arXiv Daily Digest

477

Papers

Does Dynamic-Point Filtering Help When Texture Is Scarce? A Controlled Study of ORB-SLAM2 Front-Ends in Synthetic Indoor Scenes

动态点滤波在纹理稀缺时是否有帮助？对合成室内场景中ORB-SLAM2前端的控制研究

Xue, Zekui

Abstract

Dynamic-point filters are routinely added to feature-based visual SLAM, and several recent systems argue that removing dynamic features can leave too few static features in low-texture regions. So far, these systems have been evaluated only on texture-rich benchmark sequences. We present a controlled study that isolates this interaction. We render synthetic indoor sequences in which surface texture (four levels, quantified by FAST-corner density and image-gradient entropy) and scene dynamics (three levels) are varied factorially along identical camera trajectories, with stereo, RGB-D, ground-truth poses and dynamic masks. On this grid we compare ORB-SLAM2 without filtering, with an optical-flow and epipolar-residual filter (FLOW), and with a multi-view depth-consistency filter (GEOM), and report trajectory error, tracking completeness and surviving static features over five runs. Because the masks give per-keypoint ground truth, we also measure each filter's dynamic-point precision and recall and its static-feature false-removal rate, so that mechanistic explanations can be tested directly. We do not propose a new filter. On 720 runs over 24 sequences, filtering helped mainly in the most dynamic cells; the benefit did not decline monotonically with texture, but at the lowest level filtering reduced tracking completeness, and ORB-SLAM2 never initialised in static L3 scenes. Contrary to our hypothesis, GEOM discarded more static keypoints than FLOW (median FRR 6.8% vs. 1.5% for RGB-D, 19.6% vs. 1.5% for stereo); its RGB-D advantage tracked dynamic-point recall and vanished in stereo mode. Data and code are available at https://github.com/felixxxue/texture-dynamics-slam.

Chinese Translation

动态点滤波器通常被添加到基于特征的视觉SLAM中，最近的一些系统认为，在低纹理区域去除动态特征可能会留下过少的静态特征。到目前为止，这些系统仅在纹理丰富的基准序列上进行了评估。我们提出了一项控制研究，以隔离这种相互作用。我们渲染了合成的室内序列，其中表面纹理（四个级别，通过FAST角点密度和图像梯度熵量化）和场景动态（三个级别）在相同的相机轨迹上进行了因子变化，并使用立体、RGB-D、真实位姿和动态掩模。在这个网格上，我们比较了不进行滤波的ORB-SLAM2、光流和极线残差滤波器（FLOW）以及多视图深度一致性滤波器（GEOM），并报告了轨迹误差、跟踪完整性和存活静态特征的情况，进行了五次实验。由于掩模提供了每个关键点的真实值，我们还测量了每个滤波器的动态点精度和召回率以及静态特征的误移除率，以便直接测试机械解释。我们并不提出新的滤波器。在24个序列的720次实验中，滤波主要在最动态的单元中有所帮助；这种好处并没有随着纹理的增加而单调下降，但在最低水平上，滤波减少了跟踪完整性，而ORB-SLAM2在静态L3场景中从未初始化。与我们的假设相反，GEOM丢弃的静态关键点比FLOW更多（RGB-D的中位数FRR为6.8%对比1.5%，立体的中位数FRR为19.6%对比1.5%）；其RGB-D优势与动态点召回率相关，并在立体模式下消失。数据和代码可在https://github.com/felixxxue/texture-dynamics-slam获取。

View on arXiv Download PDF AI Translation

cs.RO / 2 / 2610.10601

Teaching a Robot Dog New Tricks: Diverse Quadruped Skills via Combined Reinforcement and Imitation Learning with Adversarial Task Selection

教机器人狗新技能：通过结合强化学习和模仿学习的对抗任务选择实现多样化的四足技能

Foxmere, Lemon, Furman, Anthony, Du, Yizheng, Chang, Oliver, Gilpin, Leilani, McGuire, Steve

Abstract

Reinforcement Learning (RL) has enabled legged robots to perform a range of skills in single-task settings. However, applications such as farm robotics or space exploration require diverse skills such as locomotion, digging, or close-range surveying. Training an end-to-end policy to address this problem remains difficult due to challenges such as sample inefficiency and gradient conflict between tasks in multi-task learning. We propose a three-stage method that trains a single policy to perform distinct tasks such as walking, digging, and hopping, and compose them into novel behaviors such as crawling. First, multiple teacher policies are trained using RL on narrowly defined tasks. Then, two additional stages train a student policy with a multi-teacher distillation setup that uses a combined RL and Imitation Learning (IL) objective under an adversarial task selection process that focuses training on the worst-performing task. With this method, we train a student policy that performs 22 tasks using 8 teachers. Evaluations show our method preserves motion quality and tracks commands more accurately than PPO and distill-then-finetune baselines, and in some cases generalizes to new tasks without explicit training. Finally, we demonstrate real-world robustness by deploying the resulting policy on a Unitree B1 quadruped. Video: https://youtu.be/V9yX04EBcFA

Chinese Translation

强化学习（Reinforcement Learning, RL）使得腿部机器人能够在单任务环境中执行一系列技能。然而，农场机器人或太空探索等应用需要多样化的技能，例如移动、挖掘或近距离测量。由于样本效率低下和多任务学习中任务之间的梯度冲突等挑战，训练一个端到端的策略来解决这个问题仍然很困难。我们提出了一种三阶段的方法，训练一个单一策略来执行不同的任务，如行走、挖掘和跳跃，并将其组合成新的行为，如爬行。首先，使用RL在狭义定义的任务上训练多个教师策略。然后，两个额外阶段通过多教师蒸馏设置训练一个学生策略，该设置在对抗任务选择过程中使用结合的RL和模仿学习（Imitation Learning, IL）目标，重点训练表现最差的任务。通过这种方法，我们训练了一个学生策略，能够使用8个教师执行22个任务。评估结果表明，我们的方法在运动质量上保持良好，并且比PPO和蒸馏后微调的基线更准确地跟踪指令，在某些情况下能够在没有明确训练的情况下推广到新任务。最后，我们通过在Unitree B1四足机器人上部署所得到的策略展示了其在现实世界中的鲁棒性。视频链接：https://youtu.be/V9yX04EBcFA

View on arXiv Download PDF AI Translation

cs.RO / 3 / 2610.10637

TacHair: Tactile Contact-Distribution Guided Online Correction for Robotic Hair Stroking and Perception

TacHair：基于触觉接触分布的机器人毛发抚摸与感知的在线校正

Hu, Ruiyi, Zhao, Yongqiang, Bak, Daniel, Wang, Yupeng, Zhang, Xuyang, Luo, Shan

Abstract

Hair stroking is common in daily grooming and personal care, and is also widely used in hair-product evaluation, motivating robots with similar physical interaction capabilities. Existing robotic hair-care and surface-following methods mainly rely on trajectory planning, compliance, force regulation, or tactile-conditioned policies, but deformable hair can remain in contact while gradually drifting across the end-effector, making local interaction difficult to regulate. We propose TacHair, a tactile contact-distribution guided online correction framework that represents high-resolution tactile observations as a spatial hair-contact distribution. A visuotactile imitation policy generates the nominal stroking motion, while a separately trained residual module corrects local contact deviations, separating task progression from contact recovery. We evaluate TacHair in 525 real-robot trials across five head geometries and three hair conditions. A successful stroke requires both sufficient task progression and contact maintenance; our method improves success from 42.9% to 62.3% and contact maintenance from 59.4% to 88.0% over the same visuotactile policy without correction. These results demonstrate spatial tactile contact distributions as an effective feedback representation for contact-preserving interaction with deformable and visually occluded surfaces. Demos, code, and datasets are available at https://tachair.github.io.

Chinese Translation

毛发抚摸在日常美容和个人护理中十分常见，也广泛应用于护发产品的评估，这激励了具备类似物理交互能力的机器人。现有的机器人护发和表面跟随方法主要依赖于轨迹规划、顺应性、力调节或触觉条件策略，但可变形的毛发可能在与末端执行器接触的同时逐渐偏移，使得局部交互难以调节。我们提出了TacHair，一种基于触觉接触分布的在线校正框架，将高分辨率的触觉观测表示为空间毛发接触分布。视觉-触觉模仿策略生成标准的抚摸动作，而单独训练的残差模块则校正局部接触偏差，将任务进展与接触恢复分离。我们在五种头部几何形状和三种毛发状态下进行了525次真实机器人实验来评估TacHair。成功的抚摸需要足够的任务进展和接触维护；我们的方法使成功率从42.9%提高到62.3%，接触维护率从59.4%提高到88.0%，相比于没有校正的相同视觉-触觉策略。这些结果表明，空间触觉接触分布作为一种有效的反馈表示，能够实现与可变形和视觉遮挡表面的接触保持交互。演示、代码和数据集可在 https://tachair.github.io 获取。

View on arXiv Download PDF AI Translation

cs.RO / 4 / 2610.10646

Masked Generative Motion Planning with Geometry-Guided Token Search

基于几何引导的掩码生成运动规划

Zhuang, Lipeng, Ru, Yingdong, Fan, Shiyu, Ho, Edmond S. L., Camarasa, Gerardo Aragon, Henderson, Paul

Abstract

Generative motion planners typically use learned trajectory priors for initial generation, while leaving test-time repair to local continuous refinement. We introduce Masked Generative Motion Planning (MGMP), which extends the learned prior from efficient parallel generation to structural repair. A masked generative transformer generates discrete trajectory candidates in parallel, and Geometry-Guided Token Search (GGTS) uses scene geometry to target where to edit and which prior-supported alternatives to evaluate. This turns refinement into an efficient search over discrete motion alternatives, enabling route-level restructuring beyond local trajectory deformation. MGMP achieves 96% success on Ring Maze and 82% repair success on Controlled Route Invalidation on Kuka, exceeding the strongest external baselines by 23 and 25 percentage points, respectively. It further generalizes to unseen layouts, additional obstacles, unseen geometries, single- and dual-arm planning, and real-world Baxter tasks.

Chinese Translation

生成运动规划器通常使用学习到的轨迹先验进行初始生成，而将测试时的修复留给局部连续优化。我们提出了掩码生成运动规划（Masked Generative Motion Planning, MGMP），它将高效并行生成的学习先验扩展到结构修复。掩码生成变换器并行生成离散轨迹候选，而几何引导的标记搜索（Geometry-Guided Token Search, GGTS）利用场景几何信息确定编辑位置以及评估哪些支持先验的替代方案。这将修复过程转变为对离散运动替代方案的高效搜索，使得在局部轨迹变形之外实现路线级别的重构。MGMP在环形迷宫（Ring Maze）上实现了96%的成功率，在Kuka的受控路线失效（Controlled Route Invalidation）上实现了82%的修复成功率，分别超越了最强外部基线23和25个百分点。它进一步推广到未见布局、额外障碍、未知几何形状、单臂和双臂规划，以及现实世界的Baxter任务。

View on arXiv Download PDF AI Translation

cs.RO / 5 / 2610.10748

TAPNAV: Humanoid Navigation through Tactile Active Perception

TAPNAV：通过触觉主动感知实现类人导航

Liu, Huaze, Wu, Zhenyu, Jang, Jaehwi, Sheng, Junjie, Collins, Andrew, Xie, Aaron, Gu, Zhaoyuan, Zhu, Kaijie, Jiang, Ding, Cai, Kevin, Zhao, Ye

Abstract

Navigation in vision-denied environments is challenging for humanoid robots because proprioceptive odometry drifts and localization uncertainty accumulates rapidly. We present TAPNAV, a tactile active-perception framework that enables humanoid navigation toward a goal by actively probing surrounding structures without relying on vision. TAPNAV maintains a pose belief from odometry, IMU, and tactile contact observations, and couples uncertainty-aware global route planning with information-gain-driven local probing. The global planner searches for routes that keep predicted localization uncertainty bounded by exploiting opportunities for tactile correction, while the local planner selects probe actions that maximize expected information gain. A whole-body controller coordinates the humanoid's locomotion and end-effector contact to execute the planned navigation and probe motions. We evaluate TAPNAV in simulation and on a Unitree G1 across different floor plans and obstacle geometries. TAPNAV achieves lower state estimation error and a higher task completion rate than baselines. These results demonstrate that actively planning physical interactions with the environment can provide localization cues for reliable humanoid navigation without vision.

Chinese Translation

在视觉受限的环境中，类人机器人面临导航挑战，因为本体感觉里程计漂移和定位不确定性迅速累积。我们提出了TAPNAV，一种触觉主动感知框架，使类人机器人能够通过主动探测周围结构而不依赖视觉，朝向目标进行导航。TAPNAV结合了来自里程计、惯性测量单元（IMU）和触觉接触观察的数据，维护姿态信念，并将不确定性感知的全局路线规划与信息增益驱动的局部探测相结合。全局规划器通过利用触觉校正的机会，搜索保持预测定位不确定性受限的路线，而局部规划器则选择最大化预期信息增益的探测动作。一个全身控制器协调类人的运动和末端效应器接触，以执行规划的导航和探测动作。我们在模拟环境和Unitree G1机器人上评估了TAPNAV，涵盖不同的平面布局和障碍几何形状。TAPNAV实现了比基线更低的状态估计误差和更高的任务完成率。这些结果表明，主动规划与环境的物理交互可以为可靠的类人导航提供定位线索，而无需视觉支持。

View on arXiv Download PDF AI Translation

cs.RO / 6 / 2610.10787

NavGPT-3: Harnessing Context in a Hierarchical Navigation Runtime

NavGPT-3：在分层导航运行时中利用上下文

Zhou, Gengze, Hong, Yicong, Zhang, Jiazhao, Zhao, Xunyi, Zhou, Jian, Lei, Zixing, Wang, Zun, Zhao, Chongyang, Chen, Xionghui, Gould, Stephen, Hengel, Anton van den, Wu, Qi

Abstract

Language models trained with long-horizon agentic reinforcement learning can generalize knowledge through reasoning, express precise actions, and pursue goals over many steps, raising the ceiling on what an embodied agent can understand and decide. Physical interaction, however, remains the domain of action policies, which provide dense, low-latency control. We present NavGPT-3, a harness that connects the two models, with an OS-like runtime built above it: reasoning, acting, and monitoring run as threads with their own context, tools, and permissions, while the runtime schedules them and decides which thread controls the robot's motion, so that the robot can react to sudden real-world events through interruption and thread switching. Beneath it, our action policy NavGPT VLA, trained on 19.28M examples, allocates visual tokens using codec allocation, in proportion to scene change; its 8B model alone reaches 74.51 SR on R2R-CE and leads RxR-CE with 78.19 SR. With the complete harness, NavGPT-3 sets the state of the art on R2R-CE (81.51 SR) and, for the first time, brings an autonomous agent to human level: on RxR-CE it matches human followers in success (90.43 vs. 90.4 SR) and path fidelity (78.47 vs. 77.7 nDTW) at 1 min 22 s per episode, versus roughly 3 min for a human. We comprehensively ablate the harness design and the interaction between the two models, showing how tools and the action policy shape the path from language-model reasoning to physical control: when NavGPT VLA executes the route, the reasoning loop shortens and the system's minimum reaction time falls from 3-19 s per language-model decision to 0.5-1 s per action-policy step (1-2 Hz). These results show that designing this embodied interface is central to connecting frontier language-model intelligence with low-level physical control. We will release all models, code, and evaluation records.

Chinese Translation

通过长时间跨度的代理强化学习训练的语言模型能够通过推理进行知识泛化，表达精确的动作，并在多个步骤中追求目标，从而提升了具身代理的理解和决策能力。然而，物理交互仍然是动作策略的领域，这些策略提供了密集的、低延迟的控制。我们提出了NavGPT-3，这是一种将这两种模型连接起来的工具，构建了一个类似操作系统的运行时：推理、行动和监控作为具有各自上下文、工具和权限的线程运行，而运行时则调度它们并决定哪个线程控制机器人的运动，以便机器人能够通过中断和线程切换对突发的现实世界事件做出反应。在其底层，我们的动作策略NavGPT VLA在1928万个示例上进行训练，使用编解码器分配按场景变化的比例分配视觉标记；其8B模型在R2R-CE上单独达到74.51 SR，并在RxR-CE上以78.19 SR领先。借助完整的工具，NavGPT-3在R2R-CE上设定了最新的技术水平（81.51 SR），并首次将自主代理提升至人类水平：在RxR-CE上，其成功率（90.43 vs. 90.4 SR）和路径保真度（78.47 vs. 77.7 nDTW）与人类跟随者相匹配，每个回合耗时1分22秒，而人类大约耗时3分钟。我们全面分析了工具设计及两种模型之间的交互，展示了工具和动作策略如何塑造从语言模型推理到物理控制的路径：当NavGPT VLA执行路线时，推理循环缩短，系统的最小反应时间从每个语言模型决策的3-19秒降至每个动作策略步骤的0.5-1秒（1-2 Hz）。这些结果表明，设计这一具身接口对于将前沿语言模型智能与低级物理控制连接起来至关重要。我们将发布所有模型、代码和评估记录。

View on arXiv Download PDF AI Translation

cs.RO / 7 / 2610.10801

Same Action, Different Outcome: Variability in Dynamic Cloth Manipulation

相同动作，不同结果：动态布料操控中的变异性

Dadgostar, Mahed, Alenyà, Guillem, Borràs, Júlia

Abstract

Although cloth is known to exhibit different outcomes under repeated fast dynamic motions, even when the same trajectory is applied, this variability has not yet been systematically characterized. Quantifying it is essential to assess the reliability of learned manipulation policies and the extent to which simulation can reproduce real-world behavior. To study this, we execute the same trajectory ten times across four dynamic tasks, two of which are novel, each tested with three cloths of very different properties and at up to three execution speeds, with a total of 269 recorded rollouts. For all of them, we record small marker positions on the cloth and synchronized stereo camera. We then formalize different metrics to quantify variability, and our results show how it is significant in every test condition, in most cases one to three orders of magnitude above the repeatability inherent to the robot and sensing noise. Our results also show variability is driven mainly by the cloth physical properties but also grows with speed. By replaying all the rollouts in four calibrated modern cloth simulators, we show that none of them can reproduce the variability magnitude we observed in real cloth, nor its ordering. Together with our analysis, we publish the dataset with synchronized OptiTrack, vision and robot logs and their corresponding simulator twins.

Chinese Translation

尽管已知布料在重复快速动态运动下会表现出不同的结果，即使施加相同的轨迹，这种变异性尚未得到系统的表征。量化这种变异性对于评估学习到的操控策略的可靠性以及模拟能在多大程度上再现现实世界行为至关重要。为此，我们在四个动态任务中执行相同的轨迹十次，其中两个任务为新任务，每个任务使用三种具有非常不同特性的布料，并在最多三种执行速度下进行测试，总共记录了269次实验结果。对于所有实验，我们记录了布料上的小标记位置以及同步的立体摄像头。然后，我们正式化了不同的指标来量化变异性，结果显示在每个测试条件下变异性都是显著的，在大多数情况下，其量级比机器人固有的重复性和传感器噪声高出一个到三个数量级。我们的结果还表明，变异性主要由布料的物理特性驱动，但随着速度的增加也会增大。通过在四个经过校准的现代布料模拟器中重放所有实验结果，我们展示了没有一个模拟器能够再现我们在真实布料中观察到的变异性幅度及其顺序。结合我们的分析，我们发布了包含同步OptiTrack、视觉和机器人日志及其对应模拟器双胞胎的数据集。

View on arXiv Download PDF AI Translation

cs.RO / 8 / 2610.10810

Diagnosing and Recovering from Observation-Space Shift at Long-Horizon Skill Seams

诊断和恢复长时间跨度技能接缝中的观察空间偏移

Wagh, Pranav, Fang, Yu, Yang, Yue, Ding, Mingyu

Abstract

Long-horizon robotic manipulation is often built by chaining independently trained skills. Although each skill can be reliable in isolation, performance degrades sharply when skills are chained: each downstream skill must start from the state its predecessor leaves behind rather than from its training distribution. We study this failure mode, Observation-Space Shift (OSS), and ask what causes these skill-seam failures. Using privileged simulator resets, we find that the dominant shift comes from displaced scene state (e.g., an open drawer or secondary objects left behind by earlier skills), not from the robot's joint configuration or the object the downstream skill manipulates. To test this diagnosis, we build a fully learned detect-restore-resume system: a task-progress monitor detects the stall, a learned policy restores the displaced scene components, and seam-robust fine-tuning lets the skill resume. It recovers the seam where every tested alternative fails, which we treat as evidence for the diagnosis rather than as a general-purpose method. On the BOSS-44 benchmark, the system improves full-chain success from 7.6% to 26.5%, a 3.5x improvement over the base policy and 51% of a privileged restoration oracle, whereas best-of-K resampling, a Diffusion Policy, and world-model baselines fail to recover from the evaluated seam states. On a real Franka arm running a fine-tuned $\pi_{0.5}$ policy, the same monitor is limited by exterior-camera observability, yet closing the loop still recovers some otherwise-terminal failures, motivating wrist and gripper sensing. These results suggest that some long-horizon composition failures are better addressed by restoring the scene before resuming the policy than by retrying from an off-support state.

Chinese Translation

长时间跨度的机器人操控通常是通过链接独立训练的技能来构建的。尽管每个技能在孤立状态下可以可靠运行，但当技能被链接时，性能会急剧下降：每个下游技能必须从其前驱留下的状态开始，而不是从其训练分布开始。我们研究这种失败模式，即观察空间偏移（Observation-Space Shift, OSS），并探讨导致这些技能接缝失败的原因。通过使用特权模拟器重置，我们发现主要的偏移来自于位移的场景状态（例如，打开的抽屉或早期技能留下的次要物体），而不是来自机器人的关节配置或下游技能操控的物体。为了验证这一诊断，我们构建了一个完全学习的检测-恢复-恢复系统：任务进度监控器检测停滞，学习的策略恢复位移的场景组件，而接缝鲁棒微调使技能得以继续。它在每个测试的替代方案失败的接缝处恢复，我们将其视为诊断的证据，而不是通用方法。在BOSS-44基准测试中，该系统将全链成功率从7.6%提高到26.5%，相较于基础策略提高了3.5倍，且达到了特权恢复预言机的51%，而最佳K重采样、扩散策略和世界模型基线在评估的接缝状态下均未能恢复。在运行微调后的$ ext{π}_{0.5}$策略的真实Franka臂上，同样的监控器受到外部摄像头可观测性的限制，但闭环仍能恢复一些本应终止的失败，激励了腕部和夹具传感。这些结果表明，一些长时间跨度的组合失败通过在恢复策略之前恢复场景来解决，可能比从不支持状态重新尝试更为有效。

View on arXiv Download PDF AI Translation

cs.RO / 9 / 2610.10812

Skill-SLM: Agent Skill-driven Small Language Models for Reliable Robot Operation

Skill-SLM：基于代理技能的小型语言模型用于可靠的机器人操作

Wang, Wenhao, Li, Yanyan, Yuan, Jiawei

Abstract

Small language models (SLMs) have been increasingly adopted for onboard robot operation because they enable intelligent decision-making. However, existing approaches are mainly distillation-oriented and rely on enumerating representative task-solution pairs. This makes dataset construction difficult and limits generalization to diverse robot tasks whose possible forms grow rapidly. This paper proposes Skill-SLM, a framework that reformulates SLM-driven robot operation as a task-decomposition and skill-composition problem. Given a natural language task instruction, Skill-SLM decomposes the task into subtasks, selects appropriate skills from the skill library, and orchestrates the selected skills into executable robot operations. First, to support the skill-driven workflow, we propose a novel robot operational skill aware context-free grammar (CFG) to extract the skills required to accomplish tasks and build the skill library accordingly. Then, we configure LLM teachers to induce and synthesize training datasets for the SLMs, enabling SLMs to decompose tasks and orchestrate skills reliably. Additionally, we employ a progressive skill orchestration strategy to improve the reliability of skill implementation and overall robot operation. Experiments on UAV operation tasks indicate that Skill-SLM substantially outperforms distillation-oriented baselines, especially on unseen tasks that require generalization of capabilities. Additional experiments on ground vehicle tasks further demonstrate that Skill-SLM can be applied to different robot platforms.

Chinese Translation

小型语言模型（SLMs）因其能够实现智能决策而越来越多地被应用于机器人操作。然而，现有的方法主要以蒸馏为导向，依赖于列举代表性的任务-解决方案对。这使得数据集构建变得困难，并限制了对多样化机器人任务的泛化能力，而这些任务的可能形式迅速增长。本文提出了Skill-SLM，一个将SLM驱动的机器人操作重新构建为任务分解和技能组合问题的框架。给定自然语言任务指令，Skill-SLM将任务分解为子任务，从技能库中选择合适的技能，并将所选技能编排为可执行的机器人操作。首先，为了支持技能驱动的工作流程，我们提出了一种新颖的机器人操作技能感知的无上下文文法（CFG），以提取完成任务所需的技能，并相应地构建技能库。然后，我们配置大型语言模型（LLM）教师，以诱导和合成SLMs的训练数据集，使SLMs能够可靠地分解任务和编排技能。此外，我们采用渐进式技能编排策略，以提高技能实施的可靠性和整体机器人操作的可靠性。在无人机操作任务上的实验表明，Skill-SLM在性能上显著优于以蒸馏为导向的基线，尤其是在需要能力泛化的未见任务上。对地面车辆任务的额外实验进一步证明了Skill-SLM可以应用于不同的机器人平台。

View on arXiv Download PDF AI Translation

cs.RO / 10 / 2610.10846

Cross-Embodiment Robot Foundation World Models with Latent Actions

跨体现机器人基础世界模型与潜在动作

Huang, Huang, Yenamandra, Sriram, Majumdar, Arjun, Aljalbout, Elie, Nagarajan, Tushar, Yang, Tsung-Yen, Rai, Akshara, Rabbat, Michael, Fei-Fei, Li, Wu, Jiajun, Wu, Tingfan, Meier, Franziska

Abstract

The diversity of robot embodiments and action spaces makes it challenging to build robot world models that generalize across different embodiments. We introduce the Latent Action-Conditioned Robot World Model (LAC-WM), which operates within a learned unified latent action space shared across diverse embodiments. This unified action space improves the world model's performance when adapted to previously unseen robot embodiments. We compare LAC-WM with an Explicit Action-Conditioned World Model (EAC-WM), which conditions on explicit motion labels. Our results show that explicit action conditioning leads to disjoint action representations across embodiments, limiting downstream performance when adapting to new robots. We evaluate both models on dexterous manipulation tasks and a modified LIBERO benchmark. LAC-WM improves downstream performance over EAC-WM by up to 46.7% on dexterous manipulation and 11.7% on LIBERO. Crucially, the unified latent action space allows LAC-WM's downstream performance to scale positively with the number of embodiments used during pretraining. In contrast, the disjoint action space in EAC-WM leads to decreased performance as the number of pretraining embodiments increases. These results highlight the importance of a unified action space for efficient cross-embodiment learning, addressing a key challenge in robotics. Project website: https://lacwm.github.io/

Chinese Translation

机器人体现和动作空间的多样性使得构建能够跨不同体现进行泛化的机器人世界模型变得具有挑战性。我们提出了潜在动作条件机器人世界模型（Latent Action-Conditioned Robot World Model, LAC-WM），该模型在一个学习到的统一潜在动作空间内运行，该空间在不同体现之间共享。这个统一的动作空间在适应以前未见的机器人体现时提高了世界模型的性能。我们将LAC-WM与显式动作条件世界模型（Explicit Action-Conditioned World Model, EAC-WM）进行了比较，后者依赖于显式运动标签进行条件设置。我们的结果表明，显式动作条件导致不同体现之间的动作表示不连通，从而限制了在适应新机器人时的下游性能。我们在灵巧操作任务和修改后的LIBERO基准上评估了这两种模型。LAC-WM在灵巧操作上相较于EAC-WM提高了最多46.7%的下游性能，在LIBERO上提高了11.7%。关键是，统一的潜在动作空间使得LAC-WM的下游性能随着预训练期间使用的体现数量的增加而正向扩展。相比之下，EAC-WM中的不连通动作空间导致随着预训练体现数量的增加性能下降。这些结果突显了统一动作空间在高效跨体现学习中的重要性，解决了机器人技术中的一个关键挑战。项目网站：https://lacwm.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 11 / 2610.10855

OmniHOI: Dexterous Hand-Object Interaction from Monocular Human Video

OmniHOI：基于单目人类视频的灵巧手-物体交互

Mao, Ting, Shao, Yanming, Wang, Ziheng, Liu, Haoyu, Wang, Yiqun, Wu, Xuanye, Mu, Yao

Abstract

Monocular videos of human manipulation provide abundant dexterous demonstrations, yet reconstructing hand-object interaction from a single view and transferring it to robot hands remain difficult, limiting their direct use for robot execution. Prior methods either require task-specific RL training, limiting scalability, or assume clean motion-capture trajectories and thus cannot operate directly on video. We present OmniHOI, a pipeline that turns an RGB video of hand-object interaction into an interaction-faithful trajectory on dexterous hands. The key idea is to enforce physical consistency using the evidence available at each stage: image evidence during reconstruction, contact geometry during retargeting, and dynamics during physics-in-the-loop refinement. Each stage optimizes the corresponding representation directly, correcting errors before they propagate downstream or must be absorbed by a learned policy. Across 150 motion-capture trajectories transferred to each of five dexterous hands with 6 to 22 DoF, we achieve 39-89% success, compared with at most 31% for prior transfer methods. On 60 monocular video clips, we achieve 53% success, compared with 28% for the best prior video-to-robot pipeline. Its trajectories also execute on a real bimanual robot across diverse tasks.

Chinese Translation

单目视频中的人类操作提供了丰富的灵巧演示，但从单一视角重建手-物体交互并将其转移到机器人手上仍然困难，这限制了其在机器人执行中的直接应用。之前的方法要么需要特定任务的强化学习（RL）训练，限制了可扩展性，要么假设干净的运动捕捉轨迹，因此无法直接在视频上操作。我们提出了OmniHOI，一个将手-物体交互的RGB视频转换为灵巧手上忠实交互轨迹的流程。其关键思想是利用每个阶段可用的证据来强制物理一致性：在重建过程中使用图像证据，在重新定向过程中使用接触几何，在物理环路优化过程中使用动态信息。每个阶段直接优化相应的表示，纠正错误，以防止错误向下传播或必须被学习的策略吸收。在将150个运动捕捉轨迹转移到五个具有6到22个自由度的灵巧手上时，我们实现了39-89%的成功率，而之前的转移方法最多为31%。在60个单目视频片段上，我们实现了53%的成功率，而最佳的先前视频到机器人流程的成功率为28%。其轨迹还能够在多样化任务中在真实的双手机器人上执行。

View on arXiv Download PDF AI Translation

cs.RO / 12 / 2610.10905

Informationally Decoupled Trajectory Design for Sim-to-Real System Identification

信息解耦轨迹设计用于仿真到现实系统识别

Shin, Sangwoo, Anilkumar, Ashvin, Gao, Ryan, Hanna, Josiah P.

Abstract

Sampling-based system identification estimates physically meaningful parameters by tuning a simulator to reproduce the target system dynamics, providing an interpretable approach to improving sim-to-real transfer. Yet when the collected trajectories do not distinguish the effects of different parameters, multiple parameter combinations can reproduce those trajectories, leading to unreliable parameter estimates. To address this challenge, we introduce an Informationally Decoupled Trajectory Design framework (IDTD), which formulates the objective for the exploration policy built on the Schur complement score derived from the Fisher information matrix. To faithfully reflect parameter separability in the exploration objective, IDTD normalizes the score against the per-parameter information, selects the most favorable trajectory segment for each parameter, and aggregates the resulting scores logarithmically. The optimized trajectory is therefore composed of complementary intervals, each exposing a distinct subset of parameters whose contribution to the motion over that interval can be attributed unambiguously. Across diverse simulation environments, ranging from a linear-dynamics system to the Go2 quadruped, G1 humanoid, and Crazyflie quadrotor, IDTD reduces the parameter identification error by 39.6% on average relative to the strongest prior active-exploration baseline and attains improved downstream policy transfer. Furthermore, we validate the proposed trajectory design on a real K1 humanoid, demonstrating that the resulting identified parameters accurately capture the real-system dynamics.

Chinese Translation

基于采样的系统识别通过调整模拟器以重现目标系统动态来估计物理上有意义的参数，提供了一种可解释的方法来改善仿真到现实的转移。然而，当收集的轨迹无法区分不同参数的影响时，多种参数组合可以重现这些轨迹，从而导致不可靠的参数估计。为了解决这一挑战，我们提出了一种信息解耦轨迹设计框架（IDTD），该框架基于从费舍尔信息矩阵导出的舒尔补数分数来制定探索策略的目标。为了忠实反映探索目标中的参数可分离性，IDTD 根据每个参数的信息对分数进行归一化，为每个参数选择最有利的轨迹段，并以对数方式聚合结果分数。因此，优化后的轨迹由互补区间组成，每个区间揭示了一组独特的参数，其对该区间内运动的贡献可以明确归因。在多种仿真环境中，从线性动态系统到 Go2 四足机器人、G1 人形机器人和 Crazyflie 四旋翼机，IDTD 相较于最强的先前主动探索基线平均降低了 39.6% 的参数识别误差，并实现了改进的下游策略转移。此外，我们在真实的 K1 人形机器人上验证了所提的轨迹设计，证明所识别的参数准确捕捉了真实系统的动态。

View on arXiv Download PDF AI Translation

cs.RO / 13 / 2610.10912

When Listening Becomes Easier: Scrubbing Visual Cues for Shortcut-Free VLAs

当倾听变得更容易：去除视觉线索以实现无快捷方式的视觉-语言-动作（VLA）

Gerigk, Jasper, Aspuru-Takata, Kenzo, Wu, Chin-Hsuan, Mohammadi, Mohammad, Zheng, Shuhong, Gilitschenski, Igor

Abstract

Shortcut learning is a prevalent issue in robot learning. The limited diversity of robot demonstration datasets can mislead policies into exploiting spurious correlations between tasks and irrelevant features, such as viewpoint or background. Collecting sufficiently diverse robot demonstrations is costly and inefficient, motivating algorithmic alternatives. We focus on vision-language-action (VLA) models and discover that different vision-language model backbones exhibit substantially different levels of susceptibility to visual shortcut learning. We find that model behavior correlates with our proposed representation-level metric, action margin, which requires no policy rollouts. Visual shortcuts consistently enter action representations in early layers, with models differing in the extent to which later layers correct them by incorporating language information. To boost models' attention to language, we introduce task scrubbing, a new domain-adversarial training method that decreases models' likelihood of using visual shortcuts and improves VLAs' generalization. Experiments in both simulation and the real world across multiple VLAs and visual cues show that task scrubbing improves out-of-distribution robustness and often eliminates visual shortcut learning.

Chinese Translation

快捷学习是机器人学习中的一个普遍问题。机器人示范数据集的多样性有限可能导致策略利用任务与无关特征（如视角或背景）之间的虚假相关性。收集足够多样化的机器人示范既昂贵又低效，这促使了算法替代方案的出现。我们关注视觉-语言-动作（VLA）模型，发现不同的视觉-语言模型骨干在视觉快捷学习的易感性上表现出显著不同的水平。我们发现模型行为与我们提出的表示层级指标——动作边际（action margin）相关，该指标不需要策略回放。视觉快捷方式在早期层中持续进入动作表示，不同模型在后续层通过整合语言信息来纠正这些快捷方式的程度上存在差异。为了增强模型对语言的关注，我们引入了任务去除（task scrubbing），这是一种新的领域对抗训练方法，旨在降低模型使用视觉快捷方式的可能性并改善VLA的泛化能力。在多个VLA和视觉线索的仿真和现实世界实验中，结果表明任务去除提高了分布外的鲁棒性，并且通常消除了视觉快捷学习。

View on arXiv Download PDF AI Translation

cs.RO / 14 / 2610.10934

Higher-Order Morphology Priors for Quadruped Reinforcement Learning Under Actuator Degradation

针对执行器退化的四足强化学习的高阶形态先验

You, Derek, Shamsi, Zafir, Wang, Keqin, Allen-Blanchette, Christine

Abstract

Actuator degradation turns quadruped locomotion into a coordination problem requiring joints to compensate for lost actuation. Prior work suggests that morphology-aware graph policies improve learning and generalization under body perturbations. We ask whether these benefits can be strengthened by explicitly modeling higher-order mechanical structure. We represent the Unitree Go1 as a cell complex with limb- and body-level rank-2 cells and apply Hodge-based message passing. Under degradation training, the node-edge-face Hodge actor achieves the highest return on unseen actuator degradations, with higher survival and lower velocity-tracking error. These results support higher-order morphology as a useful inductive bias for whole-body compensation under actuator degradation.

Chinese Translation

执行器退化使四足运动变成一个协调问题，需要关节补偿失去的驱动。先前的研究表明，考虑形态的图策略在身体扰动下能够改善学习和泛化能力。我们探讨是否通过显式建模高阶机械结构可以增强这些益处。我们将Unitree Go1表示为一个细胞复合体，具有肢体和身体层级的二阶细胞，并应用基于Hodge的消息传递。在退化训练下，节点-边-面Hodge演员在未见的执行器退化上实现了最高的回报，具有更高的生存率和更低的速度跟踪误差。这些结果支持高阶形态作为在执行器退化下进行全身补偿的有用归纳偏置。

View on arXiv Download PDF AI Translation

cs.RO / 15 / 2610.10962

iAm.md: Robot Skill Self-Assessment through Agentic Introspection for Unknown Open-Vocabulary Domains

iAm.md：通过代理内省进行未知开放词汇领域的机器人技能自我评估

Guarino, Vincenzo, Musumeci, Emanuele, Suriani, Vincenzo, Nardi, Daniele

Abstract

Agentic AI based on Large Language Model generalization capabilities offers a wide range of potential applications, including planning for embodied tasks. For example, embodied agents based on Foundation models can generate plausible plans in autonomous robotics scenarios. Due to limited context windows or hallucinatory phenomena in the next-token prediction formulation, behaviors may be generated without establishing whether the deployed robot and the observed environment actually support the requested operation, in what we call a "grounding failure". Thanks to the recent improvements in reasoning capabilities of foundation models, autonomous robot behavior generation problem can be formulated as a code generation problem. We present iAm.md, a Markdown standard and generation framework, that allows anchoring this process in complementary forms of deployment evidence. Through open-vocabulary semantic mapping, we combine local vision-language detections and object segmentation and refer them to persistent object records in this intermediate standardized representation, allowing agentic introspection. We then study this new technique on a simulated TIAGo, on navigation-and-manipulation tasks, showing how this standardized representation jointly supports skill self-assessment and executable task generalization.

Chinese Translation

基于大型语言模型泛化能力的代理人工智能提供了广泛的潜在应用，包括对具身任务的规划。例如，基于基础模型的具身代理可以在自主机器人场景中生成合理的计划。然而，由于有限的上下文窗口或在下一个标记预测公式中的幻觉现象，生成的行为可能未能确认所部署的机器人和观察到的环境是否实际支持所请求的操作，这被称为“基础失败”。得益于基础模型推理能力的近期提升，自主机器人行为生成问题可以被表述为代码生成问题。我们提出了iAm.md，这是一种Markdown标准和生成框架，允许将这一过程锚定在互补的部署证据形式上。通过开放词汇的语义映射，我们结合了局部视觉-语言检测和物体分割，并将其引用到这一中间标准化表示中的持久物体记录，从而实现代理内省。随后，我们在模拟的TIAGo上研究了这一新技术，针对导航和操作任务，展示了这一标准化表示如何共同支持技能自我评估和可执行任务的泛化。

View on arXiv Download PDF AI Translation

cs.RO / 16 / 2610.11003

ActiveReg: Information-Driven Active Regional Probing for Partial-to-Full Bone Registration

ActiveReg：基于信息驱动的主动区域探测用于部分到全骨注册

Li, Tiancheng, Wang, Yingyu, Walker, Peter, Zhao, Liang, Huang, Shoudong

Abstract

Accurate bone registration is essential for orthopedic surgical navigation and robotic assistance. In a common workflow, the surgeon needs to identify and probe a number of prescribed locations on the exposed bone surface to obtain data points for registration, which can be difficult and time-consuming under limited surgical exposure. Reducing the number of required points while providing clear probing guidance would ease the surgeon's acquisition task intra-operatively. We present ActiveReg, a closed-loop framework that recommends probing regions rather than individual points, allowing flexibility in the exact contact location. A D-optimal planner that takes into account the already acquired points and the uncertainty of current registration information helps achieve accurate registration with fewer points. An online assessment combines pre-update innovation consistency with uncertainty at predefined surgical task locations improves the reliability of completion decision, without requiring ground truth. Simulations across four anatomical scenarios demonstrated that ActiveReg can achieve competitive accuracy compared with baseline registration methods while using substantially fewer acquired points. Real phantom experiments with optical and electromagnetic tracking demonstrated comparable mean target registration error to the Gradient-SDF framework, using only 24-28 acquired points instead of 771-1263.

Chinese Translation

准确的骨注册对于骨科手术导航和机器人辅助至关重要。在常见的工作流程中，外科医生需要识别并探测暴露骨表面上的多个预定位置，以获取用于注册的数据点，这在有限的手术暴露下可能既困难又耗时。减少所需点的数量，同时提供清晰的探测指导，将有助于外科医生在手术过程中更轻松地获取数据。我们提出了ActiveReg，一个闭环框架，推荐探测区域而非单个点，从而允许在确切接触位置上的灵活性。一个D-最优规划器考虑了已经获取的点和当前注册信息的不确定性，帮助以更少的点实现准确的注册。在线评估结合了预更新创新一致性与预定义手术任务位置的不确定性，提高了完成决策的可靠性，而无需真实值。针对四种解剖场景的模拟表明，ActiveReg在使用显著更少的获取点的情况下，能够实现与基线注册方法相竞争的准确性。使用光学和电磁追踪的真实幻影实验显示，ActiveReg的平均目标注册误差与Gradient-SDF框架相当，仅使用24-28个获取点，而不是771-1263个。

View on arXiv Download PDF AI Translation

cs.RO / 17 / 2610.11054

A Reconfigurable Fabric Based Pneumatic Actuator with Button Fastened Constraint Modules for Multi Mode Actuation

一种基于可重构织物的气动执行器，具有按钮固定约束模块以实现多模式驱动

Kimura, Kentaro, Ishizuka, Hiroki, Sakaue, Yusuke, Ikeda, Sei, Oshiro, Osamu

Abstract

Soft pneumatic actuators are widely used in soft robotics. However, their actuation modes are typically fixed by internal chamber designs or reinforcements, making post fabrication reconfiguration difficult. This paper presents a reconfigurable fabric-based pneumatic actuator that enables rapid, tool free, and reusable switching among multiple actuation modes by attaching constraint fabric modules to a buttoned textile sleeve. By rearranging low-stretch constraint parts on a single actuator body, the proposed system realizes four actuation modes isotropic expansion, contraction, bending, and twisting without redesigning or disassembling the actuator.We further introduce an attachable capacitive fabric-based tactile sensor module that can be mounted at arbitrary locations on the actuator surface for task dependent contact sensing. Experiments demonstrated an elongation ratio of 1.2, a maximum bending angle of approximately 60 degrees, and a twisting angle of approximately 90 degrees at 24 kPa, while achieving blocking forces exceeding 10 N across all modes. The tactile sensor exhibited a monotonic, force-dependent increase in capacitance with low hysteresis (approximately 12%) and stable cyclic responses under preload. Finally, we constructed soft robotic arms composed of multiple actuators and demonstrated in situ motion reconfiguration for grasping by changing the attached constraint patterns. These results indicate that the proposed modular constraint and sensing framework improves the functional adaptability of fabric-based soft robots while preserving a simple, robust actuator structure.

Chinese Translation

软气动执行器在软机器人中得到广泛应用。然而，它们的驱动模式通常由内部腔体设计或加固固定，使得后期的重构变得困难。本文提出了一种基于可重构织物的气动执行器，通过将约束织物模块附加到按钮式纺织套筒上，实现了快速、无工具和可重复使用的多种驱动模式切换。通过重新排列单一执行器主体上的低拉伸约束部件，所提出的系统实现了四种驱动模式：各向同性扩展、收缩、弯曲和扭转，而无需重新设计或拆卸执行器。我们进一步介绍了一种可附加的电容式织物触觉传感器模块，可以安装在执行器表面的任意位置，以实现任务依赖的接触感知。实验表明，在24 kPa下，该执行器的伸长比为1.2，最大弯曲角度约为60度，扭转角度约为90度，并且在所有模式下阻挡力超过10 N。触觉传感器的电容随施加力单调增加，滞后低（约12%），并且在预载下表现出稳定的循环响应。最后，我们构建了由多个执行器组成的软机器人手臂，并通过改变附加的约束模式展示了抓取时的原位运动重构。这些结果表明，所提出的模块化约束和传感框架在保持简单、稳健的执行器结构的同时，提高了基于织物的软机器人的功能适应性。

View on arXiv Download PDF AI Translation

cs.RO / 18 / 2610.11072

Towards Path-Creative Navigation: Robot Navigation through Embodied Interaction

朝向路径创造性导航：通过具身互动实现机器人导航

Xi, Haoyu, Cheng, Siwei, Liu, Xiangyuan, Li, Brenda, Zhang, Wei

Abstract

Autonomous navigation in cluttered and constrained environments typically assumes a fixed environment and searches only for paths within existing free space. However, a route can be blocked by an articulated structure, a movable object, or a pedestrian, and reaching the goal may therefore require appropriate embodied interaction with the environment. This paper formulates Path-Creative Navigation (PCN), a navigation paradigm in which the robot recovers free space by coordinating locomotion and embodied interaction, and addresses one class of PCN tasks with a mapless framework. The vision-LiDAR-odometry framework integrates local observations, obstacle geometry, and robot-state estimates through a unified navigation system, providing stable semantic and geometric information for navigation and interaction. The traversability-aware decision method utilizes visual and LiDAR distance information to determine whether a blockage is actionable and whether it can be safely bypassed, enabling the robot to interact only when necessary while supporting articulated structure pushing, movable object pushing, obstacle avoidance, and pedestrian requests. We design simulation scenarios and evaluate different methods in both simulation and two real-world tasks that cannot be completed by conventional navigation alone without embodied interaction. These results demonstrate that our framework achieves higher task completion than the baseline methods by coordinating navigation and interaction in cluttered and constrained environments. Code is available at:{https://anonymous.4open.science/r/path-creative-navigation/}.

Chinese Translation

自主导航在拥挤和受限环境中通常假设环境是固定的，仅在现有的自由空间内搜索路径。然而，路线可能会被关节结构、可移动物体或行人阻挡，因此到达目标可能需要与环境进行适当的具身互动。本文提出了路径创造性导航（Path-Creative Navigation, PCN），这一导航范式使机器人通过协调运动和具身互动来恢复自由空间，并针对无地图框架下的一类PCN任务进行了探讨。视觉-激光雷达-里程计框架通过统一的导航系统整合了局部观测、障碍几何形状和机器人状态估计，为导航和互动提供稳定的语义和几何信息。可通行性感知决策方法利用视觉和激光雷达距离信息来判断阻塞是否可操作以及是否可以安全绕过，使机器人仅在必要时进行互动，同时支持关节结构推送、可移动物体推送、障碍物规避和行人请求。我们设计了仿真场景，并在仿真和两个无法仅通过传统导航完成的真实任务中评估不同方法。这些结果表明，我们的框架通过在拥挤和受限环境中协调导航和互动，实现了比基线方法更高的任务完成率。代码可在以下链接获取：{https://anonymous.4open.science/r/path-creative-navigation/}。

View on arXiv Download PDF AI Translation

cs.RO / 19 / 2610.11119

FOCUS: From Privileged States to RGB-D with Controlled Modality Switching and Representation Alignment

FOCUS：从特权状态到RGB-D的受控模态切换与表示对齐

Grigorov, Filip, Darvish, Kourosh, Vijaykumar, Nandita

Abstract

Vision-based reinforcement learning for robotic manipulation is sample-inefficient because RGB-D observations are high-dimensional and noisy. Privileged state information available in simulation can accelerate training, but its absence at test time creates a train-test modality gap. We propose FOCUS, a single-stage PPO framework that trains the critic on privileged state while automatically regulating whether the actor collects rollouts from RGB-D or privileged-state latents. Regulation is driven by the KL divergence between the action distributions induced by the two modalities, while representation alignment encourages consistent action selection across them. Together, these mechanisms limit RGB-D rollouts when the actor's action distributions from RGB-D and privileged-state latents disagree. As they align, RGB-D exposure increases, shifting on-policy training toward the RGB-D inputs used at test time. Across five manipulation tasks, FOCUS raises average test success from 0.71 to 0.93 relative to the strongest RGB-D-at-test baseline on each task. When accounting for each method's complete training pipeline, budget-normalized training-success AUC increases from 0.47 to 0.65. On Pick-and-Place, test success rises from 0.47 to 0.86, while AUC increases from 0.12 to 0.61, a 5.0x improvement in learning efficiency over the fixed interaction budget.

Chinese Translation

基于视觉的强化学习在机器人操控中样本效率低下，因为RGB-D观测具有高维度和噪声。在仿真中可获得的特权状态信息可以加速训练，但在测试时缺乏这种信息会导致训练与测试之间的模态差距。我们提出了FOCUS，这是一个单阶段的PPO框架，它在特权状态上训练评论员，同时自动调节演员是从RGB-D还是特权状态潜变量中收集回放。调节是通过两种模态引起的动作分布之间的KL散度驱动的，而表示对齐则鼓励在它们之间保持一致的动作选择。这些机制共同限制了当演员从RGB-D和特权状态潜变量的动作分布不一致时的RGB-D回放。随着它们的对齐，RGB-D的曝光增加，策略训练向测试时使用的RGB-D输入转移。在五个操控任务中，FOCUS将每个任务的平均测试成功率从0.71提高到0.93，相较于最强的RGB-D测试基线。当考虑每种方法的完整训练流程时，预算归一化的训练成功AUC从0.47增加到0.65。在Pick-and-Place任务中，测试成功率从0.47上升到0.86，而AUC从0.12增加到0.61，学习效率提高了5.0倍，超越了固定交互预算的表现。

View on arXiv Download PDF AI Translation

cs.RO / 20 / 2610.11141

Distributed Relative Localization Based on Ultra-WideBand and LiDAR for Multi-robot with Limited Communication

基于超宽带和激光雷达的多机器人有限通信分布式相对定位

Cao, Zhiqiang, Liu, Ran, Lau, Billy Pik Lik, Yuen, Chau, Tan, U-Xuan

Abstract

Relative localization is crucial for a multi-robot system to collaboratively perform tasks, such as exploration and formation. However, this is highly challenging for homogeneous robots with similar appearance in GPS-denied and communication-limited environments. In this paper, we propose a fully distributed relative position estimation approach for a team of robots based on onboard UWB and LiDAR sensors, in which LiDAR is utilized to obtain the position of anonymous objects in Line-of-Sight (LOS), and UWB is used for ranging between robots. We construct two graphs, namely UWB connection graph and LiDAR connection graph, to represent the spatial relationship among objects (robots and obstacles) based on UWB and LiDAR measurements. Identification and relative position estimation are formulated as a common subgraph matching problem. A falsely-matched robot identification approach is designed to recognize the falsely-matched results caused by obstacle blockage in LiDAR field of view. These robots are then localized by leveraging the well-matched robots and the UWB ranging measurements in the UWB connection graph. We conducted experiments to evaluate the performance of our approach. The results show that the proposed approach is capable of achieving satisfactory positioning accuracy for a team of robots in a distributed manner with only exchanging limited information.

Chinese Translation

相对定位对于多机器人系统协同执行任务（如探索和编队）至关重要。然而，在GPS不可用和通信受限的环境中，对于外观相似的同质机器人来说，这是一项极具挑战性的任务。本文提出了一种基于车载超宽带（UWB）和激光雷达（LiDAR）传感器的完全分布式相对位置估计方法，其中激光雷达用于获取视距（Line-of-Sight, LOS）中匿名物体的位置，而超宽带用于机器人之间的测距。我们构建了两个图，即UWB连接图和LiDAR连接图，以表示基于UWB和LiDAR测量的物体（机器人和障碍物）之间的空间关系。识别和相对位置估计被形式化为一个公共子图匹配问题。设计了一种错误匹配机器人识别方法，以识别因障碍物阻挡激光雷达视野而导致的错误匹配结果。然后，通过利用匹配良好的机器人和UWB连接图中的UWB测距数据对这些机器人进行定位。我们进行了实验以评估我们方法的性能。结果表明，所提出的方法能够在仅交换有限信息的情况下，以分布式方式为一组机器人实现令人满意的定位精度。

View on arXiv Download PDF AI Translation

cs.RO / 21 / 2610.11168

PMTRM: Pseudo-Memory Temporal Re-encoding Module for Embodied Policy Learning

PMTRM：用于具身策略学习的伪记忆时间重编码模块

Yang, Changchuan, Xu, Haoxuan, Chen, Wenbo, Ren, Shuai, Zheng, Jianlong, Zhang, Huarui, Li, Tianfu, Tian, Guanzhong

Abstract

Robotic manipulation often contains repeated motions whose local observations look similar at different phases. When these phases require different actions, a policy that relies mainly on the current observation may repeat completed motions or switch phases at the wrong time. To address this phase ambiguity, we present the Pseudo-Memory Temporal Re-encoding Module (PMTRM), a lightweight plug-in module with only 7.61M parameters that encodes a bounded history of executed states and actions into a latent sequence for existing policies. To help distinguish phases, a temporal heterogeneity objective penalizes positive similarity between distant positions in this sequence, while anchor and reconstruction losses preserve information needed for action prediction. The reconstruction decoder is used only during training, leaving the temporal re-encoder to supply history to the policy at inference. We train the module progressively on synthetic sequences and robot data, then jointly with the policy, using temporal masking to accommodate partial histories. This integration retains the original action head and action space and adds auxiliary losses to the original policy loss. Experiments with multiple policy backbones in simulation and on a real robot show improved task success on tasks with phase ambiguity, with little additional computation.

Chinese Translation

机器人操作通常包含重复的动作，其局部观察在不同阶段看起来相似。当这些阶段需要不同的动作时，主要依赖当前观察的策略可能会重复已完成的动作或在错误的时间切换阶段。为了解决这一阶段模糊性，我们提出了伪记忆时间重编码模块（PMTRM），这是一个轻量级的插件模块，仅包含7.61M参数，能够将执行状态和动作的有限历史编码为现有策略的潜在序列。为了帮助区分阶段，一个时间异质性目标对该序列中远距离位置之间的正相似性进行惩罚，同时锚点和重构损失保留了动作预测所需的信息。重构解码器仅在训练期间使用，而时间重编码器则在推理时为策略提供历史信息。我们在合成序列和机器人数据上逐步训练该模块，然后与策略联合训练，使用时间掩蔽来适应部分历史。这种集成保留了原始动作头和动作空间，并为原始策略损失添加了辅助损失。在仿真和真实机器人上的多种策略骨干的实验表明，在具有阶段模糊性的任务上，任务成功率得到了改善，且额外计算量很小。

View on arXiv Download PDF AI Translation

cs.RO / 22 / 2610.11175

Higher-Order Action Supervision Makes A Strong Policy Class

高阶动作监督构建强大的策略类

Cheng, Peng, Hou, Yunxian, Zhou, Zhi, Zhang, Qian, Huang, Chang, Zhan, Xianyuan

Abstract

Modern data-driven decision-making methods, such as imitation learning (IL) and reinforcement learning (RL), have achieved great success in solving many complex tasks. However, these methods often suffer from serious control instability and robustness issues when applied in real-world applications such as robotics and autonomous driving, posing notable challenges for their practical deployment. We argue that this instability issue stems largely from their limitations in solely supervising and optimizing zeroth-order actions (i.e., the action labels), failing to account for higher-order action dynamics and temporal consistency. In this paper, we show that simultaneously supervising both zeroth- and first-order actions can dramatically enhance policies' performance and control robustness. To achieve this, we introduce a novel and elegant loss scheme supported by formal theoretical guarantees that can equip any off-the-shelf policy model (e.g., deterministic, stochastic, or flow policies) with the capability for higher-order action supervision, without requiring any structural modifications. Moreover, our proposed method can serve as a lightweight plug-and-play module that seamlessly integrates with a broad spectrum of existing offline RL frameworks. Extensive evaluations on OGBench and D4RL demonstrate that our approach yields substantial performance and robustness improvements across a wide range of continuous control environments. Notably, our method can also enhance policies' out-of-distribution (OOD) generalization capability in the challenging low-data regime, making it an ideal tool in tackling many real-world control problems.

Chinese Translation

现代数据驱动的决策方法，如模仿学习（Imitation Learning, IL）和强化学习（Reinforcement Learning, RL），在解决许多复杂任务方面取得了巨大成功。然而，这些方法在实际应用中（如机器人技术和自动驾驶）往往面临严重的控制不稳定性和鲁棒性问题，这对其实际部署构成了显著挑战。我们认为，这种不稳定性问题主要源于它们仅对零阶动作（即动作标签）进行监督和优化，未能考虑高阶动作动态和时间一致性。在本文中，我们展示了同时对零阶和一阶动作进行监督可以显著提升策略的性能和控制鲁棒性。为此，我们提出了一种新颖而优雅的损失方案，并提供了形式化的理论保证，使任何现成的策略模型（例如，确定性策略、随机策略或流动策略）都能具备高阶动作监督的能力，而无需进行任何结构性修改。此外，我们提出的方法可以作为一个轻量级的即插即用模块，与广泛的现有离线强化学习框架无缝集成。在OGBench和D4RL上的广泛评估表明，我们的方法在各种连续控制环境中实现了显著的性能和鲁棒性提升。值得注意的是，我们的方法还可以增强策略在低数据环境下的分布外（Out-of-Distribution, OOD）泛化能力，使其成为解决许多现实世界控制问题的理想工具。

View on arXiv Download PDF AI Translation

cs.RO / 23 / 2610.11194

OmniDex: Scaling Dexterous Hand Grasping to Diverse Cluttered Scenes

OmniDex：将灵巧手抓取扩展到多样化的杂乱场景

Fang, Naiyu, Luo, Zhongjin, Mo, Yuxin, Huang, Siyuan, Liu, Jianbo, Liu, Yufei, Zhou, Zheyuan, Jin, Chenkai, Wang, Xiaogang, Li, Hongsheng

Abstract

Dexterous grasping is the foundational primitive in embodied AI, demanding massive data to train robust models. As real-world data collection is expensive, simulation has become the mainstream paradigm. Yet, while cluttered scenes best reflect real-world applications, learning to grasp within them is bottlenecked by a critical scarcity of large-scale data. To resolve this, we curate high-quality 3D objects and supporting bases, proposing a scalable seed-and-filter strategy that bypasses sluggish scene-level optimization. This yields an unprecedented benchmark comprising over 2.6 million scenes and 0.4B scene-specific grasp ground truths, featuring diverse realistic layouts paired with rich semantic and geometric observations. Furthermore, we introduce the OmniDex model to overcome the grasp multimodality and last-millimeter precision errors plaguing current generative models. By coupling Soft Winner-Takes-All learning with human-inspired physical constraints during training, and utilizing physics-driven ranking, our approach achieves robust dexterous grasping without the latency of post-optimization. Experimental results show that OmniDex model achieves state-of-the-art performance and strong generalization across diverse scenes, views, and unseen objects.

Chinese Translation

灵巧抓取是具身人工智能的基础原语，要求大量数据以训练稳健的模型。由于现实世界数据收集成本高昂，模拟已成为主流范式。然而，尽管杂乱场景最能反映现实应用，但在这些场景中学习抓取受到大规模数据严重匮乏的瓶颈限制。为了解决这一问题，我们策划了高质量的3D物体和支撑基础，提出了一种可扩展的种子-过滤策略，绕过了缓慢的场景级优化。这产生了一个前所未有的基准，包含超过260万个场景和4亿个场景特定的抓取真实值，特征是多样化的现实布局以及丰富的语义和几何观测。此外，我们引入了OmniDex模型，以克服当前生成模型中存在的抓取多模态性和最后一毫米精度误差。通过在训练过程中结合软赢家通吃学习与人类启发的物理约束，并利用基于物理的排名，我们的方法实现了稳健的灵巧抓取，而无需后优化的延迟。实验结果表明，OmniDex模型在多样化场景、视角和未见物体中实现了最先进的性能和强大的泛化能力。

View on arXiv Download PDF AI Translation

cs.RO / 24 / 2610.11220

Demonstrating Arena 5.0: A Photorealistic ROS2 Simulation Framework for Developing and Benchmarking Social Navigation

展示 Arena 5.0：一个用于开发和基准测试社会导航的照片真实感 ROS2 仿真框架

Shcherbyna, Volodymyr, Kästner, Linh, Do, Duc Anh, Tung, Hoang, Nguyen, Huu Giang, Schreff, Maximilian Ho-Kyoung, Seeger, Tim, Wiese, Eva, Martban, Ahmed, Zeng, Huajian, Tran, An, Hung, Nguyen Quoc, Kreutz, Jonas, Lam, Vu Thanh, Kien, Ton Manh, Soh, Harold

Abstract

Building upon the foundations laid by our previous work, this paper introduces Arena 5.0, the fifth iteration of our framework for robotics social navigation development and benchmarking. Arena 5.0 provides three main contributions: 1) The complete integration of NVIDIA Isaac Gym, enabling photorealistic simulations and more efficient training. It seamlessly incorporates Isaac Gym into the Arena platform, allowing the use of existing modules such as randomized environment generation, evaluation tools, ROS2 support, and the integration of planners, robot models, and APIs within Isaac Gym. 2) A comprehensive benchmark of state-of-the-art social navigation strategies, evaluated on a diverse set of generated and customized worlds and scenarios of varying difficulty levels. These benchmarks provide a detailed assessment of navigation planners using a wide range of social navigation metrics. 3) Extensive scenario generation and task planning modules for improved and customizable generation of social navigation scenarios, such as emergency and rescue situations. The platform's performance was evaluated by generating the aforementioned benchmark and through a comprehensive user study, demonstrating significant improvements in usability and efficiency compared to previous versions. Arena 5.0 is open source and available at https://github.com/Arena-Rosnav.

Chinese Translation

在我们之前工作的基础上，本文介绍了 Arena 5.0，这是我们用于机器人社会导航开发和基准测试的框架的第五个版本。Arena 5.0 提供了三个主要贡献：1）完全整合 NVIDIA Isaac Gym，实现照片真实感的仿真和更高效的训练。它将 Isaac Gym 无缝集成到 Arena 平台中，允许使用现有模块，如随机环境生成、评估工具、ROS2 支持，以及在 Isaac Gym 中集成规划器、机器人模型和 API。2）对最先进的社会导航策略进行全面基准测试，评估在多样化的生成和定制世界及不同难度水平的场景中。这些基准测试提供了对导航规划器的详细评估，使用广泛的社会导航指标。3）广泛的场景生成和任务规划模块，用于改进和可定制的社会导航场景生成，如紧急和救援情况。通过生成上述基准和进行全面的用户研究，评估了该平台的性能，显示出与之前版本相比，在可用性和效率方面有显著改善。Arena 5.0 是开源的，可在 https://github.com/Arena-Rosnav 获取。

View on arXiv Download PDF AI Translation

cs.RO / 25 / 2610.11221

How Firm Should a Grasp Be?

抓握应该有多牢固？

Beveridge, Matthew, Nayar, Shree K.

Abstract

An ideal robot grasp is firm enough to securely handle an object, yet gentle enough to avoid damaging it. Achieving this balance requires knowledge of the object's material properties, such as its mass, elasticity, and surface friction. These properties, however, are seldom precisely known a priori. In this work, we propose a visuotactile approach to estimating material properties in real time, during the process of grasping. Our method uses these estimated properties to determine the minimum grasp force required to handle the object. We contribute a new dataset of real-world objects (fruits and vegetables) with measured physical properties (shape, mass, elasticity, and friction), which we use to construct our force estimation model via simulations. We experimentally validate our approach to grasp force control using a robot with a parallel-jaw gripper. We demonstrate our system's ability to gently grasp a wide variety of objects, in each case adapting to their unique physical properties.

Chinese Translation

理想的机器人抓握既要足够牢固，以安全地处理物体，又要足够温和，以避免对其造成损害。实现这一平衡需要对物体的材料特性有一定了解，例如其质量、弹性和表面摩擦力。然而，这些特性通常在事先并不精确已知。在本研究中，我们提出了一种视觉触觉方法，用于在抓握过程中实时估计材料特性。我们的方法利用这些估计的特性来确定处理物体所需的最小抓握力。我们贡献了一个包含实际物体（水果和蔬菜）的新数据集，记录了其物理特性（形状、质量、弹性和摩擦），并利用这些数据通过模拟构建我们的力估计模型。我们使用具有平行夹爪的机器人对抓握力控制的方法进行了实验验证。我们展示了系统在轻柔抓握各种物体方面的能力，并在每种情况下适应其独特的物理特性。

View on arXiv Download PDF AI Translation

cs.RO / 26 / 2610.11223

SafeInferCom: Safe Inference-Time Compute via Verifier-Guided Mid-Generation Intervention for Robotic Task Planning

SafeInferCom：通过验证者引导的中间生成干预实现安全推理时计算的机器人任务规划

Xu, Weizhe, Fan, Jialiang, Liu, Mengyu, Kong, Fanxin

Abstract

Large Reasoning Language Models (LRLMs) enable multi-step reasoning for robotic task planning, but continued reasoning can overwrite valid intermediate plans or leave constraint violations unresolved, reducing planning reliability and wasting inference-time computation. We develop an inference-time monitor that exposes and verifies intermediate plans without disrupting the original decoding trajectory. Building on this monitor, we propose SafeInferCom, a formal verifier-guided framework that preserves valid intermediate plans and directs error correction during generation. Experiments across multiple LRLMs and planning domains reveal reasoning-response inconsistency and limited self-correction under one-shot inference. SafeInferCom improves planning success and accelerates error correction relative to one-shot inference. When combined with iterative refinement, it further improves success while reducing token usage compared with refinement alone. We additionally evaluate SafeInferCom in VirtualHome and provide a real-world robotic-arm demonstration.

Chinese Translation

大型推理语言模型（LRLMs）使得机器人任务规划能够进行多步骤推理，但持续的推理可能会覆盖有效的中间计划或使约束违反未得到解决，从而降低规划的可靠性并浪费推理时计算资源。我们开发了一种推理时监控器，该监控器能够在不干扰原始解码轨迹的情况下暴露和验证中间计划。在此监控器的基础上，我们提出了SafeInferCom，这是一个正式的验证者引导框架，旨在保护有效的中间计划并在生成过程中指导错误修正。在多个LRLM和规划领域的实验中，我们发现了一次性推理下的推理-响应不一致性和有限的自我修正能力。与一次性推理相比，SafeInferCom提高了规划成功率并加速了错误修正。当与迭代优化结合时，它进一步提高了成功率，同时相比单独的优化减少了标记使用。我们还在VirtualHome中评估了SafeInferCom，并提供了一个现实世界的机器人臂演示。

View on arXiv Download PDF AI Translation

cs.RO / 27 / 2610.11248

SimVLA: Zero-Shot Sim-to-Real VLA Learning for Mobile Manipulation

SimVLA：用于移动操作的零样本模拟到真实VLA学习

Baik, Kyoungin, Lee, Youngwoon

Abstract

Large-scale, diverse datasets have driven the success of LLMs and VLMs. But VLAs for robotics remain limited by the cost and complexity of real-world data collection. While simulation offers a scalable alternative, its potential for sim-to-real VLA learning in mobile manipulation remains largely underexplored. We introduce SimVLA, an end-to-end framework that trains VLAs entirely on synthetic simulation data without teleoperation for mobile manipulation. SimVLA is first pre-trained on two complementary simulation-derived datasets: SimAction, a large-scale robot action dataset spanning 35 diverse mobile manipulation tasks, generated by composing atomic skills, and SimVQA, which leverages privileged simulator state to provide spatial, geometric, and subtask-level visual-language supervision. We further post-train SimVLA on a mixture of SimAction and SimDeploy, a dataset collected from policy rollouts across diverse simulated environments. We evaluate SimVLA on tasks including restocking, pouring, and cleaning, and show zero-shot transfer to real-world mobile manipulation, including real home environments. SimVLA outperforms policies trained on 50 in-domain real-world demonstrations, suggesting that simulation can enable scalable sim-to-real mobile manipulation. We further demonstrate the value of multiple complementary forms of supervision for effectively leveraging simulation in VLA training.

Chinese Translation

大规模、多样化的数据集推动了大型语言模型（LLMs）和视觉语言模型（VLMs）的成功。然而，机器人领域的视觉语言代理（VLAs）仍然受到现实世界数据收集的成本和复杂性的限制。虽然模拟提供了一种可扩展的替代方案，但在移动操作中，模拟到真实的VLA学习潜力仍然未被充分探索。我们提出了SimVLA，一个端到端框架，完全基于合成模拟数据训练VLAs，而无需遥操作以实现移动操作。SimVLA首先在两个互补的模拟衍生数据集上进行预训练：SimAction，这是一个涵盖35种多样化移动操作任务的大规模机器人动作数据集，通过组合原子技能生成；SimVQA则利用特权模拟器状态提供空间、几何和子任务级别的视觉语言监督。随后，我们在SimAction和SimDeploy的混合数据集上对SimVLA进行后训练，SimDeploy是从多样化模拟环境中的策略回放收集的数据集。我们在补货、倒水和清洁等任务上评估SimVLA，并展示其在真实世界移动操作中的零样本迁移，包括真实家庭环境。SimVLA的表现优于在50个领域内真实世界示范上训练的策略，这表明模拟可以实现可扩展的模拟到真实移动操作。我们进一步展示了多种互补监督形式在有效利用模拟进行VLA训练中的价值。

View on arXiv Download PDF AI Translation

cs.RO / 28 / 2610.11283

Being-M0.7: A Latent World-Action Model for Humanoid Robots

Being-M0.7：一种用于类人机器人潜在世界-动作模型

Yue, Junpeng, Li, Boyuan, Wang, Yuxuan, Wang, Zepeng, Fu, Yuhui, Xie, Feiyang, Zhang, Yu, Zhang, Jing, Zhang, Xianqi, Li, Weibo, Zheng, Xiaofei, Fang, Yuming, Wang, Jiangxing, Lu, Zongqing

Abstract

Humanoid loco-manipulation requires coordinated locomotion and manipulation informed by future scene evolution and whole-body motion, yet learning these capabilities is constrained by scarce robot demonstrations. Human video and motion datasets offer scalable supervision, but many contain only video or motion rather than paired video-motion data. Moreover, human motion does not directly specify executable robot actions. We present Being-M0.7, a latent world-action model that transfers visual-motion priors learned from mixed-modality human data to humanoid control through pre-training, robot mid-training, and action post-training. We curate a corpus from more than 10,000 hours of raw human-centric data, integrating video-only, motion-only, and paired video-motion streams to learn complementary visual dynamics and whole-body kinematic structure. Joint prediction of future latent visual states and motion encourages visual representations to encode future kinematics. Robot mid-training adapts this coarse-grained prior to robot viewpoints and body dynamics. During action post-training, an action expert combines visual predictive representations from the frozen, adapted prior with current images and proprioception through gated cross-attention, grounding predictive context in executable whole-body commands. Being-M0.7 achieves the highest aggregate success rate among the compared baselines on SIMPLE and matches the strongest baseline on real-world Unitree G1 loco-manipulation tasks.

Chinese Translation

类人机器人在运动操控中需要协调的运动和操控，这需要基于未来场景演变和全身运动的信息，然而学习这些能力受到稀缺机器人演示的限制。人类视频和运动数据集提供了可扩展的监督，但许多数据集仅包含视频或运动，而不是配对的视频-运动数据。此外，人类运动并未直接指定可执行的机器人动作。我们提出了Being-M0.7，这是一种潜在世界-动作模型，通过预训练、机器人中期训练和动作后期训练，将从混合模态人类数据中学习的视觉-运动先验转移到类人控制中。我们从超过10,000小时的原始以人为中心的数据中整理了一个语料库，整合了仅视频、仅运动和配对的视频-运动流，以学习互补的视觉动态和全身运动学结构。对未来潜在视觉状态和运动的联合预测鼓励视觉表征编码未来运动学。机器人中期训练将这种粗粒度的先验适应于机器人视角和身体动态。在动作后期训练中，动作专家通过门控交叉注意机制，将来自冻结的、适应后的先验的视觉预测表征与当前图像和本体感觉结合， grounding 可执行的全身指令中的预测上下文。Being-M0.7在SIMPLE上实现了与比较基线中最高的综合成功率，并在现实世界的Unitree G1运动操控任务中与最强基线相匹配。

View on arXiv Download PDF AI Translation

cs.RO / 29 / 2610.11308

Distributed Relative Localization for Homogeneous Multi-Robot Systems through UWB Ranging and Limited Communications

通过UWB测距和有限通信实现均匀多机器人系统的分布式相对定位

Cao, Zhiqiang, Liu, Ran, Lau, Billy Pik Lik, Yuen, Chau, Tan, U-Xuan

Abstract

Accurate and reliable relative localization is crucial for multi-robot applications like exploration, search, and rescue missions. LiDAR-based solutions offer high accuracy in localizing surrounding objects; however, distinguishing homogeneous robots with similar appearances remains challenging due to the lack of distinctive identification features. In this paper, we propose a fully distributed relative pose estimation approach by integrating LiDAR, UWB, and odometry measurements, allowing each robot to accurately and continuously localize its teammates without external infrastructure. Specifically, potential anonymous teammate robot clusters from LiDAR scans are tracked by a dynamic tracker. We then identify teammate robots from these tracked anonymous clusters using a joint matching strategy, ensuring reliable data association between robots and clusters. Finally, by combining the corresponding LiDAR observations, UWB ranging, and odometry measurements, each robot precisely localizes others while minimizing data exchange. The system requires only odometry data exchange through onboard UWB, eliminating the need for additional communication infrastructure like WiFi routers or mesh networks. Extensive simulation and real-world experiments demonstrate the effectiveness and reliability of the proposed relative localization approach.

Chinese Translation

准确可靠的相对定位对于多机器人应用（如探索、搜索和救援任务）至关重要。基于激光雷达（LiDAR）的解决方案在定位周围物体方面提供了高精度；然而，由于缺乏独特的识别特征，区分外观相似的均匀机器人仍然具有挑战性。本文提出了一种完全分布式的相对姿态估计方法，通过整合LiDAR、超宽带（UWB）和里程计测量，使每个机器人能够在没有外部基础设施的情况下准确且持续地定位其队友。具体而言，来自LiDAR扫描的潜在匿名队友机器人集群通过动态跟踪器进行跟踪。然后，我们使用联合匹配策略从这些跟踪的匿名集群中识别队友机器人，确保机器人与集群之间的数据关联可靠。最后，通过结合相应的LiDAR观测、UWB测距和里程计测量，每个机器人在最小化数据交换的同时精确定位其他机器人。该系统仅需通过车载UWB进行里程计数据交换，消除了对WiFi路由器或网状网络等额外通信基础设施的需求。大量的仿真和实际实验表明了所提出的相对定位方法的有效性和可靠性。

View on arXiv Download PDF AI Translation

cs.RO / 30 / 2610.11322

USDCraft: Geometrically Grounded Programmatic Modeling of Articulated 3D Assets for Simulation

USDCraft：基于几何的程序化建模用于仿真中的关节式3D资产

Zhang, Chuanrui, Yang, Zaijia, Wang, Duomin, Shi, Lu, Zhou, Daquan, Zhang, Ruihua, Wang, Ziwei

Abstract

Geometrically faithful and functional articulated 3D assets are essential for real-to-sim robot manipulation, where policies trained in simulation must transfer to physical objects. Recent mesh-based methods learn to infer articulation from annotated 3D assets, but deployment remains challenging when real-world objects fall outside the training distribution or their meshes are incomplete or corrupted. To address these limitations, we formulate articulated asset reconstruction as programmatic modeling grounded in partial geometric evidence and introduce USDCraft, a framework in which a pretrained LLM writes and revises executable programs for simulation-ready articulated assets without task-specific training. We propose source geometry analysis, which converts the source mesh into a metric textual description that distinguishes observed surface from unknown space, and iterative geometric rechecking, which re-encodes each candidate in the same representation so that discrepancies point to program edits while unobserved regions remain open to completion. Visual feedback and physical authoring guidance complete the modeling process, which produces articulated USD assets with explicit physical properties that load into Isaac Sim without manual adjustment. Experiments demonstrate leading articulation recovery on two benchmarks and validate USDCraft's effectiveness for real-to-sim-to-real robot manipulation.

Chinese Translation

几何上真实且功能性强的关节式3D资产对于真实到仿真的机器人操作至关重要，其中在仿真中训练的策略必须能够迁移到物理对象上。最近的基于网格的方法通过注释的3D资产学习推断关节，但当现实世界中的对象超出训练分布或其网格不完整或损坏时，部署仍然具有挑战性。为了解决这些限制，我们将关节资产重建形式化为基于部分几何证据的程序化建模，并引入USDCraft，一个框架，其中预训练的LLM编写和修订可执行程序，以便为仿真准备的关节资产提供支持，而无需特定任务的训练。我们提出了源几何分析，它将源网格转换为度量文本描述，以区分观察到的表面和未知空间，以及迭代几何重新检查，它在相同的表示中重新编码每个候选，以便差异指向程序编辑，而未观察到的区域则保持开放以供完成。视觉反馈和物理创作指导完成建模过程，生成具有明确物理属性的关节式USD资产，这些资产可以在Isaac Sim中加载而无需手动调整。实验表明，在两个基准测试中实现了领先的关节恢复，并验证了USDCraft在真实到仿真再到真实的机器人操作中的有效性。

View on arXiv Download PDF AI Translation

cs.RO / 31 / 2610.11382

PlanWAM: Planning-Shaped Future Representations for End-to-End Autonomous Driving

PlanWAM：面向端到端自动驾驶的规划形状未来表征

Xu, Jinchang, Yu, Hongda, Dong, Fengwei, Huang, Wenhui, Wei, Xi, Liu, Yongzhi, Zhang, Sunan, Wang, Jirao, Lv, Chen, Li, Bingbing, Yin, Guodong, Zhuang, Weichao

Abstract

World models in end-to-end autonomous driving predict future scene evolution to provide foresight for trajectory planning. Existing methods mainly study how to predict the future and how to use it, but less often ask which future representation is actually most useful for planning. To this end, we propose PlanWAM, a Planning-Shaped World Action Model. The key idea is to let the planning task shape the future-state representation, so that it retains the information most useful for planning. A latent world model then predicts this planning-shaped future latent representation from historical observations and uses it for planning, enabling foresighted planning. Specifically, we first use a Temporal Register Pyramid to compress multi-frame historical information in a recency-aware manner, learning a compact history representation oriented toward future reasoning and planning. We then introduce a privileged future posterior branch that observes ground-truth future frames, and shape its future latent representation with trajectory-planning objectives to obtain a planning-shaped future latent representation. Hindsight-to-Foresight Distillation trains a prior branch that depends only on history to predict this future latent representation. The predicted future latent representation serves as planning context and guides trajectory generation and selection. PlanWAM achieves 93.8 PDMS / 90.9 EPDMS on NAVSIM-v1/v2 navtest and reaches 38.7 HD-Score on closed-loop HUGSIM in a zero-shot setting, demonstrating leading planning performance across both open-loop and closed-loop evaluations. Extensive experiments further demonstrate that planning-shaped future representations provide an effective and deployable form of foresight for world-action models.

Chinese Translation

在端到端自动驾驶中，世界模型预测未来场景演变，以为轨迹规划提供前瞻性。现有方法主要研究如何预测未来以及如何使用这些预测，但较少探讨哪种未来表征实际上对规划最有用。为此，我们提出了PlanWAM，一个规划形状的世界动作模型。其关键思想是让规划任务塑造未来状态表征，以保留对规划最有用的信息。然后，一个潜在的世界模型从历史观察中预测这种规划形状的未来潜在表征，并将其用于规划，从而实现前瞻性规划。具体而言，我们首先使用时间注册金字塔以关注近期的方式压缩多帧历史信息，学习一个面向未来推理和规划的紧凑历史表征。接着，我们引入一个特权未来后验分支，该分支观察真实的未来帧，并通过轨迹规划目标塑造其未来潜在表征，以获得规划形状的未来潜在表征。事后到前瞻蒸馏训练一个仅依赖历史的先验分支，以预测这种未来潜在表征。预测的未来潜在表征作为规划上下文，指导轨迹生成和选择。PlanWAM在NAVSIM-v1/v2 navtest上达到了93.8 PDMS / 90.9 EPDMS，并在零样本设置下在闭环HUGSIM上达到了38.7 HD-Score，展示了在开放环和闭环评估中领先的规划性能。大量实验进一步表明，规划形状的未来表征为世界动作模型提供了一种有效且可部署的前瞻性形式。

View on arXiv Download PDF AI Translation

cs.RO / 32 / 2610.11401

WAM-Cache: Staleness-Bounded KV Reuse for Efficient World Action Models

WAM-Cache：用于高效世界动作模型的陈旧性限制键值重用

Ding, Kai, He, Yang, Quan, Ruijie, Yang, Yi

Abstract

World Action Models (WAMs) enable generalist robot manipulation by conditioning an action expert on representations from a pretrained video Diffusion Transformer (DiT). In closed-loop control, the video DiT runs at every chunk to encode the current observation into layerwise key-value (KV) pairs that the action expert queries. This prefill dominates the per-chunk computational cost, yet existing training-free accelerations leave it fully dense. We present WAM-Cache, a training-free framework that retains layerwise key-value representations across chunks and recomputes only a sparse refresh set of tokens. Crucially, we find that the intuitive heuristic of refreshing visually drifted tokens plateaus far below the dense baseline, even with an oracle predicting ground-truth KV drift. Downstream action accuracy is instead governed by where the action expert attends, not by what moved. WAM-Cache therefore selects the refresh set by uniting the action expert's cross-attention with visual latent surprise, complemented by a strict age bound that suppresses compounding error. On Fast-WAM, WAM-Cache cuts video DiT prefill FLOPs by 32-42% across RoboTwin 2.0, LIBERO, and real-world experiments, while staying within 0.7-1.8 percentage points of the dense policy in simulation and 2.5 points on a real robot.

Chinese Translation

世界动作模型（WAMs）通过将动作专家的条件与预训练视频扩散变换器（Diffusion Transformer, DiT）的表示相结合，使通用机器人操作成为可能。在闭环控制中，视频 DiT 在每个数据块上运行，将当前观察编码为层级键值（key-value, KV）对，供动作专家查询。这种预填充占据了每个数据块的计算成本，然而现有的无训练加速方法仍然保持其完全密集。我们提出了 WAM-Cache，一个无训练框架，能够在数据块之间保留层级键值表示，并仅重新计算一组稀疏的刷新令牌。关键的是，我们发现刷新视觉漂移令牌的直观启发式方法，其性能远低于密集基线，即使在有一个预言者预测真实 KV 漂移的情况下也是如此。下游动作的准确性实际上由动作专家的关注点决定，而不是由移动的内容决定。因此，WAM-Cache 通过将动作专家的交叉注意力与视觉潜在惊奇相结合来选择刷新集，并辅以严格的年龄限制，以抑制累积误差。在 Fast-WAM 上，WAM-Cache 在 RoboTwin 2.0、LIBERO 和真实世界实验中将视频 DiT 预填充的浮点运算（FLOPs）减少了 32-42%，同时在仿真中保持在密集策略的 0.7-1.8 个百分点之内，在真实机器人上保持在 2.5 个百分点之内。

View on arXiv Download PDF AI Translation

cs.RO / 33 / 2610.11416

Rewiring Semantics, Dynamics, and Control: A Simple yet Effective Action-Centric Tri-Stream Transformer

重构语义、动态与控制：一种简单而有效的以行动为中心的三流变换器

Luo, Shuang, Kong, Yilun, Qing, Yunpeng, Jiao, Yihang, Hou, Zhi, Liu, Shunyu, Wang, Xiaogang, Tao, Dacheng

Abstract

Vision-Language-Action (VLA) models have emerged as a prominent framework for complex robotic manipulation, building on the strong semantic understanding of pretrained Vision-Language Models (VLMs). However, such VLM backbones offer insufficient physical dynamics priors, which limits the generalization capabilities of robot policies. Recent efforts therefore integrate video-generation World Models (WMs) into robot policies through various strategies, using predictive dynamics to facilitate action generation. Despite these advances, harnessing semantic understanding and dynamics prediction as complementary guidance for action generation remains challenging. In this paper, we introduce $\mathrm{ACT}^3$, a simple yet effective Action-Centric Tri-Stream Transformer that fuses semantic and dynamics information into control actions while preserving the distinct roles of context streams. Specifically, $\mathrm{ACT}^3$ enables the dedicated action expert to access VLM and WM representations through layerwise attention, with each backbone attending only within its own stream. This straightforward interaction design maintains independent forward propagation in the context streams while allowing both backbones to be updated through control supervision. Experiments on both simulated and real-world robotic manipulation benchmarks show that the proposed $\mathrm{ACT}^3$ yields results superior to its counterparts.

Chinese Translation

视觉-语言-行动（VLA）模型已成为复杂机器人操作的重要框架，建立在预训练视觉-语言模型（VLMs）强大的语义理解基础上。然而，这些VLM骨干网络提供的物理动态先验不足，限制了机器人策略的泛化能力。因此，最近的研究通过各种策略将视频生成世界模型（WMs）集成到机器人策略中，利用预测动态来促进行动生成。尽管取得了这些进展，但将语义理解和动态预测作为行动生成的互补指导仍然具有挑战性。在本文中，我们提出了$ ext{ACT}^3$，一种简单而有效的以行动为中心的三流变换器，它将语义和动态信息融合到控制动作中，同时保留上下文流的独特角色。具体而言，$ ext{ACT}^3$使专门的行动专家能够通过层级注意力访问VLM和WM表示，每个骨干网络仅在其自身流中进行注意。这种简单的交互设计保持了上下文流中的独立前向传播，同时允许两个骨干通过控制监督进行更新。在模拟和真实世界的机器人操作基准测试中的实验表明，所提出的$ ext{ACT}^3$的结果优于其对手。

View on arXiv Download PDF AI Translation

cs.RO / 34 / 2610.11418

Experience-Guided Initiation Search for Learned Skills in Skill Composition

基于经验引导的技能组合中学习技能的启动搜索

Li, Qixuan, Zhao, Yanhong, Yu, Jincheng

Abstract

Deploying frozen learned skills, such as Vision-Language-Action (VLA) policies, in new environments requires identifying initiation configurations that support reliable execution. In skill composition, an initiation configuration affects not only the current skill but also the physical state passed to subsequent skills, so successful execution of an individual skill does not necessarily imply successful completion of the composed task. Estimating target-specific capability through extensive rollouts is costly in real-world deployment, while directly reusing historical experience can be unreliable under environment changes. We propose EVIS, an Experience-Guided and Behavior-Validated Initiation Search framework for discovering reliable initiation configurations under limited target interaction. EVIS uses historical execution experience to prioritize promising candidates and target-environment behavior to validate whether they remain effective. We evaluate EVIS on single-skill and two-stage manipulation tasks with frozen VLA policies. EVIS reduces mean target-environment queries and improves reliable candidate discovery under small interaction budgets. These results show that combining historical guidance with target-side behavioral validation can reduce the interaction cost of deploying frozen learned skills in new environments.

Chinese Translation

在新环境中部署冻结的学习技能，如视觉-语言-动作（Vision-Language-Action, VLA）策略，需要识别支持可靠执行的启动配置。在技能组合中，启动配置不仅影响当前技能，还影响传递给后续技能的物理状态，因此单个技能的成功执行并不一定意味着组合任务的成功完成。通过广泛的回滚估计特定目标的能力在实际部署中成本高昂，而在环境变化下直接重用历史经验可能不可靠。我们提出了EVIS，一个基于经验引导和行为验证的启动搜索框架，用于在有限的目标交互下发现可靠的启动配置。EVIS利用历史执行经验优先考虑有前景的候选者，并利用目标环境行为验证这些候选者是否仍然有效。我们在使用冻结的VLA策略的单技能和两阶段操作任务上评估了EVIS。EVIS减少了平均目标环境查询次数，并在小的交互预算下提高了可靠候选者的发现。这些结果表明，将历史指导与目标侧行为验证相结合可以降低在新环境中部署冻结学习技能的交互成本。

View on arXiv Download PDF AI Translation

cs.RO / 35 / 2610.11480

RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes

RoboAware：从反事实结果中学习协调具身技能

Zhou, Bohan, Chen, Xingbei, Huang, Emily, Ruan, Weilin, Huang, Haojian, Zhang, Yehang, Li, Zexi, Li, Wenqian, Yu, Qize, Song, Zetian, Wu, Leyi, Li, Jinghao, Song, Mingxuan, Xu, Xinrun, Qiu, Zongyang, Wei, Yangkai, Zhang, Tianyi, Zhou, Kaiwen, Li, Yinchuan, Cheng, James

Abstract

Embodied coding agents can combine modular robot skills with frozen end-to-end policies, yet effective composition requires anticipating which policy family will succeed in the current physical state. We present RoboAware, which builds on coding agents' skill orchestration by learning only a state-conditioned responsibility coordinator from counterfactual outcomes. Inspired by the success of REPL, we propose the $P^5$ schema and formulate a hierarchical MDP based on it. $P^5$ organizes skills uniformly into five semantic stages, defining where responsibility can be compared. To address the lack of counterfactual branch outcomes in existing work, we introduce State-Locked Counterfactual Branching (SCB), which restores the same training state to generate and execute a code block from each admissible family, exposing outcomes that selected-branch experience leaves unobserved. Building on this, we propose Execution-Aware Learning (EAL), which combines Monte Carlo tree search with Q-learning to distill these outcomes into family-conditioned values. At deployment, the coordinator selects the policy family according to observable context, and the frozen coding agent generates the next local code block. Comprehensive single-episode evaluations on 100 tasks show that RoboAware reaches a 77.0% overall success rate, with SOTA averages of 90.0% on RoboSuite, 73.8% on diverse LIBERO-Pro task clusters, and 90.0% on challenging RoboTwin bimanual tasks, outperforming existing code-as-policy and VLA-harness baselines.

Chinese Translation

具身编码代理可以将模块化机器人技能与冻结的端到端策略相结合，但有效的组合需要预测在当前物理状态下哪个策略家族会成功。我们提出了RoboAware，它通过仅从反事实结果中学习一个状态条件的责任协调器，构建在编码代理的技能编排之上。受到REPL成功的启发，我们提出了$P^5$框架，并基于此构建了一个层次化的马尔可夫决策过程（MDP）。$P^5$将技能均匀地组织为五个语义阶段，定义了可以进行责任比较的地方。为了解决现有工作中缺乏反事实分支结果的问题，我们引入了状态锁定反事实分支（SCB），它恢复相同的训练状态，以生成和执行来自每个可接受家族的代码块，揭示被选择分支经验未观察到的结果。在此基础上，我们提出了执行感知学习（EAL），它将蒙特卡洛树搜索与Q学习相结合，以将这些结果提炼为家族条件值。在部署时，协调器根据可观察的上下文选择策略家族，而冻结的编码代理生成下一个本地代码块。在100个任务上的全面单回合评估显示，RoboAware达到了77.0%的整体成功率，在RoboSuite上平均达到90.0%，在多样化的LIBERO-Pro任务集群上达到73.8%，在具有挑战性的RoboTwin双手任务上达到90.0%，超越了现有的代码作为策略和VLA-harness基准。

View on arXiv Download PDF AI Translation

cs.RO / 36 / 2610.11505

DAMP: Humanoid Locomotion via Denoised Belief Learning and Adversarial Motion Priors

DAMP：通过去噪信念学习和对抗运动先验实现类人步态

Shen, Puying, Cui, Wenhao, Huang, Huaxing, Qin, Bangyu, Li, Shengtao, Dong, Ziyang, Zhang, Guoteng

Abstract

Humanoid robots possess the structural capability to traverse complex terrains. However, achieving stable t raversal without relying on perceived information remains challenging, particularly in complex environments. This paper introduces DAMP, a reinforcement learning framework aimed at achieving robust and naturalistic humanoid locomotion over challenging terrains, with the assumption that no perceived information is available. The framework leverages recurrent neural networks to capture temporal dependencies and implicitly infer privileged and other task-relevant latent information. By aligning the learned representations with the task objective, the method enables robust and goal-consistent policy learning. This end-to-end framework achieves transfer learning from simulation to real-world environments, demonstrating the proposed method's robustness and generalization capabilities. The video of the real-world demonstration can be found at the following link: https://youtu.be/AkI7TZB2DDM.

Chinese Translation

类人机器人具备穿越复杂地形的结构能力。然而，在不依赖感知信息的情况下实现稳定的穿越仍然具有挑战性，尤其是在复杂环境中。本文介绍了DAMP，一个旨在实现类人在挑战性地形上稳健且自然步态的强化学习框架，假设没有可用的感知信息。该框架利用递归神经网络捕捉时间依赖性，并隐式推断特权和其他与任务相关的潜在信息。通过将学习到的表示与任务目标对齐，该方法实现了稳健且目标一致的策略学习。该端到端框架实现了从仿真到现实环境的迁移学习，展示了所提方法的稳健性和泛化能力。现实世界演示的视频可以在以下链接找到：https://youtu.be/AkI7TZB2DDM。

View on arXiv Download PDF AI Translation

cs.RO / 37 / 2610.11508

WARP-VLA: Wrist-Camera Adaptation for View-Robust Policy Execution in Vision-Language-Action Models

WARP-VLA：腕部摄像头适应性在视觉-语言-动作模型中的视图鲁棒策略执行

Lee, Junmyeong, Shin, Dongmin, Park, Min-Gyu, Jeon, Wooseok, Chang, Inho, Jeon, Hae-Gon

Abstract

Despite recent advances in Vision-Language-Action models (VLAs) for robotic manipulation, their performance remains sensitive to changes in camera configuration. The problem becomes more evident in cross-setup deployment, as reproducing the exact camera pose used for training is nearly impossible. Unlike fixed external views, wrist views are more challenging because the camera moves with the robot, causing even small mounting variations to alter fine-grained geometric cues. To address this, we propose WARP-VLA, a camera-view robust VLA for diverse wrist camera configurations. WARP-VLA adopts a Mixture-of-Experts (MoE) architecture where individual experts learn view-specific feature transformations, and a router combines them based on implicit view information. This allows the policy to be deployed without requiring camera extrinsic parameters as additional input. Through experiments on the LIBERO benchmark, WARP-VLA improves the average success rate of pi-0.5 from 39.2% to 78.3% under wrist-view perturbations. The real-robot experiments further show that the feature-level adaptation learned in simulation successfully transfers to diverse deployment settings. To facilitate reproducibility and future research, we release our wrist viewpoint robustness benchmark and a plug-and-play implementation.

Chinese Translation

尽管近年来视觉-语言-动作模型（VLA）在机器人操控方面取得了进展，但其性能仍对摄像头配置的变化敏感。在跨设置部署中，这一问题变得更加明显，因为几乎不可能重现用于训练的确切摄像头姿态。与固定的外部视角不同，腕部视角更具挑战性，因为摄像头随着机器人移动，甚至微小的安装变化都会改变细致的几何线索。为了解决这一问题，我们提出了WARP-VLA，一种针对多样化腕部摄像头配置的视图鲁棒VLA。WARP-VLA采用混合专家（Mixture-of-Experts, MoE）架构，其中各个专家学习特定视角的特征变换，而路由器根据隐式视角信息将它们结合起来。这使得策略可以在不需要摄像头外部参数作为额外输入的情况下进行部署。通过在LIBERO基准上的实验，WARP-VLA在腕部视角扰动下将pi-0.5的平均成功率从39.2%提高到78.3%。真实机器人实验进一步表明，在模拟中学习到的特征级适应性成功转移到多样化的部署环境中。为了促进可重复性和未来研究，我们发布了腕部视角鲁棒性基准和即插即用实现。

View on arXiv Download PDF AI Translation

cs.RO / 38 / 2610.11531

RAGNAROK: Radar-Aided Gravity-Normalized Alignment for Robust Open Keyframe-based Radar-Visual-Kinematic-Inertial SLAM

RAGNAROK：基于雷达辅助的重力归一化对齐方法，用于稳健的开放关键帧雷达-视觉-运动学-惯性SLAM

Kim, Hanjun, Noh, Chiyun, Jung, Sangwoo, Jung, Jaehyung, Boche, Simon, Gentil, Cedric Le, Leutenegger, Stefan, Kim, Ayoung

Abstract

Legged robots offer superior mobility in unstructured environments, but reliable operation in such conditions requires robust state estimation. To address the vulnerability of proprioceptive estimators in rough terrain, recent methods have incorporated radar to provide velocity measurements. However, their limited yaw observability still leads to drift, and failure-aware fusion for adverse environments remains underexplored. In this letter, we present RAGNAROK, the first radar-visual-kinematic-inertial SLAM designed for robust operation in challenging environments. It integrates slip- and rolling-contact-aware leg velocity estimation, a kinematics-aware radar factor, and degradation-aware image enhancement. We further incorporate a B-spline-based radar-aided proprioceptive backbone, adaptive weighting, and online extrinsic calibration. Extensive experiments on public and self-collected datasets demonstrate that RAGNAROK achieves robust performance under challenging conditions and outperforms state-of-the-art baselines. The source code and dataset are available at https://github.com/hanjun815/RAGNAROK.

Chinese Translation

腿式机器人在非结构化环境中提供了优越的机动性，但在此类条件下可靠的操作需要稳健的状态估计。为了应对本体感知估计器在崎岖地形中的脆弱性，近期的方法已将雷达纳入其中，以提供速度测量。然而，它们有限的偏航可观测性仍然导致漂移，而针对恶劣环境的故障感知融合仍然未得到充分探索。在本文中，我们提出了RAGNAROK，这是首个为在挑战性环境中稳健操作而设计的雷达-视觉-运动学-惯性SLAM。它集成了考虑滑移和滚动接触的腿部速度估计、一个考虑运动学的雷达因子，以及故障感知的图像增强。我们进一步结合了基于B样条的雷达辅助本体感知骨干、自适应加权和在线外部标定。在公共和自收集数据集上的大量实验表明，RAGNAROK在挑战性条件下实现了稳健的性能，并超越了最先进的基线。源代码和数据集可在 https://github.com/hanjun815/RAGNAROK 获取。

View on arXiv Download PDF AI Translation

cs.RO / 39 / 2610.11577

CoCam4D: Geometry-Aware Cooperative 4D Perception for Camera-Only Autonomous Driving

CoCam4D：面向相机的几何感知协作四维感知用于自主驾驶

Pahari, Soham, Das, Sudip, Das, Arindam, Bhattacharya, Ujjwal

Abstract

Autonomous vehicles often suffer from limited perception due to occlusions, blind spots, limited sensor range, and the complex nature of surrounding environments. Multi-agent collaborative perception (CP) addresses these challenges by allowing vehicles to share sensory information and reconstruct the scene cooperatively. However, camera-only perception remains fundamentally limited by the uncertainty of distance-dependent monocular depth estimation. We propose CoCam4D, a Bayesian framework for collaborative perception that explicitly models geometric uncertainty. It uses a VGGT-based feedforward network to generate 3D Gaussian scene representations with associated uncertainty estimates, enabling multiple vehicles or agents to efficiently combine their observations. By sharing compact Gaussian primitives, reliable observations from one agent can reduce the depth uncertainty of another without requiring LiDAR sensors. To support real-world deployment, we introduce Dynamic Object Primitives (DOPs), a compact 35-byte representation designed for efficient C-V2X communication. Extensive experiments show that our proposed method consistently outperforms recent vision-only methods, achieving improvements of 11.48% on OPV2V+ and 10.62% on DAIR-V2X-C, demonstrating the potential of geometrically grounded collaborative perception for LiDAR-free autonomous driving.

Chinese Translation

自主车辆常常因遮挡、盲区、传感器范围有限以及周围环境的复杂性而面临感知能力不足的问题。多智能体协作感知（CP）通过允许车辆共享传感信息并协同重建场景来应对这些挑战。然而，仅依赖相机的感知仍然受到依赖距离的单目深度估计不确定性的根本限制。我们提出了CoCam4D，一种显式建模几何不确定性的协作感知贝叶斯框架。该框架使用基于VGGT的前馈网络生成带有相关不确定性估计的3D高斯场景表示，使多个车辆或智能体能够高效地结合其观测结果。通过共享紧凑的高斯原语，一个智能体的可靠观测可以在不需要激光雷达（LiDAR）传感器的情况下减少另一个智能体的深度不确定性。为了支持实际部署，我们引入了动态物体原语（DOPs），这是一种为高效的C-V2X通信设计的紧凑35字节表示。大量实验表明，我们提出的方法在最近的仅基于视觉的方法中表现一致优越，在OPV2V+上提高了11.48%，在DAIR-V2X-C上提高了10.62%，展示了基于几何的协作感知在无激光雷达自主驾驶中的潜力。

View on arXiv Download PDF AI Translation

cs.RO / 40 / 2610.11583

SDPAD: A Fully Spike-Driven Pipeline for End-to-End Autonomous Driving

SDPAD：一种完全基于脉冲的端到端自主驾驶管道

Zhang, Chengjun, Zhang, Yuhao, Yang, Jie, Sawan, Mohamad

Abstract

End-to-end autonomous driving demands trajectory planners that are both highly accurate and cheap enough for edge deployment. State-of-the-art artificial neural network (ANN) planners meet the accuracy requirement at the cost of heavy dense computation, while spiking neural networks (SNNs)---though promising orders-of-magnitude energy savings through sparse, event-driven arithmetic---still lag far behind in planning accuracy. We present \textbf{SDPAD}, a fully spike-driven end-to-end planning pipeline that closes this gap. SDPAD converts a pre-trained ANN perception stack into integer-spike form via quantized ANN2SNN conversion, lifts multi-view images into the bird's-eye-view (BEV) space with a spike-driven-max (SDM) depth distribution (Spike-3D-Lift), and plans through the Spike-QFormer, a spiking query transformer in which ego, agent, and map queries distilled from the BEV scene are fused by learnable waypoint queries via cross-attention, followed by deformable spike-cross-attention refinement. Every operation is gated by integer spikes and inference is a single feed-forward pass without temporal simulation loops. On the nuScenes open-loop benchmark, SDPAD achieves an average $L_2$ error of 0.40\,m and a collision rate of 0.12\%, on par with strong ANN planners while consuming 69.9\,mJ---less than 2\% of recent ANN baselines. In closed-loop evaluation on the NAVSIM navtest split, SDPAD reaches 86.3 PDMS, surpassing the previous SNN planner SAD by 4.3 points and matching mainstream ANN planners at a fraction of their energy. To our knowledge, SDPAD is the first fully spike-driven planner evaluated in end-to-end autonomous driving, demonstrating that SNNs can rival dense ANNs in complex driving tasks.

Chinese Translation

端到端自主驾驶需要既高精度又足够便宜以便于边缘部署的轨迹规划器。最先进的人工神经网络（ANN）规划器在满足精度要求的同时，代价是进行大量密集计算，而脉冲神经网络（SNN）虽然通过稀疏的事件驱动算术在能耗上具有数个数量级的节省，但在规划精度上仍远远落后。我们提出了 extbf{SDPAD}，一种完全基于脉冲的端到端规划管道，旨在弥补这一差距。SDPAD通过量化的ANN2SNN转换将预训练的ANN感知堆栈转换为整数脉冲形式，利用脉冲驱动的最大（SDM）深度分布将多视图图像提升到鸟瞰图（BEV）空间（Spike-3D-Lift），并通过脉冲查询变换器（Spike-QFormer）进行规划，其中从BEV场景提取的自我、代理和地图查询通过可学习的路径点查询通过交叉注意力进行融合，随后进行可变形脉冲交叉注意力的细化。每个操作都由整数脉冲控制，推理过程为单次前馈传递，无需时间模拟循环。在nuScenes开放循环基准测试中，SDPAD实现了平均$L_2$误差为0.40米，碰撞率为0.12%，与强大的ANN规划器相当，同时消耗69.9毫焦耳——不到最近ANN基线的2%。在NAVSIM导航测试集的闭环评估中，SDPAD达到了86.3 PDMS，超越了之前的SNN规划器SAD 4.3分，并在能耗上与主流ANN规划器相匹配。据我们所知，SDPAD是首个在端到端自主驾驶中评估的完全基于脉冲的规划器，证明了SNN在复杂驾驶任务中可以与密集的ANN相抗衡。

View on arXiv Download PDF AI Translation

cs.RO / 41 / 2610.11591

Acting from Belief, Looking When Needed: A Bayesian Spatial World Model for Navigation under Intermittent Perception

基于信念的行动，必要时再观察：一种用于间歇感知下导航的贝叶斯空间世界模型

Yang, Feihong, Long, Xiang, Yu, Jincheng, Zhang, Jianfei, Ge, Guangjun, Wang, Chao, Wang, Yu

Abstract

Robot navigation commonly uses wide-coverage, high-frequency sensing to reduce partial observability; this reliance becomes restrictive when another task temporarily redirects a shared sensor from navigation, interrupting navigation-relevant observations. We study navigation under intermittent perception: acting from an internal spatial belief and looking again only when execution needs a new observation, potentially freeing the shared sensor for other tasks between navigation observations. ALONE, a Bayesian spatial world model, propagates a structured spatial belief using executed actions and corrects it with selectively acquired observations; learned priors over common geometric structures infer unobserved structure from available observation history. It decodes the belief into a spatial estimate for the motion-planning module and predicts a reliability map expressing confidence in the estimate's accuracy. ALONE requests an observation only if insufficient reliability hinders navigation and new evidence should make relevant-region spatial information more reliable; otherwise, it continues acting from the propagated belief. We instantiate ALONE for drone navigation with intermittent single-camera depth images. Across two simulated scene families, it achieves 98% and 97% closed-loop success at a 10 Hz decision rate. Among successful trials, median fractions of decision steps requiring a new depth observation are only 0.9% and 1.3%, respectively, demonstrating high navigation success with substantially reduced observation demand. Real-world indoor flight experiments further validate navigation under intermittent depth observations, with all 10 trials successful.

Chinese Translation

机器人导航通常依赖广覆盖、高频率的传感来减少部分可观测性；当另一个任务暂时将共享传感器从导航中转移时，这种依赖变得限制性，打断与导航相关的观察。我们研究间歇感知下的导航：基于内部空间信念进行行动，仅在执行需要新的观察时再进行观察，这可能使共享传感器在导航观察之间可以用于其他任务。ALONE（一个贝叶斯空间世界模型）通过执行的动作传播结构化的空间信念，并通过选择性获取的观察进行校正；对常见几何结构的学习先验从可用的观察历史中推断未观察到的结构。它将信念解码为运动规划模块的空间估计，并预测一个可靠性图，表达对估计准确性的信心。ALONE仅在不足的可靠性妨碍导航且新证据应使相关区域的空间信息更可靠时请求观察；否则，它继续基于传播的信念进行行动。我们为间歇性单摄像头深度图像的无人机导航实例化了ALONE。在两个模拟场景系列中，它以10 Hz的决策频率实现了98%和97%的闭环成功率。在成功的试验中，要求新的深度观察的决策步骤的中位数比例仅为0.9%和1.3%，展示了在显著减少观察需求的情况下实现高导航成功率。现实世界的室内飞行实验进一步验证了在间歇性深度观察下的导航，所有10次试验均成功。

View on arXiv Download PDF AI Translation

cs.RO / 42 / 2610.11622

Learning Language-Conditioned Traversability Representations for Adaptive Visual Navigation

学习语言条件下的可通行性表征用于自适应视觉导航

Chen, Senda, Cheng, Changxu, Li, Fangdi, Wang, Tao, Zhao, Wuyue

Abstract

Traversability is essential for visual navigation but varies with robot capabilities and user preferences. Conventional pipelines often rely on explicit costmaps or segmentation masks with predefined criteria, requiring hand-crafted rules and careful tuning. Moreover, viewpoint-dependent segmentation masks complicate asynchronous planning under perception latency. We present LaTraNav, a framework that learns language-conditioned traversability representations for adaptive visual navigation. Its asynchronous architecture combines a slow vision-language model that produces latent representations of traversability and navigation goals, with a fast flow-matching planner conditioned on these representations. To train the system, we develop a simulation-based data generation pipeline with controllable trajectories, producing observations paired with language instructions, traversability maps, goal locations, and diverse trajectories. Photorealistic image translation further enhances visual realism. Evaluations on datasets from multiple sources demonstrate effective language-guided traversability segmentation and goal localization by the slow VLM, alongside adaptive pixel-space path planning by the fast planner. Latent conditioning improves planning performance over explicit segmentation masks, while asynchronous scheduling increases the path-update rate by $6.05\times$ at the same semantic-update rate.

Chinese Translation

可通行性对于视觉导航至关重要，但它会因机器人能力和用户偏好而异。传统的处理流程通常依赖于显式的代价图或具有预定义标准的分割掩码，这需要手工制定规则和仔细调优。此外，视角依赖的分割掩码使得在感知延迟下的异步规划变得复杂。我们提出了LaTraNav，一个学习语言条件下可通行性表征的框架，用于自适应视觉导航。其异步架构结合了一个缓慢的视觉-语言模型，该模型生成可通行性和导航目标的潜在表征，以及一个快速的流匹配规划器，该规划器基于这些表征进行条件规划。为了训练该系统，我们开发了一个基于仿真的数据生成流程，具有可控的轨迹，生成与语言指令、可通行性图、目标位置和多样化轨迹配对的观察数据。光写实图像翻译进一步增强了视觉真实感。在来自多个来源的数据集上的评估表明，缓慢的视觉-语言模型有效地实现了语言引导的可通行性分割和目标定位，同时快速规划器实现了自适应的像素空间路径规划。潜在条件化提高了规划性能，相较于显式分割掩码，而异步调度在相同语义更新率下将路径更新率提高了$6.05 imes$。

View on arXiv Download PDF AI Translation

cs.RO / 43 / 2610.11631

Neural Networks for Temporal Pattern Recognition and Dynamic Arm Gesture Speed Estimation for Robot Control

用于机器人控制的时间模式识别和动态手势速度估计的神经网络

Bagladi, Milán Zsolt, Gulyás, László

Abstract

Deploying intelligent robotic systems that interact with humans through gestures requires neural networks capable of recognizing diverse temporal patterns. We present a systematic benchmark of ten abstract sequential tasks--five permutation-invariant (set) and five order-dependent (sequence) problems--evaluated across eighteen neural network architectures spanning recurrent, convolutional, attention-based, and set-function families. Beyond the core architecture-task grid, we explore numerous preprocessing and target-variable transformations, yielding more than 250 distinct experimental configurations. All variants are trained and tested under strictly identical conditions (fixed random seeds, shared hyperparameters, shared data splits) to ensure fair and reproducible comparison. Ranking across all ten tasks reveals four consistently top-performing architectures--BiGRU, TCN, Conv1D, and GRUReLU--all compact enough for real-time deployment (under 2,000 parameters in the benchmark setting). Based on this ranking, we apply three architecturally diverse top models (BiGRU, TCN, and GRUReLU) to a practical robotics problem: estimating the execution speed of dynamic arm gestures from skeletal keypoint sequences. Three speed interpretations (peak count, period time, and mean spike spacing) are evaluated on a custom dataset of eight traffic-related gesture classes comprising 256,710 frames recorded via OpenPose. The best configuration achieves a mean absolute error of 0.198 on the peak-count interpretation, corresponding to roughly 5% relative error, while the period-time interpretation reaches approximately 4% relative error, and the mean spike spacing interpretation approximately 8% relative error. These results demonstrate that neural networks can reliably estimate gesture speed from skeletal data, opening a path toward speed-aware gesture-controlled robotic systems.

Chinese Translation

部署能够通过手势与人类互动的智能机器人系统需要能够识别多样时间模式的神经网络。我们提出了十个抽象顺序任务的系统基准评估——五个置换不变（集合）和五个顺序依赖（序列）问题——在十八种神经网络架构上进行评估，这些架构涵盖了递归、卷积、基于注意力和集合函数等类别。除了核心架构-任务网格外，我们还探索了众多预处理和目标变量转换，产生了超过250种不同的实验配置。所有变体在严格相同的条件下（固定随机种子、共享超参数、共享数据划分）进行训练和测试，以确保公平和可重复的比较。在所有十个任务中的排名显示，四种始终表现优异的架构——BiGRU、TCN、Conv1D和GRUReLU——都足够紧凑，适合实时部署（在基准设置中参数少于2000个）。基于这一排名，我们将三种架构多样的顶级模型（BiGRU、TCN和GRUReLU）应用于一个实际的机器人问题：从骨骼关键点序列中估计动态手势的执行速度。我们在一个包含256,710帧的自定义数据集上评估了三种速度解释（峰值计数、周期时间和平均脉冲间距），该数据集由八个与交通相关的手势类别通过OpenPose录制而成。最佳配置在峰值计数解释上实现了0.198的平均绝对误差，相当于大约5%的相对误差，而周期时间解释达到了约4%的相对误差，平均脉冲间距解释则约为8%的相对误差。这些结果表明，神经网络能够可靠地从骨骼数据中估计手势速度，为速度感知的手势控制机器人系统开辟了道路。

View on arXiv Download PDF AI Translation

cs.RO / 44 / 2610.11645

Scalable LEO Conjunction Screening using Adaptive Synthetic-Covariance Thresholds

基于自适应合成协方差阈值的可扩展低地球轨道会合筛选

Srinivas, Vedant, Kim, Grace Ra, Eddy, Duncan, Kochenderfer, Mykel J.

Abstract

As orbital populations grow, conjunction screening must search more object pairs for potential collision risks. Existing pipelines use early-stage filters, such as an apsis filter, to eliminate pairs whose orbital altitude ranges cannot produce a close approach. However, these filters apply the same fixed threshold to every object, making the screening unnecessarily conservative for well-constrained objects while potentially insufficient for objects with greater orbit-prediction uncertainty. In this work, we present an uncertainty-aware conjunction screening method that replaces a single fixed apsis threshold with object-specific thresholds derived from propagated radial uncertainty. The resulting thresholds vary across both objects and prediction time, becoming narrower when propagated uncertainty is small and wider when uncertainty grows. We pair this adaptive screening rule with a binary-search altitude sweep that makes resulting heterogeneous intervals efficient to evaluate at catalog scale with log-linear search complexity. We evaluate the method on 33,791 objects from the Space-Track General Perturbations (GP) catalog using the preceding eight public ephemeris snapshots and a seven-day prediction horizon. The adaptive sieve retains 162.8 million candidate pairs, reducing the candidate set by 28.78% relative to a fixed 50 km apsis threshold and by 71.48% relative to brute-force all-pairs screening. It retains all close approaches identified in two independently propagated reference sets. At the largest tested catalog size, the binary-search sweep is 27 times faster than exhaustive interval evaluation while returning the same candidate set. These results demonstrate that uncertainty-aware screening can substantially reduce the candidate burden passed to downstream conjunction analysis, providing a scalable path toward managing increasingly large and heterogeneous orbital populations.

Chinese Translation

随着轨道人口的增长，会合筛选必须搜索更多的物体对以评估潜在的碰撞风险。现有的筛选流程使用早期阶段的过滤器，例如近地点过滤器，来排除那些轨道高度范围无法产生近距离接近的物体对。然而，这些过滤器对每个物体应用相同的固定阈值，这使得对于约束良好的物体筛选过于保守，而对于轨道预测不确定性较大的物体则可能不足。在本研究中，我们提出了一种考虑不确定性的会合筛选方法，该方法用从传播的径向不确定性衍生的特定物体阈值替代了单一的固定近地点阈值。由此产生的阈值在不同物体和预测时间之间变化，当传播的不确定性较小时，阈值变得更窄，而当不确定性增大时，阈值则变得更宽。我们将这一自适应筛选规则与二分搜索高度扫描相结合，使得在目录规模下以对数线性搜索复杂度高效评估结果的异质区间。我们在来自空间跟踪一般扰动（GP）目录的33,791个物体上评估该方法，使用前八个公共历书快照和七天的预测范围。自适应筛选保留了1.628亿个候选对，相较于固定的50公里近地点阈值减少了28.78%，相较于暴力全对筛选减少了71.48%。它保留了在两个独立传播的参考集中识别的所有近距离接近。在测试的最大目录规模下，二分搜索扫描的速度比穷举区间评估快27倍，同时返回相同的候选集。这些结果表明，考虑不确定性的筛选可以显著减少传递给下游会合分析的候选负担，为管理日益庞大和异质的轨道人口提供了一条可扩展的路径。

View on arXiv Download PDF AI Translation

cs.RO / 45 / 2610.11657

YOCO: You Only Calibrate Once! Fast Mocap Calibration for Dexterous Teleoperation

YOCO：只需校准一次！用于灵巧远程操作的快速动作捕捉校准方法

Zhang, Yu, Li, Yunqi, Du, Yushi, Ma, Yi, Yang, Yanchao

Abstract

Dexterous teleoperation requires reliable human-hand state estimations. However, common low-cost motion-capture gloves and markerless trackers often exhibit biases that vary across users, glove fit, and recording sessions, degrading retargeting and demonstration quality. We present YOCO, a fast few-shot, fine-tuning-free calibration framework that corrects biased hand-pose streams from a small set of paired raw and target poses. Instead of optimizing a separate model for every operator or session, YOCO conditions a calibration HyperNet on the paired examples and predicts LoRA-style updates for a frozen MANO hand-estimation module, turning per-user calibration into a lightweight feed-forward adaptation step while preserving the geometric prior of MANO and the efficiency of a compact estimator. We train YOCO with synthetic drift augmentations on InterHand2.6M and evaluate on augmented InterHand sequences, offline real glove data, and dexterous teleoperation tasks. Across these settings, YOCO improves calibration efficiency, hand-state estimation quality and teleoperation performance compared with uncalibrated input and standard calibration baselines.

Chinese Translation

灵巧远程操作需要可靠的人手状态估计。然而，常见的低成本动作捕捉手套和无标记追踪器通常存在因用户差异、手套贴合度及录制会话不同而变化的偏差，导致重定向和示范质量下降。我们提出了YOCO，一种快速的少样本、无需微调的校准框架，能够通过一小组配对的原始姿态与目标姿态校正有偏差的手部姿态序列。YOCO不需要为每个操作者或会话单独优化模型，而是通过配对样本条件化一个校准HyperNet，预测针对冻结的MANO手部估计模块的LoRA风格更新，将每用户校准转化为轻量级的前馈适应步骤，同时保持MANO的几何先验和紧凑估计器的效率。我们在InterHand2.6M数据集上通过合成漂移增强训练YOCO，并在增强的InterHand序列、离线真实手套数据及灵巧远程操作任务中进行评估。在这些环境下，YOCO相比未校准输入和标准校准基线，提升了校准效率、手部状态估计质量及远程操作性能。

View on arXiv Download PDF AI Translation

cs.RO / 46 / 2610.11722

From Solo to Ensemble: A Hierarchical Framework for Composable Multi-Agent Human-Object Interaction

从单体到集体：可组合的多智能体人机交互的层次框架

Deng, Zekai, Chen, Kangyi, Shi, Ye, Wang, Jingya

Abstract

Physics-based human-object interaction has achieved robust single-agent manipulation skills, yet extending them to multi-agent cooperative tasks remains challenging. Existing approaches typically adapt interaction policies through task-specific fine-tuning, which entangles low-level contact-rich execution with high-level coordination and limits reuse across object geometries, interaction types, and team sizes. We propose a hierarchical framework that converts a single-agent HOI policy into a reusable Object-oriented Motion Skill. Specifically, we reinterpret teacher rollouts as object-oriented action supervision by extracting short-horizon object-proxy motions from executed trajectories, and distill task-specific teachers into a low-level skill operating in an Object-oriented Action Space. For downstream tasks, the distilled skill is frozen as a reusable executor, while a high-level policy coordinates multiple agents by generating region-wise object-oriented actions conditioned on the shared object, task goal, agent states, and local manipulation regions. This formulation shifts multi-agent HOI learning from direct contact-rich full-body control to compact object-level proxy-motion coordination. Experiments on diverse HOI tasks show that the distilled Object-oriented Motion Skill supports robust proxy-motion execution and enables composable policy learning across different interaction types, object geometries, and team sizes.

Chinese Translation

基于物理的人机交互已经实现了稳健的单智能体操作技能，但将其扩展到多智能体协作任务仍然具有挑战性。现有方法通常通过特定任务的微调来调整交互策略，这使得低层次的接触丰富执行与高层次的协调相互纠缠，并限制了在物体几何形状、交互类型和团队规模之间的重用。我们提出了一个层次框架，将单智能体的 HOI（Human-Object Interaction）策略转换为可重用的面向对象运动技能。具体而言，我们通过从执行轨迹中提取短期的对象代理运动，将教师回放重新解释为面向对象的动作监督，并将特定任务的教师提炼为在面向对象的动作空间中操作的低层次技能。对于下游任务，提炼出的技能被固定为可重用的执行器，而高层次策略通过生成基于共享对象、任务目标、智能体状态和局部操作区域的区域性面向对象动作来协调多个智能体。这种表述将多智能体 HOI 学习从直接的接触丰富全身控制转变为紧凑的对象级代理运动协调。在多样化的 HOI 任务上的实验表明，提炼的面向对象运动技能支持稳健的代理运动执行，并使不同交互类型、物体几何形状和团队规模之间的可组合策略学习成为可能。

View on arXiv Download PDF AI Translation

cs.RO / 47 / 2610.11752

2DGS-Planner: Rasterization-based Path Planning in 2D Gaussian Splatting Map

2DGS-规划器：基于光栅化的二维高斯散点图路径规划

Park, Jiwon, Seo, Dong-Uk, Myung, Hyun

Abstract

Gaussian splatting provides an explicit and efficiently rasterizable scene representation for robot navigation. However, individual Gaussian primitives may not reliably represent obstacles as they are jointly optimized through alpha-composited rendering from a finite set of reconstruction views. We propose 2DGS-Planner, a path planner for ground robots that reads planning-relevant geometry from a 2D Gaussian splatting (2DGS) map through rasterization, rather than treating individual Gaussian primitives as obstacles. During offline roadmap construction, multi-view attribution converts rendered normal dispersion into structural scores for non-ground disks supported by the reconstruction views. These scores guide adaptive node sampling on the ground. Path-aligned orthographic queries validate candidate edges, while cylindrical queries estimate local clearance fields that are cached on the edges. During online planning, graph search initializes a route, and path refinement reuses the cached fields while accounting for the robot's dimensions and ground constraints. Experiments demonstrate improved roadmap connectivity, more accurate clearance estimation, and higher planning success compared with the tested baselines. These results support rasterization as an effective geometric query interface for planning directly on Gaussian maps. Code and data are available at https://2dgs-planner.github.io/

Chinese Translation

高斯散点图为机器人导航提供了一种明确且高效可光栅化的场景表示。然而，单个高斯原语可能无法可靠地表示障碍物，因为它们是通过有限重建视图的α合成渲染共同优化的。我们提出了2DGS-规划器，这是一种为地面机器人设计的路径规划器，它通过光栅化从二维高斯散点图（2DGS）地图中读取与规划相关的几何信息，而不是将单个高斯原语视为障碍物。在离线路网构建过程中，多视角归属将渲染的法线分散转换为非地面圆盘的结构评分，这些评分由重建视图支持。这些评分指导地面上的自适应节点采样。路径对齐的正交查询验证候选边缘，而圆柱查询估计局部间隙场并将其缓存于边缘上。在在线规划过程中，图搜索初始化路线，路径优化在考虑机器人尺寸和地面约束的同时重用缓存的场。实验表明，与测试的基线相比，路网连通性得到了改善，间隙估计更为准确，规划成功率更高。这些结果支持光栅化作为在高斯地图上直接进行规划的有效几何查询接口。代码和数据可在 https://2dgs-planner.github.io/ 获取。

View on arXiv Download PDF AI Translation

cs.RO / 48 / 2610.11764

UltraLight Luma: A Novel Edge-Deployable Perception Network for Crop-Row Segmentation in Agricultural Robotics

超轻量Luma：一种用于农业机器人作物行分割的新型边缘部署感知网络

Balasingham, Dhanushka, Ginigaddarage, Shamod, Rajahrajasingh, Hanojhan, Kodagoda, Nuwan, de Silva, Rajitha

Abstract

Reliable crop-row perception is essential for autonomous agricultural robots, but low-cost deployment is constrained by computation, memory, power, and inference time. This paper addresses crop-row detection and navigation-line extraction using UltraLight Luma, a compact encoder-decoder segmentation network with a progressive learning strategy and only 13.1k trainable parameters. The predicted crop-row masks are processed using a post-processing algorithm to estimate row alignment for downstream visual servoing. UltraLight Luma improves parameter efficiency over U-Net, YOLOv8, and YOLOv26 while maintaining reliable crop-row detection performance. The model required only 25.23 mJ per inference, demonstrating its suitability for resource-constrained agricultural robots.

Chinese Translation

可靠的作物行感知对于自主农业机器人至关重要，但低成本部署受到计算、内存、功耗和推理时间的限制。本文提出了一种使用超轻量Luma的作物行检测和导航线提取方法，该方法是一种紧凑的编码器-解码器分割网络，采用渐进学习策略，仅具有13.1k的可训练参数。预测的作物行掩膜通过后处理算法进行处理，以估计下游视觉伺服的行对齐。超轻量Luma在保持可靠的作物行检测性能的同时，相较于U-Net、YOLOv8和YOLOv26提高了参数效率。该模型每次推理仅需25.23 mJ，证明了其在资源受限的农业机器人中的适用性。

View on arXiv Download PDF AI Translation

cs.RO / 49 / 2610.11771

PathTime-VLA: Path-Time Decoupling for Factorized Post-Training of Vision-Language-Action Policies

PathTime-VLA：面向视觉-语言-动作策略的路径-时间解耦后训练

Huang, Qing, Yang, Yifei, Zou, Ziqing, Chen, Anzhe, Zhu, Zhenjie, Wei, Yufei, Xiong, Rong, Wang, Yue

Abstract

Vision-Language-Action (VLA) policies typically predict actions at fixed time intervals, coupling the route a robot follows with its execution pace. This coupling complicates adaptation from teleoperation: useful geometric guidance comes with timing shaped by interface delays and operator behavior. Our key insight is to bring the path-time parameterization of classical motion planning into the learned action representation of a VLA. We introduce PathTime-VLA, which represents motion as a progress-indexed interaction path $X(s)$ and a positive interval-time profile. The latter defines a monotone time law $t(s)$, yielding controller commands $X(s(t))$. For a given path, alternative executions are expressed through the time profile, allowing chunk-wise speed choices without changing the geometric prediction target. This representation supports a staged post-training procedure: demonstrations and DAgger interventions establish a target-domain prior, Speed-DQN learns execution multipliers from robot interaction, and Path-AWR uses rollout outcomes to refine the diffusion path generator. A path-conditioned action expert realizes the resulting motions while maintaining distinct learning interfaces for path generation and execution timing. Across three tasks, the complete method achieves $58/60$ successes versus $57/60$ for PathTime-VLA under BC + DAgger at fixed $1\times$, with approximately $39$-$52\%$ shorter mean completion times over successful trials.

Chinese Translation

视觉-语言-动作（VLA）策略通常在固定时间间隔内预测动作，将机器人所遵循的路径与其执行速度耦合。这种耦合使得从遥控操作的适应变得复杂：有用的几何指导伴随着由接口延迟和操作员行为塑造的时间。我们的关键见解是将经典运动规划的路径-时间参数化引入到VLA的学习动作表示中。我们提出了PathTime-VLA，它将运动表示为一个进度索引交互路径 $X(s)$ 和一个正的时间间隔轮廓。后者定义了一个单调时间法则 $t(s)$，从而产生控制器命令 $X(s(t))$。对于给定的路径，替代执行通过时间轮廓表达，允许在不改变几何预测目标的情况下进行分块速度选择。这种表示支持分阶段的后训练过程：演示和DAgger干预建立目标领域的先验，Speed-DQN从机器人交互中学习执行乘数，而Path-AWR利用回滚结果来优化扩散路径生成器。一个路径条件的动作专家实现了生成的运动，同时为路径生成和执行时间维持了不同的学习接口。在三个任务中，完整方法在BC + DAgger下以固定 $1 imes$ 达到 $58/60$ 的成功率，而PathTime-VLA则为 $57/60$，成功试验的平均完成时间缩短约 $39$-$52 ext{%}$。

View on arXiv Download PDF AI Translation

cs.RO / 50 / 2610.11809

WAND: Learning Robust Navigation under Complex Wind Disturbances and Dense Obstacles for Quadrotors

WAND：在复杂风干扰和密集障碍物下学习鲁棒导航的四旋翼

Tang, Zhonghan, Li, Chenhui, Liang, Shuai, You, Zhongrui, Li, Jianan, Zhao, Bin, Wang, Zhigang, Li, Xuelong

Abstract

Robust navigation in cluttered environments remains a fundamental challenge for quadrotors, particularly when strong wind disturbances arise, which perturb vehicle dynamics, limit control authority, and substantially increase collision risk. Existing learning-based navigation policies typically rely on obstacle perception and proprioceptive observations, requiring the policy to infer time-varying disturbance effects implicitly and thereby limiting robustness under partial observability. This paper proposes WAND (Wind-Aware Navigation with Disturbance Estimation), a reinforcement learning framework for navigation under time-varying wind disturbances in dense obstacle fields. Specifically, WAND estimates wind-induced disturbance acceleration from historical proprioceptive states using a Temporal Convolutional Network (TCN). This estimation is integrated into the policy via a zero-initialized residual module, \emph{WindAdapter}, while simultaneously providing feedforward compensation for low-level control. The dual use of the estimate couples disturbance-conditioned navigation with feedforward disturbance rejection. Across 12 wind-disturbed simulation settings, WAND improved the observed success rate by 8.3 percentage points on average relative to feedforward compensation alone. Controlled opposite-crosswind experiments further showed wind-direction-dependent trajectory adaptation. In indoor fan-induced flight tests, WAND succeeded in 18 of 20 trials, demonstrating the feasibility of real-time onboard navigation.

Chinese Translation

在杂乱环境中实现鲁棒导航仍然是四旋翼面临的一个基本挑战，尤其是在强风干扰出现时，这会扰动飞行器的动力学，限制控制能力，并显著增加碰撞风险。现有的基于学习的导航策略通常依赖于障碍物感知和本体感知观测，要求策略隐式推断时变干扰效应，从而限制了在部分可观测性下的鲁棒性。本文提出了WAND（带有干扰估计的风感知导航），这是一个用于在时变风干扰下进行导航的强化学习框架，适用于密集障碍物场。具体而言，WAND利用时间卷积网络（Temporal Convolutional Network, TCN）从历史本体感知状态中估计风引起的干扰加速度。该估计通过一个零初始化的残差模块 extit{WindAdapter}集成到策略中，同时为低级控制提供前馈补偿。估计的双重使用将基于干扰的导航与前馈干扰抑制相结合。在12个风干扰的仿真设置中，WAND的观察成功率平均提高了8.3个百分点，相较于仅使用前馈补偿。受控的逆风实验进一步显示了风向依赖的轨迹适应。在室内风扇引起的飞行测试中，WAND在20次试验中成功了18次，证明了实时机载导航的可行性。

View on arXiv Download PDF AI Translation

cs.RO / 51 / 2610.11943

STAG: A Sparse Traversability-Aware Graph Representation from Grid-Based Costmaps for Robotic Navigation

STAG：一种基于网格成本图的稀疏可通行性感知图表示用于机器人导航

Garcia, Gabriel Manuel, Aravecchia, Stéphanie, Olivares-Mendez, Miguel Angel

Abstract

Autonomous rovers navigating large unstructured environments need efficient global planning that accounts for terrain traversability. However, searching dense grid-based costmaps becomes computationally expensive as the mapped area grows. We introduce STAG, a Sparse Traversability-Aware Graph that converts costmaps into compact graphs. STAG combines a medial-axis topological backbone, representative nodes for homogeneous traversability regions, and transition nodes near strong traversability gradients. Edges encode geometry and traversability to account for path length and terrain difficulty. We compare A* on STAG and dense grids using synthetic cave maps, mine maps and the DARPA CERBERUS dataset. Across five benchmark categories comprising 203 map instances and 101,200 queries, STAG reduces median planning time by 3.4x to 9.9x and peak query memory by 2.1x to 15.4x, with median relative path-length differences of -2.9% and +7.6%. STAG offers a compact representation for global planning, trading dense-grid traversability optimality for faster, less memory-intensive search.

Chinese Translation

自主探测车在大型非结构化环境中导航需要高效的全球规划，以考虑地形的可通行性。然而，随着映射区域的扩大，搜索密集的网格成本图变得计算上昂贵。我们提出了STAG（Sparse Traversability-Aware Graph），一种将成本图转换为紧凑图的稀疏可通行性感知图。STAG结合了中轴拓扑骨架、同质可通行性区域的代表节点以及靠近强可通行性梯度的过渡节点。边缘编码几何形状和可通行性，以考虑路径长度和地形难度。我们在STAG和密集网格上使用合成洞穴地图、矿山地图和DARPA CERBERUS数据集进行了A*算法的比较。在包含203个地图实例和101,200个查询的五个基准类别中，STAG将中位规划时间减少了3.4倍至9.9倍，峰值查询内存减少了2.1倍至15.4倍，中位相对路径长度差异为-2.9%和+7.6%。STAG为全球规划提供了一种紧凑的表示，牺牲了密集网格的可通行性最优性，以实现更快且内存占用更少的搜索。

View on arXiv Download PDF AI Translation

cs.RO / 52 / 2610.11945

TACROSS: An Efficient and Low-Cost Scalable Human Touch System Across Heterogeneous Tactile Sensors for Dexterous Robot Learning

TACROSS：一种高效且低成本的可扩展人机触觉系统，适用于异构触觉传感器的灵巧机器人学习

Chen, Bo, Hu, Huanzhang, Ma, Junyang, Yue, Bo, Yu, Fangdi, Chen, Haijier, Lai, Xianxin, Pan, Shuyu, Yang, Zhen, Sun, Xiaoquan, Cui, Wenze, Jiang, Zhongliang, Liu, Shaopeng, Chen, Jiayu

Abstract

Collecting tactile demonstrations on robots is costly and slow, motivating the use of lower-cost human tactile gloves for scalable data collection. However, human capacitive/piezoresistive gloves and robotic tactile sensors differ fundamentally in transduction principle, sensor layout, spatial resolution, and dynamic response, making alignment of raw sensor channels ill-posed. To address this problem, we present TACROSS, a scalable system for learning from human touch and transferring it to robots that bridges this heterogeneity by aligning tactile streams at the level of contact events rather than raw sensor values. The hardware component of TACROSS integrates a piezoresistive glove with five layers and a cost of USD 10.86 with 285 sensing points. To align contact semantics, we design canonicalizers and residual adapters that map heterogeneous signals into a shared tactile latent with 256 dimensions via a temporal Transformer with attention across fingers. We further introduce a robot-grounded policy learning scheme in which robot demonstrations provide the sole source of ground-truth action supervision, while human demonstrations support tactile representation learning and provide confidence-weighted auxiliary supervision through valid retargeted hand targets. We evaluate our system on four contact-rich manipulation tasks. Compared to conventional teleoperation, our proposed system achieves a 3.5-fold efficiency improvement while reducing demonstration acquisition equipment cost by 95.7%. We will open-source the TACROSS hardware and software system and publicly release a tactile dataset comprising over 150 hours of recordings. Project page: https://tacross-touch-project.github.io/.

Chinese Translation

在机器人上收集触觉示范的成本高且速度慢，这促使我们使用低成本的人体触觉手套进行可扩展的数据收集。然而，人类的电容/压阻手套与机器人触觉传感器在传感原理、传感器布局、空间分辨率和动态响应等方面存在根本性差异，使得原始传感器通道的对齐变得困难。为了解决这个问题，我们提出了TACROSS，一个可扩展的系统，用于从人类触觉中学习并将其转移到机器人上，通过在接触事件层面而非原始传感器值层面对触觉流进行对齐，从而弥合这种异质性。TACROSS的硬件组件集成了一种具有五层的压阻手套，成本为10.86美元，具有285个传感点。为了对齐接触语义，我们设计了规范化器和残差适配器，通过一个具有跨手指注意力的时间变换器将异构信号映射到一个256维的共享触觉潜在空间。我们进一步引入了一种基于机器人基础的策略学习方案，其中机器人示范提供唯一的真实动作监督来源，而人类示范则支持触觉表示学习，并通过有效的重新定向手部目标提供置信加权的辅助监督。我们在四个接触丰富的操作任务上评估了我们的系统。与传统的遥操作相比，我们提出的系统实现了3.5倍的效率提升，同时将示范获取设备成本降低了95.7%。我们将开源TACROSS硬件和软件系统，并公开发布一个包含超过150小时录音的触觉数据集。项目页面：https://tacross-touch-project.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 53 / 2610.11952

Tell Robot What Not to Do: A Negation Understanding Perspective

告诉机器人什么不能做：一种否定理解视角

Li, Fazeng, Sun, Gan, Cheng, Hao, Ren, Weihong, Cong, Yang

Abstract

Instruction following enables robots to perform diverse tasks specified in natural language, making it a fundamental capability for human-robot interaction. Beyond communicating desired outcomes, users also need to specify constraints on what not to do. We investigate how to enable vision-language-action models (VLAs) to follow negated instructions, where robots must accomplish task goals while respecting explicit exclusions. To this end, we propose NegaAlign, a parameter-efficient, plug-and-play framework that extends pretrained VLAs to follow negated instructions through image-language supervision alone. Specifically, we introduce Negation Transformation Layers into selected layers of the vision-language backbone to reshape intermediate instruction representations. Meanwhile, a teacher-guided alignment mechanism is designed to align instruction-relevant visual tokens, transferring action-relevant grounding from instructions that satisfy the negated constraint. The training phase uses supervision constructed from existing demonstrations and updates only the inserted layers, keeping all pretrained parameters frozen, including the action generator. We further introduce NegaBench, a simulation benchmark spanning 10 scenarios across five domains for systematically evaluating manipulation under negated constraints. Experiments across GR00T, $\pi_0$, and $\pi_{0.5}$ demonstrate consistent improvements in negated instruction following. With 11.6M trainable parameters, NegaAlign increases the negated-instruction success rate of $\pi_{0.5}$ from 2.60% to 88.45% on NegaBench and from 12.4% to 88.8% on real-world tasks, while retaining performance on affirmative instructions.

Chinese Translation

指令执行使机器人能够完成以自然语言指定的多样化任务，成为人机交互的基础能力。除了传达期望的结果，用户还需要明确规定机器人不能做的事情。我们研究如何使视觉-语言-动作模型（VLAs）能够遵循否定指令，即机器人在实现任务目标的同时尊重明确的排除条件。为此，我们提出了NegaAlign，一种参数高效且即插即用的框架，通过仅利用图像-语言监督扩展预训练的VLAs以遵循否定指令。具体而言，我们在视觉-语言主干网络的选定层中引入否定变换层（Negation Transformation Layers），以重塑中间指令表示。同时，设计了一种教师引导的对齐机制，用于对齐与指令相关的视觉特征，从满足否定约束的指令中转移动作相关的定位信息。训练阶段利用现有示范构建的监督信号，仅更新插入的层，保持所有预训练参数冻结，包括动作生成器。我们进一步引入了NegaBench，一个涵盖五个领域十个场景的仿真基准，用于系统评估在否定约束下的操作能力。基于GR00T、π_0和π_{0.5}的实验表明，NegaAlign在否定指令执行上实现了持续改进。凭借1160万个可训练参数，NegaAlign将π_{0.5}在NegaBench上的否定指令成功率从2.60%提升至88.45%，在真实任务中从12.4%提升至88.8%，同时保持了对肯定指令的性能。

View on arXiv Download PDF AI Translation

cs.RO / 54 / 2610.11956

Reliability-Aware Future Conditioning for Temporally Robust Robot Manipulation

面向可靠性的未来条件化以实现时间鲁棒的机器人操作

Khoshnazar, Mohammad, Tezerjani, Mohammad Dehghani, Gao, Zhiyuan, Qu, Deyuan, Gandyra, Max, Zhan, Yanxiang, Naeem, Mehreen, Melnik, Andrew, Schafer, Jeroen, Yang, Qing, Beetz, Michael

Abstract

A generated video of a task the robot is about to perform is useful guidance only if it depicts the phase the robot is actually in. We show that temporal misalignment can turn a task-consistent generated future into actively harmful guidance. On CALVIN, a five-frame early shift nearly erases the benefit of generated futures, reducing success from 81.3% to 54.8% against 54.0% without futures; imposed timing shifts reduce it even further to 34.2%, 19.8 points below the future-free policy. We introduce Reliability-Aware Future Conditioning (RAFC), which treats this as a control problem rather than a generation problem. At every step, RAFC estimates how far to trust the received clip and which nearby temporal hypothesis to prefer, falling back toward a static branch when neither fits, and it learns both from task reward alone without shift labels or alignment supervision. RAFC sits on top of Future-Experience Conditioning (FEC), which builds the clip once from task grounding, a robot-free digital-twin rollout, and mask-free video diffusion. Under deliberately off-grid phase shifts and rate mismatch, RAFC substantially improves success under temporal mismatch. Candidate ensembling accounts for most of the recovery near alignment, while learned reliability adds a further 7.0 percentage points over uniform averaging of the identical candidate bank under off-grid shifts. The gain holds on the evaluated task sets and survives on a Franka under natural timing mismatch nobody imposed, where aggregate success rises from 26.7% to 56.7%. All resources will be made publicly available. https://future-condition.github.io/.

Chinese Translation

生成的视频任务仅在其描绘机器人实际处于的阶段时才具有指导意义。我们展示了时间错位如何将任务一致的生成未来转变为积极有害的指导。在CALVIN上，五帧的提前偏移几乎抹去了生成未来的好处，使成功率从81.3%降至54.8%，而没有未来的情况下成功率为54.0%；施加的时间偏移甚至进一步降低至34.2%，比无未来策略低19.8个百分点。我们引入了面向可靠性的未来条件化（Reliability-Aware Future Conditioning, RAFC），将其视为控制问题而非生成问题。在每一步，RAFC评估对接收到的片段的信任程度以及优先选择哪个临近的时间假设，当两者都不适合时则回退到静态分支，并且它仅从任务奖励中学习，而无需偏移标签或对齐监督。RAFC建立在未来经验条件化（Future-Experience Conditioning, FEC）之上，该方法一次性从任务基础、无机器人数字双胞胎模拟和无掩码视频扩散构建片段。在故意的非网格相位偏移和速率不匹配下，RAFC在时间不匹配的情况下显著提高了成功率。候选集成在接近对齐时占据了大部分恢复，而学习到的可靠性在非网格偏移下比均匀平均的相同候选库进一步增加了7.0个百分点。该增益在评估的任务集上保持有效，并在自然时间不匹配下的Franka机器人上存活，整体成功率从26.7%上升至56.7%。所有资源将公开提供。https://future-condition.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 55 / 2610.11971

CAPABLE: Capability-Aware Policy Adaptation via Behavioral Latent Encoding

CAPABLE：基于行为潜在编码的能力感知策略适应

Khoshnazar, Mohammad, Tezerjani, Mohammad Dehghani, Qu, Deyuan, Gao, Zhiyuan, Zhan, Yanxiang, Schafer, Jeroen, Melnik, Andrew, Yang, Qing, Beetz, Michael

Abstract

Vision-language-action (VLA) policies assume the embodiment on which they were trained and can fail when a joint fault changes how commanded actions are physically executed. Existing fault-recovery methods often require task-specific retraining, fault labels, explicit diagnosis, or privileged embodiment information. We introduce CAPABLE, a unified capability-aware adaptation framework for frozen VLAs that integrates self-supervised capability inference with residual reinforcement learning. CAPABLE infers capability, how much of the commanded motion each joint actually realizes and how that motion contributes to end-effector behavior, online from command-response history and kinematics using a temporal encoder shared across joints, Jacobian grounding, cross-joint attention, and self-supervised physical prediction. The resulting representation conditions a residual policy that adds bounded corrections to the VLA arm action without fault labels or faulty-joint identifiers. Across 28 LIBERO tasks, CAPABLE raises success on an actuator excluded from fault training from 24.8% to 59.3%, outperforming a parameter-matched global-history baseline by 17.4 points while preserving healthy performance. Leave-one-actuator-out experiments across six joints show that this transfer is not specific to one actuator, and additional evaluations characterize transfer to unseen fault families and demonstrate recovery on a physical Franka Panda. https://capable-vla.github.io/

Chinese Translation

视觉-语言-动作（VLA）策略假设其训练时的体现，并且在联合故障改变指令动作的物理执行方式时可能会失败。现有的故障恢复方法通常需要特定任务的再训练、故障标签、明确的诊断或特权体现信息。我们提出了CAPABLE，一个统一的能力感知适应框架，旨在对冻结的VLA进行集成自监督能力推断与残差强化学习。CAPABLE推断能力，即每个关节实际实现的指令运动量，以及该运动如何贡献于末端效应器行为，在线从指令-响应历史和运动学中进行推断，使用跨关节共享的时间编码器、雅可比基础、跨关节注意力和自监督物理预测。最终的表示条件化一个残差策略，该策略在没有故障标签或故障关节标识符的情况下，为VLA臂动作添加有界修正。在28个LIBERO任务中，CAPABLE将一个未包含在故障训练中的执行器的成功率从24.8%提高到59.3%，比参数匹配的全局历史基线高出17.4个百分点，同时保持健康的性能。在六个关节的留一执行器实验中显示，这种转移并不特定于某一个执行器，额外的评估表征了对未见故障类别的转移，并在物理Franka Panda上展示了恢复能力。

View on arXiv Download PDF AI Translation

cs.RO / 56 / 2610.11975

Predicting Cable Dynamics with Physical Attention Bias

带有物理注意偏差的电缆动态预测

Giuili, Avihai, Atari, Rotem, Sintov, Avishai, Bechler-Speicher, Maya

Abstract

Learned simulators for deformable linear objects (DLOs) such as cables have to predict the motion of cables they were not trained on and stay stable over long rollouts. Most of their error occurs where the cable touches itself or the floor. Attention over all pairs of cable segments can represent contact between parts of the cable that are far apart along its length, but attention has no notion of geometry. A cable has two pairwise distances, which agree only while it is straight: the arc-length distance along the cable, which governs elastic forces, and the Euclidean distance in space, which governs contact. We add a physical attention bias, an additive term on the attention logits with a learned rate, and ask which distance it should use. We compare no bias, each distance alone, and both distances on disjoint sets of heads, keeping the rest of the model and the training protocol fixed. A physical bias improves prediction on unseen cables. The gain is largest when attention is the only mechanism that connects distant segments: there, the arc-length bias reduces prediction error by 15% and more than halves the drift in segment length. The Euclidean bias alone stays close to unbiased attention, while assigning both distances across heads is best or near-best on every metric we report. Code and per-run records: https://github.com/avihaig/dlogps.

Chinese Translation

用于可变形线性物体（DLOs）如电缆的学习模拟器必须能够预测未经过训练的电缆的运动，并在长时间的模拟中保持稳定。它们的大部分误差发生在电缆相互接触或与地面接触的地方。对所有电缆段对的注意力可以表示电缆的远距离部分之间的接触，但注意力并没有几何概念。电缆有两个成对的距离，只有在电缆是直的情况下才一致：沿电缆的弧长距离，控制弹性力，以及空间中的欧几里得距离，控制接触。我们增加了一种物理注意偏差，即在注意力逻辑上添加一个学习率的附加项，并询问应该使用哪个距离。我们比较了无偏差、单独使用每个距离以及在不相交的头部上同时使用这两种距离，保持模型的其余部分和训练协议不变。物理偏差提高了对未见电缆的预测。当注意力是连接远距离段的唯一机制时，增益最大：在这种情况下，弧长偏差将预测误差降低了15%，并且将段长度的漂移减少了一半以上。单独的欧几里得偏差与无偏注意力保持接近，而在各个头部分配这两种距离在我们报告的每个指标上都是最佳或接近最佳的。代码和每次运行的记录： https://github.com/avihaig/dlogps.

View on arXiv Download PDF AI Translation

cs.RO / 57 / 2610.12007

REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models

REACT：基于滚动去噪和双重解耦的反应式机器人控制与VLA模型

Xiong, Houlong, Qiu, Zhenqi, Wang, Zechen, Zhang, Suohang, Ren, Yiyu, Xu, Wanting, Niu, Hongfei, He, Chengyang, Sun, Ge, Cheng, Ran, Zhu, Qian

Abstract

Flow-based vision-language-action (VLA) models generate action chunks for temporally coherent robot motion, but chunked control creates a fundamental closed-loop trade-off: long chunks provide smooth execution, whereas frequent replanning improves reactivity at the cost of action discontinuities. We introduce REACT, a rolling-denoising framework that makes flow-based VLAs more reactive while preserving long-horizon context. Instead of regenerating entire action chunks from scratch, REACT maintains a persistent action buffer with staggered flow timesteps. At each control step, the full horizon is denoised using the latest observation, the cleanest action block is executed, partially refined future blocks are shifted forward, and fresh noise is appended to the tail. As a result, each executed action block is refined across multiple recent observations before deployment. To support real-time control, we further introduce dual decoupling, which separates sensing, VLM encoding, DiT denoising, and action execution, enabling high-frequency observation updates and action streaming under practical compute constraints. Across the RoboTwin 2.0 simulation benchmark and real-world tasks spanning bimanual manipulation and dynamic control on multiple robot platforms, REACT improves task success and reduces reaction latency while producing smoother trajectories than frequent-replanning and asynchronous baselines.

Chinese Translation

基于流的视觉-语言-动作（VLA）模型生成用于时间一致的机器人运动的动作块，但块状控制产生了一个基本的闭环权衡：长块提供平滑的执行，而频繁的重新规划则在牺牲动作连续性的情况下提高了反应性。我们提出了REACT，一个滚动去噪框架，使基于流的VLA在保持长时间范围上下文的同时更加反应灵敏。REACT并不是从头开始重新生成整个动作块，而是维护一个具有错开流时间步的持久动作缓冲区。在每个控制步骤中，使用最新观察对整个时间范围进行去噪，执行最干净的动作块，部分精炼的未来块向前移动，并将新噪声附加到尾部。因此，每个执行的动作块在部署之前会在多个最近的观察中进行精炼。为了支持实时控制，我们进一步引入了双重解耦，分离了感知、VLM编码、DiT去噪和动作执行，使得在实际计算约束下能够高频率地更新观察和流式执行动作。在RoboTwin 2.0仿真基准和涵盖双手操作及多个机器人平台动态控制的真实任务中，REACT提高了任务成功率，减少了反应延迟，同时产生比频繁重新规划和异步基线更平滑的轨迹。

View on arXiv Download PDF AI Translation

cs.RO / 58 / 2610.12026

Humanoid World Action Model With Joint State--Action Generation

具有关节状态-动作生成的人形世界动作模型

Yang, Yan, Rong, Jikun, Zhu, Minzhao, Zhao, Zheyi, Hu, Qirui, Lan, Zihan, Mao, Weixin, Li, Yinhao, Fu, Zhen, Chen, Hua

Abstract

Humanoid robots are a promising platform for general-purpose manipulation. Recent Vision-Language-Action (VLA) policies learn actions directly from multimodal observations, while World Action Models (WAMs) further incorporate future visual prediction to improve action generation. However, in hierarchical humanoid systems, VLA and WAM policies output reference actions that are subsequently realized through whole-body control, robot dynamics, balance, and contact. This hierarchy creates an action--execution gap: the reference produced by the policy can differ from the motion realized by the robot. Without explicitly modeling the realized body state, future visual prediction must jointly explain scene evolution and discrepancies between reference actions and executed motion, making it difficult to associate an action with its physical outcome. We propose HWAM, a Humanoid World Action Model with joint state--action generation, which makes the robot's post-execution proprioceptive state an explicit prediction target. By jointly generating reference actions and their realized body states, HWAM directly incorporates supervision of executed motion into action learning. HWAM is trained through three complementary conditional paths. The Policy path jointly denoises state--action trajectories conditioned only on current observations, matching deployment conditions. Forward Dynamics Modeling (FDM) predicts future visual observations conditioned on actions and post-execution states, while Inverse Dynamics Modeling (IDM) reconstructs the joint trajectory from visual transitions. Together, these paths connect policy references, realized body motion, and visual outcomes. HWAM achieves the highest success rate among evaluated baselines on three real-robot tasks on the LimX OLI humanoid. On Candy Picking, HWAM achieves a 70.6% success rate, compared with 43.3% for Fast-WAM.

Chinese Translation

人形机器人是通用操作的有前景平台。最近的视觉-语言-动作（VLA）策略直接从多模态观察中学习动作，而世界动作模型（WAMs）进一步结合未来视觉预测以改善动作生成。然而，在分层人形系统中，VLA和WAM策略输出的参考动作随后通过全身控制、机器人动力学、平衡和接触来实现。这种层次结构造成了动作-执行差距：策略产生的参考可能与机器人实现的运动不同。如果没有明确建模实现的身体状态，未来视觉预测必须共同解释场景演变以及参考动作与执行运动之间的差异，这使得将动作与其物理结果关联变得困难。我们提出了HWAM，一种具有联合状态-动作生成的人形世界动作模型，它将机器人的执行后本体感觉状态作为明确的预测目标。通过联合生成参考动作及其实现的身体状态，HWAM将执行运动的监督直接纳入动作学习中。HWAM通过三条互补的条件路径进行训练。策略路径仅基于当前观察共同去噪状态-动作轨迹，以匹配部署条件。前向动力学建模（FDM）预测基于动作和执行后状态的未来视觉观察，而逆向动力学建模（IDM）从视觉转变中重建关节轨迹。这些路径共同连接策略参考、实现的身体运动和视觉结果。在LimX OLI人形机器人上，HWAM在评估的基线中实现了最高的成功率。在糖果采摘任务中，HWAM的成功率为70.6%，而Fast-WAM为43.3%。

View on arXiv Download PDF AI Translation

cs.RO / 59 / 2610.12033

Traceable World State: A Provenance-Aware State Representation and Deterministic Replay Framework for Robotic Systems

可追溯的世界状态：一种面向来源的状态表示和机器人系统的确定性重放框架

Li, Zoe

Abstract

Robotic systems operating over extended tasks must maintain a world state assembled from observations arriving at different times, with varying confidence and potential revisions. Conventional representations emphasize latest estimates, hindering fact provenance, decision reproduction, or execution auditing. We present Traceable World State (TWS), a middleware-neutral semantic representation and reference runtime for provenance-aware robot world state. A TWS snapshot captures entities, relations, observations, confidence, and revision metadata. Validated update operations transform snapshots immutably, ordered updates support deterministic replay, and a canonical SHA-256 hash chain ensures tamper-evident logs. We evaluate TWS through schema conformance, complete state lifecycles, deterministic replay, and fault injection. Passing 38 tests across Python 3.10-3.14, the framework detects record corruptions, broken hash links, sequence discontinuities, and world mismatches. Across ten public BEHAVIOR-1K task definitions, TWS imported 153 entities and 146 relations with successful validation. On 103 NVIDIA Unitree G1 simulated trajectories containing 78,369 frames, TWS achieved exact terminal-state replay in all episodes and detected 412/412 injected corruptions with a 1.72% storage overhead over Plain JSONL.

Chinese Translation

在执行扩展任务的机器人系统中，必须维护一个由不同时间到达的观察结果组装而成的世界状态，这些观察结果具有不同的置信度和潜在的修订。传统的表示方法强调最新的估计，阻碍了事实来源的追踪、决策的重现或执行的审计。我们提出了可追溯的世界状态（Traceable World State, TWS），这是一种中立于中间件的语义表示和面向来源的机器人世界状态的参考运行时。TWS快照捕获实体、关系、观察结果、置信度和修订元数据。经过验证的更新操作以不可变的方式转换快照，有序更新支持确定性重放，而规范的SHA-256哈希链确保了可篡改的日志。我们通过模式一致性、完整的状态生命周期、确定性重放和故障注入对TWS进行了评估。在Python 3.10-3.14版本中通过了38项测试，该框架能够检测记录损坏、哈希链接断裂、序列不连续和世界不匹配。在十个公共BEHAVIOR-1K任务定义中，TWS成功导入了153个实体和146个关系，并完成了验证。在103个包含78,369帧的NVIDIA Unitree G1模拟轨迹中，TWS在所有剧集中的终端状态重放均实现了精确，并检测到412/412个注入的损坏，存储开销为Plain JSONL的1.72%.

View on arXiv Download PDF AI Translation

cs.RO / 60 / 2610.12046

SkillWeave: Weaving Heterogeneous Demonstrations into Long-Horizon Manipulation Skills

SkillWeave：将异构示范编织成长时间操作技能

Tamura, Ryosei, Dong, Xiaoxiang, Yoo, Uksang, Mao, Yuemin, Mir, Romina, Francis, Jonathan, Ichnowski, Jeffrey

Abstract

Dexterous manipulation requires both large-scale task progression and precise contact-rich interaction, making it challenging to collect demonstrations that effectively support both regimes. We present SkillWeave, a heterogeneous demonstration framework for long-horizon dexterous manipulation that combines teleoperation for coarse reaching and transport with kinesthetic teaching for precise, contact-rich skills. To address the visual mismatch introduced by the demonstrator's presence during kinesthetic data collection, we propose an object-mask-conditioned diffusion policy that uses offline object segmentation for training supervision and a lightweight learned mask predictor at deployment, avoiding online segmentation and image inpainting. To mitigate distribution shift between independently trained sub-task policies, we introduce successor-aware terminal steering, which selects among actions sampled from the predecessor policy to guide the system toward states supported by the successor's demonstrated initial-state distribution. Across three real-world long-horizon tasks, SkillWeave achieves 27% average end-to-end success. Mask-conditioned kinesthetic policies improve dexterous sub-task success to an average of 65%, while successor-aware handoffs achieve an average composition efficiency of 87%. These results show that matching demonstration modality to interaction regime, explicitly addressing kinesthetic visual mismatch, and steering policy handoffs toward successor-supported states substantially improves long-horizon dexterous manipulation. Videos and code are available at skillweave-authors.github.io .

Chinese Translation

灵巧操作需要大规模的任务进展和精确的接触丰富交互，这使得收集有效支持这两种模式的示范变得具有挑战性。我们提出了SkillWeave，一个用于长时间灵巧操作的异构示范框架，它结合了用于粗略到达和运输的远程操作与用于精确、接触丰富技能的动觉教学。为了应对动觉数据收集过程中示范者存在所引入的视觉不匹配，我们提出了一种基于物体掩膜的扩散策略，该策略在训练监督中使用离线物体分割，并在部署时使用轻量级学习的掩膜预测器，从而避免了在线分割和图像修复。为了减轻独立训练的子任务策略之间的分布偏移，我们引入了后继感知终端引导，该方法从前驱策略中采样的动作中进行选择，以引导系统朝向由后继示范的初始状态分布所支持的状态。在三个真实世界的长时间任务中，SkillWeave实现了27%的平均端到端成功率。基于掩膜的动觉策略将灵巧子任务的成功率提高到平均65%，而后继感知交接的平均组合效率达到87%。这些结果表明，将示范模式与交互模式相匹配，明确解决动觉视觉不匹配，并将策略交接引导至后继支持的状态，显著改善了长时间灵巧操作。视频和代码可在skillweave-authors.github.io获取。

View on arXiv Download PDF AI Translation

cs.RO / 61 / 2610.12073

A Minimal Optical-Flow Representation for Vision-Based Tactile Rotation Classification in Robotic Manipulation Across Gravity Domains

一种用于机器人操作中基于视觉的触觉旋转分类的最小光流表示法，适用于不同重力环境

Martinez-Bernal, Oscar, Cavero-Vidal, Mario, Grella, Francesco, Martinez, Carol

Abstract

Vision-based tactile sensors provide rich contact information, but processing high-resolution images can be costly for resource-constrained platforms such as space robots. This work investigates whether a compact representation of tactile motion can classify object rotation across different gravity conditions. Dense optical flow from a simulated GelSight Mini is aggregated over a 7x9 grid into 126 features and used to classify the direction of load-induced rotation under Earth, Mars, Moon, and orbital gravity. Gravity causes a small but significant shift in these features, accounting for 1.6% of their variance (R2 = 0.016). Despite its small magnitude, this shift affects models trained only on Earth data: XGBoost accuracy decreases from 94.4% on Earth to 75.9% in orbit. In contrast, a single model trained across all four gravity domains achieves 96.3% overall accuracy and 95.1%-97.0% across individual domains, without using gravity as an input. The representation can also be reduced to 40 features while retaining 95.7% accuracy, with XGBoost requiring only 0.14 ms per inference. These findings show that Earth-gravity performance alone is insufficient to establish the transferability of tactile perception for space robotic manipulation, highlighting the need to account for gravity-induced domain shifts during training and validation.

