cs.RO / 1 / 2610.06862
Generalizable Robustness Testing of DNN-Based Robotic Navigation Systems via XAI-Guided Search
基于可解释性人工智能引导搜索的深度神经网络机器人导航系统的可泛化鲁棒性测试
Abstract
**Context:** Deep Neural Networks (DNNs) increasingly control Cyber-Physical Systems (CPSs), yet small input perturbations can cause unsafe system-level behavior. Existing approaches often optimize perturbations for individual images and evaluate them only in simulation, limiting their generalizability and practical validity. **Objectives:** This work aims to generate robustness tests that remain effective across operational observations and to evaluate whether the resulting failures transfer from simulation to a physical robot. **Methods:** We propose an explainability-guided multi-objective evolutionary approach that generates sparse perturbations over representative images selected through visual and behavioral clustering. Aggregated Integrated Gradients guide mutations toward influential image regions. We evaluate the approach on a DNN-controlled LeoRover in Gazebo, conduct an ablation study, and validate a stratified subset of perturbations on the physical robot. **Results:** The approach achieved a median success rate of 70.0%, compared with 53.85% for unguided search, and increased median hypervolume from 0.65 to 0.73. Multi-image optimization improved the success rate from 50.0% to 57.5%, while XAI guidance further increased it to 70.0%. In the sim-to-real evaluation, simulation achieved 0.95 precision and 0.67 recall, and simulated and physical failure times showed a significant positive correlation of 0.617. **Conclusion:** Combining multi-image optimization with explainability-guided search improves robustness testing for DNN-controlled robotic systems. Simulation effectively identifies and prioritizes transferable failures, but physical validation remains necessary because some real-world failures are not reproduced in simulation.
Chinese Translation
背景：深度神经网络（DNN）越来越多地控制网络物理系统（CPS），然而小的输入扰动可能导致系统级别的不安全行为。现有方法通常针对单个图像优化扰动，并仅在仿真中进行评估，这限制了其可泛化性和实际有效性。目标：本研究旨在生成在操作观察中仍然有效的鲁棒性测试，并评估由此产生的失败是否能够从仿真转移到物理机器人。方法：我们提出了一种基于可解释性引导的多目标进化方法，通过视觉和行为聚类选择代表性图像，生成稀疏扰动。聚合的积分梯度引导突变朝向影响力较大的图像区域。我们在Gazebo中对DNN控制的LeoRover进行了评估，进行了消融研究，并在物理机器人上验证了一组分层的扰动。结果：该方法实现了70.0%的中位成功率，而无引导搜索的成功率为53.85%，中位超体积从0.65提高到0.73。多图像优化将成功率从50.0%提高到57.5%，而可解释性引导进一步将其提高到70.0%。在仿真到现实的评估中，仿真达到了0.95的精确度和0.67的召回率，仿真和物理失败时间显示出显著的正相关性（0.617）。结论：将多图像优化与可解释性引导搜索相结合，提高了DNN控制机器人系统的鲁棒性测试。仿真有效地识别并优先考虑可转移的失败，但由于一些现实世界的失败在仿真中未被重现，物理验证仍然是必要的。
cs.RO / 2 / 2610.06863
RMRRT: Riemannian Barrier Metric RRT for Inequality-Aware Steering on Equality Manifolds
RMRRT：用于平等流形上考虑不等式的引导的黎曼障碍度量RRT
Abstract
This paper presents a motion planning framework that unifies equality and inequality constraints within a single geometric formulation for sampling-based planning in high-dimensional robotic systems. In conventional sampling-based planners, equality constraints are typically enforced through projection, whereas inequality constraints are handled separately through binary validity checks such as collision testing, often leading to inefficient exploration. To address this limitation, we propose Riemannian Barrier Metric RRT (RMRRT), which constructs a unified local geometry for planning on equality-constrained manifolds. RMRRT first builds an ambient barrier metric from inequality-sensitive barrier terms and then induces a tangent-space metric via a (G)-orthogonal projection associated with the equality constraints. The resulting tangent-space metric is used consistently in both steering and nearest-neighbor selection, biasing exploration away from nearby inequality boundaries while preserving first-order equality consistency. In this work, the metric is instantiated from signed-distance-based geometric proxy inequalities to provide collision-informative tangent-space directions; hard feasibility is enforced separately through standard validity checks. Experimental results show that RMRRT achieves a 100% success rate across diverse constrained manipulation tasks in both simulation and real-world settings, while reducing planning time relative to representative constrained planning baselines. Ablation studies further demonstrate that the proposed metric improves exploration quality by reducing rejected samples and shortening path length. Experiment videos and source code are available at: https://rmrrt-anonymous.github.io
Chinese Translation
本文提出了一种运动规划框架，将平等约束和不等式约束统一在一个几何形式中，以便在高维机器人系统中进行基于采样的规划。在传统的基于采样的规划器中，平等约束通常通过投影来强制执行，而不等式约束则通过二元有效性检查（如碰撞测试）单独处理，这往往导致探索效率低下。为了解决这一局限性，我们提出了黎曼障碍度量RRT（RMRRT），该方法为在平等约束流形上进行规划构建了统一的局部几何。RMRRT首先从对不等式敏感的障碍项构建一个环境障碍度量，然后通过与平等约束相关的（G）正交投影引导切空间度量。所得到的切空间度量在引导和最近邻选择中一致使用，偏向于远离附近的不等式边界，同时保持一阶平等一致性。在本研究中，该度量是从基于符号距离的几何代理不等式实例化的，以提供与碰撞相关的切空间方向；硬可行性则通过标准有效性检查单独强制执行。实验结果表明，RMRRT在模拟和现实环境中的多样化约束操作任务中实现了100%的成功率，同时相对于代表性的约束规划基线减少了规划时间。消融研究进一步表明，所提出的度量通过减少被拒样本和缩短路径长度来提高探索质量。实验视频和源代码可在以下网址获取：https://rmrrt-anonymous.github.io
cs.RO / 3 / 2610.06882
Geometric Coherence via Weighted Matching for 3D Heterogeneous Multi-Agent Reach-Avoid Games
通过加权匹配实现三维异质多智能体规避-追逐游戏的几何一致性
Abstract
We study assignment quality in 3D heterogeneous multi-agent reach-avoid games and identify a recurring failure mode of cardinality-only matching in geometrically structured scenarios, which we term \emph{Geometric Sprawl}. In these cases, multiple maximum-cardinality assignments are available, but some induce spatially incoherent pairings and inefficient pursuit trajectories. Building on the evasion-space framework of Yan et al.~\cite{yan2022}, we introduce a cardinality-first weighted sequential matching method in which the Hamilton--Jacobi--Isaacs interception value $z_I(s,j)$ is used as a secondary assignment weight. Each sequential stage is solved with a min-cost max-flow backend, while the unweighted baseline uses the same solver with the weight term removed. We evaluate both methods on a deterministic 35-scenario benchmark (7 families & 5 initialization variants) under two regimes: a diagnostic stationary-unmatched setting and a hybrid saddle-point setting with goal-directed unmatched evaders. On this benchmark, the weighted method resolves 13 of 15 geometric stress-test instances that cause repeated timeouts for the unweighted baseline in the diagnostic regime, and under the hybrid regime improves mean captures from $3.20$ to $3.91$ (22\% increase) and mean interception height from $3.87$ to $5.36$ (40\% increase). We also report lower path tortuosity and lower angular-effort proxy values, suggesting smoother pursuit trajectories in this first-order simulation model. We release the simulator and benchmark suite at \href{https://github.com/Prajwal-Vijay/geometric-coherence-weighted-matching}{github.com/Prajwal-Vijay/geometric-coherence-weighted-matching} to support reproducible evaluation of assignment strategies for 3D reach-avoid games.
Chinese Translation
我们研究了三维异质多智能体规避-追逐游戏中的任务分配质量，并识别出在几何结构场景中，基于基数的匹配存在一种反复出现的失效模式，我们称之为“几何蔓延”（Geometric Sprawl）。在这些情况下，存在多个最大基数的分配，但其中一些会导致空间不一致的配对和低效的追逐轨迹。基于Yan等人的逃避空间框架，我们提出了一种基于基数优先的加权顺序匹配方法，其中哈密尔顿-雅可比-艾萨克斯拦截值 $z_I(s,j)$ 被用作次要分配权重。每个顺序阶段通过最小成本最大流后端求解，而未加权的基线使用相同的求解器，但去除了权重项。我们在一个确定性的35场景基准（7个家族和5个初始化变体）上评估了这两种方法，分为两种模式：诊断性静态未匹配设置和具有目标导向未匹配逃避者的混合鞍点设置。在该基准上，加权方法解决了15个几何压力测试实例中的13个，这些实例在诊断模式下导致未加权基线反复超时，而在混合模式下，平均捕获次数从 $3.20$ 提高到 $3.91$（增加22%），平均拦截高度从 $3.87$ 提高到 $5.36$（增加40%）。我们还报告了较低的路径扭曲度和较低的角度努力代理值，表明在这一阶模拟模型中追逐轨迹更加平滑。我们在 exttt{https://github.com/Prajwal-Vijay/geometric-coherence-weighted-matching} 发布了模拟器和基准套件，以支持对三维规避-追逐游戏的任务分配策略进行可重复评估。
cs.RO / 4 / 2610.06921
Does a Learned Corrector Beat a Simple Retreat? Evidence from a Frozen VLA
学习纠正器是否优于简单撤退？来自冻结的视觉-语言-行动（VLA）的证据
Abstract
Before deploying runtime recovery for a frozen vision-language-action (VLA) policy, one must establish that an intervention improves success beyond ordinary run-to-run variation and that its complexity adds value over a simple action. We evaluate these questions on frozen $\pi_{0.5}$ across four RoboTwin tasks. For each test seed, we pair rollouts with and without correction and include a same-seed base-policy re-run as a placebo. Seed-cluster intervals and prespecified comparison rules assess net gains against stochastic outcome changes. Across 3,888 paired episodes, the full pipeline raises success on beat_allowbreak block_allowbreak hammer by $+13.5$\,pp (95\% interval $[+9.4,+17.7]$), with no detectable gain on the other three tasks at the deployed weight. Among failed base episodes on the responsive task, $43.2\%$ succeed on a plain re-run, compared with $63.5\%$ after correction; many nominal rescues therefore reflect the base policy's own variability. A fixed-time trigger and scripted return to an earlier joint configuration produce a net gain with no detected difference from the learned pipeline across two rounds, although our prespecified equivalence criterion is not met consistently. Pausing and a constant-action control do not yield comparable gains. On this benchmark, the decision to intervene depends strongly on the task, and a paired placebo plus a simple retreat baseline are needed to establish what learned correction contributes.
Chinese Translation
在为冻结的视觉-语言-行动（VLA）策略部署运行时恢复之前，必须确定干预是否能在超越普通的运行间变化的基础上提高成功率，并且其复杂性是否相较于简单动作具有附加价值。我们在四个RoboTwin任务上评估了这些问题，针对冻结的$ ext{π}_{0.5}$进行实验。对于每个测试种子，我们将带有纠正和不带纠正的回放配对，并包含同种子基准策略的重跑作为安慰剂。种子聚类间隔和预设比较规则评估相对于随机结果变化的净收益。在3,888个配对回合中，完整的管道在beat_allowbreak block_allowbreak hammer任务上将成功率提高了$+13.5$个百分点（95 ext{ }%区间$[+9.4,+17.7]$），而在部署权重下，其他三个任务没有检测到收益。在响应任务中，失败的基准回合中，$43.2 ext{ }%$在简单重跑中成功，而在纠正后成功率为$63.5 ext{ }%$；因此，许多名义上的救援反映了基准策略自身的变异性。固定时间触发和脚本化返回到早期联合配置产生了净收益，但在两轮中未检测到与学习管道的差异，尽管我们的预设等效标准并未始终满足。暂停和恒定动作控制未能产生可比的收益。在这一基准上，干预的决策在很大程度上依赖于任务，并且需要配对安慰剂加上简单撤退基线来确定学习纠正所贡献的内容。
cs.RO / 5 / 2610.06926
SWAP: Stepwise Action Policy Routing for Vision-Language-Action Models
SWAP：针对视觉-语言-动作模型的逐步动作策略路由
Abstract
Robot manipulation systems using Vision-Language-Action (VLA) model backbones typically use just one VLA for task execution. However, individual VLAs do not perform well across different task states and environments. We introduce a framework for dynamically composing multiple VLA policies during execution: StepWise Action Policy Routing (SWAP). SWAP formulates policy routing as an offline reinforcement learning problem, learning a routing critic that selects the most appropriate policy at each decision step given the current observation. SWAP enables robots to select new policies to execute online rather than committing to a single policy for the duration of an episode. We evaluate SWAP on both real-world DROID manipulation tasks and LIBERO simulation experiments. SWAP improves over fixed-policy execution and routing baselines, giving absolute improvements in real-world task success up to 33% while reducing successful trajectory robot action step length by 28.3%.
Chinese Translation
使用视觉-语言-动作（VLA）模型骨干的机器人操控系统通常仅使用一个VLA进行任务执行。然而，单个VLA在不同任务状态和环境下的表现并不理想。我们提出了一种在执行过程中动态组合多个VLA策略的框架：逐步动作策略路由（SWAP）。SWAP将策略路由形式化为一个离线强化学习问题，学习一个路由评估器，该评估器在每个决策步骤中根据当前观察选择最合适的策略。SWAP使机器人能够在线选择新的策略进行执行，而不是在整个阶段内坚持使用单一策略。我们在真实世界的DROID操控任务和LIBERO仿真实验中评估了SWAP。SWAP在固定策略执行和路由基线之上取得了改进，在真实世界任务成功率上绝对提升达33%，同时将成功轨迹的机器人动作步骤长度减少了28.3%。
cs.RO / 6 / 2610.06929
Physical Twins: Accelerating and Enabling Robot Learning with Phantom Platforms
物理双胞胎：通过幻影平台加速和促进机器人学习
Abstract
Improvements in human-robot physical interaction (pHRI) can have major implications for physical therapy, search and rescue, and telemedicine. However, a major challenge concerns human constraints and safety in human-robot physical experiments. Concerns about human studies also include repeatability, scalability, and participant diversity. To conduct such experiments, an IRB and willing human participants are required. In this work, we present an improved phantom device, a physical twin, that enables real-world RL-type testing for physically interactive algorithms. The new device not only replicates the ball-and-socket motion of the shoulder but also renders scapular and protraction/retraction motions. The experiments showcase the device's ability to render multiple 3D joint limits and demonstrate a Franka Panda arm sensing limits and interacting with the device as an arm for ADLs (activities of daily living) and pHRI tasks.
