long-document tasks, yet it remains unclear how reliably they turn complete documents into accurate structured outputs. We study this question using protocol digitization as a testbed. We introduce TrialPD-Bench, an expert-annotated benchmark of 150 publicly available protocol PDFs spanning four task categories. Beyond final-answer performance, our evaluation measures evidence recall from agent traces, agent behavior, and cost. Across nine proprietary and open-weight models in five coding-agent harnesses, even the strongest agents do not fully solve protocol digitization. Our analysis further shows that agents usually locate the required evidence but fail in the last mile of transcribing and structuring it. We further develop TrialScribe, a domain-specific harness that equips coding agents with task skills and a verification loop. TrialScribe improves a locally served open-weight model by 6.2-59.4 points across eight metrics, enabling it to match or surpass proprietary models such as GPT-5.6-Sol.

Chinese Translation

临床试验方案作为研究设计的真实来源，通常以非结构化的PDF文档形式编写和使用。数字化方案将静态内容转化为结构化的、机器可读的数据，供临床软件进行分析、验证和执行。编码代理选择工具、执行代码并在工作过程中调整其方法，越来越多地应用于此类长文档任务，但它们将完整文档转化为准确结构化输出的可靠性仍不清楚。我们以方案数字化为测试平台研究这一问题。我们引入了TrialPD-Bench，这是一个包含150个公开可用的方案PDF的专家注释基准，涵盖四个任务类别。除了最终答案性能外，我们的评估还衡量了代理痕迹中的证据召回、代理行为和成本。在五个编码代理框架中，九个专有和开放权重模型中，即使是最强的代理也未能完全解决方案数字化问题。我们的分析进一步表明，代理通常能够定位所需证据，但在转录和结构化的最后一公里上失败。我们进一步开发了TrialScribe，这是一个领域特定的框架，为编码代理提供任务技能和验证循环。TrialScribe在八个指标上使本地服务的开放权重模型提高了6.2-59.4分，使其能够与专有模型如GPT-5.6-Sol相媲美或超越。

View on arXiv Download PDF AI Translation

cs.CL / 221 / 2610.18517

Imprinting: Knowledge Distillation Can Shift Student Behavior

印记效应：知识蒸馏可以改变学生行为

Tang, Yu-Chien, Yen, An-Zi

Abstract

Knowledge distillation trains student models on teacher-generated responses to transfer task capability, but those responses may also carry behavioral tendencies of their source. We investigate whether the source teacher can shape student behavior beyond the capability that distillation is intended to transfer. We construct harm-accepting and harm-avoiding variants of the same base teacher, use them to generate responses to ordinary reasoning and knowledge questions unrelated to safety, and distill those responses into students with different base models. Students generally acquire task capability while their downstream decisions shift toward the source teacher's safety-pragmatism disposition, a phenomenon we call imprinting. Distillation from the harm-accepting teacher reduces harm avoidance, whereas distillation from the harm-avoiding teacher increases it. Imprinting varies across teacher-student pairs and does not extend uniformly across different forms of safety behavior. Excluding a sparse set of divergence tokens reduces imprinting, whereas response rewriting preserves the source-dependent effect and can introduce additional rewriter-dependent shifts. Post-distillation preference optimization raises harm avoidance but does not eliminate the source-dependent difference. These results show that behavioral tendencies can be intentionally imprinted through the source teacher, even with ordinary task supervision.

Chinese Translation

知识蒸馏通过教师生成的响应训练学生模型，以转移任务能力，但这些响应也可能携带其来源的行为倾向。我们研究了源教师是否能够在超出蒸馏所意图转移的能力范围内塑造学生行为。我们构建了接受伤害和避免伤害的同一基础教师的变体，利用它们生成与安全无关的普通推理和知识问题的响应，并将这些响应蒸馏到不同基础模型的学生中。学生通常获得任务能力，同时他们的下游决策向源教师的安全务实倾向转变，这一现象我们称之为印记效应。从接受伤害的教师进行蒸馏会减少对伤害的回避，而从避免伤害的教师进行蒸馏则会增加这种回避。印记效应在教师-学生对之间有所不同，并且在不同形式的安全行为中并不均匀扩展。排除一小部分的偏差标记会减少印记效应，而响应重写则保留了源依赖效应，并可能引入额外的重写器依赖性变化。蒸馏后的偏好优化提高了对伤害的回避，但并未消除源依赖的差异。这些结果表明，即使在普通任务监督下，行为倾向也可以通过源教师有意地印记。

View on arXiv Download PDF AI Translation

cs.CL / 222 / 2610.18520

The Geometry of Language Model Logits: Generic Structure and Learned Departures

语言模型逻辑的几何结构：通用结构与学习偏差

Narashiman, Swathi Shree, Venkatramanan, Aravind Ramana, Seshadri, Nambi

Abstract

Autoregressive language models produce, at every prediction step, a vocabulary-sized logit vector whose structure determines the conditional next-token distribution. We empirically identify that the ordered vocabulary-head logits follow a remarkably consistent log-linear relation \[ z_{(r)} \approx \mu - \alpha \log r. \] across prediction contexts, architectures, and scales, emerging progressively during pretraining. To our knowledge, this is the first systematic analysis of the logit geometry underlying the distributional assumptions used in the prior decoding work. While controlled order-statistic baselines spanning bounded, light-tailed, and heavier-tailed distributions can largely explain the observed log-linear geometry, they do not explain the systematic shift in the slope $\alpha$: trained language models exhibit $\alpha \approx 1.4$--$1.7$, sitting well above the initialization baseline ($\alpha \approx 0.2$) . Moreover, $\alpha$ correlates with predictive performance, suggesting that it captures learned structure beyond generic extreme-value effects.

Chinese Translation

自回归语言模型在每个预测步骤生成一个大小为词汇表的逻辑向量，其结构决定了条件下一个词的分布。我们通过实证研究发现，排序的词汇头逻辑在不同的预测上下文、架构和规模中遵循一种显著一致的对数线性关系 [z_{(r)} ext{ 近似 } u - eta ext{ log } r]，这一关系在预训练过程中逐渐显现。据我们所知，这是对先前解码工作中所用分布假设的逻辑几何的首次系统分析。虽然涵盖有界、轻尾和重尾分布的控制顺序统计基线在很大程度上可以解释观察到的对数线性几何，但它们无法解释斜率 $eta$ 的系统性变化：训练后的语言模型表现出 $eta ext{ 近似 } 1.4$--$1.7$，远高于初始化基线 ($eta ext{ 近似 } 0.2$)。此外，$eta$ 与预测性能相关，表明它捕捉到了超越通用极值效应的学习结构。

View on arXiv Download PDF AI Translation

cs.CL / 223 / 2610.18526

Retrieval Disagreement as a Signal: Diagnosing and Steering Multilingual Dense Embedders

检索分歧作为信号：诊断与引导多语言密集嵌入器

Ghorbanpour, Faeze, Kolli, Shaghayegh, Diesner, Jana, Fraser, Alexander

Abstract

Multilingual dense retrievers place all languages in one shared space so that a question finds relevant documents, regardless of which language they are written in. We show that the same question, asked in different languages, retrieves substantially different documents, and this disagreement is not because of weak language understanding. Each language performs well on its own and ranks candidates consistently with the other languages; divergence appears only when languages share an index. We trace the mechanism: each language keeps its own ranking, but in the shared index, its correct answers are pushed far down, buried under non-relevant documents that are either in the query's own language or in English, depending on the model. We introduce disagreement-guided steering: for each language, we contrast the documents that other languages agree on for a question against those that language retrieves idiosyncratically, and take the difference of their representations as a language-specific direction, added to that language's queries at inference. Against baselines of language-mean removal, subspace removal, and per-language rotation, it most strongly reduces same-language crowding while remaining competitive on recall, trading off against mean removal, which better preserves ranking quality.