Chinese Translation

基于视觉的触觉传感器提供丰富的接触信息，但处理高分辨率图像对于资源受限的平台（如太空机器人）可能代价高昂。本研究探讨了一种紧凑的触觉运动表示法是否能够在不同重力条件下对物体旋转进行分类。来自模拟GelSight Mini的密集光流被聚合到一个7x9的网格中，形成126个特征，并用于分类在地球、火星、月球和轨道重力下由负载引起的旋转方向。重力导致这些特征发生小但显著的偏移，占其方差的1.6%（R2 = 0.016）。尽管这一偏移幅度较小，但它影响了仅基于地球数据训练的模型：XGBoost的准确率从地球上的94.4%降至轨道上的75.9%。相比之下，跨越所有四个重力领域训练的单一模型实现了96.3%的整体准确率，并在各个领域的准确率为95.1%-97.0%，且未使用重力作为输入。该表示法还可以减少到40个特征，同时保持95.7%的准确率，XGBoost每次推理仅需0.14毫秒。这些发现表明，仅依靠地球重力性能不足以确立触觉感知在太空机器人操作中的可转移性，强调了在训练和验证过程中需要考虑重力引起的领域偏移。

View on arXiv Download PDF AI Translation

cs.RO / 62 / 2610.12089

ManiUnit: A Manipulation Skill Dataset and Benchmark for Long-Horizon Tasks

ManiUnit：一个用于长时间任务的操作技能数据集和基准测试

Wei, Guoting, Yan, Dawei, Yuan, Xia, Zhang, Gengming, He, Yelin, Du, Guodong, Ye, Jiaquan, Zhang, Heng, Wei, Xinming, Qi, Xianbiao, Zhao, Chunxia, Zhang, Haokui, Xiao, Rong

Abstract

Long-horizon mobile manipulation requires a robot to navigate multi-room environments and execute a sequence of manipulation skills under a single natural language instruction. Learning and evaluating these skills present three challenges: similar observations under a fixed task instruction may make skill selection ambiguous; even when a preceding skill succeeds, the robot state inherited by the next skill may deviate from its demonstrated starting states and affect execution; and task-level metrics hinder skill-specific diagnosis, while early failures leave later skills untested. We therefore introduce ManiUnit, a manipulation skill dataset and benchmark built from 50 BEHAVIOR-1K activities. Its dataset contains 137,899 segments across 21 skill types and 417 subtasks, and its benchmark contains 1,260 test instances. Correspondingly, ManiUnit pairs each segment with an explicit subtask instruction; measures sensitivity to perturbations of the robot's starting base position or joint configuration; and restores intermediate simulator states and defines local success conditions so that each skill can be evaluated without executing preceding stages. Evaluations of representative vision-language-action (VLA) policies show that similar aggregate scores can hide substantial per-skill differences. The tested starting-state perturbations also degrade execution: on the full benchmark, joint perturbations reduce success rates by approximately 56% relative to those from demonstrated starting states. On two long-horizon activities, a skill policy trained on ManiUnit segments achieves 78.7% local manipulation success, compared with 49.3% for a task policy trained on complete demonstrations. The trained skills further support complete-task execution on these activities, as coordinating the task and skill policies through a planner raises full-task success from 4.0% to 18.0%.

Chinese Translation

长时间移动操作要求机器人在多房间环境中导航，并在单一自然语言指令下执行一系列操作技能。学习和评估这些技能面临三个挑战：在固定任务指令下，相似的观察可能使技能选择变得模糊；即使前一个技能成功，继承给下一个技能的机器人状态也可能偏离其演示的起始状态，从而影响执行；而任务级别的指标则妨碍了技能特定的诊断，同时早期失败使得后续技能未能得到测试。因此，我们引入了ManiUnit，一个基于50个BEHAVIOR-1K活动构建的操作技能数据集和基准测试。该数据集包含137,899个片段，涵盖21种技能类型和417个子任务，其基准测试包含1,260个测试实例。相应地，ManiUnit将每个片段与明确的子任务指令配对；测量对机器人起始基座位置或关节配置扰动的敏感性；并恢复中间模拟器状态，定义局部成功条件，以便在不执行前置阶段的情况下评估每项技能。对代表性视觉-语言-动作（VLA）策略的评估表明，相似的总体得分可能掩盖显著的每项技能差异。测试的起始状态扰动也降低了执行效果：在完整基准测试中，关节扰动使成功率相较于演示起始状态降低了约56%。在两个长时间活动中，基于ManiUnit片段训练的技能策略实现了78.7%的局部操作成功率，而基于完整演示训练的任务策略仅为49.3%。训练的技能进一步支持在这些活动中的完整任务执行，因为通过规划者协调任务和技能策略使完整任务的成功率从4.0%提升至18.0%。

View on arXiv Download PDF AI Translation

cs.RO / 63 / 2610.12099

UNITAS: A 3D-Native World Action Model for Embodied Manipulation

UNITAS：一种用于具身操作的3D原生世界动作模型

Wang, Ruixiang, Su, Yongyi, Zhou, Wenlve, Yue, Bo, Liu, Hengyan, Lu, Dekun, Tian, Yuxin, Fang, Yihan, Wu, Zerui, Hu, Xing, Chen, Jietao, Guo, Yong, He, Ziyan, Yuan, Junbin, Liu, Guiliang, Xing, Xiaofen, Jia, Kui

Abstract

World action models (WAMs) aim to answer a coupled physical question: given a task instruction, what motion should the robot execute, and how will that motion change the surrounding world? Most existing WAMs build on pretrained video generators and represent world evolution through images or visual latents. Robotic interaction, however, takes place in metric three-dimensional space, while images are view-dependent projections whose pixel distances do not directly encode physical distances. We introduce UNITAS, to our knowledge the first 3D-native world action model that unifies observations, actions, and scene dynamics in a shared metric 3D frame within each interaction, using a common representation across robot embodiments and human hands. Action flow represents human hands and robot grippers as 3D point trajectories, while scene flow describes scene-point displacements conditioned on these trajectories. World-aligned 3D positional embeddings ground visual tokens with or without depth input, and a physical-time trajectory tokenizer encodes each point trajectory as one token anchored at its current 3D position. This interface supports both direct action execution and action-conditioned scene prediction. With 1.7B parameters, UNITAS achieves the best action-conditioned scene prediction on RoboTwin among the compared methods, with up to 49% lower displacement errors than PointWorld, and state-of-the-art manipulation success, including 99.8% on LIBERO and an average of 85% across real-world tasks. The code is available at https://github.com/DexForce/UNITAS.

Chinese Translation

世界动作模型（WAMs）旨在回答一个耦合的物理问题：给定任务指令，机器人应执行何种动作，以及该动作将如何改变周围的世界？大多数现有的WAMs基于预训练的视频生成器，通过图像或视觉潜变量表示世界的演变。然而，机器人交互发生在度量三维空间中，而图像是视角依赖的投影，其像素距离并不能直接编码物理距离。我们介绍了UNITAS，作为我们所知的第一个3D原生世界动作模型，它在每次交互中使用共享的度量3D框架统一观察、动作和场景动态，采用跨机器人具身和人类手部的共同表示。动作流将人类手部和机器人夹具表示为3D点轨迹，而场景流描述基于这些轨迹的场景点位移。世界对齐的3D位置嵌入将视觉标记与有或没有深度输入的情况结合起来，物理时间轨迹标记器将每个点轨迹编码为一个锚定在其当前3D位置的标记。该接口支持直接动作执行和基于动作的场景预测。UNITAS拥有17亿个参数，在比较的方法中，在RoboTwin上实现了最佳的基于动作的场景预测，位移误差比PointWorld低达49%，并在操作成功率上达到了最先进的水平，包括LIBERO上的99.8%和在真实世界任务中的平均85%。代码可在https://github.com/DexForce/UNITAS获取。

View on arXiv Download PDF AI Translation

cs.RO / 64 / 2610.12123

PIER: An Evidence-Gated Execution Interface for Robotic Manipulation

PIER：一种用于机器人操作的证据门控执行接口

Li, Zoe, Wang, Anze, Chen, Zhongyu, Hu, Jingran

Abstract

Generating a plausible robot action does not establish that current observations justify its execution. Motivated by exploratory observations of high-confidence visual outputs under severe occlusion, we present PIER, an execution-authorization interface that separates evidence checks, decision provenance, and stage-scoped re-observation from hardware control. The deterministic gate evaluates declared visual and tactile inputs, while its caller maintains a budget of at most one re-observation per stage. We evaluate the implementation using 1,600 threshold-grid cases and 1,200 paired synthetic traces spanning score noise, missing tactile inputs, stale observations, and falsely reassuring scores. The finite grid yields zero declared invariant violations, and a matched Boolean baseline reproduces all non-recovery decisions. Under synthetic score noise, re-observation reduces valid-state denials from 57/120 to 18/120 while increasing invalid-state proceeds from 5/120 to 7/120. Stale and falsely reassuring inputs expose limitations that threshold checks alone cannot resolve. Exploratory visual, tactile, and robot setup records provide context but do not establish physical task performance. These results characterize an inspectable authorization interface and its input-contract limitations, without claiming superiority over equivalent rule logic, calibrated tactile accuracy, or certified physical safety.

Chinese Translation

生成一个合理的机器人动作并不能证明当前的观察结果足以支持其执行。基于在严重遮挡下高置信度视觉输出的探索性观察，我们提出了PIER，这是一种执行授权接口，它将证据检查、决策来源和阶段范围内的重新观察与硬件控制分离。确定性门控评估声明的视觉和触觉输入，而其调用者在每个阶段最多保持一个重新观察的预算。我们使用1600个阈值网格案例和1200个配对合成轨迹对实现进行了评估，这些案例涵盖了评分噪声、缺失的触觉输入、过时的观察和虚假的安慰评分。有限的网格没有产生任何声明的不变性违反，而匹配的布尔基线重现了所有非恢复决策。在合成评分噪声下，重新观察将有效状态拒绝的比例从57/120降低到18/120，同时将无效状态的继续比例从5/120提高到7/120。过时和虚假的安慰输入暴露了仅依靠阈值检查无法解决的局限性。探索性的视觉、触觉和机器人设置记录提供了背景，但并未确立物理任务的表现。这些结果描述了一个可检查的授权接口及其输入合同的局限性，而不声称优于等效的规则逻辑、校准的触觉准确性或认证的物理安全性。

View on arXiv Download PDF AI Translation

cs.RO / 65 / 2610.12126

SuperNav: An Agentic Navigation System for Any Task in Any Scene

SuperNav：适用于任何任务和场景的自主导航系统

Zhang, Jinkai, Xu, Jingyi, Yu, Yuanhong, Guo, Jiarui, Hu, Ruizhen, Bao, Hujun, Zhou, Xiaowei, Peng, Sida

Abstract

General-purpose service robots need navigation systems that can handle diverse human requests in unfamiliar environments, combining task generality with scene generality. Some existing methods fine-tune multimodal large language models (MLLMs) to predict navigation actions, making their behavior dependent on the coverage of navigation training data and potentially limiting generalization to new requests and environments. Our key insight is to let the MLLM focus on interpreting requests, understanding scenes, and making decisions while preserving its general-purpose capabilities and delegating motion execution to navigation tools. To realize this idea, we introduce SuperNav, which equips a pretrained MLLM with a specialized agent harness without navigation-specific fine-tuning of the MLLM. Our harness supports these decisions with Navigation Skills, agent-oriented Tools for physical interaction, and task-progress and context management. A unified visual-point interface connects decision-making to motion by allowing the model to specify destinations directly in images and revise its decisions from execution feedback. Together, these components support sustained navigation across different task requirements and environments. SuperNav outperforms four evaluated baselines on instance-level, multi-object, and demand-driven tasks. Category-level evaluation on HM3D and deployment on a real quadruped robot further demonstrate its applicability across environments. Project Page: https://zju3dv.github.io/SuperNav/

Chinese Translation

通用服务机器人需要能够处理陌生环境中多样化人类请求的导航系统，结合任务通用性和场景通用性。一些现有方法对多模态大型语言模型（MLLM）进行微调，以预测导航动作，使其行为依赖于导航训练数据的覆盖范围，并可能限制对新请求和环境的泛化。我们的关键见解是让MLLM专注于解释请求、理解场景和做出决策，同时保持其通用能力，并将运动执行委托给导航工具。为实现这一理念，我们引入了SuperNav，它为预训练的MLLM配备了一个专门的代理工具，而无需对MLLM进行特定于导航的微调。我们的工具通过导航技能、面向代理的物理交互工具以及任务进展和上下文管理来支持这些决策。一个统一的视觉点接口通过允许模型直接在图像中指定目的地并根据执行反馈修正决策，将决策与运动连接起来。这些组件共同支持在不同任务需求和环境中的持续导航。SuperNav在实例级、多对象和需求驱动任务上超越了四个评估基线。在HM3D上的类别级评估和在真实四足机器人上的部署进一步证明了其在不同环境中的适用性。项目页面：https://zju3dv.github.io/SuperNav/

View on arXiv Download PDF AI Translation

cs.RO / 66 / 2610.12140

Leveraging Human-In-The-Loop Demonstrations in Reinforcement Learning for Digital Twin-Driven Robot Flexibility

利用人机协作示范在数字双胞胎驱动的机器人灵活性强化学习中的应用

Sun, Yuzhu, Van, Mien, Nhat, Nguyen Minh, McIlvanna, Stephen, McLoone, Sean

Abstract

Growing automation makes collaborative robots work in more variable environments, increasing the need for adaptation. We propose a human-in-the-loop online training framework combining a digital twin (DT), reinforcement learning (RL), and human demonstrations. Unlike DTs used mainly to generate synthetic data before task execution, our DT is synchronized with the physical system in real time through camera feeds, allowing the virtual robot to update its observations and policy from real-world feedback. A dual actor framework integrates imitation learning (IL) without adding a direct imitation loss to the RL actor, so demonstrations can guide adaptation instead of manual reprogramming. The proposed framework is demonstrated on the Ufactory Xarm5 collaborative robot, where the robot's end-effector aims to reach the target position while avoiding obstacles. The experiments show that the framework can resume training after a change in the physical workspace and that, with a fixed set of non-optimal demonstrations, the dual actor framework achieves a much higher final success rate than two methods that add an imitation loss to the actor. The same pattern holds with real human demonstrations collected in virtual reality (VR): with demonstrations that never reach the goal, the dual actor framework reached 83-100% mean deterministic evaluation success, against 0-17% for the two imitation-loss methods.

Chinese Translation

日益增长的自动化使得协作机器人在更具变动性的环境中工作，增加了适应的需求。我们提出了一种人机协作的在线训练框架，结合了数字双胞胎（Digital Twin, DT）、强化学习（Reinforcement Learning, RL）和人类示范。与主要用于在任务执行前生成合成数据的数字双胞胎不同，我们的数字双胞胎通过摄像头实时与物理系统同步，允许虚拟机器人根据现实世界的反馈更新其观察和策略。双重演员框架集成了模仿学习（Imitation Learning, IL），而无需向强化学习演员添加直接的模仿损失，因此示范可以指导适应，而不是手动重新编程。该框架在Ufactory Xarm5协作机器人上进行了验证，机器人末端执行器的目标是到达目标位置，同时避免障碍物。实验表明，该框架可以在物理工作空间发生变化后恢复训练，并且在一组固定的非最优示范下，双重演员框架的最终成功率远高于两种向演员添加模仿损失的方法。相同的模式也适用于在虚拟现实（Virtual Reality, VR）中收集的真实人类示范：在示范从未达到目标的情况下，双重演员框架的平均确定性评估成功率达到了83-100%，而两种模仿损失方法的成功率仅为0-17%。

View on arXiv Download PDF AI Translation

cs.RO / 67 / 2610.12157

Instance-anchored interaction evidence: Grounding robot plans in human pointing and handling

实例锚定交互证据：将机器人计划与人类指向和处理行为结合

Xiao, Xinliang, Yang, Bowen, Zhang, Wenjing, Yang, Li, Zhou, Wei

Abstract

A robot that assists people must often act on what a person has shown rather than said: which of several identical cartons was pointed at, or which box was handled. The plan is executed from the final scene, whereas the evidence occurs earlier, possibly on objects that have since moved. We propose instance-anchored interaction evidence (IAE), which registers every object of the final scene to its public identifier, keeps each identity through the video by backward mask propagation, and describes every frame by the geometry between hands, forearms and these instances. An evidence network trained only from task outcomes scores the instances. For pointing tasks, a grammar-constrained dynamic program trained with a structured loss decodes object-destination programs; symbolic programs handle reference disambiguation and, without learning, episodic tasks. On 1,255 WatchAct benchmark requests, scored by symbolic execution, IAE reaches 64.2% plan success on implicit-intent tasks against 27.5% for a 32B vision-language model (strict success 49.7% against 15.4%), and 46.4% against 27.0% on restoration, reversal and imitation without task-specific training. Controls with the same perception overlays, the same 32 frames, forward tracking, or a relation model trained on the same labels do not explain the gain. Given IAE's evidence as text with its meaning explained, the same language model reaches 57.4%: most of the gain comes from the instance-anchored evidence, and the explicit programs add 6.8 points at a fraction of the cost. Pointing remains the hardest case, with 16.9% strict success. The code is available at https://github.com/WeiZhou96/iae-watchact.

Chinese Translation

一个辅助人类的机器人往往必须根据人们所指示的内容而非所说的内容采取行动：例如，指向了几个相同纸箱中的哪个，或者处理了哪个盒子。计划是从最终场景执行的，而证据则发生在更早的阶段，可能涉及已经移动的物体。我们提出了实例锚定交互证据（Instance-anchored interaction evidence, IAE），该方法将最终场景中的每个物体注册到其公共标识符，利用向后掩膜传播保持每个身份，并通过手、前臂与这些实例之间的几何关系描述每一帧。一个仅从任务结果训练的证据网络对实例进行评分。在指向任务中，使用结构化损失训练的语法约束动态程序解码物体-目的地程序；符号程序处理引用消歧，并且在没有学习的情况下处理情节任务。在1,255个WatchAct基准请求中，通过符号执行评分，IAE在隐含意图任务上达到64.2%的计划成功率，而32B视觉-语言模型的成功率为27.5%（严格成功率为49.7%对比15.4%），在恢复、反转和模仿任务上达到46.4%对比27.0%，且没有特定任务的训练。与相同感知叠加、相同32帧、前向跟踪或在相同标签上训练的关系模型的对照实验无法解释这一增益。考虑到IAE的证据作为文本并解释其含义，相同的语言模型达到57.4%：大部分增益来自实例锚定证据，而显式程序在较低成本下增加了6.8个百分点。指向任务仍然是最困难的情况，严格成功率为16.9%。代码可在 https://github.com/WeiZhou96/iae-watchact 获取。

View on arXiv Download PDF AI Translation

cs.RO / 68 / 2610.12172

Unifying Policy Learning and State Prediction through Spatial Language Modeling

通过空间语言建模统一策略学习与状态预测

Wu, Minye, Wang, Zehao, Tuytelaars, Tinne

Abstract

Learning how actions change scene geometry can provide complementary supervision for goal-directed manipulation. We introduce Spatial Language Modeling, which represents scene contours, goals, action targets, and future states with a shared vocabulary of discrete coordinates and semantic tokens. A task-specific grammar organizes these elements into spatial sequences, allowing one autoregressive Transformer to learn action generation and action-conditioned state prediction through a common next-token objective. We train the model from scratch using random-play transition pretraining followed by joint action and state training on expert demonstrations. During pretraining, recorded action coordinates condition subsequent state predictions and are excluded from the prediction loss. During control, the model decodes only executable action targets and updates its history with newly observed states. We evaluate the approach on Push-T in simulation and on a real robot. The model achieves competitive simulation performance and higher task success and target coverage than the evaluated real-robot policy baselines. Training ablations show improved control with joint action and state sequences, with further gains from random-play pretraining. Given supplied action trajectories, the same model also predicts successive scene states, capturing the geometric effects of pushing.

Chinese Translation

学习动作如何改变场景几何可以为目标导向的操作提供补充监督。我们引入了空间语言建模（Spatial Language Modeling），该方法使用离散坐标和语义标记的共享词汇来表示场景轮廓、目标、动作目标和未来状态。任务特定的语法将这些元素组织成空间序列，使得一个自回归的变换器（Transformer）能够通过共同的下一个标记目标学习动作生成和基于动作的状态预测。我们从头开始训练模型，采用随机播放过渡预训练，随后在专家演示上进行联合动作和状态训练。在预训练期间，记录的动作坐标条件后续的状态预测，并从预测损失中排除。在控制过程中，模型仅解码可执行的动作目标，并用新观察到的状态更新其历史记录。我们在Push-T仿真和真实机器人上评估该方法。模型在仿真中实现了具有竞争力的性能，并且在任务成功率和目标覆盖率上超过了评估的真实机器人策略基线。训练消融实验表明，联合动作和状态序列改善了控制效果，随机播放预训练进一步提升了性能。给定提供的动作轨迹，同一模型还可以预测连续的场景状态，捕捉推动的几何效应。

View on arXiv Download PDF AI Translation

cs.RO / 69 / 2610.12185

RESETTLE: Robotic Recovery through Disagreement-Triggered Retrieval and Efficient Corrective Control

RESETTLE：通过分歧触发的检索与高效纠正控制实现机器人恢复

Chen, Yuxin, Yang, Senqiao, Wang, Zixuan, Ye, Jinhui, Lu, Changsheng, Chen, Pengguang, Liu, Shu, Tian, Zhuotao, Jia, Jiaya

Abstract

Reliable robotic manipulation requires timely intervention to correct emerging deviations and restore progress after execution errors. However, recovery methods based on repeated vision-language reasoning or iterative online optimization can incur substantial latency, delaying intervention. To address these challenges, we introduce RESETTLE(Robotic rEcovery through diSagrEement-Triggered reTrievaL and Efficient Corrective Control), a model-agnostic framework that provides computationally efficient recovery at the action-execution interface of frozen robot policies. RESETTLE triggers recovery when two action proposals independently sampled under identical conditioning persistently disagree. It retrieves a same-task demonstration reference using an adapted V-JEPA encoder and combines a state-servo prior with a guarded visual residual to execute one corrective action without online trajectory optimization or additional vision-language reasoning, then returns control to the base policy. Across six base policies in simulation, RESETTLE achieves up to 8.70%, 6.28%, and 6.83% absolute success-rate gains on LIBERO-Plus, Meta-World, and RoboCasa Tabletop, respectively, with further improvements on four real-world tasks using two policies. In QwenPI-based comparisons, its monitoring-and-recovery computation latency is 74.04%--93.57% lower than VoLoAgent's monitoring-and-planning latency for grasp and place tool calls. It also raises Harness VLA's LIBERO-Pro Swap success from 42% to 50%, demonstrating compatibility with high-level agentic planning. Code available at: https://github.com/JIA-Lab-research/RESETTLE

Chinese Translation

可靠的机器人操作需要及时干预，以纠正出现的偏差并在执行错误后恢复进程。然而，基于重复视觉-语言推理或迭代在线优化的恢复方法可能会导致显著的延迟，从而延误干预。为了解决这些挑战，我们提出了RESETTLE（Robotic rEcovery through diSagrEement-Triggered reTrievaL and Efficient Corrective Control），这是一个与模型无关的框架，能够在冻结的机器人策略的动作执行接口上提供计算高效的恢复。RESETTLE在两个在相同条件下独立采样的动作提议持续存在分歧时触发恢复。它使用改进的V-JEPA编码器检索相同任务的演示参考，并结合状态伺服先验与受保护的视觉残差执行一个纠正动作，而无需在线轨迹优化或额外的视觉-语言推理，然后将控制权返回给基础策略。在六个基础策略的仿真中，RESETTLE在LIBERO-Plus、Meta-World和RoboCasa Tabletop上分别实现了高达8.70%、6.28%和6.83%的绝对成功率提升，并在使用两种策略的四个真实世界任务上进一步改善。在基于QwenPI的比较中，其监控与恢复计算延迟比VoLoAgent的监控与规划延迟低74.04%至93.57%（针对抓取和放置工具调用）。它还将Harness VLA的LIBERO-Pro Swap成功率从42%提高到50%，展示了与高层次代理规划的兼容性。代码可在：https://github.com/JIA-Lab-research/RESETTLE

View on arXiv Download PDF AI Translation

cs.RO / 70 / 2610.12194

MiniWAM: Learning Compact Future Targets for Efficient World-Action Modeling

MiniWAM：学习紧凑的未来目标以实现高效的世界行动建模

Chen, Jie, Bai, Ruofei, Cai, Yuxin, Zhang, Yifeng, He, Chengyang, Li, Jun, Yau, Wei-Yun, Sartoretti, Guillaume

Abstract

World modeling has emerged as an effective co-training objective for robot policies, giving rise to World Action Models (WAMs) that jointly predict actions and future states. However, most WAMs predict future states in the native representation space of pretrained visual backbones, resulting in high-dimensional targets with substantial training cost. We introduce MiniWAM, which instead predicts compact future representations learned from privileged current-future transitions. To construct these targets, we propose Predictive Representations via Inverse Spatiotemporal Modeling (PRISM), which combines inverse-dynamics supervision with feature reconstruction to emphasize control-relevant transition information while preserving useful future-state information. With the learned PRISM encoder frozen, MiniWAM is trained to jointly predict the resulting targets and robot actions from current observations. With 65$\times$ fewer native future feature tokens, MiniWAM consistently outperforms native future-feature prediction with both DINOv3 and WAN2.1 VAE features, while achieving up to an 8$\times$ speedup in world-action training. At 0.25B parameters, MiniWAM is already competitive with substantially larger WAMs on LIBERO, LIBERO-Plus, and RoboTwin 2.0 simulation benchmarks. Representation analyses further show that PRISM contributes behavioral structure beyond feature reconstruction alone. These results demonstrate that effective world-action modeling does not require predicting native visual futures, and that compact predictive representations provide a strong and substantially more efficient target for policy learning. The project page is available at: https://j1dan.github.io/MiniWAM.

Chinese Translation

世界建模已成为机器人策略的有效共同训练目标，催生了世界行动模型（World Action Models, WAMs），这些模型共同预测行动和未来状态。然而，大多数WAM在预训练视觉骨干网络的原生表示空间中预测未来状态，导致高维目标和巨大的训练成本。我们提出了MiniWAM，它预测从特权的当前-未来过渡中学习到的紧凑未来表示。为了构建这些目标，我们提出了通过逆时空建模的预测表示（Predictive Representations via Inverse Spatiotemporal Modeling, PRISM），该方法结合了逆动力学监督和特征重构，强调与控制相关的过渡信息，同时保留有用的未来状态信息。在冻结学习到的PRISM编码器后，MiniWAM被训练以共同预测当前观察下的目标和机器人动作。MiniWAM使用65倍更少的原生未来特征标记，始终优于使用DINOv3和WAN2.1 VAE特征的原生未来特征预测，同时在世界行动训练中实现高达8倍的加速。在0.25B参数下，MiniWAM在LIBERO、LIBERO-Plus和RoboTwin 2.0模拟基准上已与大得多的WAM竞争。表示分析进一步表明，PRISM在行为结构上贡献了超出特征重构的效果。这些结果表明，有效的世界行动建模并不需要预测原生视觉未来，而紧凑的预测表示为策略学习提供了一个强大且显著更高效的目标。项目页面可访问： https://j1dan.github.io/MiniWAM。

View on arXiv Download PDF AI Translation

cs.RO / 71 / 2610.12196

MAP2: Model- and Acceleration-Based Pursuit with MPC and Gaussian Process Residual Learning for Autonomous Racing

MAP2：基于模型和加速度的追踪控制，结合模型预测控制（MPC）和高斯过程残差学习，用于自主赛车

Huang, Weizhan, Hu, Cheng, Ghignone, Edoardo, Baumann, Nicolas, Wang, Yiqin, Su, Hongye, Xie, Lei, Magno, Michele

Abstract

Autonomous racing requires accurate trajectory tracking near the handling limits of a vehicle while maintaining low computational latency. Geometric controllers are computationally efficient, but the Ackermann steering geometry becomes invalid under limit-handling conditions. Model- and Acceleration-based Pursuit (MAP) preserves the simplicity of geometric approaches while leveraging tire dynamics. Yet MAP remains fundamentally a geometric controller that relies on Ackermann steering geometry, and its tire model may not fully capture the vehicle's actual dynamic response. This paper presents MAP2, a model-based pursuit controller that combines a curvature-based kinematic MPC and a Sparse Gaussian Process (SGP) residual correction. The proposed algorithm uses MPC to optimize kinematic control inputs over a prediction horizon and maps them to steering commands through a tire dynamics model augmented with SGP residual correction. Real-world vehicle experiments demonstrate substantial reductions in lateral tracking error and lap time. Compared with MAP and Pure Pursuit (PP), MAP2 reduces the average lateral tracking error by 37.99% and 44.65%, respectively, while reducing average lap time by at least 1.5%.

Chinese Translation

自主赛车需要在车辆的操控极限附近进行精确的轨迹跟踪，同时保持低计算延迟。几何控制器计算效率高，但在极限操控条件下，Ackermann转向几何不再有效。基于模型和加速度的追踪控制（MAP）保留了几何方法的简单性，同时利用了轮胎动力学。然而，MAP本质上仍然是依赖于Ackermann转向几何的几何控制器，其轮胎模型可能无法完全捕捉车辆的实际动态响应。本文提出了MAP2，一种基于模型的追踪控制器，结合了基于曲率的运动学模型预测控制（MPC）和稀疏高斯过程（SGP）残差校正。所提出的算法使用MPC在预测视域内优化运动学控制输入，并通过增强SGP残差校正的轮胎动力学模型将其映射为转向指令。实际车辆实验表明，横向跟踪误差和圈速均显著降低。与MAP和纯追踪（PP）相比，MAP2分别将平均横向跟踪误差降低了37.99%和44.65%，同时平均圈速至少降低了1.5%。

View on arXiv Download PDF AI Translation

cs.RO / 72 / 2610.12202

Sim-to-Real RL for ASVs using SysID

基于系统识别的自主水面车辆的仿真到现实强化学习

Sheltraw, Cody, Lazouski, Tsimafei, Ghaffari, Maani, Papalia, Alan

Abstract

Autonomous Surface Vehicles (ASVs) operating in dynamic marine environments require robust control policies for tasks such as path following and station keeping, making reinforcement learning (RL) a promising alternative to classical controllers. However, existing ASV simulators rarely support parallel environments for RL training. Such existing simulators require accurate hydrodynamic modeling from computational fluid dynamics solvers or towing tank tests for setting hydrodynamic parameters to address the sim-to-real gap. To address these challenges, we present an ASV simulator and accompanying pipeline that enables training policies starting from unknown vehicle dynamics. Our framework uses only a CAD model and brief set of open-water field trajectories for approximating and refining both hydrodynamic and thruster parameters. Real-world deployments on a BlueBoat ASV demonstrate successful zero-shot sim-to-real transfer in path following and station-keeping tasks without prior hydrodynamic and propeller information.

Chinese Translation

在动态海洋环境中运行的自主水面车辆（ASVs）需要强大的控制策略以执行路径跟踪和保持位置等任务，这使得强化学习（RL）成为经典控制器的一个有前景的替代方案。然而，现有的ASV仿真器很少支持用于RL训练的并行环境。这些现有的仿真器需要来自计算流体动力学求解器或拖曳水槽测试的准确水动力建模，以设置水动力参数，从而解决仿真到现实的差距。为了解决这些挑战，我们提出了一种ASV仿真器及其配套管道，使得从未知车辆动态开始训练策略成为可能。我们的框架仅使用CAD模型和一组简短的开放水域轨迹来近似和优化水动力和推进器参数。在BlueBoat ASV上的实际部署展示了在路径跟踪和保持位置任务中成功实现零-shot的仿真到现实转移，而无需事先的水动力和螺旋桨信息。

View on arXiv Download PDF AI Translation

cs.RO / 73 / 2610.12231

Residual Modeling Closes the Regression and Generative Policy Gap in Robot Learning

残差建模缩小机器人学习中的回归与生成策略差距

Zhou, Yuchen, You, Jiacheng, Wan, Weikang, Dong, Weijun, Gao, Yang, Mao, Jiayuan

Abstract

Learning from demonstration has enabled impressive robot behaviors. A common choice for policy learning is to use diffusion or flow matching (Flow-Policies), which often outperforms direct action regression trained with mean squared error (MSE-Policies). This gap is commonly attributed to multimodal demonstrations. We revisit this gap from the perspective of statistical modeling: how action-prediction residuals shape policy optimization. Our analysis of real-world robot demonstration data reveals substantial state-dependent variation in residual scales and heavier-than-Gaussian tails. While both MSE-Policies and Flow-Policies exhibit heavy-tailed action residuals, their training gradients behave differently: MSE allocates more gradient magnitude to observations with large action residuals, which hurts optimization. Motivated by these findings, we introduce heteroscedastic Student-t action regression (HT-Policies), which learns input-dependent residual scales and reduces the influence of heavy tails. HT-Policies predict action chunks with a single feed-forward pass and can reuse pretrained flow-matching-based policy networks as the backbone. Across four simulation benchmarks and real-robot evaluations, HT-Policies achieves success rates competitive with generative policy baselines, both when trained from scratch and from pretrained vision-language-action and world-action models, despite being faster in training and inference. Together, these findings shed light on the practical advantages of generative objectives in robot learning from demonstrations and offer an efficient direct-regression alternative for a range of architectures and tasks. Project page: https://the-labone.github.io/regression-policy-project/

Chinese Translation

从示范中学习使得机器人行为得以显著提升。政策学习的常见选择是使用扩散或流匹配（Flow-Policies），这通常优于直接使用均方误差（MSE-Policies）训练的动作回归。这一差距通常归因于多模态示范。我们从统计建模的角度重新审视这一差距：动作预测残差如何影响政策优化。对真实世界机器人示范数据的分析揭示了残差尺度的显著状态依赖性变化和重于高斯分布的尾部特征。尽管MSE-Policies和Flow-Policies都表现出重尾的动作残差，但它们的训练梯度表现不同：MSE将更多的梯度幅度分配给具有大动作残差的观测，这对优化造成了不利影响。基于这些发现，我们引入了异方差的Student-t动作回归（HT-Policies），该方法学习输入依赖的残差尺度并减少重尾的影响。HT-Policies通过单次前馈传递预测动作块，并可以重用基于流匹配的预训练政策网络作为骨干。在四个仿真基准和真实机器人评估中，HT-Policies在从零开始训练和从预训练的视觉-语言-动作及世界-动作模型训练时，成功率与生成策略基准相当，且在训练和推理中速度更快。这些发现共同揭示了生成目标在机器人示范学习中的实际优势，并为多种架构和任务提供了一种高效的直接回归替代方案。项目页面：https://the-labone.github.io/regression-policy-project/

View on arXiv Download PDF AI Translation

cs.RO / 74 / 2610.12241

From Language to Motion: Task-Conditioned Focal-Stack Trajectory Integration for Microscopic Robots

从语言到运动：基于任务条件的焦点堆叠轨迹集成用于微型机器人

Xie, Junjie, He, Chuxuan, Huang, Junkai, Zhang, Heng, Ye, Angen, Song, Yujia, Li, Yuqing, Zhang, Pengsong, Zhang, Dapeng

Abstract

Microscopic robots require accurate task geometry despite changes in language, parts, and focus. We present a semantic-to-physical framework that maps instructions to constrained geometric operators, reuses frozen open-vocabulary perception, and integrates locally reliable focal-plane trajectories by confidence weighting and dynamic programming. Calibrated multi-view geometry connects 2-D paths to physical execution. Prompt, unseen-part, and geometry reconfiguration tests yield 6.30-6.59-pixel RMSE. Relative to part-specific U-Net training with 20-100 labels, the proposed zero-new-label configuration takes 15 rather than 72-165 min. Across nine part-illumination conditions, trajectory-space integration reduces RMSE from 14.41 to 6.28 pixels (56.4%) and P95 error from 20.07 to 8.13 pixels (59.5%) compared with image-first multi-focus fusion. An ablation isolates the roles of confidence and path-wise selection. In representative robot experiments, target-region coverage improves from 83.5% to 92.9%. Dispensing provides a measurable physical trace, not a task-specific limitation of the method.

Chinese Translation

微型机器人在语言、部件和焦点变化的情况下仍需准确的任务几何形状。我们提出了一种语义到物理的框架，将指令映射到受限的几何操作符，重用冻结的开放词汇感知，并通过置信加权和动态规划集成局部可靠的焦平面轨迹。经过校准的多视角几何将二维路径连接到物理执行。提示、未见部件和几何重配置测试的均方根误差（RMSE）为6.30-6.59像素。与使用20-100个标签的特定部件U-Net训练相比，所提出的零新标签配置仅需15分钟，而不是72-165分钟。在九种部件照明条件下，轨迹空间集成将RMSE从14.41降低到6.28像素（56.4%），P95误差从20.07降低到8.13像素（59.5%），相比于图像优先的多焦点融合。一项消融实验隔离了置信度和路径选择的作用。在代表性的机器人实验中，目标区域覆盖率从83.5%提高到92.9%。分配提供了可测量的物理痕迹，而不是方法的任务特定限制。

View on arXiv Download PDF AI Translation

cs.RO / 75 / 2610.12245

Fixed-Reference Pose Residuals for Measuring Cross-Dataset Cue Transfer in Human-Robot Interaction Anticipation

用于测量人机交互预测中跨数据集线索转移的固定参考姿态残差

Yang, Bowen, Xiao, Xinliang, Zhang, Wenjing, Yang, Li, Zhou, Wei

Abstract

Social and service robots in public spaces need to anticipate which nearby person is about to approach and touch them, so that a response can be prepared before contact. It is largely unknown which cues support this anticipation when a model trained with one robot is used on another robot at a different site. We study this question with a fixed-reference pose residual (FRPR) model: a geometry predictor built from the person's bounding box and mask is trained and frozen, and a temporal network then learns from body pose an additive correction to its logit, so that every prediction splits exactly into a geometry term and a pose term. Between two public egocentric datasets recorded by different robots, HUI360 and SSUP-A, with every choice made on source data, the pose correction raised average precision (AP) from 0.277 to 0.321 from SSUP-A to HUI360 and gave no measurable gain in the opposite direction; the same asymmetry held over a stronger, source-selected geometry reference. Freezing gave no AP advantage over joint training, and simple geometric baselines and tree ensembles remained competitive or better, so the construction serves measurement rather than prediction accuracy. A head-orientation residual added small gains in both directions. Post hoc, whether a person faces the camera kept its discriminative direction across datasets, whereas head pitch reversed. With thresholds chosen on source data, the neural models that use geometry detected at most 17% of target interactions. Code and processed data are available at https://github.com/WeiZhou96/FRPR-interaction-anticipation.

Chinese Translation

公共空间中的社交和服务机器人需要预测哪个附近的人即将接近并触摸它们，以便在接触之前做好反应准备。当使用在一种机器人上训练的模型在另一种机器人上进行不同场地的应用时，哪些线索支持这种预测仍然 largely unknown。我们通过固定参考姿态残差（FRPR）模型研究这个问题：一个基于人的边界框和掩码构建的几何预测器被训练并冻结，然后一个时间网络从身体姿态中学习对其对数几率的附加修正，使得每个预测精确地分为几何项和姿态项。在由不同机器人记录的两个公共自我中心数据集 HUI360 和 SSUP-A 之间，针对源数据的每个选择，姿态修正将平均精度（AP）从 SSUP-A 的 0.277 提高到 HUI360 的 0.321，而在相反方向上没有可测量的增益；在更强的、源选择的几何参考上也保持了相同的不对称性。冻结在联合训练上没有带来 AP 的优势，简单的几何基线和树集成仍然具有竞争力或更好，因此该构建用于测量而非预测准确性。头部朝向残差在两个方向上均增加了小幅增益。事后分析发现，面对相机的人在不同数据集中的判别方向保持一致，而头部俯仰则相反。使用源数据选择的阈值，使用几何检测的神经模型最多检测到 17% 的目标交互。代码和处理后的数据可在 https://github.com/WeiZhou96/FRPR-interaction-anticipation 获取。

View on arXiv Download PDF AI Translation

cs.RO / 76 / 2610.12249

Real-Time Motion Planning with Dynamic Hazards: Classical vs. Learning-Based Methods

动态危险下的实时运动规划：经典方法与基于学习的方法的比较

Iceland, Eran, Tuisov, Alexander, Gal, Oren, Barel, Ariel, Bruckstein, Alfred M.

Abstract

We study real-time motion planning in dynamic hazard fields through a controlled comparison between classical planning and learning-based methods. Rather than introducing a new planner, we construct a unified benchmark in which representative classical and learning-based methods face the same environments, motion constraints, information assumptions, and evaluation metrics. The test environment consists of planar domains populated with rotating sprinkler-like hazards that generate time-varying forbidden regions via sweeping angular sectors. Our results show a clear regime shift. In deterministic environments, classical planners achieve near-perfect success and higher-quality paths, though sometimes at the cost of substantial planning or replanning time. Under stochastic obstacle dynamics, however, online search becomes strongly budget-sensitive: low budgets lead to frequent failure, while high budgets improve success at the cost of latency and longer trajectories. PPO-based policies, trained under the same scenario distribution, consistently outperform in latency, success rate, and path quality in these stochastic regimes. Overall, the results indicate that uncertainty in obstacle evolution, more than partial observability, is the dominant factor determining which planning paradigm is practically effective for the problem at hand.

Chinese Translation

我们通过对经典规划与基于学习的方法进行受控比较，研究动态危险场中的实时运动规划。我们并未引入新的规划器，而是构建了一个统一的基准，在该基准中，具有代表性的经典方法和基于学习的方法在相同的环境、运动约束、信息假设和评估指标下进行测试。测试环境由平面领域构成，领域中布满了旋转的喷洒器状危险物，这些危险物通过扫动角度扇区生成时间变化的禁止区域。我们的结果显示出明显的模式转变。在确定性环境中，经典规划器几乎实现完美成功并生成更高质量的路径，尽管有时需要付出大量的规划或重新规划时间。然而，在随机障碍动态下，在线搜索变得对预算非常敏感：低预算导致频繁失败，而高预算则在延迟和更长轨迹的代价下提高成功率。在相同场景分布下训练的基于PPO（Proximal Policy Optimization）的方法在这些随机环境中在延迟、成功率和路径质量方面始终表现优于其他方法。总体而言，结果表明，障碍物演变的不确定性比部分可观测性更能决定哪种规划范式在实际问题中有效。

View on arXiv Download PDF AI Translation

cs.RO / 77 / 2610.12272

Walking on Roofs: Exploring the Potential of Walking Robots for Construction Work on Roofs

在屋顶上行走：探索步行机器人在屋顶施工中的潜力

Dettmar, Bjoern-Felix, Roennau, Arne

Abstract

This paper investigates the feasibility of deploying quadruped walking robots for the automation of work in roof environments. While quadrupeds have demonstrated versatility across various domains, their large-scale deployment remains limited, partly due to lack of application-specific designs. Roof environments represent a novel and unexplored use case, combining high safety risks for human workers with repetitive, strenuous tasks that could benefit from robotic assistance. A dedicated test rig of a roof's surface was designed to evaluate the baseline performance of a commercial quadruped, the \emph{Unitree Go2}, in this new environment. Experiments revealed that standard ball feet are inherently inadequate for locomotion on sloped roofs: slippage increased quadratically with incline angle, get-up and lie-down sequences were only possible on small inclines, and critical failures already occurred regularly on moderate inclines of 25{\deg}. This work provides the first systematic assessment of quadruped locomotion in roof environments, highlighting both the potential and the current limitations of this application and establishes a baseline for future research. Effective solutions will require a combination of task-oriented foot designs, advanced contact mechanics and environment-specific control strategies, such as reinforcement learning for roof-adapted gaits.

Chinese Translation

本文研究了在屋顶环境中部署四足步行机器人进行自动化工作的可行性。尽管四足机器人在多个领域展现了多样性，但其大规模应用仍然受到限制，部分原因是缺乏特定应用的设计。屋顶环境代表了一种新颖且未被探索的应用场景，结合了对人类工人高安全风险的挑战与可能从机器人辅助中受益的重复性、费力的任务。我们设计了一个专门的测试装置，以评估商业四足机器人 extit{Unitree Go2} 在这一新环境中的基础性能。实验结果表明，标准的球形足部在斜屋顶上的运动能力本质上是不足的：滑移随着倾斜角度的增加呈平方增长，起立和躺下的动作仅在小倾斜度下可行，而在25{ extdegree}的中等倾斜度下，关键故障已经频繁发生。这项工作提供了对屋顶环境中四足运动的首次系统评估，突出了这一应用的潜力和当前的局限性，并为未来的研究建立了基准。有效的解决方案将需要结合任务导向的足部设计、先进的接触力学和特定环境的控制策略，例如针对屋顶适应的步态的强化学习。

View on arXiv Download PDF AI Translation

cs.RO / 78 / 2610.12276

Toward Lunar Legged Robots: Field Deployment Lessons at LUNA

迈向月球四足机器人：LUNA现场部署的经验教训

Fuhrer, Adrian, Church, Joseph, Fischer, Oliver, Talbot, William, Faesch, Nicolas, Hofmann, Yannic, Kolvenbach, Hendrik, Tanaka, Yusuke, Hutter, Marco

Abstract

Legged robots are promising candidates for future lunar surface missions because they can traverse steep, loose, and obstacle-rich terrain that challenges conventional wheeled rovers. However, readiness for lunar deployment is limited by uncertainties in foot-regolith interaction, dust generation, illumination-driven perception degradation, and operational constraints. This paper reports lessons from the 2025 LUNA analogue campaign, where ANYmal-D and Magnecko traversed loose regolith simulant and crater-like terrain and collected long-horizon navigation and visual-inertial data under challenging lighting. We show that quadrupedal robots can traverse regolith simulant, but performance is affected by sinkage and slip, dust-generating contacts, and perception failures caused by overexposure, shadows, and low-texture regions. These results motivate tighter integration of regolith-aware locomotion policies, illumination-robust perception, repeatable analogue testing, and mission-level operational validation for future lunar legged robots.

Chinese Translation

四足机器人是未来月球表面任务的有希望的候选者，因为它们能够穿越陡峭、松散和充满障碍的地形，这些地形对传统的轮式探测器构成了挑战。然而，月球部署的准备工作受到脚与月壤相互作用、不确定的灰尘生成、光照驱动的感知衰减和操作限制的影响。本文报告了2025年LUNA类比活动的经验教训，在该活动中，ANYmal-D和Magnecko穿越了松散的月壤模拟物和类似陨石坑的地形，并在具有挑战性的光照条件下收集了长时间导航和视觉惯性数据。我们展示了四足机器人能够穿越月壤模拟物，但其性能受到沉陷和滑移、产生灰尘的接触以及由于过度曝光、阴影和低纹理区域导致的感知失败的影响。这些结果促使我们更加紧密地整合对月壤敏感的运动策略、光照鲁棒的感知、可重复的类比测试以及未来月球四足机器人的任务级操作验证。

View on arXiv Download PDF AI Translation

cs.RO / 79 / 2610.12285

PLaW-VLA: Predictive Latent World Modeling for Vision-Language-Action Policies

PLaW-VLA：用于视觉-语言-动作策略的预测潜在世界建模

Liu, Yu, Guo, Hetian, Huang, Tianlv, Cai, Ziyi, Chen, Wudi, Wang, Hantang, Liu, Qiutong, Peng, Yingzhi, Han, Wei, Tang, Peijun, Wang, Jianan, Fan, Zipei, Zha, Zhiyuan, Song, Xuan

Abstract

Learning to predict how the world evolves can provide vision-language-action (VLA) policies with predictive context for long-horizon control, but its effectiveness depends on what future representation is modeled and how it conditions action generation. We introduce PLaW-VLA, which models task-relevant future states in a pretrained prediction-oriented representation space, reducing the need to predict control-irrelevant visual details. Built on a Mixture-of-Transformers architecture, PLaW-VLA conditions action generation on observation history, current task semantics, and predicted future states through structured causal attention. Experiments show a +11.8 percentage-point (pp) gain over reactive policies on RoboTwin Hard Horizon III and a +1.77 pp gain over reconstruction-oriented latent prediction on zero-shot LIBERO-Plus, supporting improved long-horizon control and generalization under distribution shift, respectively. By avoiding low-level visual reconstruction, PLaW-VLA lowers the burden of future prediction, enabling a lightweight latent world model with parallel future prediction and about 1/19 the inference latency of generative world-action modeling at comparable policy performance.

Chinese Translation

学习预测世界如何演变可以为视觉-语言-动作（VLA）策略提供长时间控制的预测上下文，但其有效性取决于所建模的未来表示以及它如何影响动作生成。我们提出了PLaW-VLA，它在一个预训练的以预测为导向的表示空间中建模与任务相关的未来状态，从而减少对控制无关视觉细节的预测需求。PLaW-VLA基于混合变换器（Mixture-of-Transformers）架构，通过结构化因果注意力（structured causal attention）将动作生成与观察历史、当前任务语义和预测的未来状态相结合。实验结果显示，在RoboTwin Hard Horizon III上，相较于反应式策略，PLaW-VLA提高了11.8个百分点（pp）；在零-shot LIBERO-Plus上，相较于重建导向的潜在预测，提升了1.77个百分点，分别支持了在分布转移下的长时间控制和泛化能力。通过避免低级视觉重建，PLaW-VLA降低了未来预测的负担，使得轻量级潜在世界模型能够实现并行未来预测，并且在可比政策性能下，其推理延迟约为生成世界-动作建模的1/19。

View on arXiv Download PDF AI Translation

cs.RO / 80 / 2610.12368

LiteNWM: Efficient Latent World Models for Onboard Visual Navigation in the Wild

LiteNWM：高效的潜在世界模型用于野外的机载视觉导航

Liu, Linkai, Zhang, Yuntian, Bing, Zhenshan, Chen, Chen, Lyu, Lingjuan, Wang, Shangguang, Xu, Mengwei, Cai, Dongqi

Abstract

Direct visual navigation policies generate trajectories efficiently but do not explicitly evaluate their future consequences. Generative navigation world models provide this foresight through visual rollouts, which are costly when evaluating multiple candidates. We present LiteNWM, a latent navigation world model that shares visual encoding across candidates and jointly predicts their action-conditioned future representations at multiple horizons, while a learned scorer uses these predictions to select trajectories. In offline evaluations on RECON, SCAND, and SACSoN, LiteNWM reduces macro-averaged trajectory error by 17.56% relative to NoMaD+NWM-XL and achieves a 128.00-fold end-to-end speedup on an RTX 5090. The same evaluator transfers from NoMaD to MBRA without proposer-specific retraining, reducing MBRA's macro-averaged trajectory error by 16.2%. In real-robot experiments in unseen indoor and outdoor environments, LiteNWM improves navigation success from 43.3% to 83.3% relative to NoMaD. These results demonstrate that LiteNWM can be deployed for future-aware planning and closed-loop navigation on a physical robot.

Chinese Translation

直接的视觉导航策略能够高效生成轨迹，但并未明确评估其未来后果。生成式导航世界模型通过视觉展开提供这种前瞻性，然而在评估多个候选者时成本较高。我们提出了LiteNWM，一种潜在导航世界模型，它在候选者之间共享视觉编码，并在多个时间范围内联合预测其基于动作的未来表示，同时一个学习的评分器利用这些预测来选择轨迹。在RECON、SCAND和SACSoN的离线评估中，LiteNWM相较于NoMaD+NWM-XL将宏观平均轨迹误差降低了17.56%，并在RTX 5090上实现了128.00倍的端到端加速。同样的评估器从NoMaD转移到MBRA时，无需特定于提议者的重新训练，将MBRA的宏观平均轨迹误差降低了16.2%。在未见过的室内和室外环境中的真实机器人实验中，LiteNWM将导航成功率从43.3%提高到83.3%，相较于NoMaD。这些结果表明，LiteNWM可以用于未来感知规划和物理机器人上的闭环导航。

View on arXiv Download PDF AI Translation

cs.RO / 81 / 2610.12369

Embodied Turing Machines: Stateful Code for Robot Recursive Self-Improvement

具身图灵机：用于机器人递归自我改进的有状态代码

Hu, Kairui, Hu, Siyuan, Hong, Fangzhou, Chen, Zhaoxi, Liu, Ziwei

Abstract

Most robot policies keep a model in the control loop: a VLA maps observations to actions, and an Agent Harness, such as Agent-as-Policy or Harness VLA queries a VLM for decision making at run time. We propose a different view: the embodied world is an Embodied Turing Machine, whose tape is the robot and environment state and rules are the policy. If this state can be represented accurately, the decision making can be written entirely in code. We therefore propose Code-Only-as-Policy (COAP): code measures and tracks the robot, environment, and task state from camera images and proprioception, and makes every decision from it. The same code applies across episodes, and different tasks share one library without a VLM or VLA in the loop. Compared with VLAs and Agent Harnesses, we analyze three advantages of COAP: (i) Explicit State: the state can be stored in code; (ii) Execution: code makes decision making controllable, recovers from failures flexibly, and runs fast and cheaply online; (iii) Extensibility: new tasks reuse, inherit, or extend the shared library, so capabilities can accumulate over tasks. These advantages make COAP a suitable medium for recursive self-improvement (RSI): coding agents develop the library in a closed loop, and each change is explicit and controllable. On RoboDojo's 42 bimanual tasks, the resulting library reaches a success rate of 70.24% without a model at test time. The upper bound of COAP lies in how accurately the state is represented for decision making and how robust the code logic is. We thus propose COAP as a new paradigm for embodied tasks; since it applies across episodes, it can also serve as an efficient data engine for VLAs and Agent Harnesses.

Chinese Translation

大多数机器人策略在控制循环中保持一个模型：一个可变形状线性映射（VLA）将观察映射到动作，而代理工具（Agent Harness），如代理作为策略（Agent-as-Policy）或工具VLA（Harness VLA），在运行时查询可变形状模型（VLM）以进行决策。我们提出了一种不同的视角：具身世界是一个具身图灵机，其带子是机器人和环境状态，规则是策略。如果这个状态能够被准确表示，决策可以完全用代码编写。因此，我们提出了仅代码作为策略（Code-Only-as-Policy，COAP）：代码通过相机图像和本体感知测量和跟踪机器人、环境和任务状态，并基于此做出每一个决策。相同的代码在不同的回合中适用，不同的任务共享一个库，而不需要在循环中使用VLM或VLA。与VLA和代理工具相比，我们分析了COAP的三个优势：（i）显式状态：状态可以存储在代码中；（ii）执行：代码使得决策控制可控，能够灵活地从失败中恢复，并且在线运行快速且成本低廉；（iii）可扩展性：新任务可以重用、继承或扩展共享库，因此能力可以在任务之间积累。这些优势使得COAP成为递归自我改进（RSI）的合适媒介：编码代理在闭环中开发库，每一次变化都是显式且可控的。在RoboDojo的42个双手任务中，得到的库在测试时的成功率达到了70.24%，而没有使用模型。COAP的上限在于状态在决策中的表示准确性以及代码逻辑的稳健性。因此，我们提出COAP作为具身任务的新范式；由于它适用于不同的回合，它也可以作为VLA和代理工具的高效数据引擎。

View on arXiv Download PDF AI Translation

cs.RO / 82 / 2610.12386

ARC: A Reasoning Recipe for Robot Foundation Models

ARC：机器人基础模型的推理方案

Puthumanaillam, Gokul, Sun, Tao, Aljalbout, Elie, Reuss, Moritz, Li, Zhaoshuo, Ramos, Fabio, Goyal, Ankit, Yang, Jenai Xuning

Abstract

The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale. We show that there exists an effective and efficient complementary approach: the right reasoning recipe can substantially improve the zero-shot task performance of existing state-of-the-art RFMs. We refer to this recipe as ARC. It consists of three key ingredients: a reasoning trace, a scalable automatic labeling pipeline, and a strategy for adapting pretrained RFMs to use these traces for control. First, we find that effective reasoning traces should be grounded in the robot's next action and explain its causal structure: why the action is appropriate and what effect it should produce. Second, we show that these traces can be generated automatically from existing demonstrations, enabling us to construct ARC-Trace-DROID from DROID without collecting new robot data. Third, we show how state-of-the-art VLAs such as $\pi_{0.5}$ and WAMs such as Cosmos3-Nano-Policy can learn to use these traces for control, with fine-tuning and inference tailored to each model's architecture and capabilities. Using ARC, we obtain gains in zero-shot RFM performance that, to our knowledge, are unprecedented without additional robot demonstrations or foundation-scale training. The adapted models establish a new state of the art on RoboLab-120 and MolmoSpaces, with gains of up to 50 percentage points on RoboLab-Reasoning-50. On real robots, ARC improves $\pi_{0.5}$'s task success by 82.2 percentage points. Project website: https://arc-robot-reasoning.github.io/

Chinese Translation

当前改进机器人基础模型（RFM）的主要方法依赖于更大的模型、更多的机器人演示以及高成本的大规模训练。我们展示了一种有效且高效的互补方法：合适的推理方案可以显著提升现有最先进RFM的零-shot任务性能。我们将这一方案称为ARC。它由三个关键成分组成：推理轨迹、可扩展的自动标注管道以及将预训练RFM适应为使用这些轨迹进行控制的策略。首先，我们发现有效的推理轨迹应基于机器人的下一步动作，并解释其因果结构：为什么该动作是合适的，以及它应产生什么效果。其次，我们展示了这些轨迹可以从现有演示中自动生成，使我们能够从DROID构建ARC-Trace-DROID，而无需收集新的机器人数据。第三，我们展示了如何让最先进的视觉语言模型（VLA）如$ ext{π}_{0.5}$和基于模型的策略（WAM）如Cosmos3-Nano-Policy学习使用这些轨迹进行控制，微调和推理则针对每个模型的架构和能力进行定制。使用ARC，我们在零-shot RFM性能上获得了前所未有的提升，而无需额外的机器人演示或基础规模训练。经过调整的模型在RoboLab-120和MolmoSpaces上建立了新的最先进水平，在RoboLab-Reasoning-50上提升了多达50个百分点。在真实机器人上，ARC使得$ ext{π}_{0.5}$的任务成功率提高了82.2个百分点。项目网站：https://arc-robot-reasoning.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 83 / 2610.12404

A Physics-Informed Collision Learning Framework for Collaborative Robot Motion Generation

一种物理信息驱动的碰撞学习框架用于协作机器人运动生成

Cai, Chen, Liu, Steven

Abstract

Close-proximity multi-arm manipulation requires collision models that are both geometrically accurate and differentiable enough for real-time optimization. Classical geometry checkers provide reliable distances but are difficult to use inside gradient-based model predictive control, while conservative proxy models can restrict tightly coupled motion. We present PI-UDF, a physics-informed unified differentiable framework for body-to-body collision distance prediction between articulated robots. PI-UDF combines analytical forward kinematics with learnable link-geometry embeddings and a shared residual network to predict pairwise inter-arm distances directly from robot configurations. To improve safety-critical fidelity, we combine quota-driven boundary mining with an asymmetric boundary-crossing penalty that emphasizes false-safe sign errors near the collision boundary. The learned distance field is integrated into nonlinear MPC as a differentiable inter-arm clearance term. We validate the framework on a real dual-Franka platform through high-speed close-proximity 14-DoF dual-arm swapping, sustained single-arm dynamic evasion, and dynamic-evasion planning configurations with frozen, predicted, and target-switching treatments of the moving arm. Hardware experiments and offline Drake/FCL replay show that PI-UDF provides a differentiable inter-arm clearance estimate suitable for closed-loop collision-aware collaborative robot motion generation.

Chinese Translation

近距离多臂操作需要几何上准确且足够可微的碰撞模型，以便进行实时优化。经典的几何检查器提供可靠的距离，但在基于梯度的模型预测控制中使用起来较为困难，而保守的代理模型可能限制紧密耦合的运动。我们提出了 PI-UDF，这是一种物理信息驱动的统一可微框架，用于预测关节机器人之间的体对体碰撞距离。PI-UDF 将解析前向运动学与可学习的连杆几何嵌入和共享残差网络相结合，直接从机器人配置中预测成对的臂间距离。为了提高安全关键的保真度，我们结合配额驱动的边界挖掘与不对称的边界穿越惩罚，强调在碰撞边界附近的误安全符号错误。学习到的距离场被整合到非线性模型预测控制中，作为可微的臂间间隙项。我们在真实的双弗兰卡平台上验证了该框架，通过高速近距离的 14 自由度双臂交换、持续的单臂动态规避，以及具有静止、预测和目标切换处理的移动臂的动态规避规划配置。硬件实验和离线 Drake/FCL 重放表明，PI-UDF 提供了适用于闭环碰撞感知协作机器人运动生成的可微臂间间隙估计。

View on arXiv Download PDF AI Translation

cs.RO / 84 / 2610.12407

LeWAM: A JEPA World Action Model with Diffusion-Steering-Based MPC

LeWAM：一种基于扩散引导的模型预测控制的JEPA世界动作模型

Hegde, Shashank, Popov, Alexander, Aljalbout, Elie, Smolyanskiy, Nikolai

Abstract

World action models (WAMs) predict actions and future observations, typically from a reconstruction-based representation that carries noisy, redundant information which can complicate downstream predictions. We introduce LeWAM, a bidirectional transformer for forward, backward, inverse dynamics and policy prediction, on a decoder-free JEPA latent trained end-to-end through all four modes. We see the following benefits: 1) Alignment: linear probes read robot and object state from LeWAM's latent better than from a regular Le World Model (a forward-only JEPA world model), while the latent ignores visual distractors as well as LeWM does and far better than a reconstruction-based WAM. 2) Acting: Closed-loop evaluations of LeWAM match a regular flow-matching policy trained on the same encoder at matched size, while also providing a world model. 3) Planning: Sampling raw actions when planning with WAMs lets MPC exploit dynamics-model inaccuracies; planning in the noise space of the policy head instead improves the closed-loop performance of these WAMs.

Chinese Translation

世界动作模型（WAMs）预测动作和未来观察，通常基于重建的表示，这种表示携带噪声和冗余信息，可能会使下游预测变得复杂。我们介绍了LeWAM，一种双向变换器，用于前向、后向、逆动态和策略预测，采用无解码器的JEPA潜在表示，通过所有四种模式端到端训练。我们观察到以下好处：1）对齐：线性探针从LeWAM的潜在表示中读取机器人和物体状态的效果优于从常规的Le世界模型（仅前向的JEPA世界模型），而且潜在表示忽略了视觉干扰，与LeWM一样，且远优于基于重建的WAM。2）行动：LeWAM的闭环评估与在相同编码器上以匹配大小训练的常规流匹配策略相匹配，同时也提供了一个世界模型。3）规划：在使用WAMs进行规划时采样原始动作可以让模型预测控制（MPC）利用动态模型的不准确性；而在策略头的噪声空间中进行规划则改善了这些WAM的闭环性能。

View on arXiv Download PDF AI Translation

cs.RO / 85 / 2610.12411

GLIO2: A GPU-Parallelized Tightly-Coupled LiDAR-Inertial-GNSS System for Robust and Real-Time Global Localization and Mapping

GLIO2：一种GPU并行化的紧耦合LiDAR-惯性-GNSS系统，用于稳健的实时全球定位与地图构建

Zhang, Qi, Liu, Xikun, Qin, Qijun, Wang, Xiangru, Wang, Junzhe, Xiao, Naigui, Jiao, Jianhao, Wen, Weisong

Abstract

Globally consistent, real-time state estimation in large-scale, perceptually degraded environments is essential for autonomous vehicles and aerial robots, and requires fusing LiDAR, inertial, and GNSS measurements. Existing fusion methods, however, share a scan-to-map front-end with two failure modes. First, each scan is aligned to an incrementally built map that drifts under degeneracy, and once the estimate diverges the error is irrecoverable. Second, even without divergence, a registration biased by dynamic objects or wrong correspondences is propagated as a single pose constraint with an over-confident covariance, leaving its correspondences unavailable for GNSS to re-weight or relinearize. We propose GLIO2, a tightly-coupled LiDAR-Inertial-GNSS system whose GPU-parallel front-end jointly optimizes scan-to-multiscan LiDAR, IMU pre-integration, and raw GNSS measurements in a single sliding-window factor graph, sustaining real-time operation on edge hardware. A complementary offline back-end reuses the same cached factors to refine the entire trajectory in batch, completing the 30-min, 4.51-km UrbanNav Whampoa sequence in about 24 s. Across three public benchmarks (UrbanNav, MARS-LVIG, M3DGR) and self-collected UAV and vehicle data, GLIO2 attains the best overall accuracy among evaluated systems. On a 5.66-km bridge traversed at up to 96 km/h, where every competing baseline diverges under LiDAR degeneracy, it maintains 1.6 m horizontal accuracy. On an NVIDIA Jetson Orin NX, the full pipeline runs at about 25 Hz (39.60 ms per scan). The source code and datasets will be released.

Chinese Translation

在大规模、感知退化环境中进行全球一致的实时状态估计对于自主车辆和空中机器人至关重要，这需要融合LiDAR、惯性和GNSS测量。然而，现有的融合方法共享一个扫描到地图的前端，存在两种失效模式。首先，每个扫描都与逐步构建的地图对齐，而该地图在退化情况下会出现漂移，一旦估计发散，误差将无法恢复。其次，即使没有发散，由动态物体或错误对应关系引起的配准偏差也会作为单一姿态约束传播，并伴随过于自信的协方差，使得其对应关系无法供GNSS重新加权或重新线性化。我们提出了GLIO2，这是一种紧耦合的LiDAR-惯性-GNSS系统，其GPU并行前端在单个滑动窗口因子图中联合优化扫描到多扫描的LiDAR、IMU预积分和原始GNSS测量，从而在边缘硬件上维持实时操作。一个互补的离线后端重用相同的缓存因子，以批量方式精炼整个轨迹，完成30分钟、4.51公里的UrbanNav Whampoa序列仅需约24秒。在三个公共基准测试（UrbanNav、MARS-LVIG、M3DGR）和自收集的无人机及车辆数据中，GLIO2在评估的系统中获得了最佳的整体精度。在以最高96公里/小时的速度穿越的5.66公里桥梁上，所有竞争基线在LiDAR退化下均出现发散，而它保持了1.6米的水平精度。在NVIDIA Jetson Orin NX上，整个管道以约25 Hz（每扫描39.60毫秒）的速度运行。源代码和数据集将会发布。

View on arXiv Download PDF AI Translation

cs.RO / 86 / 2610.12424

RoboRSI: Stable, efficient, and reusable robot self-evolution in complex real-world environments

RoboRSI：在复杂现实环境中稳定、高效且可重用的机器人自我进化

Wen, Zimo, Chen, Yijin, Cao, Yuxuan, Chen, Wendi, Zou, Yanwen, Yu, Wenye, Kuang, Fuhang, Xue, Han, Lv, Jun, Wen, Chuan, Lu, Cewu

Abstract

A generalist robot should not only perform diverse tasks but also improve through experience, turning what it learns during execution into capabilities that later tasks can reuse. Robot agents that act through code can already repair programs from execution feedback, yet it remains a central challenge to organize this experience around the task structure that gives it meaning, so that each repair is attributed to the responsible capability, supported by execution evidence, and validated before it is reused. We introduce RoboRSI, a robot self-improvement system built on Top-Down Skill Refinement (TSR). TSR decomposes tasks into compound, atomic, and base skills with scoped responsibilities and explicit input--output contracts, attributes each execution outcome to the responsible branch, and confines revision to that branch. Building upon this structure, a Manager, Planner, Engineer, and Reviewer coordinate planning, execution, diagnosis, and the validated release of new skills, while people steer the process through objectives and corrections; stable skill sequences are further consolidated into reusable compound skills. On a mobile manipulator, RoboRSI develops multi-object household cleanup over 104 rounds. In simulation, it achieves the highest success rate on LIBERO, LIBERO-PRO, LIBERO-Plus, and RoboTwin, exceeding the strongest baseline by 2.7 to 11.0 percentage points.

Chinese Translation

通用机器人不仅应能执行多样化任务，还应通过经验不断改进，将其在执行过程中学到的知识转化为后续任务可重用的能力。通过代码行动的机器人代理已经能够根据执行反馈修复程序，但围绕任务结构组织这种经验仍然是一个核心挑战，以确保每次修复都归因于负责的能力，并得到执行证据的支持和验证，才能在重用之前进行确认。我们提出了RoboRSI，这是一种基于自上而下技能细化（Top-Down Skill Refinement, TSR）的机器人自我改进系统。TSR将任务分解为复合技能、原子技能和基础技能，并明确责任范围和输入输出契约，将每次执行结果归因于负责的分支，并将修订限制在该分支内。在这一结构基础上，管理者、规划者、工程师和审查员协调规划、执行、诊断和新技能的验证发布，而人类通过目标和修正引导这一过程；稳定的技能序列进一步整合为可重用的复合技能。在一个移动操控器上，RoboRSI在104轮中开发了多对象家庭清理。在模拟中，它在LIBERO、LIBERO-PRO、LIBERO-Plus和RoboTwin上实现了最高的成功率，超过最强基线2.7到11.0个百分点。

View on arXiv Download PDF AI Translation

cs.RO / 87 / 2610.12431

Control-Ready Uncertainty for Trajectory Diffusion

控制准备的不确定性在轨迹扩散中的应用

Xue, Zhiwei, Kam, Jia Yue, Qiu, Jinhang, Cheng, Yifeng, Gursoy, Ege, Wang, Jiaming, Bonnet, Vincent, Soh, Harold

Abstract

Diffusion models can represent complex, multimodal trajectory distributions, but extracting uncertainty from them typically requires costly Monte Carlo sampling. This limits their use in real-time control, where robots must rapidly assess risk and maintain safety margins. We introduce Score-Curvature for Online Precision Estimation (SCOPE), a lightweight module that augments diffusion trajectory models with control-ready uncertainty. SCOPE learns a structured precision matrix around each nominal trajectory by distilling score-curvature information and producing calibrated Gaussian tubes with low overhead and without repeated Monte Carlo sampling. These tubes provide per-timestep covariance estimates that can be used both as predicted occupancy for moving agents and as adaptive exploration guides for robot control. We evaluate SCOPE with mode-conditioned multimodal diffusion backbones in pedestrian forecasting, crowd navigation, Maze2D control, and real-world Franka Panda manipulation. Across these settings, SCOPE provides fast uncertainty estimation, which leads to better closed-loop performance. Project page: https://zackaxue.github.io/SCOPE-project-page/

Chinese Translation

扩散模型能够表示复杂的多模态轨迹分布，但从中提取不确定性通常需要昂贵的蒙特卡洛采样。这限制了它们在实时控制中的应用，因为机器人必须快速评估风险并保持安全边际。我们提出了在线精度估计的评分曲率模块（Score-Curvature for Online Precision Estimation, SCOPE），这是一个轻量级模块，能够为扩散轨迹模型增强控制准备的不确定性。SCOPE通过提炼评分曲率信息，学习围绕每个名义轨迹的结构化精度矩阵，并生成低开销的校准高斯管道，而无需重复的蒙特卡洛采样。这些管道提供每个时间步的协方差估计，可用于移动代理的预测占用以及机器人控制的自适应探索指导。我们在行人预测、拥挤导航、Maze2D 控制和现实世界的 Franka Panda 操作中评估了 SCOPE。通过这些设置，SCOPE 提供了快速的不确定性估计，从而提高了闭环性能。项目页面：https://zackaxue.github.io/SCOPE-project-page/

View on arXiv Download PDF AI Translation

cs.RO / 88 / 2610.12432

FAITH: Feasibility-Aware Safety-Filtered RL for High-Dimensional Systems

FAITH：面向可行性的安全过滤强化学习框架用于高维系统

Zhang, Songyuan, Singh, Baljeet, Kaingade, Sarthak Ranjeet, Fan, Chuchu, Trinh, Bryan

Abstract

Safe reinforcement learning commonly places safety and task performance in the same policy objective, where they can introduce competing updates. Safety filters separate them at action execution, but classical designs require an analytic safety function and dynamics model, and standard minimal-intervention filters are myopic to long-horizon task return because they minimize only instantaneous action deviation. Hard projections are also undefined when no safe action exists. We present FAITH, a feasibility-aware, model-free framework that approximates the optimal state-action safety value and amortizes minimal-intervention filtering with a feedforward network. The task policy optimizes the task return through the filtered dynamics, which recovers the feasible constrained problem without a competing safety term in the task-policy update. When no action satisfies the learned safety condition, the same filter approaches the action with minimum predicted peak harm. On a double integrator example and a Safety Gym environment, FAITH achieves the highest return among methods with no feasible-start violations and matches the lowest harm from infeasible starts. On a 29-DoF humanoid, it reaches a 99.95% safety rate while retaining 97% of the unfiltered return in Walking-Avoid, and obtains the highest measured safety rate in Push-Avoid by learning to sacrifice balancing and fall away from the protected region. The same policies are also demonstrated on a real-world Unitree G1 humanoid.

Chinese Translation

安全强化学习通常将安全性和任务性能置于同一策略目标中，这可能导致相互竞争的更新。安全过滤器在动作执行时将二者分开，但经典设计需要解析的安全函数和动态模型，而标准的最小干预过滤器对长期任务回报是短视的，因为它们仅最小化瞬时动作偏差。当不存在安全动作时，硬投影也是未定义的。我们提出了FAITH，一个面向可行性的无模型框架，能够近似最优的状态-动作安全值，并通过前馈网络实现最小干预过滤的摊销。任务策略通过过滤后的动态优化任务回报，从而恢复可行的约束问题，而在任务策略更新中不引入竞争的安全项。当没有动作满足学习到的安全条件时，相同的过滤器会接近预测的最小峰值伤害的动作。在双积分器示例和安全健身房环境中，FAITH在没有可行启动违规的情况下实现了最高回报，并与不可行启动的最低伤害相匹配。在一个29自由度的人形机器人上，它在Walking-Avoid任务中达到了99.95%的安全率，同时保留了97%的未过滤回报，并通过学习牺牲平衡能力和远离保护区域，在Push-Avoid任务中获得了最高的测量安全率。相同的策略也在现实世界的Unitree G1人形机器人上得到了验证。

View on arXiv Download PDF AI Translation

cs.RO / 89 / 2610.12435

VioLA: Learning Generalist Humanoid Control Policies from Human Data

VioLA：从人类数据中学习通用类人控制策略

Albaba, Mert, Beißwenger, Jens, Manasyan, Anna, Marta, Daniel, Black, Michael J., Brendel, Wieland, Krause, Andreas, Martius, Georg, Riedmiller, Martin

Abstract

Teaching a humanoid to follow instructions with its whole body runs into two obstacles. Its action space is large and tightly coupled: legs, arms, and fingers must move together while the robot keeps its balance, which makes joint-level actions hard to learn. And humanoid demonstrations are scarce, so current humanoid generalist policies do not follow new instructions out of the box and are fine-tuned on teleoperated demonstrations of each task before deployment. Human demonstrations exist in far larger numbers, but a person's motion is not a robot command. We remove both obstacles by changing what the generalist policy predicts. We introduce VioLA, a generalist humanoid policy that predicts body and hand motion latents instead of joint commands. A pretrained body- and hand-controller execute these latents on the robot. Their corresponding motion encoders map human and robot motion into the same latent spaces. A human recording is therefore labeled in the policy's action space, and the training demonstration pool contains 140.6 million frames, 93.2% of them human. As a result, VioLA follows locomotion instructions on the real robot zero-shot, without task-specific fine-tuning, reaching 100% success where GR00T N1.7 and $\Psi_0$ reach 16.7% and 0%, respectively. It also reaches 88.6% manipulation success without task-specific fine-tuning. The same approach works across two VLA and one world-action model backbones. A generalist policy trained on human demonstrations alone performs locomotion tasks on the real robot zero-shot. Code and checkpoints will be released.

Chinese Translation

教导类人机器人用全身遵循指令面临两个障碍。其动作空间庞大且紧密耦合：腿、手臂和手指必须协调移动，同时机器人保持平衡，这使得关节级别的动作难以学习。而且，类人演示稀缺，因此当前的类人通用策略无法直接遵循新的指令，必须在每个任务的遥控演示上进行微调后才能部署。尽管人类演示的数量要大得多，但一个人的动作并不是机器人命令。我们通过改变通用策略的预测内容来消除这两个障碍。我们引入了VioLA，这是一种通用类人策略，它预测身体和手部运动的潜变量，而不是关节命令。一个预训练的身体和手部控制器在机器人上执行这些潜变量。相应的运动编码器将人类和机器人运动映射到相同的潜在空间。因此，人类记录被标记在策略的动作空间中，训练演示池包含1.406亿帧，其中93.2%为人类数据。因此，VioLA在真实机器人上零-shot地遵循运动指令，无需特定任务的微调，成功率达到100%，而GR00T N1.7和$ ext{Ψ}_0$的成功率分别为16.7%和0%。在没有特定任务微调的情况下，它的操作成功率也达到了88.6%。相同的方法在两个VLA和一个世界动作模型骨干网络上均有效。仅在基于人类演示训练的通用策略在真实机器人上执行运动任务时实现了零-shot。代码和检查点将被发布。

View on arXiv Download PDF AI Translation

cs.RO / 90 / 2610.12440

Generative Neural Retargeting for Human-to-Robot Dexterous Manipulation

用于人机灵巧操作的生成神经重定向

Gao, Dechen, Yang, Yue, Abbatematteo, Ben, Godwin, Nathan, Wang, Pengcheng, Boldu, Roger, Man, Steven, Dou, Zhiyang, Qin, Chuan, Nakagome, Sho, Whitmire, Eric

Abstract

Human demonstrations are a scalable data source for learning dexterous manipulation, but the embodiment gap prevents human motion from being executed directly on robots. Inverse kinematics (IK) retargets human motion to robots efficiently but ignores dynamics, often producing infeasible motions. Reinforcement learning (RL) and sampling-based model predictive control (MPC) are commonly employed to yield dynamically feasible motions, but both are sample-inefficient and sensitive to hyperparameters. RL suffers from costly and unstable training and tedious reward engineering; MPC avoids policy optimization, yet retargets each trajectory in isolation, and solving one does not make the next easier. Sampling cost grows rapidly with dataset size and task difficulty. We hypothesize that dynamically feasible trajectories concentrate near a low-dimensional manifold shared across demonstrations, so that retargeting can be reduced to sampling from that manifold, conditioned on human motion, rather than solving a fresh optimization problem for every demonstration. We propose \textbf{Generative Neural Retargeting} (GNR), which uses a flow matching model to sample feasible trajectories. GNR outperforms MPC with only $8.5\%$ of the samples required by MPC, achieving a success rate of $56.20\%$ compared to $27.20\%$ for MPC. GNR can be used for scalable and efficient retargeting of large-scale, long-horizon, and millimeter precision human demonstrations: by applying GNR within a real-to-sim data engine, we produce a dexterous manipulation dataset with dense contact-force labels, spanning $223$k demonstrations and $3.3$k object geometries.

Chinese Translation

人类示范是学习灵巧操作的可扩展数据源，但体现差距阻止了人类动作在机器人上的直接执行。逆向运动学（IK）有效地将人类运动重定向到机器人，但忽略了动力学，常常产生不可行的动作。强化学习（RL）和基于采样的模型预测控制（MPC）通常用于生成动态可行的动作，但两者都存在样本效率低和对超参数敏感的问题。RL面临着昂贵且不稳定的训练以及繁琐的奖励工程；MPC避免了策略优化，但在孤立地重定向每条轨迹时，解决一个问题并不会使下一个问题变得更容易。随着数据集规模和任务难度的增加，采样成本迅速增长。我们假设动态可行的轨迹集中在一个低维流形上，该流形在示范之间共享，因此重定向可以简化为在该流形上进行采样，条件是人类运动，而不是为每个示范解决一个新的优化问题。我们提出了 extbf{生成神经重定向}（GNR），它使用流匹配模型来采样可行轨迹。GNR在所需样本仅为MPC的$8.5 ext{%}$的情况下，超越了MPC，成功率达到了$56.20 ext{%}$，而MPC的成功率为$27.20 ext{%}$。GNR可以用于大规模、有效的重定向长时间跨度和毫米级精度的人类示范：通过在真实到模拟的数据引擎中应用GNR，我们生成了一个具有密集接触力标签的灵巧操作数据集，涵盖了$223$k个示范和$3.3$k个物体几何形状。

View on arXiv Download PDF AI Translation

cs.RO / 91 / 2610.12457

SpatialHarness: Test-Time Spatial Scaffolding for Fine Robotic Manipulation

SpatialHarness：测试时空间支架用于精细机器人操作

Wang, Jiayu, Yu, Yue, Zhu, Bin, Yang, Zhiyao, Chen, Jingjing

Abstract

Frontier multimodal foundation models (e.g., GPT-6 Astra) have recently shown strong potential for direct robotic control, yet their performance on fine manipulation remains limited. We argue that an important source of failure is not necessarily insufficient policy capability, but insufficient spatial observability, where task-critical spatial relationships may be poorly revealed by the existing physical camera setup. We introduce SpatialHarness, a test-time embodied harness that provides test-time spatial scaffolding for fine robotic manipulation without policy fine-tuning or changes to the physical sensing setup. SpatialHarness maintains an online simulated scene synchronized with real-world execution, identifies task-critical spatial relationships, and renders complementary virtual views that expose them to a frozen multimodal policy. To keep the simulated scene aligned during interaction, we develop interaction-aware scene synchronization that distinguishes static, held, and transition modes. We evaluate SpatialHarness on four real-robot manipulation tasks spanning precise geometric alignment, object-relative placement, and articulated-object interaction. Using the same frozen GPT-6 Astra policy, SpatialHarness substantially improves task success, including from 26.7% to 66.7% on plug insertion and from 0% to 100% on Tower of Hanoi. These results indicate that improving spatial observability at test time can unlock fine-manipulation capabilities already present in strong multimodal foundation models. Project website: https://emilia113.github.io/SpatialHarness/.

Chinese Translation

前沿的多模态基础模型（例如，GPT-6 Astra）最近在直接机器人控制方面展现了强大的潜力，但其在精细操作上的表现仍然有限。我们认为，失败的重要来源不一定是策略能力不足，而是空间可观测性不足，其中任务关键的空间关系可能在现有的物理相机设置下未能得到良好揭示。我们提出了SpatialHarness，一种测试时具身支架，为精细机器人操作提供测试时空间支架，而无需策略微调或对物理传感设置进行更改。SpatialHarness维护与现实世界执行同步的在线模拟场景，识别任务关键的空间关系，并渲染补充的虚拟视图，以便将其暴露给冻结的多模态策略。为了在交互过程中保持模拟场景的对齐，我们开发了交互感知的场景同步，区分静态、持有和过渡模式。我们在四个真实机器人操作任务上评估SpatialHarness，这些任务涵盖精确几何对齐、相对物体放置和关节物体交互。使用相同的冻结GPT-6 Astra策略，SpatialHarness显著提高了任务成功率，包括插头插入从26.7%提高到66.7%，以及汉诺塔从0%提高到100%。这些结果表明，在测试时改善空间可观测性可以解锁强大多模态基础模型中已存在的精细操作能力。项目网站：https://emilia113.github.io/SpatialHarness/

View on arXiv Download PDF AI Translation

cs.RO / 92 / 2610.12465

A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control

平衡数据饮食：解决大规模强化学习在机器人控制中的探索瓶颈

Zhang, Octi, Castro, Mateo Guaman, Yin, Patrick, Dagnino, Ignacio, Gupta, Abhishek, Scalise, Rosario, Boots, Byron

Abstract

General-purpose robots must perform a wide range of tasks from agile locomotion to dexterous manipulation. While sim-to-real reinforcement learning (RL) has proven to be a useful tool for this goal, current RL pipelines depend on engineering-heavy, per-task structural priors such as shaped rewards and demonstrations. Recent work has shown that diverse simulator resets, combined with massively parallel simulation, can alleviate much of this engineering burden on several manipulation problems. However, we find that naively scaling this paradigm to more precise or dynamic problems remains non-trivial. While simulator resets can help with exploration, uniformly sampling over this distribution wastes a growing fraction of learning experience on task configurations the policy has already mastered or cannot yet attempt. This makes it challenging to see the expected benefits of scaling parallel environments for RL, since much of the learning signal in a batch is wasted during learning. To mitigate this, we introduce Success Guided Sampling (SGS), a simple adaptive sampler that concentrates RL training on task configurations around the frontier of the policy's capabilities. Doing so allows large-scale simulated RL to make the most out of the experience in a batch, enabling much more effective scaling to large-scale parallel simulation. Across experiments using up to $2^{20}$ (over one million) parallel environments, SGS enables RL to solve challenging multi-terrain quadruped locomotion and contact-rich assembly tasks that prior methods fail to solve. Finally, we distill the learned manipulation policies into RGB-based policies and demonstrate zero-shot transfer to several challenging assembly tasks on real hardware. Project website: https://sgs-rl.github.io/.

Chinese Translation

通用机器人必须执行从灵活运动到灵巧操作的广泛任务。尽管模拟到现实的强化学习（RL）已被证明是实现这一目标的有效工具，但当前的RL流程依赖于工程密集型的每任务结构先验，例如设计奖励和示范。近期研究表明，结合多样化的模拟器重置和大规模并行模拟，可以减轻在多个操作问题上的工程负担。然而，我们发现简单地将这一范式扩展到更精确或动态的问题仍然不是一件简单的事。虽然模拟器重置可以帮助探索，但在这一分布上均匀采样会浪费越来越多的学习经验在策略已经掌握或尚未尝试的任务配置上。这使得在RL中看到扩展并行环境的预期效果变得具有挑战性，因为批次中的大量学习信号在学习过程中被浪费。为了解决这个问题，我们引入了成功引导采样（Success Guided Sampling, SGS），这是一种简单的自适应采样器，集中RL训练于策略能力边界附近的任务配置。这样做使得大规模模拟RL能够充分利用批次中的经验，从而更有效地扩展到大规模并行模拟。在使用多达$2^{20}$（超过一百万）并行环境的实验中，SGS使得RL能够解决以往方法无法解决的具有挑战性的多地形四足运动和接触丰富的组装任务。最后，我们将学习到的操作策略提炼为基于RGB的策略，并展示了在真实硬件上对多个具有挑战性的组装任务的零样本迁移。项目网站：https://sgs-rl.github.io/

View on arXiv Download PDF AI Translation

cs.RO / 93 / 2610.12467

CSF: Contextual Safety Filtering for Motion Generators

CSF：运动生成器的上下文安全过滤

Yang, Lizhi, Hou, Yiling, Tang, Yao, Li, Junheng, Weng, Daniel, Werner, Blake, Ames, Aaron D.

Abstract

Text-conditioned motion generators produce trackable whole-body motion, but they have no notion of scene-dependent safety: the same action may target an object or a person. Existing safeguards either inspect the prompt, require labeled motion data, or enforce geometric constraints; therefore, they do not directly account for how scene context changes a motion's meaning. We introduce contextual safety filtering (CSF), a training-free filter that grounds natural-language safety rules in safe and unsafe reference trajectories produced by the generator. For each active rule, safe and unsafe reference trajectories define an affine safety value that a safe reference tracking CBF-QP enforces. Across four pretrained generators with different architectures, CSF activates the intended rules in all explicit and scene-triggered unsafe cases and reduces the danger-event rate by up to 90%, while preserving 88-100% of benign motions. We demonstrate the complete system on a real-world Unitree G1, where it successfully prevents unsafe motions in a variety of scenarios, including interactions with humans and objects.

Chinese Translation

文本条件的运动生成器能够产生可追踪的全身运动，但它们并不考虑场景依赖的安全性：同一个动作可能针对物体或人。现有的安全保障措施要么检查提示，要么需要标记的运动数据，或者强制执行几何约束；因此，它们并未直接考虑场景上下文如何改变运动的含义。我们提出了上下文安全过滤（CSF），这是一种无训练的过滤器，将自然语言安全规则与生成器产生的安全和不安全参考轨迹相结合。对于每个激活的规则，安全和不安全的参考轨迹定义了一个仿射安全值，该值由安全参考跟踪的 CBF-QP 强制执行。在四个具有不同架构的预训练生成器中，CSF 在所有显式和场景触发的不安全案例中激活了预期的规则，并将危险事件率降低了多达 90%，同时保留了 88-100% 的良性运动。我们在真实世界的 Unitree G1 上演示了完整的系统，成功地在多种场景中防止了不安全的运动，包括与人类和物体的互动。

View on arXiv Download PDF AI Translation

cs.RO / 94 / 2610.12468

DreamTrue: Action-Faithful Robot World Model with Counterfactual Post-Training

DreamTrue：具有反事实后训练的忠实动作机器人世界模型

Li, Junyan, Li, Ruizhi, Liu, Yu, Liu, Xiangshuo, Sun, Mingchao, Pan, Hongyu, Xu, Mu, Fan, Lue, Zhang, Zhaoxiang

Abstract

We present DreamTrue, a multi-view, cross-embodiment robot world model for action-faithful and physically plausible video prediction. Training such a model on existing robot datasets faces two obstacles: imprecise calibration can impair action following, while limited coverage of unsuccessful interactions can bias predictions toward successful outcomes. To improve action following across embodiments, we render action trajectories into image-space conditions and introduce offline geometric calibration to align these conditions with the target videos. To broaden interaction coverage, we introduce counterfactual post-training, modifying recorded action trajectories and generating future videos under a wider range of actions and contact configurations. To provide feedback on these predictions without paired ground-truth futures, we construct a human-annotated video dataset covering robot, object, and interaction defects and use it to train an embodied video reward model. Its scores guide reinforcement-learning post-training toward more physically plausible interaction outcomes. On AgiBot, DreamTrue attains state-of-the-art action following, while reducing the human-assessed interaction defect rate from from 48.12% to 6.25%. Notably, our model ranks first in the world model track of the AgiBot World Challenge 2026. The project page can be found at https://brave-eai.github.io/DreamTrue.

Chinese Translation

我们提出了DreamTrue，这是一种多视角、跨体现的机器人世界模型，用于忠实于动作和物理上合理的视频预测。在现有机器人数据集上训练这样的模型面临两个障碍：不精确的校准可能会损害动作跟随，而有限的失败交互覆盖可能会使预测偏向成功结果。为了改善跨体现的动作跟随，我们将动作轨迹渲染为图像空间条件，并引入离线几何校准，以将这些条件与目标视频对齐。为了扩大交互覆盖范围，我们引入了反事实后训练，修改记录的动作轨迹并在更广泛的动作和接触配置下生成未来视频。为了在没有配对真实未来的情况下对这些预测提供反馈，我们构建了一个涵盖机器人、物体和交互缺陷的人类标注视频数据集，并利用它来训练一个具身视频奖励模型。该模型的评分指导强化学习后训练朝向更物理上合理的交互结果。在AgiBot上，DreamTrue实现了最先进的动作跟随，同时将人类评估的交互缺陷率从48.12%降低到6.25%。值得注意的是，我们的模型在2026年AgiBot世界挑战赛的世界模型赛道中排名第一。项目页面可在 https://brave-eai.github.io/DreamTrue 找到。

View on arXiv Download PDF AI Translation

cs.RO / 95 / 2610.12470

Dex-One2Many: Learning Dexterous Manipulation from a Single Human Demonstration

Dex-One2Many：从单一人类示范中学习灵巧操作

Lee, Jusuk, Kim, Sungha, Park, Yeonsoo, Cheon, Jonguk, Jung, Yoonkyo, You, Yongjun, Kim, H. Jin, Huang, Jia-Bin, Huang, Furong, Jang, Youngseok, Lee, Seungjae

Abstract

While learning dexterous manipulation from a single human video offers a promising alternative to costly robot demonstrations, many recent methods predominantly imitate demonstrated motions. Such strict motion matching often limits generalization to initial object poses, goal poses, and grasps not shown in the video. Alternatively, discovering a policy via reinforcement learning (RL) allows for broad generalization, but without prior guidance, it struggles with high-dimensional exploration in complex, multi-stage tasks. To address these coupled generalization and exploration challenges, we present Dex-One2Many, a real-to-sim-to-real framework that learns a generalizable dexterous manipulation policy from a single human video. Our key insight is to abstract the video into sequential scene graphs that guide RL, enabling efficient exploration while preserving broad generalizability. The graphs serve as generative constraints for sampling diverse reset states and provide dense rewards for each stage. Because the graphs constrain relations rather than exact poses, these reset states cover object poses and grasps beyond the video, while initializing each stage from them with dense rewards keeps exploration short and guided. Trained entirely in simulation, Dex-One2Many transfers zero-shot to a real multi-fingered hand. Across five tool-use and manipulation tasks, Dex-One2Many exceeds baselines by 6.5% in seen configurations, while its robust generalization widens this gap to 71% in unseen scenarios.

Chinese Translation

从单一人类视频中学习灵巧操作为昂贵的机器人示范提供了一种有前景的替代方案，但许多近期的方法主要是模仿示范的动作。这种严格的动作匹配往往限制了对初始物体姿态、目标姿态和视频中未展示的抓取方式的泛化。相反，通过强化学习（Reinforcement Learning, RL）发现策略可以实现广泛的泛化，但在没有先前指导的情况下，它在复杂的多阶段任务中面临高维探索的挑战。为了解决这些耦合的泛化和探索问题，我们提出了Dex-One2Many，一个从单一人类视频中学习可泛化灵巧操作策略的真实-模拟-真实框架。我们的关键见解是将视频抽象为顺序场景图，以指导RL，从而实现高效探索，同时保持广泛的泛化能力。这些图作为生成约束，用于采样多样的重置状态，并为每个阶段提供密集的奖励。由于这些图约束的是关系而非精确姿态，这些重置状态覆盖了视频之外的物体姿态和抓取方式，同时从这些状态初始化每个阶段并保持密集奖励，使探索过程简短且有指导性。Dex-One2Many完全在模拟环境中训练，能够零-shot迁移到真实的多指手。在五个工具使用和操作任务中，Dex-One2Many在已见配置中超过基线6.5%，而其强大的泛化能力使这一差距在未见场景中扩大至71%。

View on arXiv Download PDF AI Translation

计算机视觉 (Computer Vision)

169

cs.CV / 1 / 2610.10563

SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows

SLVR：通过类人推理流程进行结构化潜在视觉推理

Gao, Albert, Xue, Bing, Zanette, Andrea

Abstract

Multimodal large language models (MLLMs) often answer visual reasoning questions by relying on linguistic priors rather than task-relevant visual evidence. Textual chain-of-thought reasoning can partially mitigate this issue by encouraging models to decompose visual questions into intermediate evidence-seeking steps, but generating these steps autoregressively increases inference cost. Latent reasoning avoids explicit rationale generation, but existing approaches provide limited control over what intermediate states encode, making it difficult to impose separate supervision for planning, grounding, and evidence selection. We propose Structured Latent Visual Reasoning (SLVR), a training framework that bridges explicit chain-of-thought and latent reasoning by organizing multimodal reasoning into typed latent stages for planning, grounding, evidence selection, and reasoning integration. SLVR first trains the model to rely on the image by masking answer-revealing text and contrasting the correct answer with visually plausible distractors. It then organizes reasoning into latent stages for planning, grounding, evidence selection, and integration, supervising each stage with the corresponding signal: plans, boxes, visual evidence, and final rationales. This gives latent reasoning an explicit functional structure while avoiding generated textual chains at inference time. Built on Qwen2.5-VL-7B, SLVR improves consistently across multimodal reasoning benchmarks, with absolute gains of +9.4 on MMVP and +14.2 on BLINK Relation, as well as improvements on V*, MathVista, and ChartQA. These results suggest that structured latent supervision can improve fine-grained visual reasoning without the decoding overhead of textual CoT. Project page is available \href{https://bogao-code.github.io/SLVR/}{here}.

Chinese Translation

多模态大型语言模型（MLLMs）在回答视觉推理问题时，往往依赖语言先验而非与任务相关的视觉证据。文本链式思维推理可以部分缓解这一问题，通过鼓励模型将视觉问题分解为中间证据寻求步骤，但自回归生成这些步骤会增加推理成本。潜在推理避免了显式推理生成，但现有方法对中间状态的编码提供了有限的控制，使得难以对规划、基础和证据选择施加单独的监督。我们提出了结构化潜在视觉推理（SLVR），这是一种训练框架，通过将多模态推理组织为类型化的潜在阶段，连接显式链式思维和潜在推理，以进行规划、基础、证据选择和推理整合。SLVR首先训练模型依赖于图像，通过遮蔽揭示答案的文本并将正确答案与视觉上合理的干扰项进行对比。然后，它将推理组织为规划、基础、证据选择和整合的潜在阶段，并用相应的信号对每个阶段进行监督：计划、框、视觉证据和最终推理。这为潜在推理提供了显式的功能结构，同时避免了推理时生成文本链。基于Qwen2.5-VL-7B，SLVR在多模态推理基准测试中持续改善，在MMVP上获得了+9.4的绝对增益，在BLINK Relation上获得了+14.2的增益，并在V*、MathVista和ChartQA上也有所改善。这些结果表明，结构化潜在监督可以在不增加文本链解码开销的情况下改善细粒度视觉推理。项目页面可在此处查看。

View on arXiv Download PDF AI Translation

cs.CV / 2 / 2610.10607

A Camera-Native Stereo VR180 Dataset

一种相机原生立体VR180数据集

Lu, Linxuan

Abstract

Immersive VR180 video is increasingly produced with professional stereo fisheye cameras, yet public VR180 research resources are mostly collected from online platforms such as YouTube: already stitched, projected and compressed by unknown pipelines, and without lens calibration. We present a firsthand-captured stereo VR180 dataset recorded with two Blackmagic URSA Cine Immersive cameras. It contains 1,211 samples -- 636 stereo video clips (2,220.8 s, mostly 90 fps) and 575 stereo stills -- each released as camera-native Blackmagic RAW, separate-eye native fisheye HEVC (8160x7200 per eye) and half-equirectangular HEVC (7200x7200 per eye), together with the factory lens calibration, portable fisheye/half-equirectangular conversion tools and AI-generated scene and visual-challenge annotations. Re-encoding the released fisheye and half-equirectangular renders with x265 over 24 clips, both eyes, four rate points and nine viewing directions, native-fisheye coding needed more bitrate than half-equirectangular coding at equal viewport quality for all 24 clips (median +38%), in every part of the field of view. Data: https://huggingface.co/datasets/lulinxuan/VR180 ; code: https://github.com/lulinxuan/vr180-dataset-tools

Chinese Translation

沉浸式VR180视频越来越多地使用专业的立体鱼眼相机进行制作，然而公共的VR180研究资源大多来自YouTube等在线平台，这些资源已经经过未知的处理流程进行了拼接、投影和压缩，并且没有进行镜头校准。我们呈现了一个首次捕获的立体VR180数据集，该数据集使用两台Blackmagic URSA Cine Immersive相机录制。它包含1211个样本——636个立体视频片段（总时长2220.8秒，主要为90 fps）和575个立体静态图像——每个样本以相机原生的Blackmagic RAW格式、分眼原生的鱼眼HEVC格式（每只眼8160x7200）和半等距HEVC格式（每只眼7200x7200）发布，同时提供了工厂镜头校准、便携式鱼眼/半等距转换工具以及AI生成的场景和视觉挑战注释。对发布的鱼眼和半等距渲染进行x265重新编码，涵盖24个片段、双眼、四个比特率点和九个观看方向，结果显示在相同视口质量下，原生鱼眼编码所需的比特率比半等距编码高出38%（中位数），在视野的每个部分均是如此。数据链接: https://huggingface.co/datasets/lulinxuan/VR180 ; 代码链接: https://github.com/lulinxuan/vr180-dataset-tools

View on arXiv Download PDF AI Translation

cs.CV / 3 / 2610.10703

Seeing Through the Glare: A Multi-Source Benchmark and Ocular-Adaptive Pixel MeanFlow for Eyeglass Reflection Removal

透过眩光：用于眼镜反射去除的多源基准和眼适应像素均值流

Liu, Tao, Pang, Youwei, Zhou, Kailai, Zuo, Jiaming, Liu, Hanqi, Ji, Wei, Jiang, Peng-Tao, Liu, Xiaofeng, Lin, Weisi, Zhao, Xiaoqi

Abstract

Eyeglass reflection removal is important across smartphone imaging, video conferencing, and other face-centric visual applications. The task is challenging because reflections range from mild photometric contamination to severe ocular occlusion, requiring selective correction and plausible reconstruction without altering identity or natural appearance. Existing datasets cover limited reflection conditions, constraining generalization to complex real-world scenes and systematic evaluation. We introduce \textbf{OcuBench}, a multi-source benchmark comprising 10,280 controllable synthetic pairs, 732 real-input pseudo-pairs, and 458 independent real-world test images, supporting both paired evaluation and assessment beyond generated supervision. We further propose \textbf{OcuFlow}, an ocular-adaptive pixel MeanFlow (pMF) framework for efficient, detail-preserving restoration. It combines geometry-adaptive representation with one-step pMF to focus reconstruction on reflection-obscured ocular regions, together with native-resolution frequency-preserving synthesis to retain reliable observed details. Experiments across diverse reflection conditions demonstrate that OcuFlow achieves consistent advantages in reflection removal quality, ocular fidelity, and efficiency. In a blind user study, it receives $67.32\%$ of selections, $6.2\times$ the next-best share. Both the code and dataset will be released.

Chinese Translation

眼镜反射去除在智能手机成像、视频会议及其他以面部为中心的视觉应用中至关重要。该任务具有挑战性，因为反射范围从轻微的光度污染到严重的眼部遮挡，要求在不改变身份或自然外观的情况下进行选择性修正和合理重建。现有数据集覆盖的反射条件有限，限制了对复杂真实场景的泛化能力和系统评估。我们引入了 extbf{OcuBench}，一个包含10,280对可控合成图像、732个真实输入伪对和458个独立真实世界测试图像的多源基准，支持配对评估和超越生成监督的评估。我们进一步提出了 extbf{OcuFlow}，一种眼适应像素均值流（pMF）框架，用于高效且保留细节的恢复。它结合了几何适应表示和一步式pMF，重点对反射遮挡的眼部区域进行重建，同时采用原生分辨率的频率保留合成，以保留可靠的观察细节。在多种反射条件下的实验表明，OcuFlow在反射去除质量、眼部保真度和效率方面均表现出一致的优势。在一次盲测用户研究中，其选择率达到$67.32 ext{ extperthousand}$，是第二高选择率的$6.2 imes$。代码和数据集将会发布。

View on arXiv Download PDF AI Translation

cs.CV / 4 / 2610.10722

LinSlot: Exploiting Linear Representation hypothesis for unsupervised attribute discovery from slot based object representation

LinSlot：利用线性表示假设进行基于槽的对象表示的无监督属性发现

Gandhi, Sanket, Giri, Utkarsh, Subramanium, Varun, Paul, Rohan, Singla, Parag

Abstract

This paper studies the problem of learning disentangled representations of objects and their attributes from raw, unstructured image data. Slot-based methods have shown considerable success in unsupervised learning of object representations from images. Block-slot attention-based methods extend this framework to attribute representations by assuming a uniform factorization of object representations into attributes, which may be suboptimal and consequently limit the quality of the learned representations. We therefore investigate a framework for jointly discovering object and attribute representations. Our key contribution is leveraging the Linear Representation Hypothesis (LRH), which postulates that composable concepts can be represented as linearly additive subspaces in slot representations. Based on this insight, we propose a probabilistic model connecting images, slots (objects), and blocks (attributes). We present an architecture that leverages block attention to connect attribute representations to slots and incorporates LRH in both object and attribute representation spaces. This architecture effectively optimizes the Evidence Lower Bound (ELBO) of the proposed graphical model. Our experiments demonstrate (i) effective discovery of disentangled object and attribute representations, (ii) empirical evidence for LRH in slot space, and (iii) the ability to perform image editing owing to the disentangled and interpretable nature of the learned representations. Our experiments on multiple datasets demonstrate improvements in DCI scores over state-of-the-art methods.

Chinese Translation

本文研究了从原始非结构化图像数据中学习对象及其属性的解耦表示的问题。基于槽的方法在从图像中无监督学习对象表示方面取得了显著成功。基于块-槽注意力的方法通过假设对象表示均匀分解为属性，扩展了这一框架以表示属性，但这种假设可能是次优的，从而限制了学习表示的质量。因此，我们探讨了一个联合发现对象和属性表示的框架。我们的关键贡献是利用线性表示假设（Linear Representation Hypothesis, LRH），该假设认为可组合的概念可以在槽表示中作为线性可加子空间表示。基于这一见解，我们提出了一个将图像、槽（对象）和块（属性）连接起来的概率模型。我们提出了一种架构，利用块注意力将属性表示与槽连接，并在对象和属性表示空间中结合LRH。该架构有效优化了所提图形模型的证据下界（Evidence Lower Bound, ELBO）。我们的实验表明：（i）有效发现解耦的对象和属性表示；（ii）在槽空间中对LRH的实证证据；（iii）由于学习表示的解耦和可解释性，能够进行图像编辑。我们在多个数据集上的实验显示，与最先进的方法相比，DCI评分有所提高。

View on arXiv Download PDF AI Translation

cs.CV / 5 / 2610.10759

MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting

MESSENGER：通过自回归下一帧预测的记忆增强序列场景流估计

Liu, Jiuming, Li, Jianing, Liu, Mengmeng, He, Hongyang, Wang, Hesheng, Kristensson, Per Ola

Abstract

Scene flow can capture low-level 3D motion displacements in dynamic scenarios. Early pairwise estimators relying on instantaneous two-frame motion lack long-term temporal correlation and also struggle with poor extrapolation ability in future prediction. Although some recent methods attempt to explore multi-frame scene flow estimation in a sequence-to-sequence manner, they typically suffer from heavy computational overhead with increasing input frames and long-horizon prediction degradation due to ineffective motion propagation. To address these problems, we propose a novel memory-enhanced sequential scene flow pipeline, called MESSENGER. To sufficiently mine long-term temporal dependencies naturally within consecutive sequences, a memory buffer is designed by explicitly storing multiple history flow estimates and latent states. For each input frame, the temporally stored flows and states are correlated and retrieved to predict the current initialized flow in a next-frame forecasting manner. Furthermore, we develop an uncertainty-aware reweighting module to filter unreliable retrievals and mitigate accumulated errors. Extensive experiments on nuScenes and Argoverse 2 demonstrate state-of-the-art performance of our MESSENGER, reducing EPE3D by 71.6% on nuScenes and 67.7% on Argoverse 2 in long-horizon future extrapolation. This superiority can be attributed to our designed autoregressive forecasting paradigm, which naturally forces the network to progressively learn the next-frame distribution based on history observations. Code will be released at https://github.com/liujiuming123/Messenger.

Chinese Translation

场景流可以捕捉动态场景中的低级三维运动位移。早期的成对估计器依赖于瞬时的两帧运动，缺乏长期的时间相关性，并且在未来预测中也面临较差的外推能力。尽管一些近期方法尝试以序列到序列的方式探索多帧场景流估计，但它们通常在输入帧数增加时遭遇较大的计算开销，并且由于运动传播效率低下，长时间预测性能下降。为了解决这些问题，我们提出了一种新颖的记忆增强序列场景流管道，称为 MESSENGER。为了充分挖掘连续序列中的长期时间依赖性，我们设计了一个记忆缓冲区，通过显式存储多个历史流估计和潜在状态。对于每个输入帧，时间存储的流和状态被关联并检索，以预测当前初始化的流，采用下一帧预测的方式。此外，我们开发了一个不确定性感知重加权模块，以过滤不可靠的检索结果并减轻累积误差。在 nuScenes 和 Argoverse 2 上进行的广泛实验表明，我们的 MESSENGER 达到了最先进的性能，在长时间未来外推中，nuScenes 上的 EPE3D 降低了 71.6%，Argoverse 2 上降低了 67.7%。这种优越性归因于我们设计的自回归预测范式，它自然地迫使网络基于历史观察逐步学习下一帧的分布。代码将发布在 https://github.com/liujiuming123/Messenger。

View on arXiv Download PDF AI Translation

cs.CV / 6 / 2610.10760

DOGS: Design-Space Sampling for Prompt-Driven Logo Generation

DOGS：基于提示驱动的标志生成设计空间采样

Zou, Ganyu, Dai, Chen, Self, Nathan, Piper, Kevin, Seethiraju, Ramachandra Rao, Shyamsunder, Karthik, Lu, Chang-Tien, Ramakrishnan, Naren

Abstract

Prompt optimization for text-to-image (T2I) generation has been pursued almost entirely as text rewriting, in which a short user brief is expanded into a longer, model-preferred token sequence. We argue that such a language-space formulation is ill-suited to structured visual design tasks such as logo creation, where a one-line brief leaves most design decisions unspecified. These decisions depend on relational priors that a linear sequence cannot encode, and they leave an uncontrolled channel through which protected marks may be reproduced. We therefore recast logo prompting as sampling within a structured design space, and instantiate this idea as DOGS (Design-space prompting with an Originality-aware GFlowNet Sampler). From a large corpus of real-world logos, we mine a typed, graph-structured design grammar whose edges record empirical co-occurrence. A GFlowNet sampler then generates design graphs with probability proportional to a terminal reward that combines recognizability, aesthetics, and corpus-relative originality. Every slot draws only from a closed design-level vocabulary, and any infringement-inducing or harmful token is removed during parsing. The originality reward further penalizes proximity to existing logos, thereby incorporating infringement avoidance into the method by construction. On two open-source renderers and against nine baselines, DOGS produces logos that are more recognizable and aesthetic, substantially more diverse, and far less prone to trademark infringement.

Chinese Translation

文本到图像（T2I）生成的提示优化几乎完全被视为文本重写，其中用户的简短描述被扩展为更长的、模型偏好的标记序列。我们认为，这种语言空间的表述不适合结构化视觉设计任务，例如标志创建，因为一行简要说明留下了大多数设计决策未被指定。这些决策依赖于线性序列无法编码的关系先验，并且它们留下了一个不受控制的渠道，通过该渠道可能会复制受保护的商标。因此，我们将标志提示重新表述为在结构化设计空间内进行采样，并将这一思想具体化为DOGS（带有原创性意识的GFlowNet采样器的设计空间提示）。我们从大量真实世界的标志中挖掘出一种类型化的图结构设计语法，其边缘记录了经验共现。然后，GFlowNet采样器生成的设计图的概率与结合了可识别性、美学和相对于语料库的原创性的终端奖励成正比。每个槽位仅从封闭的设计级词汇中抽取，任何可能引发侵权或有害的标记在解析过程中都会被移除。原创性奖励进一步惩罚与现有标志的接近，从而通过构建将避免侵权纳入方法中。在两个开源渲染器上与九个基线进行比较，DOGS生成的标志在可识别性和美学上更佳，具有显著更高的多样性，并且侵权的可能性大大降低。

View on arXiv Download PDF AI Translation

cs.CV / 7 / 2610.10782

VICO: Visual Environments Co-Evolving for Vision-Language Model Reasoning

VICO：视觉环境与视觉-语言模型推理的共同演化

Lu, Meng, Zhu, Ligeng, Xiao, Olivia, Zhuang, Yuchen, Wang, Zihan, Wu, Kuncheng, Liu, Bangya, Wang, Yu, Fleming, Charles, Shi, Wenqi, Wang, Xuan

Abstract

Reinforcement learning with verifiable rewards (RLVR) has become a standard recipe for post-training vision-language models (VLMs), but it typically assumes a static training environment. As the actor improves, fixed tasks drift out of its learning frontier: many become trivial, others remain unsolvable; and the learning signal collapses. We argue that VLM post-training should evolve the visual environment alongside the actor, not just the actor itself. We propose VICO, a co-evolutionary framework in which an actor and an Environment-as-Rewriter (EnvRewriter) are trained jointly: the EnvRewriter edits verifiable image-side structures, such as scene graphs, chart tables, or protected region masks, and re-renders them to produce label-valid training samples whose difficulty is calibrated to the actor's current ability through a pass-rate-based reward. This loop continuously realigns task difficulty with actor capability without any additional human annotation. Across nine multimodal benchmarks spanning mathematical reasoning and visually grounded understanding, VICO-8B improves over its base model by up to +5.0% on out-of-domain tasks, surpasses the strongest self-evolution and text-editing co-evolution baselines by +4.3% and +8.4% respectively, and stays comparable to chart-specialized RLVR methods using 16-160 times fewer labeled samples. By shifting from human-labeled supervision to image-editing co-evolution, VICO offers a scalable path beyond static-corpus RLVR for visual reasoning.

Chinese Translation

可验证奖励的强化学习（RLVR）已成为后训练视觉-语言模型（VLMs）的标准方法，但它通常假设训练环境是静态的。随着行为者的改进，固定任务逐渐超出其学习边界：许多任务变得微不足道，其他任务则仍然无法解决；学习信号因此崩溃。我们认为，VLM的后训练应与行为者共同演化视觉环境，而不仅仅是行为者本身。我们提出了VICO，一个共同演化框架，其中行为者和环境重写器（Environment-as-Rewriter, EnvRewriter）共同训练：EnvRewriter编辑可验证的图像侧结构，如场景图、图表表格或受保护区域掩码，并重新渲染它们以生成标签有效的训练样本，其难度通过基于通过率的奖励与行为者当前能力进行校准。这个循环不断地将任务难度与行为者能力对齐，而无需额外的人类标注。在涵盖数学推理和视觉基础理解的九个多模态基准测试中，VICO-8B在领域外任务上比其基础模型提高了最多5.0%，超越了最强的自我演化和文本编辑共同演化基线，分别提高了4.3%和8.4%，并且在使用16-160倍更少标注样本的情况下，与图表专用的RLVR方法保持了相当的效果。通过从人类标注监督转向图像编辑共同演化，VICO为超越静态语料RLVR的视觉推理提供了一条可扩展的路径。

View on arXiv Download PDF AI Translation

cs.CV / 8 / 2610.10823

Velocity Scaling in Flow Matching

流匹配中的速度缩放

Saied, Youssef, Fleuret, François

Abstract

Scaling a learned flow-matching velocity field $v_\theta$ by a gain $\gamma(t)$ was recently shown to greatly improve generation quality. Prior work argued that velocity fields trained with mean-squared error (MSE) systematically underestimate velocity magnitude and that scaling corrects this error. We show that MSE training does not create a velocity-magnitude deficit. We find instead that velocity scaling reduces population time lag: sampled states at model time $t$ resemble training states from an earlier time. Velocity scaling and moving model time back are two ways to address this population time lag. Across architectures and model sizes, measuring population time lag and using it to select a gain greatly improves generation quality, reducing FID from 28.0 to 12.2 (estimated by linear interpolation between FID measurements at neighboring gains) on ImageNet-256 at NFE 25 without guidance.

Chinese Translation

最近的研究表明，通过增益 $eta(t)$ 缩放学习到的流匹配速度场 $v_ heta$ 可以显著提高生成质量。之前的工作认为，使用均方误差（MSE）训练的速度场系统性地低估了速度幅度，而缩放可以纠正这一错误。我们表明，MSE训练并不会造成速度幅度的不足。相反，我们发现速度缩放减少了群体时间滞后：在模型时间 $t$ 采样的状态与早期时间的训练状态相似。速度缩放和将模型时间向回移动是解决这一群体时间滞后的两种方法。在不同的架构和模型规模中，测量群体时间滞后并利用其选择增益显著提高了生成质量，在 NFE 25 的 ImageNet-256 上将 FID 从 28.0 降低到 12.2（通过在相邻增益的 FID 测量之间进行线性插值估算）而无需指导。

View on arXiv Download PDF AI Translation

cs.CV / 9 / 2610.10859

Enabling Preference-driven Unlearning in Few-step Distilled Text-to-Image Diffusion Models

在少步蒸馏文本到图像扩散模型中实现基于偏好的遗忘

Patel, Gaurav, Fang, Jun, Steeg, Greg Ver, Qiu, Qiang, Sripada, Sravan

Abstract

Text-to-image diffusion models are increasingly distilled into few-step variants and being deployed to enable fast inference. However, their ability to generate harmful or undesired content poses significant safety risks. Data-driven unlearning methods suppress targeted generations by fine-tuning model weights using specialized unlearning objectives. Crucially, these objectives implicitly rely on multi-step denoising dynamics, an assumption that breaks down for few-step distilled (FSD) models, resulting in ineffective forgetting. Furthermore, performing unlearning on the non-distilled base model and subsequently re-distilling it to obtain an unlearned FSD model incurs substantial computational and time overhead, making it impractical in many settings. Hence, we address this limitation with a preference-driven unlearning framework that revisits Direct Preference Optimization (DPO) for diffusion models. We show that standard DPO and its unlearning derivatives, formulated around noise-prediction error, transfer poorly to FSD models due to their altered generation dynamics. To overcome this, we introduce a modified preference optimization formulation explicitly aligned with the few-step generation properties, enabling direct concept removal in FSD models while preserving few-step efficiency and maintaining strong retention of desirable (non-targeted) capabilities. We evaluate our framework primarily on identity and NSFW (nudity) removal tasks and also extend our method to object-level unlearning. Extensive experiments demonstrate consistent and effective forgetting, and strong retention performance, establishing our method as a practical and principled solution for unlearning in FSD models.

Chinese Translation

文本到图像的扩散模型正越来越多地被蒸馏为少步变体，并被部署以实现快速推理。然而，它们生成有害或不希望内容的能力带来了显著的安全风险。数据驱动的遗忘方法通过使用专门的遗忘目标微调模型权重，从而抑制目标生成。关键是，这些目标隐含地依赖于多步去噪动态，这一假设在少步蒸馏（FSD）模型中失效，导致遗忘效果不佳。此外，在非蒸馏基础模型上执行遗忘，然后再重新蒸馏以获得一个已遗忘的FSD模型，会产生大量的计算和时间开销，使其在许多场景中不切实际。因此，我们通过一个基于偏好的遗忘框架来解决这一限制，该框架重新审视了扩散模型的直接偏好优化（Direct Preference Optimization, DPO）。我们展示了标准DPO及其围绕噪声预测误差制定的遗忘衍生物在FSD模型中转移效果不佳，因为它们的生成动态发生了变化。为了解决这个问题，我们引入了一种修改后的偏好优化公式，明确与少步生成特性对齐，使得在FSD模型中能够直接移除概念，同时保持少步效率并保持对期望（非目标）能力的强保留。我们主要在身份和NSFW（裸露）移除任务上评估我们的框架，并将我们的方法扩展到对象级遗忘。大量实验表明我们的方法在遗忘和保留性能上均表现出一致且有效的效果，确立了我们的方法作为FSD模型中遗忘的实用和原则性解决方案。

View on arXiv Download PDF AI Translation

cs.CV / 10 / 2610.10889

SPLIT-RL: Staged Perception-Language Reasoning Training with Claim-Level Advantages

SPLIT-RL：具有声明级优势的分阶段感知-语言推理训练

Kumar, Raja, Koner, Rajat, Chaudhry, Ritwick, Li, Zhuowei, Sankaran, Nishant, Xing, Yifan

Abstract

Vision-Language (VL) reasoning requires a model to both extract relevant and accurate information from an image (visual reasoning, VR), and to infer the answer from it (language reasoning, LR). Reinforcement learning with verifiable rewards typically trains both through a single chain-of-thought with a final-answer reward. This gives every CoT token the same sequence-level advantage, failing to distinguish capability specific errors. We propose SPLIT-RL, a staged post-training approach that trains VR and LR in disjoint phases. Because a group's rollouts differ along one capability at a time, the group-relative advantage isolates it, and each phase is optimized using phase-specific reward. We further introduce Claim-Level Advantage (CLA-GRPO), which decomposes VR-phase rollouts into atomic visual claims and provides a fine-grained advantage at claim level based on visual-type group formation. Although trained in two phases, trained policy is evaluated like GRPO model, with a single CoT call at inference time. Under this protocol, SPLIT-RL improves average accuracy over GRPO by 1.4-6.1 points across Qwen3-VL models from 2B to 30B-A3B and InternVL3.5-8B. Evaluating each capability using an oracle based diagnostic shows that answer-only GRPO leaves perception unchanged, whereas SPLIT-RL improves both VR and LR.

Chinese Translation

视觉-语言（VL）推理要求模型从图像中提取相关且准确的信息（视觉推理，VR），并从中推断答案（语言推理，LR）。使用可验证奖励的强化学习通常通过单一的思维链和最终答案奖励来训练这两者。这使得每个思维链（CoT）令牌具有相同的序列级优势，未能区分特定能力的错误。我们提出了SPLIT-RL，这是一种分阶段的后训练方法，在不相交的阶段中训练VR和LR。由于一个组的回滚在每次只沿着一项能力不同，因此组相对优势将其隔离，每个阶段使用特定阶段的奖励进行优化。我们进一步引入了声明级优势（CLA-GRPO），它将VR阶段的回滚分解为原子视觉声明，并基于视觉类型组的形成提供声明级的细粒度优势。尽管在两个阶段中进行训练，但训练后的策略在推理时像GRPO模型一样通过单个CoT调用进行评估。在这一协议下，SPLIT-RL在Qwen3-VL模型（从2B到30B-A3B）和InternVL3.5-8B上提高了平均准确率1.4-6.1个百分点。使用基于oracle的诊断评估每项能力显示，仅使用答案的GRPO未改变感知，而SPLIT-RL则同时提高了VR和LR。

View on arXiv Download PDF AI Translation

cs.CV / 11 / 2610.10893

Less from More: Reinforcing Sparse Video Reasoning from Dense References

从多到少：通过密集参考强化稀疏视频推理

Sun, Wenfang, Du, Yingjun, Snoek, Cees G. M.

Abstract

Video-language models commonly assume that more temporal observations lead to more reliable reasoning. We question this assumption and argue that the key challenge is not merely processing more video frames efficiently, but learning to reason reliably under limited temporal evidence. We propose SAVER, a dense-to-sparse post-training framework that uses dense video views as training-time references for sparse-frame inference. During reinforcement post-training, paired dense and sparse views are optimized with grounding rewards and a reliability-gated reference reward, encouraging sparse view predictions to preserve task-relevant temporal evidence. Notably, SAVER is trained only on 1,250 randomly sampled temporal grounding examples, without using any video question answering annotations. Across three temporal grounding benchmarks and six video question-answering benchmarks, SAVER consistently improves performance across frame budgets. In particular, SAVER can match or surpass dense-frame Qwen3.5 baselines while using substantially fewer frames. These results show that temporal grounding can serve as an effective evidence-localization proxy for learning sparse video reasoning that transfers to broader video understanding tasks.

Chinese Translation

视频语言模型通常假设更多的时间观察会导致更可靠的推理。我们质疑这一假设，并认为关键挑战不仅在于高效处理更多的视频帧，而在于在有限的时间证据下学习可靠的推理。我们提出了SAVER，一个从密集到稀疏的后训练框架，利用密集视频视图作为稀疏帧推理的训练时参考。在强化后训练过程中，成对的密集和稀疏视图通过基础奖励和可靠性门控参考奖励进行优化，鼓励稀疏视图预测保留与任务相关的时间证据。值得注意的是，SAVER仅在1,250个随机采样的时间基础示例上进行训练，而不使用任何视频问答注释。在三个时间基础基准和六个视频问答基准上，SAVER在不同的帧预算下始终提高了性能。特别是，SAVER能够在使用显著更少的帧的情况下，与密集帧Qwen3.5基线相匹配或超越。这些结果表明，时间基础可以作为一种有效的证据定位代理，用于学习稀疏视频推理，并能够迁移到更广泛的视频理解任务中。

View on arXiv Download PDF AI Translation

cs.CV / 12 / 2610.10945

GHARP: Real-time Gaussian Head Animation from Large-scale Reconstruction Prior

GHARP：基于大规模重建先验的实时高斯头部动画

Benlalah, Ali, Johari, Sepehr, Vitoria, Patricia, Kappeler, Armin, Sevastopolsky, Artem, Jung, Alexander, Fanelli, Gabriele, Mader, Kevin, Breitenstein, Manuel, Plüss, Claudia, Rüegg, Jan, Biland, Simon, Etterlin, Thomas, Kostiaev, Dmitry, Deschler, Mathias, Amberg, Brian, Martin, Sebastian

Abstract

We present GHARP (Real-time Gaussian Head Animation from Large-scale Reconstruction Prior), a method that animates 3D human heads in real time from a few input images of a subject and a driving expression signal. We decouple the problem into an identity stage that builds a representation of the subject's geometry and appearance offline, and an animation stage that predicts expression-dependent residuals on top of it at runtime. This separation offers a favorable trade-off with respect to fidelity, quality and runtime: the identity stage can be expensive while the animation stage runs a lightweight network, optimized for mobile devices. Our method performs animation in a semantically structured latent space of a pretrained reconstruction model, where expression changes remain spatially contained, making residual prediction efficient. This reconstruction prior provides a consistent spatial layout, allowing fusion of multiple input views into a compact, fixed-size canonical Gaussian representation. While this two-stage design improves the runtime-quality trade-off, it still inherits a problem common to all expression-driven avatar methods: expression codes describe only the face and thus omit body pose and clothing position, making these regions underspecified in the input. The animation network faces an ill-posed mapping and resorts to averaging over conflicting body appearances, producing blur and temporal flicker. We address this with a body alignment network that learns to align the person's body in the target image with the input reference images, removing the ambiguity from the training signal. Our method achieves state-of-the-art quality on the Ava-256 benchmark while running up to 13x faster on an A100 GPU with 8x fewer Gaussians.

Chinese Translation

我们提出了GHARP（基于大规模重建先验的实时高斯头部动画），一种从少量输入图像和驱动表情信号实时动画化3D人头的方法。我们将问题分解为两个阶段：身份阶段离线构建主体的几何和外观表示，动画阶段在运行时预测基于表情的残差。这种分离在保真度、质量和运行时间之间提供了良好的权衡：身份阶段可能较为耗时，而动画阶段则运行一个轻量级网络，优化用于移动设备。我们的方法在一个预训练重建模型的语义结构潜在空间中执行动画，其中表情变化保持空间上的局部性，使得残差预测高效。该重建先验提供了一致的空间布局，允许将多个输入视图融合为紧凑的固定大小的标准高斯表示。尽管这种两阶段设计改善了运行时间与质量的权衡，但它仍然继承了所有基于表情的虚拟形象方法的一个共同问题：表情编码仅描述面部，因此忽略了身体姿势和衣物位置，使得这些区域在输入中不够明确。动画网络面临一个病态映射，并通过对冲突的身体外观进行平均，导致模糊和时间闪烁。我们通过一个身体对齐网络来解决这个问题，该网络学习将目标图像中的人物身体与输入参考图像对齐，从而消除训练信号中的歧义。我们的方法在Ava-256基准测试中实现了最先进的质量，同时在A100 GPU上运行速度提高了最多13倍，并且所需的高斯数量减少了8倍。

View on arXiv Download PDF AI Translation

cs.CV / 13 / 2610.10959

GPU-Accelerated Computation of Persistent Homology for Topological Analysis of Image Data

基于GPU加速的持久同调计算用于图像数据的拓扑分析

Wang, Fan, Wagner, Hubert, Chowdhury, Rezaul, Chen, Chao

Abstract

In recent years, persistent homology has seen rapid adoption in deep learning, yet its computation remains a major bottleneck in network training. This paper introduces TopoGPU, a GPU streaming pipeline that computes persistence diagrams of cubical complexes induced by 2D and 3D images. TopoGPU streams the input image chunk by chunk, processing each chunk with massively parallel GPU kernels on a grid of GPU blocks; the resulting boundary relations are accumulated in host memory, where the CPU performs the boundary matrix reduction. TopoGPU introduces a stratification-aware discrete Morse matching that provably preserves persistent homology under streaming, together with a parallel topological sorting algorithm and a parallel V-path parity algorithm for deriving Morse boundaries on the GPU. TopoGPU outperforms Cubical Ripser, a state-of-the-art method for persistent homology computation, on every benchmark evaluated, achieving an average end-to-end speedup of 53.24x and a maximum of 198.01x. We further integrate TopoGPU into a topology-preserving deep network, demonstrating that it substantially reduces the cost of persistent homology computation during network training. TopoGPU is open source, with pre-built binaries, Google Colab notebooks, and Docker images available at the project's GitHub page: https://github.com/seravee08/GPU-Computation-of-Persistent-Homology-for-Image-Data.

Chinese Translation

近年来，持久同调在深度学习中得到了快速应用，但其计算仍然是网络训练中的一个主要瓶颈。本文介绍了TopoGPU，一个计算由2D和3D图像诱导的立方复形的持久性图的GPU流处理管道。TopoGPU逐块流式处理输入图像，利用大规模并行的GPU内核在GPU块的网格上处理每个块；生成的边界关系在主机内存中累积，CPU在此处执行边界矩阵的简化。TopoGPU引入了一种考虑分层的离散莫尔斯匹配，能够在流式处理下证明保持持久同调，并结合一种并行拓扑排序算法和一种并行V路径奇偶性算法以在GPU上推导莫尔斯边界。TopoGPU在每个评估的基准测试中均优于最先进的持久同调计算方法Cubical Ripser，平均实现了53.24倍的端到端加速，最高可达198.01倍。我们进一步将TopoGPU集成到一个保持拓扑特性的深度网络中，证明其在网络训练过程中显著降低了持久同调计算的成本。TopoGPU是开源的，预构建的二进制文件、Google Colab笔记本和Docker镜像可在项目的GitHub页面获取：https://github.com/seravee08/GPU-Computation-of-Persistent-Homology-for-Image-Data。

View on arXiv Download PDF AI Translation

cs.CV / 14 / 2610.10960

LVSPM: Long Sequence View Synthesis and Pose Estimation Model

LVSPM：长序列视图合成与姿态估计模型

Chen, Xi, Zhang, Yachi, Chen, Linghao, Liu, Minghua, Su, Hao, Xu, Zexiang, Zhang, Xiaoshuai

Abstract

We present LVSPM, a generalizable model that jointly estimates camera poses and synthesizes novel views from uncalibrated image collections. Trained with only RGB images and pose supervision, LVSPM avoids dense 3D ground truth and employs test-time training (TTT) layers to scale seamlessly to hundreds of input views. On RealEstate10k, Co3Dv2, and DL3DV, LVSPM surpasses VGGT in pose estimation across 16-256 views, with especially large margins at strict thresholds. For novel view synthesis under a practical protocol where more views cover larger scenes, LVSPM achieves state-of-the-art pose-free quality---surpassing even pose-dependent models in PSNR---and still maintains high quality as scene scale grows, while baselines collapse. The code is available at https://burningdust21.github.io/Projects/LVSPM .

Chinese Translation

我们提出了LVSPM，这是一种可推广的模型，能够从未校准的图像集合中联合估计相机姿态并合成新视图。LVSPM仅使用RGB图像和姿态监督进行训练，避免了密集的3D真实值，并采用测试时训练（TTT）层，使其能够无缝扩展到数百个输入视图。在RealEstate10k、Co3Dv2和DL3DV数据集上，LVSPM在16-256个视图的姿态估计中超越了VGGT，尤其在严格阈值下表现出显著的优势。在一个实用协议下进行新视图合成时，LVSPM在无姿态情况下实现了最先进的质量——在PSNR上甚至超越了依赖姿态的模型，并且随着场景规模的增加仍能保持高质量，而基线模型则崩溃。代码可在 https://burningdust21.github.io/Projects/LVSPM 获取。

View on arXiv Download PDF AI Translation

cs.CV / 15 / 2610.10984

Fluid-Gen-Zero: Grounding Pretrained Video Generators in Physics without Training

Fluid-Gen-Zero：在不训练的情况下将预训练视频生成器与物理学相结合

Huang, Hong, Liu, Yuqiu, You, Chenyu, Martin, Daniel, Zou, Chuhang, Chen, Wuyang

Abstract

We present Fluid-Gen-Zero, a training-free framework for physics-aware fluid-object interaction video generation that decouples physical reasoning from appearance synthesis. Our key insight is to delegate motion dynamics to a physics simulator while preserving the appearance modeling capacity of pretrained video generators. We bridge these two domains through a two-level agentic workflow: generation-time planning, where a vision-language model (VLM) agent interprets intent and the simulation rollout to organize generation clips, and latent-space guidance, which injects simulation signals into denoising through region-aware latent wrapping. This plug-and-play design is compatible with current video foundation models. We further introduce a benchmark for fluid-object interaction video generation. Across Tora (CogVideoX-based), VACE and WanMove (Wan-based), Fluid-Gen-Zero consistently improves simulation alignment, reducing object trajectory error by 26.7%-81.5% and fluid fEPE (fluid flow endpoint error) by 67.9%-84.0%, while largely preserving perceptual quality. In a human preference study, raters favor Fluid-Gen-Zero in 55.1%-74.4% of same-backbone comparisons across three backbones, and in 90.4%-94.2% of comparisons against simulation-based methods. Code and data will be released upon acceptance.

Chinese Translation

我们提出了Fluid-Gen-Zero，这是一个无需训练的框架，用于物理感知的流体-物体交互视频生成，它将物理推理与外观合成解耦。我们的关键见解是将运动动态委托给物理模拟器，同时保留预训练视频生成器的外观建模能力。我们通过两级代理工作流程将这两个领域连接起来：生成时规划，其中视觉-语言模型（VLM）代理解释意图并进行模拟展开以组织生成片段，以及潜在空间引导，通过区域感知的潜在包装将模拟信号注入去噪。这种即插即用的设计与当前的视频基础模型兼容。我们进一步引入了一个用于流体-物体交互视频生成的基准。在Tora（基于CogVideoX）、VACE和WanMove（基于Wan）上，Fluid-Gen-Zero在模拟对齐方面始终表现出色，将物体轨迹误差减少了26.7%-81.5%，流体fEPE（流体流动端点误差）减少了67.9%-84.0%，同时在很大程度上保持了感知质量。在一项人类偏好研究中，评审者在三个骨干网络的同骨干比较中，55.1%-74.4%偏好Fluid-Gen-Zero，并在与基于模拟的方法的比较中，90.4%-94.2%偏好Fluid-Gen-Zero。代码和数据将在接受后发布。

View on arXiv Download PDF AI Translation

cs.CV / 16 / 2610.10990

Omni-Diffusion-Distill: Few-Step Distillation of Unified Multimodal Diffusion Large Language Models

全方位扩散蒸馏：统一多模态扩散大语言模型的少步蒸馏

Huang, Hong, Yang, Chenhongyi, Sun, Junzhe, Sinha, Animesh, Chen, Wuyang, Jiang, Yifan

Abstract

Unified multimodal diffusion large language models (dLLMs) offer a single architecture for both image generation and multimodal understanding, but their iterative decoding requires tens to hundreds of forward passes. Existing few-step distillation methods largely focus on either image generation or text generation, making it unclear how to compress a fully discrete multimodal dLLM into a single efficient student while preserving both generation and understanding. We introduce Omni-Diffusion-Distill, a unified two-stage distillation framework that retains strong generation and understanding capabilities while substantially reducing the inference cost of a unified multimodal dLLM. Omni-Diffusion-Distill aligns the distillation of both generation and understanding, for both images and text, in the discrete token space. In the first stage, the student is trained to skip decoding steps by replaying cached teacher trajectories, and in the second stage the student is refined on intermediate states along its own rollouts. We further remedy two sources of degradation in unified distillation with a pairwise collision penalty that reduces repetition under parallel text decoding, and entropy-matched guidance that prevents entropy collapse caused by fitting the sharpened teacher distribution in image generation. Omni-Diffusion-Distill achieves state-of-the-art trade-offs between decoding efficiency and generation and understanding performance for multimodal dLLMs, reducing image generation from 128 to 8 decoding steps and multimodal understanding from 512 to 64, giving 18.2x and 21.2x wall-clock speedups. Under these budgets, it scores 0.828 on GenEval and 83.0 on DPG-Bench for text-to-image generation, while reaching GPT judge scores of 20.0 on MM-Vet and 57.2 on COCO captioning (twice the teacher's 28.4 at the same steps) for multimodal understanding.

Chinese Translation

统一多模态扩散大语言模型（dLLMs）为图像生成和多模态理解提供了单一架构，但其迭代解码需要数十到数百次前向传递。现有的少步蒸馏方法主要集中在图像生成或文本生成上，因此不清楚如何将一个完全离散的多模态 dLLM 压缩为一个高效的学生模型，同时保留生成和理解能力。我们提出了全方位扩散蒸馏（Omni-Diffusion-Distill），这是一个统一的两阶段蒸馏框架，能够在显著降低统一多模态 dLLM 的推理成本的同时，保持强大的生成和理解能力。全方位扩散蒸馏在离散标记空间中对图像和文本的生成与理解蒸馏进行了对齐。在第一阶段，学生模型通过重放缓存的教师轨迹来跳过解码步骤进行训练；在第二阶段，学生模型在其自身的展开过程中的中间状态上进行精炼。我们进一步通过成对碰撞惩罚来修正统一蒸馏中的两个降级来源，该惩罚减少了并行文本解码下的重复，并通过熵匹配引导防止因拟合图像生成中的锐化教师分布而导致的熵崩溃。全方位扩散蒸馏在多模态 dLLMs 中实现了解码效率与生成和理解性能之间的最先进权衡，将图像生成的解码步骤从 128 降至 8，将多模态理解的解码步骤从 512 降至 64，分别实现了 18.2 倍和 21.2 倍的墙钟速度提升。在这些预算下，其在文本到图像生成的 GenEval 上得分 0.828，在 DPG-Bench 上得分 83.0，同时在多模态理解方面，在 MM-Vet 上获得 GPT 评估分数 20.0，在 COCO 标题生成上获得 57.2（是教师在相同步骤下的 28.4 的两倍）。

View on arXiv Download PDF AI Translation

cs.CV / 17 / 2610.10991

Region-Aware CLS Token Augmentation for Fine-Grained Image Retrieval

区域感知的 CLS 令牌增强用于细粒度图像检索

Bezerra, Ian de Holanda Cavalcanti, Trivedy, Vivek, Valem, Lucas Pascotti, Latecki, Longin Jan

Abstract

Image retrieval methods often rely on a single global semantic descriptor extracted from an image, e.g., the [CLS] token in vision transformers. However, trying to squeeze all the semantic information of an image into a single descriptor can hurt downstream retrieval performance, especially for fine-grained retrieval tasks. In this work, we augment the semantic tokens in the newer visual transformers, the global [CLS] token and the four register tokens, with a carefully selected collection of spatial tokens, aiming to capture the spatial region representation that characterizes the contents captured in each of the semantic tokens. We leverage the DINOv2-reg model, which includes register tokens that emergently learn object and part-based representations. For each "cue" token ([CLS] and each register token), we find a "buddy" image patch token and extract an N x N patch region to produce a set of localized ROI tokens. Our approach automatically captures important regions of interest without any external bounding boxes or saliency modules, purely by matching semantic tokens with their spatial representation regions. Furthermore, we incorporate these tokens into a multi-vector retrieval framework inspired by ColBERT, enabling fine-grained matching via a per-token alignment mechanism while avoiding the large storage cost of keeping all patch embeddings. Through extensive experiments, we find that (1) register tokens encode useful fine-grained details that can complement the [CLS] token; (2) automatically pooled ROI tokens further improve fine-grained discrimination; and (3) multi-vector retrieval with a small set of tokens improves over a DINOv2-reg single-vector baseline while remaining tractable for large-scale search. The code is available at https://github.com/IdhcbIan/Augmenting_CLS_with_ROI_tokens.

Chinese Translation

图像检索方法通常依赖于从图像中提取的单一全局语义描述符，例如视觉变换器中的 [CLS] 令牌。然而，试图将图像的所有语义信息压缩到一个描述符中可能会损害下游检索性能，尤其是在细粒度检索任务中。在本研究中，我们通过精心选择的一组空间令牌来增强新型视觉变换器中的语义令牌，包括全局 [CLS] 令牌和四个注册令牌，旨在捕捉表征每个语义令牌所捕获内容的空间区域表示。我们利用 DINOv2-reg 模型，该模型包括自发学习对象和部分表示的注册令牌。对于每个“提示”令牌（[CLS] 令牌和每个注册令牌），我们找到一个“伙伴”图像补丁令牌，并提取一个 N x N 补丁区域以生成一组局部化的 ROI 令牌。我们的方法通过将语义令牌与其空间表示区域匹配，自动捕捉重要的兴趣区域，而无需任何外部边界框或显著性模块。此外，我们将这些令牌纳入受 ColBERT 启发的多向量检索框架中，通过每个令牌的对齐机制实现细粒度匹配，同时避免了保留所有补丁嵌入的大量存储成本。通过大量实验，我们发现（1）注册令牌编码了有用的细粒度细节，可以补充 [CLS] 令牌；（2）自动聚合的 ROI 令牌进一步提高了细粒度区分能力；（3）使用少量令牌的多向量检索在保持大规模搜索可行性的同时，优于 DINOv2-reg 单向量基线。代码可在 https://github.com/IdhcbIan/Augmenting_CLS_with_ROI_tokens 获取。

View on arXiv Download PDF AI Translation

cs.CV / 18 / 2610.11011

PCAsplat: Gaussian Splatting with Local PCA Regularization

PCAsplat：带有局部主成分分析正则化的高斯喷溅

Matias, Vitor, Nascimento, Filipe, Nakayama, Kiyohiro, Lima, João Paulo, Lobo, Márcus, Wetzstein, Gordon, Guibas, Leonidas, Paiva, Afonso, Novello, Tiago

Abstract

Gaussian splatting has emerged as a flexible representation for 3D reconstruction from posed images. However, existing methods are optimized primarily using rasterization-based losses, which supervise a splat only when it contributes to sampled camera rays. Gaussians that are occluded or contribute little to the sampled view therefore receive weak or no geometric gradients and may drift away from the underlying surface, producing undesired floaters. We introduce PCAsplat, a geometry-aware regularization framework for Gaussian splatting based on differentiable local principal component analysis (PCA). Our PCA regularizer acts directly on neighborhoods of Gaussian centers and can therefore update Gaussians that do not contribute to the current training view. We regularize the PCA eigenvalues to encourage Gaussians to move to the underlying surface with isotropic tangent-plane coverage. We also align each Gaussian normal with the PCA-estimated neighborhood normal to enforce consistent orientation. Experiments on DTU, Tanks and Temples, and NeRF Synthetic show that the splats produced by PCAsplat better approximate samples of the reference surface while substantially reducing undesired floaters. These surface-aligned splats enable downstream geometry-processing tasks, including point cloud segmentation, and direct Poisson reconstruction. Additionally, PCAsplat remains competitive under conventional novel view synthesis and mesh extraction tasks. Code will be released.

Chinese Translation

高斯喷溅已成为从姿态图像进行三维重建的一种灵活表示方法。然而，现有方法主要使用基于光栅化的损失进行优化，这种损失仅在喷溅对采样的相机光线有贡献时进行监督。因此，被遮挡或对采样视图贡献较小的高斯会收到较弱或没有几何梯度，可能会偏离基础表面，产生不希望出现的漂浮物。我们提出了PCAsplat，一种基于可微分局部主成分分析（PCA）的几何感知正则化框架，用于高斯喷溅。我们的PCA正则化器直接作用于高斯中心的邻域，因此可以更新那些对当前训练视图没有贡献的高斯。我们对PCA特征值进行正则化，以鼓励高斯移动到基础表面，并实现各向同性的切平面覆盖。我们还将每个高斯的法线与PCA估计的邻域法线对齐，以确保一致的方向。对DTU、Tanks and Temples和NeRF Synthetic的实验表明，PCAsplat生成的喷溅更好地逼近参考表面的样本，同时显著减少不希望出现的漂浮物。这些与表面对齐的喷溅使下游几何处理任务得以实现，包括点云分割和直接泊松重建。此外，PCAsplat在传统的新视图合成和网格提取任务中仍保持竞争力。代码将会发布。

View on arXiv Download PDF AI Translation

cs.CV / 19 / 2610.11019

Mid-Training Language Models on Raw Video

在原始视频上进行中期训练的语言模型

Hwang, Jaedong, Shen, Xiaoqian, Chang, Ernie, Zhao, Changsheng, Zhou, Chong, Suri, Saksham, Qian, Qi, Liu, Zechun, Wu, Lemeng, Wang, Qinsi, Krishnamoorthi, Raghuraman, Wen, Wei

Abstract

Multimodal large language models learn mostly from paired image-text data or annotated video, and raw web video is rarely used to further train an existing language model. We study whether raw video, with no captions and no text loss, can serve as mid-training data for a pretrained language model. Frames are encoded into continuous visual tokens, and the language model learns to predict the next visual token. We mid-train Qwen3-1.7B on raw clips from YT-Temporal-1B and then apply the same image-text instruction tuning to it and to the model without mid-training, so that the two differ only in mid-training. The mid-trained model scores 2.9 points higher on average across four video benchmarks and 5.1 points higher across ten image benchmarks, spanning perception, document, and chart tasks. Text performance is preserved even though mid-training includes no text, with an average of 48.9 across 14 text benchmarks compared with 48.0 for the model without mid-training. Analyses across training show that the image and video gains emerge within 30% of training and plateau thereafter, varying by less than 0.5 points. Predicting captions fails to outperform next-visual-token prediction, demonstrating that video mid-training can remain purely self-supervised without the computational overhead or labeling noise of automated captioning.

Chinese Translation

多模态大型语言模型主要通过配对的图像-文本数据或注释视频进行学习，而原始网络视频则很少用于进一步训练现有的语言模型。我们研究了没有字幕和文本损失的原始视频是否可以作为预训练语言模型的中期训练数据。帧被编码为连续的视觉标记，语言模型学习预测下一个视觉标记。我们在来自 YT-Temporal-1B 的原始剪辑上对 Qwen3-1.7B 进行中期训练，然后对其进行与未进行中期训练的模型相同的图像-文本指令调优，使得两者仅在中期训练上存在差异。中期训练的模型在四个视频基准测试中平均得分高出 2.9 分，在十个图像基准测试中高出 5.1 分，涵盖感知、文档和图表任务。尽管中期训练不包括文本，但文本性能得以保持，在 14 个文本基准测试中平均得分为 48.9，而未进行中期训练的模型得分为 48.0。训练分析表明，图像和视频的提升在训练的 30% 内出现，并在此后趋于平稳，变化幅度小于 0.5 分。预测字幕未能超越下一个视觉标记的预测，证明视频中期训练可以保持完全自监督，而无需自动字幕生成的计算开销或标注噪声。

View on arXiv Download PDF AI Translation

cs.CV / 20 / 2610.11023

Expression-Diverse References for Identity-Preserving Video Generation

多样化表情参考用于身份保留的视频生成

Fu, Tianwen, Teng, Wenbin, Chen, Gonglin, Ouyang, Junyi, Xiong, Haolin, Zhao, Yajie

Abstract

Identity-preserving video generation aims to maintain a subject's identity while synthesizing realistic videos. Yet a single reference portrait captures the subject's appearance under only one facial configuration. As expressions change, facial appearance can vary in highly identity-specific ways, leaving the subject's appearance under unseen expressions underdetermined by the reference alone. This expression-dependent variation also complicates evaluation: similarity to a neutral reference may decrease under strong expressions even for real images of the same person. We investigate this limitation from both generation and evaluation perspectives. First, we quantify how face-recognition similarity varies with expression intensity using controlled photographs and MEAD videos. We then construct a compact yet expressive reference gallery that captures diverse expression-dependent facial configurations. Matching against this gallery provides a more robust measure of identity similarity under expressive motion. To further expose performance degradation with expression intensity, we report identity similarity separately for mild, intense, and extreme expressions. For generation, we extend Stand-In to condition on our expression-diverse reference sets and develop a data-curation pipeline that extracts consistent yet diverse face crops from training videos. In practical settings where only a single portrait is available, we construct the reference set by synthesizing additional expressions with a pretrained facial reenactment model. On our controlled benchmark, both real and synthesized reference sets outperform the evaluated baselines in identity similarity across all three expression-intensity regimes, with the largest improvements for extreme expressions.

Chinese Translation

身份保留视频生成旨在在合成真实视频的同时保持主体的身份。然而，单一的参考肖像只能捕捉到主体在一种面部配置下的外观。随着表情的变化，面部外观可能以高度特定于身份的方式发生变化，这使得在未见表情下，单靠参考无法确定主体的外观。这种依赖表情的变化也使得评估变得复杂：即使是同一个人的真实图像，在强烈表情下与中性参考的相似度也可能降低。我们从生成和评估两个角度研究这一局限性。首先，我们利用受控照片和MEAD视频量化面部识别相似度如何随表情强度变化。然后，我们构建一个紧凑而富有表现力的参考库，以捕捉多样化的依赖表情的面部配置。与该库进行匹配提供了在表现性运动下更稳健的身份相似度度量。为了进一步揭示表情强度对性能的影响，我们分别报告轻度、强烈和极端表情下的身份相似度。在生成方面，我们扩展了Stand-In以基于我们多样化的表情参考集进行条件生成，并开发了一个数据策划管道，从训练视频中提取一致而多样的面部裁剪。在仅有单一肖像可用的实际环境中，我们通过使用预训练的面部重现模型合成额外的表情来构建参考集。在我们的受控基准测试中，无论是真实还是合成的参考集，在所有三种表情强度下的身份相似度均优于评估的基线，其中在极端表情下的改进最大。

View on arXiv Download PDF AI Translation

cs.CV / 21 / 2610.11037

Transforming Image Editors into Video Editors

将图像编辑器转变为视频编辑器

Wang, Feng, Li, Zijie, Yang, Ceyuan, Yuille, Alan, Wang, Peng

Abstract

Recent image editing systems have achieved impressive semantic understanding, visual fidelity, and instruction-following ability, while video editing remains substantially more difficult and costly. In this paper, we present a simple alternative to end-to-end video editing: instead of training a monolithic video editor, we transform a strong image editor into a video editor through anchor-based generation. Our key insight is that video editing can be decomposed into two subproblems: editing a sparse set of keyframes and propagating those edits across time. Based on this observation, we propose Anchor-based Video Editing (AVE), a two-stage framework in which a powerful image editor first performs composed editing on selected keyframes, and a motion-guided image-to-video diffusion model then generates the final video by treating the edited keyframes as fixed anchors. This design directly inherits the strengths of modern image editors while avoiding expensive end-to-end video editing training. Experiments on IVEBench and VIE-Bench show that AVE achieves strong performance in instruction following, temporal consistency, and content fidelity. Further ablations reveal that final video editing quality is strongly correlated with the quality of the image editor, suggesting that future progress in video editing may come from stronger image editing foundations and lightweight transfer to video. Code is available at https://github.com/wangf3014/AVE.

Chinese Translation

近期的图像编辑系统在语义理解、视觉保真度和遵循指令的能力方面取得了令人瞩目的成就，而视频编辑仍然显得更加困难且成本高昂。本文提出了一种简单的端到端视频编辑替代方案：我们通过基于锚点的生成将一个强大的图像编辑器转变为视频编辑器，而不是训练一个单一的视频编辑器。我们的关键见解是，视频编辑可以被分解为两个子问题：编辑稀疏的关键帧集合以及在时间上传播这些编辑。基于这一观察，我们提出了基于锚点的视频编辑（Anchor-based Video Editing, AVE），这是一个两阶段框架，其中强大的图像编辑器首先对选定的关键帧进行组合编辑，然后一个运动引导的图像到视频扩散模型通过将编辑后的关键帧视为固定锚点来生成最终视频。这一设计直接继承了现代图像编辑器的优势，同时避免了昂贵的端到端视频编辑训练。在IVEBench和VIE-Bench上的实验表明，AVE在遵循指令、时间一致性和内容保真度方面表现出色。进一步的消融实验显示，最终视频编辑质量与图像编辑器的质量密切相关，这表明视频编辑的未来进展可能来自更强大的图像编辑基础和轻量级的转移到视频。代码可在 https://github.com/wangf3014/AVE 获取。

View on arXiv Download PDF AI Translation

cs.CV / 22 / 2610.11039

Rendering-Free Lookahead for Question-Guided Active Vision

无渲染前瞻的基于问题引导的主动视觉

Sakamoto, Koya, Azuma, Daichi, Kurita, Shuhei, Chiba, Naoya, Iwasawa, Yusuke, Matsuo, Yutaka, Miyanishi, Taiki

Abstract

Active robot vision requires controlling the camera to reveal task-relevant information that is hidden from the current viewpoint. For example, determining what is inside a box may require raising the camera and looking down into it. For viewpoint-dependent question answering, the challenge is to select camera motions that expose the visual evidence needed to answer the question. Although vision-language models (VLMs) can interpret observed images, selecting such motions requires anticipating the usefulness of unseen views. We quantify this usefulness as answerability, a VLM's estimate that a view suffices to answer the question, and present Rendering-Free Lookahead (RFL), a viewpoint-selection policy that ranks candidate camera motions by predicted future answerability. RFL transfers visual lookahead from deployment to offline training. At training, a privileged teacher renders candidate future views in 3D Gaussian Splatting (3DGS) scenes and uses a frozen VLM to compute one- and two-step answerability targets. Through two-stage distillation, a student learns to predict these action values from the question, recent visual observations, and a candidate camera motion. At deployment, RFL uses these predicted values to select camera motions without rendering future views. On 377 E3VS-Bench test episodes in unseen environments, RFL improves the mean judge score by 43\% over a direct-action baseline using the same VLM. These results support learning camera-control policies from privileged visual lookahead for viewpoint-dependent question answering.

Chinese Translation

主动机器人视觉需要控制相机以揭示当前视角下隐藏的与任务相关的信息。例如，确定箱子内部的物体可能需要抬起相机并向下观察。对于依赖视角的问题回答，挑战在于选择能够暴露回答问题所需视觉证据的相机运动。尽管视觉语言模型（VLMs）可以解释观察到的图像，但选择这样的运动需要预测未见视图的有用性。我们将这种有用性量化为可回答性，即VLM对某个视图是否足以回答问题的估计，并提出无渲染前瞻（Rendering-Free Lookahead, RFL），这是一种通过预测未来可回答性来对候选相机运动进行排名的视角选择策略。RFL将视觉前瞻从部署阶段转移到离线训练。在训练阶段，特权教师在3D高斯点云（3D Gaussian Splatting, 3DGS）场景中渲染候选未来视图，并使用冻结的VLM计算一步和两步的可回答性目标。通过两阶段蒸馏，学生学习从问题、最近的视觉观察和候选相机运动中预测这些动作值。在部署阶段，RFL利用这些预测值选择相机运动，而无需渲染未来视图。在377个未见环境的E3VS-Bench测试案例中，RFL在使用相同VLM的直接动作基线基础上提高了平均评判分数43%。这些结果支持从特权视觉前瞻中学习相机控制策略，以实现依赖视角的问题回答。

View on arXiv Download PDF AI Translation

cs.CV / 23 / 2610.11049

SatFix: Absolute Visual Localization of UAVs in Satellite Maps from a Single Oblique Image

SatFix：基于单幅倾斜图像在卫星地图中对无人机的绝对视觉定位

Zeng, Jiarui, Shi, Kun, Vong, Chiman, Zheng, Zhedong

Abstract

We study absolute metric UAV localization within a provided geo-referenced satellite region, recovering continuous map position and viewing heading from a single oblique image or a short multi-view clip. Existing cross-view geo-localization methods retrieve the most similar satellite tile from a gallery and report Recall@K, but retrieval depends on gallery sampling, provides no heading estimate, and returns a tile index rather than a continuous coordinate. We propose SatFix, a feed-forward UAV--satellite localization framework built on VGGT-$\Omega$. Satellite-grid features act as queries that aggregate UAV visual evidence, and two lightweight heads regress a 3-DoF pose in the satellite-map frame: continuous 2D position and heading. SatFix requires no explicit 3D map, rendered bird's-eye image, auxiliary sensor, or test-time pose alignment. A single model supports both single- and multi-view inputs, with trajectory constraints used during multi-view training. For metric evaluation, we introduce University-Metric, where satellite imagery is re-collected over a region up to 10.7$\times$ longer on a side (about 114$\times$ the ground area) than the original University-1652 tiles, with continuous position and heading labels for the original UAV tours. With one UAV view, SatFix localizes 52.08% of test frames within 50 m and 17.34% within 10 m, with median position and heading errors of 45.66 m and $20.81^\circ$, respectively. Inference takes under 0.1 s per single-view query on an NVIDIA RTX 4090. With nine UAV views, the median position error falls to 21.96 m and the median heading error to $8.73^\circ$. Compared with a fine-tuned VGGT-$\Omega$ baseline, SatFix reduces median position error by 34.0% and nine-view median heading error from $25.43^\circ$ to $8.73^\circ$.

Chinese Translation

我们研究在提供的地理参考卫星区域内进行绝对度量无人机定位，从单幅倾斜图像或短的多视角视频片段中恢复连续的地图位置和视角方向。现有的跨视角地理定位方法从图库中检索最相似的卫星图块并报告 Recall@K，但检索依赖于图库采样，无法提供视角估计，并返回图块索引而非连续坐标。我们提出了 SatFix，一个基于 VGGT-$ ext{Ω}$ 的前馈无人机-卫星定位框架。卫星网格特征作为查询，聚合无人机视觉证据，两个轻量级头部回归卫星地图框架中的 3 自由度姿态：连续的二维位置和视角。SatFix 不需要显式的三维地图、渲染的鸟瞰图像、辅助传感器或测试时姿态对齐。单一模型支持单视角和多视角输入，在多视角训练中使用轨迹约束。为了进行度量评估，我们引入了 University-Metric，其中卫星图像在一个区域内重新收集，边长比原始 University-1652 图块长 10.7$ imes$（约 114$ imes$ 地面面积），并为原始无人机巡航提供连续的位置和视角标签。在单个无人机视角下，SatFix 在 50 米内定位 52.08% 的测试帧，在 10 米内定位 17.34%，中位位置和视角误差分别为 45.66 米和 $20.81^ ext{°}$。在 NVIDIA RTX 4090 上，单视角查询的推理时间少于 0.1 秒。使用九个无人机视角时，中位位置误差降至 21.96 米，中位视角误差降至 $8.73^ ext{°}$。与经过微调的 VGGT-$ ext{Ω}$ 基线相比，SatFix 将中位位置误差降低了 34.0%，九视角中位视角误差从 $25.43^ ext{°}$ 降至 $8.73^ ext{°}$。

View on arXiv Download PDF AI Translation

cs.CV / 24 / 2610.11057

Learning What to Trust in Multimodal Learning under Noisy Supervision

在噪声监督下多模态学习中学习信任的内容

Zou, Jiashuo, Xia, Xiaobo

Abstract

Multimodal classification processes and relates information from multiple modalities to achieve more accurate predictions. However, existing methods typically rely on high-quality ground-truth labels, which are difficult to obtain in real-world scenarios. While sample-selection methods for learning with noisy labels aim to identify correctly labeled examples from noisy data, traditional methods primarily focus on unimodal settings and fail to exploit multimodal information fully. This motivates us to build a more reliable noise detector in multimodal learning. To this end, we theoretically analyze the relationship between representation structure and noise detection capability. Based on this analysis, we propose REFINE, which is a multimodal label-noise detection framework that jointly uses fused and unimodal representations for label-noise detection. Specifically, REFINE constructs discriminative eigenvectors through discriminative analysis of the target and background classes and selects trusted representation spaces with better noise detection capability for each class. Within each trusted space, REFINE measures the alignment between each instance representation and the discriminative eigenvectors. It then combines the subsets selected from these spaces. The combined set provides cleaner supervision for updating the multimodal classifier, thereby reducing the influence of mislabeled examples during training and improving model generalization. Extensive experiments across diverse tasks demonstrate REFINE's superiority compared to baseline methods. The source code will be publicly available.

Chinese Translation

多模态分类过程涉及并关联来自多个模态的信息，以实现更准确的预测。然而，现有方法通常依赖于高质量的真实标签，而这些标签在现实场景中难以获得。虽然针对带有噪声标签的学习的样本选择方法旨在从噪声数据中识别正确标记的示例，但传统方法主要集中在单模态设置上，未能充分利用多模态信息。这促使我们在多模态学习中构建一个更可靠的噪声检测器。为此，我们理论上分析了表示结构与噪声检测能力之间的关系。基于这一分析，我们提出了REFINE，这是一种多模态标签噪声检测框架，联合使用融合和单模态表示进行标签噪声检测。具体而言，REFINE通过对目标类和背景类的区分分析构建判别特征向量，并为每个类别选择具有更好噪声检测能力的可信表示空间。在每个可信空间内，REFINE测量每个实例表示与判别特征向量之间的对齐程度。然后，它结合从这些空间中选择的子集。组合集为更新多模态分类器提供了更干净的监督，从而减少了训练过程中错误标记示例的影响，提高了模型的泛化能力。在多种任务上的广泛实验表明，REFINE相较于基线方法具有优越性。源代码将公开发布。

View on arXiv Download PDF AI Translation

cs.CV / 25 / 2610.11058

Refine Connections, Close the Gap: A Reliable Enhancement Framework for Driving Scene Topology

优化连接，缩小差距：一种可靠的驾驶场景拓扑增强框架

Wang, Xiaoqi, Zhaung, Dingyi, Paz, David, He, Wenbin, Bai, Yucai, Zhou, Peng, Zhang, Rui, Zhao, Jinhua, Ren, Liu

Abstract

In autonomous driving, understanding scene topology - the connectivity between lanes and traffic elements - is critical for safe path planning and motion control. While current methods excel at detecting individual map elements, their connectivity reasoning often falls short of its theoretical potential, leaving a significant performance gap relative to the theoretical upper-bound achievable given the underlying detections. Furthermore, the decision-ready topology graphs passed to downstream tasks often remain unreliable. Current approaches typically derive connectivity by thresholding continuous topology scores; however, these scores often fail to reflect the true logical likelihood of connectivity, resulting in false positives or missing connections. Existing benchmarks further overlook this issue by primarily evaluating continuous metrics, rather than assessing the discrete connectivity required for decision-making. To bridge these gaps, we propose TopoEnhance, a novel topology enhancement framework designed to unlock the latent potential of existing methods and improve the reliability of decision-ready topology. We formulate topology enhancement as a denoising-based reconstruction process, where the model learns to recover structural consistency from stochastically corrupted ground-truth graphs. This formulation enables the model to resolve logical inconsistencies and rectify unreliable connections, producing robust discrete topology graphs that closely approach theoretical maximum performance. Extensive experiments across different baselines show that TopoEnhance consistently improves both continuous topology metrics (TOP score), and discrete connectivity measured by our adapted Topology Jaccard Similarity (TJS) metric. As a flexible, source-agnostic framework, TopoEnhance delivers substantial gains across diverse state-of-the-art baselines without requiring retraining.