Chinese Translation
人机物理交互（pHRI）的改善对物理治疗、搜索与救援以及远程医疗具有重要影响。然而，一个主要挑战是人类在机器人物理实验中的限制和安全性。关于人类研究的担忧还包括可重复性、可扩展性和参与者多样性。进行此类实验需要伦理审查委员会（IRB）和愿意参与的志愿者。在本研究中，我们提出了一种改进的幻影设备，即物理双胞胎，能够为物理交互算法提供现实世界的强化学习（RL）类型测试。该新设备不仅复制了肩关节的球窝运动，还能呈现肩胛骨及前伸/后缩运动。实验展示了该设备能够呈现多个三维关节限制，并演示了Franka Panda机械臂感知限制并与该设备进行交互，执行日常生活活动（ADLs）和pHRI任务。
cs.RO / 7 / 2610.06955
ROMA: LLM System for Real-World Object-Centric Multi-Sensory Active Perception
ROMA：用于现实世界物体中心多感知主动感知的LLM系统
Abstract
Humans inherently understand the physical world through an active process. When sensory evidence is insufficient to infer physical properties, we naturally interact with the environment by deciding what information is missing, how to acquire it, and when sufficient evidence has been obtained. In stark contrast, existing multi-sensory robot systems mainly integrate sensory inputs rather than actively acquiring missing evidence through interactions. In this work, we introduce ROMA, an LLM-based system for Real-World Object-Centric Multi-Sensory Active Perception. ROMA integrates vision, audio, tactile, and force sensing into a reasoning-interaction-feedback loop. The model identifies missing evidence and determines the target objects, interactions, and modalities, while a physical interface executes the selected interactions and collects the multi-sensory feedback. To support this capability, we construct ROMI-2K, a large-scale real-world multi-sensory object interaction dataset covering nearly 2,000 objects and 6 atomic interactions with synchronized sensory feedback. Building on these data, we develop a two-stage training framework that aligns sensory modalities and equips the LLM to assess evidence sufficiency, select informative interactions, and reason over the multi-sensory feedback. We further characterize active perception as perception chains, where acquired evidence guides subsequent interactions and reasoning, and establish ROMA Bench to evaluate single-attribute, long-horizon multi-attribute, and intent-driven active perception. Experiments show that ROMA can actively acquire missing evidence and solve complex, long-chain multi-sensory perception tasks that existing methods struggle to handle, laying a strong perceptual foundation for active multi-sensory embodied agents.
Chinese Translation
人类通过主动过程固有地理解物理世界。当感官证据不足以推断物理属性时，我们自然会通过决定缺失的信息、如何获取这些信息以及何时获得足够的证据来与环境互动。与此形成鲜明对比的是，现有的多感知机器人系统主要整合感官输入，而不是通过互动主动获取缺失的证据。在本研究中，我们介绍了ROMA，一个基于LLM的现实世界物体中心多感知主动感知系统。ROMA将视觉、音频、触觉和力感知整合到一个推理-互动-反馈循环中。该模型识别缺失的证据，并确定目标物体、互动和感知方式，同时物理接口执行所选的互动并收集多感知反馈。为了支持这一能力，我们构建了ROMI-2K，一个大规模的现实世界多感知物体互动数据集，涵盖近2000个物体和6种原子互动，并具有同步的感官反馈。在这些数据的基础上，我们开发了一个两阶段的训练框架，旨在对齐感知方式，并使LLM能够评估证据的充分性、选择信息丰富的互动并对多感知反馈进行推理。我们进一步将主动感知表征为感知链，其中获取的证据指导后续的互动和推理，并建立了ROMA Bench来评估单属性、长时间跨度的多属性和意图驱动的主动感知。实验表明，ROMA能够主动获取缺失的证据，并解决现有方法难以处理的复杂、长链的多感知感知任务，为主动多感知具身代理奠定了坚实的感知基础。
cs.RO / 8 / 2610.06958
Learning Modular Policy for Multi-Floor Object Navigation:A Factorized Framework for Diagnostic Study
多层楼物体导航的模块化策略学习：用于诊断研究的分解框架
Abstract
Object-goal navigation (ObjectNav) in multi-floor scenarios presents a challenge due to sparse rewards caused by long-horizon decision-making. In this paper, we propose a diagnostic study based on a modular framework with an effective learnable policy to analyze failure factors in multi-floor scenarios. To achieve an effective policy for diagnosis, we design the hierarchical factorization policy that deconstructs a single global policy into an intra-floor exploration policy and an inter-floor switching policy. To providing an effective initialization for Reinforcement Learning (RL), the lightweight intra-floor policy is learned by distilling the exploration logic of Visual Language Models (VLMs). Under idealized assumptions, we show that the factorized policy is theoretically equivalent to a single global policy at the policy-representation level. Experiment results indicate that perception performance and stair climbing stability are the primary bottlenecks in multi-floor navigation.
Chinese Translation
在多层楼场景中，物体目标导航（ObjectNav）因长时间决策导致的稀疏奖励而面临挑战。本文提出了一项基于模块化框架的诊断研究，旨在通过有效的可学习策略分析多层楼场景中的失败因素。为了实现有效的诊断策略，我们设计了分层分解策略，将单一的全局策略分解为楼层内探索策略和楼层间切换策略。为了为强化学习（Reinforcement Learning, RL）提供有效的初始化，轻量级的楼层内策略通过提炼视觉语言模型（Visual Language Models, VLMs）的探索逻辑进行学习。在理想化假设下，我们证明了分解策略在策略表示层面上与单一全局策略是理论等价的。实验结果表明，感知性能和楼梯攀爬稳定性是多层楼导航的主要瓶颈。
cs.RO / 9 / 2610.06965
ACG-WAM: World-Action Modeling via Action-Conditioned Geometric Latent Prediction
ACG-WAM：通过动作条件几何潜在预测进行世界动作建模
Abstract
World action models jointly learn visual predictionand robot actions, providing a way to use observations ofscene evolution for policy learning. Their video and actionlosses, however, provide no explicit target for the geometricconsequences of a demonstrated action sequence. Moreover,visual features taken after temporal attention can contain futureobservations, making them unsuitable as the sole current visualinput to an auxiliary predictor. We introduce ACG-WAMand its auxiliary objective, the Action-Conditioned GeometricJoint-Embedding Predictive Architecture (ACG-JEPA), whichpredicts geometric features at several horizons from the currentobservation and intervening actions, using the future slot of afrozen VGGT encoding of each current and future image pairas the target. We apply this supervision from the head and wristcameras to a shared visual embedding before temporal mixing,and remove the teacher and auxiliary modules at inference.On 50 RoboTwin 2.0 tasks, ACG-WAM achieves 93.46%success in clean scenes, with the best randomized success(92.68%) and mean across both settings (93.07%) among thecompared methods; across three tasks on a real robot, itachieves 85.00% success and 91.67% partial completion score,exceeding Motus by 10.00 and 9.17 percentage points, respec-tively. Code:https://github.com/RoboOpus/ACG-WAM.Website:https://RoboOpus.github.io/ACG-WAM.
Chinese Translation
世界动作模型共同学习视觉预测和机器人动作，为利用场景演变的观察进行策略学习提供了一种方法。然而，它们的视频和动作损失并未为演示的动作序列的几何后果提供明确的目标。此外，经过时间注意力处理后的视觉特征可能包含未来的观察，使其不适合作为辅助预测器的唯一当前视觉输入。我们提出了ACG-WAM及其辅助目标——动作条件几何联合嵌入预测架构（ACG-JEPA），该架构使用当前观察和介入动作预测多个时间范围内的几何特征，并将每对当前和未来图像的冻结VGGT编码的未来时隙作为目标。在推理时，我们将来自头部和手腕摄像头的监督应用于共享视觉嵌入，然后进行时间混合，并去除教师和辅助模块。在50个RoboTwin 2.0任务中，ACG-WAM在干净场景中实现了93.46%的成功率，在比较方法中获得了最佳随机成功率（92.68%）和两种设置下的均值（93.07%）；在真实机器人上的三个任务中，它实现了85.00%的成功率和91.67%的部分完成分数，分别超过了Motus 10.00和9.17个百分点。代码：https://github.com/RoboOpus/ACG-WAM。网站：https://RoboOpus.github.io/ACG-WAM。
cs.RO / 10 / 2610.06999
ProactiveVLA: Augmenting Embodied Memory through Proactive Environment Exploration
ProactiveVLA：通过主动环境探索增强具身记忆
Abstract
Rapid adaptation to a new environment requires a robot to acquire useful knowledge about local objects, states, and interactions from limited experience. Systems that combine a reasoning agent with a frozen vision-language-action model (VLA) can adapt through execution feedback and memory, making the choice of experience central to their effectiveness. Repeated practice of a target task may refine a familiar solution while leaving other interactions relevant to changed conditions untested. We introduce ProactiveVLA, which uses proactive environment exploration to acquire reusable knowledge for deployment-time adaptation. After completing an initial task, the agent allocates the remaining interaction budget to self-proposed goals covering object affordances, state-changing interactions, and compositions of interactions. It verifies execution outcomes and consolidates both task-directed and exploratory experience into memory that guides subsequent planning and control. ProactiveVLA outperforms the baselines under the same turn budget on LIBERO-Pro and RoboCasa365 Composite-Seen. On LIBERO-Pro Goal-T, with at most one VLA primitive invocation allowed during evaluation, ProactiveVLA completes 48% of instances, compared with 19% for the state-of-the-art task-refinement baseline.
Chinese Translation
快速适应新环境要求机器人从有限的经验中获取关于当地物体、状态和交互的有用知识。将推理代理与冻结的视觉-语言-行动模型（VLA）相结合的系统可以通过执行反馈和记忆进行适应，使经验的选择成为其有效性的核心。对目标任务的重复练习可能会精炼熟悉的解决方案，而未能测试与变化条件相关的其他交互。我们提出了ProactiveVLA，它利用主动环境探索来获取可重用的知识，以便在部署时进行适应。在完成初始任务后，代理将剩余的交互预算分配给自我提出的目标，这些目标涵盖物体可用性、状态变化交互和交互组合。它验证执行结果，并将任务导向和探索性经验整合到记忆中，以指导后续的规划和控制。在相同的回合预算下，ProactiveVLA在LIBERO-Pro和RoboCasa365 Composite-Seen上超越了基线。在LIBERO-Pro Goal-T上，评估期间最多允许一次VLA原语调用，ProactiveVLA完成了48%的实例，而最先进的任务精炼基线仅完成了19%。
cs.RO / 11 / 2610.07015
LEAP: Making Privileged Geometry Supervision Effective for Visuomotor Learning
LEAP：使特权几何监督在视觉运动学习中有效
Abstract
Privileged 3D supervision uses additional geometric information during training to guide RGB-based visuomotor policy learning, without requiring geometric inputs at deployment. However, low reconstruction error does not ensure that visual representations capture geometry useful for control. We identify three limitations that weaken this supervision: proprioceptive shortcuts, dominant-view reliance, and reconstruction objectives dominated by task-irrelevant geometry. To address these limitations, we propose Latent Encoding with Aligned Privileged Geometry (LEAP). Our framework uses an auxiliary decoder to reconstruct point clouds within the manipulation workspace from visual features alone, while retaining proprioception for action prediction. Alongside full reconstruction, we introduce wrist-view dropout and partial reconstruction targets matched to the retained wrist, encouraging the encoder to capture complementary local geometry. The auxiliary decoder is removed at inference, leaving only RGB observations and robot state as policy inputs. Extensive experiments on RoboTwin, ManiSkill, and real-world tasks demonstrate consistent and substantial improvements over Diffusion Policy and ACT, with only a small increase in parameter count and no reduction in inference speed.
Chinese Translation
特权三维监督在训练过程中使用额外的几何信息来指导基于RGB的视觉运动策略学习，而在部署时不需要几何输入。然而，低重建误差并不能确保视觉表征捕捉到对控制有用的几何信息。我们识别出三种限制因素，削弱了这种监督：本体感觉捷径、对主视图的依赖，以及由与任务无关的几何信息主导的重建目标。为了解决这些限制，我们提出了与对齐特权几何的潜在编码（Latent Encoding with Aligned Privileged Geometry，LEAP）。我们的框架使用辅助解码器仅从视觉特征重建操作工作空间内的点云，同时保留本体感觉用于动作预测。除了完全重建外，我们引入了腕部视图丢弃和与保留的腕部匹配的部分重建目标，鼓励编码器捕捉互补的局部几何信息。在推理时，辅助解码器被移除，仅保留RGB观察和机器人状态作为策略输入。在RoboTwin、ManiSkill和现实世界任务上的大量实验表明，与扩散策略（Diffusion Policy）和ACT相比，我们的方法在参数数量仅略有增加且推理速度没有降低的情况下，取得了一致且显著的改进。
cs.RO / 12 / 2610.07052
BRACE: Adapting Whole-Body References for Force and Terrain Aware Humanoid Motion Tracking
BRACE：适应全身参考的力与地形感知类人运动跟踪
Abstract
Whole-body tracking has become the interface through which operators drive humanoid robots, yet the references it consumes are recorded on level ground and carrying nothing, so the tracker is aware of neither the forces the robot must exchange with objects nor the terrain it must stand on. Existing controllers address one side of this gap: force-capable policies command an end-effector force but prescribe no whole-body pose, while terrain-adaptive trackers treat loads as disturbances to reject rather than wrenches to command. We present BRACE, a whole-body tracker that exerts and compensates commanded hand forces from diverse poses while following a flat-ground reference on sloped terrain. Rather than leaving the tracker to absorb the load and the slope, BRACE folds both into the reference it follows: terrain conformance lifts footholds and root onto the local surface, and a wrench transformation resolves the hand displacement that produces a force jointly with the center-of-mass and center- of-pressure shifts it induces, bounded by teacher-specific arm-effort limits. Separate exertion and compensation teachers are distilled by DAgger into one flow-matching student that runs on proprioception alone, without a height map or measured wrench, so a teleoperator (even in remote locations) can supply a flat- ground trajectory and the robot resolves slope and load onboard. We extensive experiments on the Unitree-G1 to demonstrate the ability of BRACE to exert, compensate force and handle diverse terrain in simulation and real robot experiments.