Chinese Translation

多语言密集检索器将所有语言置于一个共享空间中，以便无论文档使用何种语言撰写，问题都能找到相关文档。我们展示了同一个问题用不同语言提问时，检索到的文档存在显著差异，而这种分歧并不是由于语言理解能力弱。每种语言在其自身的检索中表现良好，并且与其他语言一致地对候选文档进行排名；分歧仅在语言共享索引时出现。我们追踪了这一机制：每种语言保持自己的排名，但在共享索引中，其正确答案被推到较低的位置，埋在与查询语言相同或英文的非相关文档之下，具体取决于模型。我们引入了基于分歧的引导：对于每种语言，我们对比其他语言在某个问题上达成一致的文档与该语言独特检索到的文档，并将它们表示的差异作为特定于语言的方向，添加到该语言在推理时的查询中。在语言均值去除、子空间去除和每种语言旋转的基线对比中，该方法最有效地减少了同语言的拥挤，同时在召回率上保持竞争力，权衡了均值去除，这更好地保留了排名质量。

View on arXiv Download PDF AI Translation

cs.CL / 224 / 2610.18529

How Reliable Are System-One Decisions in Medicine? A Comparative Study of Jev, Jev-like Models, and LLMs

系统一决策在医学中的可靠性如何？Jev、类Jev模型与大型语言模型的比较研究

Du, Yanrui, Zhao, Danyang, Zhao, Sendong, Li, Wenhao, Gao, Yibo, Wang, Haochun, Xu, Zhuoer, Cui, Shiwen, Qin, Bing

Abstract

Jev has attracted considerable attention for its rapid, generalizable decision-making, yet its medical reliability remains under explored. To address this gap, our study compares various types of system-one decision models across 15 medical datasets comprising 8,742 instances under a unified protocol. Beyond accuracy, we assess decision stability and the ability to recognize unreliable decisions. Our results show that Jev rivals leading models on primarily text-driven tasks but is less effective on numerically driven tasks, revealing a distinct capability boundary. Under question-format variations, Jev demonstrates strong robustness, achieving the lowest decision reversal rate and relatively small accuracy fluctuations. Among evaluated open-source alternatives, only StartLux exhibits comparable robustness. But these strengths do not extend to reliable abstention awareness: most evaluated models show limitations in both recognizing missing correct answers and using optional abstention. Moreover, we find that Jev's probability-based answer selection can be improved by changing how decisions are scored: normalizing option-wise binary scores yields higher retained accuracy than choice scoring. Taken together, our findings provide early guidance for improving and applying decision models in medicine.

Chinese Translation

Jev因其快速且可推广的决策能力而受到广泛关注，但其在医学领域的可靠性仍然未得到充分探讨。为了解决这一问题，我们的研究在统一协议下比较了15个医学数据集中的各种系统一决策模型，共计8742个实例。除了准确性外，我们还评估了决策的稳定性和识别不可靠决策的能力。我们的结果表明，Jev在主要以文本驱动的任务中与领先模型相媲美，但在以数字驱动的任务中效果较差，揭示了其能力的明显边界。在问题格式的变化下，Jev表现出强大的鲁棒性，达到了最低的决策反转率和相对较小的准确性波动。在评估的开源替代方案中，只有StartLux展现出可比的鲁棒性。然而，这些优势并未延伸至可靠的弃权意识：大多数评估模型在识别缺失正确答案和使用可选弃权方面均显示出局限性。此外，我们发现，通过改变决策评分方式，可以改善Jev基于概率的答案选择：对选项的二元评分进行归一化比选择评分能保持更高的准确性。综合来看，我们的研究结果为改善和应用医学中的决策模型提供了初步指导。

View on arXiv Download PDF AI Translation

cs.CL / 225 / 2610.18541

Empathy Is Not the Scapegoat for Declines in Factuality and Reliability

同理心并不是事实性和可靠性下降的替罪羊

Shi, Letian, Mehreen, Kanwal, Zhang, Deheng, Wang, Yuxia

Abstract

Language models are increasingly expected to support users in stressful or uncertain situations, where warmth and empathy are desirable. While recent work associates warmer models with reduced reliability, we show that this conclusion is incomplete. Prior studies often train primarily on empathetic dialogue, making it difficult to separate tone effects from SFT-induced distribution shift. We revisit this issue at both the model and evaluation levels. We study prompting, SFT, and RL for inducing warm behavior, and comprehensively evaluate capability preservation using an 11-dataset benchmark suite and EMOFACT under emotional contexts. Across multiple model families, we find that warmth effects are task- and model-dependent, while task-aligned adaptive tone can improve empathy and targeted reliability, although some capability trade-offs remain. We further introduce human-validated empathy evaluators specifically developed to assess empathetic model responses. Our results suggest that models can be more supportive in social interactions while remaining reliable.

Chinese Translation

语言模型越来越被期望在压力或不确定的情况下支持用户，此时温暖和同理心是可取的。尽管近期研究将更温暖的模型与降低的可靠性联系在一起，但我们表明这一结论是不完整的。之前的研究通常主要训练于同理心对话，这使得很难将语调效应与SFT（监督微调）引起的分布偏移区分开来。我们在模型和评估层面重新审视了这一问题。我们研究了提示、SFT和RL（强化学习）以诱导温暖行为，并使用11个数据集的基准套件和EMOFACT在情感背景下全面评估能力保留。在多个模型系列中，我们发现温暖效应依赖于任务和模型，而与任务对齐的自适应语调可以改善同理心和目标可靠性，尽管仍然存在一些能力权衡。我们进一步引入了经过人类验证的同理心评估工具，专门用于评估同理心模型的响应。我们的结果表明，模型在社会互动中可以更加支持用户，同时保持可靠性。

View on arXiv Download PDF AI Translation

cs.CL / 226 / 2610.18569

CREATE: Creativity Evaluation of VLMs using the Visual Remote Associates Test

CREATE：使用视觉远程联想测试评估视觉语言模型的创造力

Yang, Ziqing, Jangra, Anubhav, Muresan, Smaranda

Abstract

Vision language models (VLMs) are used in creative tasks such as visual storytelling, design ideation, art interpretation, and advertising. Yet we lack principled ways to measure the lateral reasoning of these models compared to human potential. We introduce CREATE, a semi-automatically constructed benchmark containing 155 visual Remote Associates Test problems. These problems require a solver to find an underlying connection among three seemingly dissimilar images. We conduct a human study (N=25) to validate and ground the quality of our benchmark, and evaluate 11 VLMs. Our findings show that most VLMs underperform humans on this task. Our experiments indicate that the associative ability of VLMs depends on the cue concepts provided in the text and does not universally extend to visual puzzles.

Chinese Translation

视觉语言模型（VLMs）被用于视觉叙事、设计构思、艺术解读和广告等创造性任务。然而，我们缺乏原则性的方法来衡量这些模型的侧向推理能力与人类潜力的比较。我们引入了CREATE，这是一个半自动构建的基准，包含155个视觉远程联想测试问题。这些问题要求解题者在三幅看似不相关的图像中找到潜在的联系。我们进行了一项人类研究（N=25）以验证和确定我们基准的质量，并评估了11个VLMs。我们的研究结果表明，大多数VLMs在这一任务中的表现不及人类。我们的实验表明，VLMs的联想能力依赖于文本中提供的线索概念，并不普遍适用于视觉难题。

View on arXiv Download PDF AI Translation

cs.CL / 227 / 2610.18573

The Many Morals of AI: The cautionary bias of North American news narratives in global perspective

人工智能的多重道德：北美新闻叙事在全球视角下的警示性偏见

Piper, Andrew, Pearman, Edie

Abstract

The stories told about artificial intelligence (AI) play a central role in shaping public understanding, policy debates, and expectations about technological change. News media are especially influential in this process. Using a cross-regional corpus of 12,924 English-language news articles about AI published since 2022, we examine how storytelling about AI varies across media systems in four regions: North America, China, India, and the Middle East. Leveraging large language models, we extract the implicit narrative lessons conveyed by each article---commonly known as story morals---capturing value-laden interpretations that extend beyond thematic content or sentiment. We find that North American AI coverage is uniquely characterized by narratives of caution, while under-emphasizing narratives of innovation and collaboration that are more prevalent elsewhere. However, this pattern is largely explained by a broader regional negativity bias in North American news sentiment, rather than by heightened concern about AI itself. Our findings bring into view the culturally conditioned nature of AI news narratives.

Chinese Translation

关于人工智能（AI）的叙述在塑造公众理解、政策辩论和对技术变革的期望中起着核心作用。新闻媒体在这一过程中尤其具有影响力。通过分析自2022年以来发布的12,924篇关于AI的英语新闻文章的跨区域语料库，我们考察了不同地区的媒体系统中关于AI的叙事如何变化：北美、中国、印度和中东。利用大型语言模型，我们提取了每篇文章所传达的隐含叙事教训——通常称为故事道德——捕捉那些超越主题内容或情感的价值观解释。我们发现，北美的AI报道独特地以警示性叙事为特征，而对创新和合作的叙事则相对被低估，这在其他地区更为普遍。然而，这一模式在很大程度上是由北美新闻情感中更广泛的消极偏见所解释，而不是对AI本身的高度关注。我们的研究结果揭示了AI新闻叙事的文化条件性特征。

View on arXiv Download PDF AI Translation

cs.CL / 228 / 2610.18588

Decomposing and Predicting Annotation Disagreement in Harmful Content Across Languages

跨语言有害内容注释分歧的分解与预测

Yelkanat, Irem Ecem, Ghorbanpour, Faeze, Fraser, Alexander

Abstract

Annotator disagreement is increasingly recognized as a meaningful property of subjective tasks, rather than merely noise to be removed. Predicting which items are likely to be contested could help allocate annotation effort and route difficult cases to human review. We propose predicting disagreement from interpretable semantic properties of the text. An LLM answers a set of binary questions covering harmful-content characteristics, disagreement-related cues such as borderline severity and missing context, and language-sensitive cues such as local slurs. A classifier predicts disagreement from these answers, and we analyze which questions contribute most to the predictions. Across four datasets in English, German, and Turkish and three open-weight LLMs, our structured question features outperform direct prompting by 10.5 ROC-AUC points and the strongest text-based baselines by 5.0 points. A compact subset of eight questions performs within 1.5 ROC-AUC points of the full twenty-question set. In leave-one-dataset-out evaluation, models retain roughly 90-95% of their in-domain ROC-AUC. Across models and datasets, questions about derogation, missing context, and decision boundaries serve as the strongest predictors, suggesting that what makes harmful content contested is partly observable in the text.