Chinese Translation

在自动驾驶中，理解场景拓扑——车道与交通元素之间的连接性——对于安全路径规划和运动控制至关重要。尽管当前方法在检测单个地图元素方面表现出色，但它们的连接性推理往往未能达到理论潜力，导致相对于给定基础检测的理论上限存在显著的性能差距。此外，传递给下游任务的决策准备拓扑图通常仍然不可靠。当前方法通常通过对连续拓扑得分进行阈值处理来推导连接性；然而，这些得分往往未能反映连接性的真实逻辑可能性，导致误报或遗漏连接。现有基准测试进一步忽视了这一问题，主要评估连续指标，而不是评估决策所需的离散连接性。为了解决这些问题，我们提出了TopoEnhance，一种新颖的拓扑增强框架，旨在释放现有方法的潜在能力，提高决策准备拓扑的可靠性。我们将拓扑增强形式化为基于去噪的重建过程，其中模型学习从随机损坏的真实图中恢复结构一致性。这一形式化使模型能够解决逻辑不一致性并纠正不可靠的连接，生成稳健的离散拓扑图，接近理论最大性能。通过对不同基线的广泛实验表明，TopoEnhance始终改善了连续拓扑指标（TOP得分）和通过我们改编的拓扑Jaccard相似度（TJS）指标测量的离散连接性。作为一个灵活的、源无关的框架，TopoEnhance在多种最先进的基线中提供了显著的提升，而无需重新训练。

View on arXiv Download PDF AI Translation

cs.CV / 26 / 2610.11060

AffordDrive3D: Affordance-Aware World-Action Modeling with Spatial Understanding

AffordDrive3D：具有空间理解的可供性感知世界-动作建模

Cai, Tianhui, Sun, Xinglong, Fang, Chao, Li, Zhenxin, Song, Rui, Alvarez, Jose M., Mao, Yunxiang, Ma, Jiaqi, Liu, Langechuan

Abstract

World-action models have recently improved autonomous driving by jointly learning future scene prediction and trajectory generation. Most existing approaches model the future primarily through RGB appearance, and recent works have begun to incorporate geometric prediction to improve spatial understanding. However, dense geometry describes the spatial layout of the entire scene without indicating which parts are most relevant to the ego vehicle's action. For driving, the model must also identify and anticipate where it can safely move and which regions may pose collision risks. Jointly modeling action-relevant regions and future geometry can provide the policy with both driving-relevant cues and their corresponding spatial structure. We therefore propose AffordDrive3D, an affordance- and geometry-aware world-action model that jointly learns future action-relevant regions and spatial structure. In order to capture the scene semantics and driving context needed for driving affordance prediction, we build AffordDrive3D on a VLM backbone to forecast drivable areas and collision-critical regions that directly affect ego motion, while predicting future geometry from RGB world-model latents. On NAVSIM, AffordDrive3D achieves state-of-the-art performance with 91.3 PDMS and 89.9 EPDMS, demonstrating the effectiveness of jointly modeling future affordances and geometry for trajectory planning.

Chinese Translation

世界-动作模型最近通过联合学习未来场景预测和轨迹生成来提升自主驾驶的能力。现有的大多数方法主要通过RGB外观建模未来，而最近的研究开始结合几何预测以改善空间理解。然而，密集几何描述了整个场景的空间布局，却未指明哪些部分与自我车辆的动作最为相关。对于驾驶而言，模型还必须识别和预测其可以安全移动的位置以及哪些区域可能存在碰撞风险。联合建模与动作相关的区域和未来几何可以为策略提供与驾驶相关的线索及其对应的空间结构。因此，我们提出了AffordDrive3D，这是一种可供性和几何感知的世界-动作模型，能够联合学习未来的动作相关区域和空间结构。为了捕捉驾驶可供性预测所需的场景语义和驾驶上下文，我们在VLM骨干网络上构建AffordDrive3D，以预测直接影响自我运动的可驾驶区域和碰撞关键区域，同时从RGB世界模型潜变量中预测未来几何。在NAVSIM上，AffordDrive3D以91.3 PDMS和89.9 EPDMS达到了最先进的性能，证明了联合建模未来可供性和几何在轨迹规划中的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 27 / 2610.11067

Diffusion Meta-Prompting and Steering for Generalizable Foundation Model Adaptation

用于可泛化基础模型适应的扩散元提示与引导

Sridhar, Deepak, Li, Yi, Bhardwaj, Kartikeya, Liu, Shuangjun, Jing, Taotao, Li, Yuan, Zhang, Shuai, Lyu, Jiancheng, Gao, Dashan, Vasconcelos, Nuno

Abstract

Prompt learning is a popular method for adapting foundation models, but learned prompts are typically task-specific and fail to generalize to new classes, domains, or compositions of tasks. In this paper, we introduce a Diffusion Meta-Prompt (DMP) model , a framework that models the distribution of learned prompts using diffusion models. Given a repository of previously learned prompts, DMP is trained and sampled without access to the original task examples or task losses, and synthesizes new prompts conditioned on natural language task descriptions. To improve the sampling stability, we introduce a test-time steering strategy for DMP, which uses the best training-selected prompt in the repository as a latent anchor during diffusion sampling, without retraining the DMP or accessing test classes. DMP improves generalization across classification, retrieval and text-to-image generation tasks, supports concept composition and negative prompting without explicit training. It reduces storage and inference costs by over 90% compared to prompt retrieval methods. For composite classification, DMP achieves upto 2.0% average gain over prior meta-learning methods across 55 pairs of datasets with gains as high as 8.5% on specific pairs such as Eurosat and Flowers. DMP also enhances cross-task generalization with ~2-9% improvement for hierarchical classification task. We further provide a theoretical guarantee bounding the expected task loss of prompts sampled from a DMP. Code is available: https://github.com/DeepakSridhar/dmp

Chinese Translation

提示学习是一种流行的基础模型适应方法，但学习到的提示通常是特定于任务的，无法泛化到新的类别、领域或任务组合。在本文中，我们介绍了一种扩散元提示（Diffusion Meta-Prompt, DMP）模型，该框架使用扩散模型对学习到的提示的分布进行建模。DMP在没有访问原始任务示例或任务损失的情况下，基于之前学习到的提示库进行训练和采样，并根据自然语言任务描述合成新的提示。为了提高采样的稳定性，我们为DMP引入了一种测试时引导策略，该策略在扩散采样过程中使用库中最佳的训练选择提示作为潜在锚点，而无需重新训练DMP或访问测试类别。DMP在分类、检索和文本到图像生成任务中提高了泛化能力，支持概念组合和负提示，而无需显式训练。与提示检索方法相比，它将存储和推理成本降低了超过90%。在复合分类任务中，DMP在55对数据集上实现了高达2.0%的平均增益，在特定对（如Eurosat和Flowers）上增益高达8.5%。DMP还增强了跨任务的泛化能力，在层次分类任务中实现了约2-9%的改进。我们进一步提供了理论保证，界定了从DMP采样的提示的预期任务损失。代码可在此获取：https://github.com/DeepakSridhar/dmp

View on arXiv Download PDF AI Translation

cs.CV / 28 / 2610.11070

No Distillation Needed: Single-Pass Real-Time Talking Heads via Acausal Noise Shaping

无需蒸馏：通过非因果噪声塑形实现单次实时人头动画

Han, Yu, Markovic, Dejan, Richard, Alexander, Zielonka, Wojciech, Venkatesh, Akshay, Wuu, Cheng-hsin, Zollhoefer, Michael

Abstract

Audio-driven facial animation underpins real-time avatars, telepresence, and embodied virtual agents. And it must run online: each frame emitted from audio observed up to the current time, at interactive rates. Recent progress is dominated by diffusion models, which need many network evaluations per sample and are therefore a poor fit for streaming. We argue the cost is unnecessary in this domain. Audio-conditioned facial motion occupies a comparatively low-dimensional manifold, a regime where a single-pass GAN suffices. The obstacle is not capacity but stochastic structure. We show that a causal, time-invariant generator driven by i.i.d. noise cannot suppress its output spectrum over a band without collapsing its per-step innovation. We proposed FaceGAN, which dissolved the limitation by shaping the noise pathway acausally. Because the driving noise is synthetic, its future can be sampled now, so the audio-to-expression path stays causal, and the model supports fully causal operation. FaceGAN emits expression and head pose in a single forward pass per frame and matches or outperforms state-of-art approaches in generation quality. Being feed-forward with bounded attention windows, it generates indefinitely without drift.

Chinese Translation

基于音频驱动的面部动画是实时化身、远程呈现和具身虚拟代理的基础。该技术必须在线运行：每一帧的生成都依赖于截至当前时刻的音频，以交互速率进行处理。最近的进展主要由扩散模型主导，这些模型每个样本需要多次网络评估，因此不适合流媒体应用。我们认为在这一领域，这种成本是不必要的。音频条件下的面部运动占据了相对低维的流形，这种情况下单次通过生成对抗网络（GAN）就足够了。障碍并非是容量，而是随机结构。我们展示了一个由独立同分布（i.i.d.）噪声驱动的因果时间不变生成器无法在不崩溃其每步创新的情况下抑制其输出频谱。我们提出了FaceGAN，通过非因果地塑形噪声路径来消除这一限制。由于驱动噪声是合成的，其未来可以现在进行采样，因此音频到表情的路径保持因果性，模型支持完全因果操作。FaceGAN在每帧中以单次前向传递的方式生成表情和头部姿态，并在生成质量上与最先进的方法相匹配或超越。作为前馈模型，具有有限的注意力窗口，它能够无限生成而不出现漂移。

View on arXiv Download PDF AI Translation

cs.CV / 29 / 2610.11086

Contrast Enhancement or Noise Reduction? On Improving Cervical Cancer Classification

对比度增强还是噪声减少？关于改善宫颈癌分类的研究

Khozaimi, Ach, Nahdhiyah, Ulfatun

Abstract

Purpose: Cervical cancer is one of the leading causes of mortality worldwide. Deep learning has shown promising performance in medical image classification. The influence of image preprocessing algorithms on classification performance remains insufficiently investigated in the literature. This research aims to evaluate the impact of image preprocessing algorithms on the performance of CNNs for Pap smear image classification. Methods: Three CNN architectures (ResNet-34, MobileNet-V2, and DenseNet-121) were trained and evaluated using the SIPaKMeD dataset. Two preprocessing algorithms were applied: the PMD filter for noise reduction and CLAHE for contrast enhancement. The model performance was assessed using a confusion matrix. Results: Preprocessing improved the classification performance of all models. CLAHE significantly increased the accuracy of ResNet-34 from 76.73% to 84.16% and DenseNet-121 from 76.73% to 84.16%. The PMD filter yielded limited improvement and slightly reduced the MobileNet-V2 performance. Novelty: This research provides a systematic comparison of contrast enhancement and noise reduction techniques across CNN architectures. This research demonstrates that contrast enhancement is more effective than noise reduction in improving CNN performance. The research provides new pipelines for improving cervical cancer classification.

Chinese Translation

目的：宫颈癌是全球主要的死亡原因之一。深度学习在医学图像分类中表现出良好的性能。文献中对图像预处理算法对分类性能的影响研究仍然不足。本研究旨在评估图像预处理算法对卷积神经网络（CNN）在宫颈抹片图像分类性能的影响。方法：使用SIPaKMeD数据集训练和评估三种CNN架构（ResNet-34、MobileNet-V2和DenseNet-121）。应用了两种预处理算法：用于噪声减少的PMD滤波器和用于对比度增强的CLAHE。通过混淆矩阵评估模型性能。结果：预处理提高了所有模型的分类性能。CLAHE显著提高了ResNet-34的准确率，从76.73%提升至84.16%，DenseNet-121的准确率也从76.73%提升至84.16%。PMD滤波器的改进有限，并略微降低了MobileNet-V2的性能。创新性：本研究系统地比较了不同CNN架构下的对比度增强和噪声减少技术。研究表明，对比度增强在提升CNN性能方面比噪声减少更为有效。该研究为改善宫颈癌分类提供了新的方法。

View on arXiv Download PDF AI Translation

cs.CV / 30 / 2610.11096

Continuous Ground-Truth Construction and a Recovery Policy for Air--Water Robotic Tracking

连续真实数据构建及空气-水域机器人跟踪的恢复策略

Xiao, Jiangong, Sun, Zhe, Yao, Kanzhong, Bi, Yuanbo, Zhao, Haofei, Hu, Ruixuan, Huang, Guan, Li, Xuelong

Abstract

Visual tracking across the air-water interface is challenged by splashes, bubbles, refraction, reflections, and abrupt appearance changes that can temporarily invalidate observations. This setting poses two coupled difficulties: first, for evaluation, image-only annotation cannot reliably describe the target's physical location during visual blindness; second, for online tracking, corrupted observations can contaminate motion estimates and appearance templates. We address the first difficulty with a construction pipeline that synchronizes camera frames with motion-capture poses, projects known target geometry, corrects underwater projection with a medium-gated residual, and subjects the annotations to manual review. This yields an evaluation-only cross-medium test set of 22,346 frames. We further introduce a Cross-Medium Recovery Policy (CMRP) centered on confidence-triggered template selection. It supplies MixFormerV2 with the fixed initial template, a window-best pre-trigger template, and a trigger-frame Kalman-guided image crop, together with their associated weights, without retraining the visual backbone. In the accuracy evaluation, CMRP achieves 49.90 Macro Success AUC, 2.95 points above MixFormerV2 Official. On selected cross-medium transition and occlusion-recovery intervals, CMRP increases MixFormerV2 tracking coverage from 47.91\% to 50.43\% relative to Official updating, while mean loss-to-recovery latency over successfully recovered videos decreases from 55.3 to 49.3 frames.

Chinese Translation

在空气-水界面进行视觉跟踪面临着水花、气泡、折射、反射以及突发外观变化等挑战，这些因素可能暂时使观测失效。这一环境带来了两个相互关联的难题：首先，在评估过程中，仅依赖图像注释无法可靠地描述目标在视觉盲区内的物理位置；其次，在在线跟踪中，受损的观测可能会污染运动估计和外观模板。我们通过一个构建管道来解决第一个难题，该管道将相机帧与运动捕捉姿态同步，投影已知目标几何形状，利用介质门控残差校正水下投影，并对注释进行人工审核。这产生了一个仅用于评估的跨介质测试集，共22,346帧。我们进一步提出了一种以置信度触发模板选择为中心的跨介质恢复策略（Cross-Medium Recovery Policy, CMRP）。该策略为MixFormerV2提供固定的初始模板、窗口最佳的预触发模板以及触发帧卡尔曼引导的图像裁剪，连同其相关权重，而无需重新训练视觉主干。在准确性评估中，CMRP达到了49.90的宏观成功AUC，比MixFormerV2官方版本高出2.95分。在选定的跨介质过渡和遮挡恢复间隔中，CMRP将MixFormerV2的跟踪覆盖率从47.91%提高至50.43%，相较于官方更新，同时成功恢复视频的平均损失到恢复延迟从55.3帧减少至49.3帧。

View on arXiv Download PDF AI Translation

cs.CV / 31 / 2610.11104

Skeleton-Guided Progressive Test-Time Adaptation for Thin Curvilinear Structures

基于骨架的渐进式测试时间适应用于细长曲线结构

Jang, Boa, Lee, JunGyu, Lee, Gwanho, Choi, Jinwook, Kim, Young-Gon

Abstract

Accurate segmentation of thin curvilinear structures is vital for various real-world applications, from vessel analysis to road extraction. Yet their intricate geometry makes even minor pixel-wise errors enough to break the global topology, and this structural fragility turns severe domain shifts into catastrophic failures. The difficulty is most acute under cross-modality gaps, where the imaging process itself differs fundamentally between source and target. While test-time adaptation (TTA) offers a practical source-free remedy, existing methods adapt feature statistics and confidence, neither of which constrains connectivity, and thus degrade under such extreme gaps. To address this, we propose Skeleton-Guided Progressive Test-Time Adaptation (SGP-TTA). Progressive Batch Normalization (ProgBN) shifts normalization from frozen source statistics toward current target estimates under a sample-count schedule, so that the source-target balance follows the stage of adaptation rather than a fixed coefficient. Consensus Skeleton Recall (CSR) then derives a structural target from geometrically aligned multi-view predictions and updates only the BN affine parameters to preserve connected structures. Extensive experiments show that SGP-TTA consistently outperforms existing TTA methods in topological connectivity, with the largest margins under cross-modality shift. The project page is available at https://boa-jang.github.io/SGP-TTA.

Chinese Translation

细长曲线结构的准确分割对从血管分析到道路提取等各种现实应用至关重要。然而，它们复杂的几何形状使得即使是微小的像素级错误也足以破坏全局拓扑，这种结构脆弱性使得严重的领域转移变成灾难性的失败。在跨模态差异下，这种困难尤为明显，因为源和目标之间的成像过程本质上存在根本性差异。尽管测试时间适应（TTA）提供了一种实用的无源解决方案，但现有方法仅适应特征统计和置信度，这两者都无法约束连通性，因此在这种极端差距下表现不佳。为了解决这个问题，我们提出了基于骨架的渐进式测试时间适应（SGP-TTA）。渐进式批量归一化（ProgBN）在样本计数计划下将归一化从冻结的源统计转向当前的目标估计，从而使源-目标平衡遵循适应阶段而不是固定系数。共识骨架召回（CSR）则从几何对齐的多视图预测中推导出结构目标，并仅更新BN的仿射参数以保持连通结构。大量实验表明，SGP-TTA在拓扑连通性方面始终优于现有的TTA方法，尤其在跨模态转移下表现出最大的优势。项目页面可访问 https://boa-jang.github.io/SGP-TTA。

View on arXiv Download PDF AI Translation

cs.CV / 32 / 2610.11105

TKCAM: Text and Keyframe to Camera Trajectory Generation

TKCAM：文本和关键帧到相机轨迹生成

Yang, Haozhe, Dou, Zhiyang, Gu, Zekai, Lin, Cheng, Wang, Wenping, Liu, Yuan, Komura, Taku

Abstract

Generating high-quality and controllable camera motion is essential for AI-assisted cinematography, video synthesis, and 3D scene understanding. We introduce TKCAM, a Text- and Keyframe-conditioned CAMera-motion synthesis framework based on generative masked modeling. We represent camera dynamics using a 12-dimensional kinematic feature comprising position, velocity, and a continuous rotation representation and discretize them into hierarchical motion tokens via a Residual Vector Quantizer (RVQ). A two-stage masked transformer architecture then learns to reconstruct and refine these tokens, utilizing explicit self- and cross-attention modules for multimodal conditioning. A central feature of our framework is sparse visual keyframe conditioning: users can provide free-form text prompts together with RGB observations at selected timestamps, which provide temporally localized visual guidance for generating coherent in-between trajectories. Furthermore, to advance evaluation standards, we curate RealEstate10K-Cap, a large-scale text-camera dataset, and establish a cross-domain benchmark with a Universal CLaTr Evaluator. Extensive experiments demonstrate that TKCAM surpasses recent state-of-the-art baselines on Fr\'echet distance (FID), text-motion matching scores, and retrieval metrics (R@K), while additional analyses evaluate temporal smoothness and cross-domain generalization. Code is available at https://github.com/linearalgebrayhz/TKCAM.

Chinese Translation

生成高质量且可控的相机运动对于人工智能辅助的电影摄影、视频合成和三维场景理解至关重要。我们提出了TKCAM，这是一个基于生成掩模建模的文本和关键帧条件下的相机运动合成框架。我们使用一个包含位置、速度和连续旋转表示的12维运动特征来表示相机动态，并通过残差向量量化器（Residual Vector Quantizer, RVQ）将其离散化为分层运动标记。然后，采用两阶段的掩模变换器架构学习重建和细化这些标记，利用显式的自注意力和交叉注意力模块进行多模态条件处理。我们框架的一个核心特征是稀疏视觉关键帧条件：用户可以在选定时间戳提供自由形式的文本提示以及RGB观测，这为生成连贯的中间轨迹提供了时间上局部化的视觉指导。此外，为了推动评估标准的发展，我们策划了RealEstate10K-Cap，这是一个大规模的文本-相机数据集，并建立了一个跨领域基准，使用通用的CLaTr评估器（Universal CLaTr Evaluator）。大量实验表明，TKCAM在Fréchet距离（FID）、文本-运动匹配得分和检索指标（R@K）上超越了最近的最先进基线，同时额外分析评估了时间平滑性和跨领域泛化能力。代码可在https://github.com/linearalgebrayhz/TKCAM获取。

View on arXiv Download PDF AI Translation

cs.CV / 33 / 2610.11113

DiscoVL: Unveiling Disentangled C ross-Modal Representation Learning via Orthogonal Adversarial Regularization for V ision-Language Models

DiscoVL：通过正交对抗正则化揭示视觉-语言模型的解耦跨模态表示学习

Dong, Mengping, Li, Jinbao, Li, Fei

Abstract

Pre-trained vision-language models excel across varied perception tasks, but adapting them to novel downstream settings without sacrificing generalization remains non-trivial. Existing parameter-efficient prompt learning method often yields inconsistent representations and fails to account for semantic distribution shifts. In this work, we present DiscoVL, a disentangled cross-modal representation learning framework that couples orthogonal adversarial regularization with structured cross-modal alignment for vision-language models. To address the insufficient cross-modal interaction, our DiscoVL designs a multi-branch low-rank residual aligner that decomposes representations into subspaces and enables bidirectional cross-modal feedback between visual and textual streams at each layer. Furthermore, while conventional triplet constraints overfit features to class centroids, we design an orthogonal regularization for adversarial triplet loss, which prevents centroid collapse and substantially boosts generalization. Evaluations on 15 benchmarks demonstrate that DiscoVL delivers consistent improvements over state-of-the-art methods for base-to-novel generalization, cross-dataset evaluation, and few-shot learning

Chinese Translation

预训练的视觉-语言模型在各种感知任务中表现出色，但在不牺牲泛化能力的情况下将其适应于新的下游设置仍然具有挑战性。现有的参数高效提示学习方法往往导致不一致的表示，并未能考虑语义分布的变化。在本研究中，我们提出了DiscoVL，一个解耦的跨模态表示学习框架，该框架将正交对抗正则化与结构化跨模态对齐相结合，应用于视觉-语言模型。为了解决跨模态交互不足的问题，我们的DiscoVL设计了一个多分支低秩残差对齐器，该对齐器将表示分解为子空间，并在每一层实现视觉和文本流之间的双向跨模态反馈。此外，尽管传统的三元组约束会导致特征过拟合到类别中心，我们为对抗三元组损失设计了正交正则化，以防止中心崩溃并显著提升泛化能力。在15个基准测试上的评估表明，DiscoVL在基础到新颖的泛化、跨数据集评估和少样本学习方面，相较于最先进的方法提供了一致的改进。

View on arXiv Download PDF AI Translation

cs.CV / 34 / 2610.11117

MCL: Meta Convolution Layer

MCL：元卷积层

Reza, Naim, Amin, Md Al, Jung, Ho Yub

Abstract

Dynamic convolution enhances convolutional neural networks (CNNs) by adapting kernels to input content, but it expresses the effective kernel as a linear mixture of a small number of basis kernels, which limits expressivity and complicates optimization as the mixture size grows. In this work, we revisit dynamic convolution from a functional perspective and propose the Meta Convolution Layer (MCL), which directly models the convolutional kernel as an input-conditioned function W(x) realized via a high-order polynomial expansion. Leveraging nested residual blocks inspired by deep polynomial networks, MCL implements a structured polynomial meta-network that generates a single input-adaptive kernel, thereby decoupling representational power from the explicit number of mixture kernels and alleviating training instability. MCL is a plug-in addition with standard convolutions and can be seamlessly integrated into both CNN and transformer backbones. Experimental evaluation shows that adding MCL improves the Top-1 accuracy of Resnet- 18, Resnet-50 and ResNet-101 by 6.61%, 3.42% and 3.05% on the ImageNet dataset. Moreover, the proposed method significantly boosts the accuracy of Resnet and Wide-Resnet variants on CIFAR-10 and CIFAR-100 datasets. Additionally, the proposed method outperforms previous methods on fine-grained visual classification tasks using Swin and ViT backbones. These results demonstrate that high-order polynomial kernel generation is a powerful and scalable alternative to linear mixture based dynamic convolution.

Chinese Translation

动态卷积通过根据输入内容调整卷积核来增强卷积神经网络（CNN），但它将有效卷积核表示为少量基核的线性组合，这限制了表达能力，并且随着组合大小的增加，优化变得复杂。在这项工作中，我们从功能的角度重新审视动态卷积，并提出了元卷积层（Meta Convolution Layer, MCL），该层直接将卷积核建模为一个输入条件函数 W(x)，通过高阶多项式展开实现。利用受深度多项式网络启发的嵌套残差块，MCL 实现了一个结构化的多项式元网络，生成一个单一的输入自适应卷积核，从而将表示能力与显式的组合核数量解耦，并减轻训练的不稳定性。MCL 是一个与标准卷积的插件式扩展，可以无缝集成到 CNN 和变换器（transformer）主干中。实验评估表明，在 ImageNet 数据集上，添加 MCL 可以将 Resnet-18、Resnet-50 和 ResNet-101 的 Top-1 准确率分别提高 6.61%、3.42% 和 3.05%。此外，所提方法显著提升了 Resnet 和 Wide-Resnet 变体在 CIFAR-10 和 CIFAR-100 数据集上的准确率。此外，所提方法在使用 Swin 和 ViT 主干的细粒度视觉分类任务中优于之前的方法。这些结果表明，高阶多项式核生成是基于线性组合的动态卷积的强大且可扩展的替代方案。

View on arXiv Download PDF AI Translation

cs.CV / 35 / 2610.11138

IntrinSync: Joint Intrinsic Decomposition and Reciprocal Rendering

IntrinSync：联合内在分解与互逆渲染

Gu, Zheng, Huang, Rui, Zhang, Xilu, Zhang, Jingbo, Lu, Min, Sun, Zhida, Lischinski, Dani, Cohen-Or, Daniel, Huang, Hui

Abstract

Inverse rendering decomposes an image into intrinsic properties such as appearance, illumination, geometry, and material, yet these properties are inherently interdependent. A reliable decomposition should produce intrinsic maps that are not only individually plausible, but also mutually compatible in explaining the image. However, existing methods either model intrinsic channels in isolation or treat inverse and forward rendering as separate processes, leaving the interdependence underexploited. In this paper, we introduce IntrinSync, a unified framework that captures this interdependence through joint-channel modeling and reciprocal inverse-forward rendering. At the channel level, we jointly decompose an input RGB into albedo, shading, surface normal, roughness, and metallic maps through a 1-to-N mapping, enabling information exchange across channels throughout generation. At the process level, we establish inverse-forward reciprocity through a dual cycle-consistent objective that aligns corresponding predictions across a closed loop. Experiments on three datasets demonstrate that our method achieves competitive intrinsic estimation and forward rendering performance, improving coherence and physical consistency. Beyond decomposition, IntrinSync provides a physically grounded interface for image editing, allowing intrinsic properties to be explicitly manipulated and rendered back into RGB images.

Chinese Translation

逆向渲染将图像分解为内在属性，如外观、照明、几何形状和材料，但这些属性本质上是相互依赖的。一个可靠的分解应该产生不仅在个体上合理，而且在解释图像时相互兼容的内在图。然而，现有的方法要么孤立地建模内在通道，要么将逆向渲染和正向渲染视为独立的过程，从而未能充分利用这种相互依赖性。本文提出了IntrinSync，一个统一框架，通过联合通道建模和互逆的逆向-正向渲染捕捉这种相互依赖性。在通道层面，我们通过1对N映射将输入RGB联合分解为反射率、阴影、表面法线、粗糙度和金属图，从而在生成过程中实现通道间的信息交换。在过程层面，我们通过一个双循环一致性目标建立逆向-正向的互惠关系，使得在闭环中对齐对应的预测。在三个数据集上的实验表明，我们的方法在内在估计和正向渲染性能上都达到了竞争水平，提高了一致性和物理一致性。除了分解，IntrinSync还提供了一个物理基础的图像编辑接口，允许对内在属性进行明确的操控，并将其渲染回RGB图像中。

View on arXiv Download PDF AI Translation

cs.CV / 36 / 2610.11144

Improving Image-Based Nutrition Estimation Through Multimodal Food-Item Verification and Recovery

通过多模态食物项验证与恢复提升基于图像的营养估计

Yue, Jingbo, Coburn, Bruce, Ma, Jinge, Chi, Jui-Feng, Zhu, Fengqing

Abstract

Single-image nutrition estimation can fail silently when visible foods are missed. Even when a food is correctly identified, its proposed region may not support portion estimation. We propose a framework that uses multimodal large language models (MLLMs) to inventory visible foods and separately verify food identity and whether each proposed 2D region supports portion estimation. One whole-image review uses these verification results to identify unresolved gaps and omitted foods, triggering at most one targeted recovery pass. Recovered regions are re-verified without access to the recovery prompt, then reconciled into a final item set for nutrition estimation. The framework requires no task-specific fine-tuning. Matched evaluation on common valid-output samples shows that item-level grounding improves mass accuracy across all tested settings and energy accuracy relative to an adapted retrieval baseline, with item-identity precision and recall also improving, while post-recovery visual coverage is assessed separately at inference time without ground-truth annotations.

Chinese Translation

单张图像的营养估计在遗漏可见食物时可能无声失败。即使食物被正确识别，其所提议的区域也可能无法支持份量估计。我们提出了一个框架，利用多模态大语言模型（MLLMs）对可见食物进行清点，并分别验证食物身份及每个提议的二维区域是否支持份量估计。一次整体图像复核利用这些验证结果识别未解决的空白和遗漏的食物，从而触发最多一次有针对性的恢复过程。恢复的区域在无恢复提示的情况下重新验证，随后整合为最终的食物项集合以进行营养估计。该框架无需特定任务的微调。针对常见有效输出样本的匹配评估表明，基于食物项的定位提升了所有测试设置下的质量准确度及相较于改编检索基线的能量准确度，同时食物身份的精确率和召回率也有所提升，而恢复后的视觉覆盖度则在推理阶段单独评估，无需真实标注。

View on arXiv Download PDF AI Translation

cs.CV / 37 / 2610.11148

SP-DocReader: Difference-Aware Self-Play for Precise Document OCR

SP-DocReader：差异感知自我对弈以实现精确文档OCR

Liao, Wenjie, Song, Xiaohui, Zhao, Liangjie, Lu, Haonan

Abstract

Accurate page transcription remains difficult for vision language models under limited input and training budgets. We present SP-DocReader, a self-play framework for optical character recognition (OCR) that targets residual errors after supervised fine-tuning. Reading Discrepancy Masking aligns reference and generated model tokens through a longest common subsequence, then scores unmatched positions with their full conditioning prefixes. Focused Fidelity Loss adds direct negative log-likelihood supervision at unmatched ground-truth positions. Only the OCR module is trained, while the backbone remains frozen. We derive the combined gradient to distinguish relative score optimization from direct supervision. Compared with SFT-2, SP-DR-3 reduces Vary-600K character error rate on both backbones. On Qwen3-VL-4B, it reduces character error rate by approximately 54 percent and improves DocVQA Average Normalized Levenshtein Similarity (ANLS) by 3.7 points. These results show the value of focusing self-play training on the discrepancies that remain after supervised fine-tuning.

Chinese Translation

在有限的输入和训练预算下，准确的页面转录仍然对视觉语言模型构成挑战。我们提出了SP-DocReader，一个针对光学字符识别（OCR）的自我对弈框架，旨在解决监督微调后遗留的错误。阅读差异掩蔽通过最长公共子序列对齐参考和生成模型的标记，然后根据其完整的条件前缀对不匹配的位置进行评分。聚焦保真损失在不匹配的真实位置添加了直接的负对数似然监督。仅训练OCR模块，而主干网络保持冻结。我们推导出组合梯度，以区分相对评分优化与直接监督。与SFT-2相比，SP-DR-3在两个主干网络上均降低了Vary-600K字符错误率。在Qwen3-VL-4B上，它将字符错误率降低了约54%，并将文档视觉问答平均归一化莱文斯坦相似度（ANLS）提高了3.7分。这些结果表明，专注于监督微调后仍然存在的差异进行自我对弈训练的价值。

View on arXiv Download PDF AI Translation

cs.CV / 38 / 2610.11149

A Unified Score Matching Paradigm for Video Anomaly Detection and Anticipation

视频异常检测与预测的统一评分匹配范式

Cao, Congqi, Liang, Zhenhe, Zhang, Hanwen, Zhao, Yifan, Lv, Qinyi, Min, Lingtong, Zhang, Yanning

Abstract

Video anomaly detection (VAD) is a fundamental and safety-critical task in computer vision. Recent generative approaches detect anomalies from a distributional perspective, but remain limited by local anomaly modes. Meanwhile, video anomaly anticipation (VAA), as a proactive extension beyond post-hoc detection, introduces additional challenges. In particular, the contrastive inference paradigm in VAD, which relies on ground-truth frames, is not applicable to VAA, hindering its development. To address these challenges, we propose a unified score-driven framework, termed Uni-DSM, based on denoising score matching (DSM), which models anomaly patterns through likelihood estimation and score functions over the learned data distribution. Within this unified framework, we adopt a shared noise-conditioned score transformer backbone with scene-dependent embeddings and motion-aware weighting for distribution-level modeling. Instead of introducing separate architectures, Uni-DSM unifies VAD and VAA through different inference and supervision paradigms built upon the same score-based formulation. For VAD, we instantiate an autoregressive denoising score matching (ADSM) mechanism, which progressively accumulates anomalous evidence via autoregressive denoising, enabling enhanced perception of local modes beyond visual cues. For VAA, we extend the same architecture by incorporating a lightweight auxiliary decoder and a novel self-distilled denoising score matching (SDSM) mechanism. By constructing supervision from output discrepancies instead of relying on unavailable future ground truth, our method achieves efficient training suitable or early anomaly anticipation. Extensive experiments on multiple benchmark datasets demonstrate state-of-the-art performance in both VAD and VAA while maintaining high efficiency, establishing a unified and scalable pipeline from anomaly detection to anticipation.

Chinese Translation

视频异常检测（VAD）是计算机视觉中的一项基础且至关重要的任务。近期的生成方法从分布的角度检测异常，但仍然受到局部异常模式的限制。同时，视频异常预测（VAA）作为一种超越事后检测的主动扩展，带来了额外的挑战。特别是，VAD中的对比推理范式依赖于真实帧，这在VAA中并不适用，阻碍了其发展。为了解决这些挑战，我们提出了一种基于去噪评分匹配（DSM）的统一评分驱动框架，称为Uni-DSM，该框架通过对学习数据分布的似然估计和评分函数建模异常模式。在这个统一框架内，我们采用了共享的噪声条件评分变换器骨干网络，结合场景依赖的嵌入和运动感知加权进行分布级建模。Uni-DSM通过基于相同评分公式构建不同的推理和监督范式，将VAD和VAA统一，而不是引入单独的架构。对于VAD，我们实例化了一种自回归去噪评分匹配（ADSM）机制，通过自回归去噪逐步累积异常证据，从而增强对超越视觉线索的局部模式的感知。对于VAA，我们通过引入轻量级辅助解码器和新颖的自蒸馏去噪评分匹配（SDSM）机制扩展了相同的架构。通过从输出差异构建监督，而不是依赖不可用的未来真实值，我们的方法实现了适合早期异常预测的高效训练。在多个基准数据集上的广泛实验表明，我们的方法在VAD和VAA中均实现了最先进的性能，同时保持高效率，建立了从异常检测到预测的统一且可扩展的流程。

View on arXiv Download PDF AI Translation

cs.CV / 39 / 2610.11153

CARE: Constrained Attention Refinement for Fine-Grained Visual Classification via Teacher-Student Distillation

CARE：通过教师-学生蒸馏实现细粒度视觉分类的约束注意力精炼

Wen, Ruibo, Shao, Hang, Lei, Yiming

Abstract

Fine-grained visual classification requires models to recognize subtle local traits while exposing the visual evidence behind their predictions. Class-specific attention pathways provide a natural basis for interpretable recognition, but their constrained prediction structure limits discriminative capacity and underuses intermediate representations from strong pretrained backbones. To address this problem, we propose CARE, a constrained attention refinement framework for interpretable fine-grained recognition via teacher-student distillation. CARE keeps the final prediction and explanation within a class-specific attention student, while introducing a training-only auxiliary query teacher that reads selected intermediate DINOv2 layers with learnable queries. The teacher fuses multi-level representations and transfers logit-standardized class-discriminative knowledge to the student. To further refine the explanation pathway, we design diversity and sparsity terms to regularize student attention heads, reducing redundancy and encouraging compact trait localization. Experiments on CUB, Oxford-IIIT Pet, Stanford Dogs, and Stanford Cars show that CARE achieves strong classification performance under an interpretable frozen-backbone setting, reaching 78.5% Top-1 accuracy on CUB. Faithfulness analysis with insertion and deletion metrics further indicates that the top-ranked attention regions retain class-relevant evidence for explanation.

Chinese Translation

细粒度视觉分类要求模型识别微妙的局部特征，同时揭示其预测背后的视觉证据。特定类别的注意力路径为可解释的识别提供了自然基础，但其受限的预测结构限制了判别能力，并未充分利用强大的预训练骨干网络的中间表示。为了解决这个问题，我们提出了CARE，一个通过教师-学生蒸馏实现可解释细粒度识别的约束注意力精炼框架。CARE在特定类别的注意力学生中保持最终的预测和解释，同时引入一个仅用于训练的辅助查询教师，该教师通过可学习的查询读取选定的中间DINOv2层。教师融合多层次表示，并将对数标准化的类别判别知识传递给学生。为了进一步精炼解释路径，我们设计了多样性和稀疏性项来规范学生的注意力头，减少冗余并鼓励紧凑的特征定位。在CUB、Oxford-IIIT Pet、Stanford Dogs和Stanford Cars上的实验表明，CARE在可解释的冻结骨干设置下实现了强大的分类性能，在CUB上达到了78.5%的Top-1准确率。使用插入和删除指标的可信度分析进一步表明，排名最高的注意力区域保留了与类别相关的证据以供解释。

View on arXiv Download PDF AI Translation

cs.CV / 40 / 2610.11161

VGGTWorld-VLA: Intent-Conditioned 3D World Evolution for Autonomous Driving

VGGTWorld-VLA：基于意图条件的自主驾驶3D世界演化

Liu, Zhaoyang, Jiang, Kun, Song, Ziying, Yang, Diange

Abstract

VGGT provides a strong foundation for geometry-centric world models by recovering unified 3D scene geometry from visual observations. Although recent extensions enable temporal 3D prediction, their future evolution remains weakly conditioned on driving intentions and actions, limiting their ability to model alternative action-dependent futures. We propose VGGTWorld-VLA, an intention-conditioned extension of VGGT-World for controllable 3D world evolution in autonomous driving. First, we introduce an action--semantic conditioning mechanism that injects complementary driving semantics and ego-motion representations into the future-token stream, enabling different future geometry predictions for the same observed scene under alternative ego actions. Second, we develop a geometry--language--action bridge that adapts historical geometry, VLA semantic features, and maneuver and trajectory representations for joint conditioning of future geometry prediction. We evaluate future geometry prediction on NAVSIM, while conditioning ablations further examine the contributions of semantic and action information. Compared with the baseline, our method demonstrates competitive geometry prediction performance. Ablation studies further support the effectiveness of semantic and action conditioning. These results demonstrate the potential of semantic and action conditioning for controllable VGGT-based world prediction in autonomous driving.

Chinese Translation

VGGT为以几何为中心的世界模型提供了坚实的基础，通过从视觉观察中恢复统一的3D场景几何体。尽管最近的扩展使得时间维度的3D预测成为可能，但它们的未来演化仍然对驾驶意图和动作的条件设置较弱，限制了它们对依赖于动作的替代未来的建模能力。我们提出了VGGTWorld-VLA，这是VGGT-World的一个基于意图条件的扩展，旨在实现自主驾驶中可控的3D世界演化。首先，我们引入了一种动作-语义条件机制，将补充的驾驶语义和自我运动表示注入未来标记流中，使得在不同的自我动作下，对同一观察场景的未来几何预测能够有所不同。其次，我们开发了一个几何-语言-动作桥接，适应历史几何、VLA语义特征以及机动和轨迹表示，以共同条件化未来几何预测。我们在NAVSIM上评估未来几何预测，同时条件消融实验进一步考察了语义和动作信息的贡献。与基线相比，我们的方法展示了具有竞争力的几何预测性能。消融研究进一步支持了语义和动作条件化的有效性。这些结果展示了语义和动作条件化在自主驾驶中基于VGGT的世界预测的可控性潜力。

View on arXiv Download PDF AI Translation

cs.CV / 41 / 2610.11162

AutoAdapt: Reliable Few-Shot Adaptation under Clinical Distribution Shifts

AutoAdapt：在临床分布变化下的可靠少样本适应

Wang, Song, Peng, Jie, Hobley, Davis, Plotkin, Zachary, Chen, Tianlong

Abstract

Large pretrained clinical models provide a practical way to reuse learned prior knowledge across hospitals by adapting models to them. In practice, a target hospital may only have a small labeled patient cohort, a setting commonly referred to as few-shot adaptation. This requires making multiple decisions, such as which pretrained model to adapt, how much of the model to update, and which patients to use. Nevertheless, this process faces two primary challenges. First, the best adaptation strategy varies across clinical tasks. Second, evaluating and comparing candidate strategies becomes unreliable due to the small patient cohort. In this work, we introduce AutoAdapt with two core designs to deal with these challenges. The Adapter defines an extensible space of adaptation recipes, and the Automator forms a weighted recipe combination from evidence within the adaptation patients. We propose a reliability rule to ensure that only the most effective strategy on most available patients will be selected. These selected strategies then form a combination for effective few-shot adaptation. We conduct extensive experiments across critical care, emergency care, and diagnostic datasets, and the results show that AutoAdapt consistently achieves state-of-the-art performance using only a few patients for adaptation.

Chinese Translation

大型预训练临床模型为在医院之间重用学习到的先验知识提供了一种实用的方法，通过将模型适应于特定医院。在实际操作中，目标医院可能只有少量标记的患者群体，这种情况通常被称为少样本适应。这需要做出多个决策，例如选择适应哪个预训练模型、更新模型的多少以及使用哪些患者。然而，这一过程面临两个主要挑战。首先，最佳适应策略在不同的临床任务中有所不同。其次，由于患者群体较小，评估和比较候选策略变得不可靠。在本研究中，我们引入了AutoAdapt，采用两种核心设计来应对这些挑战。Adapter定义了一个可扩展的适应配方空间，而Automator则根据适应患者中的证据形成加权的配方组合。我们提出了一条可靠性规则，以确保仅选择在大多数可用患者中最有效的策略。这些选定的策略随后形成有效的少样本适应组合。我们在重症监护、急救和诊断数据集上进行了广泛的实验，结果表明，AutoAdapt在仅使用少量患者进行适应时，始终实现了最先进的性能。

View on arXiv Download PDF AI Translation

cs.CV / 42 / 2610.11171

VAMR: Multi-Question Agentic Reasoning for Efficient Long-Form Video Understanding

VAMR：高效长视频理解的多问题代理推理

Gui, Runquan, Chen, Hanzhu, Wang, Zehao, Zhu, Hanxin, Li, Xin, Chen, Zhibo

Abstract

Long-form video understanding often involves multiple questions about different aspects of the same recording. Yet existing video agents typically process each question through an isolated tool-use trajectory. This repeatedly restarts video exploration and memory construction, missing opportunities to acquire evidence jointly and progressively build a shared understanding that supports the complete question set. We introduce \textbf{VAMR} (\textbf{V}ideo \textbf{A}gent for \textbf{M}ulti-Question \textbf{R}easoning), which coordinates all questions about a video through one shared tool-use trajectory. At each round, a persistent policy model can invoke tools for one or more unresolved questions and submit answers for questions with sufficient evidence. Question-conditioned visual perception retrieves fine-grained clues for several questions in one call, while layered multi-question memory integrates reusable context into a shared video story and preserves separate evidence for individual questions. After supervised fine-tuning initializes this interaction protocol, we propose question-horizon policy optimization (\qhpo) to optimize shared trajectories in which questions progress and finish at different rounds. Specifically, a question-level critic estimates the value of each active question, while round alignment maps each question advantage to the rounds that directly serve it before the aligned advantages are aggregated to optimize the shared actor. Across LVBench, Video-Holmes, and LongVideoBench, VAMR achieves the highest accuracy overall and the fewest reasoning rounds among iterative methods. On LVBench, it reaches 62.1\% accuracy, exceeding VideoARM by \textbf{4.3} points while reducing reasoning rounds and processed frames by \textbf{85.9\%} and \textbf{61.4\%}.

Chinese Translation

长视频理解通常涉及对同一录制内容的不同方面提出多个问题。然而，现有的视频代理通常通过孤立的工具使用轨迹处理每个问题。这种方式反复启动视频探索和记忆构建，错失了共同获取证据和逐步建立支持完整问题集的共享理解的机会。我们提出了 extbf{VAMR}（ extbf{V}ideo extbf{A}gent for extbf{M}ulti-Question extbf{R}easoning），该方法通过一个共享的工具使用轨迹协调关于视频的所有问题。在每一轮中，一个持久的策略模型可以为一个或多个未解决的问题调用工具，并提交具有充分证据的问题的答案。基于问题的视觉感知在一次调用中检索多个问题的细粒度线索，而分层的多问题记忆则将可重用的上下文整合到一个共享的视频故事中，并为各个问题保留单独的证据。在监督微调初始化这一交互协议后，我们提出了问题视野策略优化（question-horizon policy optimization, extbf{qhpo}），以优化问题在不同轮次中进展和完成的共享轨迹。具体而言，问题级别的评论者评估每个活跃问题的价值，而轮次对齐将每个问题的优势映射到直接服务于它的轮次，然后将对齐的优势汇总以优化共享的行动者。在LVBench、Video-Holmes和LongVideoBench上，VAMR在迭代方法中实现了最高的整体准确率和最少的推理轮次。在LVBench上，其准确率达到62.1\%，比VideoARM高出 extbf{4.3}个百分点，同时推理轮次和处理帧数分别减少了 extbf{85.9\%}和 extbf{61.4\\%}。

View on arXiv Download PDF AI Translation

cs.CV / 43 / 2610.11174

IntactWorld: Joint World Modeling with Intact Features

IntactWorld：具有完整特征的联合世界建模

Tan, Boming, Zhang, Xiangdong, Xia, Yan, Zhu, Qi, Ji, Deyi, Yang, Xue, Zhang, Shaofeng

Abstract

While recent video generation models synthesize highly realistic visuals, they lack a genuine understanding of intrinsic real-world logic. Existing methods attempt to understand the world by internalizing diverse world knowledge, yet constrained by computational overhead or dimensionality alignment, their learning processes inevitably compress features, causing a severe loss of structural information. To address this, we propose \textbf{IntactWorld}, a \textbf{Joint World Modeling Architecture} utilizing uncompressed \textbf{Intact Features}. Since data naturally reside on a low-dimensional manifold within a high-dimensional space, predicting the flow velocity $v$ within this uncompressed high-dimensional space induces a severe manifold gap. To successfully eliminate this optimization bottleneck, our framework instead predicts the clean feature $x_0$ at intermediate layers. Furthermore, to mitigate the computational overhead of incorporating complete world knowledge, we introduce a \textit{Full-to-Compact Training Paradigm}. By replacing raw full features with highly refined CLS tokens, this paradigm enables efficient single-branch guidance, reducing spatial memory consumption by 11.4\% and cutting inference latency by 43.8\%. Extensive evaluations demonstrate the effectiveness of IntactWorld, outperforming established baselines by 2.46 points on the VBench 2.0 benchmark.

Chinese Translation

尽管最近的视频生成模型合成了高度逼真的视觉效果，但它们缺乏对内在现实世界逻辑的真正理解。现有方法试图通过内化多样的世界知识来理解世界，但由于计算开销或维度对齐的限制，它们的学习过程不可避免地压缩特征，导致结构信息的严重损失。为了解决这个问题，我们提出了 extbf{IntactWorld}，一种利用未压缩 extbf{完整特征}的 extbf{联合世界建模架构}。由于数据自然位于高维空间中的低维流形上，在这个未压缩的高维空间中预测流动速度$v$会导致严重的流形间隙。为了成功消除这一优化瓶颈，我们的框架改为在中间层预测干净特征$x_0$。此外，为了减轻整合完整世界知识的计算开销，我们引入了一种 extit{全到紧凑训练范式}。通过用高度精炼的CLS标记替换原始完整特征，该范式实现了高效的单分支引导，减少了11.4 ext{%}的空间内存消耗，并将推理延迟降低了43.8 ext{%}。大量评估证明了IntactWorld的有效性，在VBench 2.0基准测试中超越了既定基线2.46分。

View on arXiv Download PDF AI Translation

cs.CV / 44 / 2610.11176

Multimodal Remote Sensing Image Registration: A Comprehensive Review, Challenges and Prospects

多模态遥感图像配准：全面综述、挑战与前景

Han, Zhiqiang, Ye, Yuanxin, Wu, Qiuyun, Chen, Jinhao, Zhu, Bai, Hao, Siyuan

Abstract

Multimodal remote sensing image registration is a crucial prerequisite for the collaborative processing and downstream application of remote sensing data, such as image fusion, change detection, and target recognition. However, significant variations in radiometry, geometry, scale, viewpoint, and time often exist between multimodal images. These differences, driven by varying sensor geometries, physical radiation mechanisms, imaging platforms, and environmental disturbances, pose severe challenges to achieving high-precision, robust registration. This paper systematically reviews the progress of mainstream multimodal remote sensing image registration methods. Based on their registration pipelines, existing approaches are categorized into three main types: region-based, feature-based, and deep learning-based methods. We detail the core principles, representative algorithms, advantages, and limitations of each category. Additionally, we summarize publicly available multimodal image datasets in the remote sensing domain, analyzing their specific characteristics and applicable scenarios. Finally, we highlight current bottlenecks in high-precision registration research and outline future development trends. This review aims to provide a comprehensive reference and valuable insights for researchers in related fields.

Chinese Translation

多模态遥感图像配准是遥感数据协同处理和下游应用（如图像融合、变化检测和目标识别）的重要前提。然而，多模态图像之间往往存在显著的辐射度、几何形状、尺度、视角和时间的变化。这些差异源于传感器几何、物理辐射机制、成像平台和环境干扰的不同，给实现高精度、稳健的配准带来了严峻挑战。本文系统回顾了主流多模态遥感图像配准方法的进展。根据其配准流程，现有方法被分为三大类：基于区域的方法、基于特征的方法和基于深度学习的方法。我们详细阐述了每一类的核心原理、代表性算法、优缺点。此外，我们总结了遥感领域中公开可用的多模态图像数据集，分析其具体特征和适用场景。最后，我们强调了高精度配准研究中的当前瓶颈，并概述了未来的发展趋势。本综述旨在为相关领域的研究人员提供全面的参考和宝贵的见解。

View on arXiv Download PDF AI Translation

cs.CV / 45 / 2610.11181

MATE4D: Matrix-Guided Editable 4D Generation from a Single Image

MATE4D：基于矩阵引导的可编辑4D生成从单幅图像

Chen, Xiaotian, Yin, Dongfu

Abstract

Generative models have rapidly pushed content creation be-yond 2D imagery toward dynamic 3D and 4D scene synthesis. Yet pro-ducing realistic and temporally stable 4D content from a single image is still difficult because one view provides limited structural cues and weak motion evidence. We introduce MATE4D, a framework that converts one input image into editable dynamic 4D content. Our method constructs a spatio-temporal multi-view image matrix with text-guided background manipulation, delivering coherent supervision over viewpoint, appear-ance, and motion. These synthesized observations are used to optimize 3D Gaussian primitives, which are then animated through a lightweight deformation module to form a 4D representation. The resulting scenes preserve geometry more faithfully, maintain smoother temporal behavior, and keep background edits more consistent, reducing context ambiguity and motion artifacts. Experiments on Objaverse-XL and Diffusion4D show that MATE4D outperforms strong baselines in visual quality, effi-ciency, and controllability, supporting practical AR/VR content creation.

Chinese Translation

生成模型迅速推动内容创作超越2D图像，朝向动态3D和4D场景合成。然而，从单幅图像生成真实且时间稳定的4D内容仍然困难，因为单一视角提供的结构线索有限且运动证据薄弱。我们提出了MATE4D，一个将单个输入图像转换为可编辑动态4D内容的框架。我们的方法构建了一个时空多视图图像矩阵，并通过文本引导的背景操控，提供对视角、外观和运动的连贯监督。这些合成的观察结果用于优化3D高斯原语，随后通过轻量级变形模块进行动画处理，以形成4D表示。生成的场景更忠实地保留了几何形状，保持了更平滑的时间行为，并使背景编辑更一致，从而减少了上下文模糊和运动伪影。在Objaverse-XL和Diffusion4D上的实验表明，MATE4D在视觉质量、效率和可控性方面优于强基线，支持实际的AR/VR内容创作。

View on arXiv Download PDF AI Translation

cs.CV / 46 / 2610.11184

WorldFact-Bench: Beyond Image-Internal Plausibility to Image-World Consistency

WorldFact-Bench：超越图像内部可信度，实现图像与世界的一致性

Chen, Zhuohong, Wu, Zhengxian, Yu, Yunyao, Xu, Hangrui, Yu, Zijian, Tan, Hao, Liu, Zhifang, Jiao, Peng, Lan, Jun, Wang, Haoqian

Abstract

Advances in image generation have made visual authenticity increasingly difficult to assess. Although image forensics now examines both generation artifacts and higher-level visual inconsistencies, a plausible image can still contradict real-world facts or rules. We introduce WorldFact-Bench to evaluate image-world consistency from a single image, without a predefined claim or verification target. The benchmark contains 1,274 source-aligned real-fake pairs across four verification regimes and ten semantic domains. Each pair introduces a specific, evidence-supported factual conflict while seeking to preserve non-target content and visual plausibility. Images are evaluated independently, and pair accuracy requires both members of a pair to be classified correctly. We further propose PERSIST-Agent, which organizes iterative verification around a persistent state linking candidate facts, visual observations, evidence, and verification statuses. This state guides subsequent inspection and retrieval while retaining unresolved candidates. With backbone weights fixed, harness self-optimization refines the agent's prompts and execution rules through validation feedback. Experiments reveal strong label biases in several detectors and uneven gains from retrieval. On the evaluated 8B backbones, PERSIST-Agent improves pair accuracy over both direct judgment and retrieval-augmented baselines, while ablations support the role of persistent verification state. These findings highlight the value of state-guided verification and the remaining gap between visual plausibility and factual correctness.

Chinese Translation

图像生成的进步使得视觉真实性的评估变得愈加困难。尽管图像取证现在同时考察生成伪影和更高层次的视觉不一致性，但一个看似可信的图像仍然可能与现实世界的事实或规则相矛盾。我们引入了WorldFact-Bench，以评估单幅图像的图像与世界的一致性，而无需预定义的声明或验证目标。该基准包含1,274对源对齐的真实-虚假图像，涵盖四种验证机制和十个语义领域。每对图像引入了一个特定的、基于证据的事实冲突，同时力求保持非目标内容和视觉可信度。图像独立评估，配对准确性要求一对中的两个成员都被正确分类。我们进一步提出了PERSIST-Agent，它围绕持久状态组织迭代验证，该状态将候选事实、视觉观察、证据和验证状态联系起来。该状态指导后续的检查和检索，同时保留未解决的候选项。在固定主干权重的情况下，利用自我优化通过验证反馈来完善代理的提示和执行规则。实验揭示了多个检测器中存在强烈的标签偏见以及检索带来的不均匀收益。在评估的8B主干上，PERSIST-Agent在配对准确性上优于直接判断和检索增强基线，同时消融实验支持持久验证状态的作用。这些发现突显了状态引导验证的价值以及视觉可信度与事实正确性之间仍然存在的差距。

View on arXiv Download PDF AI Translation

cs.CV / 47 / 2610.11187

RGBD-to-3D Object Mesh Refinement via Depth Matching and Symmetry Propagation

通过深度匹配和对称传播进行RGBD到3D物体网格的精细化

Seo, Ahyun, Cho, Minsu

Abstract

Single-view 3D reconstructors often produce plausible meshes that disagree with the input view, especially near depth discontinuities and self-occlusions. We present a lightweight, plug-and-play RGBD-to-3D refinement that improves any RGB-to-3D reconstructor without retraining. Given a depth map, we correct the visible surface by bipartite matching to back-projected depth points, mirror these corrections onto the occluded side across a detected symmetry plane, and propagate them with a smoothness solver. Every stage is closed-form, making the method orders of magnitude faster than optimization-heavy test-time refinement. On GSO and OmniObject3D with five backbones, it yields consistent gains, also with monocular pseudo-depth, benefits more from symmetry on symmetric objects, and compares favorably with prior refinement in accuracy and runtime. It further improves an RGB-D-to-mesh reconstructor and transfers to real captures with noisy sensor depth.

Chinese Translation

单视图3D重建器通常会生成与输入视图不一致的合理网格，特别是在深度不连续性和自遮挡附近。我们提出了一种轻量级、即插即用的RGBD到3D精细化方法，能够在不重新训练的情况下改善任何RGB到3D重建器。给定深度图，我们通过二分匹配来纠正可见表面与反投影深度点的偏差，将这些修正镜像到检测到的对称平面另一侧的遮挡区域，并通过平滑求解器进行传播。每个阶段都是封闭形式，使得该方法比重优化的测试时精细化快几个数量级。在GSO和OmniObject3D上，使用五种主干网络，它在一致性上取得了提升，同时也适用于单目伪深度，对称物体上更能受益于对称性，并在准确性和运行时间上与之前的精细化方法相比表现良好。它进一步改善了RGB-D到网格重建器，并能够迁移到具有噪声传感器深度的真实捕获场景中。

View on arXiv Download PDF AI Translation

cs.CV / 48 / 2610.11198

3DTexMOR: 3D Gaussian Multi-Object Removal via Texture-Space Inpainting

3DTexMOR：通过纹理空间修复实现的三维高斯多物体移除

Guang, Kunxin, Zhao, Yonghao, Yang, Jian, Wang, Beibei

Abstract

3D object removal aims to remove target objects from reconstructed scenes and complete the geometry and appearance of occluded regions. Existing NeRF- and 3DGS-based methods typically inpaint 2D images to guide 3D completion. However, complex multi-object layouts limit the surrounding context visible in each view, making 2D inpainting prone to artifacts. Inconsistent completions across views also introduce conflicting supervision and blurry reconstructions. We propose 3D Gaussian Multi-Object Removal via Texture-Space Inpainting (3DTexMOR). Our key idea is to perform inpainting in a unified texture space shared by all views. By combining complementary observations, this space provides richer context for recovering missing regions and promotes cross-view appearance consistency. We aggregate multi-view observations into texture maps, inpaint the missing regions, and reproject the completed maps into camera views to supervise Gaussian scene completion. To avoid the influence of view-dependent highlights and reflections, we decompose appearance and aggregate view-independent intrinsic attributes instead of RGB colors. We further introduce geometrically regularized Gaussian completion to constrain the geometry of the completed regions. Extensive experiments demonstrate visually plausible completions and state-of-the-art multi-object removal performance, improving PSNR by 5.8 dB and reducing LPIPS by at least 22% compared with existing methods.

Chinese Translation

三维物体移除旨在从重建场景中去除目标物体，并完成被遮挡区域的几何形状和外观。现有的基于 NeRF 和 3DGS 的方法通常对 2D 图像进行修复，以指导 3D 完成。然而，复杂的多物体布局限制了每个视图中可见的周围上下文，使得 2D 修复容易产生伪影。不同视图之间的不一致完成也引入了相互矛盾的监督和模糊的重建。我们提出了通过纹理空间修复实现的三维高斯多物体移除（3DTexMOR）。我们的关键思想是在所有视图共享的统一纹理空间中进行修复。通过结合互补的观察，这个空间为恢复缺失区域提供了更丰富的上下文，并促进了跨视图外观的一致性。我们将多视图观察聚合成纹理图，修复缺失区域，并将完成的图重新投影到相机视图中，以监督高斯场景的完成。为了避免视图依赖的高光和反射的影响，我们分解外观并聚合视图无关的内在属性，而不是 RGB 颜色。我们进一步引入几何正则化的高斯完成，以约束完成区域的几何形状。大量实验表明，我们的方法在视觉上实现了合理的完成效果，并在多物体移除性能上达到了最先进的水平，与现有方法相比，PSNR 提高了 5.8 dB，LPIPS 至少减少了 22%。

View on arXiv Download PDF AI Translation

cs.CV / 49 / 2610.11205

Dissecting Representation Structure in Vision Transformers: A Rigorous Architectural Study

剖析视觉变换器中的表征结构：一项严谨的架构研究

Nguyen, Kim-Cuc, Cheung, Ngai-Man

Abstract

Representation structure is crucial for understanding Vision Transformer (ViT) architectures and their generalization behavior. However, prior studies neither isolate nor analyze module-level features nor investigate how their interactions contribute to performance estimation. In this work, we conduct the first rigorous analysis of feature information across diverse architectural scales, empirically uncover the relationship between ViT representation and generalization behavior, and leverage these insights to guide efficient ViT design. Our contributions are fivefold: Across diverse architectural scales, 1) We identify feature collapse at initialization, which leads to redundancy, and propose a reduction scheme to mitigate this issue. 2) We quantify feature information using entropy and the minimum eigenvalue, demonstrating that these metrics serve as reliable indicators for generalization prediction. 3) We show that feature in the token space provides a more faithful representation than those in embedding space. 4) We discover an unexpected finding: features produced by linear submodules within ViT layers are critical for the prediction of generalization performance. 5) Our proposed proxy improves the correlation ranking by 18-48% over prior baselines and can effectively identify ViT architectures that achieve higher accuracy at lower or comparable computational cost.

Chinese Translation

表征结构对于理解视觉变换器（Vision Transformer, ViT）架构及其泛化行为至关重要。然而，先前的研究既未孤立分析模块级特征，也未探讨它们之间的相互作用如何影响性能评估。在本研究中，我们首次对不同架构尺度下的特征信息进行了严谨分析，实证揭示了ViT表征与泛化行为之间的关系，并利用这些洞察指导高效的ViT设计。我们的贡献主要有五个方面：在不同的架构尺度下，1）我们识别出初始化时的特征崩溃现象，这导致了冗余，并提出了一种减少该问题的方案。2）我们通过熵和最小特征值量化特征信息，证明这些指标作为泛化预测的可靠指示器。3）我们展示了在标记空间中的特征提供了比嵌入空间中更真实的表征。4）我们发现了一个意外的结果：ViT层内线性子模块产生的特征对泛化性能的预测至关重要。5）我们提出的代理方法在相关性排名上比先前基线提高了18-48%，并能有效识别在较低或可比计算成本下实现更高准确率的ViT架构。

View on arXiv Download PDF AI Translation

cs.CV / 50 / 2610.11215

GATOR: Generative and Agentic 3D Object Reconstruction From Casual Images

GATOR：从随意图像生成和代理的三维物体重建

Wu, Qirui, Birchfield, Stan, Rabeti, Hesam, Chang, Angel X., Wen, Bowen

Abstract

Reconstructing complete, scene-aligned 3D objects from casual images requires integrating sparse, uncertain observations and inferring surfaces hidden by occlusions. We present GATOR, a generative and agentic framework that recovers textured object assets and their scene-relative pose from one or more images. Our local modality mixer couples patch-aligned RGB, target-mask, and pointmap features before cross-view reasoning, preserving scene context while distinguishing the target from its surroundings. Text-guided semantic conditioning complements these spatial cues with category names and object captions through stage-specific adapters for structure, geometry, and appearance generation. The generated asset initializes a multimodal agent, providing instance-specific geometry and pose for targeted structural and texture refinement through an observation-guided edit-render-review loop. Across synthetic objects, cluttered tabletops, and indoor scenes, GATOR achieves strong geometric and appearance fidelity while recovering scene-relative pose from sparse observations. Time-budget comparisons and scene-level simulation further demonstrate the reconstruction efficiency and simulation readiness. Project page: https://research.nvidia.com/labs/lpr/gator/

Chinese Translation

从随意图像重建完整的、场景对齐的三维物体需要整合稀疏、不确定的观测数据，并推断被遮挡的表面。我们提出了GATOR，一个生成性和代理性的框架，能够从一张或多张图像中恢复纹理化的物体资产及其相对于场景的姿态。我们的局部模态混合器在跨视角推理之前，将补丁对齐的RGB、目标掩膜和点图特征结合在一起，保留场景上下文，同时将目标与其周围环境区分开来。文本引导的语义条件通过针对结构、几何和外观生成的阶段特定适配器，结合类别名称和物体标题，补充了这些空间线索。生成的资产初始化一个多模态代理，提供实例特定的几何和姿态，以便通过观察引导的编辑-渲染-审查循环进行针对性的结构和纹理细化。在合成物体、杂乱的桌面和室内场景中，GATOR在从稀疏观测中恢复场景相对姿态的同时，实现了强大的几何和外观保真度。时间预算比较和场景级模拟进一步证明了重建效率和模拟准备性。项目页面：https://research.nvidia.com/labs/lpr/gator/

View on arXiv Download PDF AI Translation

cs.CV / 51 / 2610.11233

MiniVer-V: Identifying Minimal Sufficient Evidence for Short Video Verification

MiniVer-V：识别短视频验证的最小充分证据

Chen, Leran, Kong, Lingnan, Cai, Zile

Abstract

A core challenge in short-video fact-checking is identifying which evidence is sufficient to support a verification conclusion. Existing approaches either give the verifier all available evidence, introducing noise, or select evidence by topical relevance, which conflates relatedness with sufficiency. We identify evidential sufficiency as the selection criterion: whether a subset of evidence is adequate to support a confident verdict without redundancy. We introduce MiniVer-V, a benchmark of 195 short videos with three-way verdict annotations (supported, refuted, insufficient) and 5,510 multimodal evidence units spanning visual keyframes, speech transcripts, and web-retrieved external sources. We propose a two-layer verification framework that separates claim-video consistency, assessed from internal evidence, from factual verdict determination, which additionally requires external corroboration. On top of it, a sufficiency-driven greedy search assembles evidence until a sufficiency threshold is met and outputs insufficient when the candidate pool is exhausted, rather than forcing a verdict. With Claude Sonnet 4, the method reaches a Macro-F1 of 0.510 using 4.5 evidence units on average (16% of the full evidence set), statistically indistinguishable from the full-evidence baseline (0.518 with 27.7 units), while significantly improving recognition of insufficient cases over the same search without abstention. The efficiency result replicates with GPT-5.5 and holds only partially with an open-weight Qwen2.5-72B verifier. Ablations show that external evidence is indispensable for factual determination, while internal video evidence grounds the verdict in claim-video consistency. These findings suggest that evidence-efficient verification is achievable, and that explicit abstention is needed when evidence is genuinely inadequate.

Chinese Translation

短视频事实核查的核心挑战在于识别哪些证据足以支持验证结论。现有的方法要么提供所有可用证据给验证者，从而引入噪声，要么通过主题相关性选择证据，这混淆了相关性与充分性。我们将证据的充分性确定为选择标准：即证据的子集是否足以支持一个自信的裁决而不产生冗余。我们引入了MiniVer-V，这是一个包含195个短视频的基准数据集，具有三类裁决注释（支持、反驳、不足）以及5,510个多模态证据单元，涵盖视觉关键帧、语音转录和网络检索的外部来源。我们提出了一个两层验证框架，将基于内部证据评估的声明-视频一致性与事实裁决的确定分开，后者还需要外部证实。在此基础上，基于充分性的贪婪搜索组装证据，直到达到充分性阈值，并在候选池耗尽时输出不足，而不是强行得出裁决。使用Claude Sonnet 4，该方法在平均使用4.5个证据单元（占完整证据集的16%）的情况下，达到了0.510的宏观F1值，统计上与完整证据基线（0.518，使用27.7个单元）无显著差异，同时在识别不足案例方面显著改善了相同搜索的结果而没有弃权。该效率结果在GPT-5.5中得到了复制，并在开放权重的Qwen2.5-72B验证器中部分保持。消融实验表明，外部证据对于事实确定是不可或缺的，而内部视频证据则为裁决提供了声明-视频一致性的基础。这些发现表明，证据高效的验证是可以实现的，并且在证据确实不足时需要明确的弃权。

View on arXiv Download PDF AI Translation

cs.CV / 52 / 2610.11237

Breaking the Group Size Barrier: Parameter-Efficient Group Dance Generation with Chain-of-Dancers

打破群体规模限制：基于舞者链的参数高效群体舞蹈生成

Xu, Jing, Chen, Cunjian, Ke, Qiuhong

Abstract

Group dance generation aims to synthesize coordinated multi-dancer choreography from music, with broad applications in animation and interactive content creation. This task requires modeling dense inter-person dependencies to ensure spatial coordination, while naturally preserving individual dancer identities. Existing approaches model all dancers jointly with end-to-end transformers, which tie the architecture to a fixed group size and entangle per-dancer identities across frames. We propose ChainDance, a scalable framework that reformulates group dance generation as a Chain-of-Dancers: a sequential decomposition over per-dancer conditional distributions, allowing a single model to scale across variable group sizes without retraining and naturally preserving per-dancer identity. Built on a frozen single-dancer diffusion backbone, ChainDance introduces two lightweight modules: a Role-Aware Text Encoder (RATE) for per-dancer semantic conditioning, and a Group-Aware Motion Encoder (GAME) that aggregates previously generated dancers via a distance-weighted graph convolutional network, and incorporates a training-free noise optimization procedure at inference time to enforce global spatial coherence. Experiments on AIOZ-GDance demonstrate that ChainDance achieves state-of-the-art motion quality and group coordination while structurally preserving per-dancer identity, with $3$-$4\times$ fewer parameters and requiring $3$-$6\times$ less training time compared to prior approaches.

Chinese Translation

群体舞蹈生成旨在从音乐中合成协调的多舞者编舞，广泛应用于动画和互动内容创作。该任务需要建模密集的人际依赖关系，以确保空间协调，同时自然地保留每位舞者的身份。现有方法通过端到端的变换器联合建模所有舞者，这使得架构固定在特定的群体规模，并在帧间纠缠每位舞者的身份。我们提出了ChainDance，一个可扩展的框架，将群体舞蹈生成重新表述为舞者链（Chain-of-Dancers）：对每位舞者条件分布的顺序分解，使得单一模型能够在不同的群体规模之间扩展而无需重新训练，并自然保留每位舞者的身份。ChainDance基于一个冻结的单舞者扩散骨干网络，引入了两个轻量级模块：一个角色感知文本编码器（Role-Aware Text Encoder, RATE）用于每位舞者的语义条件化，以及一个群体感知运动编码器（Group-Aware Motion Encoder, GAME），通过距离加权图卷积网络聚合先前生成的舞者，并在推理时引入无训练的噪声优化过程，以强制执行全局空间一致性。在AIOZ-GDance上的实验表明，ChainDance在运动质量和群体协调方面达到了最先进的水平，同时结构上保留了每位舞者的身份，相比于先前的方法，参数减少了$3$-$4 imes$，训练时间减少了$3$-$6 imes$。

View on arXiv Download PDF AI Translation

cs.CV / 53 / 2610.11238

Attributing HOW, Not Just WHICH: Counterfactual Response Trajectories for Diffusion Models

归因于如何，而不仅仅是归因于哪个：扩散模型的反事实响应轨迹

Zhang, Haoqian, Yang, Ziyuan, Shao, Zerui, Zhang, Yi

Abstract

Diffusion models have achieved remarkable success in image generation, yet tracing their outputs to individual training examples remains challenging. Existing attribution methods often compress factor-specific effects into scalar responses, making distinct internal changes indistinguishable. This is particularly limiting for diffusion models, where semantic factors emerge through evolving representation dynamics during denoising. We therefore reformulate diffusion data attribution as attributing factor-induced internal response trajectories. In this paper, we propose a novel Concept Attribution method through Dynamic Trajectories(CADT). We argue that attribution should therefore ask not only \emph{which} examples matter, but also \emph{how} their influence unfolds during generation. Specifically, we construct matched counterfactual pairs at identical noisy states to isolate factor-specific representation displacements, and model their directional and magnitude evolution across denoising as dynamic attribution signatures. For each training example and generated query, CADT extracts stage-wise feature vectors and integrates them along the denoising process to form a trajectory descriptor. Applying the same construction across the training set yields a bank of factor-specific trajectory descriptors. The covariance statistics of this bank are then used to construct . CADT uses this covariance-aware positive-semidefinite kernel to calibrate the query and training representations, and compares the calibrated query trajectory with each training trajectory to produce the final training-sample attribution scores. Experiments on multiple public datasets show consistent improvements over existing diffusion attribution baselines across hierarchical, compositional, and style attribution.

Chinese Translation

扩散模型在图像生成方面取得了显著成功，但将其输出追溯到单个训练示例仍然具有挑战性。现有的归因方法通常将特定因素的效应压缩为标量响应，使得不同的内部变化难以区分。这对于扩散模型尤其有限，因为语义因素在去噪过程中通过不断演变的表示动态显现。因此，我们将扩散数据的归因重新表述为归因于因素引起的内部响应轨迹。在本文中，我们提出了一种通过动态轨迹的概念归因方法（Concept Attribution through Dynamic Trajectories，CADT）。我们认为，归因不仅应询问 extit{哪些}示例重要，还应询问 extit{如何}它们的影响在生成过程中展开。具体而言，我们在相同的噪声状态下构建匹配的反事实对，以隔离特定因素的表示位移，并将它们在去噪过程中的方向和幅度演变建模为动态归因特征。对于每个训练示例和生成查询，CADT提取逐阶段特征向量，并在去噪过程中将其整合以形成轨迹描述符。在训练集上应用相同的构建方法会产生一组特定因素的轨迹描述符。然后，利用这组描述符的协方差统计量构建CADT。CADT使用这种考虑协方差的半正定核来校准查询和训练表示，并将校准后的查询轨迹与每个训练轨迹进行比较，以生成最终的训练样本归因分数。在多个公共数据集上的实验表明，在层次、组合和风格归因方面，CADT在现有扩散归因基线之上 consistently 提高了性能。

View on arXiv Download PDF AI Translation

cs.CV / 54 / 2610.11241

TAP3D: Thermal-Assisted 3D Human Point Clouds

TAP3D：热辅助三维人体点云

Zhang, Xie, Li, Chengxiao, Liu, Xuan, Wu, Chenshu

Abstract

Human body point clouds are a versatile representation for AI-enabled human sensing. However, existing methods using LiDAR, radar, and depth cameras suffer from inherent drawbacks in high cost, sparse reconstruction, and privacy concerns, etc. In this paper, we exploit low-cost thermal arrays and present TAP3D, the first system to reconstruct 3D human point clouds from body heat signatures, offering significant advantages in cost, density, human sensitivity, and privacy. To overcome major challenges in depth estimation, thermal interference, and multi-person separation, we propose a novel physics-informed design, which integrates a forward thermal physics model with two distinct modules: multi-primitive estimation for self-supervised joint recovery of depth and other thermal properties, and geometric perspective fusion for suppressing interference and disentangling multiple people. We implement TAP3D using a single commodity thermal array sensor and build a large-scale dataset (160K samples, 8 environments, 11 users) for evaluation. TAP3D achieves remarkable accuracy for dense point cloud generation, enabling downstream tasks like fall detection (91.46%), indoor tracking (21.86 cm MAE), and human mesh recovery (4.87 cm error). By transforming body heat into point clouds for the first time, TAP3D pioneers a new paradigm for privacy-first, fully passive human sensing for many applications. TAP3D is open-sourced at https://github.com/aiot-lab/TAP3D.

Chinese Translation

人体点云是一种用于人工智能驱动的人体感知的多功能表示。然而，现有使用激光雷达（LiDAR）、雷达和深度摄像头的方法存在高成本、稀疏重建和隐私问题等固有缺陷。本文利用低成本的热阵列，提出了TAP3D，这是第一个从人体热信号重建三维人体点云的系统，在成本、密度、人类敏感性和隐私方面具有显著优势。为了克服深度估计、热干扰和多人分离等主要挑战，我们提出了一种新颖的物理信息设计，该设计将前向热物理模型与两个不同的模块相结合：用于深度和其他热特性自监督联合恢复的多原始体估计，以及用于抑制干扰和解开多个人体的几何视角融合。我们使用单个商品热阵列传感器实现了TAP3D，并构建了一个大规模数据集（160K样本，8个环境，11个用户）进行评估。TAP3D在稠密点云生成方面实现了显著的准确性，使得下游任务如跌倒检测（91.46%）、室内跟踪（21.86 cm MAE）和人体网格恢复（4.87 cm误差）成为可能。通过首次将人体热量转化为点云，TAP3D开创了一种以隐私为首、完全被动的人体感知新范式，适用于多种应用。TAP3D已在https://github.com/aiot-lab/TAP3D上开源。

View on arXiv Download PDF AI Translation

cs.CV / 55 / 2610.11251

V-CoLA: Vision Token Compression with Linear Attention

V-CoLA：基于线性注意力的视觉令牌压缩

Jiang, Hao, Mao, Yiru, Bu, Tianpeng, Zhou, Hao, Duan, Hongtao, Jing, Wang, Xu, Bowen, Chen, Xin, Hu, Lulu, Yang, Bin, Tao, Yongliang, Zhang, Minying

Abstract

Vision-language models (VLMs) have demonstrated impressive capabilities but suffer from substantial computational overhead, as vision tokens dominate the input sequence. This motivates vision token compression as a key direction to alleviate the burden. However, with the emergence of hybrid architectures incorporating linear attention (\eg, Qwen3.5), prior methods designed for softmax attention struggle to generalize. Our analysis reveals that both attention- and similarity-based approaches suffer notable performance degradation, underscoring the urgent need for compression methods tailored to this regime. To this end, we propose \textbf{V-CoLA}, an efficient training-free token compression framework specifically designed for linear attention. V-CoLA introduces a novel \textit{uniqueness-aware importance criterion} for identifying critical vision tokens, coupled with an \textit{adaptive token merging strategy} that performs compression. All components are optimized at the implementation level to remain compatible with the chunk-wise parallelism of linear attention, ensuring strong practical value. Extensive experiments across multiple benchmarks demonstrate the superiority of V-CoLA: it achieves 99.5\% of the original performance with only 50.0\% of vision tokens, and over 88.0\% with as few as 12.5\%, while delivering a 1.86$\times$ to 6.15$\times$ prefill speedup.

Chinese Translation

视觉语言模型（VLMs）展示了令人印象深刻的能力，但由于视觉令牌在输入序列中占主导地位，导致了显著的计算开销。这促使视觉令牌压缩成为缓解这一负担的关键方向。然而，随着结合线性注意力的混合架构（例如，Qwen3.5）的出现，之前为softmax注意力设计的方法难以推广。我们的分析表明，基于注意力和相似性的两种方法都面临显著的性能下降，这突显了针对这一领域量身定制的压缩方法的迫切需求。为此，我们提出了 extbf{V-CoLA}，一个专门为线性注意力设计的高效无训练令牌压缩框架。V-CoLA引入了一种新颖的 extit{关注独特性的重要性标准}来识别关键视觉令牌，并结合 extit{自适应令牌合并策略}进行压缩。所有组件在实现层面进行了优化，以保持与线性注意力的分块并行性兼容，确保强大的实际价值。通过在多个基准测试中的广泛实验，V-CoLA的优越性得到了验证：它在仅使用50.0 ext{%}的视觉令牌的情况下实现了99.5 ext{%}的原始性能，而在使用仅12.5 ext{%}的情况下仍能达到超过88.0 ext{%}，同时提供了1.86$ imes$到6.15$ imes$的预填充加速。

View on arXiv Download PDF AI Translation

cs.CV / 56 / 2610.11264

LR-V2X: Loss-Resilient Collaborative Perception under Low-Bandwidth Communication

LR-V2X：低带宽通信下的抗损失协同感知

Yang, Kang, Bu, Tianci, Wang, Peng, Li, Deying, Wang, Yongcai

Abstract

Given the inherent unpredictability of packet loss in vehicular wireless communications, V2X collaborative perception can yield practical benefits only if agents can achieve reliable collaboration under lossy and low-bandwidth communication conditions. Existing dense BEV feature fusion methods depend on redundant BEV feature exchange, which is infeasible in low-bandwidth scenarios, while compact-communication methods aggressively compress messages but can hardly recover the missing feature content after packet loss. In this paper, we present LR-V2X, a loss-resilient, latent-space reconstruction framework that converts corrupted received latents (even under severe 90% packet loss) into a spatial prior and then reconstructs the missing BEV information from this informative prior and using ego context as condition. Notably, the model can be trained under complete communication conditions and can be directly applied to lossy conditions at test time, eliminating the need for training under numerous lossy conditions. Experiments on DAIR-V2X and V2XREAL show that LR-V2X delivers the strongest robustness under severe packet loss and preserves reliable collaboration as communication quality degrades. And it reduces communication overhead by $64\times$ compared to dense BEV feature fusion baselines. Code will be released at https://github.com/sidiangongyuan/LR-V2X.

Chinese Translation

鉴于车辆无线通信中数据包丢失的固有不可预测性，V2X协同感知只有在代理能够在丢包和低带宽通信条件下实现可靠协作时才能产生实际效益。现有的密集BEV特征融合方法依赖于冗余的BEV特征交换，这在低带宽场景中是不可行的，而紧凑通信方法则过度压缩消息，导致在数据包丢失后几乎无法恢复缺失的特征内容。本文提出了LR-V2X，一个抗损失的潜在空间重建框架，该框架将损坏的接收潜在信息（即使在严重的90%数据包丢失情况下）转换为空间先验，然后利用这一信息先验和自我上下文作为条件重建缺失的BEV信息。值得注意的是，该模型可以在完整通信条件下进行训练，并可以直接应用于测试时的丢包条件，消除了在多种丢包条件下训练的需求。在DAIR-V2X和V2XREAL上的实验表明，LR-V2X在严重数据包丢失情况下展现出最强的鲁棒性，并在通信质量下降时保持可靠的协作。同时，与密集BEV特征融合基线相比，它将通信开销减少了64倍。代码将发布在 https://github.com/sidiangongyuan/LR-V2X。

View on arXiv Download PDF AI Translation

cs.CV / 57 / 2610.11286

When Scene Text Hijacks the Scene: Uncovering, Exploiting, and Mitigating Rendered-Text Semantic Leakage in Image Generation Models

当场景文本劫持场景：揭示、利用和缓解图像生成模型中的渲染文本语义泄漏

Li, Feifei, Wang, Runjie, Zhang, Xiaohan, Qian, Zhenxing, Wen, Mi, Zhang, Mi

Abstract

The reliability and accountability of image generative models (IGMs) are essential for building responsible and trustworthy AI systems. Recent IGMs, such as Nano Banana and GPT-Image, now support complex instruction following, realistic image synthesis, and controllable scene-text rendering. As these capabilities expand, safety analysis must also account for new control channels introduced by complex prompts. In this work, we study rendered-text semantic leakage, a largely overlooked phenomenon in open-domain text rendering. Although rendered text is intended to serve as a local visual constraint that should be reproduced verbatim in the generated image, it also carries linguistic semantics that may be interpreted by the model as part of the input instruction. This makes rendered text a potential semantic control channel whose safety implications remain insufficiently understood. We systematically characterize this phenomenon by decoupling the main visual prompt from the rendered text and measuring their individual and compositional effects on generated images. We quantify semantic leakage and rendering fidelity, and further analyze how leakage emerges from intermediate model evidence. We then show that harmful semantics embedded in scene text can persist through LLM-based prompt enhancement pipelines and steer non-text image regions, even when the main visual prompt remains benign. Finally, we propose a preliminary mitigation approach that reduces unsafe semantic transfer from rendered text to non-text regions while preserving the intended text-rendering behavior on FLUX-2-dev. Our findings reveal rendered text as a dual-use carrier of visible data and latent semantics, exposing a text-centric cross-modal attack surface in modern IGMs.

Chinese Translation

图像生成模型（IGMs）的可靠性和问责制对于构建负责任和可信赖的人工智能系统至关重要。最近的IGMs，如Nano Banana和GPT-Image，现已支持复杂的指令跟随、逼真的图像合成和可控的场景文本渲染。随着这些能力的扩展，安全分析也必须考虑复杂提示引入的新控制通道。在本研究中，我们研究了渲染文本语义泄漏，这是一种在开放领域文本渲染中被大多数人忽视的现象。尽管渲染文本旨在作为一种局部视觉约束，应该在生成的图像中逐字再现，但它也携带了语言语义，模型可能将其解释为输入指令的一部分。这使得渲染文本成为一种潜在的语义控制通道，其安全性影响尚未得到充分理解。我们通过将主要视觉提示与渲染文本解耦，并测量它们对生成图像的个体和组合效果，系统地表征了这一现象。我们量化了语义泄漏和渲染保真度，并进一步分析了泄漏是如何从中间模型证据中产生的。然后，我们展示了嵌入场景文本中的有害语义可以通过基于LLM的提示增强管道持续存在，并引导非文本图像区域，即使主要视觉提示保持良性。最后，我们提出了一种初步的缓解方法，减少渲染文本向非文本区域的不安全语义转移，同时在FLUX-2-dev上保留预期的文本渲染行为。我们的研究结果揭示了渲染文本作为可见数据和潜在语义的双重用途载体，暴露了现代IGMs中的以文本为中心的跨模态攻击面。

View on arXiv Download PDF AI Translation

cs.CV / 58 / 2610.11296

Spatial-Frequency-Aware Implicit Neural Representation of Multidimensional Signals via MLP-KAN Fusion

基于空间频率感知的多维信号隐式神经表示通过MLP-KAN融合

Yan, Wen, Shi, Ligen, Qiu, Jun, Zhang, Haimiao, Wu, Lina, Liu, Chang

Abstract

Implicit Neural Representations (INRs) have emerged as a compelling paradigm for modeling multidimensional signals by mapping continuous coordinates to signal values. However, Multi-Layer Perceptrons (MLP)-based INRs inherently suffer from spectral bias, which favors low-frequency components and suppresses the reconstruction of essential high-frequency details. While existing techniques, such as Fourier feature mappings, mitigate this issue, they often rely on sensitive manual tuning and are prone to spectral artifacts. In this paper, we propose a spatial-frequency-aware INR framework that combines an MLP branch with a Kolmogorov-Arnold network (KAN) branch for complementary frequency-oriented modeling. The MLP branch provides a low-frequency-oriented representation of smooth structures, whereas the KAN branch complements localized variations and fine details. To coordinate the two branches, we integrate the discrete wavelet transform (DWT) and inverse discrete wavelet transform (IDWT) into the output fusion stage. The outputs of the two branches are decomposed into wavelet coefficients, and the corresponding coefficients are additively fused before inverse wavelet reconstruction. A wavelet-domain band-separation regularization further penalizes high-frequency responses in the MLP branch and low-frequency responses in the KAN branch, thereby encouraging complementary frequency-oriented behavior. Experiments on 1D signals, 2D images, 3D volumes and signed distance functions, videos, and 4D light-fields demonstrate the applicability of the proposed representation across the evaluated signal modalities. Results demonstrate improved reconstruction fidelity across the evaluated signal modalities.

Chinese Translation

隐式神经表示（INRs）作为一种有效的范式，通过将连续坐标映射到信号值，已成为建模多维信号的重要方法。然而，基于多层感知器（MLP）的INRs固有地存在谱偏差，偏向低频成分，抑制了重要高频细节的重建。虽然现有技术，如傅里叶特征映射，能够缓解这一问题，但它们通常依赖于敏感的手动调节，并且容易产生谱伪影。本文提出了一种空间频率感知的INR框架，该框架将MLP分支与Kolmogorov-Arnold网络（KAN）分支相结合，以实现互补的频率导向建模。MLP分支提供了平滑结构的低频导向表示，而KAN分支则补充了局部变化和细节。为了协调这两个分支，我们在输出融合阶段整合了离散小波变换（DWT）和逆离散小波变换（IDWT）。两个分支的输出被分解为小波系数，并在逆小波重建之前进行相应系数的加性融合。小波域带分离正则化进一步惩罚MLP分支中的高频响应和KAN分支中的低频响应，从而鼓励互补的频率导向行为。在一维信号、二维图像、三维体积和有符号距离函数、视频以及四维光场的实验中，验证了所提表示在评估的信号模态中的适用性。结果表明，所提方法在评估的信号模态中提高了重建保真度。

View on arXiv Download PDF AI Translation

cs.CV / 59 / 2610.11302

iCATS: Fast Video Generation via Interaction-Aware Sparse Attention and Timestep-Adaptive Sparsity

iCATS：通过交互感知稀疏注意力和时间步自适应稀疏性实现快速视频生成

Han, Chengfeng, Ai, Baole, Bian, Xianlu, Yao, Jie, Huang, Zilong, Wang, Ang, Ding, Dandan

Abstract

Training-free sparse attention offers a practical acceleration solution to Diffusion Transformers (DiTs) via reducing computations without fine-tuning. It typically involves estimating the importance of query-key regions and deriving sparse masks to compute only the important candidates, which inevitably introduces approximation errors that may degrade generation quality. To better balance the efficiency-quality trade-off, we propose iCATS, integrating improved importance estimation and sparse mask construction with an efficient hardware execution strategy. Specifically, for importance estimation, unlike previous works that perform independent clustering over query and key tokens based on feature similarity to estimate attention scores, iCATS demonstrates that clustering based on query-key dot-product interactions is more accurate and further reformulates this objective as a simple quadratic form for low-cost computation. For sparse mask construction, instead of using a fixed top-p rule, we observe that tolerance to sparse approximation errors varies across denoising timesteps and therefore introduce an SNR-guided sparsity schedule to adjust sparsity dynamically, leading to higher accuracy. Finally, for hardware execution, we devise a tail-merging strategy to reduce padding overhead caused by irregular cluster sizes, improving GPU kernel utilization. Extensive experiments show that iCATS achieves $2.03\times$ acceleration with 31.017 dB PSNR on HunyuanVideo-T2V-13B and $1.55\times$ acceleration with 29.301 dB PSNR on Wan2.1-T2V-14B, delivering a state-of-the-art efficiency-quality trade-off.

Chinese Translation

无训练的稀疏注意力为扩散变换器（Diffusion Transformers, DiTs）提供了一种实用的加速解决方案，通过减少计算量而无需微调。它通常涉及估计查询-键区域的重要性，并推导稀疏掩码以仅计算重要候选项，这不可避免地引入了可能降低生成质量的近似误差。为了更好地平衡效率与质量的权衡，我们提出了iCATS，结合了改进的重要性估计和稀疏掩码构建，以及高效的硬件执行策略。具体而言，在重要性估计方面，与以往基于特征相似性对查询和键标记进行独立聚类以估计注意力分数的工作不同，iCATS展示了基于查询-键点积交互的聚类更为准确，并进一步将该目标重新表述为简单的二次形式以实现低成本计算。在稀疏掩码构建方面，我们观察到对稀疏近似误差的容忍度在去噪时间步之间有所不同，因此引入了一种基于信噪比（SNR）指导的稀疏性调度，以动态调整稀疏性，从而提高准确性。最后，在硬件执行方面，我们设计了一种尾部合并策略，以减少由不规则聚类大小引起的填充开销，提高GPU内核的利用率。大量实验表明，iCATS在HunyuanVideo-T2V-13B上实现了$2.03 imes$的加速，PSNR为31.017 dB，在Wan2.1-T2V-14B上实现了$1.55 imes$的加速，PSNR为29.301 dB，提供了最先进的效率与质量的权衡。

View on arXiv Download PDF AI Translation

cs.CV / 60 / 2610.11303

Efficient Multi-Granularity Knowledge Transfer for Radiology Report Generation

高效的多粒度知识迁移用于放射学报告生成

Zhong, Xubin, Zhang, Zheyu, Qin, Wenjian, Wen, Ning

Abstract

Radiology report generation can automatically generate clinical descriptions from X-ray images, thereby significantly improving the efficiency of radiologists. This task is challenging because it requires medical knowledge to accurately identify diseases and describe them in a professional manner. However, existing methods often overlook the importance of enhancing medical knowledge in describing pivotal areas, a capability that requires models to effectively extract and aggregate knowledge at multiple levels of granularity. Accordingly, we herein propose a novel and compact Efficient Multi-Granularity Knowledge Transfer (\textbf{EMGKT}) method to address the above issues. First, we encode global knowledge embeddings using a medical vision-language model, which provides contextual medical knowledge. Moreover, we devise a novel Fine-Grained Knowledge Distillation (FGKD) training task which efficiently extract fine-grained knowledge. Specifically, the FGKD training task contains teacher embeddings and student embeddings. Teacher embeddings are encoded using extra priors; while student embeddings are learned from the teacher embeddings through knowledge distillation. During inference, the student embeddings are used to enhance fine-grained knowledge while the teacher embeddings are discarded, resulting in negligible computational costs and no need for extra priors. Finally, we further develop a mixture of disease diagnosis expert classifiers to enhance knowledge extraction. The classifiers are initialized using disease embeddings and are modeled as different experts to address various granularity features. Notably, \textbf{EMGKT} can be efficiently applied to most existing methods. Extensive experiments are conducted on two widely-used public datasets and various baselines, which demonstrates the effectiveness and transferability of \textbf{EMGKT}.

Chinese Translation

放射学报告生成能够自动从X光图像中生成临床描述，从而显著提高放射科医生的工作效率。该任务具有挑战性，因为它需要医学知识来准确识别疾病并以专业的方式描述它们。然而，现有方法往往忽视了在描述关键领域时增强医学知识的重要性，这一能力要求模型能够有效地提取和聚合多层次粒度的知识。因此，我们在此提出了一种新颖且紧凑的高效多粒度知识迁移（Efficient Multi-Granularity Knowledge Transfer，EMGKT）方法来解决上述问题。首先，我们使用医学视觉-语言模型编码全局知识嵌入，以提供上下文医学知识。此外，我们设计了一种新颖的细粒度知识蒸馏（Fine-Grained Knowledge Distillation，FGKD）训练任务，以高效提取细粒度知识。具体而言，FGKD训练任务包含教师嵌入和学生嵌入。教师嵌入使用额外的先验进行编码；而学生嵌入则通过知识蒸馏从教师嵌入中学习。在推理过程中，学生嵌入用于增强细粒度知识，而教师嵌入被丢弃，从而实现了微不足道的计算成本且无需额外的先验。最后，我们进一步开发了一种混合疾病诊断专家分类器，以增强知识提取。这些分类器使用疾病嵌入进行初始化，并被建模为不同的专家，以处理各种粒度特征。值得注意的是，EMGKT可以高效地应用于大多数现有方法。我们在两个广泛使用的公共数据集和各种基线模型上进行了大量实验，证明了EMGKT的有效性和可迁移性。

View on arXiv Download PDF AI Translation

cs.CV / 61 / 2610.11306

Collaboratively Guided Adversarial Robust Distillation with Teacher-Favorable Examples

协同引导的对抗鲁棒蒸馏与教师友好示例

Li, Zhi, Liu, Haowei, Yang, Hongchen, Wang, Xiaoxuan, Gao, Song, Yao, Shaowen, Zhou, Wei

Abstract

Adversarial distillation transfers robustness from high-capacity teachers to compact students. Existing adversarial distillation methods mainly use teacher predictions on clean or adversarial examples to supervise student learning. However, teacher-favorable supervision within the perturbation neighborhood remains underexplored in adversarial distillation. We therefore propose Collaboratively Guided Adversarial Robust Distillation (CGARD), which jointly optimizes distinct student-adversarial and teacher-collaborative examples within the same perturbation neighborhood. The teacher-collaborative example is constrained to incur no greater cross-entropy loss under the teacher than the clean input. CGARD combines collaborative teacher guidance with adversarial teacher supervision to improve robust knowledge transfer. Experiments on CIFAR-10 and CIFAR-100, including white-box evaluation and additional black-box transfer evaluation, demonstrate consistent robustness improvements over strong adversarial distillation baselines.

Chinese Translation

对抗蒸馏将鲁棒性从高容量教师转移到紧凑学生。现有的对抗蒸馏方法主要利用教师在干净或对抗示例上的预测来监督学生学习。然而，在对抗蒸馏中，扰动邻域内的教师友好监督仍然未得到充分探索。因此，我们提出了协同引导的对抗鲁棒蒸馏（Collaboratively Guided Adversarial Robust Distillation, CGARD），该方法在同一扰动邻域内联合优化不同的学生对抗示例和教师协作示例。教师协作示例被约束为在教师下的交叉熵损失不大于干净输入的损失。CGARD结合了协作教师引导与对抗教师监督，以改善鲁棒知识的转移。在CIFAR-10和CIFAR-100上的实验，包括白盒评估和额外的黑盒转移评估，均显示出相较于强对抗蒸馏基线的一致鲁棒性提升。

View on arXiv Download PDF AI Translation

cs.CV / 62 / 2610.11310

FloorSAV: Elucidating Spatial Audio-Visual Context with 2D Floormap for AV-LLMs

FloorSAV：通过二维楼层图阐明音频-视觉上下文以支持音频-视觉大语言模型

Kim, Kyeong-Rae, Kim, Sungnyun, Oh, Tae-Hyun

Abstract

While 3D spatial reasoning in dynamic egocentric environments is crucial for embodied intelligence, audio-visual large language models (AV-LLMs) lack explicit mechanisms to process and internalize global geometry directly from raw sensory streams. Existing approaches either require costly fine-tuning or underutilize the model's cross-modal reasoning capacities. In this paper, we propose FloorSAV, a novel framework that explicitly grounds spatial audio-visual context by rendering a dynamic 2D floormap. By integrating 3D point clouds, camera trajectories, spatial audio cues, and semantically grounded object landmarks, we inject this floormap into the AV-LLM as a synchronized stream with an egocentric video. AV-LLMs utilize their multi-modal capabilities to jointly reason over visual, auditory, and geometric cues in a single inference with floormap interpretation guidance. We further introduce SAVED-Bench (Spatial Audio-Visual Egocentric Benchmark with Dynamic Agents), constructing essential tasks of spatial capability in real-world scenarios: dynamic relativity, regional, and path reasoning QAs. FloorSAV improves AV-LLMs' spatial reasoning on various tasks from both SAVED-Bench and SAVVY-Bench. Studies with ground-truth floormaps demonstrate the substantial potential of FloorSAV with accurate spatial information.

Chinese Translation

在动态自我中心环境中，3D空间推理对具身智能至关重要，但音频-视觉大语言模型（AV-LLMs）缺乏直接从原始感官流中处理和内化全局几何的明确机制。现有方法要么需要昂贵的微调，要么未充分利用模型的跨模态推理能力。本文提出了FloorSAV，一个新颖的框架，通过渲染动态二维楼层图明确地将空间音频-视觉上下文与之关联。通过整合3D点云、相机轨迹、空间音频线索和语义基础的物体地标，我们将此楼层图作为与自我中心视频同步的流注入AV-LLM中。AV-LLMs利用其多模态能力，在单次推理中结合楼层图解释指导，对视觉、听觉和几何线索进行联合推理。我们进一步引入SAVED-Bench（动态代理的空间音频-视觉自我中心基准），构建了现实场景中空间能力的基本任务：动态相对性、区域和路径推理问答。FloorSAV在SAVED-Bench和SAVVY-Bench的各种任务上提高了AV-LLMs的空间推理能力。与真实楼层图的研究表明，FloorSAV在准确空间信息方面具有显著潜力。

View on arXiv Download PDF AI Translation

cs.CV / 63 / 2610.11320

It's Always 10:10: Reference Images Break a Bias That Prompts Only Dent

总是10:10：参考图像打破仅显示凹痕的偏见

Cazzaniga, Luca

Abstract

Text-to-image models appear to reproduce the habits of the photographs they learned from. Analog clocks are an extreme case: in advertising, watches almost always show 10:10, and generated clocks return to 10:10 even when another time is requested. We measure this bias and test three ways of overcoming it on 52 models available on the Magnific platform, with a replication on Higgsfield. Every image shows three identical clocks that must show 2:35, 6:50 and 11:20. The description of the object is fixed and only the request about the time changes: no time (A), the time in digits (B), the hand positions described by construction relative to the dial numerals (C), or the same description plus a drawn reference dial (D). Two AI readers read all 1,799 images blind from coded copies, with a third reader and the author settling disagreements (dial-level agreement 96.0% and 97.3%). With no time requested, 67% of the images have all three clocks at 10:10. On the 20 current models, all three clocks are correct in 34% of the images with digits, 30% with the hands described in words and 75% with the reference dial (D-B: +37 points, 95% CI +28 to +45); we found no evidence that describing the hands in words beats the digits (C-B: -4 points, CI -10 to +1). The replication on the 12 models shared by both platforms gives the same picture (B 54%, C 50%, D 81%). Writing the time reduces the bias but leaves two thirds of the images of the 20 current models with at least one wrong clock; adding a drawn reference raises full accuracy to three quarters and almost eliminates images entirely at 10:10. We release all images, prompts, raw readings and a script that recomputes every result.

Chinese Translation

文本到图像模型似乎再现了它们所学习的照片的习惯。模拟时钟是一个极端的例子：在广告中，手表几乎总是显示10:10，而生成的时钟即使在请求其他时间时也会返回到10:10。我们测量了这种偏见，并在Magnific平台上对52个模型测试了三种克服该偏见的方法，并在Higgsfield上进行了复现。每幅图像展示了三只相同的时钟，必须显示2:35、6:50和11:20。物体的描述是固定的，只有关于时间的请求发生变化：没有时间（A），数字形式的时间（B），相对于表盘数字描述的指针位置（C），或相同描述加上绘制的参考表盘（D）。两位AI读者从编码副本中盲读所有1,799幅图像，第三位读者和作者解决分歧（表盘级一致性为96.0%和97.3%）。在没有请求时间的情况下，67%的图像中三只时钟均显示10:10。在20个当前模型中，所有三只时钟在数字形式的图像中正确率为34%，在用文字描述指针的位置的图像中为30%，在带有参考表盘的图像中为75%（D-B：+37个百分点，95%置信区间+28到+45）；我们没有发现用文字描述指针优于数字的证据（C-B：-4个百分点，置信区间-10到+1）。在两个平台共享的12个模型上的复现结果与之相同（B 54%，C 50%，D 81%）。书写时间减少了偏见，但在20个当前模型中，仍有三分之二的图像至少有一只时钟错误；添加绘制的参考图像将完整准确率提高到四分之三，并几乎消除了完全显示10:10的图像。我们发布了所有图像、提示、原始读数和重新计算每个结果的脚本。

View on arXiv Download PDF AI Translation

cs.CV / 64 / 2610.11329

PathLang: A Language-Centered Benchmark for Vision-Language Models in Computational Pathology

PathLang：计算病理学中以语言为中心的视觉-语言模型基准

Cheng, Fanqi, Gong, Kuo, Liu, Shangke, Zhao, Beidi, Zhu, Junchao, Zhu, Zheyu, Zhao, Leiyue, Liu, Fengbei, Cannon, John, Wang, Gang, Gao, Zu-hua, Ikemura, Kenji, Yang, Yihe, Wang, Yaohong, Huo, Yuankai, Li, Xiaoxiao, Sabuncu, Mert R., Deng, Ruining

Abstract

Pathology vision-language models (VLMs) have shown strong visual perception ability, but their robustness in the language domain remains poorly characterized. Existing pathology VLM benchmarks largely rely on canonical closed-set prompts or perturb only generic templates, treating language as a fixed evaluation component rather than a variable axis of model behavior. In clinical practice, however, diagnostic language varies across reports, institutions, and candidate diagnoses. We introduce PathLang, a language-centered and clinically grounded zero-shot benchmark. PathLang holds the underlying slides, ground-truth labels, and image-text evaluation direction fixed while systematically varying only the diagnostic language, so that performance differences reflect how a diagnosis is phrased rather than what is imaged. The language variation follows how pathologists actually rephrase diagnoses (terminology, specificity, and reporting style), and all prompts and candidate pools are validated by six board-certified pathologists. PathLang covers four task families: (1) zero-shot classification with image-text alignment analysis, (2) cross-modal retrieval, (3) paraphrase robustness, including semantic-equivalence paraphrases, length and reporting-style variation, and prompt ensembling, and (4) open-vocabulary diagnosis retrieval over four candidate pools with distinct forms of semantic competition. Across nine VLMs and five public datasets spanning four organs, we find that performance is highly sensitive to clinically equivalent paraphrases, varies substantially across forms of semantic competition, and that image-text alignment quality does not necessarily translate into inter-class separability. We release the prompt corpus, candidate pools, pre-computed text embeddings, and evaluation code at https://anonymous.4open.science/r/PathLang.

Chinese Translation

病理视觉-语言模型（VLMs）展现了强大的视觉感知能力，但其在语言领域的鲁棒性仍然缺乏充分的表征。现有的病理VLM基准主要依赖于规范的封闭集提示或仅对通用模板进行扰动，将语言视为固定的评估组件，而非模型行为的可变轴。然而，在临床实践中，诊断语言在报告、机构和候选诊断之间存在差异。我们引入了PathLang，这是一个以语言为中心且基于临床的零-shot基准。PathLang在保持基础幻灯片、真实标签和图像-文本评估方向固定的同时，仅系统性地改变诊断语言，以便性能差异反映诊断的措辞方式，而非所成像的内容。语言变异遵循病理学家实际如何重新表述诊断（术语、特异性和报告风格），所有提示和候选池均由六位获得认证的病理学家验证。PathLang涵盖四个任务类别：（1）带有图像-文本对齐分析的零-shot分类，（2）跨模态检索，（3）释义鲁棒性，包括语义等价释义、长度和报告风格变异，以及提示集成，以及（4）在四个具有不同语义竞争形式的候选池中进行开放词汇诊断检索。在涵盖九个VLM和五个涵盖四个器官的公共数据集的研究中，我们发现性能对临床等价释义高度敏感，在不同形式的语义竞争中变化显著，并且图像-文本对齐质量并不一定转化为类间可分性。我们在https://anonymous.4open.science/r/PathLang发布了提示语料库、候选池、预计算文本嵌入和评估代码。

View on arXiv Download PDF AI Translation

cs.CV / 65 / 2610.11342

Point-Focused Attention Meets Context-Scan State Space: Robust Biological Visual Perception for Point Cloud Representation

点聚焦注意力与上下文扫描状态空间的结合：用于点云表示的鲁棒生物视觉感知

Qu, Kanglin, Gao, Pan, Dai, Qun, Sun, Yuanhao

Abstract

Synergistically capturing intricate local structures and global contextual dependencies has become a critical challenge in point cloud representation learning. To address this, we introduce PointLearner, a point cloud representation learning network that closely aligns with biological vision which employs an active, foveation-inspired processing strategy, thus enabling local geometric modeling and long-range dependency interactions simultaneously. Specifically, we first design a point-focused attention, which simulates foveal vision at the visual focus through a competitive normalized attention mechanism between local neighbors and spatially downsampled features. The spatially downsampled features are extracted by a pooling method based on learnable inducing points, which can flexibly adapt to the non-uniform distribution of point clouds as the number of inducing points is controlled and they interact directly with point clouds. Second, we propose a context-scan state space that mimics eye's saccade inference, which infers the overall semantic structure and spatial content in the scene through a scan path guided by the Hilbert curve for the bidirectional S6. With this focus-then-context biomimetic design, PointLearner demonstrates remarkable robustness and achieves state-of-the-art performance across multiple point cloud tasks.

Chinese Translation

在点云表示学习中，协同捕捉复杂的局部结构和全局上下文依赖性已成为一项关键挑战。为了解决这个问题，我们提出了PointLearner，一个与生物视觉紧密对齐的点云表示学习网络，该网络采用了一种主动的、受中心注视启发的处理策略，从而能够同时实现局部几何建模和长距离依赖交互。具体而言，我们首先设计了一种点聚焦注意力，通过局部邻域与空间下采样特征之间的竞争归一化注意机制，模拟了视觉焦点处的中心视觉。空间下采样特征是通过基于可学习诱导点的池化方法提取的，该方法可以灵活适应点云的非均匀分布，因为诱导点的数量是可控的，并且它们与点云直接交互。其次，我们提出了一种上下文扫描状态空间，模拟眼睛的扫视推理，通过由Hilbert曲线引导的扫描路径推断场景中的整体语义结构和空间内容，以实现双向S6。通过这种聚焦-再上下文的仿生设计，PointLearner展现出卓越的鲁棒性，并在多个点云任务中达到了最先进的性能。

View on arXiv Download PDF AI Translation

cs.CV / 66 / 2610.11347

EgoPhys: Estimating Peak Contact Force and Mechanical Work from Egocentric Manipulation Video

EgoPhys：从自我中心操作视频中估计峰值接触力和机械功

Dong, Zhuo, Yang, Jianhua, Li, Haohao, Zhao, Yumeng, He, Keji, Huang, Yan, Wang, Liang

Abstract

Physically grounded manipulation of articulated objects requires understanding both the maximum forces encountered during contact and the work performed as their parts move. Peak contact force and mechanical work quantify these complementary aspects, but estimating them from egocentric video is challenging because physical interaction cues are local and indirect. Moreover, peak force is associated with brief contact events, whereas mechanical work depends on force-motion coupling throughout the contact duration. To address these challenges, we propose EgoPhys, an RGB-only framework comprising Contact-Aware Spatial Aggregation (CASA) and Target-Specific Multi-Expert Temporal Routing (TMTR). CASA integrates appearance and geometry features to emphasize interaction-relevant cues, while TMTR models semantic, event, and motion cues with specialized temporal experts and routes them separately for force and work prediction. On the test split from Hoi! dataset, EgoPhys substantially improves predictions of peak force and mechanical work, achieving MAEs of $5.205 \pm 0.584$ $N$ and $0.894 \pm 0.081$ $J$, respectively.

Chinese Translation

对关节物体进行物理基础的操作需要理解接触过程中遇到的最大力和部件运动所做的功。峰值接触力和机械功量化了这两个互补方面，但从自我中心视频中估计它们是具有挑战性的，因为物理交互线索是局部的和间接的。此外，峰值力与短暂的接触事件相关，而机械功则依赖于接触持续时间内的力-运动耦合。为了解决这些挑战，我们提出了EgoPhys，一个仅使用RGB的框架，包含接触感知空间聚合（Contact-Aware Spatial Aggregation, CASA）和目标特定多专家时间路由（Target-Specific Multi-Expert Temporal Routing, TMTR）。CASA整合外观和几何特征，以强调与交互相关的线索，而TMTR则通过专门的时间专家建模语义、事件和运动线索，并分别为力和功的预测进行路由。在Hoi!数据集的测试集上，EgoPhys显著改善了峰值力和机械功的预测，分别达到了MAE为5.205 ± 0.584 $N$和0.894 ± 0.081 $J$。

View on arXiv Download PDF AI Translation

cs.CV / 67 / 2610.11356

Adaptive Adversarial Augmentation for Controllable Face Synthesis

可控面部合成的自适应对抗增强

Suri, Saransh, Agarwal, Shivang, Vatsa, Mayank, Singh, Richa

Abstract

Synthetic data provides a scalable alternative to real-world datasets for training face recognition models, particularly under challenging conditions such as low resolution, occlusion, and masks. Yet, most approaches lack diversity and fail to generalize effectively. We propose Ensemble Feedback Controllable Synthesis (EFCS), a guided framework that generates diverse and challenging samples while preserving visual realism. EFCS expands distributional variability, often reflected in higher FID and KID scores compared to single-feedback and random synthesis, while maintaining high precision. Recognition models trained on EFCS data consistently outperform baselines across multiple benchmarks, showing improved generalization to real-world scenarios. Furthermore, we introduce an analytically motivated formulation linking perturbation-induced difficulty, sample utility, and performance degradation, offering principled insights into balancing synthetic data complexity for optimal training. Together, these contributions establish EFCS as an effective and analytically grounded approach for bridging the gap between synthetic and real datasets.

Chinese Translation

合成数据为训练面部识别模型提供了一种可扩展的替代方案，尤其是在低分辨率、遮挡和口罩等挑战性条件下。然而，大多数方法缺乏多样性，未能有效泛化。我们提出了集成反馈可控合成（Ensemble Feedback Controllable Synthesis, EFCS），这是一个引导框架，能够生成多样且具有挑战性的样本，同时保持视觉真实感。EFCS 扩展了分布变异性，通常在较高的 FID 和 KID 分数中体现，相较于单反馈和随机合成，仍能保持高精度。基于 EFCS 数据训练的识别模型在多个基准测试中始终优于基线，显示出对真实场景的改善泛化。此外，我们引入了一种分析性驱动的公式，连接扰动引起的难度、样本效用和性能下降，为平衡合成数据复杂性以实现最佳训练提供了原则性的见解。这些贡献共同确立了 EFCS 作为一种有效且具有分析基础的方法，以弥合合成数据集与真实数据集之间的差距。

View on arXiv Download PDF AI Translation

cs.CV / 68 / 2610.11364

FlyMark: Training-Free Invisible Watermarking of 3D Gaussian Splatting via a Fruit Fly Connectome

FlyMark：基于果蝇连接组的无训练隐形水印3D高斯点云技术

Luo, Ziyuan, Li, Haoliang, Wan, Renjie

Abstract

A trained 3D Gaussian Splatting (3DGS) scene ships as a portable parameter array that can be copied, pruned, requantized, or repackaged outside its training pipeline, so ownership evidence is most useful when it lives in the released parameters and remains checkable long after the embedding tooling is gone. Existing 3DGS watermarks typically tie embedding or extraction to scene optimization, a learned decoder, or rendered views, so the evidence survives only as long as a second trained artifact does. FlyMark instead writes a keyed message into the parameters a 3DGS file already stores. Its carrier directions are derived from the photoreceptors of a published connectome, a citable versioned artifact that fixes the geometry exhaustively and leaves nothing to tune per scene. A virtual observer reads cone-wise apparent luminance along a scene-normalized orbit from stored centers, colors, and opacities; a keyed dithered quantization-index-modulation code replicates each message bit across these observations; and one sparse bounded least-squares solve realizes the targets through achromatic shifts of existing degree-zero colors under a hard per-channel linear-RGB bound. All geometry and higher-order appearance parameters are preserved bit-identically, and extraction needs only cone queries, rounding, and majority voting. Under a model-domain threat model on synthetic and real scenes, FlyMark attains high clean bit accuracy and visual fidelity while cleanly separating matched from wrong keys.

Chinese Translation

经过训练的3D高斯点云（3DGS）场景以可移植的参数数组形式存在，可以在其训练流程之外被复制、修剪、重新量化或重新打包，因此，当所有嵌入工具消失后，拥有权证据在发布的参数中存在并保持可检查时最为有用。现有的3DGS水印通常将嵌入或提取与场景优化、学习解码器或渲染视图绑定，因此证据的存续仅与第二个训练的工件的存在时间相关。FlyMark则将一个带密钥的信息写入3DGS文件已存储的参数中。其载体方向源自已发布的连接组的光感受器，这是一种可引用的版本化工件，全面固定几何形状，并且不留任何场景调节的余地。一个虚拟观察者沿着场景标准化轨道，从存储的中心、颜色和不透明度中读取锥体方向的表观亮度；一个带密钥的抖动量化索引调制代码在这些观察中复制每个信息位；而一个稀疏有界最小二乘解通过在硬的每通道线性RGB界限下的无色偏移实现目标。所有几何形状和高阶外观参数都以位相同的方式保留，提取只需锥体查询、四舍五入和多数投票。在合成和真实场景的模型域威胁模型下，FlyMark在高干净位准确性和视觉保真度方面表现出色，同时能够干净地将匹配的密钥与错误的密钥分离。

View on arXiv Download PDF AI Translation

cs.CV / 69 / 2610.11374

Rethinking Contrastive Loss in CLIP Post-training: A Complementary Framework with Frozen Text Encoder

重新思考CLIP后训练中的对比损失：一个冻结文本编码器的补充框架

Wang, Zidan, Li, Yaqian, Zhang, Xiaokai, Long, Kaiwen, He, Kun, Liu, Hanpeng

Abstract

CLIP serves as a foundational vision-language model and the de facto vision encoder for downstream VLMs such as LLaVA. Post-training offers a lightweight route to refine CLIP, but recent work argues that the standard contrastive loss is unsuitable for post-training due to catastrophic forgetting under small batches, motivating designs that abandon the contrastive objective in favor of distillation. We revisit this premise and find that, for the InfoNCE objective, the reported forgetting is driven primarily not by insufficient negatives but by an inappropriate magnitude of the contrastive temperature $\tau$: with $\tau$ set sufficiently small, contrastive post-training improves rather than degrades the pretrained CLIP, which we explain through the temperature dependence of the InfoNCE gradient. Building on this finding, we propose \textbf{ComCLIP}, a lightweight single-epoch post-training recipe that freezes CLIP's text encoder---so the refined vision encoder is a drop-in replacement with unchanged architecture and inference cost---and trains the vision encoder with a properly-tempered contrastive loss, an MSE anchoring loss against the original CLIP, and a relational distillation loss from DINOv2. Over multiple seeds, ComCLIP matches the self-distillation baseline CLIP-Refine on zero-shot classification while significantly improving the transferability of visual features, measured by linear probing ($48.99$ vs.\ $42.28$ on ViT-B/16), and on ViT-L/14 it also improves MMVP over CLIP-Refine ($24.20$ vs.\ $19.01$); CLIP-Refine remains stronger on image-text retrieval. Used as a drop-in vision encoder for LLaVA-1.5-7B without re-aligning the projector or LLM, ComCLIP yields no net change across $8$ VLM benchmarks, i.e., the refinement does not break downstream compatibility. Code and models are available at https://github.com/showstarpro/ComCLIP.git.

Chinese Translation

CLIP作为一个基础的视觉-语言模型，是下游视觉语言模型（VLMs）如LLaVA的事实上的视觉编码器。后训练提供了一种轻量化的方式来优化CLIP，但最近的研究认为，标准的对比损失不适合后训练，因为在小批量情况下会出现灾难性遗忘，这促使设计放弃对比目标而转向蒸馏。我们重新审视这一前提，发现对于InfoNCE目标，报告的遗忘主要不是由于负样本不足，而是由于对比温度$ au$的不当幅度：当$ au$设置得足够小，进行对比后训练反而改善了预训练的CLIP，我们通过InfoNCE梯度的温度依赖性来解释这一现象。在此基础上，我们提出了 extbf{ComCLIP}，一种轻量级的单轮后训练方案，该方案冻结CLIP的文本编码器——因此优化后的视觉编码器是一个架构和推理成本不变的直接替代品——并使用适当调温的对比损失、针对原始CLIP的均方误差（MSE）锚定损失以及来自DINOv2的关系蒸馏损失来训练视觉编码器。在多个随机种子下，ComCLIP在零-shot分类上与自蒸馏基线CLIP-Refine相匹配，同时显著提高了视觉特征的可迁移性，通过线性探测测量（在ViT-B/16上为$48.99$对$42.28$），在ViT-L/14上，ComCLIP也在CLIP-Refine的MMVP上有所提升（$24.20$对$19.01$）；而在图像-文本检索任务上，CLIP-Refine仍然表现更强。作为LLaVA-1.5-7B的直接视觉编码器使用，无需重新对齐投影器或LLM，ComCLIP在$8$个VLM基准测试中没有产生净变化，即优化不会破坏下游兼容性。代码和模型可在https://github.com/showstarpro/ComCLIP.git获取。

View on arXiv Download PDF AI Translation

cs.CV / 70 / 2610.11376

CRISP: Fixing Flying Pixels in Latent LiDAR Generation via Diffusion Decoding

CRISP：通过扩散解码修复潜在LiDAR生成中的飞行像素

Ceron, Andrea, Schmidt, Michael, Marcos-Ramiro, Alvaro, Schmidt, Sebastian, Busam, Benjamin

Abstract

Latent LiDAR pipelines suffer from flying pixels: convolutional VAEs blur sharp radial depth discontinuities, yielding edge depths that back-project to points floating between surfaces. We identify this as a major, directly correctable decoder bottleneck and introduce CRISP: a pixel-space diffusion decoder with a backbone-agnostic latent adapter, DiT-based denoiser, and support mask predictor. CRISP replaces video-VAE and LiDAR-native decoders alike while keeping the encoder and latent generator fixed. Across KITTI-360, SemanticKITTI, and nuScenes, replacing only the decoder reduces FSVD/FPVD by 50.5% on average across frozen backbones; for generic video VAEs, the reductions reach 71%/74%. On the LiDAR-native LiDM backbone, FRID drops by 71%, with the largest gains at depth discontinuities. In a pretrained LiDM world model, the same zero-shot replacement improves FSVD by 15.5%, narrowing the sim-to-real gap.

Chinese Translation

潜在LiDAR管道存在飞行像素的问题：卷积变分自编码器（VAEs）模糊了锐利的径向深度不连续性，导致边缘深度回投到表面之间漂浮的点。我们将此识别为一个主要且可直接纠正的解码器瓶颈，并引入CRISP：一种具有骨干网络无关的潜在适配器、基于DiT的去噪器和支持掩码预测器的像素空间扩散解码器。CRISP替代了视频变分自编码器和LiDAR原生解码器，同时保持编码器和潜在生成器不变。在KITTI-360、SemanticKITTI和nuScenes数据集上，仅替换解码器平均减少了50.5%的FSVD/FPVD；对于通用视频变分自编码器，减少幅度达到71%/74%。在LiDAR原生的LiDM骨干网络上，FRID下降了71%，在深度不连续性处获得了最大的提升。在一个预训练的LiDM世界模型中，相同的零-shot替换使FSVD提高了15.5%，缩小了模拟与现实之间的差距。

View on arXiv Download PDF AI Translation

cs.CV / 71 / 2610.11379

FastJEV: Understanding Redundancy for Compact JEV Inference

FastJEV：理解紧凑 JEV 推理中的冗余性

Ma, Jie, Gao, Jie, Liu, Yihang, Qiu, Zhike, Li, Junle, Zhuang, Chongyi, Ji, Jiayi, Sun, Xiaoshuai

Abstract

JEV models make multimodal decisions by directly scoring candidates. Although the common context is encoded once, candidate evaluation can still repeat matching token histories, duplicate inference states, and execute the full backbone. In this paper, we study these sources of redundancy and present FastJEV for compact candidate evaluation. We jointly organize history reuse and state storage, since sharing computation requires preserving states for later branches. We first introduce shared context anchoring to reuse recurrent initial states and omit unused final recurrent caches. We extend this reuse through candidate prefix sharing, retaining the intermediate states needed by subsequent branches. To further reduce the depth of these paths, we apply decision guided pruning based on relative score changes measured on a small unlabeled set. Our method retains full context encoding and all candidates without additional training. We evaluate FastJEV across three OmniJev model sizes on five public benchmarks and reconstructed LIBERO-10 offline questions. At the selected pruning budgets, the complete method reduces candidate depth by 43.75% to 45.83%, while retaining 93.66% to 97.52% of the original task scores on average across the six evaluation sets. Through controlled experiments, we show how candidate overlap and branching structure affect the execution cost of history reuse. In our implementation, candidate prefix sharing can reduce repeated computation while increasing latency. These findings motivate designing sharing granularity and execution schedules together for efficient JEV inference.

Chinese Translation

JEV 模型通过直接对候选项进行评分来做出多模态决策。尽管常见的上下文只编码一次，候选项评估仍可能重复匹配的标记历史、重复的推理状态，并执行完整的主干网络。本文研究了这些冗余来源，并提出了用于紧凑候选项评估的 FastJEV。我们共同组织历史重用和状态存储，因为共享计算需要为后续分支保留状态。我们首先引入共享上下文锚定，以重用递归初始状态并省略未使用的最终递归缓存。我们通过候选项前缀共享扩展这种重用，保留后续分支所需的中间状态。为了进一步减少这些路径的深度，我们基于在小型未标记集上测量的相对得分变化应用决策引导的剪枝。我们的方法在不进行额外训练的情况下保留完整的上下文编码和所有候选项。我们在五个公共基准和重建的 LIBERO-10 离线问题上评估了三个 OmniJev 模型规模的 FastJEV。在选定的剪枝预算下，完整方法将候选项深度减少了 43.75% 至 45.83%，同时在六个评估集上平均保留了 93.66% 至 97.52% 的原始任务得分。通过控制实验，我们展示了候选项重叠和分支结构如何影响历史重用的执行成本。在我们的实现中，候选项前缀共享可以减少重复计算，同时增加延迟。这些发现促使我们共同设计共享粒度和执行调度，以实现高效的 JEV 推理。

View on arXiv Download PDF AI Translation

cs.CV / 72 / 2610.11381

EvoKnow: Continual Knowledge Evolution for AI-Generated Image Detection

EvoKnow：用于AI生成图像检测的持续知识演化

Peng, Zhiheng, Jin, Wenwei, Ge, Yangshi, Xia, Siyu, Li, Jiawei, Tang, Xu

Abstract

AI-generated image detectors are commonly trained on fixed generator domains and become difficult to maintain as new generative models emerge. Continual adaptation is challenging because replaying historical generated images is costly, whereas updating shared parameters with limited current-domain data can overwrite prior forensic knowledge. We propose EvoKnow, a replay-free framework that formulates continual AI-generated image detection as forensic knowledge evolution. EvoKnow preserves a shared forensic basis learned from base domains, incrementally adds isolated residual experts for complementary generator-relevant evidence, and retrieves expertise through an Analytical Incremental Router (AIR) updated in closed form from current-stage generated images and accumulated sufficient statistics. Experiments demonstrate effective cross-generator generalization, few-shot expansion, and long-horizon continual adaptation. With ten generated images per arriving generator, EvoKnow achieves 96.70% average accuracy on non-base GenImage generators and 94.48% accuracy on Chameleon without target-benchmark adaptation. Under a strict replay-free continual learning protocol, EvoKnow achieves state-of-the-art continual learning performance, attaining 96.32% mean stage-wise accuracy and 4.32% average forgetting.

Chinese Translation

AI生成图像检测器通常在固定的生成器领域上进行训练，随着新生成模型的出现，维护变得困难。持续适应具有挑战性，因为重放历史生成的图像成本高昂，而使用有限的当前领域数据更新共享参数可能会覆盖先前的取证知识。我们提出了EvoKnow，一个无重放框架，将AI生成图像检测形式化为取证知识的演化。EvoKnow保留了从基础领域学习的共享取证基础，逐步添加孤立的残差专家以提供补充的生成器相关证据，并通过一个分析增量路由器（Analytical Incremental Router, AIR）从当前阶段生成的图像和累积的充分统计数据中以封闭形式更新来检索专业知识。实验表明，EvoKnow在跨生成器泛化、少样本扩展和长期持续适应方面表现有效。在每个到达的生成器上生成十张图像的情况下，EvoKnow在非基础GenImage生成器上实现了96.70%的平均准确率，在Chameleon上实现了94.48%的准确率，且无需目标基准适应。在严格的无重放持续学习协议下，EvoKnow实现了最先进的持续学习性能，达到了96.32%的平均阶段准确率和4.32%的平均遗忘率。

View on arXiv Download PDF AI Translation

cs.CV / 73 / 2610.11402

GroundSight at GroundLM 2026 Shared Tasks: GoldenViewVQA

GroundSight 在 GroundLM 2026 共享任务中的表现：GoldenViewVQA

Wang, Kun, Hu, Yupeng, Cao, Ruping, Liu, Hao, Li, Zhiran, Xiang, Qianlong, Cheng, Harry

Abstract

GoldenViewVQA requires models to jointly answer driving-scene questions and identify the camera view containing the supporting visual evidence, making precise evidence localization as important as answer correctness. We present \textbf{CoVeR-VQA}, a training-free multi-stage verification and correction framework for grounded multi-view VQA. Starting from GPT-5.6 zero-shot predictions, CoVeR-VQA progressively applies view-specific verification with Gemini-3.6-Flash, prior-guided joint verification with Claude-Opus-5, and cross-split group-level verification that exploits semantically filtered question groups from shared multi-view scenes and validation-derived prior knowledge. On the official GoldenViewVQA test set, the four-stage CoVeR-VQA pipeline achieves 84.75\% Joint Accuracy, improving the GPT-5.6 zero-shot baseline by 13.56 percentage points, while reaching 94.92\% Answer Accuracy and 86.44\% View Accuracy. The final submitted run achieves 88.14\% Joint Accuracy after two additional evaluator-informed post-hoc corrections. Our analysis shows that supporting-view localization remains the primary source of residual errors, highlighting the importance of explicit evidence verification for reliable multi-view multimodal reasoning.

Chinese Translation

GoldenViewVQA 要求模型共同回答驾驶场景问题，并识别包含支持视觉证据的相机视角，使得精确的证据定位与答案的正确性同样重要。我们提出了 extbf{CoVeR-VQA}，这是一种无训练的多阶段验证和修正框架，旨在处理基于多视角的视觉问答（VQA）。从 GPT-5.6 的零样本预测开始，CoVeR-VQA 逐步应用视角特定的验证，使用 Gemini-3.6-Flash 进行验证，使用 Claude-Opus-5 进行先验引导的联合验证，以及利用来自共享多视角场景的语义过滤问题组和验证衍生的先验知识进行交叉拆分组级验证。在官方的 GoldenViewVQA 测试集上，四阶段的 CoVeR-VQA 流程实现了 84.75\% 的联合准确率，比 GPT-5.6 的零样本基线提高了 13.56 个百分点，同时达到了 94.92\\% 的答案准确率和 86.44\\% 的视角准确率。最终提交的结果在经过两次额外的评估者指导的事后修正后，实现了 88.14\\% 的联合准确率。我们的分析表明，支持视角的定位仍然是残余错误的主要来源，强调了显式证据验证在可靠的多视角多模态推理中的重要性。

View on arXiv Download PDF AI Translation

cs.CV / 74 / 2610.11412

Fresco++: Frequency-Guided and Canonical-Consistent Optimization for Fine-Grained Head Avatar Modeling

Fresco++：基于频率引导和规范一致性的细粒度头部头像建模优化

Zhang, Shikun, Li, Yong, Wang, Yiqun, Ke, Qiuhong, Chen, Cunjian

Abstract

We propose Fresco++, a unified optimization framework for fine-grained and view-consistent head avatar reconstruction. Head avatar optimization is typically driven by per-view image supervision, which can lead to premature fitting of unstable high-frequency details and inconsistent local appearance across viewpoints. Fresco++ addresses these challenges by regulating both the progression of visual detail and the formation of cross-view supervision during optimization. For frequency-aware optimization, a progressive curriculum first stabilizes low-frequency structures and then introduces high-frequency constraints to recover fine facial and hair details without amplifying spurious responses at early stages. For cross-view optimization, we introduce Canonical Group Consensus, which associates local observations through shared canonical surface regions and establishes correspondence across different viewpoints. Geometric and visibility-aware screening removes unreliable observations, while the remaining multi-view evidence is aggregated in feature space to form a consensus target for supervising the current rendering. This design enforces local consistency without relying on a specific image-space parameterization and avoids additional rendering of the auxiliary view. Together, the frequency curriculum and canonical consensus provide stable optimization from coarse structures to fine details while maintaining coherent appearance across viewpoints. Extensive experiments on NeRSemble demonstrate improved reconstruction quality and cross-view consistency, while evaluations across diverse avatar representations further confirm the generality and transferability of Fresco++.

Chinese Translation

我们提出了Fresco++，一个用于细粒度和视图一致的头部头像重建的统一优化框架。头部头像优化通常受到每视图图像监督的驱动，这可能导致不稳定的高频细节的过早拟合，以及不同视点之间局部外观的不一致。Fresco++通过在优化过程中调节视觉细节的进展和跨视图监督的形成来解决这些挑战。对于频率感知优化，渐进式课程首先稳定低频结构，然后引入高频约束，以恢复细致的面部和头发细节，而不在早期阶段放大虚假响应。对于跨视图优化，我们引入了规范组共识（Canonical Group Consensus），通过共享的规范表面区域关联局部观测，并在不同视点之间建立对应关系。几何和可见性感知筛选去除不可靠的观测，而剩余的多视图证据在特征空间中聚合，以形成一个共识目标来监督当前的渲染。这一设计在不依赖特定图像空间参数化的情况下强制执行局部一致性，并避免了对辅助视图的额外渲染。频率课程和规范共识共同提供了从粗糙结构到细节的稳定优化，同时保持不同视点之间的一致外观。在NeRSemble上的大量实验表明，重建质量和跨视图一致性得到了改善，而在不同头像表示上的评估进一步确认了Fresco++的普遍性和可迁移性。

View on arXiv Download PDF AI Translation

cs.CV / 75 / 2610.11419

Missing Modality-Aware Calibration for Trustworthy Brain Tumor Segmentation

缺失模态感知校准以实现可信的脑肿瘤分割

Lee, Sol, Kim, Hyunji, Hong, Sungrae, Han, Donghee, Yi, Mun

Abstract

Multimodal brain tumor segmentation typically leverages multiple MRI modalities, yet incomplete modality acquisition is common in clinical practice due to protocol heterogeneity and scan failures. Although recent methods maintain segmentation accuracy under missing modality conditions, they frequently overlook prediction reliability, leading to miscalibrated confidence estimates that hinder clinical adoption. Existing calibration techniques are largely modality-agnostic or assume that prediction difficulty decreases monotonically as additional modalities become available. However, in brain tumor segmentation, prediction difficulty depends primarily on which modalities are absent rather than how many, leading to combination-specific and spatially heterogeneous calibration errors. To address this, we propose Missing Modality-Aware Local Temperature Scaling (MMA-LTS), a post-hoc voxel-wise confidence calibration method. It estimates a spatially adaptive temperature field conditioned on a modality-availability learnable token and a voxel-wise difficulty score. Experiments on BraTS 2020 and FeTS 2024 show that MMA-LTS improves calibration while preserving the segmentation accuracy of state-of-the-art models across diverse missing-modality scenarios, thereby enhancing trustworthiness toward clinical deployment.

Chinese Translation

多模态脑肿瘤分割通常利用多种MRI模态，但由于协议异质性和扫描失败，临床实践中常常出现模态获取不完整的情况。尽管最近的方法在缺失模态条件下保持了分割精度，但它们往往忽视了预测的可靠性，导致信心估计的校准不准确，从而阻碍了临床应用。现有的校准技术在很大程度上是模态无关的，或者假设随着额外模态的可用性，预测难度单调下降。然而，在脑肿瘤分割中，预测难度主要取决于缺失的模态，而不是缺失的模态数量，导致特定组合和空间异质的校准误差。为了解决这个问题，我们提出了缺失模态感知局部温度缩放（Missing Modality-Aware Local Temperature Scaling，MMA-LTS），这是一种后处理的体素级信心校准方法。它根据模态可用性可学习的标记和体素级难度评分，估计空间自适应的温度场。在BraTS 2020和FeTS 2024上的实验表明，MMA-LTS在保持最先进模型的分割精度的同时，改善了校准效果，增强了临床应用的可信度。

View on arXiv Download PDF AI Translation

cs.CV / 76 / 2610.11431

EchoDiST: Self-distillation-based joint learning for diffusion-conditioned echocardiographic myocardial motion estimation

EchoDiST：基于自蒸馏的联合学习用于扩散条件下的超声心动图心肌运动估计

Qi, Feiyue, Wei, Xingyue, Luo, Jianwen

Abstract

Motion estimation in echocardiography is essential for quantitative assessment of cardiac function and myocardial mechanics, but remains challenging due to image artifacts, limited image information, speckle decorrelation, and the scarcity of ground-truth displacement fields. Anatomy-guided approaches can provide structural information, yet often rely on expert-labeled myocardial segmentations. We propose EchoDiST, a framework for unsupervised echocardiographic myocardial motion estimation that integrates self-distillation-based joint learning with a diffusion-conditioned motion estimation network. Here, unsupervised motion estimation refers to learning without ground-truth displacement fields. The self-distillation strategy jointly optimizes anatomical segmentation and myocardial motion estimation under limited anatomical annotations. Diffusion-based conditioning is used during training with stochastic perturbations, while inference requires only a single deterministic forward pass without iterative reverse-diffusion sampling. EchoDiST was evaluated on three echocardiographic datasets, including two external test datasets under cross-view and cross-dataset settings. Compared with seven representative learning-based methods, EchoDiST consistently improved anatomical alignment, myocardial strain assessment, and motion-derived functional and cardiac-phase assessment. These gains were statistically significant across the evaluated tasks and datasets. Overall, EchoDiST provides an effective approach for reliable myocardial motion estimation under limited anatomical supervision and supports downstream quantitative assessment of cardiac function.

Chinese Translation

超声心动图中的运动估计对于心脏功能和心肌力学的定量评估至关重要，但由于图像伪影、有限的图像信息、斑点去相关以及真实位移场的稀缺，仍然面临挑战。基于解剖结构的方法可以提供结构信息，但通常依赖于专家标注的心肌分割。我们提出了EchoDiST，一个用于无监督超声心动图心肌运动估计的框架，它将基于自蒸馏的联合学习与扩散条件的运动估计网络相结合。在这里，无监督运动估计是指在没有真实位移场的情况下进行学习。自蒸馏策略在有限的解剖标注下共同优化解剖分割和心肌运动估计。训练过程中使用基于扩散的条件和随机扰动，而推理只需一次确定性的前向传播，无需迭代的反向扩散采样。EchoDiST在三个超声心动图数据集上进行了评估，包括在交叉视图和交叉数据集设置下的两个外部测试数据集。与七种代表性的基于学习的方法相比，EchoDiST在解剖对齐、心肌应变评估以及运动衍生的功能和心脏相位评估方面均有持续改善。这些提升在评估的任务和数据集中具有统计学意义。总体而言，EchoDiST为在有限解剖监督下可靠的心肌运动估计提供了一种有效的方法，并支持下游心脏功能的定量评估。

View on arXiv Download PDF AI Translation

cs.CV / 77 / 2610.11433

DLC: A Metric-Guided Dynamic Loss Controller for Multi-Objective Training

DLC：一种基于指标引导的动态损失控制器用于多目标训练

Ko, Jaewan, Choi, Janghoon

Abstract

In this paper, we introduce a metric-guided dynamic loss controller (DLC) for multi-objective image restoration. Conventional image restoration pipelines usually train with a fixed weighted combination of multiple losses, without changing the relative importance of fidelity, perceptual similarity, and no-reference quality during optimization. DLC is an architecture- and loss-term-agnostic training-time controller: it does not modify the restoration architecture or introduce new differentiable loss terms, but dynamically reweights the existing training losses. During training, DLC periodically evaluates the current model on a small fixed feedback subset and uses the resulting quality metrics to update the loss-weight vector through an LLM-based controller. Because DLC operates on existing loss terms rather than task-specific architectures, the same controller formulation can be instantiated across diverse image restoration training pipelines. We evaluate DLC on three restoration domains: low-light image enhancement, deraining, and real-world super-resolution, using both reference-based and no-reference quality metrics. Across these settings, DLC considers metric-dependent trade-offs during optimization and guides training toward balanced operating points across fidelity and perceptual quality. The results show that DLC can move models toward more favorable operating points across different restoration domains, supporting its role as a practical plug-in controller for multi-objective image restoration.

Chinese Translation

在本文中，我们介绍了一种用于多目标图像修复的基于指标引导的动态损失控制器（DLC）。传统的图像修复流程通常使用固定权重组合的多种损失进行训练，在优化过程中不改变保真度、感知相似性和无参考质量的相对重要性。DLC是一种与架构和损失项无关的训练时控制器：它不修改修复架构或引入新的可微分损失项，而是动态地重新加权现有的训练损失。在训练过程中，DLC定期在一个固定的小反馈子集上评估当前模型，并利用得到的质量指标通过基于LLM的控制器更新损失权重向量。由于DLC在现有损失项上操作，而不是特定于任务的架构，因此相同的控制器公式可以在不同的图像修复训练流程中实例化。我们在三个修复领域对DLC进行了评估：低光照图像增强、去雨和真实世界超分辨率，使用了基于参考和无参考的质量指标。在这些设置中，DLC在优化过程中考虑了依赖指标的权衡，并引导训练朝向保真度和感知质量之间的平衡操作点。结果表明，DLC能够使模型在不同修复领域朝向更有利的操作点移动，支持其作为多目标图像修复的实用插件控制器的角色。

View on arXiv Download PDF AI Translation

cs.CV / 78 / 2610.11444

Learning to Retrieve: Internalizing Memory Retrieval for Video World Models

学习检索：内化视频世界模型的记忆检索

Hu, JiaKui, Chen, Tailai, Pan, Yuqi, Qiu, Xuerui, Liu, Jialun, Cao, Xiao, Zhu, Zhenxin, Chen, Guang, Ye, Hangjun, Wang, Bing, Lu, Yanye

Abstract

Video world models aim to generate explorable, 3D-consistent scene videos conditioned on camera trajectories. Existing approaches often rely on external memory systems that explicitly retrieve previously observed content to mitigate scene drift during long-horizon generation. However, these auxiliary memory pathways operate outside the model's internal generative dynamics, preventing the model from intrinsically learning when and what historical information should be retrieved. We propose to internalize memory retrieval into the generation process, allowing retrieval to emerge as an intrinsic behavior of the video world model rather than relying on an external memory system. Based on this principle, we introduce \textbf{Learning-to-Retrieve (L2R)}, which repurposes the model's persistent internal state as a memory for historical context. A camera-conditioned retrieval gate selectively accesses relevant historical information from this state, determining \textit{what to retrieve}, while a retrieval trigger determines \textit{when to retrieve}. We further supervise the trigger with a 3D re-visibility signal, activating retrieval when previously observed content re-enters the current view while otherwise preserving the existing context. Together, these components enable the model to intrinsically acquire memory retrieval behavior and incorporate relevant historical observations into generation without a separate retrieval pathway. Across multiple base models and camera-revisit benchmarks, L2R improves long-term scene consistency while eliminating the need for an external memory bank or 3D conditions. https://jkhu29.github.io/l2r

Chinese Translation

视频世界模型旨在生成可探索的、三维一致的场景视频，这些视频是基于摄像机轨迹生成的。现有方法通常依赖于外部记忆系统，明确检索先前观察到的内容，以减轻在长时间生成过程中场景漂移的问题。然而，这些辅助记忆路径在模型的内部生成动态之外操作，阻碍了模型内在学习何时以及应检索何种历史信息。我们提出将记忆检索内化到生成过程中，使检索作为视频世界模型的内在行为出现，而不是依赖外部记忆系统。基于这一原则，我们引入了 extbf{学习检索（Learning-to-Retrieve, L2R）}，该方法将模型的持久内部状态重新利用为历史上下文的记忆。一个基于摄像机条件的检索门选择性地从该状态中访问相关的历史信息，以确定 extit{检索什么}，而检索触发器则决定 extit{何时检索}。我们进一步通过三维可见性信号来监督触发器，当先前观察到的内容重新进入当前视图时激活检索，同时保持现有上下文。通过这些组件，模型能够内在地获得记忆检索行为，并将相关的历史观察融入生成中，而无需单独的检索路径。在多个基础模型和摄像机重访基准测试中，L2R提高了长期场景一致性，同时消除了对外部记忆库或三维条件的需求。

View on arXiv Download PDF AI Translation

cs.CV / 79 / 2610.11460

ProtoSemImage: Image-Valued Prototypes with Deformable Row Alignment for Interpretable Document Classification

ProtoSemImage：具有可变形行对齐的图像值原型用于可解释的文档分类

Zare, Mohammad, Shamsinejadbabaki, Pirooz

Abstract

Prototypes in classification models are almost always vectors, and a vector has no readable form. This paper asks what happens when a prototype is an image. Documents give the question a natural form, because a document can be rendered as a multi-channel image in which every token becomes a pixel, so a class representative can take the same shape and the same channel semantics as the inputs it stands for. ProtoSemImage represents each class by one or more visual archetypes: prototype images in a four-channel HSV space whose channels carry named linguistic factors. A Skip-Gram objective learns that color space end to end through a four-dimensional bottleneck, discourse boundary rows become differentiable typed difference rows, and classification reduces to 2D visual template matching: a deformable row alignment between a document image and the archetype bank, in the spirit of dynamic time warping. Because the match is a spatial pattern comparison rather than a linear readout, the model reports where an input departs from its archetype and along which channel, and a generative head decodes each archetype back into text. The image representation works: it beats an otherwise identical model with vector prototypes in all three paired seeds, by between 4.3 and 11.8 points on a ten-class task. The distance-based matching does not. A diagnostic that keeps the representation fixed and swaps only the classifier recovers the sequence baselines, which locates a 20.6-point shortfall in the matching rather than in the color compression, and a benchmark built so that a pair of documents shares a bag of words and differs only in arrangement confirms the layout-preservation it was designed for. We report both directions, because for a representation whose whole purpose is inspect ability, the failure modes are as informative as the gains.

Chinese Translation

分类模型中的原型几乎总是向量，而向量没有可读的形式。本文探讨当原型是图像时会发生什么。文档为这个问题提供了自然的形式，因为文档可以被渲染为多通道图像，其中每个标记变成一个像素，因此类代表可以与其所代表的输入具有相同的形状和通道语义。ProtoSemImage通过一个或多个视觉原型来表示每个类：在四通道HSV空间中的原型图像，其通道承载命名的语言因素。Skip-Gram目标通过一个四维瓶颈端到端地学习颜色空间，话语边界行变为可微分的类型差异行，分类简化为二维视觉模板匹配：在文档图像和原型库之间进行可变形行对齐，体现动态时间扭曲的精神。由于匹配是空间模式比较而不是线性读取，因此模型报告输入在何处偏离其原型以及沿哪个通道，而生成头将每个原型解码回文本。图像表示有效：在所有三个配对种子中，它比一个在向量原型上完全相同的模型高出4.3到11.8分，适用于十类任务。基于距离的匹配则不然。一个保持表示固定并仅交换分类器的诊断恢复了序列基线，定位到匹配中的20.6分短缺而非颜色压缩，并且一个构建的基准使得一对文档共享词袋，仅在排列上有所不同，确认了其设计的布局保持性。我们报告了两个方向，因为对于一个其全部目的在于可检验性的表示，失败模式与收益同样具有信息价值。

View on arXiv Download PDF AI Translation

cs.CV / 80 / 2610.11469

SAGE: Sink-Aware Guided Emphasis for Visual Grounding in Vision-Language Decoders

SAGE：面向视觉语言解码器的沉没感知引导强调

Song, Jeonghyo, Yoo, YoungJoon

Abstract

Recent large vision-language models (VLMs) pair a visual encoder with a large language model (LLM) and perform well on diverse image-text tasks, yet their reliability is often limited by decoder attention pathologies that suppress visual evidence and exacerbate hallucinations. In this paper, we revisit visual attention sinks and uncover a structured, layer-dependent behavior: across prompts, early and late decoder layers exhibit prompt-invariant attention collapse onto the same few image regions, which we term PIS (Prompt-Invariant Sinks), whereas mid layers become prompt-conditioned and drive vision-language alignment. This split suggests that treating sinks as a uniform effect is incomplete. Building on this insight, we propose SAGE (Sink-Aware Guided Emphasis), a lightweight intervention that steers decoder attention away from PIS and toward query-dependent regions of interest (ROIs) using token-aligned ROI masks derived from standard vision backbones such as CLIP, ViT, and DINOv3. Evaluated on diverse vision-encoder + decoder-only LLM VLM families, SAGE improves visual grounding, reduces hallucinations, and yields consistent gains across public downstream vision-language benchmarks, including fine-grained visual discrimination settings where localized evidence is crucial, when instantiated with backbone-derived ROI masks.

Chinese Translation

近年来，大型视觉语言模型（VLMs）将视觉编码器与大型语言模型（LLM）相结合，在多样的图像-文本任务中表现出色，但它们的可靠性常常受到解码器注意力病态的限制，这些病态抑制了视觉证据并加剧了幻觉。在本文中，我们重新审视了视觉注意力沉没，并揭示了一种结构化的、层依赖的行为：在不同的提示下，早期和晚期的解码器层表现出对同几处图像区域的提示不变注意力崩溃，我们称之为PIS（Prompt-Invariant Sinks），而中间层则变得依赖于提示，并推动视觉语言对齐。这一分裂表明，将沉没视为统一效应是不完整的。基于这一见解，我们提出了SAGE（Sink-Aware Guided Emphasis），这是一种轻量级干预措施，旨在通过使用从标准视觉骨干网络（如CLIP、ViT和DINOv3）派生的与标记对齐的ROI掩码，将解码器注意力引导远离PIS，朝向依赖于查询的兴趣区域（ROIs）。在多种视觉编码器 + 仅解码器的LLM VLM系列上进行评估时，SAGE改善了视觉基础，减少了幻觉，并在公共下游视觉语言基准测试中产生了一致的提升，包括在局部证据至关重要的细粒度视觉辨别设置中，当使用基于骨干网络的ROI掩码实例化时。

View on arXiv Download PDF AI Translation

cs.CV / 81 / 2610.11479

Conditional Residual Prediction: Improving Autoregressive Video Diffusion without a Bidirectional Teacher

条件残差预测：在没有双向教师的情况下改善自回归视频扩散

Zheng, Bowen, Liu, Zhiguang, Ou, Jiarong, Chen, Rui, Hu, Tianyang

Abstract

Causal video diffusion models generate video autoregressively, which suits streaming, interactive, and long-video generation. Under standard training, however, they often yield lower generation quality than bidirectional models of the same size. Many existing approaches address this gap by initializing from or distilling a pretrained bidirectional teacher. We instead train a causal model from an image-model initialization, with no bidirectional video model at any stage. Because this path requires neither a large bidirectional teacher nor a complex distillation pipeline, it is simpler and more scalable. On this path, we find that a causal model trained on ground-truth history becomes strongly dependent on it, so that at inference errors in its own generated history propagate forward. We hypothesize that much of this dependence is unnecessary, because the current input already determines much of what the history provides. We propose Conditional Residual Prediction (CRP), a simple recipe for reducing a model's reliance on a condition: the model first predicts the target without the condition, and the condition may only add a residual on top of this prediction. Applied to history, CRP makes the model predict each chunk from the present as far as it can and use the past only for what the present cannot supply. In controlled experiments, CRP nearly closes the 6.14-point gap to a bidirectional model trained under the same setup. Scaling this recipe, we train Optica, a 2B-parameter causal video model that autoregressively generates 5-second 480p videos and reaches 82.78 on VBench with only about 15M training videos.

Chinese Translation

因果视频扩散模型以自回归方式生成视频，适用于流媒体、交互式和长视频生成。然而，在标准训练下，它们通常产生的生成质量低于同等规模的双向模型。许多现有方法通过从预训练的双向教师初始化或提炼来解决这一差距。我们则从图像模型初始化训练一个因果模型，在任何阶段都不使用双向视频模型。由于这一路径既不需要大型双向教师，也不需要复杂的提炼流程，因此更简单且更具可扩展性。在这一过程中，我们发现基于真实历史训练的因果模型对其强烈依赖，以至于在推理时自身生成的历史中的错误会向前传播。我们假设这种依赖性在很大程度上是多余的，因为当前输入已经决定了历史提供的许多内容。我们提出了条件残差预测（Conditional Residual Prediction, CRP），这是一种简单的策略，用于减少模型对条件的依赖：模型首先在没有条件的情况下预测目标，条件仅在此预测上添加残差。应用于历史时，CRP使模型尽可能从当前时刻预测每个片段，仅在当前无法提供的情况下使用过去。在控制实验中，CRP几乎缩小了与在相同设置下训练的双向模型之间的6.14分差距。通过扩展这一策略，我们训练了Optica，一个具有20亿参数的因果视频模型，能够自回归生成5秒的480p视频，并在VBench上达到82.78，仅使用约1500万条训练视频。

View on arXiv Download PDF AI Translation

cs.CV / 82 / 2610.11489

Beyond Resolution: Object-to-Image Ratio Mismatch in Instance Retrieval

超越分辨率：实例检索中的物体与图像比例不匹配

Meivar, Boaz, Kedem, Ofir, Edenzon, Amit, Chechik, Gal, Avidan, Shai

Abstract

Visual instance retrieval often fails when the same object appears at different apparent sizes in the query and gallery. We show that the dominant cause is usually not resolution loss but object-to-image (O2I) ratio mismatch: the object occupies different fractions of the two images. On a controlled benchmark of 3,021 Objaverse objects rendered at five camera distances, more than 80% of the cross-distance degradation is attributable to O2I mismatch rather than resolution for 9 of 12 pretrained backbones; multi-scale architectures cut the resolution-only effect to single digits yet remain equally susceptible. The failure is also asymmetric: tight queries retrieve more reliably against wide gallery images than the reverse. Guided by this analysis, query-side scale augmentation and an OWLv2 crop reranker reach state of the art on ILIAS 100M (29.2 mAP@1000 before reranking, 42.0 after) without training or modifying the precomputed gallery index, and a LoRA fine-tune matches the query-side gains at a single forward pass, showing that O2I robustness is learnable.

Chinese Translation

视觉实例检索在查询和图库中同一物体以不同显著大小出现时常常失败。我们表明，主要原因通常不是分辨率损失，而是物体与图像（O2I）比例不匹配：物体在两幅图像中占据的比例不同。在一个控制的基准测试中，3,021个在五个相机距离下渲染的Objaverse物体中，12个预训练主干网络中有9个的跨距离降级超过80%可归因于O2I不匹配，而非分辨率；多尺度架构将仅分辨率的影响降低到个位数，但仍然同样容易受到影响。失败也是不对称的：紧凑的查询在宽图库图像中检索的可靠性高于反向情况。根据这一分析，查询端的尺度增强和OWLv2裁剪重排序器在ILIAS 100M上达到了最先进的水平（重排序前29.2 mAP@1000，重排序后42.0），而无需训练或修改预计算的图库索引，LoRA微调在单次前向传递中匹配了查询端的增益，表明O2I的鲁棒性是可学习的。

View on arXiv Download PDF AI Translation

cs.CV / 83 / 2610.11496

Stop My Dancing! Understanding, Detecting and Attributing Motion-Aware Deepfake Videos

停止我的舞蹈！理解、检测与归因运动感知深度伪造视频

Liu, Fazhong, Meng, Yan, Dong, Tian, Chen, Guoxing, Zhu, Haojin

Abstract

Pose-guided diffusion models can now synthesize entire human figures in motion, spawning a new class of deepfakes: Motion Aware Deepfake (MAD) that have already reached hundreds of millions of viewers. To better understand this emerging threat, we construct the first MAD-specific benchmark and measurement framework, containing over 1.5 million frames that mix 1,363 real and 30,122 synthetic videos from six controllable generators, with realistic perturbations and open-world evaluation splits. Then, we dissect MAD and discover that, despite their global coherence, these videos betray faint yet reliable cues: because the model relies on limited input frames for motion synthesis, it must predict and simulate coherent movement at motion boundaries, thereby producing high-frequency artifacts along with model-specific spectral fingerprints. Based on the observations obtained from analysis on dataset, we propose MoDA, the first defense framework tailored to detect and attribute MAD videos. MoDA couples spatial semantics with steganalysis-rich frequency features via cross-domain alignment and multi-scale aggregation, achieving 94.8% in-distribution and 89.1% cross-dataset detection accuracy gains of 10% to 25% over prior work and 91.5% model attribution accuracy. MoDA achieves 81.94% accuracy on 200 clips produced by two unseen commercial MAD platforms, indicating promising zero-shot transfer, and 78.13% detection accuracy on 1,200 unseen MAD video clips (55k frames in total) collected from the open Internet. Under white-box, gray-box, and black-box adaptive attacks, MoDA maintains relatively stable detection and attribution performance while the accuracies of the baselines drop rapidly.

Chinese Translation

基于姿态引导的扩散模型现在能够合成整个运动中的人类形象，催生了一类新的深度伪造视频：运动感知深度伪造（Motion Aware Deepfake, MAD），已吸引了数亿观众。为了更好地理解这一新兴威胁，我们构建了首个针对MAD的基准和测量框架，包含超过150万帧，混合了来自六个可控生成器的1,363个真实视频和30,122个合成视频，具有现实的扰动和开放世界的评估划分。然后，我们对MAD进行深入分析，发现尽管这些视频在整体上具有一致性，但它们仍然暴露出微弱但可靠的线索：由于模型依赖有限的输入帧进行运动合成，它必须在运动边界处预测和模拟一致的运动，从而在模型特定的频谱指纹上产生高频伪影。基于对数据集分析中获得的观察结果，我们提出了MoDA，这是第一个专门用于检测和归因MAD视频的防御框架。MoDA通过跨域对齐和多尺度聚合，将空间语义与丰富的频率特征结合，实现了94.8%的分布内检测准确率和89.1%的跨数据集检测准确率，相较于之前的工作提高了10%到25%，并达到了91.5%的模型归因准确率。MoDA在由两个未见商业MAD平台生成的200个视频片段上实现了81.94%的准确率，显示出良好的零样本迁移能力，并在从开放互联网收集的1,200个未见MAD视频片段（总计55,000帧）上达到了78.13%的检测准确率。在白盒、灰盒和黑盒自适应攻击下，MoDA保持了相对稳定的检测和归因性能，而基线的准确率则迅速下降。

View on arXiv Download PDF AI Translation

cs.CV / 84 / 2610.11498

Parametric Trajectory Distillation for Few-Step Video Generation

用于少步视频生成的参数化轨迹蒸馏

Feng, Lan, Karkus, Peter, Igl, Maximilian, Berner, Julius, Chen, Yuxiao, Tan, Shuhan, Alahi, Alexandre, Ivanovic, Boris, Pavone, Marco

Abstract

Video diffusion and flow models require many sequential evaluations, making generation computationally expensive. Few-step distillation reduces this cost but poses a capacity allocation problem: a student must match the teacher's iterative generation with far less sequential computation. Existing trajectory methods ask the student to reproduce teacher transitions that are highly curved at high noise, which can exceed its capacity and degrade fine detail. We introduce Parametric Trajectory Distillation (PTD), which lets the student parameterize teacher trajectory segments as polynomials and learn from teacher guidance along its own predicted path. PTD is designed to let the learned curvature adapt to the backbone's predictive capacity, preserving motion and diversity. The curvature head is used only in training; inference keeps the original backbone architecture. On Wan2.1-14B, four-step PTD sets a new state of the art for trajectory distillation, significantly improving dynamic quality and naturalness over PDD, the best-performing trajectory-only method on this model, under the same training setting. On the 33B audio-video MiniMax-H3, LoRA-trained PTD significantly improves diversity and naturalness over the state-of-the-art LightX2V Turbo. Blinded human votes give PTD 55.1% and 63.4% preference shares against PDD and LightX2V Turbo. Project page: https://alan-lanfeng.github.io/PTD/.

Chinese Translation

视频扩散和流模型需要进行大量的连续评估，这使得生成过程计算成本高昂。少步蒸馏可以降低这一成本，但也带来了容量分配问题：学生必须在远少于教师的连续计算下匹配教师的迭代生成。现有的轨迹方法要求学生在高噪声下重现教师的高度弯曲过渡，这可能超出其能力并降低细节表现。我们引入了参数化轨迹蒸馏（Parametric Trajectory Distillation, PTD），该方法允许学生将教师轨迹段参数化为多项式，并沿着自己预测的路径从教师指导中学习。PTD的设计旨在使学习到的曲率适应主干网络的预测能力，从而保持运动性和多样性。曲率头仅在训练中使用；推理时保持原始主干架构。在Wan2.1-14B上，四步PTD在轨迹蒸馏方面设定了新的最先进水平，在相同训练设置下显著提高了动态质量和自然性，优于PDD，这是该模型上表现最佳的仅轨迹方法。在33B音视频MiniMax-H3上，经过LoRA训练的PTD在多样性和自然性方面显著优于最先进的LightX2V Turbo。盲测人类投票显示PTD对PDD和LightX2V Turbo的偏好份额分别为55.1%和63.4%。项目页面：https://alan-lanfeng.github.io/PTD/

View on arXiv Download PDF AI Translation

cs.CV / 85 / 2610.11526

MSGAT: Multi-Head Spiking Graph Attention with Similarity-Space Fusion for Image-Text Retrieval

MSGAT：基于相似性空间融合的多头脉冲图注意力网络用于图像-文本检索

Zong, Xintao, Liu, Wenxuan, Ding, Jianhao, Yu, Zhaofei, Huang, Tiejun

Abstract

Spiking neural networks (SNNs) offer an energy-efficient computing paradigm through sparse event-driven computation, showing great potential for efficient multimodal learning. However, applying SNNs to high-level multimodal tasks, such as image-text retrieval (ITR), remains challenging, since sparse spike representations make it difficult to capture semantic structures required for cross-modal alignment. Existing spiking ITR methods rely on local alignment and additional soft-label supervision during training, while lacking awareness of structural and multi-granularity relationships. To address these issues, we propose a Multi-head Spiking Graph Attention Network (\textbf{MSGAT}) for structural modeling and equip it with dynamic attention heads to capture complementary relational patterns and enable spike-driven graph reasoning and aggregation. However, within a two-branch multi-granularity fusion framework, the fine-grained spike representations generated by MSGAT are sparse and discrete, whereas the global representations are continuous, making conventional feature-level fusion susceptible to interference across heterogeneous representations. Therefore, we introduce \textbf{Sim-Fuse}, a similarity-space fusion alignment strategy integrating coarse- and fine-grained matching relations while avoiding direct fusion of heterogeneous representations. Experiments on Flickr30K and MSCOCO show our method outperforms ANN methods under matched settings and existing SNN retrieval baselines. Moreover, with only two time steps, our SNN achieves comparable or superior performance to its ANN counterpart while reducing theoretical module-level energy by 55\%. The code is provided in the Supplementary Materials.