Chinese Translation
全身跟踪已成为操作员驱动类人机器人所使用的接口，然而其所依赖的参考是在平坦地面上记录的，并且没有任何负载，因此跟踪器既无法感知机器人必须与物体交换的力，也无法感知其必须站立的地形。现有控制器解决了这一差距的某一方面：具备力控制能力的策略指令终端执行器施加力，但并未规定全身姿态，而地形适应型跟踪器则将负载视为需要拒绝的干扰，而非需要指令的扭矩。我们提出了BRACE，这是一种全身跟踪器，能够在倾斜地形上遵循平坦地面的参考，同时施加和补偿来自不同姿态的手部指令力。BRACE并未让跟踪器自行吸收负载和坡度，而是将两者都融入其遵循的参考中：地形适应性将足部支撑点提升至局部表面，而扭矩变换则解决了产生力的手部位移，同时考虑到其引起的质心和压力中心的位移，受限于特定教师的臂力限制。通过DAgger将独立的施力和补偿教师提炼为一个仅依赖本体感知的流匹配学生，该学生无需高度图或测量的扭矩，因此远程操作员（即使在偏远地点）可以提供平坦地面的轨迹，机器人则在内部解决坡度和负载。我们在Unitree-G1上进行了广泛实验，以展示BRACE在仿真和真实机器人实验中施加、补偿力以及处理多样地形的能力。
cs.RO / 13 / 2610.07056
Behavioral Cloning Mystery
行为克隆之谜
Abstract
Behavioral cloning (BC), despite its simplicity, exhibits many counterintuitive phenomena in the real world. For example, the performance of BC often keeps increasing as the model overfits more to the dataset, and fully closed-loop policies often completely fail without action chunking. Unfortunately, properly studying these anecdotal phenomena ("behavioral cloning mysteries") is challenging: in the real world, datasets and experiments are costly and not fully controllable; in simulation with synthetic data, these phenomena are often not easily observed partly due to the discrepancy between scripted policies and human demonstrations. In this work, we propose OCBench, a robotic manipulation benchmark with controllable scripted policies that have similar properties to human demonstrations. We show that, by mimicking key properties of human demonstrations, OCBench reproduces many anecdotal BC-related phenomena in controlled settings. With its GPU-accelerated environments and scripted policies, we demonstrate how OCBench enables scientific studies of previously reported BC-related phenomena by analyzing and refuting various hypotheses. Project page: https://seohong.me/projects/ocbench
Chinese Translation
尽管行为克隆（Behavioral Cloning, BC）方法简单，但在现实世界中却表现出许多反直觉的现象。例如，BC的性能往往随着模型对数据集的过拟合而不断提高，而完全闭环的策略在没有动作分块的情况下常常会完全失败。不幸的是，正确研究这些轶事现象（“行为克隆之谜”）是具有挑战性的：在现实世界中，数据集和实验成本高昂且不可完全控制；而在使用合成数据的模拟中，这些现象往往不易观察，部分原因是脚本化策略与人类演示之间存在差异。在本研究中，我们提出了OCBench，这是一个具有可控脚本化策略的机器人操作基准，这些策略具有与人类演示相似的特性。我们展示了通过模仿人类演示的关键特性，OCBench能够在受控环境中重现许多与BC相关的轶事现象。借助其GPU加速的环境和脚本化策略，我们展示了OCBench如何通过分析和反驳各种假设，促进对先前报告的BC相关现象的科学研究。项目页面：https://seohong.me/projects/ocbench
cs.RO / 14 / 2610.07081
Demo: Vision-Language Model-Guided Online Calibration of an Electromagnetic Digital Twin
演示：基于视觉-语言模型的电磁数字双胞胎在线校准
Abstract
An electromagnetic (EM) digital twin gives mobile robots wireless situational awareness but depends on material conductivities that change with the environment. Online calibration faces initialization sensitivity and measurement travel costs. We demonstrate a vision-language model (VLM)-guided framework using a Unitree G1 robot and NVIDIA Sionna, with two VLM calls: material classification maps visible materials through ITU-R P.2040 to conductivity priors for Sionna's gradient descent on accumulated received signal strength (RSS) measurements; waypoint planning selects the next measurement location online using residual RSS calibration error and image coverage. In a real indoor scenario, the framework achieves a normalized mean absolute conductivity error of $1.74\times10^{-4}$ within 20 m of travel; random initialization never converges, while random waypoints require over twice the travel.
Chinese Translation
电磁（EM）数字双胞胎为移动机器人提供无线环境感知，但依赖于随环境变化的材料导电性。在线校准面临初始化敏感性和测量旅行成本的问题。我们展示了一种基于视觉-语言模型（VLM）指导的框架，使用Unitree G1机器人和NVIDIA Sionna，进行两次VLM调用：材料分类图通过ITU-R P.2040将可见材料映射到Sionna的导电性先验，并在累积接收信号强度（RSS）测量上进行梯度下降；路径规划在线选择下一个测量位置，使用残余RSS校准误差和图像覆盖率。在一个真实的室内场景中，该框架在20米的旅行范围内实现了标准化平均绝对导电性误差为$1.74 imes10^{-4}$；随机初始化从未收敛，而随机路径点的旅行需求超过两倍。
cs.RO / 15 / 2610.07096
TeleHairing: A Teleoperation Baseline for Robotic Haircutting
远程理发：机器人理发的远程操作基准
Abstract
Robotic haircutting requires controlled tool motion near the head while simultaneously accounting for communication, visual feedback, tool actuation, and interruption handling. Existing studies still lack an operator-in-the-loop reference for analyzing these coupled behaviors before human trials or stronger autonomy. This paper presents TeleHairing, a closed-loop teleoperation architecture for mannequin-based robotic haircutting evaluation under local, relay, and remote deployment conditions. Logged timing shows that the main remote latency increase occurs before the robot-side control endpoint: overall timing reached 190.5~ms in remote mode, while robot-side command queue, control processing, and control-to-robot timing remained similar across modes. Trajectory analysis shows that the larger remote command-following error was dominated by the terminal withdrawal segment rather than accumulated uniformly over the path; excluding this segment reduced remote root-mean-square error (RMSE) from 45.1 mm to 9.6 mm. Detection-loss trials further show that rebase events resumed motion without a large target jump under the tested condition. These results clarify how deployment, execution, and interruption affect the robotic haircutting teleoperation loop, providing a quantitative reference for future autonomy, safety, and user-facing studies.
Chinese Translation
机器人理发需要在靠近头部的区域控制工具运动，同时考虑通信、视觉反馈、工具驱动和中断处理。现有研究仍缺乏操作员参与的参考，以便在进行人类试验或实现更强的自主性之前分析这些耦合行为。本文提出了TeleHairing，一种闭环远程操作架构，用于在本地、中继和远程部署条件下对基于假人的机器人理发进行评估。记录的时序数据显示，主要的远程延迟增加发生在机器人端控制端点之前：在远程模式下，总体时序达到了190.5毫秒，而机器人端命令队列、控制处理和控制到机器人时序在不同模式下保持相似。轨迹分析表明，较大的远程命令跟随误差主要由末端撤回段主导，而不是在路径上均匀累积；排除该段后，远程均方根误差（RMSE）从45.1毫米降至9.6毫米。检测丢失试验进一步表明，在测试条件下，重置事件在没有大幅目标跳跃的情况下恢复了运动。这些结果阐明了部署、执行和中断如何影响机器人理发的远程操作循环，为未来的自主性、安全性和用户相关研究提供了定量参考。
cs.RO / 16 / 2610.07116
AIM: Adaptive Interaction Modeling Networks for Real-to-Sim Soft-Body Simulation
AIM：用于真实到仿真软体模拟的自适应交互建模网络
Abstract
Deformable-object manipulation is essential for robotic tasks such as folding laundry and handling food, where robots must control shape changes as well as object motion. Predictive soft-body simulation supports these tasks by anticipating deformation under external interactions. However, spatial neighborhoods can misrepresent deformation dependencies, introducing local errors that accumulate over successive predictions. Models fitted to individual scenes must also accommodate changes in object geometry and manipulation conditions. In this work, we propose AIM, an Adaptive Interaction Modeling framework that treats real-to-sim soft-body simulation as a local-global interaction modeling problem. AIM uses motion history and geometry to adapt particle relations over current spatial neighbors and retained connections, while geometry-conditioned global communication coordinates object-wide responses. A unified kinematic control-point interface represents different manipulation configurations, and multi-step autoregressive supervision trains the model on its own predicted trajectories. Experiments on PhysTwin and PGND demonstrate improved motion accuracy and visual fidelity, with a 20.0% reduction in future-prediction tracking error relative to PhysTwin and a 22.8% reduction in mean long-horizon particle error across six object categories relative to PGND. The framework further supports transfer across actions, object instances, and scenes, including zero-shot transfer from robot interactions to human manipulation without target-domain dynamics fitting.
Chinese Translation
可变形物体的操作对于机器人任务至关重要，例如折叠衣物和处理食品，机器人必须控制形状变化以及物体运动。预测性软体模拟通过预见外部交互下的变形来支持这些任务。然而，空间邻域可能会错误地表示变形依赖关系，引入局部误差，这些误差在连续预测中会累积。针对单个场景拟合的模型还必须适应物体几何形状和操作条件的变化。在本研究中，我们提出了AIM（自适应交互建模框架），将真实到仿真的软体模拟视为一个局部-全局交互建模问题。AIM利用运动历史和几何信息来调整当前空间邻域和保留连接的粒子关系，同时几何条件下的全局通信协调物体范围内的响应。一个统一的运动控制点接口表示不同的操作配置，而多步自回归监督则在模型自身预测的轨迹上进行训练。在PhysTwin和PGND上的实验表明，运动精度和视觉保真度得到了改善，相较于PhysTwin，未来预测跟踪误差减少了20.0%，而在六个物体类别中，相较于PGND，平均长时间粒子误差减少了22.8%。该框架进一步支持在动作、物体实例和场景之间的迁移，包括从机器人交互到人类操作的零样本迁移，而无需针对目标领域动态进行拟合。
cs.RO / 17 / 2610.07217
RoboCap: A New Platform for Egocentric Robot Learning
RoboCap：一种新的自我中心机器人学习平台
Abstract
Despite its promise for scaling robot learning, egocentric manipulation data is still scarce today. Collection at scale requires vertically integrating ergonomic hardware with centimeter-precise 3D algorithms, at a precision that has not been publicly demonstrated. To address this gap, we introduce RoboCap, a 250\,g six-camera dual-IMU hat designed for in-the-wild egocentric data capture, and the Grounded API, a suite of device-agnostic 3D algorithms tuned for RoboCap. In this report, we demonstrate how hardware, calibration, and 3D algorithms interact to achieve state-of-the-art performance on the public benchmarks: our SLAM across diverse settings and rigs, our depth estimation on egocentric settings, and our hand tracking when adapted to third-party devices.
Chinese Translation
尽管自我中心操控数据在扩展机器人学习方面具有潜力，但目前仍然稀缺。大规模收集需要将符合人体工程学的硬件与厘米级精度的3D算法垂直集成，而这种精度尚未得到公开验证。为了解决这一问题，我们推出了RoboCap，这是一款重250克的六摄像头双IMU帽子，旨在进行野外自我中心数据捕获，并推出了Grounded API，这是一套针对RoboCap调优的设备无关3D算法。在本报告中，我们展示了硬件、校准和3D算法如何相互作用，以在公共基准测试中实现最先进的性能：我们在不同设置和设备上的SLAM，我们在自我中心设置上的深度估计，以及我们在适配第三方设备时的手部追踪。
cs.RO / 18 / 2610.07231
Monocular Navigation Relative to Unknown Spacecraft Using a Transformer-Aided Kalman Filter
基于变换器辅助卡尔曼滤波器的单目导航相对于未知航天器
Abstract
This work presents a novel learning-based pipeline for pose estimation of unknown spacecraft using only monocular images from a single servicer. The approach combines a transformer-based neural network with a Multi-State Constraint Kalman Filter (MSCKF) to estimate the pose (i.e., position and orientation of the target spacecraft relative to the camera) throughout rendezvous and proximity operations. Unlike existing vision-based methods that require prior knowledge of the target shape or inertia properties, rely on additional sensing modalities such as depth, lidar, or stereo, or only recover translation up to scale, the proposed pipeline generalizes to previously unseen spacecraft using a single monocular camera. The transformer network estimates the odometry, the change in pose between images up to scale, from SuperPoint features matched by LightGlue. The MSCKF uses these pseudo-measurements along with an orbit and attitude kinematics model to estimate the pose of the target. In particular, the relative orbit elements, the target's attitude with respect to the servicer's camera, and the associated angular velocity are estimated directly by the filter. Given the monocular approach and short distance to the target, the full observability of the range to the target is recovered via attitude maneuvers by the servicer. The method is trained and evaluated on a re-rendered high-resolution version of the SPE3R dataset, which includes synthetic images of 103 spacecraft. Eleven of these spacecraft are held out during training to evaluate the generalization to unseen targets. Monte Carlo simulations are then used to evaluate the navigation pipeline on rendered trajectories of the held out spacecraft. The results demonstrate that learned vision pipelines as a front-end for Kalman filters provide median errors of 3.7{\deg} in attitude and 2.2% of range in ROE when navigating about unknown targets.
Chinese Translation
本研究提出了一种新颖的基于学习的管道，用于仅利用单个服务器的单目图像对未知航天器进行姿态估计。该方法将基于变换器的神经网络与多状态约束卡尔曼滤波器（Multi-State Constraint Kalman Filter, MSCKF）相结合，以在交会和接近操作中估计目标航天器相对于相机的姿态（即位置和方向）。与现有的视觉方法不同，后者需要目标形状或惯性特性的先验知识，依赖于深度、激光雷达或立体等额外传感方式，或仅能恢复比例不变的平移，所提管道能够使用单个单目相机对以前未见过的航天器进行推广。变换器网络通过LightGlue匹配的SuperPoint特征估计里程计，即图像之间的姿态变化（比例不变）。MSCKF利用这些伪测量值以及轨道和姿态运动学模型来估计目标的姿态。特别地，相对轨道元素、目标相对于服务器相机的姿态以及相关的角速度由滤波器直接估计。考虑到单目方法和与目标的短距离，服务器通过姿态机动恢复到目标的距离的完全可观测性。该方法在重新渲染的高分辨率SPE3R数据集上进行训练和评估，该数据集包括103个航天器的合成图像。在训练期间，保留其中11个航天器以评估对未见目标的推广能力。然后使用蒙特卡洛模拟评估在保留航天器的渲染轨迹上的导航管道。结果表明，作为卡尔曼滤波器前端的学习视觉管道在导航未知目标时提供了姿态的中位误差为3.7°，范围的中位误差为2.2%（相对轨道元素）。
cs.RO / 19 / 2610.07245
Robust Nonprehensile Object Transport with Quadruped Robots
四足机器人稳健的非抓取物体运输
Abstract
In this paper, we present a robust nonprehensile object transportation framework for quadruped robots. An uncertainty-aware trajectory optimization method generates object motions with minimal closed-loop sensitivity to uncertain parameters. The resulting reference trajectory is tracked using a coupled convex model predictive controller that jointly predicts the CoM dynamics of the quadruped and the payload followed by a whole-body QP that enforces ground reaction constraints. The approach is evaluated through extensive simulations and real-world experiments under variations in the object's inertial parameters. Its performance is compared with fixed-orientation and straight-line trajectories as baseline. The results show that the optimized object motion reduces the sliding by approximately 50% compared with the fixed-orientation baseline and 30% compared with the straight-line baseline, while also achieving lower robot CoM tracking errors.