Chinese Translation

注释者之间的分歧越来越被认为是主观任务的一个重要特征，而不仅仅是需要去除的噪声。预测哪些项目可能会引发争议可以帮助合理分配注释工作，并将困难案例提交给人工审核。我们提出从文本的可解释语义特征中预测分歧。一个大型语言模型（LLM）回答一组涵盖有害内容特征的二元问题，包括与分歧相关的线索，如边缘严重性和缺失上下文，以及与语言相关的线索，如地方性侮辱。分类器根据这些答案预测分歧，我们分析哪些问题对预测贡献最大。在英语、德语和土耳其语的四个数据集以及三个开放权重的LLM中，我们的结构化问题特征比直接提示提高了10.5个ROC-AUC点，且比最强的基于文本的基线提高了5.0个点。一个由八个问题组成的紧凑子集在ROC-AUC上与完整的二十个问题集相差仅1.5个点。在逐一剔除数据集的评估中，模型在其领域内保持了大约90-95%的ROC-AUC。在不同模型和数据集中，关于贬损、缺失上下文和决策边界的问题作为最强的预测因子，这表明有害内容之所以引发争议，部分特征可以在文本中观察到。

View on arXiv Download PDF AI Translation

cs.CL / 229 / 2610.18605

From Hindsight to Foresight: Dissecting Clinical Evidence Acquisition in Large Language Models

从事后分析到前瞻性分析：剖析大型语言模型中的临床证据获取

Ling, Ziyang

Abstract

Large language models (LLMs) often diagnose full sight, static clinical cases well yet struggle in foresight, dynamic diagnosis. Prior work has improved this limitation by quantifying the information models acquire and by developing stronger follow-up strategies to improve diagnostic accuracy. In contrast, we introduce a distinctive hindsight to systematically dissect why LLMs fail to acquire diagnostically critical evidence, tracing the failure from evidence identification, utilization and downstream diagnostic updating through not only diagnostic accuracy, but also evidence coverage, trajectory dynamics, and elicited probability and information entropy. Starting experiments from the breakdown at the level of capabilities, to the diagnosis of the pre-event state before the evidence arrives, and to the post-event state transition after the evidence arrives, our research has revealed the profound limitations and internal causes existing in the dynamic diagnostic process. At the same time, we have also conducted sufficient verification and generalization experiments to prove that these limitations are widespread. Together, these findings shift the central challenge from whether LLMs possess relevant clinical knowledge to whether they can identify, acquire, and effectively use the right evidence under clinical diagnosis and conduct a more detailed mechanism analysis to identify the exact location of the capability gap.

Chinese Translation

大型语言模型（LLMs）通常能够很好地诊断完整、静态的临床案例，但在前瞻性、动态诊断方面却面临挑战。之前的研究通过量化模型获取的信息并开发更强的后续策略来提高诊断准确性，从而改善了这一局限性。相较之下，我们引入了一种独特的事后分析方法，系统性剖析LLMs为何未能获取诊断上关键的证据，追踪从证据识别、利用到下游诊断更新的失败，不仅通过诊断准确性，还包括证据覆盖率、轨迹动态、引发的概率和信息熵。我们的实验从能力层面的崩溃开始，诊断证据到达前的事件状态，再到证据到达后的事件状态转变，揭示了动态诊断过程中存在的深刻局限性和内在原因。同时，我们还进行了充分的验证和推广实验，以证明这些局限性是普遍存在的。这些发现将核心挑战从LLMs是否具备相关临床知识转变为它们是否能够在临床诊断中识别、获取并有效利用正确的证据，并进行更详细的机制分析，以确定能力差距的确切位置。

View on arXiv Download PDF AI Translation

cs.CL / 230 / 2610.18610

Jev-ToM: Evaluating Theory of Mind in Typed Decision Models

Jev-ToM：在类型化决策模型中评估心智理论

Chung, Kawai, Chan, Chunkit, Liu, Yuxuan, Yim, Yauwai, Yang, Xi, Hu, Wenbin, Jing, Huihao, Tang, Chiu Yeung, Fan, Changxuan, Hung, Ki Sen, Kwok, Tsun On, Li, Haoran

Abstract

Dialogue and collaborative assistants repeatedly judge what a partner knows, believes, or intends. Typed decision models offer a potential alternative to generative large language models for bounded Theory of Mind (ToM) selection. We present Jev-ToM, comparing five typed decision models and seven generative baselines on 11,117 English text questions from nine ToM benchmarks across six capability groups. TypeSafe Jev reaches 73.2% source-balanced accuracy including native controls and ranks first or second across all six capabilities, leading beliefs, goals/desires/intentions, and emotions. Other typed models show distinct strengths across tasks. Separate resource comparisons suggest that Jev could reduce the cost and waiting time of repeated ToM selection. Confidence-based selection remains benchmark-dependent, and supplied reasons can repair errors or overturn correct answers depending on their attachment. Candidate expansion reveals distinct selection and output-validity problems. Related-question cues account for most of Jev's downstream gains, with task-dependent effects from predicted states. The findings motivate testing Jev as a ToM selection component in assistants, with task-specific confidence calibration, candidate design, and output validation as integration priorities.

Chinese Translation

对话和协作助手不断判断合作伙伴的知识、信念或意图。类型化决策模型为有限的心智理论（Theory of Mind, ToM）选择提供了一种潜在的替代方案，取代生成性大型语言模型。我们提出了Jev-ToM，比较了五种类型化决策模型和七个生成性基线模型在来自九个ToM基准的11,117个英语文本问题上的表现，涵盖六个能力组别。TypeSafe Jev在包括本地控制在内的源平衡准确率上达到了73.2%，并在所有六个能力组中排名第一或第二，尤其在信念、目标/欲望/意图和情感方面表现突出。其他类型化模型在不同任务中展现出独特的优势。单独的资源比较表明，Jev可能降低重复ToM选择的成本和等待时间。基于信心的选择仍然依赖于基准，而提供的理由可以修正错误或推翻正确答案，具体取决于其附加方式。候选扩展揭示了独特的选择和输出有效性问题。相关问题提示占据了Jev下游收益的大部分，且预测状态对任务的影响具有依赖性。这些发现激励我们在助手中测试Jev作为ToM选择组件，任务特定的信心校准、候选设计和输出验证被视为整合的优先事项。

View on arXiv Download PDF AI Translation

cs.CL / 231 / 2610.18611

Legal-R1: Agentic Retrieval for Evidence-Based Legal Reasoning

Legal-R1：基于证据的法律推理的自主检索

Pang, Huaiwen, Han, Jiale, Chen, Junying, Xiao, Wenjun, Wang, Benyou

Abstract

Legal question answering requires answers that are not only plausible, but also grounded in legally valid and citable evidence. Standard retrieval-augmented generation often relies on one-shot semantic matching, which is insufficient when legal evidence must be selected according to authority, applicability, and diversity across heterogeneous sources. We introduce Legal-R1, an agentic retrieval framework for evidence-based legal reasoning. Legal-R1 decomposes legal reasoning into a multi-agent workflow, where a Lawyer agent learns when to retrieve, which legal source to query, what evidence to seek, and when the collected evidence is sufficient. A Legal Assistant agent retrieves from normative legal sources and precedent cases with authority-aware and diversity-aware selection, while a Judge agent generates the final answer from the complete evidence trajectory. We post-train only the Lawyer agent with masked supervised fine-tuning and trajectory-aware reinforcement learning. Experiments on multiple Chinese legal benchmarks show that Legal-R1 improves over general-purpose, legal-specific, and retrieval-augmented baselines.