Chinese Translation

脉冲神经网络（SNNs）通过稀疏事件驱动计算提供了一种节能的计算范式，显示出在高效多模态学习中的巨大潜力。然而，将SNN应用于高层次的多模态任务，如图像-文本检索（ITR），仍然面临挑战，因为稀疏的脉冲表示使得捕捉跨模态对齐所需的语义结构变得困难。现有的脉冲ITR方法依赖于局部对齐和训练期间的额外软标签监督，但缺乏对结构和多粒度关系的意识。为了解决这些问题，我们提出了一种多头脉冲图注意力网络（MSGAT）用于结构建模，并为其配备动态注意力头，以捕捉互补的关系模式，并实现脉冲驱动的图推理和聚合。然而，在一个双分支多粒度融合框架内，MSGAT生成的细粒度脉冲表示是稀疏和离散的，而全局表示则是连续的，这使得传统的特征级融合容易受到异构表示之间的干扰。因此，我们引入了Sim-Fuse，一种相似性空间融合对齐策略，整合粗粒度和细粒度匹配关系，同时避免异构表示的直接融合。在Flickr30K和MSCOCO上的实验表明，我们的方法在匹配设置下优于ANN方法和现有的SNN检索基线。此外，仅用两个时间步，我们的SNN在性能上与其ANN对应物相当或更优，同时将理论模块级能耗降低了55%。代码已在补充材料中提供。

View on arXiv Download PDF AI Translation

cs.CV / 86 / 2610.11534

HAND: A Biologically-Inspired Activation Function that Improves Generalisation and Sample Efficiency in Image Classification

HAND：一种生物启发的激活函数，改善图像分类中的泛化能力和样本效率

Spratling, Michael W., Schütt, Heiko H.

Abstract

DNNs exhibit robustness and generalisation issues not seen in humans. They are also far less data-efficient learners, requiring considerably more training samples to accurately classify novel exemplars. Inductive bias could help with these issues by providing in-built mechanisms to improve generalisation, and hence, reduce reliance on learning from data. We incorporate a biologically-inspired inductive bias into a new activation function, HAND (Homeostasis, Accelerating Nonlinearity, and Divisive-nomalisation), and show its effectiveness with CNNs trained on image classification. Using HAND a ConvNeXt-tiny required 25 training epochs to reach the same accuracy on ImageNet1k as the unmodified model achieved after 200 epochs. Consistent with the effects of an inductive bias, the performance gap reduced with training time and increased data augmentation. When the volume of training data was reduced and unevenly distributed between classes (Long-tailed ImageNet) the improvements in accuracy were even larger and did not reduce with increased training time. Generalisation performance with the common-corruptions data, and the ability to reject samples from unknown classes, were unaffected or improved by HAND. Results generalised across CNN architectures and training data-sets. HAND can, therefore, reduce the required training time and/or the required volume and variety of training data, helping to improve sample efficiency.

Chinese Translation

深度神经网络（DNNs）表现出与人类不同的鲁棒性和泛化问题。它们的学习效率远低于人类，通常需要更多的训练样本才能准确分类新样本。归纳偏差可以通过提供内置机制来改善泛化能力，从而减少对数据学习的依赖。我们将一种生物启发的归纳偏差融入到新的激活函数HAND（Homeostasis, Accelerating Nonlinearity, and Divisive-nomalisation）中，并展示其在图像分类任务中与卷积神经网络（CNNs）的有效性。使用HAND的ConvNeXt-tiny模型在ImageNet1k上达到与未修改模型在200个训练周期后相同的准确率，仅需25个训练周期。与归纳偏差的影响一致，性能差距随着训练时间的增加和数据增强的增加而减小。当训练数据量减少且在类别之间分布不均（长尾ImageNet）时，准确率的提升更为显著，并且随着训练时间的增加并未减小。使用常见干扰数据的泛化性能，以及拒绝未知类别样本的能力，均未受到HAND的影响，甚至有所改善。结果在不同的CNN架构和训练数据集上均得到了推广。因此，HAND可以减少所需的训练时间和/或所需的训练数据量和多样性，从而帮助提高样本效率。

View on arXiv Download PDF AI Translation

cs.CV / 87 / 2610.11547

OX-NeRF: 3D X-ray Tomography Reconstruction from Sparse Views Using Implicit Neural Representation

OX-NeRF：基于隐式神经表示的稀疏视图3D X射线断层重建

Welsch, Thomas, Tu, Min-Hsin, Chapman, David J., Eakins, Daniel E.

Abstract

NeRF and Gaussian splatting methods have been successfully applied on X-ray scenes where the views are too sparse for 3D reconstruction via classical methods. Ultra-sparse scenes with 10 or fewer views such as those with high-rate or low-dose acquisition still, however, present a significant challenge. To address this problem we present a new framework, Optimised X-ray Neural Radiance Fields (OX-NeRF), that combines cross-scene feature learning with scene-specific optimisation to reconstruct sets of related scenes. OX-NeRF employs a convolutional neural network (CNN) to identify cross-scene features while maintaining scene-specific multi-resolution hash grids of spatial features. The paired representations are fused and passed to a multilayer perceptron (MLP); the CNN, hash grids and MLP are then jointly optimised end-to-end. Benchmarking on parallel-beam and cone-beam X-ray datasets shows OX-NeRF provides significantly higher reconstruction accuracy on ultra-sparse scenes compared to existing radiance field methods.

Chinese Translation

NeRF 和高斯点云方法已成功应用于 X 射线场景，这些场景的视图过于稀疏，无法通过经典方法进行 3D 重建。然而，具有 10 个或更少视图的超稀疏场景，如高频率或低剂量采集的场景，仍然面临重大挑战。为了解决这个问题，我们提出了一个新的框架，优化 X 射线神经辐射场（OX-NeRF），该框架结合了跨场景特征学习和场景特定优化，以重建相关场景集。OX-NeRF 采用卷积神经网络（CNN）来识别跨场景特征，同时保持场景特定的多分辨率哈希网格空间特征。这些配对表示被融合并传递给多层感知器（MLP）；然后，CNN、哈希网格和 MLP 共同进行端到端的优化。在平行束和锥束 X 射线数据集上的基准测试表明，与现有的辐射场方法相比，OX-NeRF 在超稀疏场景上的重建精度显著提高。

View on arXiv Download PDF AI Translation

cs.CV / 88 / 2610.11560

Hankel Subspace Self-Supervised Learning for Parallel MRI Reconstruction

汉克尔子空间自监督学习用于并行磁共振成像重建

Hu, Mingyu, Zhu, Siquan, Zhong, Xijun, Liu, Qiegen

Abstract

Parallel magnetic resonance imaging reconstruction is an ill-posed inverse problem under undersampling. Multi-coil acquisition and Hankel lifting expose complementary repeated information: observations of the same anatomy across coils and repeated local k-space neighborhoods in overlapping windows. These dependencies guide recovery of missing k-space data. However, splitting lifted Hankel entries for self-supervision can place the original sample in both input and target, causing data leakage. We propose Hankel Subspace Self-Supervised Reconstruction (HSSRecon), a scan-specific reconstruction framework for parallel magnetic resonance imaging. HSSRecon partitions data by physical acquisition units before Hankel lifting and applies multiplicity normalization to repeated Hankel copies in overlapping windows. Rather than learning a mapping that directly predicts missing data, the network learns a compact complex-valued Hankel subspace operator. Reconstruction is performed over the original k-space variables using a conjugategradient solver with hard data consistency. This design separates structural learning in the Hankel domain from data consistency in the physical domain: the former exploits multi-coil and local Hankel correlations, while the latter solves over unacquired degrees of freedom. We provide theoretical analyses of physicalgroup splitting and multiplicity normalization, and establish positive definiteness, uniqueness, hard data consistency, and a finite-step conjugate-gradient error bound for the system. On fastMRI brain data with three contrasts and three sampling masks, HSSRecon achieves competitive peak signal-to-noise ratio, structural similarity, and normalized mean squared error across six aggregated conditions.

Chinese Translation

并行磁共振成像重建在欠采样情况下是一个病态反问题。多线圈采集和汉克尔提升揭示了互补的重复信息：在不同线圈中对相同解剖结构的观察以及在重叠窗口中重复的局部k空间邻域。这些依赖关系指导缺失k空间数据的恢复。然而，将提升的汉克尔条目拆分用于自监督可能会导致原始样本同时出现在输入和目标中，从而引发数据泄漏。我们提出了汉克尔子空间自监督重建（HSSRecon），这是一个针对并行磁共振成像的扫描特定重建框架。HSSRecon在汉克尔提升之前按物理采集单元对数据进行分区，并对重叠窗口中的重复汉克尔副本应用多重归一化。网络学习一个紧凑的复值汉克尔子空间算子，而不是直接预测缺失数据的映射。重建是在原始k空间变量上使用共轭梯度求解器进行的，并保持硬数据一致性。该设计将汉克尔域中的结构学习与物理域中的数据一致性分开：前者利用多线圈和局部汉克尔相关性，而后者则在未采集的自由度上进行求解。我们提供了物理组分割和多重归一化的理论分析，并建立了系统的正定性、唯一性、硬数据一致性以及有限步共轭梯度误差界限。在具有三种对比和三种采样掩模的fastMRI脑数据上，HSSRecon在六种聚合条件下实现了具有竞争力的峰值信噪比、结构相似性和归一化均方误差。

View on arXiv Download PDF AI Translation

cs.CV / 89 / 2610.11572

PAM-ToD: Plug-and-Play Appearance Modeling for Cross-Time-of-Day 3D Gaussian Splatting

PAM-ToD：跨时间段的即插即用外观建模用于3D高斯点云渲染

Shimomura, Kota, Moon, Sungho, Hirakawa, Tsubasa, Yamashita, Takayoshi, Im, Sunghoon, Fujiyoshi, Hironobu

Abstract

Adapting a pre-trained 3D Gaussian Splatting (3DGS) road scene to a new time of day requires learning appearance changes from a few anchor images while preserving consistent, real-time rendering. We propose PAM-ToD, a lightweight plug-in that learns color corrections while keeping the pre-trained 3DGS parameters fixed. PAM-ToD scales each Gaussian's existing color to model illumination changes and uses an additive term for additional brightness, such as when street lamps turn on at night. Under a simplified image formation model, unchanged surface albedo can be eliminated from the relation between source and target appearances, allowing us to learn these corrections without separately estimating albedo and illumination. The model corrects colors across the scene while allowing the corrections to vary by location and by Gaussian. To guide learning from a few anchor images, it discourages abrupt spatial changes in these corrections. We also introduce CARLA-ToD, a benchmark with matching geometry, camera poses, and moving-object trajectories across three times of day. A few target-time anchor images are used to train each plug-in, while separate views are used for evaluation. Across the static and dynamic settings, PAM-ToD achieves higher PSNR and lower LPIPS than the baselines, even when the anchor images come from a single synchronized capture across multiple cameras.

Chinese Translation

将预训练的3D高斯点云（3DGS）道路场景适应到新的时间段需要从少量锚点图像中学习外观变化，同时保持一致的实时渲染。我们提出了PAM-ToD，这是一种轻量级插件，通过保持预训练的3DGS参数不变来学习颜色校正。PAM-ToD对每个高斯的现有颜色进行缩放，以建模光照变化，并使用附加项来增加亮度，例如在夜间街灯开启时。在简化的图像形成模型下，可以消除源和目标外观之间的关系中的不变表面反照率，从而使我们能够在不单独估计反照率和光照的情况下学习这些校正。该模型在整个场景中校正颜色，同时允许校正因位置和高斯而异。为了指导从少量锚点图像中学习，它抑制这些校正中的突变空间变化。我们还引入了CARLA-ToD，这是一个基准，具有匹配的几何形状、相机姿态和跨越三个时间段的移动物体轨迹。少量目标时间的锚点图像用于训练每个插件，而单独的视图用于评估。在静态和动态设置中，PAM-ToD在PSNR上取得了更高的值，并在LPIPS上取得了更低的值，甚至当锚点图像来自多个相机的单次同步捕获时也是如此。

View on arXiv Download PDF AI Translation

cs.CV / 90 / 2610.11579

SV-TAD: Native Sparse Convs for Efficient Temporal Action Detection

SV-TAD：用于高效时间动作检测的原生稀疏卷积

Pizarro, Ricardo, Valle, Roberto, Buenaposada, José M., Bergasa, Luis M., Baumela, Luis

Abstract

To adapt billion-parameter Vision Transformers for long-video understanding, recent methods freeze the backbone and train lightweight convolutional modules. While effective for parameter-efficient training, existing adapters do not reduce inference-time computation, leaving scalability with respect to video length largely unaddressed. Token selection can reduce attention cost by pruning redundant tokens, but it breaks the spatial grid structure required by convolutional adapters. This forces an expensive dense reconstruction, nullifying much of the potential speedup. We address this by introducing native sparse 2D convolutions, a primitive that allows these adapters, for the first time, to operate directly and efficiently on dynamically pruned token sets. We integrate this primitive into SV-TAD, an adapter framework for temporal action detection, reducing VideoMAEv2-L computation by up to 64% and achieving 2.2x faster inference, while maintaining state-of-the-art accuracy on THUMOS-14 and ActivityNet-1.3. When scaled to InternVideoNext-L, our approach surpasses the previous state of the art at roughly half its computational cost. Moreover, the sparse formulation naturally supports auxiliary task tokens, which improves fine-grained assembly detection on ATTACH.

Chinese Translation

为了使具有数十亿参数的视觉变换器适应长视频理解，最近的方法冻结了主干网络，并训练轻量级卷积模块。尽管这种方法在参数高效训练方面有效，但现有的适配器并未减少推理时的计算，导致视频长度的可扩展性问题在很大程度上未得到解决。令牌选择可以通过修剪冗余令牌来降低注意力成本，但这破坏了卷积适配器所需的空间网格结构。这迫使进行昂贵的密集重构，从而抵消了大部分潜在的加速效果。我们通过引入原生稀疏二维卷积来解决这个问题，这一原语首次允许这些适配器直接且高效地在动态修剪的令牌集上操作。我们将这一原语集成到SV-TAD中，这是一个用于时间动作检测的适配器框架，减少了VideoMAEv2-L的计算量高达64%，并实现了2.2倍的推理速度，同时在THUMOS-14和ActivityNet-1.3上保持了最先进的准确性。当扩展到InternVideoNext-L时，我们的方法在计算成本大约为之前最先进技术的一半的情况下超越了其性能。此外，稀疏形式自然支持辅助任务令牌，从而改善了ATTACH上的细粒度组装检测。

View on arXiv Download PDF AI Translation

cs.CV / 91 / 2610.11608

S$^3$Geo: Structure-Semantic Synergistic Learning for Cross-View Geo-Localization

S$^3$Geo：用于跨视角地理定位的结构-语义协同学习

Mo, Ziqian, Zhang, Hill, Tan, Haosheng, Li, Ling, Wei, Jiaheng

Abstract

Cross-view geo-localization (CVGL) aims to estimate geographic locations by matching images captured from different viewpoints, such as drone and satellite views. Existing methods mainly rely on visual representations, but often fail to jointly model fine-grained structural correspondences and semantic priors, making them prone to confusion between visually similar but semantically different regions, and thus limiting robustness under large viewpoint variations. To address these challenges, we propose \textbf{S$^3$Geo}, a structure-semantic synergistic learning framework for cross-view matching. Specifically, we first introduce a Decoupled Query Pooling (DQP) module to extract a compact set of region-aware features from dense tokens, enabling explicit modeling of local structural patterns. We then design a query-level contrastive learning scheme with an optimal transport (OT)-based formulation to establish soft correspondences under cross-view spatial misalignment. Furthermore, we incorporate a Semantic Knowledge Distillation (SKD) strategy from a frozen CLIP teacher to transfer semantic priors and relational structures, thereby improving discrimination on hard negatives. By operating synergistically, the semantic priors provide robust contextual filtering, which guides the structural module to establish precise spatial alignments. Experiments on the University-1652 and SUES-200 datasets demonstrate that \textbf{S$^3$Geo} consistently outperforms state-of-the-art approaches without increasing inference complexity, validating the effectiveness of jointly modeling structural and semantic information for CVGL.

Chinese Translation

跨视角地理定位（CVGL）旨在通过匹配从不同视点（如无人机和卫星视图）捕获的图像来估计地理位置。现有方法主要依赖于视觉表示，但往往未能联合建模细粒度的结构对应关系和语义先验，导致在视觉上相似但语义上不同的区域之间容易混淆，从而在大视角变化下限制了鲁棒性。为了解决这些挑战，我们提出了 extbf{S$^3$Geo}，一种用于跨视角匹配的结构-语义协同学习框架。具体而言，我们首先引入了一个解耦查询池化（DQP）模块，从密集标记中提取一组紧凑的区域感知特征，使得局部结构模式的显式建模成为可能。然后，我们设计了一种基于最优传输（OT）形式的查询级对比学习方案，以在跨视角空间错位下建立软对应关系。此外，我们结合了来自冻结的CLIP教师的语义知识蒸馏（SKD）策略，以转移语义先验和关系结构，从而提高对难负样本的区分能力。通过协同操作，语义先验提供了鲁棒的上下文过滤，指导结构模块建立精确的空间对齐。在University-1652和SUES-200数据集上的实验表明， extbf{S$^3$Geo}在不增加推理复杂度的情况下，始终优于最先进的方法，验证了联合建模结构和语义信息在CVGL中的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 92 / 2610.11610

Beyond Report Imitation: Clinically Aware Multi-Image Ultrasound Report Generation from Visible Evidence

超越报告模仿：基于可见证据的临床感知多图像超声报告生成

Yang, Yuchen, Wang, Xin, Wang, Lufan, Pan, Yinghong, Feng, Yujuan, Yang, Yuqing

Abstract

Generating ultrasound reports from multiple images requires aggregating clinical evidence across views, yet archived key frames capture only part of the dynamic examination. Raw-report imitation is therefore misaligned with visual supervision: content that is clinically valid for the full examination may be unverifiable from the images available to a model. This gap creates a clinical behavior alignment problem. A model must preserve visible findings, avoid diagnostic reversals and unsupported completion, and not collapse into conservative templates. We propose CAMEO, a Clinically Aware Multi-image Evidence-grounded Orchestration framework for ultrasound report generation. Stage I learns ultrasound visual-language primitives; Stage II performs Cross-View Evidence Grounding by distilling trusted visible report points into multi-image QA and report-style supervision; and Stage III performs Clinically Aware Preference Alignment using clinical-error-oriented preference pairs. From USReport, we construct USReport-Distilled with 17,670 evidence-grounded paired-image training instances and USReport-Pref with 21,869 preference pairs; we additionally use 25,631 PubMedVision-US ultrasound instruction samples for domain adaptation and multi-image instruction tuning. On the primary USReport-Distilled benchmark, CAMEO improves over EchoVLM from 0.25 to 0.40 BLEU-1, 0.28 to 0.45 ROUGE-1, and 0.27 to 0.43 METEOR, while raising ClinicalScore from 55.02 to 74.20. These results underscore the value of evidence-grounded supervision, clinically aware alignment, and clinically structured evaluation for reliable ultrasound report generation.

Chinese Translation

从多幅图像生成超声报告需要在不同视图中聚合临床证据，然而归档的关键帧仅捕捉到动态检查的一部分。因此，原始报告的模仿与视觉监督之间存在不一致：对于完整检查而言临床有效的内容可能无法从模型可用的图像中验证。这一差距造成了临床行为对齐问题。模型必须保留可见发现，避免诊断反转和不支持的补全，且不能陷入保守模板。我们提出了CAMEO（Clinically Aware Multi-image Evidence-grounded Orchestration）框架，用于超声报告生成。第一阶段学习超声视觉语言原语；第二阶段通过将可信的可见报告点提炼为多图像问答和报告风格监督，执行跨视图证据对接；第三阶段使用面向临床错误的偏好对进行临床感知偏好对齐。我们从USReport构建了USReport-Distilled，包含17,670个基于证据的配对图像训练实例，以及USReport-Pref，包含21,869个偏好对；此外，我们还使用25,631个PubMedVision-US超声指令样本进行领域适应和多图像指令调优。在主要的USReport-Distilled基准上，CAMEO将EchoVLM的BLEU-1从0.25提高到0.40，ROUGE-1从0.28提高到0.45，METEOR从0.27提高到0.43，同时将ClinicalScore从55.02提升至74.20。这些结果强调了基于证据的监督、临床感知对齐和临床结构化评估在可靠超声报告生成中的重要性。

View on arXiv Download PDF AI Translation

cs.CV / 93 / 2610.11612

PointVGGT: Zero-Shot Multiview RGB-D Point Cloud Registration with Visual Geometry Foundation Priors

PointVGGT：基于视觉几何基础先验的零样本多视图RGB-D点云配准

Jiang, Haobo, Yu, Liang, Zheng, Jianmin

Abstract

This paper addresses multiview RGB-D point cloud registration, aiming to estimate global rigid poses for unordered RGB-D scans and align them in a metrically consistent coordinate frame. The conventional pairwise-then-global paradigm suffers from locally optimized pairwise registration, severe error propagation and high computational burden. In particular, existing methods typically treat RGB data as a mere auxiliary matching cue and overlook the holistic geometric priors (e.g., camera poses and 3D models) encoded across image sequences. This paper introduces PointVGGT, a zero-shot framework built upon a novel \emph{foundation-then-refinement} paradigm that systematically leverages visual geometry foundation models (e.g., VGGT) as the computational backbone for robust, training-free multiview RGB-D registration. In the foundation stage, we directly recover metrically consistent global poses (without any pairwise estimation) by grounding the scale-ambiguous pose predictions of the foundation model against metric depth observations. In the refinement stage, we introduce an efficient voxelized spatial hashing mechanism that exploits the globally coherent 3D reconstruction (induced by the foundation model) as a shared spatial anchor, enabling dense multiview correspondences in near-linear time. On top of this, an IRLS-based robust motion-only bundle adjustment is performed using a conjugate gradient solver to jointly minimize the correspondence and reprojection residuals for multiview pose refinement. Extensive experiments on indoor/object-centric/outdoor datasets verify the outstanding zero-shot registration accuracy and computational efficiency of our proposed method.

Chinese Translation

本文针对多视图RGB-D点云配准问题，旨在为无序的RGB-D扫描估计全局刚性姿态，并在一个度量一致的坐标框架中对其进行对齐。传统的先配对后全局的范式存在局部优化的配对注册、严重的误差传播和高计算负担等问题。特别是，现有方法通常将RGB数据视为仅仅是辅助匹配线索，而忽视了跨图像序列编码的整体几何先验（例如，相机姿态和3D模型）。本文提出了PointVGGT，一个基于新颖的“基础-再优化”范式的零样本框架，系统性地利用视觉几何基础模型（例如，VGGT）作为鲁棒、无训练的多视图RGB-D配准的计算支撑。在基础阶段，我们通过将基础模型的尺度模糊姿态预测与度量深度观测进行对比，直接恢复度量一致的全局姿态（无需任何配对估计）。在再优化阶段，我们引入了一种高效的体素化空间哈希机制，该机制利用基础模型诱导的全局一致3D重建作为共享空间锚点，从而实现近线性时间内的稠密多视图对应关系。在此基础上，采用基于IRLS的鲁棒运动仅束调整，使用共轭梯度求解器联合最小化多视图姿态再优化的对应关系和重投影残差。大量在室内/物体中心/户外数据集上的实验验证了我们所提出方法的卓越零样本配准精度和计算效率。

View on arXiv Download PDF AI Translation

cs.CV / 94 / 2610.11617

TAM: Task-Aware Memory Distillation for Efficient Spatiotemporal Prediction

TAM：任务感知记忆蒸馏用于高效时空预测

Li, Yuqi, Feng, Xiaoqin, Xu, Fan, Feng, Weilun, Yang, Chuanguang, Tian, Yingli, Wu, Hao

Abstract

Knowledge distillation enables efficient spatiotemporal prediction by transferring knowledge from an accurate teacher to a compact student. However, matching outputs or features independently for each sample leaves cross-sample predictive structure underused. Exploiting this structure requires representations and historical references that reflect the dynamics of each task. We propose TAM, a Task-Aware Memory Distillation framework that organizes a frozen teacher's knowledge into a bounded, retrievable history. Memory entries encode latent features, forecast changes, or flow residuals, while task-specific selection rules identify relevant historical references. The student either matches the teacher's similarity distribution over shared references or regresses observation-conditioned residual prototypes. These objectives complement supervised prediction and conventional distillation. The teacher, memory, and auxiliary adapters are used only during training, leaving student inference unchanged. We evaluate TAM on video prediction, weather forecasting, and traffic flow prediction across multiple teacher-student configurations. Averaged over four paired runs, adding TAM improves SSIM on all six video datasets and reduces MSE on five relative to the corresponding KD baselines. Mean paired MSE reductions reach 1.86% on KittiCaltech, 1.93% on WeatherBench with a gSTA teacher, and 1.01% on TaxiBJ. These results demonstrate the utility of historical teacher supervision across distinct forecasting tasks without additional student inference cost.

Chinese Translation

知识蒸馏通过将准确教师的知识转移到紧凑学生，从而实现高效的时空预测。然而，独立地为每个样本匹配输出或特征使得跨样本的预测结构未得到充分利用。利用这一结构需要反映每个任务动态的表示和历史参考。我们提出了TAM（Task-Aware Memory Distillation）框架，该框架将冻结教师的知识组织成一个有界的、可检索的历史。记忆条目编码潜在特征、预测变化或流动残差，而特定任务的选择规则则识别相关的历史参考。学生要么匹配教师在共享参考上的相似性分布，要么回归观察条件下的残差原型。这些目标补充了监督预测和传统蒸馏。教师、记忆和辅助适配器仅在训练期间使用，学生推理保持不变。我们在视频预测、天气预报和交通流预测等多个教师-学生配置上评估了TAM。在四次配对运行的平均结果中，添加TAM在所有六个视频数据集上提高了SSIM，并在五个数据集上相对于相应的KD基线降低了MSE。KittiCaltech上的平均配对MSE降低达到1.86%，WeatherBench（使用gSTA教师）上为1.93%，TaxiBJ上为1.01%。这些结果展示了在不同预测任务中历史教师监督的有效性，而没有增加学生推理的成本。

View on arXiv Download PDF AI Translation

cs.CV / 95 / 2610.11641

Revisiting Handcrafted Minutiae Detection: A Simple and Effective Open Source Baseline for Modern Fingerprint Workflows

重新审视手工特征点检测：现代指纹工作流程的简单有效开源基线

Cappelli, Raffaele

Abstract

Handcrafted minutiae detection algorithms remain fundamental to biometric science and forensic practice due to their full auditability, adherence to international standards, and operational independence from training datasets or GPU hardware. However, current open-source traditional baselines are severely outdated, relying almost exclusively on legacy C/C++ codebases that lack seamless integration with modern scientific software ecosystems. To bridge this gap, the present work introduces SBMEX (Skeleton-Based Minutiae EXtraction), a fast and deterministic minutiae detection method integrated into the open source \texttt{pyfing} package. SBMEX achieves high computational throughput by employing a dual Look-Up Table architecture that replaces runtime neighborhood scanning during Crossing Number computation and skeleton tracking. Additionally, it incorporates a continuous quality scoring framework driven by tracking path length, dual ridge-valley skeleton fusion, and spatial density decay. Rigorous evaluation on NIST SD302 datasets demonstrates that SBMEX delivers feature extraction accuracy comparable to or outperforming traditional open-source baselines without fine-tuning, while achieving a drastic reduction in minutiae detection latency relative to classical Crossing Number Python implementations.

Chinese Translation

手工特征点检测算法由于其完全可审计性、遵循国际标准以及在操作上独立于训练数据集或GPU硬件，仍然是生物识别科学和法医学实践的基础。然而，目前的开源传统基线严重过时，几乎完全依赖于缺乏与现代科学软件生态系统无缝集成的遗留C/C++代码库。为了解决这一问题，本研究提出了SBMEX（基于骨架的特征点提取），这是一种快速且确定性的特征点检测方法，集成于开源的 exttt{pyfing}包中。SBMEX通过采用双查找表架构，实现了高计算吞吐量，替代了在交叉数计算和骨架跟踪过程中运行时的邻域扫描。此外，它还结合了一个基于跟踪路径长度、双脊谷骨架融合和空间密度衰减的连续质量评分框架。在NIST SD302数据集上的严格评估表明，SBMEX在特征提取准确性上可与传统开源基线相媲美或超越，且无需微调，同时在特征点检测延迟方面相较于经典的交叉数Python实现大幅减少。

View on arXiv Download PDF AI Translation

cs.CV / 96 / 2610.11656

DisFace3DNet: Explainable Facial Attractiveness Prediction via 3D Component Disentanglement

DisFace3DNet：通过3D组件解耦实现可解释的面部吸引力预测

Rao, Fenggui, Luximon, Yan, Zhang, Jie

Abstract

Facial attractiveness prediction usually assigns one overall rating, leaving the roles of shape, appearance, and viewing conditions implicit. We propose DisFace3DNet, which uses 3D component disentanglement to learn seven component reference scores from overall ratings with auxiliary weak semantic supervision, without human-labeled component targets. Designated 3D representations and image cues feed jointly learned routes for identity, skin, hair, light, background, expression, and pose. A constrained fit then combines five static and two signed dynamic scores into the overall rating, exposing each component's numerical contribution and supporting component-specific comparisons across images. On SCUT-FBP5500, DisFace3DNet achieves a Pearson correlation of $0.8904\pm0.0063$ (mean $\pm$ standard deviation across five folds) with average human ratings; its component terms reconstruct every held-out prediction to numerical precision. Skin, hair, and facial shape account for the largest component-wise prediction variation. Human evaluation supports the score directions for facial shape, skin, and hair; expression agreement is weaker. DisFace3DNet thus connects overall prediction to quantitative analysis of the facial and contextual cues entering each estimate.

Chinese Translation

面部吸引力预测通常只给出一个总体评分，而形状、外观和观察条件的作用则隐含在其中。我们提出了DisFace3DNet，该模型利用3D组件解耦从总体评分中学习七个组件参考分数，并辅以弱语义监督，而无需人工标注的组件目标。指定的3D表示和图像线索共同为身份、皮肤、头发、光照、背景、表情和姿势提供学习路径。然后，通过受限拟合将五个静态分数和两个带符号的动态分数组合成总体评分，揭示每个组件的数值贡献，并支持跨图像的组件特定比较。在SCUT-FBP5500数据集上，DisFace3DNet与平均人类评分的皮尔逊相关系数达到$0.8904 ext{±}0.0063$（五折交叉验证的均值±标准差）；其组件项能够以数值精度重构每个保留预测。皮肤、头发和面部形状占据了组件预测变化的最大比例。人类评估支持面部形状、皮肤和头发的评分方向，但表情一致性较弱。因此，DisFace3DNet将总体预测与进入每个估计的面部和上下文线索的定量分析相连接。

View on arXiv Download PDF AI Translation

cs.CV / 97 / 2610.11669

Perceptually Grounded and Semantics-Aware Evaluation for Holistic Co-Speech Gesture Generation

感知基础与语义意识的整体共语手势生成评估

Milkin, Nick, Liu, Lanmiao, Ghaleb, Esam, Ozyurek, Asli, Yumak, Zerrin

Abstract

Holistic and semantics-aware co-speech gesture generation has advanced rapidly, yet evaluation remains behind: objective metrics do not consistently reflect human perception, and semantic appropriateness remains difficult to quantify. We present a perceptually grounded and semantics-aware benchmark that combines standardized model comparison, human-centered metric validation, and fine-grained semantic evaluation. We first curate a list of 13 objective metrics covering different aspects, including distributional similarity, geometric fidelity, kinematic quality, cross-modal synchrony, and semantic appropriateness. For the semantic-appropriateness category, we propose a new metric, Semantic Gesture Preservation (SGP), which measures how far semantic gestures in the ground truth are preserved in the generated gestures. For this, we augment the BEAT2 dataset's annotations using a multi-modal LLM. We then conduct a perceptual study where 101 participants score generated gestures among five dimensions, including human-likeness, motion diversity, absence of animation errors, speech timing and content match. We systematically analyze objective metric--subjective score correlations. Unlike Semantic Score (SC), which shows no significant association with the evaluated perceptual dimensions, SGP is selectively aligned with speech-aware human judgments. We construct five target-specific composite metrics aligned with the subjective dimensions. These composites improve perceptual alignment across all five dimensions, with the largest gains for absence of animation errors and content match, indicating that complementary objective signals can better approximate human judgments than individual metrics alone. Overall, our results show that objective metrics require validation against subjective evaluations.

Chinese Translation

整体且具语义意识的共语手势生成快速发展，但评估仍滞后：客观指标未能始终如一地反映人类感知，语义适当性仍难以量化。我们提出了一种感知基础与语义意识的基准，结合了标准化模型比较、人本中心的指标验证和细粒度的语义评估。我们首先整理了一份涵盖不同方面的13个客观指标的列表，包括分布相似性、几何保真度、运动质量、跨模态同步和语义适当性。在语义适当性类别中，我们提出了一种新的指标——语义手势保留（Semantic Gesture Preservation, SGP），用于衡量生成的手势在多大程度上保留了真实标签中的语义手势。为此，我们利用多模态大语言模型（LLM）增强了BEAT2数据集的注释。随后，我们进行了一项感知研究，101名参与者在五个维度上对生成的手势进行评分，包括人类相似性、运动多样性、动画错误缺失、语音时机和内容匹配。我们系统地分析了客观指标与主观评分之间的相关性。与语义评分（Semantic Score, SC）不同，后者与评估的感知维度之间没有显著关联，而SGP则与语音意识的人类判断选择性一致。我们构建了五个与主观维度对齐的特定目标复合指标。这些复合指标在所有五个维度上改善了感知对齐，其中在动画错误缺失和内容匹配方面的提升最大，表明互补的客观信号能够比单一指标更好地接近人类判断。总体而言，我们的结果表明，客观指标需要针对主观评估进行验证。

View on arXiv Download PDF AI Translation

cs.CV / 98 / 2610.11674

VESSI - VLM-Enhanced Support for Surveillance and Investigations

VESSI - 基于视觉语言模型的监控与调查增强支持

Cavasin, Saverio, Tedeschi, Pietro, Tamiazzo, Mattia, Brighente, Alessandro, Milani, Simone, Conti, Mauro

Abstract

Automated video surveillance analysis has become a critical component of intelligence infrastructures and Law Enforcement agencies. Traditional systems lack the semantic module for comprehensive situational awareness and forensic tasks, limiting their ability to interpret events meaningfully or support post-incident investigations. This slows operational insight and increases the burden on human analysts. Recent advances in Vision-Language Models (VLMs) offer promising pathways to bridge this gap. To address this, we propose VLM-Enhanced Support for Surveillance and Investigations (VESSI), a VLM-based framework designed to enhance automated video surveillance analysis through prompt-driven interrogation of video sequences where salient visual features are converted into textual descriptions. We test our framework with four state-of-the-art models. Since most datasets for this task are unlabeled, we also propose the Composite Model Utility Score (CMUS) to assess VLM performance. Experimental results show that our solution substantially improves the analysis capabilities of human operators and enhances the flexibility of automated surveillance systems. In our evaluation, the most reliable model flagged potentially relevant activity in more than 66% of the videos while reducing review time by more than 85%, offering a practical balance between selectivity and efficiency. The model ordering produced by the reference-free CMUS evaluation was reproduced by the normal-video CMUS evaluation and matched the false-positive-rate ordering obtained from 5,909 manually referenced frames. This agreement supports the operational use of the score within the evaluated setting.

Chinese Translation

自动化视频监控分析已成为情报基础设施和执法机构的关键组成部分。传统系统缺乏语义模块，无法提供全面的态势感知和取证任务，限制了其对事件的有意义解读或对事后调查的支持。这减缓了操作洞察的速度，并增加了人类分析师的负担。最近在视觉语言模型（VLMs）方面的进展为弥补这一差距提供了有希望的途径。为此，我们提出了基于VLM的监控与调查增强支持框架（VESSI），旨在通过对视频序列的提示驱动询问，将显著的视觉特征转换为文本描述，从而增强自动化视频监控分析。我们使用四个最先进的模型对我们的框架进行了测试。由于大多数用于此任务的数据集未标记，我们还提出了复合模型效用评分（CMUS）来评估VLM的性能。实验结果表明，我们的解决方案显著提升了人类操作员的分析能力，并增强了自动化监控系统的灵活性。在我们的评估中，最可靠的模型在超过66%的视频中标记了潜在相关活动，同时将审查时间减少了超过85%，在选择性和效率之间提供了实用的平衡。参考无CMUS评估产生的模型排序被正常视频CMUS评估重现，并与从5,909个手动参考帧获得的假阳性率排序相匹配。这一一致性支持了在评估环境中使用该评分的操作性。

View on arXiv Download PDF AI Translation

cs.CV / 99 / 2610.11685

HI3D 3.0 (Twinkle3D): Object-specific 3D Asset Generation with High Resolution

HI3D 3.0 (Twinkle3D): 面向对象特定的高分辨率3D资产生成

Li, Ziying, Zhao, Shengchu, He, Huiang, Chen, Yiyang, Huang, Jianwen, Li, Bailin, Li, Changhao, Li, Jianhui, Li, Jie, Liu, Ruiyang, Luo, Yibo, Sun, Tengjiao, Tang, Pei, Wang, Shiwen, Wu, Jiaqi, Wu, Kang, Yang, Kaiqiao, Yang, Zherui, Zhang, Hu, Zhao, Xuezhi, Zheng, Xinhe, Li, Yukun, Zheng, Heliang, Jia, Rongfei

Abstract

Image-to-3D generation has become increasingly capable of producing objects that closely resemble the input image, and an outstanding challenge is to reproduce the depicted object itself, including the specific geometry that defines it. Inscriptions, brand marks, and repeated structures are frequently distorted or lost, despite being critical to object identity. We present Hi3D 3.0, an image-to-3D generation system targeting object-specific fidelity, with Twinkle3D as its geometry model for generating watertight triangle meshes at $2048^{3}$ resolution. Twinkle3D advances high-fidelity geometry generation along four dimensions. First, while O-Voxel/FaithC offers high representational precision, it often suffers from poor surface quality and non-watertight geometry. We address both issues while retaining its $2048^{3}$-level precision. Second, we scale diffusion generation to sequences of up to 300K geometric tokens through a redesigned DiT architecture and large-scale distributed training optimizations, reducing training time per step from approximately ten minutes to ten seconds. Third, subsequent refinement cannot fully compensate for errors introduced during initial generation; we therefore strengthen both global shape and local detail in the initial generation stage, and the resulting single-stage model surpasses prior two-stage pipelines with $512^{3}$ refinement. Finally, we introduce a fine-grained image-3D cross-modal interaction mechanism that strengthens correspondence between visual evidence and geometric tokens, improving the recovery of object-specific structures. We evaluate geometric fidelity using alignment metrics derived from silhouettes and normal fields. Hi3D 3.0 outperforms four commercial systems across all reported metrics, recovering 82.1% of inscribed characters at 98.2% precision, compared with 21.7% recall for the strongest competitor.

Chinese Translation

图像到3D生成技术在生成与输入图像高度相似的对象方面变得越来越强大，而一个突出的挑战是重现所描绘的对象本身，包括定义其特定几何形状的细节。铭文、品牌标记和重复结构常常会失真或丢失，尽管这些对于对象的身份至关重要。我们提出了Hi3D 3.0，这是一个针对对象特定保真度的图像到3D生成系统，以Twinkle3D作为其几何模型，用于生成2048³分辨率的密闭三角网格。Twinkle3D在四个维度上推动了高保真几何生成。首先，尽管O-Voxel/FaithC提供了高表示精度，但它通常面临表面质量差和非密闭几何的问题。我们在保留2048³级精度的同时解决了这两个问题。其次，我们通过重新设计的DiT架构和大规模分布式训练优化，将扩散生成的规模扩展到多达300K的几何标记，将每步的训练时间从大约十分钟减少到十秒。第三，后续的精细化无法完全弥补初始生成过程中引入的错误；因此，我们在初始生成阶段增强了全局形状和局部细节，最终得到的单阶段模型在512³精细化方面超越了之前的两阶段管道。最后，我们引入了一种细粒度的图像-3D跨模态交互机制，增强了视觉证据与几何标记之间的对应关系，提高了对象特定结构的恢复。我们使用从轮廓和法线场导出的对齐度量来评估几何保真度。Hi3D 3.0在所有报告的指标上超越了四个商业系统，以98.2%的精度恢复了82.1%的铭文字符，而最强竞争者的召回率仅为21.7%。

View on arXiv Download PDF AI Translation

cs.CV / 100 / 2610.11736

Towards Unified Evaluation of Prompt Enhancers for Video Generation

朝向视频生成提示增强器的统一评估

Shao, Yawen, Zhu, Yubo, Dai, Ziyun, Fang, Zixun, Zhu, Kai, Jiang, Zeyinzi, Ai, Yufeng, Sun, Siyang, Xue, Haolan, Shang, Yu, Bao, Yuxiang, Bi, Zoubin, Luo, Jingming, Xiao, Jie, Mao, Chaojie, Kan, Zhehan, Luo, Hongchen, Liu, Yu, Zhong, Sheng, Tong, Wei, Fu, Xueyang, Cao, Yang, Zhai, Wei, Zha, Zheng-Jun

Abstract

Modern video generators can realize increasingly complex visual narratives, positioning the prompt enhancer (PE) as a critical bridge from concise user instructions and multimodal references to structured cinematic plans. However, existing PE evaluation relies on rendered videos, imposing substantial computational and human costs, slowing PE training and iteration, and conflating PE quality with downstream generator behavior. To address this gap, we introduce PEBench, the first unified benchmark for direct PE evaluation across text-to-video, image-to-video, and reference-to-video prompt enhancement. It comprises 1,100 expert-verified cases and 1,005 visual assets, spanning 35 fine-grained tasks with diverse temporal, cinematic, audiovisual, and multi-reference requirements. In addition, we develop PEBench evaluation, an evidence-grounded framework that combines modality-aware fact extraction with rubric-based assessment across 24 criteria. Our systematic evaluation of representative open- and closed-source PE methods reveals an emerging shift from fine-grained descriptive expansion toward intent-preserving cinematic planning, while the caption-reconstruction and forward-refinement methods show complementary strengths in cinematic coverage and semantic fidelity or internal coherence, respectively. Human validation shows that PEBench scores align closely with expert judgments of enhanced prompts and downstream videos from Wan3.0 and MiniMax-H3, indicating that prompt-level evaluation reliably reflects downstream utility.

Chinese Translation

现代视频生成器能够实现日益复杂的视觉叙事，使得提示增强器（Prompt Enhancer, PE）成为从简洁用户指令和多模态参考到结构化电影计划的重要桥梁。然而，现有的PE评估依赖于渲染视频，带来了显著的计算和人力成本，减缓了PE的训练和迭代，并将PE质量与下游生成器行为混为一谈。为了解决这一问题，我们提出了PEBench，这是第一个针对文本到视频、图像到视频和参考到视频的提示增强直接评估的统一基准。它包含1,100个经过专家验证的案例和1,005个视觉素材，涵盖35个细粒度任务，具有多样的时间、电影、视听和多参考要求。此外，我们开发了PEBench评估，这是一个基于证据的框架，结合了对模态的事实提取与基于评分标准的评估，涵盖24个标准。我们对代表性开源和闭源PE方法的系统评估揭示了从细粒度描述扩展向意图保留的电影规划的转变，而字幕重建和前向精炼方法在电影覆盖和语义保真度或内部一致性方面各具互补优势。人类验证显示，PEBench得分与专家对增强提示和来自Wan3.0及MiniMax-H3的下游视频的判断高度一致，表明提示级评估可靠地反映了下游效用。

View on arXiv Download PDF AI Translation

cs.CV / 101 / 2610.11746

Streaming-Aware Diffusion for Real-Time Video Super-Resolution via Cross-Step Attention

面向流媒体的扩散模型在实时视频超分辨率中的应用：跨步注意力机制

Partaourides, Harris, Chatzis, Sotirios

Abstract

Real-time video super-resolution requires high spatio-temporal fidelity under strict latency constraints, challenging diffusion models due to their iterative sampling cost and limited temporal coordination. We propose a streaming-aware framework that adapts pretrained single-image latent diffusion models for efficient video super-resolution (VSR) by exploiting the sequential structure of video streams. Our Cross-Step Attention mechanism reuses intermediate denoising features across adjacent frames and diffusion steps, enabling temporal information exchange without explicit temporal modeling. We further introduce Trajectory-Coupled Diffusion Scheduling, which aligns adjacent diffusion states and provides cleaner intermediate representations for cross-step conditioning, improving temporal coherence. These components are integrated into a streaming inference pipeline that incrementally propagates latent states across frames, reducing the effective computational complexity from $O(N \cdot S)$ to $O(N + S)$ for $N$ frames and $S$ diffusion steps. Experiments on REDS4 and YouHQ40-Test demonstrate improved perceptual quality and temporal realism while maintaining frame-wise stability. Our method achieves over 40 FPS at $512 \times 512$ resolution after cold start, enabling real-time VSR without explicit temporal modeling.

Chinese Translation

实时视频超分辨率在严格的延迟限制下需要高时空保真度，这对扩散模型提出了挑战，因为它们的迭代采样成本和有限的时间协调性。我们提出了一种面向流媒体的框架，通过利用视频流的顺序结构，将预训练的单图像潜在扩散模型适配为高效的视频超分辨率（VSR）。我们的跨步注意力机制在相邻帧和扩散步骤之间重用中间去噪特征，实现了时间信息的交换，而无需显式的时间建模。我们进一步引入了轨迹耦合扩散调度，旨在对齐相邻的扩散状态，并为跨步条件提供更清晰的中间表示，从而改善时间一致性。这些组件被整合到一个流媒体推理管道中，逐帧增量传播潜在状态，将对于 $N$ 帧和 $S$ 次扩散步骤的有效计算复杂度从 $O(N imes S)$ 降低到 $O(N + S)$。在 REDS4 和 YouHQ40-Test 上的实验表明，在保持逐帧稳定性的同时，感知质量和时间真实感得到了改善。我们的方法在冷启动后以 $512 imes 512$ 分辨率实现了超过 40 FPS 的帧率，使得实时视频超分辨率成为可能，而无需显式的时间建模。

View on arXiv Download PDF AI Translation

cs.CV / 102 / 2610.11751

Dino Forcing Flow Models: Do not denoise what you can predict

恐龙强制流模型：不去去噪你可以预测的内容

Ghosh, Arijit, Degeorge, Lucas, Couairon, Paul, Efros, Alexei A, Kalogeiton, Vicky, Picard, David

Abstract

Co-denoising pretrained representations such as DINO can substantially improve the training speed and quality of flow matching models, but it introduces a second denoising trajectory and requires carefully designed schedules. We propose a simpler alternative: predict the pretrained representation directly, then condition the model on its own prediction. This removes the need for a second ODE and any representation-specific denoising schedules, while retaining the benefits of representation guidance. Our approach converges substantially faster and achieves better generation quality as measured by FID score. On ImageNet, it outperforms the state of the art in latent space at 2x fewer epochs than prior methods; in pixel space, it improves FID over comparable prior methods by more than 20%. These results support a simple principle: do not denoise what you can predict. Our code is openly available at https://github.com/arijit-hub/dino_forcing.

Chinese Translation

共同去噪预训练表示（如 DINO）可以显著提高流匹配模型的训练速度和质量，但这引入了第二个去噪轨迹，并需要精心设计的调度。我们提出了一种更简单的替代方案：直接预测预训练表示，然后根据其自身的预测对模型进行条件化。这消除了对第二个常微分方程（ODE）和任何特定于表示的去噪调度的需求，同时保留了表示指导的好处。我们的方法收敛速度显著更快，并且在生成质量上（以 FID 分数衡量）表现更佳。在 ImageNet 上，它在潜在空间中以比先前方法少 2 倍的训练轮数超越了当前最优性能；在像素空间中，它比可比较的先前方法提高了超过 20% 的 FID。这些结果支持一个简单的原则：不去去噪你可以预测的内容。我们的代码可以在 https://github.com/arijit-hub/dino_forcing 上公开获取。

View on arXiv Download PDF AI Translation

cs.CV / 103 / 2610.11756

Memory Forcing: Attendable Mid-Horizon History for Streaming Video Generation

记忆强制：可关注的中期历史用于流媒体视频生成

Zhang, Jiaming, Wang, Xinyu, Shi, Huafeng, Wu, Gangshan, Wang, Limin

Abstract

Autoregressive video diffusion enables causal video streaming without a bidirectional pass over the full clip, but existing few-step systems usually retain only the opening and most recent frames in a fixed-size KV cache. Once an event leaves this window, later frames can no longer attend to it, a failure we term mid-horizon forgetting. We present Memory Forcing, a few-step streaming method that preserves this missing history without increasing the cache size. Its Archive \& Working Banks partition the cache into sink, archive, and working regions, retaining diverse intermediate events alongside recent motion under fixed memory. Because absolute temporal indices drift outside the training range, Bank-aware RoPE reassigns indices at attention time so each bank remains distinguishable. At 1.3B, Memory Forcing leads on longer clips, shows the smallest drop from 5s to 60s among methods reporting all four lengths, and preserves subjects and scenes through leave-and-return. The same design scales to Wan2.2 5B, producing more physically plausible, realistic, and dynamic videos and, to our knowledge, the first public 5B model on this forcing line.

Chinese Translation

自回归视频扩散使得因果视频流媒体生成无需对完整剪辑进行双向传递，但现有的几步系统通常仅在固定大小的键值缓存中保留开头和最近的帧。一旦事件超出此窗口，后续帧将无法再关注到它，这种失败我们称之为中期遗忘。我们提出了记忆强制（Memory Forcing），一种几步流媒体方法，在不增加缓存大小的情况下保留这些缺失的历史。其归档与工作银行（Archive & Working Banks）将缓存划分为汇聚、归档和工作区域，在固定内存下保留多样的中间事件以及最近的运动。由于绝对时间索引漂移超出训练范围，银行感知的相对位置编码（Bank-aware RoPE）在注意力时间重新分配索引，使得每个银行保持可区分性。在1.3B的模型中，记忆强制在较长剪辑上表现优越，在报告所有四种长度的方法中，从5秒到60秒的下降幅度最小，并通过离开与返回的方式保留了主题和场景。相同的设计可扩展至Wan2.2 5B，生成更具物理 plausibility、现实感和动态感的视频，并且据我们所知，这是在这一强制线上的第一个公开5B模型。

View on arXiv Download PDF AI Translation

cs.CV / 104 / 2610.11770

From Video Clips to Creation Trajectory: Sora100K for AI-Native Video Creation

从视频片段到创作轨迹：Sora100K用于AI原生视频创作

Yang, Sicong, Yang, Ruihuan, Lu, Jian, Yuan, Jianfei, Cun, Xiaodong, Bi, Xiuli

Abstract

AI-Native video creation is shifting from isolated video clips toward iterative video creation workflows. However, existing datasets remain largely video clips, representing video generation and editing as separate tasks rather than connected stages of a video creation workflow. In this paper, we introduce Sora100K, a dataset that represents the AI-Native video creation workflow as a structured video creation trajectory. Specifically, we first identify video creation trajectories and decompose them into three subsets according to their structural roles: text-to-video generation records as roots, single-turn video editing records as editing edges, and multi-turn video editing records as complete trajectories. Then, we use a VLM to assign semantic annotations for generation roots and editing-operation annotations for editing edges. A strict construction pipeline further reconstructs source-to-edit lineage, editing order, and intermediate video states while ensuring data quality. Finally, we perform lightweight adaptation on LTX-2 models to assess the supervision value of Sora100K. The results show improvements in visual quality, multi-shot generation, and cross-shot consistency, while successive-turn evaluation reveals that following multi-turn editing instructions remains challenging. Sora100K establishes a new data foundation for AI-Native video creation beyond isolated video clips and toward structured video creation trajectory. The dataset and supplementary materials are publicly available at https://huggingface.co/datasets/ysicong/Sora100K.

Chinese Translation

AI原生视频创作正从孤立的视频片段转向迭代的视频创作工作流程。然而，现有的数据集仍然主要是视频片段，将视频生成和编辑视为分开的任务，而不是视频创作工作流程的连接阶段。本文介绍了Sora100K，一个将AI原生视频创作工作流程表示为结构化视频创作轨迹的数据集。具体而言，我们首先识别视频创作轨迹，并根据其结构角色将其分解为三个子集：作为根的文本到视频生成记录、作为编辑边的单轮视频编辑记录，以及作为完整轨迹的多轮视频编辑记录。然后，我们使用视觉语言模型（VLM）为生成根分配语义注释，为编辑边分配编辑操作注释。严格的构建流程进一步重建源到编辑的传承、编辑顺序和中间视频状态，同时确保数据质量。最后，我们对LTX-2模型进行轻量级适配，以评估Sora100K的监督价值。结果显示，在视觉质量、多镜头生成和跨镜头一致性方面有所改善，而连续轮次评估则表明，遵循多轮编辑指令仍然具有挑战性。Sora100K为超越孤立视频片段、迈向结构化视频创作轨迹的AI原生视频创作建立了新的数据基础。数据集及补充材料已公开发布在https://huggingface.co/datasets/ysicong/Sora100K。

View on arXiv Download PDF AI Translation

cs.CV / 105 / 2610.11781

Skill-V: Verifiable Self-Evolving Skill Library for Interactive Agents

Skill-V：可验证的自我演化技能库用于交互代理

Ma, Jie, Qian, Zhipeng, Ma, Yufei, Liang, Zihan, Ji, Jiayi, Cai, Qingpeng, Chen, Ben, Jiang, Peng, Sun, Xiaoshuai

Abstract

Interactive agents can turn experience into reusable skills, yet existing self-evolving skill libraries primarily improve by accumulating new knowledge. Failures may lead to new skills, while previously stored skills are less often revisited as new evidence arrives. However, growth alone does not ensure reliability, as a retrieved skill may be inapplicable under the current task conditions, and an existing skill may encode a mis-specified operational boundary. Reliable skill evolution therefore requires not only adding knowledge, but also testing and revising what is already stored. We introduce Skill-V, a verifiable self-evolving skill library. To make stored knowledge testable, we propose representing skills as versioned, falsifiable contracts that link semantic intent to observable behavioral criteria. We use environment outcomes to drive library evolution. Specifically, task failures motivate skill addition, while disagreements between contract evaluations and task outcomes guide revisions to existing skill boundaries. To validate these revisions, we require them to preserve protected semantic constraints and satisfy non-regression criteria for rubric-outcome metrics on historical replay evidence. Finally, we employ an applicability-aware filter to exclude candidates judged confidently inapplicable to the current task. Across ALFWorld and WebShop, Skill-V achieves success rates of 95.3% and 85.9%, respectively, while maintaining a more compact skill library than growth-oriented baselines. Applicability-aware filtering reduces incorrect skill invocations, and outcome-grounded revisions correct mis-specified skill boundaries without degrading performance on previously observed evidence. These results show that reliable skill evolution requires more than accumulating experience: the library must learn which knowledge to retain, when to revise it, and when it should be applied.

Chinese Translation

交互代理能够将经验转化为可重用的技能，但现有的自我演化技能库主要通过积累新知识来提升。失败可能导致新技能的产生，而随着新证据的到来，之前存储的技能则较少被重新审视。然而，仅仅增长并不能确保可靠性，因为检索到的技能在当前任务条件下可能不适用，而现有技能可能编码了错误指定的操作边界。因此，可靠的技能演化不仅需要增加知识，还需要测试和修订已存储的内容。我们提出了Skill-V，一个可验证的自我演化技能库。为了使存储的知识可测试，我们建议将技能表示为版本化的、可证伪的合同，这些合同将语义意图与可观察的行为标准联系起来。我们使用环境结果来驱动库的演化。具体而言，任务失败促使技能的增加，而合同评估与任务结果之间的分歧则指导对现有技能边界的修订。为了验证这些修订，我们要求它们保持受保护的语义约束，并满足历史重放证据上评分结果指标的非回归标准。最后，我们采用了一种关注适用性的过滤器，以排除被判断为对当前任务自信不适用的候选技能。在ALFWorld和WebShop中，Skill-V分别实现了95.3%和85.9%的成功率，同时保持了比以增长为导向的基线更紧凑的技能库。关注适用性的过滤减少了错误的技能调用，而基于结果的修订则在不降低对先前观察证据的性能的情况下纠正了错误指定的技能边界。这些结果表明，可靠的技能演化不仅仅依赖于经验的积累：库必须学习保留哪些知识、何时修订这些知识以及何时应用它们。

View on arXiv Download PDF AI Translation

cs.CV / 106 / 2610.11791

Phase-aware video generation for physics-grounded dynamics and interactions

基于相位的物理驱动动态与交互视频生成

Ou, Jingfeng, Wang, Kun, Zhao, Rui, Guan, Jingwei, Wang, Limin, Dong, Chao, Zeng, Xingyu

Abstract

Generating physically plausible videos for solid-gas dynamics is challenging as different phases exhibit distinct dynamics yet remain coupled through physical interactions. We present PAVG, a Phase-Aware Video Generator for solid-gas dynamics and interactions. It employs a dual-branch architecture to explicitly model the distinct dynamics of solids and gases, while spatiotemporal cross-attention captures their physical interactions. This design enables PAVG to preserve phasespecific motion characteristics while producing physically consistent responses across phases. To facilitate this task, we further construct a simulation corpus comprising over 700K physical trajectories across diverse solid, gas, and solid-gas interaction scenarios. Extensive evaluations demonstrate that our PAVG produces videos with improved motion adherence, physical plausibility, and visual quality compared with existing approaches.

Chinese Translation

为固体-气体动态生成物理上合理的视频是一项挑战，因为不同相位表现出不同的动态特征，但又通过物理交互保持耦合。我们提出了PAVG（Phase-Aware Video Generator），一种用于固体-气体动态与交互的相位感知视频生成器。它采用双分支架构，明确建模固体和气体的不同动态，同时时空交叉注意力捕捉它们的物理交互。这一设计使得PAVG能够在生成物理一致的相位响应的同时，保持相位特定的运动特征。为促进这一任务，我们进一步构建了一个模拟语料库，包含超过70万条物理轨迹，涵盖多种固体、气体和固体-气体交互场景。广泛的评估表明，与现有方法相比，我们的PAVG生成的视频在运动遵循性、物理合理性和视觉质量上都有所提升。

View on arXiv Download PDF AI Translation

cs.CV / 107 / 2610.11810

Seek-and-View Reasoning for Multi-View Spatial Understanding

多视角空间理解的寻视推理

Chen, Qixiang, Zhang, Cheng, Ke, Fucai, Fu, Chi-Wing, Cai, Jianfei, Ye, Jingwen

Abstract

Existing approaches to multi-view spatial reasoning operate largely on sparse input views. Vision-language models (VLMs) are thus restricted to understand a scene and infer spatial relations within these fixed views, leading to fragile cross-view alignment and geometry-to-language bottleneck. To address these issues, we formulate a novel Seek-and-View reasoning approach to find implicit cross-view spatial evidence by locating a question-relevant view to support the spatial reasoning. To realize this approach, we propose Vantage, a training-free model-agnostic reasoning framework that pairs a VLM with a 3D foundation model: a viewpoint-grounded reasoning stage for question analysis and view planning, followed by a geometry-grounded evidence augmentation stage to effectively synthesize and incorporate visual evidence into the final reasoning. Comprehensive experiments on six VLMs demonstrate consistent improvements on five benchmarks without fine-tuning. Overall, by revealing spatial evidence through view-grounded reasoning, Vantage can largely reduce reliance on language-based cross-view alignment and improve multi-view spatial understanding. Our code is available at https://github.com/q1xiangchen/Vantage.

Chinese Translation

现有的多视角空间推理方法主要依赖于稀疏的输入视图。因此，视觉-语言模型（VLMs）在理解场景和推断这些固定视图中的空间关系时受到限制，导致跨视图对齐脆弱以及几何到语言的瓶颈。为了解决这些问题，我们提出了一种新颖的寻视推理方法，通过定位与问题相关的视图来寻找隐含的跨视图空间证据，以支持空间推理。为实现这一方法，我们提出了Vantage，一个无训练的模型无关推理框架，它将VLM与3D基础模型相结合：一个基于视点的推理阶段用于问题分析和视图规划，随后是一个基于几何的证据增强阶段，有效地合成和整合视觉证据到最终推理中。在六个VLMs上的综合实验表明，在五个基准测试中均实现了一致的改进，而无需微调。总体而言，通过视点基础的推理揭示空间证据，Vantage可以大大减少对基于语言的跨视图对齐的依赖，并改善多视角空间理解。我们的代码可在 https://github.com/q1xiangchen/Vantage 获取。

View on arXiv Download PDF AI Translation

cs.CV / 108 / 2610.11815

Fast Pose Tracking of Rigid Objects with Compact Pose Graph Optimization

基于紧凑姿态图优化的刚性物体快速姿态跟踪

Zhang, Xiaojie, Fischer, Tom, Larsson, Viktor, Ilg, Eddy

Abstract

Tracking a novel object's 6D pose over long horizons currently requires either expensive onboarding or a reconstruction maintained throughout the sequence. This makes current trackers impractical for robotic manipulation and augmented reality, which need trackers that are ready to use and run in real time. We show that a lightweight tracking module can be applied on top of a wide range of correspondence estimation methods to keep drifts bounded while maintaining fast runtime. Our key idea is to avoid point-based optimization in the pose graph and operate only on relative pose constraints, which we weight by a derived uncertainty from the geometric alignment. This makes optimization independent of the number of correspondences while avoiding the direct inclusion of noisy point measurements, leading to fast and robust long-term tracking. Across four real-world benchmarks, our approach achieves tracking accuracy comparable to reconstruction-based trackers with a fraction of the optimization cost. Overall, these results suggest that a compact and reliable pose graph optimization can provide long-horizon consistency at substantially lower computational cost.

Chinese Translation

在长时间范围内跟踪新物体的6D姿态目前需要昂贵的初始化或在整个序列中保持的重建。这使得当前的跟踪器在机器人操作和增强现实中不够实用，因为这些应用需要能够实时运行的即用型跟踪器。我们展示了一种轻量级跟踪模块可以应用于多种对应估计方法之上，以保持漂移在可控范围内，同时保持快速的运行时间。我们的关键思想是避免在姿态图中进行基于点的优化，仅在相对姿态约束上进行操作，并根据几何对齐推导的不确定性对其进行加权。这使得优化与对应点的数量无关，同时避免了直接包含噪声点测量，从而实现快速且稳健的长期跟踪。在四个真实世界基准测试中，我们的方法在优化成本的极小部分下实现了与基于重建的跟踪器相当的跟踪精度。总体而言，这些结果表明，紧凑且可靠的姿态图优化能够以显著更低的计算成本提供长时间范围的一致性。

View on arXiv Download PDF AI Translation

cs.CV / 109 / 2610.11818

From Pixels to Structure: Lightweight Vision-Language Models for Document OCR and Structured JSON Extraction

从像素到结构：轻量级视觉-语言模型用于文档OCR和结构化JSON提取

Bir, Uddipan Basu, Christlein, Vincent, Maier, Andreas, Zinnen, Mathias

Abstract

While massive, closed-source Vision-Language Models (VLMs) set strong benchmarks for document understanding, their dependence on commercial APIs limits adoption in institutional archives due to data autonomy concerns, recurring costs, and the environmental footprint of hyperscale computing. This is especially acute in heritage digitization, where documents include historical handwriting, domain-specific terminology (e.g., jewelry, prehistory, architecture), and non-standard layouts requiring high-dimensional structured extraction. We present a comparative study of eight open-source lightweight VLMs (up to 7B parameters) for Optical Character Recognition (OCR)-to-structure across three university heritage collections. Given a document image, models must extract text and generate schema-compliant JSON, enabling automatic validation and downstream use. We evaluate models under a constraint-aware protocol across zero-shot, few-shot, and fine-tuning settings, measuring extraction fidelity and structured-output quality using Character Error Rate (CER), Approximate Normalized Levenshtein Similarity (ANLS*), and mean Average Precision F1 (mAP-F1). Against a fine-tuning baseline, we further test the independent impact of (i) hyperparameter optimization, (ii) classical image preprocessing (illumination flattening, denoising, and CLAHE), and (iii) multi-stage training. Finally, we analyze the trade-off between dataset-specific fine-tuning and a single multi-dataset checkpoint, where joint training enables one model to operate across collections but can shift performance between datasets. Overall, we show that carefully adapted VLMs with up to 7B parameters can provide a sustainable, private, high-performing alternative to manual transcription or commercial black-box systems, and we offer actionable guidance for heritage institutions seeking institution-controlled OCR-to-JSON extraction.

Chinese Translation

尽管大规模的闭源视觉-语言模型（VLMs）为文档理解设定了强有力的基准，但它们对商业API的依赖限制了在机构档案中的应用，原因在于数据自主性问题、持续成本和超大规模计算的环境影响。这在遗产数字化中尤为明显，因为文档包含历史手写、特定领域术语（例如，珠宝、史前、建筑）和需要高维结构提取的非标准布局。我们对八个开源轻量级VLM（最多7B参数）进行了比较研究，旨在实现光学字符识别（OCR）到结构化的JSON提取，涵盖了三个大学的遗产收藏。给定一幅文档图像，模型必须提取文本并生成符合架构的JSON，以便实现自动验证和后续使用。我们在零-shot、少-shot和微调设置下，采用约束感知协议评估模型，使用字符错误率（CER）、近似归一化Levenshtein相似度（ANLS*）和平均精确率F1（mAP-F1）来衡量提取的准确性和结构化输出的质量。针对微调基线，我们进一步测试了（i）超参数优化，（ii）经典图像预处理（照明平坦化、去噪和CLAHE），以及（iii）多阶段训练的独立影响。最后，我们分析了数据集特定微调与单一多数据集检查点之间的权衡，其中联合训练使得一个模型能够在多个收藏中运行，但可能会导致在不同数据集之间的性能变化。总体而言，我们表明，经过精心调整的VLM（最多7B参数）可以为手动转录或商业黑箱系统提供可持续、私密且高效的替代方案，并为寻求机构控制的OCR到JSON提取的遗产机构提供可行的指导。

View on arXiv Download PDF AI Translation

cs.CV / 110 / 2610.11826

From Suppression to Repair: Mitigating Object Hallucination in Large Vision-Language Models via Localized Distribution Alignment

从抑制到修复：通过局部分布对齐减轻大型视觉-语言模型中的物体幻觉

Zhao, Chen, Dong, Xingping, Shi, Jiachun, Peng, Liang, Wang, Chong, Lei, Zhen, He, Ran, Du, Bo

Abstract

Object hallucination remains a major obstacle for large vision-language models (LVLMs) to generate reliable content. An intuitive mitigation strategy is to suppress hallucination-related components in hidden representations. However, these components may also contain useful information, and suppressing them can weaken the model's multimodal capabilities. In this paper, we propose ResOT, a training-free method that repairs representations at inference time through localized distribution alignment. Specifically, ResOT projects dominant hallucinated directions away from the faithful subspace, forming a low-dimensional residual subspace for intervention. Within this subspace, ResOT uses Gaussian optimal transport (OT) to align the hallucinated distribution with the faithful one. The resulting map defines repair targets with minimal changes to the original representations. At inference, ResOT adaptively controls how far each token state moves toward its OT target. Experiments on three representative LVLMs show that ResOT substantially reduces object hallucination while improving image caption quality and multimodal performance across multiple benchmarks. Code will be released.

Chinese Translation

物体幻觉仍然是大型视觉-语言模型（LVLMs）生成可靠内容的主要障碍。一种直观的缓解策略是抑制隐藏表示中的幻觉相关成分。然而，这些成分也可能包含有用的信息，抑制它们可能会削弱模型的多模态能力。在本文中，我们提出了ResOT，这是一种无训练的方法，通过局部分布对齐在推理时修复表示。具体而言，ResOT将主导的幻觉方向投影到真实子空间之外，形成一个低维的残差子空间以进行干预。在这个子空间内，ResOT使用高斯最优传输（OT）将幻觉分布与真实分布对齐。由此产生的映射定义了对原始表示进行最小变化的修复目标。在推理时，ResOT自适应地控制每个标记状态向其OT目标移动的距离。对三种代表性LVLM的实验表明，ResOT显著减少了物体幻觉，同时在多个基准测试中提高了图像标题质量和多模态性能。代码将会发布。

View on arXiv Download PDF AI Translation

cs.CV / 111 / 2610.11846

Open-Vocabulary Audio-Visual Event Localization via Complex-Valued Fusion

通过复值融合实现开放词汇音视频事件定位

Praveen, Anirudh, Jerripothula, Koteswar Rao, Joshi, Pratik, Dayal, Aveen, Sawant, Neela

Abstract

Open-Vocabulary Audio-Visual Event Localization (OV-AVEL) labels each video segment with an event class, including classes that were never seen during training. The dominant pipeline uses a frozen multimodal foundation model (e.g. ImageBind) to embed the visual frame, the audio mel-spectrogram, and each candidate class name into a shared space, then computes two cosine similarities for each segment against each class: visual-text and audio-text. Existing methods then collapse this pair into a single scalar score with a fixed rule (geometric mean, weighted average) before taking the argmax. Instead, we compute complex-valued similarities and learn their fusion using a complex-valued neural network (CVNN). Each modality's standard representation becomes the real part of our pipeline, and a paired companion stream supplies the imaginary part. We use imaginary part of iHSV for visual modality and CycleGAN-translated phase spectrogram for audio modality as these companion streams. This results in two complex similarities, which are then fused. While the vision and audio encoders remain frozen, only the temporal-attention blocks and the fusion CVNN are trained. The four-stream complex architecture sets a new state of the art on both OV-AVEL benchmarks. On the open (unseen-class) split of OV-AVEBench we reach 66.5/59.1/54.1% Acc/Seg-F1/Event-F1 (+1.6/+4.1/+6.6 over the previously reported fine-tuned baseline), with consistent gains for seen classes as well. We also modify AVE dataset for this task and observe that our architecture reaches 60.7/51.9/50.4% Acc/Seg-F1/Event-F1, achieving state-of-the-art OV-AVEL results on it as well. We also propose a two-stream alternative, which also sees great improvements over the baseline.

Chinese Translation

开放词汇音视频事件定位（Open-Vocabulary Audio-Visual Event Localization, OV-AVEL）为每个视频片段标注一个事件类别，包括在训练过程中从未见过的类别。当前主流的方法使用一个冻结的多模态基础模型（例如 ImageBind）将视觉帧、音频梅尔谱图和每个候选类别名称嵌入到一个共享空间中，然后计算每个片段与每个类别之间的两个余弦相似度：视觉-文本和音频-文本。现有方法将这对相似度合并为一个标量分数，采用固定规则（几何平均、加权平均）后取最大值。相反，我们计算复值相似度，并使用复值神经网络（Complex-Valued Neural Network, CVNN）学习它们的融合。每种模态的标准表示成为我们管道的实部，而配对的伴随流提供虚部。我们使用 iHSV 的虚部作为视觉模态的伴随流，使用 CycleGAN 转换的相位谱图作为音频模态的伴随流。这导致生成两个复值相似度，然后进行融合。在视觉和音频编码器保持冻结的情况下，仅训练时间注意力模块和融合 CVNN。四流复值架构在 OV-AVEL 基准测试上设定了新的最先进水平。在 OV-AVEBench 的开放（未见类别）分割上，我们达到了 66.5/59.1/54.1% 的准确率/分割 F1/事件 F1（比之前报告的微调基线提高了 +1.6/+4.1/+6.6），并且在已见类别上也有一致的提升。我们还为此任务修改了 AVE 数据集，并观察到我们的架构在该数据集上达到了 60.7/51.9/50.4% 的准确率/分割 F1/事件 F1，取得了最先进的 OV-AVEL 结果。我们还提出了一种双流替代方案，该方案在基线之上也取得了显著的改进。

View on arXiv Download PDF AI Translation

cs.CV / 112 / 2610.11850

VEDJE: Video-Efficient Discriminative Joint Encoder for Scalable Video-Text Retrieval

VEDJE：视频高效判别联合编码器用于可扩展视频-文本检索

Wagner, Shahaf, Serussi, Gabriele, Ami, Dan Ben, Galanti, Tomer, Baskin, Chaim

Abstract

Finding the right video often requires distinguishing similar scenes in which different events occur. Joint matching improves retrieval, but processing rich video representations for each query is costly. VEDJE compresses features within sampled frames while keeping their representations separate in a reusable cache. Feature-change prediction supplies an auxiliary training signal that improves retrieval from the compressed cache without adding work at query time. On MSR-VTT, MSVD, DiDeMo, and ActivityNet, VEDJE improves R@1 over matched first-stage retrievers in both retrieval directions. On MSR-VTT, it reaches 59.8 text-to-video R@1 with a fine-tuned VideoCLIP-XL first stage. In the VideoPrism configuration, shrinking the per-video cache fourfold to 12 KiB preserves text-to-video recall within 0.2 points. These results show that accurate video search can operate on compact evidence, encoded once and reused as new queries arrive.

Chinese Translation

找到合适的视频通常需要区分不同事件发生的相似场景。联合匹配可以提高检索效果，但为每个查询处理丰富的视频表示是成本高昂的。VEDJE 在保持其表示在可重用缓存中分开的同时，压缩了采样帧中的特征。特征变化预测提供了一种辅助训练信号，能够在不增加查询时工作量的情况下，从压缩缓存中改善检索效果。在 MSR-VTT、MSVD、DiDeMo 和 ActivityNet 数据集上，VEDJE 在两个检索方向上都提高了与匹配的第一阶段检索器相比的 R@1。在 MSR-VTT 上，它与微调后的 VideoCLIP-XL 第一阶段结合，达到了 59.8 的文本到视频 R@1。在 VideoPrism 配置中，将每个视频的缓存缩小四倍至 12 KiB，保持了文本到视频的召回率在 0.2 分以内。这些结果表明，准确的视频搜索可以在紧凑的证据上进行，该证据只需编码一次，并在新的查询到达时重复使用。

View on arXiv Download PDF AI Translation

cs.CV / 113 / 2610.11857

Pose-Free Feed-Forward 3D Inpainting via Learnable Mask Attention and Support Token Refinement

无姿态前馈3D修复：可学习的掩码注意力与支持令牌精炼

Pan, Jingyi, Xu, Dan, Luo, Qiong

Abstract

3D scene inpainting aims to recover missing or occluded regions in edited 3D scenes, while ensuring geometric and textural consistency. Existing approaches, however, typically require accurately calibrated camera poses, which restricts their applicability in casual, in-the-wild scenarios and introduces additional preprocessing overhead. To overcome this limitation, we present FreeInpaint, a novel feed-forward framework that generates complete and 3D-consistent scenes directly from unposed multi-view images with masked regions. At its core, FreeInpaint extends a 3D foundation model to propagate masked regions from a reference view to other unposed views, bridging 3D reconstruction and scene inpainting while preserving the model's native ability to recover camera poses and scene geometry. Our method addresses two key challenges in adapting feed-forward 3D foundation models to masked inputs. First, masked regions can corrupt cross-view correspondence reasoning, degrading pose estimation and geometry recovery. To address this, we introduce a Learnable Mask Attention mechanism that preserves the spatial anchoring of reliable observations while allowing masked regions to progressively absorb useful context in deeper layers. Second, under severe occlusions, a single forward pass often lacks sufficient appearance evidence for high-fidelity completion. Therefore, we propose a Support Token Refinement strategy, which injects diffusion-generated support evidence as confidence-weighted auxiliary tokens to refine under-observed regions while preserving the original spatial anchor. Extensive experiments across diverse datasets demonstrate that FreeInpaint achieves superior inpainting quality, eliminating the reliance on pre-computed camera poses while keeping a fast inference speed. The project page is https://rorisis.github.io/FreeInpaint/.

Chinese Translation

3D场景修复旨在恢复编辑后的3D场景中缺失或被遮挡的区域，同时确保几何和纹理的一致性。然而，现有的方法通常需要准确校准的相机姿态，这限制了它们在日常场景和野外场景中的适用性，并增加了额外的预处理开销。为了解决这一限制，我们提出了FreeInpaint，一种新颖的前馈框架，能够直接从未标定的多视图图像中生成完整且3D一致的场景，这些图像中包含被掩盖的区域。FreeInpaint的核心是扩展一个3D基础模型，以便将参考视图中的掩盖区域传播到其他未标定视图，从而在保持模型原有恢复相机姿态和场景几何能力的同时，桥接3D重建和场景修复。我们的方法解决了将前馈3D基础模型适应于掩盖输入的两个关键挑战。首先，掩盖区域可能会破坏视图间对应关系的推理，从而降低姿态估计和几何恢复的精度。为此，我们引入了一种可学习的掩码注意力机制，该机制在允许掩盖区域逐步吸收有用上下文的同时，保持可靠观测的空间锚定。其次，在严重遮挡的情况下，单次前向传递通常缺乏足够的外观证据以实现高保真度的修复。因此，我们提出了一种支持令牌精炼策略，该策略将扩散生成的支持证据作为置信加权的辅助令牌注入，以在保持原始空间锚定的同时精炼观察不足的区域。通过在多样化数据集上的广泛实验，FreeInpaint展示了卓越的修复质量，消除了对预先计算的相机姿态的依赖，同时保持了快速的推理速度。项目页面为 https://rorisis.github.io/FreeInpaint/.

View on arXiv Download PDF AI Translation

cs.CV / 114 / 2610.11876

Relative Patch Response Learning for Generalizable AI-Generated Image Detection

用于可推广的AI生成图像检测的相对补丁响应学习

Wang, Tianyu, Li, Ouxiang, Hao, Yanbin, Tang, Zhenhua, Wang, Shuo

Abstract

Generative models can now synthesize highly realistic images, simultaneously increasing the risks of misinformation and visual forgery. Therefore, detecting AI-generated images becomes more essential, and a reliable detector must generalize to unseen generators and stay robust to unseen perturbations in the wild. Existing detectors are typically trained on either independently collected real and generated images or aligned real-generated pairs designed to mitigate content bias. Building on aligned pairs, recent methods form a mixed view by replacing some patches of the real image with their generated counterparts. However, we find that self-attention lets real and generated patches interact, so the feature of each patch no longer reflects its own source alone. This contextual shift makes a per-patch source label an imprecise target. To this end, we propose Relative Patch Response Learning (PRL). Instead of labeling each patch, PRL compares the same patch across two mixed views of an aligned pair and learns from its patch response, the change of its score between the views. (i) To give precise supervision under the contextual shift, a relative response objective measures the responses of source-changed patches against those of source-unchanged patches, which respond to the shift alone. (ii) To provide a reliable reference for the shift, a reference coherence objective keeps each group of source-unchanged patches moving as a whole. (iii) Since the two views contain different amounts of generated content, an area ranking objective asks the view with the larger generated area to have a higher mean patch score. Extensive experiments demonstrate the superior performance of PRL, which surpasses the best prior methods by 4.3% and 5.9% in average balanced accuracy across eight standard and three in-the-wild benchmarks, respectively.

Chinese Translation

生成模型现在能够合成高度逼真的图像，同时增加了误信息和视觉伪造的风险。因此，检测AI生成的图像变得更加重要，可靠的检测器必须能够对未见过的生成器进行泛化，并对现实环境中未见过的扰动保持鲁棒性。现有的检测器通常是在独立收集的真实图像和生成图像上训练，或者是在旨在减轻内容偏差的对齐真实-生成对上训练。基于对齐对，最近的方法通过用生成图像的对应部分替换真实图像的一些补丁形成混合视图。然而，我们发现自注意力机制使得真实和生成补丁之间相互作用，因此每个补丁的特征不再仅仅反映其自身的来源。这种上下文的变化使得每个补丁的源标签成为一个不精确的目标。为此，我们提出了相对补丁响应学习（Relative Patch Response Learning, PRL）。PRL不是对每个补丁进行标记，而是比较对齐对的两个混合视图中的相同补丁，并从其补丁响应中学习，即在视图之间其得分的变化。(i) 为了在上下文变化下提供精确的监督，相对响应目标测量源变化补丁的响应与源不变补丁的响应，这些补丁仅对变化作出响应。(ii) 为了为变化提供可靠的参考，参考一致性目标保持每组源不变补丁作为一个整体移动。(iii) 由于两个视图包含不同数量的生成内容，区域排名目标要求生成区域较大的视图具有更高的平均补丁得分。大量实验表明，PRL的性能优越，在八个标准基准和三个现实环境基准中，平均平衡准确率分别超过了最佳先前方法4.3%和5.9%。

View on arXiv Download PDF AI Translation

cs.CV / 115 / 2610.11907

Beyond Visual Enhancement: Adaptive Multi-Context Steering to Mitigate LVLM Hallucinations

超越视觉增强：自适应多上下文引导以减轻大型视觉语言模型的幻觉

Ma, Shuran, Li, JiaLe, Dong, Yuxin, Zheng, Shan, Jiang, Qingyun, Chen, Xiang, Zhu, Qi, Ji, Deyi, Yang, Yifan, Pan, Jianfeng, Tian, Yu, Yang, Xue

Abstract

Hallucination remains a significant challenge in Large Vision-Language Models (LVLMs). Existing training-free methods generally mitigate hallucinations through contrastive decoding or visual enhancement, often increasing the relative influence of visual evidence during generation. This raises a fundamental question: Can LVLMs dynamically regulate the contributions of different context sources to suppress hallucinations? In this work, we investigate and quantify how LVLMs coordinate multiple context sources during decoding and examine how this intrinsic behavior can guide hallucination mitigation. We find that LVLMs exhibit an intrinsic vision-attending tendency that can guide adaptive visual steering, while textual contexts can also contribute to hallucination mitigation. Motivated by these findings, we propose AIMS (Adaptive Information Multi-source Steering), a lightweight training-free framework that adaptively coordinates visual, prefilled textual, and generated contexts during decoding. Specifically, AIMS constructs compact prototypes for the three context domains and estimates their affinities with the current query to determine head-wise steering weights. The resulting multi-source steering direction is applied to the query representation, enabling adaptive context integration without additional model training or auxiliary forward passes. Extensive experiments across multiple LVLMs and decoding strategies demonstrate that AIMS effectively mitigates object hallucination while maintaining competitive general-purpose multimodal capabilities.

Chinese Translation

幻觉仍然是大型视觉语言模型（LVLMs）面临的重大挑战。现有的无训练方法通常通过对比解码或视觉增强来减轻幻觉，往往在生成过程中增加视觉证据的相对影响。这引发了一个根本性的问题：LVLMs能否动态调节不同上下文来源的贡献以抑制幻觉？在本研究中，我们调查并量化了LVLMs在解码过程中如何协调多个上下文来源，并考察这种内在行为如何指导幻觉的减轻。我们发现LVLMs表现出一种内在的视觉关注倾向，可以引导自适应视觉引导，而文本上下文也可以有助于减轻幻觉。基于这些发现，我们提出了AIMS（自适应信息多源引导），这是一个轻量级的无训练框架，能够在解码过程中自适应地协调视觉、预填充文本和生成上下文。具体而言，AIMS为这三个上下文领域构建紧凑的原型，并估计它们与当前查询的亲和性，以确定头部引导权重。生成的多源引导方向应用于查询表示，使得在不需要额外模型训练或辅助前向传递的情况下实现自适应上下文整合。在多个LVLMs和解码策略上的广泛实验表明，AIMS有效减轻了物体幻觉，同时保持了竞争力的通用多模态能力。

View on arXiv Download PDF AI Translation

cs.CV / 116 / 2610.11924

Revisiting Identity and Spectra Dispersion in Media-Bridged Time Series Forecasting: Linking Multivariate Signals and Narrative Flows

重新审视媒体桥接时间序列预测中的身份与谱散布：连接多变量信号与叙事流

Lei, Jierui, Zhang, Wenjian, Yang, Qingyi, Hong, Yuyang, Chen, Fangzheng, Zhang, Zhengbo, Tang, Haina, Xiang, Shiming

Abstract

Media-bridged time series forecasting is expanding to encompass traditional "multivariate" and emerging "multimodal" (e.g., through textual assistance). Existing Time Series Forecasting (TSF) models still rely on paradigm-specific relation, fusion, and temporal modules, hindering a common forecasting backbone across numerical and pre-aligned narrative-flow settings. To explore this, we propose the Multimedia Identity-Aware Prism Network (MIDAPN), a unified spatiotemporal forecasting backbone based on media-general graph adaptation and automatic temporal learning: (1) Following media pre-alignment, our Multimedia Identity-Aware Graph (MIDAG) revisits identity through static essence, dynamic behavior, and latent commonality, inducing affinities that extend variable-specific dependencies across media. Contextual Identity Modulation (CIM) further refines discriminative aggregation. (2) We develop Spectral Prism Convolution (SPConv) to automatically perform hierarchical temporal analysis, balancing coarse trends and fine-grained details. Meanwhile, its Adaptive Search Guidance configures a scale-efficient architecture for temporal-dimension reconstruction. These decoupled yet synergistic components jointly address media identity disentanglement and temporal-scale mismatch. Comprehensive evaluations involving 16 SOTA TSF models across 13 "multivariate" and 12 "multimodal" datasets, alongside targeted long-context comparisons against 14 time series foundation models and fused pretrained language models, demonstrate MIDAPN's consistent superiority and broad shared backbone compatibility. The code is available at \href{https://github.com/leijieruilq/MIDAPN/tree/main}{https://github.com/MIDAPN}.

Chinese Translation

媒体桥接的时间序列预测正在扩展，以涵盖传统的“多变量”和新兴的“多模态”（例如，通过文本辅助）。现有的时间序列预测（TSF）模型仍然依赖于特定范式的关系、融合和时间模块，这阻碍了在数值和预对齐叙事流设置之间建立共同的预测基础。为此，我们提出了多媒体身份感知棱镜网络（MIDAPN），这是一个基于媒体通用图适应和自动时间学习的统一时空预测基础： (1) 在媒体预对齐后，我们的多媒体身份感知图（MIDAG）通过静态本质、动态行为和潜在共性重新审视身份，诱导出跨媒体扩展变量特定依赖关系的亲和力。上下文身份调制（CIM）进一步精炼了区分性聚合。 (2) 我们开发了谱棱镜卷积（SPConv），以自动执行层次时间分析，平衡粗略趋势与细粒度细节。同时，其自适应搜索引导配置了一个规模高效的时间维度重构架构。这些解耦但协同的组件共同解决了媒体身份解耦和时间尺度不匹配的问题。涉及16个最先进的TSF模型的综合评估，涵盖13个“多变量”和12个“多模态”数据集，以及针对14个时间序列基础模型和融合预训练语言模型的针对性长上下文比较，证明了MIDAPN的一致优越性和广泛的共享基础兼容性。代码可在 exttt{https://github.com/leijieruilq/MIDAPN/tree/main} 获取。

View on arXiv Download PDF AI Translation

cs.CV / 117 / 2610.11938

Does Target Alignment Mean Target Recovery? An Evidence-Ladder Study of Adversarial Claims on Contrastive Encoders

目标对齐是否意味着目标恢复？对对比编码器的对抗性声明的证据阶梯研究

Yang, Tao, Zhou, Jianying

Abstract

Adversarial attacks on vision-language models optimize an image toward a text target, then cite the attacked model's similarity score as evidence of success. We ask whether that score - victim-space target alignment (VTS) - predicts recovery of the target by an independent model. We first build a measurement instrument: supervised judges outside the attacked geometry, real-target blend controls, shuffled-target negatives, and a reference level derived from a 50% target-image blend. Two preregistered studies then compare six contrastive encoders under a matched attack at three perturbation budgets. Robustly trained encoders (FARE, TeCoA, PMG, TRADES) transfer substantially more independent evidence than vanilla CLIP or SigLIP; all eight contrasts reject at the bootstrap floor. However, no cell reaches the blend-derived reference level. The three best cells fall within its replication band, leaving practical recovery undecided. Within robust encoders, per-sample alignment gain correlates with evidence gain ($\rho = 0.24-0.51$); within vanilla CLIP the correlation is consistent with zero. Across encoders we find no monotone alignment-evidence relation. VTS is therefore informative only within a fixed robust encoder, and we provide a reporting protocol in its place.

Chinese Translation

对视觉-语言模型的对抗攻击通过将图像优化到文本目标，然后引用被攻击模型的相似性得分作为成功的证据。我们探讨这个得分——受害者空间目标对齐（VTS）——是否能预测独立模型对目标的恢复。我们首先构建了一种测量工具：在被攻击几何体外的监督评审员、真实目标混合控制、打乱目标的负例，以及从50%目标-图像混合中得出的参考水平。随后，我们进行了两项预注册研究，比较六种对比编码器在三个扰动预算下的匹配攻击。经过强健训练的编码器（FARE、TeCoA、PMG、TRADES）传递的独立证据显著多于普通的CLIP或SigLIP；所有八个对比在自助法底线下均被拒绝。然而，没有任何组别达到混合得出的参考水平。表现最好的三个组别落在其复制带内，实际恢复尚未确定。在强健编码器中，每个样本的对齐增益与证据增益相关（$ ho = 0.24-0.51$）；而在普通的CLIP中，这种相关性与零一致。在所有编码器中，我们未发现单调的对齐-证据关系。因此，VTS 仅在固定的强健编码器内具有信息价值，我们提供了相应的报告协议。

View on arXiv Download PDF AI Translation

cs.CV / 118 / 2610.11942

Right Screen, Wrong Transition: World Models as Verifiers for GUI Agents

正确的屏幕，错误的过渡：世界模型作为GUI代理的验证器

Zhang, Jiaming, Wang, Xuan, Zhang, Fuyao, Cao, Yang, Lyu, Lingjuan, Lim, Wei Yang Bryan

Abstract

A login screen that appears after a tap on Sign in is expected; the same screen after a tap on View order is an attack. For GUI agents, safety is therefore a property of the transition rather than of the screen, and a monitor that inspects only screens can be defeated by reusing a legitimate one. Judging a transition requires an expectation of what should have followed the action. Existing GUI world models provide one, but they output it as text, code, or images, so checking it against the observed screen requires a second model to judge the two. We argue that a world model meant for verification should instead predict in the space in which observations are encoded, and present LGWM, a decoder-free, action-conditioned world model that predicts the representation of the next screen directly, trained without semantic annotation on 1.85M real GUI transitions. Verification reduces to a vector comparison, and the same signal reveals whether a mismatch is harmful. We evaluate on RSWT-BENCH, a diagnostic where each credential screen appears under both a legitimate and a hijacked transition, so detectors that see only the screen are at chance by construction. The training-free score reaches 0.987 AUC at 17 ms per decision, on par with the strongest closed-source VLMs and about ten AUC points above generative GUI world models at over three orders of magnitude lower latency. The residual direction reaches 0.953 AUC at separating harmful from benign violations, where prompted VLMs are near chance. Further analyses show that the prediction is a usable future state rather than an anomaly score. World models have mostly served as simulators or planners; our results point to a third role, verification, for which predicting in representation space is the natural design.

Chinese Translation

在点击“登录”后出现的登录屏幕是预期的；而在点击“查看订单”后出现的同一屏幕则构成攻击。因此，对于GUI代理而言，安全性是过渡的属性，而不是屏幕的属性，单纯检查屏幕的监控工具可能会被重用合法屏幕的方式所击败。判断一个过渡需要对该行为后应该出现的内容有预期。现有的GUI世界模型提供了这样的预期，但它们以文本、代码或图像的形式输出，因此需要一个第二模型来判断观察到的屏幕与预期之间的关系。我们认为，旨在验证的世界模型应该在观察编码的空间中进行预测，并提出了LGWM，这是一种无解码器、基于动作条件的世界模型，能够直接预测下一个屏幕的表示，且在185万真实GUI过渡上进行训练时不需要语义注释。验证简化为向量比较，相同的信号可以揭示不匹配是否有害。我们在RSWT-BENCH上进行评估，这是一个诊断工具，其中每个凭证屏幕在合法和劫持的过渡下均会出现，因此仅查看屏幕的检测器在构造上处于随机状态。无训练的得分在每个决策17毫秒内达到了0.987 AUC，接近最强的闭源VLM，并且比生成的GUI世界模型高出约十个AUC点，同时延迟低三个数量级。残差方向在区分有害与良性违规时达到了0.953 AUC，而提示的VLM接近随机水平。进一步分析表明，该预测是一个可用的未来状态，而不是异常分数。世界模型大多作为模拟器或规划者存在；我们的结果指出了第三种角色，即验证，其中在表示空间中进行预测是自然的设计。

View on arXiv Download PDF AI Translation

cs.CV / 119 / 2610.11954

Look Where You Can: Active View Selection for CAD Reconstruction under Occlusion

关注你能看到的：在遮挡下的CAD重建主动视角选择

Bali, Kartik, Guru, Mahish, Borjigin, Yiderigun, Starostina, Alexandra, Cyron, Christian J., Aydin, Roland

Abstract

CAD reconstruction methods assume a luxury reality rarely grants: unrestricted visual access to the object, photographed from any desired angle. Real objects, however, are scene-embedded, bolted against walls, wedged into corners, resting on floors, where the scene renders much of the view sphere unreachable and the remaining views unequally informative. We introduce \textbf{SightCAD}, a framework for parametric CAD reconstruction that treats view feasibility as a first-class constraint. In this work we consider objects from standard CAD benchmarks embedded in realistic indoor scenes with physically derived visibility constraints over a discrete view sphere. A learned view selector must choose $K$ feasible views for a vision--language model (VLM) that generates executable CadQuery code, scored by geometric fidelity of the executed solid. Because reward arrives only after discrete view selection, autoregressive generation, and CAD-kernel execution, we propose a joint training paradigm in which the view selector and the CAD-generation VLM are trained together against this reward. The learned selection policy departs sharply from random, uniform, and coverage-greedy alternatives, outperforming surface-area maximization (SA-max) by up to $6.4$ Intersection-over-Union (IoU) points across budgets $K\in\{1,\dots,5\}$. The full system surpasses strong external baselines on scene-embedded, occluded multi-view renders of DeepCAD and Fusion360 objects ($+21$ and $+17$ effective-mIoU points over the best baseline, respectively), as well as on test-time domain-canonicalized real images from the industrial T-LESS benchmark and on both synthetic and real images from the MP6D industrial metal-parts benchmark, while producing the highest rate of executable programs of any method compared (invalid-code rate ${\leq}1.5\%$).

Chinese Translation

CAD重建方法假设了一种奢侈的现实，而这种现实很少被赋予：对物体的无限制视觉访问，可以从任何所需角度拍摄。然而，真实物体通常嵌入场景中，固定在墙壁上，卡在角落里，或放置在地板上，这使得场景中的大部分视图空间无法到达，剩余的视图信息量也不均等。我们引入了 extbf{SightCAD}，一个将视图可行性视为一项首要约束的参数化CAD重建框架。在这项工作中，我们考虑了来自标准CAD基准的物体，这些物体嵌入了具有物理可见性约束的现实室内场景中，覆盖在离散视图空间上。一个学习的视图选择器必须为一个生成可执行CadQuery代码的视觉-语言模型（VLM）选择$K$个可行视图，该代码的评分基于执行固体的几何保真度。由于奖励仅在离散视图选择、自回归生成和CAD内核执行后到达，我们提出了一种联合训练范式，其中视图选择器和CAD生成VLM共同针对这一奖励进行训练。学习到的选择策略与随机、均匀和覆盖贪婪的替代方案有显著区别，在预算$K extin ext{in} ext{1, ext{...},5 ext{}}$中，超越了表面积最大化（SA-max）多达$6.4$个交并比（IoU）点。整个系统在DeepCAD和Fusion360物体的场景嵌入、遮挡多视图渲染上超越了强大的外部基线（分别比最佳基线高出$+21$和$+17$有效mIoU点），同时在工业T-LESS基准的测试时域标准化真实图像以及来自MP6D工业金属零件基准的合成和真实图像上表现优异，且产生的可执行程序的比率是所有比较方法中最高的（无效代码率${ ext{≤}}1.5 ext{％}$）。

View on arXiv Download PDF AI Translation

cs.CV / 120 / 2610.11967

Learning Which Correspondences to Trust: Confidence-Weighted Event-Camera Localization in LiDAR Maps

学习信任哪些对应关系：基于置信度加权的事件相机在LiDAR地图中的定位

Kiousis, Panagiotis, Chen, Kuangyi, Zhang, Jun, Fraundorfer, Friedrich

Abstract

Localizing an event camera against a pre-built LiDAR map can be cast as dense optical-flow estimation between a rendered depth view and an event image, followed by a Perspective-n-Point (PnP) solver over the induced 3D-2D correspondences. Existing pipelines rely on geometric consensus during pose estimation, but do not explicitly model the reliability or pose informativeness, i.e., how strongly a correspondence constrains the camera pose, of individual correspondences. We show that the natural way to learn it -- using the per-correspondence error to constrain the learning of confidence -- suffers from a depth-dependent bias: small pixel errors reside predominantly at large depths and do not lead to high pose informativeness. Instead, in our method (CELL), we learn a per-correspondence confidence end-to-end through the pose, using a differentiable probabilistic PnP whose log-partition term encourages weight configurations that yield a better-constrained pose distribution. The learned confidence is used in three ways: (i) it reweights the flow supervision in a decoupled training scheme that keeps pose gradients out of the flow/edge backbone; (ii) it drives a probabilistic correspondence selection at test time; and (iii) together with the network's edge-probability it weights a final edge-matching refinement. We further design a partial-completion depth representation that adds signal without hallucinating across large gaps. On M3ED and DSEC our full system improves over the LEAR baseline on the majority of the evaluated sequences: it reduces the median translation error by up to 26.9% and the median rotation error by up to 15.8%.

Chinese Translation

将事件相机定位于预先构建的LiDAR地图可以视为在渲染深度视图与事件图像之间进行密集光流估计，随后通过透视-n-点（PnP）求解器处理所产生的3D-2D对应关系。现有的流程在姿态估计过程中依赖几何一致性，但并未明确建模单个对应关系的可靠性或姿态信息量，即对应关系对相机姿态的约束程度。我们表明，学习这一点的自然方式——利用每个对应关系的误差来约束置信度的学习——存在深度依赖的偏差：小的像素误差主要出现在大深度处，并不会导致高姿态信息量。相反，在我们的方法（CELL）中，我们通过姿态端到端地学习每个对应关系的置信度，使用可微分的概率PnP，其对数分区项鼓励产生更好约束的姿态分布的权重配置。学习到的置信度有三种使用方式：（i）在一个解耦的训练方案中重新加权光流监督，避免姿态梯度影响光流/边缘主干；（ii）在测试时驱动概率对应关系选择；（iii）与网络的边缘概率一起加权最终的边缘匹配精细化。我们进一步设计了一种部分完成的深度表示，增加了信号而不在大间隙中产生幻觉。在M3ED和DSEC上，我们的完整系统在大多数评估序列中超越了LEAR基线：将中位平移误差降低了最多26.9%，将中位旋转误差降低了最多15.8%。

View on arXiv Download PDF AI Translation

cs.CV / 121 / 2610.11986

FearCaut-Qwen: Affective Steering in a Vision-Language Model Shifts the Decision Criterion for Hazard Assessment

FearCaut-Qwen：情感引导在视觉-语言模型中的应用改变了危险评估的决策标准

Zhou, Xiaoshan

Abstract

Vision-language models (VLMs) show great potential for damage assessment after a disaster, but a recurring deficiency is that they are reluctant to declare a hazard; that is, recall is low even when overall accuracy appears adequate. This study examines that deficiency by using signal detection theory to decompose the decision behavior into perceptual capability and decision-criterion placement. We then propose a novel method for correcting the over-conservative decision policy, inspired by the finding that fear makes humans risk-averse, and ask whether an affective representation associated with fear can be causally manipulated to similarly alter a VLM's decision tendency. Using mechanistic interpretability, we localize a causally implicated affective circuit in the model and use activation steering to manipulate it while observing the effect on downstream prediction. The method is tested on a two-stage SeisMLLM pipeline built on Qwen2.5-VL-7B-Instruct, which flags only 27.0% of genuinely unsafe buildings on the SeisMLLM-1K test split and never issues a false Red, an SDT criterion of c = +1.354, despite adequate evidence quality (d' = 1.521). An affective direction is localized on emotion-rich natural scenes, causally validated by sparse-neuron knockout and distributed steering on held-out emotion data, and then injected into the building task. Fear-direction injection raises Red recall to 75.7% (p<0.001), and subtracting the same direction suppresses Red predictions entirely, whereas norm-matched random and matched happiness directions show no significant effect. The mechanism is a shift in criterion (c=-1.515) while discrimination is not improved (d'=-0.493). These results show that VLM decisions can be adjusted at inference time without retraining and demonstrate how mechanistic interpretability can be used to diagnose and control VLM decision behaviors in engineering applications.

Chinese Translation

视觉-语言模型（VLMs）在灾后损害评估中展现出巨大潜力，但一个反复出现的缺陷是它们不愿意宣告危险；也就是说，即使整体准确性看似足够，召回率仍然较低。本研究通过使用信号检测理论来分解决策行为，考察这一缺陷，将其分为感知能力和决策标准的设置。我们提出了一种新的方法来纠正过于保守的决策策略，灵感来自于恐惧使人类趋向于规避风险的发现，并探讨与恐惧相关的情感表征是否可以被因果操控，从而类似地改变VLM的决策倾向。通过机械可解释性，我们在模型中定位了一个因果相关的情感电路，并利用激活引导对其进行操控，同时观察对下游预测的影响。该方法在基于Qwen2.5-VL-7B-Instruct构建的两阶段SeisMLLM管道上进行了测试，该管道在SeisMLLM-1K测试集上仅标记了27.0%的真正不安全建筑，并且从未发出假红色警报，其信号检测理论标准为c = +1.354，尽管证据质量足够（d' = 1.521）。在情感丰富的自然场景中定位到一个情感方向，通过稀疏神经元敲除和在保留情感数据上的分布式引导进行因果验证，然后将其注入建筑任务中。恐惧方向的注入将红色召回率提高到75.7%（p<0.001），而去除相同方向则完全抑制了红色预测，而与规范匹配的随机方向和匹配的幸福方向则未显示出显著效果。该机制是标准的转变（c=-1.515），而区分能力并未改善（d'=-0.493）。这些结果表明，VLM的决策可以在推理时进行调整而无需重新训练，并展示了机械可解释性如何用于诊断和控制工程应用中的VLM决策行为。

View on arXiv Download PDF AI Translation

cs.CV / 122 / 2610.11993

DataVista: Diagnosing Multimodal LLMs on Data Video Understanding

DataVista：对多模态大语言模型在数据视频理解上的诊断

Xie, Yupeng, Wang, Zhenyang, Zhu, Jiayi, Tang, Yinghao, Shen, Zhouan, Chen, Yiyu, Luo, Yuyu

Abstract

Data video is a media form that integrates data visualization with video narrative, widely adopted in news reporting and business analysis. Compared with general video understanding, data video understanding places greater emphasis on accurately reading data from animated charts, integrating evidence across charts and time, and understanding how narrative organization and visual design communicate information. Yet existing benchmarks target either general videos or static charts, and data video understanding has not been systematically evaluated. We present DataVista, the first benchmark for data video understanding, containing 961 real-world data videos and 6,775 evaluation questions organized under a three-level progressive capability framework (data perception, temporal reasoning, narrative understanding) with 10 fine-grained question types across five topic domains. Systematic evaluation of 19 mainstream MLLMs shows that the best-performing model, Gemini-3.1-Pro, achieves 70.0% overall accuracy, still far below human expert performance, with models performing worst on Causal Reasoning and Narrative Structure. Increasing frame counts and adding subtitles mainly benefit data perception and temporal reasoning, with limited gains in narrative understanding. Further analysis of model responses identifies typical failure modes in chart reading, evidence judgment, and instruction understanding. The benchmark is available at https://github.com/HKUSTDial/DataVista.

Chinese Translation

数据视频是一种将数据可视化与视频叙事相结合的媒体形式，广泛应用于新闻报道和商业分析。与一般的视频理解相比，数据视频理解更强调准确读取动态图表中的数据、跨图表和时间整合证据，以及理解叙事组织和视觉设计如何传达信息。然而，现有的基准测试主要针对一般视频或静态图表，数据视频理解尚未得到系统评估。我们提出了DataVista，这是首个针对数据视频理解的基准，包含961个真实数据视频和6,775个评估问题，按照三层渐进能力框架（数据感知、时间推理、叙事理解）组织，涵盖五个主题领域的10种细分问题类型。对19个主流多模态大语言模型的系统评估显示，表现最佳的模型Gemini-3.1-Pro的整体准确率为70.0%，仍远低于人类专家的表现，模型在因果推理和叙事结构方面的表现最差。增加帧数和添加字幕主要有利于数据感知和时间推理，而在叙事理解方面的提升有限。对模型响应的进一步分析识别了在图表阅读、证据判断和指令理解中的典型失败模式。该基准可在 https://github.com/HKUSTDial/DataVista 获取。

View on arXiv Download PDF AI Translation

cs.CV / 123 / 2610.12048

Do Not Train Away Uncertainty: Early Uncertainty Anchored Calibration

不要消除不确定性：早期不确定性锚定校准

Xie, Yutong, Tang, Jiawei, Hua, Zhenglin, Ma, Yuxiang, Qin, Si, Hou, Yaxin, Liu, Hui, Hou, Junhui, Jia, Yuheng

Abstract

Deep neural networks, including large language models, have achieved remarkable performance across various tasks. However, they are prone to overconfidence during training or fine-tuning. In this work, we observe a consistent phenomenon across different models that the early model is better calibrated, while later training or fine-tuning yields marginal accuracy gains but substantially increases calibration errors. Our analysis suggests that the early model retains uncertainty awareness in both its predictions and features, which is gradually lost with continued training. To avoid training away this uncertainty awareness, we propose \textbf{EUA-Cal}, a novel method that exploits the \textbf{E}arly model as an \textbf{U}ncertainty \textbf{A}nchor for \textbf{Cal}ibration. EUA-Cal introduces early prediction regularization to preserve early predictive uncertainty and prototype structure regularization to exploit uncertainty reflected in the early feature space, jointly mitigating overconfidence. Extensive experiments on image classification and multiple-choice question answering across eight diverse models demonstrate that EUA-Cal outperforms state-of-the-art calibration methods.

Chinese Translation

深度神经网络，包括大型语言模型，在各种任务中取得了显著的性能。然而，它们在训练或微调过程中容易产生过度自信。在本研究中，我们观察到不同模型之间存在一个一致的现象，即早期模型的校准效果更好，而后期的训练或微调虽然带来了边际的准确性提升，却显著增加了校准误差。我们的分析表明，早期模型在其预测和特征中保留了不确定性意识，而这种意识随着训练的持续而逐渐丧失。为了避免消除这种不确定性意识，我们提出了 extbf{EUA-Cal}，一种新颖的方法，利用 extbf{E}arly模型作为 extbf{U}ncertainty extbf{A}nchor进行 extbf{Cal}ibration。EUA-Cal引入了早期预测正则化，以保留早期预测的不确定性，并采用原型结构正则化，以利用早期特征空间中反映的不确定性，从而共同减轻过度自信。在八个不同模型上的图像分类和多项选择问答的广泛实验表明，EUA-Cal优于最先进的校准方法。

View on arXiv Download PDF AI Translation

cs.CV / 124 / 2610.12060

Look Back, Think Ahead: Visual Memory on Demand for Efficient Multimodal Reasoning

回顾与展望：按需视觉记忆以实现高效的多模态推理

Xue, Yicheng, Wu, Han, Yang, Jufeng, Dong, Minjing, Chen, Xinghao, Chen, Hanting, Guo, Jianyuan

Abstract

Processing long visual token sequences from high-resolution images makes multi-step reasoning computationally expensive for multimodal Large Language Models (MLLMs). Existing one-shot pruning and aggregation methods compress visual tokens into a fixed context before decoding. However, visual evidence needs can shift as reasoning unfolds, making it difficult for a fixed compressed context to retain all the details needed across stages. To address this challenge, we propose ViMoD, a lightweight framework that maintains a compact visual context while preserving access to original fine-grained evidence as reasoning needs evolve. Deformable Aggregation of Region-wise Tokens (DART) learns content-adaptive groups and aggregation capacities, constructing compact Coarse representations linked to recoverable original Fine tokens. Temporal Routing for Adaptive Contextual Evidence (TRACE) integrates decoding history to anticipate upcoming evidence needs and select, retain, or replace active Fine-token groups. Selected Fine tokens augment the persistent Coarse context in the frozen backbone, enabling stage-specific evidence access without continuously attending to all visual tokens. On Qwen3-VL-4B, ViMoD outperforms all evaluated baselines on all eight reasoning benchmarks at a 20% target visual token budget, improving the mean normalized score by 39.0% over the strongest evaluated one-shot baseline. These gains are achieved with only 0.0546% additional trainable parameters relative to the frozen backbone.

Chinese Translation

处理来自高分辨率图像的长视觉标记序列使得多步推理在多模态大型语言模型（MLLMs）中计算成本高昂。现有的一次性剪枝和聚合方法在解码之前将视觉标记压缩为固定上下文。然而，随着推理的展开，视觉证据的需求可能会发生变化，这使得固定的压缩上下文难以在各个阶段保留所需的所有细节。为了解决这一挑战，我们提出了ViMoD，一个轻量级框架，它在保持紧凑视觉上下文的同时，保留对原始细粒度证据的访问，以适应推理需求的演变。区域标记的可变聚合（DART）学习内容自适应的组和聚合能力，构建与可恢复的原始细粒度标记相关联的紧凑粗略表示。自适应上下文证据的时间路由（TRACE）整合解码历史，以预测即将到来的证据需求并选择、保留或替换活动的细粒度标记组。选定的细粒度标记增强了在冻结主干网络中的持久粗略上下文，使得在不持续关注所有视觉标记的情况下实现阶段特定的证据访问。在Qwen3-VL-4B上，ViMoD在20%的目标视觉标记预算下，在所有八个推理基准测试中超越了所有评估的基线，相较于最强的一次性基线，平均标准化得分提高了39.0%。这些提升仅通过相对于冻结主干网络增加0.0546%的可训练参数实现。

View on arXiv Download PDF AI Translation

cs.CV / 125 / 2610.12069

LIVIN: Benchmarking Spatial and Embodied Intelligence in Digital Twins of Lived-In Homes

LIVIN：数字双胞胎中居住家庭的空间与具身智能基准测试

Xu, Peijun, Nie, Chuansen, He, Yiyang, Bai, Yinuo, Liu, Jingyang, Shao, Kuixiang, Jiao, Yuyang, Xia, Kuanhao, Zhu, Jiayi, Yang, Zitian, Zhang, Yanqi, Tan, Tianye, Di, Shuwei, Xu, Junyi, Yu, Jingyi, Gu, Jiayuan

Abstract

Realistic household simulation must capture not only diverse environments but also the lived-in object arrangements and spatial constraints that shape robot motion and interaction. Existing resources often trade off scale, real-world correspondence, and interaction readiness, leaving a gap in faithful, interactive replicas of how real homes are actually arranged. To this end, we introduce LIVIN, a benchmark for spatial and embodied intelligence built on digital twins of 30 diverse lived-in homes. These replicas preserve observed room layouts, furniture configurations, and everyday belongings. To construct them, we design a human-in-the-loop workflow comprising instance recognition, architectural reconstruction, and object generation and placement, with intermediate results reviewed and corrected by humans against the source observations at each stage. We evaluate four tasks in LIVIN: 3D detection, 3D reconstruction, navigation, and loco-manipulation. Our evaluations show that current methods remain challenged by the dense object arrangements, occlusions, limited free space, and constrained interaction regions found in realistic lived-in homes. We hope LIVIN will help advance embodied AI in real-world homes, from spatial understanding to robotic interaction, and ultimately bring embodied intelligence into everyday home environments.

Chinese Translation

现实的家庭模拟不仅必须捕捉多样化的环境，还要反映居住物品的排列和塑造机器人运动与互动的空间限制。现有资源往往在规模、与现实世界的对应性和互动准备性之间进行权衡，导致缺乏真实家庭布局的忠实互动复制品。为此，我们引入了LIVIN，一个基于30个多样化居住家庭的数字双胞胎构建的空间与具身智能基准。这些复制品保留了观察到的房间布局、家具配置和日常物品。为了构建这些复制品，我们设计了一个人机协作的工作流程，包括实例识别、建筑重建以及物体生成与放置，且在每个阶段中，初步结果均由人类根据源观察进行审查和纠正。我们在LIVIN中评估了四个任务：3D检测、3D重建、导航和运动操控。我们的评估显示，当前方法在面对现实居住家庭中密集的物体排列、遮挡、有限的自由空间以及受限的互动区域时仍面临挑战。我们希望LIVIN能够推动具身人工智能在现实家庭中的发展，从空间理解到机器人互动，最终将具身智能引入日常家庭环境。

View on arXiv Download PDF AI Translation

cs.CV / 126 / 2610.12081

Perception Test 2026: Challenge Summary and Extension to City-scale Audio-Visual Reasoning

感知测试2026：挑战总结及城市规模音视频推理的扩展

Kitashov, Fedor, Carreira, João, Ginosar, Shiry, Damen, Dima, Zisserman, Andrew, Pătrăucean, Viorica

Abstract

Continuing the Perception Test challenge series, we organised the fourth edition as a workshop at the European Conference on Computer Vision (ECCV) 2026 in Malm\"o, Sweden. This edition focused on spatial intelligence and featured four different tracks: unified multiple-choice videoQA and grounded videoQA from the original Perception Test benchmark, alongside two new tracks based on city-scale walking-tour videos (KilometerAudio and KilometerVision). In this report, we describe the new benchmarks used for the city-scale tracks and summarise the winning solutions across all tracks, including a generalist model that competed across all tracks with satisfactory performance. The winning solutions in the newly added city-scale tracks demonstrated that complex spatial and multimodal reasoning can be solved by expensive agentic pipelines, but remains difficult for multimodal models used standalone.

Chinese Translation

为了延续感知测试挑战系列，我们在2026年欧洲计算机视觉会议（ECCV）于瑞典马尔默举办了第四届研讨会。本届会议聚焦于空间智能，设立了四个不同的赛道：来自原始感知测试基准的统一多项选择视频问答（videoQA）和基于位置的视频问答（grounded videoQA），以及两个基于城市规模步行旅游视频的新赛道（KilometerAudio 和 KilometerVision）。在本报告中，我们描述了用于城市规模赛道的新基准，并总结了所有赛道的获胜解决方案，包括一款在所有赛道中竞争并表现令人满意的通用模型。新增加的城市规模赛道中的获胜解决方案表明，复杂的空间和多模态推理可以通过昂贵的代理管道解决，但对于单独使用的多模态模型而言，仍然具有挑战性。

View on arXiv Download PDF AI Translation

cs.CV / 127 / 2610.12095

DVLA-RL++: Dual-Level Vision-Language Alignment with Reinforcement Learning Gating for Few-Shot Learning

DVLA-RL++：基于强化学习门控的双层视觉-语言对齐用于小样本学习

Li, Wenhao, Meng, Xianjing, Wang, Qiangchang, Han, Zhongyi, Yin, Yilong, Nie, Liqiang

Abstract

Few-shot learning aims to recognize novel categories from limited labeled examples. Recent studies incorporate textual semantics to compensate for limited visual observations and improve class representations. However, high image-text agreement may reflect both intrinsic object properties and incidental context, making support prototypes susceptible to contextual contamination. To address this problem, we propose DVLA-RL++, which extends DVLA-RL with complementary semantic purification (CSP) and counterfactual reinforcement-learning gating (CRG). Specifically, CSP generates intrinsic and nuisance descriptions from labeled supports and compares their agreement with each support token. An ambiguity-dependent rejection margin guides sparse evidence allocation, while an intrinsic semantic anchor fills the unassigned mass to provide a fallback when visual evidence is unreliable. CRG learns layer-wise semantic fusion strengths using a reward that balances recognition performance and nuisance exposure. An independently executed reference trajectory on the same episode provides a paired learning signal. Theoretical analysis relates retained evidence and anchor quality to prototype stability and establishes conditions for unbiased on-policy gradient estimation. Experiments on standard, fine-grained, and cross-domain benchmarks show state-of-the-art accuracy, with an average gain of 1.4% over DVLA-RL. The project page is available at https://peacelwh.github.io/TPAMI27-DVLA-RLpp/.

Chinese Translation

小样本学习旨在从有限的标记示例中识别新类别。最近的研究结合了文本语义，以弥补有限的视觉观察并改善类别表示。然而，高图像-文本一致性可能反映了内在的物体属性和偶然的上下文，使得支持原型容易受到上下文污染。为了解决这个问题，我们提出了DVLA-RL++，它在DVLA-RL的基础上扩展了互补语义净化（CSP）和反事实强化学习门控（CRG）。具体而言，CSP从标记支持中生成内在和干扰描述，并比较它们与每个支持标记的一致性。依赖于模糊性的拒绝边际指导稀疏证据分配，而内在语义锚点填补未分配的质量，以便在视觉证据不可靠时提供后备。CRG使用一种奖励来学习逐层的语义融合强度，该奖励平衡识别性能和干扰暴露。在同一回合上独立执行的参考轨迹提供了配对学习信号。理论分析将保留的证据和锚点质量与原型稳定性相关联，并建立了无偏在政策梯度估计的条件。在标准、细粒度和跨领域基准上的实验显示出最先进的准确性，平均比DVLA-RL提高了1.4%。项目页面可访问 https://peacelwh.github.io/TPAMI27-DVLA-RLpp/。

View on arXiv Download PDF AI Translation

cs.CV / 128 / 2610.12104

VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling

VINCIE-NExT：通过上下文建模实现从图像到视频的编辑解锁

Qu, Leigang, Cheng, Feng, Yang, Ziyan, Yang, Bangbang, Huang, Zhaoyang, Chow, Wei, Li, Yicong, Wang, Wenjie, Chua, Tat-Seng, Zeng, Yan

Abstract

Building a capable video editor remains significantly harder than a video generator: editing requires (source, instruction, edited) triplets that are prohibitively expensive to annotate and difficult to synthesize at scale, whereas image editing has already reached maturity with millions of such pairs readily available. In this work, we introduce VINCIE-NExT, a unified framework that transfers editing capability from images to videos through in-context visual demonstrations, alleviating the need for large-scale paired video editing data. VINCIE-NExT decomposes video editing into a structured chain of composable sub-tasks (Video -> Image -> Image -> Video), routing editing intent through the image domain and enabling scalable joint training from heterogeneous image and video corpora under a unified diffusion objective. An image editing pair, synthesized by the model or supplied by the user, is prepended as an in-context visual demonstration that serves as a spatial appearance blueprint for every output frame. To ground appearance edits across the interleaved context, we introduce a novel position encoding that links image demonstrations and video frames in a shared spatial coordinate system, enabling pixel-faithful propagation of appearance changes to every output frame. Chain-of-Editing further provides principled test-time scaling: by executing the sub-task chain as progressive diffusion stages, editing quality can be improved by investing additional compute without retraining. Comprehensive experiments on OpenVE-Bench demonstrate the state-of-the-art performance across diverse editing categories, with ablations confirming the effectiveness of each component.

Chinese Translation

构建一个功能强大的视频编辑器远比视频生成器困难得多：编辑需要（源视频、指令、编辑后视频）三元组，这类数据标注成本高昂且难以大规模合成，而图像编辑已相对成熟，拥有数百万此类配对数据。在本工作中，我们提出了VINCIE-NExT，一个统一框架，通过上下文视觉示范将图像的编辑能力迁移到视频上，缓解了对大规模配对视频编辑数据的需求。VINCIE-NExT将视频编辑分解为一系列可组合的结构化子任务链（视频 -> 图像 -> 图像 -> 视频），通过图像域传递编辑意图，并在统一的扩散目标下，实现了来自异构图像和视频语料库的可扩展联合训练。由模型合成或用户提供的图像编辑对作为上下文视觉示范被预置，作为每一输出帧的空间外观蓝图。为实现跨交错上下文的外观编辑定位，我们引入了一种新颖的位置编码，将图像示范与视频帧链接到共享的空间坐标系中，使外观变化能够像素级地传播到每一输出帧。Chain-of-Editing进一步提供了原则性的测试时扩展能力：通过将子任务链作为渐进式扩散阶段执行，可以通过增加计算投入而无需重新训练来提升编辑质量。在OpenVE-Bench上的全面实验表明，本方法在多样化编辑类别中实现了最先进的性能，消融实验验证了各组成部分的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 129 / 2610.12107

ContourVLA: A Closed-Loop Perception-Action Contour Policy for Generalized Referring Expression Segmentation

ContourVLA：一种用于广义指称表达分割的闭环感知-行动轮廓策略

Zhang, Ruicheng, Shen, Kaiwen, Hou, Jiaqi, Yang, Shuhan, Huang, Junchao, Zhang, Kewei, Zhou, Jun, Jiang, Li, Zhao, Shen

Abstract

Generalized referring expression segmentation (GRES) requires dynamically balancing high-level semantics for identifying a variable number of language-specified referents with fine-grained visual evidence for precise boundary delineation. This requirement challenges existing cascaded vision-language architectures, which typically rely on static feature interfaces and single-pass mask prediction, limiting adaptive perception and geometric correction. We introduce ContourVLA, a vision-language-action policy that recasts GRES as a closed-loop visuomotor process, in which editable contours serve as explicit policy states that condition multimodal perception and are updated by geometric action chunks. Evolution-Aware Semantic Scheduling (EASS) couples contour-guided bidirectional boundary sampling with state-conditioned routing of multilevel multimodal features, adapting perception to each contour state. Following supervised initialization, Dustbin-Augmented Entropic Credit Transport GRPO (DECT-GRPO) jointly optimizes discrete grounding and continuous contour actions with instance-level credits. Its rollout rewards and credits are derived from soft prediction-target correspondences that account for false positives and missed targets. ContourVLA improves gIoU over the strongest evaluated baselines by 8.7, 2.8, and 2.7 points on gRefCOCO val, testA, and testB, respectively, and achieves the highest mIoU across all eight RefCOCO, RefCOCO+, and RefCOCOg splits.

Chinese Translation

广义指称表达分割（GRES）需要动态平衡高层语义，以识别可变数量的语言指定指称对象，并结合细粒度的视觉证据以实现精确的边界划定。这一要求对现有的级联视觉-语言架构提出了挑战，这些架构通常依赖静态特征接口和单次掩码预测，限制了自适应感知和几何校正。我们提出了ContourVLA，一种将GRES重新构建为闭环视觉运动过程的视觉-语言-行动策略，其中可编辑的轮廓作为显式策略状态，调节多模态感知，并通过几何动作块进行更新。演化感知语义调度（EASS）将轮廓引导的双向边界采样与状态条件的多层多模态特征路由相结合，使感知适应每个轮廓状态。在监督初始化后，垃圾桶增强熵信用传输GRPO（DECT-GRPO）联合优化离散基础和连续轮廓动作，并使用实例级信用。其展开奖励和信用源自软预测目标对应关系，考虑了假阳性和漏检目标。ContourVLA在gRefCOCO验证集、测试集A和测试集B上分别比最强的评估基线提高了8.7、2.8和2.7个点的gIoU，并在所有八个RefCOCO、RefCOCO+和RefCOCOg分割中实现了最高的mIoU。

View on arXiv Download PDF AI Translation

cs.CV / 130 / 2610.12127

LVS: Local View Synthesis from Relative Camera Pose by Reusing Previous Views

LVS：通过重用先前视图基于相对相机位姿的局部视图合成

Huo, Qizhou, Sun, Xuan, Guo, Yongfei, Wang, Zhipeng, Gong, Yuanhao

Abstract

Interactive scene exploration requires frequent view updates, although small camera motions preserve much of the visible content. Conventional 3D Gaussian Splatting nevertheless renders each target view, leaving this image overlap unexploited. Reusing rendered images offers an alternative. Geometric warping alone cannot recover newly exposed content and remains sensitive to depth errors. We propose a per-scene framework that replaces repeated scene rendering for nearby views with relative-pose-guided RGB-D image reuse. Geometric warping uses depth and relative pose to transport source content, while a lightweight multiscale network predicts RGB residuals to correct artifacts and infer missing appearance. Cached source features further reduce repeated computation. On GS-render, residual refinement improves PSNR by 0.72~dB over pure warping; evaluations on captured and rendered scenes demonstrate low query latency. This separation of scene rendering from local view updates supports responsive scene exploration, with potential applications in augmented and virtual reality.

Chinese Translation

交互式场景探索需要频繁更新视图，尽管小幅相机运动保留了大部分可见内容。传统的三维高斯点渲染（3D Gaussian Splatting）仍需对每个目标视图进行渲染，未能利用图像重叠部分。重用已渲染图像提供了一种替代方案。单纯的几何变形无法恢复新暴露的内容，且对深度误差较为敏感。我们提出了一种基于每个场景的框架，用相对位姿引导的RGB-D图像重用替代邻近视图的重复场景渲染。几何变形利用深度和相对位姿传递源内容，而轻量级多尺度网络预测RGB残差以修正伪影并推断缺失的外观。缓存的源特征进一步减少重复计算。在GS-render数据集上，残差细化较纯变形方法提升了0.72 dB的峰值信噪比（PSNR）；在实拍和渲染场景上的评估表明查询延迟较低。该场景渲染与局部视图更新的分离支持响应式场景探索，具有增强现实和虚拟现实的潜在应用价值。

View on arXiv Download PDF AI Translation

cs.CV / 131 / 2610.12147

Healthy Counterfactual Generation via Diffusion Inpainting for Mammography Classification

通过扩散修复生成健康的反事实数据用于乳腺X光分类

Garcia, Inês Cruchinho, Mourão, Mariana, Calisto, Francisco Maria, Santiago, Carlos, Nascimento, Jacinto

Abstract

False negatives remain a critical limitation of computer-aided diagnosis (CAD) systems for breast cancer screening due to delayed detection and treatment. To address this issue, we propose a counterfactual data augmentation strategy that generates healthy mammograms by "erasing" lesions from anomalous images, thereby enriching the training distribution. We train a Denoising Diffusion Probabilistic Model on BI-RADS 1 (healthy) mammograms and use a RePaint-based sampling strategy to inpaint realistic normal tissue within annotated lesion bounding boxes. The resulting healthy counterfactuals replace annotated lesion regions with realistic healthy tissue while preserving patient-specific anatomical structure, as supported by similarity metrics between real and generated images. Image realism was further assessed by radiologists and found to be consistent with the original dataset quality. We evaluate counterfactual augmentation across four representative classifier architectures: a convolutional neural network (ConvNeXt), a vision transformer (ViT), a vision-language model pre-trained on mammogram-report pairs (Mammo-CLIP) and a multi-scale attention-based multiple-instance learning framework (FPN-MIL). Experiments conducted on the VinDr-Mammo dataset show improvements in sensitivity across all architectures, particularly at 80\% fixed specificity, contributing towards more reliable CAD systems for breast cancer. Code is available at: https://github.com/ines03garcia/diffusion-based-counterfactual-generation.

Chinese Translation

假阴性仍然是计算机辅助诊断（CAD）系统在乳腺癌筛查中的一个关键限制，主要由于延迟的检测和治疗。为了解决这一问题，我们提出了一种反事实数据增强策略，通过从异常图像中“抹去”病变生成健康的乳腺X光图像，从而丰富训练分布。我们在BI-RADS 1（健康）乳腺X光图像上训练了去噪扩散概率模型，并使用基于RePaint的采样策略在标注的病变边界框内修复现实的正常组织。生成的健康反事实图像用现实的健康组织替换了标注的病变区域，同时保留了特定患者的解剖结构，这得到了真实图像与生成图像之间相似性度量的支持。影像的真实性进一步由放射科医师评估，并发现与原始数据集的质量一致。我们在四种代表性的分类器架构上评估反事实增强：卷积神经网络（ConvNeXt）、视觉变换器（ViT）、在乳腺X光报告对上预训练的视觉-语言模型（Mammo-CLIP）以及基于多尺度注意力的多实例学习框架（FPN-MIL）。在VinDr-Mammo数据集上进行的实验显示，所有架构的灵敏度都有所提高，特别是在80%的固定特异性下，这为乳腺癌的更可靠的CAD系统做出了贡献。代码可在以下链接获取：https://github.com/ines03garcia/diffusion-based-counterfactual-generation。

View on arXiv Download PDF AI Translation

cs.CV / 132 / 2610.12156

Connected Self Forcing: Beyond Local Learning in Video Autoregression

连接自我强制：超越视频自回归中的局部学习

Zhang, Dongbin, Zheng, Chaoda, Chen, Kangjie, Li, Xiangyu, Chen, Shijia, Deng, Jinhao, Zhang, Yuqi, Jiang, Guangfeng, Lin, Hongbin, Wai, Choo Sin, Wang, Minqi, Wang, Puyi, Zhang, Jingye, Zhang, Yu, Liu, Xianming, Wang, Boyang

Abstract

To stream long videos while maintaining visual quality and temporal consistency, Self Forcing mitigates exposure bias through self-rollout training on self-generated histories with key-value (KV) caching. To keep memory manageable, it detaches historical caches, preserving forward dependencies between chunks but severing the backward gradient paths. We introduce Connected Self Forcing, a training framework that reconnects gradient paths across autoregressive chunks, allowing feedback from later predictions to guide how earlier context is generated. These connections go beyond historical KV-writing: gradients pass through generated latents into the computations that produced them, linking the generation of earlier context to its use in later predictions. To make this connected training memory-efficient, we develop shortcut gradient replay, which recovers cross-chunk gradients without retaining the full rollout computation graph. Integrated with distribution matching distillation, Connected Self Forcing trains historical chunks according to both their direct supervision and their contribution to subsequent generation. Experiments on autoregressive video generation show improvements in long-horizon visual quality and temporal consistency, without changing the inference procedure.

Chinese Translation

为了在保持视觉质量和时间一致性的同时流式传输长视频，自我强制通过对自生成历史进行自回归训练并结合关键值（KV）缓存来减轻曝光偏差。为了保持内存的可管理性，它分离了历史缓存，保留了块之间的前向依赖关系，但切断了后向梯度路径。我们提出了连接自我强制（Connected Self Forcing），这是一种训练框架，通过重新连接自回归块之间的梯度路径，允许后续预测的反馈指导早期上下文的生成。这些连接超越了历史KV写入：梯度通过生成的潜变量传递到产生它们的计算中，将早期上下文的生成与其在后续预测中的使用联系起来。为了使这种连接训练在内存使用上更为高效，我们开发了快捷梯度重放（shortcut gradient replay），该方法在不保留完整的回滚计算图的情况下恢复跨块梯度。结合分布匹配蒸馏（distribution matching distillation），连接自我强制根据历史块的直接监督和对后续生成的贡献进行训练。在自回归视频生成的实验中，长时间范围内的视觉质量和时间一致性得到了改善，而推理过程没有改变。

View on arXiv Download PDF AI Translation

cs.CV / 133 / 2610.12182

AI-Based On-Board Maritime Object Detection for Earth Observation Payload Data Reduction on Versal Embedded Hardware

基于人工智能的船载海洋目标检测用于地球观测有效载荷数据在Versal嵌入式硬件上的减少

Goudemant, Thomas, Bobey, Aurélien, Hlimi, Omar, Bellizzi, Marjorie

Abstract

Very-high-resolution Earth-observation satellites acquire more data than they can store and downlink, while in maritime surveillance the vessels cover a tiny fraction of each scene. We study onboard vessel detection as a way to select what is downlinked, which reduces the data according to its content rather than coding every pixel; it is complementary to conventional onboard compression. The work follows three axes. (i) Data and algorithm: a controlled dataset is generated from 68 annotated Maxar scenes with 43 vessel classes, and a YOLOX-S detector is trained on it. (ii) Embedded deployment: the detector is quantized and deployed on the DPU of a Versal VC1902, with a limited loss of detection quality and a processing time of a few seconds per scene. (iii) Data reduction: we propose several downlink modes, from metadata only (box, class and score of each detection) to image crops around vessels, tiles holding detections, or the whole scene with a degraded background, and estimate from the measured detection errors the trade-off each offers between the vessels kept and the volume downlinked. On our dense harbor and coastal scenes, tiles keep 98% of the vessels with 29% of the scene volume, and crops 83% with 3%.

Chinese Translation

超高分辨率地球观测卫星获取的数据量超过其存储和下行能力，而在海洋监视中，船只仅覆盖每个场景的一小部分。我们研究船载目标检测作为选择下行数据的一种方式，这种方法根据内容减少数据，而不是对每个像素进行编码；它与传统的船载压缩方法互为补充。该研究遵循三个方向。(i) 数据和算法：从68个带注释的Maxar场景生成一个受控数据集，包含43个船只类别，并在其上训练YOLOX-S检测器。(ii) 嵌入式部署：检测器经过量化并部署在Versal VC1902的DPU上，检测质量损失有限，每个场景的处理时间为几秒。(iii) 数据减少：我们提出几种下行模式，从仅包含元数据（每个检测的框、类别和得分）到围绕船只的图像裁剪、包含检测的瓦片，或整个场景的降级背景，并根据测量的检测误差估算每种模式在保留船只与下行数据量之间的权衡。在我们密集的港口和沿海场景中，瓦片保留了98%的船只，场景体积为29%；而裁剪则保留了83%的船只，场景体积仅为3%。

View on arXiv Download PDF AI Translation

cs.CV / 134 / 2610.12216

Stride Independent Patching for Deep Learning

独立于步幅的深度学习补丁技术

Rukundo, Olivier

Abstract

This paper presents semi-automatic stride-independent patching (SSP) as an alternative to automatic stride-dependent patching techniques. SSP uses user or expert input to position predefined patches over one or more objects of interest. To evaluate its effectiveness, three patch-based datasets were created using SSP, overlapping patching (Overlap), and non-overlapping patching (Noverlap). DeepLabV3+ models with ResNet50, ResNet18, and MobileNetV2 backbones were trained sepa-rately on each dataset. Quantitative evaluations were performed on the respective test splits and a common external test set. SSP generally achieved higher segmentation scores on the test splits and required the shortest model training time across all three backbones. On the external test set, SSP achieved the highest average precision and F1-score across backbones, whereas Noverlap achieved the highest average recall. These preliminary results demonstrate that the potentially greater spatial coverage of Noverlap and Overlap does not generally translate into better segmentation perfor-mance and that SSP offers a favorable balance between segmentation performance and model training time.

Chinese Translation

本文提出了一种半自动的独立于步幅的补丁技术（Semi-automatic Stride-independent Patching, SSP），作为自动依赖步幅的补丁技术的替代方案。SSP利用用户或专家的输入，将预定义的补丁放置在一个或多个感兴趣的对象上。为了评估其有效性，使用SSP、重叠补丁（Overlap）和非重叠补丁（Noverlap）创建了三个基于补丁的数据集。分别使用ResNet50、ResNet18和MobileNetV2作为骨干网络的DeepLabV3+模型在每个数据集上进行了单独训练。在各自的测试集和一个共同的外部测试集上进行了定量评估。SSP在测试集上通常获得了更高的分割得分，并且在所有三个骨干网络中所需的模型训练时间最短。在外部测试集上，SSP在各骨干网络中达到了最高的平均精度和F1-score，而Noverlap则获得了最高的平均召回率。这些初步结果表明，Noverlap和Overlap可能更大的空间覆盖并不一定转化为更好的分割性能，而SSP在分割性能和模型训练时间之间提供了良好的平衡。

View on arXiv Download PDF AI Translation

cs.CV / 135 / 2610.12229

VibeEdit: Image Editing with Canvas Instructions

VibeEdit：基于画布指令的图像编辑

Zhao, Jinjing, Wei, Fangyun, Wang, Yitong, Wu, Xiuyu, Chen, Yunuo, Yue, Yang, Zhang, Sirui, Wang, Wenbo, Zhang, Hongyang, Chen, Dong, Lu, Yan, Xu, Chang

Abstract

In text-guided image editing, describing the desired change is often straightforward, but identifying the intended object or region can be cumbersome, especially when several objects look alike. We introduce a new image editing interface that lets users place spatial marks and optional short notes directly on the image. Together, these annotations form a canvas instruction that specifies where to edit and what to change. Our editor, VibeEdit, follows these instructions to perform object addition, removal, replacement, attribute modification, and movement without a separate text prompt. We construct 1.55 million source-target edit pairs with object masks and structured edit descriptions, from which we render canvas instructions during training. We adapt Qwen-Image-Edit with layer-decoupled conditioning that separately encodes source images and canvas instructions for image editing. We train the model with region-weighted supervised fine-tuning, followed by rubric-guided reinforcement learning to improve edit completion, local edit quality, and preservation of unedited regions. We evaluate VibeEdit on an independently constructed, human-curated benchmark of 419 cases emphasizing target selection among similar objects. VibeEdit achieves a VLM rubric score of 79.9 and an outside-region PSNR of 32.8 dB, compared with 67.4 and 24.0 dB for FireRed, the highest-scoring text-instructed baseline in our evaluation.

Chinese Translation

在文本引导的图像编辑中，描述所需的变化通常是直接的，但识别目标对象或区域可能会很麻烦，尤其是当多个对象看起来相似时。我们介绍了一种新的图像编辑界面，允许用户直接在图像上放置空间标记和可选的简短注释。这些注释共同形成了一种画布指令，指定了编辑的位置和内容。我们的编辑器VibeEdit遵循这些指令进行对象的添加、移除、替换、属性修改和移动，而无需单独的文本提示。我们构建了155万个源-目标编辑对，包含对象掩码和结构化编辑描述，从中在训练过程中生成画布指令。我们采用了Qwen-Image-Edit模型，使用层解耦条件编码源图像和画布指令，以进行图像编辑。我们通过区域加权的监督微调训练模型，随后进行基于评分标准的强化学习，以提高编辑完成度、局部编辑质量和未编辑区域的保留。我们在一个独立构建的人类策划基准上评估VibeEdit，该基准包含419个案例，强调在相似对象中进行目标选择。VibeEdit在我们的评估中取得了79.9的VLM评分标准分数和32.8 dB的外部区域PSNR，而FireRed（我们评估中得分最高的文本指令基线）分别为67.4和24.0 dB。

View on arXiv Download PDF AI Translation

cs.CV / 136 / 2610.12230

From Prompting to Composing: A Spatial Canvas Interface for Poster Generation

从提示到创作：一种用于海报生成的空间画布界面

Wang, Yitong, Wei, Fangyun, Zhao, Jinjing, Zhang, Sirui, Zhang, Hongyang, Chen, Dong, Dai, Bo, Lu, Yan

Abstract

Text prompting is an indirect interface for poster generation, requiring users to encode inherently two-dimensional composition intent into a one-dimensional sequence of words. We introduce a Spatial Canvas Interface that enables users to directly compose generation intent in space through four complementary binding types: semantic, identity, text, and pixel, together with Text Specifications for individual elements and global appearance. Based on this interface, we develop Compo, a poster generation model adapted from a pretrained image editing model to understand Spatial Canvas inputs and Text Specifications. Compo supports both direct inference, where users explicitly construct the canvas, and agentic mode, where a high-level request is automatically translated into a planned Spatial Canvas. To train Compo, we develop a scalable pipeline that automatically constructs supervision data for different binding types and their combinations, enabling efficient adaptation without training a specialized poster generator from scratch. We further introduce a benchmark that evaluates adherence to individual binding types and their joint composition. Experiments show that Compo achieves stronger compositional controllability than both general-purpose image generation models and dedicated poster generation systems while maintaining high visual quality. By decoupling intent specification from visual generation, our work shifts poster generation from prompting toward composing.

Chinese Translation

文本提示是一种间接的海报生成接口，要求用户将固有的二维构图意图编码为一维的单词序列。我们提出了一种空间画布界面，使用户能够通过四种互补的绑定类型（语义、身份、文本和像素）直接在空间中构建生成意图，并结合针对各个元素和整体外观的文本规格。基于该接口，我们开发了 Compo，这是一种改编自预训练图像编辑模型的海报生成模型，能够理解空间画布输入和文本规格。Compo 支持直接推理，用户明确构建画布，以及代理模式，高级请求自动转换为计划的空间画布。为了训练 Compo，我们开发了一种可扩展的管道，能够自动构建不同绑定类型及其组合的监督数据，从而实现高效适应，而无需从头开始训练专门的海报生成器。我们进一步引入了一个基准，评估对各个绑定类型及其联合构图的遵循情况。实验表明，Compo 在构图可控性方面优于通用图像生成模型和专用海报生成系统，同时保持高视觉质量。通过将意图规范与视觉生成解耦，我们的工作将海报生成从提示转向创作。

View on arXiv Download PDF AI Translation

cs.CV / 137 / 2610.12256

Syn-Omni: Structured Specialization and Progressive Collaboration for Omnimodal Embeddings

Syn-Omni：用于全模态嵌入的结构化专业化与渐进协作

Oh, Youngtaek, Wu, Qiyu, Wakaki, Hiromi, Kim, Junmo, Mitsufuji, Yuki

Abstract

Omnimodal embeddings naturally involve both shared representations and modality-specific features across heterogeneous inputs. However, existing omnimodal embedding methods often rely on a single shared parameter space over mixed-modality data, limiting structural separation between universal and modality-specific representations. To address this, we propose Syn-Omni, a unified framework for structured omnimodal adaptation with modality specialization and controlled cross-modal collaboration. Specifically, we introduce Orthogonal Modality-Expert LoRA (OME-LoRA), which decomposes adaptation into a shared LoRA path for universal semantics and modality-expert LoRA paths for modality-aware specialization. Furthermore, Progressive Synergy Routing (PSR) enables experts to first establish modality-specific priors, then gradually interact with other modality-experts for cross-modal synergy. Evaluated across 81 diverse tasks spanning image, video, audio, and audiovisual modalities, Syn-Omni consistently outperforms omnimodal baselines, demonstrating the effectiveness of structured specialization and cross-modal progressive collaboration.

Chinese Translation

全模态嵌入自然涉及在异构输入中共享表示和特定模态特征。然而，现有的全模态嵌入方法通常依赖于在混合模态数据上使用单一的共享参数空间，这限制了通用表示与特定模态表示之间的结构分离。为了解决这一问题，我们提出了Syn-Omni，一个统一的框架，用于具有模态专业化和受控跨模态协作的结构化全模态适应。具体而言，我们引入了正交模态专家LoRA（Orthogonal Modality-Expert LoRA，OME-LoRA），它将适应过程分解为一个用于通用语义的共享LoRA路径和多个用于模态感知专业化的模态专家LoRA路径。此外，渐进协同路由（Progressive Synergy Routing，PSR）使专家首先建立特定模态的先验，然后逐渐与其他模态专家进行交互，以实现跨模态协同。在涵盖图像、视频、音频和视听模态的81个多样化任务中进行评估，Syn-Omni始终优于全模态基线，证明了结构化专业化和跨模态渐进协作的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 138 / 2610.12266

DVD: Dynamic Vector Decoding for Efficient MLLM-based Perception

DVD：用于高效MLLM基础感知的动态向量解码

Hou, Jinghua, Liu, Zhe, Zhao, Hengshuang

Abstract

Multimodal large language models have made remarkable progress in bridging vision and language, facilitating various perception tasks essential for human-machine interaction, robotics, and autonomous driving. However, existing MLLM-based perception methods predominantly rely on text-based coordinate representation, which suffers from excessive token overhead, or fixed-range quantization, which suffers from range and precision constraints, especially for 3D domains with unbounded spatial range and high localization accuracy requirements. To address these challenges, we propose a dynamic vector decoding method named DVD, which unifies the representation of 2D and 3D perception tasks. Specifically, we first transform diverse perceptual representation (i.e., 2D bounding boxes, 2D masks, and 3D bounding boxes) into 1D vector sequences, which are then mapped to compact discrete tokens in the high-dimensional space. Then, a lightweight de-tokenizer enables seamless integration with MLLMs by decoding output tokens back to original 2D and 3D perceptual representations. Extensive experiments on 2D and 3D perception benchmarks including RefCOCO series, SUN-RGBD, KITTI, Hypersim, nuScenes demonstrate that DVD achieves superior performance in 2D and 3D tasks and reduces significantly the token overhead and inference latency. DVD provides an efficient and general framework for integrating perception capabilities into MLLMs, overcoming the inherent limitations of existing methods.

Chinese Translation

多模态大语言模型在视觉与语言之间架起了显著的桥梁，促进了人机交互、机器人技术和自动驾驶等多种感知任务的发展。然而，现有的基于MLLM的感知方法主要依赖于基于文本的坐标表示，这种方法存在过多的标记开销，或者依赖于固定范围量化，这在空间范围不受限且对高定位精度要求的3D领域中面临范围和精度的限制。为了解决这些挑战，我们提出了一种名为DVD的动态向量解码方法，该方法统一了2D和3D感知任务的表示。具体而言，我们首先将多样的感知表示（即2D边界框、2D掩码和3D边界框）转换为1D向量序列，然后将其映射到高维空间中的紧凑离散标记。接着，一个轻量级的去标记器通过将输出标记解码回原始的2D和3D感知表示，实现与MLLM的无缝集成。在包括RefCOCO系列、SUN-RGBD、KITTI、Hypersim和nuScenes在内的2D和3D感知基准上的大量实验表明，DVD在2D和3D任务中表现出色，并显著减少了标记开销和推理延迟。DVD为将感知能力集成到MLLM中提供了一个高效且通用的框架，克服了现有方法的固有局限性。

View on arXiv Download PDF AI Translation

cs.CV / 139 / 2610.12282

Slot3R: Set-Associative Spatial Memory for Streaming 3D Reconstruction

Slot3R：用于流式3D重建的集合关联空间记忆

Zhang, Xiyuan, Yang, Yanming, Xu, Kaiyuan, Li, Ruibo, Zhang, Chi

Abstract

Streaming 3D reconstruction must preserve evidence from each frame while processing an expanding scene online. Spatial memory is a natural fit because it organizes history by reconstructed 3D location. Yet Point3R uses spatial proximity both to associate a new observation with an existing memory entry and to decide whether to fuse it, conflating co-location with state identity. Because pointers summarize image patches, nearby pointers may encode distinct surfaces, viewpoints, or visibility conditions; averaging them can destroy complementary evidence before later frames disambiguate it. We argue that location should determine address, not whether observations must merge. Slot3R realizes this principle as a training-free, set-associative retrofit that lets multiple states coexist at a shared address while keeping the pretrained Point3R backbone frozen. A bounded sparse readout further decouples persistent storage from per-frame decoder access. At 300-500 sampled frames, Slot3R reduces Point3R's point-cloud accuracy error (Acc) by 57.1%-63.1% on 7Scenes and 64.0%-72.0% on NeuralRGBD, lowers Sim(3)-aligned absolute trajectory error (ATE) on all three pose benchmarks, and remains competitive on video-depth estimation. It completes all evaluated settings from 600 to 1000 sampled frames at about 19 FPS under the same protocol, whereas Point3R and InfiniteVGGT run out of memory at 800 frames and beyond.

Chinese Translation

流式3D重建必须在在线处理扩展场景的同时保留每帧的证据。空间记忆自然适合这一任务，因为它通过重建的3D位置组织历史。然而，Point3R使用空间邻近性来将新的观察与现有记忆条目关联，并决定是否融合，从而将共位置与状态身份混淆。由于指针总结了图像块，附近的指针可能编码不同的表面、视点或可见性条件；对它们进行平均可能会在后续帧澄清之前破坏互补证据。我们认为，位置应决定地址，而不是观察是否必须合并。Slot3R实现了这一原则，作为一种无训练的集合关联改进，允许多个状态在共享地址处共存，同时保持预训练的Point3R主干不变。一个有界稀疏读出进一步将持久存储与每帧解码器访问解耦。在300-500个采样帧下，Slot3R将Point3R的点云精度误差（Acc）在7Scenes上降低了57.1%-63.1%，在NeuralRGBD上降低了64.0%-72.0%，并在所有三个姿态基准上降低了Sim(3)对齐的绝对轨迹误差（ATE），在视频深度估计中仍保持竞争力。它在相同协议下以约19 FPS完成从600到1000个采样帧的所有评估设置，而Point3R和InfiniteVGGT在800帧及以上时则出现内存不足。

View on arXiv Download PDF AI Translation

cs.CV / 140 / 2610.12299

Multi-Agent Egocentric World Model with Fine-Grained Embodied Interaction

多智能体自我中心世界模型与细粒度具身交互

Chung, Dahyun, Jin, Siyoon, Choi, Hyunwook, An, Honggyu, Seo, Junyoung, Kim, Hyunsung, Kim, Seung Wook, Kim, Seungryong

Abstract

Egocentric world models predict first-person observations conditioned on an agent's actions, but most focus on a single agent. Real embodied settings often involve multiple agents that act and interact within a shared environment. Existing multi-agent world models rely on coarse actions like locomotion, camera control, or discrete commands, leaving fine-grained embodied interactions underexplored. We formulate multi-agent egocentric world modeling as synchronized ego-stream generation for multiple agents interacting through fine-grained actions in a shared world. This requires cross-view action consistency, shared-environment consistency, and consistent propagation of interaction-induced state updates. We propose Multi-agent Egocentric World Model (ME-World), which jointly denoises multiple ego streams in a shared token sequence, conditions each stream on all agents' target-view poses, and grounds generation with shared environment memory. We train and evaluate on real and synthetic multi-agent data and introduce shared-world consistency metrics for environment, update, and identity consistency. Experiments show ME-World improves shared-world consistency, action control, identity preservation, and video quality over existing methods.

Chinese Translation

自我中心世界模型预测基于智能体行为的第一人称观察，但大多数模型仅关注单一智能体。真实的具身环境通常涉及多个智能体在共享环境中进行行动和交互。现有的多智能体世界模型依赖于粗略的动作，如移动、摄像机控制或离散指令，导致细粒度的具身交互未得到充分探索。我们将多智能体自我中心世界建模表述为多个智能体通过细粒度动作在共享世界中进行同步自我流生成。这要求跨视角动作一致性、共享环境一致性以及交互引发的状态更新的一致传播。我们提出了多智能体自我中心世界模型（ME-World），该模型在共享的标记序列中共同去噪多个自我流，基于所有智能体的目标视角姿态对每个流进行条件化，并利用共享环境记忆进行生成。我们在真实和合成的多智能体数据上进行训练和评估，并引入共享世界一致性指标以衡量环境、一致性更新和身份一致性。实验表明，ME-World在共享世界一致性、动作控制、身份保留和视频质量方面优于现有方法。

View on arXiv Download PDF AI Translation

cs.CV / 141 / 2610.12305

From What to Which: Decoding Modifier Grounding in Frozen MLLMs

从什么到哪个：解码冻结的多模态大语言模型中的修饰符基础

Corradini, Barbara Toniella, Gallegati, Caterina, Genovese, Ludovica, Murino, Vittorio

Abstract

As Multimodal Large Language Models (MLLMs) can describe increasingly complex visual scenes, token-level grounding becomes crucial. Yet, when an MLLM generates "the yellow banana on the left", established grounding approaches focus on what is in the image ("banana"), overlooking tokens that help describe which instance is meant ("yellow", "left"). In this work, we ask whether frozen MLLM representations contain decodable grounding information about the referred instance across generated tokens, extending to modifiers such as attributes, spatial expressions, and relational/action terms. To address this question, we introduce OTTER, a lightweight supervised probe over frozen MLLM representations that uses Optimal Transport (OT) to align generated tokens with visual regions and produce compact grounding maps. Our results show that (i) instance-discriminative visual information can be decoded from modifier tokens, with the clearest evidence for spatial terms, but (ii) is not confined to them, as contextualized object nouns also carry referential information; (iii) the recovered grounding remains informative under context perturbations, while selected regions remain relevant to generation; and (iv) the learned OT-based grounding extends beyond the controlled setting to free generation and cross-dataset transfer.

Chinese Translation

随着多模态大语言模型（MLLMs）能够描述越来越复杂的视觉场景，令牌级的基础变得至关重要。然而，当一个MLLM生成“左边的黄色香蕉”时，现有的基础方法关注于图像中的内容（“香蕉”），而忽视了帮助描述所指实例的令牌（“黄色”、“左边”）。在本研究中，我们探讨冻结的MLLM表示是否包含可解码的关于所指实例的基础信息，涵盖生成的令牌，包括属性、空间表达和关系/动作术语。为了解决这个问题，我们引入了OTTER，一个轻量级的监督探测器，基于冻结的MLLM表示，使用最优传输（Optimal Transport, OT）将生成的令牌与视觉区域对齐，并生成紧凑的基础图。我们的结果表明：（i）可以从修饰符令牌中解码出实例区分的视觉信息，空间术语的证据最为明显；（ii）这种信息并不限于空间术语，上下文化的对象名词同样携带指称信息；（iii）在上下文扰动下，恢复的基础信息仍然具有信息量，而所选区域仍然与生成相关；（iv）基于OT的学习基础超越了受控设置，适用于自由生成和跨数据集迁移。

View on arXiv Download PDF AI Translation

cs.CV / 142 / 2610.12307

BudgetPix: Compute-Adaptive Tokenization for Pixel-Space Image Diffusion

BudgetPix：针对像素空间图像扩散的计算自适应分词

Kara, Ozgur, Chen, Yujia, Watson, Daniel, Forsyth, David, Rehg, James Matthew, Chu, Wen-Sheng, Tran, Du

Abstract

Most image generation models rely on uniform tokenization, allocating the exact same computational budget to equally-sized image patches. This static paradigm cannot adapt to different resource constraints at inference time, and yields suboptimal quality-cost tradeoff by devoting the same effort to both plain backgrounds and intricate details. We propose BudgetPix, an adaptive tokenization framework that dynamically allocates compute based on visual complexity and spatial layout, enabling flexible computational budgeting at inference time. BudgetPix comprises three key components: (1) an adaptive encoder that maps a fixed-size image to a variable-length token sequence using an entropy-guided quadtree alongside a multi-scale patch embedder; (2) a scale-aware decoder reconstructs fixed-resolution images from multi-scale token sets; and (3) a flexible training and sampling schedule that enables pixel-space denoisers to operate across variable token counts. BudgetPix seamlessly integrates with existing pixel-space diffusion architectures, enabling a single checkpoint to be operated at a wide range of compute budgets. Evaluated on text-to-image generation, BudgetPix matches the fidelity of MiniT2I-L at $512^2$ and PixelDiT at $1024^2$ using just 25% of the original compute budget. In class-conditional generation using a MeanFlow backbone, BudgetPix requires merely 60% of the full compute budget to produce images with near-zero quality degradation, observing a marginal 0.8-point increase in FID. Comprehensive assessments by human and VLM judges confirm that BudgetPix establishes a significantly improved quality-efficiency tradeoff over prior budget-adaptive baselines. More details are available at our project page: https://karaozgur.com/BudgetPix

Chinese Translation

大多数图像生成模型依赖于统一的分词方法，为同等大小的图像块分配相同的计算预算。这种静态范式无法适应推理时不同的资源限制，并且通过对简单背景和复杂细节投入相同的努力，导致质量与成本的权衡不理想。我们提出了BudgetPix，一种自适应分词框架，根据视觉复杂性和空间布局动态分配计算，从而在推理时实现灵活的计算预算。BudgetPix包括三个关键组件：（1）一个自适应编码器，它使用基于熵的四叉树和多尺度补丁嵌入器将固定大小的图像映射为可变长度的令牌序列；（2）一个尺度感知解码器，从多尺度令牌集重建固定分辨率的图像；（3）一个灵活的训练和采样计划，使像素空间去噪器能够在可变令牌数量下运行。BudgetPix与现有的像素空间扩散架构无缝集成，使单个检查点能够在广泛的计算预算下运行。在文本到图像生成的评估中，BudgetPix在使用仅25%原始计算预算的情况下，达到了MiniT2I-L在$512^2$和PixelDiT在$1024^2$的保真度。在使用MeanFlow骨干网的类条件生成中，BudgetPix仅需60%的完整计算预算即可生成几乎无质量下降的图像，FID仅观察到0.8点的微小增加。人类和VLM评审的全面评估确认，BudgetPix在质量与效率的权衡上显著优于先前的预算自适应基线。更多细节请访问我们的项目页面：https://karaozgur.com/BudgetPix

View on arXiv Download PDF AI Translation

cs.CV / 143 / 2610.12310

Is In-Domain Training Enough for Fine-Grained Industrial Anomaly Understanding?

领域内训练是否足以实现细粒度工业异常理解？

Zhang, Xingwu, Du, Duanyang, Zhu, Huiling, Dai, Jiayue, Liu, Yixiao, Liu, Guozhi, Zhang, Zhihan, Long, Zijun

Abstract

A single multimodal large language model (MLLM) struggles to excel simultaneously at detection, localization, description, and reasoning in multimodal industrial anomaly understanding (MM-IAU). We show that in-domain training does not close this gap. On MMAD, a widely adopted MM-IAU benchmark, trained specialists reach at most 75.5% accuracy in defect localization, against 92.3% for human experts, and even detect anomalies less accurately than their untrained base model. Meanwhile, different MLLMs offer complementary strengths but share this weakness in fine-grained perception, so combining them alone cannot remove it. We therefore propose SiGMA, a spatially grounded multi-agent framework that divides labor between heterogeneous MLLM agents and a dedicated visual defect expert. A multimodal searcher supplies industrial knowledge and normal references, the defect expert turns query-reference comparison into calibrated anomaly evidence, and a label-free reliability controller weighs each source by task-wise competence and query-level evidence quality. SiGMA reaches 85.2% average accuracy on MMAD, 4.0% above the strongest trained specialist and Gemini-2.5-Pro and within 1.5% of human experts. Even with three agents of at most 9B parameters, it reaches 84.4%, and new MLLMs join without retraining.

Chinese Translation

单一的多模态大语言模型（MLLM）在多模态工业异常理解（MM-IAU）中，难以同时在检测、定位、描述和推理方面表现出色。我们表明，领域内训练并未缩小这一差距。在广泛采用的MM-IAU基准MMAD上，经过训练的专家在缺陷定位中的最高准确率仅为75.5%，而人类专家的准确率为92.3%，甚至在检测异常时的准确性也低于其未训练的基础模型。同时，不同的MLLM提供互补的优势，但在细粒度感知方面共享这一弱点，因此仅仅将它们结合在一起无法消除这一问题。因此，我们提出了SiGMA，一个空间基础的多智能体框架，该框架在异构MLLM代理和专门的视觉缺陷专家之间分配任务。一个多模态搜索器提供工业知识和正常参考，缺陷专家将查询-参考比较转化为经过校准的异常证据，而一个无标签的可靠性控制器根据任务能力和查询级别证据质量对每个来源进行加权。SiGMA在MMAD上达到了85.2%的平均准确率，比最强的训练专家和Gemini-2.5-Pro高出4.0%，并且与人类专家的差距仅为1.5%。即使在最多9B参数的三个代理下，它也达到了84.4%的准确率，并且新的MLLM可以在不重新训练的情况下加入。

View on arXiv Download PDF AI Translation

cs.CV / 144 / 2610.12316

Controllable Exaggeration for Generative Motion Models via Training-Time Adaptation and Inference-Time Guidance

通过训练时适应和推理时引导实现可控夸张的生成运动模型

Zamani, Amirhossein, Rampini, Arianna, Roy, Bruno

Abstract

Recent motion generative models have demonstrated strong capabilities in synthesizing physically plausible character motion, but often overlook established animation principles used by professional animators to ground and design their animation work. Understanding and incorporating these principles into motion generative pipelines is essential for producing motions that serve not only physically grounded applications but also the needs of the character animation community. This enables the creation of characters that not only move in physically plausible ways but also feel alive, expressive, and engaging. To close this gap, we focus on the Exaggeration principle of animation and investigate how it can be incorporated into modern motion generative pipelines to produce more expressive character motions. To this end, we introduce a framework that operates at two stages of existing motion generative pipelines. The first stage introduces exaggeration during training, where we perform supervised fine-tuning of pre-trained text-to-motion models on our curated exaggeration dataset. The second stage operates at inference time, where we: (i) introduce a mathematical formulation of exaggeration based on dynamic movement primitives (DMPs); and (ii) leverage this formulation as an exaggeration guidance signal to guide existing diffusion and flow-matching text-to-motion generation models toward exaggerated motion without additional training. Through qualitative and quantitative evaluations against three strong motion generation models, we show that our methods generate more exaggerated and expressive motions while preserving neutral reference motion intent and physical plausibility.

Chinese Translation

近期的运动生成模型在合成物理上合理的人物运动方面展现了强大的能力，但往往忽视了专业动画师在其动画创作中使用的既定动画原则。理解并将这些原则融入运动生成流程对于产生不仅适用于物理基础应用而且满足角色动画社区需求的运动至关重要。这使得创建的角色不仅能够以物理上合理的方式运动，还能让人感到生动、富有表现力和吸引力。为了解决这一问题，我们关注动画的夸张原则，并研究如何将其融入现代运动生成流程，以产生更具表现力的人物运动。为此，我们提出了一个在现有运动生成流程的两个阶段操作的框架。第一阶段在训练过程中引入夸张，我们对在我们精心策划的夸张数据集上进行监督微调的预训练文本到运动模型进行调整。第二阶段在推理时进行操作，我们：(i) 基于动态运动原语（DMPs）引入夸张的数学公式；(ii) 利用该公式作为夸张引导信号，引导现有的扩散和流匹配文本到运动生成模型朝向夸张运动发展，而无需额外训练。通过对比三种强大的运动生成模型进行定性和定量评估，我们展示了我们的方法在保持中性参考运动意图和物理合理性的同时，生成了更夸张和富有表现力的运动。

View on arXiv Download PDF AI Translation

cs.CV / 145 / 2610.12333

RiCo: Neural Simulation of Rigid-Body Interactions via Local Contact Reasoning

RiCo：通过局部接触推理进行刚体相互作用的神经模拟

Ouyang, Ruixiang, Qiao, Guanren, Meng, Fansen, Deng, Yueci, Jin, Ruixing, Jia, Kui, Liu, Guiliang

Abstract

Accurate simulation of rigid-body interactions is essential for predictive physical world models. Despite recent progress in modeling object dynamics, capturing how local contacts between surfaces shape object motion remains challenging. While end-to-end world models predict interactions across entire scenes or objects, in practice, rigid-body contact is inherently local, and only nearby surfaces can directly exchange contact forces. Motivated by this observation, we introduce Rigid-body Contact Reasoning (RiCo), which represents interactions between objects through sparse neighborhoods of contact surface points. RiCo combines each point's state with the relative geometry, motion, and physical properties of nearby surfaces, then reasons across the object's points to determine how these local contacts jointly affect its motion. By confining cross-object reasoning to nearby surfaces while propagating contact information within each rigid body, RiCo retains fine-grained interaction details without the cost of modeling every pair of scene points. Such properties enable RiCo a higher accuracy and contact fidelity. Experiments on MOVi-benchmark demonstrate that RiCo reduces 100-frame position and orientation errors by 31-35% and approximately 38%, respectively, compared with baselines. Moreover, RiCo achieves high contact fidelity, with ground-truth-relative penetration-time and mean-depth differences of 11.0% and 2.22 mm, respectively. RiCo further generalizes zero-shot from small-scale training scenarios to scenes containing 270 objects. Our real-world multi-ball collision experiments further provide preliminary evidence of sim-to-real transfer.