Chinese Translation
本文提出了一种针对四足机器人的稳健非抓取物体运输框架。该框架采用一种考虑不确定性的轨迹优化方法，以生成对不确定参数具有最小闭环灵敏度的物体运动。生成的参考轨迹通过一个耦合的凸模型预测控制器进行跟踪，该控制器共同预测四足机器人和负载的质心（CoM）动力学，随后通过一个全身二次规划（QP）来强制执行地面反作用力约束。该方法通过广泛的仿真和现实世界实验进行了评估，实验中考虑了物体惯性参数的变化。其性能与固定方向和直线轨迹作为基线进行了比较。结果表明，与固定方向基线相比，优化的物体运动减少了约50%的滑动，与直线基线相比减少了30%，同时也实现了更低的机器人质心跟踪误差。
cs.RO / 20 / 2610.07275
Fluorescence-enhanced Whisker Array with Vision-based Deformation Analysis for Underwater Source Localization
基于视觉变形分析的荧光增强触须阵列用于水下源定位
Abstract
Deep-water biological observation is essential for understanding marine organisms and their interactions with the environment. However, conventional optical and acoustic approaches can introduce stimuli that alter animal behavior and bias biological observations. This paper proposes a fluorescence-enhanced whisker array sensing system that pinpoints underwater hydrodynamic sources through local optical readout rather than direct source imaging. Five spatially oriented whiskers, fabricated with nitinol cores and fluorescent urethane shells, are integrated with ultraviolet excitation and a monocular camera. Image enhancement and segmentation are applied to track the whisker deformation. A lightweight convolutional neural network captures temporal and cross-whisker features from 2 s sequences to estimate source localization. Pool experiments achieve a mean spatial localization error of 88 mm, with 73.5 mm in radius and $2.5^\circ$ in angle, across a test region of 600 mm with $\pm30^\circ$. Real-time localization of a moving thruster demonstrates the capability of the proposed method in dynamic scenarios, highlighting its potential for integration into underwater robots for hydrodynamic source detection, localization, and tracking in low-light environments.
Chinese Translation
深水生物观察对于理解海洋生物及其与环境的相互作用至关重要。然而，传统的光学和声学方法可能引入刺激，从而改变动物行为并偏倚生物观察。本文提出了一种荧光增强触须阵列传感系统，通过局部光学读出而非直接源成像来精确定位水下水动力源。该系统集成了五个空间定向的触须，采用镍钛合金核心和荧光聚氨酯外壳，并结合紫外线激发和单目相机。应用图像增强和分割技术跟踪触须变形。轻量级卷积神经网络从2秒序列中捕获时间和跨触须特征，以估计源定位。在600毫米的测试区域内，池实验实现了平均空间定位误差为88毫米，半径为73.5毫米，角度为$2.5^ heta$。实时定位移动推进器展示了该方法在动态场景中的能力，突显了其在低光环境下集成到水下机器人中进行水动力源检测、定位和跟踪的潜力。
cs.RO / 21 / 2610.07277
Distribution-Transfer Safe-Horizon MPC under Mode Uncertainty
模式不确定性下的分布-转移安全视界模型预测控制
Abstract
Scenario-based MPC is an attractive strategy for chance-constrained motion planning that approximates uncertainty via a finite set of sampled scenarios. As a sampling-based method, scenario-based MPC is sensitive to distribution mismatch. We address this problem in the context of Safe-Horizon Model Predictive Control (SH-MPC) with obstacles governed by switching dynamic modes. From finite mode observations, we construct a confidence set for the unknown categorical mode law and derive a multiplicative domination bound that transfers a Safe-Horizon collision-risk certificate from a selected scenario-sampling distribution to every law in the confidence set. Wasserstein geometry is used to regularize probability reallocation among modes according to the similarity of their induced trajectory predictions, while a collision-risk surrogate biases sampling toward dangerous modes. The resulting certificate explicitly quantifies the additional tightening required under distribution mismatch and exposes the multiplicative conservatism that arises when several obstacle-wise transfer factors are combined
Chinese Translation
基于情景的模型预测控制（MPC）是一种吸引人的机会约束运动规划策略，通过有限的采样情景集来近似不确定性。作为一种基于采样的方法，基于情景的MPC对分布不匹配非常敏感。我们在安全视界模型预测控制（Safe-Horizon Model Predictive Control, SH-MPC）的背景下解决了这个问题，该控制方法涉及由切换动态模式主导的障碍物。通过有限的模式观察，我们为未知的分类模式法则构建了一个置信集，并推导出一个乘法主导界限，该界限将从选定的情景采样分布转移到置信集中的每个法则的安全视界碰撞风险证书。Wasserstein几何用于根据模式引起的轨迹预测的相似性来规范概率在模式之间的重新分配，而碰撞风险替代物则使采样偏向于危险模式。最终的证书明确量化了在分布不匹配下所需的额外收紧，并揭示了当多个障碍物转移因子结合时所产生的乘法保守性。
cs.RO / 22 / 2610.07278
Embedded Bare-Metal Radar-Inertial Odometry
嵌入式裸金属雷达惯性里程计
Abstract
Compact extraplanetary rovers and micro aerial vehicles require robust state estimation frameworks designed to operate under strict computational constraints in unforgiving environments. Typical solutions involving vision- or LiDAR-based sensing are computationally expensive and vulnerable to environments with perceptual degradation, making them poorly suited for resource-constrained platforms and austere conditions. Alternatively, Frequency Modulated Continuous Wave (FMCW) radar offers both robustness and computational efficiency by directly providing velocity measurements coupled with inherent resilience to perceptual degradation. These factors enable robust estimation, thereby reducing reliance on human operators for supervision to ensure platform safety in challenging environments. In this manuscript, we propose an embedded radar-inertial estimator tailored for low-compute platforms. All sensor drivers, data processing, and aided inertial navigation are performed on a single-core microcontroller, demonstrating its computational efficiency. Flight experiments show translational APE of $0.51-0.82\,\mathrm{m}$ and RPE of $<$3\,\% against motion capture, alongside closed-loop flight through an unmodified PX4 stack. The firmware and printed circuit board design are openly available on GitHub at \href{https://github.com/ntnu-arl/embedded_rio}{ntnu-arl/embedded\_rio} and \href{https://github.com/ntnu-arl/embedded_rio-pcb}{ntnu-arl/embedded\_rio-pcb} respectively.
Chinese Translation
紧凑型星际探测车和微型无人机需要在严苛环境下，设计出能够在严格计算限制下运行的稳健状态估计框架。典型的基于视觉或激光雷达（LiDAR）的解决方案计算成本高，并且在感知退化的环境中容易受到影响，因此不适合资源受限的平台和恶劣条件。相对而言，频率调制连续波（FMCW）雷达通过直接提供速度测量，并具备对感知退化的固有抗性，提供了稳健性和计算效率。这些因素使得稳健的估计成为可能，从而减少对人类操作员的监督依赖，以确保平台在挑战性环境中的安全。在本文中，我们提出了一种针对低计算平台的嵌入式雷达惯性估计器。所有传感器驱动程序、数据处理和辅助惯性导航均在单核微控制器上执行，展示了其计算效率。飞行实验显示，与运动捕捉相比，平移绝对定位误差（APE）为$0.51-0.82 \, ext{m}$，相对定位误差（RPE）小于3\, ext{ extperthousand}，并通过未修改的PX4堆栈实现了闭环飞行。固件和印刷电路板设计已在GitHub上公开，分别可在 exttt{https://github.com/ntnu-arl/embedded_rio} 和 exttt{https://github.com/ntnu-arl/embedded_rio-pcb} 获取。
cs.RO / 23 / 2610.07319
Lego-Like Stiffness Configuration of Planar Compliant Modules for Task-Specific Flexible Interfaces
类乐高的平面柔性模块刚度配置用于特定任务的灵活接口
Abstract
Compliant mechanisms provide compact and intrinsic structural compliance for regulating physical interactions between mechanisms and environments. However, different tasks demand distinct stiffness characteristics, often requiring task-specific optimization and redesign due to limited geometric design space and inherent coupling among multiple stiffness components. This paper presents a Lego-like stiffness configuration approach using stackable planar compliant modules. Three complementary module geometries are introduced, with their stiffness characteristics further regulated through beam width, plate thickness, and module orientation. A unified stiffness model is established for quantitative analysis of individual and composed modules. Further, a two-stage optimization method is presented to achieve desired stiffness profiles, combining a genetic algorithm for configuration and sequential quadratic programming for parameter refinement. Experimental verification shows deviations below 6.5% for simulated stiffness. A flexible wrist is further developed as a representative implementation, exhibiting distinct compliant and dynamic responses under different stiffness characteristics. An optimized modular composition realizes prescribed stiffness values and maintains compliant obstacle interaction during high-speed motion at 1 m/s, with a maximum tested angular compliance of approximately $15^\circ$. The proposed framework provides a systematic approach for constructing flexible interfaces with task-specific stiffness characteristics.
Chinese Translation
柔性机构为调节机制与环境之间的物理交互提供了紧凑且内在的结构柔性。然而，不同的任务要求不同的刚度特性，通常需要针对特定任务进行优化和重新设计，因为几何设计空间有限且多个刚度组件之间存在固有耦合。本文提出了一种使用可堆叠平面柔性模块的类乐高刚度配置方法。介绍了三种互补的模块几何形状，通过梁宽、板厚和模块方向进一步调节其刚度特性。建立了一个统一的刚度模型，以便对单个模块和组合模块进行定量分析。此外，提出了一种两阶段优化方法，以实现所需的刚度特征，结合遗传算法进行配置和顺序二次规划进行参数优化。实验验证表明，模拟刚度的偏差低于6.5%。进一步开发了一个灵活的手腕作为代表性实现，在不同刚度特性下表现出明显的柔性和动态响应。优化的模块组合实现了规定的刚度值，并在1 m/s的高速运动中保持柔性障碍交互，最大测试角度柔性约为$15^ heta$。所提出的框架为构建具有特定任务刚度特性的灵活接口提供了系统的方法。
cs.RO / 24 / 2610.07320
Towards Decentralized Formation of Minimum-Length Communication Networks Using Robot Swarms
基于机器人群体的最小长度通信网络去中心化构建研究
Abstract
Multi-robot missions in infrastructure-denied environments frequently rely on reliable communication links between spatially separated locations. We propose a fully decentralized framework for constructing and dynamically maintaining communication networks without centralized topology planning or global positioning infrastructure. Driven strictly by local interactions, robots reconfigure local network topologies and adjust their physical positions to minimize overall network length while adhering to communication constraints. Formal analysis shows that our local reconfiguration operations guarantee continuous network connectivity, strictly decrease network length with every branch transfer, and bound worst-case performance to the shortest starlike tree. Embodied simulations and physical multi-robot experiments confirm that our approach forms networks near the length of centrally computed Euclidean Steiner trees. Additionally, the system dynamically adapts to moving targets and optimizes deployment by utilizing only necessary connectors, preserving excess robots for auxiliary tasks. This work enables autonomous swarms to self-organize adaptive ad hoc communication infrastructure in communication-denied environments, which could support applications varying from subterranean exploration to planetary missions.
Chinese Translation
在基础设施缺失的环境中，多机器人任务通常依赖于空间分隔位置之间可靠的通信链路。我们提出了一种完全去中心化的框架，用于构建和动态维护通信网络，无需集中式拓扑规划或全球定位基础设施。机器人仅通过局部交互来驱动，重新配置局部网络拓扑，并调整其物理位置，以最小化整体网络长度，同时遵循通信约束。形式分析表明，我们的局部重配置操作保证了网络的连续连接性，在每次分支转移中严格减少网络长度，并将最坏情况性能限制在最短的星状树。具体的仿真和物理多机器人实验确认了我们的方法形成的网络接近于集中计算的欧几里得斯坦纳树的长度。此外，该系统能够动态适应移动目标，并通过仅使用必要的连接器来优化部署，保留多余的机器人用于辅助任务。这项工作使自主群体能够在通信受限的环境中自组织适应性临时通信基础设施，支持从地下探索到行星任务等多种应用。
cs.RO / 25 / 2610.07327
SharedKV-BT: Node-Local Typed Decisions for Behavior-Tree Agents
SharedKV-BT：行为树代理的节点本地类型决策
Abstract
Agent tasks require sequences of interdependent decisions. Autoregressive models support more flexible decision interfaces than conventional classifiers but incur the latency of token-by-token generation. Recent shared-prefix methods reduce this cost by reusing encoded context and scoring multiple decisions in parallel, but do not model decision dependencies or verify execution. We propose SharedKV-BT, where each active node of a behavior tree (BT) exposes stage-local fields and candidates, and Shared-KV scores the candidates in parallel and passes the selected decision to a separate execution system. We tested SharedKV-BT on robot manipulation, mobile navigation, and computer-use tasks. Across three tasks, SharedKV-BT made typed decisions 2.36-4.15 times faster than prompt-matched autoregressive decoding. On the manipulation task, node-local Shared-KV improved joint decision accuracy from 75% to 94% and closed-loop success from 0% to 60%. Fixed-score policy replay showed that stage gating prevented out-of-order actions and external postconditions prevented premature completion.