Chinese Translation

法律问答不仅需要合理的答案，还需要基于合法有效且可引用的证据。标准的检索增强生成方法通常依赖于一次性语义匹配，但在法律证据必须根据权威性、适用性和来自异构来源的多样性进行选择时，这种方法显得不足。我们提出了Legal-R1，一个用于基于证据的法律推理的自主检索框架。Legal-R1将法律推理分解为一个多智能体工作流程，其中律师智能体学习何时检索、查询哪个法律来源、寻找什么证据以及何时收集的证据足够。法律助理智能体从规范法律来源和先例案例中进行权威性和多样性意识的选择进行检索，而法官智能体则从完整的证据轨迹中生成最终答案。我们仅对律师智能体进行了掩码监督微调和轨迹意识的强化学习后训练。在多个中文法律基准上的实验表明，Legal-R1在通用、法律特定和检索增强基线之上有所提升。

View on arXiv Download PDF AI Translation

cs.CL / 232 / 2610.18616

Knowledge Graphs as Files: An Agent Harness Where Tool Creation and Tool Use Co-Evolve

知识图谱作为文件：工具创建与工具使用共同演化的代理框架

Bai, Yuyang, Wang, Yu, Zhang, Yu

Abstract

Searching text-rich knowledge graphs requires following relations while examining textual evidence. Predefined graph interfaces constrain what evidence agents can access, even when they write code to combine tool calls or process retrieved subgraphs. We propose direct knowledge graph interaction (DKGI), a harness in which language model agents create graph access tools directly over complete node and edge files. The agent writes Python code to follow relations, combine node sets, and select textual evidence, interleaving execution with shell search and file inspection. This lets agents process large node sets in bulk while returning selected text for the model to assess. Execution feedback guides tool creation and revision, while the evolving tools determine what evidence the agent gathers next. On benchmark datasets spanning three domains, training-free DKGI exceeds the strongest published baseline by at least 6.7 Hit@1 points on every graph, even with a minimal harness. Controlled ablations show the benefits of multi-turn programming over shell-only access and predefined graph tools, while trajectories show agents revising code to extend searches or address errors. These findings highlight the value of allowing tool creation and tool use to co-evolve during graph search. Our code and data are available at https://github.com/leopoldwhite/DKGI.

Chinese Translation

在文本丰富的知识图谱中进行搜索需要在检查文本证据的同时跟踪关系。预定义的图接口限制了代理可以访问的证据，即使他们编写代码以组合工具调用或处理检索到的子图。我们提出了直接知识图谱交互（DKGI），这是一个框架，其中语言模型代理直接在完整的节点和边文件上创建图访问工具。代理编写Python代码以跟踪关系、组合节点集并选择文本证据，同时将执行与shell搜索和文件检查交错进行。这使得代理能够批量处理大量节点集，同时返回选定的文本供模型评估。执行反馈指导工具的创建和修订，而不断演化的工具决定了代理接下来收集的证据。在涵盖三个领域的基准数据集上，无需训练的DKGI在每个图上超越了最强的已发布基线至少6.7个Hit@1点，即使在一个最小的框架下。控制性消融实验显示了多轮编程相较于仅使用shell访问和预定义图工具的优势，而轨迹显示代理修订代码以扩展搜索或解决错误。这些发现强调了在图搜索过程中允许工具创建与工具使用共同演化的价值。我们的代码和数据可在https://github.com/leopoldwhite/DKGI获取。

View on arXiv Download PDF AI Translation

cs.CL / 233 / 2610.18627

Breakdown of Conversational Involvement: A Joint Multi-Modal Analysis of Conversation Floors and F-formation Detection with Multiple Features

对话参与的崩溃：基于多特征的对话层和 F-formation 检测的联合多模态分析

Li, Zonghuan, Raman, Chirag, Hung, Hayley

Abstract

Interactions happen universally and spontaneously in daily lives. Studying involvement in interactions helps us analyze who can influence who and has a wide potential in applications. Specifically, we consider conversational involvement in free-standing conversations. We study two operationalizations of an FCG: F-formations and conversation floors, and propose that current assumption that F-formations capture single conversations is insufficient. To illustrate how different implementations could impact the detected F-formations, we employ three analyses. First, we propose that speaking status can be used for estimating conversation floors, and leveraged it to analyze floors within F-formations. Second, we analyze spatial characteristics of the detection results by applying different feature sets on chosen baselines, and find that groups resulting from head features are more likely to be subsumed by groups detected using other features. Finally, we do a joint analysis which combines the spatial features and floor patterns in the conversation. We observe that certain pipelines can detect groups with single floors from upper body features, while retaining F-formations with lower body features. We conclude that a more detailed theory and framework is needed to holistically demonstrate conversational involvement within interactions.

Chinese Translation

互动在日常生活中普遍且自发地发生。研究互动中的参与性有助于我们分析谁能够影响谁，并在应用中具有广泛的潜力。具体而言，我们考虑独立对话中的参与性。我们研究了 FCG 的两种操作化方式：F-formation 和对话层，并提出当前假设 F-formation 捕捉单一对话的观点是不够的。为了说明不同实现如何影响检测到的 F-formation，我们进行了三项分析。首先，我们提出可以利用发言状态来估计对话层，并利用这一点分析 F-formation 内的对话层。其次，我们通过在选定基线应用不同特征集来分析检测结果的空间特征，发现由头部特征生成的群体更可能被使用其他特征检测到的群体所包含。最后，我们进行了一项联合分析，将空间特征和对话中的层模式结合在一起。我们观察到某些流程能够从上半身特征中检测到具有单一层的群体，同时保留下半身特征的 F-formation。我们得出结论，需要一个更详细的理论和框架，以全面展示互动中的对话参与性。

View on arXiv Download PDF AI Translation

cs.CL / 234 / 2610.18669

Known-Voice Cloning and the Limits of Personality Transfer for Character Design

已知声音克隆与角色设计中的个性转移限制

Kuch, Johanna Magdalena, van Lier, Maud, Don, Daksitha Senel Withanage, Gebhard, Patrick, Schneeberger, Tanja, André, Elisabeth

Abstract

Recent advances in zero-shot voice cloning systems make it possible to reproduce a recognizable voice from a short recording and attach it to arbitrary virtual characters. Prior work suggests that familiar vocal identity can carry over associated personality impressions to new targets such as robots or virtual characters. However, these works have mainly examined familiarity and personalization effects rather than directly testing whether a cloned familiar voice systematically transfers personality impressions to an embodied virtual character. Such transference raises the concern that cloned voices could be used to manipulate people into engaging with virtual characters as capable social actors. Across two studies, we investigated whether personality impressions transfer when a virtual character speaks with the cloned voice of a familiar person. Study 1 examined this question in a real-world university setting using the cloned voice of a lecturer familiar to the participating students. Study 2 examined the same question online using voices of social media content creators and members of their communities. Drawing on the concepts of identity activation, trait transfer, and person reconstruction, we show that evidence of personality transfer was limited in both studies. Observed effects were selective and did not generalize across traits or study contexts. In the online study in particular, recognizable voice identity cues activated expectations about the speaker but did not reconstruct the broader person impression. Qualitative responses further indicated that judgments were frequently shaped by perceived artificiality and by whether voice and appearance fit together. Our findings moderate common concerns that cloned voices straightforwardly manipulate people into perceiving artificial characters as capable social actors.

Chinese Translation

最近在零样本声音克隆系统方面的进展使得从短录音中重现可识别的声音并将其附加到任意虚拟角色上成为可能。先前的研究表明，熟悉的声音身份可以将相关的个性印象转移到新的目标，例如机器人或虚拟角色。然而，这些研究主要考察了熟悉度和个性化效果，而不是直接测试克隆的熟悉声音是否系统性地将个性印象转移到具身的虚拟角色上。这种转移引发了一个担忧，即克隆的声音可能被用来操控人们与虚拟角色互动，使其被视为具有社交能力的行为者。在两项研究中，我们调查了当虚拟角色使用熟悉人物的克隆声音说话时，个性印象是否会转移。研究1在一个真实的大学环境中考察了这个问题，使用了参与学生熟悉的讲师的克隆声音。研究2在线上考察了同样的问题，使用了社交媒体内容创作者及其社区成员的声音。基于身份激活、特质转移和人物重建的概念，我们发现两项研究中个性转移的证据有限。观察到的效果是选择性的，并未在特质或研究背景之间普遍适用。特别是在在线研究中，可识别的声音身份线索激活了对说话者的期望，但并未重建更广泛的人物印象。定性反馈进一步表明，判断常常受到感知的人工性以及声音与外貌是否相符的影响。我们的研究结果缓解了人们普遍担忧的克隆声音简单地操控人们将人工角色视为具有社交能力的行为者。

View on arXiv Download PDF AI Translation

cs.CL / 235 / 2610.18672

SMI: Ontology-Grounded Semantic IDs for Hierarchy-Aware Medical Concept Representation

SMI：基于本体的语义标识符用于层次感知的医学概念表示

Song, Ziyang, Shen, Lia, Li, Yixuan, Lu, Qincheng, Li, Weihao, Xu, Haoran, Zhu, He, Wang, Xinyu, Li, Yue

Abstract

Biomedical language models (LMs) provide strong representations for clinical tasks, but they often fail to capture the hierarchical structure among medical concepts. This limitation is especially pronounced for rare and fine-grained clinical codes in long-tailed vocabularies, where hierarchy-agnostic LM representations become less informative as concept frequency decreases. We introduce Semantic Medical Identifier (SMI), a training-free framework for constructing ontology-grounded Semantic IDs and hierarchy-aware medical concept representations. For each diagnosis, procedure, or medication concept, SMI defines its Semantic ID as the coarse-to-fine path through an expert-defined medical ontology. SMI then aggregates frozen biomedical LM embeddings across the corresponding ontology levels to form hierarchy-aware medical concept representations that preserve coarse-to-fine clinical semantics. The base SMI uses uniform aggregation across ontology levels, while a level-weighted variant selects global level weights on validation data. Our theoretical analysis shows that SMI yields more informative medical concept representations than hierarchy-agnostic LM embeddings, particularly for rare and fine-grained clinical concepts. Experiments show that SMI captures meaningful hierarchical structure and improves clinical relation prediction. On MIMIC-IV, SMI improves mortality prediction, readmission prediction, drug recommendation, and rare drug prediction. On eICU, SMI yields more robust patient representations under cross-hospital distribution shift.