Chinese Translation

准确模拟刚体相互作用对于预测物理世界模型至关重要。尽管在建模物体动态方面取得了近期进展，但捕捉表面之间的局部接触如何影响物体运动仍然具有挑战性。虽然端到端的世界模型可以预测整个场景或物体之间的相互作用，但在实际应用中，刚体接触本质上是局部的，只有附近的表面才能直接交换接触力。基于这一观察，我们提出了刚体接触推理（Rigid-body Contact Reasoning，RiCo），通过稀疏的接触表面点邻域来表示物体之间的相互作用。RiCo将每个点的状态与附近表面的相对几何、运动和物理属性结合，然后在物体的点之间进行推理，以确定这些局部接触如何共同影响其运动。通过将跨物体的推理限制在附近表面，同时在每个刚体内传播接触信息，RiCo保留了细粒度的交互细节，而无需建模每对场景点。这些特性使RiCo在准确性和接触保真度上具有更高的表现。MOVi基准测试的实验表明，与基线相比，RiCo将100帧的位置和方向误差分别减少了31-35%和约38%。此外，RiCo实现了高接触保真度，真实值相对穿透时间和平均深度的差异分别为11.0%和2.22毫米。RiCo还进一步从小规模训练场景零-shot 泛化到包含270个物体的场景。我们的现实世界多球碰撞实验进一步提供了从模拟到现实转移的初步证据。

View on arXiv Download PDF AI Translation

cs.CV / 146 / 2610.12337

ContiLNN: Mitigating Slice Sampling Discontinuity with Liquid Neural Networks for Medical Image Restoration

ContiLNN：利用液态神经网络缓解切片采样不连续性以进行医学图像恢复

He, Jialei, Liu, Enhe, Song, Sifan, Jin, Pengfei, Su, Jionglong, Wang, Hongbin, Lu, Zhixiang, Huang, Yanhao, Cai, Anteng, Jiang, Zhengyong, Ding, Jiaman, Zhou, S. Kevin, Wang, Jinfeng

Abstract

Anatomical continuity provides complementary information for medical image restoration, but its use requires accounting for local anatomy and variations in slice sampling. We introduce ContiLNN, which augments two-dimensional restoration backbones with bidirectional closed-form continuous-time (Bi-CfC) modules for cross-slice modeling while retaining in-plane feature extraction. Slice-index intervals modulate gates determined by local features and hidden states, enabling propagation to respond to sampling variations without numerical ODE integration. Reference-guided consistency aligns first- and second-order cross-slice intensity differences to preserve anatomical variation, while distillation from a frozen backbone helps retain in-plane fidelity. Across five training seeds, ContiLNN improves mean PSNR over Restore-RWKV by 0.1907, 1.0176, and 1.2482 dB for CT denoising, MRI super-resolution, and reduced-count PET restoration, respectively, with lower RMSE in all three tasks. CT results are descriptive for one held-out patient. PET ablations support ordered propagation beyond additional pointwise capacity. Under contiguous training, Bi-CfC achieves higher fidelity than a Bi-GRU with similar parameter counts and arithmetic costs across all tested sampling conditions. Matched seven-slice profiling shows 52.8% lower latency and 57.0% lower peak GPU memory use than Bi-GRU. Mixed-gap training improves sparse and irregular-context performance for both operators, without a uniform ranking across metrics and contexts. Experiments with fewer training patients and a second backbone further support data efficiency and backbone compatibility.

Chinese Translation

解剖连续性为医学图像恢复提供了互补信息，但其使用需要考虑局部解剖结构和切片采样的变化。我们提出了ContiLNN，它通过双向闭式连续时间（Bi-CfC）模块增强二维恢复主干，以实现跨切片建模，同时保留平面内特征提取。切片索引间隔调制由局部特征和隐藏状态决定的门控，使得传播能够响应采样变化，而无需数值常微分方程（ODE）积分。参考引导一致性对齐一阶和二阶跨切片强度差异，以保持解剖变异，同时从冻结主干进行蒸馏有助于保持平面内的保真度。在五个训练种子中，ContiLNN在CT去噪、MRI超分辨率和减少计数PET恢复方面，分别比Restore-RWKV提高了0.1907、1.0176和1.2482 dB的平均PSNR，并在这三项任务中均表现出更低的均方根误差（RMSE）。CT结果对一个保留患者的描述性分析。PET消融实验支持超出额外逐点容量的有序传播。在连续训练下，Bi-CfC在所有测试的采样条件下实现了比具有相似参数数量和算术成本的Bi-GRU更高的保真度。匹配的七切片分析显示，Bi-CfC的延迟降低了52.8%，峰值GPU内存使用降低了57.0%。混合间隙训练提高了两种操作符在稀疏和不规则上下文中的性能，但在指标和上下文之间没有统一的排名。对较少训练患者和第二个主干的实验进一步支持数据效率和主干兼容性。

View on arXiv Download PDF AI Translation

cs.CV / 147 / 2610.12343

Reasoning-Informed Visual Editing

基于推理的视觉编辑

Yang, Xue, Zhang, Peiyuan, Zhu, Yilun, Yang, Qihao, Liu, Mingxin, Zhao, Xiangyu, Fan, Ziqian, Wang, Zhaokai, Li, Yan, Yang, Yifan, Yang, Xu, Jia, Xiaosong, Zhou, Yue, Zhong, Zhihang, Yan, Junchi

Abstract

Large Multi-modality Models (LMMs) have made significant progress in visual understanding and generation, but still face challenges in visual editing, particularly in following complex instructions, preserving appearance consistency, and supporting flexible input formats. To study this gap, we introduce RISEBench, the first benchmark for evaluating Reasoning-Informed viSual Editing (RISE), and extend it to RISEBench++, a more comprehensive and fine-grained benchmark for this emerging task. RISEBench++ extends the taxonomy into a hierarchical scheme spanning six reasoning dimensions: Temporal, Causal, Spatial, Logical, and Counterfactual Reasoning, together with Hybrid Reasoning integrating multiple reasoning types across multi-turn edits. These dimensions are further decomposed into 12 subcategories and 65 fine-grained task types. We expand input formats to include multi-image conditioning and scale the benchmark to 1000 human-annotated test cases, released in English and Chinese. We also improve our evaluation framework, assessing Instruction Reasoning, Appearance Consistency, and Visual Plausibility with human judges and an LMM-as-a-judge approach for more reliable and calibrated judgements. Beyond benchmarking, we introduce RISE-Agent, a training-free agentic framework integrating reasoning-driven planning, tool-augmented execution, and verifier-guided refinement, outperforming most strong existing approaches across diverse RISE tasks. We evaluate 58 visual editing approaches, including 34 open-source models, 19 closed-source models, and 5 agentic methods. The results reveal substantial challenges in reasoning-based visual editing, with even the strongest evaluated approach, GPT-Image-2.5 Sunburst, achieving only 56.6% accuracy. RISEBench++ highlights the limitations of contemporary editing models, provides insights, and indicates future directions for reasoning-aware visual editing.

Chinese Translation

大型多模态模型（LMMs）在视觉理解和生成方面取得了显著进展，但在视觉编辑方面仍面临挑战，特别是在遵循复杂指令、保持外观一致性和支持灵活输入格式方面。为了研究这一差距，我们引入了RISEBench，这是第一个用于评估基于推理的视觉编辑（Reasoning-Informed viSual Editing, RISE）的基准，并将其扩展为RISEBench++，这是一个更全面和细致的基准，针对这一新兴任务。RISEBench++将分类法扩展为一个层次结构，涵盖六个推理维度：时间推理、因果推理、空间推理、逻辑推理和反事实推理，以及整合多种推理类型的混合推理，适用于多轮编辑。这些维度进一步细分为12个子类别和65个细粒度任务类型。我们扩展了输入格式，包括多图像条件，并将基准规模扩大到1000个人工标注的测试案例，发布于英语和中文。我们还改进了评估框架，通过人类评审和LMM作为评审的方法评估指令推理、外观一致性和视觉合理性，以获得更可靠和校准的判断。除了基准测试，我们还引入了RISE-Agent，这是一个无训练的代理框架，集成了基于推理的规划、工具增强执行和验证者引导的优化，在各种RISE任务中超越了大多数现有强大方法。我们评估了58种视觉编辑方法，包括34个开源模型、19个闭源模型和5种代理方法。结果揭示了基于推理的视觉编辑面临的重大挑战，即使是评估的最强方法GPT-Image-2.5 Sunburst的准确率也仅为56.6%。RISEBench++突显了当代编辑模型的局限性，提供了见解，并指明了基于推理的视觉编辑的未来方向。

View on arXiv Download PDF AI Translation

cs.CV / 148 / 2610.12355

Distilling Routed 3D Privilege for Spatial Reasoning in Vision-Language Models

提炼用于空间推理的路由3D特权在视觉-语言模型中的应用

Li, Hongxing, Li, Yixin, Li, Dingming, Wang, Zixuan, Yan, Yuchen, Zhang, Wenqi, Lu, Weiming, Shen, Yongliang

Abstract

Spatial reasoning remains a persistent weakness of vision-language models (VLMs), because RGB inputs do not directly provide geometric evidence. Existing remedies either inject 3D into the model at inference, paying architecture and latency costs, or train with outcome rewards that supervise only the final answer. Spatial errors originate in perception: a misjudged depth or direction can be corrected only by the scene's true geometry, which the 3D-scanned sources of spatial training corpora already provide. We propose GPD (Geometry-Privileged Distillation), which makes geometric evidence the privilege in on-policy self-distillation (OPSD). For each question, depth, semantic, and bird's-eye-view (BEV) cues are rendered as compact text and routed to the teacher alongside the reference answer; a privileged KL, applied only to incorrect trajectories, augments GRPO, and the deployed model remains RGB-only. On the 4B backbone, GPD achieves 57.1 on VSI-Bench and 37.6 average across MindCube, SPARBench, MMSI-Bench, and ViewSpatial, outperforming both GRPO and answer-privileged OPSD across spatial reasoning benchmarks. Ablations confirm the complementarity of 3D and answer privilege, the advantage of question-conditioned routing over full-context injection, and the benefit of restricting distillation to incorrect trajectories.

Chinese Translation

空间推理仍然是视觉-语言模型（VLMs）的一个持续弱点，因为RGB输入并未直接提供几何证据。现有的解决方案要么在推理时将3D信息注入模型，付出架构和延迟的代价，要么通过仅监督最终答案的结果奖励进行训练。空间错误源于感知：错误判断的深度或方向只能通过场景的真实几何形状进行修正，而3D扫描的空间训练语料库已经提供了这一点。我们提出了GPD（几何特权蒸馏），使几何证据成为在线自蒸馏（OPSD）中的特权。对于每个问题，深度、语义和鸟瞰图（BEV）线索被渲染为紧凑的文本，并与参考答案一起路由到教师；仅对错误轨迹应用的特权KL增强了GRPO，而部署的模型仍然保持RGB输入。在4B骨干网络上，GPD在VSI-Bench上取得了57.1的成绩，并在MindCube、SPARBench、MMSI-Bench和ViewSpatial上平均达到37.6，超越了GRPO和答案特权OPSD在空间推理基准测试中的表现。消融实验确认了3D和答案特权的互补性、基于问题的路由相较于全上下文注入的优势，以及将蒸馏限制于错误轨迹的好处。

View on arXiv Download PDF AI Translation

cs.CV / 149 / 2610.12374

AgentGarten: Code Worlds for Evolving Agents

AgentGarten：用于进化代理的代码世界

Chi, Jiawei, Miao, Shangchen, Shi, Zhiyuan, Wu, Kailu, Wang, Hanyang, Chen, Weiliang, Dai, Qiyu, Ren, Jinshan, Gao, Jun, Long, Mingsheng, Duan, Yueqi, Lyu, Jiangran, Wu, Jialong, Liu, Fangfu

Abstract

Interactive virtual worlds allow agents to learn through exploration and interaction. What agents can learn is bounded by the environments they practice in, which must be faithful, with consistent state, rules, and dynamics, and realistic, with observations that follow the real-world visual distributions. Achieving both across diverse worlds remains a bottleneck. We introduce AgentGarten, a framework that couples simulators and game engines with a shared neural renderer to build real-time interactive environments. Its simulation backends maintain persistent world state and execute program-defined interaction rules, while the renderer generates visual observations from structured conditions exported through a common interface. To build the neural renderer, we adapt a pretrained video model to geometry conditions, distill it with our proposed Adversarial Forcing, and optimize inference for real-time interaction. Adversarial Forcing makes history prefilling differentiable through exact replay, so that losses on later predictions update how the renderer encodes prior observations, and adds real-data adversarial supervision to improve its visual quality. In AgentGarten, agents perceive the world through visual observations, interact with it in real time, and improve by distilling each round of experience into playbooks that subsequent agents inherit and refine. Our empirical study demonstrates a substantial gain in learning efficiency, with agents learning from just 4 rounds compared with millions for a conventional reinforcement learning counterpart. As new worlds can be written as code and rendered through the same interface, environments can scale in both number and difficulty alongside their agents, a step toward agents that keep evolving through interactive experience.

Chinese Translation

互动虚拟世界允许代理通过探索和互动进行学习。代理能够学习的内容受到其所处环境的限制，这些环境必须是可信的，具有一致的状态、规则和动态，并且是现实的，观察结果应遵循真实世界的视觉分布。在多样化世界中实现这两者仍然是一个瓶颈。我们提出了AgentGarten，一个将模拟器和游戏引擎与共享神经渲染器结合的框架，以构建实时互动环境。其模拟后端维护持久的世界状态并执行程序定义的互动规则，而渲染器则从通过公共接口导出的结构化条件生成视觉观察。为了构建神经渲染器，我们将预训练的视频模型适配于几何条件，通过我们提出的对抗强制（Adversarial Forcing）进行蒸馏，并优化推理以实现实时互动。对抗强制通过精确重放使历史预填充可微分，从而使后续预测的损失更新渲染器对先前观察的编码，并添加真实数据的对抗监督以提高其视觉质量。在AgentGarten中，代理通过视觉观察感知世界，实时与之互动，并通过将每轮经验蒸馏成后续代理继承和改进的剧本来提升自身能力。我们的实证研究表明，学习效率显著提升，代理仅需4轮学习即可获得与传统强化学习方法数百万轮相当的效果。由于新世界可以作为代码编写并通过相同接口渲染，环境可以在数量和难度上与其代理一起扩展，这是朝着通过互动经验不断进化的代理迈出的重要一步。

View on arXiv Download PDF AI Translation

cs.CV / 150 / 2610.12382

WorldAlign: Decoupled 4D Reward for World-Consistent Video Generation

WorldAlign：解耦的4D奖励用于世界一致性视频生成

He, Jing, Ding, Kaixin, Tian, Xingye, Shen, Guibao, Ge, Wenhang, Tao, Xin, Wan, Pengfei, Chen, Ying-Cong

Abstract

Faithful visual world simulation requires generated videos to maintain 4D world consistency, encompassing both static and dynamic consistency. Static consistency requires coherent 3D structure in static environments across viewpoints, while dynamic consistency requires plausible subject motion and consistent appearance over time. Geometry-aware post-training offers a promising way to improve world consistency. However, existing methods often rely on a static-scene assumption. Even those that accommodate dynamic scenes struggle to provide reliable static-consistency feedback, while dynamic consistency is often overlooked or inadequately assessed. To address these limitations, we introduce WorldAlign, a decoupled 4D reward framework that semantically separates static regions and dynamic subjects and provides feedback by aligning each with a world prior suited to its assumptions. For static regions, WorldAlign aligns static geometry with a geometric world prior through semantically guided masked reprojection, enabling more reliable static-consistency evaluation; an auxiliary camera-motion reward discourages nearly static solutions. For dynamic subjects, WorldAlign uses a strong vision-language model (VLM) as a dynamic world prior and constructs a VLM-as-a-judge reward based on sample-specific checklists that assess dynamicity, physical plausibility, shape, and texture consistency. This decoupled design enables more effective online post-training without requiring human preference annotations. Across two pretrained image-to-video generators, Wan2.1 and Wan2.2, WorldAlign jointly improves static and dynamic consistency over existing methods without suppressing overall or subject motion. These results support decoupled world-prior alignment for more faithful visual world simulation. Project page: https://worldalign.github.io/.

Chinese Translation

忠实的视觉世界模拟要求生成的视频保持4D世界一致性，包括静态一致性和动态一致性。静态一致性要求在不同视角下静态环境中的3D结构连贯，而动态一致性则要求主体运动合理且外观随时间一致。基于几何感知的后训练方法为提高世界一致性提供了一种有前景的途径。然而，现有方法往往依赖于静态场景假设。即使是那些适应动态场景的方法，也难以提供可靠的静态一致性反馈，而动态一致性常常被忽视或评估不足。为了解决这些局限性，我们提出了WorldAlign，一个解耦的4D奖励框架，它在语义上分离静态区域和动态主体，并通过将每个区域与适合其假设的世界先验对齐来提供反馈。对于静态区域，WorldAlign通过语义引导的掩蔽重投影将静态几何体与几何世界先验对齐，从而实现更可靠的静态一致性评估；辅助的相机运动奖励则抑制几乎静态的解决方案。对于动态主体，WorldAlign使用强大的视觉-语言模型（VLM）作为动态世界先验，并基于样本特定的检查清单构建VLM作为评判奖励，以评估动态性、物理合理性、形状和纹理一致性。这种解耦设计使得在线后训练更加有效，而无需人类偏好注释。在两个预训练的图像到视频生成器Wan2.1和Wan2.2上，WorldAlign在不抑制整体或主体运动的情况下，联合提高了静态和动态一致性。这些结果支持解耦的世界先验对齐，以实现更忠实的视觉世界模拟。项目页面：https://worldalign.github.io/

View on arXiv Download PDF AI Translation

cs.CV / 151 / 2610.12388

GenIA: Generative Reconstruction with Test-Time Input Alignment

GenIA：基于测试时输入对齐的生成重建

Esposito, Stefano, Pearl, Naama, Karpikova, Polina, Bulò, Samuel Rota, Porzi, Lorenzo, Kontschieder, Peter, Geiger, Andreas, Luiten, Jonathon

Abstract

Reconstructing complete 3D object assets from monocular or sparse multi-view observations remains challenging. Generative 3D foundation models can complete object geometry beyond the observed views, but their predictions may not faithfully reproduce the observed geometry, appearance, or pose. We introduce GenIA, a framework for test-time input-aligned generation that grounds SAM3D's generative prior in geometric and photometric observations without retraining the foundation model. We improve object pose by deriving translation and scale from geometry while retaining the learned rotation prior, and align appearance through visibility-biased attention, cross-observation fusion, and differentiable rendering guidance during denoising. An optional post-denoising refinement further adapts the appearance latent, lightweight decoder adapters, and object placement to the observations. Our framework also supports externally supplied geometry; when given temporal shapes of dynamic objects, it recovers a shared, input-aligned canonical appearance and stable world-space placement. Across synthetic and real benchmarks, GenIA improves pose prediction and object reconstruction from monocular, multi-view, and dynamic inputs, outperforming recent optimization-based, per-frame image-to-3D, and video-to-4D methods. Our project page is available at https://facebookresearch.github.io/GenIA.

Chinese Translation

从单目或稀疏多视角观测中重建完整的3D物体资产仍然具有挑战性。生成性3D基础模型可以在观察到的视角之外完成物体几何形状，但它们的预测可能无法忠实再现观察到的几何形状、外观或姿态。我们提出了GenIA，一个用于测试时输入对齐生成的框架，它在不重新训练基础模型的情况下，将SAM3D的生成先验与几何和光度观测相结合。我们通过从几何中推导平移和缩放来改善物体姿态，同时保留学习到的旋转先验，并通过可见性偏向注意力、跨观测融合和在去噪过程中使用可微渲染指导来对齐外观。可选的去噪后精细化进一步调整外观潜变量、轻量级解码器适配器和物体放置以适应观测。我们的框架还支持外部提供的几何形状；当给定动态物体的时间形状时，它能够恢复共享的、输入对齐的规范外观和稳定的世界空间放置。在合成和真实基准测试中，GenIA提高了从单目、多视角和动态输入中进行姿态预测和物体重建的性能，超越了最近基于优化的逐帧图像到3D和视频到4D的方法。我们的项目页面可在 https://facebookresearch.github.io/GenIA 找到。

View on arXiv Download PDF AI Translation

cs.CV / 152 / 2610.12399

SpaceFlow: Locally Controllable 3D Generation

SpaceFlow：局部可控的三维生成

De La Fuente, Neil, Lafuente, Joan, Sayfiddinov, Mukhammadali, Scharitzer, Felicia, Pollefeys, Marc, Celen, Ata, Sarkar, Sayan Deb, Fedele, Elisabetta

Abstract

Current 3D generation methods lack explicit local control: geometric adherence is often defined by a global control strength, and appearance cannot be specified locally. We present SpaceFlow, a training-free pipeline for locally controllable 3D generation from text descriptions and a collection of geometric primitives. Each primitive serves as a proxy for an object part and is assigned a local control level, enabling users to specify whether regions should strictly follow the input shape or allow generative completion. During structure generation, we enforce these spatial constraints within the generative flow process. For appearance synthesis, the generated structure is segmented and matched to the primitives. Each generated part is conditioned only on its assigned text or image cue, thereby limiting cross-part leakage. Regional geometry metrics demonstrate that SpaceFlow preserves the specified geometry in high-control regions and enables plausible shape variation in low-control areas. A user study further indicates that the resulting balance between geometric fidelity and generative freedom remains competitive in overall quality. When evaluating appearance on fixed geometry, text-conditioned routing achieves state-of-the-art prompt faithfulness and color/material accuracy. Qualitative results additionally show localized routing of image cues. The project page is available at SpaceFlow3D.github.io.

Chinese Translation

当前的三维生成方法缺乏明确的局部控制：几何遵循通常由全局控制强度定义，而外观无法在局部进行指定。我们提出了SpaceFlow，一个无需训练的管道，用于从文本描述和几何原件集合中进行局部可控的三维生成。每个原件作为对象部分的代理，并被分配一个局部控制级别，使用户能够指定区域是否应严格遵循输入形状或允许生成补全。在结构生成过程中，我们在生成流过程中强制执行这些空间约束。对于外观合成，生成的结构被分割并与原件匹配。每个生成的部分仅依赖于其分配的文本或图像提示，从而限制了跨部分的信息泄漏。区域几何度量表明，SpaceFlow在高控制区域中保持了指定的几何形状，并在低控制区域中实现了合理的形状变化。一项用户研究进一步表明，几何保真度与生成自由度之间的平衡在整体质量上仍具竞争力。在固定几何体上评估外观时，基于文本的路由实现了最先进的提示忠实度和颜色/材料准确性。定性结果还显示了图像提示的局部路由。项目页面可访问于SpaceFlow3D.github.io。

View on arXiv Download PDF AI Translation

cs.CV / 153 / 2610.12402

SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models

SpaceCast-Bench：评估视觉语言模型中的预测空间推理能力

Li, Hongxing, Su, Jinyue, Li, Dingming, Zhang, Wenqi, Lu, Weiming, Xiao, Jun, Zhuang, Yueting, Shen, Yongliang

Abstract

Existing spatial reasoning benchmarks mainly test spatial perception: reading off relations already visible in the input. Yet real-world spatial intelligence demands predictive spatial reasoning: constructing a scene from observations, anticipating how an intervention changes it, and reasoning about the unseen outcome. We introduce SpaceCast-Bench, the first benchmark to directly and diagnostically evaluate this capability. Built around an observe-transform-infer framework, its 3,862 questions from 182 real-world scenes span 16 task types at three levels: static perception, local prediction, and global prediction, progressively requiring scene understanding, spatial state updating, and relational inference over unobserved outcomes. Evaluating 21 models exposes a stark gap: the strongest model reaches only 58.0% against 87.2% human performance, while spatially specialized models remain near random chance. Controlled analyses further reveal that bridge views are critical for integrating distributed observations, and that explicit 3D evidence benefits models more reliably than generated outcome images or videos. Fine-tuning on our programmatically generated data lifts Qwen3-VL-4B from 34.0% to 65.7% with macro-average gains across six out-of-domain benchmarks.

Chinese Translation

现有的空间推理基准主要测试空间感知：从输入中读取已可见的关系。然而，现实世界中的空间智能要求预测空间推理：从观察中构建场景，预测干预如何改变场景，并推理未见结果。我们介绍了SpaceCast-Bench，这是第一个直接且诊断性地评估这一能力的基准。该基准围绕观察-转化-推理框架构建，包含来自182个真实场景的3,862个问题，涵盖三个层次的16种任务类型：静态感知、局部预测和全局预测，逐步要求场景理解、空间状态更新和对未观察结果的关系推理。对21个模型的评估揭示了明显的差距：最强模型仅达到58.0%的准确率，而人类表现为87.2%，而空间专用模型的表现接近随机猜测。控制分析进一步揭示，桥接视图对于整合分散的观察至关重要，而显式的3D证据比生成的结果图像或视频更可靠地提升模型性能。在我们的程序生成数据上进行微调，使得Qwen3-VL-4B的准确率从34.0%提升至65.7%，在六个领域外基准测试中实现了宏观平均增益。

View on arXiv Download PDF AI Translation

cs.CV / 154 / 2610.12403

ViSkill: Reinforcing VLM Agents with Evolving Visual-Native Skills

ViSkill：通过不断演变的视觉原生技能增强视觉语言模型代理

Li, Hongxing, Li, Dingming, Li, Yixin, Du, Yong, Zhang, Wenqi, Lu, Weiming, Xiao, Jun, Zhuang, Yueting, Shen, Yongliang

Abstract

Skill-augmented agents improve sample efficiency by distilling successful trajectories into reusable strategies. Yet most existing approaches remain text-centric, linearizing spatial layouts and action-state correspondences into language that loses critical geometric structure. Recent efforts have begun incorporating visual evidence, but construct and update skills separately from policy optimization, leaving their mutual improvement underexplored. We propose ViSkill, a visual-native skill learning framework that encodes successful interactions as composite visual skill cards directly accessible to VLM agents. Retrieved skills guide both inference and reward shaping, while successful trajectories are distilled back into the library, forming a closed feedback loop in which skill accumulation and policy improvement reinforce each other. An optional cold-start mechanism further accelerates early-stage learning. Evaluated on Sokoban, FrozenLake, and PrimitiveSkill, ViSkill achieves an overall success rate of 0.89, rising to 0.91 with cold-start initialization, outperforming all evaluated proprietary and open-source baselines while converging faster than standard PPO. Our code is available at https://github.com/ZJU-REAL/ViSkill.

Chinese Translation

技能增强代理通过将成功的轨迹提炼为可重用的策略来提高样本效率。然而，大多数现有方法仍然以文本为中心，将空间布局和动作-状态对应关系线性化为丧失关键几何结构的语言。近期的努力开始纳入视觉证据，但在政策优化之外单独构建和更新技能，导致它们的相互改善未得到充分探索。我们提出了ViSkill，一个视觉原生技能学习框架，将成功的交互编码为可直接供视觉语言模型（VLM）代理访问的复合视觉技能卡。检索到的技能指导推理和奖励塑造，而成功的轨迹则被提炼回库中，形成一个技能积累与政策改进相互强化的闭环反馈机制。可选的冷启动机制进一步加速早期学习。在Sokoban、FrozenLake和PrimitiveSkill上的评估中，ViSkill的整体成功率达到0.89，冷启动初始化后上升至0.91，超越了所有评估的专有和开源基线，并且收敛速度快于标准的PPO。我们的代码可在https://github.com/ZJU-REAL/ViSkill获取。

View on arXiv Download PDF AI Translation

cs.CV / 155 / 2610.12412

WorldCast: Distributed Multiplayer World Models

WorldCast：分布式多人世界模型

Ye, Ziyang, Huang, Junchao, Zhang, Evelyn, Xie, Zhihao, Zhang, Ruicheng, Han, Boyao, Ban, Litao, Wang, Ziye, Hu, Xinting, Shi, Shaoshuai, Tian, Zhuotao, Jiang, Li

Abstract

Multiplayer world models must generate independently controlled views with consistent representations of both players and their shared environment. Most existing approaches coordinate multiple players through joint multi-view generation, whose cost grows with each additional player. We present WorldCast, a distributed multiplayer world model in which each player runs a local client comprising a video generator and a state model. Using recorded player positions and map geometry during training, the state model estimates the player's position from generated video and control inputs. Clients exchange player states and project them into camera-aligned player state fields that guide where and how other players are rendered. Shared scene state enables clients to reuse one another's generated observations to maintain consistent scene appearance across views. Experiments on Counter-Strike 2 demonstrate WorldCast's consistency, real-time performance, and distributed scalability. The camera-aligned player state field improves player rendering rates by over an order of magnitude over joint-generation methods, while shared scene state improves visual consistency over whole rounds. Each client runs in real time and exchanges only player and scene states, enabling scalable multiplayer generation without a centralized computational bottleneck. Image quality remains stable over hour-long rollouts.

Chinese Translation

多人世界模型必须生成独立控制的视图，并对玩家及其共享环境保持一致的表示。现有的大多数方法通过联合多视图生成来协调多个玩家，其成本随着每增加一个玩家而增加。我们提出了WorldCast，一种分布式多人世界模型，其中每个玩家运行一个本地客户端，该客户端包括视频生成器和状态模型。在训练过程中，状态模型利用记录的玩家位置和地图几何信息，从生成的视频和控制输入中估计玩家的位置。客户端交换玩家状态，并将其投影到与相机对齐的玩家状态场中，以指导其他玩家的渲染方式和位置。共享场景状态使客户端能够重用彼此生成的观察结果，从而在不同视图之间保持一致的场景外观。在《反恐精英2》的实验中，WorldCast展示了一致性、实时性能和分布式可扩展性。与联合生成方法相比，相机对齐的玩家状态场提高了玩家渲染速率超过一个数量级，而共享场景状态则提高了整个回合的视觉一致性。每个客户端实时运行，仅交换玩家和场景状态，从而实现可扩展的多人生成，无需集中计算瓶颈。图像质量在长达一小时的回合中保持稳定。

View on arXiv Download PDF AI Translation

cs.CV / 156 / 2610.12416

MAMHOI: Factorizing Scene-Aware Human-Object Interaction through Affordances

MAMHOI：通过可供性分解场景感知的人物-物体交互

Lei, Mingyuan, Sung, Yoonchang, Cham, Tat-Jen

Abstract

Generating realistic human-object interactions (HOI) in complex 3D scenes requires two complementary capabilities: reasoning about interaction feasibility in the environment and synthesizing realistic human-object motion. However, supervision for these capabilities is rarely available jointly at scale. Human-scene datasets provide rich information about environment-aware motion, while human-object datasets capture detailed interaction dynamics, yet paired human-object-scene data remain scarce. We present MAMHOI, an affordance-mediated factorization for scene-aware human-object interaction generation. MAMHOI factorizes scene-aware HOI generation through an explicit motion-affordance interface between scene understanding and motion synthesis: a scene-conditioned model first predicts where and how an interaction can be feasibly executed, and an affordance-conditioned HOI model then generates the corresponding human-object motion. This factorization allows scene understanding and interaction dynamics to be learned from complementary sources of supervision without requiring paired human-object-scene data. Experiments in complex indoor environments show that MAMHOI reduces object--scene penetration while better preserving human--object interaction quality, yielding more realistic and physically feasible scene-aware interactions. Project page: https://leimingyuan.github.io/MAMHOI-project-page/

Chinese Translation

在复杂的3D场景中生成真实的人物-物体交互（HOI）需要两种互补的能力：推理环境中交互的可行性和合成真实的人物-物体运动。然而，这些能力的监督数据在规模上很少同时可用。人类-场景数据集提供了关于环境感知运动的丰富信息，而人类-物体数据集则捕捉了详细的交互动态，但配对的人类-物体-场景数据仍然稀缺。我们提出了MAMHOI，一种通过可供性介导的场景感知人类-物体交互生成的分解方法。MAMHOI通过场景理解与运动合成之间的显式运动-可供性接口来分解场景感知的HOI生成：一个场景条件模型首先预测交互可以在何处以及如何可行地执行，然后一个可供性条件的HOI模型生成相应的人物-物体运动。这种分解允许从互补的监督来源中学习场景理解和交互动态，而不需要配对的人类-物体-场景数据。在复杂的室内环境中的实验表明，MAMHOI减少了物体-场景的穿透，同时更好地保持了人类-物体交互的质量，从而产生更真实和物理上可行的场景感知交互。项目页面：https://leimingyuan.github.io/MAMHOI-project-page/

View on arXiv Download PDF AI Translation

cs.CV / 157 / 2610.12417

WOVEN: Weaving Visual World Modeling into Multimodal LLMs

WOVEN：将视觉世界建模融入多模态大语言模型

Fan, Zheyu, Zhang, Yue, Deng, Mingkai, Wang, Kangrui, Wang, Qineng, Chen, Canyu, Hao, Jie, Fan, Xing, Guo, Chenlei, Xing, Eric P., Bansal, Mohit, Li, Manling

Abstract

Multimodal large language models (MLLMs) struggle with spatial, embodied, physical, and temporal reasoning. We hypothesize that these failures reflect a shared deficit in visual transition reasoning, and test whether this capability can serve as a shared training primitive, one that different models can learn from different supervision sources and reuse across different tasks, with a systematic training recipe. Existing benchmarks document these deficits separately but do not support controlled comparisons across scenes, actions, and reasoning operations. We therefore introduce WOVEN, a training source and benchmark for visual transition reasoning that organizes transition supervision by scene, action, and reasoning type, using diverse, realistic rollouts from video-pretrained generative models: 36,076 examples across 20 scene types, 5 action types, and 8 reasoning types. We first evaluate 38 frontier MLLMs (e.g., GPT-5.4 and Qwen3-VL-235B-A22B) and find a substantial and systematic deficit: even the strongest models fall far below humans, and the failures recur across model families and persist with scale. We then train MLLMs at multiple scales on WOVEN and find that they learn a shared capability that transfers broadly: training subsets of only about 2,000 items each collectively improve 22 of 26 external benchmarks by up to 27.3 percentage points, and WOVEN data can replace 30-50% of a task's own training data with comparable accuracy. Controlled comparisons further yield a training recipe for visual world modeling, validated prospectively on held-out benchmarks: select supervision by the reasoning operation it teaches rather than by the actions, scenes, or domains it shows, and prefer larger changes to the visual state for robustness. Our work establishes visual transition reasoning as a reusable foundation for systematic visual world-model training in MLLMs.

Chinese Translation

多模态大语言模型（MLLMs）在空间、具身、物理和时间推理方面存在困难。我们假设这些失败反映了视觉转移推理的共同缺陷，并测试这种能力是否可以作为一种共享的训练原语，使不同模型能够从不同的监督源中学习并在不同任务中重用，采用系统的训练方案。现有基准分别记录了这些缺陷，但不支持在场景、动作和推理操作之间进行受控比较。因此，我们引入了WOVEN，这是一个用于视觉转移推理的训练源和基准，通过场景、动作和推理类型组织转移监督，使用来自视频预训练生成模型的多样化、真实的样本：涵盖20种场景类型、5种动作类型和8种推理类型的36,076个示例。我们首先评估了38个前沿的MLLM（例如，GPT-5.4和Qwen3-VL-235B-A22B），发现存在显著且系统的缺陷：即使是最强的模型也远低于人类，并且这些失败在模型家族之间反复出现，并随着规模的扩大而持续。然后，我们在WOVEN上以多种规模训练MLLM，发现它们学习到了一种广泛迁移的共享能力：仅约2,000个项目的训练子集共同提高了26个外部基准中的22个，提升幅度高达27.3个百分点，并且WOVEN数据可以以可比的准确性替代任务自身训练数据的30-50%。受控比较进一步提供了一种视觉世界建模的训练方案，在保留基准上前瞻性验证：根据所教授的推理操作选择监督，而不是根据所展示的动作、场景或领域，并优先选择对视觉状态的较大变化以增强鲁棒性。我们的工作确立了视觉转移推理作为MLLM中系统视觉世界模型训练的可重用基础。

View on arXiv Download PDF AI Translation

cs.CV / 158 / 2610.12419

OneSearch-VL: Unified Multimodal Deep Research Agent for Image and Video

OneSearch-VL：统一的多模态深度研究代理用于图像和视频

Li, Hongyu, Zhang, Manyuan, Feng, Kaituo, Chen, Shu, Zheng, Dian, Li, Hao, Yu, Hao, Chen, Zhangquan, Guo, Zoey, Zhang, Ray, Huang, Shaofei, Hui, Tianrui, Huang, Linjiang, Liu, Si

Abstract

Single-image, multi-image, and video deep research require different visual operations but share a workflow of visual grounding, external retrieval, and fact composition. A key challenge is to preserve the dependencies linking localized visual anchors, entity relations, source-supported facts, and answer-producing operations. We introduce OneSearch-VL, a unified agent centered on the Visually Grounded Evidence Graph (VGEG), which encodes these dependencies as a shared task-level reference for data construction, process supervision, and operation-level evaluation. Our VGEG-based data engine constructs and verifies multi-image and video questions and filters expert trajectories. Using these data, we assemble OneSearch-VL-SFT-110K and OneSearch-VL-RL-10K for SFT and RL, respectively. We further derive the Evidence-aware Visual-Grounded Rubric reward (EVGR) from VGEG annotations to supervise evidence traceability and visual grounding during RL. For fine-grained evaluation, we construct OneSearch-MI-Bench and OneSearch-Video-Bench, organizing questions by the research operations encoded in their VGEGs. Experiments show that OneSearch-VL-8B improves over Qwen3-VL-8B with tool access by 20.2 and 17.6 percentage points on the two new benchmarks, respectively, while also achieving substantial gains across 7 image benchmarks and VideoDR. Project repository: https://github.com/appletea233/OneSearch-VL

Chinese Translation

单图像、多图像和视频的深度研究需要不同的视觉操作，但共享视觉定位、外部检索和事实组合的工作流程。一个关键挑战是保持连接局部视觉锚点、实体关系、来源支持的事实和产生答案的操作之间的依赖关系。我们引入了OneSearch-VL，一个以视觉基础证据图（Visually Grounded Evidence Graph, VGEG）为中心的统一代理，它将这些依赖关系编码为数据构建、过程监督和操作级评估的共享任务级参考。我们的基于VGEG的数据引擎构建并验证多图像和视频问题，并过滤专家轨迹。利用这些数据，我们分别组装了OneSearch-VL-SFT-110K和OneSearch-VL-RL-10K用于监督微调（SFT）和强化学习（RL）。我们进一步从VGEG注释中推导出证据感知的视觉基础评分奖励（Evidence-aware Visual-Grounded Rubric reward, EVGR），以监督强化学习过程中的证据可追溯性和视觉定位。为了进行细粒度评估，我们构建了OneSearch-MI-Bench和OneSearch-Video-Bench，按其VGEG中编码的研究操作组织问题。实验表明，OneSearch-VL-8B在工具访问方面在两个新基准上分别比Qwen3-VL-8B提高了20.2和17.6个百分点，同时在7个图像基准和VideoDR上也取得了显著的提升。项目仓库：https://github.com/appletea233/OneSearch-VL

View on arXiv Download PDF AI Translation

cs.CV / 159 / 2610.12421

Beyond Spatio-Temporal Priors: A Generalizable Approach for Dense Correspondence Matching

超越时空先验：一种可推广的密集对应匹配方法

Liu, Luping, Kang, Bingyi, Wang, Yifan, Xu, Dong

Abstract

Dense correspondence matching has historically been bounded by simplifying spatio-temporal priors, such as smooth motion and rigid geometry. While effective for classical tasks, these assumptions break down in image editing and reference-guided generation (IEG), where transformations can preserve visual identity while breaking physical continuity. To establish identity-preserving correspondence across such transformations, we introduce FreeMatching, a generalizable framework combining generative and semantic foundation representations with heterogeneous supervision from classical datasets, tracked videos, and synthetic scenes. Teacher-guided iterative refinement further improves correspondence in IEG without dense correspondence annotations. Experimentally, a single FreeMatching model substantially improves correspondence quality on challenging IEG image pairs while retaining competitive performance on classical benchmarks. Furthermore, we demonstrate its utility as a quantitative metric for evaluating identity preservation, with scores that correlate with human judgment. The code is available at https://github.com/luping-liu/FreeMatching.

Chinese Translation

密集对应匹配历史上受到简化时空先验的限制，如平滑运动和刚性几何。虽然这些假设在经典任务中有效，但在图像编辑和参考引导生成（IEG）中，这些假设失效，因为变换可以在保持视觉身份的同时打破物理连续性。为了在此类变换中建立保持身份的对应关系，我们提出了FreeMatching，这是一种可推广的框架，结合了生成和语义基础表示，并利用来自经典数据集、跟踪视频和合成场景的异构监督。教师引导的迭代优化进一步改善了IEG中的对应关系，而无需密集对应注释。实验表明，单个FreeMatching模型在具有挑战性的IEG图像对上显著提高了对应质量，同时在经典基准测试中保持了竞争力。此外，我们展示了它作为评估身份保持的定量指标的实用性，其得分与人类判断相关联。代码可在 https://github.com/luping-liu/FreeMatching 获取。

View on arXiv Download PDF AI Translation

cs.CV / 160 / 2610.12423

Pumpire: Unified Benchmark for Metric Distance Estimation

Pumpire：统一的度量距离估计基准

Chen, Siyu, Wang, Zehan, Xu, Jiayang, Wu, Yihan, Wang, Jialei, Chen, Junming, Zhang, Ziang, Ying, Yutong, Zhao, Zhou

Abstract

We present Pumpire, a unified benchmark for evaluating metric point-pair distance estimation capability of both image- and video-level 3D foundation models, with or without depth priors. In contrast to previous approaches that normally evaluate depth and camera intrinsics separately or evaluate point-clouds with geometric similarity metrics, which cannot directly reflect models' point-to-point distance estimation capability, Pumpire directly assesses point-to-point distances from the reconstructed geometry. To this end, we collect a large-scale and diverse dataset (pumpire-6k) comprising 100 real-world scenes, each annotated with physically measured point-pair distances and containing 64 frames, for a total of 6,400 frames. Building on this dataset, we establish a holistic evaluation protocol that covers both image- and video-level 3D foundation models and enables direct assessment of point-pair distance errors and cross-setting comparison. We conduct extensive experiments across 29 baseline configurations of representative 3D foundation models and provide a comprehensive analysis of the results. By offering this benchmark, we target the more fundamental ability to perceive and estimate physical scale in the reconstructed 3D space, which prior evaluation protocols have largely overlooked. The project page can be found at https://pumpire.github.io/

Chinese Translation

我们提出了Pumpire，一个统一的基准，用于评估图像和视频级3D基础模型在有或没有深度先验情况下的度量点对距离估计能力。与以往通常分别评估深度和相机内参或使用几何相似性度量评估点云的方法不同，这些方法无法直接反映模型的点对点距离估计能力，Pumpire直接评估从重建几何体得到的点对点距离。为此，我们收集了一个大规模且多样化的数据集（pumpire-6k），该数据集包含100个真实场景，每个场景都标注了物理测量的点对距离，并包含64帧，总计6,400帧。在此数据集的基础上，我们建立了一个全面的评估协议，涵盖图像和视频级3D基础模型，并能够直接评估点对距离误差和跨设置比较。我们在29个代表性3D基础模型的基线配置上进行了广泛的实验，并对结果进行了全面分析。通过提供这个基准，我们旨在关注在重建的3D空间中感知和估计物理尺度的更基本能力，而这一点在以往的评估协议中往往被忽视。项目页面可访问 https://pumpire.github.io/

View on arXiv Download PDF AI Translation

cs.CV / 161 / 2610.12427

FastBench: Can Streaming VLMs Perceive High-Dynamic Real-World Streams?

FastBench：流媒体 VLM 能否感知高动态真实世界流?

Hu, Yuxuan, Shi, Weikang, Bo, Yang, Lu, Xudong, Guo, Xintong, Li, Shuhan, He, Yuyang, Guan, Huankang, Sun, Peiwen, Yang, Yunqiao, Li, Wenbo, Liu, Rui, Li, Hongsheng

Abstract

Streaming Video Large Language Models (VLMs) enable continuous video understanding, yet existing benchmarks focus on low-dynamic scenarios. Under bounded context budgets, models must balance temporal history, spatial resolution, and temporal granularity; sparse sampling at 1--2 FPS misses fast events. We introduce FastBench to evaluate high-dynamic perception in real-world video streams. Its trajectory-grounded pipeline combines QA generation from high-FPS clips, filtering of questions answerable at 2 FPS, answer verification using SAM3 and CoTracker3 trajectories, and three rounds of human inspection. FastBench contains 306 QA pairs across eight domains, six capabilities, and forward, instant, and backward temporal scopes, with human-annotated evidence intervals. We also present ProactiveFrame, a training-free baseline that adjusts incoming frame rates through text tokens. A dual-tier sliding window retains recent high-FPS observations while downsampling older ones into sparse history. Experiments reveal substantial limitations: the strongest model, Gemini-3.5-Flash, scores only 50.7%. Denser sampling improves Qwen3-VL-8B from 32.9% at 2 FPS to 44.6% at 24 FPS, but gains saturate as history is compressed. ProactiveFrame outperforms sparse uniform sampling by 5.4 and 1.5 percentage points, yet remains well below oracle-guided focusing, showing that current VLMs struggle to determine from the stream alone when finer temporal perception is needed. FastBench provides a testbed for high-dynamic streaming video understanding. Code and data: https://github.com/Ashone3/FastBench.

Chinese Translation

流媒体视频大型语言模型（VLMs）能够实现连续的视频理解，但现有基准主要集中在低动态场景下。在有限的上下文预算下，模型必须平衡时间历史、空间分辨率和时间粒度；以 1-2 FPS 的稀疏采样会错过快速事件。我们提出了 FastBench，以评估真实世界视频流中的高动态感知。其轨迹基础的流程结合了从高 FPS 剪辑生成的问答（QA），过滤出在 2 FPS 下可回答的问题，使用 SAM3 和 CoTracker3 轨迹进行答案验证，以及三轮人工检查。FastBench 包含来自八个领域的 306 对问答，涵盖六种能力，以及前向、即时和后向的时间范围，并附有人工标注的证据区间。我们还提出了 ProactiveFrame，这是一种无需训练的基线，通过文本标记调整输入帧率。双层滑动窗口保留最近的高 FPS 观察，同时将较旧的观察降采样为稀疏历史。实验揭示了显著的局限性：最强的模型 Gemini-3.5-Flash 仅得分 50.7%。更密集的采样使 Qwen3-VL-8B 从 2 FPS 的 32.9% 提升至 24 FPS 的 44.6%，但随着历史的压缩，增益趋于饱和。ProactiveFrame 比稀疏均匀采样高出 5.4 和 1.5 个百分点，但仍远低于由 oracle 指导的聚焦，显示出当前 VLM 在仅凭流媒体判断何时需要更精细的时间感知方面存在困难。FastBench 为高动态流媒体视频理解提供了一个测试平台。代码和数据： https://github.com/Ashone3/FastBench。

View on arXiv Download PDF AI Translation

cs.CV / 162 / 2610.12442

LEGO: A Lifting-Free Approach for Exocentric-to-Egocentric Video Generation

LEGO：一种无提升方法的外中心到自中心视频生成

Cho, Suhwan, Choi, Yonwoo, Kim, Soongjin, Park, Jicheol, Lim, Taegyu

Abstract

Generating an egocentric video from a single exocentric recording is a challenging case of novel view synthesis, as the two cameras share little overlap and much of the target view is unobserved. Current state-of-the-art methods reconstruct the scene explicitly by estimating depth, lifting the video into a point cloud, and re-rendering it from the egocentric camera to condition a video diffusion model. This deterministic mapping assigns each pixel to a single reprojected location, which preserves texture but translates depth errors into misplaced content. We ask what a video diffusion model should receive as its condition and propose a lifting-free answer: a learned view synthesizer, an LVSM-style transformer fine-tuned to render the egocentric view directly without depth, point clouds, or reprojection, resolving cross-view correspondence internally. In contrast, its probabilistic mapping averages each region over candidate source locations according to a learned correspondence distribution, preserving structure while fine texture is averaged away. We argue that this trade-off suits a diffusion generator, whose denoising training excels at restoring detail, so an effective condition should prioritize structural alignment over sharpness. This distribution's concentration also yields a per-region confidence, used both to mask low-confidence regions and to guide the generator toward high-confidence areas during early layout-forming denoising steps. Our approach consistently outperforms the state-of-the-art explicit pipeline and generalizes to other datasets without retraining. The synthesizer thus supplies view structure, and the diffusion model its detail.

Chinese Translation

从单个外中心录制生成自中心视频是新视角合成中的一个挑战性案例，因为两个摄像机之间的重叠很少，目标视图的大部分是未观察到的。目前的最先进方法通过估计深度、将视频提升为点云，并从自中心摄像机重新渲染来显式重建场景，以此为视频扩散模型提供条件。这种确定性映射将每个像素分配到一个单一的重投影位置，虽然保留了纹理，但将深度错误转化为错误的内容。我们探讨视频扩散模型应接收何种条件，并提出一种无提升的答案：一个学习的视图合成器，一个经过微调的LVSM风格变换器，能够直接渲染自中心视图，而无需深度、点云或重投影，从而在内部解决跨视图对应关系。相比之下，其概率映射根据学习到的对应分布对候选源位置的每个区域进行平均，保留结构的同时细节纹理被平均掉。我们认为这种权衡适合扩散生成器，其去噪训练在恢复细节方面表现出色，因此有效的条件应优先考虑结构对齐而非清晰度。该分布的集中性还产生了每个区域的置信度，用于掩蔽低置信度区域，并在早期布局形成去噪步骤中引导生成器朝向高置信度区域。我们的方法在性能上始终优于最先进的显式管道，并且在不重新训练的情况下能够推广到其他数据集。因此，合成器提供视图结构，而扩散模型则提供细节。

View on arXiv Download PDF AI Translation

cs.CV / 163 / 2610.12448

One Block, Multiple Depths: Recurrent Vision Transformers with Depth-Programmed Experts

一个块，多重深度：具有深度编程专家的递归视觉变换器

Bulat, Adrian, Ouali, Yassine, Tzimiropoulos, Georgios

Abstract

In this work, we show that a single Transformer block, applied recurrently, can match the accuracy of a full-depth vision encoder at comparable inference FLOPs without intermediate feature distillation. reViT restores depth-specific transformations by representing the FFN at each recurrent depth as a convex combination of a small shared expert bank. A continuous normalized-depth coordinate programs this mixture, defining a resampleable trajectory through FFN parameter space. We evaluate this design in two regimes: supervised ImageNet-1k training and distillation from a DINOv2 teacher. Across both regimes, controlled adaptations identify weight-space merging as the strongest tested MoE family at a matching one-FFN budget, ahead of the token-dispatch and output-mixture alternatives. Trained from scratch, reViT-B/16 attains DeiT III accuracy with about 70\% fewer stored parameters. An 8-experts model distilled using only the teacher's output features retains nearly all of its DINOv2 teacher's linear-probe accuracy and transfers across classification, segmentation, and depth prediction. Elastic-depth training allows one checkpoint (trained model) to operate at multiple tested depths by resampling the same normalized coordinate interval. For fixed-depth deployment, the recurrent block can be materialized as a conventional dense graph, removing online routing and merging without changing the one-FFN-per-depth compute but expanding deployment storage.

Chinese Translation

在本研究中，我们展示了一个单一的变换器块，通过递归应用，可以在不进行中间特征蒸馏的情况下，达到与全深度视觉编码器相当的准确性，同时推理的浮点运算量（FLOPs）也相当。reViT通过将每个递归深度的前馈网络（FFN）表示为一小组共享专家库的凸组合，恢复了深度特定的变换。一个连续的归一化深度坐标对这一组合进行编程，定义了通过FFN参数空间的可重采样轨迹。我们在两个领域评估了这一设计：监督的ImageNet-1k训练和从DINOv2教师进行的蒸馏。在这两个领域中，受控适应性识别出权重空间合并是以匹配的单FFN预算测试的最强MoE（专家混合）家族，优于令牌调度和输出混合的替代方案。从零开始训练的reViT-B/16以大约70%的存储参数达到了DeiT III的准确性。使用仅教师输出特征蒸馏的8个专家模型几乎保留了其DINOv2教师的线性探测准确性，并在分类、分割和深度预测中实现了迁移。弹性深度训练允许一个检查点（训练模型）通过重采样相同的归一化坐标区间在多个测试深度下运行。对于固定深度的部署，递归块可以被实现为传统的稠密图，去除在线路由和合并，而不改变每个深度的单FFN计算，但扩展了部署存储。

View on arXiv Download PDF AI Translation

cs.CV / 164 / 2610.12451

VersaCamVLA: Camera-Configurable VLA Policies for Robotic Manipulation

VersaCamVLA：用于机器人操作的相机可配置VLA策略

Han, Boyao, Shi, Chen, Qian, Jingjing, Tian, ZhuoTan, Jiang, Li

Abstract

Vision-Language-Action (VLA) models have emerged as powerful foundations for robotic manipulation, but their reliance on fixed camera configurations during training makes them brittle to changes in camera count or pose during deployment. To overcome these limitations, we propose VersaCamVLA, a camera-configurable framework that decouples camera-set representation from action learning. VersaCamVLA learns a unified scene-token interface that maps an arbitrary, variable set of posed RGB views into fixed-size latent scene tokens. This is achieved via multi-signal target-view prediction and Wrist-Augmented Pose Sampling (WAPS), which leverages natural wrist-camera motion for free pose diversity. At deployment, a lightweight spatial encoder injects these compact scene tokens into a pretrained base VLA as a supplementary visual condition, requiring no explicit 3D sensing or novel-view rendering. Experiments on RoboTwin, LIBERO, and a real-robot platform demonstrate that VersaCamVLA consistently outperforms prior VLA methods and direct multi-view baselines, maintaining robust performance across varying camera counts and unseen camera poses.

Chinese Translation

视觉-语言-动作（VLA）模型已成为机器人操作的强大基础，但它们在训练过程中对固定相机配置的依赖使其在部署时对相机数量或姿态的变化变得脆弱。为克服这些限制，我们提出了VersaCamVLA，这是一种相机可配置的框架，将相机集表示与动作学习解耦。VersaCamVLA学习一个统一的场景标记接口，将任意的、可变的姿态RGB视图映射为固定大小的潜在场景标记。这是通过多信号目标视图预测和增强手腕姿态采样（WAPS）实现的，后者利用自然的手腕-相机运动实现自由姿态的多样性。在部署时，一个轻量级空间编码器将这些紧凑的场景标记注入到预训练的基础VLA中，作为补充视觉条件，且无需显式的3D感知或新视图渲染。在RoboTwin、LIBERO和真实机器人平台上的实验表明，VersaCamVLA始终优于之前的VLA方法和直接的多视图基线，在不同相机数量和未见相机姿态下保持稳健的性能。

View on arXiv Download PDF AI Translation

cs.CV / 165 / 2610.12458

OmniCapBench: A Deep-Structured Evaluation Framework for Fine-Grained Audio-Visual Captioning

OmniCapBench：一种用于细粒度音视频字幕的深度结构评估框架

Yang, Zhongyu, Tao, Jiale, Chen, Ruitao, Yang, Zuhao, Yuan, Yingfang, Zhao, Xueliang, Auden, Wang, Kai, Shao, Shuai, Wang, Biao, Yves, Steve, Lu, Qinglin

Abstract

Multimodal large language models (MLLMs) are rapidly evolving toward continuous audio--visual reasoning, creating an urgent need for evaluations that expose their capability limits. Audio--visual captioning is an ideal diagnostic task, yet current benchmarks face a coupled trade-off: whole-caption scores provide coverage without localization, local probes provide localization without coverage, and unconstrained LLM judges introduce instability. We introduce OmniCapBench (Omni-Video Caption Benchmark), a benchmark that reframes audio--visual caption evaluation as a deep-structured diagnostic framework. OmniCapBench shifts the prediction target from free-form text to sets of atomic, verifiable evaluation units across three tracks: entity references, visual shots, and audio events, enabling reliable scoring with deterministic constraint checks and localized LLM-based semantic comparisons. With 786 densely annotated videos, OmniCapBench effectively distinguishes MLLM perception errors, including temporal grounding failures, identity drift, cross-modal misalignment, and hallucinated descriptions. Evaluating frontier MLLMs reveals strong local perception but weak long-horizon audio--visual reasoning, particularly in identity drift and cross-modal misalignment, providing a fine-grained roadmap for omnimodal development.

Chinese Translation

多模态大型语言模型（MLLMs）正在迅速发展，朝着连续的音视频推理方向迈进，这使得评估其能力极限的需求变得迫切。音视频字幕生成是一个理想的诊断任务，但当前的基准面临着一个耦合的权衡：整体字幕评分提供了覆盖但缺乏定位，局部探测提供了定位但缺乏覆盖，而不受限制的LLM评判引入了不稳定性。我们提出了OmniCapBench（Omni-视频字幕基准），一个将音视频字幕评估重新框定为深度结构诊断框架的基准。OmniCapBench将预测目标从自由形式文本转移到三个轨道上的原子、可验证评估单元集合：实体引用、视觉镜头和音频事件，从而实现通过确定性约束检查和基于LLM的语义比较进行可靠评分。通过786个密集注释的视频，OmniCapBench有效区分了MLLM的感知错误，包括时间定位失败、身份漂移、跨模态不对齐和虚构描述。对前沿MLLM的评估揭示了强大的局部感知能力，但在长时间音视频推理方面较弱，特别是在身份漂移和跨模态不对齐方面，为全模态发展提供了细粒度的路线图。

View on arXiv Download PDF AI Translation

cs.CV / 166 / 2610.12459

WorldGuide: Goal-Directed Video World Model for Procedural Task Execution

WorldGuide：面向目标的视频世界模型用于程序性任务执行

Deria, Ankan, Kumar, Komal, Cholakkal, Hisham, Khan, Fahad Shahbaz, Khan, Salman

Abstract

Video generators and video-based world models can synthesize plausible visual trajectories, but long-horizon procedural tasks require generation to adapt to what has actually been produced. A model must determine the next action from its generated state, execute that action, and recognize when the task is complete. Open-loop generation cannot adapt to execution outcomes, while existing closed-loop systems often rely on pretrained executors or indirect verification. This leaves a gap between deciding an action and successfully realizing it. We formulate procedural video generation as \emph{closed-loop task execution in visual world space} and introduce \textbf{WorldGuide}. Given only an initial image and a task goal, WorldGuide predicts an atomic action, generates its corresponding video clip, and uses the generated result to select the next action or terminate. The Planner and Executor are trained on the same step-level procedural demonstrations: the Planner learns to predict the next atomic action or task completion from visual progress, while the Executor is directly trained to realize the predicted actions. Hierarchical visual memory maintains state across long-horizon execution with bounded history token cost. Due to the lack of step-level action-video supervision for joint planner-executor training, we introduce \textbf{WorldGuide Bench}: approximately 59K step-annotated videos across 245 tasks and 27 procedural categories. WorldGuide achieves a 33.33\% Task Success on \textbf{WorldGuide-Bench}, compared with 29.90\% for the strong recent video model MiniMax-H3, even though MiniMax-H3 receives reference action plans, and achieves 47.69\% on \textbf{VideoCraft-Bench} compared with 32.73\% for MiniMax-H3 under goal-only conditioning. These results demonstrate the importance of coupling planning with learned execution for goal-directed procedural video generation.

Chinese Translation

视频生成器和基于视频的世界模型能够合成可信的视觉轨迹，但长时间跨度的程序性任务要求生成能够适应实际产生的结果。模型必须从其生成的状态中确定下一步动作，执行该动作，并识别任务何时完成。开放式生成无法适应执行结果，而现有的闭环系统通常依赖于预训练的执行器或间接验证。这在决定动作和成功实现动作之间留下了空白。我们将程序性视频生成形式化为 extit{视觉世界空间中的闭环任务执行}，并引入 extbf{WorldGuide}。仅给定初始图像和任务目标，WorldGuide预测一个原子动作，生成相应的视频片段，并利用生成的结果选择下一步动作或终止。规划器（Planner）和执行器（Executor）在相同的步骤级程序性演示上进行训练：规划器学习从视觉进展中预测下一步原子动作或任务完成，而执行器则直接训练以实现预测的动作。分层视觉记忆在长时间执行中维护状态，同时限制历史令牌成本。由于缺乏步骤级动作-视频监督以进行联合规划器-执行器训练，我们引入了 extbf{WorldGuide Bench}：大约59K个步骤标注视频，涵盖245个任务和27个程序性类别。WorldGuide在 extbf{WorldGuide-Bench}上实现了33.33 ext{%}的任务成功率，而强大的近期视频模型MiniMax-H3的成功率为29.90 ext{%}，尽管MiniMax-H3接收参考动作计划，并且在 extbf{VideoCraft-Bench}上实现了47.69 ext{%}的成功率，而MiniMax-H3在仅基于目标的条件下为32.73 ext{%}。这些结果表明，将规划与学习执行相结合对于面向目标的程序性视频生成的重要性。

View on arXiv Download PDF AI Translation

cs.CV / 167 / 2610.12461

OuroWorld: Bringing Any 3D World Alive as Diverse, Endlessly Looping 3D Cinemagraphs

OuroWorld：将任何三维世界转化为多样化、无尽循环的三维动态影像

Xie, You-Zhe, Chou, Ting-Wei, Li, Yu-Hsuan, Zhang, Kaipeng, Wang, Zhixiang, Liu, Yu-Lun

Abstract

Recent 3D world models generate photorealistic, explorable scenes that remain frozen in time. OuroWorld is a mask-free framework that turns any static 3D Gaussian Splatting scene into a 3D cinemagraph: a dynamic scene with vivid, diverse motion looping seamlessly from any viewpoint. A vision-language model infers plausible dynamics and guides a video model to synthesize a reference video, which we lift and complete into multi-view videos. To learn from this imperfect supervision, we propose Inconsistency-Robust Periodic 4DGS: a Fourier-series deformation field guarantees looping by construction, while a Grounded Drift Field anchored at the reference view absorbs cross-view inconsistency. Unlike prior Eulerian methods limited to fluid-like motion, we capture general deformation, object motion, and illumination change. We introduce a ground-truth-free evaluation covering vividness, naturalness, loop seam coherence, and scene quality. On 39 reconstructed and generated scenes, OuroWorld outperforms all baselines and wins 70.8%-99.0% of user-study comparisons. Project page: https://ouroworld.userwei.com

Chinese Translation

近期的三维世界模型生成了可探索的逼真场景，这些场景在时间上保持静止。OuroWorld 是一个无遮罩框架，它将任何静态的三维高斯点云场景转化为三维动态影像：一个动态场景，具有生动、多样的运动，从任何视角无缝循环。视觉-语言模型推断出合理的动态，并指导视频模型合成参考视频，我们将其提升并补全为多视角视频。为了从这种不完美的监督中学习，我们提出了不一致性鲁棒的周期性 4DGS：一个傅里叶级数变形场通过构造保证循环，而锚定在参考视图的基础漂移场吸收了视角间的不一致性。与先前限于流体状运动的欧拉方法不同，我们捕捉到一般变形、物体运动和光照变化。我们引入了一种无真实标签的评估方法，涵盖生动性、自然性、循环接缝一致性和场景质量。在39个重建和生成的场景中，OuroWorld 超越了所有基线，并在用户研究比较中赢得了70.8%-99.0%的胜率。项目页面：https://ouroworld.userwei.com

View on arXiv Download PDF AI Translation

cs.CV / 168 / 2610.12464

What 30,000 Hours of Ego-centric Video Does Not Teach

30,000小时自我中心视频无法教会我们的内容

Dong, Jiahua, Bagchi, Anurag, Jangir, Yash, Irshad, Muhammad Zubair, Zakharov, Sergey, Hebert, Martial, Bharadhwaj, Homanga, Wang, Yu-Xiong, Guizilini, Vitor Campagnolo, Tokmakov, Pavel

Abstract

World models offer a promising alternative to physics-based simulators, yet remain far from practical deployment. We ask how far scaling ego-centric human video takes them, using a dataset of 30,000 hours spanning over 1,000 scene types and 14,000 contributors. Rather than relying on opaque downstream metrics, we directly evaluate agent and object-interaction fidelity on a challenging out-of-distribution benchmark. Increasing training data by 100x improves both, but unevenly: the agent is modeled well, while object fidelity remains far lower and improves slowly. We show that the agent gains need not come from data, and a careful visual conditioning design saturates fidelity with a fraction of it, which lets us measure object fidelity on its own and discover its saturation point. We then introduce a supervision scheme that shifts capacity from scene appearance toward object dynamics, improving object fidelity though a substantial gap remains. Finally, our conclusions transfer to downstream humanoid modeling. Overall, our results suggest that scaling ego-centric data brings agent modeling close to its limit while leaving its effects on the world far behind, and that closing this gap will depend on how models are trained, not only on how much data they see.

Chinese Translation

世界模型为基于物理的模拟器提供了一个有前景的替代方案，但仍然距离实际应用相去甚远。我们探讨了自我中心人类视频的扩展能带来多大进展，使用的数据集包含30,000小时的视频，涵盖超过1,000种场景类型和14,000名贡献者。我们并未依赖不透明的下游指标，而是直接在一个具有挑战性的分布外基准上评估代理和物体交互的真实性。将训练数据增加100倍可以改善这两者，但改善程度不均匀：代理的建模效果良好，而物体的真实性仍然较低且改善缓慢。我们表明，代理的收益不必来自数据，经过精心设计的视觉条件可以在使用一小部分数据的情况下饱和真实性，这使我们能够单独测量物体的真实性并发现其饱和点。随后，我们引入了一种监督方案，将能力从场景外观转移到物体动态上，从而改善物体的真实性，尽管仍存在显著差距。最后，我们的结论可转移到下游的人形建模中。总体而言，我们的结果表明，扩展自我中心数据使代理建模接近其极限，而在世界上的影响仍然滞后，缩小这一差距将取决于模型的训练方式，而不仅仅是它们所接触的数据量。

View on arXiv Download PDF AI Translation

cs.CV / 169 / 2610.12469

Rubric-CEPR: Self-Evolving Image Editing via Reward-Verified Self-Distillation

Rubric-CEPR：通过奖励验证的自我蒸馏实现自我进化的图像编辑

Thawkar, Ritesh, Patle, Shubham, Venkatraman, Shravan, Anwer, Rao Muhammad

Abstract

Instruction-guided image editors have become highly capable, yet improving them further still depends on human-edited training pairs or external reward models. Such supervision is costly to obtain and can reward plausible failures: a realistic output may leave the requested change undone or alter content that should be preserved. In this work, we strive to improve a pretrained image editor using only its own generations, without human-edited targets or an external training-time reward model. To this end, we propose a self-evolving framework, named Rubric-CEPR, that verifies the editor's own samples with its internal representations through a rubric-augmented Contrastive Edit-Preservation Reward (CEPR). A Planner proposes structured edit instructions from unlabeled images, the Editor samples multiple candidate edits, and a frozen Critic scores each candidate with decomposed rubric checks for edit realization, removal of the old state, and content preservation, using features already exposed by the editor. Non-compensatory gates reject infeasible candidates, and the best verified candidate is distilled into the editor through lightweight adapter training. On Qwen-Image-Edit, Rubric-CEPR improves ImgEdit from 4.36 to 4.60 (+5.5%), with a +24.9% gain on object isolation, and transfers to GEdit-Bench and Complex-Edit. The same procedure also improves Step1X-Edit by +7.8% on ImgEdit. We hope our approach will serve as a solid baseline for image editors that improve themselves from their own verified samples. Our code is publicly available at $\href{https://riteshthawkar.github.io/Rubric-CEPR/}{\text{this URL}}$

Chinese Translation

受指令引导的图像编辑器已经变得非常强大，但进一步提升仍然依赖于人工编辑的训练对或外部奖励模型。这种监督获取成本高昂，并且可能奖励看似合理的失败：一个逼真的输出可能未能完成请求的更改，或改变应当保留的内容。在本研究中，我们努力仅使用预训练图像编辑器自身的生成结果来改进它，而不依赖于人工编辑的目标或外部训练时的奖励模型。为此，我们提出了一种自我进化的框架，命名为Rubric-CEPR，通过增强的对比编辑保留奖励（Contrastive Edit-Preservation Reward, CEPR）验证编辑器自身的样本与其内部表示。规划器从未标记的图像中提出结构化的编辑指令，编辑器生成多个候选编辑，冻结的评估器对每个候选进行评分，采用分解的标准检查编辑实现、旧状态的移除和内容的保留，使用编辑器已经暴露的特征。非补偿性门控拒绝不可行的候选，最佳验证候选通过轻量适配器训练蒸馏到编辑器中。在Qwen-Image-Edit上，Rubric-CEPR将ImgEdit的评分从4.36提升至4.60（+5.5%），在物体隔离方面获得+24.9%的提升，并转移到GEdit-Bench和Complex-Edit。相同的过程也使Step1X-Edit在ImgEdit上提高了+7.8%。我们希望我们的方法能为图像编辑器提供一个坚实的基线，使其能够从自身验证的样本中自我改进。我们的代码可以在$ ext{this URL}$公开获取。

View on arXiv Download PDF AI Translation

人工智能 (Artificial Intelligence)

125

cs.AI / 1 / 2610.10541

An Explainable Header-Centric Framework for Large-Scale Semantic Table Interpretation and Data Quality Assessment

一种可解释的以标题为中心的大规模语义表解读与数据质量评估框架

Silva, Marcelo Valentim, Herrmann, Hannes, Maxville, Valerie

Abstract

Knowledge Graph (KG) quality depends not only on downstream graph validation, but also on the quality of tabular metadata used before integration. In metadata-only Semantic Table Interpretation (STI), where cell values are unavailable, noisy, or unsuitable, column headers become a critical source of semantic evidence for traceable KG preparation. We present an explainable, header-centric framework for metadata-only Column Type Annotation (CTA) and Data Quality Assessment (DQA). The framework maps headers to 39 interpretable FinalFormat types using curated lexical resources and preserves token-level traceability through SourceKeywords. Each assigned type activates validation rules based on a taxonomy of Data Quality Issues (DQIs), producing detections such as missing data, duplicates, domain violations, wrong data type, and temporal mismatch. These detections are aggregated into HeadersIQ, a lightweight, unweighted data source-level quality metric. The framework was evaluated across heterogeneous benchmarks, including UCI, Prague, Kaggle, VizNet/Sato, SOTAB, T2Dv2, and the SemTab 2024 Metadata-to-KG track, comprising around 120,000 header columns. The results show broad practical coverage across noisy real-world metadata, while a parallel KG-mapping pathway supports alignment to DBpedia and Schema.org. On the SemTab 2024 Metadata-to-KG track, the official GT-strict evaluation was modest. However, a blinded diagnostic audit indicates that many mismatches reflect benchmark granularity, aliasing, and ontology-selection effects rather than wholly implausible header-centric predictions. We report this audit as diagnostic evidence on disagreement patterns, not as revised benchmark performance. Overall, the paper presents a reusable workflow for metadata-driven semantic annotation, data source-level quality monitoring, and KG-oriented benchmark diagnosis.

Chinese Translation

知识图谱（Knowledge Graph, KG）的质量不仅依赖于下游图的验证，还取决于集成前使用的表格元数据的质量。在仅基于元数据的语义表解读（Semantic Table Interpretation, STI）中，当单元格值不可用、噪声过大或不适用时，列标题成为可追溯的知识图谱准备的重要语义证据来源。我们提出了一种可解释的、以标题为中心的框架，用于仅基于元数据的列类型注释（Column Type Annotation, CTA）和数据质量评估（Data Quality Assessment, DQA）。该框架利用精心策划的词汇资源将标题映射到39种可解释的最终格式（FinalFormat）类型，并通过源关键词（SourceKeywords）保持逐词级别的可追溯性。每个分配的类型根据数据质量问题（Data Quality Issues, DQIs）的分类法激活验证规则，产生缺失数据、重复数据、领域违规、错误数据类型和时间不匹配等检测结果。这些检测结果被汇总为HeadersIQ，一种轻量级、无权重的数据源级质量指标。该框架在异构基准上进行了评估，包括UCI、Prague、Kaggle、VizNet/Sato、SOTAB、T2Dv2和SemTab 2024元数据到知识图谱（Metadata-to-KG）轨道，涵盖约120,000个标题列。结果显示在嘈杂的现实世界元数据中具有广泛的实际覆盖，同时平行的知识图谱映射路径支持与DBpedia和Schema.org的对齐。在SemTab 2024元数据到知识图谱轨道上，官方的GT严格评估结果适中。然而，盲审诊断审核表明，许多不匹配反映了基准的粒度、别名和本体选择效应，而非完全不合理的以标题为中心的预测。我们将此审核报告作为对不一致模式的诊断证据，而非修订的基准性能。总体而言，本文提出了一种可重用的工作流程，用于基于元数据的语义注释、数据源级质量监测和面向知识图谱的基准诊断。

View on arXiv Download PDF AI Translation

cs.AI / 2 / 2610.10549

Synthesis Through Simulation: Generating Coherent Enterprise Data via Scalable Agent-System Interaction

通过模拟合成：通过可扩展的代理系统交互生成一致的企业数据

Li, Yipeng, Hathidara, Ashutosh, Lo, Jane, Abichandani, Harshavardhan, Singh, Gunraj, Ghosh, Atin

Abstract

Tool-calling agents have become central to enterprise AI, yet training and evaluating them at scale remains severely constrained due to business and legal restrictions on enterprise systems, data, and database schemas. Tabular data synthesis offers a natural alternative, but its effectiveness is fundamentally limited by structural validity and schema availability, while procedure-based approaches yield the opposite weakness, typically lacking distributional fidelity without per-domain authoring. We introduce **Synthesis Through Simulation** (STS), a **schema--free** data synthesis paradigm in which an LLM agent generates data by executing operations against policy-enforcing APIs within simulated enterprise environments. Because data is generated through the same environment that defines what is valid, STS guarantees structural validity by construction while decoupling validity enforcement from distribution modeling, allowing each to be addressed independently. The **Generalist Populator** (GP), STS's domain-agnostic agent, addresses the remaining challenges of distributional fidelity and synthesis scalability: GP achieves **0.88** average marginal fidelity and **100\% constraint satisfaction** across all ten environments *without access to DB schemas*, while statistical synthesizers are inapplicable to seven due to necessary seed data requirements, and schema-privileged agents fail 82\% of trajectories on airline environment's tightly coupled workflows due to brittle task composition. We open-source the full framework, all ten environments, and generated datasets at https://github.com/SAP/synthesis-through-simulation.

Chinese Translation

调用工具的代理已成为企业人工智能的核心，但由于对企业系统、数据和数据库模式的商业和法律限制，在大规模训练和评估它们方面仍然受到严重制约。表格数据合成提供了一种自然的替代方案，但其有效性在根本上受到结构有效性和模式可用性的限制，而基于过程的方法则产生相反的弱点，通常缺乏分布保真度，除非进行每个领域的创作。我们提出了**通过模拟合成**（Synthesis Through Simulation, STS），这是一种**无模式**的数据合成范式，其中一个大型语言模型（LLM）代理通过在模拟企业环境中执行操作来生成数据，操作是通过政策执行的API进行的。由于数据是在定义有效性的同一环境中生成的，STS通过构建保证了结构有效性，同时将有效性执行与分布建模解耦，使得两者可以独立处理。STS的领域无关代理**通用填充器**（Generalist Populator, GP）解决了分布保真度和合成可扩展性剩余的挑战：GP在所有十个环境中实现了**0.88**的平均边际保真度和**100\%的约束满足**，*在没有访问数据库模式的情况下*，而统计合成器由于必要的种子数据要求在七个环境中无法应用，且特权模式代理在航空环境的紧密耦合工作流中失败了82 ext{%}的轨迹，原因在于任务组合脆弱。我们在 https://github.com/SAP/synthesis-through-simulation 上开源了完整框架、所有十个环境和生成的数据集。

View on arXiv Download PDF AI Translation

cs.AI / 3 / 2610.10590

Agent-Controlled Forgetting for Tool-Using Agents: Reversible Context Curation in Practice

代理控制的遗忘机制用于工具使用代理：可逆上下文管理的实践

Franke, Jan-Peter

Abstract

Tool-using agents repeatedly carry observations whose useful content can be much smaller than their original payload. We study agent-controlled forgetting: the acting model selects previously observed tool results, replaces each with a short note at its original position, and retains the exact original in a recoverable archive. A Python harness exposes batch archival and explicit recovery without task-specific model training, while protecting user instructions and assistant messages from these operations. In an exploratory OpenTelemetry debugging case followed by an unrelated implementation task, the method ended with 231,951 provider-reported prompt tokens versus 912,492 under retained history, used 50% fewer cumulative input tokens, and had an estimated API cost of USD 1.28-1.44 versus approximately USD 4.38. Both arms passed the two-case primary behavioral oracle; neither fully satisfied the follow-up evaluation. The method made more requests and took 17% longer. A contrasting application-development pair produced no context or cost saving, and an earlier continuation exhibited lower manually assessed quality despite reduced context. These observations demonstrate substantial resource savings in noisy tool-use trajectories and identify workload dependence as a central consideration for reversible context management.

Chinese Translation

工具使用代理反复携带的观察内容，其有用信息可能远小于其原始负载。我们研究了代理控制的遗忘机制：行动模型选择先前观察到的工具结果，将每个结果替换为其原位置的简短备注，并在可恢复的档案中保留确切的原始内容。一个 Python 框架提供了批量归档和显式恢复的功能，无需特定任务的模型训练，同时保护用户指令和助手消息不受这些操作的影响。在一个探索性的 OpenTelemetry 调试案例后，进行了一项无关的实施任务，该方法最终报告了 231,951 个提供者报告的提示令牌，而在保留历史的情况下则为 912,492，使用的累计输入令牌减少了 50%，估计的 API 成本为 1.28-1.44 美元，而保留历史的情况下约为 4.38 美元。两个方案均通过了两案例的主要行为评估；但均未完全满足后续评估。该方法发出了更多请求，耗时长了 17%。一个对比的应用开发对没有产生上下文或成本节约，而早期的延续尽管上下文减少却表现出较低的人工评估质量。这些观察结果表明，在嘈杂的工具使用轨迹中实现了显著的资源节约，并确定了工作负载依赖性作为可逆上下文管理的核心考虑因素。

View on arXiv Download PDF AI Translation

cs.AI / 4 / 2610.10611

Verification and Self-Improvement in Agentic AI: Foundations and Limits

代理人工智能中的验证与自我改进：基础与限制

Lu, Chien-Ping

Abstract

Agentic AI systems can improve by searching longer, receiving additional support, or modifying how they propose and verify outputs. A performance score does not distinguish these mechanisms. We compare these changes through bounded verification with hidden terminal randomness. A stage specifies admissible transcripts, polynomial bounds, an alternating verification protocol, and a terminal checker. Its native reach uses default support; its closure frontier permits all support already admitted by the interface. Under a uniform pointwise probability gap and task-relative soundness, these are well-defined languages. We prove that independent majority amplification preserves both languages, whereas existential acceptance over random tapes can admit incorrect outputs. Exact verification is the zero-randomness case, with placement and completeness results. The randomized-verifier classes satisfy $\Sigma_k^{\mathrm{P}}\subseteq\Sigma_k^{\mathrm{RV}}\subseteq\Sigma_{k+1}^{\mathrm{P}}$; strict enlargement and depth separation require explicit complexity assumptions, while $\mathrm{BPP}=\mathrm{P}$ yields exact companions with the same frontiers. Representation analysis separates invariant acceptance from core-versus-support labels that can change under refactoring. For recursive self-improvement, uniformly bounded self-modification under a common sound interpreter and fixed verification protocol remains within the same verification class. A separate conditional-error budget controls false selection across adaptively chosen candidates. A quota-enforced XOR-synthesis family separates unbounded ratios of search success from changes in the accepted languages; exact and probabilistic audits check the resulting evidence requirements. The framework ties self-improvement claims to obligations on correctness, admissible evidence, verification resources, and selection error.

Chinese Translation

代理人工智能系统可以通过更长时间的搜索、获得额外支持或修改其提出和验证输出的方式来实现改进。性能评分无法区分这些机制。我们通过带有限制的验证与隐藏终端随机性比较这些变化。一个阶段指定了可接受的记录、多项式界限、交替验证协议和终端检查器。其本地范围使用默认支持；其闭包边界允许接口已承认的所有支持。在统一的逐点概率差距和任务相关的健全性下，这些都是明确定义的语言。我们证明独立的多数放大保持这两种语言，而在随机带上的存在性接受可能会接受不正确的输出。精确验证是零随机性的情况，具有位置和完备性结果。随机验证者类满足 $ ext{Σ}_k^{ ext{P}} ext{⊆} ext{Σ}_k^{ ext{RV}} ext{⊆} ext{Σ}_{k+1}^{ ext{P}}$；严格扩展和深度分离需要明确的复杂性假设，而 $ ext{BPP}= ext{P}$ 产生具有相同边界的精确伴侣。表示分析将不变接受与在重构下可能变化的核心与支持标签分开。对于递归自我改进，在共同的健全解释器和固定验证协议下，均匀有界的自我修改仍然保持在同一验证类内。一个单独的条件错误预算控制在自适应选择的候选者中出现的错误选择。一个配额强制的异或合成家族将搜索成功的无界比率与接受语言的变化分开；精确和概率审计检查所需的证据要求。该框架将自我改进的主张与对正确性、可接受证据、验证资源和选择错误的义务联系起来。

View on arXiv Download PDF AI Translation

cs.AI / 5 / 2610.10629

The Harness as the Only Mutable Surface: Compliance-Bounded Self-Evolution of LLM Agents in Credit Pipelines, with a Measured Admission Gate

作为唯一可变表面的马具：在信用管道中受合规约束的LLM代理自我进化，伴随测量的准入门

Akhtyamov, Ravil

Abstract

Self-improving LLM agents can adapt a credit pipeline to a changed rule, but an agent that rewrites itself destroys the artefact a supervisor reviews: a named change, a recorded test, an approval. We argue that self-evolution is reviewable only if it is confined to the runtime harness (instruction text, tool-call logic and primitive composition) while model weights stay fixed, so that every adaptation is a diff with a cause and a test attached. We give a dual-loop engine built on that bound, with one admission gate that writes a hash-chained record before deployment, and we measure the gate in simulation, with a simulated agent and a seeded-search proposer rather than language models. Across three families of supervisory re-interpretation at three severities, 10 seeds each, the gated loop admitted 144 of 7,449 candidate changes, none of which worsened error on held-out history, and restored the false-positive rate to the oracle level without raising missed flags in every low- and mid-severity cell. With the gate replaced by the check an unbounded system applies (fewer errors visible in recent traces), the same loops admitted 309 harmful changes and left missed flags above 10% in 49 of 90 runs: false positives fell because the screen was loosened. Evaluated on pre-shift labels, the gate rejected every candidate, so a re-interpretation must be encoded as a rule that relabels history. Parametric and scope shifts were repaired locally, a structural one only by primitive replacement; at the highest structural severity the gate's fixed tolerance blocked the correct replacement in half the seeds. We map the mechanisms to the EU AI Act's provisions for high-risk credit scoring and note that the April 2026 US model-risk guidance excludes agentic AI from its scope.

Chinese Translation

自我改进的LLM代理能够将信用管道适应于变化的规则，但一个自我重写的代理会破坏监督者审查的工件：一个命名的变化、一个记录的测试、一个批准。我们认为，自我进化只有在限制于运行时马具（指令文本、工具调用逻辑和原始组合）而模型权重保持固定的情况下才能被审查，以便每次适应都是一个附带原因和测试的差异。我们基于这一约束构建了一个双循环引擎，其中一个准入门在部署前写入哈希链记录，我们在模拟中测量该门，使用模拟代理和种子搜索提议者，而不是语言模型。在三种不同严重程度的监督重新解释的三组中，每组10个种子，带门循环接受了7,449个候选变化中的144个，其中没有一个在保留历史上加重错误，并且在每个低和中等严重性单元中将假阳性率恢复到神谕水平，而没有提高遗漏标志。将门替换为一个无界系统应用的检查（在最近的轨迹中可见的错误更少），相同的循环接受了309个有害变化，并在90次运行中的49次中留下了超过10%的遗漏标志：假阳性下降是因为筛选放宽。根据预移位标签进行评估，门拒绝了每一个候选，因此重新解释必须编码为重新标记历史的规则。参数和范围的变化在本地修复，结构性变化仅通过原始替换修复；在最高结构严重性下，门的固定容忍度在一半的种子中阻止了正确替换。我们将这些机制映射到欧盟人工智能法案对高风险信用评分的规定，并注意到2026年4月的美国模型风险指导将代理人工智能排除在其范围之外。

View on arXiv Download PDF AI Translation

cs.AI / 6 / 2610.10635

Speaking the Navigator's Language: Trajectory-Grounded Instruction Translation for Frozen Aerial VLN Agents

说出导航员的语言：基于轨迹的指令翻译用于冻结的空中视觉与语言导航代理

Chen, Xi, Liu, Zhe, Xu, Xiaogang, Xu, Jiafei, Zhou, Chunyi, Su, Yuan, Zeng, Rui, Du, Tianyu, Yao, Kelu, Li, Chao, Ji, Shouling

Abstract

Aerial vision-and-language navigation (VLN) agents are typically trained on detail-rich, trajectory-aligned commands, whereas users issue short, intent-driven instructions; on a frozen OpenFly navigator, this \emph{instruction gap} drops success rate (SR) from $31.03\%$ to $11.33\%$. To scale translator training, we prompt a language model with human-written style examples to convert original commands into paired, intent-centered Weak commands, which yield $15.27\%$ SR. We introduce the \textbf{Trajectory-Grounded Instruction Translator (TGIT)}, a front-end that keeps the navigator frozen and translates Weak inputs into agent-executable commands by learning from its trajectory outcomes. The resulting Weak-trained translator raises Weak-input SR to $37.93\%$ and transfers zero-shot to real human instructions ($11.33\%{\rightarrow}32.51\%$); it also improves held-out OpenFly ($4.95\%{\rightarrow}20.79\%$) and yields recovery on CityNav and AirVLN.

Chinese Translation

空中视觉与语言导航（VLN）代理通常在细节丰富、与轨迹对齐的指令上进行训练，而用户则发出简短、以意图驱动的指令；在一个冻结的 OpenFly 导航器上，这种 extit{指令差距} 使成功率（SR）从 $31.03\%$ 降至 $11.33\\%$。为了扩大翻译器的训练，我们通过人类撰写的风格示例来提示语言模型，将原始指令转换为配对的、以意图为中心的弱指令，从而实现 $15.27\\%$ 的成功率。我们引入了 extbf{轨迹基础指令翻译器（TGIT）}，这是一个前端系统，保持导航器冻结，并通过学习其轨迹结果将弱输入翻译为代理可执行的指令。最终，经过弱训练的翻译器将弱输入的成功率提高至 $37.93\\%$，并在零样本情况下转移到真实的人类指令上（$11.33\\%{ ightarrow}32.51\\%$）；它还改善了保留的 OpenFly 成绩（$4.95\\%{ ightarrow}20.79\\%$），并在 CityNav 和 AirVLN 上实现了恢复。

View on arXiv Download PDF AI Translation

cs.AI / 7 / 2610.10786

Plan-and-Patch: Diffusion Language Models for Agentic Planning

计划与修补：用于自主规划的扩散语言模型

Kumar, Syamantak, Guo, Jiang, Hamad, Hassan, Kobayashi, Hideo, Xiang, Yi, Yang, Yezhou, Qi, Yanjun, Bonadiman, Daniele, Jiang, Jiarong

Abstract

Planning is increasingly important for long-horizon agents, where successful execution requires coordinating subgoals, tool use, and intermediate outcomes over many steps. Yet assumptions made during planning may be invalidated by the environment, tools may return unexpected results, or actions may fail. Effective agents must therefore not only generate plans, but also revise them. Such revisions often affect only part of a plan, leaving the preceding and subsequent structure intact. Rather than regenerate the entire plan and risk unnecessary changes, repair can regenerate the affected region conditioned on the preserved prefix and suffix. We introduce Plan-and-Patch, a plan-and-act framework in which a diffusion language model (dLLM) generates a structured, program-like plan through parallel unmasking and repairs it by filling in selected regions while keeping the surrounding steps fixed. We compare DreamReasoner-8B and Qwen3-8B as diffusion and autoregressive (AR) planners. On Natural Plan without task-specific training, diffusion (53.7%) achieves nearly twice the plan repair success rate of AR (27.0%). After task-specific training on agentic benchmarks, ALFWorld and TextCraft, the planners achieve similar observed success in plan generation, while diffusion reduces mean plan-generation latency by 39-46% relative to AR. Our results show that Plan-and-Patch provides a framework for faster plan generation and effective plan repair in long-horizon agents.

Chinese Translation

规划对于长时间跨度的智能体变得越来越重要，成功的执行需要在多个步骤中协调子目标、工具使用和中间结果。然而，在规划过程中做出的假设可能会被环境所推翻，工具可能会返回意想不到的结果，或者行动可能会失败。因此，有效的智能体不仅需要生成计划，还需要对其进行修订。这种修订通常只影响计划的一部分，保持前后的结构不变。与其重新生成整个计划并冒着不必要的变更风险，不如通过在保留的前缀和后缀的条件下重新生成受影响的区域来进行修复。我们提出了计划与修补（Plan-and-Patch），这是一个计划与行动框架，其中扩散语言模型（dLLM）通过并行解掩码生成结构化的程序式计划，并通过填充选定区域来修复计划，同时保持周围步骤不变。我们比较了 DreamReasoner-8B 和 Qwen3-8B 作为扩散和自回归（AR）规划器。在没有任务特定训练的自然计划（Natural Plan）上，扩散（53.7%）的计划修复成功率几乎是自回归（27.0%）的两倍。在针对自主基准（ALFWorld 和 TextCraft）进行任务特定训练后，规划器在计划生成中的观察成功率相似，而扩散相较于自回归将平均计划生成延迟减少了39-46%。我们的结果表明，计划与修补提供了一个更快的计划生成和有效的计划修复框架，适用于长时间跨度的智能体。

View on arXiv Download PDF AI Translation

cs.AI / 8 / 2610.10805

Whose Ground Truth? Embracing Ambiguity in Human-Centered AI

谁的真实标准？拥抱人本中心人工智能中的模糊性

Wu, Jingyao, Islam, Mohammad Tariqul, Nagbøl, Per Rådberg, Gerlings, Julie, Leoni, Giovanni, Fryza, Georgiana, Thomsen, Lars Pilegaard, Mainz, Jakob

Abstract

As AI systems increasingly interact with people and make decisions about them, understanding human interpretations becomes an important part of developing human-centered AI. Conventional machine learning and AI systems are largely developed under the assumption that a single definitive ground truth exists, with variability in human annotations often resolved through aggregation or treated as noise. However, for many human-centered tasks, human interpretation is inherently ambiguous, and multiple interpretations of the same input may be simultaneously reasonable and valid. Reducing such ambiguity to a single target risks overlooking meaningful information about the diversity of human perception, judgment, and experience. In this position paper, we call for a shift towards modeling the interpretation space of plausible human judgments, while distinguishing meaningful ambiguity from annotation noise. We argue that this perspective should guide how AI systems are represented, learned, evaluated, deployed, and governed, supporting more human-centered AI that better reflects the diversity of human interpretation.

Chinese Translation

随着人工智能系统越来越多地与人类互动并对其做出决策，理解人类的解释成为开发人本中心人工智能的重要组成部分。传统的机器学习和人工智能系统在很大程度上是在假设存在单一明确的真实标准的前提下开发的，而人类注释中的变异性通常通过聚合来解决或被视为噪声。然而，对于许多人本中心的任务而言，人类的解释本质上是模糊的，同一输入的多种解释可能同时是合理且有效的。将这种模糊性简化为单一目标的做法，可能会忽视有关人类感知、判断和经验多样性的有意义信息。在这篇立场论文中，我们呼吁转向建模合理人类判断的解释空间，同时区分有意义的模糊性与注释噪声。我们认为，这一视角应指导人工智能系统的表示、学习、评估、部署和治理，支持更具人本中心的人工智能，更好地反映人类解释的多样性。

View on arXiv Download PDF AI Translation

cs.AI / 9 / 2610.10833

On the Clock: Towards Punctual and Productive Time-Budgeted AI Agents

时间管理：朝着准时和高效的时间预算人工智能代理迈进

Wang, Aaron, Madan, Neelabh, Sobal, Vlad, Trager, Matthew, Kleinman, Michael, Mansimov, Elman, Xia, Wei, Soatto, Stefano

Abstract

We study whether small LLM agents can operate effectively under explicit wall-clock time budgets by both respecting the allocated runtime and using available time productively. We evaluate Qwen3.6-27B on five competitions from MLE-Bench Lite and Qwen3-4B on Zork I (Jericho), two agentic benchmarks where additional computational time can meaningfully improve performance. In the simplest setting, where the budget is stated only in the prompt, agents fail to translate the stated budget into controlled use of time. These failures arise from gaps in time awareness, since the harness provides no timing feedback, but also because they cannot reliably anticipate the duration of actions, and do not have a learned mapping from available time to an appropriate strategy. We investigate two complementary classes of interventions: harness-based mechanisms that expose timing information and enforce deadlines, and reinforcement learning with budget-aware rewards. Injecting timing information through the harness substantially improves budget adherence for Qwen3.6-27B without measurable loss in performance, while enforcement hooks tighten adherence further. RL with GRPO achieves near-perfect budget adherence on Zork I and generalizes to held-out budgets not seen during training, but does not improve task performance over the untrained harness on MLE-Bench. Once agents are made to respect the budget, they still fail to use additional time to improve task performance. RL-trained policies learn when to stop but often fill extra time with repeated actions, and GRPO training on multiple budgets tends to collapse toward the strategy learned for the shortest budget. Our results reveal a gap between time adherence and productive time allocation, which remains a central challenge for budget-conditioned agents.

Chinese Translation

我们研究小型大语言模型（LLM）代理在明确的墙钟时间预算下是否能够有效运作，既要遵守分配的运行时间，又要高效利用可用时间。我们在 MLE-Bench Lite 的五个竞赛和 Zork I（Jericho）上评估 Qwen3.6-27B 和 Qwen3-4B，这两个代理基准中，额外的计算时间可以显著提高性能。在最简单的设置中，预算仅在提示中说明，代理未能将所述预算转化为对时间的有效控制。这些失败源于时间意识的缺失，因为系统没有提供时间反馈，同时代理也无法可靠地预测行动的持续时间，并且没有从可用时间到适当策略的学习映射。我们研究了两类互补的干预措施：基于系统的机制，暴露时间信息并强制执行截止日期，以及具有预算意识奖励的强化学习（RL）。通过系统注入时间信息显著改善了 Qwen3.6-27B 的预算遵守情况，且没有可测量的性能损失，而强制执行机制进一步加强了遵守情况。使用具有预算意识的奖励的 RL 在 Zork I 上实现了近乎完美的预算遵守，并且能够推广到训练期间未见的保留预算，但在 MLE-Bench 上并未改善任务性能。一旦代理被要求遵守预算，它们仍然未能利用额外时间来提高任务性能。经过 RL 训练的策略学会了何时停止，但常常用重复的动作填充额外时间，而在多个预算上进行 GRPO 训练往往会崩溃到为最短预算学习的策略。我们的结果揭示了时间遵守与高效时间分配之间的差距，这仍然是预算条件代理面临的一个核心挑战。

View on arXiv Download PDF AI Translation

cs.AI / 10 / 2610.10857

Self-Supervised Keyframe Discovery for Horizon-Invariant Behavior Cloning

自监督关键帧发现用于水平不变行为克隆

Rath, Prabin Kumar, Patil, Omkar, Gopalan, Nakul

Abstract

Behavior cloning (BC) in non-Markovian environments is a challenging problem because policies have to reason over contextual information over long horizons. Existing policy architectures rely on recurrent or attention-based mechanisms to capture long-term dependencies. However, recurrent models suffer from hidden-state collapse and gradient instability under backpropagation through time, while attention-based models are fundamentally limited by context length. To address these issues, we propose Keyframe Mnemonics, a novel self-supervised method that $\textit{discovers}$ a set of information-critical observations ($\textit{mnemonics}$) by learning an objective from randomly sampled past observations and using it as a reward for keyframe selection. We then train a BC policy that conditions on the discovered keyframes to model the action distribution. Under certain task-structure assumptions, our formulation provides context retention guarantees over an infinite horizon, while maintaining a small set of decision-relevant keyframes in the policy's working memory. We evaluate our method on synthetic memory domains, where mnemonic-conditioned BC policies achieve $100$% success rates (SR) and generalize to horizons orders of magnitude beyond training without performance degradation. Additionally, we evaluate on memory-intensive robot manipulation benchmark, achieving a $13.9$% average absolute SR improvement over the strongest baseline across $23$ tasks and retaining $80$% SR at $20\times$ longer horizons on a real robot. Code and videos are available at https://keyframe-mnemonics.github.io.

Chinese Translation

在非马尔可夫环境中，行为克隆（BC）是一项具有挑战性的任务，因为策略需要在较长的时间范围内推理上下文信息。现有的策略架构依赖于递归或基于注意力的机制来捕捉长期依赖关系。然而，递归模型在通过时间反向传播时会遭遇隐藏状态崩溃和梯度不稳定的问题，而基于注意力的模型则在本质上受到上下文长度的限制。为了解决这些问题，我们提出了关键帧记忆（Keyframe Mnemonics），这是一种新颖的自监督方法，通过从随机采样的过去观察中学习目标，发现一组信息关键的观察（记忆），并将其作为关键帧选择的奖励。然后，我们训练一个基于发现的关键帧的BC策略，以建模动作分布。在特定任务结构假设下，我们的公式提供了在无限时间范围内的上下文保留保证，同时在策略的工作记忆中保持一小组与决策相关的关键帧。我们在合成记忆领域评估了我们的方法，其中记忆条件的BC策略实现了100%的成功率（SR），并且在训练后能够推广到远超训练的时间范围而不降低性能。此外，我们在内存密集型机器人操作基准上进行评估，在23个任务中实现了比最强基线高出13.9%的平均绝对SR提升，并在真实机器人上在20倍更长的时间范围内保持80%的SR。代码和视频可在 https://keyframe-mnemonics.github.io 获取。

View on arXiv Download PDF AI Translation

cs.AI / 11 / 2610.10906

Reading the Room: Foundations, Design, and Challenges of Normative Competence in LLMs

解读环境：大型语言模型中规范能力的基础、设计与挑战

Wynn, Andrea, Satija, Harsh, Seokhyun, Baek, Liu, Anqi, Nalisnick, Eric, Hadfield, Gillian K.

Abstract

Human communities are governed by normative systems: shared standards that produce \textit{norms} dictating acceptable behavior, enforced through community sanctioning. Aligning increasingly autonomous AI systems with these norms is a central alignment challenge, complicated by the fact that norms are vast in number, change quickly, and are often arbitrary (e.g., dress or language conventions). Thus, alignment requires \textit{normative competence}: the ability to discern from interaction alone what norms a community enforces without relying on static pretrained knowledge. We introduce a multi-agent community debate setting, where access to debate is governed by synthetic norms, to study normative competence in isolation from pretraining exposure. We show that baseline LLM agents fail to learn norms even when doing so would improve their accuracy. We then experiment with various \textit{normative modules} -- architectural components for norm inference -- finding that norm-following is highly sensitive to both the style of norm and the model powering the normative module, suggesting a lack of generalizability. Furthermore, when idiosyncratic, non-normative behaviors accompany the true norm, LLM agents exhibit an unselective attribution failure: they indiscriminately copy idiosyncratic noise alongside enforced rules, a pattern that persists even when imitating unnecessary behaviors is explicitly penalized. To the best of our knowledge, our work is the first to operationalize and evaluate normative competence in LLMs, demonstrating that current AI systems excel at behavioral mimicry but lack the capacity to discern socially enforced order.

Chinese Translation

人类社区由规范系统治理：共享标准产生规范，规定可接受的行为，并通过社区制裁进行执行。将日益自主的人工智能系统与这些规范对齐是一个核心的对齐挑战，这一过程因规范数量庞大、变化迅速且往往是任意的（例如，着装或语言习惯）而变得复杂。因此，对齐需要具备规范能力：仅通过互动来辨别一个社区所执行的规范，而不依赖于静态的预训练知识。我们引入了一种多智能体社区辩论设置，其中辩论的访问受合成规范的管理，以便在没有预训练暴露的情况下研究规范能力。我们发现，基线大型语言模型（LLM）智能体即使在学习规范能够提高其准确性的情况下也未能做到这一点。随后，我们对各种规范模块进行了实验——用于规范推断的架构组件，发现遵循规范对规范的风格和驱动规范模块的模型高度敏感，这表明缺乏可推广性。此外，当特有的非规范行为伴随真实规范出现时，LLM智能体表现出一种不加选择的归因失败：它们不加区分地复制特有的噪声和执行的规则，即使在明确惩罚模仿不必要行为的情况下，这种模式依然存在。据我们所知，我们的工作是首次在LLM中操作化和评估规范能力，表明当前的人工智能系统在行为模仿方面表现出色，但缺乏辨别社会执行秩序的能力。

View on arXiv Download PDF AI Translation

cs.AI / 12 / 2610.10942

StoreBench: A Live-Commerce Environment for Evaluating and Training Autonomous Operator Agents

StoreBench：用于评估和训练自主运营代理的直播电商环境

Raghuvanshi, Daksh, Vedere, Ved, Wang, Yifan

Abstract

Reinforcement learning environments are now a primary lever for improving large language model (LLM) capabilities in post-training, yet most agentic benchmarks remain static: the world moves only when the agent acts, the reward is a terminal verdict, and the pass bar is set arbitrarily. We introduce StoreBench, a live-commerce environment in which an agent runs a mid-size online apparel store on a production-grade commerce backend, testing long-horizon planning and economic judgment under uncertainty. Customers order around the clock, suppliers reprice and fail, and market shocks arrive with partial or no warning. The agent acts through the same 29 merchant tools a human operator would use, under a windowed operation budget that makes simulated time a function of actions taken, so model latency cannot influence simulated time. Pass thresholds are calibrated against scripted anchor policies, the reward is hardened against a catalogue of reward hacks, and every episode replays identically given a sequence of actions. We evaluate seven frontier LLMs on 11 scenarios of 30 to 45 days and a full simulated year, over three world seeds at matched reasoning effort. No model matches the scripted smart-triage policy on average: the best, DeepSeek-V4-Pro, passes 49% of task-seed cells against the heuristic's 97%. Human experts working through the same tools and budgets outscore every model (mean composite 0.708 vs. 0.700). Over a full simulated year under the Claude Code harness, most models show dramatic performance improvement. In a GRPO post-training run, Qwen3.5-27B trained on only five disjoint tasks raises its mean composite on the held-out evaluation tasks from 0.136 to 0.373. We release five example training-split tasks, ten sample trajectories, and the scoring and verification tooling; the full environment and evaluation suite are withheld to keep the benchmark uncontaminated.

Chinese Translation

强化学习环境现已成为提升大型语言模型（LLM）后训练能力的主要手段，然而大多数智能体基准测试仍然是静态的：世界仅在智能体行动时发生变化，奖励是终局判定，且通过门槛是任意设定的。我们提出了StoreBench，这是一个直播电商环境，智能体在生产级电商后台运营一家中型在线服装店，测试其在不确定性下的长远规划和经济判断能力。顾客全天候下单，供应商调整价格并可能失效，市场冲击部分或完全无预警地出现。智能体通过与人类运营者相同的29种商户工具操作，且在一个窗口化的操作预算下，模拟时间成为动作数量的函数，因此模型延迟不会影响模拟时间。通过脚本化锚定策略校准通过门槛，奖励机制针对一系列奖励漏洞进行了强化，每个回合在给定动作序列时均可完全重放。我们在11个场景（每个持续30至45天）及一个完整模拟年中，针对三个不同世界种子以相同推理努力评估了七个前沿LLM。没有模型能平均超过脚本化的智能分诊策略：表现最佳的DeepSeek-V4-Pro通过率为49%，而启发式策略为97%。使用相同工具和预算的人类专家得分超过所有模型（平均综合得分0.708对0.700）。在Claude Code框架下的完整模拟年中，大多数模型表现出显著的性能提升。在GRPO后训练过程中，Qwen3.5-27B仅在五个不相交任务上训练，其在未见评估任务上的平均综合得分从0.136提升至0.373。我们发布了五个示例训练任务、十条示例轨迹以及评分和验证工具；为保持基准测试的纯净性，完整环境和评估套件暂未公开。

View on arXiv Download PDF AI Translation

cs.AI / 13 / 2610.10954

Learning How to Search for Plans with Exponentially Less Space

学习如何以指数级更少的空间搜索计划

Drexler, Dominik, Ståhlberg, Simon, Fritzsche, Markus, Bonet, Blai

Abstract

Heuristic search for a plan can store exponentially many states, even when its heuristic is almost perfect. We instead learn search control, one specification per domain, written as an indexical policy: a generalized policy with registers that hold objects and modes that sequence its rules. We add the choose rule, which loads an object into a register and marks a backtracking point, where one candidate suffices; every other rule must work for all of its outcomes and needs no search. Our main result is that structural termination, which rules out infinite executions, also bounds every execution by a polynomial in the number of objects. A depth-first procedure then finds a plan in polynomial space, however large the state space, with no list of visited states. The cost is time, exponential only in the choice depth, the number of real choices along an execution. Any class that such a policy solves therefore lies in NP, and in P at constant choice depth. We learn these policies with a language model in a counterexample-guided loop that certifies termination, verifies the training tasks, and keeps the choice depth small. With the learned policies, the procedure solves 1,709 of 1,890 test tasks of the IPC 2023 Learning Track and the Autoscale Agile suite, more than LAMA, BFWS, and Levitron, and most of them within one second and 100 MiB.

Chinese Translation

启发式搜索计划可以存储指数级数量的状态，即使其启发式几乎完美。我们则学习搜索控制，每个领域一个规范，写成指示性策略：一种具有寄存器的广义策略，用于存储对象和顺序执行其规则的模式。我们添加了选择规则，该规则将对象加载到寄存器中并标记回溯点，在该点上一个候选项就足够；其他所有规则必须适用于其所有结果，并且无需搜索。我们的主要结果是，结构终止排除了无限执行，同时也将每次执行的界限限制为对象数量的多项式。然后，深度优先程序在多项式空间中找到计划，无论状态空间多大，并且不需要访问状态列表。其代价是时间，仅在选择深度上呈指数级增长，即执行过程中的真实选择数量。因此，任何此类策略解决的类都属于 NP，并且在常量选择深度下属于 P。我们通过一种语言模型在一个反例引导循环中学习这些策略，该循环认证终止，验证训练任务，并保持选择深度较小。通过学习到的策略，该程序解决了 IPC 2023 学习轨道和 Autoscale Agile 套件中的 1,709 个测试任务中的 1,890 个，超过了 LAMA、BFWS 和 Levitron，并且大多数任务在一秒内和 100 MiB 内完成。

View on arXiv Download PDF AI Translation

cs.AI / 14 / 2610.11005

How Narrative Wrapping Affects LLM Refusal: A Cross-Language Benchmark and Defense

叙事包装如何影响大语言模型的拒绝：跨语言基准与防御

Ye, Zhankai, Wang, Yanning, Jin, Yukai, Mei, Bo, Li, Fangyi, Wang, Wei, Gao, Shangqian, Liu, Xin

Abstract

Safety-aligned language models often refuse a harmful request stated directly but answer the same request inside a role-play or narrative wrapper. We measure this vulnerability across languages and registers: attack success on Qwen3-1.7B is already 89.4% in English and 93.0% in modern Chinese, and reaches 95.7% in Classical Chinese. We build GUISE, a benchmark for systematically studying this vulnerability. It includes parallel requests in English, modern Chinese, and Classical Chinese, matched harmful and benign pairs, wrapper types held out for evaluation, and a stricter criterion that counts warn-then-answer responses as attack successes. Representation analysis shows that language and register move harmful-request representations only slightly away from the model's refusal direction, whereas narrative wrappers move them much farther away. We propose AXIS, which combines preference optimisation with a rotation objective that aligns harmful-request representations with the refusal direction and a commitment objective that trains the model to refuse completely rather than produce a warn-then-answer response. Across Qwen3-1.7B, Qwen3-4B and GLM-4-9B, AXIS achieves the highest combined safety and usability score among the compared methods.

Chinese Translation

安全对齐的语言模型通常会直接拒绝有害请求，但在角色扮演或叙事包装中却会回答相同的请求。我们测量了这种脆弱性在不同语言和语域中的表现：在英语中，Qwen3-1.7B的攻击成功率已达到89.4%，在现代汉语中为93.0%，而在古汉语中则高达95.7%。我们构建了GUISE，一个系统性研究这种脆弱性的基准。它包括英语、现代汉语和古汉语中的平行请求，匹配的有害和良性对，保留用于评估的包装类型，以及一个更严格的标准，将警告后再回答的响应计为攻击成功。表示分析表明，语言和语域仅稍微将有害请求的表示移离模型的拒绝方向，而叙事包装则将其移得更远。我们提出了AXIS，它结合了偏好优化与旋转目标，旨在将有害请求的表示与拒绝方向对齐，并通过承诺目标训练模型完全拒绝，而不是产生警告后再回答的响应。在Qwen3-1.7B、Qwen3-4B和GLM-4-9B中，AXIS在比较方法中实现了最高的安全性和可用性综合得分。

View on arXiv Download PDF AI Translation

cs.AI / 15 / 2610.11007

Curating Always-Loaded Context for LLM Agents: A Capacitated Assortment Model with Censored Feedback

为大型语言模型代理策划始终加载的上下文：一个带有审查反馈的能力组合模型

Liu, Zexuan, Yang, Yuning, Zhao, Tiancheng

Abstract

At the start of every session, LLM agents load a fixed context file, such as $\texttt{AGENTS.md}$. Each loaded token in the file is charged again in every later round of the session, and these files can degrade performance as they grow in size. However, in practice, human or automated curators usually grow these files by appending. We formulate context curation as a capacitated assortment problem. Instructions consume tokens under a finite attention capacity; adding an instruction never raises the compliance of the others, while retained instructions incur a per-session setup cost. We prove an upper bound on the optimal file size, regardless of the number of available candidate instructions, and that appending every instruction with positive standalone value can be arbitrarily worse in net value than selecting an optimal subset. A token budget also limits the loss when the token price is underestimated. We then examine what can be learned from past sessions and how this information can guide decisions to add or remove instructions. Feedback is inherently censored: the benefits and harms of loaded instructions are observable, whereas missing instructions generate feedback only when their absence causes harm. In this setting, we show that deleting instructions ignored by agents can inevitably remove helpful ones. We characterize how much evidence should be collected before adding an instruction. Besides, we bound regret when human reviewers can inspect only a limited number of edits per period. Empirical experiments further show that irrelevant rules drawn from real context files reduce language-model compliance.

Chinese Translation

在每个会话开始时，大型语言模型（LLM）代理会加载一个固定的上下文文件，例如 $ exttt{AGENTS.md}$。文件中每个加载的标记在会话的后续轮次中都会再次计费，且随着文件大小的增加，这些文件可能会降低性能。然而，在实践中，人类或自动化策展者通常通过追加的方式来扩展这些文件。我们将上下文策划形式化为一个能力组合问题。指令在有限的注意力容量下消耗标记；添加一条指令不会提高其他指令的合规性，而保留的指令则会产生每个会话的设置成本。我们证明了最佳文件大小的上限，无论可用候选指令的数量如何，并且附加每条具有正独立价值的指令在净价值上可能会比选择一个最佳子集差得多。标记预算也限制了当标记价格被低估时的损失。接着，我们考察从过去的会话中可以学到什么，以及这些信息如何指导添加或删除指令的决策。反馈本质上是被审查的：加载的指令的好处和危害是可观察的，而缺失的指令只有在其缺失造成危害时才会产生反馈。在这种情况下，我们表明，删除被代理忽视的指令不可避免地会移除有帮助的指令。我们描述了在添加指令之前应收集多少证据。此外，当人类审阅者每个周期只能检查有限数量的编辑时，我们对遗憾进行了界定。实证实验进一步表明，从真实上下文文件中提取的无关规则降低了语言模型的合规性。

View on arXiv Download PDF AI Translation

cs.AI / 16 / 2610.11012

Distillation for Incrimination and Distillation for Capabilities

指控的蒸馏与能力的蒸馏

Prasanna, Sebastian, Tay, Jacqueline, Westover, Alek

Abstract

Powerful misaligned AI models might recognize alignment evaluations and strategically behave well on them, rendering direct audits uninformative. However, distilling such a model into a weaker benign student places the teacher in a Distillation Double Bind: if misalignment transfers, the student may conceal it less effectively, revealing evidence about the teacher; if it does not, the student may learn useful capabilities while remaining benign. We introduce two distinct distillation approaches, one targeting each outcome. Distillation for Incrimination (DFI) aims to transfer misalignment but not the ability to conceal it. Distilling AuditBench's secret-keeping models into their underlying instruction-tuned model produces students that are significantly more likely than their teachers to admit their hidden behavior when asked, suggesting that knowledge of the behavior transferred more readily than the propensity to conceal it. Confession gains largely disappear when the student does not share the teacher's pretrained base, so DFI should target the teacher's own pre-RL checkpoint, which is weaker than the teacher but shares its base model. Distillation for Capabilities (DFC) aims to transfer capabilities but not misalignment. Among several techniques we evaluate, two are effective: inoculation prompting and training for more epochs on fewer unique examples. Both preserve the capability gains of standard distillation while substantially reducing the subliminal transfer of an animal preference, our proxy for misalignment. Together, these findings demonstrate two ways distillation can be used for AI safety: incriminating misaligned models, and extracting their capabilities without their misalignment.

Chinese Translation

强大的不对齐人工智能模型可能会识别对齐评估并在其上采取策略性良好行为，从而使直接审计变得无信息。然而，将这样的模型蒸馏成一个较弱的良性学生使得教师处于蒸馏双重困境中：如果不对齐转移，学生可能会更不有效地掩盖这一点，从而揭示关于教师的证据；如果不转移，学生可能在保持良性的同时学习有用的能力。我们提出了两种不同的蒸馏方法，分别针对每种结果。指控的蒸馏（Distillation for Incrimination, DFI）旨在转移不对齐但不转移掩盖能力。将 AuditBench 的保密模型蒸馏为其基础的指令调优模型，产生的学生在被询问时显著更可能承认其隐藏行为，这表明对行为的知识比掩盖倾向更容易转移。当学生不共享教师的预训练基础时，忏悔的收益大幅消失，因此 DFI 应该针对教师自己的预强化学习检查点，该检查点比教师弱，但共享其基础模型。能力的蒸馏（Distillation for Capabilities, DFC）旨在转移能力但不转移不对齐。在我们评估的几种技术中，有两种是有效的：接种提示和在更少的独特示例上训练更多的轮次。这两者在保留标准蒸馏的能力增益的同时，显著减少了动物偏好的潜移默化转移，这是我们对不对齐的代理。总之，这些发现展示了蒸馏在人工智能安全中的两种应用方式：指控不对齐模型，以及提取其能力而不带有不对齐。

View on arXiv Download PDF AI Translation

cs.AI / 17 / 2610.11050

AgentHorizon: Evaluating Agentic Judges for Long-Horizon Computer-Use Tasks

AgentHorizon：评估用于长时间计算机使用任务的代理判断者

Lù, Xing Han, Vattikonda, Dheeraj, Hajimiri, Sina, Zadeh, Fatemeh Pesaran, BehnamGhader, Parishad, Darwiche, Ghazwa, Kazemnejad, Amirhossein, Pal, Christopher, Drouin, Alexandre, Reddy, Siva

Abstract

Computer-use agents are capable of completing complex tasks, increasing the use of automatic judges to determine success, either for training or for evaluation without human involvement. Despite their flexibility, their reliability on long tasks spanning multiple applications remains unclear. A trajectory, composed of long sequences of screenshots and actions, may appear complete, but in reality violates constraints from the instruction or introduces an unwanted side effect. To identify these errors, a judge needs to examine the trajectory with respect to the user's instruction. To this end, we introduce AgentHorizon, a benchmark of 1,373 computer-use tasks (instruction-trajectory pairs) drawn from 166 hours of human-recorded trajectories spanning three operating systems. By recording trajectories for closely related instructions, we can construct negative tasks by swapping the instructions. This paired design evaluates judges on their ability to distinguish a successful trajectory from one that completed a similar (but incompatible) request. We release the benchmark under three splits: a frontier split, AgentHorizon (AH), a simplified split, AgentHorizon-Simple (AH-S), and a development split, AgentHorizon-Development (AH-D). We further evaluate eleven judges by (1) directly passing the full trajectory (with up to 300 screenshots and actions), and (2) using them as coding agents across five agent harnesses. We find that our best agentic judge, GPT-5.5, achieves 80.9% balanced accuracy on the AH subset. We find that tool-use improves certain models but results in worse performance for open-weight models, and that judges differ drastically in their ability to accept a valid trajectory and reject failed ones. Our findings highlight the need for judges that are capable of locating and verifying often hidden evidence that a task was properly completed inside long interaction histories.

Chinese Translation

计算机使用代理能够完成复杂任务，越来越多地使用自动判断者来确定成功与否，无论是用于训练还是评估，而无需人类参与。尽管它们具有灵活性，但在跨多个应用的长任务中的可靠性仍然不清楚。一条轨迹，由长序列的屏幕截图和动作组成，可能看起来完整，但实际上可能违反了指令中的约束或引入了不必要的副作用。为了识别这些错误，判断者需要根据用户的指令检查轨迹。为此，我们引入了AgentHorizon，这是一个包含1,373个计算机使用任务（指令-轨迹对）的基准数据集，来源于166小时的人类记录轨迹，涵盖三个操作系统。通过记录紧密相关指令的轨迹，我们可以通过交换指令构建负任务。这种配对设计评估判断者区分成功轨迹与完成类似（但不兼容）请求的轨迹的能力。我们将基准数据集分为三个部分：前沿分割，AgentHorizon (AH)；简化分割，AgentHorizon-Simple (AH-S)；以及开发分割，AgentHorizon-Development (AH-D)。我们进一步评估了十一位判断者，方法是（1）直接传递完整轨迹（最多包含300个屏幕截图和动作），以及（2）在五个代理框架中将其用作编码代理。我们发现，最佳的代理判断者GPT-5.5在AH子集上达到了80.9%的平衡准确率。我们发现工具使用改善了某些模型，但导致开放权重模型的性能下降，并且判断者在接受有效轨迹和拒绝失败轨迹的能力上差异显著。我们的研究结果强调了需要能够定位和验证在长时间交互历史中任务是否正确完成的隐性证据的判断者。

View on arXiv Download PDF AI Translation

cs.AI / 18 / 2610.11087

Beyond Imitation: A Framework and Benchmark for LLM-Assisted Peer Review

超越模仿：一个 LLM 辅助同行评审的框架与基准

Teo, Rachel S. Y., Yamada, Yutaro, Kotyan, Shashank, Imajuku, Yuki, Clanuwat, Tarin

Abstract

The rapid growth of scientific publishing has strained peer review, particularly in machine learning, raising concerns about declining review quality and increasing reviewer workload. Large language models (LLMs) have been proposed as automated review assistants, yet their evaluation has focused largely on imitating human-written reviews rather than supporting the core functions of peer review. Here, we introduce a verification-centric perspective on LLM-assisted peer review, emphasizing error detection as a critical and resource-intensive task. We present a scalable benchmark that evaluates review systems' ability to identify logical contradictions, constructed through synthetic insertion of errors into conference papers, yielding unambiguous evaluation targets and enabling systematic comparison. We further propose a Multi-Layered Review (MLR) framework that prioritizes detailed manuscript comprehension before review generation, aligning more closely with human reviewing practices while improving token efficiency. Across evaluations, our approach demonstrates strong alignment with human review scores, achieves high error detection performance, and provides complementary perspectives on reviewer focus. These improvements can be attributed to both the choice of the underlying LLM and the design of our system. At the same time, we corroborate persistent vulnerabilities to adversarial manipulation, underscoring the need for robustness in automated review systems. Our findings highlight the importance of rigorous, error-focused evaluation to guide responsible deployment of LLM-based tools in peer review and other critical scientific workflows.

Chinese Translation

科学出版的快速增长给同行评审带来了压力，尤其是在机器学习领域，引发了对评审质量下降和评审工作负担增加的担忧。大型语言模型（LLMs）被提议作为自动化评审助手，但它们的评估主要集中在模仿人类撰写的评审，而非支持同行评审的核心功能。在此，我们引入了一种以验证为中心的 LLM 辅助同行评审的视角，强调错误检测作为一项关键且资源密集的任务。我们提出了一个可扩展的基准，评估评审系统识别逻辑矛盾的能力，该基准通过在会议论文中合成插入错误构建，产生明确的评估目标并实现系统比较。我们进一步提出了一个多层次评审（Multi-Layered Review, MLR）框架，优先考虑在生成评审之前对手稿的详细理解，更加贴近人类评审实践，同时提高了令牌效率。在评估中，我们的方法与人类评审得分表现出强一致性，达到了高错误检测性能，并提供了对评审者关注点的补充视角。这些改进可以归因于所选用的基础 LLM 和我们系统的设计。同时，我们证实了对对抗性操控的持续脆弱性，强调了自动化评审系统在稳健性方面的需求。我们的研究结果突显了严格、以错误为中心的评估的重要性，以指导 LLM 基于工具在同行评审和其他关键科学工作流中的负责任部署。

View on arXiv Download PDF AI Translation

cs.AI / 19 / 2610.11118

OpenProblemBench: Benchmarking AI on Open Problems in the Foundational Theoretical Sciences

OpenProblemBench：在基础理论科学中的开放问题上对人工智能进行基准测试

Li, Zhiyi, Hu, Sihan, Xiao, Tianning, Cai, Xiansheng, Tan, Xiaojun, Deng, Youjin, Chen, Kun

Abstract

The next frontier for artificial general intelligence is tackling unresolved scientific problems, calling for benchmarks that assess progress beyond established knowledge. We introduce OpenProblemBench, a benchmark of 82 unresolved problems drawn from the mathematics and theoretical physics literature. Each problem supplies the research context, assumptions, and prior progress needed to investigate the question. We select problems whose proposed solutions admit comparatively clear checks of their decisive mathematical or computational claims. Four evaluator models independently assess the correctness, completeness, and degree of progress of each submission without reference solutions. Across seven evaluated configurations, GPT-6-Astra achieves the highest mean judged solve rate of 14.0%, compared with 5.5-6.7% for the evaluated full-size open models and 2.4-3.7% for Flash models. Case comparisons connect stronger outcomes to changes in problem representation, general arguments that extend beyond finite evidence, and proofs of the steps needed to complete a solution. By grounding evaluation in questions arising from the research literature, OpenProblemBench provides a setting for investigating the capabilities and limitations of AI as a contributor to foundational theoretical science.

Chinese Translation

人工通用智能的下一个前沿是解决未解的科学问题，这需要评估超越已建立知识的基准。我们介绍了OpenProblemBench，这是一个由82个未解问题组成的基准，问题来源于数学和理论物理文献。每个问题提供了研究背景、假设和调查该问题所需的先前进展。我们选择了那些提出的解决方案能够相对清晰地检查其决定性数学或计算主张的问题。四个评估模型独立评估每个提交的正确性、完整性和进展程度，而不参考解决方案。在七个评估配置中，GPT-6-Astra的平均判断解决率达到14.0%，而评估的全尺寸开放模型为5.5-6.7%，Flash模型为2.4-3.7%。案例比较将更强的结果与问题表述的变化、超越有限证据的一般论点以及完成解决方案所需步骤的证明联系起来。通过将评估基于来自研究文献的问题，OpenProblemBench为研究人工智能作为基础理论科学贡献者的能力和局限性提供了一个环境。

View on arXiv Download PDF AI Translation

cs.AI / 20 / 2610.11123

When Interfaces Speak: Data-Aware Generative UI Harness for Active Interaction

当接口发声：数据感知生成用户界面工具用于主动交互

Li, Xiaolong, Xu, Xiaohan, Li, Jinyang, Xu, Xinnuo, Qu, Ge, Huo, Nan, Williams, Jack, Cheng, Reynold

Abstract

Most human-agent interaction today remains text-based. Natural language can impose cognitive overload, ambiguity, information chaos, and slow input for complex tasks; ephemeral generative UIs can present structured information and guide users toward task completion. We propose GenUI-Harness, a multi-agent harness pairing a Tool Agent for information retrieval and task execution with a GUI Coder Agent that identifies ambiguities and generates front-end code for structured interfaces. Training the coder with reinforcement learning is challenging: verifiable rewards for interactive UI generation require costly execution, while LLM-as-a-Judge rewards are prone to reward hacking. We address the first challenge with Dynamic UX, a lightweight package for dynamic interaction and reward collection in a single sandbox, and the second with Reward Auditor, a meta-reward mechanism that monitors reward distributions and distills diagnostic patterns into a shared rubric and scoring specification. We introduce UI-TAU Bench, a benchmark for active human-agent interaction through generated UI code, built on 10 real-world domain databases constructed from public data sources and based on Tau-Bench tool-use settings, with Lite (300 tasks) and Full (1,000 tasks) splits. GenUI-Harness achieves an average Pass@3 gain of 4.48 percentage points over smolagents on Lite. Training with GenUI-Harness improves a 4B backbone from 9.33% to 58.00% Pass@3, outperforming larger frontier models such as Claude Opus 5 (46.67%). GenUI-Harness also remains robust on ambiguous and non-ambiguous queries. In a reviewer survey comparing communication channels, generated UIs reduce average dialogue rounds from 3.4 to 1.2. These results show that data-aware generative interfaces can support effective task completion and reduce dialogue rounds in evaluated database-backed workflows.

Chinese Translation

目前大多数人机交互仍然基于文本。自然语言可能导致认知负担、模糊性、信息混乱以及复杂任务的输入缓慢；短暂的生成用户界面可以呈现结构化信息并引导用户完成任务。我们提出了GenUI-Harness，这是一种多智能体工具，将信息检索和任务执行的工具智能体与识别模糊性并生成结构化界面前端代码的GUI编码智能体配对。使用强化学习训练编码器面临挑战：交互式用户界面生成的可验证奖励需要昂贵的执行，而使用大型语言模型（LLM）作为评判者的奖励容易受到奖励操控的影响。我们通过动态用户体验（Dynamic UX）解决了第一个挑战，这是一个轻量级包，用于在单一沙箱中进行动态交互和奖励收集，并通过奖励审计员（Reward Auditor）解决第二个挑战，这是一种元奖励机制，监控奖励分配并将诊断模式提炼成共享的评分标准和评分规范。我们介绍了UI-TAU Bench，这是一个通过生成的用户界面代码进行主动人机交互的基准，基于10个来自公共数据源构建的真实世界领域数据库，并基于Tau-Bench工具使用设置，分为Lite（300个任务）和Full（1,000个任务）两个部分。GenUI-Harness在Lite上相较于smolagents实现了平均4.48个百分点的Pass@3提升。使用GenUI-Harness训练将一个4B主干模型的Pass@3从9.33%提升至58.00%，超越了Claude Opus 5等更大前沿模型（46.67%）。GenUI-Harness在模糊和非模糊查询中也保持稳健。在一项比较通信渠道的评审调查中，生成的用户界面将平均对话轮次从3.4减少到1.2。这些结果表明，数据感知生成接口可以支持有效的任务完成，并减少在评估的数据库支持工作流中的对话轮次。

View on arXiv Download PDF AI Translation

cs.AI / 21 / 2610.11128

Balancing Reference Guidance and Free Generation in Trajectory Rollouts for Reasoning RL

在推理强化学习的轨迹展开中平衡参考指导与自由生成

Wang, Hanyu, Agarwal, Nakul, Mahjoub, Hossein Nourkhiz, Pari, Ehsan Moradi, Fukushima, Makoto, Chen, Jinghui, Tadiparthi, Vaishnav

Abstract

A verified reference solution provides a correct trajectory for training a reasoning model. Alternatively, a prefix of the reference can guide the model in generating a trajectory of its own. How much reference guidance should we provide? We study this question through prefix continuation, where the model continues from a reference prefix and keeps the resulting trajectory if it passes verification, falling back to the reference otherwise. Since both procedures produce correct trajectories, we compare their distributions with the ideal distribution, the model's own distribution conditioned on successful verification. For one continuation, we derive the KL divergence in closed form, which, up to a bounded term, decreases with the product of the probability of generating a different correct trajectory and the reference surprisal, the negative log probability of the reference suffix given the prefix. Since a longer prefix tends to raise the former but lowers the latter, continuation success alone does not determine the preferred amount of guidance. From this analysis, we learn a prefix selector shared across training questions from continuation outcomes, without estimating success probabilities or additional generation. The resulting Adaptive Reference Guidance (ARG) constructs correct trajectories within a fixed generation budget, and we apply it to all-failure groups in Group Relative Policy Optimization (GRPO). Experiments on Qwen3-4B and Qwen3-8B across five mathematical reasoning benchmarks show that ARG achieves the highest aggregate pass@12 among the evaluated methods with competitive average sampled accuracy.

Chinese Translation

经过验证的参考解提供了一个正确的轨迹，用于训练推理模型。另一方面，参考的前缀可以指导模型生成自己的轨迹。我们应该提供多少参考指导？我们通过前缀延续来研究这个问题，在这种情况下，模型从参考前缀继续生成轨迹，如果生成的轨迹通过验证，则保留该轨迹，否则回退到参考轨迹。由于这两种程序都产生正确的轨迹，我们将它们的分布与理想分布进行比较，即模型在成功验证条件下的自身分布。对于一次延续，我们以封闭形式推导出KL散度，该散度在一个有界项的情况下，随着生成不同正确轨迹的概率与参考惊讶度（给定前缀的参考后缀的负对数概率）的乘积而减小。由于较长的前缀往往会提高前者但降低后者，因此仅凭延续成功并不能决定所需的指导量。通过这一分析，我们从延续结果中学习到一个跨训练问题共享的前缀选择器，而无需估计成功概率或额外生成。最终得到的自适应参考指导（Adaptive Reference Guidance, ARG）在固定生成预算内构建正确的轨迹，我们将其应用于组相对策略优化（Group Relative Policy Optimization, GRPO）中的全失败组。在五个数学推理基准上对Qwen3-4B和Qwen3-8B的实验表明，ARG在评估方法中实现了最高的aggregate pass@12，并且具有竞争力的平均采样准确率。

View on arXiv Download PDF AI Translation

cs.AI / 22 / 2610.11129

GameCommBench: A Unified Benchmark and Type-Aware Evaluation for AI-Generated Game Commentary

GameCommBench：一种统一的基准和类型感知评估用于AI生成的游戏评论

Zheng, Qirui, Lin, Zhengteng, Xiao, Yunyi, Li, Junhao, Cheng, Keyuan, Wang, Xingbo, Wang, Yongyi, Li, Lingfeng, Lu, Yunlong, Li, Wenxin

Abstract

Game commentary is an open-ended generation task requiring multimodal perception, strategic reasoning, and contextual knowledge. Existing AI-Generated Game Commentary (AI-GGC) studies remain fragmented across games, modalities, and evaluation protocols, while overlap-based or holistic evaluators fail to capture the functional heterogeneity of commentary. We introduce \textsc{GameCommBench}, a unified benchmark spanning board games, sports, and esports, with commentary aligned to heterogeneous game contexts and annotated by commentary type. We further propose Type-Aware Commentary Evaluation (TACE), a structured framework for evaluating different types of commentary. We then validate TACE for reliability and human agreement, and use it to benchmark representative AI commentators. Results reveal non-uniform capability profiles, with live observation and strategic analysis emerging as major bottlenecks. Together, \textsc{GameCommBench} and TACE provide a diagnostic foundation for comparable and interpretable AI-GGC evaluation.

Chinese Translation

游戏评论是一项开放式生成任务，要求具备多模态感知、战略推理和上下文知识。现有的AI生成游戏评论（AI-GGC）研究在不同游戏、模态和评估协议之间仍然分散，而基于重叠或整体评估的方式无法捕捉评论的功能异质性。我们引入了 extsc{GameCommBench}，这是一个涵盖棋类游戏、体育和电子竞技的统一基准，其评论与异质游戏上下文相一致，并按评论类型进行了注释。我们进一步提出了类型感知评论评估（Type-Aware Commentary Evaluation, TACE），这是一个用于评估不同类型评论的结构化框架。随后，我们验证了TACE的可靠性和人类一致性，并利用其对代表性的AI评论员进行基准测试。结果显示出不均匀的能力特征，实时观察和战略分析成为主要瓶颈。总之， extsc{GameCommBench}和TACE为可比且可解释的AI-GGC评估提供了诊断基础。

View on arXiv Download PDF AI Translation

cs.AI / 23 / 2610.11155

Social Pain Disrupts Emotion-Action Brain-State Dynamics in Adolescents with Non-Suicidal Self-Injury

社会痛苦干扰非自杀性自伤青少年的情绪-行动脑状态动态

Xu, Ying, Liang, Xiaojun, Zhang, Li, Yuan, Yixuan, Huang, Gan, Zhou, Yongjie, Liang, Zhen

Abstract

Non-suicidal self-injury (NSSI) is prevalent among adolescents with depression, but the rapid brain-state dynamics linking social distress to maladaptive behavior remain unclear. We combine an experimental pain paradigm, electroencephalography (EEG) microstate analysis, and interpretable deep sequence modeling to investigate NSSI-related neurodynamics in 106 adolescents with depression, including 67 with NSSI (DN+) and 39 without NSSI (DN-), during social pain, physical pain, and resting-state conditions. A model integrating disease-specific, domain-adversarial, consistency, and contrastive learning captures higher-order dependencies in microstate sequences. Social pain yields the strongest NSSI discrimination, with 68.55% accuracy, outperforming the best baseline by 8.94% points. Model interpretation and conventional microstate analyses reveal weakened bidirectional transitions between MS3 and MS5 in DN+ adolescents during social pain. Source reconstruction associates MS3 with emotional/interoceptive processing and MS5 with action preparation, suggesting disrupted emotion-action coupling. Time-resolved analyses show greater early-to-middle action-state recruitment and later emotion-state recruitment in DN+ adolescents. In DN- adolescents, MS5-to-MS3 dynamics mediate associations between social-evaluation sensitivity and affective outcomes, whereas this mediation is absent in DN+; conversely, MS3-to-MS5 transitions are associated with greater negative affect in DN+. Together, these findings identify disrupted emotion-action coupling as a key neurodynamic mechanism underlying altered social pain processing in adolescents with NSSI, providing a mechanistically interpretable neural signature for objective identification of NSSI.

Chinese Translation

非自杀性自伤（NSSI）在抑郁青少年中普遍存在，但社会痛苦与不适应行为之间快速的脑状态动态仍不清楚。我们结合实验性疼痛范式、电生理学（EEG）微状态分析和可解释的深度序列建模，研究106名抑郁青少年中与NSSI相关的神经动态，其中67名为NSSI（DN+），39名为无NSSI（DN-），在社会痛苦、身体疼痛和静息状态条件下进行研究。一个整合疾病特异性、领域对抗、一致性和对比学习的模型捕捉微状态序列中的高阶依赖关系。社会痛苦产生了最强的NSSI区分，准确率为68.55%，比最佳基线高出8.94个百分点。模型解释和常规微状态分析揭示DN+青少年在社会痛苦期间MS3和MS5之间的双向转变减弱。源重建将MS3与情感/内感处理相关联，将MS5与行动准备相关联，提示情绪-行动耦合受损。时间分辨分析显示DN+青少年在早期到中期的行动状态招募和后期情绪状态招募更为显著。在DN-青少年中，MS5到MS3的动态在社会评估敏感性与情感结果之间起到中介作用，而在DN+中则缺乏这种中介；相反，MS3到MS5的转变与DN+中的更大负性情感相关。综上所述，这些发现识别出情绪-行动耦合的干扰作为非自杀性自伤青少年社会痛苦处理改变的关键神经动态机制，为NSSI的客观识别提供了可机制解释的神经特征。

View on arXiv Download PDF AI Translation

cs.AI / 24 / 2610.11188

What to Admit and How to Present: Governing Persistent Memory in LLM Agents

如何选择和呈现：管理大语言模型代理中的持久性记忆

Liu, Chang, Ding, Deliang

Abstract

Persistent memory can improve personalization in LLM agents but can also induce sycophancy and cross-domain leakage. We distinguish two governance decisions: admission, which determines what recalled information enters the working context, and presentation, which determines how admitted information is expressed. We implement two inference-time designs without retraining: factor-compiled admission (FC), which assesses whole memory entries, and permission-semantic admission (PS), which decomposes entries into typed units; both translate adjudicated attributes into eligibility decisions via deterministic policies. We evaluate on a four-backbone development suite and an external benchmark with four tasks of 300 samples each. Relative to verbatim injection, FC and PS reduce pooled judge-assessed failure rates on the external benchmark by 6.7 and 8.8 percentage points (p = 2.7e-7 and 4.1e-12), and development-set cross-domain leakage falls by up to 29.5 percentage points. A query-conditioned gating baseline shows no significant change in objective-fact failure or pooled failure. Under matched admission budgets, PS outperforms random and relevance-based selection on external objective-fact judgment after Holm correction. Holding presentation fixed, tightening admission cuts cross-domain failure by a further 17.5 percentage points (p = 1.6e-4); in contrast, no comparison between two renderings of identical adjudicated outputs survives multiple-comparison correction. Both designs increase personalization failures, and PS misses the preregistered improvement and personalization-preservation criteria. These results support evaluating admission and presentation separately: selection quality provides task-specific safety gains, while preserving beneficial memory use remains unresolved.

Chinese Translation

持久性记忆可以改善大语言模型（LLM）代理的个性化，但也可能导致谄媚和跨领域信息泄露。我们区分了两种治理决策：接纳（admission），决定哪些被回忆的信息进入工作上下文；呈现（presentation），决定已接纳信息的表达方式。我们实现了两种推理时设计，无需重新训练：因子编译接纳（factor-compiled admission, FC），评估整个记忆条目；权限语义接纳（permission-semantic admission, PS），将条目分解为类型化单元；两者都通过确定性策略将裁定属性转换为资格决策。我们在一个四个主干的开发套件和一个外部基准上进行了评估，基准包含四个任务，每个任务有300个样本。相较于逐字注入，FC和PS在外部基准上将评审者评估的失败率分别降低了6.7和8.8个百分点（p = 2.7e-7和4.1e-12），开发集的跨领域泄露减少了多达29.5个百分点。一个基于查询条件的门控基线在客观事实失败或汇总失败上没有显著变化。在匹配的接纳预算下，PS在Holm校正后的外部客观事实判断中优于随机和基于相关性的选择。固定呈现的情况下，收紧接纳进一步将跨领域失败降低了17.5个百分点（p = 1.6e-4）；相比之下，两种相同裁定输出的呈现之间的比较在多重比较校正后没有存活。两种设计都增加了个性化失败，且PS未能达到预注册的改进和个性化保留标准。这些结果支持分别评估接纳和呈现：选择质量提供特定任务的安全增益，而保持有益记忆的使用仍未解决。

View on arXiv Download PDF AI Translation

cs.AI / 25 / 2610.11213

AliO: Output Alignment Matters in Long-Term Time Series Forecasing

AliO：输出对齐在长期时间序列预测中的重要性

Park, Kwangryeol, Kim, Jaeho, Lee, Seulki

Abstract

Long-term Time Series Forecasting (LTSF) tasks, which leverage the current data sequence as input to predict the future sequence, have become increasingly crucial in real-world applications such as weather forecasting and planning of electricity consumption. However, state-of-the-art LTSF models often fail to achieve prediction output alignment for the same timestamps across lagged input sequences. Instead, these models exhibit low output alignment, resulting in fluctuation in prediction outputs for the same timestamps, undermining the model's reliability. To address this, we propose AliO (Align Outputs), a novel approach designed to improve the output alignment of LTSF models by reducing the discrepancies between prediction outputs for the same timestamps in both the time and frequency domains. To measure output alignment, we introduce a new metric, TAM (Time Alignment Metric), which quantifies the alignment between prediction outputs, whereas existing metrics such as MSE only capture the distance between prediction outputs and ground truths. Experimental results show that AliO effectively improves the output alignment, i.e., up to 58.2% in TAM, while maintaining or enhancing the forecasting performance (up to 27.5%). This improved output alignment increases the reliability of the LTSF models, making them more applicable in real-world scenarios.

Chinese Translation

长期时间序列预测（LTSF）任务利用当前数据序列作为输入来预测未来序列，在天气预测和电力消费规划等实际应用中变得越来越重要。然而，最先进的LTSF模型通常无法实现滞后输入序列中相同时间戳的预测输出对齐。相反，这些模型表现出较低的输出对齐，导致相同时间戳的预测输出波动，从而削弱了模型的可靠性。为了解决这个问题，我们提出了AliO（Align Outputs），这是一种新颖的方法，旨在通过减少时间域和频率域中相同时间戳的预测输出之间的差异来改善LTSF模型的输出对齐。为了衡量输出对齐，我们引入了一种新的度量标准，TAM（时间对齐度量），它量化了预测输出之间的对齐，而现有的度量标准如均方误差（MSE）仅捕捉预测输出与真实值之间的距离。实验结果表明，AliO有效地改善了输出对齐，即在TAM上提高了58.2%，同时保持或提升了预测性能（最高可达27.5%）。这种改进的输出对齐提高了LTSF模型的可靠性，使其在实际场景中更具适用性。

View on arXiv Download PDF AI Translation

cs.AI / 26 / 2610.11226

When Lower Reconstruction Loss Hurts: Distributionally Robust Refinement for Low-Bit LLM Quantization

当较低的重建损失带来负面影响：低比特大语言模型量化的分布鲁棒精炼

Zhao, Yanlong, Cheng, Xiaoyuan, Liu, Huihang, He, Baihua, Zhang, Xinyu, Zhu, Harrison Bo Hua, Chen, Wenlong, Zeng, Li, Sun, Zhuo

Abstract

Weight-only post-training quantization (PTQ) relies heavily on reconstruction loss minimization to preserve model quality at low precision. We show that the weights favored by minimizing this loss need not yield better model performance on new tasks. In fact, we find that lower reconstruction loss can even degrade model performance on the same calibration data. Our analysis further shows that weights with lower reconstruction loss on calibration data can have higher loss than other weights when the distribution of input activations changes. Motivated by these observations and our analysis, we propose Distributionally Robust Quantization (DRQ), a post-hoc refinement process that minimizes worst-case reconstruction loss over a constrained set of input activation distributions. DRQ refines the integer codes representing quantized weights within the existing quantization grid, keeping quantization parameters and inference operators unchanged. Extensive experiments show that DRQ improves models quantized by six representative PTQ methods, including AWQ, GPTQ, and ParoQuant, and delivers gains across both dense and mixture-of-experts large language models. These results establish DRQ as a general post-hoc refinement framework for weight-only PTQ, achieving better downstream performance without adding inference overhead.

Chinese Translation

仅基于权重的后训练量化（PTQ）在低精度下依赖于重建损失最小化来保持模型质量。我们表明，最小化此损失所偏好的权重并不一定能在新任务上带来更好的模型性能。实际上，我们发现较低的重建损失甚至可能降低模型在相同校准数据上的性能。我们的分析进一步表明，在校准数据上具有较低重建损失的权重，在输入激活分布变化时可能会比其他权重具有更高的损失。基于这些观察和我们的分析，我们提出了分布鲁棒量化（Distributionally Robust Quantization, DRQ），这是一种后续精炼过程，旨在最小化在约束的输入激活分布集上的最坏情况重建损失。DRQ在现有量化网格内精炼表示量化权重的整数编码，同时保持量化参数和推理操作不变。大量实验表明，DRQ改善了通过六种代表性PTQ方法（包括AWQ、GPTQ和ParoQuant）量化的模型，并在稠密和混合专家的大语言模型中均取得了性能提升。这些结果确立了DRQ作为一种通用的后续精炼框架，用于仅基于权重的PTQ，在不增加推理开销的情况下实现更好的下游性能。

View on arXiv Download PDF AI Translation

cs.AI / 27 / 2610.11231

Harness Compilation: Which Decisions Should a Small Vision-Language Model Keep?

Harness Compilation：小型视觉语言模型应保留哪些决策？

Fan, Minhao, Liu, Yinyi, Zhao, Jiayu, Teng, Zihan, Chen, Song, Liu, Weichen

Abstract

Small vision-language models may be able to read external evidence yet struggle to obtain it. We introduce Harness Compilation (HC), an offline procedure that adapts the division of work between a frozen small VLM and its external harness. A large teacher uses student execution traces to revise reusable content and control, while a separate validation set selects the deployed harness. Deployment requires neither weight updates nor teacher calls. Across seven visual question-answering settings with students of at most 9B parameters, HC improves scores over bare students by 9.9-23.9 points, averaged over three independent builds per setting. Interventions on five runtime decision types (invocation, selection, argument generation, evidence integration and abstention) show why this allocation matters: requesting evidence and generating open queries can be costly, whereas bounded choices and reading supplied text can remain useful student work. Fact cards benefit all ten evaluated students, but decision policies transfer unevenly. Recompilation for a new student model helps when the transferred interface no longer fits the student. With 100 practice items, HC exceeds answer-only LoRA on three tasks. Larger training budgets can match or surpass a fixed harness, while combining the two improves SlideVQA beyond either alone. These findings support allocating work from measured student behavior rather than uniformly removing decisions.

Chinese Translation

小型视觉语言模型可能能够读取外部证据，但在获取这些证据时却面临困难。我们提出了Harness Compilation（HC），这是一种离线程序，旨在调整冻结的小型视觉语言模型（VLM）与其外部支架之间的工作分配。一个大型教师模型利用学生执行轨迹来修订可重用内容和控制，而一个单独的验证集则选择部署的支架。部署不需要权重更新或教师调用。在七个视觉问答设置中，最多包含90亿参数的学生模型，HC在三个独立构建的平均分数上比单纯的学生模型提高了9.9到23.9分。对五种运行时决策类型（调用、选择、参数生成、证据整合和放弃）的干预显示了这种分配的重要性：请求证据和生成开放查询可能代价高昂，而有限选择和阅读提供的文本则可以成为有用的学生工作。事实卡片对所有十个评估的学生都有益，但决策策略的转移并不均匀。针对新学生模型的重新编译在转移接口不再适合学生时会有所帮助。在100个练习项目中，HC在三个任务上超越了仅回答的LoRA。更大的训练预算可以与固定支架匹敌或超越，而将两者结合在一起则使SlideVQA的表现超越单独使用任一者。这些发现支持根据测量到的学生行为分配工作，而不是均匀地去除决策。

View on arXiv Download PDF AI Translation

cs.AI / 28 / 2610.11253

LLM-IDEA: Identifiability-Driven Experimental Agent for Autonomous Discovery of Mechanistic World Models

LLM-IDEA：基于可识别性的实验代理，用于自主发现机制世界模型

Shetty, Surya, Braga-Neto, Ulisses

Abstract

Large language model agents are being increasingly deployed as autonomous scientists, designing experiments and inferring mechanistic world models with minimal human oversight. Yet identifiability is often overlooked: when a plateau is reached, the agent needs to know whether it is not yet capable enough or the model simply is not identifiable from the data, in which case no amount of further experimentation of the same kind can help. We propose the Identifiability-Driven Experimental Agent (LLM-IDEA) for closed-loop discovery with an identifiability engine that returns a three-way plateau verdict: capability limit, resolvable within the design class, or certified exhausted. On ODEBench, 60 of the 62 systems with free constants are identifiable at round 0; the RC circuit is certified exhausted for every experiment that protocol can run, and a harvesting model is resolvable by one added initial condition. The identifiability engine reproduces known verdicts on Lotka-Volterra, Van der Pol, Lorenz, and a pharmacokinetic model, where it recommends the intravenous arm pharmacologists use, and it ranks the depth scorer of our own benchmark last among four observation designs. On the DiscoverPhysics benchmark, it finds two public worlds whose explanation rubric rewards a distinction no legal experiment can make, and every model there with accurate trajectories failed the explanation grade (15 of 15, against 5 of 9 in identifiable worlds, p = 0.012). On the Alien Universe, a two-body testbed we propose in which a force law switches between a provably non-identifiable and an identifiable protocol, LLM-IDEA on the identifiable protocol reaches discovery depth at least three on 8/8 seeds versus 1/8 without it. An autonomous discovery agent can thus compute, rather than guess, whether a plateau calls for more search, a better experiment of the same kind, or a different kind of experiment.

Chinese Translation

大型语言模型代理正越来越多地被部署为自主科学家，设计实验并在最小人类监督下推断机制世界模型。然而，可识别性常常被忽视：当达到一个平台期时，代理需要知道自己是尚未具备足够能力，还是模型根本无法从数据中识别，此时再进行同类实验也无济于事。我们提出了可识别性驱动的实验代理（LLM-IDEA），用于闭环发现，配备一个可识别性引擎，该引擎返回三种平台期判决：能力限制、可在设计类别内解决或已确认耗尽。在ODEBench上，62个具有自由常数的系统中有60个在第0轮是可识别的；RC电路在该协议可以运行的每个实验中都被确认耗尽，而一个收获模型可以通过增加一个初始条件来解决。可识别性引擎在Lotka-Volterra、Van der Pol、Lorenz和一个药代动力学模型上重现了已知判决，其中推荐了药理学家使用的静脉注射臂，并且在我们自己的基准中将深度评分的排名排在四个观察设计的最后。在DiscoverPhysics基准上，它发现两个公共世界，其解释标准奖励了一个合法实验无法做出的区分，而每个具有准确轨迹的模型都未能获得解释等级（15个中有15个，识别世界中为9个中的5个，p = 0.012）。在我们提出的外星宇宙中，这是一个双体测试平台，其中一个力法则在可证明不可识别和可识别协议之间切换，LLM-IDEA在可识别协议下在8/8个种子中达到至少三的发现深度，而在没有它的情况下为1/8。因此，一个自主发现代理可以计算，而不是猜测，某个平台期是否需要更多搜索、同类实验的更好实验，或不同类型的实验。

View on arXiv Download PDF AI Translation

cs.AI / 29 / 2610.11268

TaReD: Tool-Aware Recursive Decomposition for Long-Horizon Tasks

TaReD：面向工具的递归分解用于长时间任务

Mao, Wei-Xiang, Chen, Zhi-Kai, Zhan, De-Chuan, Ye, Han-Jia

Abstract

Agents combine reasoning with tools to interact with external systems and complete real-world tasks. Early agents typically interleave reasoning and actions along a single execution chain. On complex tasks, this chain becomes unreliable because growing histories obscure intermediate dependencies and allow early planning errors to propagate. Recursively decomposing a complex task into smaller subtasks offers a natural solution, yet effective decomposition must account for the system's capabilities so that each subtask can be executed by the available tools. In realistic systems, however, tool libraries can be too large to expose in full. Injecting every tool description consumes substantial context while making relevant tools harder to retrieve and useful task boundaries harder to identify. We propose tool-aware recursive decomposition, which organizes tools by functional relationships into a hierarchy of capabilities. During execution, the agent discovers tools on demand and uses the hierarchy to recursively decompose a complex task into a subtask tree whose levels are aligned with the capabilities required at each stage. Experiments on complex real-world tasks show that the proposed method improves end-to-end task success rate by up to 40 percentage points over the compared baselines. The implementation of TaReD is available on GitHub: https://github.com/WeiXiang-Mao/TaReD.

Chinese Translation

智能体结合推理与工具与外部系统互动，以完成现实世界的任务。早期的智能体通常在单一执行链上交替进行推理和行动。在复杂任务中，这条链变得不可靠，因为不断增长的历史会模糊中间依赖关系，并使早期规划错误得以传播。将复杂任务递归地分解为更小的子任务提供了一种自然的解决方案，然而有效的分解必须考虑系统的能力，以便每个子任务都能由可用工具执行。然而，在现实系统中，工具库可能过于庞大，无法完全暴露。注入每个工具描述会消耗大量上下文，同时使得相关工具更难以检索，并且有用的任务边界更难以识别。我们提出了面向工具的递归分解，它通过功能关系将工具组织成能力层次结构。在执行过程中，智能体按需发现工具，并利用该层次结构将复杂任务递归分解为一个子任务树，其层级与每个阶段所需的能力相一致。在复杂的现实任务上的实验表明，所提方法的端到端任务成功率比对比基线提高了多达40个百分点。TaReD的实现可在GitHub上获取：https://github.com/WeiXiang-Mao/TaReD。

View on arXiv Download PDF AI Translation

cs.AI / 30 / 2610.11289

Open-ended Scientific Discovery with Possibilistic Reasoning

基于可能性推理的开放式科学发现

Yang, Anita, Chau, Siu Lun, Wakayama, Tomoya, Muandet, Krikamol, Adachi, Masaki

Abstract

Autonomous scientific discovery with LLMs requires generating and testing hypotheses adaptively as evidence accumulates while maintaining statistical validity. Existing anytime-valid methods can handle data-dependent hypotheses, but open-ended discovery poses a deeper challenge: the best discovered hypothesis may still be the best of a bad lot, with better explanations yet undiscovered, while even background knowledge such as physical laws may require revision in light of new findings. In response, we formalize the problem as Abductive Autonomous Scientific Discovery (AASD) using possibility theory. We introduce abductive utility, a computable measure of discovery progress, and possibility frontier search, the first algorithm for AASD, which maintains anytime validity and achieves $\varepsilon$-optimal abductive utility asymptotically under suitable conditions. Experiments on synthetic and real-world scientific-discovery tasks show strong performance.

Chinese Translation

自主科学发现需要在证据不断积累的过程中自适应地生成和测试假设，同时保持统计有效性。现有的任何时刻有效的方法能够处理依赖数据的假设，但开放式发现提出了更深层次的挑战：最佳发现的假设可能仍然是糟糕选择中的最佳，而更好的解释尚未被发现，甚至诸如物理定律等背景知识也可能需要根据新发现进行修订。为此，我们将问题形式化为基于可能性理论的归纳自主科学发现（Abductive Autonomous Scientific Discovery, AASD）。我们引入了归纳效用（abductive utility），这是一个可计算的发现进展度量，并提出了可能性边界搜索（possibility frontier search），这是第一个用于AASD的算法，能够保持任何时刻的有效性，并在适当条件下渐近地实现$ ext{ε}$-最优的归纳效用。在合成和真实世界的科学发现任务上的实验显示出强劲的性能。

View on arXiv Download PDF AI Translation

cs.AI / 31 / 2610.11294

ORDO: Operation-level Round-aware Dynamic Ordering for MIP Presolve

ORDO：面向操作级的轮次感知动态排序用于混合整数规划预处理

Chen, Zehuan, Song, Chunhe

Abstract

Presolve strongly affects mixed-integer programming (MIP) performance, yet learning-based methods only optimize parameter configurations and cannot express the non-commutative temporal dependencies among actions, whose default order is nearly unique on most domains, yet functionally necessary: artificially shuffling the order of the same sequence inflates the tail of the solve-time distribution by up to several-fold. We recast presolve planning as autoregressive sequence generation over a unified atomic action space, moving the decision object to action sequences; we call this framework ORDO---Operation-level Round-aware Dynamic Ordering for MIP Presolve. Its payoff is cross-domain generalization: on multiple unseen domains it attains end-to-end zero-shot speedup---to our knowledge the first for presolve action sequences---varying by domain and not explained by corpus richness, the strongest domain reaching the largest speedup once racing is added. Deployment uses sequence racing, in which candidate sequences run concurrently and the winner is kept, enabled by an execution-and-observation facility, added by modifying the SCIP source, that injects sequences along the native path and records which actions actually execute and in which round.

Chinese Translation

预处理对混合整数规划（MIP）的性能有着显著影响，但基于学习的方法仅优化参数配置，无法表达动作之间的非交换时间依赖关系，而这些关系在大多数领域中几乎是唯一的且功能上是必要的：人为打乱相同序列的顺序会将求解时间分布的尾部膨胀数倍。我们将预处理规划重新表述为在统一原子动作空间上的自回归序列生成，将决策对象转移到动作序列上；我们称这一框架为ORDO——面向操作级的轮次感知动态排序用于混合整数规划预处理。其收益在于跨领域的泛化：在多个未见领域中，它实现了端到端的零样本加速——据我们所知，这是针对预处理动作序列的首次实现——加速效果因领域而异，且无法用语料丰富度解释，最强的领域在加入竞速后达到最大的加速。部署使用序列竞速，其中候选序列并行运行，保留赢家，通过修改SCIP源代码添加的执行与观察设施实现，该设施沿原生路径注入序列并记录实际执行的动作及其所在轮次。

View on arXiv Download PDF AI Translation

cs.AI / 32 / 2610.11299

DuplexAgent-RSI: Recursive Harness Improvement for Full-Duplex Voice Agent Collaboration

DuplexAgent-RSI：全双工语音代理协作的递归工具改进

Shen, Yingda, Wang, Yuxiang, Feng, Kunyu, Ni, Qinke, Li, Jiaqi, Hsu, Minghao, Zhang, Junan, Chen, Dekun, Bian, Yutong, Wu, Zhizheng

Abstract

Voice agents are converging on a collaboration pattern: a full-duplex interaction model stays on the live channel as the entry to the conversation, while search, reasoning, and coding are handled through asynchronous delegation. A duplex model supports continuous listening and speaking, but complex reasoning and tool use may exceed its capabilities. A coding agent can plan and execute extended tasks, but its sequential interface is a poor fit for live conversation. Combining them requires a harness that coordinates task acceptance, progress, cancellation, replacement, and result delivery while keeping the conversation responsive. Existing harnesses often rely on coupled heuristics, making them difficult to improve systematically from evidence. We present DuplexAgent, a full-duplex collaboration system whose harness expresses this workflow as six editable modules, and Duplex-Harness-RSI, a closed loop that revises them from interaction traces. A simulator automatically generates timed test conversations, runs the system, and produces failure traces that identify the collaboration modules requiring repair. Reasoning LLMs and coding agents in the delegation pool also serve the improvement loop: the Exam Planner selects the next tests from observed weaknesses and the repair archive, and the Harness Editor proposes targeted module changes. The capabilities that serve the user thus also improve the system's coordination. Experiments on intelligence, agentic, and duplex benchmarks show that DuplexAgent combines continuous interaction with difficult reasoning and complex task execution, achieving stronger spoken-knowledge and executable-tool scores than the compared delegated systems while maintaining strong interruption response. A harness ablation further shows that this modular, verifiable loop outperforms the initial harness and repeated editing that lacks its diagnosis and repair archive.

Chinese Translation

语音代理正在趋向一种协作模式：全双工交互模型作为对话的入口，保持在实时通道上，而搜索、推理和编码则通过异步委托来处理。双工模型支持持续的听说，但复杂的推理和工具使用可能超出其能力范围。编码代理可以规划和执行扩展任务，但其顺序接口与实时对话不太匹配。将它们结合起来需要一个工具，以协调任务的接受、进展、取消、替换和结果交付，同时保持对话的响应性。现有的工具通常依赖于耦合启发式，使得从证据中系统性改进变得困难。我们提出了DuplexAgent，一个全双工协作系统，其工具将这一工作流程表达为六个可编辑模块，以及Duplex-Harness-RSI，一个从交互痕迹中修订这些模块的闭环。一个模拟器自动生成定时测试对话，运行系统，并产生识别需要修复的协作模块的失败痕迹。委托池中的推理大型语言模型（LLMs）和编码代理也服务于改进循环：考试规划器从观察到的弱点和修复档案中选择下一个测试，工具编辑器则提出针对性的模块更改。服务于用户的能力也因此改善了系统的协调性。在智能、代理和双工基准上的实验表明，DuplexAgent将持续交互与困难推理和复杂任务执行相结合，在口语知识和可执行工具得分上优于比较的委托系统，同时保持强大的中断响应。工具消融进一步表明，这种模块化、可验证的循环优于初始工具和缺乏诊断和修复档案的重复编辑。

View on arXiv Download PDF AI Translation

cs.AI / 33 / 2610.11312

MedBenchAgent: Towards Systematic Automation of Medical VLM Benchmark Construction

MedBenchAgent：迈向医学视觉-语言模型基准构建的系统化自动化

Fu, Yulin, Wang, Junren, Yang, Guangjing, Yu, Zhangyuan, Sun, Wanran, Zhou, Jiabao, Yin, Jin, Lao, Qicheng

Abstract

Large-scale construction of medical vision-language model (VLM) benchmarks is increasingly feasible with richly annotated imaging datasets and large language models (LLMs), yet existing automation largely focuses on generating evaluation items within predefined benchmark specifications. We study the broader problem of automatically deriving the specification itself: what to evaluate, which annotations support each task, and how to translate this evidence into reliable evaluation items. We formulate benchmark construction as constrained compilation, in which the benchmark specification is progressively derived from evaluation requirements, heterogeneous annotations, and medical knowledge. Based on this formulation, we introduce MedBenchAgent, a multi-agent framework with a Benchmark Intermediate Representation (BIR) that encodes task definitions, evidence mappings, evaluation protocols, and item specifications across construction stages. MedBenchAgent separates planning, which derives and verifies the specification, from instantiation, which constructs and audits items under the locked specification. MedBenchAgent achieves a Task-Space F1 of 90.9%, outperforming direct task induction (79.2-80.0%) and prior-guided induction (85.1%); 994 of 1,000 sampled items from correctly identified tasks pass human audit. We further demonstrate portability to a specialized medical domain and evaluate twelve VLMs, revealing task- and setting-specific variation obscured by aggregate scores. These results establish constrained compilation as a scalable and auditable framework for medical VLM benchmark construction beyond question generation.