Chinese Translation
代理任务需要一系列相互依赖的决策。自回归模型支持比传统分类器更灵活的决策接口，但会带来逐个生成的延迟。近期的共享前缀方法通过重用编码上下文并平行评分多个决策来降低这一成本，但未能建模决策依赖关系或验证执行。我们提出了SharedKV-BT，其中行为树（BT）的每个活动节点暴露阶段本地字段和候选项，Shared-KV平行评分候选项并将选定的决策传递给一个独立的执行系统。我们在机器人操作、移动导航和计算机使用任务上测试了SharedKV-BT。在这三项任务中，SharedKV-BT的类型决策速度比提示匹配的自回归解码快2.36-4.15倍。在操作任务中，节点本地的Shared-KV将联合决策准确率从75%提高到94%，闭环成功率从0%提高到60%。固定评分策略重放显示，阶段门控防止了无序操作，外部后置条件防止了过早完成。
cs.RO / 26 / 2610.07346
ScanSTL: Parallel Robustness Evaluation for Signal Temporal Logic
ScanSTL：信号时序逻辑的并行鲁棒性评估
Abstract
Repeated evaluation and differentiation of Signal Temporal Logic (STL) robustness can become a computational bottleneck in robot planning and control. Sequential temporal recurrences limit parallelism, while dense masking increases memory requirements. We propose ScanSTL, which combines associative temporal aggregation with parallel scans and ordered block reductions. Eventually and Always use range extrema, while inclusive strong Until composes compact segment representations. A common range engine handles bounded and shifted intervals, including the guards required before delayed Until witnesses. Each exact temporal operator computes complete robustness traces with linear work and storage and logarithmic parallel depth on uniformly sampled finite signals. An open source JAX implementation supports automatic differentiation, batching, and compilation. We compare ScanSTL with STLCG and STLCG++ using CPU and GPU operator benchmarks and nine composed specifications. Across these nine specifications at 512 samples, ScanSTL achieves geometric mean speedups of 243 times for forward evaluation and 104 times for gradient computation over STLCG++ in JAX on the CPU. On an RTX~5090 GPU, ScanSTL evaluates unbounded Until over more than two million samples with median times below 0.1 ms for forward evaluation and 0.25 ms for gradient computation. Simulated escort and patrol experiments with a robot dog further demonstrate faster repair of violating plans and greater solver capacity in model predictive control.
Chinese Translation
信号时序逻辑（STL）鲁棒性的重复评估和微分可能成为机器人规划和控制中的计算瓶颈。顺序时间递归限制了并行性，而密集掩码增加了内存需求。我们提出了ScanSTL，它结合了关联时间聚合、并行扫描和有序块归约。最终和始终使用范围极值，而包含强直到（inclusive strong Until）则组合紧凑的段表示。一个通用范围引擎处理有界和偏移区间，包括在延迟的直到（Until）见证之前所需的守卫。每个精确的时间算子以线性工作和存储以及在均匀采样的有限信号上以对数并行深度计算完整的鲁棒性轨迹。一个开源的JAX实现支持自动微分、批处理和编译。我们使用CPU和GPU算子基准测试以及九个组合规范对ScanSTL与STLCG和STLCG++进行了比较。在这九个规范中，在512个样本下，ScanSTL在CPU上的前向评估和梯度计算相对于STLCG++实现了243倍和104倍的几何平均加速。在RTX~5090 GPU上，ScanSTL在超过两百万个样本上评估无界的直到（Until），前向评估的中位时间低于0.1毫秒，梯度计算的中位时间为0.25毫秒。与机器人狗的模拟护送和巡逻实验进一步展示了更快的违规计划修复和更大的模型预测控制求解器能力。
cs.RO / 27 / 2610.07371
Expressiveness, Equivalence, and Uncertainty in Velocity Obstacles and Closest Point of Approach Metrics
速度障碍与最近接点接近度度量中的表达能力、等价性与不确定性
Abstract
Time to Closest Point of Approach (TCPA), Distance to Closest Point of Approach (DCPA), and Velocity Obstacles (VOs), are widely used to assess and mitigate collision risk in autonomous navigation, yet their relationship and behavior under uncertainty remain largely unexplored. Assuming perfect state information, we establish a relationship between these representations over finite and infinite prediction horizons and derive conditions under which they provide equivalent characterizations of collision risk. Under bounded uncertainty, we extend the Closest Point of Approach (CPA) metrics and VO to convex relative-state sets. We show that in this setting, independently computed TCPA and DCPA bounds lose the joint relationship required for VO membership, while uncertainty-aware VOs preserve this relationship through a set-valued representation of collision-inducing velocities.
Chinese Translation
最近接点接近度时间（TCPA）、最近接点接近度距离（DCPA）和速度障碍（VOs）广泛用于评估和降低自主导航中的碰撞风险，但它们在不确定性下的关系和行为仍然很大程度上未被探索。在假设完美状态信息的情况下，我们建立了这些表示在有限和无限预测视野下的关系，并推导出它们提供碰撞风险等价表征的条件。在有界不确定性下，我们将最近接点接近度（CPA）度量和速度障碍扩展到凸相对状态集。我们表明，在这种情况下，独立计算的TCPA和DCPA界限失去了VO成员资格所需的联合关系，而考虑不确定性的VO通过对碰撞诱导速度的集合值表示保留了这种关系。
cs.RO / 28 / 2610.07382
Toward Trustworthy Physical AI for Human Interaction
迈向可信赖的物理人工智能以促进人机互动
Abstract
Robots are entering human spaces faster than we can establish when they deserve trust. We propose a framework for trustworthy physical AI that integrates Safety, Behavioral Intelligibility, and Perceptual Alignment across embodiment, control, cognition, and design. Trustworthiness emerges from aligning physical capabilities, observable behavior, and expectations people form during interaction.
Chinese Translation
机器人进入人类空间的速度超过了我们建立其可信赖性的能力。我们提出了一个可信赖的物理人工智能框架，该框架整合了安全性、行为可理解性和感知一致性，涵盖了体现、控制、认知和设计等方面。可信赖性源于物理能力、可观察行为以及人们在互动过程中形成的期望之间的对齐。
cs.RO / 29 / 2610.07390
AeroBuoy: A Drone Deployable, 3D Printed, Autonomous Robotic Buoy for Environmental Inspection in Remote and Hazardous River Systems
AeroBuoy：一种可由无人机部署的、3D 打印的、用于偏远和危险河流系统环境检查的自主机器人浮标
Abstract
Monitoring of waterways such as remote and hazardous rivers and streams is important so as to assess the impact of external factors including construction runoff or climate change. Versatile, autonomous robotic boats can offer excellent environmental inspection and monitoring solutions for remote, dangerous, or access protected water bodies but they have several shortcomings in terms of maneuverability. This paper proposes an environmental inspection system consisting of an autonomous data collection buoy which is designed to be deployed to inaccessible river systems using a drone. The system can perform a drop off and pickup of the buoy depending on the requirements of a particular location and monitoring task. Utilising the natural flow of the river the buoy autonomously steers down, using GPS and magnetometers so as to maintain the desired trajectory. The buoy is capable of measuring water temperature but it can also be equipped with a range of sensors such as water oxygen meter, sonar for river bed inspection, or turbidity for water clarity. This paper describes the system design, presents an analysis of the self-righting capabilities of the buoy, and shows a full system demonstration at the \=Orewa River in Auckland, New Zealand.
Chinese Translation
对偏远和危险的河流及溪流等水道的监测至关重要，以评估包括建筑径流或气候变化在内的外部因素的影响。多功能的自主机器人船只能够为偏远、危险或受保护的水体提供出色的环境检查和监测解决方案，但在机动性方面存在若干不足。本文提出了一种环境检查系统，该系统由一个自主数据收集浮标组成，旨在通过无人机部署到无法到达的河流系统。该系统可以根据特定位置和监测任务的要求进行浮标的投放和回收。浮标利用河流的自然流动自主导航，使用 GPS 和磁力计以保持所需的轨迹。浮标能够测量水温，但也可以配备多种传感器，如水氧计、用于河床检查的声纳或用于水质清晰度的浊度传感器。本文描述了系统设计，分析了浮标的自立能力，并展示了在新西兰奥克兰的 Orewa 河进行的完整系统演示。
cs.RO / 30 / 2610.07396
What the Elevation Map Cannot See: Semantic-Aware Locomotion and Execution-Aware Navigation for Humanoid Robot
高程图无法看到的：面向语义的运动和面向执行的导航在类人机器人中的应用
Abstract
Navigation for humanoid robots is critical, yet large-scale evaluation on physical hardware is often impractical due to cost and safety concerns, making simulation benchmarks essential. Existing VLN benchmarks achieve physically executable navigation, but still assume (1) all hazards are observable from elevation maps; (2) realized motions closely match desired motions. In real environments, however, fallen bottles may be ambiguous in elevation maps, while phones and water spills may be difficult to differentiate; hazard avoidance by the locomotion policy can cause the robot's actual trajectory to deviate from the path intended by the VLN policy. Such command-execution mismatch can accumulate and lead the robot toward unintended locations. To expose these failure modes, we introduce a benchmark that models both elevation-subtle hazards and execution deviations, together with a closed-loop VLN + locomotion control framework that continuously realigns high-level navigation with the robot's actual state. We evaluate navigation in simulation and further validate the locomotion policy on a physical Unitree G1 humanoid robot. Results show that semantic input reduces contact with hazards poorly represented in elevation maps, while anti-deviation improves navigation success. These findings highlight the need to evaluate humanoid navigation jointly in terms of route completion and hazard avoidance.
Chinese Translation
类人机器人的导航至关重要，但由于成本和安全问题，在物理硬件上进行大规模评估往往不切实际，因此模拟基准测试显得尤为重要。现有的视觉导航（VLN）基准测试实现了可物理执行的导航，但仍然假设（1）所有危险都可以从高程图中观察到；（2）实际运动与期望运动密切匹配。然而，在真实环境中，倒下的瓶子在高程图中可能模糊不清，而手机和水洒可能难以区分；运动策略的危险规避可能导致机器人实际轨迹偏离VLN策略所意图的路径。这种命令执行不匹配可能会累积并使机器人朝向意外位置移动。为了揭示这些失败模式，我们引入了一个基准测试，模型同时考虑高程图中微妙的危险和执行偏差，并结合一个闭环的VLN + 运动控制框架，持续将高层导航与机器人的实际状态对齐。我们在模拟中评估导航，并进一步在物理Unitree G1类人机器人上验证运动策略。结果表明，语义输入减少了与高程图中表现不佳的危险的接触，而反偏差策略提高了导航成功率。这些发现强调了在路线完成和危险规避方面共同评估类人导航的必要性。
cs.RO / 31 / 2610.07398
An Autonomous, 3D Printed, Waterjet-Powered, Open-Source Robotic Trimaran for Environmental Inspection and Monitoring
一种自主的、3D打印的、水射流驱动的开源机器人三体船用于环境检查与监测
Abstract
Versatile, autonomous robotic boats can offer excellent environmental inspection and monitoring solutions for remote, dangerous, hard to reach, or access protected water bodies. This paper introduces such a platform in the form of an autonomous, cost-effective, waterjet-powered robotic trimaran. Motivated by the need for an efficient aquatic monitoring, particularly in Aotearoa - New Zealand's diverse environments, the trimaran provides an efficient, low-cost, and easy to replicate alternative to resource-intensive research vessels. The proposed platform, costs $600-1,500 USD to develop (depending on the sensing system configuration), weighs under 5 kg, and excels in bathymetry and water quality testing. The trimaran can reach speeds of up to 2 m/s offering obstacle avoidance of natural features, such as rocks. Utilizing off-the-shelf components and 3D printing technology, the proposed platform offers excellent reproducibility and robustness while operating in shallow waters with its jet propulsion system. The paper presents in detail the design characteristics, the sensing system employed, testing results focusing on bathymetry, and highlights the ability of vessel and the potential for future research and data collection.
Chinese Translation
多功能的自主机器人船可以为偏远、危险、难以到达或受保护的水域提供卓越的环境检查和监测解决方案。本文介绍了一种这样的平台，即一种自主的、经济高效的水射流驱动机器人三体船。该三体船的设计受到对高效水域监测需求的驱动，特别是在新西兰多样化环境中的应用，提供了一种高效、低成本且易于复制的替代方案，取代资源密集型的研究船。所提议的平台开发成本为600-1500美元（具体取决于传感系统配置），重量低于5千克，并在水深测量和水质测试方面表现出色。该三体船的速度可达2米/秒，能够避开自然障碍物，如岩石。利用现成组件和3D打印技术，所提议的平台在浅水区操作时提供了卓越的可重复性和稳健性。本文详细介绍了设计特征、所采用的传感系统、以水深测量为重点的测试结果，并强调了该船的能力及其未来研究和数据收集的潜力。
cs.RO / 32 / 2610.07409
RACER: Residual-Adaptive Closed-Loop Estimation for Sampling-Based Planning in Wheeled-Quadruped Racing
RACER：用于轮式四足动物竞赛的残差自适应闭环估计的采样基础规划
Abstract
We present RACER, a hierarchical control framework for wheel-based quadruped racing that combines an MPPI planner with a learned residual dynamics model and a low-level RL velocity tracker. The planner augments a nominal unicycle kinematic model with a neural residual term to capture the closed-loop tracking behavior of the RL policy. To train this residual model under limited real-world data, we propose Low-Rank Residual Adaptation (LoRRA), a two-stage approach that pre-trains on large-scale simulation data for broad coverage and then fine-tunes on a small real-world dataset with a low-rank constraint. In simulation, we empirically validate our engineering choices by showing (A) Residual dynamics improve the overall performance of our pipeline by capturing the tracking error of RL velocity tracker at high-speed cornering. (B) Residual dynamics trained with both source-domain and target-domain data gives racing performance significantly better than the residual dynamics trained with only target-domain data. (C) Low-rank constraint at target-domain adaptation gives higher success rates and higher performance than full-tune and from-scratch when domain gap in ground coefficient or joint gain increases.