Chinese Translation

生物医学语言模型（LM）为临床任务提供了强大的表示能力，但它们往往无法捕捉医学概念之间的层次结构。这一局限性在长尾词汇中的稀有和细粒度临床编码中尤为明显，因为与概念频率降低相关的无层次LM表示变得信息量较少。我们提出了语义医学标识符（SMI），这是一个无训练的框架，用于构建基于本体的语义标识符和层次感知的医学概念表示。对于每个诊断、程序或药物概念，SMI将其语义标识符定义为通过专家定义的医学本体的粗到细路径。然后，SMI在相应的本体层次上聚合冻结的生物医学LM嵌入，以形成保留粗到细临床语义的层次感知医学概念表示。基础SMI在本体层次上使用均匀聚合，而一个层次加权变体则在验证数据上选择全局层次权重。我们的理论分析表明，SMI比无层次LM嵌入产生更具信息量的医学概念表示，特别是对于稀有和细粒度的临床概念。实验表明，SMI捕捉到了有意义的层次结构，并改善了临床关系预测。在MIMIC-IV上，SMI提高了死亡率预测、再入院预测、药物推荐和稀有药物预测。在eICU上，SMI在跨医院分布变化下产生了更稳健的患者表示。

View on arXiv Download PDF AI Translation

cs.CL / 236 / 2610.18674

From Relation-Specific Heads to Shared Interventions: Contextual Entrainment in Language Models

从关系特定的头到共享干预：语言模型中的上下文同步

Zheng, Jinghan, Shi, Jiajun

Abstract

Large language models rely on prompt context to perform in-context learning, follow formats, and adapt to local evidence. This ability, however, also raises a robustness concern: models may reuse irrelevant tokens from earlier context when answering a later query. Prior work identifies this behavior as contextual entrainment and finds attention heads whose ablation reduces it, arguing that these heads are largely task- or relation-specific. But are the mechanisms behind contextual entrainment truly isolated by relation, or do relation-conditioned searches recover reusable components shared across relations? Using factual-distractor relation-completion prompts across 19 lexical-semantic subtypes, we identify attention heads whose ablation reduces entrainment and test how the resulting head sets transfer recovery ability across exact relations, semantic families, and mixed relation distributions. Same-family transfer exceeds different-family transfer, family residuals weaken once globally shared heads are removed, and subtype-specific searches repeatedly rediscover many global heads. Clean-completion and context-use probes further suggest that this shared intervention can reduce harmful copying while preserving ordinary relation completion and simple in-context abilities. Our findings provide a more precise perspective on contextual entrainment and its mitigation: harmful context following is not best addressed by searching for isolated relation-specific circuits, but by using heterogeneous attack prompts to reveal reusable intervention targets whose effects are shaped by semantic relation structure.

Chinese Translation

大型语言模型依赖于提示上下文进行上下文学习、遵循格式和适应局部证据。然而，这种能力也引发了稳健性问题：模型在回答后续查询时可能会重用早期上下文中的无关标记。先前的研究将这种行为称为上下文同步，并发现某些注意力头的消融可以减少这种现象，认为这些头主要是任务或关系特定的。但上下文同步背后的机制是否真的仅由关系孤立，还是关系条件搜索恢复了跨关系共享的可重用组件？通过使用19种词汇语义子类型的事实干扰关系补全提示，我们识别出消融某些注意力头可以减少同步现象，并测试结果头集在精确关系、语义家族和混合关系分布之间的恢复能力转移。同一语义家族的转移超过不同语义家族的转移，一旦全球共享头被移除，家族残余减弱，而子类型特定的搜索则反复重新发现许多全球头。干净补全和上下文使用探测进一步表明，这种共享干预可以减少有害复制，同时保持普通关系补全和简单的上下文能力。我们的发现为上下文同步及其缓解提供了更精确的视角：有害的上下文跟随并不最好通过寻找孤立的关系特定电路来解决，而是通过使用异质攻击提示来揭示可重用的干预目标，其效果受到语义关系结构的影响。

View on arXiv Download PDF AI Translation

cs.CL / 237 / 2610.18726

Code-Switching Detection via Language-Selective Dimensions in Speech Foundation Models

通过语言选择性维度在语音基础模型中检测代码切换

Chi, Jie, Aldeneh, Zakaria, Fedzechkina, Masha, de Seyssel, Maureen, Schluter, Natalie

Abstract

Code-switching (CS), the alternation between languages within an utterance, is common in multilingual speech but underrepresented in pretraining corpora. Increasing CS coverage requires finding relevant utterances in large speech collections so that they can be included in training data. One way to find such utterances is to transcribe the speech and detect language alternation in the text, but this depends on obtaining accurate multilingual transcriptions. We propose a lightweight method that detects CS directly from speech. It uses frame-level language information in pretrained speech encoders to determine whether both languages are present within an utterance. Specifically, we identify language-selective hidden-state dimensions using monolingual recordings and aggregate their frame-level activations into a CS prediction. With a small CS set used to tune a decision rule, we show that the Whisper-small encoder achieves a mean F1 of $0.798$ across 14 language pairs in CS-FLEURS, compared with $0.776$ for a CNN and $0.751$ for a HuBERT linear baseline trained for monolingual language identification. Our findings suggest that pretrained speech encoders already encode the information needed to identify CS utterances, and they provide a lightweight tool for building more representative multilingual corpora.

Chinese Translation

代码切换（CS）是指在一个话语中交替使用多种语言的现象，这在多语言语音中十分常见，但在预训练语料中却表现不足。增加代码切换的覆盖范围需要在大型语音集合中找到相关话语，以便将其纳入训练数据。找到这些话语的一种方法是对语音进行转录并在文本中检测语言交替，但这依赖于获得准确的多语言转录。我们提出了一种轻量级的方法，直接从语音中检测代码切换。该方法利用预训练语音编码器中的帧级语言信息来判断一个话语中是否同时存在两种语言。具体而言，我们使用单语录音识别语言选择性隐藏状态维度，并将其帧级激活聚合为代码切换预测。通过使用一个小型代码切换数据集来调整决策规则，我们展示了Whisper-small编码器在CS-FLEURS数据集中对14种语言对的平均F1值达到0.798，而CNN的平均F1值为0.776，HuBERT线性基线的平均F1值为0.751，这些基线是为单语语言识别训练的。我们的研究结果表明，预训练的语音编码器已经编码了识别代码切换话语所需的信息，并为构建更具代表性的多语言语料库提供了一种轻量级工具。

View on arXiv Download PDF AI Translation

cs.CL / 238 / 2610.18762

Looking Complete Is Not Being Grounded: Evaluating LLM Sitreps Against Human-Authored Situation Reports

看似完整并不意味着扎根：评估大型语言模型（LLM）生成的情况报告与人类撰写的情况报告的对比

Kumar, Sachin

Abstract

Evaluation for automatic crisis reporting should measure what practitioners need, not what is convenient to compare. Through LLM-assisted, author-approved content analysis, we derive a 15-slot information schema from professional humanitarian situation reports using qualitative content analysis, then evaluate 121 human and 261 LLM-generated reports across five disasters and three hazard types. Under google/gemini-3.7-flash on Cyclone Idai, the best machine arm is nominally above the empirical human ceiling (0.827 vs. 0.821), but the difference is statistically indistinguishable from zero (95\% bootstrap CI $[-.049,.059]$; permutation $p=.852$), and the ordering does not replicate under openai/gpt-5.6-luna. On a blind 100-item Idai sample, the author agrees with the Gemini judge at 0.910 accuracy and unweighted Cohen's $\kappa=.799$. All four judges clear the plan's raw-agreement criterion of .7; only Gemini reaches .7 on chance-corrected $\kappa$. Across the three primary events, machine reports reproduce 5.4-11.0% of same-day official figures. Across those same events, raw exact-number matching links 14.3-19.6% of official figures and 85.4-90.1% of machine figures to the social stream; manual audit confirms 45/50 declared absences, 4/25 official matches, and 24/25 machine matches semantically. The issuing-provenance gap is positive in all 12 observed event-judge cells: report date, organisation, series, and authority were not supplied at professional-report coverage in any cell. On the three primary events, ROUGE-L is uninformative, BERTScore is negatively associated with completeness after multiplicity correction, and a generic LLM judge changes sign.