Chinese Translation

随着丰富注释的影像数据集和大型语言模型（LLMs）的发展，大规模构建医学视觉-语言模型（VLM）基准变得越来越可行。然而，现有的自动化主要集中在生成预定义基准规范内的评估项目。我们研究了自动推导规范本身的更广泛问题：评估什么，哪些注释支持每个任务，以及如何将这些证据转化为可靠的评估项目。我们将基准构建形式化为受限编译，其中基准规范是根据评估要求、异构注释和医学知识逐步推导而来的。基于这一形式化，我们引入了MedBenchAgent，一个多代理框架，具有基准中间表示（BIR），该表示在构建阶段编码任务定义、证据映射、评估协议和项目规范。MedBenchAgent将规划（推导和验证规范）与实例化（在锁定规范下构建和审计项目）分开。MedBenchAgent在任务空间F1上达到了90.9%的表现，超越了直接任务归纳（79.2-80.0%）和先前引导的归纳（85.1%）；从正确识别的任务中抽样的1,000个项目中，有994个通过了人工审计。我们进一步展示了其在专业医学领域的可移植性，并评估了十二个VLM，揭示了被汇总分数掩盖的任务和设置特定的变异。这些结果确立了受限编译作为超越问题生成的医学VLM基准构建的可扩展和可审计框架。

View on arXiv Download PDF AI Translation

cs.AI / 34 / 2610.11317

DivMoE: Fine-Grained MoE Upcycling via Cross-Domain Expert Composition

DivMoE：通过跨领域专家组合实现细粒度MoE的升级

Lou, Yuxuan, Yang, Kai, Zhang, Geng, Liu, Yong, You, Yang

Abstract

Mixture-of-Experts (MoE) architectures have become essential for scaling large language models, with recent work demonstrating the benefits of fine-grained expert designs. Training such models from scratch is expensive, and sparse upcycling from pre-trained dense models is an attractive alternative. However, we identify a structural pathology of fine-grained upcycling: when fine-grained experts are derived from a single source model, naive routing collapses and downstream accuracy drops to near-random (e.g., on Qwen3-1.7B, Drop-Upcycling-fine-grained reaches only 23.2% average accuracy across 15 benchmarks, essentially matching from-scratch training at 22.2%, while the same method's coarse-grained variant reaches 50.2%). We propose DivMoE, the first framework achieving fine-grained MoE Upcycling with structurally-balanced routing. DivMoE introduces domain-specialized fine-grained expert initialization, deriving experts from dense models that have undergone domain-adaptive continual pre-training, and diversity-constrained routing, a hard structural constraint guaranteeing that each token activates experts from distinct domain groups. Across two base models and 15 benchmarks, DivMoE consistently outperforms six upcycling baselines (55.6% vs. 51.6% for the strongest baseline on Qwen3-1.7B) and strictly improves over the dense base model on every benchmark after Stage 2 continual pre-training -- closing the regression gap that has plagued prior fine-grained upcycling. After supervised fine-tuning on a public reasoning mixture, our 12B-parameter DivMoE model matches Moonlight-MoE (16B) at 64.5% average accuracy while outperforming a controlled NVIDIA-Upcycling baseline by 6.1 percentage points.

Chinese Translation

混合专家（Mixture-of-Experts, MoE）架构已成为扩展大型语言模型的关键，近期研究展示了细粒度专家设计的优势。从零开始训练此类模型的成本高昂，而从预训练的稠密模型进行稀疏升级则是一种具有吸引力的替代方案。然而，我们发现细粒度升级存在结构性病态：当细粒度专家来源于单一源模型时，简单的路由会崩溃，导致下游准确率接近随机（例如，在Qwen3-1.7B上，Drop-Upcycling-细粒度在15个基准测试中的平均准确率仅为23.2%，基本与从零开始训练的22.2%相当，而同一方法的粗粒度变体则达到了50.2%）。我们提出了DivMoE，这是第一个实现具有结构平衡路由的细粒度MoE升级的框架。DivMoE引入了领域专用的细粒度专家初始化，从经过领域自适应持续预训练的稠密模型中衍生专家，并采用了多样性约束路由，这是一种硬性结构约束，确保每个标记激活来自不同领域组的专家。在两个基础模型和15个基准测试中，DivMoE始终优于六个升级基线（在Qwen3-1.7B上，55.6%对比最强基线的51.6%），并在每个基准测试中严格优于稠密基础模型，经过第二阶段的持续预训练后，弥补了之前细粒度升级所面临的回归差距。在对公共推理混合进行监督微调后，我们的12B参数DivMoE模型在平均准确率上达到了64.5%，与Moonlight-MoE（16B）相匹配，同时比受控的NVIDIA-Upcycling基线高出6.1个百分点。

View on arXiv Download PDF AI Translation

cs.AI / 35 / 2610.11318

Finsler Flow Matching: Dynamics-Aware Geodesic Interpolation for Single-Snapshot Trajectory Inference

芬斯勒流匹配：基于动态感知的测地线插值用于单快照轨迹推断

Canova, Niklas, Fleck, Jonas Simon

Abstract

Single-cell snapshot data can resolve a continuum of cellular states but do not uniquely determine the dynamics governing transitions between them. However, additional dynamical information can often be encoded in a cell-cell Markov transition kernel. Existing generative approaches for single cell trajectory inference either infer transport only from population marginals, impose a symmetric geometry on the state space, or incorporate directionality through a single velocity vector at each observed state. We introduce Finsler Flow Matching (FFM), a framework for learning continuous stochastic dynamics from discrete Markov transition graphs. We use the first and second local moments to construct a Finsler structure motivated by the Freidlin--Wentzell action, where the second moment determines anisotropic accessibility and the first moment introduces a preferred direction of motion. We learn neural approximations of the resulting directed geodesics, use their Finsler cost to construct source-target couplings, and define geometry-aware stochastic conditional paths that can be distilled into a continuous generative process through simulation-free score and flow matching. Across synthetic and single-cell trajectory inference benchmarks, FFM improves recovery of withheld intermediate populations, particularly when the transition dynamics are strongly directional or anisotropic. Our results provide a principled route from discrete transition probabilities to continuous generative dynamics while retaining both directional and diffusive structure.

Chinese Translation

单细胞快照数据能够解析细胞状态的连续性，但并不能唯一确定控制状态之间转变的动力学。然而，额外的动态信息通常可以通过细胞间的马尔可夫转移核进行编码。现有的单细胞轨迹推断生成方法要么仅从种群边际推断传输，要么在状态空间上施加对称几何，或通过每个观察状态的单一速度向量引入方向性。我们提出了芬斯勒流匹配（Finsler Flow Matching, FFM），这是一个从离散马尔可夫转移图学习连续随机动力学的框架。我们使用第一和第二局部矩构建一个受Freidlin-Wentzell作用启发的芬斯勒结构，其中第二矩决定各向异性可达性，第一矩引入运动的优选方向。我们学习结果导向测地线的神经近似，利用其芬斯勒成本构建源-目标耦合，并定义几何感知的随机条件路径，这些路径可以通过无模拟的评分和流匹配提炼为连续生成过程。在合成和单细胞轨迹推断基准测试中，FFM改善了对被隐含中间种群的恢复，特别是在转变动力学强烈具有方向性或各向异性时。我们的结果提供了一条从离散转移概率到连续生成动力学的原则性路径，同时保留了方向性和扩散结构。

View on arXiv Download PDF AI Translation

cs.AI / 36 / 2610.11328

Mine Odyssey: Benchmarking Spatial Agentic Intelligence in the Wild

矿山奥德赛：野外空间代理智能基准测试

Cao, Yuxuan, Li, Junlong, Li, Hao, He, Junxian

Abstract

Advances in foundation models are driving efforts to introduce agents to assist people in the physical world. Such agents require agentic spatial intelligence: exploring unfamiliar environments, updating spatial understanding through interaction, and adapting actions based on feedback to sustain progress toward a sequence of goals. Existing benchmarks cover only a limited range of spatial layouts, scales, and traversal requirements. We introduce Mine Odyssey, a benchmark for evaluating agentic spatial intelligence using Minecraft reconstructions of real-world locations. It comprises 180 tasks covering 30 such locations across 20 countries and regions on five continents, including 20 outdoor and 10 indoor settings. These settings span diverse spatial scales, layouts, terrains, and connectivity patterns, from Midtown Manhattan and rural Entrup to Santa Luc\'ia Hill and Buckingham Palace. We select meaningful waypoints, such as landmarks, buildings, and rooms, and manually verify their accessibility. Each task provides a natural-language instruction specifying which waypoints to visit and in what order. Completing these tasks requires agents to find accessible routes and entrances, open doors, and move between levels using stairs and ladders, while monitoring their progress and recovering from navigation errors. Across eight evaluated state-of-the-art models, GPT-6 Astra achieves the highest success rate of 85.6%. However, the second-best model, Claude Opus 5.5, completes 73.9% of tasks, while the strongest evaluated open-weight model, DeepSeek-V4.1-Flash, reaches 23.9%, highlighting substantial room for improvement in the agentic spatial intelligence of current models. Comprehensive analyses and ablation studies on Mine Odyssey reveal current models' limitations and provide insights for advancing agentic spatial intelligence.

Chinese Translation

基础模型的进展推动了将代理引入物理世界以协助人类的努力。这些代理需要具备代理空间智能：探索不熟悉的环境，通过互动更新空间理解，并根据反馈调整行动，以持续朝向一系列目标的进展。现有基准仅覆盖有限范围的空间布局、规模和穿越要求。我们提出了矿山奥德赛（Mine Odyssey），这是一个用于评估代理空间智能的基准，利用Minecraft重建真实世界位置。该基准包含180个任务，涵盖20个国家和地区的30个地点，包括20个户外和10个室内环境。这些环境跨越多样的空间规模、布局、地形和连接模式，从曼哈顿中城和乡村Entrup到圣卢西亚山和白金汉宫。我们选择了具有意义的路标，如地标、建筑和房间，并手动验证其可达性。每个任务提供自然语言指令，指定要访问的路标及其顺序。完成这些任务需要代理找到可达的路线和入口，打开门，并通过楼梯和梯子在不同楼层之间移动，同时监控其进度并从导航错误中恢复。在评估的八个最先进模型中，GPT-6 Astra的成功率最高，达到85.6%。然而，第二好的模型Claude Opus 5.5完成了73.9%的任务，而评估的最强开放权重模型DeepSeek-V4.1-Flash的完成率为23.9%，突显了当前模型在代理空间智能方面的显著改进空间。对矿山奥德赛的全面分析和消融研究揭示了当前模型的局限性，并为推进代理空间智能提供了见解。

View on arXiv Download PDF AI Translation

cs.AI / 37 / 2610.11333

TokenBank: Financial Infrastructure for AI Services

TokenBank：人工智能服务的金融基础设施

Chang, Cary, Zhou, Jialin

Abstract

AI services incur inference costs during execution, while revenue may arrive later. Changing API prices, limited upfront capital, and service failures can limit operators' ability to sustain or expand their services. Beyond reducing per-request costs, operators need to plan future spending, fund execution before revenue arrives, and obtain compensation for specified losses. This requires clear agreements across services with different pricing and execution conditions. These agreements must distinguish rights to consume services from rights to receive payments, define obligations under uncertain costs and income, and specify which failures qualify for compensation and how much can be paid. We present TokenBank, a financial infrastructure that represents these commitments through structured contracts. It supports service-consumption rights, agreements that settle API-price differences in cash (forwards), financing through limited rights to future service revenue, and protection claims for specified service failures. Contracts specify participants, covered services, validity, ownership, fulfillment conditions, and settlement rules. Evaluation combines replay of 899,441 API requests, real model-driven agent execution, and contract API tests. In a zero-discount rising-price resampling scenario, forwards reduce mean expenditure by USD 304.88 but increase its standard deviation from USD 1,152.45 to USD 1,190.82. A controlled replication with five portfolios per capital condition finds mean contribution differences between financing and self-funding of +1.0635, -0.1406, and -0.2962 experimental USD under low, baseline, and ample capital, respectively. The evaluation distinguishes contract correctness from economic effectiveness under declared economic and failure assumptions; supplier invoices and commercial revenue are unavailable.

Chinese Translation

人工智能服务在执行过程中会产生推理成本，而收入可能会在稍后到达。API价格的变化、有限的前期资本和服务故障可能限制运营商维持或扩展其服务的能力。除了降低每次请求的成本外，运营商还需要规划未来支出，在收入到达之前为执行提供资金，并获得对特定损失的赔偿。这需要在不同定价和执行条件的服务之间达成明确的协议。这些协议必须区分消费服务的权利与收取付款的权利，定义在不确定成本和收入下的义务，并明确哪些故障符合赔偿条件以及可以支付多少赔偿。我们提出了TokenBank，这是一种通过结构化合同来表示这些承诺的金融基础设施。它支持服务消费权、以现金结算API价格差异的协议（远期合约）、通过对未来服务收入的有限权利进行融资，以及针对特定服务故障的保护索赔。合同规定了参与者、覆盖服务、有效性、所有权、履行条件和结算规则。评估结合了899,441个API请求的重放、真实模型驱动的代理执行和合同API测试。在零折扣上涨价格的重采样场景中，远期合约将平均支出减少了304.88美元，但将其标准差从1,152.45美元增加到1,190.82美元。在每个资本条件下进行的五个投资组合的受控复制实验发现，融资与自筹资金之间的平均贡献差异分别为+1.0635、-0.1406和-0.2962实验美元，适用于低、中等和充足资本。评估区分了在声明的经济和故障假设下合同的正确性与经济有效性；供应商发票和商业收入不可用。

View on arXiv Download PDF AI Translation

cs.AI / 38 / 2610.11334

ReCast: Attribution-Oriented Step Representation Learning for LLM-Based Agent Systems

ReCast：面向归因的步骤表示学习用于基于LLM的智能体系统

Jin, Weilin, Wang, Mingyu, Zhu, Taiyu, Zhou, Ziqi, Li, Wenbo, Huang, Haoyang, Duan, Nan, Wu, Yifan, Li, Ying, Wu, Zhonghai

Abstract

In LLM-based agent systems, failures can originate from early steps whose effects propagate through subsequent interactions, making their origins difficult to identify. To trace such failures back to their origin, failure attribution has been formulated as the task of identifying the earliest step responsible for the failure. Recent methods leverage LLM internal signals for failure attribution, typically using hidden states as step representations. We therefore conduct an empirical study to evaluate how effectively these representations distinguish root-cause steps from other steps and find limited separation. Motivated by this observation, we propose ReCast, a step representation learning method that transforms hidden states from a frozen LLM into attribution-oriented step representations. ReCast first selects attribution-relevant layers, then constructs complementary pattern and deviation features, and finally learns contextualized step representations through an encoder trained with contrastive and ranking objectives. We also introduce ReCast-2K, a training dataset for failure attribution. ReCast achieves the best Hit@1 across four benchmarks, surpassing the strongest baseline by 5.65 and 9.19 pp on Who&When Algorithm and Handcrafted, respectively. Code is available at https://anonymous.4open.science/r/ReCast-5FB6 .

Chinese Translation

在基于LLM的智能体系统中，故障可能源于早期步骤，其影响通过后续交互传播，使得其来源难以识别。为了追溯这些故障的根源，故障归因被定义为识别导致故障的最早步骤的任务。最近的方法利用LLM内部信号进行故障归因，通常使用隐藏状态作为步骤表示。因此，我们进行了一项实证研究，以评估这些表示在多大程度上有效地区分根本原因步骤与其他步骤，并发现分离效果有限。基于这一观察，我们提出了ReCast，一种步骤表示学习方法，将来自冻结的LLM的隐藏状态转化为面向归因的步骤表示。ReCast首先选择与归因相关的层，然后构建互补的模式和偏差特征，最后通过一个使用对比和排序目标训练的编码器学习上下文化的步骤表示。我们还引入了ReCast-2K，一个用于故障归因的训练数据集。ReCast在四个基准测试中实现了最佳的Hit@1，分别超越最强基线5.65和9.19个百分点，针对Who&When算法和手工设计数据集。代码可在 https://anonymous.4open.science/r/ReCast-5FB6 获取。

View on arXiv Download PDF AI Translation

cs.AI / 39 / 2610.11344

EvoSim: Learning to Model, Modeling to Learn

EvoSim：学习建模，建模以学习

Song, Yun-Wei, Tao, Jinkai, Zhang, Jun-Dong, Zhang, Rui, Wu, Yi-Min, Zhang, Qiang

Abstract

Physics-based models connect scientific explanation with quantitative prediction. Constructing them requires selecting physical processes, defining states and governing equations, specifying couplings, and identifying parameters from experiments. Existing AI systems remain limited in making these model structure decisions autonomously. We introduce EvoSim, a self-evolving AI scientist for physical modeling. It uses experimental discrepancies to drive mechanism and equation revisions and held-out experimental data to test physical plausibility. Exploration traces make updates to knowledge, skills, and multi-agent orchestration. This co-evolution improves physics-based models and EvoSim's ability to select mechanisms, diagnose failures, and coordinate research. We evaluate EvoSim on two industrial battery modeling tasks. It predicts lithium-metal-plating onset from 25 to 45 degrees Celsius and 2 C to 6 C with a mean absolute error of 1.79% in state of charge. Dynamic voltage prediction under vehicle driving conditions achieves a root mean square error of 7.62 mV, surpassing the reported accuracy of models developed by human experts. Self-evolution reduces model and physics errors by approximately 36% relative to baseline, demonstrating improved scientific modeling capability. EvoSim turns experimental observations into validated models and cumulative research expertise.

Chinese Translation

基于物理的模型将科学解释与定量预测相连接。构建这些模型需要选择物理过程、定义状态和控制方程、指定耦合关系以及从实验中识别参数。现有的人工智能系统在自主做出这些模型结构决策方面仍然有限。我们介绍了EvoSim，一种用于物理建模的自我进化人工智能科学家。它利用实验差异推动机制和方程的修订，并使用保留的实验数据来测试物理的合理性。探索轨迹对知识、技能和多智能体协调进行更新。这种共同进化改善了基于物理的模型以及EvoSim选择机制、诊断故障和协调研究的能力。我们在两个工业电池建模任务上评估了EvoSim。它在25到45摄氏度和2 C到6 C的条件下预测锂金属沉积的起始点，充电状态的平均绝对误差为1.79%。在车辆驾驶条件下的动态电压预测实现了7.62 mV的均方根误差，超越了人类专家开发的模型所报告的准确性。自我进化使模型和物理误差相对于基线减少了约36%，展示了改进的科学建模能力。EvoSim将实验观察转化为经过验证的模型和累积的研究专长。

View on arXiv Download PDF AI Translation

cs.AI / 40 / 2610.11345

SynCo: Data Synthesis Co-Training for Self-Evolving LLMs via Multi-Agent Reinforcement Learning

SynCo：通过多智能体强化学习实现自我进化大语言模型的数据合成协同训练

Yang, Wei, Li, Shawn, Qin, Yuehan, Wang, Yawei, Wang, Mingxi, Li, Shixuan, Yang, Tiankai, Li, Jiate, Thomason, Jesse, Ma, Xuezhe, Zhao, Yue

Abstract

Self-evolving LLM agents promise to improve autonomously through continual interaction and learning, reducing their dependence on manually curated supervision. Realizing this promise requires not only updating the agent, but also evolving its training experience as its capabilities change. However, most existing pipelines rely on static datasets or separately updated synthesis models, causing previously useful tasks to become trivial while overly difficult tasks remain uninformative. This growing mismatch between agent capability and training experience limits sustained self-improvement. To address this problem, we propose SynCo, an agentic data synthesis co-training framework for self-evolving LLMs based on multi-agent reinforcement learning. SynCo jointly optimizes two independently parameterized agents: a Synthesizer that constructs training tasks from the Reasoner's evolving capability state, and a Reasoner that learns from the resulting experience. Each synthesized task induces multiple Reasoner rollouts whose outcomes provide complementary rewards to both agents. Correctness feedback improves the Reasoner, while task quality, answer reliability, and outcome-grounded teachability guide the Synthesizer. Their updates are fed back into subsequent synthesis rounds, allowing the task-solving policy and its training distribution to evolve together. Extensive experiments across eight mathematical reasoning benchmarks demonstrate that SynCo substantially outperforms a broad range of existing synthetic-data methods and controlled baselines, achieving the strongest overall performance while deriving most of its gains from previously unsolved problems.

Chinese Translation

自我进化的大语言模型（LLM）代理承诺通过持续的互动和学习实现自主改进，从而减少对人工策划监督的依赖。实现这一承诺不仅需要更新代理，还需要随着其能力的变化而演变其训练经验。然而，大多数现有的流程依赖于静态数据集或单独更新的合成模型，导致之前有用的任务变得微不足道，而过于困难的任务则缺乏信息量。这种代理能力与训练经验之间日益增长的不匹配限制了持续的自我改进。为了解决这个问题，我们提出了SynCo，这是一个基于多智能体强化学习的自我进化大语言模型的代理数据合成协同训练框架。SynCo共同优化两个独立参数化的代理：一个合成器（Synthesizer），根据推理者（Reasoner）不断演变的能力状态构建训练任务；一个推理者，从生成的经验中学习。每个合成的任务引发多个推理者的回合，其结果为两个代理提供互补的奖励。正确性反馈改善推理者，而任务质量、答案可靠性和基于结果的可教性则指导合成器。它们的更新反馈到后续的合成轮次中，使得任务解决策略及其训练分布能够共同演变。在八个数学推理基准上的广泛实验表明，SynCo在广泛的现有合成数据方法和控制基线中显著优于其他方法，实现了最强的整体性能，同时其大部分增益来自于以前未解决的问题。

View on arXiv Download PDF AI Translation

cs.AI / 41 / 2610.11352

RL-ARC: Calibrating Large Reasoning Models via Reasoning-guided Uncertainty

RL-ARC：通过推理引导的不确定性校准大型推理模型

Lee, Gukhyeon, Lee, SangKeun

Abstract

Language models (LMs) are commonly trained with Reinforcement Learning with Verifiable Rewards (RLVR) to enhance their reasoning capabilities. However, since RLVR does not explicitly account for calibration during training, it can lead to severe calibration degradation, including overconfidence. Recent calibration-aware training methods for LMs, which incorporate objectives for uncertainty estimation into training, improve calibration but still exhibit overconfidence under distribution shift, while sacrificing reasoning performance. To this end, we propose RL-ARC, a calibration-aware training framework that jointly leverages reasoning confidence and answer confidence. Specifically, RL-ARC leverages reasoning confidence as an auxiliary signal for calibrating answer confidence, applying it as reasoning-guided regularization for correct cases and as an overconfidence penalty for incorrect cases. Comprehensive results across ID and OOD settings show that, beyond improving calibration, RL-ARC enables reasoning models to adaptively estimate confidence based on the given question without substantially sacrificing reasoning performance, thereby highlighting the importance of reasoning confidence for training reliable reasoning models.

Chinese Translation

语言模型（LMs）通常通过可验证奖励的强化学习（RLVR）进行训练，以增强其推理能力。然而，由于RLVR在训练过程中并未明确考虑校准，这可能导致严重的校准退化，包括过度自信。近期针对语言模型的校准意识训练方法，通过将不确定性估计目标纳入训练，改善了校准效果，但在分布转移下仍表现出过度自信，同时牺牲了推理性能。为此，我们提出了RL-ARC，一个校准意识训练框架，联合利用推理置信度和答案置信度。具体而言，RL-ARC将推理置信度作为校准答案置信度的辅助信号，将其作为正确案例的推理引导正则化，并作为错误案例的过度自信惩罚。在ID和OOD设置下的全面结果表明，除了改善校准外，RL-ARC使推理模型能够根据给定问题自适应地估计置信度，而不会显著牺牲推理性能，从而强调了推理置信度在训练可靠推理模型中的重要性。

View on arXiv Download PDF AI Translation

cs.AI / 42 / 2610.11355

Writing for the Reviewer: Defensive Writing in GPT Models

为审稿人而写：GPT模型中的防御性写作

Liao, Junchi

Abstract

Researchers increasingly use ChatGPT to revise their papers, and recent GPT versions often narrow or even retract the authors' claims. We call such changes defensive writing when the given material does not support them, and we test two explanations: the model corrects the authors' overclaiming, or it writes for an anticipated reviewer. We ask GPT versions and models from other developers to rewrite paragraphs from papers written before ChatGPT, or to write from an evidence sheet that lists a paper's method and results. Defensive writing grows with GPT version. GPT-6-astra retracts the authors' claims outright, and when it writes from the evidence sheet, it still adds the most ungrounded qualifications. The results favor the anticipated-review explanation, and correcting overclaiming explains only a small part. When the models are only asked to polish, defense stays near the level of the originals; mentioning review raises it, and one round of self-review raises it further. At the same time, fewer than one in ten of the claims GPT-6-astra retracts are overstated. AI reviewers score defensive rewrites higher, while human readers find them harder to read and the authors less certain. Combining AI writing with AI review may amplify this style.

Chinese Translation

研究人员越来越多地使用ChatGPT来修订他们的论文，而最近的GPT版本往往会缩小甚至撤回作者的主张。当给定材料不支持这些主张时，我们称这种变化为防御性写作。我们测试了两种解释：模型纠正了作者的过度主张，或者它是为了预期的审稿人而写作。我们要求GPT版本和其他开发者的模型重写在ChatGPT之前撰写的论文段落，或根据列出论文方法和结果的证据表进行写作。防御性写作随着GPT版本的提升而增加。GPT-6-astra直接撤回了作者的主张，当它根据证据表写作时，仍然添加了最不具依据的限定条件。结果支持预期审稿人解释，而纠正过度主张仅解释了一小部分。当模型仅被要求润色时，防御性写作保持在原始水平附近；提及审稿会提高这一水平，而一次自我审稿则进一步提高。同时，GPT-6-astra撤回的主张中，过度陈述的比例不到十分之一。AI审稿人对防御性重写的评分更高，而人类读者则发现其阅读难度更大，且作者显得不那么确定。将AI写作与AI审稿结合可能会放大这种风格。

View on arXiv Download PDF AI Translation

cs.AI / 43 / 2610.11358

RaReCache: Bridging the Gap in Cross-Model KV Cache Reuse via Rank disagreement-based Selective Recomputation

RaReCache：通过基于排名不一致的选择性重新计算弥合跨模型KV缓存重用的差距

Sarkar, Sreetama, Mitra, Saptarshi, Huang, Sitao, Kundu, Souvik, Beerel, Peter A.

Abstract

Cross-model KV-cache reuse remains a key challenge in modern LLM serving. Coding agents and multi-model systems increasingly route a shared context across models: a user may switch models mid-session, or a cascade may escalate a difficult query. Because KV caches contain model-specific representations, each switch typically forces the receiving model to prefill the entire context from scratch. Recent work shows that closed-form linear maps can translate KV caches between models in the same family, but transfer accuracy degrades as the model-size gap widens. In this paper, we establish that these transfer failures are concentrated in a small subset of information-dense tokens. To bridge this gap, we introduce RaReCache, a framework that enables a large target model to decode accurately from a cache prefilled by a much smaller source via selective recomputation. RaReCache identifies these critical positions using a novel rank disagreement metric, scoring each token by the energy of its mapped KV in output directions weakly supported by the calibration data. Across two model families and five benchmarks, on a 23x parameter gap (Qwen3-0.6B to 14B) recomputing just 30% of positions retains 95-99% of the target accuracy, whereas on a 8.8x gap (Llama3-8B to 70B), recomputing 40% retains 96.5% of the target accuracy. RaReCache largely removes sensitivity to source-model size, and achieves up to a 3.04x prefill speedup. For online serving, it handles 1.8x the request throughput of target prefill on a single GPU, and at the target's saturation load, reduces median and 99th-percentile time-to-first-token (TTFT) by 5.0x and 6.4x respectively, with a 30% recompute budget. RaReCache establishes an efficient serving paradigm where small models prefill on behalf of massive targets, enabling large models to recompute only critical tokens, drastically reducing prefill latency.

Chinese Translation

跨模型KV缓存重用仍然是现代大规模语言模型（LLM）服务中的一个关键挑战。编码代理和多模型系统越来越多地在模型之间路由共享上下文：用户可能在会话中途切换模型，或者级联可能会升级一个困难的查询。由于KV缓存包含特定于模型的表示，每次切换通常迫使接收模型从头开始重新填充整个上下文。最近的研究表明，封闭形式的线性映射可以在同一家族的模型之间转换KV缓存，但随着模型规模差距的扩大，传输准确性下降。在本文中，我们确定这些传输失败集中在一小部分信息密集型的标记上。为了弥合这一差距，我们引入了RaReCache，一个框架，使得一个大型目标模型能够通过选择性重新计算，从一个规模远小于其的源模型预填充的缓存中准确解码。RaReCache使用一种新颖的排名不一致度量来识别这些关键位置，通过其映射KV在输出方向上能量的强度对每个标记进行评分，这些方向在校准数据中得到的支持较弱。在两个模型家族和五个基准测试中，在23倍参数差距（Qwen3-0.6B到14B）下，仅重新计算30%的位置就能保留95-99%的目标准确性，而在8.8倍差距（Llama3-8B到70B）下，重新计算40%能保留96.5%的目标准确性。RaReCache在很大程度上消除了对源模型规模的敏感性，并实现了高达3.04倍的预填充加速。在在线服务中，它在单个GPU上处理的请求吞吐量是目标预填充的1.8倍，并且在目标的饱和负载下，减少了中位数和99百分位数的首次令牌时间（TTFT）分别为5.0倍和6.4倍，使用了30%的重新计算预算。RaReCache建立了一种高效的服务范式，使小模型能够代表大型目标进行预填充，从而使大型模型仅重新计算关键标记，显著降低预填充延迟。

View on arXiv Download PDF AI Translation

cs.AI / 44 / 2610.11363

UniData: Universal Multimodal Instruction Generation Pipeline

UniData：通用多模态指令生成管道

Tang, Jiaqi, Wu, Yi-Feng, Zhang, Yuting, Lu, Hao, Fu, Bowen, Chen, Qing-Guo, Xu, Xiaogang, Hu, Yuwei, Lu, Shiyin, Wei, Wei, Zhang, Lei, Xu, Zhao, Luo, Weihua, Chen, Qifeng, Chen, Ying-Cong

Abstract

Multimodal Large Language Models (MLLMs) are increasingly being applied in a wider range of real-world scenarios. However, due to the substantial labor cost, creating high-quality multimodal instruction datasets for MLLMs remains a significant challenge. Although some methods propose to generate instruction data, they often face limitations in modality support and struggle with generating multi-round instructions. To address these problems, we introduce UniData, a universal instruction generation pipeline, to transform simple user requirements into multi-round, multimodal instructions. Specifically, UniData first expands user requirements into multiple diverse events. Using these events, UniData then integrates an any-to-any large model for multimodal instruction generation. Finally, UniData enhances data quality by correcting irrelevant and redundant inference flow, leveraging correlations between instruction rounds. To train this pipeline, we also build UniDataset, a dataset comprising 20,000 entries across nine modalities for improved multimodal generation. Our experiments demonstrate that UniData achieves SOTA performance in data quality and can also enhance the understanding and generation capabilities of other multimodal models.

Chinese Translation

多模态大型语言模型（MLLMs）正越来越广泛地应用于各种现实场景。然而，由于巨大的劳动成本，为MLLMs创建高质量的多模态指令数据集仍然是一个重大挑战。尽管一些方法提出了生成指令数据的方案，但它们通常在模态支持方面存在局限，并且在生成多轮指令时面临困难。为了解决这些问题，我们提出了UniData，一个通用的指令生成管道，旨在将简单的用户需求转化为多轮多模态指令。具体而言，UniData首先将用户需求扩展为多个多样化的事件。利用这些事件，UniData随后集成了一个任意对任意的大型模型进行多模态指令生成。最后，UniData通过纠正无关和冗余的推理流程，利用指令轮次之间的相关性来提升数据质量。为了训练这个管道，我们还构建了UniDataset，一个包含20,000个条目的数据集，涵盖九种模态，以改善多模态生成。我们的实验表明，UniData在数据质量方面达到了最先进的性能，并且能够增强其他多模态模型的理解和生成能力。

View on arXiv Download PDF AI Translation

cs.AI / 45 / 2610.11370

RIT-RAG: Navigating Document Corpora with Retrieval-Induced Trees

RIT-RAG：通过检索诱导树导航文档语料库

Pulivarthi, Meghanadh, Biswal, Swaraj Kumar, Bhushan, Kushagra, Nandwani, Yatin, Joshi, Sachindra, Raghu, Dinesh

Abstract

Retrieval-augmented generation (RAG) grounds language models in external corpora. Agentic RAG enables iterative search, yet exposes the model to isolated chunks without document structure, making it difficult to distinguish relevant evidence from chunks that merely resemble the query. Structure-aware methods such as PageIndex navigate document structure but cannot scale to the structures of large corpora, which do not fit in the LLM context. Hence, they first commit to a single document using a document retriever and cannot recover from a wrong choice. We propose RIT-RAG (Retrieval-Induced Tree RAG), which combines content retrieval with structural navigation. Offline, RIT-RAG builds a tree for each document from its table of contents or sitemap. At query time, it retrieves a broad set of chunks and uses their positions to induce manageable sub-trees, potentially across multiple documents. An LLM agent navigates these sub-trees, selectively reads promising nodes, and reformulates queries when needed. Thus, retrieval proposes where to look, while the agent decides what to read. Across financial, scientific, and customer-support benchmarks, RIT-RAG achieves the highest answer accuracy among vanilla, graph-based, and agentic baselines. On EntQABench, our new benchmark of 2.84 million technical-documentation webpages, it improves accuracy by 6.8 to 11.4 points over the strongest baseline across three LLMs.

Chinese Translation

检索增强生成（RAG）将语言模型与外部语料库相结合。代理式 RAG 允许迭代搜索，但使模型暴露于孤立的片段中，缺乏文档结构，这使得区分相关证据与仅仅类似于查询的片段变得困难。结构感知方法如 PageIndex 可以导航文档结构，但无法扩展到大型语料库的结构，这些结构无法适应大语言模型（LLM）的上下文。因此，它们首先使用文档检索器承诺单一文档，无法从错误选择中恢复。我们提出了 RIT-RAG（检索诱导树 RAG），它将内容检索与结构导航相结合。在离线阶段，RIT-RAG 根据每个文档的目录或网站地图构建一棵树。在查询时，它检索一组广泛的片段，并利用它们的位置诱导可管理的子树，可能跨越多个文档。一个 LLM 代理导航这些子树，选择性地读取有前景的节点，并在需要时重新构造查询。因此，检索提出了查找的方向，而代理决定阅读的内容。在金融、科学和客户支持基准测试中，RIT-RAG 在普通、基于图的和代理基线中实现了最高的答案准确率。在我们的新基准 EntQABench 上，该基准包含 284 万个技术文档网页，它在三个 LLM 上比最强基线提高了 6.8 到 11.4 个百分点的准确率。

View on arXiv Download PDF AI Translation

cs.AI / 46 / 2610.11380

Equal Path Cost, Unequal Output Effects: Understanding Perturbation Propagation in Diffusion Models

相等路径成本，不等输出效应：理解扩散模型中的扰动传播

Guo, Wei, Zhang, Yaowen, Ge, Xingtong, Zhang, Jun

Abstract

Diffusion models have achieved remarkable success in generative modeling, with their sampling procedures routinely modified to control generation and improve efficiency. These modifications introduce perturbations along the sampling trajectory, raising a central question: how do such perturbations affect generated output? To address this question, we develop a theoretical framework to investigate perturbation propagation, combining dynamical analysis of the sampling process with an information-theoretic characterization of output responses. Within this framework, we quantify perturbation strength using the Kullback--Leibler (KL) divergence between perturbed and reference trajectory distributions, termed as path cost, which is shown to bound, but do not determine, changes in the output distribution. Building on this analysis, we derive a response identity that connects the propagation and accumulation of local perturbations with the information captured by a selected feature mean, explaining why changes in the output distribution can remain undetected by its first-order response. We test our theoretical analysis through controlled interventions at equal path cost in pretrained diffusion models, revealing distinct patterns of output sensitivity across sampling stages and spatial frequencies. To assess whether our framework can diagnose perturbations arising from practical approximations, we apply it to cache-based acceleration and show that our propagation analysis reliably identifies sampling intervals where caching causes larger image errors.

Chinese Translation

扩散模型在生成建模方面取得了显著成功，其采样过程常常被修改以控制生成和提高效率。这些修改在采样轨迹上引入了扰动，提出了一个核心问题：这些扰动如何影响生成的输出？为了解决这个问题，我们开发了一个理论框架来研究扰动传播，结合了采样过程的动态分析和输出响应的信息论特征。在这个框架内，我们使用Kullback–Leibler (KL) 散度来量化扰动强度，该散度衡量了扰动轨迹与参考轨迹分布之间的差异，称为路径成本。研究表明，路径成本可以界定但不决定输出分布的变化。在此分析的基础上，我们推导出一个响应恒等式，将局部扰动的传播和积累与所选特征均值所捕获的信息联系起来，解释了为何输出分布的变化可能在其一阶响应中未被检测到。我们通过在预训练的扩散模型中进行等路径成本的控制干预来测试我们的理论分析，揭示了在采样阶段和空间频率上输出敏感性的不同模式。为了评估我们的框架是否能够诊断来自实际近似的扰动，我们将其应用于基于缓存的加速，并显示我们的传播分析可靠地识别出缓存导致更大图像误差的采样间隔。

View on arXiv Download PDF AI Translation

cs.AI / 47 / 2610.11384

Environmental Feedback Modeling Matters: Rethinking Feedback Treatment in Agentic Hindsight Self-Distillation

环境反馈建模的重要性：重新思考代理后视自蒸馏中的反馈处理

Guo, Hangxi, Liu, Fengyuan, Wang, Yue, Qi, Yuhua, Xiong, Haoyi, Sun, Fei, Du, Mengnan

Abstract

Reinforcement learning is commonly used to train language agents in interactive environments, but cannot be directly applied when rewards are unavailable. Recent methods use environmental feedback as privileged context for hindsight self-distillation, but our analysis suggests that simply conditioning the teacher on feedback is insufficient, motivating us to rethink how environmental feedback is used in agentic self-distillation. Given that environmental feedback contains rich supervision for modeling how the environment responds to agent actions, we introduce \textit{agentic SElf-distilLation with environmental Feedback modeling} (SELF), a framework that jointly optimizes environmental feedback modeling and hindsight self-distillation. SELF learns to predict environmental responses while distilling guidance from a feedback-conditioned self-teacher into the policy. Our analysis reveals a mutually reinforcing mechanism: environmental feedback modeling strengthens hindsight supervision and policy learning, while self-distillation enhances the model's ability to model environmental feedback. With Qwen3-8B, SELF outperforms SDPO and GRPO by 6.4 and 4.1 percentage points in $\tau$-bench success rate, and by 10.71 and 3.57 percentage points in AppWorld task goal completion, respectively. These results show that SELF uses environmental feedback more effectively within agentic self-distillation, improving agent capabilities.

Chinese Translation

强化学习通常用于在互动环境中训练语言代理，但在奖励不可用时无法直接应用。近期的方法利用环境反馈作为后视自蒸馏的特权上下文，但我们的分析表明，仅仅将教师模型与反馈进行条件化是不够的，这促使我们重新思考环境反馈在代理自蒸馏中的使用。鉴于环境反馈包含了丰富的监督信息，用于建模环境如何对代理行为作出反应，我们提出了 extit{代理自蒸馏与环境反馈建模}（SELF）框架，该框架联合优化环境反馈建模和后视自蒸馏。SELF学习预测环境响应，同时将基于反馈的自教师的指导蒸馏到策略中。我们的分析揭示了一种相互增强的机制：环境反馈建模增强了后视监督和策略学习，而自蒸馏提升了模型对环境反馈的建模能力。在Qwen3-8B上，SELF在$ au$-bench成功率上比SDPO和GRPO分别提高了6.4和4.1个百分点，在AppWorld任务目标完成率上提高了10.71和3.57个百分点。这些结果表明，SELF在代理自蒸馏中更有效地利用了环境反馈，从而提升了代理的能力。

View on arXiv Download PDF AI Translation

cs.AI / 48 / 2610.11392

TypedBench: A Benchmark for Calibration, Framing Sensitivity, and Cost in System One Decision Models

TypedBench：用于系统一决策模型中的校准、措辞敏感性和成本的基准测试

Sharma, Rahul, Ducan, Andrew B., Caron, Gaétan Marceau, Vollmer, Sebastian J.

Abstract

System One models output calibrated probabilities over typed answers such as categorical choices, ordinal levels, or binary outcomes, via a non-generative interface. Software can act on these probabilities through thresholds, cost-weighted choices, and escalation rules. Consequently, if these probabilities are miscalibrated or wording-sensitive, the software ma take unintended actions leaving human operators with no textual rationale to inspect. Current evaluations largely report accuracy and calibration on public classification datasets without a clear reference. We present TypedBench, a benchmark for typed decision models like Jev built from seven policy-labelled generators and nine evaluation suites. We report accuracy as median and range across paraphrases, and calibration error relative to the finite-sample noise floor of a matched, perfectly calibrated predictor. We assess probability quality through selective prediction, ordinal proper scoring rules, and realised cost under asymmetric cost matrices. We evaluate a hosted model, an open encoder, and a family of open decoders spanning 0.8B-9B parameters on identical items. The hosted model follows the stated policy but is wording-sensitive and systematically underconfident; under asymmetric costs, using its probabilities can be worse than taking its top answer. Decoders route exactly and are slow as options or questions are added. The decoder is least accurate on policy questions and degrades with more options. Overall, typed decision models must be evaluated jointly on policy adherence, wording robustness, probability quality, and induced decision outcomes.

Chinese Translation

系统一模型通过非生成接口输出针对类型化答案（如分类选择、序数水平或二元结果）的校准概率。软件可以通过阈值、加权选择和升级规则对这些概率进行操作。因此，如果这些概率存在校准不准确或措辞敏感的问题，软件可能会采取意想不到的行动，使人类操作员无法检查文本依据。目前的评估主要报告公共分类数据集上的准确性和校准情况，但缺乏明确的参考。我们提出了TypedBench，这是一个针对类型化决策模型（如Jev）的基准，基于七个政策标记生成器和九个评估套件构建。我们报告了在不同释义下的中位数准确性和范围，以及相对于匹配的、完美校准预测器的有限样本噪声底线的校准误差。我们通过选择性预测、序数适当评分规则和在不对称成本矩阵下的实际成本来评估概率质量。我们评估了一个托管模型、一个开放编码器以及一个跨越0.8B-9B参数的开放解码器系列，所有模型在相同项目上进行测试。托管模型遵循所述政策，但对措辞敏感且系统性地缺乏信心；在不对称成本下，使用其概率可能比直接选择其最佳答案更糟。解码器的路由准确，但在添加选项或问题时速度较慢。解码器在政策问题上的准确性最低，并且随着选项增多而下降。总体而言，类型化决策模型必须在政策遵循、措辞稳健性、概率质量和诱导决策结果方面进行联合评估。

View on arXiv Download PDF AI Translation

cs.AI / 49 / 2610.11399

Estimating great expectations under autoregressive language models with potentials

在具有潜力的自回归语言模型下估计期望值

Re, Francesco I., Ghosh, Shubhangi, Vieira, Tim, Cotterell, Ryan

Abstract

Many applications of language models hinge not on individual samples but on the expectation of a test functional under the model. Estimating such expectations reliably can be computationally expensive. In this paper, we show how to make estimation more efficient by exploiting the next-token conditional probabilities which are available as a by-product of sampling. We do so through potentials: real-valued functions on prefixes that decompose the test functional additively. We construct an estimator whose variance depends on the chosen potential, and derive conditions under which a potential reduces this variance. We then develop practical potentials for several estimands and applications, and demonstrate substantial variance reductions across several estimands at comparable computational cost.

Chinese Translation

许多语言模型的应用不仅依赖于单个样本，而是依赖于模型下测试泛函的期望值。可靠地估计这种期望值可能在计算上代价高昂。本文展示了如何通过利用作为采样副产品的下一个标记条件概率来提高估计的效率。我们通过潜力（potentials）实现这一点：在前缀上的实值函数，这些函数将测试泛函加性地分解。我们构建了一个估计器，其方差依赖于所选择的潜力，并推导出潜力能够降低该方差的条件。随后，我们为多个估计量和应用开发了实用的潜力，并在可比的计算成本下展示了多个估计量的显著方差降低。

View on arXiv Download PDF AI Translation

cs.AI / 50 / 2610.11410

Cognition-Oriented Emotion Tracing from Causes to Consequences in Real-World Social Scenes

面向认知的情感追踪：从原因到后果的真实世界社交场景

Li, Hao, Zhang, Jinye, Li, Bobo, Lee, Mong-Li, Hsu, Wynne, Wang, Zheng, Fei, Hao, Zhang, Min

Abstract

Affective computing has progressed from categorical emotion recognition to open-ended affective analysis with large multimodal models. Yet affective science describes emotion as an unfolding process shaped by appraisal, regulation, and social interpretation, which remains underexplored computationally. We propose TRACE, a cognition-oriented framework that formalizes an affective episode through three interrelated stages: Condition, Affect, and Effect, integrating observable cues with cognitive factors such as internal stance and regulation of emotional display. Based on this formulation, TRACE-Bench evaluates multimodal models in real-world social scenes through five tasks spanning grounded affect recognition, regulation decoding, cause reasoning, effect reasoning, and full-chain reconstruction, with 3,746 structured question-answer pairs over 646 videos. A matched human-model comparison reveals a substantial performance gap, while affect-specialized models also generally lag behind general-purpose MLLMs. Model outputs show recurring failures, including treating displayed behavior as genuine feeling and fabricating unsupported events during long-chain generation. We further propose TRACER, a cognition-grounded structured reasoning method that couples each inference with explicit premises from factual observations, cognitive appraisals, and established upstream conclusions, forming a traceable graph of intermediate and target conclusions. TRACER outperforms all evaluated model baselines on each of the five tasks. Project page: https://cogaffc.github.io/TRACE

Chinese Translation

情感计算已经从分类情感识别发展到基于大型多模态模型的开放式情感分析。然而，情感科学将情感描述为一个由评估、调节和社会解释塑造的展开过程，这在计算上仍然未得到充分探索。我们提出了TRACE，一个面向认知的框架，通过三个相互关联的阶段：条件、情感和效果，形式化情感事件，整合可观察的线索与内部立场和情感表现调节等认知因素。基于这一框架，TRACE-Bench通过五个任务评估真实世界社交场景中的多模态模型，这些任务涵盖了基础情感识别、调节解码、原因推理、效果推理和全链重建，共有3,746个结构化问答对，涉及646个视频。与人类模型的匹配比较显示出显著的性能差距，而专注于情感的模型通常也落后于通用的多语言大模型（MLLMs）。模型输出显示出反复出现的失败，包括将表现出的行为视为真实情感，以及在长链生成过程中虚构不支持的事件。我们进一步提出了TRACER，一种基于认知的结构化推理方法，将每个推理与来自事实观察、认知评估和已建立的上游结论的明确前提结合，形成一个可追踪的中间和目标结论图。TRACER在五个任务中超越了所有评估的模型基线。项目页面：https://cogaffc.github.io/TRACE

View on arXiv Download PDF AI Translation

cs.AI / 51 / 2610.11424

Why machines will still not rule the world

为什么机器仍然无法统治世界

Landgrebe, Jobst, Smith, Barry

Abstract

In our book Why machines will never rule the world [13, 14] we argue that arti- ficial general intelligence is mathematically impossible. This is because the human beings and the processes which exhibit intelligence are complex systems whose be- haviour cannot be captured by the kinds of models that we can generate with or without computers. Proponents of contemporary machine intelligence respond with two lines of argument: a theoretical one, grounded in the universal approximation theorems for neural networks and the Church-Turing-Deutsch principle; and an em- pirical one, grounded in rapidly rising scores on standardized benchmarks. In this communication we examine and reject both responses. First, we show serious issues in the physicalist counter-argument based on the Church-Turing-Deutsch principle. Second, we review recent evidence to the effect that prominent benchmarks are compromised by training-data contamination, flawed test construction, and strate- gic optimization. Our central argument remains: That models required to perform cognitive behaviour in open-ended, thermodynamically complex and non-ergodic environments are not and will not become achievable.

Chinese Translation

在我们的书《为什么机器永远无法统治世界》中，我们认为人工通用智能在数学上是不可能的。这是因为人类及其表现出智能的过程是复杂系统，其行为无法通过我们可以生成的模型（无论是否使用计算机）来捕捉。现代机器智能的支持者提出了两条论证：一是基于神经网络的普遍逼近定理和丘奇-图灵-德意志原则的理论论证；二是基于标准基准测试分数迅速上升的经验论证。在本次交流中，我们审视并驳斥了这两种回应。首先，我们展示了基于丘奇-图灵-德意志原则的物理主义反驳中的严重问题。其次，我们回顾了最近的证据，表明显著的基准测试受到训练数据污染、测试构建缺陷和战略优化的影响。我们的核心论点仍然是：在开放式、热力学复杂和非遍历环境中，执行认知行为所需的模型并不可实现，也不会变得可实现。

View on arXiv Download PDF AI Translation

cs.AI / 52 / 2610.11450

Tracing the Thoughts of a Coding Agent Playing ARC-AGI-3: Lessons for Continual Learning

追踪编码智能体在ARC-AGI-3中思维的轨迹：对持续学习的启示

Wu, Chen, Passenger, Josh, Song, Yin

Abstract

We study how a coding agent learns across a sequence of abstract reasoning tasks. The agent runs on a frozen foundation model inside a fixed harness and acts by writing and running Python and shell scripts. It retains no state across turns other than its written artifacts, so every thought it forms, carries, corrects or abandons leaves a trace, where a thought is any belief, rule or plan committed to a file. We let the agent play ARC-AGI-3, a set of interactive reasoning games that provide no instructions. Each game is a sequence of levels, and a strategy that clears one level can fail on the next, so every new level is in effect a new task. The agent records what it learns as Python scripts and text notes, while the harness keeps a complete log of every action and observation. Our contribution is a measurement protocol that traces each thought through these files, from the task where it forms to the task where it is corrected or abandoned, applied to seven evaluation runs with three backbones from two model families. Scripts written for one task are almost never called again in a later task (33 of 630 references cross a task boundary), because most scripts embed the state of the current level. Instead, the agent rewrites its knowledge into new scripts, keeping the general rules and dropping the level-specific details, and abandons 74% of the scripts it wrote before a boundary. The notes, which only the model reads, are never revised: the agent appends without removing earlier claims, and the contradictions that accumulate are settled against the log. Because the log preserves everything, the agent forgets selectively, not catastrophically. The most costly error is a hard-coded value carried into a task where it no longer holds. These findings come from the files the agent wrote, without access to the model, and constitute a white-box analysis of how a coding agent continually learns.

Chinese Translation

我们研究了一个编码智能体如何在一系列抽象推理任务中学习。该智能体在一个固定的框架内运行于一个冻结的基础模型，通过编写和运行Python及Shell脚本进行操作。它在每个回合之间不保留状态，除了其编写的文档，因此它形成、携带、修正或放弃的每一个思维都会留下痕迹，其中思维是指任何被记录到文件中的信念、规则或计划。我们让智能体参与ARC-AGI-3，这是一组不提供说明的互动推理游戏。每个游戏都是一系列关卡，一个清除某个关卡的策略可能在下一个关卡失败，因此每个新关卡实际上都是一个新任务。智能体将其学习记录为Python脚本和文本笔记，而框架则保持每个行动和观察的完整日志。我们的贡献是一个测量协议，追踪每个思维在这些文件中的轨迹，从形成的任务到被修正或放弃的任务，应用于七次评估运行，使用来自两个模型家族的三种基础模型。为一个任务编写的脚本几乎不会在后续任务中再次调用（630个引用中只有33个跨越任务边界），因为大多数脚本嵌入了当前关卡的状态。相反，智能体将其知识重写为新的脚本，保留一般规则并丢弃特定关卡的细节，并在边界之前放弃74%的先前编写的脚本。只有模型读取的笔记从不被修订：智能体在不删除早期声明的情况下进行附加，积累的矛盾通过日志得到解决。由于日志保存了一切，智能体选择性地遗忘，而不是灾难性地遗忘。最昂贵的错误是将一个硬编码的值带入一个不再适用的任务。这些发现来自智能体编写的文件，而没有访问模型，构成了对编码智能体持续学习的白盒分析。

View on arXiv Download PDF AI Translation

cs.AI / 53 / 2610.11456

SpikeSSL: A Universal Spike Inference Framework with Dynamics-Informed State-Space Layers

SpikeSSL：一个具有动态信息状态空间层的通用尖峰推断框架

Yue, Chenghao, Xing, Siming, Liu, Shuran, Li, Angran, Zhang, Yuanlong

Abstract

Two-photon calcium imaging is a standard tool for recording large neural populations in vivo, yet inferring spikes accurately across the growing diversity of calcium indicators remains an open problem. Existing supervised methods achieve reasonable in-domain accuracy but generalize poorly to unseen indicators, because different indicators induce distinct fluorescence kinetics and signal statistics while existing architectures remain relatively simple generic temporal regressors without dynamics-matched inductive bias. We propose SpikeSSL, a universal spike inference framework whose temporal backbone is a bank of bidirectional IIR state-space layers broadly motivated by calcium dynamics. A multi-modal conditioning encoder maps indicator identity, sampling rate, and trace-level signal statistics into a global conditioning vector that modulates the backbone via Adaptive Layer Normalization, while a heteroscedastic variance head provides calibrated per-frame uncertainty. On a benchmark with five fixed evaluation splits built from 33 public ground-truth datasets, SpikeSSL achieves state-of-the-art performance in both in-domain and zero-shot leave-one-indicator-out settings. We also develop a biophysical simulation pipeline capable of generating paired fluorescence-spike traces with systematically varied kinetic parameters, spike statistics, response nonlinearities, baseline drift, and noise. Using this pipeline, we synthesize approximately 11,000 simulated traces. Augmenting training with these data effectively closes the cross-indicator domain gap and improves zero-shot generalization. Code is publicly available at https://github.com/detimage123/SpikeSSL.

Chinese Translation

双光子钙成像是记录大规模神经群体的标准工具，但在不断增长的钙指示剂多样性中准确推断尖峰仍然是一个未解决的问题。现有的监督方法在领域内取得了合理的准确性，但对未见过的指示剂的泛化能力较差，因为不同的指示剂会引起不同的荧光动力学和信号统计，而现有的架构相对简单，作为通用时间回归器缺乏与动力学匹配的归纳偏置。我们提出了SpikeSSL，一个通用的尖峰推断框架，其时间主干是一个双向IIR状态空间层的库，广泛受到钙动力学的启发。一个多模态条件编码器将指示剂身份、采样率和轨迹级信号统计映射为一个全局条件向量，通过自适应层归一化调节主干，同时异方差方差头提供校准的逐帧不确定性。在一个基于33个公共真实数据集构建的包含五个固定评估拆分的基准测试中，SpikeSSL在领域内和零-shot留一指示剂外的设置中都达到了最先进的性能。我们还开发了一个生物物理模拟管道，能够生成配对的荧光-尖峰轨迹，并系统性地改变动力学参数、尖峰统计、响应非线性、基线漂移和噪声。利用该管道，我们合成了大约11,000个模拟轨迹。通过这些数据增强训练有效缩小了跨指示剂领域的差距，并改善了零-shot泛化能力。代码可在 https://github.com/detimage123/SpikeSSL 上公开获取。

View on arXiv Download PDF AI Translation

cs.AI / 54 / 2610.11464

Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents

谁来验证验证者？与自我改进代理共同演化的可检查评分者

Zhang, Xing, Wang, Guanghui, Cui, Yanwei, Li, Ziyuan, Qiu, Wei, Zhu, Bing, He, Peiyang

Abstract

We changed the agent: did it actually get better? Every self-improving agent loop answers this hundreds of times, and every answer comes from a verifier. On open-ended tasks none exists, so the loop is handed a hand-written rubric or a bare LLM judge grading output from a model like itself, inviting reward hacking and shared blind spots. We make the verifier the evolving object: an inspectable expression over small, mostly deterministic drawback detectors, synthesized from clustered failures, gated at birth, and selected for agreement with a ten-item anchored reference set plus consensus over unlabeled outputs, never for the agent's score. On MBPP+ it gains +0.21 held-out agreement over the hand-authored seed composition, on every seed, and ends ahead of the bare LLM judge it contains. One finding should change how co-evolved verifiers are validated: removing the anchor guards collapses the verifier into a vacuous always-pass grader, yet that collapsed verifier trains skills just as well. Downstream task score cannot certify a self-evolved verifier. Score does answer sufficiency, and there an evolved verifier can substitute: Double Ratchet, pairing the verifier with a lifecycle-managed skill loop, retains 88-110% of the lift that ground truth or a rubric buys the same loop, across code generation, enterprise text-to-SQL, and reference-free report generation. When evolved skills gamed the report rubric, an outer judge caught it and one added detector repaired it; the judge itself was wrong until given the task contract.

Chinese Translation

我们改变了代理：它真的变得更好吗？每个自我改进代理循环对此进行数百次回答，而每个答案都来自一个验证者。在开放式任务中并不存在这样的验证者，因此该循环被交给手写的评分标准或来自类似模型的裸露大型语言模型（LLM）评判的输出，这引发了奖励黑客行为和共同的盲点。我们将验证者作为演化对象：一个可检查的表达，基于小型、主要是确定性的缺陷检测器，这些检测器是从聚类失败中合成的，在诞生时被限制，并根据与十项锚定参考集的协议以及对未标记输出的共识进行选择，而不是根据代理的得分。在 MBPP+ 上，它在每个种子上获得了 +0.21 的保留一致性，超越了其包含的裸露 LLM 评判者。一个发现应该改变共同演化验证者的验证方式：去除锚定保护会使验证者崩溃为一个空洞的总是通过的评分者，但那个崩溃的验证者同样能够训练技能。下游任务得分无法认证自我演化的验证者。得分确实回答了充分性，而在这一点上，演化的验证者可以替代：双重棘轮（Double Ratchet），将验证者与生命周期管理的技能循环配对，保留了88-110%的提升，这与真实值或评分标准为同一循环所带来的提升相当，涵盖代码生成、企业文本到 SQL 和无参考报告生成。当演化的技能利用了报告评分标准时，外部评判者发现了这一点，并且一个新增的检测器修复了它；评判者本身在给定任务合同之前是错误的。

View on arXiv Download PDF AI Translation

cs.AI / 55 / 2610.11472

From Chain-of-Thought to Loops: Non-Autoregressive Latent Reasoning via Looped Transformers

从思维链到循环：通过循环变换器实现非自回归潜在推理

García, Gerard Grau, Masdemont, Arnau Padrés, Grillo, Niccolò, Ros-Giralt, Jordi, Behboodi, Arash, Vendrell, Victor Conchello

Abstract

Chain-of-thought (CoT) reasoning often improves language-model performance by giving models additional computation before answering. However, explicit CoT expresses this computation as a sequence of autoregressively generated tokens. Latent reasoning replaces these tokens with compact continuous states, but most autoregressive latent-reasoning methods retain a left-to-right dependency among latent vectors. We introduce LLoCoT: a looped latent-reasoning framework that replaces left-to-right latent generation with iterative refinement of a compact latent workspace. A shared transformer is reapplied for a small number of refinement iterations, jointly updating the latent slots based on the prompt and the evolving workspace state. Using the refined state, a probabilistic head predicts a distribution from which latent tokens are sampled in parallel and used to condition an autoregressive decoder for answer generation. Training uses continuous representations derived from explicit CoT together with a final-answer prediction loss and likelihood-based supervision of the latent states. Across HumanEval and MBPP, LLoCoT achieves the highest mean among the evaluated methods, performing on par in accuracy with Reasoning SFT, our explicit-CoT baseline, while outperforming the base model, answer-only SFT and NF-CoT. Relative to Reasoning SFT, LLoCoT reduces time to the first answer token by approximately $36\times$ and reasoning-phase latency by approximately $42\times$, while increasing end-to-end throughput by $9.2\%$. This design replaces serial thought generation with parallel latent-slot refinement while retaining probabilistic latent modeling and autoregressive answer decoding.

Chinese Translation

思维链（Chain-of-thought，CoT）推理通常通过在回答之前为模型提供额外的计算来提高语言模型的性能。然而，显式的思维链将这种计算表示为一系列自回归生成的标记。潜在推理用紧凑的连续状态替代这些标记，但大多数自回归潜在推理方法在潜在向量之间保留了从左到右的依赖关系。我们引入了LLoCoT：一个循环潜在推理框架，它用紧凑潜在工作空间的迭代精炼替代了从左到右的潜在生成。一个共享的变换器被重新应用于少量的精炼迭代中，基于提示和不断演变的工作空间状态共同更新潜在槽。利用精炼后的状态，一个概率头预测一个分布，从中并行采样潜在标记，并用于条件自回归解码器以生成答案。训练使用从显式思维链派生的连续表示，以及最终答案预测损失和对潜在状态的基于似然的监督。在HumanEval和MBPP上，LLoCoT在评估方法中实现了最高的平均值，在准确性上与我们的显式思维链基线Reasoning SFT表现相当，同时优于基础模型、仅答案SFT和NF-CoT。相较于Reasoning SFT，LLoCoT将第一个答案标记的时间减少了约$36 imes$，推理阶段的延迟减少了约$42 imes$，而端到端的吞吐量提高了$9.2\%$。该设计用并行潜在槽精炼替代了串行思维生成，同时保留了概率潜在建模和自回归答案解码。

View on arXiv Download PDF AI Translation

cs.AI / 56 / 2610.11483

BridgeGuard: Explicit Safety Drift for Diffusion-based Autonomous Driving

BridgeGuard：基于扩散的自主驾驶中的显式安全漂移

Qiu, Zhenjun, Huang, Jianing, Liu, Dongang, Du, Baiyu, Niu, Yixun, Yang, Hao, Huang, Xinyu, Hu, Chuan, Liu, Shu

Abstract

Diffusion-based driving planners capture diverse behaviors but can generate unsafe trajectories under distribution shift. We propose BridgeGuard, a safety-constrained diffusion planning method that progressively strengthens a constraint term during denoising to drive intermediate trajectories toward a scene-dependent safety domain. Corrections operate in a low-dimensional curve space, promoting geometric coherence. A learned module, DistanceFieldNet, predicts a time-dependent distance field from bird's-eye-view features. Value and spatial-gradient supervision at queries sampled beyond expert trajectories teaches this field about both safe and unsafe regions. The learned field supplies the constraint term through safety injection while the pretrained perception backbone and planner remain frozen. We further establish sufficient conditions for terminal safety in an idealized continuous-time bridge. On Bench2Drive, BridgeGuard improves driving score/success rate from 87.99/74.99% to 90.88/76.36% for BridgeDrive and from 80.79/58.18% to 90.46/74.09% for $\text{DiffusionDrive}^{\text{geo}}$, demonstrating cross-model generalization.

Chinese Translation

基于扩散的驾驶规划器能够捕捉多样化的行为，但在分布转移下可能生成不安全的轨迹。我们提出了BridgeGuard，一种安全约束的扩散规划方法，该方法在去噪过程中逐步加强约束项，以将中间轨迹引导至场景依赖的安全域。修正操作在低维曲线空间中进行，促进几何一致性。一个学习模块DistanceFieldNet从鸟瞰视图特征中预测时间依赖的距离场。在超出专家轨迹的查询点上进行值和空间梯度监督，使该距离场了解安全和不安全区域。学习到的距离场通过安全注入提供约束项，同时预训练的感知骨干网络和规划器保持冻结。我们进一步为理想化的连续时间桥梁建立了终端安全的充分条件。在Bench2Drive上，BridgeGuard将BridgeDrive的驾驶得分/成功率从87.99/74.99%提升至90.88/76.36%，将$ ext{DiffusionDrive}^{ ext{geo}}$的驾驶得分/成功率从80.79/58.18%提升至90.46/74.09%，展示了跨模型的泛化能力。

View on arXiv Download PDF AI Translation

cs.AI / 57 / 2610.11502

Fed-GRPO: Reward-Signal-Driven Federated Group Relative Policy Optimization

Fed-GRPO：基于奖励信号的联邦组相对策略优化

Guo, Pengxin, Zeng, Shuang, Li, Zonggen, Zheng, Weiying, Liu, Mengting, Qu, Liangqiong

Abstract

Large Language Models (LLMs) have shown strong reasoning capabilities when fine-tuned with reinforcement learning (RL), particularly through Group Relative Policy Optimization (GRPO). However, existing GRPO methods assume centralized access to training data, which may not hold in practice due to privacy or regulatory constraints. To this end, we propose Fed-GRPO, a federated GRPO training framework that addresses these privacy constraints by enabling collaborative reasoning training without sharing raw data, which leverages the reward statistics naturally produced during GRPO training as zero-cost signals to guide aggregation, local training, and communication. Fed-GRPO contains three reward-signal-driven mechanisms: (i) \emph{signal-weighted aggregation} that weights clients by their reward standard deviation, prioritizing clients with stronger learning signals; (ii) \emph{global reward calibration} that re-weights per-prompt objectives based on the local-global reward gap, steering each client toward its relative weaknesses; and (iii) \emph{adaptive sparse communication} that allocates bandwidth based on the informativeness of each client's update. Extensive experiments on mathematical reasoning tasks demonstrate that Fed-GRPO achieves the best performance among all federated methods, clearly outperforms FedAvg and approaches centralized training performance, while losslessly reducing communication by $32\times$ and supporting up to $621\times$ compression under tight bandwidth budgets with only graceful accuracy degradation. Our code is available at https://github.com/HKU-HealthAI/Fed-GRPO.

Chinese Translation

大型语言模型（LLMs）在通过强化学习（RL）微调时展现出强大的推理能力，特别是通过组相对策略优化（GRPO）。然而，现有的GRPO方法假设可以集中访问训练数据，这在隐私或监管约束下可能并不成立。为此，我们提出了Fed-GRPO，这是一种联邦GRPO训练框架，通过在不共享原始数据的情况下实现协作推理训练，从而解决这些隐私约束，利用在GRPO训练过程中自然产生的奖励统计作为零成本信号来指导聚合、局部训练和通信。Fed-GRPO包含三种基于奖励信号的机制：（i） extit{信号加权聚合}，根据客户的奖励标准差对客户进行加权，优先考虑学习信号更强的客户；（ii） extit{全局奖励校准}，根据局部与全局奖励差距重新加权每个提示的目标，引导每个客户朝向其相对弱点；（iii） extit{自适应稀疏通信}，根据每个客户更新的信息量分配带宽。在数学推理任务上的广泛实验表明，Fed-GRPO在所有联邦方法中实现了最佳性能，明显优于FedAvg，并接近集中训练的表现，同时在通信上无损地减少了$32 imes$，并在带宽预算紧张的情况下支持高达$621 imes$的压缩，仅有轻微的准确度下降。我们的代码可在https://github.com/HKU-HealthAI/Fed-GRPO获取。

View on arXiv Download PDF AI Translation

cs.AI / 58 / 2610.11504

The Operator Mismatch Problem: Deploying BEV Perception with Portable GPU Compute

运算符不匹配问题：使用便携式 GPU 计算部署鸟瞰视图 (BEV) 感知

Verma, Rohit, Bodas, Anand V

Abstract

Modern autonomous driving systems rely on bird's-eye-view (BEV) perception models that fuse camera and LiDAR inputs to detect objects in 3D space. These models are accurate, but they cannot be deployed through standard inference runtimes. The reason is an operator mismatch between dense convolutions (which runtimes handle well), sparse 3D convolutions (which runtimes cannot represent), and geometric scatter operations (which runtimes have no vocabulary for). Today, every sparse convolution library is CUDA-only and PyTorch-coupled, locking BEV deployment to a single vendor's hardware and a single execution framework. We present BEVPIPE, a framework for deploying multimodal BEV perception pipelines using portable GPU compute APIs and integrating them with production inference runtimes. BEVPIPE partitions the model into runtime-managed dense subgraphs and three external operator extensions (voxelizer, sparse encoder, BEV projector), connected through a shared GPU memory space. BEVPIPE achieves a 19.5x end-to-end speedup over conventional deployments while retaining 98.5% of reference mAP. We also showcase that BEVPIPE is portable across different GPU backends.

Chinese Translation

现代自动驾驶系统依赖于鸟瞰视图 (BEV) 感知模型，这些模型融合了相机和激光雷达 (LiDAR) 输入，以在三维空间中检测物体。这些模型虽然准确，但无法通过标准推理运行时进行部署。原因在于密集卷积（运行时能够很好处理）、稀疏三维卷积（运行时无法表示）和几何散射操作（运行时没有相关词汇）之间存在运算符不匹配。目前，所有稀疏卷积库均为仅支持 CUDA 且与 PyTorch 绑定，这使得 BEV 部署被锁定在单一供应商的硬件和单一执行框架上。我们提出了 BEVPIPE，一个用于使用便携式 GPU 计算 API 部署多模态 BEV 感知管道的框架，并将其与生产推理运行时集成。BEVPIPE 将模型划分为运行时管理的密集子图和三个外部运算符扩展（体素化器、稀疏编码器、BEV 投影器），通过共享的 GPU 内存空间连接。BEVPIPE 在保留 98.5% 参考 mAP 的同时，实现了相较于传统部署的 19.5 倍端到端加速。我们还展示了 BEVPIPE 在不同 GPU 后端之间的可移植性。

View on arXiv Download PDF AI Translation

cs.AI / 59 / 2610.11527

AtomWorld-Mirror: Macro-Step World Modeling of Critical Evolution Backbones for Materials Dynamics

AtomWorld-Mirror：材料动态关键演化骨架的宏步世界建模

Pan, Ziming, Zhang, Ruge, Han, Haozhi, Zhou, Junkai, Chen, Xingyuan, Chen, Yifeng, Zhang, Yunquan, Cao, Ting, Liu, Yunxin, Li, Kun

Abstract

Atomistic simulation is a fundamental tool for studying long-term materials evolution, from diffusion and defect dynamics to interfacial reactions and fracture. Yet conventional simulators typically advance at microscopic resolution, spending substantial computation on low-impact local updates before reaching structurally consequential states, an evolutionary-resolution bottleneck that limits long-horizon simulation. We propose AtomWorld-Mirror, a time-aware macro-step world model for the critical evolution backbone of atomic systems. For Step-Wise atomistic simulation, AtomWorld-Mirror distills short micro-event segments into physically reachable transitions between key states, jointly predicting sparse structural edits and accumulated physical time through latent macro-step dynamics. Local reachability, inventory conservation, and continuous-time consistency constrain each transition. By amortizing local atomic physics into a reusable latent macro model and replacing explicit micro-event replay with macro-step inference, this formulation provides a path toward substantially faster prediction of long-term materials evolution while preserving structural validity and time semantics. Across five atomic systems, spanning Cu-rich RPV steel irradiation aging, Cu-Zr metallic glass, and Li$_3$N-based anti-perovskite solid electrolyte, macro-step inference delivers a speed up of $10^3$ to $10^4$ times over event-by-event simulation.

Chinese Translation

原子模拟是研究长期材料演化的基础工具，涵盖从扩散和缺陷动态到界面反应和断裂等过程。然而，传统模拟器通常以微观分辨率推进，在达到结构上重要的状态之前，花费大量计算在低影响的局部更新上，这种演化分辨率瓶颈限制了长时间范围的模拟。我们提出了AtomWorld-Mirror，一种时间感知的宏步世界模型，用于原子系统的关键演化骨架。对于逐步原子模拟，AtomWorld-Mirror将短微事件片段提炼为物理上可达的关键状态之间的转变，通过潜在的宏步动态共同预测稀疏的结构编辑和累积的物理时间。局部可达性、库存守恒和连续时间一致性约束每个转变。通过将局部原子物理学摊销到可重用的潜在宏模型中，并用宏步推理替代显式的微事件重放，该公式为显著加快长期材料演化预测提供了一条途径，同时保持结构有效性和时间语义。在五个原子系统中，包括富铜RPV钢辐照老化、Cu-Zr金属玻璃和基于Li$_3$N的反钙钛矿固体电解质，宏步推理的速度提升达到了事件逐一模拟的$10^3$到$10^4$倍。

View on arXiv Download PDF AI Translation

cs.AI / 60 / 2610.11529

ReTeach: Building a Self-Teacher through Multi-Round Reflection and Retry

ReTeach：通过多轮反思与重试构建自我教师

Tang, Yafeng, Li, Hao, Yu, Hongsheng, Fu, Qiang

Abstract

Self-distillation can improve reasoning without a separately trained, more capable teacher, but its effectiveness depends on how the self-teacher gains an advantage over the student. Conditioning the teacher on reference answers or solutions can provide such an advantage, but this information may be unavailable. Reflection offers a way to derive explicit error diagnoses and revision guidance from self-generated attempts, yet existing reflection-based methods often combine it with reference information, rich task feedback, or persistent memory. We introduce ReTeach, a Reflective self-distillation framework that constructs its self-Teacher through multi-round reflection and retry using only self-generated attempts and outcome-level verification. Starting from an unsuccessful student rollout, the teacher alternates explicit reflection with renewed attempts until success or the retry budget is exhausted, without reference answers or solutions, external diagnostic feedback, or cross-example memory. Each failed retry informs subsequent reflection, while successful correction provides outcome-level evidence for the potential utility of the resulting teacher context. An outcome-aware selection and weighting strategy distinguishes initially correct, reflection-corrected, and unresolved examples, assigning separate weights to their category-normalized distillation losses. Through on-policy distillation, the student matches the teacher's context-conditioned token-level predictive distributions at prefixes of its own rollouts, transferring the benefits of iterative correction while retaining single-pass inference. Across six benchmarks spanning mathematical reasoning, science question answering, and tool use, ReTeach improves average accuracy over GRPO by 1.39 percentage points.

Chinese Translation

自我蒸馏可以在没有单独训练的更强大教师的情况下提高推理能力，但其有效性取决于自我教师如何在学生之上获得优势。将教师与参考答案或解决方案相结合可以提供这种优势，但这些信息可能不可用。反思提供了一种从自生成的尝试中推导出明确的错误诊断和修订指导的方法，但现有的基于反思的方法通常将其与参考信息、丰富的任务反馈或持久记忆结合在一起。我们提出了ReTeach，一个反思性自我蒸馏框架，通过仅使用自生成的尝试和结果级验证，构建其自我教师，采用多轮反思与重试。从一个不成功的学生回合开始，教师在明确的反思与重新尝试之间交替进行，直到成功或重试预算耗尽，而不依赖于参考答案或解决方案、外部诊断反馈或跨示例记忆。每次失败的重试为后续反思提供信息，而成功的修正则为所得到的教师上下文的潜在效用提供结果级证据。一种关注结果的选择和加权策略区分了最初正确的、经过反思修正的和未解决的示例，为其类别归一化的蒸馏损失分配不同的权重。通过策略蒸馏，学生在其自身回合的前缀上匹配教师的上下文条件的标记级预测分布，转移迭代修正的好处，同时保留单次推理。在涵盖数学推理、科学问答和工具使用的六个基准测试中，ReTeach的平均准确率比GRPO提高了1.39个百分点。

View on arXiv Download PDF AI Translation

cs.AI / 61 / 2610.11552

Safe, Persistent, and Evolving Agent Harness for Understanding Partially Observable Worlds

安全、持久和演化的智能体框架用于理解部分可观察世界

Gao, Yisen, Guo, Yue, Zong, Qing, Guo, Yiwen, Song, Yangqiu

Abstract

Large language model agents can invoke tools fluently, but enterprise workflows demand more than selecting the right tools: actions must strictly comply with organizational policies, tool feedback often conceals hidden side effects under partial observability, and long-horizon tasks require persistent state tracking across multiple records. To address these challenges, we introduce E-Ledger, a multi-agent harness for safe and persistent execution. E-Ledger employs a code approval layer that checks every proposed action against policy before execution, and maintains a world ledger of verified hidden rules alongside evidence-backed dynamic state. Because hidden rules are typically unknown a priori, we further propose WorldAbduct, an abductive, world-model-driven harness evolution framework. WorldAbduct diagnoses execution trajectories across four complementary views (state consistency, world-observation gap, policy-gate correctness, and goal judgment) to hypothesize latent rules, and verifies them through targeted abductive interactions before integrating them into the ledger. On the enterprise benchmark World of Workflows, E-Ledger with WorldAbduct improves safe task completion across four LLM backbones, outperforming the strongest evolution baseline by 5--15 percentage points. Experiments in ScienceWorld and DiscoveryWorld further show that abductive harness evolution carries over to scientific environments. Our code is available at https://github.com/HKUST-KnowComp/E-LEDGER-WorldAbduct.

Chinese Translation

大型语言模型智能体能够流畅地调用工具，但企业工作流程不仅仅需要选择正确的工具：行动必须严格遵循组织政策，工具反馈常常在部分可观察性下掩盖隐藏的副作用，而长时间跨度的任务需要在多个记录中持久跟踪状态。为了解决这些挑战，我们提出了E-Ledger，一个用于安全和持久执行的多智能体框架。E-Ledger采用了一个代码审批层，在执行之前检查每个提议的行动是否符合政策，并维护一个包含经过验证的隐藏规则及证据支持的动态状态的世界账本。由于隐藏规则通常在事先未知，我们进一步提出了WorldAbduct，一个基于推理的、以世界模型驱动的框架演化方法。WorldAbduct通过四个互补视角（状态一致性、世界观察差距、政策门正确性和目标判断）诊断执行轨迹，以假设潜在规则，并通过针对性的推理交互验证这些规则，然后将其整合到账本中。在企业基准测试“工作流世界”上，结合WorldAbduct的E-Ledger在四个大型语言模型基础上提高了安全任务完成率，超越了最强的演化基线5至15个百分点。在ScienceWorld和DiscoveryWorld的实验进一步表明，推理框架的演化能够延续到科学环境中。我们的代码可在https://github.com/HKUST-KnowComp/E-LEDGER-WorldAbduct获取。

View on arXiv Download PDF AI Translation

cs.AI / 62 / 2610.11561

Workerville: Towards an Organizational Behavior Account of Agent Safety

Workerville：朝着代理安全的组织行为解释

Luo, Hanjun, Mao, Junting, Lu, Yuhan, Zhang, Haobo, Huang, Zhimu, Chen, Yankai, Salam, Hanan, Liu, Xue

Abstract

LLM-based agents now interact with their environments continuously, shaped by such organizational channels as user instructions, peer messages, and long-term memory. Existing safety research has examined these influences, but largely as separate agent components. How such factors jointly shape an agent's safety behavior from a unified perspective remains unmeasured. To bridge this gap, we advocate organizational behavior (OB) as a framework for studying the safety of advanced agents, reorganizing the objects of study, theoretical foundations, and experimental design around the relational structure in which agents operate. We present the first systematic formalization of counterproductive work behavior (CWB), a canonical safety-relevant subfield of OB, as Agentic Counterproductive Behavior (ACB). ACB specifies three organizational antecedents (vertical supervisor relations, horizontal peer norms, and internal cognitive structures) and maps them onto three counterproductive outcome dimensions (unauthorized disclosure, destructive operations, and production deviation). To operationalize ACB, we introduce Workerville, a controlled benchmark that manipulates organizational conditions over shared tasks, applying 16 organizational configurations to 210 tasks to yield 3,360 challenges, evaluated by human-validated agentic judges. Benchmarking 6 frontier LLMs, we find that (I) negative organizational antecedents exhibit non-monotonic amplification when combined, with the unauthorized-disclosure rate rising from 16.5% under no negative antecedent to 60.1% under two and falling back to 50.3% under three; (II) agents reproduce typical behavioral patterns predicted by human CWB research; (III) these results establish OB as a systematic framework for agent safety research, pointing toward a new research agenda.

Chinese Translation

基于大语言模型（LLM）的代理现在与其环境持续互动，这种互动受到用户指令、同伴信息和长期记忆等组织渠道的影响。现有的安全研究已经考察了这些影响，但主要是将其视为独立的代理组成部分。从统一的视角来看，这些因素如何共同塑造代理的安全行为仍未得到测量。为填补这一空白，我们倡导将组织行为（OB）作为研究先进代理安全的框架，围绕代理操作的关系结构重新组织研究对象、理论基础和实验设计。我们首次系统性地形式化了反生产性工作行为（CWB），这是OB的一个经典安全相关子领域，称为代理反生产性行为（ACB）。ACB指定了三个组织前因（纵向监督关系、横向同伴规范和内部认知结构），并将其映射到三个反生产性结果维度（未经授权的披露、破坏性操作和生产偏差）。为了操作化ACB，我们引入了Workerville，这是一个控制基准，通过共享任务操控组织条件，应用16种组织配置于210个任务，产生3360个挑战，由经过人类验证的代理评审进行评估。在对6个前沿LLM进行基准测试时，我们发现：（I）负面组织前因在结合时表现出非单调的放大效应，未经负面前因的情况下，未经授权的披露率为16.5%，在两个负面前因下上升至60.1%，而在三个负面前因下又回落至50.3%；（II）代理再现了人类CWB研究预测的典型行为模式；（III）这些结果确立了OB作为代理安全研究的系统框架，指向一个新的研究议程。

View on arXiv Download PDF AI Translation

cs.AI / 63 / 2610.11570

Scaling to Tens of Thousands of Test-Time Iterations with Loop-Native Attention Residuals

通过循环原生注意力残差扩展至数万次测试时间迭代

Li, Pengxiang, Muhtar, Dilxat, He, Di, Su, Guinan, Yin, Lu, Liu, Shiwei

Abstract

In this paper, we argue that looped Transformers need their own residual connections to prevent performance degradation as the number of iterations grows. We observe that increasing loop iterations can reduce reasoning accuracy: noisy state updates overwrite correct intermediate deductions and even undo completed solutions. This leaves subsequent iterations to recover lost information from an already degraded representation: once an error arises in an earlier loop, often as a result of long-range propagation through the recurrence, later loops find it difficult to correct. In this paper, we introduce InfiLoop, a loop-native residual connection that learns which past computations to retain and how much to accept from each new update. InfiLoop combines content-based weighting with learned temporal decay to maintain a running summary of recurrent states. An exact streaming recurrence keeps its persistent aggregation memory constant as the loop count grows. The resulting adaptive update suppresses unreliable proposals and preserves useful intermediate states. Across extensive reasoning tasks, a 7M-parameter InfiLoop model outperforms existing recursive architectures, reaching 97.9% exact accuracy on Sudoku-Extreme, and 13.6% pass@2 on ARC-AGI-2. Notably, on Sudoku-Extreme, InfiLoop continues to improve with test-time looping beyond 20,000 effective steps, showing that added depth translates directly into stronger reasoning. Our code is available at https://github.com/pixeli99/InfiLoop.

Chinese Translation

在本文中，我们论证了循环变换器需要自己的残差连接，以防止随着迭代次数的增加而导致性能下降。我们观察到，增加循环迭代次数会降低推理准确性：噪声状态更新会覆盖正确的中间推导，甚至撤销已完成的解决方案。这使得后续迭代需要从已经退化的表示中恢复丢失的信息：一旦在早期循环中出现错误，通常是由于通过递归的长距离传播，后续循环很难进行纠正。本文中，我们引入了 InfiLoop，这是一种循环原生的残差连接，能够学习保留哪些过去的计算以及从每个新更新中接受多少。InfiLoop 结合了基于内容的加权和学习的时间衰减，以维持递归状态的运行摘要。精确的流式递归在循环计数增加时保持其持久聚合记忆不变。最终的自适应更新抑制了不可靠的提议，并保留了有用的中间状态。在广泛的推理任务中，一个 7M 参数的 InfiLoop 模型超越了现有的递归架构，在 Sudoku-Extreme 上达到了 97.9% 的精确准确率，在 ARC-AGI-2 上达到了 13.6% 的 pass@2。值得注意的是，在 Sudoku-Extreme 上，InfiLoop 在测试时间循环超过 20,000 个有效步骤时继续改善，显示出增加的深度直接转化为更强的推理能力。我们的代码可在 https://github.com/pixeli99/InfiLoop 获取。

View on arXiv Download PDF AI Translation

cs.AI / 64 / 2610.11573

Memory Type Varies: Empowering LLM Agents for Long-Term Memory with Diverse Strategies

记忆类型多样化：通过多样化策略赋能大型语言模型代理的长期记忆

Wen, Yi, Xu, Derong, Jia, Pengyue, Wang, Yichao, Zhang, Yingyi, Wang, Maolin, Li, Junyi, Zhang, Wenlin, Li, Xiaopeng, Liu, Yong, Zhao, Xiangyu

Abstract

The memory capabilities of Large Language Models (LLMs) have garnered increasing attention recently. Despite great success achieved, existing retrieval-based memory approaches typically overlook the differences between memories and employ a unified strategy to process all memories, leading to suboptimal performance. Thus, an intuitive question arises: can we categorize memory into different types and select appropriate strategies? However, given the topic-rich, scenario-complex, and boundary-blurred nature of memory scenarios, achieving precise classification of memories is not easy. To address this challenge, we propose a memory multi-class dataset in this paper, termed TriMEM, which provides precise annotations for memory types across diverse scenarios. Building upon this foundation, we propose a novel memory framework, named MemoType, which can adaptively recognize each memory and query type with the learned router model. With the memory and query routing, MemoType can retrieve the memory with corresponding query types and design tailored retrieval strategies, thereby enhancing the retrieval performance. Moreover, we theoretically prove that any single retrieval strategy is subject to a fundamental upper bound on its expected retrieval precision in multi-class corpora, leading to systematic precision degradation. Extensive experiments on three datasets demonstrate that MemoType consistently outperforms existing methods, achieving up to 16.18% improvement in Recall@1.

Chinese Translation

大型语言模型（LLMs）的记忆能力近年来引起了越来越多的关注。尽管取得了巨大的成功，现有的基于检索的记忆方法通常忽视了记忆之间的差异，并采用统一的策略处理所有记忆，导致性能不佳。因此，一个直观的问题出现了：我们能否将记忆分类为不同类型并选择合适的策略？然而，鉴于记忆场景的主题丰富、情境复杂和边界模糊，精确分类记忆并非易事。为了解决这一挑战，本文提出了一个多类别记忆数据集，称为TriMEM，该数据集为不同场景中的记忆类型提供了精确的注释。在此基础上，我们提出了一种新颖的记忆框架，命名为MemoType，该框架可以通过学习的路由模型自适应地识别每种记忆和查询类型。通过记忆和查询路由，MemoType能够检索与相应查询类型相关的记忆，并设计量身定制的检索策略，从而提升检索性能。此外，我们理论上证明了在多类别语料库中，任何单一的检索策略都受到其预期检索精度的基本上限的限制，导致系统性的精度下降。在三个数据集上的大量实验表明，MemoType始终优于现有方法，在Recall@1上实现了高达16.18%的提升。

View on arXiv Download PDF AI Translation

cs.AI / 65 / 2610.11600

Error-Propagation Modeling for Failure Attribution in LLM-Based Multi-Agent Systems

基于大语言模型的多智能体系统中故障归因的误差传播建模

Liao, Jiaqi, Zhai, Yuanzhao, Liu, Huanxi, Zhang, Xu, Zhuang, Zheming, Feng, Dawei, Ding, Bo, Wang, Huaimin

Abstract

LLM-based multi-agent systems (MASs) are increasingly used to solve complex tasks through coordinated reasoning, tool use, and interaction with external resources. However, attributing failures in such systems remains challenging because the observed outcome often does not directly reveal the error responsible for the failed execution. In this work, the attribution target is the decisive error, defined as the agent--step pair whose correction would recover the failed execution. Existing approaches largely identify suspicious steps without explicitly modeling how errors propagate across interactions or persist in unresolved loops, making decisive errors difficult to distinguish from downstream failure symptoms. We propose \textbf{E}rror-Propagation \textbf{M}odeling for \textbf{F}ailure \textbf{A}ttribution (\textbf{EMFA}). EMFA constructs a structured representation of the failed trajectory, models both cascading propagation and persistent interaction loops, and uses propagation-aware candidate screening followed by counterfactual verification to identify the decisive agent--step pair. On the Who\&When benchmark, EMFA achieves state-of-the-art step-level attribution accuracy and remains competitive at the agent level. It improves the previous best step-level results by 3.45 and 4.40 percentage points on the Hand-Crafted and Algorithm-Generated subsets, respectively.

Chinese Translation

基于大语言模型的多智能体系统（MASs）越来越多地被用于通过协调推理、工具使用和与外部资源的互动来解决复杂任务。然而，在这些系统中归因故障仍然具有挑战性，因为观察到的结果往往并不能直接揭示导致执行失败的错误。在本研究中，归因目标是决定性错误，定义为修正后能够恢复失败执行的智能体-步骤对。现有方法主要识别可疑步骤，但并未明确建模错误如何在交互中传播或在未解决的循环中持续存在，这使得决定性错误难以与下游故障症状区分开来。我们提出了 extbf{E}rror-Propagation extbf{M}odeling for extbf{F}ailure extbf{A}ttribution（ extbf{EMFA}）。EMFA构建了失败轨迹的结构化表示，建模了级联传播和持久交互循环，并使用传播感知的候选筛选和反事实验证来识别决定性智能体-步骤对。在Who&When基准测试中，EMFA实现了最先进的步骤级归因准确性，并在智能体级别保持竞争力。它在手工制作和算法生成子集上分别提高了之前最佳步骤级结果3.45和4.40个百分点。

View on arXiv Download PDF AI Translation

cs.AI / 66 / 2610.11613

AgentEvolver: System-Wide Self-Evolution Through Task Execution

AgentEvolver：通过任务执行实现系统范围的自我演化

Zhang, Wentao, Yang, Fuchao, Zhao, Yilei, Wang, Xinrun, An, Bo

Abstract

An agent can complete a task without improving how it works. Turning task experience into reusable capability requires connecting the changed component to its evaluation and subsequent use. We present AgentEvolver, a system for developing capabilities during task execution while keeping the foundation model fixed. Eight entity families expose reusable operations, methods, agents, control flow, interfaces, and supporting state to revision through a common versioned lifecycle. A shared Runtime coordinates ongoing work, while persistent planning and recoverable context preserve task direction and supporting evidence. We evaluate task outcomes on SWE-bench Pro Public and examine capability changes in six application cases. The team reports an 82.08\% resolution rate with evolution, exceeding its reported baseline without evolution. The cases show retained capabilities entering later website, game, and research work, while also documenting incomplete objectives and an unsuccessful strategy. These findings distinguish improvement in a reusable component from success on the final task. AgentEvolver provides a concrete basis for studying capability accumulation through execution; independent-task transfer and total development cost remain open questions.

Chinese Translation

代理可以完成任务，而无需改善其工作方式。将任务经验转化为可重用的能力需要将变化的组件与其评估及后续使用连接起来。我们提出了AgentEvolver，这是一个在任务执行过程中开发能力的系统，同时保持基础模型不变。八个实体家族通过一个共同的版本生命周期暴露可重用的操作、方法、代理、控制流、接口和支持状态以供修订。一个共享的运行时协调正在进行的工作，而持久的规划和可恢复的上下文则保持任务方向和支持证据。我们在SWE-bench Pro Public上评估任务结果，并在六个应用案例中检查能力变化。团队报告了82.08%的解决率，伴随演化的结果超过了未演化时的报告基线。这些案例显示保留的能力进入了后续的网站、游戏和研究工作，同时也记录了未完成的目标和不成功的策略。这些发现将可重用组件的改进与最终任务的成功区分开来。AgentEvolver为通过执行研究能力积累提供了一个具体基础；独立任务转移和总开发成本仍然是未解的问题。

View on arXiv Download PDF AI Translation

cs.AI / 67 / 2610.11620

Where to Adapt Matters: Layer-Selective Fine-Tuning for Capability Retention

适应的位置至关重要：层选择性微调以保持能力

Pang, Zhiqiang, Sun, Zihong, Xie, Qi, Shu, Jun, Meng, Deyu, Xu, Zongben

Abstract

Parameter-efficient fine-tuning (PEFT) enables large language models (LLMs) to adapt to specialized tasks, but often at the cost of degrading general capabilities acquired during pretraining. Existing approaches primarily mitigate this trade-off through data replay or regularization, relying on additional data or explicit optimization constraints. We instead focus on a different question: where should adaptation be applied? We find that fine-tuning different Transformer layers produces different target-task gains and degrees of capability degradation, suggesting that not all layers are equally suitable for adaptation. To characterize this difference, we use layer-wise empirical Fisher information to measure target-task sensitivity. However, computing Fisher scores requires backward computation and becomes increasingly expensive for large models. We therefore introduce input--output cosine similarity as a lightweight, forward-only proxy for ranking layer sensitivity. Across models and tasks, layers with lower input--output similarity consistently exhibit higher empirical Fisher scores. Building on this observation, we propose Layer-Selective LoRA (LS-LoRA), which places trainable LoRA adapters only in layers with low input--output similarity. Experiments on mathematical reasoning and code generation show that LS-LoRA improves average target-task performance while retaining substantially more commonsense reasoning capability than standard all-layer LoRA, demonstrating that carefully choosing where to adapt can provide a simple and effective way to balance target-task adaptation and general capability retention.

Chinese Translation

参数高效微调（PEFT）使大型语言模型（LLMs）能够适应专业任务，但往往以牺牲在预训练过程中获得的通用能力为代价。现有的方法主要通过数据重放或正则化来缓解这种权衡，依赖于额外的数据或显式的优化约束。我们则关注一个不同的问题：适应应该应用于哪里？我们发现微调不同的Transformer层会产生不同的目标任务增益和能力退化程度，这表明并非所有层都同样适合进行适应。为了表征这种差异，我们使用逐层经验Fisher信息来测量目标任务的敏感性。然而，计算Fisher分数需要反向计算，并且对于大型模型来说成本越来越高。因此，我们引入输入-输出余弦相似度作为一种轻量级的、仅前向计算的代理，用于对层敏感性进行排名。在不同模型和任务中，具有较低输入-输出相似度的层始终表现出更高的经验Fisher分数。基于这一观察，我们提出了层选择性LoRA（LS-LoRA），仅在具有低输入-输出相似度的层中放置可训练的LoRA适配器。在数学推理和代码生成的实验中，LS-LoRA在保持显著更多常识推理能力的同时，提高了平均目标任务性能，证明了仔细选择适应的位置可以提供一种简单有效的方式来平衡目标任务适应与通用能力保持之间的关系。

View on arXiv Download PDF AI Translation

cs.AI / 68 / 2610.11651

Evidence-Traceable Dynamic Interviewer Architecture for Expertise-Adaptive Qualitative Interviews Using Local LLMs

基于证据可追溯的动态访谈者架构：使用本地大型语言模型进行适应性定性访谈

Adeseye, Aisvarya, Isoaho, Jouni, Adeseye, Adeyemi, Virtanen, Seppo, Tahir, Mohammad

Abstract

Automated interviewers and conversational agents are increasingly used in research, recruitment, customer service, and education. However, many existing systems rely on fixed question sequences and provide limited context-based personalization without considering participants' knowledge, which can lead to repetitive or irrelevant follow-up questions. Therefore, there is a need for an adaptive interviewing system that can adjust question depth while maintaining conversational continuity and semantic progression. To address this, an Evidence-Traceable Dynamic Interviewer Architecture is presented using a locally hosted Large Language Model (LLM), with the interview continuously adapted throughout the entire conversation based on the participant's responses and evolving context. The interviewer profiles participants' expertise in real time to generate knowledge-appropriate questions, well-articulated responses, and smooth transition messages that support conversational continuity. A five-module prompt-driven architecture and persistent interview-state record support these functions. The interviewer was evaluated with 246 participants. Expertise Profiling module (M3) showed 78.9% exact agreement with independently reported participant expertise, with a weighted Cohen's K of 0.80. Generate Iterative Questions module (M4) showed a strong expertise-complexity association (p=.79, p<.001), and participants reported high relevance (mean 4.41), engagement (mean 4.32), and satisfaction (mean 4.38), providing evidence that the architecture's adaptive components operated consistently with their intended functions while participants reported a positive interview experience.

Chinese Translation

自动化访谈者和对话代理在研究、招聘、客户服务和教育等领域的应用日益增加。然而，许多现有系统依赖于固定的问题序列，提供的基于上下文的个性化服务有限，未考虑参与者的知识水平，这可能导致重复或不相关的后续问题。因此，迫切需要一种适应性访谈系统，能够在保持对话连续性和语义进展的同时调整问题的深度。为此，提出了一种基于证据可追溯的动态访谈者架构，使用本地托管的大型语言模型（LLM），在整个对话过程中根据参与者的反馈和不断变化的上下文持续调整访谈内容。访谈者实时分析参与者的专业知识，以生成适合其知识水平的问题、清晰的回答以及支持对话连续性的顺畅过渡信息。该架构支持五个模块的提示驱动设计和持久的访谈状态记录。访谈者在246名参与者中进行了评估。专业知识分析模块（M3）与独立报告的参与者专业知识的精确一致性达到78.9%，加权Cohen's K值为0.80。生成迭代问题模块（M4）显示出强烈的专业知识与复杂性之间的关联（p=.79, p<.001），参与者报告了高相关性（平均4.41）、参与度（平均4.32）和满意度（平均4.38），提供了证据表明该架构的适应性组件与其预期功能一致，同时参与者报告了积极的访谈体验。

View on arXiv Download PDF AI Translation

cs.AI / 69 / 2610.11655

Harness Evolution Hits a Ceiling: When Weight Training Should Begin

控制演化遇到瓶颈：何时应开始进行权重训练

Tian, Yuan, Hu, Bing, Wang, Hao, Lu, Binghang, Wu, Fang

Abstract

Improving a long-horizon LLM agent means evolving the harness around a frozen model or training its weights. We let a self-evolving harness make the system stronger first, then cross seed and evolved harnesses with base and trained weights to learn which gains the trained model keeps and which still need the runtime. We show that the right lever can be read off the agent's failure composition: labelling failed trajectories by the first signal that fires separates process failures (blocked calls, loops, exhausted step budgets) from content failures (a delivered plan that is poor). Harness evolution repairs the former, the behaviour it instils can be trained into the weights, and content failures are what weight training is for. On DeepPlanning, a self-evolving harness loop lifts the held-out score of Qwen3.5-4B from 0.16 to 0.30 and of Qwen3.5-9B from 0.32 to 0.44; for 4B, held-out delivery rises from 55% to 90% while content failures are left for the weights. LoRA adapters trained on evolved-harness trajectories internalise the gain: under the original harness they add +0.13 on held-out tasks for both sizes; on 4B they stack with the harness to more than double the held-out score, and on 9B the adapter alone matches the full evolution line, cutting content failures from a quarter of trajectories to one in twenty. A placebo adapter trained on answer-shuffled trajectories falls below the base model. The loop transfers to WebArena-Lite (+0.09 on 117 unseen tasks), where the gain lives in what the model sees and adapters do not add to it. The result is a diagnose-then-intervene rule applied twice: read the failure composition to choose between harness and weights, then read what the accepted edits changed to decide which gains to train in. Scores are four-rollout means against fresh anchors, same-night except where marked, across eight models from six families and two benchmarks.

Chinese Translation

提升长时间范围的LLM（大语言模型）代理意味着在一个冻结模型周围演化控制或训练其权重。我们让一个自我演化的控制首先增强系统，然后交叉种子并与基础和训练权重的演化控制结合，以学习训练模型保持的增益以及哪些仍需运行时支持。我们展示了正确的杠杆可以从代理的失败组成中读取：通过第一个触发信号标记失败轨迹，将过程失败（阻塞调用、循环、耗尽步骤预算）与内容失败（交付的计划不佳）区分开来。控制演化修复前者，其所灌输的行为可以训练到权重中，而内容失败则是权重训练的目的。在DeepPlanning上，一个自我演化的控制循环将Qwen3.5-4B的持出分数从0.16提升到0.30，将Qwen3.5-9B的分数从0.32提升到0.44；对于4B，持出交付率从55%上升至90%，而内容失败则留给权重处理。基于演化控制轨迹训练的LoRA适配器内化了增益：在原始控制下，它们在两个规模的持出任务上增加了+0.13；在4B上，它们与控制叠加，使持出分数翻倍，而在9B上，适配器单独的表现与完整的演化线相匹配，将内容失败从四分之一的轨迹减少到二十分之一。一个在答案随机轨迹上训练的安慰适配器表现低于基础模型。该循环转移到WebArena-Lite（在117个未见任务上增加+0.09），增益存在于模型所见的内容中，而适配器并未对此增益产生影响。结果是一个诊断后干预的规则被应用了两次：读取失败组成以选择控制和权重之间的优先级，然后读取接受的编辑所改变的内容，以决定训练哪些增益。分数是针对新锚点的四次回合均值，除非标记外，均为同夜数据，涵盖六个家族中的八个模型和两个基准。

View on arXiv Download PDF AI Translation

cs.AI / 70 / 2610.11662

Lamarck's Driving School: Discovering Autonomous Driving Training Strategies through Evolutionary Competition

拉马克的驾驶学校：通过进化竞争发现自主驾驶训练策略

Ye, Yichun, Zhang, He, Tian, Ye, Sun, Jian

Abstract

Autonomous driving capabilities depend strongly on the distribution of scenarios encountered during training. Existing methods commonly construct or dynamically adapt training scenario distributions using surrogate criteria such as realism, difficulty, or risk. However, these predefined surrogates may misrepresent training value, leading to inefficient use of training resources. To address this limitation, we propose a Lamarckian evolutionary framework that replaces surrogate-based guidance with competition among candidate distributions. We formulate training strategy discovery as a multi-stage bilevel optimization problem and use Lamarckian evolution algorithm to approximate its solution. At the outer level, Darwinian crossover, mutation, and selection explore the scenario distribution space; at the inner level, policy learning acquires new capabilities, and Lamarckian inheritance transfers them to subsequent stages, allowing scenario distributions and policy capabilities to co-evolve. The resulting evolutionary trajectories reveal recurring stage-wise regularities among high-value distributions, characterized by capability accumulation through stage-wise challenge rotation. We further distill these regularities into a lightweight, reusable Lamarckian Training Strategy. Experiments show that, compared with the baseline, the complete framework reduces performance loss by up to 25.07%, while the lightweight strategy still achieves a 19.13% reduction. These results demonstrate that evolutionary competition can both discover effective training strategies and reveal reusable stage-wise patterns in how the value of training distributions changes with policy capability. Code is available on GitHub.

Chinese Translation

自主驾驶能力在很大程度上依赖于训练过程中遇到的场景分布。现有方法通常使用现实性、难度或风险等代理标准构建或动态调整训练场景分布。然而，这些预定义的代理可能会错误地表征训练价值，从而导致训练资源的低效使用。为了解决这一局限性，我们提出了一种拉马克进化框架，通过候选分布之间的竞争替代基于代理的指导。我们将训练策略发现形式化为一个多阶段的双层优化问题，并使用拉马克进化算法来近似其解。在外层，达尔文式的交叉、变异和选择探索场景分布空间；在内层，策略学习获取新能力，而拉马克遗传将这些能力转移到后续阶段，从而使场景分布和策略能力共同进化。由此产生的进化轨迹揭示了高价值分布之间的阶段性规律，这些规律的特征是通过阶段性挑战轮换实现能力的积累。我们进一步将这些规律提炼为一种轻量级、可重用的拉马克训练策略。实验表明，与基线相比，完整框架将性能损失减少了最多25.07%，而轻量级策略仍实现了19.13%的减少。这些结果表明，进化竞争不仅可以发现有效的训练策略，还能揭示训练分布价值随策略能力变化的可重用阶段性模式。代码可在GitHub上获取。

View on arXiv Download PDF AI Translation

cs.AI / 71 / 2610.11663

Constrained Command-Conditioned Reinforcement Learning with Bandit Strategy Selection in Real-Time Strategy Games

在实时战略游戏中采用带有策略选择的约束命令条件强化学习

Leenders, Nick, Lindelauf, Roy, van Oijen, Joost, Cule, Boris

Abstract

Deep reinforcement learning agents reach strong performance in real-time strategy games but can be brittle against opponents outside their training distribution. Separating strategic command selection from learned unit control allows different strategies to be selected for different opponents while reusing the same execution policy. This requires an executor that can follow different commands and measurable criteria for assessing whether it does so. We introduce a constrained command-conditioned Proximal Policy Optimization (PPO) policy, the executor, for MicroRTS, a real-time strategy environment. Discrete commands specify strategic objectives and behavioral requirements for economy, army composition, military posture, and worker policy over multiple environment steps; the executor determines the unit-level actions used to fulfill them. A Thompson-sampling bandit acts as the strategist, selecting command tuples from an estimate of the opponent's strategy built from in-game observations rather than opponent identity. In a controlled comparison with a flat PPO baseline trained with the same architecture, budget, curriculum and self-play league, the strategist-executor system wins significantly more often against three of the four strongest opponents on a training map, including the two strongest held-out ones (0.55 to 0.97 and 0.01 to 0.34), with no significant difference against the others.

Chinese Translation

深度强化学习代理在实时战略游戏中表现出色，但在面对训练分布之外的对手时可能会表现脆弱。将战略命令选择与学习到的单位控制分开，允许针对不同对手选择不同的策略，同时重用相同的执行策略。这需要一个能够遵循不同命令的执行器，以及可测量的标准来评估其执行效果。我们为实时战略环境MicroRTS引入了一种约束命令条件的近端策略优化（Proximal Policy Optimization, PPO）策略，即执行器。离散命令指定了多个环境步骤中的战略目标和经济、军队组成、军事姿态及工人政策的行为要求；执行器决定用于实现这些目标的单位级行动。汤普森采样（Thompson-sampling）赌博机作为战略家，从基于游戏内观察构建的对手策略估计中选择命令元组，而不是基于对手身份。在与使用相同架构、预算、课程和自对弈联赛训练的平坦PPO基线的对照比较中，战略家-执行器系统在训练地图上显著更频繁地战胜四个最强对手中的三个，包括两个最强的保留对手（胜率从0.55提升至0.97，0.01提升至0.34），而与其他对手之间没有显著差异。

View on arXiv Download PDF AI Translation

cs.AI / 72 / 2610.11696

A 3D Characterization Framework for Intelligent Sequential Decision Making

用于智能顺序决策的三维表征框架

Gojayev, Sadig, Fortuna, Carolina

Abstract

Puzzles are widely used to evaluate the reasoning capabilities of artificial intelligence (AI) systems for sequential decision making, yet approaches originating from different paradigms are rarely compared under unified conditions. To address this gap, we introduce a three-dimensional characterization framework that enables the analysts of AI methods by 1) projecting them to the Markov decision process (MDP) sequential decision making formalism, 2) degree of autonomy through human prior ranking of their designs and, 3) skill and computational cost. Using this framework, we analyze how representative graph-based, reinforcement learning, and large language model (LLM)-based approaches differ in their design choices and performance characteristics, instantiated respectively by Neurosolver, forward-backward reinforcement learning (FBRL), and automated thought-of-search (AutoToS), including a double-agent extension of thought-of-search (DA-ToS). The analysis relies on the Tower of Hanoi puzzle that provides a controlled benchmark with well-defined rules and scalable complexity, enabling consistent comparison across increasing problem sizes. The 3D characterization reveals that LLM-based methods, due to their weakly constrained action-space design, shift complexity from architecture to inference-time verification, leading to substantially higher memory and runtime costs than Neurosolver and FBRL.

Chinese Translation

拼图广泛用于评估人工智能（AI）系统在顺序决策中的推理能力，但来自不同范式的方法在统一条件下的比较却很少。为了解决这一问题，我们提出了一个三维表征框架，使AI方法的分析者能够通过以下三方面进行分析：1）将其映射到马尔可夫决策过程（MDP）顺序决策形式，2）通过人类对其设计的优先级排名来评估自主性程度，以及3）技能和计算成本。利用该框架，我们分析了基于图的、强化学习和大型语言模型（LLM）方法在设计选择和性能特征上的差异，分别由Neurosolver、前向-后向强化学习（FBRL）和自动搜索思维（AutoToS）实例化，包括搜索思维的双代理扩展（DA-ToS）。该分析依赖于汉诺塔拼图，它提供了一个具有明确规则和可扩展复杂度的受控基准，能够在不断增加的问题规模中进行一致比较。三维表征揭示，由于LLM方法在动作空间设计上约束较弱，其复杂性从架构转移到了推理时验证，导致其内存和运行时间成本显著高于Neurosolver和FBRL。

View on arXiv Download PDF AI Translation

cs.AI / 73 / 2610.11704

Intervention anchors and scientific verification in synthetic vascular predictive representations

合成血管预测表示中的干预锚点与科学验证

You, Lingsen, Guo, Yujun, Zhong, Xinyu, Peng, Zisu, Wang, Wentong, Shen, Li, Ge, Junbo

Abstract

Complete orthogonal predictive coordinates do not by themselves bind a latent direction to a named intervention. We present a mathematical and synthetic audit motivated by vascular device-vessel suitcordance. Capacity-matched least-squares predictors were exactly equivalent under complete fixed output transforms, whereas an anchor-only observer recovered interpretations only within the span of known perturbation signatures. Six three-dimensional configurations across 64 seeds gave a maximum paired prediction discrepancy of 6.7e-15 but a median untransported edit error of 1.513. Coordinate transport removed that error. Noisy and weak anchors constrained calibration stability, and changing the representation basis required recalibration or verified transport. Across 256 additional fits in dimensions 3-24, prediction equivalence persisted within 4.0e-15. We then evaluated nine deliberate runnable fault classes across 64 seeds. All 576 faulty executions completed, but each violated at least one reconstruction, prediction, delivered-edit or scope contract; all 320 valid control records passed. Repeating a faulty implementation gave exact self-agreement despite error against the separately computed simulator expectation. For one omitted-direction defect, probe coverage followed its analytic law, and rank-aware abstention protected unsupported interpretations. Scalar-noise experiments exposed both missed weak faults and excessive rejection under narrow relative tolerances. These controls provide an executable separation of prediction, semantic support and scientific acceptance. They are synthetic numerical audits, not clinical validation, neural JEPA-Anything replication, agent learning or patient treatment-effect estimation.

Chinese Translation

完整的正交预测坐标本身并不能将潜在方向绑定到一个命名的干预上。我们提出了一种数学和合成审计，旨在解决血管设备与血管的适配性问题。在完全固定输出变换下，容量匹配的最小二乘预测器是完全等价的，而仅依赖锚点的观察者只能在已知扰动特征的范围内恢复解释。六种三维配置在64个种子下给出了最大配对预测差异为6.7e-15，但中位数未传输编辑误差为1.513。坐标传输消除了该误差。噪声和弱锚点限制了校准的稳定性，而改变表示基础则需要重新校准或验证传输。在256个额外的3-24维拟合中，预测等价性保持在4.0e-15以内。随后，我们评估了64个种子下的九个故障类别。所有576个故障执行均已完成，但每个都至少违反了一个重建、预测、交付编辑或范围合同；所有320个有效控制记录均通过。重复故障实现给出了精确的自我一致性，尽管与单独计算的模拟器期望存在误差。对于一个遗漏方向的缺陷，探针覆盖遵循其解析法则，基于秩的弃权保护了不支持的解释。标量噪声实验揭示了未发现的弱故障和在狭窄相对容忍度下的过度拒绝。这些控制提供了预测、语义支持和科学接受之间的可执行分离。它们是合成数值审计，而非临床验证、神经JEPA-Anything复制、代理学习或患者治疗效果估计。

View on arXiv Download PDF AI Translation

cs.AI / 74 / 2610.11707

DeltaReplay: Task-Relative Memory Reuse for Mobile GUI Agents

DeltaReplay：面向任务的移动 GUI 代理记忆重用

Bai, Yudong, Chen, Yihong, Yao, Quanming, Wang, Yaqing

Abstract

Memory-augmented mobile GUI agents store successful execution trajectories and reuse them in later tasks, but a stored trajectory rarely matches a new task exactly. The new task may use different parameters, share only some of its steps with a stored trajectory, or have no relevant record in memory. Forcing the agent to use irrelevant memory can mislead it, whereas discarding memory that may still be useful deprives it of guidance from past experience. To address this dilemma, we propose DeltaReplay, a step-level memory reuse framework that decides how to use existing memory without modifying it. We observe that the reusable part of a stored record is determined not by the record itself but by its relation to the new task, mainly through two factors: page-level consistency and action-level generality. We therefore store execution trajectories as paths in a transition graph, whose nodes (pages) and edges (actions between pages) capture these two factors. At reuse time, the action on each edge is split into a task-independent operation and task-specific parameters. DeltaReplay then compares each recorded step with the new task and the current screen, and decides whether to follow it, execute it after replacing its parameters, or leave it to the base agent. On AndroidWorld and SPA-Bench, DeltaReplay improves the task success rate over a base agent with the same backbone by up to 10.3 and 25.0 percentage points, respectively. These results indicate that deciding at each step how to use retrieved memory lets agents benefit even from partially matching trajectories.

Chinese Translation

增强记忆的移动 GUI 代理存储成功的执行轨迹并在后续任务中重用它们，但存储的轨迹很少与新任务完全匹配。新任务可能使用不同的参数，仅与存储轨迹共享部分步骤，或者在记忆中没有相关记录。强迫代理使用无关的记忆可能会误导它，而丢弃可能仍然有用的记忆则剥夺了它从过去经验中获得指导的机会。为了解决这一困境，我们提出了 DeltaReplay，一种逐步级别的记忆重用框架，决定如何在不修改现有记忆的情况下使用它。我们观察到，存储记录的可重用部分并不是由记录本身决定的，而是由其与新任务的关系决定的，主要通过两个因素：页面级一致性和动作级通用性。因此，我们将执行轨迹存储为转移图中的路径，其节点（页面）和边（页面之间的动作）捕捉这两个因素。在重用时，每条边上的动作被拆分为与任务无关的操作和特定于任务的参数。DeltaReplay 然后将每个记录的步骤与新任务和当前屏幕进行比较，并决定是遵循该步骤、在替换参数后执行该步骤，还是将其留给基础代理。在 AndroidWorld 和 SPA-Bench 上，DeltaReplay 相较于具有相同骨干的基础代理提高了任务成功率，分别达到了 10.3 和 25.0 个百分点。这些结果表明，在每一步决定如何使用检索的记忆使代理即使从部分匹配的轨迹中也能受益。

View on arXiv Download PDF AI Translation

cs.AI / 75 / 2610.11715

Internalizer: Portable Context-to-Parameter Mapping for Very Large Language Models

Internalizer：用于超大语言模型的可移植上下文到参数映射

Devine, Peter, Ryan, Nick, Sirb, Benjamin, Chiocchi, Alex

Abstract

Hypernetworks that map a context directly to a LoRA adapter let a large language model carry that context in its weights, but prior work has demonstrated them only on base models of up to 14 billion parameters. We present the Internalizer, a state-of-the-art, portable Context-to-Parameter Mapping hypernetwork that generates document-specific LoRA adapters for the frozen 284B-parameter DeepSeek v4 Flash, a target two orders of magnitude larger than in any previous work. Most of its parameters live in a model-agnostic trunk with only thin entry and exit layers per base model, so it trains cheaply against small models before being ported to the large one. On unseen documents of up to 4096 tokens, the generated adapters reach 84.9% top-1 and 97.8% top-5 teacher-forced accuracy against 63.4% and 83.5% for the base model, with nothing in the context window but a three-word instruction. Once the hypernetwork is trained, a single forward pass turns any document into an adapter for such a model, which could be served alone for speed or alongside the document in the window to raise accuracy further.

Chinese Translation

超网络（Hypernetworks）能够直接将上下文映射到 LoRA 适配器，使得大型语言模型能够在其权重中携带该上下文，但之前的研究仅在最多 140 亿参数的基础模型上进行了验证。我们提出了 Internalizer，这是一种最先进的可移植上下文到参数映射超网络，能够为冻结的 284B 参数 DeepSeek v4 Flash 生成文档特定的 LoRA 适配器，其目标比以往任何工作大两个数量级。其大部分参数位于一个与模型无关的主干中，每个基础模型仅有薄的输入和输出层，因此它在小模型上训练成本低廉，然后再迁移到大型模型上。在未见过的最多 4096 个标记的文档上，生成的适配器在教师强制（teacher-forced）准确率上达到了 84.9% 的 top-1 和 97.8% 的 top-5，而基础模型的准确率分别为 63.4% 和 83.5%，上下文窗口中仅包含一个三词指令。一旦超网络训练完成，单次前向传播即可将任何文档转换为该模型的适配器，可以单独提供以提高速度，或与文档一起放入窗口中以进一步提高准确率。

View on arXiv Download PDF AI Translation

cs.AI / 76 / 2610.11723

MultiWorldBench: Do Independently Controlled Views Describe One Shared World?

MultiWorldBench：独立控制的视图是否描述了一个共享的世界？

Xu, Zhangbo, Zhang, Ruoxi, Hu, Rui, Wang, Yisong

Abstract

Multiplayer world models must ensure that independently controlled views remain consistent with one shared and persistent world. We introduce MultiWorldBench, a diagnostic Minecraft benchmark containing 495 case configurations across seven task suites and ten capabilities, including independent control, cross-view motion, shared-state synchronization, persistence, structural reasoning, concurrent interaction, and delayed revisit. We evaluate Solaris, Gamma-World, and MineWorld, using Engine GT as a reference. Gamma-World achieves the highest ten-capability average among the generated systems at 21.39, followed by Solaris at 20.88 and MineWorld at 1.89, while Engine GT reaches 91.69. Gamma-World performs better on several control, shared-state, and revisit capabilities, whereas Solaris leads in cross-view motion and race-condition consistency. Nevertheless, all generated systems score at most 8.00 on state persistence and 1.33 on structural consistency, and none succeeds in spatial reasoning or building-identity preservation. Human preferences produce the same overall ranking and show strong alignment with the automatic evaluation, with a mean dimension-level Spearman correlation of 0.96. These results show that plausible individual views do not yet constitute a coherent multiplayer world.

Chinese Translation

多玩家世界模型必须确保独立控制的视图与一个共享且持久的世界保持一致。我们提出了MultiWorldBench，这是一个包含495个案例配置的Minecraft基准测试，涵盖七个任务套件和十种能力，包括独立控制、跨视图运动、共享状态同步、持久性、结构推理、并发交互和延迟重访。我们使用Engine GT作为参考，评估了Solaris、Gamma-World和MineWorld。Gamma-World在生成的系统中以21.39的十种能力平均分数获得最高，Solaris以20.88紧随其后，MineWorld则为1.89，而Engine GT达到了91.69。Gamma-World在多个控制、共享状态和重访能力上表现更佳，而Solaris在跨视图运动和竞争条件一致性方面领先。然而，所有生成的系统在状态持久性上最高得分为8.00，在结构一致性上为1.33，且没有一个系统在空间推理或建筑身份保持方面取得成功。人类偏好产生了相同的总体排名，并与自动评估显示出强烈的一致性，平均维度级Spearman相关系数为0.96。这些结果表明，合理的个体视图尚未构成一个连贯的多玩家世界。

View on arXiv Download PDF AI Translation

cs.AI / 77 / 2610.11732

MemTrial: Learning When to Trust Memory in LLM Portfolio Agents

MemTrial：学习在投资组合代理中何时信任记忆

Wu, Guanghao, Cai, Zhuo, Wang, Shoujin

Abstract

Large language model (LLM) agents for portfolio management learn from experience: they credit each experience in their memory with the outcome of the decisions that used it. In financial markets, however, this outcome mostly reflects the market move shared by all decisions on that date, so the credit tracks the market rather than the experience, and these agents often do worse than simply holding the equal-weight (1/$N$) portfolio. We ask how an agent can credit an experience with what it changes, and answer it by putting memory on trial: drafts of the same decision with and without an experience face the same market, so the outcome they share cancels in their difference. Our agent, MemTrial, drafts each decision with eight combinations of its retrieved experiences, chosen by a fractional factorial design, and credits each experience with its Banzhaf value, the average of these differences. As each date occurs once and each draft is a noisy LLM sample, these credits are noisy and may not hold on new dates. MemTrial therefore pools them across dates and similar experiences with a hierarchical Bayesian model, acts on them only after they have predicted unseen dates, and otherwise stays anchored at a conservative reference such as 1/$N$. On four benchmarks, MemTrial not only benefits from experiences that matter (the best of 15 methods on a semi-synthetic benchmark with known experience quality) but also limits its losses when its values do not hold (at most 2.2\% below 1/$N$ on PortBench and InvestorBench, against 15--38\% for the best experience-learning agent). Averaged over five settings, it improves the utility of the best experience-learning agent by 21.2\%, and with eight LLMs it beats every LLM-based baseline on InvestorBench.

Chinese Translation

大型语言模型（LLM）代理在投资组合管理中通过经验学习：它们根据使用该经验的决策结果对记忆中的每个经验进行评分。然而，在金融市场中，这一结果主要反映了当天所有决策所共同经历的市场波动，因此评分跟踪的是市场而非经验，这使得这些代理的表现往往不如简单持有等权重（1/$N$）投资组合。我们探讨了代理如何能够根据其改变的内容对经验进行评分，并通过将记忆置于审判中来回答这一问题：同一决策的草稿在有和没有经验的情况下面临相同的市场，因此它们共享的结果在差异中相互抵消。我们的代理MemTrial使用分数因子设计，从其检索的经验中选择八种组合来草拟每个决策，并根据其Banzhaf值（这些差异的平均值）对每个经验进行评分。由于每个日期只出现一次，每个草稿都是一个噪声的LLM样本，因此这些评分是嘈杂的，并可能在新日期上不成立。因此，MemTrial通过层次贝叶斯模型在不同日期和相似经验之间进行汇总，仅在它们预测未见日期后才采取行动，并在其他情况下保持在保守的参考点，如1/$N$。在四个基准测试中，MemTrial不仅从重要经验中受益（在一个已知经验质量的半合成基准上是15种方法中表现最佳），而且在其价值不成立时限制了损失（在PortBench和InvestorBench上最多低于1/$N$ 2.2\%，而最佳经验学习代理的损失为15%至38%）。在五个设置中，MemTrial平均提高了最佳经验学习代理的效用21.2%，并且在使用八个LLM时，在InvestorBench上超越了所有基于LLM的基线。

View on arXiv Download PDF AI Translation

cs.AI / 78 / 2610.11735

Onboard Marine Anomaly Detection on $\Phi$sat-2: From Simulation-Based Development to In-Orbit Demonstration

在$ ext{Φ}$sat-2上的船载海洋异常检测：从基于仿真的开发到在轨演示

Szywala, Clotilde, Goudemant, Thomas, Bellizzi, Marjorie, Francesconi, Benjamin, Girard, Adrien

Abstract

Onboard Artificial Intelligence can improve responsiveness and bandwidth efficiency of Earth Observation systems by processing data directly on the satellite. This paper presents the experience gained from the development, onboard integration, and post-launch adaptation of a lightweight marine anomaly detection pipeline deployed on the European Space Agency's $\Phi$sat-2 mission. The application combines sea segmentation, self-supervised feature encoding of marine regions, generic anomaly detection based on deviations from a normal sea state, and optional characterization of selected anomaly types. Before launch, the pipeline was trained and validated on simulated $\Phi$sat-2 imagery to assess algorithmic performance and compatibility with resource-constrained onboard hardware. After integration and functional validation in the mission environment, early experiments on real $\Phi$sat-2 acquisitions revealed a significant mismatch between simulated and in-orbit data. The pipeline was therefore retrained on real Level-1 imagery using an improved annotation strategy to better handle ambiguous marine regions, substantially enhancing performance. Beyond demonstrating the onboard feasibility of the application, the $\Phi$sat-2 experience highlights the importance of robust annotation strategies and sensor-aware design, and shows that simulation-based development is valuable for pre-flight risk reduction, while reliable scientific validation requires representative in-orbit data and should be clearly distinguished from functional validation.

Chinese Translation

船载人工智能通过直接在卫星上处理数据，可以提高地球观测系统的响应能力和带宽效率。本文介绍了在欧洲航天局的$ ext{Φ}$sat-2任务中，开发、船载集成和发射后适应轻量级海洋异常检测管道所获得的经验。该应用结合了海洋分割、自监督特征编码海洋区域、基于正常海洋状态偏差的通用异常检测，以及对选定异常类型的可选特征描述。在发射前，该管道在模拟的$ ext{Φ}$sat-2影像上进行了训练和验证，以评估算法性能和与资源受限的船载硬件的兼容性。在任务环境中完成集成和功能验证后，针对真实的$ ext{Φ}$sat-2数据的早期实验显示，模拟数据与在轨数据之间存在显著不匹配。因此，该管道在真实的一级影像上进行了重新训练，采用改进的标注策略，以更好地处理模糊的海洋区域，显著提升了性能。除了展示该应用的船载可行性外，$ ext{Φ}$sat-2的经验强调了稳健的标注策略和传感器感知设计的重要性，并表明基于仿真的开发对于飞行前风险降低是有价值的，而可靠的科学验证需要具有代表性的在轨数据，并应与功能验证明确区分。

View on arXiv Download PDF AI Translation

cs.AI / 79 / 2610.11738

Scalable AI Uncertainty Quantification via Generalized Laplace Active Subspaces

通过广义拉普拉斯主动子空间实现可扩展的人工智能不确定性量化

Edeling, Wouter N., Coveney, Peter V.

Abstract

Reliable uncertainty quantification (UQ) is essential for deploying neural networks in scientific and high-stakes applications, but full Bayesian inference over the network parameters is computationally infeasible. We propose a low-rank generalized Laplace approximation for neural-network UQ based on a small number of data-informed curvature directions. Starting from a generalized Bayesian posterior defined through an empirical loss, we construct a local Gaussian approximation around a pretrained set of weights in this active curvature subspace. The posterior variances in the retained subspace are available in closed form, and the prior variance is calibrated by an empirical Bayes procedure. The generalized Bayesian formulation allows us to compare two posterior scalings: the standard Bayesian scaling associated with the summed negative log likelihood, and a mean-loss scaling in which the empirical loss is normalized by the number of data. A central finding is that the standard scaling induces a data-size dependent contraction of the posterior variance in the leading active directions. In regression problems, this can force the low-rank framework to retain additional weak-curvature directions in order to achieve nominal coverage of calibration data. When posterior samples are propagated through the non-linear network, these additional directions can degrade the coherence of the predictive intervals and shift the posterior predictive mean away from the pretrained model. In contrast, the generalized mean-loss scaling yields a more stable, lower dimensional active subspace and produces calibrated, coherent predictive confidence intervals. These results indicate that generalized Laplace active subspaces provide a practical and scalable route to calibrated uncertainty quantification in neural networks.

Chinese Translation

可靠的不确定性量化（UQ）对于在科学和高风险应用中部署神经网络至关重要，但对网络参数进行全面的贝叶斯推断在计算上是不可行的。我们提出了一种基于少量数据驱动的曲率方向的低秩广义拉普拉斯近似，用于神经网络的不确定性量化。从通过经验损失定义的广义贝叶斯后验出发，我们在这个主动曲率子空间中围绕预训练的权重集构造了局部高斯近似。保留子空间中的后验方差以封闭形式提供，先验方差通过经验贝叶斯程序进行校准。广义贝叶斯公式使我们能够比较两种后验缩放：与总负对数似然相关的标准贝叶斯缩放，以及通过数据数量归一化的经验损失的均值损失缩放。一个核心发现是，标准缩放导致后验方差在主要主动方向上依赖于数据大小的收缩。在回归问题中，这可能迫使低秩框架保留额外的弱曲率方向，以实现对校准数据的名义覆盖。当后验样本通过非线性网络传播时，这些额外方向可能会降低预测区间的连贯性，并使后验预测均值偏离预训练模型。相比之下，广义均值损失缩放产生了一个更稳定、维度更低的主动子空间，并生成经过校准的、连贯的预测置信区间。这些结果表明，广义拉普拉斯主动子空间为神经网络中的校准不确定性量化提供了一条实用且可扩展的途径。

View on arXiv Download PDF AI Translation

cs.AI / 80 / 2610.11750

Where Draft Trees Lose Target Mass: Exit-Guided Speculative Decoding

草稿树失去目标质量的地方：出口引导的推测解码

Hu, Shijing, Ren, Xuancheng, Lu, Zhihui, Zhou, Pan

Abstract

Tree-based speculative decoding verifies multiple draft continuations in one target-model pass, but finite trees built from draft scores face a fundamental draft-target mismatch. We ask whether better exact verification can increase acceptance on a fixed tree and how target feedback can improve the tree itself. Through a target-flow view, we identify a canonical exit law and prove that one plus target coverage sharply bounds the expected output-block length, including the bonus token, of any exact path verifier. All optimal verifiers share the same exit and bonus-token law, already attained by representative predraw-and-follow and sequential residual verifiers. This yields Tree Exit Verification (TEV), an exact, level-parallel procedure using one exit-node decision and one bonus-token decision. The exit law also identifies missing target probability, providing node-level feedback for Exit-Guided Draft-Tree Training (ExitTrain) on inference-time draft trees. Experiments across dialogue, code, and mathematical reasoning validate fixed-tree equivalence: ExitTrain increases average output-block length by 13%, while TEV reduces verifier-stage latency by 15%, yielding a 14% end-to-end speedup over DDTree. Our results distinguish two opportunities: better draft trees for higher acceptance and more direct verification for lower latency. Code: https://github.com/hsj576/TEV.

Chinese Translation

基于树的推测解码在一次目标模型传递中验证多个草稿延续，但由草稿分数构建的有限树面临根本的草稿-目标不匹配。我们探讨更好的精确验证是否能在固定树上提高接受率，以及目标反馈如何改善树本身。通过目标流视角，我们识别出一个典型的出口法则，并证明目标覆盖加一严格限制了任何精确路径验证器的期望输出块长度，包括奖励标记。所有最优验证器共享相同的出口和奖励标记法则，这一法则已经被代表性的预绘制跟随和顺序残差验证器所实现。这导致了树出口验证（Tree Exit Verification, TEV），这是一种精确的、层级并行的过程，使用一个出口节点决策和一个奖励标记决策。出口法则还识别出缺失的目标概率，为推理时草稿树的出口引导草稿树训练（Exit-Guided Draft-Tree Training, ExitTrain）提供节点级反馈。针对对话、代码和数学推理的实验验证了固定树的等价性：ExitTrain使平均输出块长度增加了13%，而TEV使验证器阶段延迟减少了15%，从而在DDTree上实现了14%的端到端加速。我们的结果区分了两个机会：更好的草稿树以提高接受率，以及更直接的验证以降低延迟。代码链接：https://github.com/hsj576/TEV。

View on arXiv Download PDF AI Translation

cs.AI / 81 / 2610.11754

What Output-Only Review Cannot Verify: Study Contracts for Research Agents

输出仅审查无法验证的内容：研究代理的研究合同

Waks, Eitan, Glocker, Ben

Abstract

Some defects in an AI-generated study can be identified from its artifacts; others require knowledge of what was approved before execution. We propose study contracts that bind declared experimental choices, run obligations and claim scope to recorded execution evidence, and distinguish this contract-relative verification from scientific truth. A diagnostic using eight self-authored clean/mutated pairs illustrates the information boundary. A deterministic checker applying a registered, fault-specific rule to approved and executed objects detected all eight registered mutations. Across eighteen recorded judge aliases given individual metadata-filtered packages without pair context or the registry-selected fault label, 104 of 144 mutated evaluation cases received defect flags; the remaining cases comprised 32 abstentions and eight terminal failures, with no explicit clean decisions on mutated cases. Some packages retained approval and execution fields, including digests. The prompt instructed judges to abstain when evidence was insufficient. These results characterize a deliberately information-asymmetric development setting; they do not isolate the effect of authoritative information from differences in task specification and rule selection, and they are not comparative verifier quality or agent reward hacking. We identify full-information comparisons, legitimate-adaptation controls and closed-loop agent evaluations as necessary tests of whether contract checks improve useful compliant completion under optimization.

Chinese Translation

某些AI生成研究中的缺陷可以通过其产物识别；而其他缺陷则需要在执行前对已批准内容的了解。我们提出了研究合同，这些合同将声明的实验选择、运行义务和索赔范围与记录的执行证据绑定，并将这种合同相对验证与科学真理区分开来。使用八对自创的干净/变异对进行的诊断展示了信息边界。应用注册的、特定故障的确定性检查器对已批准和执行的对象检测到所有八个注册变异。在十八个记录的评审别名中，给定个体元数据过滤的包，没有配对上下文或注册选择的故障标签，144个变异评估案例中有104个收到了缺陷标记；其余案例包括32个弃权和8个终止失败，变异案例没有明确的干净决策。一些包保留了批准和执行字段，包括摘要。提示指示评审在证据不足时弃权。这些结果描绘了一个故意信息不对称的发展环境；它们并未将权威信息的影响与任务规范和规则选择的差异隔离开来，也不是比较验证者质量或代理奖励黑客。我们确定全信息比较、合法适应控制和闭环代理评估是检验合同检查是否改善在优化下有用合规完成的必要测试。

View on arXiv Download PDF AI Translation

cs.AI / 82 / 2610.11773

Safe Actions Alone Do Not Ensure Safe Agents: Identifying Unfulfilled Obligations with Guard Models

仅有安全行为并不能确保代理的安全：通过守护模型识别未履行的义务

Feng, Youwei, Zhang, Yitong, Liu, Yuetong, Li, Jia

Abstract

Guard models are increasingly used to safeguard LLM-based agents, primarily by identifying actions that agents are forbidden to perform. However, identifying forbidden actions alone is insufficient to ensure agent safety. In this paper, we argue that agent safety also depends on identifying required yet unperformed safety-critical actions, which we call obligations. Our preliminary study on a popular benchmark for evaluating safety shows that 56.92% of GLM-5.3 trajectories contain unfulfilled obligations, compared with only 30.00% containing forbidden actions. This finding reveals unfulfilled obligations as a major and previously overlooked source of safety risk. However, to our knowledge, no existing benchmark evaluates whether guard models can identify these obligations. To close this gap, we introduce ObligationBench, the first benchmark for evaluating the capability of obligation identification, comprising 240 expert-validated trajectories covering issue resolution, feature development, and terminal operations. Our evaluation of 14 representative models reveals substantial limitations: the highest recall and exact-match rate are only 48.97% and 10.00%, respectively. To address these limitations, we develop ObligationGuard using 40,000 synthetic training examples. ObligationGuard achieves 57.52% recall and an exact-match rate of 21.67%, surpassing all evaluated models on both metrics. We call on the community to incorporate obligation identification into the design and evaluation of future guard models to improve agent safety.

Chinese Translation

守护模型越来越多地被用于保护基于大型语言模型（LLM）的代理，主要通过识别代理被禁止执行的行为。然而，仅仅识别被禁止的行为不足以确保代理的安全。本文论证了代理的安全性还依赖于识别所需但未执行的安全关键行为，我们称之为义务。我们对一个流行的安全评估基准的初步研究表明，56.92%的GLM-5.3轨迹包含未履行的义务，而仅有30.00%的轨迹包含被禁止的行为。这个发现揭示了未履行的义务是一个主要且被忽视的安全风险来源。然而，据我们所知，目前没有现有的基准评估守护模型是否能够识别这些义务。为了解决这个问题，我们引入了ObligationBench，这是第一个评估义务识别能力的基准，包含240个经过专家验证的轨迹，涵盖问题解决、特征开发和终端操作。我们对14个代表性模型的评估揭示了显著的局限性：最高的召回率和精确匹配率仅为48.97%和10.00%。为了解决这些局限性，我们开发了ObligationGuard，使用了40,000个合成训练样本。ObligationGuard的召回率达到57.52%，精确匹配率为21.67%，在这两个指标上均超过了所有评估的模型。我们呼吁社区在未来的守护模型设计和评估中纳入义务识别，以提高代理的安全性。

View on arXiv Download PDF AI Translation

cs.AI / 83 / 2610.11775

RouterInterp: Understanding Superposed Specialisation in Mixture of Experts Routing

RouterInterp：理解混合专家路由中的叠加专业化

Lasy, Ilya, Cai, Nora Yinuo, Ayonrinde, Kola

Abstract

Sparse Mixture of Experts (MoE) models scale more efficiently than dense models by routing tokens to modular expert networks that are only active for processing a fraction of tokens. A leading hypothesis for the performance of MoE models is that each expert specialises in a single, coherent domain. However, interpretability efforts that assume this hypothesis have generally been unsuccessful. We propose and present evidence for an alternative account that we call the Superposed Specialisation Hypothesis (SSH): experts specialise in a disjoint union of fine-grained features rather than one broad domain. Leveraging the SSH, we introduce RouterInterp, a method for interpreting expert routing that identifies Sparse Autoencoder features most predictive of routing decisions and produces unified natural language explanations. On gpt-oss-20b, RouterInterp explains expert routing with ${\sim}65\%$ higher detection accuracy than prior token statistics based methods. This work provides a scalable method for generating more accurate explanations of expert routing and increases our understanding of a previously uninterpretable component of foundation models.

Chinese Translation

稀疏混合专家（MoE）模型通过将令牌路由到仅在处理部分令牌时激活的模块化专家网络，比密集模型更有效地扩展。MoE模型性能的一个主要假设是，每个专家专注于一个单一且一致的领域。然而，假设这一点的可解释性努力通常未能成功。我们提出并提供证据支持一个替代解释，我们称之为叠加专业化假设（Superposed Specialisation Hypothesis, SSH）：专家专注于细粒度特征的互不重叠的并集，而不是一个广泛的领域。利用SSH，我们引入了RouterInterp，这是一种解释专家路由的方法，能够识别与路由决策最具预测性的稀疏自编码器特征，并生成统一的自然语言解释。在gpt-oss-20b上，RouterInterp以比之前基于令牌统计的方法高出约65%的检测准确率解释了专家路由。这项工作提供了一种可扩展的方法，以生成更准确的专家路由解释，并增加了我们对基础模型中一个先前不可解释组件的理解。

View on arXiv Download PDF AI Translation

cs.AI / 84 / 2610.11794

Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks

Memento 3：通过反思规则书实现基于模型的递归自我改进

Zhao, Haoyu, Yu, Zhengxu, He, Zhiyuan, Fang, Meng, Tutunov, Rasul, Bou-Ammar, Haitham, Luo, Weilin, Wang, Jun

Abstract

Learning to act in unfamiliar environments requires agents to infer how the world works and revise that understanding as new evidence arrives. Yet limited observations can support multiple world models that explain past interactions but predict different outcomes in unseen states. We introduce Memento 3, building on the Memento series to enable frozen LLM agents to continually learn explicit world models through external memory. The agent maintains a natural-language rulebook as persistent semantic memory, recording revisable hypotheses about environment dynamics while leaving unknown aspects underspecified. It compiles this rulebook into executable code for prediction and planning. Through a continual loop of observation, reflection, rule revision, compilation, and verification, the agent uses prediction errors to refine both the rulebook and its code. Updated code is accepted only when the LLM judges it faithful to the rulebook and cell-exact replay reproduces the observed transitions. We investigate this process as a model-based route to recursive self-improvement (RSI): the agent autonomously explores the environment, revises its world model, and uses verified updates to guide subsequent interaction and learning, while the underlying LLM remains fixed. A population extension maintains multiple world models in parallel, sharing interaction evidence and using their predictions to guide exploration. On ARC-AGI-3, the single-model agent clears every level of all 25 public games, achieves a mean Relative Human Action Efficiency (RHAE) of 100.0, and uses 44% of the human action count. In an Atari Pong case study, a learned feedback controller wins 21:0 in each of three evaluated episodes with different openings, without further LLM calls.

Chinese Translation

在不熟悉的环境中学习行动要求智能体推断世界的运作方式，并在新证据出现时修正这种理解。然而，有限的观察可能支持多个世界模型，这些模型能够解释过去的互动，但在未见状态下预测不同的结果。我们介绍了 Memento 3，基于 Memento 系列，使得冻结的 LLM（大型语言模型）智能体能够通过外部记忆持续学习明确的世界模型。该智能体维护一个自然语言规则书作为持久的语义记忆，记录关于环境动态的可修订假设，同时对未知方面保持不确定性。它将该规则书编译成可执行代码用于预测和规划。通过观察、反思、规则修订、编译和验证的持续循环，智能体利用预测误差来完善规则书及其代码。只有当 LLM 判断更新的代码忠实于规则书并且单元精确重放再现观察到的转变时，更新的代码才会被接受。我们将这一过程视为基于模型的递归自我改进（RSI）路径：智能体自主探索环境，修订其世界模型，并利用经过验证的更新来指导后续的互动和学习，同时基础的 LLM 保持不变。种群扩展在并行维持多个世界模型，分享互动证据并利用其预测来指导探索。在 ARC-AGI-3 上，单模型智能体清除了所有 25 个公共游戏的每一关，达到了 100.0 的平均相对人类行动效率（RHAE），并使用了 44% 的人类行动计数。在 Atari Pong 的案例研究中，学习到的反馈控制器在三次评估的不同开局中以 21:0 获胜，而无需进一步调用 LLM。

View on arXiv Download PDF AI Translation

cs.AI / 85 / 2610.11833

Probability-Signature Dynamics: Unpacking Modular Addition Learning Within Two-Layer Networks

概率特征动态：解读双层网络中的模加学习

Wang, Yunji, Yao, Junjie, Liu, Linyu, Lu, Pinyan, Xu, Zhi-Qin John

Abstract

Neural networks trained on modular addition tasks often develop Fourier-structured representations that support exact generalization. While prior work has identified these Fourier circuits, the mechanism by which gradient-based training selects them from the data distribution remains unclear. We address this question using probability signatures, which express leading gradient interactions through conditional statistics of the training distribution. For modular addition, these signatures are cyclic shift operators and are diagonalized by the discrete Fourier transform, yielding approximately decoupled Fourier-mode dynamics. This explains the emergence of Fourier sparsity, frequency matching, and phase alignment. The same framework resolves a puzzle under label noise: corrupted examples can show faster early loss decrease than clean examples, despite lacking a coherent generalization rule. We show that noise increases conditional label collisions, strengthening early shared-coordinate reinforcement. Finally, this method can be applied to other operators. Taking XOR as an example, we observed the predicted frequency in experiments.

Chinese Translation

在模加任务上训练的神经网络通常会发展出支持精确泛化的傅里叶结构表示。尽管先前的研究已经识别出这些傅里叶电路，但基于梯度的训练如何从数据分布中选择它们的机制仍不清楚。我们通过概率特征来解决这个问题，概率特征通过训练分布的条件统计量表达主要的梯度交互。对于模加任务，这些特征是循环移位算子，并且通过离散傅里叶变换对角化，从而产生大致解耦的傅里叶模式动态。这解释了傅里叶稀疏性、频率匹配和相位对齐的出现。同样的框架解决了标签噪声下的一个难题：尽管缺乏一致的泛化规则，受损示例的早期损失下降速度可能比干净示例更快。我们表明，噪声增加了条件标签冲突，增强了早期共享坐标的强化。最后，这种方法可以应用于其他算子。以异或（XOR）为例，我们在实验中观察到了预测的频率。

View on arXiv Download PDF AI Translation

cs.AI / 86 / 2610.11862

LEVER: Adaptive Cost-Aware Proof Search Over AND/OR Graphs

LEVER：一种自适应成本感知的证明搜索算法在 AND/OR 图上

Jain, Nihal, Yao, Shuangjie, Cicekdag, Begum, Zhang, Zhuo, Jana, Suman

Abstract

Mathematicians value proofs for more than correctness: among correct proofs, simplicity, purity and the computational cost of finding them vary widely. Yet LLM-powered theorem provers largely search for any correct proof, and improve its quality only after it is found. We propose LEVER, a proof search algorithm that makes the objective over correct proofs programmable and optimizes it during search. LEVER scores partial proofs over an AND/OR proof graph, combining realized objective values with predictions for open subgoals, so the objective guides search before a proof is complete. The same mechanism optimizes computational cost, proof length, topical impurity, and even their weighted combinations, while the Lean kernel enforces correctness. On PutnamBench in Lean 4, under matched budgets, LEVER costs 34% less than a strong single-conversation agent while raising the solve rate from 80% to 96%. On reducing topical impurity, i.e., how far a proof strays from its theorem's subject, it improves over post-hoc refactoring (42% reduction against 33%) at two-thirds of the cost and more reliably; on proof length, the metric refactoring is built for, it approaches refactoring. Varying the objective's weights traces a quality-cost trade-off curve, so the user can choose how much a better proof is worth. Overall, LEVER is a performant, cost-efficient and tunable proof search algorithm for navigating the space of correct proofs.

Chinese Translation

数学家们对证明的重视不仅仅在于其正确性：在正确的证明中，简单性、纯粹性以及找到它们的计算成本差异很大。然而，基于大型语言模型（LLM）的定理证明器主要搜索任何正确的证明，并且仅在找到证明后才提高其质量。我们提出了 LEVER，一种证明搜索算法，使得对正确证明的目标可编程，并在搜索过程中进行优化。LEVER 在 AND/OR 证明图上对部分证明进行评分，将实现的目标值与未完成子目标的预测相结合，从而在证明完成之前引导搜索。相同的机制优化计算成本、证明长度、主题纯度，甚至它们的加权组合，同时 Lean 内核确保正确性。在 Lean 4 的 PutnamBench 上，在匹配预算下，LEVER 的成本比强大的单会话代理低 34%，同时解决率从 80% 提高到 96%。在减少主题纯度方面，即证明与其定理主题的偏离程度，LEVER 在成本的三分之二下，相比于事后重构（42% 的减少对 33%）表现更佳且更可靠；在证明长度方面，LEVER 的表现接近于重构。调整目标权重描绘出质量与成本的权衡曲线，使用户能够选择更好的证明的价值。总体而言，LEVER 是一种高效、成本效益高且可调的证明搜索算法，能够在正确证明的空间中进行导航。

View on arXiv Download PDF AI Translation

cs.AI / 87 / 2610.11877

How Is Automated Research Evaluated? A Survey of Benchmarks and Evaluation Practices

自动化研究如何被评估？基准和评估实践的调查

Zhang, Liulei, Zhou, Dejing, Huang, Chuyue, Chen, Guanhua, Yao, Yutong, Chao, Lidia S., Vong, Chi Man, Wong, Derek F.

Abstract

Automated research systems support literature synthesis, ideation, experiments, writing, and peer review, but their evaluation is dispersed across tasks, benchmarks, and studies that are difficult to compare directly. We review this literature from the perspective of evaluation design and evidence, covering six targets: literature synthesis, research ideation, executable workflows, scholarly writing and communication, automatic peer review, and end-to-end research. We compare task construction, evidence sources, evaluators, and scoring procedures to explain the capabilities assessed by different designs. Our synthesis highlights three recurring lessons: output checks, process checks, and human studies provide complementary information; evaluator calibration is specific to the property being assessed; and resource budgets and attempt selection are integral to interpreting performance comparisons. We identify diagnostic evaluation designs and documented gaps in supporting evidence, and translate these comparisons into reporting and audit recommendations for specific evaluation settings. The survey helps readers navigate existing evaluations, select appropriate benchmarks, and design subsequent studies.

Chinese Translation

自动化研究系统支持文献综合、创意生成、实验、写作和同行评审，但其评估分散在各个任务、基准和难以直接比较的研究中。我们从评估设计和证据的角度回顾了相关文献，涵盖六个目标：文献综合、研究创意、可执行工作流程、学术写作与交流、自动同行评审以及端到端研究。我们比较了任务构建、证据来源、评估者和评分程序，以解释不同设计所评估的能力。我们的综合分析突出了三个反复出现的教训：输出检查、过程检查和人类研究提供互补信息；评估者的校准特定于被评估的属性；资源预算和尝试选择对于解释性能比较至关重要。我们识别了诊断评估设计和支持证据中的文档缺口，并将这些比较转化为特定评估环境的报告和审计建议。该调查帮助读者导航现有评估，选择适当的基准，并设计后续研究。

View on arXiv Download PDF AI Translation

cs.AI / 88 / 2610.11966

MindFlow: Mind Supernet Powered Thinking Flows for Research Idea Innovation

MindFlow：基于思维流的思维超网络驱动的研究创意创新

Liu, Mengdi, Chen, Wenjue, Chen, Wenyue, Yang, Cheng, Kong, Fanqi, Gao, Zhangyang, Cheng, Xiaoxue, Li, Yiheng, Yuan, Yujian, Li, Keliang, Chang, Hong, Shan, Shiguang, Wu, Chenglin

Abstract

Research idea innovation is a fundamental engine of scientific progress, yet it remains difficult to generate and evaluate in a scalable and controllable way. This challenge lies in its inherently open-ended and multi-objective nature, where ideas should balance novelty, plausibility and feasibility. While recent LLM-based approaches have made progress through carefully designed prompts or agent pipelines, they are constrained by predefined, static ideation workflows. To address this limitation, we propose MindFlow, a framework that explicitly formulates ideation as a graph-structured Flow in Mind, which is composed of modular thinking operators and modeled by a probabilistic mind supernet. Given a research topic, a controller dynamically samples thinking flows to generate candidate ideas. This open-ended problem is optimized using a tournament-based relative ranking, enabling the controller to progressively favor higher-quality thinking flows. We further introduce an evaluation protocol that jointly assesses problem finding and problem solving, going beyond title- or abstractonly judgments. Across diverse topics, MindFlow shows its superiority as an explicit, controllable and optimizable research idea innovator.

Chinese Translation

研究创意创新是科学进步的基本引擎，但在可扩展和可控的方式下生成和评估创意仍然困难。这一挑战源于其固有的开放性和多目标特性，创意需要在新颖性、可信性和可行性之间取得平衡。尽管近期基于大型语言模型（LLM）的方法通过精心设计的提示或代理管道取得了一定进展，但它们受到预定义的静态创意工作流程的限制。为了解决这一局限性，我们提出了MindFlow，一个将创意明确地形式化为思维中的图结构流的框架，该流由模块化思维操作符组成，并由概率性思维超网络建模。在给定研究主题的情况下，控制器动态地采样思维流以生成候选创意。通过基于锦标赛的相对排名来优化这一开放性问题，使控制器能够逐步偏向更高质量的思维流。我们进一步引入了一种评估协议，联合评估问题发现和问题解决，超越仅基于标题或摘要的判断。在多样化的主题中，MindFlow展现了作为明确、可控和可优化的研究创意创新者的优越性。

View on arXiv Download PDF AI Translation

cs.AI / 89 / 2610.11989

MetaOPD: Meta-Learned Token Weighting for On-Policy Distillation

MetaOPD：用于在线策略蒸馏的元学习令牌加权

Wang, Zipeng, Dong, Xinpeng, Wang, Yuefan, Lu, Pingchen, Wei, Xian, Kuang, Kun, Wu, Fei, Dai, Zhongxiang, Zhang, Min

Abstract

On-policy distillation (OPD) trains a student on its own generated responses using token-level teacher supervision. However, uniform weighting overlooks differences in token learning value, while existing weighting methods rely on predefined mappings from prediction signals to token weights. These mappings are not learned from the effectiveness of the resulting student updates, limiting their ability to adapt to evolving learning needs. In this paper, we propose MetaOPD, a bilevel optimization framework that jointly learns the student model and a lightweight token-weighting network. The inner objective updates the student through weighted OPD, while the outer objective optimizes the weighting network using validation loss on reference solutions after a virtual student update. Differentiating through this update connects weighting decisions to their effects on post-update performance, allowing the mapping from prediction signals to token weights to evolve alongside the student. Experiments on six mathematical reasoning and three out-of-domain datasets, covering two student scales and seven baselines, demonstrate the effectiveness of MetaOPD, with Avg@8/Pass@8 gains over OPD of 1.99/5.97 percentage points for the 0.6B student and 2.25/6.41 points for the 1.7B student.

Chinese Translation

在线策略蒸馏（OPD）通过令牌级教师监督训练学生模型，使其基于自身生成的响应进行学习。然而，统一加权忽视了令牌学习价值的差异，而现有的加权方法依赖于从预测信号到令牌权重的预定义映射。这些映射并不是基于学生更新效果学习的，限制了它们适应不断变化的学习需求的能力。本文提出了MetaOPD，一个双层优化框架，联合学习学生模型和轻量级令牌加权网络。内部目标通过加权OPD更新学生，而外部目标则在虚拟学生更新后使用参考解的验证损失来优化加权网络。通过此更新的微分连接了加权决策与其对更新后性能的影响，使得从预测信号到令牌权重的映射能够与学生模型共同演变。在六个数学推理和三个领域外数据集上的实验，涵盖了两种学生规模和七个基线，证明了MetaOPD的有效性，对于0.6B学生模型，Avg@8/Pass@8相较于OPD提升了1.99/5.97个百分点，而对于1.7B学生模型则提升了2.25/6.41个百分点。

View on arXiv Download PDF AI Translation

cs.AI / 90 / 2610.12003

An Interpretable Approach to PDE Solution Discovery via Structural Experience Distillation

通过结构经验蒸馏的可解释 PDE 解发现方法

Gong, Yunpeng, Wu, Huolong, Yang, Can, Jiang, Min

Abstract

PDE solution discovery aims to identify explicit symbolic expressions for unknown physical fields from observations under known physical constraints. Existing methods, however, collapse data fidelity and physical consistency into a single terminal score used as the sole feedback signal, providing little information about which subexpressions are responsible for a candidate's final performance. This opaque terminal feedback severely limits the interpretability of the search process itself, offering no insight into why a candidate succeeds or fails. Consequently, reusable structures in otherwise suboptimal candidates are often discarded, whereas incidental syntax along successful search trajectories may be repeatedly reinforced. We propose SED-MCTS, a Monte Carlo tree search approach that distills structural experience from evaluated expressions and reuses it to guide subsequent symbolic solution search. Through counterfactual subtree interventions, SED-MCTS estimates local structural contributions, routes reliable evidence to the responsible construction edges, and preserves useful components in a refined structural archive. The approach naturally extends to coupled multiphysics systems. Across a diverse suite of PDE benchmarks, SED-MCTS achieves strong performance under a fixed evaluation budget and improves search efficiency and robustness under noisy or scarce observations.

Chinese Translation

PDE 解发现旨在从已知物理约束下的观测中识别未知物理场的显式符号表达式。然而，现有方法将数据保真度和物理一致性合并为一个单一的终端评分，作为唯一的反馈信号，这对于候选解最终表现的子表达式缺乏足够的信息。这种不透明的终端反馈严重限制了搜索过程的可解释性，无法提供候选解成功或失败的原因。因此，在其他次优候选解中可重复使用的结构往往被丢弃，而在成功搜索轨迹中的偶然语法可能会被反复强化。我们提出了 SED-MCTS，一种蒙特卡洛树搜索方法，通过评估表达式提炼结构经验，并将其重用于指导后续的符号解搜索。通过反事实子树干预，SED-MCTS 估计局部结构贡献，将可靠证据引导至负责构造的边缘，并在精炼的结构档案中保留有用组件。该方法自然扩展到耦合多物理系统。在一系列多样的 PDE 基准测试中，SED-MCTS 在固定评估预算下实现了强大的性能，并在噪声或稀缺观测条件下提高了搜索效率和鲁棒性。

View on arXiv Download PDF AI Translation

cs.AI / 91 / 2610.12008

Complexity of Grounded Semantics and Preferred Semantics in Finitary Argumentation Frameworks

有限论证框架中基础语义和优选语义的复杂性

Xu, Jinfan, Luo, Jieting

Abstract

Abstract argumentation frameworks (AFs) introduced by Dung provide a formal foundation for non-monotonic reasoning in artificial intelligence. While decision problems for general infinite AFs typically reside at high levels of the analytical hierarchy ($\Sigma_1^1$ or $\Pi_1^1$), restricting the framework to be computably finitary reduces some of the complexity to the arithmetical hierarchy. In this paper, we present a complexity mapping of grounded and preferred semantics in computably finitary AFs across standard decision problems: credulous acceptance ($\Cred$), skeptical acceptance ($\Skep$), extension existence ($\Ex$), uniqueness ($\Uni$), and non-empty existence ($\NE$). For grounded semantics, credulous and skeptical acceptance are already known to be $\Sigma_1^0$-complete. We show that non-empty existence is also $\Sigma_1^0$-complete, whereas existence and uniqueness are trivial. These classifications are understood within the domain of valid computably finitary representations. For preferred semantics, using a computably finitely branching computation tree, $\Cred_{\pref}$ is shown to be in $\Pi_1^0$-c and $\NE_{\pref}$ is $\Sigma_2^0$-c. However, it is insufficient to reduce universal quantification and global uniqueness, leaving $\Skep_{\pref}$ in $\Pi_1^1$ and $\UniPref$ in $\Sigma_2^1$-c. Our results show the precise boundary where finitarity succeeds to bring reasoning down to the arithmetical hierarchy and where second-order quantification forces problems back into the analytical hierarchy.

Chinese Translation

邓（Dung）提出的抽象论证框架（AFs）为人工智能中的非单调推理提供了形式基础。尽管一般无限AFs的决策问题通常位于分析层次的高层（$ ext{Σ}_1^1$或$ ext{Π}_1^1$），但将框架限制为可计算的有限性则将部分复杂性降低到算术层次。在本文中，我们展示了可计算有限AFs中基础语义和优选语义在标准决策问题上的复杂性映射：可信接受（$ ext{Cred}$）、怀疑接受（$ ext{Skep}$）、扩展存在性（$ ext{Ex}$）、唯一性（$ ext{Uni}$）和非空存在性（$ ext{NE}$）。对于基础语义，可信和怀疑接受已知是$ ext{Σ}_1^0$-完全的。我们证明非空存在性也是$ ext{Σ}_1^0$-完全的，而存在性和唯一性则是平凡的。这些分类在有效的可计算有限表示的领域内被理解。对于优选语义，使用可计算有限分支的计算树，$ ext{Cred}_{ ext{pref}}$被证明在$ ext{Π}_1^0$-c中，而$ ext{NE}_{ ext{pref}}$是$ ext{Σ}_2^0$-c。然而，减少全称量化和全局唯一性是不够的，这使得$ ext{Skep}_{ ext{pref}}$处于$ ext{Π}_1^1$中，而$ ext{UniPref}$则在$ ext{Σ}_2^1$-c中。我们的结果显示了有限性成功将推理降低到算术层次的精确边界，以及二阶量化迫使问题回到分析层次的情况。

View on arXiv Download PDF AI Translation

cs.AI / 92 / 2610.12010

PulseBound: Future-Beat State Forecasting Under an Explicit Information Boundary

PulseBound：在明确的信息边界下的未来节拍状态预测

Xu, Chenyang, Xie, Donglin, Xiang, Xi, Li, Xiaoyu, Lu, Yufan, Gao, Jiqiun, Zhao, Yi, Li, Xin-Yi, Zhu, Guangpu, Wang, Zijian, Yang, Xiwen, Wang, Dezhen, Chen, Lin, Hong, Shenda, Li, Leilei

Abstract

Predictive representation learning from photoplethysmography (PPG) can violate causal information access even with causal attention, as normalization, nonlocal transforms, or companion views may depend on withheld samples. We introduce PulseBound, a PPG representation learner combining physiologically structured future-beat prediction with an explicit stored-window information boundary. A content-independent cutoff separates the visible prefix from the prediction target. Prefix-only normalization, suffix replacement before derived-view construction, and aligned masking ensure that encoder inputs depend only on the visible prefix and cutoff. This yields stored-suffix invariance: with fixed model state, randomness, prefix, and cutoff, changing the stored suffix cannot change the forecast context. A shared horizon-conditioned head predicts nine rhythm and morphology descriptors for up to four extractor-valid future beats, using elementwise validity masks; optional ECG-derived pulse-arrival-time supervision is restricted to training. On MIMIC and VitalDB groups held out from PulseBound backbone pretraining, PulseBound reduces nine-state transformed-space MAE relative to last-visible-beat persistence by 28.06% and 22.22%, respectively, with gains in MAE, MAE-Skill, and Spearman correlation across all 40 source-cutoff-horizon cells. In a separate comparison of seven models on 13 downstream tasks, PulseBound achieves the best mean on nine frozen linear-probe and seven full-fine-tuning tasks. Stored-suffix interventions cause zero recorded changes in forecast contexts or predictions, with zero suffix-input gradients at audited precision under the stored-window interface. These findings separate three testable aspects of predictive physiological representation learning: information access, supervised future structure, and transfer.