Chinese Translation
我们提出了RACER，一个用于轮式四足动物竞赛的分层控制框架，该框架结合了MPPI规划器、学习的残差动力学模型和低级强化学习（RL）速度跟踪器。该规划器通过神经残差项增强了名义上的单轮车运动学模型，以捕捉RL策略的闭环跟踪行为。为了在有限的真实世界数据下训练这个残差模型，我们提出了低秩残差自适应（LoRRA），这是一种两阶段的方法，首先在大规模仿真数据上进行预训练以实现广泛覆盖，然后在具有低秩约束的小规模真实世界数据集上进行微调。在仿真中，我们通过以下方式实证验证了我们的工程选择：（A）残差动力学通过捕捉RL速度跟踪器在高速转弯时的跟踪误差，提高了我们管道的整体性能。（B）使用源域和目标域数据训练的残差动力学，其竞赛性能显著优于仅使用目标域数据训练的残差动力学。（C）在目标域适应中施加低秩约束时，当地面系数或关节增益的领域差距增加时，成功率和性能均高于完全调优和从头开始训练。
cs.RO / 33 / 2610.07432
Entropy-Gated Belief Coordination for Decentralized Multi-Agent Search Under Intermittent Communication
基于熵门控的信念协调在间歇通信下的去中心化多智能体搜索
Abstract
We study decentralized multi-agent target search where homogeneous agents communicate intermittently at Poisson-distributed times. Standard unconditional belief fusion wastes communication opportunities by synchronizing agents during high-entropy exploration, when diverse independent beliefs provide better coverage than a premature consensus. We introduce \emph{entropy-gated belief coordination}, in which agents skip fusion while their collective entropy ratio exceeds a threshold~\(\theta\) and merge only during exploitation, consistent with the bifurcation structure of nonlinear opinion dynamics and the submodular structure of the per-step information gain objective. We further derive \(\Istep(\alpha,\beta)\), the expected mutual information per observation step between two agents' binary sensors, as an interpretable, communication-free measure of sensor informativeness that motivates the gating design and guides system-level analysis. Experiments across 103{,}680 trials (nine grid sizes up to \(100{\times}100\), Poisson communication timing, four target movement patterns) show that the Entropy-Gated Trust-Decay Planner (\textsc{EG-TDP}), which adds a detection-probability planner switch in exploitation mode, achieves mean belief quality \(\bar{Q}=0.300\) (mean belief mass at the true target cell, averaged across all trials and steps), a \(58.9\%\) gain over arithmetic mean and a \(37.4\%\) gain over visit-weighted fusion. On representative configurations, EG-TDP also outperforms a joint-Bayesian reference that uses all agents'~observations at every step, despite operating under random intermittent contact only.
Chinese Translation
我们研究了去中心化的多智能体目标搜索，其中同质智能体在泊松分布的时间间隔内进行间歇性通信。标准的无条件信念融合在高熵探索期间浪费了通信机会，此时多样化的独立信念提供的覆盖效果优于过早达成共识。我们引入了 extit{熵门控信念协调}，在此框架下，当智能体的集体熵比率超过阈值 heta时，智能体跳过信念融合，仅在利用阶段进行融合，这与非线性意见动态的分叉结构以及每步信息增益目标的子模块结构相一致。我们进一步推导了 extit{Istep}( extalpha, extbeta)，这是两个智能体的二元传感器之间每次观察步骤的预期互信息，作为一种可解释的、无通信的传感器信息量度，激励了门控设计并指导系统级分析。在103,680次实验中（九种网格大小，最大为100×100，泊松通信时机，四种目标运动模式），熵门控信任衰减规划器（ extsc{EG-TDP}）在利用模式下添加了检测概率规划器切换，实现了平均信念质量ar{Q}=0.300（在所有实验和步骤中，真实目标单元的平均信念质量），比算术平均提高了58.9%，比访问加权融合提高了37.4%。在代表性配置中，EG-TDP的表现也优于一个联合贝叶斯参考，该参考在每一步都使用所有智能体的观察，尽管仅在随机间歇接触下操作。
cs.RO / 34 / 2610.07456
Dynamics Modeling of a Multi-UAV Slung Load System Using a Discrete-Link Cable Approach
基于离散链缆方法的多无人机吊载系统动力学建模
Abstract
A common assumption to simplify the problem of controlling a multi-UAV slung load system (MUSLS) is that the flexible cables can be modeled as massless rigid rods. In this work, we propose an alternative Euler-Newton derived dynamical model which uses a series of rigid links to model the flexible cables. The model is specifically designed to allow efficient simulation using Featherstone's articulated body algorithm. We perform real-world validation of this model on gentle, aggressive, and tension-engagement maneuvers and run a parameter sweep to determine the number of links, joint damping, and joint friction to achieve the greatest model fidelity. The model closely matches real-world flight data with mean load translation errors below 132 mm (5.5% of the cable length) and orientation errors below 11.4 degrees. We make the real-world flight data publicly available for the development of future cable models.
Chinese Translation
为了简化多无人机吊载系统（MUSLS）控制问题，通常假设柔性缆绳可以被建模为无质量的刚性杆。在本研究中，我们提出了一种替代的基于欧拉-牛顿推导的动力学模型，该模型使用一系列刚性链节来模拟柔性缆绳。该模型专门设计用于通过Featherstone的关节体算法进行高效仿真。我们在温和、激进和张力接触的飞行动作中对该模型进行了实际验证，并进行了参数扫描，以确定链节数量、关节阻尼和关节摩擦，以实现最大的模型保真度。该模型与实际飞行数据高度吻合，平均载荷位移误差低于132毫米（占缆绳长度的5.5%），方向误差低于11.4度。我们将实际飞行数据公开，以促进未来缆绳模型的开发。
cs.RO / 35 / 2610.07472
SURGE: Sonar-fUsed Reconstruction and localization via image-gated Graph Estimation
SURGE：通过图像引导的图估计进行声纳驱动的重建与定位
Abstract
Remotely operated vehicles (ROVs) are widely used to explore and inspect underwater environments such as caves, shipwrecks, and submerged infrastructure. These missions require accurate 3D understanding of the surrounding environment, which depends on both reliable vehicle localization and metric scene reconstruction. However, external positioning is often unavailable underwater, requiring small ROVs to rely primarily on onboard perception. Optic vision provides rich visual and geometric information but suffers from scale ambi- guity and trajectory drift, whereas 2D imaging sonar provides metric range but incomplete 3D geometry. Existing underwater reconstruction approaches typically address these limitations separately or assume known sensor poses, leaving localization and reconstruction disconnected. We present SURGE, a camera sonar framework that jointly estimates the ROV trajectory and target location by integrating visual and acoustic observations within a factor graph, then uses the recovered metric poses for sonar Gaussian splatting. Experiments on real underwater RGB sonar observations show that SURGE substantially improves localization consistency over conventional vision based pose estimation and produces a more compact, natively metric reconstruction than RGB Gaussian splatting baselines.
Chinese Translation
遥控水下机器人（ROVs）广泛用于探索和检查水下环境，如洞穴、沉船和水下基础设施。这些任务需要对周围环境进行准确的三维理解，这依赖于可靠的机器人定位和度量场景重建。然而，在水下通常无法获得外部定位，这使得小型ROV主要依赖于机载感知。光学视觉提供丰富的视觉和几何信息，但存在尺度歧义和轨迹漂移的问题，而二维成像声纳提供度量范围但缺乏完整的三维几何信息。现有的水下重建方法通常分别解决这些局限性或假设已知传感器姿态，从而使定位和重建脱节。我们提出了SURGE，一种相机声纳框架，通过在因子图中整合视觉和声学观测，联合估计ROV轨迹和目标位置，然后利用恢复的度量姿态进行声纳高斯点云处理。在真实的水下RGB声纳观测实验中，SURGE显著提高了相较于传统视觉基础姿态估计的定位一致性，并生成了比RGB高斯点云基线更紧凑、原生度量的重建。
cs.RO / 36 / 2610.07474
Risk-Sensitive Crowd Navigation with Adaptive Ellipsoidal Conformal Prediction
风险敏感的群体导航与自适应椭球体保守预测
Abstract
Safe crowd navigation under distribution shift requires uncertainty representations that capture structured human-motion prediction errors and safety objectives that account for rare but consequential failures. Existing uncertainty-aware methods typically represent prediction errors using isotropic regions, which can be either overly conservative or poorly aligned with directional motion uncertainty. We introduce a risk-aware navigation framework that uses anisotropic conformal ellipsoids to translate structured prediction uncertainty into an episode-level conditional value-at-risk (CVaR) signal that regulates the Lagrangian safety penalty. In particular, adaptive ellipsoidal conformal prediction (AECP) captures directional prediction errors and adaptively calibrates uncertainty regions under distribution shift, while the resulting CVaR-regulated navigation policy is optimized using Lagrangian proximal policy optimization. We evaluate our proposed method under both in-distribution settings and out-of-distribution (OOD) settings involving shifts in pedestrian motion patterns. Compared with state-of-the-art baselines, our method maintains competitive in-distribution performance while improving success rates by 5.68-7.44 percentage points and reducing collision rates by 5.44-6.64 percentage points across OOD settings. We further deploy the trained policy without fine-tuning on a physical robot with onboard perception and CPU-only inference, showing that the full pipeline is feasible in physical crowd navigation.
Chinese Translation
在分布变化下的安全群体导航需要能够捕捉结构化人类运动预测误差的 uncertainty 表示，以及考虑到罕见但后果严重的失败的安全目标。现有的关注不确定性的方法通常使用各向同性区域来表示预测误差，这可能过于保守或与方向性运动不确定性不匹配。我们提出了一种风险意识导航框架，该框架使用各向异性保守椭球体将结构化预测不确定性转化为一个情节级条件风险价值（CVaR）信号，从而调节拉格朗日安全惩罚。特别地，自适应椭球体保守预测（AECP）捕捉方向性预测误差，并在分布变化下自适应地校准不确定性区域，而由此产生的 CVaR 调节导航策略则通过拉格朗日近端策略优化进行优化。我们在分布内设置和涉及行人运动模式变化的分布外（OOD）设置下评估了我们提出的方法。与最先进的基线相比，我们的方法在保持竞争力的分布内性能的同时，在 OOD 设置中提高了成功率 5.68-7.44 个百分点，并减少了 5.44-6.64 个百分点的碰撞率。我们进一步在具有板载感知和仅 CPU 推理的物理机器人上部署了训练好的策略，而无需微调，显示出完整的流程在物理群体导航中是可行的。
cs.RO / 37 / 2610.07511
MobileVISTA: Generative Data Augmentation for Pose Generalization in Mobile Manipulation
MobileVISTA：用于移动操作中的姿态泛化的生成数据增强
Abstract
Mobile manipulators such as humanoid robots are increasingly deployed in dynamic, unstructured environments to perform dexterous manipulation tasks. However, end-to-end manipulation policies trained to imitate demonstration data collected from a single robot pose are brittle: even centimeter-scale deviations in robot pose at deployment can drive ego-centric observations and end-effector trajectories out of the training distribution, leading to sharp drops in performance. We introduce MobileVISTA, a data generation framework that transforms demonstrations captured at canonical poses into diverse, pose-perturbed training data by jointly (1) augmenting egocentric visual observations and (2) retargeting actions to compensate for base pose changes. Unlike prior methods, which assume a camera rigidly mounted off the actuated chain or non-trivial articulated robot geometry largely out of frame, MobileVISTA targets compatibility with egocentric platforms (e.g., humanoids) where the camera is both influenced by and must observe the robot's kinematic chain as it moves. We study MobileVISTA in simulated tasks spanning humanoid and bimanual embodiments, and on a real Galaxea R1 Pro. We find policies trained on MobileVISTA-augmented data demonstrate improved robustness to previously out-of-distribution poses encountered at test time, without additional demonstration collection or a trained generative model. Additionally, we find MobileVISTA's benefit is largest on tested humanoids, where the camera rides the actuated chain and the robot fills much of the frame. Additional videos and appendix can be found on our website: https://mobilevista.github.io
Chinese Translation
移动操作器，例如类人机器人，越来越多地被部署在动态、非结构化的环境中，以执行灵巧的操作任务。然而，端到端的操作策略在模仿从单一机器人姿态收集的演示数据时表现出脆弱性：即使是机器人姿态在部署时的厘米级偏差，也会使自我中心的观察和末端执行器轨迹偏离训练分布，导致性能急剧下降。我们提出了MobileVISTA，一个数据生成框架，通过联合（1）增强自我中心的视觉观察和（2）重新定向动作以补偿基础姿态变化，将在标准姿态下捕获的演示转化为多样化的姿态扰动训练数据。与先前的方法不同，后者假设相机刚性固定在驱动链之外或非平凡的关节机器人几何形状大部分超出框架，MobileVISTA旨在与自我中心平台（例如类人机器人）兼容，其中相机受机器人运动的运动链影响并且必须观察该运动链。我们在模拟任务中研究了MobileVISTA，涵盖类人和双手操作的实现，并在真实的Galaxea R1 Pro上进行测试。我们发现，基于MobileVISTA增强数据训练的策略在测试时对先前超出分布的姿态表现出更强的鲁棒性，而无需额外的演示收集或训练生成模型。此外，我们发现MobileVISTA的优势在测试的类人机器人上最大，因为相机与驱动链相连，机器人占据了大部分画面。更多视频和附录可以在我们的网站上找到：https://mobilevista.github.io
cs.RO / 38 / 2610.07515
ProCut: Probabilistic Cutting Topology for Autonomous Electrosurgical Tissue Dissection
ProCut：用于自主电外科组织解剖的概率切割拓扑
Abstract
Accurately modeling and tracking the deformation of soft tissue is critical for a wide range of interventional and surgical procedures. However, current methods struggle in scenarios involving topological changes, such as cutting and dissection, due to the inherent non-linearity and discontinuity introduced by explicit changes in connectivity. In this work, we present a novel, fully differentiable framework that enables robust estimation and modeling of topological changes during deformable tracking. Our method introduces a continuous, sigmoid-based formulation to smooth the otherwise discrete event of tissue cutting, making it amenable to gradient-based optimization within a differentiable Position-Based Dynamics (PBD) simulation. To account for uncertainty and improve robustness in the presence of noisy visual data, we incorporate Stein Variational Gradient Descent (SVGD) for particle-based probabilistic inference, generating multiple hypotheses for topological state estimation. Building on this foundation, we develop an autonomous dissection algorithm for thin-shell tissues that leverages topological updates to guide closed-loop cutting trajectory control. We evaluate our approach in both simulated and real-world electrosurgical environments, demonstrating significant improvements in topological estimation accuracy and dissection precision over existing methods. Our results highlight the potential of this framework to advance automation in soft-tissue surgical procedures by enabling reliable perception and control in the presence of complex structural changes.