Chinese Translation

自动危机报告的评估应当衡量从业者所需的信息，而非便于比较的内容。通过大型语言模型（LLM）辅助的、作者批准的内容分析，我们利用定性内容分析从专业人道主义情况报告中提取了一个包含15个信息槽的信息架构，并对121份人类报告和261份LLM生成的报告进行了评估，涵盖五个灾害和三种风险类型。在使用google/gemini-3.7-flash评估的赛风伊代（Cyclone Idai）案例中，最佳机器模型的表现名义上高于经验性的人工上限（0.827对比0.821），但这一差异在统计上与零无显著差异（95\%自助法置信区间 $[-.049,.059]$；置换 $p=.852$），且在openai/gpt-5.6-luna下的排序并未重现。在一个盲测的100项伊代样本中，作者与Gemini评审的准确率为0.910，未加权的Cohen's $ ext{kappa}=.799$。所有四位评审均达到了计划的原始一致性标准0.7；只有Gemini在机会校正的$ ext{kappa}$上达到了0.7。在三次主要事件中，机器报告重现了5.4-11.0%的同日官方数据。在这些事件中，原始的精确数字匹配将14.3-19.6%的官方数据和85.4-90.1%的机器数据与社交流关联；人工审核确认了50个声明缺失中的45个，25个官方匹配中的4个，以及25个机器匹配中的24个在语义上的一致性。所有12个观察到的事件-评审单元中的发行来源差距均为正值：在任何单元中，专业报告覆盖时未提供报告日期、组织、系列和权威信息。在三次主要事件中，ROUGE-L无信息量，BERTScore在多重性校正后与完整性呈负相关，而一个通用的LLM评审的结果则发生了变化。

View on arXiv Download PDF AI Translation

cs.CL / 239 / 2610.18766

Rubrics from the Task, Verdicts from Evidence: Pointwise Rubric Rewards with Tool-Augmented Judges

基于任务的评分标准，基于证据的裁决：工具增强评审的逐点评分标准奖励

Liu, Tianci, Wang, Haoyu

Abstract

Rubric-based reward models (RMs) evaluate responses against explicit criteria. However, existing methods often derive criteria from observed response candidates, introducing response-induced bias that may limit rubric generalizability. Their judges also struggle with mechanical constraints when relying on reading alone, and commonly used pairwise judges provide only relative preferences, complicating their direct use for policy optimization. We propose ReVeR, a pointwise rubric RM that derives hard rules and graded principles by reasoning about the instruction alone and evaluates responses independently through tool-augmented judging. Both components are trained through label-blind distillation, in which preference labels only filter a teacher's outputs, yielding ReVeR-Traces, a large-scale dataset of rubric-reasoning and judging trajectories. Experiments demonstrate that ReVeR outperforms pairwise rubric RMs in average preference accuracy, most clearly on instruction following, as well as in test-time scaling and downstream policy optimization. We further find that generator RL improves preference accuracy while reducing rubric fidelity, and that higher preference accuracy does not necessarily translate into better downstream policies. These findings highlight that rubric fidelity, preference accuracy, and downstream utility are distinct properties, shedding light on future RM research.

Chinese Translation

基于评分标准的奖励模型（RMs）根据明确的标准评估响应。然而，现有方法通常从观察到的响应候选中推导标准，这引入了响应引发的偏差，可能限制评分标准的普遍适用性。它们的评审在仅依赖阅读时也面临机械限制，而常用的成对评审仅提供相对偏好，使其在政策优化中的直接应用变得复杂。我们提出了ReVeR，这是一种逐点评分标准RM，通过仅对指令进行推理来推导硬规则和分级原则，并通过工具增强的评审独立评估响应。两个组件均通过标签盲蒸馏进行训练，其中偏好标签仅过滤教师的输出，产生ReVeR-Traces，这是一个大规模的评分标准推理和评审轨迹数据集。实验表明，ReVeR在平均偏好准确性上优于成对评分标准RM，尤其在遵循指令方面，以及在测试时的扩展性和下游政策优化中。我们进一步发现，生成器强化学习提高了偏好准确性，同时降低了评分标准的保真度，而更高的偏好准确性并不一定转化为更好的下游政策。这些发现强调了评分标准保真度、偏好准确性和下游效用是不同的属性，为未来的RM研究提供了启示。

View on arXiv Download PDF AI Translation

cs.CL / 240 / 2610.18786

Where Do LLM Ensemble Gains Come From? Decomposing Candidate Value, Diversity, and Aggregation

大型语言模型集成增益来自何处？候选值、差异性与聚合的分解

Kim, Donghwan

Abstract

LLM ensemble gains are often measured against a single baseline, but the value attributed to additional candidates can change with the comparison used. We study an LLM aggregator under four matched conditions: Direct (no candidates), Single-slot (one candidate in a multi-slot interface), Duplicate (one candidate repeated), and Multi-candidate (separately generated candidates). Under either non-diverse control, overall candidate gain (CIV) splits algebraically into an additional-candidate contrast (DSV) and a candidate-context contrast (CCV); the split, but not CIV, depends on the chosen control. On MultiHopRAG model diversity (k=2), the repetition gap Rep=-8.33 pp [-13.25, -3.75] quantifies the difference between gains measured against Duplicate and Single-slot. On a matched FRAMES stronger-panel extension (n=200, k=3, aggregator output limit 4096), paired Delta Rep=+9.67 pp [+6.00, +13.50]; within panels, Rep changes sign (original -6.33; stronger +3.33). In FRAMES evidence aggregation, candidate-context contrasts remain strongly negative under both non-diverse controls. Our results show why ensemble evaluations should state the reference control before attributing gains to additional candidates.

Chinese Translation

大型语言模型（LLM）集成增益通常是相对于单一基线进行测量的，但额外候选者所赋予的价值可能会随着所用比较而变化。我们在四种匹配条件下研究了LLM聚合器：直接（无候选者）、单槽（多槽界面中的一个候选者）、重复（一个候选者重复）和多候选者（单独生成的候选者）。在非多样性控制下，总体候选增益（CIV）在代数上分解为额外候选对比（DSV）和候选上下文对比（CCV）；这种分解，但不是CIV，依赖于所选择的控制。在MultiHopRAG模型的多样性（k=2）下，重复差距Rep=-8.33个百分点[-13.25, -3.75]量化了相对于重复和单槽的增益之间的差异。在匹配的FRAMES更强面板扩展（n=200，k=3，聚合器输出限制4096）中，配对Delta Rep=+9.67个百分点[+6.00, +13.50]；在面板内，Rep的符号发生变化（原始-6.33；更强+3.33）。在FRAMES证据聚合中，候选上下文对比在两种非多样性控制下仍然强烈为负。我们的结果表明，集成评估在将增益归因于额外候选者之前应声明参考控制。

View on arXiv Download PDF AI Translation

cs.CL / 241 / 2610.18818

Back-Translation Still Matters: Leveraging Monolingual Data for Low-Resource Machine Translation with LLMs

反向翻译仍然重要：利用单语数据提升低资源机器翻译的效果

Bathla, Divyanshu, Singla, Parag, Singla, Jitin

Abstract

Pre-trained Large Language Model (LLMs) are effective on many tasks in zero-shot and few-shot settings. For machine translation of low-resource languages, the obvious options are few-shot prompting and fine-tuning on a small parallel set. We ask: what if moderate amounts of low-resource monolingual data are also available? We argue that, rather than continued pretraining, a more effective use of this data is the pre-LLM era technique of back-translation (BT). The current fine-tuned LLM translates monolingual sentences in each language, and the resulting pairs train it in the reverse direction. Updates on the parallel set are interleaved for stable learning. With 1K parallel pairs on Odia and Swahili, BT beats continued pretraining into the low-resource language by 7.7 to 16.4 XCOMET-XXL points in domain. With beam-decoded intermediate translations, it also beats a model with 10 time more parallel data. In-context examples retrieved from the parallel data further improve BT. We also analyze the source, quality and amount of monolingual text. Across five low-resource languages, with Hindi as a high-resource control, and two 3B backbones, BT improves translation into the low-resource language by 16 to 173% XCOMET-XXL over parallel-only fine-tuning. Blind human evaluation on Odia and Sanskrit agrees with the automatic metrics.