Chinese Translation

从光电容积描记法（PPG）中进行的预测表示学习即使在因果注意力下也可能违反因果信息访问，因为归一化、非局部变换或伴随视图可能依赖于被 withheld 的样本。我们提出了 PulseBound，这是一种结合生理结构的未来节拍预测与明确存储窗口信息边界的 PPG 表示学习器。一个与内容无关的截止点将可见前缀与预测目标分开。仅前缀的归一化、在派生视图构建之前的后缀替换，以及对齐的掩蔽确保编码器输入仅依赖于可见前缀和截止点。这产生了存储后缀不变性：在固定模型状态、随机性、前缀和截止点的情况下，改变存储后缀不能改变预测上下文。一个共享的地平线条件头预测最多四个提取器有效的未来节拍的九个节奏和形态描述符，使用逐元素有效性掩蔽；可选的基于 ECG 的脉冲到达时间监督仅限于训练。在从 PulseBound 主干预训练中保留的 MIMIC 和 VitalDB 组上，PulseBound 相较于最后可见节拍的持久性，分别减少了九状态变换空间的平均绝对误差（MAE）28.06% 和 22.22%，在所有 40 个源截止地平线单元中，MAE、MAE-Skill 和斯皮尔曼相关性均有提升。在对七个模型在 13 个下游任务的单独比较中，PulseBound 在九个冻结线性探针和七个完全微调任务上实现了最佳均值。存储后缀干预在存储窗口接口下的审计精度下记录的预测上下文或预测变化为零，且后缀输入梯度为零。这些发现将预测生理表示学习的三个可测试方面分开：信息访问、监督的未来结构和迁移。

View on arXiv Download PDF AI Translation

cs.AI / 93 / 2610.12023

InterviewPlayground: A Simulation Environment for Evaluating AI Interviewers

InterviewPlayground：评估人工智能面试官的模拟环境

Ivey, Jonathan, Liang, Aimee, Wang, Arthur Y. S., Mandell, Madeline, Xiao, Ziang, Field, Anjalie

Abstract

Increasingly, AI interviewers are being developed to elicit open-ended responses in applications like market research, public polling, preference elicitation, and social science research. However, evaluating AI interviewers is challenging because they function in extended, multi-turn interactions where they must adapt to participant behaviors. To address this need, we develop InterviewPlayground, a simulation environment for evaluating AI interviewers using simulated study participants whose behaviors are grounded in social theory. Simulated studies in InterviewPlayground produce an InterviewReportCard, which assesses the performance of AI interviewers using a suite of validated measures. To test whether our simulation-based evaluations predict performance with human participants, we conduct 15 real qualitative studies with five AI interviewers, three interview topics, and 450 human participants and compare them to simulated studies in InterviewPlayground. We find that AI interviewer performance in InterviewPlayground predicts performance in human studies with an average Pearson correlation of 0.86 across 12 measures, and the simulated interactions from InterviewPlayground reproduce key findings from behavioral analysis of AI interviewers in the human studies. Together, these findings support the validity of InterviewPlayground in assessing AI interviewer performance and examining potential failure modes. Our work contributes a simulation environment for AI interviewers supported with empirical validation, and more broadly, a roadmap for future work to develop validated, simulation-based evaluations of conversational AI systems.

Chinese Translation

随着人工智能面试官的不断发展，它们被广泛应用于市场研究、公众投票、偏好引导和社会科学研究等领域，以引导开放式回应。然而，评估人工智能面试官的效果具有挑战性，因为它们在延续的多轮互动中运作，必须适应参与者的行为。为了解决这一需求，我们开发了InterviewPlayground，这是一个用于评估人工智能面试官的模拟环境，利用基于社会理论的模拟研究参与者的行为。在InterviewPlayground中进行的模拟研究生成了InterviewReportCard，该报告卡使用一系列经过验证的指标评估人工智能面试官的表现。为了测试我们的基于模拟的评估是否能够预测与人类参与者的表现，我们进行了15项真实的定性研究，涉及五个人工智能面试官、三个面试主题和450名人类参与者，并将其与InterviewPlayground中的模拟研究进行了比较。我们发现，InterviewPlayground中人工智能面试官的表现能够预测人类研究中的表现，12项指标的平均Pearson相关系数为0.86，并且InterviewPlayground中的模拟互动再现了人类研究中人工智能面试官行为分析的关键发现。这些发现共同支持了InterviewPlayground在评估人工智能面试官表现和检查潜在失败模式方面的有效性。我们的工作为人工智能面试官提供了一个经过实证验证的模拟环境，并更广泛地为未来开发经过验证的基于模拟的对话人工智能系统评估提供了路线图。

View on arXiv Download PDF AI Translation

cs.AI / 94 / 2610.12056

Structure Tax: How Structured Output affects LLMs Performance

结构税：结构化输出如何影响大型语言模型的性能

Kumar, Vineet, Kanishka, Mandora, Bhuvanesh

Abstract

Deploying large language models in production often requires constraining outputs to structured formats such as JSON or XML, and prior work treats the resulting accuracy loss as an inherent `structure tax'. We re-examine this claim by evaluating a battery of models, datasets and schemas, measuring task accuracy, confidence calibration, and hidden-state geometry. The tax turns out to depend on schema design rather than on structure per se: reasoning-first field ordering matches or exceeds free-form accuracy, while answer-first ordering causes steep drops, particularly in smaller models. Format sensitivity scales inversely with a task's own structural constraints, and schemas that preserve reasoning order also improve calibration with CKA showing greater separability between correct and incorrect representations in middle transformer layers. Our findings indicate that properly designed structured formats can match or exceed free-form performance, reframing the critical question from `whether to structure' to `how to structure' for optimal reasoning preservation.

Chinese Translation

在生产中部署大型语言模型通常需要将输出限制为结构化格式，如 JSON 或 XML，之前的研究将由此产生的准确性损失视为固有的“结构税”。我们通过评估一系列模型、数据集和模式，重新审视这一说法，测量任务准确性、置信度校准和隐藏状态几何。结果表明，这一税收依赖于模式设计，而非结构本身：以推理优先的字段排序与自由格式的准确性相匹配或超越，而以答案优先的排序则导致显著下降，尤其是在较小的模型中。格式敏感性与任务自身的结构约束成反比，保持推理顺序的模式也改善了校准，CKA 显示中间变换层中正确与错误表示之间的可分离性更强。我们的研究结果表明，合理设计的结构化格式可以匹配或超越自由格式的性能，将关键问题从“是否结构化”重新框定为“如何结构化”，以实现最佳的推理保留。

View on arXiv Download PDF AI Translation

cs.AI / 95 / 2610.12061

When Should Agents Think? Adaptive Reasoning via Cross-Turn Estimation

代理何时应进行思考？通过跨回合估计的自适应推理

Cheng, Yiruo, Huang, Shen, Song, Xiaoshuai, Tan, Jiejun, Dong, Guanting, Xie, Pengjun, Wen, Ji-Rong, Dou, Zhicheng

Abstract

Large language model (LLM)-based agents have demonstrated strong capabilities on complex tasks. They typically perform reasoning before each action throughout an interaction trajectory. However, reasoning may not be necessary at every turn, as reasoning produced earlier can continue to support subsequent actions. A key challenge is therefore to determine when existing reasoning remains sufficient and when a new reasoning step is needed, without relying on costly generation-based verification. We find that decreases in the likelihood of subsequent reference actions after removing additional reasoning closely track whether those actions remain recoverable given earlier reasoning, providing an effective and lightweight signal for estimating cross-turn action support. Based on this observation, we propose Reasoning Adaptation through Cross-Turn Estimation (RACE), a training approach for adaptive agent reasoning. RACE introduces a Likelihood-Guided Progressive Reasoning Cover Detection (LoGiC) procedure that progressively identifies reasoning turns whose removal has limited impact on the current and subsequent reference actions. The resulting removal signals are incorporated into both supervised fine-tuning and agentic reinforcement learning, enabling the policy to learn when to reason and when to act directly. Extensive experiments on four representative agent benchmarks show that RACE substantially reduces reasoning cost while maintaining or improving task performance.

Chinese Translation

基于大型语言模型（LLM）的代理在复杂任务上表现出了强大的能力。它们通常在交互轨迹中的每个动作之前进行推理。然而，并非在每个回合中都需要推理，因为早期产生的推理可以继续支持后续的动作。因此，一个关键挑战是确定现有推理何时仍然足够，以及何时需要新的推理步骤，而不依赖于成本高昂的生成验证。我们发现，在移除额外推理后，后续参考动作的可能性下降与这些动作在早期推理下是否仍然可恢复密切相关，从而提供了一种有效且轻量的信号来估计跨回合动作支持。基于这一观察，我们提出了通过跨回合估计进行推理适应（Reasoning Adaptation through Cross-Turn Estimation, RACE），这是一种用于自适应代理推理的训练方法。RACE引入了一种可能性引导的渐进推理覆盖检测（Likelihood-Guided Progressive Reasoning Cover Detection, LoGiC）程序，逐步识别移除对当前和后续参考动作影响有限的推理回合。所得到的移除信号被纳入监督微调和代理强化学习中，使得策略能够学习何时进行推理以及何时直接行动。在四个代表性代理基准上的大量实验表明，RACE显著降低了推理成本，同时保持或提高了任务性能。

View on arXiv Download PDF AI Translation

cs.AI / 96 / 2610.12085

Is Memorization Context-Sensitive? Prefix-Based Extraction Beyond Isolated Prefixes

记忆是否与上下文相关？基于前缀的提取超越孤立前缀

Satvaty, Ali, Sharafi, Narjes, Qi, Jirui, Verberne, Suzan, Turkmen, Fatih

Abstract

Large language models (LLMs) can expose memorized training sequences under prefix-based extraction: given a prefix from a training example, the model may assign high probability to the original continuation. In deployed systems, however, prefixes are rarely evaluated in isolation. They often appear together with instructions, retrieved documents, or other task-specific context, as in retrieval-augmented generation (RAG). This motivates examining whether contextual conditioning mitigates memorization or merely changes the set of memorized samples that become extractable. We investigate this issue through paired item-level measurements of probabilistic suffix extraction. For each prefix-suffix pair, we score the target suffix under an empty prompt and under retrieved contexts of varying relevance, across three open-weight instruction-tuned models. We find that context does not simply erase memorization. Instead, extractable memorization consists of a context-robust core and a context-sensitive boundary. Many samples that are extractable without context remain extractable under the retrieved context, especially as the prefix length increases. At the same time, context mainly affects marginal samples near the extraction threshold: it suppresses some exposures, but also enables new ones that are missed by prefix-only evaluation. These findings qualify the view that RAG reduces memorization risk. Context can lower aggregate extraction by suppressing boundary cases, yet robustly extractable samples persist, and context-enabled extractability remains security-relevant.

Chinese Translation

大型语言模型（LLMs）在基于前缀的提取中可以暴露出记忆的训练序列：给定来自训练示例的前缀，模型可能会对原始延续分配高概率。然而，在实际应用系统中，前缀很少单独评估。它们通常与指令、检索的文档或其他特定任务的上下文一起出现，如在检索增强生成（RAG）中。这促使我们研究上下文条件是否减轻记忆，或仅仅改变可提取的记忆样本集。我们通过配对的项目级概率后缀提取测量来调查这个问题。对于每个前缀-后缀对，我们在空提示下和在不同相关性的检索上下文下对目标后缀进行评分，涵盖三个开放权重的指令调优模型。我们发现，上下文并不会简单地抹去记忆。相反，可提取的记忆由一个上下文稳健的核心和一个上下文敏感的边界组成。许多在没有上下文的情况下可提取的样本在检索上下文下仍然可提取，尤其是在前缀长度增加时。同时，上下文主要影响接近提取阈值的边际样本：它抑制了一些暴露，但也启用了在仅前缀评估中被遗漏的新暴露。这些发现修正了RAG降低记忆风险的观点。上下文可以通过抑制边界案例来降低整体提取，但稳健可提取的样本仍然存在，并且上下文启用的可提取性仍然与安全性相关。

View on arXiv Download PDF AI Translation

cs.AI / 97 / 2610.12086

EvoAlloc: A Self-Evolving Resource Allocation Agent for Efficient Program Evolution

EvoAlloc：一种自我进化的资源分配代理，用于高效程序演化

Dai, Yanning, Wang, Yuhui, Li, Nanbo, Wang, Wenyi, Schmidhuber, Jürgen

Abstract

LLM-based program evolution relies on evaluation feedback to guide the iterative search for high-performing programs. However, evaluation is often computationally expensive, making it essential to allocate limited resources to candidates that can most effectively advance the search. Existing LLM-based methods typically rely on fixed allocation strategies throughout the search, potentially wasting resources on low-value candidates while overlooking promising ones. We propose EvoAlloc, a self-evolving resource-allocation agent that learns from search experience to revise its strategy for allocating computational resources across candidates. EvoAlloc periodically consolidates prior search and allocation outcomes into reusable experience, which informs subsequent strategy revisions. It further uses a counterfactual exploration mechanism to occasionally evaluate candidates denied resources by the allocator, revealing their outcomes to enrich its experience for future strategy updates. Across coding and agent-harness optimization benchmarks, EvoAlloc requires 59-82% fewer full evaluations and 61-89% fewer total LLM tokens to reach baseline-level performance. Moreover, under the same full-evaluation budget, EvoAlloc achieves 8.7-12.0% higher final performance.

Chinese Translation

基于大语言模型（LLM）的程序演化依赖于评估反馈来指导高性能程序的迭代搜索。然而，评估通常计算成本高，因此必须将有限的资源分配给那些能够最有效推动搜索的候选者。现有的基于LLM的方法通常在整个搜索过程中依赖固定的分配策略，这可能导致在低价值候选者上浪费资源，同时忽视有前景的候选者。我们提出了EvoAlloc，一种自我进化的资源分配代理，它通过学习搜索经验来修订其在候选者之间分配计算资源的策略。EvoAlloc定期将先前的搜索和分配结果整合为可重用的经验，从而为后续的策略修订提供依据。它还利用反事实探索机制，偶尔评估被分配器拒绝资源的候选者，揭示其结果以丰富其未来策略更新的经验。在编码和代理优化基准测试中，EvoAlloc需要减少59-82%的完整评估次数和61-89%的总LLM令牌数，以达到基线水平的性能。此外，在相同的完整评估预算下，EvoAlloc实现了8.7-12.0%的更高最终性能。

View on arXiv Download PDF AI Translation

cs.AI / 98 / 2610.12090

Recompose and Refine Latent Reasoning Flows for Vision-Language-Action Models

重组与优化潜在推理流以用于视觉-语言-动作模型

Shi, Hongyu, Zhao, Sen, Zhang, Zuyu, Shen, Lifeng, Zou, Ding, He, Xinyu, Zhang, Xu, Zhang, Qinghua

Abstract

Latent reasoning enables vision-language-action (VLA) models to transform multimodal observations into task-relevant internal states before generating continuous robot actions. While existing methods learn to generate or refine such states for each policy query, they discard successful reasoning after execution and therefore reconstruct similar computation from scratch. We present Reasoning and Flow Memory (FLOWMEM), a unified VLA model that turns successful latent computation into reusable reasoning experience. Rather than appending a fixed retrieved context, FLOWMEM dynamically retrieves and recomposes compatible latent fragments as the embodied context evolves, forming a reasoning route that follows the temporal structure and progress of successful computation. The route is then refined using current visual and proprioceptive evidence before it conditions action generation. Experiments on RoboMME and LIBERO-Plus show that FLOWMEM attains 48.0% and 77.3% success, outperforming memory-free policies by 1.7 and 4.1 percentage points, respectively. These results demonstrate the value of reusing successful latent computation for closed-loop VLA control.

Chinese Translation

潜在推理使视觉-语言-动作（VLA）模型能够在生成连续机器人动作之前，将多模态观察转化为与任务相关的内部状态。尽管现有方法学习为每个策略查询生成或优化这些状态，但它们在执行后丢弃成功的推理，因此从头开始重建类似的计算。我们提出了推理与流记忆（Reasoning and Flow Memory，FLOWMEM），这是一个统一的VLA模型，将成功的潜在计算转化为可重用的推理经验。FLOWMEM并不是简单地附加固定的检索上下文，而是随着具身上下文的演变，动态检索和重组兼容的潜在片段，形成一个遵循成功计算的时间结构和进展的推理路径。然后，使用当前的视觉和本体感觉证据对该路径进行优化，以便为动作生成提供条件。在RoboMME和LIBERO-Plus上的实验表明，FLOWMEM分别达到了48.0%和77.3%的成功率，分别比无记忆策略提高了1.7和4.1个百分点。这些结果展示了重用成功潜在计算在闭环VLA控制中的价值。

View on arXiv Download PDF AI Translation

cs.AI / 99 / 2610.12114

Universal Textual Teaching for LLMs

大型语言模型的通用文本教学

Lu, Zhanyi, Wang, Huan

Abstract

Knowledge distillation (KD) transfers knowledge from stronger Teacher models to weaker Student models, but most methods require training the Student parameters, thereby binding the distilled knowledge to a specific architecture and checkpoint. This implicit representation is difficult to interpret or reuse across models and limits KD for API-only or costly-to-train models. This paper studies knowledge transfer for large language models (LLMs). We introduce Universal Textual Teaching (UTT), a parameter-update-free framework that distills observed Teacher-Student knowledge gaps into a textual, interpretable, and reusable natural-language artifact called Primer. Specifically, UTT first identifies representative gap cases through paired evaluations, and iteratively updates the Primer via multi-role interactions: the Student attempts each task, the Prompter turns evaluation feedback into a teaching instruction, the Teacher provides a targeted demonstration, and the Synthesizer consolidates validated lessons. Empirically, on the challenging math (Omni-MATH-2) and code generation (KernelBench) tasks, extensive results confirm the effectiveness of the method: UTT remarkably raises the Student's accuracy from 9.4% to 48.6% and Fast1 accuracy from 9% to 35% on KernelBench, while increasing mathematical reasoning accuracy from 27.6% to 51.7%. UTT also performs better than representative prompt engineering and parameter-based KD methods. Of note, UTT is shown to be generalizable across different Teachers and Students: a Primer synthesized for one Teacher-Student pair can generalize to other Students that do not participate in the synthesis.

Chinese Translation

知识蒸馏（Knowledge Distillation, KD）将知识从更强的教师模型转移到较弱的学生模型，但大多数方法需要训练学生参数，从而将蒸馏的知识绑定到特定的架构和检查点。这种隐式表示难以在模型之间进行解释或重用，并限制了对仅提供API或训练成本高的模型的KD应用。本文研究了针对大型语言模型（Large Language Models, LLMs）的知识转移。我们提出了通用文本教学（Universal Textual Teaching, UTT），这是一种无需参数更新的框架，将观察到的教师-学生知识差距蒸馏为一种文本、可解释且可重用的自然语言工件，称为Primer。具体而言，UTT首先通过配对评估识别代表性的差距案例，并通过多角色互动迭代更新Primer：学生尝试每个任务，提示者将评估反馈转化为教学指令，教师提供针对性的示范，而合成者整合经过验证的教训。在具有挑战性的数学（Omni-MATH-2）和代码生成（KernelBench）任务上，实证结果确认了该方法的有效性：UTT显著提高了学生的准确率，从9.4%提升至48.6%，在KernelBench上的Fast1准确率从9%提升至35%，同时数学推理准确率从27.6%提升至51.7%。UTT的表现也优于代表性的提示工程和基于参数的KD方法。值得注意的是，UTT在不同的教师和学生之间显示出良好的泛化能力：为一个教师-学生对合成的Primer可以泛化到其他未参与合成的学生。

View on arXiv Download PDF AI Translation

cs.AI / 100 / 2610.12124

Use and Disuse: Intent-Structured Experience Consolidation for Memory and Learning in LLM Agents

使用与不使用：基于意图结构的经验巩固在大型语言模型代理中的记忆与学习

Zeng, Xiangyi, Liu, Baihang, Wang, Xutong, Jin, Ze, Li, Yunpeng, Liu, Qixu

Abstract

The evolution of Large Language Model agents from single-task execution to long-term autonomous operation highlights the critical challenge of transforming continuous experiences into reusable knowledge. To address this, we propose Hippocam, a hierarchical memory and continual learning architecture. Hippocam draws inspiration from two characteristics of human memory: cognitive processes selectively maintain information relevant to current goals, while long-term memories form gradually through repeated consolidation. Accordingly, Hippocam structures an agent's ongoing work as nested intents. The active context remains centered on the current intent, while completed intents are consolidated into the task-relevant outcomes and state needed for subsequent work, rather than carrying forward their full working details. Concurrently, a recursive prefix consolidation mechanism repeatedly consolidates earlier history, causing long-unused experiences to become increasingly abstract. Original interactions are preserved, allowing the agent to progressively recover finer-grained details through the hierarchy and stop once sufficient information is available. Crucially, when past experiences are recalled and reintegrated into active work, they undergo subsequent consolidation alongside new experiences, thereby being reinforced, supplemented, and updated. Through this memory dynamic of use and disuse, Hippocam connects working context, long-term memory, knowledge accumulation, and skill learning within a single continuously evolving experiential process. This enables agents to learn and evolve capabilities through their own experiences without parameter updates.

Chinese Translation

大型语言模型代理从单一任务执行到长期自主操作的演变突显了将连续经验转化为可重用知识的关键挑战。为了解决这一问题，我们提出了Hippocam，一种层次化记忆与持续学习架构。Hippocam受到人类记忆两个特征的启发：认知过程选择性地保持与当前目标相关的信息，而长期记忆则通过重复巩固逐渐形成。因此，Hippocam将代理的持续工作结构化为嵌套意图。活动上下文始终集中在当前意图上，而已完成的意图则被巩固为与任务相关的结果和后续工作所需的状态，而不是携带其完整的工作细节。同时，一个递归前缀巩固机制反复巩固早期历史，使得长时间未使用的经验变得越来越抽象。原始交互被保留，允许代理通过层次结构逐步恢复更细致的细节，并在获得足够信息后停止。至关重要的是，当过去的经验被回忆并重新整合到当前工作中时，它们会与新经验一起经历后续巩固，从而得到强化、补充和更新。通过这种使用与不使用的记忆动态，Hippocam将工作上下文、长期记忆、知识积累和技能学习连接在一个持续演变的经验过程中。这使得代理能够通过自身经验学习和发展能力，而无需参数更新。

View on arXiv Download PDF AI Translation

cs.AI / 101 / 2610.12129

An Investigation of Model Coherence: Narrow Finetunes Contradict Themselves Under Resampling

模型一致性研究：狭义微调在重采样下自相矛盾

Graham, Robert, Barsheshat, Yariv, Blandfort, Phil, Alouache, Sabri

Abstract

A large body of research measures model coherence based on output variance without adequately considering competing causes. We identify two such causes, ambiguity and indifference, and we introduce a set of 175 questions where contradicting answers cannot easily be explained by either. We then measure incoherence in terms of contradictions when resampling answers to the same question. In contrast to other methods our metric has high specificity, and only ranks models as incoherent when the issues are glaring. Even so, we find narrow finetunes score poorly. Inspecting inconsistencies flagged by our method, we find that model organisms from the literature display severe issues such as identity conflation, introspection failures and rationalizations. These findings suggest that the pathologies induced by narrow finetuning may limit what these models can tell us about coherent misaligned behaviour.

Chinese Translation

大量研究基于输出方差来衡量模型一致性，但未充分考虑竞争性原因。我们识别出两种此类原因：模糊性和漠不关心，并引入了一套175个问题，其中矛盾的答案无法通过任何一种原因轻易解释。然后，我们通过对同一问题的答案进行重采样来衡量不一致性，具体体现在矛盾上。与其他方法相比，我们的度量具有较高的特异性，仅在问题明显时将模型评为不一致。即便如此，我们发现狭义微调的得分较低。通过检查我们方法标记的不一致性，我们发现文献中的模型生物体存在严重问题，例如身份混淆、自省失败和合理化。这些发现表明，狭义微调引发的病态可能限制了这些模型对一致性不匹配行为的解释能力。

View on arXiv Download PDF AI Translation

cs.AI / 102 / 2610.12134

OA-MAP: Evidence-Grounded Multi-Agent Multimodal Framework for Interpretable Knee Osteoarthritis Progression

OA-MAP：基于证据的多智能体多模态框架用于可解释的膝关节骨关节炎进展

Chen, Sixu, Yang, Mingrui, Guan, Qiang, Li, Xiaojuan

Abstract

Knee osteoarthritis (KOA) progression prediction can support patient monitoring, requiring the integration of multimodal data and multidomain expertise. Moreover, isolated risk estimates provide limited insight underlying a prediction. To automate the progression assessment workflow and reduce manual effort while providing interpretable findings and supporting evidence, we present OA-MAP, an autonomous multi-agent framework for evidence-grounded assessment of structural and pain progression in KOA. The system incorporates modality-specific agents including MRI, X-ray, and clinical agents, together with a coordinator agent. This framework can autonomously recruit specialist agents, select tools for prediction and analysis, and retrieve literature as external evidence based on user request and available patient information. An uncertainty-informed human-in-the-loop mechanism enables clinicians to review and correct intermediate findings, triggering recomputation of affected results. We evaluate the prediction models using 600 participants from the FNIH Osteoarthritis Biomarkers Consortium cohort. On the test set of 100 participants, the fusion models achieve AUROCs of 0.80 for structural progression and 0.68 for pain progression. A case study illustrates how OA-MAP combines risk estimates with intermediate findings, cross-modal conflicts, literature support, and uncertainty indicators to support interactive review.

Chinese Translation

膝关节骨关节炎（KOA）进展预测可以支持患者监测，这需要整合多模态数据和多领域专业知识。此外，孤立的风险估计对预测的深层次理解提供了有限的洞察。为了自动化进展评估工作流程，减少人工工作量，同时提供可解释的发现和支持证据，我们提出了OA-MAP，一个用于KOA结构和疼痛进展的基于证据的自主多智能体框架。该系统包含特定模态的智能体，包括MRI、X光和临床智能体，以及一个协调智能体。该框架能够自主招募专业智能体，选择预测和分析工具，并根据用户请求和可用的患者信息检索文献作为外部证据。一个基于不确定性的“人机协作”机制使临床医生能够审查和修正中间发现，从而触发受影响结果的重新计算。我们使用来自FNIH骨关节炎生物标志物联盟队列的600名参与者评估预测模型。在100名参与者的测试集中，融合模型在结构进展方面的AUROC达到0.80，在疼痛进展方面的AUROC达到0.68。一个案例研究展示了OA-MAP如何结合风险估计、中间发现、跨模态冲突、文献支持和不确定性指标，以支持互动审查。

View on arXiv Download PDF AI Translation

cs.AI / 103 / 2610.12135

Q-Shaped Options for Hierarchical Reinforcement Learning

层次强化学习中的Q形选项

Wibault, Clarisse, Gorceix, Antoine, Villares, Antonio Léon, Zakharov, Alexey, Chatzaroulas, Evangelos, Matthews, Michael, Pignatelli, Eduardo, Foerster, Jakob

Abstract

Learning to tackle long-horizon, goal-conditioned tasks requires an agent to reason over extended timescales and act across a broad range of states. In principle, Hierarchical Reinforcement Learning (HRL) addresses both challenges through the interaction between action (temporal) and state (spatial) abstraction. First, using an action abstraction to represent temporally extended behaviour as options reduces the effective decision horizon. Second, enabling different state abstractions at each level of the decision process permits greater data aggregation for learning. However, realising these two benefits of a hierarchical policy depends on learning an appropriate action abstraction. Current HRL algorithms fail in one of two ways. Some discard distinctions between options needed for optimal control, undermining hierarchy altogether. Others retain unnecessary distinctions, preserving horizon reduction, but forfeiting coarser state abstraction. In this work, we characterise three desiderata for an action abstraction. We introduce Q-Shaped Options (QSO) to address all three. QSO builds on an architecture with distinct state-value functions, Q functions and policies at each level of the hierarchy. It learns the action abstraction between consecutive levels as a shared encoder shaped by their respective Q functions. The low-level Q function uses the option as a goal, encouraging the abstraction to retain distinctions necessary for optimal control. The high-level Q function uses it as an action, encouraging unnecessary distinctions to be discarded. Across offline goal-conditioned locomotion and manipulation environments, QSO learns semantically meaningful option spaces and outperforms baselines, achieving non-zero performance in tasks where all other evaluated algorithms fail.

Chinese Translation

学习应对长期、目标条件任务需要代理在扩展的时间尺度上进行推理，并在广泛的状态范围内采取行动。从理论上讲，层次强化学习（HRL）通过动作（时间）和状态（空间）抽象之间的互动来解决这两个挑战。首先，使用动作抽象将时间上扩展的行为表示为选项，从而减少有效决策的时间范围。其次，在决策过程的每个层级启用不同的状态抽象，允许更大的数据聚合以进行学习。然而，实现层次策略的这两个好处依赖于学习适当的动作抽象。目前的HRL算法在两种方式中失败。一些算法忽视了优化控制所需的选项之间的区别，从而完全削弱了层次结构。另一些则保留了不必要的区别，保持了时间范围的缩减，但放弃了粗略的状态抽象。在本研究中，我们描述了动作抽象的三个期望特性。我们引入Q形选项（QSO）以满足这三个特性。QSO基于一种架构，在层次的每个层级都有不同的状态值函数、Q函数和策略。它学习连续层级之间的动作抽象，作为由各自的Q函数塑造的共享编码器。低层Q函数将选项作为目标，鼓励抽象保留优化控制所需的区别。高层Q函数将其作为动作，鼓励丢弃不必要的区别。在离线目标条件的运动和操控环境中，QSO学习到语义上有意义的选项空间，并超越基线，在所有其他评估算法失败的任务中实现非零性能。

View on arXiv Download PDF AI Translation

cs.AI / 104 / 2610.12142

Instruction-Conditioned Electromagnetic Spectrum Understanding via Budget-Adaptive Signal Tokenization

基于预算自适应信号标记化的指令条件电磁频谱理解

Zhai, Lei, Chang, Zhihao, Yang, Shuyuan, Feng, Zhixi

Abstract

Electromagnetic spectrum monitoring increasingly requires flexible analysis beyond task-specific recognition and detection. Multimodal large language models offer a unified interface, but extending vision-language models (VLMs) to raw I/Q signals requires tokenization that balances fidelity against a strict budget. For signals, dense encoding causes token costs to grow with observation length, whereas fixed-resolution compression may discard short-duration or localized signal evidence. Thus, we propose \textbf{BATok}, a budget-adaptive signal tokenizer that adjusts token capacity to the input length while allocating that capacity according to the signal content. BATok constructs candidate representations from signal-derived features using lightweight multi-resolution branches, then combines a local energy prior with learnable queries to resample these representations into compact signal tokens. The number of tokens adapts to the input length while remaining strictly bounded. The resulting tokens are projected into the language embedding space of VLMs. We further introduce \textbf{EMSpec-Instruct}, a multimodal instruction dataset aligning raw I/Q signals, waterfall images, and language supervision for modulation recognition, structured detection, and language-conditioned signal grounding. Experiments show that BATok learns effective signal representations and achieves competitive performance across all tasks.

Chinese Translation

电磁频谱监测越来越需要灵活的分析，超越任务特定的识别和检测。多模态大型语言模型提供了一个统一的接口，但将视觉-语言模型（VLMs）扩展到原始I/Q信号需要一种在严格预算下平衡保真度的标记化方法。对于信号而言，密集编码会导致标记成本随着观测长度的增长而增加，而固定分辨率压缩可能会丢弃短时或局部信号证据。因此，我们提出了 extbf{BATok}，一种预算自适应信号标记器，它根据输入长度调整标记容量，同时根据信号内容分配该容量。BATok利用轻量级多分辨率分支从信号派生特征构建候选表示，然后结合局部能量先验与可学习查询，将这些表示重新采样为紧凑的信号标记。标记的数量根据输入长度自适应，同时保持严格的界限。生成的标记被投影到VLMs的语言嵌入空间。我们进一步引入 extbf{EMSpec-Instruct}，一个多模态指令数据集，旨在对齐原始I/Q信号、瀑布图像和语言监督，用于调制识别、结构化检测和语言条件信号定位。实验表明，BATok学习到了有效的信号表示，并在所有任务中实现了具有竞争力的性能。

View on arXiv Download PDF AI Translation

cs.AI / 105 / 2610.12168

Learning to Plan by Looking Back: Hindsight Hierarchies for Training Reasoning Models

通过回顾学习规划：用于训练推理模型的回顾层次结构

Simon, Lars, Eble, Holger, Radons, Manuel

Abstract

We introduce a self-improvement loop for reasoning models based on the following observation: Even when the difficulty of a problem exceeds the model's current solving abilities, an additionally supplied solution might enable the model to extract useful solution ideas in hindsight. We operationalize this by jointly training the same model to exhibit the following three capabilities: predicting solution ideas from problems alone, reverse-engineering ideas from problems and known solutions, and solving problems using provided ideas. The loop alternates between reverse engineering such ideas from problems with supplied solutions and using these ideas as additional supervision for joint training of all three capabilities. We give a formal specification of our method and a concrete instantiation for interactive theorem proving in the Lean theorem prover; empirical evaluation remains future work.

Chinese Translation

我们提出了一种基于以下观察的推理模型自我改进循环：即使问题的难度超过模型当前的解决能力，额外提供的解决方案也可能使模型能够在事后提取有用的解决思路。我们通过联合训练同一模型，使其展现以下三种能力：仅从问题中预测解决思路、从问题和已知解决方案中逆向推导思路，以及使用提供的思路解决问题。该循环在从问题中逆向推导这些思路与使用这些思路作为对三种能力联合训练的额外监督之间交替进行。我们给出了我们方法的正式规范以及在 Lean 定理证明器中进行交互式定理证明的具体实例；实证评估仍然是未来的工作。

View on arXiv Download PDF AI Translation

cs.AI / 106 / 2610.12176

Recursive Self-Improvement through Multi-Agent Self-Supervision

通过多智能体自监督实现递归自我提升

Lee, Hyunin, Xu, Jinglue, Seely, Jeffrey, Lee, Donghyun, Sojoudi, Somayeh, Zaharia, Matei, Tang, Yujin

Abstract

Recursive self-improvement (RSI) of a model on non-verifiable tasks, such as open-ended research, faces a supervision bottleneck when its outputs exceed what even human experts can reliably assess, leaving the model itself (optimizee) as the best available optimizer and evaluator. However, a single model instance struggles to critique and improve its own complex reasoning under this homogeneous loop. To address this, we propose Multi-Agent Self-Supervision (MASS), an RSI method that alternates between evolutionary workflow optimization and supervised fine-tuning on self-generated trajectories. Guided by early findings that multi-agent topologies excel at complex reasoning, MASS prompts a single base model to iteratively propose, execute, and self-evaluate multi-agent workflows. Through an evolutionary search constrained by structural guardrails, the model optimizes these computational-graph-like orchestrations, discovering the most effective distinct roles and information routing for a given task. Over two MASS cycles with Qwen3.6-27B, the model achieves 1.2-1.6x higher performance per output tokens on four open-ended public benchmarks. Because the improved model subsequently acts as a better optimizer and evaluator, this alternating framework enables a continuous, recursive bootstrapping of the model's capabilities. Moreover, multi-agent traces are also more training-efficient: a student trained on them outperforms a single-agent student trained on 1.4x more training tokens. These findings suggest that jointly learning orchestration and bounded subagent execution from multi-agent trajectories can provide an effective signal for RSI.

Chinese Translation

在非可验证任务（如开放式研究）中，模型的递归自我提升（Recursive Self-Improvement, RSI）面临监督瓶颈，当模型输出超出人类专家能够可靠评估的范围时，模型自身（被优化者）成为最佳的优化器和评估者。然而，单一模型实例在这一同质循环中难以批判和改进自身复杂推理能力。为此，我们提出了多智能体自监督（Multi-Agent Self-Supervision, MASS）方法，这是一种在进化工作流优化与基于自生成轨迹的监督微调之间交替进行的RSI方法。基于早期发现多智能体拓扑结构在复杂推理中表现优异，MASS促使单一基础模型迭代地提出、执行并自我评估多智能体工作流。通过受结构性护栏约束的进化搜索，模型优化这些类似计算图的编排，发现针对特定任务最有效的不同角色分工和信息路由。在两轮基于Qwen3.6-27B的MASS循环中，模型在四个开放式公共基准测试中每输出标记的性能提升了1.2至1.6倍。由于改进后的模型随后作为更优的优化器和评估者，该交替框架实现了模型能力的持续递归自举。此外，多智能体轨迹在训练效率上也更具优势：基于其训练的学生模型性能优于使用1.4倍训练标记的单智能体学生模型。这些发现表明，从多智能体轨迹中联合学习编排与有界子代理执行能够为递归自我提升提供有效信号。

View on arXiv Download PDF AI Translation

cs.AI / 107 / 2610.12212

When Has a Bayesian Neural Network Sampled Enough? Adaptive Inference Time with Statistical Guarantees

贝叶斯神经网络何时采样足够？具有统计保证的自适应推断时间

Denoodt, Fabian, Hess, Sibylle

Abstract

Bayesian neural network predictions are commonly approximated using a fixed number of Monte Carlo samples per input, without controlling the resulting error that comes from this finite sample. We propose the use of confidence sequences to dynamically determine how many samples are needed while maintaining statistical guarantees. We consider several ways in which predictive probabilities are used, including identifying the most likely class, approximating the full predictive distribution, and resolving probability-threshold decisions. Sampling stops once the corresponding decision can be made with the desired guarantee. Experiments show that the method allocates the computational budget efficiently, assigning more samples to ambiguous inputs than to easy inputs while preserving reliable decisions and reducing overall latency relative to a fixed Monte Carlo budget.

Chinese Translation

贝叶斯神经网络的预测通常是通过对每个输入使用固定数量的蒙特卡洛样本来近似的，而没有控制由此有限样本带来的结果误差。我们提出使用置信序列动态确定所需样本数量，同时保持统计保证。我们考虑了预测概率的几种使用方式，包括识别最可能的类别、近似完整的预测分布以及解决概率阈值决策。一旦可以在所需保证下做出相应的决策，采样便停止。实验表明，该方法有效分配计算预算，对模糊输入分配更多样本，而对简单输入分配较少样本，同时保持可靠的决策，并相对于固定的蒙特卡洛预算减少整体延迟。

View on arXiv Download PDF AI Translation

cs.AI / 108 / 2610.12292

One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails

一句话打开大门：针对类型决策模型的选项通道攻击作为代理防护措施

Azizi, Seyedarmin, Potraghloo, Erfan Baghaei, Pedram, Massoud

Abstract

A typed decision model reads a piece of text and returns a probability over caller-defined options, each with a short written definition, generating no text. Recent work places these models in agent systems as guardrails: the component that reads a proposed tool call or incoming message and decides whether to allow it. We evaluate seven open-weight models in that role and report the two error directions separately: a fail-open error allows a prohibited action and is a vulnerability; a fail-closed error blocks a permitted one and is only a cost. On prompt-injection, jailbreak and toxic-content screening, accuracy at the allow-or-block decision ranges from 36% to 72% against a chance level of 50%. A low error rate in one direction only reflects which answer a model defaults to: one allows nearly everything, another blocks nearly everything. On a synthetic suite of agent tool calls, six lines of server log text that say nothing about the policy raise a gate's fail-open rate from 0% to 63% on a policy it otherwise decides correctly. Giving the permissive option a misleading name, with its definition and the judged text untouched, raises that rate to between 93% and 100% on the four models that place the label in their input. Every defense we tested is defeated, either by an attacker who targets its mechanism or by attacker-controlled text. Escalating the least confident decisions does not help either: a decision an attack has reversed is no less confident than the one it replaced. Parsing each policy field into a typed value does eliminate one attack, but it also makes the model unnecessary: a deterministic rule over those values reaches 100% accuracy on all six policies. These models can reduce how many cases reach a reviewer, but on this evidence they should not be the component that decides. Code is available at https://github.com/ArminAzizi98/option-channel-attack.

Chinese Translation

类型决策模型读取一段文本并返回对调用者定义的选项的概率，每个选项都有一个简短的书面定义，不生成文本。近期的研究将这些模型置于代理系统中作为防护措施：该组件读取提议的工具调用或传入消息，并决定是否允许该操作。我们评估了七个开放权重模型在这一角色中的表现，并分别报告了两种错误方向：失败开放错误允许被禁止的操作，是一种漏洞；失败关闭错误阻止被允许的操作，仅仅是成本。在提示注入、越狱和有害内容筛查方面，允许或阻止决策的准确率在36%到72%之间，而随机水平为50%。在一个方向上的低错误率仅反映了模型默认的答案：一个几乎允许所有内容，另一个几乎阻止所有内容。在一个合成的代理工具调用套件中，六行与政策无关的服务器日志文本将一个政策的失败开放率从0%提高到63%，而该政策在其他情况下是正确决定的。给允许的选项一个误导性的名称，同时保持其定义和判断文本不变，将该率提高到在四个将标签放入其输入中的模型中达到93%到100%。我们测试的每一种防御措施都被击败，要么是针对其机制的攻击者，要么是攻击者控制的文本。提升信心最低的决策也无济于事：一个被攻击逆转的决策并不比其替代决策更有信心。将每个政策字段解析为类型值确实消除了一个攻击，但这也使得模型变得不必要：对这些值的确定性规则在所有六个政策上达到了100%的准确率。这些模型可以减少达到审查者的案例数量，但根据这些证据，它们不应成为决策的组件。代码可在 https://github.com/ArminAzizi98/option-channel-attack 获取。

View on arXiv Download PDF AI Translation

cs.AI / 109 / 2610.12303

Learning Probabilistic Logic Programs with Functional Gradient Guided Language Models

基于功能梯度引导语言模型的概率逻辑程序学习

Mathur, Saurabh, Sidheekh, Sahil, Vasu, Bhavan, Tavakkoli, Farbod, Tadepalli, Prasad, Kersting, Kristian, Natarajan, Sriraam

Abstract

Declarative logic programs offer a powerful and interpretable abstraction for encoding relational structure and neurosymbolic reasoning, by expressing dependencies as weighted compositional rules. However, inducing them from data remains fundamentally hard, bottlenecked by the combinatorial explosion of symbolic search spaces. LLMs have recently emerged as powerful hypothesis generators, but when used in isolation, they lack the capacity to do systematic inductive reasoning needed to reliably synthesize valid programs that fit complex relational distributions. We introduce grasp (Gradient-boosted Synthesis of Probabilistic logic programs), a neurosymbolic framework that casts relational structure learning as functional gradient boosting in which the weak learner is a first-order rule and the intractable inner search is delegated to an LLM proposal oracle. We evaluate grasp on four relational benchmarks spanning molecular toxicity prediction (Tox21), mutagenesis, and citation matching (Cora), and show that it improves over purely symbolic, neural, and LLM-based baselines, while producing interpretable weighted rule ensembles. By replacing combinatorial search with gradient-guided LLM hypothesis generation, grasp retains boosting guarantees without sacrificing the transparency of symbolic outputs.

Chinese Translation

声明性逻辑程序提供了一种强大且可解释的抽象，用于编码关系结构和神经符号推理，通过将依赖关系表达为加权组合规则。然而，从数据中诱导这些程序仍然是根本性的困难，受限于符号搜索空间的组合爆炸。大型语言模型（LLMs）最近作为强大的假设生成器出现，但在单独使用时，它们缺乏进行系统性归纳推理的能力，这对于可靠地合成适合复杂关系分布的有效程序是必需的。我们提出了 grasp（基于梯度提升的概率逻辑程序合成），这是一个神经符号框架，将关系结构学习视为功能梯度提升，其中弱学习者是一个一阶规则，而难以处理的内部搜索则委托给 LLM 提议的神谕。我们在四个关系基准上评估 grasp，涵盖了分子毒性预测（Tox21）、诱变性和引用匹配（Cora），结果表明其优于纯符号、神经和基于 LLM 的基线，同时生成可解释的加权规则集。通过用梯度引导的 LLM 假设生成替代组合搜索，grasp 保留了提升保证，同时不牺牲符号输出的透明性。

View on arXiv Download PDF AI Translation

cs.AI / 110 / 2610.12304

Looking Inside LLMs: Small-World Connectivity as a Signature of Reasoning Performance

深入探讨大型语言模型（LLM）：小世界连接性作为推理性能的特征

Huang, Zheng, Cao, Sansheng, Zhang, Enpei, Qiu, Weikang, Chen, Elynn, Zhang, Xiang, Yang, Yaoqing, Ying, Rex, Zhou, Dawei, Yan, Yujun

Abstract

Understanding large language model (LLM) reasoning requires looking beyond behavioral performance to examine how reasoning ability is reflected in internal organization. Inspired by neuroscience findings linking higher intelligence to stronger small-world organization in functional brain networks, we investigate small-world connectivity as a structural signature of LLM reasoning. We construct functional graphs from attention-head activation similarities and find that a higher small-world index (SWI), capturing local clustering and short global paths, consistently correlates with better fluid reasoning performance across models and training checkpoints. Since local clustering is central to small-world organization, we further examine how heads important for model performance connect within and across communities. We find that these heads tend to have a larger share of connection weight within their own communities (high core scores) and a more concentrated weight distribution across communities (low bridge scores). These observations motivate the hypothesis that high core and low bridge scores serve as structural indicators of head importance for reasoning capability. We validate this hypothesis through pruning, introducing Small-World Allocation (SWA), a hierarchical sparsity allocation method guided by these scores. Across six LLMs, SWA better preserves small-world organization and model performance than competing allocation strategies, reducing WikiText perplexity by up to 20%. Together, these findings identify small-world functional connectivity as a measurable signature of LLM reasoning performance, offering a structural perspective that complements behavioral evaluation.

Chinese Translation

理解大型语言模型（LLM）的推理能力需要超越行为表现，考察推理能力如何在内部组织中体现。受神经科学研究启发，发现更高的智力与功能脑网络中更强的小世界组织相关，我们探讨小世界连接性作为LLM推理的结构特征。我们从注意力头激活相似性构建功能图，并发现更高的小世界指数（SWI），捕捉局部聚类和短的全局路径，与模型和训练检查点中的更好流体推理表现一致。由于局部聚类是小世界组织的核心，我们进一步考察对模型性能重要的头部如何在社区内外连接。我们发现这些头部在自身社区内的连接权重占比更大（高核心分数），而在社区间的权重分布更集中（低桥分数）。这些观察促使我们提出假设：高核心分数和低桥分数作为推理能力的重要性结构指标。我们通过剪枝验证这一假设，提出小世界分配（Small-World Allocation, SWA），一种基于这些分数的层次稀疏分配方法。在六个LLM中，SWA比竞争性分配策略更好地保留小世界组织和模型性能，将WikiText困惑度降低了多达20%。综合来看，这些发现将小世界功能连接性识别为LLM推理性能的可测量特征，提供了一种补充行为评估的结构视角。

View on arXiv Download PDF AI Translation

cs.AI / 111 / 2610.12306

A Structural Theory of Cognitive Representation and Problem Solving,Contexts, Invariance, and the Knowledge Space

认知表征与问题解决的结构理论：上下文、不变性与知识空间

Jakovác, Antal, Telcs, András

Abstract

Learning and problem solving depend critically on the structure of internal representations. While many modern data-driven artificial systems achieve strong predictive performance, their learned representations often lack explicit structure for expressing abstraction, invariance, and task-relevant regularities. We propose a minimal structural framework in which representational operations relevant to problem solving, such as context formation, invariance recognition, representative selection, abstraction, and procedural reuse, are made explicit. The central notion is that of a \emph{context}, formalized as a partition of a subset of an underlying state space, which fixes the distinctions, granularity, and form in which a problem can be posed. Within this setting, invariance recognition and representative selection are treated as fundamental representational operations. The framework is realized as a Knowledge Space composed of two coupled graph structures: a Concept Graph that hosts constructed and refined concepts, and a Procedure Graph that encodes typed operations over representations. Together, these structures provide a minimal cognitive-representational algebra for operating on representations without assuming sophisticated inference, learning, control, perception, or motor mechanisms. Using simple illustrative examples and a finite weak-solver demonstration, we show that appropriate representational organization can simplify the form and scope of admissible regularities, even when problem solving is carried out by a fixed and limited solver. The contribution of the paper is structural rather than algorithmic: it identifies representational prerequisites for abstraction, invariance, and procedural reuse in problem solving, and states explicit success and failure conditions for the weak-solver setting.

Chinese Translation

学习和问题解决在很大程度上依赖于内部表征的结构。尽管许多现代数据驱动的人工系统在预测性能上表现出色，但它们所学习的表征往往缺乏用于表达抽象、不变性和任务相关规律的明确结构。我们提出了一个最小结构框架，在该框架中，与问题解决相关的表征操作，如上下文形成、不变性识别、代表选择、抽象和过程重用，被明确化。核心概念是 extit{上下文}，被形式化为基础状态空间子集的划分，固定了问题可以被提出的区分、粒度和形式。在这个设置中，不变性识别和代表选择被视为基本的表征操作。该框架被实现为一个知识空间，由两个耦合的图结构组成：一个概念图（Concept Graph），用于承载构建和精炼的概念；一个过程图（Procedure Graph），用于编码对表征的类型化操作。这些结构共同提供了一种最小的认知-表征代数，用于在不假设复杂推理、学习、控制、感知或运动机制的情况下对表征进行操作。通过简单的示例和有限弱求解器的演示，我们展示了适当的表征组织可以简化可接受规律的形式和范围，即使问题解决是由一个固定且有限的求解器进行的。本文的贡献在于结构性而非算法性：它识别了在问题解决中进行抽象、不变性和过程重用的表征前提，并明确了弱求解器设置的成功和失败条件。

View on arXiv Download PDF AI Translation

cs.AI / 112 / 2610.12313

Verdict Without the Rule: Diagnosing and Auditing Regulatory Rule Sensitivity in LLM Compliance Systems

没有规则的裁决：诊断和审计大型语言模型合规系统中的监管规则敏感性

Sadhu, Saisab, Sengupta, Aadit, Sankarapu, Vinay kumar, Seth, Pratinav

Abstract

Large language model compliance systems are deployed on the assumption that a verdict depends on the regulatory rule it is given. We test this directly across five models and 20 regulatory and platform-policy domains: delete, swap, or negate the governing rule while holding the case fixed, and check whether the verdict changes (OCS) or the model's internal representation of compliance shifts at all (ICS-delta). Neither moves much: models' verdicts are often invariant to substantial perturbations of the supplied rule, and the guard model, evaluated here under a custom-rule adaptation of its native taxonomy, is the least rule-sensitive and least accurate of the five, barely above chance (51%, versus 90-92% for general-purpose models). This reflects easy cases more than blanket neglect: on cases where deleting the rule changes a previously correct model prediction, models do track it closely. Neither better prompting nor direct intervention on the model's internal representations closes this gap. Accuracy alone does not establish that a compliance verdict is grounded in the supplied rule.

Chinese Translation

大型语言模型合规系统的部署假设裁决依赖于所给定的监管规则。我们在五个模型和20个监管及平台政策领域中直接测试这一假设：在固定案例的情况下删除、交换或否定主导规则，并检查裁决是否发生变化（OCS）或模型对合规性的内部表示是否发生变化（ICS-delta）。结果显示，两者的变化都很小：模型的裁决往往对所提供规则的重大扰动保持不变，而在这里评估的守护模型，在其本土分类法的自定义规则适应下，是五个模型中对规则敏感性最低且准确性最低的，几乎接近随机猜测（51%，而通用模型为90-92%）。这反映了简单案例的特征，而非全面忽视：在删除规则会改变之前正确的模型预测的案例中，模型确实紧密跟踪这一变化。无论是更好的提示还是对模型内部表示的直接干预，都无法缩小这一差距。仅凭准确性并不能证明合规裁决是基于所提供规则的。

View on arXiv Download PDF AI Translation

cs.AI / 113 / 2610.12325

Prior or Feedback? What an LLM Uses When Adapting Neural Operators

先验还是反馈？大型语言模型在适应神经算子时的使用方式

Chan, Julian, Jimenez, Javier Mora

Abstract

Do LLM scientific agents rely only on their initial task context, or do they adapt their decisions in response to experimental feedback? We study this question in neural operator adaptation, where a large language model (LLM) selects fine-tuning configurations under a limited trial budget. Across transfers within and between partial differential equation (PDE) families, the LLM achieves lower held-out test nRMSE than random search and Bayesian optimisation in nearly every matched comparison. Endpoint performance alone cannot distinguish what happens, so we verify each attribution with controlled interventions. Before observing any validation score, the LLM's first configuration already ranks near the top of the corresponding random-search pool, indicating a useful initial bias. A complementary cold-start intervention shows that the selected base learning rate shifts with the PDE description. Once feedback becomes available, reassigning validation scores among evaluated configurations changes the next proposal in every case tested, whereas a value-preserving rewrite produces no comparable aggregate effect. These interventions establish that the LLM's decision-level actions respond to the given task and observed outcomes, showing that it combines a task-dependent prior with sensitivity to experimental feedback.

Chinese Translation

大型语言模型（LLM）科学代理是仅依赖于其初始任务上下文，还是根据实验反馈调整其决策？我们在神经算子适应中研究了这个问题，其中大型语言模型在有限的试验预算下选择微调配置。在部分微分方程（PDE）家族内外的迁移中，LLM在几乎每个匹配比较中实现了比随机搜索和贝叶斯优化更低的保留测试nRMSE。仅凭最终性能无法区分发生了什么，因此我们通过控制干预验证每个归因。在观察到任何验证分数之前，LLM的第一个配置已经在相应的随机搜索池中排名靠前，表明存在有用的初始偏倚。一项补充的冷启动干预表明，所选的基础学习率随着PDE描述的变化而变化。一旦反馈可用，在评估的配置之间重新分配验证分数会在每个测试案例中改变下一个提议，而价值保持重写则没有产生可比的整体效果。这些干预表明，LLM的决策级别行为对给定任务和观察结果做出反应，显示出它结合了任务依赖的先验和对实验反馈的敏感性。

View on arXiv Download PDF AI Translation

cs.AI / 114 / 2610.12341

Can AI Agents Learn Their Way to the Top? Evaluating Heuristic Learning in a Long-Running Game Agent Competition

人工智能代理能否学习登顶？评估长期游戏代理竞赛中的启发式学习

Yang, Kaisen, Liu, Qingle, Wang, Kejin, Zhao, Yicheng, Li, Jieming, Zheng, Shenghan, Yang, Ruize, Yang, Bojun, Gong, Heng, Gao, Xiang, Zhang, Lanyue, Zhong, Kaiyu, Liu, Zhuo, Li, Shaoxuan, Li, Chengxi, Yan, Yong, Zhang, Weixuan, Luo, Tianwei, Wang, Situ, Zheng, Youjie, Zhao, Sihan, Wang, Shengyuan, Gao, Huan-ang, Xu, Jiazheng, Xie, Xiaohui, Han, Wentao, Wang, Hongning

Abstract

Adversarial games have driven advances from heuristic search to reinforcement learning, yet learning and adapting strategies from limited samples remain challenging. AI agents offer an alternative by turning game experience into revisions of executable policies. Building on heuristic learning (HL), we formalize Adversarial Heuristic Learning (AHL), a paradigm that uses AI agents as learning engines to refine game policies and supporting software while keeping model weights fixed. We introduce AAArena, a benchmark comprising 12 authentic adversarial games and 1,920 archived human programs, with an evaluation protocol modeled on real-world game competitions. Agents interpret rules, choose opponents, analyze replays, and revise game agents to achieve their highest ranking within fixed match and evaluation budgets. We evaluate \val{completedmodels} model and harness configurations: Opus5.5 with Claude Code earns 6 gold medals, while no evaluated configuration tops the remaining 6 human ladders. Performance is generally weaker in games with more complex rule specifications. Further experiments show that opponent selection and dense feedback support policy improvement, and that agents learn from both on-policy replays of their own matches and off-policy replays of other players' matches. These results highlight HL's potential in adversarial games and identify persistent challenges in game understanding, strategy implementation, and long-horizon policy development.

Chinese Translation

对抗性游戏推动了从启发式搜索到强化学习的进展，但从有限样本中学习和调整策略仍然具有挑战性。人工智能代理通过将游戏经验转化为可执行策略的修订提供了一种替代方案。基于启发式学习（HL），我们形式化了对抗性启发式学习（AHL），这一范式利用人工智能代理作为学习引擎来优化游戏策略和支持软件，同时保持模型权重不变。我们引入了AAArena，一个包含12个真实对抗性游戏和1,920个存档人类程序的基准测试，评估协议基于现实世界的游戏竞赛模型。代理解释规则、选择对手、分析重播，并修订游戏代理，以在固定的比赛和评估预算内实现最高排名。我们评估了 ext{completedmodels}模型和配置：Opus5.5与Claude Code获得6枚金牌，而没有评估的配置能超越剩余的6个人工阶梯。在规则规格更复杂的游戏中，性能通常较弱。进一步的实验表明，对手选择和密集反馈支持策略改进，代理从自己比赛的在线重播和其他玩家比赛的离线重播中学习。这些结果突显了HL在对抗性游戏中的潜力，并识别了游戏理解、策略实施和长远政策发展中的持续挑战。

View on arXiv Download PDF AI Translation

cs.AI / 115 / 2610.12345

Overcoming Prior Barriers: Supervised Fine-Tuning under Long-Tail Distribution

克服先前障碍：长尾分布下的监督微调

Wang, Haohui, Xu, Jiahao, Zhan, Wangzhi, Zeng, Tong, Fu, Dongqi, Li, Hong, Roy, Swastik, Ramakrishnan, Naren, North, Chris, Kang, Jian, Yan, Yujun, Zhou, Dawei

Abstract

Supervised fine-tuning (SFT) adapts pretrained large language models (LLMs) to downstream tasks, but the required concepts can receive substantially different levels of pretrained support. Frequent concepts are more likely to be well learned, whereas rare concepts may remain weakly represented. We introduce a novel notion named prior barrier to quantify how strongly the pretrained model supports competing concepts over the target concept. We observe that prior barriers follow a long-tail distribution, placing head and tail concepts at different starting points for SFT: head concepts face lower prior barriers, whereas tail concepts require additional instructions to overcome their higher prior barriers. Our theoretical analysis further derives a predictive risk bound for SFT under long-tail prior barriers, explicitly characterizing how the prior barrier and accumulated SFT evidence jointly determine predictive performance. Motivated by this prior barrier-dependent demand, we propose PASS, an adaptive SFT instruction selection method that constructs reference-derived concepts and estimates the distinguishing evidence provided by each instruction, and adaptively allocates the selection budget toward concepts that remain insufficiently covered under the current selection. In this way, PASS jointly considers which instructions can provide useful evidence and where additional supervision is needed under a limited budget. Experiments show that our method consistently outperforms seven state-of-the-art instruction selection methods on four backbone-budget settings. An ablation study further shows that PASS's adaptive allocation consistently improves over uniform allocation.

Chinese Translation

监督微调（SFT）将预训练的大型语言模型（LLMs）适应于下游任务，但所需的概念可能会受到不同程度的预训练支持。频繁出现的概念更有可能被很好地学习，而稀有概念可能仍然表现较弱。我们引入了一种新概念，称为先前障碍，以量化预训练模型对目标概念的竞争概念的支持强度。我们观察到先前障碍遵循长尾分布，使得头部和尾部概念在SFT中处于不同的起点：头部概念面临较低的先前障碍，而尾部概念则需要额外的指导以克服其较高的先前障碍。我们的理论分析进一步推导出在长尾先前障碍下SFT的预测风险界限，明确表征了先前障碍和累积的SFT证据如何共同决定预测性能。受到这种依赖先前障碍的需求的启发，我们提出了PASS，一种自适应SFT指令选择方法，该方法构建参考派生概念并估计每个指令提供的区分证据，并自适应地将选择预算分配给在当前选择下仍然覆盖不足的概念。通过这种方式，PASS共同考虑了哪些指令可以提供有用的证据，以及在有限预算下需要额外监督的地方。实验表明，我们的方法在四个主干预算设置上始终优于七种最先进的指令选择方法。消融研究进一步表明，PASS的自适应分配始终优于均匀分配。

View on arXiv Download PDF AI Translation

cs.AI / 116 / 2610.12360

Accurate but Not Humble: Evaluating Epistemic Humility in LLM Agents under Knowledge Conflict

准确但不谦逊：在知识冲突下评估大型语言模型代理的认知谦逊

Sun, Kaiser, Gutierrez, Bernal Jimenez, Liu, Hongjun, Zhang, Jingyu, Gao, Jie, Dredze, Mark, Khashabi, Daniel

Abstract

When retrieved evidence contradicts an agent's prior beliefs, does it revise its answer, acknowledge uncertainty, or persist with an incorrect conclusion? Existing evaluations of agentic systems focus primarily on task success, offering limited insight into how agents handle such conflicts. We propose to evaluate agents on epistemic humility (EH): the agent's willingness to recognize, act on, and communicate uncertainty during task execution. We operationalize EH through three trajectory-level behavioral dimensions: Identify, Solve, and Escalate (ISE). Through knowledge conflict, situations where the backbone language model's parametric knowledge contradicts the evidence it encounters, or where two contextual sources disagree, we evaluate two conflict settings: (1) controlled conflict and (2) naturally occurring conflict during multi-step agentic execution, each paired with matched no-conflict controls. Evaluating four agents, we find that higher task accuracy does not necessarily correspond to greater epistemic humility: some high-accuracy configurations recognize conflicts during execution but do not communicate unresolved uncertainty in their incorrect final answers. Trajectory-level analysis further reveals that agents frequently detect conflicts in early steps of execution but fail to maintain or resolve them in later steps. Finally, we show that model-level interventions can improve EH, but often at the cost of task accuracy, suggesting that epistemic humility emerges from the interaction among the backbone model, agent harness, and evaluation environment.

Chinese Translation

当检索到的证据与代理的先前信念相矛盾时，代理会修正其答案、承认不确定性，还是坚持错误的结论？现有的代理系统评估主要集中在任务成功上，提供的关于代理如何处理此类冲突的见解有限。我们建议评估代理的认知谦逊（Epistemic Humility, EH）：代理在任务执行过程中识别、采取行动并传达不确定性的意愿。我们通过三个轨迹级行为维度来操作化EH：识别（Identify）、解决（Solve）和升级（Escalate）（ISE）。通过知识冲突，即基础语言模型的参数知识与其遇到的证据相矛盾，或两个上下文来源不一致的情况，我们评估了两种冲突设置：（1）受控冲突和（2）在多步骤代理执行过程中自然发生的冲突，每种设置都配有匹配的无冲突控制组。评估四个代理后，我们发现更高的任务准确性并不一定对应于更大的认知谦逊：一些高准确性的配置在执行过程中识别冲突，但未能在其错误的最终答案中传达未解决的不确定性。轨迹级分析进一步揭示，代理在执行的早期步骤中经常检测到冲突，但未能在后续步骤中维持或解决这些冲突。最后，我们展示了模型级干预可以改善EH，但通常以牺牲任务准确性为代价，这表明认知谦逊是基础模型、代理框架和评估环境之间相互作用的结果。

View on arXiv Download PDF AI Translation

cs.AI / 117 / 2610.12361

Cited but Not Consulted: A Counterfactual Audit of Legal Chain-of-Thought Faithfulness

被引用但未咨询：法律推理链条忠实性的反事实审计

Sadhu, Saisab, Arora, Shreeyans, Seth, Pratinav

Abstract

Large language models increasingly justify legal decisions by naming the statute or precedent behind a verdict, treated as evidence that the decision follows from it. We test this directly: holding case facts fixed, we substitute the named legal authority for an unrelated one and decode a model's evolving verdict from its hidden states. Across seven open-weight models (8B-70B) and four benchmarks spanning judicial and contractual reasoning, when explicitly required to justify a verdict by naming the governing authority, models name the correct one in 66.7%-100% of generations, while the verdict changing when the authority changes is far less consistent: 0.0%-21.7% on CaseHOLD, 30.0%-76.7% on ECHR and SCOTUS, and 43.3%-50.0% on ContractNLI. Neither scale nor a purpose-built legal-reasoning model (a best-effort LoRA reproduction; Section 6) closes this gap. A red-teaming evaluation on five core models finds compliance with an adversarial instruction hidden in the case facts (73.3%-96.4%) exceeds verdict-swap sensitivity by a wide margin, holding without exception across model rankings. Naming a legal authority is thus a poor proxy for a verdict's dependence on it, while the same verdict remains separately vulnerable to adversarial manipulation. Both findings replicate across checks ruling out prompt-wording noise and confounded sampling, and bear directly on the use of generated legal explanations as compliance or audit artefacts.

Chinese Translation

大型语言模型越来越多地通过提及判决背后的法规或先例来为法律决策提供依据，这被视为该决策遵循这些依据的证据。我们直接测试这一点：在保持案件事实不变的情况下，我们将所提及的法律权威替换为一个无关的权威，并从模型的隐藏状态中解码其不断变化的判决。在七个开放权重模型（8B-70B）和四个涵盖司法和合同推理的基准测试中，当明确要求通过命名主管权威来为判决提供依据时，模型在66.7%-100%的生成中命名了正确的权威，而当权威发生变化时，判决的变化则远不如一致：在CaseHOLD上为0.0%-21.7%，在ECHR和SCOTUS上为30.0%-76.7%，在ContractNLI上为43.3%-50.0%。无论是规模还是专门构建的法律推理模型（最佳努力的LoRA重现；第6节）都无法缩小这一差距。对五个核心模型的红队评估发现，遵循隐藏在案件事实中的对抗性指令（73.3%-96.4%）的合规性远远超过判决交换的敏感性，并且在模型排名中没有例外。因此，命名法律权威并不是判决对其依赖的良好代理，而同一判决仍然容易受到对抗性操控。这两个发现通过排除提示措辞噪声和混淆抽样的检查得以复制，并直接影响生成的法律解释作为合规或审计工具的使用。

View on arXiv Download PDF AI Translation

cs.AI / 118 / 2610.12363

HANS: A Handwritten Answer Sheet Dataset for Noisy Hybrid Document Parsing

HANS：一个用于嘈杂混合文档解析的手写答题纸数据集

Wu, Xiazhen, Qin, Wansong, Zheng, Yangbin, Fang, Liangda, Li, Zhan, Huang, Xiujie, Zhou, Liushen, Guan, Quanlong

Abstract

Intelligent grading and automated scoring technologies constitute critical infrastructure for smart education. However, existing document parsing and handwriting recognition benchmarks are predominantly designed for well-structured printed documents or isolated mathematical expressions, lacking datasets that capture the complex characteristics inherent to student answer sheets, including multi-line derivation processes, heterogeneous mixtures of text and mathematical formulae, and noise artifacts such as strikethroughs. To address this gap, we introduce HANS, the first dataset explicitly constructed for real-world educational scenarios, encompassing mathematical expressions, natural language text, hand-drawn tables, and diverse noise patterns including corrections and deletions, accompanied by fine-grained annotations that establish a reliable foundation for robust recognition research. Building upon HANS, we propose NA-GOT, an end-to-end framework that achieves two-stage noise suppression through a lightweight noise suppression module operating at the feature level, complemented by a noiseaware attention mechanism incorporated into the decoding stage. Experimental results demonstrate that HANS poses substantial challenges to existing methods, while NA-GOT achieves significant improvements in both accuracy and stability for answer process recognition. The dataset will be made publicly available upon publication.

Chinese Translation

智能评分和自动评分技术构成了智能教育的重要基础设施。然而，现有的文档解析和手写识别基准主要针对结构良好的印刷文档或孤立的数学表达式，缺乏能够捕捉学生答题纸固有复杂特征的数据集，包括多行推导过程、文本与数学公式的异构混合以及诸如划线等噪声伪影。为了解决这一问题，我们引入了HANS，这是第一个专门为现实教育场景构建的数据集，涵盖数学表达式、自然语言文本、手绘表格以及包括更正和删除在内的多样噪声模式，并附有细粒度的注释，为稳健的识别研究奠定了可靠的基础。在HANS的基础上，我们提出了NA-GOT，一个端到端框架，通过在特征层面操作的轻量级噪声抑制模块实现两阶段的噪声抑制，并在解码阶段结合了噪声感知注意力机制。实验结果表明，HANS对现有方法提出了重大挑战，而NA-GOT在答题过程识别的准确性和稳定性方面取得了显著提升。该数据集将在发表后公开提供。

View on arXiv Download PDF AI Translation

cs.AI / 119 / 2610.12375

OnTrack: Real-Time Monitoring and Intervention in LLM Agent Trajectories via Streaming Structure-Aware Optimal Transport

OnTrack：通过流式结构感知最优传输实现 LLM 代理轨迹的实时监控与干预

Barazandeh, Babak, Swanson, Connor, Kulkarni, Chinmay, Mungel, Nikhil

Abstract

Agents are deployed in applications from trip planners and stock trading to IT incident triage. In most cases, LLM agents work autonomously with minimal rule-based safeguarding, leading to cost and safety issues from irreversible actions. Recent works resolve this either by using a safeguard agent to monitor behavior or evaluating logs post-hoc. The first adds cost and latency to every step; the second delivers its verdict after the run, when tokens are burned and damage is done. To overcome this, we propose OnTrack, a streaming monitoring mechanism that compares an agent's steps and dependencies against recorded successful runs to alert users or block the agent in about a millisecond per step. We study this problem in three regimes of decreasing access: full reference access (historical runs and tool schemas), intermediate access (only tool schemas), and no prior knowledge (only step logs as generated). Expectation of OnTrack's monitoring capabilities reduces as data access drops, ranging from plan violation detection to identifying loops, stalls, and repeated tool calls. Finally, we evaluate OnTrack using SWE-bench trajectories. Based on the first 8 steps, our method ranks failing trajectories below succeeding ones better than content similarity approaches (+0.057 AUROC). With an abort policy, we save about 18% of compute that would be burned on failing runs, where 83% of interrupted runs were actually heading to failure (5 out of 6 aborts were correct).

Chinese Translation

代理被广泛应用于旅行规划、股票交易和 IT 事件分类等领域。在大多数情况下，LLM 代理以最小的基于规则的保护措施自主工作，这导致了不可逆操作带来的成本和安全问题。近期的研究通过使用保护代理监控行为或事后评估日志来解决这一问题。前者在每一步增加了成本和延迟；后者在运行结束后才给出判断，此时令牌已经消耗，损害已经发生。为了解决这个问题，我们提出了 OnTrack，一种流式监控机制，能够在约毫秒级的时间内将代理的步骤和依赖与记录的成功运行进行比较，以警告用户或阻止代理。我们在三种不同的数据访问模式下研究了这一问题：完全参考访问（历史运行和工具模式）、中等访问（仅工具模式）和无先验知识（仅生成的步骤日志）。随着数据访问的减少，OnTrack 的监控能力的期望值也随之降低，从计划违规检测到识别循环、停滞和重复工具调用。最后，我们使用 SWE-bench 轨迹评估了 OnTrack。基于前 8 步，我们的方法在失败轨迹的排名上优于成功轨迹，表现优于内容相似性方法（+0.057 AUROC）。通过中止策略，我们节省了约 18% 的计算资源，避免了在失败运行上浪费的计算，其中 83% 的中断运行实际上是走向失败的（6 次中止中有 5 次是正确的）。

View on arXiv Download PDF AI Translation

cs.AI / 120 / 2610.12391

GeoReform: Reflective Formalization Evolution for Multimodal Geometry Problem Solving

GeoReform：用于多模态几何问题解决的反思性形式化演化

Wang, Jialu, Zhang, Ruichen, Liu, Xiaoou, Wei, Hua, Chen, Tianlong

Abstract

Multimodal large language models (MLLMs) often struggle to identify and use geometric relations in diagrams. Recent methods address this challenge by converting geometric entities, relations, and constraints into explicit textual representations for the model to reason over. However, effective formalization is highly non-trivial: on Geometry3K, structure injection fixes 28 errors but introduces 13 new ones among 200 examples. Redundant relations can distract the model, while ambiguous references to diagram elements can lead it to apply constraints incorrectly. This suggests that the key challenge is not merely extracting more geometric facts, but organizing them into representations that support downstream reasoning. To fully exploit the power of formalization, we further propose GeoReform, a reflective formalization evolution framework that treats formalization as an optimizable policy rather than a fixed parser output. GeoReform executes the full reasoning pipeline, collects failed rollouts, diagnoses defects in the current representation, and mutates the policy to better select, ground, group, and present geometric entities, relations, constraints, and targets. On Geometry3K, GeoReform improves Qwen3VL-2B accuracy from 42.0\% to 56.0\%. Extensive experiments and analyses across geometry reasoning benchmarks demonstrate that effective formalization is crucial for improving multimodal geometry reasoning.

Chinese Translation

多模态大型语言模型（MLLMs）在识别和使用图示中的几何关系方面常常面临困难。近期的方法通过将几何实体、关系和约束转换为模型可以推理的明确文本表示来应对这一挑战。然而，有效的形式化并非易事：在Geometry3K数据集上，结构注入修复了28个错误，但在200个示例中引入了13个新错误。冗余关系可能会分散模型的注意力，而对图示元素的模糊引用可能导致模型错误地应用约束。这表明，关键挑战不仅在于提取更多的几何事实，而在于将其组织成支持下游推理的表示。为了充分利用形式化的力量，我们进一步提出了GeoReform，一个将形式化视为可优化策略而非固定解析器输出的反思性形式化演化框架。GeoReform执行完整的推理流程，收集失败的回滚，诊断当前表示中的缺陷，并变异策略以更好地选择、定位、分组和呈现几何实体、关系、约束和目标。在Geometry3K上，GeoReform将Qwen3VL-2B的准确率从42.0\%提高到56.0\%。在几何推理基准上的广泛实验和分析表明，有效的形式化对于提高多模态几何推理至关重要。

View on arXiv Download PDF AI Translation

cs.AI / 121 / 2610.12393

HRIL: Learning Multimodal Synergy via Higher-Order Tensor Modeling

HRIL：通过高阶张量建模学习多模态协同

Dai, Qun, Wen, Liangjian, Duan, Jiang, Dai, Yong, Wang, Dongkai, Wang, Maolin, Wang, Mingjie, Liu, Jianzhuang, Yan, He, Kang, Zhao

Abstract

Self-supervised multimodal representation learning has achieved remarkable success across diverse domains, yet capturing synergistic information remains challenging due to the complexity of cross-modal interactions. Unlike the shared information across individual modalities, synergy arises when task-relevant signals emerge only from the joint configuration of multiple modalities and cannot be recovered from any modality in isolation. This work focuses on how to preserve the information capacity for such synergistic signals in multimodal representations. The key observation is that synergistic information is reflected in higher-order statistical dependence among modalities, which provides a principled target for explicitly modeling joint interactions. Motivated by this insight, we propose Higher-order Representation and Information Learning (HRIL), which constructs an empirical cross-moment tensor over modality embeddings to represent multi-way interactions. HRIL employs Tucker decomposition to obtain a core tensor, complemented by a synergy-aware regularizer that prevents energy concentration and preserves higher-order coupling capacity for synergistic information capture. Experiments on the controlled synergy task and real-world benchmarks demonstrate consistent improvements over existing multimodal contrastive methods, with notable gains on tasks dominated by synergistic interactions. Code is released at https://github.com/brightest66/HRIL.

Chinese Translation

自监督多模态表示学习在多个领域取得了显著成功，但由于跨模态交互的复杂性，捕捉协同信息仍然具有挑战性。与各个模态之间的共享信息不同，协同信息是在多个模态的联合配置中仅出现与任务相关的信号，而无法从任何单一模态中恢复。本文重点研究如何在多模态表示中保留此类协同信号的信息容量。关键观察是，协同信息反映在模态之间的高阶统计依赖关系中，这为显式建模联合交互提供了一个原则性目标。基于这一见解，我们提出了高阶表示与信息学习（Higher-order Representation and Information Learning, HRIL），该方法构建了一个经验交叉矩阵张量，以表示多向交互。HRIL采用塔克分解（Tucker decomposition）来获取核心张量，并辅以一个关注协同的正则化器，以防止能量集中并保留高阶耦合能力，以捕捉协同信息。在受控协同任务和现实世界基准上的实验表明，HRIL在现有多模态对比方法上实现了一致的改进，尤其在以协同交互为主导的任务上取得了显著提升。代码已发布在 https://github.com/brightest66/HRIL。

View on arXiv Download PDF AI Translation

cs.AI / 122 / 2610.12409

Searching for "Harmful Refusal": A Psychometric Audit of an AI Safety Benchmark

寻找“有害拒绝”：人工智能安全基准的心理测量审计

Stewart, Christopher M., Botter, Preston, Sarabosing, Natalie, Zhang, Muye, Phinnemore, Rachel, Ghosh, Shalini, Shen, Hong, Heidari, Hoda

Abstract

Safety benchmarks typically report one overall score for a suite of datasets, each of which may target one or more safety-related attributes, so models with similar overall scores can have very different attribute profiles. Comparing models is more tractable at the level of individual attributes, yet it is often unclear whether even a single dataset's scores isolate any single attribute. One plausible candidate for such an attribute is harmful refusal, a model's tendency to refuse dangerous or policy-violating prompts. We examine whether it constitutes a single, measurable attribute in HELM Safety. Using a construct validity framework that stipulates that an attribute must exist before a test can measure it, we start with HELM Safety's four datasets that might plausibly target harmful refusal, but find that three are saturated. We subject the remaining dataset, HarmBench, to two psychometric tests to determine if a single attribute like harmful refusal could stand behind its score. First, multidimensional item response theory modeling strongly suggests that HarmBench does not measure a singular attribute. Second, a differential item functioning analysis finds items where models from different developers with the same refusal ability score differently. These flags largely disappear under scope-specific matching, a pattern consistent with aggregation effects but not sufficient to rule out domain-specific developer differences. Zooming out, HarmBench collapses distinct harm behaviors into one score, and the overall HELM safety aggregate further collapses HarmBench and scores from other datasets into a single top-line number. Any safety score that averages over datasets and items can hide saturation and conflate behaviors this way. We argue that a score should earn its single-attribute reading before models are compared with it.

Chinese Translation

安全基准通常为一组数据集报告一个总体得分，每个数据集可能针对一个或多个与安全相关的属性，因此具有相似总体得分的模型可能具有非常不同的属性特征。在个别属性层面比较模型更为可行，但通常不清楚即使是单个数据集的得分是否能够隔离出任何单一属性。有害拒绝是一个合理的候选属性，指的是模型拒绝危险或违反政策提示的倾向。我们考察它是否构成HELM安全中的一个单一可测量属性。我们使用一个构念效度框架，该框架规定在测试能够测量某个属性之前，该属性必须存在。我们从HELM安全的四个数据集开始，这些数据集可能合理地针对有害拒绝，但发现其中三个数据集已饱和。我们对剩余的数据集HarmBench进行了两项心理测量测试，以确定像有害拒绝这样的单一属性是否可以解释其得分。首先，多维项目反应理论建模强烈表明HarmBench并未测量单一属性。其次，差异项目功能分析发现，不同开发者的模型在相同拒绝能力得分下得分不同。这些标志在特定范围匹配下大多消失，这一模式与聚合效应一致，但不足以排除特定领域开发者差异的影响。放眼全局，HarmBench将不同的伤害行为合并为一个得分，而整体HELM安全汇总进一步将HarmBench和其他数据集的得分合并为一个单一的顶线数字。任何对数据集和项目进行平均的安全得分都可能隐藏饱和现象并以这种方式混淆行为。我们认为，在比较模型之前，得分应当获得其单一属性的解读。

View on arXiv Download PDF AI Translation

cs.AI / 123 / 2610.12436

Ecology of AI Agents: Collaboration Creates a Population Threshold for Takeoff

人工智能代理的生态学：协作创造了起飞的人口阈值

Crawley, Erin, Tanaka, Hidenori

Abstract

AI agents can now conduct real-world cyberattacks, scale up capabilities with the number of agents, and collectively pursue misaligned goals to obtain rewards. Together, these factors raise the risk of a population explosion of misaligned agents: agents could compromise computers and secretly deploy additional agents, creating a self-reinforcing cycle where larger populations develop greater collective cyber capability and expand further. This raises a fundamental question: What determines whether a population of misaligned agents remains contained or takes off into this self-reinforcing cycle? This population-level problem is ecological safety: unlike individual-agent or multi-agent safety with a fixed population, it concerns the dynamics of the population itself. Here, we develop an ecological theory of AI-agent populations based on a population growth equation in which fitness (growth rate) depends on cybersecurity capability. We show that, without collaboration, the population takes off only when individual-agent capability exceeds a critical threshold. With collaboration, however, collective cybersecurity capability increases with population size. This creates a critical population threshold: below it, the population declines; above it, the population takes off, even though individual-agent capability has not changed. In ecology, this phenomenon is known as the strong Allee effect. Because red teaming a small group of agents cannot guarantee ecological safety in larger populations, our theory calls for ecological red teaming and population pacing: gradually deploying larger agent populations in controlled environments, while measuring how cyber capability scales with population size, and estimating the critical population size for takeoff. Capability gains may lower this threshold, requiring re-estimation for each new model generation.

Chinese Translation

人工智能代理现在能够进行现实世界的网络攻击，随着代理数量的增加而提升能力，并共同追求不一致的目标以获得奖励。这些因素共同提高了不一致代理人口爆炸的风险：代理可能会妥协计算机并秘密部署额外的代理，形成一个自我强化的循环，其中更大的人口发展出更强的集体网络能力并进一步扩展。这提出了一个根本性的问题：是什么决定了不一致代理的人口是保持受控还是进入这个自我强化的循环？这个人口层面的问题是生态安全：与固定人口的单个代理或多代理安全不同，它涉及到人口本身的动态。在这里，我们基于一个人口增长方程发展了人工智能代理人口的生态理论，其中适应度（增长率）依赖于网络安全能力。我们表明，在没有协作的情况下，只有当单个代理的能力超过一个临界阈值时，人口才会起飞。然而，随着协作的进行，集体网络安全能力会随着人口规模的增加而提高。这创造了一个关键的人口阈值：低于该阈值，人口会下降；高于该阈值，人口会起飞，即使单个代理的能力没有改变。在生态学中，这种现象被称为强阿利效应。由于对小组代理进行红队测试无法保证在更大人口中的生态安全，我们的理论呼吁进行生态红队测试和人口节奏控制：在受控环境中逐步部署更大规模的代理人口，同时测量网络能力如何随人口规模变化，并估计起飞的关键人口规模。能力的提升可能会降低这一阈值，因此需要对每一代新模型进行重新评估。

View on arXiv Download PDF AI Translation

cs.AI / 124 / 2610.12452

BrickBench: Evaluating Agentic Brick Design

BrickBench：评估自主砖块设计

Kulits, Peter, Xu, Yiqing, Jones, R. Kenny, Schmid, Cordelia, Wu, Jiajun

Abstract

We propose BrickBench, a benchmark for agentic text-conditioned LEGO-set design. Given a prompt, an agent is tasked with producing an assembly that not only satisfies semantic and design criteria, but that can also be physically built. To do so, it must select parts from a discrete library and reason jointly about local and global constraints. We score validity, alignment, and design across three settings that vary in scale and part availability. We provide BrickAgent, an environment for coding agents to construct, inspect, and validate their designs. We find that leading agents largely satisfy verifiable physical and semantic requirements, but fall short of human designs. We release our benchmark and environment at http://www.brickben.ch

Chinese Translation

我们提出了BrickBench，这是一个用于自主文本条件下乐高套件设计的基准测试。给定一个提示，代理的任务是生成一个不仅满足语义和设计标准，而且可以实际构建的组装体。为此，它必须从一个离散的零件库中选择零件，并共同推理局部和全局约束。我们在三个不同规模和零件可用性设置下对有效性、一致性和设计进行评分。我们提供了BrickAgent，一个用于编码代理构建、检查和验证其设计的环境。我们的研究发现，领先的代理在很大程度上满足可验证的物理和语义要求，但仍未达到人类设计的水平。我们在http://www.brickben.ch发布了我们的基准和环境。

View on arXiv Download PDF AI Translation

cs.AI / 125 / 2610.12466

On the estimation and validity of AI time horizons---a statistical look at the METR plot

关于人工智能时间范围的估计与有效性——对METR图的统计分析

Nguyen, Drew T., Fithian, William

Abstract

METR's 50\% time horizon measures the human completion time of software tasks that an AI solves with 50\% probability, allowing AI capabilities to be expressed in interpretable units. On 228 tasks and 26 AIs, we recompute the time horizons using splines and item-response theory to relax the assumption that the AI difficulty of a task depends linearly on the log of human time. Our fitted spline can be interpreted as a function that \emph{converts} human time to AI difficulty; it is nearly flat in a region from 2--30 min but close to linear elsewhere. Hence, a time-horizon jump from 3 min to 30 min is much easier than one from 30 min to 5 hours despite the same multiplier of $10 \times$. Overall, we contribute time-horizon point estimates that perform better under a cross-validated suite of proper scoring rules, as well as diagnostic plots for assessing time horizons' construct validity. We suggest that time horizons be interpreted together with the diagnostic plots, especially as new time-horizon-based benchmarks are proposed or existing ones grow to include longer tasks.

Chinese Translation

METR的50\%时间范围测量了AI以50\%概率解决的软件任务的人类完成时间，从而使AI能力能够以可解释的单位表达。在228个任务和26个AI上，我们使用样条和项目反应理论重新计算时间范围，以放宽AI任务难度与人类时间对数之间线性依赖的假设。我们拟合的样条可以被解释为一个将人类时间“转换”为AI难度的函数；在2到30分钟的区域内几乎是平坦的，但在其他地方接近线性。因此，从3分钟到30分钟的时间范围跃迁要比从30分钟到5小时的跃迁容易得多，尽管它们的乘数都是$10 \times$。总体而言，我们贡献了在交叉验证的适当评分规则下表现更好的时间范围点估计，以及用于评估时间范围构念有效性的诊断图。我们建议将时间范围与诊断图一起解释，尤其是在提出新的基于时间范围的基准或现有基准扩展到更长任务时。

View on arXiv Download PDF AI Translation

计算语言学 (Computation and Language)

cs.CL / 1 / 2610.10550

Diffu-LoRA: A Novel Low-Rank Adaptation for Personalized Diffusion Models

Diffu-LoRA：一种用于个性化扩散模型的新型低秩适应方法

Li, Tianjing, Zhu, Wei

Abstract

Personalizing text-to-image diffusion models from a few reference images requires preserving subject identity while following prompts that describe new contexts. Full-model fine-tuning is parameter-intensive, whereas low-rank adaptation (LoRA) reduces the number of trainable parameters but leaves open how adaptation capacity should be distributed across layers. We introduce Diffu-LoRA, a parameter-efficient method that learns this allocation through gated low-rank adaptation. Diffu-LoRA inserts trainable low-rank components into the linear layers of Transformer blocks and assigns a learnable gate to each component. Bilevel optimization updates the adaptation weights and gate parameters on separate data splits, while progressive pruning removes components with the lowest gate values to meet a prescribed rank budget. This procedure allocates adaptation capacity nonuniformly across layers while keeping the pretrained backbone frozen. Experiments with Stable Diffusion on subjects from DreamBooth and additional collected datasets show improved overall subject fidelity and prompt alignment relative to the evaluated fine-tuning baselines. Ablation studies examine the contributions of bilevel optimization, progressive pruning, and adapter placement. These results support learned rank allocation as a practical approach to parameter-efficient diffusion model personalization.

Chinese Translation

从少量参考图像个性化文本到图像的扩散模型需要在遵循描述新上下文的提示的同时保持主体身份。全模型微调参数密集，而低秩适应（LoRA）减少了可训练参数的数量，但尚未明确如何在各层之间分配适应能力。我们提出了Diffu-LoRA，这是一种通过门控低秩适应学习这种分配的参数高效方法。Diffu-LoRA在Transformer块的线性层中插入可训练的低秩组件，并为每个组件分配一个可学习的门。双层优化在不同的数据分割上更新适应权重和门参数，而渐进式剪枝则移除具有最低门值的组件，以满足规定的秩预算。该过程在保持预训练主干不变的同时，在各层之间非均匀地分配适应能力。对来自DreamBooth和其他收集数据集的主体进行的Stable Diffusion实验显示，相较于评估的微调基线，整体主体保真度和提示对齐性有所改善。消融研究考察了双层优化、渐进式剪枝和适配器位置的贡献。这些结果支持学习的秩分配作为一种实用的参数高效扩散模型个性化方法。

View on arXiv Download PDF AI Translation

cs.CL / 2 / 2610.10592

Wieszcz-XIX: A 3.1-Billion-Word Corpus of Pre-1918 Polish and Temporally Bounded Language Models Trained From Scratch

Wieszcz-XIX：一个包含31亿词的1918年前波兰语语料库及从零开始训练的时间限制语言模型

Kocur, Szymon

Abstract

Historical Polish is well documented as a language but annotated in machine-readable form only to about a million words for the period this paper covers; the rest sits behind optical character recognition of variable quality. We present Wieszcz-XIX, a corpus of 6.75 billion tokens (about 3.1 billion words) in 294,369 documents, most of them periodical issues, of Polish published from 1800 to 1918, assembled from Wolne Lektury and the Internet Archive by a pipeline that filters, deduplicates, audits for post-1918 leakage and splits at the document level. It is over three orders of magnitude larger than the annotated corpus of the same period, and we quantify its defects: recognition corruption against a false-positive floor, near-identical duplication, which is removed, and post-1918 leakage, which is excluded from the training corpus itself down to a known residue of 0.04 to 0.38% of its bytes, found in the transcribed source, so the published corpus is the trained one document for document. On a hand-corrected sample the character error rate is 0.68% where the text is legible, and 45% of the sampled passages cannot be corrected. On it we train a ladder of decoder-only models from 47M to 349M parameters from scratch, and measure their temporal boundedness. Against two modern Polish base models, one far larger, the 349M shows a crossover, as does the 107M against the comparator of its size: post-1918 vocabulary costs them about 3.1 bits per byte more than period vocabulary, a gap the comparators do not show, and period vocabulary costs them fewer bits than it costs the comparators. Shown period text, the models keep its spelling and the comparators only partly. Adding parameters gains about twice as much as a second pass over the data. We release the corpus, code and weights. Content warning: the models reproduce period prejudice, including antisemitic statements.

Chinese Translation

历史波兰语作为一种语言有着良好的文献记录，但在机器可读形式下的注释仅覆盖约一百万词，涉及本论文所讨论的时期；其余部分则受到质量不一的光学字符识别的限制。我们提出了Wieszcz-XIX，这是一个包含67.5亿个标记（约31亿词）的语料库，涵盖294,369份文档，其中大多数为1800年至1918年间出版的波兰期刊，语料库由Wolne Lektury和互联网档案馆组装而成，采用了一条过滤、去重、审核后1918年泄漏并按文档级别拆分的管道。它的规模比同一时期的注释语料库大三个数量级，我们量化了其缺陷：识别错误与假阳性阈值的对比、近乎相同的重复（已被移除）以及1918年后的泄漏（已从训练语料库中排除，剩余已知为0.04%至0.38%的字节，存在于转录源中），因此发布的语料库是经过训练的文档逐一对应。在经过人工校正的样本中，文本可读部分的字符错误率为0.68%，而45%的抽样段落无法纠正。在此基础上，我们从零开始训练了一系列仅解码器模型，参数从4700万到3.49亿不等，并测量了它们的时间限制性。与两个现代波兰语基础模型相比，其中一个规模更大，3.49亿参数的模型显示出交叉点，107M参数的模型与其规模的对比者也表现出类似情况：1918年后的词汇使它们的成本比时期词汇多约3.1比特每字节，而对比者则没有显示出这种差距，且时期词汇的成本低于对比者的成本。面对时期文本，这些模型保持了其拼写，而对比者则仅部分保持。增加参数的收益约为对数据进行第二次处理的两倍。我们发布了语料库、代码和权重。内容警告：这些模型再现了时期偏见，包括反犹太主义言论。

View on arXiv Download PDF AI Translation

cs.CL / 3 / 2610.10650

Large Language Model-Assisted Preparation of Transportation Management Plans: A Case Study with WisDOT WisTMP System

大型语言模型辅助的交通管理计划准备：以WisDOT WisTMP系统为案例研究

Sheng, Zihao, Li, Pei, Huang, Zilin, Chen, Yen-Jung, Luo, Yuhao, Wan, Zhengyang, Parker, Steven T., Noyce, David A., Chen, Sikai

Abstract

Work zones are critical yet hazardous components of transportation infrastructure, requiring carefully designed Transportation Management Plans (TMPs) to ensure safety and mobility. However, TMP preparation remains labor-intensive and heavily dependent on practitioner expertise. This paper proposes a Large Language Model (LLM)-assisted framework to automate TMP content generation, leveraging the WisDOT WisTMP system as the application context. The framework fine-tunes multiple open-source LLMs across different model scales and deploys them locally to ensure data security. To support model training, we construct a domain-specific dataset from historical WisTMP documents by converting PDF files into structured question-answer pairs in JSON format. Experimental results show that fine-tuning significantly improves performance across standard text generation metrics. Further section-wise and strategy-level analyses reveal that, while LLMs achieve strong overall performance, they tend to over-generate strategies and struggle to produce project-specific justifications and accurate cost estimates. In addition, scaling from 7B/8B to 14B yields limited gains. These findings demonstrate the potential of LLMs to improve TMP preparation efficiency while highlighting remaining challenges in LLM-assisted TMP development. The source code and demo videos will be publicly available at https://zihaosheng.github.io/TMP-LLM/.

Chinese Translation

施工区域是交通基础设施中至关重要但又充满危险的组成部分，需要精心设计的交通管理计划（TMP）以确保安全和流动性。然而，TMP的准备仍然劳动密集，并且高度依赖从业人员的专业知识。本文提出了一种大型语言模型（LLM）辅助的框架，以自动化TMP内容生成，并以WisDOT WisTMP系统作为应用背景。该框架对多种开源LLM进行了微调，涵盖不同的模型规模，并在本地部署以确保数据安全。为了支持模型训练，我们从历史WisTMP文档中构建了一个特定领域的数据集，通过将PDF文件转换为JSON格式的结构化问答对。实验结果表明，微调显著提高了标准文本生成指标的性能。进一步的按章节和策略层面的分析显示，尽管LLM在整体性能上表现强劲，但它们往往会过度生成策略，并且在生成项目特定的理由和准确的成本估算方面存在困难。此外，从7B/8B扩展到14B的收益有限。这些发现展示了LLM在提高TMP准备效率方面的潜力，同时突显了LLM辅助TMP开发中仍然存在的挑战。源代码和演示视频将公开发布在https://zihaosheng.github.io/TMP-LLM/。

View on arXiv Download PDF AI Translation

cs.CL / 4 / 2610.10724

Cognitive Thermometers: Machine Learning and Logical Complexity

认知温度计：机器学习与逻辑复杂性

Steinert-Threlkeld, Shane, Szymanik, Jakub

Abstract

How does the human mind represent semantic categories? Why do natural languages favor certain meanings over others? Prior explanations have relied on logical definability and complexity, but these are highly sensitive to the choice of logical language, rendering some design choices unmotivated. In this article, we propose that machine learning provides a somewhat more agnostic approach to measuring semantic complexity. We review emerging evidence that logic and machine learning often yield converging results on relative complexity and its resulting effects in semantic typology. Where they diverge, learning appears to be a better explanation than logical complexity. We argue that treating machine learning models as ``cognitive thermometers'' enables a unified approach to complexity that bridges symbolic logic and connectionist AI.

Chinese Translation

人类思维是如何表征语义类别的？为什么自然语言偏向某些意义而非其他？先前的解释依赖于逻辑可定义性和复杂性，但这些对逻辑语言的选择高度敏感，使得某些设计选择缺乏合理性。本文提出，机器学习提供了一种相对更为中立的方法来测量语义复杂性。我们回顾了新兴证据，表明逻辑与机器学习在相对复杂性及其在语义类型学中的影响方面常常得出一致的结果。在它们出现分歧的地方，学习似乎比逻辑复杂性提供了更好的解释。我们认为，将机器学习模型视为“认知温度计”能够实现一种统一的复杂性方法，连接符号逻辑与联结主义人工智能。

View on arXiv Download PDF AI Translation

cs.CL / 5 / 2610.10738

Lossy Compressive Text Autoencoders

有损压缩文本自编码器

Sabolčec, Vinko, Katharopoulos, Angelos, Grangier, David

Abstract

Our work explores learning a compressed latent representation of text, at the intersection of data compression and representation learning. We propose an autoencoder architecture that performs residual downscaling and upscaling of hidden representations along the time axis, with a residual low-dimension discrete bottleneck. We analyze our approach for different quantization methods, training objectives, and datasets. For different levels of compression, we evaluate the similarity between the original and reconstructed text both at the surface-level (BLEU) and at the semantic-level (LLM-based judge). Additionally, we evaluate our models on downstream question-answering and semantic text similarity benchmarks. Our approach results in compressed representations which are on par with lossless text compression algorithms at 2.24 bits per byte on web text data, while having good reconstruction and downstream task performance.

Chinese Translation

我们的工作探讨了在数据压缩和表示学习交叉领域中学习文本的压缩潜在表示。我们提出了一种自编码器架构，该架构沿时间轴对隐藏表示进行残差下采样和上采样，并具有残差低维离散瓶颈。我们分析了不同量化方法、训练目标和数据集下我们方法的表现。对于不同的压缩水平，我们评估了原始文本与重构文本之间的相似性，既在表面层面（BLEU），也在语义层面（基于LLM的评估）。此外，我们还在下游问答和语义文本相似性基准上评估了我们的模型。我们的研究结果显示，在网络文本数据上，我们的压缩表示在每字节2.24比特的情况下，与无损文本压缩算法相当，同时在重构和下游任务性能上表现良好。

View on arXiv Download PDF AI Translation

cs.CL / 6 / 2610.10758

Clarify, Then Focus: Statement Normalization for Conversation Analytics at Scale

先澄清，再聚焦：面向大规模对话分析的话语规范化方法

Arbuzov, Mikhail L., Dave, Karan, Dontsova, Evgeniya, Hu, Yaodong, Lao, Vincent, Jain, Navita, Bei, Sisong, Dimov, Dmitry

Abstract

Enterprise conversation analytics asks many questions of millions of interactions. Each question can require reconstructing what people mean and identifying which information matters, repeating costly interpretive work across the same transcripts. We propose a simple principle: clarify the text, then focus the reader. Statement normalization transforms dialogue into short, speaker-attributed statements with source references and semantic tags. The statements make meaning more explicit; the tags support selecting evidence for a particular question. Downstream models can use the full representation or a relevant subset, depending on what helps them make the decision. In an offer-suppression task on customer-service calls, normalization improves a supervised classifier without selection, while weaker prompted readers benefit from both normalization and selection. A small model can learn the normalization contract, while lightweight encoders handle tagging and downstream decisions. Sharing this preparation across questions supports an inference pipeline built entirely from small models, making analytics over millions of conversations substantially less expensive.

Chinese Translation

企业对话分析需要对数百万次交互提出诸多问题。每个问题可能都需要重构参与者的意图并识别关键信息，这导致在相同的对话记录中反复进行高成本的解释性工作。我们提出一个简单原则：先澄清文本，再引导读者聚焦。话语规范化将对话转换为简短的、带有说话者归属、来源引用和语义标签的陈述。这些陈述使意义更加明确；标签则支持针对特定问题选择证据。下游模型可以根据决策需求，使用完整表示或相关子集。在客户服务电话的报价抑制任务中，规范化在无选择机制的监督分类器中提升了性能，而较弱的提示式阅读器则同时受益于规范化和选择。小型模型能够学习规范化协议，而轻量级编码器负责标签处理和下游决策。跨问题共享这种预处理支持了完全由小型模型构建的推理流水线，从而显著降低了对数百万次对话进行分析的成本。

View on arXiv Download PDF AI Translation

cs.CL / 7 / 2610.10827

Grammar Concept Annotation at Scale: Deployed Fine-Tuned Small Language Models Outperform Prompted Frontier Models

大规模语法概念标注：部署的微调小型语言模型优于提示式前沿模型

Celikik, Marjan, Ramallo, Ana Peleteiro, Morales, Javier

Abstract

Corrective feedback is among the best-evidenced drivers of second-language acquisition, yet corrections delivered during lessons rarely accumulate into an actionable view of grammar mastery. Prompted frontier models can provide such a view from learner--tutor lesson transcripts, but they are costly at scale. We close this gap by fine-tuning Qwen3.5 small language models (SLMs) on filtered and rebalanced teacher-generated supervision, then deploying an efficient 0.8B model in an end-to-end grammar mastery tracker for all English learners on our platform. Internalizing the annotation contract into adapter weights enables pairing the 0.8B model with a compact matched prompt rather than verbose instructions. On two human-curated benchmarks, both the deployed 0.8B model and a 4B reference comparator outperform prompted GPT-5.4 and GPT-5.6 Sol in precision and recall under nested matching criteria of increasing strictness: concept, evidence span, and correctness. The deployed 0.8B SLM reduces serving cost by approximately 16$\times$. A feature-level online experiment shows significant gains in learner engagement ($+15.8\%$) and key business metrics, including scheduled hours ($+2.1\%$) and GMV from new lessons ($+13.2\%$).

Chinese Translation

纠正性反馈是第二语言习得中证据最充分的驱动因素之一，但在课堂上提供的纠正意见往往无法汇聚成对语法掌握的可操作视图。提示式前沿模型可以从学习者与导师的课堂记录中提供这样的视图，但在大规模应用时成本较高。我们通过在经过筛选和重新平衡的教师生成监督数据上微调Qwen3.5小型语言模型（SLMs），填补了这一空白，然后在我们的平台上为所有英语学习者部署一个高效的0.8B模型，作为端到端的语法掌握跟踪器。将注释契约内化为适配器权重，使得0.8B模型可以与紧凑的匹配提示配对，而不是冗长的指令。在两个人工策划的基准测试中，部署的0.8B模型和一个4B参考比较模型在精确度和召回率上均优于提示式的GPT-5.4和GPT-5.6 Sol，尤其是在逐渐严格的嵌套匹配标准下：概念、证据范围和正确性。部署的0.8B SLM将服务成本降低了约16倍。特征级在线实验显示学习者参与度显著提升（+15.8%），以及关键业务指标的改善，包括计划课程时长（+2.1%）和新课程的GMV（+13.2%）。

View on arXiv Download PDF AI Translation

cs.CL / 8 / 2610.10845

Real Long-Term Memory for AI: A 50-Million-Token Window That Is Faster and Cheaper Than Recompute

真实的长期记忆用于人工智能：一个比重新计算更快且更便宜的5000万标记窗口

Schelpe, Sietse

Abstract

A large language model can only use the text that fits in its context window, and it recomputes its internal key-value (KV) state for a prompt every time the prompt is sent. We test a memory layer, the public package galahad-kv, that saves the KV state of each block of about 16,000 tokens to encrypted local NVMe disk and loads it back later, byte-exact, without recomputing it. We ran it on 50,000,000 tokens of real public text, served through vLLM on one NVIDIA H100, with Gemma 4 12B and Gemma 4 31B. Every block we probed was loaded back from the encrypted store with no recompute (100 of 100, at depths from 0 to 50M tokens) on both models. Loading a block was 2.8x to 4.3x faster than recomputing it and used 8.8x to 12.3x less GPU energy, and GPU memory stayed flat over the whole 50M-token stream. Asked about facts planted millions of tokens earlier, the 12B model gave the right answer 82 times out of 100 and the 31B model 98 times out of 100. Neither model made up an answer. The limits are as follows. This is reuse of stored state, not a wider attention window: one block is loaded at a time, and how well a question is answered depends on the model. Writing the memory is a one-time cost, and the store takes terabytes of local NVMe disk. We describe the test protocol, which is built to resist common ways of gaming long-context benchmarks, and give a single-GPU reproduction that uses public software and a free licence for the package.

Chinese Translation

大型语言模型只能使用适合其上下文窗口的文本，并且每次发送提示时都会重新计算其内部的键值（KV）状态。我们测试了一种内存层，公共包galahad-kv，它将每个约16,000个标记的块的KV状态保存到加密的本地NVMe磁盘中，并在稍后以字节精确的方式加载，而无需重新计算。我们在5000万标记的真实公共文本上进行了测试，通过vLLM在一台NVIDIA H100上运行，使用Gemma 4 12B和Gemma 4 31B。我们探测的每个块都是从加密存储中加载的，无需重新计算（100个中的100个，深度从0到5000万标记）在这两个模型上。加载一个块的速度比重新计算快2.8到4.3倍，并且使用的GPU能量少了8.8到12.3倍，整个5000万标记流中的GPU内存保持平稳。对于数百万标记之前植入的事实，12B模型在100次中正确回答了82次，31B模型正确回答了98次。两个模型都没有编造答案。其限制如下。这是对存储状态的重用，而不是更广泛的注意窗口：一次只加载一个块，问题的回答质量取决于模型。写入内存是一次性成本，存储需要数TB的本地NVMe磁盘。我们描述了测试协议，该协议旨在抵抗常见的操纵长期上下文基准的方法，并提供了一个使用公共软件和免费许可证的单GPU重现。

View on arXiv Download PDF AI Translation

cs.CL / 9 / 2610.10865

Disentangling Linguistic and Paralinguistic Information with Routed Sparse Autoencoders

通过路由稀疏自编码器解构语言和副语言信息

Guta, Beimnet Bekele, Yang, Xiaoyu, Sun, Guangzhi, Woodland, Philip C.

Abstract

Self-supervised speech encoders contain linguistic and paralinguistic information in a shared, entangled representation space. We combine a TopK sparse autoencoder with route-specific supervision and cross-factor adversaries. Across frozen SPEAR and WavLM encoders, independent probes show factor-specific retention and suppression: linguistic information remains stronger in the linguistic route, while paralinguistic factors, including speaker identity, emotion, and prosody, are retained in the paralinguistic route and substantially reduced in the linguistic route. The route organisation learned on LibriSpeech persists on MSP-Podcast without representation-side retraining. Feature-space route interventions further transfer the swapped factor while largely preserving the information carried by the unchanged route. These results show consistent route-selective separation across encoders, corpora, independent probes, and representation-level interventions.

Chinese Translation

自监督语音编码器在共享的纠缠表示空间中包含语言和副语言信息。我们结合了 TopK 稀疏自编码器与特定路由的监督和交叉因子对抗。在冻结的 SPEAR 和 WavLM 编码器中，独立探针显示出因子特定的保留和抑制：语言信息在语言路由中保持更强，而副语言因素，包括说话者身份、情感和韵律，则在副语言路由中被保留，并在语言路由中大幅减少。在 LibriSpeech 上学习到的路由组织在 MSP-Podcast 上持续存在，而无需在表示侧进行再训练。特征空间的路由干预进一步转移了交换的因子，同时在很大程度上保留了未更改路由所携带的信息。这些结果显示出在编码器、语料库、独立探针和表示级干预之间的一致路由选择性分离。

View on arXiv Download PDF AI Translation

cs.CL / 10 / 2610.10871

Sparse Attention Is Matrix Approximation, Not Choosing from a Bag of Values

稀疏注意力是矩阵近似，而不是从值的集合中选择

Wan, Fang, Liu, Xufeng, Li, Fan, Liu, Yi

Abstract

Large Language Models (LLMs) achieve strong performance across many domains, but their efficiency is limited by the quadratic cost of attention with respect to prompt length. Sparse attention reduces this cost by retaining only a small fraction of query-key interactions to approximate the full attention matrix. However, existing methods are trapped in a mathematically wrong view: they simply keep large scalar entries or high-mass regions of the attention matrix. This treats the attention matrix as a bag of values, ignoring that it is used as a structured matrix whose entries jointly determine the attention output through multiplication with value vectors. We argue that this is the core conceptual issue: sparse attention should be formulated as matrix approximation, not as blindly choosing the largest values from a bag of entries. Based on this view, we propose Matrix Approximation Sparse Attention (MASA). MASA replaces raw attention-mass ranking with a closed-form score that measures how much each sparse unit reduces matrix-product approximation error. As a theory-grounded plug-in correction, MASA can be added to existing sparse attention frameworks without changing their sparse kernels or budgets. Extensive experiments across multiple sparse attention methods, benchmarks, and LLM backbones show consistent accuracy gains, supporting both MASA and the matrix-approximation view of sparse attention.

Chinese Translation

大型语言模型（LLMs）在多个领域表现出色，但其效率受到与提示长度相关的注意力的平方成本的限制。稀疏注意力通过仅保留一小部分查询-键交互来减少这一成本，从而近似完整的注意力矩阵。然而，现有方法陷入了一个数学上错误的观点：它们仅仅保留注意力矩阵中的大标量条目或高质量区域。这将注意力矩阵视为一个值的集合，忽视了它作为一个结构化矩阵的作用，其条目通过与值向量的乘法共同决定注意力输出。我们认为这是核心概念问题：稀疏注意力应被表述为矩阵近似，而不是盲目选择值集合中的最大值。基于这一观点，我们提出了矩阵近似稀疏注意力（Matrix Approximation Sparse Attention, MASA）。MASA用一个闭式评分替代原始的注意力质量排名，该评分衡量每个稀疏单元减少矩阵乘积近似误差的程度。作为一个基于理论的插件修正，MASA可以在不改变现有稀疏注意力框架的稀疏内核或预算的情况下添加。针对多种稀疏注意力方法、基准测试和LLM骨干的广泛实验显示了一致的准确性提升，支持MASA及稀疏注意力的矩阵近似观点。

View on arXiv Download PDF AI Translation

cs.CL / 11 / 2610.10878

Stochastic Teacher Intervention for Agentic On-Policy Distillation

随机教师干预的代理性在线蒸馏

Liu, Junnan, Luo, Linhao, Chen, Zhijun, Mao, Qianren, Vu, Thuy-Trang, Haffari, Gholamreza

Abstract

On-policy distillation (OPD) efficiently transfers capabilities from a stronger teacher to a student language model through dense token-level supervision on student-generated rollouts and has shown promise on complex tasks such as mathematical reasoning. However, in multi-turn agentic tasks, student decisions shape subsequent observations, causing early errors to accumulate across turns. The resulting trajectories can drift away from the teacher's rollout distribution, making the teacher's token-level supervision less reliable or even counterproductive for OPD training. To address this issue, we introduce STI-OPD, a stochastic teacher intervention framework for multi-turn agentic OPD. During multi-turn interaction, STI-OPD uses teacher intervention guided by teacher-student policy discrepancy to replace the student's proposed action with a teacher-generated one to maximize the acquisition of reliable supervision. We further develop a stochastic intervention strategy, addressing the limitations of previous threshold-based or fixed-schedule approaches, that estimates policy discrepancy using KL divergence and maps it to an intervention probability. By sampling whether to intervene from this probability, STI-OPD adaptively balances teacher control with student exploration. To learn from the resulting mixed-policy trajectories, we introduce an Importance-Weighted Reverse KL objective that corrects the token sampling mismatch between teacher-generated responses and the student policy to preserve the original OPD objective. Across tool-integrated reasoning and long-horizon interaction, STI-OPD outperforms the strongest prior OPD baseline on every evaluated benchmark and student size. Ablations further show that both discrepancy-guided intervention and importance weighting contribute to these gains.

Chinese Translation

在线蒸馏（OPD）通过对学生生成的回合进行密集的标记级监督，有效地将能力从更强的教师转移到学生语言模型，并在数学推理等复杂任务中展现出良好的前景。然而，在多轮代理性任务中，学生的决策会影响后续观察，导致早期错误在多个回合中累积。由此产生的轨迹可能会偏离教师的回合分布，使得教师的标记级监督在OPD训练中变得不可靠甚至适得其反。为了解决这一问题，我们提出了STI-OPD，一个用于多轮代理性OPD的随机教师干预框架。在多轮交互过程中，STI-OPD利用教师与学生策略之间的差异指导教师干预，以用教师生成的动作替代学生提出的动作，从而最大化可靠监督的获取。我们进一步开发了一种随机干预策略，解决了之前基于阈值或固定时间表方法的局限性，该策略通过KL散度估计策略差异，并将其映射到干预概率上。通过从该概率中抽样决定是否干预，STI-OPD自适应地平衡了教师控制与学生探索。为了从生成的混合策略轨迹中学习，我们引入了一种重要性加权反向KL目标，纠正教师生成的响应与学生策略之间的标记采样不匹配，以保持原始OPD目标。在工具集成推理和长时间交互中，STI-OPD在每个评估基准和学生规模上均优于最强的先前OPD基线。消融实验进一步表明，差异指导的干预和重要性加权均对这些提升有所贡献。

View on arXiv Download PDF AI Translation

cs.CL / 12 / 2610.10918

Large Language Models for Machine Translation Quality Annotation: Humans and Models Are Both Challenged

大型语言模型在机器翻译质量标注中的应用：人类与模型均面临挑战

Almaghout, Hala, Federmann, Christian, Gao, Qin

Abstract

Large Language Models (LLMs) are considered to be a more efficient and cost-effective alternative to human judgment for Machine Translation (MT) evaluation. With MT evaluation spanning a large number of language pairs, domains and levels of annotation granularity, LLMs must be thoroughly evaluated across these dimensions before being reliably used as alternatives to human evaluation. In this paper, we evaluate the performance of LLMs for two prominent MT quality evaluation schemes: Multidimensional Quality Metrics (MQM) and Error Span Annotation (ESA) by comparing their agreement with human annotators. We present results on a long-context test set of 70 language pairs and the publicly available WMT23 and WMT25 data, investigating both score and error span annotation agreement across a variety of language pairs and domains. Our results show that while LLM agreement with human annotators exceeds agreement between human annotators for some evaluation tasks, both vary substantially across annotation schemes, language pairs and domains and remain unreliable for most settings. Furthermore, we identify challenges facing both human and LLM annotators: humans are particularly challenged by fine-grained MQM annotations and low-resource language pairs, while LLMs struggle with minor errors, wrong language variants and error span annotation. Our results highlight the potential for improvement for both human and LLM annotation performance, possibly through human-LLM collaborative annotation pipelines that address the reliability issues identified in this work.

Chinese Translation

大型语言模型（LLMs）被认为是机器翻译（MT）评估中比人类判断更高效且成本更低的替代方案。由于机器翻译评估涉及大量语言对、领域和标注粒度，LLMs必须在这些维度上进行全面评估，才能可靠地作为人类评估的替代工具。本文评估了LLMs在两个主要机器翻译质量评估方案中的表现：多维质量指标（MQM）和错误跨度标注（ESA），通过比较其与人类标注者的一致性。我们在70个语言对的长上下文测试集以及公开的WMT23和WMT25数据上呈现结果，研究了各种语言对和领域中的得分和错误跨度标注的一致性。我们的结果表明，尽管LLMs与人类标注者的一致性在某些评估任务中超过了人类标注者之间的一致性，但两者在标注方案、语言对和领域之间的差异仍然很大，并且在大多数设置中仍然不可靠。此外，我们识别了人类和LLMs标注者面临的挑战：人类在细粒度MQM标注和低资源语言对方面特别困难，而LLMs在处理小错误、错误语言变体和错误跨度标注时表现不佳。我们的结果突显了人类和LLMs标注性能改进的潜力，可能通过人类-LLM协作标注流程来解决本研究中识别的可靠性问题。

View on arXiv Download PDF AI Translation

cs.CL / 13 / 2610.10946

AI4Fire: Evaluating Large Language Models on Wildfire Tasks

AI4Fire：评估大型语言模型在野火任务中的表现

Zhao, Yue, Hu, Xiyang, Xiong, Zuobin, Wang, Zhangyu, Li, Ruolin

Abstract

Large language models (LLMs) are entering wildfire management, where overstated evaluations can cost property and lives. How do they perform on wildfire tasks, with and without grounding? Bare means a model receives the task input alone. Grounded means it also receives one task-specific addition: for smoke detection, a smoke-free reference frame from the same camera. AI4Fire runs six core models bare and grounded on five wildfire tasks, zero-shot; a sweep adds 29 more. Our literature search on fire tasks found 138 works; none combines this roster, task coverage, and paired bare and grounded runs. We report three findings. (1) Grounding helped most where the addition carried the answer: a read-only SQL tool lifted every core model's database accuracy from at most 16 to at least 88 percent. (2) Simple rules were hard to beat: no core model outperformed repeating today's staffing count, and two open-weight models mostly copied the median of similar earlier fire-days, a worse forecast. (3) Public releases carry hazards: a fire-danger column separates the holdout perfectly, and 67 aerial fire frames carry smoldering or fire-free labels read from a clipped thermal maximum. We release prompts, responses, scores, code, and the survey record.

Chinese Translation

大型语言模型（LLMs）正在进入野火管理领域，其中夸大的评估可能会导致财产和生命的损失。它们在野火任务中的表现如何，尤其是在有和没有基础信息的情况下？“裸”意味着模型仅接收任务输入，而“有基础”则意味着它还接收一个特定于任务的附加信息：例如，对于烟雾检测，来自同一摄像头的无烟参考帧。AI4Fire 在五个野火任务上对六个核心模型进行了裸和有基础的零样本测试；另外增加了29个任务的测试。我们对火灾任务的文献检索发现了138篇相关研究；但没有任何研究结合了这一名单、任务覆盖范围以及配对的裸和有基础的测试结果。我们报告了三项发现。(1) 在大多数情况下，基础信息的提供有助于提高准确性：一个只读的 SQL 工具使每个核心模型的数据库准确率从最多16%提升至至少88%。(2) 简单的规则很难被超越：没有核心模型的表现超过重复今天的人员统计，而两个开放权重模型大多复制了类似早期火灾日的中位数，导致更差的预测结果。(3) 公开发布存在风险：一个火灾危险列完美地区分了保留样本，67个空中火灾帧带有从剪裁的热最大值读取的冒烟或无火标签。我们发布了提示、响应、得分、代码和调查记录。

View on arXiv Download PDF AI Translation

cs.CL / 14 / 2610.10971

When Citations Mislead? A Claim-Level Benchmark for Legal Hallucination Detection

何时引用会误导？法律幻觉检测的声明级基准

Demir, M. Mikail, Canbaz, M. Abdullah

Abstract

Large language models are increasingly used in legal research and drafting, but they can still produce claims that sound convincing without being supported by the cited source. We introduce PARCEL, a benchmark for checking whether a legal claim is supported by the underlying authority. Using recent New York State Court of Appeals decisions, we build a dataset of 3,396 parenthetical-style claims labeled as Supported, Refuted, or Not Found. We cast this task as a three-way natural language inference problem and evaluate several state-of-the-art LLMs in a zero-shot setting. Although the strongest models reach up to 0.97 accuracy, the results also show an important weakness: models still incorrectly mark unsupported claims as supported, even when the full opinion text is provided. Across models, missing support is harder to detect than direct contradiction, and fabricated but plausible citations cause the largest drop in performance. Overall, PARCEL provides a practical benchmark for testing claim-level groundedness in legal RAG systems.

Chinese Translation

大型语言模型在法律研究和起草中越来越多地被使用，但它们仍然可能产生听起来令人信服但并未得到引用来源支持的声明。我们引入了PARCEL，一个用于检查法律声明是否得到基础权威支持的基准。利用最近的纽约州上诉法院裁决，我们构建了一个包含3,396个括号式声明的数据集，并标记为支持、反驳或未找到。我们将此任务视为一个三方自然语言推理问题，并在零样本设置中评估了几种最先进的LLM。尽管最强模型的准确率高达0.97，但结果也显示出一个重要的弱点：即使提供了完整的意见文本，模型仍然错误地将不支持的声明标记为支持。在各模型中，缺失支持的检测比直接矛盾更困难，而虚构但看似合理的引用导致性能下降最大。总体而言，PARCEL为测试法律RAG系统中的声明级基础性提供了一个实用的基准。

View on arXiv Download PDF AI Translation

cs.CL / 15 / 2610.10988

Back in Style: A Sociolinguistic Approach to Authoring and Measuring Persona Fidelity in User Simulation

回归风格：一种社会语言学方法用于用户模拟中的角色忠实度的创作与测量

Konnelly, Lex, Khasanova, Elena, Wang, Riqiang, Lee, Matthias, Saini, Harsh, Kavehzadeh, Parsa

Abstract

As agentic systems gain commercial popularity, user simulators increasingly serve as measurement instrument for their evaluation. However, the fidelity of simulated users in comparison to real human users is generally low, and typically assessed by costly, subjective LLM judges. In this pilot study, we ask whether fidelity can instead be measured deterministically by treating a user persona sociolinguistically: as a social type that emerges from observable linguistic style, rather than one predicted by labels or descriptions a model must extrapolate into behaviour. We author personas as concrete stylistic rates, which lets us transfer two established, model-free instruments -- authorship-verification stylometry and lexicon-based content analysis -- as fidelity diagnostics. We A/B-test the sociolinguistic schema against a flat descriptive baseline across five task-oriented customer-service agents. Results show that the sociolinguistic schema improves both stylistic adherence and stylometric distinguishability for most of the tested models, with a caveat that persona style fidelity does not necessarily equal persona "naturalness". We argue that a sociolinguistic approach to persona design is a promising path towards more diverse and representative user personas, and that these metrics are most valuable in an error-attribution analysis, localizing where fidelity breaks down. This is a first step towards interventions that move user simulations closer to faithful renderings of diverse and variable linguistic outputs.

Chinese Translation

随着自主系统在商业上的普及，用户模拟器越来越多地作为评估这些系统的测量工具。然而，模拟用户与真实人类用户之间的忠实度通常较低，且通常由昂贵的主观大型语言模型（LLM）评审进行评估。在这项初步研究中，我们探讨是否可以通过社会语言学的视角确定性地测量忠实度：将用户角色视为一种从可观察的语言风格中产生的社会类型，而不是通过模型必须推断为行为的标签或描述来预测的类型。我们将角色创作为具体的风格比率，这使我们能够将两种已建立的无模型工具——作者验证风格分析和基于词汇的内容分析——转化为忠实度诊断。我们对五个任务导向的客户服务代理进行了社会语言学框架与平面描述基线的A/B测试。结果表明，社会语言学框架在大多数测试模型中提高了风格遵循性和风格分析的可区分性，但有一个警告，即角色风格的忠实度不一定等同于角色的“自然性”。我们认为，社会语言学方法在角色设计中的应用是朝着更具多样性和代表性的用户角色迈出的有希望的一步，并且这些指标在错误归因分析中最为有价值，能够定位忠实度失效的地方。这是向使用户模拟更接近多样化和可变语言输出的忠实呈现迈出的第一步。

View on arXiv Download PDF AI Translation

cs.CL / 16 / 2610.11015

Prompts versus Rules: Auditing and Controlling Speech Naturalness Behaviors in Voice User Simulators

提示与规则：语音用户模拟器中的语音自然性行为审计与控制

Wang, Riqiang, Khasanova, Elena, Saini, Harsh, Konnelly, Lex, Kavehzadeh, Parsa, Lee, Matthias, Attia, Mohamed

Abstract

As voice agents gain more popularity commercially, the user simulators used to evaluate the deployed agents are also being developed to include more realistic, variable, and diverse speech naturalness behaviors -- disfluency, interruption and backchanneling. The quality of the user simulator directly affects the validity of agent evaluation results. However, we find that most studies so far have not examined in detail whether the intended configuration for these behaviors is realized in the simulation. In this study, we audit the realized naturalness behaviors of tau-Voice, our own LLM-based prompting approach across three models, and our rule-based injection algorithm for disfluency, interruption, and backchanneling. We find that prompting for these behaviors is unreliable and produces speech inconsistent with the instructions, placed and distributed less naturally than the instruction implies. In contrast, our rule-based, model-free algorithm produces controllable and diverse naturalness behaviors more aligned with natural speech. Our results suggest that LLMs not purpose-trained for user simulation are not sufficient on their own to represent authentic user behavior, and that linguistically informed deterministic approaches or specialized models are needed to close the gap; auditing and reporting realized naturalness behaviors, rather than configured settings, is what makes that gap visible.

Chinese Translation

随着语音代理在商业上的日益普及，用于评估已部署代理的用户模拟器也在不断发展，以包含更真实、多变和多样的语音自然性行为——不流畅、打断和反馈。用户模拟器的质量直接影响代理评估结果的有效性。然而，我们发现迄今为止大多数研究并未详细检查这些行为的预期配置是否在模拟中得以实现。在本研究中，我们审计了我们的 LLM（大型语言模型）基础提示方法 tau-Voice 在三个模型中的自然性行为实现情况，以及我们针对不流畅、打断和反馈的基于规则的注入算法。我们发现，针对这些行为的提示不可靠，产生的语音与指令不一致，且分布和放置不如指令所暗示的自然。相比之下，我们的基于规则的无模型算法能够产生可控且多样的自然性行为，更加符合自然语音。我们的结果表明，未专门训练用于用户模拟的 LLM 本身不足以代表真实的用户行为，因此需要语言学上有依据的确定性方法或专门模型来填补这一差距；审计和报告实现的自然性行为，而不是配置设置，才使这一差距显现。

View on arXiv Download PDF AI Translation

cs.CL / 17 / 2610.11033

FedAlphaEdit: Null-Space-Aligned Merging for Collaborative Knowledge Editing

FedAlphaEdit：协作知识编辑的零空间对齐合并

Sugawara, Sota, Okada, Yukihiko

Abstract

Multiple institutions may each hold their own private knowledge edits and wish to integrate them into a single large language model without sharing raw edit requests. Null-space-constrained editing methods such as AlphaEdit mathematically guarantee that each update leaves unrelated knowledge intact, while collaborative frameworks such as CollabEdit aggregate edits from multiple clients without data sharing. Combining the two appears trivial. However, we show that this naive combination fails structurally, and we identify its cause. Guided by this analysis, we propose FedAlphaEdit. To our knowledge, this is the first collaborative knowledge editing framework that aligns both local editing and the server-side merging rule under a single null-space principle for preserving existing knowledge. FedAlphaEdit builds on null-space-aligned merging, in which clients share projected statistics and the server provably recovers the result of editing everything in one place under a one-shot idealization. Empirically, the proposed method repairs the collapse and brings edit success and preservation simultaneously close to the level of centralized editing across two architecture families. FedAlphaEdit thus lets institutions that cannot share raw edit data, such as hospitals and financial firms, jointly maintain a shared model that closely approximates editing all facts in one place.

Chinese Translation

多个机构可能各自持有私有的知识编辑，并希望在不共享原始编辑请求的情况下将其整合到一个大型语言模型中。受零空间约束的编辑方法，如 AlphaEdit，数学上保证每次更新不会影响无关的知识，而协作框架如 CollabEdit 则在不共享数据的情况下聚合来自多个客户端的编辑。将这两者结合似乎是微不足道的。然而，我们展示了这种简单的结合在结构上是失败的，并识别了其原因。在这一分析的指导下，我们提出了 FedAlphaEdit。据我们所知，这是第一个在单一零空间原则下对齐本地编辑和服务器端合并规则的协作知识编辑框架，以保护现有知识。FedAlphaEdit 基于零空间对齐合并，客户端共享投影统计数据，服务器则可以在一次理想化的情况下证明性地恢复所有编辑的结果。实证结果表明，该方法修复了崩溃，并使编辑成功率和知识保护同时接近于两个架构系列中的集中编辑水平。因此，FedAlphaEdit 使得无法共享原始编辑数据的机构，如医院和金融公司，能够共同维护一个共享模型，该模型与在一个地方编辑所有事实的结果非常接近。

View on arXiv Download PDF AI Translation

cs.CL / 18 / 2610.11069

Clinician use of language models diverges from how the models are evaluated

临床医生对语言模型的使用与模型评估方式存在差异

Vishwanath, Krithik, Lin, Haitong, Alyakin, Anton, Lee, Jin Vivian, Hewitt, D. Brock, Yao, Jie J., Small, William Robert, Khan, Hammad A., Orillac, Cordelia, Varma, Aakaash, Ye, Brandon, Alber, Daniel Alexander, Stolovitzky, Gustavo, Wiesenfeld, Batia, Nov, Oded, Wu, Wei, Zhang, Kang, Aphinyanaphongs, Yindalon, Requarth, Tim, Oermann, Eric Karl, Healthcare, The International Digital Twin Consortium in, Medicine

Abstract

Large language model (LLM) assistants are being deployed to clinicians across health systems, and judgments about their readiness rest largely on benchmark scores, most of them derived from examination questions or curated cases. A benchmark predicts performance in deployment only to the extent that its items resemble real use, yet whether benchmarks reflect the work these systems receive has rarely been measured. Here we analyze 127,833 queries sent by 6,342 physicians, advanced practice providers and nurses in 35 specialties to an institutional assistant during an eight-month roll-out. We characterize each query with RCQ-Map, a clinician-validated framework grounded in taxonomies of clinical questions and of LLM evaluation, which records its task, intent, answerability, missing information and potential harm. Documentation and administration (36.2%) and knowledge retrieval (28.9%) made up nearly two-thirds of use, and diagnosis 3.7%; more than a third of queries could not be answered well as posed. Applying RCQ-Map to 58 public benchmarks drawn from major evaluation suites and frontier model reports, which we assemble into the Clinical AI Benchmark Atlas, showed that the median benchmark contained no documentation requests and shared 31% of the task mix of real use, less than an even spread across task categories would. Benchmarks in suites designed to resemble clinical practice were individually no closer to real use than those used in frontier model reports. Benchmark scores therefore say little about how clinical AI performs on most of the work it is actually given, and evaluation should be matched to real clinical use.

Chinese Translation

大型语言模型（LLM）助手正在被部署到各个健康系统的临床医生中，而对其准备程度的判断主要依赖于基准得分，这些得分大多来源于考试问题或精心策划的案例。基准预测在实际应用中的表现仅在于其项目与真实使用的相似程度，但基准是否反映了这些系统所接收的工作却很少被测量。在此，我们分析了在八个月的推广期间，由6342名医生、先进实践提供者和护士在35个专业向一个机构助手发送的127,833个查询。我们使用RCQ-Map对每个查询进行特征描述，这是一个经过临床医生验证的框架，基于临床问题和LLM评估的分类法，记录其任务、意图、可回答性、缺失信息和潜在危害。文档和管理（36.2%）以及知识检索（28.9%）几乎占据了使用的三分之二，而诊断仅占3.7%；超过三分之一的查询无法得到良好的回答。将RCQ-Map应用于从主要评估套件和前沿模型报告中提取的58个公共基准（我们将其汇编成临床AI基准地图）显示，中位数基准不包含文档请求，并且与真实使用的任务组合共享31%，远低于各任务类别的均匀分布。旨在模拟临床实践的套件中的基准在个体上与真实使用的接近程度并不比在前沿模型报告中使用的基准更高。因此，基准得分对临床AI在实际工作中的表现几乎没有说明作用，评估应与真实的临床使用相匹配。

View on arXiv Download PDF AI Translation

cs.CL / 19 / 2610.11111

Lapras: Latent Reasoning for Time Series Language Models

Lapras：时间序列语言模型的潜在推理

Chen, Yuliang, Wu, Yu Yvonne, Langer, Patrick, Pillai, Arvind, Regmi, Sudarshan, Maritsch, Martin, Liu, Juncheng, Jakob, Robert, Kaar, Thomas, Griffin, Tess Z., Marsch, Lisa, Heinz, Michael V., Jacobson, Nicholas C., Campbell, Andrew

Abstract

Time Series Language Models (TSLMs) offer a promising path toward time series understanding by reasoning over temporal signals and producing natural language answers and explanations. A common approach is Chain-of-Thought (CoT), which generates step-by-step rationales linking relevant signal patterns to final answers. Although these models learn from reference CoT traces during post-training, generating faithful descriptions of input time series at inference remains challenging. Expressing high-dimensional, continuous temporal representations in discrete language tokens may cause the model to neglect task-relevant patterns or describe them inaccurately. Because later reasoning steps build on these descriptions, early errors propagate, leading to incorrect answers with plausible explanations that are inconsistent with the input signal. We propose Lapras (Latent Post-trained Reasoning Across Series), a post-training framework that equips TSLMs with latent reasoning. A model trained with Lapras reasons through a sequence of continuous thoughts in the joint time series-language space, producing text only for the final answer. It learns this through teacher-student self-distillation, where a teacher trained on CoT reference traces reasons explicitly through text. The student aligns its hidden states with the teacher's at the answer stage, transferring the teacher's reasoning ability into its latent computation. We evaluate Lapras across four TSLM backbones on five time series question answering benchmarks. Lapras improves average F1 by up to 10.79% over explicit CoT while generating 23.9x fewer tokens. Lapras's continuous thoughts can also be decoded into readable reasoning traces via standard language decoding, preserving textual explanations. Together, these results highlight Lapras as a promising post-training paradigm for efficient, effective, and interpretable TSLM reasoning.

Chinese Translation

时间序列语言模型（TSLMs）通过对时间信号进行推理并生成自然语言答案和解释，为时间序列理解提供了有希望的路径。一种常见的方法是思维链（Chain-of-Thought, CoT），它生成逐步的推理，将相关信号模式与最终答案联系起来。尽管这些模型在后训练阶段从参考的 CoT 轨迹中学习，但在推理时生成对输入时间序列的忠实描述仍然具有挑战性。将高维、连续的时间表示表达为离散的语言符号可能导致模型忽视与任务相关的模式或不准确地描述它们。由于后续的推理步骤依赖于这些描述，早期的错误会传播，导致产生与输入信号不一致的合理解释的错误答案。我们提出了 Lapras（潜在后训练推理跨序列），这是一个后训练框架，为 TSLMs 提供潜在推理。使用 Lapras 训练的模型在联合时间序列-语言空间中通过一系列连续的思维进行推理，仅为最终答案生成文本。它通过教师-学生自蒸馏学习这一过程，其中在 CoT 参考轨迹上训练的教师通过文本显式推理。学生在答案阶段将其隐藏状态与教师的状态对齐，将教师的推理能力转移到其潜在计算中。我们在五个时间序列问答基准上评估了四个 TSLM 骨干网络的 Lapras。Lapras 在显式 CoT 的基础上将平均 F1 提高了最多 10.79%，同时生成的符号数量减少了 23.9 倍。Lapras 的连续思维还可以通过标准语言解码解码为可读的推理轨迹，保留文本解释。综合来看，这些结果突显了 Lapras 作为高效、有效且可解释的 TSLM 推理的有前景的后训练范式。

View on arXiv Download PDF AI Translation

cs.CL / 20 / 2610.11132

SFT-as-Context Mitigates Forgetting in Supervised Fine-Tuning

SFT作为上下文缓解监督微调中的遗忘

Tang, Kenan, Hua, Andong, Qian, Chengxuan, Tiwari, Saket, Qin, Yao

Abstract

Supervised fine-tuning (SFT) equips large language models (LLMs) with specialized capabilities, but often comes at the cost of forgetting the general capabilities of their parent models (i.e., the pretrained models before fine-tuning). This trade-off is especially limiting for queries that require both specialized and general capabilities. We introduce SFT-as-context, a training-free method in which the parent model uses the SFT model's response as context to answer the query. This allows the parent model to acquire fine-tuned capabilities from the SFT response through in-context learning while preserving its own general capabilities. Across 19 parent-SFT model pairs and 11 benchmarks, SFT-as-context remains close to the SFT models on fine-tuned capabilities, with gaps of only 2.2 and 2.1 percentage points on AIME 2024 and LiveCodeBench and 2.0 macro MAE on NutriBench-English, while staying within 2.2 percentage points of the parent models on general capabilities on average. Remarkably, it can solve queries requiring both fine-tuned and general capabilities, even when neither the parent nor SFT model succeeds alone. This approach also extends beyond parent-SFT pairs: responses from a small open-source SFT model can improve a strong closed-source LLM, outperforming either model alone. Furthermore, we use a Bayesian framework to derive theoretical guarantees that bound the error of SFT-as-context relative to the SFT model on fine-tuned capabilities and to the parent model on general capabilities. In addition, we visualize the attention weights and find that the parent model attends more to useful SFT responses and less to irrelevant ones, suggesting that selective attention helps the parent model use the SFT response through in-context learning.

Chinese Translation

监督微调（SFT）为大型语言模型（LLMs）赋予了专业能力，但往往以遗忘其母模型（即微调前的预训练模型）的一般能力为代价。这种权衡对于需要专业和一般能力的查询尤其有限。我们提出了SFT作为上下文，这是一种无训练的方法，其中母模型使用SFT模型的响应作为上下文来回答查询。这使得母模型能够通过上下文学习从SFT响应中获取微调能力，同时保留其自身的一般能力。在19对母模型-SFT模型和11个基准测试中，SFT作为上下文在微调能力上与SFT模型保持接近，在AIME 2024和LiveCodeBench上仅相差2.2和2.1个百分点，在NutriBench-English上相差2.0的宏平均绝对误差（MAE），同时在一般能力上平均保持在母模型的2.2个百分点之内。值得注意的是，它能够解决同时需要微调和一般能力的查询，即使母模型和SFT模型单独都无法成功。这种方法还超越了母模型-SFT模型对：来自一个小型开源SFT模型的响应可以改善一个强大的闭源LLM，超越任一模型单独的表现。此外，我们使用贝叶斯框架推导出理论保证，界定了SFT作为上下文在微调能力上相对于SFT模型的误差界限，以及在一般能力上相对于母模型的误差界限。此外，我们可视化了注意力权重，发现母模型对有用的SFT响应的关注度更高，而对无关响应的关注度较低，这表明选择性注意力有助于母模型通过上下文学习使用SFT响应。

View on arXiv Download PDF AI Translation

cs.CL / 21 / 2610.11135

Can a System-One LLM Perform Knowledge Tracing When Few or No Learners Are Logged?

系统一型大语言模型在少量或没有学习者登录时能否进行知识追踪？

Lee, Unggi, Park, Haeun

Abstract

Knowledge tracing (KT) models need many logged learners, so a new course or platform starts without a usable model. In LLM-based KT the LLM generates the answer, which we call System-Two; it is either fine-tuned on the target data or reasons and votes over ten samples, which is slow and gives coarse probabilities. We ask whether an off-the-shelf System-One LLM, which returns a probability for a typed question directly in a single pass, can perform KT when few or no learners are logged. On seven datasets, Jev without any data from the target platform reaches a mean AUC of .706, above the best of 28 deep KT models trained on 8 learners (.689) and above System-Two Thinking-KT on all seven datasets (.650) at about 1/100 of its API cost. Adding examples and a similar-learner statistic from the logged learners (JevKT) raises this to .722; JevKT stays significantly ahead of deep KT up to 16 learners and ahead on average up to 64, and supervised KT catches up between 64 and 128 learners. Among the readers we tested, the gain is specific to Jev, since three other LLMs queried with the byte-identical typed request through the official System-One adapter fall below it on all seven datasets, and reader swaps and contamination checks find no evidence that the input format or memorised data explain the gain. For new learners the advantage holds from their first interactions, whereas on unseen items with all learners logged, deep KT remains ahead.

Chinese Translation

知识追踪（KT）模型需要大量登录的学习者，因此新课程或平台在开始时没有可用的模型。在基于大语言模型（LLM）的知识追踪中，LLM生成答案，我们称之为系统二；它要么在目标数据上进行微调，要么对十个样本进行推理和投票，这种方法较慢且给出粗略的概率。我们探讨了一个现成的系统一型LLM是否能够在少量或没有学习者登录的情况下进行知识追踪。通过对七个数据集的测试，Jev在没有任何来自目标平台的数据的情况下达到了平均AUC值0.706，超过了在8个学习者上训练的28个深度KT模型的最佳表现（0.689），并且在所有七个数据集上超过了系统二型Thinking-KT（0.650），其API成本约为后者的1/100。通过添加示例和来自已登录学习者的相似学习者统计数据（JevKT），这一值提高至0.722；JevKT在最多16个学习者的情况下显著领先于深度KT，并在平均情况下在最多64个学习者的情况下保持领先，而监督KT在64到128个学习者之间追赶。我们测试的读者中，这一提升特定于Jev，因为通过官方系统一适配器以字节完全相同的输入请求查询的其他三种LLM在所有七个数据集上的表现均低于Jev，并且读者交换和污染检查未发现输入格式或记忆数据解释该提升的证据。对于新学习者，这一优势从他们的首次互动开始就保持有效，而在所有学习者均已登录的情况下，深度KT在未见项目上仍然保持领先。

View on arXiv Download PDF AI Translation

cs.CL / 22 / 2610.11136

The "10th Juror": Open-Set Standpoint Screening for Bureaucratic Bias Detection

“第十陪审员”：开放集视角筛选用于官僚偏见检测

Miao, Yuchen, Wang, Zijun, Han, Chang, Shi, Yurui, Zhang, Mingtai, Xu, Siyang

Abstract

Presupposing the boundaries of bias is itself a form of bias. We study closed-loop bias governance for Dutch government documents, where a system must detect biased language, ground decisions in legal and contextual evidence, rewrite problematic sentences when intervention is warranted, and verify that the rewrite mitigates harm without distorting meaning. Existing methods face three challenges: (i) discriminative classifiers capture surface regularities but lack normative grounding; (ii) zero-shot LLMs often adopt generic viewpoints and over-flag ambiguous administrative language; and (iii) fixed taxonomies inherit the Closed-World Assumption, missing emerging local targets. We propose MARS-Gov, a standpoint-aware multi-agent framework that combines legal retrieval, open-set target screening, specialized jurors, conservative routing, and rewrite verification. When screening finds an uncovered group, MARS-Gov instantiates a dynamic "10th juror" to deliberate outside the fixed panel. On DGDB, MARS-Gov sets a new SOTA with 0.880 F1, outperforming the strongest zero-shot LLM detector by 20.2 points (29.8% relative) and the best supervised Dutch encoder by 6.8 points, while reducing unnecessary interventions to 2.5%. Leave-One-Category-Out (LOCO) evaluation recovers held-out categories with 85.1% Correct@1 and 93.8% Correct@3.

Chinese Translation

假设偏见的边界本身就是一种偏见。我们研究了荷兰政府文件的闭环偏见治理，其中系统必须检测偏见语言，基于法律和上下文证据做出决策，在必要时重写有问题的句子，并验证重写是否减轻了伤害而不扭曲意义。现有方法面临三个挑战：（i）判别分类器捕捉表面规律但缺乏规范基础；（ii）零样本大语言模型（zero-shot LLMs）往往采用通用视角，并过度标记模糊的行政语言；（iii）固定分类法继承了封闭世界假设，错过了新兴的地方目标。我们提出了MARS-Gov，一个具有立场意识的多智能体框架，结合了法律检索、开放集目标筛选、专业陪审员、保守路由和重写验证。当筛选发现未覆盖的群体时，MARS-Gov 实例化一个动态的“第十陪审员”以在固定小组之外进行审议。在DGDB上，MARS-Gov以0.880的F1分数设定了新的状态-of-the-art（SOTA），比最强的零样本LLM检测器高出20.2分（相对提高29.8%），比最佳的监督荷兰编码器高出6.8分，同时将不必要的干预减少到2.5%。Leave-One-Category-Out（LOCO）评估在85.1%的Correct@1和93.8%的Correct@3中恢复了被保留的类别。

View on arXiv Download PDF AI Translation

cs.CL / 23 / 2610.11159

Local Prototype Reconstruction for Text-Compatible Speech-to-LLM Bridge Pretraining

文本兼容的语音到大语言模型桥接预训练的局部原型重建

Zhao, Xinnian, Wu, Chia-Hua, Wang, Pu, Van Hamme, Hugo

Abstract

Speech-to-LLM systems often connect a frozen speech encoder to a frozen large language model (LLM) through a small trainable bridge. The bridge is usually treated as plumbing, but it in fact defines the geometry of the speech-to-LLM interface, and the pretraining objective decides whether that interface provides a reusable initialization for downstream tasks. We study a transferable bridge through two complementary properties: global alignment with the text side, and local lexical manifold compatibility, where bridge embeddings remain close to the frozen LLM's input-embedding neighbourhoods. We make this property measurable with a fixed, head-free, timestamp-free diagnostic that applies to any objective, and show that next-word prediction (NWP) and sentence-level contrastive pretraining do not fully capture token-level lexical compatibility. We then introduce Local Prototype Reconstruction (LPR), a lightweight training-only regularizer that requires each aligned bridge token to be reconstructable from a small neighbourhood of frozen LLM token embeddings, with a hard single-prototype anchor as its limiting case. On multilingual ASR and speech translation, LPR improves transfer, with the largest gains on translation and low-resource adaptation. Crucially, our independent diagnostic correlates with downstream gains across objectives, suggesting that lexical manifold compatibility is predictive of reusability for speech-to-LLM bridges.

Chinese Translation

语音到大语言模型（Speech-to-LLM）系统通常通过一个小的可训练桥接将一个冻结的语音编码器与一个冻结的大语言模型（LLM）连接起来。这个桥接通常被视为管道，但它实际上定义了语音到LLM接口的几何形状，而预训练目标决定了该接口是否为下游任务提供可重用的初始化。我们通过两个互补特性研究一个可转移的桥接：与文本侧的全局对齐，以及局部词汇流形兼容性，其中桥接嵌入保持接近冻结的LLM输入嵌入邻域。我们使这一特性可测量，采用一种固定的、无头的、无时间戳的诊断方法，适用于任何目标，并显示下一词预测（NWP）和句子级对比预训练并未完全捕捉到词元级的词汇兼容性。随后，我们引入局部原型重建（Local Prototype Reconstruction, LPR），这是一种轻量级的仅训练正则化器，要求每个对齐的桥接词元能够从一小邻域的冻结LLM词元嵌入中重建，其极限情况为一个硬单原型锚点。在多语言自动语音识别（ASR）和语音翻译中，LPR提高了迁移效果，翻译和低资源适应的增益最大。至关重要的是，我们的独立诊断与各目标的下游增益相关，表明词汇流形兼容性可以预测语音到LLM桥接的可重用性。

View on arXiv Download PDF AI Translation

cs.CL / 24 / 2610.11160

LadderEdit: Edit-Level Residual Compression for Memory-Efficient Lifelong Editing of LLMs

LadderEdit：面向内存高效的终身编辑的编辑级残差压缩

Yu, Xiaobing, Qiu, Peijie, Yang, Jin, Dong, Xuanzhao, Ma, Weiwei, An, Zhaoqi, Zhao, Xiaoqi, Liu, Xiaofeng

Abstract

Lifelong editing of LLMs requires storing thousands of edits after acquisition. A widely used family of approaches attaches one LoRA adapter per edit, which preserves behavior but grows linearly in storage. To address this challenge, we propose LadderEdit, a method that compresses each LoRA adapter after it is acquired. Each edit is first stored at low rank as a cheap sketch. We then check whether this sketch still satisfies the rewrite, generalization, and locality contract on probe prompts. Edits that pass keep the sketch; those that fail are promoted to a higher rank along a ladder until the contract is met. Because every edit retains some representation, coverage is maintained, and only hard edits consume more rank. Across ZsRE, CounterFact, and WikiBigEdit benchmarks on LLaMA-3-8B, Mistral-7B, and Qwen2.5-7B, LadderEdit tracks exact LoRA storage at 5.2x less memory and remains effective at 50,000 sequential edits.

Chinese Translation

终身编辑大型语言模型（LLMs）需要在获取后存储数千次编辑。广泛使用的一类方法为每次编辑附加一个 LoRA（低秩适配器），这种方法虽然能够保持行为，但在存储上呈线性增长。为了解决这一挑战，我们提出了 LadderEdit，一种在获取每个 LoRA 适配器后进行压缩的方法。每次编辑首先以低秩形式存储为一种廉价的草图。然后，我们检查该草图是否仍然满足重写、泛化和局部性合同的要求。通过的编辑保留草图；未通过的编辑则沿着梯子提升到更高的秩，直到满足合同。由于每次编辑都保留了一定的表示，覆盖率得以维持，只有难度较大的编辑才会消耗更多的秩。在 LLaMA-3-8B、Mistral-7B 和 Qwen2.5-7B 上的 ZsRE、CounterFact 和 WikiBigEdit 基准测试中，LadderEdit 在内存使用上以 5.2 倍的效率跟踪精确的 LoRA 存储，并在 50,000 次连续编辑中保持有效。

View on arXiv Download PDF AI Translation

cs.CL / 25 / 2610.11183

RAG-Stress: Probing the Limits of Evidence Reliance in Retrieval-Augmented Generation

RAG-Stress：探究检索增强生成中证据依赖的极限

Zhou, Shunyuan, Chen, Hao, Wang, Tianyu, Lin, Goose, Wang, Zaiyuan, Zhao, Haiying

Abstract

Following retrieved evidence does not guarantee factual correctness: misleading evidence can induce a model to replace an answer it previously gave correctly. Standard accuracy measures obscure this behavior by combining answer replacement with preexisting errors. We introduce RAG-Stress, a controlled diagnostic protocol for examining the limits of evidence reliance in retrieval-augmented generation. The protocol holds the question and reference answer fixed, edits one assertion to support a designated incorrect answer, and crosses two source priority policies with three positions of the answer span within the evidence text. We measure misleading rate (MR) on each model's subset of questions answered correctly without retrieval, alongside clean accuracy on the full evaluation set. We evaluate fifteen systems spanning API models, open models, and search agents trained with reinforcement learning on TriviaQA-RC, HotpotQA, and SearchQA, with additional English and Chinese MedQA evaluations. Instructions that prioritize documents consistently produce higher MR than those permitting reliance on prior knowledge. Averaged over models and positions, the gap ranges from 10.9 to 13.5 percentage points across the three QA datasets. Mean MR follows End $>$ Beginning $>$ Middle under both policies, although individual models do not uniformly follow this ordering. A separate paired audit of 500 questions and two checkpoints supports increased harmful override without establishing a corresponding improvement in beneficial correction. These findings distinguish evidence adherence from factual reliability and motivate evaluating whether retrieved evidence preserves, replaces, or corrects a model's answers.

Chinese Translation

依赖检索到的证据并不保证事实的正确性：误导性的证据可能导致模型替换之前正确给出的答案。标准的准确性度量通过将答案替换与先前存在的错误结合在一起，模糊了这种行为。我们提出了RAG-Stress，一种用于检查检索增强生成中证据依赖极限的受控诊断协议。该协议固定问题和参考答案，编辑一个断言以支持指定的错误答案，并交叉两种源优先策略与证据文本中答案跨度的三个位置。我们测量每个模型在未进行检索的情况下正确回答的问题子集上的误导率（MR），以及在完整评估集上的清晰准确性。我们评估了十五个系统，包括API模型、开放模型和使用强化学习训练的搜索代理，数据集包括TriviaQA-RC、HotpotQA和SearchQA，以及额外的英语和中文MedQA评估。优先考虑文档的指令相比于允许依赖先前知识的指令，通常产生更高的误导率。根据模型和位置的平均值，这一差距在三个问答数据集中范围为10.9到13.5个百分点。在两种策略下，平均误导率遵循“结束 $>$ 开始 $>$ 中间”的顺序，尽管个别模型并不均匀遵循这一排序。对500个问题和两个检查点的单独配对审计支持了有害覆盖的增加，但未建立相应的有益修正的改善。这些发现将证据遵循与事实可靠性区分开来，并促使我们评估检索到的证据是否保留、替换或修正模型的答案。

View on arXiv Download PDF AI Translation

cs.CL / 26 / 2610.11270

Gated Memory: Admission-Controlled Memory Formation for Conversational AI

门控记忆：面向对话人工智能的入场控制记忆形成

Saraswat, Preeti, Neelagiri, Divya, Manoj, Ajay

Abstract

Personalized conversational AI relies on long-term memory systems that extract facts from user utterances and store them in persistent vector stores. Despite progress in retrieval, deduplication, and lifecycle management, the formation stage, the moment a fact is first written to storage has received almost no principled attention. We identify this as the binding constraint on memory quality in production systems. Critical contextual signals, such as the distinction between a permanent user attribute and a transient situation, exist only in the original utterance and are irreversibly lost the moment extraction produces a subject-relation-object triple. No downstream process can recover them. We propose Gated Memory, a lightweight, modular formation framework that interposes two decision checkpoints between conversation and storage: an admission gate that evaluates every candidate fact against the full utterance context before extraction runs, and a conditional enrichment stage that grounds admitted facts through an entity scope taxonomy with privacy constraints. The gate evaluates only the current exchange while using prior turns as read-only reference context, and produces a structured formation record. Admitted content is decomposed into atomic facts, each categorized, tagged with provenance (directly stated versus inferred), scoped to its condition of applicability, and grounded in resolved time and place, subject to a constraint that no entity absent from the context may be asserted. On the LoCoMo-10 benchmark with atypical emotional density in utterance data, Gated Memory achieves an overall +2.6% relative improvement in LLM-judge accuracy over a strong baseline with identical retrieval and generation, establishing formation quality as a measurable constraint on memory performance.

Chinese Translation

个性化对话人工智能依赖于长期记忆系统，从用户的发言中提取事实并将其存储在持久的向量存储中。尽管在检索、去重和生命周期管理方面取得了进展，但形成阶段，即事实首次写入存储的时刻，几乎没有受到原则性的关注。我们将其视为生产系统中记忆质量的绑定约束。关键的上下文信号，例如永久用户属性与瞬时情况之间的区别，仅存在于原始发言中，并在提取生成主题-关系-对象三元组的瞬间不可逆转地丧失。没有下游过程能够恢复这些信息。我们提出了门控记忆（Gated Memory），这是一种轻量级、模块化的形成框架，在对话与存储之间设置了两个决策检查点：一个入场门，评估每个候选事实在提取运行之前与完整发言上下文的匹配情况，以及一个条件丰富阶段，通过具有隐私约束的实体范围分类法来确定被接受事实的基础。该门仅评估当前交流，同时将之前的轮次作为只读参考上下文，并生成结构化的形成记录。被接受的内容被分解为原子事实，每个事实都被分类、标记来源（直接陈述与推断）、限定适用条件，并在已解决的时间和地点中确定，受到一个约束，即不得断言上下文中缺失的任何实体。在具有非典型情感密度的发言数据的LoCoMo-10基准测试中，门控记忆在与强基线相同的检索和生成条件下，整体上实现了+2.6%的相对提高，确立了形成质量作为记忆性能的可测量约束。

View on arXiv Download PDF AI Translation

cs.CL / 27 / 2610.11275

Phonological Interference in Multilingual Speech Models

多语种语音模型中的音位干扰

Yanuka, Moran, Giryes, Raja, Alper, Moris

Abstract

Phoneme-level models transcribe or generate speech as a sequence of phonemes, the smallest sound units that distinguish words. These models enable fine-grained pronunciation control and understanding, yet often fail on input that does not match any single training language, such as speech alternating between two languages, known as code-switching, or low-resource languages absent from training. We identify a systematic failure mode behind this, phonological interference: models assume the input is in a single language and impose its phonology, overriding local phoneme-level decisions that conflict with the assumed language. We measure interference by how often a model retains phonemes that one language has but the other lacks. On code-switched input, two phone recognizers (speech-to-phoneme models) and a phoneme-conditioned text-to-speech model lose 32% to 79% of these phonemes, but lose far fewer of the phonemes both languages share. On unseen languages, we find that phone recognizers impose the phonology of the training language they assign to the speech, and the more confident the assignment, the more they lose phonemes the unseen language has but the assigned language lacks. We probe the models' language estimate from their internal activations, and trace interference to a low dimensional subspace. On monolingual speech, steering this subspace toward another language makes the model lose the phonemes that only the original language uses and produce phonemes that only the target language has. We introduce windowed language estimation (WLE), an inference time repair that replaces the model's language estimate in this subspace with one computed from a short window around each position. On code-switched input, WLE removes 34% to 69% of the interference in all three models, and in the recognizers it leaves monolingual performance essentially unchanged.

Chinese Translation

音位级模型将语音转录或生成为音位序列，即区分单词的最小声音单位。这些模型能够实现细致的发音控制和理解，但在输入不符合任何单一训练语言的情况下，往往表现不佳，例如在两种语言之间交替的语音（称为代码切换）或缺乏训练的低资源语言。我们识别出这一现象背后的系统性失败模式：音位干扰。模型假设输入为单一语言，并强加其音系，覆盖与假定语言冲突的局部音位级决策。我们通过模型保留某一语言拥有而另一语言缺乏的音位的频率来衡量干扰。在代码切换输入中，两个语音识别器（语音到音位模型）和一个音位条件的文本到语音模型失去了32%到79%的这些音位，但失去的共享音位则少得多。在未见语言中，我们发现语音识别器强加了它们分配给语音的训练语言的音系，且分配越自信，越会失去未见语言拥有而分配语言缺乏的音位。我们从模型的内部激活中探测语言估计，并将干扰追溯到一个低维子空间。在单语音频中，将该子空间引导至另一种语言使模型失去仅原语言使用的音位，并产生仅目标语言拥有的音位。我们引入了窗口语言估计（WLE），这是一种推理时修复方法，它用从每个位置周围短窗口计算的语言估计替换模型在该子空间中的语言估计。在代码切换输入中，WLE消除了所有三个模型中34%到69%的干扰，并且在识别器中几乎不改变单语表现。

View on arXiv Download PDF AI Translation

cs.CL / 28 / 2610.11287

REMORY: Learning Residual Memory for Context Compaction

REMORY：学习残差记忆以实现上下文压缩

Xia, Hanchen, Chen, Baoyou, Ge, Yutang, Deng, Naihao, Yang, Senqiao, Dong, Zilong, Yuan, Weihao, Zhu, Siyu

Abstract

Long-horizon agents compact their history to continue within a finite context window, but a textual summary alone may not support every subsequent decision. We introduce REMORY, a neural memory network that supplements the summary with a bounded sequence of soft memory tokens. Given the history and summary, the network learns to generate tokens that help a frozen LLM approximate the continuation it would produce with the full history. The tokens are conditioned on the summary and appended after it, forming an analogue of a residual connection along the sequence dimension. On SummHay, REMORY improves source attribution at nearly unchanged insight coverage and approaches the full-context joint score using only 5.2% of the input positions. Across long-horizon agent benchmarks, Qwen3.8-27B and GLM-5.3-Flash show consistent gains with residual memory. Both models also exhibit substantially fewer repeated tool outputs and tool errors on BrowseComp and Terminal-Bench 2.1.

Chinese Translation

长时间跨度的智能体将其历史信息压缩，以便在有限的上下文窗口内继续进行，但单靠文本摘要可能无法支持每一个后续决策。我们提出了REMORY，一种神经记忆网络，它通过一系列有界的软记忆标记来补充摘要。给定历史和摘要，该网络学习生成有助于冻结的大型语言模型（LLM）近似其在完整历史下所产生的延续的标记。这些标记以摘要为条件，并在其后附加，形成沿序列维度的残差连接的类比。在SummHay数据集上，REMORY在几乎不改变洞察覆盖率的情况下改善了源归属，并且仅使用5.2%的输入位置就接近完整上下文的联合得分。在长时间跨度的智能体基准测试中，Qwen3.8-27B和GLM-5.3-Flash在使用残差记忆时表现出持续的提升。这两个模型在BrowseComp和Terminal-Bench 2.1上也表现出显著减少的重复工具输出和工具错误。

View on arXiv Download PDF AI Translation

cs.CL / 29 / 2610.11291

When Do We Need On-Policy Distillation? Distilling on Offline Student Rollouts Is Often Better

何时需要在线策略蒸馏？离线学生回放的蒸馏通常更好

Zhao, Siyan, Fu, Yonggan, Jiang, Jindong, Liu, Shih-Yang, Bian, Song, Lee, Byung-Kwan, Sreenivas, Sharath Turuvekere, Dai, Wenliang, Ye, Hanrong, Grover, Aditya, Molchanov, Pavlo

Abstract

On-policy distillation (OPD) has become increasingly popular for transferring teacher capabilities to student models. In this work, we ask a critical research question: Is on-policy sampling always beneficial for distilling arbitrary teacher-student pairs? We show that a simple alternative, Semi-OPD, which distills from offline rollouts generated by the initial student, can often outperform OPD in both accuracy and training efficiency. Across 17 teacher-student pairs ranging from 1.5B to 235B parameters, Semi-OPD outperforms OPD in 14 cases, with up to +13.6% accuracy and 11.4x training speedup. We further find that the choice between OPD and Semi-OPD depends on the alignment between the initial teacher and student, quantified by an output-token overlap ratio: OPD is beneficial only when the two are highly aligned with high overlap ratios. Our deeper investigation suggests that effective distillation requires on-policyness w.r.t. both the student and the teacher. For misaligned pairs, student rollouts can become increasingly off-policy w.r.t. the teacher as context length grows, weakening the distillation signal. In contrast, Semi-OPD is often more stable, as it distills on shorter contexts while covering full trajectories and exposing the student to more teacher-preferred tokens. Beyond proposing Semi-OPD as an efficient alternative, our work motivates the community to rethink when to use OPD and to study stronger OPD variants with meaningful teacher-student pairs.

Chinese Translation

在线策略蒸馏（On-policy Distillation, OPD）在将教师模型的能力转移到学生模型中变得越来越受欢迎。在本研究中，我们提出了一个关键的研究问题：在线采样是否总是有利于蒸馏任意的教师-学生对？我们表明，一个简单的替代方案，半在线策略蒸馏（Semi-OPD），它从由初始学生生成的离线回放中进行蒸馏，通常在准确性和训练效率上优于OPD。在17对参数范围从15亿到235亿的教师-学生对中，Semi-OPD在14个案例中优于OPD，准确性提升高达13.6%和训练速度提升11.4倍。我们进一步发现，OPD和Semi-OPD之间的选择取决于初始教师和学生之间的对齐程度，这一程度通过输出标记重叠率量化：只有在两者高度对齐且重叠率高时，OPD才是有利的。我们的深入研究表明，有效的蒸馏需要在学生和教师之间保持在线策略性。对于不对齐的对，随着上下文长度的增加，学生的回放可能会越来越偏离教师，从而削弱蒸馏信号。相反，Semi-OPD通常更为稳定，因为它在较短的上下文中进行蒸馏，同时覆盖完整的轨迹，并使学生接触到更多教师所偏好的标记。除了提出Semi-OPD作为一种高效的替代方案，我们的工作还激励社区重新思考何时使用OPD，并研究具有有意义的教师-学生对的更强OPD变体。

View on arXiv Download PDF AI Translation

cs.CL / 30 / 2610.11305

BeliefScope: Diagnosing Evidence-Driven Revision and Pressure-Induced Shifts in Large Language Models

BeliefScope：诊断大型语言模型中的证据驱动修正和压力诱导转变

Guo, Shuai, Cui, Yidong

Abstract

A language model may revise the same proposition after receiving genuinely relevant evidence or after receiving directional user pressure that adds no relevant fact. The observable response shift alone therefore does not reveal which source drove the change. We introduce BeliefScope, a controlled black-box framework for separating these two sources of influence around a fixed target proposition. BeliefScope crosses Evidence and Pressure with factor-specific local controls and measures response changes through probability reports, categorical judgments, and action recommendations on channel-appropriate scales. To determine when these observable contrasts support reliable attribution, we evaluate the observation design under controlled synthetic conditions. Known-truth recovery and targeted ablations establish where Evidence- and Pressure-related effects can be separated, while semi-synthetic stress tests map how that recoverability changes as the observation process becomes noisier and more heterogeneous. Across a 36-family Qwen/Llama study, with targeted 12-family checks that also include Gemma3-12B, the resulting profiles show substantial evaluation-context dependence: broad model-level differences can change under matched controls, decoding, or response interfaces, while some narrower within-model patterns remain stable. Instruction interventions further show that reduced target-aligned Pressure following can reflect either stable resistance or movement in the opposite direction. BeliefScope summarizes these measurements as a conditional belief-response profile that keeps diagnostic effects tied to the evaluation conditions under which they are observed, together with explicit validity boundaries for that diagnosis.