Chinese Translation
准确建模和跟踪软组织的变形对于广泛的介入和外科手术至关重要。然而，当前方法在涉及拓扑变化的场景中（如切割和解剖）表现不佳，因为显式的连通性变化引入了固有的非线性和不连续性。在本研究中，我们提出了一种新颖的、完全可微分的框架，能够在可变形跟踪过程中稳健地估计和建模拓扑变化。我们的方法引入了一种基于连续Sigmoid的公式，以平滑组织切割这一离散事件，使其适合于在可微分的基于位置的动力学（Position-Based Dynamics, PBD）仿真中进行基于梯度的优化。为了考虑不确定性并提高在噪声视觉数据存在下的稳健性，我们结合了斯坦因变分梯度下降（Stein Variational Gradient Descent, SVGD）进行基于粒子的概率推断，生成多个拓扑状态估计的假设。在此基础上，我们开发了一种针对薄壳组织的自主解剖算法，该算法利用拓扑更新来指导闭环切割轨迹控制。我们在模拟和真实的电外科环境中评估了我们的方法，显示出在拓扑估计精度和解剖精确度方面显著优于现有方法。我们的结果突显了该框架在软组织外科手术中推进自动化的潜力，通过在复杂结构变化的情况下实现可靠的感知和控制。
cs.RO / 39 / 2610.07525
ReDex: Repairing Sim-to-Real Dexterous Policies by Finger-Level Compliant Interaction
ReDex：通过指级合规交互修复模拟到真实的灵巧策略
Abstract
Dexterous manipulation policies trained in simulation often fail to transfer to the real world because of errors in contact timing and force regulation. Yet these policies can retain useful multi-finger coordination for task progression. We propose ReDex, a framework for adapting a simulation-trained base policy to the real world by correcting local contact failures and incorporating tactile feedback. Starting from a proprioception-only base policy, ReDex allows a human operator to physically correct contact failures at selected fingers under compliant control during real-world rollouts, while the frozen base policy continues to control the remaining fingers. These rollouts combine base policy execution, human-corrected finger motion, and fingertip force observations. We reconstruct force-informed targets from these rollouts to train a standalone force-conditioned policy via behavior cloning. This design reduces human correction effort, enables learning of contact regulation from real-world interaction, and introduces force feedback into a proprioception-only policy without tactile simulation or complex full-hand teleoperation. We evaluate ReDex on two challenging, contact-rich dexterous manipulation tasks on real hardware. Compared with sim-to-real transferred base policies, ReDex increases Object Flipping success rate from 14\% to 86\% across two objects and average Screwdriver Rotation progress from 26.0% to 95.3% across three objects.
Chinese Translation
在模拟中训练的灵巧操作策略常常无法成功转移到现实世界，原因在于接触时机和力调节的错误。然而，这些策略能够保持多指协调以推进任务。我们提出了ReDex，一个通过纠正局部接触失败并结合触觉反馈来将模拟训练的基础策略适应于现实世界的框架。从仅依赖本体感知的基础策略开始，ReDex允许人类操作员在现实世界的执行过程中，在合规控制下对选定手指的接触失败进行物理修正，同时冻结的基础策略继续控制其余手指。这些执行过程结合了基础策略的执行、人类修正的手指运动和指尖力的观察。我们从这些执行过程中重构出基于力的信息目标，通过行为克隆训练一个独立的基于力的策略。该设计减少了人类的修正工作量，使得从现实世界交互中学习接触调节成为可能，并在没有触觉模拟或复杂全手远程操作的情况下，将力反馈引入仅依赖本体感知的策略。我们在真实硬件上对ReDex进行了评估，测试了两个具有挑战性的、接触丰富的灵巧操作任务。与模拟到真实转移的基础策略相比，ReDex将物体翻转成功率从14%提高到86%，并将平均螺丝刀旋转进展从26.0%提高到95.3%。
cs.RO / 40 / 2610.07527
Task-Space Imitation Guidance for Efficient Reinforcement Learning
高效强化学习的任务空间模仿指导
Abstract
We introduce Task-Space Imitation Guidance for Efficient Reinforcement Learning (TIGER), a reward-construction and pretraining framework for sparse-reward tabletop robotic manipulation. TIGER treats an action-chunked imitation policy not as an executable controller or action prior, but as a local task-space progress estimator: predicted action chunks are converted, using controller-aware action-to-motion mapping, into short-horizon end-effector references, and the RL agent receives dense progress rewards toward these references while the sparse environment reward remains the dominant objective. During pretraining, TIGER uses imitation-guided look-ahead signals to relax conservative value penalties for actions predicted to make task-space progress, reducing off-manifold exploration during early online RL. Across simulation and real-robot experiments, TIGER improves early sample efficiency and reduces measured safety violations while matching or improving final success rates relative to prior RL and IL-RL baselines on the evaluated tasks.
Chinese Translation
我们提出了高效强化学习的任务空间模仿指导（Task-Space Imitation Guidance for Efficient Reinforcement Learning，TIGER），这是一个用于稀疏奖励桌面机器人操控的奖励构建和预训练框架。TIGER将动作分块模仿策略视为局部任务空间进展估计器，而非可执行控制器或动作先验：预测的动作块通过控制器感知的动作到运动映射转换为短期末端执行器参考，强化学习（RL）代理在这些参考上获得密集的进展奖励，同时稀疏环境奖励仍然是主要目标。在预训练过程中，TIGER利用模仿指导的前瞻信号来放宽对预测能在任务空间中取得进展的动作的保守价值惩罚，从而减少早期在线强化学习中的离散探索。在模拟和真实机器人实验中，TIGER提高了早期样本效率，并减少了测量的安全违规，同时在评估任务上与之前的强化学习和模仿强化学习（IL-RL）基线相比，保持或提高了最终成功率。
cs.RO / 41 / 2610.07558
Seeing the Invisible: Physics-Guided Visual Prompting for Temperature- and Radiation-Aware VLA Navigation
看见隐形的事物：基于物理引导的视觉提示在温度和辐射感知的视觉-语言-行动导航中的应用
Abstract
Vision-Language-Action (VLA) models have become a major paradigm for Vision-and-Language Navigation (VLN). However, in safety-critical facilities, invisible risks such as radiation or temperature spikes cannot be detected by an RGB camera, and handling each risk is expensive, requiring a new encoder, new data, and model retraining. We propose Physics-Guided Visual Prompting (PG-VP), a plug-and-play multimodal perception module that instead reuses what a frozen VLA model already does well: avoiding visible obstacles. Given a proximal radiation or thermal source, PG-VP performs a physics-guided risk assessment to determine the avoidance direction and overlays a corresponding virtual obstacle that moves across consecutive frames (Dynamic Visual Prompting). The navigation policy then naturally detours around this invisible hazard. The identical virtual obstacle is used regardless of hazard type, so the visual prompting pattern remains fixed as sensors are added. When no hazard is detected, nothing is rendered, and the policy behaves exactly as it would without PG-VP. We evaluate PG-VP on OmniNav using the val-unseen splits of R2R-CE and RxR-CE, where it guides the policy toward intended low-risk actions in 84.9% and 83.2% of cases, at a cost of 6.8 and 7.9 percentage points in navigation success rate. We further test it with distinct scenarios on a real robot in the presence of actual thermal and radiation sources, all without any retraining. The real test shows that PG-VP effectively avoids these invisible hazards, improving worst-10% average trajectory safety by 63.45% and 32.59% against thermal and radiation sources, respectively.
Chinese Translation
视觉-语言-行动（VLA）模型已成为视觉与语言导航（VLN）的主要范式。然而，在安全关键设施中，诸如辐射或温度突升等不可见风险无法通过RGB相机检测到，处理每种风险的成本昂贵，需要新的编码器、新的数据和模型重训练。我们提出了基于物理引导的视觉提示（PG-VP），这是一种即插即用的多模态感知模块，它重新利用了冻结的VLA模型已经擅长的功能：避免可见障碍物。在靠近辐射或热源的情况下，PG-VP执行物理引导的风险评估，以确定避让方向，并在连续帧中叠加一个相应的虚拟障碍物（动态视觉提示）。导航策略因此自然地绕过这个不可见的危险。无论危险类型如何，使用的虚拟障碍物都是相同的，因此随着传感器的增加，视觉提示模式保持不变。当未检测到危险时，不会渲染任何内容，策略的行为与没有PG-VP时完全一致。我们在OmniNav上评估PG-VP，使用R2R-CE和RxR-CE的val-unseen拆分，在84.9%和83.2%的情况下引导策略朝向预期的低风险行动，导航成功率分别下降6.8和7.9个百分点。我们还在实际机器人上测试了不同场景，存在实际的热源和辐射源，所有测试均未进行重训练。实际测试表明，PG-VP有效避免了这些不可见的危险，使得在热源和辐射源下，最差10%的平均轨迹安全性分别提高了63.45%和32.59%。
cs.RO / 42 / 2610.07569
OpenSplatGraph: From Dense Semantic Maps to Structured Scene Graphs for Open-Vocabulary Robot Perception
OpenSplatGraph：从密集语义地图到结构化场景图的开放词汇机器人感知
Abstract
Dense 3D mapping with semantic understanding is essential for robotic perception in complex environments. Recent 3D Gaussian Splatting-based mapping approaches enable high-fidelity geometry and efficient open-vocabulary perception, but typically represent semantics as unstructured feature fields that limit object-centric reasoning. In contrast, 3D scene graphs explicitly model objects and their relationships for structured reasoning, but are commonly constructed from sparse geometric representations that do not fully exploit dense semantic maps. In this work, we present OpenSplatGraph, a unified framework that constructs persistent 3D scene graphs directly from an online Gaussian-based open-vocabulary semantic map. The proposed framework augments the dense semantic map with a reliability-aware semantic field that maintains lightweight observation statistics for confidence-aware, query-conditioned object extraction. Extracted object instances are associated with persistent graph nodes, allowing object attributes and relationships to be incrementally updated across observations and queries. By tightly coupling dense semantic mapping with persistent object-centric representations, our framework supports both language-guided object grounding and structured relational reasoning while preserving the geometric fidelity of Gaussian-based mapping. Comprehensive evaluations on standard 3D scene understanding benchmarks and real-world robotic experiments demonstrate that OpenSplatGraph achieves competitive performance for online open-vocabulary perception and downstream robotic tasks. Project page: https://csiro-robotics.github.io/OpenSplatGraph.
Chinese Translation
在复杂环境中，具有语义理解的密集3D映射对于机器人感知至关重要。最近基于3D高斯点云映射的方法实现了高保真几何和高效的开放词汇感知，但通常将语义表示为非结构化特征场，限制了以对象为中心的推理。相比之下，3D场景图明确建模对象及其关系，以支持结构化推理，但通常是从稀疏几何表示中构建的，未能充分利用密集语义地图。在本研究中，我们提出了OpenSplatGraph，一个统一框架，能够直接从在线高斯基础的开放词汇语义地图构建持久的3D场景图。该框架通过一个可靠性感知的语义场增强了密集语义地图，维护轻量级的观察统计数据，以支持基于信心的、条件查询的对象提取。提取的对象实例与持久的图节点关联，使得对象属性和关系能够在观察和查询中逐步更新。通过将密集语义映射与持久的以对象为中心的表示紧密结合，我们的框架支持语言引导的对象定位和结构化关系推理，同时保持基于高斯映射的几何保真性。在标准3D场景理解基准和真实世界机器人实验上的全面评估表明，OpenSplatGraph在在线开放词汇感知和下游机器人任务中实现了竞争力的性能。项目页面：https://csiro-robotics.github.io/OpenSplatGraph。
cs.RO / 43 / 2610.07594
BiGym 2.0: Benchmarking Learned and Agent-Developed Policies for Humanoid Household Manipulation
BiGym 2.0：人形机器人家庭操作的学习与代理开发策略基准测试
Abstract
Humanoid household manipulation requires the arms to act while the body balances, steps and changes posture. We present BiGym 2.0, an adaptation of BiGym for the Unitree G1 across 20 household tasks using a unified whole-body controller for demonstration and evaluation. The suite provides 60 native human virtual-reality demonstrations per task with synchronised multi-camera views and full-body execution records. We benchmark vision-language-action fine-tuning, imitation learning, demo-driven reinforcement learning, and cold-start coding agents given the interaction budget of online reinforcement learning. With the same onboard views, proprioception and whole-body controller for every method, vision-language-action fine-tuning has the highest nine-task mean, and agent-developed programs outperform every demo-driven reinforcement learning baseline on this mean and lead on bimanual reaching. Cross-workspace stacking remains open, $\pi_{0.5}$ stays low on pick-box, and multi-object transport is hard for imitation learning, demo-driven reinforcement learning and coding agents. All environments, human demonstrations, and evaluation traces are open-sourced at https://github.com/swirl-uk/BiGym2.
Chinese Translation
人形机器人家庭操作要求手臂在身体保持平衡、行走和改变姿势的同时进行动作。我们提出了BiGym 2.0，这是BiGym在Unitree G1上的适配版本，涵盖20项家庭任务，采用统一的全身控制器进行演示和评估。该套件为每个任务提供60个原生人类虚拟现实演示，配有同步的多摄像头视角和全身执行记录。我们基准测试了视觉-语言-动作微调、模仿学习、演示驱动的强化学习以及在在线强化学习的交互预算下的冷启动编码代理。在相同的机载视角、身体感知和全身控制器下，视觉-语言-动作微调在九项任务中的平均表现最高，而代理开发的程序在该平均值上超越了每个演示驱动的强化学习基线，并在双手到达任务中领先。跨工作空间堆叠仍然是一个开放问题，$ ext{π}_{0.5}$在取箱任务上保持较低，而多物体运输对模仿学习、演示驱动的强化学习和编码代理来说都很困难。所有环境、人类演示和评估记录均已开源，网址为 https://github.com/swirl-uk/BiGym2。
cs.RO / 44 / 2610.07597
The Robot Is Not Its Description: GaugeBench for Representation Robustness in Morphology-Aware Policies
机器人并非其描述：用于形态感知策略的 GaugeBench 表示鲁棒性
Abstract
A robot description does more than specify a physical mechanism: it also encodes arbitrary conventions, such as joint-axis direction, joint-angle zero, and the order and names of links and joints. Morphology-aware policies consume interfaces built from these descriptions, yet cross-embodiment evaluation typically changes the robot while keeping those conventions fixed. This leaves a simple question unanswered: does behavior survive when the robot stays fixed but its description changes? GaugeBench isolates this case by rewriting a fixed mechanism under physically equivalent conventions, verifying that its physics and policy interface are preserved, and then evaluating the same policy weights. The result is stark: three MetaMorph policies score 4030.6 on 80 familiar robots, but only 51.6 when those same robots are equivalently re-described, while 98 genuinely held-out robots score 1489.6. A new description can therefore be more damaging than a new robot. Tracing the failure reveals that axis reversal alone reproduces the collapse, joint-angle zero changes are nearly harmless, and reordering lies between them; moreover, changing joint-state and torque coordinates alone is sufficient to cause the failure, while changing description-derived features alone is not. The same phenomenon appears in ModuMorph and an unrelated PyBullet framework. Yet it is not irreversible: exact two-description transport restores the original controller, and training across equivalent axis conventions raises retained return under axis reversal from 3.6% to 80.6%. Together, these results separate mechanism robustness from representation robustness and show that cross-embodiment evaluation should test both.