Chinese Translation

预训练的大型语言模型（LLMs）在零样本和少样本设置下对许多任务表现出色。对于低资源语言的机器翻译，显而易见的选择是少样本提示和在小规模平行数据集上的微调。我们提出：如果还有适量的低资源单语数据可用，该如何处理？我们认为，与其继续预训练，更有效的利用方式是采用反向翻译（Back-Translation, BT）这一预LLM时代的技术。当前的微调LLM在每种语言中翻译单语句子，生成的句对用于反向训练。平行数据集的更新交替进行，以确保学习的稳定性。在Odia和Swahili语言上使用1000个平行句对，BT在领域内超越了继续预训练的低资源语言，提升了7.7到16.4个XCOMET-XXL点。通过束搜索解码的中间翻译，BT的表现也超过了拥有10倍平行数据的模型。从平行数据中检索的上下文示例进一步改善了BT的效果。我们还分析了单语文本的来源、质量和数量。在五种低资源语言中，以印地语作为高资源对照，结合两个3B骨干网络，BT在低资源语言的翻译效果上比仅使用平行微调提高了16%到173%的XCOMET-XXL分数。对Odia和梵语的盲人评估结果与自动评测指标一致。

View on arXiv Download PDF AI Translation

cs.CL / 242 / 2610.18861

Hybrid Reasoning Agents

混合推理智能体

Zou, Jiaru, Qi, Yunzhe, Chen, Sirui, Bianchi, Federico, Zhu, Shang, Lu, Pan, He, Jingrui, Zou, James

Abstract

How can we build an agent that thinks beyond human language, yet still acts through it in the external environment? Modern LLM agents still reason and interact through text, while latent reasoning offers an efficient internal medium but lacks native tool access and reasoning transparency. We bridge these two directions with Hybrid Reasoning Agents (HybridAgents), which dynamically navigate between latent reasoning and text-based interaction throughout a unified generation process. Architecturally, a lightweight HybridLink enables seamless transitions between text and latent modes, while the agent's continuous working memory preserves lossless hybrid context across multiple turns. These designs preserve tool-integrated agentic capabilities while making latent thoughts directly inspectable. Building on this architecture, we further develop dedicated post-training algorithms for hybrid reasoning: HybridSFT stabilizes tool-integrated hybrid generation by teaching the backbone to coordinate with HybridLink, then HybridRL scales reasoning by learning when to think in latent space and when to act through text. We evaluate HybridAgent at 3 model scales across 11 benchmarks covering mathematics, science, medicine, search, and agentic reasoning. Compared with frontier LLM Agents and agentic frameworks spanning both text-based and latent methods, HybridAgent consistently improves performance by 10.4%, while delivering 35.5%-54.7% token reduction, 1.8x-2.6x faster inference, and 2.7x-3.7x cost savings. Code and data are available at https://github.com/togethercomputer/HybridAgent.

Chinese Translation

我们如何构建一个超越人类语言思维的智能体，同时仍然通过语言在外部环境中进行行动？现代的大型语言模型（LLM）智能体仍然通过文本进行推理和交互，而潜在推理提供了一种高效的内部媒介，但缺乏原生工具访问和推理透明性。我们通过混合推理智能体（Hybrid Agents，HybridAgents）将这两个方向结合起来，HybridAgents在统一的生成过程中动态地在潜在推理和基于文本的交互之间切换。在架构上，一个轻量级的HybridLink使得文本模式和潜在模式之间的无缝过渡成为可能，同时智能体的连续工作记忆在多个回合中保持无损的混合上下文。这些设计保留了工具集成的智能能力，同时使潜在思维可以直接检查。在此架构基础上，我们进一步开发了专门的后训练算法以实现混合推理：HybridSFT通过教导主干与HybridLink协调，稳定工具集成的混合生成；然后，HybridRL通过学习何时在潜在空间思考以及何时通过文本行动来扩展推理。我们在11个涵盖数学、科学、医学、搜索和智能推理的基准测试中，对3种模型规模的HybridAgent进行了评估。与前沿的LLM智能体和涵盖文本和潜在方法的智能框架相比，HybridAgent的性能一致提高了10.4%，同时实现了35.5%-54.7%的令牌减少、1.8倍至2.6倍的推理速度提升和2.7倍至3.7倍的成本节约。代码和数据可在 https://github.com/togethercomputer/HybridAgent 获取。

View on arXiv Download PDF AI Translation

cs.CL / 243 / 2610.18862

DoubtLess: Training-Free Contrastive Decoding for Efficient Reasoning

DoubtLess：无训练的对比解码以实现高效推理

Luo, Minghao, Chen, Liang

Abstract

Reasoning models achieve strong performance through long chains of thought (CoT), but often spend more tokens than needed to solve a problem. Existing ways to shorten the CoT require additional training for each model, add a fixed steering vector to the model's hidden states, or use decoding heuristics such as stopping early or banning words like "wait", which can reduce accuracy. We propose DoubtLess, a training-free decoding method in which a much smaller reasoning model steers the large one towards brevity. The small model reads the large model's CoT so far twice, once prompted to be brief (the brief thinker) and once prompted to be verbose (the verbose thinker), and the difference between its two logit vectors, scaled by a strength, is added to the large model's logits at every step of the CoT. On seven benchmarks of competition mathematics, graduate-level science and competitive programming, DoubtLess shortens the CoT of Qwen3-32B by 31% with unchanged accuracy, outperforming five inference-time baselines. We also show that DoubtLess reduces thinking tokens per turn by 20% in a self-debugging coding agent with unchanged task performance. It also generalizes to two other model families with unchanged accuracy. When combined with training-based compression, it further shortens the CoT of a model already trained for brevity. Ablations and behavioral analysis show that DoubtLess reduces repeated checking, with most token savings coming after the correct answer first appears in the CoT. Our code is available at https://github.com/leoluolol/DoubtLess.

Chinese Translation

推理模型通过长链思维（CoT）实现了强大的性能，但往往在解决问题时消耗的标记超过所需。现有的缩短CoT的方法需要对每个模型进行额外训练，向模型的隐藏状态添加固定的引导向量，或使用解码启发式方法，如提前停止或禁止使用“wait”等词，这可能会降低准确性。我们提出了DoubtLess，这是一种无训练的解码方法，其中一个更小的推理模型引导大型模型朝着简洁性发展。小模型对大型模型的CoT进行两次读取，一次是被提示简洁（简洁思考者），一次是被提示冗长（冗长思考者），并将其两个logit向量之间的差异，按一定强度缩放后，添加到大型模型在CoT每一步的logits中。在七个竞争数学、研究生级科学和竞争编程的基准测试中，DoubtLess将Qwen3-32B的CoT缩短了31%，而准确性保持不变，超越了五个推理时基线。我们还展示了DoubtLess在自我调试编码代理中将每轮思考的标记减少了20%，而任务性能保持不变。它还在另外两个模型系列中推广，准确性保持不变。当与基于训练的压缩相结合时，它进一步缩短了已经经过简洁训练的模型的CoT。消融实验和行为分析表明，DoubtLess减少了重复检查，大多数标记节省发生在正确答案首次出现在CoT之后。我们的代码可在 https://github.com/leoluolol/DoubtLess 获取。

View on arXiv Download PDF AI Translation

cs.CL / 244 / 2610.18868

Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards

Tev：通过密集进展奖励扩展长时间跨度的自主环境

Hu, Wenbo, Pramanick, Shraman, Singh, Krishna Kumar, Peng, Nanyun, Chang, Kai-Wei, Lee, Yong Jae

Abstract

Long-horizon agents have drawn growing attention for their ability to work autonomously -- from hours to days -- on economically valuable tasks, a capability essential to automated discovery and recursive self-improvement (RSI). Yet open-source models still struggle on long-horizon tasks, largely for lack of both training data and a working recipe, and scaling both is hard: long, complex tasks are difficult to curate and verify, and their rewards are sparse. We address both. First, we introduce Tev-Gym, the first long-horizon agent training dataset with verifiable, dense, progressive rewards -- 1,170 verified task environments across 7 domains. It is built on a latent-world-first approach that first constructs a latent world: a complete and correct hidden specification of the task domain, workflow, and expected state after each step. Each task is then progressively rendered from this specification into an agent-facing environment, enabling controllable complexity and a natural progress metric. Second, we turn this metric into a training recipe: via progress-potential reward shaping (PPRS), we convert progress into a dense per-step reward that improves training efficiency and makes otherwise-intractable tasks trainable. Training with this recipe yields Tev-27B, which achieves substantial gains on the Agents' Last Exam computing and math sciences domains (ALE-CM), matching GPT-6 Astra and Claude Opus 5 at Pass@5. Despite training only on ALE-related data, it generalizes to the Terminal-Bench (TB) 3.0 and 4.0 subsets, surpassing GLM 5.2 and Gemini 3.7 Flash, and reaches Claude Opus 4.8-level performance on middle-horizon tasks such as TB 2.1.