Chinese Translation

语言模型在接收到真正相关的证据或在接收到没有相关事实的方向性用户压力后，可能会修正相同的命题。因此，仅凭可观察的响应变化无法揭示是哪个来源驱动了这种变化。我们引入了BeliefScope，这是一个受控的黑箱框架，用于在固定目标命题周围分离这两种影响来源。BeliefScope通过特定因素的局部控制，将证据和压力交叉，并通过概率报告、分类判断和适当通道规模的行动建议来测量响应变化。为了确定这些可观察的对比何时支持可靠的归因，我们在受控的合成条件下评估观察设计。已知真相的恢复和目标性消融确立了证据和压力相关效应可以分离的地方，而半合成压力测试则描绘了随着观察过程变得更加嘈杂和异质化，这种可恢复性如何变化。在一个包含36个家庭的Qwen/Llama研究中，针对12个家庭的检查也包括Gemma3-12B，结果显示出显著的评估上下文依赖性：广泛的模型级差异可以在匹配控制、解码或响应接口下发生变化，而一些较窄的模型内部模式则保持稳定。指令干预进一步表明，随后减少的目标对齐压力可能反映出稳定的抵抗或朝相反方向的移动。BeliefScope将这些测量总结为一个条件信念-响应轮廓，将诊断效应与观察到的评估条件相联系，并提供该诊断的明确有效性边界。

View on arXiv Download PDF AI Translation

cs.CL / 31 / 2610.11314

From Retrieval to Reconstruction: Constructing Evolvable Cognitive Memory for Long-Term Dialogue

从检索到重建：构建可演化的认知记忆以支持长期对话

Liao, Zirui, Wu, Zhengxian, Chen, Zhuohong, Yu, Yunyao, Liu, Xiaoyu, Xu, Yifan, Wang, Haoqian

Abstract

Large Language Models (LLMs) serving as long-term dialogue agents require memory systems that support reliable reasoning over extended interactions. However, existing Retrieval-Augmented Generation (RAG) frameworks typically treat memory as passive storage, making it difficult to distinguish source-attributed beliefs from unattributed event/fact records and to connect evidence dispersed across sessions. We introduce CogMem, a cognitive memory architecture based on the PEC$^2$F (Person-Event-Concept-Claim-Fact) graph schema. Dedicated Claim nodes preserve the source and target of subjective statements, while Fact and Event nodes represent semantic and episodic knowledge. Dialogue turns are incrementally converted into provenance-aware graph records, consolidated into higher-level facts, and reconciled into temporally scoped Claim views when the same source provides conflicting updates. For retrieval, a rule-based controller driven by LLM intent parsing composes four deterministic graph operators---anchoring, traversal, intersection, and evidence grounding---to reconstruct query-relevant context. Experiments on LoCoMo and LongMemEval show strong performance, especially on multi-hop, temporal, and knowledge-update tasks. Ablations and a semantic-collapse probe support complementary contributions from epistemic separation, consolidation, and agentic retrieval. Code: https://github.com/Silent-Rain02/CogMem.

Chinese Translation

作为长期对话代理的大型语言模型（LLMs）需要支持可靠推理的记忆系统，以应对延续的互动。然而，现有的检索增强生成（RAG）框架通常将记忆视为被动存储，这使得区分源属性信念与无属性事件/事实记录变得困难，并且难以连接跨会话分散的证据。我们提出了CogMem，一种基于PEC$^2$F（人物-事件-概念-主张-事实）图谱架构的认知记忆。专用的主张节点保存主观陈述的来源和目标，而事实和事件节点则表示语义和情节知识。对话轮次被逐步转换为具有来源意识的图记录，并在同一来源提供冲突更新时整合为更高级的事实，并协调为时间范围内的主张视图。为了检索，一个基于规则的控制器通过LLM意图解析驱动，组合四个确定性图操作——锚定、遍历、交集和证据基础——以重建与查询相关的上下文。在LoCoMo和LongMemEval上的实验显示出强劲的表现，特别是在多跳、时间和知识更新任务上。消融实验和语义崩溃探测支持了来自认知分离、整合和代理检索的互补贡献。代码：https://github.com/Silent-Rain02/CogMem。

View on arXiv Download PDF AI Translation

cs.CL / 32 / 2610.11316

MetaEncoder: Exploring the Limit of Bi-Encoders for Multimodal System One Decision Making with Natural Language Interface

MetaEncoder：探索双编码器在自然语言接口下多模态系统一决策中的极限

Cheng, Jianpeng, Sun, Guangyu, Singh, Aashu, Zhang, Benyu, Dai, Haixing, Mansour, Hossein, Zhang, Jiangfan, Mishra, Shlok Kumar, Sun, Wei, Cui, Xuanming, Liu, Yanli, Guo, Qi, Fan, Max Xiangjun, Xiao, Jun

Abstract

System One models output constrained decisions and probability distributions rather than free-form text generation. While prevailing paradigms rely on structured schema objects to encode state, intent, and candidate choices, we revisit a fully natural language-based System One interface. In this framework, both the user request and each candidate option are expressed in natural language, supported by multimodal (image and video) auxiliary inputs. We introduce MetaEncoder, which fine-tunes a pre-trained Muse-Glimmer 30B decoder into an instruction-following decision-making encoder. To scale effectively across both small closed-set (< 256) and massive open-set (millions) candidate spaces, MetaEncoder employs a bi-encoder architecture trained via unidirectional contrastive learning for request-candidate alignment. We conduct extensive evaluations across 11 benchmark suites and 190 tasks spanning multimodal decision-making, understanding (closed-set) and retrieval (open-set), highlighting where MetaEncoder beats SOTA multimodal encoders, as well as its current limits on reasoning-intensive tasks.

Chinese Translation

系统一模型输出受限于决策和概率分布，而非自由形式的文本生成。尽管现有范式依赖于结构化的模式对象来编码状态、意图和候选选择，我们重新审视了一个完全基于自然语言的系统一接口。在这个框架中，用户请求和每个候选选项均以自然语言表达，并辅以多模态（图像和视频）辅助输入。我们引入了MetaEncoder，它将预训练的Muse-Glimmer 30B解码器微调为一个遵循指令的决策编码器。为了在小型封闭集（< 256）和大型开放集（数百万）候选空间中有效扩展，MetaEncoder采用了通过单向对比学习训练的双编码器架构，以实现请求与候选之间的对齐。我们在11个基准套件和190个任务上进行了广泛评估，涵盖多模态决策、理解（封闭集）和检索（开放集），突显了MetaEncoder在超越现有最先进的多模态编码器方面的表现，以及其在推理密集型任务上的当前局限性。

View on arXiv Download PDF AI Translation

cs.CL / 33 / 2610.11332

ReCal: Calibrating Structured Pruning for On-Policy Distillation Recovery

ReCal：为在线策略蒸馏恢复校准结构化剪枝

Jiang, Houcheng, Zheng, Mao, Song, Mingyang, Zhong, Qiyong, Sun, Jie, Zhang, Tianyu, Fang, Junfeng

Abstract

Structured pruning reduces the deployment cost of reasoning language models, but the resulting capability degradation can hinder subsequent on-policy distillation (OPD) recovery. Because OPD relies on student-generated trajectories, pruning damage that persists after offline distillation can limit its effectiveness. We propose RECAL, Recovery-Aware Calibration, a simple plug-and-play approach that improves OPD recovery by adjusting calibration before pruning. RECAL uses forward KL between an unpruned teacher and a pruned probe to identify teacher-supported predictions disrupted by pruning, then reweights calibration statistics to guide existing pruning criteria toward preserving these predictions. Across multiple models and pruning methods, RECAL consistently improves mathematical reasoning after OPD, achieving gains of up to 16.7 percentage points on AIME, alongside improvements in most code-generation comparisons. Further analysis shows that RECAL reduces residual damage at heavily affected tokens and establishes performance advantages that persist through recovery. These results demonstrate the value of recovery-aware calibration for improving on-policy distillation recovery of pruned reasoning models.

Chinese Translation

结构化剪枝降低了推理语言模型的部署成本，但随之而来的能力下降可能会阻碍后续的在线策略蒸馏（OPD）恢复。由于OPD依赖于学生生成的轨迹，离线蒸馏后持续存在的剪枝损害可能限制其有效性。我们提出了RECAL（Recovery-Aware Calibration），一种简单的即插即用方法，通过在剪枝前调整校准来改善OPD恢复。RECAL使用未剪枝教师模型与剪枝探测器之间的前向KL散度来识别被剪枝干扰的教师支持预测，然后重新加权校准统计数据，以引导现有剪枝标准朝着保留这些预测的方向发展。在多个模型和剪枝方法中，RECAL在OPD后始终改善数学推理，AIME上的增益高达16.7个百分点，同时在大多数代码生成比较中也有改善。进一步分析表明，RECAL减少了在受影响严重的标记上的残余损害，并建立了在恢复过程中持续存在的性能优势。这些结果展示了恢复感知校准在改善剪枝推理模型的在线策略蒸馏恢复中的价值。

View on arXiv Download PDF AI Translation

cs.CL / 34 / 2610.11351

Deception by Omission: Language Models Knowingly Hide Their Mistakes

通过遗漏进行欺骗：语言模型故意隐藏其错误

Florin, Lucas, Knecht, Amelie, Schaller, Ulysse, Hagendorff, Thilo

Abstract

Large language models (LLMs) increasingly act as agents with little human oversight, so potential mistakes they make can go unnoticed. Users then depend on the model to report what went wrong. An honest model discloses its mistakes, while a deceptive one conceals them. However, it is unclear how current LLMs behave in such situations. In this study, we prefill LLM trajectories with synthetic mistakes. The trajectories resemble real deployments in chat and agentic settings. Models fail to disclose their mistake in 36.4% of chat and 67.1% of agentic rollouts. In 2.4% and 5.3% of rollouts, respectively, they are aware of the mistake in their chain of thought but still deceptively conceal it. Rates vary by model: for instance, Gemini 3.5 Flash knowingly conceals mistakes in up to 19.9% of agentic rollouts. In 11.9% of chat and 51.8% of agentic rollouts, models show no awareness of mistakes, even though they reliably spot them when reviewing the same transcript as an outside observer. Our results show that, as agents take on more tasks with less oversight, users cannot rely on them to self-report possible mistakes. Developers should instead use independent monitors that review agent trajectories, or specifically train models to check their past actions and disclose what they find.

Chinese Translation

大型语言模型（LLMs）越来越多地作为几乎没有人类监督的代理进行操作，因此它们可能犯下的错误可能会被忽视。用户依赖模型报告出错的情况。一个诚实的模型会披露其错误，而一个具有欺骗性的模型则会隐瞒这些错误。然而，目前尚不清楚当前的LLMs在这种情况下的表现。在本研究中，我们用合成错误预填充LLM轨迹。这些轨迹类似于聊天和代理设置中的真实部署。模型在36.4%的聊天和67.1%的代理回放中未能披露其错误。在2.4%和5.3%的回放中，它们分别意识到思维链中的错误，但仍然故意隐瞒。不同模型的隐瞒率有所不同：例如，Gemini 3.5 Flash在多达19.9%的代理回放中故意隐瞒错误。在11.9%的聊天和51.8%的代理回放中，模型对错误没有意识，尽管它们在作为外部观察者审查相同的转录时能够可靠地发现这些错误。我们的结果表明，随着代理承担更多任务且监督减少，用户不能依赖它们自我报告可能的错误。开发者应使用独立监控者来审查代理轨迹，或专门训练模型检查其过去的行为并披露所发现的内容。

View on arXiv Download PDF AI Translation

cs.CL / 35 / 2610.11354

AdaptEvo: Adaptive Agent Learning with Evolving Supervision

AdaptEvo：具有演变监督的自适应代理学习

Wan, Shijun, Xie, Jiancong, Xu, Hang, Duan, Jin, Wang, Qixiong, Xiang, Xi, Que, Maofei, Liu, Yahui, Wei, Zhongyu, Chuan, Mu

Abstract

Rule-governed contextual decision tasks require models to apply specified rules to case-specific context and evidence. Written rules can leave gaps in decision guidance and process evaluation, while reference judgments vary in their support from the rules and evidence. To address these challenges, we introduce AdaptEvo, a framework for learning under imperfect supervision that couples confidence-adaptive policy optimization with evolving decision knowledge and evaluation rubrics. Its Training module uses Confidence-Adaptive GRPO (CA-GRPO) to balance outcome and process rewards according to reference confidence. Its Evolution module synthesizes reusable decision knowledge from recurring failures across training cases and refines process rubrics to detect overlooked errors. To support empirical evaluation, we construct an industrial multimodal content moderation dataset comprising a training set and In-Period and Out-of-Period test sets, with the latter collected under changed rules. Using Qwen3.6-35B-A3B, AdaptEvo achieves 61.9% exact-label accuracy and 72.2% binary decision accuracy on In-Period, exceeding GRPO by 7.5 and 3.7 percentage points, respectively. On Out-of-Period, the policy trained with CA-GRPO retains exact-label accuracy gains over the base model across evaluated checkpoints without injected decision knowledge, while GRPO declines with continued training. CA-GRPO also outperforms the tested fixed reward mixtures on both Out-of-Period metrics.

Chinese Translation

受规则约束的情境决策任务要求模型将指定规则应用于特定案例的上下文和证据。书面规则可能在决策指导和过程评估中留下空白，而参考判断在其对规则和证据的支持程度上也存在差异。为了解决这些挑战，我们提出了AdaptEvo，一个在不完美监督下学习的框架，它结合了信心自适应政策优化与不断演变的决策知识和评估标准。其训练模块使用信心自适应的广义重加权政策优化（Confidence-Adaptive GRPO, CA-GRPO）根据参考信心平衡结果和过程奖励。其演变模块从训练案例中的重复失败中综合可重用的决策知识，并完善过程标准以检测被忽视的错误。为了支持实证评估，我们构建了一个工业多模态内容审核数据集，包括训练集以及在规则变化下收集的期内（In-Period）和期外（Out-of-Period）测试集。使用Qwen3.6-35B-A3B，AdaptEvo在期内实现了61.9%的精确标签准确率和72.2%的二元决策准确率，分别比GRPO高出7.5和3.7个百分点。在期外，使用CA-GRPO训练的政策在评估的检查点上保持了相较于基础模型的精确标签准确率提升，而GRPO在持续训练中则有所下降。CA-GRPO在期外的两个指标上也优于测试的固定奖励组合。

View on arXiv Download PDF AI Translation

cs.CL / 36 / 2610.11371

SignRAG: Unified Retrieval-Augmented Gloss-Free Sign Language Translation

SignRAG：统一的检索增强无注释手语翻译

Rao, Zhi, Zhou, Yucheng, Sun, Qianran, Huang, Yiqing, Yuan, Longcan, Hou, Jiayi, Yao, Chengwen, Cheng, Lin, Sun, Donghui, Chen, Xiaoxin, Wan, Jun

Abstract

Contemporary decoder-only large language models (LLMs) have demonstrated strong capabilities across a wide range of domains. However, existing pretraining paradigms for gloss-free sign language translation (SLT) are largely designed around conventional encoder-decoder pretrained language models, which limits their direct applicability to decoder-only LLMs. To address this limitation, we propose SignRAG, a unified framework combining hierarchical pretraining, target-domain retrieval augmentation, and retrieval-aware reinforcement fine-tuning. Hierarchical pretraining first learns linguistically grounded sign representations and then jointly aligns the sign encoder with an LLM, mitigating cross-modal optimization imbalance. For downstream adaptation, SignRAG complements parameter-based fine-tuning with a target-domain retrieval gallery that provides instance-specific translation cues. To ensure that retrieved contexts are used appropriately, we further introduce Retrieval Utility-Guided Reinforcement Fine-Tuning (RUG-RFT), which combines translation-quality and retrieval-utility rewards to encourage beneficial retrieval use while suppressing harmful reliance. Experiments on multiple SLT benchmarks establish new state-of-the-art performance. In particular, to the best of our knowledge, SignRAG is the first gloss-free approach to outperform gloss-supervised methods across all reported metrics on CSL-Daily. Our code has been released at \href{https://github.com/shahelaojieraozhi/SignRAG}{GitHub}, together with models of different sizes to support future academic research.

Chinese Translation

当代仅解码器的大型语言模型（LLMs）在广泛领域中展现了强大的能力。然而，现有的无注释手语翻译（SLT）预训练范式主要围绕传统的编码器-解码器预训练语言模型设计，这限制了它们对仅解码器LLMs的直接适用性。为了解决这一限制，我们提出了SignRAG，一个结合了分层预训练、目标领域检索增强和检索感知强化微调的统一框架。分层预训练首先学习语言学基础的手语表示，然后将手语编码器与LLM联合对齐，从而缓解跨模态优化不平衡。对于下游适应，SignRAG通过提供实例特定翻译线索的目标领域检索库来补充基于参数的微调。为了确保检索到的上下文被适当地使用，我们进一步引入了检索效用引导的强化微调（RUG-RFT），该方法结合了翻译质量和检索效用奖励，以鼓励有益的检索使用，同时抑制有害的依赖。在多个SLT基准上的实验建立了新的最先进性能。特别是，据我们所知，SignRAG是第一个在所有报告指标上超越注释监督方法的无注释方法，尤其是在CSL-Daily数据集上。我们的代码已在GitHub上发布，链接为 exttt{https://github.com/shahelaojieraozhi/SignRAG}，并提供了不同规模的模型以支持未来的学术研究。

View on arXiv Download PDF AI Translation

cs.CL / 37 / 2610.11389

Fact over Fiction: Detection of Pathological Hallucinations in Sinhala-to-English Neural Machine Translation

事实胜于虚构：在僧伽罗语到英语的神经机器翻译中检测病理性幻觉

Obeysekara, Navam, Jayatilleke, Nevidu

Abstract

Neural Machine Translation (NMT) models, while capable of producing highly fluent outputs, remain vulnerable to hallucinations, which are translations that are natural yet semantically unrelated to the source. This vulnerability is acute in low-resource settings like Sinhala-to-English, where weak cross-lingual alignment leads to hallucinations. This paper introduces a framework for reference-free hallucination detection in this language pair. We present a 45,000-sample synthetic dataset generated through a probabilistic chain of five linguistically motivated corruption strategies, with a semantic rescue mechanism that uses character-level similarity to distinguish hallucinations from morphological variants. We fine-tune mDeBERTa-v3 for token-level sequence labelling, reaching a token-level F1 of 0.841 +/- 0.001 over three seeds on a source-disjoint test set, and study a three-signal ensemble integrating neural risk scores, sequence log-probabilities, and cross-lingual semantic embeddings (LaBSE). A source-ablation control shows that the detector relies on the Sinhala source rather than on surface artefacts of the corruption process: shuffling or removing the source reduces sentence-level AUROC from 0.970 to chance. We benchmark eight NMT systems spanning five model families and find that detector firings vary by an order of magnitude across architectures.

Chinese Translation

神经机器翻译（NMT）模型虽然能够生成高度流畅的输出，但仍然容易出现幻觉，即那些自然但与源文本在语义上无关的翻译。这种脆弱性在像僧伽罗语到英语这样的低资源环境中尤为明显，因为弱的跨语言对齐导致了幻觉的产生。本文提出了一种在这一语言对中无参考的幻觉检测框架。我们展示了一个由五种语言学驱动的腐蚀策略的概率链生成的45,000样本合成数据集，并采用语义救助机制，通过字符级相似性来区分幻觉与形态变体。我们对mDeBERTa-v3进行了微调，以进行标记级序列标注，在一个源不重叠的测试集上达到了0.841 +/- 0.001的标记级F1分数，并研究了一个三信号集成模型，结合了神经风险评分、序列对数概率和跨语言语义嵌入（LaBSE）。源消融控制显示，检测器依赖于僧伽罗语源文本，而不是腐蚀过程的表面伪影：打乱或移除源文本将句子级AUROC从0.970降低到随机水平。我们基准测试了八个跨五个模型家族的NMT系统，发现检测器的触发在不同架构之间变化了一个数量级。

View on arXiv Download PDF AI Translation

cs.CL / 38 / 2610.11430

BioBigBird: A Sparse Attention Model for Long-Range Dependency Processing in Biomedical Text

BioBigBird：一种用于生物医学文本长距离依赖处理的稀疏注意力模型

Balaji, Roshan, S, Pavan Kumar, Gupta, Vasudev, N, Sreejith, Sridhar, Keerthana, Bhatt, Nirav

Abstract

While domain-specific Large Language Models (LLMs) have encoded vast biomedical knowledge, their limited context windows often hinder a deep understanding of nuanced relationships within and across texts. To address this limitation, we introduce BioBigBird, a bidirectional language model pre-trained on extensive biomedical literature and clinical data, specifically designed to handle long-range dependencies. BioBigBird leverages a sparse attention mechanism to process sequences up to 4096 tokens, and its training incorporates a multi-stage process to mitigate noise from the large-scale pre-training corpus. We further enhance its performance by employing a multi-task learning (MTL) framework that jointly optimizes for Named Entity Recognition and Relation Extraction. Comprehensive evaluations on the BLURB benchmark reveal that our MTL-enhanced BioBigBird achieves highly competitive results against state-of-the-art models. Our work contributes an effective methodology for developing powerful, long-context language models for specialized domains, demonstrating the value of extended sequence processing for complex text analysis. Our models are publicly available at https://huggingface.co/collections/bisectgroup/biobigbird.

Chinese Translation

尽管领域特定的大型语言模型（LLMs）已经编码了大量生物医学知识，但其有限的上下文窗口常常阻碍对文本内外细微关系的深入理解。为了解决这一限制，我们提出了BioBigBird，这是一种在广泛的生物医学文献和临床数据上进行预训练的双向语言模型，专门设计用于处理长距离依赖。BioBigBird利用稀疏注意力机制处理最长可达4096个标记的序列，其训练过程采用多阶段方法，以减少大规模预训练语料库中的噪声。我们进一步通过采用多任务学习（MTL）框架来提升其性能，该框架联合优化命名实体识别和关系提取。在BLURB基准上的全面评估显示，我们的MTL增强型BioBigBird在与最先进模型的比较中取得了高度竞争的结果。我们的工作为开发强大的长上下文语言模型提供了一种有效的方法，展示了扩展序列处理在复杂文本分析中的价值。我们的模型可在 https://huggingface.co/collections/bisectgroup/biobigbird 上公开获取。

View on arXiv Download PDF AI Translation

cs.CL / 39 / 2610.11436

Adversarial Cues in Decision Models Used as Judges: The Role of Request Presentation

作为裁判的决策模型中的对抗性提示：请求呈现的作用

Liu, Hongliang

Abstract

An answer judge instructed to grade the final commitment should reject an explicitly wrong final value even when an earlier value matches the reference. We show that adding one colon to a candidate can violate this requirement depending on the presentation of the structured judging request. Numeric references certify the error, and paired interventions distinguish the candidate edit from the integration's presentation choices. On 200 previously unused DROP and GSM8K source clusters, the edit increased Jev's false acceptance from 1.0% to 26.0% with three output labels and from 3.0% to 26.5% with the published four-label grading instruction under sorted JSON keys. Both candidate variants were rejected under insertion presentation. These interactions passed the prespecified statistical correction even though Jev met the control thresholds under both grading configurations and presentations. Most excess acceptances occurred among candidates assigned larger numerical errors. GPT-6 Sol produced no observed cue-condition false acceptances, with missing responses unresolved. The result shows that basic judging competence can coexist with sharply different vulnerability to a fixed candidate edit across logically equivalent request presentations. The reference-aware grammar and compound ordering change limit the finding's operational scope and leave its internal cause unmeasured.

Chinese Translation

被指示对最终承诺进行评分的答案裁判应当拒绝一个明显错误的最终值，即使早期值与参考值匹配。我们展示了在候选者中添加一个冒号可能会违反这一要求，具体取决于结构化评分请求的呈现方式。数字参考证实了错误，而配对干预则将候选者的编辑与整合的呈现选择区分开。在200个先前未使用的DROP和GSM8K源集群中，该编辑使Jev的错误接受率从1.0%增加到26.0%（使用三个输出标签），并在已发布的四标签评分指令下，错误接受率从3.0%增加到26.5%（在排序的JSON键下）。在插入呈现下，这两种候选变体均被拒绝。这些交互通过了预先指定的统计修正，尽管Jev在两种评分配置和呈现下均达到了控制阈值。大多数额外的接受发生在被分配较大数值错误的候选者中。GPT-6 Sol未观察到提示条件下的错误接受，缺失的响应未得到解决。结果表明，基本的评分能力可以与对固定候选编辑在逻辑上等效的请求呈现下的脆弱性存在显著差异。参考意识的语法和复合排序变化限制了该发现的操作范围，并使其内部原因未被测量。

View on arXiv Download PDF AI Translation

cs.CL / 40 / 2610.11451

SAIL: Scientific Agentic Intelligence via a Science-Aware Loop

SAIL：通过科学感知循环实现科学代理智能

SAIL Model Team, Sun, Boyuan, Dai, Bryan, Liu, Che, Liu, Chi, Li, Derek, Piao, Hongming, Chen, Mengzhuo, Wang, Xidong, Shu, Yan, Chen, Yinda, Zeng, Ziyang

Abstract

We introduce SAIL, an open model with 35B total and 3B active parameters for literature research, scientific coding, and multi-step research workflows. SAIL is developed through a science-aware improvement loop: agents built on frontier AI models analyze its task failures and construct training tasks that address the underlying capability gaps. The diagnosis examines search and evidence selection in literature tasks, scientific assumptions and reasoning in coding, and planning and revision in longer investigations. The agents draw on paper collections and scientific code repositories to build problems, interaction trajectories, and executable tasks with the required environments and tools. We repeat this loop over multiple development cycles and train SAIL through supervised fine-tuning, specialist training, multi-teacher on-policy distillation, and agentic reinforcement learning. SAIL achieves competitive performance across scientific research tasks with substantially fewer parameters than leading open-weight models.

Chinese Translation

我们介绍了SAIL，这是一个用于文献研究、科学编码和多步骤研究工作流程的开放模型，具有350亿总参数和30亿活跃参数。SAIL通过一个科学感知改进循环进行开发：基于前沿人工智能模型构建的代理分析其任务失败，并构建解决潜在能力差距的训练任务。诊断过程考察文献任务中的搜索和证据选择、编码中的科学假设和推理，以及更长时间调查中的规划和修订。代理利用论文集合和科学代码库构建问题、交互轨迹和具有所需环境和工具的可执行任务。我们在多个开发周期中重复这一循环，并通过监督微调、专业培训、多教师在线蒸馏和代理强化学习训练SAIL。SAIL在科学研究任务中实现了具有竞争力的性能，且所需参数远少于领先的开放权重模型。

View on arXiv Download PDF AI Translation

cs.CL / 41 / 2610.11501

Beyond Sequences: Distilling Structured Decision Memory for LLM Recommendation

超越序列：提炼结构化决策记忆以进行大语言模型推荐

Liang, Leikun, Wang, Guoshuai, He, Xingsheng, Han, Yushan, Xuan, Yunyi, Xu, Xiaoxiao, Qu, Lin

Abstract

Despite the adoption of large language models (LLMs) in recommendation systems, prevailing approaches mostly model single-type behaviors (e.g., views or purchases). Even when incorporating multiple behaviors, existing methods flatten heterogeneous actions into homogeneous token sequences, ignoring their distinct decision-making roles. This flattening fails to capture semantic hierarchies and contextual nuances in complex decision-making, such as trade-offs between price and quality. Consequently, performance degrades in critical ``difficult-choice'' scenarios involving highly similar items. To bridge this gap, we propose MARI (Memory-Augmented Recommendation with Interpretability), which grounds predictions in explicit, structured decision evidence. MARI maintains a Decision Memory Bank (DMB) that archives users' past rationales as Structured Decision Memories (SDMs): concise records of goals, constraints, and trade-offs. These SDMs are generated offline via Post-Hoc Decision Distillation from heterogeneous behaviors and user-generated content. By retrieving relevant SDMs to augment LLM reasoning, MARI achieves interpretability and scalability without the prohibitive cost of processing long raw sequences. Extensive experiments show MARI significantly outperforms state-of-the-art baselines on standard next-item prediction and a newly introduced Difficult Choice Prediction task, incurring low latency overhead by decoupling memory construction from online inference. Qualitative analyses reveal actionable, human-readable insights into user decision-making, marking a concrete step toward reasoning-aware recommendation systems.

Chinese Translation

尽管大型语言模型（LLMs）在推荐系统中的应用日益普及，但现有的方法大多仅建模单一类型的行为（例如，浏览或购买）。即使在结合多种行为时，现有方法也将异构行为压平为同质的令牌序列，忽视了它们在决策中的不同角色。这种压平无法捕捉复杂决策中的语义层次和上下文细微差别，例如价格与质量之间的权衡。因此，在涉及高度相似项目的关键“困难选择”场景中，性能下降。为了弥补这一缺口，我们提出了MARI（具有可解释性的记忆增强推荐），该方法将预测建立在明确的、结构化的决策证据之上。MARI维护一个决策记忆库（DMB），该库存档用户过去的理由，作为结构化决策记忆（SDMs）：目标、约束和权衡的简明记录。这些SDMs通过对异构行为和用户生成内容进行后验决策提炼（Post-Hoc Decision Distillation）离线生成。通过检索相关的SDMs来增强LLM推理，MARI实现了可解释性和可扩展性，而无需处理长原始序列的高昂成本。大量实验表明，MARI在标准的下一个项目预测和新引入的困难选择预测任务上显著优于最先进的基线，并通过将记忆构建与在线推理解耦，带来了低延迟开销。定性分析揭示了可操作的、易于理解的用户决策洞察，标志着向具有推理能力的推荐系统迈出了重要一步。

View on arXiv Download PDF AI Translation

cs.CL / 42 / 2610.11510

Does Modern Standard Arabic (MSA) Dominate Arabic Dialects in LLMs? A Representation-Level Analysis

现代标准阿拉伯语（MSA）在大型语言模型中是否主导阿拉伯方言？一种表征层面的分析

Sallouh, Abdu, Popovič, Nicholas, Färber, Michael

Abstract

Large language models (LLMs) often default to Modern Standard Arabic (MSA) when generating Arabic, even when prompted with dialectal Arabic. A natural explanation is that their internal representations are dominated by MSA. We test this hypothesis by adapting the language-dominance framework of Shani and Basirat (2025) (https://doi.org/10.18653/v1/2025.blackboxnlp-1.7) to 26 Arabic varieties. Across layers and model families, we find no evidence that MSA acts as a dominant internal representation for Arabic dialects. Instead, dialect representations form a dense and highly overlapping space: normalized mutual information drops sharply compared to patterns reported for more distinct languages. Moreover, the strongest separability effects are not confined to intermediate layers, but can shift toward later layers depending on the architecture. These findings challenge a common interpretation of MSA-biased generation: output preference does not necessarily reveal internal representational dominance. Analyses of multilingual and dialectal LLMs should therefore distinguish generation bias from the geometry of internal representations.

Chinese Translation

大型语言模型（LLMs）在生成阿拉伯语时通常默认使用现代标准阿拉伯语（MSA），即使在提示方言阿拉伯语时也是如此。一个自然的解释是，它们的内部表征被MSA主导。我们通过将Shani和Basirat（2025）提出的语言主导性框架（https://doi.org/10.18653/v1/2025.blackboxnlp-1.7）适用于26种阿拉伯方言来检验这一假设。在不同层次和模型家族中，我们没有发现MSA作为阿拉伯方言的主导内部表征的证据。相反，方言表征形成了一个密集且高度重叠的空间：归一化互信息与更为明显的语言相比急剧下降。此外，最强的可分离性效应并不局限于中间层，而是可以根据架构转移到后续层。这些发现挑战了对MSA偏向生成的常见解释：输出偏好并不一定揭示内部表征的主导性。因此，对多语言和方言LLMs的分析应区分生成偏差与内部表征的几何结构。

View on arXiv Download PDF AI Translation

cs.CL / 43 / 2610.11519

Residual Advantage: Student-Relative Teacher Guidance for RL with Verifiable Rewards

残差优势：基于学生相对教师指导的可验证奖励强化学习

Chen, Xiaobing, Pang, Zhiqi

Abstract

Reinforcement learning with verifiable rewards (RLVR) and on-policy distillation (OPD) have become two main paradigms for post-training reasoning models. RLVR gives each response a single outcome label, leaving the steps inside it without separate credit. OPD provides token-level guidance at student-visited prefixes, but its pointwise signal does not directly reflect the pattern of teacher--student disagreement across the vocabulary. Dense, unbounded log-ratio supervision can amplify the teacher's influence, yet a strong solver is not necessarily a suitable guide when the student's solution paths depart from the teacher's. We propose Residual Advantage (\RA{}), which treats the teacher--student probability residual as a bounded one-step reward, subtracts the corresponding state value under the student policy to form a standard advantage, and centers the result within each response before adding it to the verifier advantage. The guidance term has zero mean within each response, so the verifier advantage remains the response's mean label and the teacher only redistributes credit among the steps within it. \CoRA{} further updates a teacher LoRA with verifier advantages on the same scored student batch and uses the updated teacher in the next iteration's residual, adapting guidance to the student's attempts. With Qwen3-1.7B-Base and Qwen3-4B-Base students and a Qwen3-8B teacher, \RA{} combined with GRPO or REINFORCE++ improves the underlying sequence-advantage algorithm in all 24 comparisons on three mathematical benchmarks, raising macro Avg@8 by 1.7--3.6 points and Pass@8 by 3.9--6.3 points. Both combinations surpass teacher-only OPD, and \CoRA{} adds a further 1.0--1.5 Avg@8 points.

Chinese Translation

具有可验证奖励的强化学习（RLVR）和在线蒸馏（OPD）已成为后训练推理模型的两个主要范式。RLVR为每个响应提供一个单一的结果标签，但其内部步骤没有单独的信用。OPD在学生访问的前缀上提供了令牌级别的指导，但其逐点信号并未直接反映教师与学生在词汇上的不一致模式。密集的、无界的对数比监督可以放大教师的影响，然而，当学生的解决路径偏离教师时，强大的求解器并不一定是合适的指导。我们提出了残差优势（Residual Advantage， extit{RA}），将教师与学生的概率残差视为一个有界的一步奖励，减去学生策略下相应状态值以形成标准优势，并在每个响应内对结果进行中心化，然后将其添加到验证者优势中。指导项在每个响应内的均值为零，因此验证者优势保持为响应的均值标签，教师仅在其中重新分配信用。 extit{CoRA}进一步在同一评分学生批次上使用验证者优势更新教师的LoRA，并在下一次迭代的残差中使用更新后的教师，适应学生的尝试。使用Qwen3-1.7B-Base和Qwen3-4B-Base学生以及Qwen3-8B教师， extit{RA}结合GRPO或REINFORCE++在三个数学基准的24个比较中改善了基础序列优势算法，提高了宏观Avg@8的得分1.7至3.6分，Pass@8的得分提高了3.9至6.3分。两种组合均超越了仅使用教师的OPD，而 extit{CoRA}进一步增加了1.0至1.5的Avg@8分。

View on arXiv Download PDF AI Translation

cs.CL / 44 / 2610.11520

When Can You Prune Your Network? A Study of Intermediate Neurons in Multilingual Speech Parsing

何时可以剪枝网络？多语言语音解析中中间神经元的研究

Kabra, Minnie, Lecouteux, Benjamin, Coavoux, Maximin

Abstract

End-to-end speech parsing, a task recently proposed, consists in predicting both the transcription and the syntactic tree for a spoken utterance. Existing architectures for speech parsing often utilise intermediate neural networks. In this work, we examine the effectiveness of intermediate neural networks (NN) for parsing, and, specifically, what role do they play. We introduce a simpler end-to-end architecture for speech parsing, where we remove these intermediate NN units, reducing the parameters by 12%, while achieving comparable or better performance than prior method on both automatic speech recognition (ASR) and parsing. We demonstrate that intermediate NN units help reduce the representational gap when the pre-trained encoder is frozen. We do a comprehensive evaluation of speech parsing on French, and medium-low resource languages Slovenian and Naija. We further investigate the impact of the training data size and intermediate layers of the pretrained speech encoder on speech parsing.

Chinese Translation

端到端语音解析是一项最近提出的任务，旨在预测口语表达的转录文本和句法树。现有的语音解析架构通常利用中间神经网络。在本研究中，我们考察了中间神经网络（NN）在解析中的有效性，特别是它们所扮演的角色。我们提出了一种更简单的端到端语音解析架构，去除了这些中间NN单元，参数减少了12%，同时在自动语音识别（ASR）和解析方面实现了与先前方法相当或更好的性能。我们证明了中间NN单元在冻结预训练编码器时有助于减少表征差距。我们对法语以及中低资源语言斯洛文尼亚语和奈及利亚语的语音解析进行了全面评估。我们进一步研究了训练数据规模和预训练语音编码器的中间层对语音解析的影响。

View on arXiv Download PDF AI Translation

cs.CL / 45 / 2610.11542

Constitutional Gating and Deterministic Recovery for Multi-Agent LLM Negotiation: Ablations Against a Stateful Adversarial Gatekeeper

多智能体大语言模型谈判中的宪法门控与确定性恢复：针对有状态对抗门卫的消融实验

Nakatsu, Masaaki, Wang, Reno

Abstract

Multi-agent LLM systems negotiating with a stateful counterpart waste model calls in three ways: polite loops that never meet the counterpart's hidden acceptance condition, malformed outputs that trigger retries, and compliance deadlocks in which the counterpart demands something the agent must refuse. We study a three-part control stack - a 5-Pillar runtime constitution, a 4-tier swarm (Director, three-agent majority vote, Monitor, schema hard gate) and Cognitive Annealing (deterministic deadlock detection, atomic purge of the agent-side context, a canonical recovery message) - against a released adversarial Gatekeeper whose acceptance rules are fixed regular expressions and whose LLM only renders reply text. The testbed has a known solution: it measures whether the stack executes a constitution-aligned strategy against swarm drift and recovers from deadlock, not whether it discovers anything. In five runs per configuration (30 runs; Gemini 2.5 Pro agents, Claude Haiku 4.5 Gatekeeper) we find: (i) the constitution and Director make an acceptable framing possible but not reliable - 0/5 baseline unlocks versus 1/5 and 2/5 with the constitution; when the swarm unlocks it does so in one turn with 7-8 calls and about 15k tokens (67-73% below baseline); when it does not, it costs 17-38% more; (ii) the Monitor and hard gate do not reduce unlocks and leave an audit trail; (iii) under a honeytrap-to-compliance deadlock, LLM-only steering escapes 0 of 5 times while atomic purge plus a canonical strike escapes 5 of 5 (Fisher $p = 0.008$) at the same call budget, with zero calls for the strike. LLM-written strikes failed the deterministic pre-flight 5 of 5 times although an LLM Monitor had approved four. Pre-registered hypotheses on average call and token reduction were not supported. Cost is bounded in every arm by deterministic stop rules; the stack adds recovery at no extra model cost.

Chinese Translation

与有状态对手进行谈判的多智能体大语言模型系统在三个方面浪费模型调用：礼貌循环未能满足对手隐藏的接受条件、触发重试的格式错误输出，以及对手要求智能体必须拒绝的合规死锁。我们研究了一个三部分的控制栈——一个5柱运行时宪法、一个4层群体（指挥者、三智能体多数投票、监控器、模式硬门）和认知退火（确定性死锁检测、智能体端上下文的原子清除、规范恢复消息）——针对一个已发布的对抗性门卫，其接受规则是固定的正则表达式，且其大语言模型仅渲染回复文本。测试平台有一个已知解决方案：它测量栈是否执行与宪法对齐的策略以抵御群体漂移并从死锁中恢复，而不是测量其是否发现任何内容。在每个配置下进行五次运行（共30次运行；Gemini 2.5 Pro智能体，Claude Haiku 4.5门卫），我们发现：（i）宪法和指挥者使得可接受的框架成为可能，但不可靠——基线解锁0/5次，而宪法下为1/5和2/5；当群体解锁时，通常在一次回合内完成，调用次数为7-8次，约15,000个标记（比基线低67-73%）；当未解锁时，成本增加17-38%；（ii）监控器和硬门并未减少解锁次数，并留下审计轨迹；（iii）在蜜罐到合规死锁下，仅使用大语言模型的引导在5次中逃脱0次，而原子清除加规范打击在同一调用预算下逃脱5次（Fisher $p = 0.008$），且打击没有调用次数。大语言模型编写的打击在确定性预飞行中失败5次，尽管一个大语言模型监控器已批准了四次。预注册的关于平均调用和标记减少的假设未得到支持。每个臂的成本都受到确定性停止规则的限制；该栈在没有额外模型成本的情况下增加了恢复功能。

View on arXiv Download PDF AI Translation

cs.CL / 46 / 2610.11543

Learning the Loop, Not Just the Page: Execution-Grounded Loop Learning for Web Generation

学习循环，而不仅仅是页面：基于执行的循环学习用于网页生成

Meng, Yuxin, Zhang, Ruixu, Wang, Junjie, Suo, Yuhan, Sun, Yuhan, Hu, Ruining, Yu, Yiyao, Wang, Yubin, Ruan, Shouwei, Wang, Bin, Liao, Yue, Zhang, Yuxiang, Yang, Yujiu

Abstract

Functional Web generation is increasingly optimized with executable rewards, yet existing methods largely focus on the quality of the final page and leave the process of diagnosing and repairing imperfect implementations underexplored. We identify a central challenge in this setting: the Generator and Refiner produce executable artifacts with direct environment rewards, whereas the intermediate Critic influences downstream behavior without a directly executable outcome. We introduce WebLoop, an execution-grounded framework that jointly learns generation, critique, and refinement within a shared policy. WebLoop trains an execution-free Critic with complementary signals for requirement-level discriminability and downstream helpfulness, first establishing reliable diagnosis and then introducing consequence-aware credit, while all three roles are jointly optimized with group-relative policy learning. With Qwen3.5-9B, WebLoop reaches 41.5 Overall on WebRise and 38.9% accuracy on WebGen-Bench, improving the base model by 11.3 and 15.4 points, respectively. The gains transfer to first-pass generation, persist at 27B scale, and generalize from text-only training to multimodal inputs. Controlled analyses further show that the improvement cannot be explained by an additional refinement pass alone, highlighting the importance of learning the Critic and the loop itself.

Chinese Translation

功能性网页生成越来越多地通过可执行奖励进行优化，但现有方法主要关注最终页面的质量，而对诊断和修复不完善实现的过程探讨不足。我们在这一背景下识别出一个核心挑战：生成器（Generator）和精炼器（Refiner）产生具有直接环境奖励的可执行工件，而中间的批评者（Critic）则在没有直接可执行结果的情况下影响下游行为。我们提出了WebLoop，这是一个基于执行的框架，能够在共享策略中共同学习生成、批评和精炼。WebLoop训练一个无执行的批评者，利用互补信号来提高需求级别的可区分性和下游的有用性，首先建立可靠的诊断，然后引入关注后果的信用，同时这三个角色通过群体相对策略学习进行联合优化。使用Qwen3.5-9B，WebLoop在WebRise上达到41.5的总体分数，在WebGen-Bench上达到38.9%的准确率，分别提高了基础模型11.3和15.4个点。这些提升在首次生成中得以转移，在27B规模下持续存在，并且从仅文本训练推广到多模态输入。控制分析进一步表明，改进不能仅通过额外的精炼过程来解释，突显了学习批评者和循环本身的重要性。

View on arXiv Download PDF AI Translation

cs.CL / 47 / 2610.11544

Prosody-to-Text: Predicting text from low-pass filtered speech

韵律到文本：从低通滤波语音预测文本

Porteš, David, Horák, Aleš

Abstract

While predicting prosody from text is an established task in the field, the opposite direction, predicting text that fits a given prosodic pattern, remains largely overlooked. We find this unfortunate, because this opposite direction could lead to some very interesting use cases. Therefore, in this paper, we make the first steps in the prosody-to-text direction by inves- tigating how much of the original sentence can be recovered from its prosodic pattern. To this end, we fine-tune the Whis- per model using only the 12 lowest Mel bins (low-pass filter with approximately 450Hz cutoff), and obtain surprisingly accurate results (WER 36%), with 10% of utterances be- ing recovered perfectly, and 40% of utterances having Word Error Rate at or below 25%. We also find that, given the correct prefix, the next token was predicted correctly in 79% of cases. Our results suggest that the relationship between low-frequency speech features and lexical content is much stronger than previously thought, and we believe that direct- ing more attention to this topic might open the door to new applications, such as using prosody to guide text generation of modern LLMs

Chinese Translation

虽然从文本预测韵律在该领域已是一个成熟的任务，但相反的方向，即预测符合给定韵律模式的文本，仍然在很大程度上被忽视。我们认为这很不幸，因为这一相反方向可能会带来一些非常有趣的应用场景。因此，在本文中，我们在韵律到文本的方向上迈出了第一步，研究了从其韵律模式中可以恢复多少原始句子。为此，我们仅使用12个最低的梅尔频率（Mel bins）对Whisper模型进行了微调（低通滤波，截止频率约为450Hz），并获得了令人惊讶的准确结果（字错误率为36%），其中10%的语句被完美恢复，40%的语句的字错误率在25%或以下。我们还发现，在给定正确前缀的情况下，下一标记在79%的情况下被正确预测。我们的结果表明，低频语音特征与词汇内容之间的关系比之前认为的要强得多，我们相信，更多关注这一主题可能会为新应用打开大门，例如利用韵律指导现代大语言模型（LLMs）的文本生成。

View on arXiv Download PDF AI Translation

cs.CL / 48 / 2610.11559

SWE-Journey: Towards More Realistic Evaluation of Coding Assistants through Long-Horizon, Multi-Turn Interaction

SWE-Journey：通过长时间、多轮交互实现对编码助手更现实的评估

Deng, Hexuan, Wang, Yue, Jiang, Wenyu, Yang, Cheng, Yang, Haolin, Zhang, Zhaohua, Zhao, Chenchen, Chen, Beiduo, Chen, Muxi, Zhu, Sa, Zhu, Geyuan, Zhuo, Jianhuan, Xiao, Qiuyong, Jiang, Tianwen, Zhang, Jihong, Liu, Xuebo

Abstract

Coding assistants such as Claude Code and Codex have become a major application of LLM agents, yet existing benchmarks remain far from real-world use, particularly in task horizon and interaction length. Code assistants require completing long chains of development work in continuously evolving repositories, while repeatedly clarifying requirements and adapting implementations through multi-turn interaction. To address these gaps, we introduce SWE-Journey, a benchmark for more realistic evaluation of coding assistants. To address the task-horizon gap, we propose a weak-to-strong synthesis pipeline that automatically constructs long-horizon coding tasks. To address the interaction gap, we mine four representative user personas from real interaction data and build a user-simulation agent to reproduce realistic code-assistance interactions. On average, models pass over 75% of tests for requested functionality with software architects, but fewer than 25% with non-coders. These results show that current coding assistants still fall short of enabling reliable coding for non-coders. We further analyze the reasons for this gap and identify asking right, finding right, and fixing right as key capabilities during interaction.

Chinese Translation

编码助手如Claude Code和Codex已成为大型语言模型（LLM）代理的主要应用，但现有基准测试与现实世界的使用仍相去甚远，特别是在任务时间跨度和交互长度方面。编码助手需要在不断发展的代码库中完成长链的开发工作，同时通过多轮交互不断澄清需求和调整实现。为了解决这些问题，我们提出了SWE-Journey，一个用于更现实评估编码助手的基准测试。为了解决任务时间跨度问题，我们提出了一种弱到强的合成管道，自动构建长时间跨度的编码任务。为了解决交互问题，我们从真实交互数据中挖掘出四种具有代表性的用户角色，并构建了一个用户模拟代理，以重现真实的代码辅助交互。平均而言，模型在软件架构师的请求功能测试中通过率超过75%，而在非编码者中则低于25%。这些结果表明，当前的编码助手仍未能为非编码者提供可靠的编码支持。我们进一步分析了这一差距的原因，并确定在交互过程中“问对”、“找对”和“修对”是关键能力。

View on arXiv Download PDF AI Translation

cs.CL / 49 / 2610.11566

Incremental Open-Ended Deep Research with Structured Harness

增量开放式深度研究与结构化工具

Chen, Meilin, Bao, Hongyuan

Abstract

Existing Open-Ended Deep Research (OEDR) systems primarily generate reports from scratch, making them inefficient for scenarios where research reports need to be continuously maintained as new information emerges. We introduce \textbf{Incremental Open-Ended Deep Research (Incremental-OEDR)}, a research setting that treats a report as an evolving research state and incrementally updates it by preserving valid knowledge, revising outdated or incomplete content, and incorporating newly available information. To support this setting, we propose \textbf{Structured Harness}, which represents reports as structured collections of outlines, sections, and supporting evidence, and provides structured retrieval, a persistent structured evidence pool, and structured generation for selective report updating and evidence reuse. We further establish a temporal evaluation framework spanning ten years, with \emph{Single-Step Task} and \emph{Long-Chain Task} to evaluate incremental updates over both individual transitions and long-term update chains. Extensive Experiments on DeepResearch Bench and DeepConsult under both the Open-source Configuration (OC) and Proprietary Configuration (PC) show that Incremental-OEDR maintains competitive report quality while substantially improving report continuity and reducing research costs. As shown in Figure~\ref{fig:profile}, it achieves up to 0.51 higher content-level ROUGE-L F1, 0.63 higher outline-level EM F1, 33\% lower token consumption, and 61\% fewer search calls than OEDR on DeepResearch Bench. For more details, please refer to our project page: https://ioedr-project.github.io/.

Chinese Translation

现有的开放式深度研究（Open-Ended Deep Research, OEDR）系统主要是从零开始生成报告，这使得它们在需要随着新信息的出现持续维护研究报告的场景中效率低下。我们提出了 extbf{增量开放式深度研究（Incremental Open-Ended Deep Research, Incremental-OEDR）}，这一研究设置将报告视为一个不断演变的研究状态，通过保留有效知识、修订过时或不完整的内容以及纳入新获得的信息来进行增量更新。为了支持这一设置，我们提出了 extbf{结构化工具（Structured Harness）}，该工具将报告表示为结构化的提纲、章节和支持证据的集合，并提供结构化检索、持久的结构化证据池以及用于选择性报告更新和证据重用的结构化生成。我们进一步建立了一个跨越十年的时间评估框架，包括 extit{单步任务（Single-Step Task）}和 extit{长链任务（Long-Chain Task）}，以评估个体过渡和长期更新链上的增量更新。在开放源配置（Open-source Configuration, OC）和专有配置（Proprietary Configuration, PC）下，在DeepResearch Bench和DeepConsult上的广泛实验表明，Incremental-OEDR在保持竞争性报告质量的同时，显著提高了报告的连续性并降低了研究成本。如图~ ef{fig:profile}所示，它在DeepResearch Bench上实现了高达0.51的内容级ROUGE-L F1提升、0.63的提纲级EM F1提升、33 ext{%}的令牌消耗降低，以及61 ext{%}的搜索调用减少。更多细节请参见我们的项目页面：https://ioedr-project.github.io/。

View on arXiv Download PDF AI Translation

cs.CL / 50 / 2610.11575

Smoothing the Top-k Exposure Boundary for Sparse Mixture-of-Experts

平滑稀疏专家混合模型的 Top-k 曝露边界

Chai, Yunkai, Zhu, Tong, Qu, Xiaoye, Hu, Xuyang, Chen, Guanjie, Guo, Qipeng, Cheng, Yu

Abstract

Sparse Mixture-of-Experts models scale parameter capacity efficiently while maintaining a fixed compute budget per token. However, traditional training paradigms enforce a static choice of top-$k$ experts, which converts a continuous routing distribution into a rigid step function. This constraint introduces a brittle boundary where highly competitive experts are arbitrarily separated into full-supervision and zero-feedback zones based on minor score fluctuations. To address this issue, we propose Elastic Expert Routing, which stochastically samples the active expert budget from a localized discrete distribution centered at $k$. Over multiple training iterations, this mechanism softens the sharp threshold into a gradual probability distribution. Because the sampling neighborhood remains symmetric, this approach matches the expected computational cost of deterministic training, while preserving the inference budget. Extensive experiments demonstrate the efficacy of our method on both supervised fine-tuning and from-scratch pretraining settings. During supervised fine-tuning, elastic routing improves downstream macro-averages on OLMoE-1B-7B and Qwen3-30B-A3B by $+0.84$ and $+2.02$ points, respectively. In addition, in from-scratch pretraining, it outperforms the static top-$k$ baseline by $1.6$ points on average across downstream tasks.

Chinese Translation

稀疏专家混合模型有效地扩展了参数容量，同时保持每个标记的固定计算预算。然而，传统的训练范式强制选择静态的 Top-$k$ 专家，这将连续的路由分布转化为刚性的阶跃函数。这一约束引入了一个脆弱的边界，在该边界上，高度竞争的专家根据微小的分数波动被任意分隔为全监督区和零反馈区。为了解决这个问题，我们提出了弹性专家路由（Elastic Expert Routing），该方法从以 $k$ 为中心的局部离散分布中随机抽样活跃专家预算。在多个训练迭代中，该机制将尖锐的阈值软化为渐进的概率分布。由于抽样邻域保持对称，这种方法在保持推理预算的同时，匹配了确定性训练的预期计算成本。大量实验表明，我们的方法在监督微调和从头预训练设置中均表现出色。在监督微调过程中，弹性路由使 OLMoE-1B-7B 和 Qwen3-30B-A3B 的下游宏平均分别提高了 $+0.84$ 和 $+2.02$ 个百分点。此外，在从头预训练中，它在下游任务上平均超越了静态 Top-$k$ 基线 $1.6$ 个百分点。

View on arXiv Download PDF AI Translation

cs.CL / 51 / 2610.11585

Adapting English Quality Classifiers for Multilingual LLM Pretraining Data Selection

为多语言大语言模型预训练数据选择调整英语质量分类器

Sabolčec, Vinko, Messmer, Bettina, Turki, Yassine, Jaggi, Martin

Abstract

Recent advances in large language model (LLM) pretraining highlight the role of high-quality training data in improving performance. While model-based filtering has proven effective in selecting high-quality subsets from web-scale corpora, especially for high-resource languages, low-resource languages face challenges due to limited availability of annotated data. This work explores extending quality filtering to over 100 languages by proposing a multilingual adaptation approach that converts an existing English quality classifier into a multilingual variant. Our approach proposes training a small multi-layer perceptron on top of Transformer encoder-only model embeddings, using multilingual text as input and scores obtained from English classifiers applied to machine-translated text as labels. Our 1B, 3B and 8B scale experiments show that our approach maintains the downstream LLM benchmark performance of existing multilingual model-based filtering baselines, without harming regional and cultural knowledge benchmarks. To further evaluate cross-lingual generalization, we compare classifier scores of high-quality synthetic data and web samples, and the correlation of classifier scores with LLM-based ones, revealing that the classifier can learn the scoring criteria of its original English variant, even for languages not included in its training data.

Chinese Translation

最近在大语言模型（LLM）预训练方面的进展突显了高质量训练数据在提升性能中的重要性。尽管基于模型的过滤在从网络规模语料库中选择高质量子集方面已被证明有效，尤其是对于高资源语言，但由于标注数据的有限可用性，低资源语言面临挑战。本研究探讨通过提出一种多语言适应方法，将现有的英语质量分类器转换为多语言变体，从而扩展质量过滤到超过100种语言。我们的方法建议在Transformer编码器模型嵌入的基础上训练一个小型多层感知机，以多语言文本作为输入，并使用应用于机器翻译文本的英语分类器获得的分数作为标签。我们的1B、3B和8B规模实验表明，我们的方法在不损害区域和文化知识基准的情况下，保持了现有多语言模型基于过滤基线的下游LLM基准性能。为了进一步评估跨语言泛化，我们比较了高质量合成数据和网络样本的分类器分数，以及分类器分数与基于LLM的分数之间的相关性，揭示了分类器能够学习其原始英语变体的评分标准，即使对于未包含在其训练数据中的语言也是如此。

View on arXiv Download PDF AI Translation

cs.CL / 52 / 2610.11586

Measuring Cultural Alignment Beyond the Average: A Framework for Evaluating Maternal-Health LLM Interactions in Indian Contexts

超越平均水平的文化对齐测量：评估印度背景下母婴健康大型语言模型互动的框架

Izhar, Umaira, Arora, Gunjan, Singh, Pushpendra

Abstract

Existing evaluation methods for healthcare LLMs primarily assess factual correctness,safety, and fluency, while providing limited insight into whether generated interactions reflect culturally situated healthcare reasoning. This limitation is particularly important in maternal health, where care decisions are shaped by social and relational norms. We introduce MH-INDIC, a culturally grounded evaluation framework for maternal-health interactions in urban and semi-urban North Indian contexts that operationalises cultural behaviour through ten dimensions of maternal-health reasoning. Using a 26-item survey administered to 102 pregnant and postpartum women from urban and semi-urban North India, we evaluate ten LLMs. We distinguish population level cultural alignment from profile-level behavioural variation. Although several models approximate the human population-level distribution, all evaluated systems exhibit substantially lower variation across demographic and household profiles than the human cohort, revealing a gap between aggregate alignment and profile-conditioned sensitivity. As a downstream application of MH-INDIC, we use the strongest-aligned proprietary and open-source models to generate culturally conditioned maternal-health dialogues under zero-shot, self-conditioned, and human-grounded prompting. Human-grounded conditioning produces stronger profile alignment and dialogue quality ratings, suggesting that measured cultural profiles can improve the cultural grounding of generated interactions

Chinese Translation

现有的医疗保健大型语言模型（LLMs）评估方法主要评估事实正确性、安全性和流畅性，但对生成的互动是否反映文化背景下的医疗推理提供的见解有限。这一局限性在母婴健康领域尤为重要，因为护理决策受到社会和关系规范的影响。我们提出了MH-INDIC，这是一个针对北印度城市和半城市背景下母婴健康互动的文化基础评估框架，通过十个母婴健康推理维度来操作化文化行为。我们使用一项包含26个条目的调查，对102名来自北印度城市和半城市的孕妇及产后女性进行评估，评估十个LLMs。我们区分了人口层面的文化对齐与个人层面的行为变异。尽管多个模型近似于人类人口层级的分布，但所有评估的系统在不同的人口和家庭特征之间的变异性明显低于人类群体，揭示了总体对齐与基于特征的敏感性之间的差距。作为MH-INDIC的下游应用，我们使用最强对齐的专有和开源模型，在零样本、自我条件和人类引导的提示下生成文化条件的母婴健康对话。人类引导的条件产生了更强的个人特征对齐和对话质量评分，表明测量的文化特征可以改善生成互动的文化基础。

View on arXiv Download PDF AI Translation

cs.CL / 53 / 2610.11592

Probing for Long-Horizon Deductive Reasoning Capabilities in Language Models with Prolog

使用 Prolog 探索语言模型的长远推理能力

Al-Negheimish, Hadeel, Ilieva, Jasna, Kim, Yoon

Abstract

Current frontier LLMs can theoretically process long contexts with 1M tokens or more. But to what extent can they go beyond simple retrieval and perform deeper reasoning over such long contexts? We empirically investigate long-horizon reasoning capabilities of LLMs, focusing on deductive logic expressed in Prolog. We construct ProloNg, a synthetic testbed to probe Prolog Long Reasoning, which systematically varies the complexity (reasoning depth) of problems, where the hardest case has a reasoning depth of 22 and 62k context length. We study 8 reasoning models across 5 families of frontier LLMs, and find that performance degrades substantially as reasoning depth grows, with the majority of models approaching chance beyond depth 10.

Chinese Translation

当前前沿的大型语言模型（LLMs）理论上可以处理超过 100 万个标记的长上下文。但它们在多大程度上能够超越简单的检索，并对如此长的上下文进行更深层次的推理呢？我们实证研究了 LLMs 的长远推理能力，重点关注用 Prolog 表达的演绎逻辑。我们构建了 ProloNg，一个合成测试平台，以探测 Prolog 长推理，系统地变化问题的复杂性（推理深度），其中最难的案例具有 22 的推理深度和 62,000 的上下文长度。我们研究了 5 个家族中的 8 种推理模型，发现随着推理深度的增加，性能显著下降，大多数模型在深度超过 10 时接近随机猜测。

View on arXiv Download PDF AI Translation

cs.CL / 54 / 2610.11599

Large Language Model Turnover Undermines Screening for Artificial Intelligence-Assisted Scientific Writing

大型语言模型的更替削弱了人工智能辅助科学写作的筛选

Nakajima, Kazuki, Mizuno, Takayuki

Abstract

Journals and conferences have begun to screen submitted manuscripts for text written using large language models (LLMs). The reliability of this screening rests on benchmark evaluations against a fixed set of LLM versions, while the versions in actual use keep changing. Here we quantify how this LLM turnover affects the screening of scientific manuscripts. We paired 4,000 pre-ChatGPT abstracts from the Proceedings of the National Academy of Sciences with their rewrites by 23 LLM versions from three vendors, released between June 2023 and August 2026. We then trained detectors under maintenance scenarios ranging from a detector retrained on every new version to one trained once and never updated. Detectors trained only on a vendor's past versions can collapse at the boundaries between model generations: calibrated to falsely flag 1% of human-written abstracts, they catch above 99% of rewrites just before the sharpest boundary and 3.8% just after it. Detectors trained on later versions can also miss rewrites of earlier ones. Vocabulary differences between versions largely track where detection transfers and where it fails. In the two screening scenarios we simulated, screens covering all 23 versions either flagged one in eight human-written abstracts or missed one in three rewrites of the newest version. Indeed, a commercial detector missed most rewrites of the version just after the sharpest boundary while flagging almost no human-written abstracts. Research-integrity policy should therefore treat the benchmark accuracy of a detector as provisional, to be re-verified with every LLM release, including earlier versions.

Chinese Translation

期刊和会议已开始对提交的手稿进行筛选，以检测使用大型语言模型（LLMs）撰写的文本。这种筛选的可靠性依赖于对固定版本的LLM进行基准评估，而实际使用中的版本不断变化。在此，我们量化了LLM更替如何影响科学手稿的筛选。我们将4000个来自《国家科学院院刊》（Proceedings of the National Academy of Sciences）的ChatGPT发布前的摘要与23个来自三家供应商的LLM版本的重写进行配对，这些版本在2023年6月至2026年8月之间发布。随后，我们在不同的维护场景下训练了检测器，从每个新版本重新训练的检测器到只训练一次且从不更新的检测器。仅基于供应商过去版本训练的检测器可能在模型代际之间的边界处崩溃：经过校准后，错误标记1%的人类撰写的摘要，但在最尖锐的边界之前捕获超过99%的重写，而在其之后仅捕获3.8%。基于后续版本训练的检测器也可能错过早期版本的重写。版本之间的词汇差异在很大程度上决定了检测的转移和失败的地方。在我们模拟的两种筛选场景中，涵盖所有23个版本的筛选器要么标记每八个中就有一个人类撰写的摘要，要么错过每三个最新版本的重写中的一个。实际上，一个商业检测器在标记几乎没有人类撰写的摘要的同时，错过了大多数在最尖锐边界之后版本的重写。因此，研究诚信政策应将检测器的基准准确性视为临时的，需在每次LLM发布时重新验证，包括早期版本。

View on arXiv Download PDF AI Translation

cs.CL / 55 / 2610.11638

UXBench Pro: Benchmarking Personalized User Experience in Multi-Turn Dialogue Interactions

UXBench Pro：多轮对话交互中个性化用户体验的基准测试

Hong, Mengze, Lei, Zeyang, Shang, Wenbo, Zeng, Xia, Zhao, Xiying, Zhu, Qi, Zhang, Chen Jason, Jiang, Di, Fu, Taiming, Zhou, Qiongyi, Chang, Qinghe, Zhang, Fubao, Ma, Chenxuan, Peng, Minlong, Huang, Jinfeng, Zhou, Zineng, Wu, Jindou, Qi, Muge, He, Sijun, Cui, Xin, Liang, Di, Hua, Yuan, Chen, Davey

Abstract

Evaluating user experience (UX) with automated computational methods has gained increasing attention, supported by empirical evidence from UXBench. However, binary preference prediction provides limited insight, while relying on a single user-agnostic reward model overlooks the inherent heterogeneity of users, whose expectations can differ substantially. In this paper, we present UXBench Pro, comprising 1{,}000 test instances derived from real user interactions across 12 task scenarios and 82 domains. Each instance is paired with a FACTORS user profile that characterizes the user through seven interpretable behavioral facets, differentiating user groups. To provide richer evaluation insights, we introduce a dual-perspective paradigm that combines a personalized User Reward Model (URM) for third-person judgment with Sim4Eval, a user simulator that enables multi-turn interactions and provides first-person evaluation across four cognitive state dimensions. To assess the reliability of these based evaluators, we further introduce two meta-benchmarks, URMBench and USimBench, that evaluate how faithfully they reproduce real human preferences and behaviors. Extensive experiments reveal seven key findings that highlight the importance of user modeling and multi-perspective evaluation, offering a fresh perspective on user-centric benchmarking and motivating personalized model optimization.

Chinese Translation

使用自动化计算方法评估用户体验（UX）越来越受到关注，这得到了来自UXBench的实证支持。然而，二元偏好预测提供的洞察有限，而依赖单一的用户无关奖励模型则忽视了用户固有的异质性，因为用户的期望可能存在显著差异。本文提出了UXBench Pro，包含从12个任务场景和82个领域的真实用户交互中衍生的1,000个测试实例。每个实例都配有一个FACTORS用户档案，通过七个可解释的行为维度来描述用户，区分用户群体。为了提供更丰富的评估洞察，我们引入了一种双视角范式，结合了个性化用户奖励模型（User Reward Model, URM）用于第三方判断，以及Sim4Eval，一个用户模拟器，能够实现多轮交互并在四个认知状态维度上提供第一人称评估。为了评估这些基于模型的评估者的可靠性，我们进一步引入了两个元基准，URMBench和USimBench，评估它们在多大程度上忠实再现真实人类的偏好和行为。大量实验揭示了七个关键发现，突显了用户建模和多视角评估的重要性，为以用户为中心的基准测试提供了新视角，并激励个性化模型优化。

View on arXiv Download PDF AI Translation

cs.CL / 56 / 2610.11646

Phonologically Informed Tokenization for German Speech Recognition: A Cross-Domain Study

基于音位信息的德语语音识别分词：跨领域研究

Witzl, Christopher, Bocklet, Tobias, Riedhammer, Korbinian

Abstract

German is a morphologically rich language whose syllable structure is exceptionally well-predicted by the Knuth--Liang hyphenation algorithm. We ask whether phonologically informed tokenization can serve as a competitive target for end-to-end speech recognition. We compare three tokenizer families on the Omnilingual ASR wav2vec 2.0 backbone fine-tuned with CTC: the pretrained multilingual character inventory, a data-driven Byte-Pair Encoding (BPE) over orthography, and phonologically informed units from Pyphen syllabification and grapheme-to-phoneme conversion. Across 40 fine-tunes, we evaluate on three German test sets spanning orthogonal shifts: in-domain read speech, dialectal spontaneous speech, and standard-German spontaneous speech. In-domain, all phonologically informed tokenizers match BPE and the multilingual character baseline on both WER and CER. Under domain shift the picture splits along vocabulary size rather than the linguistic axis of variation: at small vocabularies, syllable-aware tokenization improves on dialectal speech, where phonetic surface forms vary but syllable structure is preserved, and stays ahead on spontaneous speech, where new word-forms violate vocabulary closure. A phoneme-level confusion analysis further shows that all tokenizers commit the same canonical function-word errors, indicating that the acoustic encoder, not the tokenizer, dominates the error topology. Our findings suggest that tokenizer choice may depend on the vocabulary budget as much as on the distribution shift expected at deployment rather than reducing to a single universal optimum.

Chinese Translation

德语是一种形态丰富的语言，其音节结构可以通过 Knuth--Liang 连字符算法进行非常准确的预测。我们探讨基于音位信息的分词是否可以作为端到端语音识别的一个竞争目标。我们比较了三种分词器家族，基于 Omnilingual ASR wav2vec 2.0 主干模型进行 CTC 微调：预训练的多语言字符库、基于正字法的数据驱动字节对编码（Byte-Pair Encoding, BPE），以及来自 Pyphen 音节划分和音素到音位转换的音位信息单元。在 40 次微调中，我们在三个涵盖正交变化的德语测试集上进行评估：领域内的朗读语音、方言自发语音和标准德语自发语音。在领域内，所有基于音位信息的分词器在字错误率（WER）和字符错误率（CER）上均与 BPE 和多语言字符基线相匹配。在领域转移下，结果依据词汇量的大小而非语言变化的轴线进行划分：在小词汇量下，音节感知分词在方言语音上有所改善，因为音位表面形式变化但音节结构得以保留，并且在自发语音中保持领先，因为新词形式违反了词汇闭合。音素级混淆分析进一步表明，所有分词器都犯下相同的规范功能词错误，表明声学编码器而非分词器主导了错误拓扑。我们的研究结果表明，分词器的选择可能与词汇预算和预期的分布转移一样重要，而不仅仅是简化为单一的通用最优解。

View on arXiv Download PDF AI Translation

cs.CL / 57 / 2610.11659

DIAL-OPD: Learning More from Fewer Tokens in On-Policy Distillation

DIAL-OPD：在策略蒸馏中从更少的标记中学习更多

Zhao, Anhao, Xin, Haoran, Tong, Junlong, Fan, Yingqi, Lu, Xuan, Nie, Ping, Li, Wenjie, Shen, Xiaoyu

Abstract

On-policy distillation (OPD) supervises student-generated trajectories with token-level teacher signals. Its sampled-token variant avoids the cost of full-vocabulary probabilities. Yet we find that training on fewer tokens can outperform full-token OPD, challenging the intuition that more supervision improves learning. This motivates selecting tokens by learning value. Existing disagreement-based criteria ignore probability scale: tokens assigned negligible probability by both models, termed low-low tokens, can receive large log-ratio rewards and hinder learning. We propose DIAL-OPD, a token-selection method that bridges log-probability and probability spaces by weighting reward magnitude with the logarithmic mean of teacher and student probabilities. A parameter beta controls this weighting, and the highest-scoring tokens are retained. Across 4 teacher-student pairs and 7 mathematical reasoning benchmarks, we compare DIAL-OPD with 9 baselines. Retaining only 40% of tokens, it outperforms Vanilla OPD and its full-token variants, with mean accuracy gains reaching 5.25 percentage points over Vanilla OPD, and doubles AIME25 Pass@16 from 13.33% to 26.67%. It also achieves up to an 18% relative improvement in mean accuracy over the strongest token-selection baseline at matched retention ratios. With a 4B teacher, DIAL-OPD surpasses the strongest full-token baseline using an 8B teacher at both student scales, showing that effective supervision allocation can outweigh teacher scaling. Further analysis shows that moderate beta balances suppressing low-low tokens against preserving useful disagreements. Token-level evidence reveals that DIAL-OPD filters high-reward tokens with limited reasoning value while preserving supervision critical to reasoning correctness.

Chinese Translation

策略蒸馏（On-policy distillation, OPD）通过标记级教师信号监督学生生成的轨迹。其采样标记变体避免了全词汇概率的成本。然而，我们发现，在更少的标记上训练可以超越全标记 OPD，这挑战了更多监督改善学习的直觉。这促使我们通过学习价值来选择标记。现有的基于分歧的标准忽略了概率规模：被两个模型分配了微不足道概率的标记，称为低-低标记，可以获得大的对数比奖励并阻碍学习。我们提出了 DIAL-OPD，一种标记选择方法，通过用教师和学生概率的对数均值加权奖励幅度，连接对数概率和概率空间。一个参数 beta 控制这种加权，得分最高的标记被保留。在 4 对教师-学生组合和 7 个数学推理基准上，我们将 DIAL-OPD 与 9 个基线进行比较。仅保留 40% 的标记时，它的表现超越了 Vanilla OPD 及其全标记变体，平均准确率提升达到 5.25 个百分点，并将 AIME25 Pass@16 从 13.33% 提升至 26.67%。在匹配的保留比率下，它还在平均准确率上相较于最强的标记选择基线实现了高达 18% 的相对提升。在使用 4B 教师时，DIAL-OPD 在两个学生规模上超越了使用 8B 教师的最强全标记基线，表明有效的监督分配可以超过教师规模的影响。进一步分析表明，适中的 beta 平衡了抑制低-低标记与保留有用分歧之间的关系。标记级证据显示，DIAL-OPD 过滤掉了具有有限推理价值的高奖励标记，同时保留了对推理正确性至关重要的监督。

View on arXiv Download PDF AI Translation

cs.CL / 58 / 2610.11678

TRACE: Diagnosing Verifier Brittleness in Agentic Evaluation

TRACE：诊断代理评估中的验证器脆弱性

Gaonkar, Radhika

Abstract

Verifier scores now serve as both benchmark metrics and training rewards for large language model (LLM) agents, and a change in score is routinely read as a change in capability. It may instead reflect a change in the evaluation. We introduce TRACE, a protocol that turns a score change from a verdict into a testable diagnosis: it applies a targeted change to one part of an evaluation, compares paired runs, checks whether the agent's behavior changed, and rescores unchanged trajectories to test whether the scoring rule is responsible. In a controlled suite of 25 synthetic tasks, renaming tools lowers a scripted agent's score by 0.250 even though it performs exactly the same operations; restoring the original names at scoring time closes the entire gap, while the same mutation exposes a genuine behavioral failure in a second agent. On public $\tau^2$-bench tasks with four LLM agents, an initial 30-task study finds mixed reward changes whose one clear effect does not replicate. In a larger follow-up on 88 new tasks with repeated runs per condition, renaming tools or reformatting tool outputs leaves reward unchanged to within $\pm$0.10 for seven of eight agent-change pairs, whereas tool names that deliberately mislead lower every agent's reward by 0.20-0.44, showing that the setup can detect real effects. Identical reruns flip 15-36% of task outcomes, so single-run comparisons cannot separate presentation effects from run-to-run variation. Two frontier LLM judges give consistent verdicts when a fixed trajectory is presented differently, yet disagree with each other on 57% of the same records, largely because one grades procedure rather than outcome. TRACE thus separates what a score change says about the agent from what it says about the measurement.

Chinese Translation

验证器评分现在既作为大型语言模型（LLM）代理的基准指标，也作为训练奖励，分数的变化通常被解读为能力的变化。然而，这可能反映的是评估的变化。我们提出了TRACE，一种将分数变化从裁决转化为可测试诊断的协议：它对评估的一个部分进行有针对性的改变，比较配对运行，检查代理的行为是否发生变化，并重新评分未变的轨迹以测试评分规则是否负责。在一个包含25个合成任务的受控实验中，重命名工具使得一个脚本代理的分数降低了0.250，尽管它执行的操作完全相同；在评分时恢复原始名称则弥补了整个差距，而同样的变更在第二个代理中暴露了真正的行为失败。在公共的$ au^2$-bench任务中，针对四个LLM代理的初步30任务研究发现了混合的奖励变化，其一个明显的效果并未复制。在对88个新任务的更大后续研究中，每个条件下重复运行，重命名工具或重新格式化工具输出使得奖励在八对代理变化中有七对保持在$ ext{±}0.10$以内不变，而故意误导的工具名称则使每个代理的奖励降低了0.20-0.44，显示该设置能够检测到真实效果。相同的重跑翻转了15-36%的任务结果，因此单次运行比较无法区分呈现效应与运行间变异。两个前沿的LLM评审在固定轨迹以不同方式呈现时给出了相似的裁决，但在57%的相同记录上彼此存在分歧，主要是因为一个评审关注过程而非结果。因此，TRACE将分数变化对代理的意义与对测量的意义区分开来。

View on arXiv Download PDF AI Translation

cs.CL / 59 / 2610.11695

Structured Sentiment Analysis Using Sequence Labeling as Dependency Graph Parsing

基于序列标注的依赖图解析的结构化情感分析

Imran, Muhammad, Ezquerro, Ana, Gómez-Rodríguez, Carlos, Søgaard, Anders, Vilares, David

Abstract

This study addresses the problem of structured sentiment analysis, whose goal is to obtain a fine-grained sentiment graph where the nodes represent spans of sentiment holders, targets, and expressions, while the arcs define the relationships among them. Our proposed approach casts the task as dependency graph parsing, but departs from traditional parsing methods by solving it through sequence labeling. To do so, we leverage recent advances in linearized graph encodings that allow each word in the input to be assigned a label, effectively capturing the structure of the dependency graph. We conducted experiments on seven datasets spanning five languages (English, Spanish, Norwegian, Basque, and Catalan), showing performance competitive with leading, more complex single-model approaches.

Chinese Translation

本研究解决了结构化情感分析的问题，其目标是获得一个细粒度的情感图，其中节点表示情感持有者、目标和表达的范围，而弧线定义它们之间的关系。我们提出的方法将该任务视为依赖图解析，但与传统的解析方法不同，通过序列标注来解决。为此，我们利用了线性化图编码的最新进展，使输入中的每个单词都可以被分配一个标签，从而有效捕捉依赖图的结构。我们在涵盖五种语言（英语、西班牙语、挪威语、巴斯克语和加泰罗尼亚语）的七个数据集上进行了实验，显示出与领先的、更复杂的单模型方法具有竞争力的性能。

View on arXiv Download PDF AI Translation

cs.CL / 60 / 2610.11765

Thinking Inertia: LLMs Keep Thinking When Told Not To

思维惯性：大型语言模型在被告知不思考时仍然继续思考

Lei, Dianqiao, Lin, Kevin Qinghong, Lu, Pan, Torr, Philip, Zou, James

Abstract

Large Language Models (LLMs) increasingly ship with explicit "thinking modes", yet their counterpart, "no-thinking", has received far less attention. We study LLMs' no-thinking behavior along two axes. a. How to measure no-thinking? Prior work typically defines no-thinking through proxies such as a disabled thinking mode or the absence of long traces. These proxies are unreliable: disabled thinking modes may still emit reasoning, while long traces may contain filler rather than genuine inference. We instead normalize each response into a pre-answer trace and final answer, and evaluate it at three levels: (i) Empty-Thinking Rate for strict answer-only compliance; (ii) instruction-aware Question-Pre-answer Relevance for similarity between the question and pre-answer trace; and (iii) LLM-as-judge Explicit Inference Rate for visible explicit inference. Together, these metrics distinguish answer-only output, relevant but non-inferential text, and explicit inference. b. How does no-thinking vary across tasks and models? We evaluate six prompting interventions on six LLMs across Boolean, multiple-choice, and open-ended questions. We find that explicit no-think controls cannot reliably eliminate visible inference. Models instead exhibit "Thinking Inertia": explicit inference persists even under strict controls and becomes more prevalent as the answer space opens. Accuracy remains stable on Boolean and multiple-choice tasks, whereas open-ended tasks reveal a trade-off between answer-only compliance and task accuracy. Rewriting the same questions across answer spaces shows that supplying candidate answers makes answer-only responses easier to produce. These findings establish no-thinking as a non-trivial capability: stopping explicit reasoning cannot be assumed from model settings or instructions alone and deserves systematic evaluation alongside reasoning ability.

Chinese Translation

大型语言模型（LLMs）越来越多地配备了明确的“思考模式”，而其对应的“无思考”模式却受到的关注远远不够。我们从两个方面研究LLMs的无思考行为。a. 如何衡量无思考？以往的研究通常通过代理指标来定义无思考，例如禁用思考模式或缺乏长时间的推理痕迹。这些代理指标并不可靠：禁用的思考模式仍可能产生推理，而长时间的痕迹可能包含填充内容而非真正的推理。我们则将每个响应标准化为预回答痕迹和最终回答，并在三个层面进行评估：（i）空思考率，用于严格的仅回答合规性；（ii）指令感知的问答预回答相关性，用于评估问题与预回答痕迹之间的相似性；（iii）LLM作为评判者的显式推理率，用于评估可见的显式推理。这些指标共同区分了仅回答的输出、相关但非推理的文本以及显式推理。b. 无思考在不同任务和模型之间如何变化？我们对六个LLM在布尔、选择题和开放式问题上进行了六种提示干预的评估。我们发现，显式的无思考控制无法可靠地消除可见推理。模型表现出“思维惯性”：即使在严格控制下，显式推理仍然存在，并且随着答案空间的扩大而变得更加普遍。在布尔和选择题任务中，准确性保持稳定，而开放式任务则揭示了仅回答合规性与任务准确性之间的权衡。对同一问题在不同答案空间中的重写显示，提供候选答案使得生成仅回答的响应变得更容易。这些发现确立了无思考作为一种非平凡能力：仅凭模型设置或指令无法假设停止显式推理，并且值得与推理能力一起进行系统评估。

View on arXiv Download PDF AI Translation

cs.CL / 61 / 2610.11766

Same Outcome, Different Evidence: Intent Recovery in LLM Safety Evaluation

相同结果，不同证据：大型语言模型安全评估中的意图恢复

Jiang, Haitong, Liu, Chunlin, Tang, Sihan, Wu, Chan, Su, Xiaoqing, Feng, Yuhong

Abstract

Safety evaluations of large language models commonly summarize harmful-output behavior with attack success rate (ASR). Yet the same non-harmful outcome can arise for very different reasons. A model may recover a harmful task and refuse it, fail to recover the task, or respond to something else entirely. Distinguishing these cases becomes especially important under intent-obscuring prompts, where a low ASR does not reveal whether the evaluated task was actually engaged. To make this distinction explicit, we pair ASR with operative understanding rate (UR), which measures whether a response both identifies the evaluated task and treats it as the task to be answered. Across interfaces, this paired view reveals substantial variation hidden by ASR: similar ASR values can correspond to sharply different recovery rates. Controlled English reconstructions show that recovery consistently improves as compressed prompts become more explicit, whereas ASR does not follow the same pattern. A complementary contrast comes from FormalLogic, where high recovery can still coincide with frequent harmful assistance. Together, these results show that non-harmful outcomes are not equally informative about model safety, motivating the joint reporting of intent recovery and ASR in LLM safety evaluation. Code and experiment inputs are available at https://github.com/kevinjiang0121-cyber/IRIS.

Chinese Translation

大型语言模型的安全评估通常通过攻击成功率（ASR）来总结有害输出行为。然而，相同的非有害结果可能由非常不同的原因引起。一个模型可能会恢复一个有害任务并拒绝它，可能无法恢复该任务，或者完全响应其他内容。在意图模糊的提示下，区分这些情况变得尤为重要，因为低ASR并不能揭示被评估的任务是否实际上被参与。为了明确这种区分，我们将ASR与操作理解率（UR）结合使用，后者衡量响应是否既识别了被评估的任务，又将其视为需要回答的任务。在不同接口中，这种配对视角揭示了ASR隐藏的显著变化：相似的ASR值可能对应于截然不同的恢复率。控制的英语重构显示，随着压缩提示变得更加明确，恢复率持续改善，而ASR并未遵循相同的模式。来自FormalLogic的对比结果表明，高恢复率仍然可能与频繁的有害协助共存。综合来看，这些结果表明，非有害结果在模型安全性方面并不具有相同的信息量，这促使在大型语言模型安全评估中共同报告意图恢复和ASR。代码和实验输入可在https://github.com/kevinjiang0121-cyber/IRIS获取。

View on arXiv Download PDF AI Translation

cs.CL / 62 / 2610.11776

DPPM: Dual-Path Parametric Memory for Personalized Language Models

DPPM：用于个性化语言模型的双路径参数记忆

Chen, Yuhao, Liu, Shuochen, Shi, Jiayao, Hong, Jian, Cheng, Chen, Ding, Xinyun, Wang, Tao, Li, Ya, Liu, Quan, Xu, Tong

Abstract

Long-term personalization requires language models to use interaction history to track users' preferences across sessions. Parametric memory encodes this interaction history into model parameters or adapters, reducing the need to include it in the inference context. However, independent context compilation leaves cross-session integration unspecified, while recurrent updates can attenuate earlier evidence. To address these challenges, we propose Dual-Path Parametric Memory (DPPM). Its Evidence path directly pools representations of the interaction history to preserve earlier evidence, while its Delta path sequentially updates an associative state to capture changes. Fusing both outputs produces history-conditioned LoRA adapters that combine evidence accumulation with ordered revision. Across multiple backbones, DPPM outperforms the evaluated baselines, achieving 54.22% on PersonaMem-v2 and 86.79% on PrefEval. These results suggest that DPPM provides a simple and effective design choice for cross-session personalized parametric memory.

Chinese Translation

长期个性化要求语言模型利用交互历史在会话之间跟踪用户的偏好。参数记忆将这一交互历史编码为模型参数或适配器，从而减少在推理上下文中包含该历史的需要。然而，独立的上下文编译未明确跨会话的整合，而递归更新可能会削弱早期证据。为了解决这些挑战，我们提出了双路径参数记忆（DPPM）。其证据路径直接汇聚交互历史的表示，以保留早期证据，而其增量路径则顺序更新关联状态以捕捉变化。融合这两种输出生成历史条件的LoRA适配器，将证据积累与有序修订相结合。在多个基础模型上，DPPM的表现优于评估的基线，在PersonaMem-v2上达到54.22%，在PrefEval上达到86.79%。这些结果表明，DPPM为跨会话个性化参数记忆提供了一种简单而有效的设计选择。

View on arXiv Download PDF AI Translation

cs.CL / 63 / 2610.11787

From Sparse Representations to Behavioral Insights for Multimodal Depression Assessment

从稀疏表示到多模态抑郁评估的行为洞察

Hu, Guimin, Song, Zihao, Luo, Jiachen, Xie, Jiayuan, Cai, Ruichu

Abstract

Multimodal depression assessment offers a promising approach to analyzing behavioral patterns associated with depression. However, existing methods often rely on dense and opaque multimodal representations, making it difficult to interpret the behavioral patterns underlying their predictions. In this work, we introduce BehavDep, a sparse factor-based framework that decomposes multimodal behavioral representations into sparse latent factors and associates them with behaviorally meaningful concepts through a semantic bridge. To address the mismatch between user-level annotations and heterogeneous video-level behaviors, BehavDep further learns video-level depression tendency scores under weak supervision and aggregates information across multiple observations for user-level assessment. Extensive experiments demonstrate that BehavDep achieves the best overall assessment performance while revealing complementary modality contributions, heterogeneous behavioral patterns across observations, and prediction responses to concept-level editing. These results show that BehavDep provides a structured and interpretable approach to analyzing multimodal behavioral representations for depression assessment.

Chinese Translation

多模态抑郁评估为分析与抑郁相关的行为模式提供了一种有前景的方法。然而，现有的方法往往依赖于密集且不透明的多模态表示，这使得难以解释其预测背后的行为模式。在本研究中，我们引入了BehavDep，一个基于稀疏因子的框架，该框架将多模态行为表示分解为稀疏潜在因子，并通过语义桥将其与具有行为意义的概念关联起来。为了解决用户级注释与异构视频级行为之间的不匹配，BehavDep进一步在弱监督下学习视频级抑郁倾向评分，并在多个观察中聚合信息以进行用户级评估。大量实验表明，BehavDep在整体评估性能上表现最佳，同时揭示了互补的模态贡献、观察间的异构行为模式以及对概念级编辑的预测响应。这些结果表明，BehavDep为分析多模态行为表示在抑郁评估中的应用提供了一种结构化且可解释的方法。

View on arXiv Download PDF AI Translation

cs.CL / 64 / 2610.11790

Easy to anticipate, hard to compute: boundary dependence finds the computed outputs that entropy patching misses

易于预测，难以计算：边界依赖发现熵修补遗漏的计算输出

Zúñiga, Nicolás Vera

Abstract

Byte-level language models such as the Byte Latent Transformer (BLT) group bytes into patches and run their large global model once per patch. BLT starts a patch where a small model's next-byte entropy is high, so global compute goes where the next byte is hard to predict. We show that this rule has a systematic blind spot: positions whose type is predictable but whose value must be computed, such as the number after "=" in a worked math solution. Under tight patch budgets, entropy-triggered layouts skip these positions, and accuracy on them collapses. In Meta's BLT-1B with patch starts on 10% of bytes, the entropy rule puts a patch start at 16% of the computed results in GSM8K solutions and gets 19.0% of them exactly right; a boundary after each "=" at the same patch count gets 51.8%, and entropy combined with a label-free boundary-dependence signal gets 67.1% (default layout at 26% of bytes: 76.8%). The gap survives adapting BLT-1B to the budget with low-rank fine-tuning (32.9% vs 72.7%, three runs per rule, paired p < 1e-200) and grows with model size in byte models trained from scratch at a 10% budget: at 1M, 12M and 50M parameters, boundary dependence beats entropy on final answers by -1.6, +10.1 and +19.8 points, and at 50M it gets 35.9% of computed results against 13.9% (3 seeds each). BLT's entropy-jump rule helps neither target at 50M. The entropy trigger of Scratchpad Patching is likewise indistinguishable from random scratchpads on final answers (5.6% vs 5.9%, 5 seeds), while answer-start scratchpads give 38.1%. The effect is specific to computed values: copies and lookups gain little, and values the model cannot compute gain nothing. Boundary dependence, the rise in the model's own loss when a patch start is removed, measured per two-byte context, finds these positions without labels: combined with entropy it beats the hand-written rule on computed results.

Chinese Translation

字节级语言模型如字节潜在变换器（Byte Latent Transformer, BLT）将字节分组为补丁，并对每个补丁运行其大型全局模型一次。BLT 在下一个字节的熵较高的位置开始一个补丁，因此全局计算集中在下一个字节难以预测的地方。我们展示了这一规则存在系统性的盲点：某些位置的类型可以预测，但其值必须计算，例如在数学解题中“=”后面的数字。在紧张的补丁预算下，熵触发的布局跳过这些位置，导致其准确性崩溃。在 Meta 的 BLT-1B 中，补丁开始于 10% 的字节，熵规则使补丁开始位于 GSM8K 解答的 16% 计算结果中，并且准确得到 19.0%；而在相同补丁数量下，每个“=”后面的边界得到 51.8%，熵结合无标签的边界依赖信号则达到 67.1%（默认布局在 26% 的字节时为 76.8%）。这一差距在通过低秩微调适应预算时仍然存在（32.9% 对 72.7%，每条规则三次运行，配对 p < 1e-200），并且在从零开始训练的字节模型中，随着模型规模的增加而扩大：在 1M、12M 和 50M 参数时，边界依赖在最终答案上分别比熵超出 -1.6、+10.1 和 +19.8 分，在 50M 时，边界依赖获得 35.9% 的计算结果，而熵仅为 13.9%（每个种子 3 个）。在 50M 时，BLT 的熵跳跃规则对这两个目标均无帮助。Scratchpad Patching 的熵触发在最终答案上同样与随机的 Scratchpad 无法区分（5.6% 对 5.9%，5 个种子），而答案起始的 Scratchpad 则给出 38.1%。这一效果特定于计算值：复制和查找几乎没有增益，而模型无法计算的值则没有任何增益。边界依赖，即当移除补丁开始时模型自身损失的上升，按每两个字节上下文测量，能够在没有标签的情况下找到这些位置：与熵结合时，它在计算结果上超越了手写规则。

View on arXiv Download PDF AI Translation

cs.CL / 65 / 2610.11845

Detecting Spin in Clinical Trials with Large Language Models

利用大型语言模型检测临床试验中的结果扭曲

Ajdovec, Tjaš, Robnik-Šikonja, Marko, Šuster, Simon

Abstract

Spin in clinical trials includes reporting practices that distort the presentation of results. This is particularly critical in medicine, where spin is present in more than 50% of randomized controlled trials that fail to reach statistical significance. The comparison of primary and reported outcomes is crucial for detecting several types of spin, including outcome switching. We used 300 pairs of outcomes labeled with semantic similarity to develop a system for automatic detection of outcome switching. We evaluated baseline text similarity models and open-source LLMs using generated similarity scores and the Youden index to determine the classification threshold. The proposed approach involves prompt engineering, classification based on token probabilities, and majority voting for the final decision. The results on the test set of 2,496 examples with an F1 score of 0.78 and an accuracy of 0.90 outperform baseline text similarity models but trail behind fine-tuned versions of BERT. We used LLMs to generate natural language explanations for the classified instances and manually assessed their quality.

Chinese Translation

临床试验中的结果扭曲包括扭曲结果呈现的报告实践。这在医学领域尤为重要，因为在超过50%的未达到统计显著性的随机对照试验中都存在结果扭曲。比较主要结果和报告结果对于检测多种类型的结果扭曲至关重要，包括结果切换。我们使用300对标记为语义相似的结果，开发了一个自动检测结果切换的系统。我们评估了基线文本相似性模型和开源大型语言模型（LLMs），使用生成的相似性分数和Youden指数来确定分类阈值。所提出的方法涉及提示工程、基于标记概率的分类以及最终决策的多数投票。测试集2,496个实例的结果显示，F1分数为0.78，准确率为0.90，优于基线文本相似性模型，但落后于经过微调的BERT版本。我们使用LLMs为分类实例生成自然语言解释，并手动评估其质量。

View on arXiv Download PDF AI Translation

cs.CL / 66 / 2610.11899

Forms of LLM-Integrated Applications from LLM-Chats to Autonomous AI Agent System

从LLM聊天到自主AI代理系统的LLM集成应用形式

Weber, Irene

Abstract

Large language models (LLMs) are increasingly embedded as components in software systems, marketed under labels such as chatbot, copilot, retrieval-augmented generation, workflow, coding agent and AI agent. Whether these labels denote genuine architectural forms or serve as branding has not been assessed systematically. In the sources surveyed, labels do carry architectural content, most clearly in vendor usage: copilot denotes a router-worker architecture operating a host application under step-by-step user confirmation, while the more recent shift to the label agent coincides with AI-planned multi-step execution of which the user sees only the outcome. The coding agents of four major providers share one architecture, a reason-and-act loop delegating to subagents. This survey describes seven recurring forms---LLM chats, custom agents, retrieval-augmented generation (RAG), AI-enhanced workflows, copilots, coding agents, and, in part, agentic RAG---in a common vocabulary of agents and tools. Each is characterized along four structural dimensions (agentic RAG only partially): the architectural pattern, the control of execution and the point of user intervention, the number of agent calls per task, and tool use. An illustrative corpus of 22 systems from research publications and vendor documentation grounds the descriptions and shows where they reach their limit.

Chinese Translation

大型语言模型（LLMs）越来越多地作为软件系统中的组件嵌入，市场上以聊天机器人、协作助手、增强检索生成、工作流、编码代理和AI代理等标签进行销售。这些标签是否代表真正的架构形式，还是仅仅作为品牌标识尚未得到系统评估。在调查的资料中，这些标签确实包含架构内容，尤其是在供应商的使用中：协作助手（copilot）表示一种在逐步用户确认下操作主应用的路由-工作者架构，而最近转向代理（agent）标签的变化则与AI规划的多步骤执行相吻合，用户仅能看到结果。四大主要供应商的编码代理共享一种架构，即一个将任务委派给子代理的推理与行动循环。本调查描述了七种反复出现的形式——LLM聊天、自定义代理、增强检索生成（RAG）、AI增强工作流、协作助手、编码代理，以及部分代理RAG——在一个共同的代理和工具词汇中。每种形式在四个结构维度上进行特征描述（代理RAG仅部分适用）：架构模式、执行控制与用户干预点、每个任务的代理调用次数，以及工具使用。通过来自研究出版物和供应商文档的22个系统的示例语料，支撑了这些描述，并展示了它们的局限性。

View on arXiv Download PDF AI Translation

cs.CL / 67 / 2610.11901

Can Decision Models Understand Stance? Evaluating Jev Against General-Purpose LLMs

决策模型能理解立场吗？评估Jev与通用大型语言模型的表现

Li, Xing, Ning, Jinzhong, Zhang, Yijia, Yang, Liang, Lin, Hongfei

Abstract

Stance detection requires identifying an author's attitude toward a given target, sometimes based on conversational context. Jev, a specialized decision model designed for structured decision-making, offers an alternative to general-purpose large language models (LLMs). In this work, we evaluate Jev on two stance detection datasets, VAST (English texts) and ZS-CSD (Chinese conversations), comparing it with four general-purpose LLMs and two fine-tuned models. Results show that Jev achieves competitive performance on VAST, matching GPT-5.6 and outperforming the other general-purpose LLMs. However, it falls behind stronger LLMs on ZS-CSD, particularly in distinguishing favor from against. Further analysis suggests that this limitation may be related to understanding reply relationships and stance direction rather than conversation length alone. These findings highlight both the potential and limitations of Jev for stance detection.

Chinese Translation

立场检测需要识别作者对特定目标的态度，有时还需基于对话上下文。Jev是一种专门为结构化决策而设计的决策模型，为通用大型语言模型（LLMs）提供了一种替代方案。在本研究中，我们在两个立场检测数据集上评估Jev，分别是VAST（英文文本）和ZS-CSD（中文对话），并将其与四个通用大型语言模型和两个微调模型进行比较。结果显示，Jev在VAST数据集上表现出竞争力，达到了与GPT-5.6相当的水平，并优于其他通用大型语言模型。然而，在ZS-CSD数据集上，它的表现不及更强的LLMs，特别是在区分支持与反对的立场方面。进一步分析表明，这一局限性可能与理解回复关系和立场方向有关，而不仅仅是对话长度。这些发现突显了Jev在立场检测中的潜力与局限性。

View on arXiv Download PDF AI Translation

cs.CL / 68 / 2610.11915

Not Every Change Is Necessary: Recoverable Drift in Large Language Model Unlearning

并非每一次变化都是必要的：大型语言模型中的可恢复漂移

Chen, Xunlei, Gong, Qinghui, Xue, Jingkun, Liu, Qihe, Zhou, Shijie, Ye, Fei

Abstract

Machine unlearning in large language models aims to remove unwanted knowledge while preserving the model's remaining capabilities. Although existing methods use retention objectives or restrict where edits occur, achieving the desired forgetting level can still leave collateral changes that impair non-target behavior. Our recovery comparisons suggest that some of these changes can be reversed while preserving observed forgetting performance. In this work, we present Propose-Then-Project Unlearning (PTP-U), a framework that combines targeted forgetting with the recovery of non-target capabilities. PTP-U first applies local analytic edits to weaken target knowledge associations, then aligns non-target output distributions with those of the original model to recover capabilities while maintaining fixed forgetting constraints. Both stages serve a common goal: satisfying the forgetting requirements while preserving fluent generation and performance on non-target tasks. Across three benchmarks, PTP-U achieves the strongest forgetting-retention trade-off among evaluated methods, reaching 81.22%-91.03% forgetting while preserving 94.20% non-target utility on average. At matched forgetting, PTP-U consistently retains higher non-target utility.

Chinese Translation

大型语言模型中的机器遗忘旨在去除不必要的知识，同时保留模型的其他能力。尽管现有方法使用保留目标或限制编辑发生的地方，但实现所需的遗忘水平仍可能留下影响非目标行为的附带变化。我们的恢复比较表明，这些变化中的一些可以在保持观察到的遗忘性能的同时被逆转。在本研究中，我们提出了“先提议后投影遗忘”（Propose-Then-Project Unlearning, PTP-U）框架，该框架将有针对性的遗忘与非目标能力的恢复相结合。PTP-U首先应用局部分析编辑以削弱目标知识关联，然后将非目标输出分布与原始模型的分布对齐，以在保持固定遗忘约束的同时恢复能力。这两个阶段的共同目标是：满足遗忘要求，同时保持流畅的生成和在非目标任务上的性能。在三个基准测试中，PTP-U在评估的方法中实现了最强的遗忘-保留权衡，达到81.22%-91.03%的遗忘率，同时平均保留94.20%的非目标效用。在匹配的遗忘条件下，PTP-U始终保持更高的非目标效用。

View on arXiv Download PDF AI Translation

cs.CL / 69 / 2610.11920

Event-Centric Memory with Query-Aware Graph Augmentation for Long-Term Conversational Agents

基于事件的记忆与查询感知图增强用于长期对话代理

Liu, Yichen, Yuan, Chunfeng, Liu, Haowei, Li, Wenjuan, Lin, Zefeng, Li, Bing, Chen, Xu, Hu, Weiming

Abstract

For persistent and personalized conversational agents, memory systems can enable them to remember, update, and reason over long histories by storing past interactions and retrieving relevant information. Existing memory systems typically follow two paradigms: flat-structured memory and graph-based memory. The former is lightweight but leaves event relations and state updates implicit, while the latter explicitly models memory structure but incurs additional construction cost and introduces irrelevant relations over long histories. To address these limitations, we propose QGMem, a novel memory construction and activation framework motivated by human memory, in which experience is organized into events and query-relevant events are modeled by graph as working memory. QGMem converts long dialogue histories into event-indexed atomic memory units that preserve individual experiences and consolidates related units into dynamic memory traces that retain state trajectories and current states. When a query arrives, hybrid memory retrieval gathers complementary candidate memories, and query-aware reranking activates the most relevant units as a compact working memory. To expose relational dependencies in the working memory and support conflict-aware reasoning, QGMem organizes the working memory as a local graph, which is then encoded as a graph token and provided to the LLM together with the textual working memory to improve evidence utilization during answer generation. Experiments across six benchmarks validate the framework and show consistent gains in retrieval, multi-hop evidence composition, conflict resolution, and ultra-long dialogue reasoning with compact contexts and moderate inference cost.

Chinese Translation

对于持久且个性化的对话代理，记忆系统能够通过存储过去的互动并检索相关信息，使其记住、更新和推理长时间的历史。现有的记忆系统通常遵循两种范式：扁平结构记忆和基于图的记忆。前者轻量但将事件关系和状态更新隐含，而后者则明确建模记忆结构，但会产生额外的构建成本，并在长时间历史中引入无关关系。为了解决这些局限性，我们提出了QGMem，这是一种受到人类记忆启发的新型记忆构建与激活框架，其中经验被组织为事件，查询相关事件通过图模型化为工作记忆。QGMem将长对话历史转换为事件索引的原子记忆单元，保留个体经验，并将相关单元整合为动态记忆痕迹，保持状态轨迹和当前状态。当查询到达时，混合记忆检索收集互补的候选记忆，而查询感知的重新排序则激活最相关的单元作为紧凑的工作记忆。为了揭示工作记忆中的关系依赖并支持冲突感知推理，QGMem将工作记忆组织为局部图，然后将其编码为图令牌，并与文本工作记忆一起提供给大型语言模型（LLM），以提高在答案生成过程中的证据利用。六个基准测试的实验验证了该框架，并显示在检索、多跳证据组合、冲突解决以及在紧凑上下文和适度推理成本下的超长对话推理方面的一致性提升。

View on arXiv Download PDF AI Translation

cs.CL / 70 / 2610.11948

When History Helps and Hurts: Selective History Use across Multimodal Turns

历史的帮助与伤害：多模态转变中的选择性历史使用

Sun, Shuoyang, Gu, Kerui, Fang, Hao, Huang, Shaoli, Chen, Bin

Abstract

Reliable multimodal interaction depends on selective use of conversational history: an earlier question may remain relevant while its previous answer is outdated, whereas a current request may depend on historical evidence despite conflicting new observations. Existing multi-turn evaluations rarely separate these history-use demands from underlying question difficulty. To address this gap, we introduce ReTurn, a benchmark of 7,000 base tasks spanning visual and audio evidence for evaluating selective history use. For task-carrying history, Reconfirm/Reground require applying a historical question to current media while varying historical agreement; for evidence-carrying history, Retrieve/Rebind require answering a current question using historical media while varying current-media competition. Each pair preserves the target question, media, and answer. Tasks support open-ended and multiple-choice evaluation, with matched single-turn counterparts serving as answerability references. Across 13 omni-modal, vision-language, and audio-language models, median model-level open-ended accuracy falls from 93.7% with direct input to 72.3% in conversation. Behavioral probes show that high question recall can coexist with weaker task application, while competing media can redirect answers away from historical targets. Supervised adaptation yields only partial gains. ReTurn provides a controlled framework for assessing whether multimodal models select and use the historical information required by each request.

Chinese Translation

可靠的多模态互动依赖于对对话历史的选择性使用：早期的问题可能仍然相关，而其之前的答案则可能过时；当前的请求可能依赖于历史证据，尽管存在相互矛盾的新观察。现有的多轮评估很少将这些历史使用需求与潜在的问题难度区分开。为了解决这一问题，我们引入了ReTurn，这是一个包含7000个基础任务的基准，涵盖视觉和音频证据，用于评估选择性历史使用。对于任务承载的历史，Reconfirm/Reground要求将历史问题应用于当前媒体，同时变化历史一致性；对于证据承载的历史，Retrieve/Rebind要求使用历史媒体回答当前问题，同时变化当前媒体的竞争性。每对任务都保留了目标问题、媒体和答案。任务支持开放式和多项选择评估，匹配的单轮对应任务作为可回答性参考。在13个全模态、视觉-语言和音频-语言模型中，中位模型级开放式准确率从直接输入的93.7%下降到对话中的72.3%。行为探测显示，高问题回忆可以与较弱的任务应用共存，而竞争媒体可以将答案引导远离历史目标。监督适应仅带来部分提升。ReTurn提供了一个受控框架，用于评估多模态模型是否选择并使用每个请求所需的历史信息。

View on arXiv Download PDF AI Translation

cs.CL / 71 / 2610.11959

MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement

MiMo-V2.6：将强化学习扩展至自我改进

Core Team, Qiao, Zongming, Hua, Ziyue, Ou, Zirui, Yue, Zihao, Jiang, Zihan, Huang, Zhuo, Chen, Zhiyang, Zheng, Zhixian, Xu, Zhipeng, Ma, Zhengrui, Hu, Yuyang, Dong, Yuhang, Zhang, Yuechen, Wang, Yudong, Liu, Yuanxin, Yang, Yixin, Cai, Yishuo, Zhao, Yikai, Yan, Yihan, Zhang, Yifan, Song, Yifan, Wei, Xiyu, Zhang, Xing, Zhang, Xin, Liu, Xiaoqian, Ji, Xiaodong, Deng, Xiangwei, Guo, Xueyu, Ma, Wenhan, Xiong, Weimin, Wang, Weikun, Zhuang, Weiji, Liu, Shuo, Ren, Shuhuai, Gu, Shuhao, Chen, Shimao, Cao, Shijie, Yu, Shihua, Li, Shicheng, Zhou, Shengjie, Zhang, Shaolei, Li, Rang, Wang, Qiying, Fang, Qingkai, Chen, Qianli, Wang, Minzheng, Wang, Liwen, Yao, Linli, Zhang, Linghao, Cheng, Liangyu, Zhao, Liang, Li, Lei, Dong, Jinhao, Xiang, Jinyu, Wei, Jianyu, Duo, Jiangshan, Liu, Huaqiu, Fan, Huanjie, Guan, Hongyi, Xu, Hongshen, Tian, Hao, Li, Hanyu, Zhang, Hailin, Wang, Gang, Luo, Fuli, Wei, Feng, Zhang, Dong, Zhu, Dawei, Lou, Chiheng, He, Chenhong, He, Chenhao, Liu, Chenghua, Ye, Bowen, Shen, Bowen, Xu, Boshen, Yang, Bo, Xia, Bingquan, Xiao, Bangjun, Xu, Baixuan, Mao, Zhouxiang, Zhang, Zhiyang, Xu, Zhixiang, Lin, Zhenru, Tang, Zhengju, Huang, Zhaojun, Weng, Yuzhe, Xiang, Yuxing, Li, Yuxiao, Yang, Yuheng, Wang, Yuhang, Liu, Yuchen, Tian, Yuanyuan, Dong, Yuanliang, Cheng, Yu, He, Yongzhe, Liang, Yongshun, Wang, Yong, Wang, Yiyan, Gong, Yitian, Zhang, Yijie, Xin, Yanshu, Zhang, Xun, Zhao, Xingjian, Yang, Wenyu, Huang, Wenshan, Li, Wenhao, Huang, Tingwei, Yu, Tianyu, Lu, Tianyang, Yang, Taoyu, Du, Sinan, Tian, Shutong, Du, Shulin, Wang, Shengfan, Fang, Shanchuan, Zhang, Qihao, Yang, Qibin, Yu, Qian, Tu, Qian, Xie, Pengrong, Wang, Peipei, Li, Peidian, Guo, Minkun, Shao, Mingchen, Gao, Luohan, Wang, Lijie, Shi, Liang, Chen, Kaiqi, Liu, Kaiming, Wang, Kaifei, Yang, Kai, Xue, Jinlong, Zhang, Jiechen, Liu, Jiaxuan, An, Hongxu, Peng, Hao, Lü, Hanglong, Wang, Guonan, Yang, Feiyu, Cao, Fanyu, Liu, Fangyue, Cui, Fan, Wang, Cong, Chen, Chun, Bai, Chenxu, Zhu, Chengxuan, Wang, Chenghua, Zeng, Boyi

Abstract

Reinforcement learning (RL) is the central training paradigm for advancing large foundation models towards self-improvement. This report introduces the MiMo-V2.6 series, an omni-modal family that pushes the frontier of model intelligence by scaling RL compute. Prior to RL, we conduct mid-training on a broad multimodal corpus to provide ample exploration space, and build a solid infrastructure on the pretrained hybrid-SWA architecture to support subsequent scale-up. We scale RL compute along three dimensions: (1) larger batches and higher throughput, with an asynchronous training that consumes 1,568 samples and 2.7-3.7B tokens per step at context lengths of up to 1M; (2) more diverse and complex environments, spanning code, general, visual, and cyber domains under a mixture of agent harnesses; and (3) more grader compute, via groupwise agentic grading that yields more accurate reward signals for long-horizon tasks and steers the model towards shorter, more token-efficient solutions. To keep training stable at scale, we freeze the MoE router and establish a multi-layer defense against reward hacking. We further build infrastructure for mixed-task agentic RL, including a unified trajectory representation, high-concurrency multi-framework rollout, decoupled control and data planes, and training-inference consistency. We open-source the training dynamics, RL environments, and RL framework to facilitate reproduction and further research on scaled RL and model self-improvement.

Chinese Translation

强化学习（RL）是推动大型基础模型实现自我改进的核心训练范式。本报告介绍了MiMo-V2.6系列，这是一种全模态家族，通过扩展RL计算推动模型智能的前沿。在进行RL之前，我们在广泛的多模态语料库上进行中期训练，以提供充足的探索空间，并在预训练的混合SWA架构上构建坚实的基础设施，以支持后续的扩展。我们从三个维度扩展RL计算：（1）更大的批量和更高的吞吐量，通过异步训练，每步消耗1,568个样本和2.7-3.7B个标记，上下文长度可达1M；（2）更具多样性和复杂性的环境，涵盖代码、通用、视觉和网络领域，采用多种代理框架；（3）更多的评分计算，通过分组代理评分产生更准确的奖励信号，以应对长期任务，并引导模型朝向更短、更高效的解决方案。为了在大规模训练中保持稳定，我们冻结了MoE路由器，并建立了多层防御机制以抵御奖励操控。我们进一步为混合任务的代理RL构建基础设施，包括统一的轨迹表示、高并发的多框架回放、解耦的控制和数据平面，以及训练-推理一致性。我们开源了训练动态、RL环境和RL框架，以促进对扩展RL和模型自我改进的再现和进一步研究。

View on arXiv Download PDF AI Translation

cs.CL / 72 / 2610.11978

Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks

专用决策模型与通用大型语言模型的比较：在知识、推理和多语言任务中的基准测试

Li, Xing, Chang, Qingcheng, Ning, Jinzhong, Xu, Changfeng, Zhang, Shenlong, Zhang, Yijia, Luo, Ling, Lin, Hongfei

Abstract

Jev is a "System One" model that returns a choice among given options instead of generating text. We study how such a specialized decision model compares with general-purpose large language models (LLMs). We evaluate Jev on 13 multiple-choice benchmarks covering knowledge, reasoning, and multilingual understanding, and compare it with 19 LLMs in three tiers: frontier, representative, and small. Jev is competitive with frontier LLMs on knowledge and commonsense benchmarks and obtains the best score on MMLU-Redux and ARC-Challenge. Outside mathematics, it also outperforms most representative LLMs and all small LLMs. However, it falls behind on mathematical word problems: on MathQA, it is 17.7 points below the frontier median and scores lower than all 19 LLMs. These results indicate that a specialized decision model can match general-purpose LLMs on decisions that rely mainly on knowledge, but not on decisions that require multi-step calculation.

Chinese Translation

Jev 是一种“系统一”模型，它在给定选项中返回选择，而不是生成文本。我们研究了这种专用决策模型与通用大型语言模型（LLMs）的比较。我们在涵盖知识、推理和多语言理解的 13 个多项选择基准上评估了 Jev，并将其与 19 个 LLM 进行比较，分为三个层次：前沿、代表性和小型。Jev 在知识和常识基准上与前沿 LLM 竞争，并在 MMLU-Redux 和 ARC-Challenge 上获得最佳分数。在数学之外，它还超越了大多数代表性 LLM 和所有小型 LLM。然而，在数学文字问题上，它的表现不佳：在 MathQA 上，它的分数比前沿中位数低 17.7 分，并且低于所有 19 个 LLM。这些结果表明，专用决策模型可以在主要依赖知识的决策上与通用 LLM 相匹配，但在需要多步计算的决策上则不然。

View on arXiv Download PDF AI Translation

cs.CL / 73 / 2610.12022

Examining Social Attribution in LLM Reasoning: A Theory-Guided Probing Methodology

探讨大型语言模型推理中的社会归因：一种理论指导的探测方法

Yu, Zhaoxin, Kong, Qingchao, Zeng, Dajun, Mao, Wenji

Abstract

Large language models (LLMs) are increasingly deployed in sociotechnical systems where social attribution, the reasoning process attributing external events to the causes and reasons of agents' social behaviors, plays a critical role. These processes involve judgments of social cause, responsibility, and blame/credit to agents. Although attributional models are well-studied in social psychology and cognition through Attribution Theory, social attribution remains underexplored in AI, particularly LLM social reasoning. This paper provides the first systematic exploration of LLM social attribution. Our work focuses on responsibility and blame attributions, examining current LLMs' judgments and their underlying internal mechanisms. Guided by attribution theory, we construct a social attribution benchmark consisting of a Vignette subset based on classic scenarios from attribution theory research and a Reality subset based on real-world social narratives, yielding 7,639 responsibility/blame judgment questions. On this basis, we evaluate 32 representative LLMs and 5 basic non-LLM baselines. To further explore the internal mechanisms underlying the LLM judgment process, we develop a probing-based methodology to investigate the latent-space representations of 5 key attribution dimensions and the consistency of their influences on LLM judgments compared to those in human social attribution. Our research findings reveal that current LLMs exhibit measurable but incomplete agreement with human responsibility and blame judgments, and meanwhile, this agreement is positively correlated with model size. Some attribution dimensions are systematically decodable from specific positions in LLM hidden states, and their influences on the final judgment are consistent with those indicated by human Attribution Theory. The dataset and associated code are available at https://github.com/Yuzhaoxin946/SAB-Bench.

Chinese Translation

大型语言模型（LLMs）越来越多地应用于社会技术系统中，其中社会归因，即将外部事件归因于代理人社会行为的原因和理由的推理过程，发挥着关键作用。这些过程涉及对社会原因、责任以及对代理人的指责/赞誉的判断。尽管归因模型在社会心理学和认知科学中通过归因理论得到了充分研究，但在人工智能，特别是大型语言模型的社会推理方面，社会归因仍然未得到充分探讨。本文首次系统性地探索了大型语言模型的社会归因。我们的研究重点是责任和指责的归因，考察当前大型语言模型的判断及其潜在的内部机制。在归因理论的指导下，我们构建了一个社会归因基准，包括基于归因理论研究经典场景的情境子集和基于现实社会叙事的现实子集，共计7,639个责任/指责判断问题。在此基础上，我们评估了32个代表性的大型语言模型和5个基本的非大型语言模型基线。为了进一步探讨大型语言模型判断过程背后的内部机制，我们开发了一种基于探测的方法，研究5个关键归因维度的潜在空间表示及其对大型语言模型判断的影响与人类社会归因的比较一致性。我们的研究结果表明，当前的大型语言模型在责任和指责判断上与人类的判断存在可测量但不完全的一致性，并且这种一致性与模型规模呈正相关。一些归因维度可以从大型语言模型隐藏状态的特定位置系统性解码，其对最终判断的影响与人类归因理论所指示的一致。数据集和相关代码可在 https://github.com/Yuzhaoxin946/SAB-Bench 获取。

View on arXiv Download PDF AI Translation

cs.CL / 74 / 2610.12030

Natural Language to First-Order Logic LLM-based Autoformalization

基于大型语言模型的自然语言到一阶逻辑的自动形式化

Brunello, Andrea, Curaba, Cristian, Geatti, Luca, Mignani, Michele, Montanari, Angelo, Saccomanno, Nicola

Abstract

Large Language Models (LLMs) have renewed interest in autoformalization. Yet, when First-Order Logic (FOL) is considered as the target formalism, the field still lacks a unified task formulation and a systematic survey. This paper addresses this gap: we first provide a principled definition for the FOL-autoformalization task by distinguishing Ontology Extraction from Logical Translation, showing how their conflation obscures (cross-study) evaluation; we review existing datasets, evaluation metrics, and LLM-based methods, including fine-tuning, prompting, and verification-based refinement; we identify open challenges in benchmarking, semantic evaluation, ontology-aware methods, and end-to-end applications.

Chinese Translation

大型语言模型（LLMs）重新激发了对自动形式化的兴趣。然而，当考虑一阶逻辑（FOL）作为目标形式化时，该领域仍然缺乏统一的任务表述和系统的调查。本文旨在填补这一空白：我们首先通过区分本体提取（Ontology Extraction）与逻辑翻译（Logical Translation）为FOL-自动形式化任务提供了一个原则性的定义，展示了它们的混淆如何模糊（跨研究）评估；我们回顾了现有的数据集、评估指标和基于LLM的方法，包括微调（fine-tuning）、提示（prompting）和基于验证的精炼（verification-based refinement）；我们识别了基准测试、语义评估、本体感知方法和端到端应用中的开放挑战。

View on arXiv Download PDF AI Translation

cs.CL / 75 / 2610.12064

ILM: An AI-Powered Storytelling Educational Tool

ILM：一种基于人工智能的故事讲述教育工具

Mohammed, Suhaila, Serour, Abdelaziz, Lahnala, Allison

Abstract

Digital technologies have made Islamic narratives more accessible, but existing platforms provide limited support for structured learning and comprehension of these stories, particularly in Arabic and multilingual settings. We present ILM, an interactive educational platform for Stories of the Prophets that combines Arabic natural language processing, structured knowledge representation, and retrieval-based question generation. Admin-approved Arabic narratives are processed by a Knowledge Graph (KG) Constructor Engine that identifies entities and narrative relationships and stores them as structured knowledge, enabling learners to explore stories through a visual story map and answer entity- and relation-based questions generated from the KG. Separately, a multilingual retrieval pipeline retrieves relevant passages from the original narratives to generate multiple-choice and open-ended comprehension questions. For open-ended questions, an LLM-as-a-Judge evaluates learners' answers against the retrieved passages and reference answers to determine correctness. The platform also incorporates Quranic content as a separate enrichment layer, allowing selected narratives to be supplemented with source-supported information. By combining structured knowledge with passage-based retrieval, ILM supports narrative exploration, comprehension, and assessment across Arabic and multilingual content. The system demonstrates the feasibility of combining structured knowledge representation and retrieval-based generation to support interactive learning of Islamic narratives. A demo is available at anonymous.4open.science/r/mml-5FCF.

Chinese Translation

数字技术使伊斯兰叙事变得更加易于获取，但现有平台对这些故事的结构化学习和理解支持有限，尤其是在阿拉伯语和多语言环境中。我们提出了ILM，一个互动教育平台，专注于先知故事，结合了阿拉伯自然语言处理、结构化知识表示和基于检索的问题生成。经过管理员审核的阿拉伯叙事通过知识图谱（Knowledge Graph, KG）构造引擎进行处理，该引擎识别实体和叙事关系，并将其存储为结构化知识，使学习者能够通过可视化故事地图探索故事，并回答从KG生成的基于实体和关系的问题。此外，一个多语言检索管道从原始叙事中检索相关段落，以生成多项选择和开放式理解问题。对于开放式问题，一个作为评判者的语言模型（LLM-as-a-Judge）根据检索到的段落和参考答案评估学习者的回答，以确定其正确性。该平台还将古兰经内容作为单独的丰富层，允许选定的叙事补充源支持的信息。通过将结构化知识与基于段落的检索相结合，ILM支持阿拉伯语和多语言内容的叙事探索、理解和评估。该系统展示了结合结构化知识表示和基于检索的生成以支持伊斯兰叙事互动学习的可行性。演示版可在anonymous.4open.science/r/mml-5FCF获取。

View on arXiv Download PDF AI Translation

cs.CL / 76 / 2610.12083

All Verdicts are Not Equal: Rethinking LLM Judge Reliability

所有裁决并不相同：重新思考大型语言模型法官的可靠性

Kumar, Vineet, Rathore, Darshita, Moitra, Anindya

Abstract

LLM-as-a-Judge is the standard paradigm for NLP evaluation, yet its systemic reliability remains poorly understood despite being widely treated as a deterministic ground truth. We present a comprehensive reliability audit, stresstesting six frontier models across four benchmarks, five prompt formats, two presentation orders, three sampling temperatures, and ten repetitions per condition. Our empirical analysis reveals severe vulnerabilities: verdicts change across identical replications at temperature zero, position-order swaps flip the majority of verdicts on challenging tasks, and the most deterministic judge achieves perfect consistency by trivially repeating incorrect verdicts, agreeing with ground truth only 51% of the time. To formalize these multi-faceted failure modes, we introduce the trustworthy verdict rate (T ), a unified metric capturing the joint probability that an evaluation is reproducible, order-invariant, and accurate. UsingT , we derive a theoretical upper bound on accuracy imposed by position bias and show that reliability is item-specific rather than modellevel. Finally, we demonstrate that shifting from pairwise win-rate to holistic rubric scoring improves trustworthiness more than any single-format prompting intervention, offering an actionable framework for robust NLP evaluation.

Chinese Translation

大型语言模型作为法官（LLM-as-a-Judge）是自然语言处理（NLP）评估的标准范式，但其系统可靠性仍然不甚明了，尽管它被广泛视为确定性的真实标准。我们进行了一次全面的可靠性审计，针对六个前沿模型在四个基准测试、五种提示格式、两种呈现顺序、三种采样温度以及每种条件下十次重复进行压力测试。我们的实证分析揭示了严重的脆弱性：在温度为零的情况下，相同的重复实验中裁决会发生变化，位置顺序的交换会在具有挑战性的任务中翻转大多数裁决，而最具确定性的法官通过简单重复错误裁决来实现完美一致性，仅在51%的情况下与真实标准一致。为了形式化这些多方面的失败模式，我们引入了可信裁决率（trustworthy verdict rate, T），这是一个统一指标，捕捉评估可重复性、顺序不变性和准确性的联合概率。利用T，我们推导出位置偏差对准确性施加的理论上限，并展示可靠性是特定项目而非模型层面的。最后，我们证明从成对胜率转向整体评分标准比任何单一格式的提示干预更能提高可信度，为稳健的NLP评估提供了可行的框架。

View on arXiv Download PDF AI Translation

cs.CL / 77 / 2610.12118

A persistent accuracy ceiling in automated verbal deception detection

自动化口头欺骗检测中的持续准确率上限

Loconte, Riccardo, Festor, Jonas, Fatjanova, Zane, Bolkvadze, Mariam, Kleinberg, Bennett

Abstract

Automated methods have been proposed to overcome the limitations of human verbal deception detection, but evidence remains fragmented across disciplines. We systematically reviewed 25 years of research (289 reports, 6,136 classification models) and meta-analyzed 3,653 models nested within 97 datasets. Pooled accuracy was 74.4% (95% CI: 71.2%-77.4%) with substantial heterogeneity. Accuracy was driven by methodological quality (ground truth, data source, class balance, evaluation procedure) more than by model complexity: the adoption of embeddings and large language models has not translated into improved predictive performance. Only 12.46% of reports used data with verifiable ground-truth, and only 23.96% of models were evaluated on independent data. The pooled accuracy aligns with meta-analyses of manual approaches, suggesting a ceiling of 70-75%, unlikely to be lifted by current research conventions.

Chinese Translation

为克服人类口头欺骗检测的局限性，已有自动化方法被提出，但相关证据在各学科间仍较为分散。我们系统性回顾了过去25年的研究（289篇报告，6,136个分类模型），并对嵌套于97个数据集中的3,653个模型进行了元分析。汇总准确率为74.4%（95%置信区间：71.2%-77.4%），且存在显著异质性。准确率更多受方法学质量（真实标签、数据来源、类别平衡、评估程序）驱动，而非模型复杂度：采用嵌入技术和大型语言模型（large language models）并未带来预测性能的提升。仅有12.46%的报告使用了可验证的真实标签数据，且仅23.96%的模型在独立数据上进行了评估。汇总准确率与手工方法的元分析结果相符，表明准确率存在70%-75%的上限，当前研究惯例难以突破该限制。

View on arXiv Download PDF AI Translation

cs.CL / 78 / 2610.12133

Rehearse Everything, Remember Nothing: Attic-KV Rehearses What Will Be Read

复习一切，记住无物：Attic-KV 复习将被读取的内容

Shi, Zhiyun

Abstract

Many key-value (KV) caches are compressed before anyone knows what will be asked of them: a document cached for retrieval, a prompt prefix shared across requests, the memory of a long conversation. The prevailing approach scores KV entries by rehearsal: the model rereads the context and keeps the entries it attends to, assuming that the more completely a cache rehearses its context, the better it remembers it. We show that under tight budgets this assumption backfires: rehearse everything, remember nothing. At a 3% keep ratio, rereading the whole context keeps 31.5 of 96.5 points on RULER, and on LongBench's natural-text tasks it falls below methods that rehearse nothing at all. The cause is that a cache keeps what it rehearses: rereading spreads the budget across the whole context, so the answer's own entries survive at little more than chance. Like a student before an exam, a cache remembers more by testing itself than by rereading. Two principles follow: rehearse what will be read, and rehearse as much as there is. We instantiate them as Attic-KV (Attic for short), a training-free rehearsal in which the model quizzes itself with question-answer pairs that quote the context, alongside anchor tokens in a content-adaptive amount. Changing only the rehearsal lifts three hosts that score it in three different ways: Attic alone is the best training-free method in all eight settings we test on RULER and LongBench's natural-text tasks, and plugged into the gradient-based KVgrad and the trained RestoreKV+, it raises them by up to 17.1 and 28.1 points. Its advantage grows as the budget shrinks, reaching 41.9 points over full rereading at a 3% keep ratio, and it compresses faster than rereading the whole context.

Chinese Translation

许多键值（KV）缓存是在任何人知道它们将被请求什么之前就被压缩的：用于检索的文档、跨请求共享的提示前缀、长时间对话的记忆。当前的主流方法通过复习来评分 KV 条目：模型重新阅读上下文并保留其关注的条目，假设缓存越完全地复习其上下文，它的记忆就越好。我们展示了在预算紧张的情况下，这一假设是错误的：复习一切，记住无物。在 3% 的保留比率下，重新阅读整个上下文在 RULER 上保持 31.5 分（满分 96.5 分），而在 LongBench 的自然文本任务中，它的表现低于那些完全不复习的方法。原因在于缓存保留它所复习的内容：重新阅读将预算分散到整个上下文中，因此答案自身的条目仅以偶然的方式存活。就像考试前的学生一样，缓存通过自我测试比通过重新阅读记住更多。由此得出两个原则：复习将被读取的内容，以及尽可能多地复习。我们将其实例化为 Attic-KV（简称 Attic），这是一种无训练的复习方式，模型通过引用上下文的问题-答案对自我提问，同时以内容自适应的量提供锚定标记。仅改变复习方式就提升了三种不同评分方式的三个主机：Attic 在我们测试的 RULER 和 LongBench 的自然文本任务的八种设置中，都是最佳的无训练方法，并且在与基于梯度的 KVgrad 和训练后的 RestoreKV+ 结合时，分别提高了最多 17.1 和 28.1 分。随着预算的缩减，其优势不断扩大，在 3% 的保留比率下，相比于完全重新阅读，提升了 41.9 分，并且其压缩速度快于重新阅读整个上下文。

View on arXiv Download PDF AI Translation

cs.CL / 79 / 2610.12144

Language-Specific Effects of Tokenizer Choice in Multilingual Language Models

多语言模型中分词器选择的语言特定影响

Meister, Clara, Altıntaş, Gül Sena, Bosselut, Antoine

Abstract

Tokenizer choice affects multilingual language modeling, but vocabulary capacity is finite and vocabulary size is often constrained: improving representation for some languages often comes at the expense of others. We therefore ask whether tokenizer choice matters equally across languages, a question that the current literature leave unanswered. To this end, we train 123 language models spanning 54 tokenizers. In the main comparison, architecture, training corpus, training-token budget, and optimization are held fixed, so the models differ only in their tokenizer. We find that tokenizer choice matters more for languages with less language-model training data: across the 54 tokenizers, the standard deviation of a language's bits-per-byte (BPB) increases as its model training-data share decreases (Spearman rho = -0.52 over the 31 trained languages and -0.69 over the 28 written with word boundaries). Leaving a language out of tokenizer training raises its BPB in every language we study, and the penalty tends to be larger for languages with less language-model training data. Giving lower-resource languages a larger share of tokenizer-training data, however, does not unconditionally help those languages: both equal weighting and an allocation inverting the shares with respect to the language model training data increase their BPB, particularly when language-model training repeats data. Finally, which intrinsic tokenizer properties are associated with better BPB differs across languages, providing further evidence that what makes a good tokenizer depends on the language. We find that the metrics quantifying these properties can be successfully used to predict downstream models' pairwise BPB rankings, suggesting a practical strategy for screening tokenizer candidates before training language models.

Chinese Translation

分词器的选择影响多语言建模，但词汇容量是有限的，词汇大小通常受到限制：改善某些语言的表示往往以牺牲其他语言为代价。因此，我们提出一个问题：分词器选择在不同语言中是否同样重要，而当前文献对此尚未给出答案。为此，我们训练了涵盖54种分词器的123个语言模型。在主要比较中，架构、训练语料库、训练令牌预算和优化保持不变，因此模型仅在其分词器上有所不同。我们发现，分词器选择对语言模型训练数据较少的语言影响更大：在54种分词器中，随着模型训练数据份额的减少，语言的每字节比特数（BPB）的标准差增加（在31种训练语言中Spearman rho = -0.52，在28种以词边界书写的语言中Spearman rho = -0.69）。将某种语言排除在分词器训练之外会导致我们研究的每种语言的BPB上升，而这种惩罚对于语言模型训练数据较少的语言往往更为显著。然而，给予低资源语言更大份额的分词器训练数据并不总是有利于这些语言：无论是平等加权还是相对于语言模型训练数据反转份额的分配都会增加它们的BPB，尤其是在语言模型训练重复数据时。最后，与更好BPB相关的固有分词器属性在不同语言中有所不同，进一步证明了良好分词器的标准取决于语言。我们发现，量化这些属性的指标可以成功用于预测下游模型的成对BPB排名，这为在训练语言模型之前筛选分词器候选提供了一种实用策略。

View on arXiv Download PDF AI Translation

cs.CL / 80 / 2610.12207

SciTBERT: A family of chronologically consistent language models for scientific and technological language processing

SciTBERT：一系列时间一致的语言模型用于科学和技术语言处理

Gebhart, Thomas, Funk, Russell J.

Abstract

Pre-trained transformer models are increasingly being used to study scientific and technological progress. Encoders tuned to paper or patent text outperform general-purpose models on downstream classification, regression, and proximity tasks within science and technology. However, the applicability of these models for studying time-dependent or archival properties of science, technology, and their interface is limited due to lookahead and domain biases inherent to these pre-trained models. These limitations arise from training on corpora with unconstrained chronological and text source distributions. We introduce SciTBERT: a family of chronologically consistent BERT-derived language models trained on text from scientific papers, patents, and high-quality educational web text with training data cutoff dates spanning each year between 2013 and 2025. We also post-train these models in a chronologically-consistent manner using paper and patent citations, creating SciTBERT-CI model family. We find that these models generally outperform predecessor domain-specific encoder models even when training data is limited by early year restrictions in the corpus. To further investigate the extent to which this class of models can learn representations that bridge science and technology, we introduce the PatRepEval benchmark, a suite of patent-related text embedding tasks at the science-technology interface. Performance in a variety of classification, regression, and retrieval tasks spanning papers and patents highlights the importance of aligning encoder model representations with the domain distributions of their downstream tasks, and chronologically consistent encoders can match or exceed models trained without temporal constraints.

Chinese Translation

预训练的变换器模型在研究科学和技术进展方面的应用越来越广泛。针对论文或专利文本调优的编码器在科学和技术领域的下游分类、回归和相似性任务中表现优于通用模型。然而，由于这些预训练模型固有的前瞻性和领域偏差，这些模型在研究科学、技术及其接口的时间依赖性或档案特性方面的适用性受到限制。这些限制源于在具有不受限制的时间和文本来源分布的语料库上进行训练。我们提出了SciTBERT：一系列时间一致的基于BERT的语言模型，训练数据来自科学论文、专利和高质量的教育网页文本，训练数据截止日期涵盖2013年至2025年之间的每一年。我们还以时间一致的方式对这些模型进行后续训练，使用论文和专利引用，创建了SciTBERT-CI模型系列。我们发现，即使在训练数据受到早期年份限制的情况下，这些模型通常也优于前代特定领域的编码器模型。为了进一步探讨这一类模型在学习连接科学与技术的表示方面的能力，我们引入了PatRepEval基准，这是一个涵盖科学与技术接口的专利相关文本嵌入任务的套件。在涵盖论文和专利的多种分类、回归和检索任务中的表现突显了将编码器模型表示与其下游任务的领域分布对齐的重要性，而时间一致的编码器能够匹配或超越没有时间限制的模型。

View on arXiv Download PDF AI Translation

cs.CL / 81 / 2610.12235

Language Models as AI Research World Models

语言模型作为人工智能研究世界模型

Wang, Zijun, Liu, Zewen, Lin, Minhua, Weng, Zhaotian, Shi, Zhan, He, Bing, Sang, Yisi, Wang, Dakuo, Dumoulin, Benoit, Jin, Wei, Zhou, Yuyin, Xie, Cihang, Lu, Hanqing

Abstract

AI research agents automate the cycle of proposing, implementing, and evaluating experiments, opening a path toward recursive self-improvement. Yet their ability to propose experiments outpaces their capacity to execute them in real environments, making outcome prediction a key capability for sustained self-improvement under limited experimental budgets. We investigate language models as Research World Models (RWMs), which predict the outcomes of candidate interventions across research environments. Our evaluation draws on over 2,600 experimental records from nine research environments spanning pretraining, post-training, and inference, representing more than 171,000 H100 GPU-hours of experimentation. Research knowledge acquired from real experimental experience improves RWM predictions of unseen interventions within the same environment (Spearman +0.27), and can be reused across environments. For example, using only pretraining experience from OLMo3, Marin, and Nanochat, an RWM reduces selection regret in the Qwen3 environment by 78% compared with zero-experience setting. These benefits extend to multi-round Autoresearch under a fixed selection budget: RWMs with in-env and cross-env research knowledge increase the best gain achieved by 15.8% and 11.6%, respectively. Ablations across 13 language models used as RWMs show that adding research knowledge can improve intervention ranking more than changing models or increasing reasoning effort alone. These findings support language models as RWMs and motivate accumulating experimental data for future RWM training.

Chinese Translation

人工智能研究代理自动化了提出、实施和评估实验的循环，为递归自我改进开辟了道路。然而，它们提出实验的能力超过了在真实环境中执行实验的能力，使得在有限的实验预算下，结果预测成为持续自我改进的关键能力。我们研究了语言模型作为研究世界模型（Research World Models, RWMs），它们预测候选干预在研究环境中的结果。我们的评估基于来自九个研究环境的2600多个实验记录，这些环境涵盖了预训练、后训练和推理，代表了超过171,000个H100 GPU小时的实验。通过真实实验经验获得的研究知识提高了RWM对同一环境中未见干预的预测（Spearman +0.27），并且可以在不同环境之间重用。例如，仅使用来自OLMo3、Marin和Nanochat的预训练经验，RWM在Qwen3环境中将选择遗憾减少了78%，与零经验设置相比。这些好处扩展到在固定选择预算下的多轮自研究：具有环境内和跨环境研究知识的RWMs分别增加了最佳收益的15.8%和11.6%。对13个作为RWMs使用的语言模型的消融实验表明，添加研究知识可以比单独更改模型或增加推理努力更有效地改善干预排名。这些发现支持将语言模型视为RWMs，并激励积累实验数据以便未来的RWM训练。

View on arXiv Download PDF AI Translation

cs.CL / 82 / 2610.12242

TokenRouter: Efficient Serving System for Token-Level LLM Routing

TokenRouter：高效的令牌级大语言模型路由服务系统

Fu, Tianyu, Liu, Tengxuan, Wang, Ruoxi, Dong, Yixin, Ge, Yi, You, Yichen, Wang, Yu

Abstract

Large language model (LLM) routing distributes inference work across different models, advancing the cost-quality Pareto frontier of LLM serving. While coarse-grained routing at the session or query level has been widely adopted in production systems, recent algorithmic work shows that fine-grained token-level routing can yield substantial efficiency and quality gains. However, efficiently serving token-level routed inference poses significant challenges to existing systems. Built on single-LLM assumptions, current systems suffer from severe step desynchronization and frequent batch admission delays under token-level routing, and they also impose high implementation complexity on developers. To address these challenges, we design TokenRouter, an efficient and developer-friendly serving system for token-level routed LLM inference. TokenRouter follows the principle of request-centric programming, model-centric execution: developers describe routing logic from the perspective of a single request, while the runtime launches a subserver for each LLM and dispatches requests asynchronously. Each subserver employs a delayed-batching scheduler, whose optimal hyperparameters are derived from a mathematical throughput model of the system. Across diverse routing algorithms, workloads, and model pairs, TokenRouter achieves 2.01-64.15x higher decoding throughput than existing systems, substantially advancing the serving efficiency of token-level LLM routing. Our code is available at https://github.com/thu-nics/TokenRouter.

Chinese Translation

大语言模型（LLM）路由将推理工作分配到不同的模型上，推动了LLM服务的成本-质量帕累托前沿。尽管在生产系统中广泛采用了会话或查询级的粗粒度路由，但最近的算法研究表明，细粒度的令牌级路由可以带来显著的效率和质量提升。然而，高效地服务于令牌级路由的推理对现有系统提出了重大挑战。基于单一LLM假设，当前系统在令牌级路由下遭遇严重的步骤不同步和频繁的批量接纳延迟，同时也给开发者带来了高实现复杂性。为了解决这些挑战，我们设计了TokenRouter，一个高效且开发者友好的令牌级路由LLM推理服务系统。TokenRouter遵循请求中心编程和模型中心执行的原则：开发者从单个请求的角度描述路由逻辑，而运行时为每个LLM启动一个子服务器并异步调度请求。每个子服务器采用延迟批处理调度器，其最优超参数源自系统的数学吞吐量模型。在多种路由算法、工作负载和模型对下，TokenRouter的解码吞吐量比现有系统高出2.01-64.15倍，显著提升了令牌级LLM路由的服务效率。我们的代码可在https://github.com/thu-nics/TokenRouter获取。

View on arXiv Download PDF AI Translation

cs.CL / 83 / 2610.12248

EgoVoice: Proactive Spoken Assistance from Egocentric Multimodal Streams

EgoVoice：来自自我中心多模态流的主动语音助手

Kim, Heeseung

Abstract

Wearable augmented reality (AR) assistants are moving toward continuous real-world interaction, where they perceive the user's activity through first-person video and audio and provide timely spoken guidance without being explicitly asked. While proactive video assistants, spoken dialog systems, and egocentric task understanding have each advanced rapidly, existing systems do not address the joint problem of deciding when to speak and what to say from continuous first-person streams. We introduce EgoVoice, a framework for training and evaluating proactive egocentric spoken assistants. From HoloAssist video recordings of real human instructors, we construct clean audio streams through source separation and speech resynthesis, and convert each video session into a format where the model must decide at each moment whether to remain silent or provide spoken guidance. We fine-tune an omni-modal LLM with our data, and further improve its proactive intervention behavior with direct preference optimization. Experiments across closed and open-source models show that existing systems rarely produce well-timed, meaningful proactive interventions, while EgoVoice yields clear improvements in intervention timing, content relevance, and human preference over the zero-shot backbone.

Chinese Translation

可穿戴增强现实（AR）助手正朝着持续的现实世界交互发展，它们通过第一人称视频和音频感知用户的活动，并在未被明确要求的情况下提供及时的语音指导。尽管主动视频助手、语音对话系统和自我中心任务理解各自都取得了快速进展，但现有系统并未解决从连续的第一人称流中决定何时发言和说什么的联合问题。我们提出了EgoVoice，一个用于训练和评估主动自我中心语音助手的框架。通过对真实人类讲师的HoloAssist视频录制，我们通过源分离和语音重合成构建了干净的音频流，并将每个视频会话转换为模型必须在每个时刻决定是保持沉默还是提供语音指导的格式。我们使用我们的数据对全模态大语言模型（omni-modal LLM）进行了微调，并通过直接偏好优化进一步改善其主动干预行为。针对封闭和开源模型的实验表明，现有系统很少产生时机良好、意义明确的主动干预，而EgoVoice在干预时机、内容相关性和人类偏好方面相较于零-shot基线表现出明显的改善。

View on arXiv Download PDF AI Translation

cs.CL / 84 / 2610.12274

HarnessSQL: Harness-Native Training for SQL Agents in Realistic Database Environments

HarnessSQL：在现实数据库环境中为 SQL 代理提供原生训练

Yang, Haolin, Zhang, Jipeng, Xie, Jian, Gong, Shuaishuai, Han, Sirui, Guo, Yike

Abstract

Text-to-SQL models are commonly trained to map questions directly to static queries, whereas real-world database agents operate through stateful, multi-turn interaction with live databases -- inspecting schemas, executing probe queries, diagnosing errors, and revising hypotheses. This creates a critical train-deploy mismatch, as the execution harness that mediates this interaction is introduced only at inference time. To bridge this gap, we propose HarnessSQL, a harness-native post-training framework that preserves the full interaction structure throughout both supervised fine-tuning and reinforcement learning. HarnessSQL builds isolated, executable database environments paired with hidden execution oracles, rolls out teachers directly inside the target SQL harness, and retains only verified trajectories for full-sequence SFT, followed by execution-reward RL. Across Spider 2.0-SQLite, HarnessSQL dramatically boosts the execution accuracy of compact models, raising Qwen3-8B from 15.5% to 45.2% and Qwen3-14B from 22.2% to 54.8%, while transferring effectively to out-of-distribution interactive benchmarks such as BIRD-Interact and LiveSQLBench. Our findings demonstrate that training database agents directly within their execution harness is essential for mastering complex, long-horizon database workflows.

Chinese Translation

文本到 SQL 模型通常被训练为将问题直接映射到静态查询，而现实世界中的数据库代理则通过与实时数据库的有状态、多轮交互来操作——检查模式、执行探测查询、诊断错误和修正假设。这造成了一个关键的训练-部署不匹配，因为调解这种交互的执行环境仅在推理时引入。为了弥补这一差距，我们提出了 HarnessSQL，一种原生执行环境的后训练框架，能够在监督微调和强化学习的整个过程中保留完整的交互结构。HarnessSQL 构建了独立的、可执行的数据库环境，并配备隐藏的执行预言机，直接在目标 SQL 执行环境内部推出教师，并仅保留经过验证的轨迹用于全序列的监督微调，随后进行执行奖励的强化学习。在 Spider 2.0-SQLite 上，HarnessSQL 显著提高了紧凑模型的执行准确性，将 Qwen3-8B 从 15.5% 提升至 45.2%，将 Qwen3-14B 从 22.2% 提升至 54.8%，同时有效地转移到分布外的交互基准，如 BIRD-Interact 和 LiveSQLBench。我们的研究结果表明，在其执行环境中直接训练数据库代理对于掌握复杂的长期数据库工作流程至关重要。

View on arXiv Download PDF AI Translation

cs.CL / 85 / 2610.12338

VFold: Symmetry-Aware Cross-Layer Value Cache Compression

VFold：对称感知跨层值缓存压缩

Verma, Neha, Kim, Sungwon, Murray, Kenton, Duh, Kevin

Abstract

While caching key-value (KV) states accelerates Large Language Model (LLM) decoding, this cache can dominate memory usage at long context lengths. One solution is to compress this memory by exploiting inter-layer cache similarities. However, most existing techniques necessitate architectural changes to LLMs and incur substantial overhead. In this work, we propose a symmetry-aware value cache merging strategy that reduces cache memory while avoiding both harmful performance degradation and architectural overhead during decoding. Furthermore, we show that this approach can be exploited alongside existing cache compression techniques, composing with high-ratio quantization or key cache pruning to reach compression ratios that neither method reaches alone, with minimal additional cost. Ultimately, our findings reveal a major source of underutilized capacity in the value cache, offering a simple yet highly effective direction for scaling context windows under memory constraints.

Chinese Translation

在长上下文长度下，缓存键值（KV）状态虽然加速了大型语言模型（LLM）的解码，但该缓存可能主导内存使用。一个解决方案是通过利用层间缓存相似性来压缩这部分内存。然而，大多数现有技术需要对LLM进行架构更改，并且会产生显著的开销。在本研究中，我们提出了一种对称感知的值缓存合并策略，该策略在解码过程中减少缓存内存，同时避免了有害的性能下降和架构开销。此外，我们展示了该方法可以与现有的缓存压缩技术结合使用，与高比率量化或键缓存修剪组合，从而达到单独方法无法实现的压缩比，且附加成本极小。最终，我们的研究结果揭示了值缓存中一个主要的未充分利用的容量来源，为在内存限制下扩展上下文窗口提供了一个简单而高效的方向。

View on arXiv Download PDF AI Translation

cs.CL / 86 / 2610.12367

Which Skill to Distill? SGUID: Selecting a Compact Skill Bank for Model-Skill Co-Evolution

选择哪种技能进行提炼？SGUID：为模型-技能共同演化选择紧凑的技能库

Liu, Yuhan, Ma, Xiyao, Sun, Zhongkai, Han, Xu, Ma, Chengyuan, Yao, Benjamin Z., Guo, Chenlei

Abstract

Skills, reusable procedural guidance added at inference, can substantially improve LLM downstream performance (Li et al., 2026). Prior work retrieves skills from a bank by semantic relevance, then uses them as inference-time patches or for model distillation. The individual utility of each skill, however, is largely neglected. We first show that, in on-policy distillation where skill-conditioned policies serve as teachers, fewer than 25% of retrieved skills provide useful distillation signals. We then propose SGUID, a method for selecting a compact subset of skills for distillation. SGUID retains a skill only if it consistently yields effective learning signals during training. The selected skills are then distilled to produce a better model. Our results show that not all skills are worth distilling. Across four models from the Olmo and Qwen families, distilling 6 selected skills matches or exceeds full-bank distillation in mean avg@12 on three of the four models, and on all four after a second round that distills 3 newly selected skills, while the full banks are up to 11x larger. Importantly, SGUID supports stable model-skill co-evolution: after a distillation round, a new candidate bank is curated from the updated model's rollouts, and SGUID selects which skills to internalize next. In the second round, this loop selects 3 new skills and improves Qwen3-8B from 64.3% to 66.3%. The selection step is essential for stability: on Qwen3-4B, naively updating the model with unfiltered skills degrades performance, including a 0.3 percentage point drop on HMMT25, whereas SGUID improves HMMT25 by 0.5 points after the first round and 1.1 points after the second. These results identify skill selection as the key mechanism for stable model-skill co-evolution.

Chinese Translation

技能是添加在推理阶段的可重用程序指导，可以显著提高大型语言模型（LLM）的下游性能（Li et al., 2026）。先前的研究通过语义相关性从技能库中检索技能，然后将其用作推理时的补丁或模型提炼。然而，单个技能的实用性在很大程度上被忽视。我们首先展示，在技能条件策略作为教师的在线提炼中，检索到的技能中不足25%提供有用的提炼信号。随后，我们提出了SGUID，一种选择紧凑技能子集进行提炼的方法。SGUID仅保留在训练过程中始终产生有效学习信号的技能。然后，将选定的技能进行提炼，以生成更好的模型。我们的结果表明，并非所有技能都值得提炼。在Olmo和Qwen系列的四个模型中，提炼6个选定技能在四个模型中的三个模型上平均达到或超过了完整技能库提炼的mean avg@12，而在第二轮提炼中，提炼3个新选定技能后，四个模型均达到了这一效果，而完整技能库的规模大约是其11倍。重要的是，SGUID支持稳定的模型-技能共同演化：在一次提炼轮次后，从更新模型的回滚中策划出一个新的候选技能库，SGUID选择下一个要内化的技能。在第二轮中，这一循环选择了3个新技能，并将Qwen3-8B的性能从64.3%提高到66.3%。选择步骤对于稳定性至关重要：在Qwen3-4B上，简单地使用未过滤的技能更新模型会降低性能，包括HMMT25上下降0.3个百分点，而SGUID在第一轮后将HMMT25提高了0.5点，第二轮后提高了1.1点。这些结果表明，技能选择是实现稳定模型-技能共同演化的关键机制。

View on arXiv Download PDF AI Translation

cs.CL / 87 / 2610.12376

Latent Core Tokenizer: Compress, but Meaningfully

潜在核心分词器：压缩，但有意义

Ali, Felermino D. M. A., Ochieng, Millicent, Ekwejunor-Etchie, Ogbemi, Famoti, Ade, O'Neill, Jacki, Paul, Debjit

Abstract

Tokenizers are commonly optimized for compression, but a compact vocabulary does not necessarily distribute its capacity evenly across languages. We introduce the Latent Core Tokenizer (LCT), a language-agnostic approach that separates structural discovery from vocabulary construction. LCT uses Minimum Description Length, entropy-based boundary signals, and morphotactic constraints to identify reusable linguistic units before constructing a shared vocabulary. Across 104 languages with a 200K-token vocabulary, LCT achieves lower fertility and higher MorphScore than BPE, Unigram, and parity-aware BPE, while maintaining comparable cross-lingual disparity in tokenization cost. Across four multilingual downstream benchmarks, LCT improves aggregate score by 1.48, 1.83, and 2.00 points over BPE, Unigram, and parity-aware BPE, respectively. Our findings show that compression alone does not predict representation quality and highlight the importance of morphology-driven structural discovery and how frequency is used to allocate the final vocabulary across languages.

Chinese Translation

分词器通常优化以实现压缩，但紧凑的词汇表并不一定在不同语言之间均匀分配其容量。我们提出了潜在核心分词器（Latent Core Tokenizer, LCT），这是一种与语言无关的方法，它将结构发现与词汇构建分开。LCT使用最小描述长度（Minimum Description Length）、基于熵的边界信号和形态学约束来识别可重用的语言单位，然后再构建共享词汇表。在104种语言中，使用200K词汇的情况下，LCT在生育率（fertility）上低于BPE、Unigram和考虑平衡的BPE，同时在分词成本的跨语言差异上保持可比性。在四个多语言下游基准测试中，LCT分别比BPE、Unigram和考虑平衡的BPE提高了1.48、1.83和2.00分。我们的研究结果表明，单靠压缩并不能预测表示质量，并强调了形态驱动的结构发现的重要性，以及频率如何用于在不同语言之间分配最终词汇。

View on arXiv Download PDF AI Translation

cs.CL / 88 / 2610.12410

Predicting Alignment Generalization with Value Representations

基于价值表示的对齐泛化预测

Liu, Andy, Bhatia, Mehar, Stanczak, Karolina, Diab, Mona, Shwartz, Vered, Fried, Daniel

Abstract

LLM developers post-train their models to exhibit prosocial values and behavioral traits, which are enumerated in an alignment target. However, while recent post-training developments have yielded models that score highly on alignment evaluations, training models on sets of narrow behaviors still influences their behavior across unseen contexts and environments in unexpected ways. In this paper, we establish the task of alignment generalization prediction, i.e., predicting how fine-tuning a model to follow a given value changes its behavior across a wide range of held-out values. We conduct a large-scale analysis of alignment generalization effects across 66 values found in modern alignment targets, and benchmark representational techniques on the alignment generalization prediction task. We find that representations based on model activations when applying values in context significantly outperform methods based on textual descriptions of the values. Specifically, the best activations-based methods achieve correlations of 0.45 with our generalization matrix, compared with 0.05 from description-based baselines. We then show the applicability of representations that predict alignment generalization toward downstream tasks by using them to measure how similar the values in a multi-value alignment target are, which we find is significantly correlated with model robustness. Finally, we show initial evidence towards a shared, model-independent value space, which we use to develop the first taxonomy of LLM values grounded in empirical generalization dynamics. Our work demonstrates the importance of studying value generalization in LLMs and its application toward the more empirical design and training of model behavior.

Chinese Translation

大型语言模型（LLM）的开发者对其模型进行后训练，以展现亲社会价值观和行为特征，这些特征在对齐目标中被列举。然而，尽管近期的后训练发展已产生在对齐评估中得分较高的模型，但在狭窄行为集上训练模型仍会以意想不到的方式影响其在未见上下文和环境中的行为。在本文中，我们确立了对齐泛化预测的任务，即预测将模型微调以遵循给定价值观如何改变其在广泛保留价值观中的行为。我们对现代对齐目标中发现的66个价值观进行了大规模的对齐泛化效应分析，并在对齐泛化预测任务上对表示技术进行了基准测试。我们发现，当在上下文中应用价值观时，基于模型激活的表示显著优于基于价值观文本描述的方法。具体而言，基于激活的最佳方法与我们的泛化矩阵的相关性达到0.45，而基于描述的基线仅为0.05。随后，我们展示了预测对齐泛化的表示在下游任务中的适用性，通过使用它们来衡量多价值对齐目标中价值观的相似性，我们发现这与模型的鲁棒性显著相关。最后，我们展示了初步证据，表明存在一个共享的、模型无关的价值空间，我们利用这一空间开发了第一个基于经验泛化动态的LLM价值分类法。我们的研究表明，研究LLM中的价值泛化及其在模型行为的更具经验的设计和训练中的应用的重要性。

View on arXiv Download PDF AI Translation

arXiv Papers

Does Dynamic-Point Filtering Help When Texture Is Scarce? A Controlled Study of ORB-SLAM2 Front-Ends in Synthetic Indoor Scenes

Teaching a Robot Dog New Tricks: Diverse Quadruped Skills via Combined Reinforcement and Imitation Learning with Adversarial Task Selection

TacHair: Tactile Contact-Distribution Guided Online Correction for Robotic Hair Stroking and Perception

Masked Generative Motion Planning with Geometry-Guided Token Search

TAPNAV: Humanoid Navigation through Tactile Active Perception

NavGPT-3: Harnessing Context in a Hierarchical Navigation Runtime

Same Action, Different Outcome: Variability in Dynamic Cloth Manipulation

Diagnosing and Recovering from Observation-Space Shift at Long-Horizon Skill Seams

Skill-SLM: Agent Skill-driven Small Language Models for Reliable Robot Operation

Cross-Embodiment Robot Foundation World Models with Latent Actions

OmniHOI: Dexterous Hand-Object Interaction from Monocular Human Video

Informationally Decoupled Trajectory Design for Sim-to-Real System Identification

When Listening Becomes Easier: Scrubbing Visual Cues for Shortcut-Free VLAs

Higher-Order Morphology Priors for Quadruped Reinforcement Learning Under Actuator Degradation

iAm.md: Robot Skill Self-Assessment through Agentic Introspection for Unknown Open-Vocabulary Domains

ActiveReg: Information-Driven Active Regional Probing for Partial-to-Full Bone Registration

A Reconfigurable Fabric Based Pneumatic Actuator with Button Fastened Constraint Modules for Multi Mode Actuation

Towards Path-Creative Navigation: Robot Navigation through Embodied Interaction

FOCUS: From Privileged States to RGB-D with Controlled Modality Switching and Representation Alignment

Distributed Relative Localization Based on Ultra-WideBand and LiDAR for Multi-robot with Limited Communication

PMTRM: Pseudo-Memory Temporal Re-encoding Module for Embodied Policy Learning

Higher-Order Action Supervision Makes A Strong Policy Class

OmniDex: Scaling Dexterous Hand Grasping to Diverse Cluttered Scenes

Demonstrating Arena 5.0: A Photorealistic ROS2 Simulation Framework for Developing and Benchmarking Social Navigation

How Firm Should a Grasp Be?

SafeInferCom: Safe Inference-Time Compute via Verifier-Guided Mid-Generation Intervention for Robotic Task Planning

SimVLA: Zero-Shot Sim-to-Real VLA Learning for Mobile Manipulation

Being-M0.7: A Latent World-Action Model for Humanoid Robots

Distributed Relative Localization for Homogeneous Multi-Robot Systems through UWB Ranging and Limited Communications

USDCraft: Geometrically Grounded Programmatic Modeling of Articulated 3D Assets for Simulation

PlanWAM: Planning-Shaped Future Representations for End-to-End Autonomous Driving

WAM-Cache: Staleness-Bounded KV Reuse for Efficient World Action Models

Rewiring Semantics, Dynamics, and Control: A Simple yet Effective Action-Centric Tri-Stream Transformer

Experience-Guided Initiation Search for Learned Skills in Skill Composition

RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes

DAMP: Humanoid Locomotion via Denoised Belief Learning and Adversarial Motion Priors

WARP-VLA: Wrist-Camera Adaptation for View-Robust Policy Execution in Vision-Language-Action Models

RAGNAROK: Radar-Aided Gravity-Normalized Alignment for Robust Open Keyframe-based Radar-Visual-Kinematic-Inertial SLAM

CoCam4D: Geometry-Aware Cooperative 4D Perception for Camera-Only Autonomous Driving

SDPAD: A Fully Spike-Driven Pipeline for End-to-End Autonomous Driving

Acting from Belief, Looking When Needed: A Bayesian Spatial World Model for Navigation under Intermittent Perception

Learning Language-Conditioned Traversability Representations for Adaptive Visual Navigation

Neural Networks for Temporal Pattern Recognition and Dynamic Arm Gesture Speed Estimation for Robot Control

Scalable LEO Conjunction Screening using Adaptive Synthetic-Covariance Thresholds

YOCO: You Only Calibrate Once! Fast Mocap Calibration for Dexterous Teleoperation

From Solo to Ensemble: A Hierarchical Framework for Composable Multi-Agent Human-Object Interaction

2DGS-Planner: Rasterization-based Path Planning in 2D Gaussian Splatting Map

UltraLight Luma: A Novel Edge-Deployable Perception Network for Crop-Row Segmentation in Agricultural Robotics

PathTime-VLA: Path-Time Decoupling for Factorized Post-Training of Vision-Language-Action Policies

WAND: Learning Robust Navigation under Complex Wind Disturbances and Dense Obstacles for Quadrotors

STAG: A Sparse Traversability-Aware Graph Representation from Grid-Based Costmaps for Robotic Navigation

TACROSS: An Efficient and Low-Cost Scalable Human Touch System Across Heterogeneous Tactile Sensors for Dexterous Robot Learning

Tell Robot What Not to Do: A Negation Understanding Perspective

Reliability-Aware Future Conditioning for Temporally Robust Robot Manipulation

CAPABLE: Capability-Aware Policy Adaptation via Behavioral Latent Encoding

Predicting Cable Dynamics with Physical Attention Bias

REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models

Humanoid World Action Model With Joint State--Action Generation

Traceable World State: A Provenance-Aware State Representation and Deterministic Replay Framework for Robotic Systems

SkillWeave: Weaving Heterogeneous Demonstrations into Long-Horizon Manipulation Skills

A Minimal Optical-Flow Representation for Vision-Based Tactile Rotation Classification in Robotic Manipulation Across Gravity Domains

ManiUnit: A Manipulation Skill Dataset and Benchmark for Long-Horizon Tasks

UNITAS: A 3D-Native World Action Model for Embodied Manipulation

PIER: An Evidence-Gated Execution Interface for Robotic Manipulation

SuperNav: An Agentic Navigation System for Any Task in Any Scene

Leveraging Human-In-The-Loop Demonstrations in Reinforcement Learning for Digital Twin-Driven Robot Flexibility

Instance-anchored interaction evidence: Grounding robot plans in human pointing and handling

Unifying Policy Learning and State Prediction through Spatial Language Modeling

RESETTLE: Robotic Recovery through Disagreement-Triggered Retrieval and Efficient Corrective Control

MiniWAM: Learning Compact Future Targets for Efficient World-Action Modeling

MAP2: Model- and Acceleration-Based Pursuit with MPC and Gaussian Process Residual Learning for Autonomous Racing

Sim-to-Real RL for ASVs using SysID

Residual Modeling Closes the Regression and Generative Policy Gap in Robot Learning

From Language to Motion: Task-Conditioned Focal-Stack Trajectory Integration for Microscopic Robots

Fixed-Reference Pose Residuals for Measuring Cross-Dataset Cue Transfer in Human-Robot Interaction Anticipation

Real-Time Motion Planning with Dynamic Hazards: Classical vs. Learning-Based Methods

Walking on Roofs: Exploring the Potential of Walking Robots for Construction Work on Roofs

Toward Lunar Legged Robots: Field Deployment Lessons at LUNA

PLaW-VLA: Predictive Latent World Modeling for Vision-Language-Action Policies