Chinese Translation

长时间跨度的自主体因其能够在经济价值任务上自主工作（从数小时到数天）而受到越来越多的关注，这一能力对于自动化发现和递归自我改进（RSI）至关重要。然而，开源模型在长时间跨度任务上仍然面临困难，主要是缺乏训练数据和有效的工作方案，而这两者的扩展都很困难：长而复杂的任务难以策划和验证，其奖励也稀疏。我们解决了这两个问题。首先，我们介绍了Tev-Gym，这是第一个具有可验证、密集、渐进奖励的长时间跨度自主体训练数据集——涵盖7个领域的1,170个经过验证的任务环境。它基于潜在世界优先的方法，首先构建一个潜在世界：任务领域、工作流程和每一步后预期状态的完整且正确的隐藏规范。然后，每个任务从这个规范逐步渲染为面向自主体的环境，实现可控的复杂性和自然的进展度量。其次，我们将这一度量转化为训练方案：通过进展潜力奖励塑形（PPRS），我们将进展转化为密集的逐步奖励，从而提高训练效率，使得原本难以处理的任务变得可训练。使用这一方案进行训练的Tev-27B，在自主体最后考试计算和数学科学领域（ALE-CM）上取得了显著进展，达到了与GPT-6 Astra和Claude Opus 5在Pass@5上的相匹配的水平。尽管仅在ALE相关数据上进行训练，它在Terminal-Bench（TB）3.0和4.0子集上表现出良好的泛化能力，超越了GLM 5.2和Gemini 3.7 Flash，并在TB 2.1等中间时间跨度任务上达到了Claude Opus 4.8级别的表现。

View on arXiv Download PDF AI Translation

arXiv Papers

RoboHarness: A Simple Harness Outperforms VLA and World Actions Models

LabInstruct: Benchmarking Situated Instructional Video Generation for Lab Procedures

OIL: An Object Interaction Layer for Humanoid Whole-Body Control

Building Trustworthy Household Robots through Open-World Safety

Residual Compliant Policy: Learning Compliant Policies from Few Demonstrations

From bits to atoms: A survey of cross-layer safety in Embodied AI

RAIN: Robotic Task Generalization with Region-Aware Interaction Networks

Reinforcement Learning from Real-Robot Interaction without High-Quality State Estimates

Terrain Traversal Without Motion Priors: Expert-Routed Humanoid Locomotion at Deployment Fidelity

Do My Samples Help? Compute Budget Allocation for Real-World Sampling-Based MPC in Contact-Rich Manipulation

Safety-Oriented Trajectory Ranking from BEV Perception Outputs

From Hitter to Player: Separating Motor Execution and Strategy in Competitive Humanoid Table Tennis

Assessing the Marginal Utility of Exocentric Head Views in UMI-Style Large-FoV Wrist-Camera Manipulation

SenseWeather: Multimodal Sensing for Robust Robotic Perception under Extreme Conditions

When pose graphs are not enough: a failure analysis of pose-graph-based camera estimation for Gaussian splatting

Embodied Zero-Shot Task Planning via Hardware-Aware LLM-VLM Integration for Grounded Robotic Manipulation

Make {\pi}0.5 Learn to Call Astra for Reliable Robot Manipulation

Refine Where It Matters: Contact-Adaptive Meshing for Deformable Object Manipulation

Axiom-0: Hierarchical Self-Refinement for Vision-Language-Action Policies

Disentangling Task-Relevant Latent Dynamics for Reward Modeling

Observation-Referenced Tissue Deformation Forecasting with Tool-Driven and Residual Dynamics

ClothCode: Generative Modeling for Garment State Estimation and Dynamics Prediction

Mapping Tire-Road Friction Coefficients Along Driven Trajectories

Composable Dynamics for Cooperative Humanoids: A RoboCompiler Extension

From LLM Plans to Authorized Motion: Execution-Bound Authorization for Excavators

Akule: Fast and Scalable Multi-Robot Motion Planning via Sparse Interaction Diffusion

PILOT: Guiding Long-Horizon World-Model Planning for Robot Manipulation with Visual Subgoals and Policy Priors

CoPlanVLM: Coordinated Path Planning for a Robot Team using Vision Language Models

DexFLEX: Contact-Aware Foundation Controller for Command-Guided Dexterity

OATH: Object-Agnostic Tracking for Dexterous Hands, Enabling Rapid Physics-based Retargeting

Memory Without Curation: Learning Multi-Stream Latent Memory for Robot Policies

Red-Teaming of Vision-Language-Action Models via Quality Diversity Environment Generation

Shared-Control Teleoperation of Omnidirectional Aerial Robots for Continuous Valve Rotation

From Field to Planning: Learning Temporal Polar Gradient Fields for Vision-based AAV Trajectory Planning in Dynamic Environments

Multi-modal Representation Fusion for Long-term Microsurgical Anastomosis Trajectory Anticipation

Safety-Critical Quadrotor Navigation Using Fourth-Order Control Barrier Functions with Full Nonlinear Quadrotor Dynamics

Who to Trust in Multi-Robot Systems: Joint Robot and Object Hypothesis Trust Inference via Neural-Symbolic Cross-Validation

GG-Nav: Goal-oriented Grounding Chain-of-Thought Elicits Reasoning in Navigation Foundation Models

Before We Act: Learning Action-Conditioned Team-Response Beliefs for Decentralized Multi-Robot Manipulation

From Video to Vision: Learning Generalizable Humanoid Carrying from Human Motion

Chunk-wise Model-based Policy Optimization

In View or Out of View? A Vision-Aware Filter that Infers and Reacts to the Human's Field of View

Remembering What Matters for Long-Horizon Robotic World Models

PC7 Model: Dynamics of Expanding and Contracting Soft Robots using Piece-wise Constant 7D Deformation Based on Similarity Transformation Group

Flow-Support-Weighted Proposals for Test-Time Verified VLA Policies

ToF-Wheel Odometry: Navigating Degeneration of Localization With Limited Sensing Capabilities

Load-Aware Quadruped Navigation with Online Contact-based Geometry Estimation

StateSkill-Agent: An Effective and Efficient Embodied Agent via State-conditioned Skill

AnyRank: Reliable Anytime Policy Ranking via Sequential Testing

VbotNav-0: Asynchronous Cloud-Edge Dual-Policy Navigation across Indoor and Outdoor Environments

Global Optimality of Pose Graph Optimization

Successful but Unsafe: An Analytical Benchmark for Revealing Collisions in Vision-and-Language Navigation

DexCast: Learning Real-World Dexterous Control from Human Videos

BGC-Nav: Cross-Altitude Multi-UAV Collaborative Navigation via Belief-Guided Coupling

PRISM: Precision Forcing for Generalist Vision-Language-Action Policies

C2Flight: A Closed-Loop Co-Simulation Framework for Early-Stage Evaluation of Heterogeneous UAV Computing Systems

CAG-WAM: Causal Action-Gain World Action Models for Robot Manipulation

B-Human '26 - Scalable Team Play up to 11 vs. 11

BridgeWAM: Bridging Video and Action Experts via Latent Bridge Queries

Beyond Fixed Visual History: Learning Motion-Adaptive Temporal Receptive Fields for Complex-Terrain Quadruped Locomotion

Scaling Latent Motor Adaptation to Coordinated Bimanual Manipulation

Push the Limit: Learning High-Speed Sprinting and Aggressive Cornering on a Full-Scale Humanoid

Learning Dynamic Grasping

DEFT-VLA: Stage-Localized Force Specialization for Contact-Rich Robotic Manipulation

Predict in Chunks, Revise in Spans: History-conditioned Action Regeneration for Robotic Manipulation

Verify, Act, Distill: Turning Test-Time Compute into Test-Time Adaptation for Vision-Language-Action Models

LETS: Low-Overhead External-Torque Sensing for Force-Feedback Bilateral Teleoperation

Test-Time Compute for Humanoid Behavior Foundation Models

Scalable Multi-Human Cooperative Manipulation via Commander-Executor Coordination

Funnel planning and control for manipulators using neural Signed Configuration Distance Functions

Edge-Native Temporal--Graph Dual-Expert Inference for Dexterous Grasp Classification

CFG-Policy: Unified Diffusion Policy Learning Framework via Classifier-free Guidance

FoundationVIO: Enforcing Metric Scale in Feed-forward Visual-Inertial Odometry

EnactOpt: Diagnosis-Guided Harness Evolution for Embodied Agents

Barrier-Constrained Guidance For Real-time Asynchronous VLA Execution

Design and Evaluation of a Collaborative Teleoperation Interface for Multi-Operator Multi-Robot Systems in Challenging Environments

Training-Free Multi-Agent Cooperative Embodied Search from Uncertain Vision-Language Priors

Modeling and Control of a Rigid-soft Hybrid-link Humanoid Robot

Targeting the Right Situations: Mutation Testing of Fuzzy-Powered Situation Hyperspace