Chinese Translation
机器人描述不仅仅是指定一个物理机制：它还编码了任意的约定，例如关节轴方向、关节角零点，以及连杆和关节的顺序和名称。形态感知策略使用基于这些描述构建的接口，然而跨体现评估通常在保持这些约定不变的情况下改变机器人。这留下了一个简单的问题没有回答：当机器人保持不变但其描述发生变化时，行为是否依然存在？GaugeBench 通过在物理等效的约定下重写一个固定机制来隔离这种情况，验证其物理特性和策略接口是否被保留，然后评估相同的策略权重。结果非常明显：三种 MetaMorph 策略在 80 个熟悉的机器人上得分为 4030.6，但当这些相同的机器人被等效重新描述时仅得 51.6，而 98 个真正的保留机器人得分为 1489.6。因此，一个新的描述可能比一个新的机器人更具破坏性。追踪失败的原因表明，仅仅轴反转就能重现崩溃，关节角零点的变化几乎无害，而重新排序则介于两者之间；此外，仅改变关节状态和扭矩坐标就足以导致失败，而仅改变描述派生特征则不足以造成失败。相同的现象出现在 ModuMorph 和一个无关的 PyBullet 框架中。然而，这并不是不可逆的：精确的双描述传输恢复了原始控制器，而在等效轴约定下的训练使得在轴反转下的保留回报从 3.6% 提升至 80.6%。综合来看，这些结果将机制鲁棒性与表示鲁棒性区分开来，并表明跨体现评估应同时测试两者。
cs.RO / 45 / 2610.07599
Modeling Latent Disturbances for Robust Decision-Making in World Models
为世界模型中的稳健决策建模潜在干扰
Abstract
In this paper, we study robust decision-making in the latent space of world models (WMs). Robust optimization is a mathematical framework where, given explicitly specified dynamics and physically meaningful disturbances, a robot can select actions that remain effective even under worst-case disturbances. However, applying this principle to the learned latent space of WMs introduces a fundamental challenge: because WMs have fully learned state spaces and dynamics inferred from high-dimensional observations, it is unclear how to define latent-space disturbances that faithfully represent uncertainty in the underlying system. Our key idea is to model a latent-space disturbance as a perturbation to the learned latent dynamics that induces pessimistic but plausible transitions. Specifically, we construct a set of plausible latent dynamics by combining a dynamics-aware similarity metric that captures plausible transitions with out-of-distribution detection that excludes implausible latent states. We calibrate this uncertainty set over latent dynamics using conformal prediction, ensuring that WM imaginations induced by the latent disturbance remain plausible without becoming overly pessimistic. We then jointly optimize robust robot actions and the worst-case latent disturbances through game-theoretic optimization. We leverage this latent-space robust optimization to robustify policy steering, considering two paradigms: latent safety filtering and sample-and-verify steering of a generative control policy. Our controlled simulation experiments show that our latent disturbance enables robust decision-making directly in WM latent spaces, and hardware experiments with a Franka manipulator show that modeling latent disturbances enables robust policy steering, reducing failures by 70% in safety filtering and 54% in sampling-based policy steering. Project website: https://junwon.me/LatentDisturbance/.
Chinese Translation
在本文中，我们研究了世界模型（WMs）潜在空间中的稳健决策。稳健优化是一种数学框架，在该框架中，给定明确指定的动态和具有物理意义的干扰，机器人可以选择在最坏情况下仍然有效的动作。然而，将这一原则应用于WMs学习到的潜在空间时，面临一个基本挑战：由于WMs已完全学习了从高维观察中推断出的状态空间和动态，因此不清楚如何定义潜在空间中的干扰，以真实地代表基础系统中的不确定性。我们的关键思想是将潜在空间干扰建模为对学习到的潜在动态的扰动，这种扰动会引发悲观但合理的转变。具体而言，我们通过结合捕捉合理转变的动态感知相似性度量和排除不合理潜在状态的分布外检测，构建了一组合理的潜在动态。我们使用符合预测对潜在动态的这一不确定性集进行校准，确保潜在干扰引发的WM想象保持合理，而不会变得过于悲观。然后，我们通过博弈论优化共同优化稳健的机器人动作和最坏情况下的潜在干扰。我们利用这种潜在空间的稳健优化来增强策略引导，考虑两种范式：潜在安全过滤和基于采样的生成控制策略的验证引导。我们的受控仿真实验表明，我们的潜在干扰使得在WM潜在空间中直接进行稳健决策成为可能，而使用Franka操纵器的硬件实验则表明，建模潜在干扰使得稳健策略引导成为可能，在安全过滤中减少了70%的失败率，在基于采样的策略引导中减少了54%的失败率。项目网站：https://junwon.me/LatentDisturbance/.
cs.RO / 46 / 2610.07613
Learning Grasp Targeting from Point Clouds for Log Pile Clearing on a Hydraulic Crane
从点云中学习抓取目标以实现液压起重机的木材堆清理
Abstract
In mill yards, log loaders clear dense piles by a sequence of bundle grasps: hundreds of logs rest in contact, and each removal changes the pile available to the next grasp. A learned policy chooses where to place and orient the grapple from unsegmented point clouds and runs on a trailer-mounted hydraulic forestry crane. The policy classifies at which observed point to grasp and predicts depth and grapple orientation there. The same network outputs support behavior cloning (BC), reinforcement learning (RL), and deployment. BC learns from successful top-of-pile demonstrations; RL explores for improvements by fine-tuning the cloned policy (BC$\to$RL) or by training from scratch. In simulation, BC clears 98 of 100 piles of 200 logs, while BC$\to$RL improves load stability. Twelve field trials compare a geometric heuristic, RL from scratch, BC, and BC$\to$RL through complete grasp-transport-deposit cycles. BC and BC$\to$RL deposit 93.8% and 88.9% of pooled inventory, against 80.4% for the heuristic. BC$\to$RL deposits logs on 83.6% of its cycles, against 79.6% for the heuristic and 65.7% for BC, while its simulated stability gain does not carry over to the crane testbed. Trained entirely in simulation and run unchanged on the crane, the learned policies clear more than the hand-filtered heuristic while observing unfiltered clouds that still contain the storage rack's rails and poles.
Chinese Translation
在木材场中，木材装载机通过一系列捆绑抓取来清理密集的木材堆：数百根木材相互接触，每次移除都会改变下一个抓取可用的堆积情况。一个学习到的策略从未分割的点云中选择抓取的位置和方向，并在拖车安装的液压林业起重机上运行。该策略对观察到的点进行分类以决定抓取位置，并预测抓取深度和方向。相同的网络输出支持行为克隆（Behavior Cloning, BC）、强化学习（Reinforcement Learning, RL）和部署。BC从成功的顶部堆演示中学习；RL通过微调克隆策略（BC→RL）或从头开始训练来探索改进。在仿真中，BC清理了100个包含200根木材的堆中的98个，而BC→RL则改善了负载稳定性。十二次现场试验比较了几何启发式、从头开始的RL、BC和BC→RL在完整的抓取-运输-放置周期中的表现。BC和BC→RL分别放置了93.8%和88.9%的集中库存，而启发式方法为80.4%。BC→RL在其周期中放置了83.6%的木材，而启发式方法为79.6%，BC为65.7%，尽管其在仿真中的稳定性提升未能在起重机测试平台上体现。完全在仿真中训练并在起重机上不变地运行的学习策略，清理的数量超过了手动过滤的启发式方法，同时观察到的未过滤点云中仍包含储存架的轨道和柱子。
cs.RO / 47 / 2610.07616
Nine Trials to Recover: A Reproducible Benchmark for Repertoire-Free Soft-Robot Damage Adaptation
九次恢复试验：一种可重复的无库软机器人损伤适应基准
Abstract
We present a repertoire-free benchmark for soft-robot damage recovery under nine online trials. The protocol pairs damage masks within each morphology, retains a measured nominal fallback, and separates method development from evaluation on new bodies. A Gaussian-process expected-improvement (GP-EI) reference controller adapts actuator phases using three initialization probes and six feedback-selected rollouts. Across two disjoint 75-body cohorts, it outperforms random search, Sobol, CEM, and CMA-ES under equal budgets. GP-EI improves the worst-mask gain on 69/75 confirmation bodies and exceeds these four baselines by 0.235-0.310 mean worst-mask reward. A TuRBO-style local GP is the closest comparator; the paired confidence interval includes zero. Post-confirmation fixed-controller replay finds 5.06 voxel widths of mean recovery together with a 0.0031 increase in worst-mask p99 geometric edge strain; a strict no-added-demand deployment gate retains 48.7% of mean gain. In a frozen development stress test that physically removes 10% of occupied voxels, GP-EI improves 71/75 bodies and exceeds official BoTorch TuRBO by 0.356 mean worst-mask gain. Together, the replayable controllers, body-level inference, and frozen-cohort evaluation provide a reference for measuring the added value of learned damage priors and future adaptation methods.
Chinese Translation
我们提出了一种无库的软机器人损伤恢复基准，基于九次在线试验。该协议在每种形态中配对损伤掩码，保留测量的名义备用，并将方法开发与在新体上进行评估分开。高斯过程期望改进（Gaussian-process expected-improvement, GP-EI）参考控制器通过三个初始化探针和六个反馈选择的回放来调整执行器相位。在两个不重叠的75体群中，其表现优于随机搜索、Sobol、CEM和CMA-ES，在相同预算下。GP-EI在69/75确认体上改善了最差掩码增益，且在这四个基准上超出0.235-0.310的平均最差掩码奖励。TuRBO风格的局部高斯过程是最接近的比较对象；配对置信区间包括零。在确认后的固定控制器重放中，平均恢复找到5.06个体素宽度，同时最差掩码的p99几何边缘应变增加了0.0031；严格的无附加需求部署门保留了48.7%的平均增益。在一个冻结开发压力测试中，物理移除10%的占用体素，GP-EI改善了71/75个体，并在平均最差掩码增益上超出官方的BoTorch TuRBO 0.356。综上所述，可重放控制器、体级推断和冻结队列评估为测量学习损伤先验和未来适应方法的附加价值提供了参考。
cs.RO / 48 / 2610.07621
TacZero: Training-Free Peg Insertion Using a General-Purpose Vision-Language Model with Tactile Feedback
TacZero：使用通用视觉-语言模型和触觉反馈进行无训练的插销插入
Abstract
Robots that autonomously determine their actions from language instructions and sensory observations could perform new contact-rich manipulation tasks without task-specific training or hand-designed rules. To perform these tasks, robots must infer how objects contact one another and move as a result, then select actions. For contact inference and action selection, prior approaches involve designing estimation models and tactile feedback control laws, or learning models for object-motion estimation, action-outcome prediction, and action selection from tactile data. Instead, we propose TacZero, which uses a pretrained general-purpose vision-language model (VLM) to interpret visual and tactile observations and select robot actions without additional tactile or manipulation training or task-specific rules for contact interpretation or action selection. TacZero provides the VLM with camera images, robot state, and three-axis tactile responses represented as numerical values or vectors overlaid on the images. From these observations and interaction history, the VLM generates commands specifying target end-effector positions and gripper opening or closing, which a low-level controller executes. In real-world cylindrical-peg insertion experiments, TacZero succeeded in 15 of 20 trials with numerical tactile input, compared with 10 of 20 without tactile input. This study provides a concrete starting point for further research on contact-rich manipulation using general-purpose VLMs and highlights challenges in pursuing this direction.
Chinese Translation
能够根据语言指令和感官观察自主决定行动的机器人，可以在没有特定任务训练或手工设计规则的情况下执行新的接触丰富的操作任务。为了执行这些任务，机器人必须推断物体之间的接触方式及其导致的运动，然后选择相应的行动。对于接触推断和行动选择，之前的方法涉及设计估计模型和触觉反馈控制法则，或从触觉数据中学习物体运动估计、行动结果预测和行动选择的模型。相反，我们提出了TacZero，它使用预训练的通用视觉-语言模型（VLM）来解释视觉和触觉观察，并选择机器人行动，而无需额外的触觉或操作训练或特定任务规则来进行接触解释或行动选择。TacZero向VLM提供相机图像、机器人状态和以数值或向量形式叠加在图像上的三轴触觉响应。基于这些观察和交互历史，VLM生成指令，指定目标末端执行器位置以及夹爪的开合，这些指令由低级控制器执行。在实际的圆柱插销插入实验中，TacZero在20次试验中成功了15次，而没有触觉输入时成功了10次。本研究为进一步研究使用通用VLM进行接触丰富的操作提供了一个具体的起点，并突出了在这一方向上追求的挑战。
cs.RO / 49 / 2610.07629
Beyond Task Reward: A Controller-Restriction Protocol for Evaluating Embodiment-Dependent Competence
超越任务奖励：评估依赖于具身性的能力的控制器限制协议
Abstract
Co-design methods optimize a robot's body and controller jointly and judge the result by one number, the task reward of the fully optimized pair. That number cannot separate morphologies whose competence depends on the controller to very different degrees. We evaluate a morphology by restricting its controller instead, recording the task competence it retains under an explicitly declared, low-complexity controller family, environment, task and search budget. On three EvoGym locomotion tasks, task reward explains only $39\%$, $33\%$ and $10\%$ of the variance in this quantity, and geometric descriptors do not predict it under run-grouped cross-validation. The measurement is reliable across optimizer restarts (ICC$(2,k) = 0.956$--$0.986$) but depends on the declared family: phasing the drive by actuator index instead of position ranks the same morphologies at Spearman $0.50$--$0.63$ and reverses reward-matched pairs. As a second search objective the axis improved competence at matched task reward in $3$ of $5$ paired runs, short of a pre-registered bar of $4$. Used after an ordinary reward-only search instead, to choose within its top task-reward band, it selected a different body in all $8$ runs offering a choice, at a cost of at most $0.10$ reward units, and in $6$ of $8$ that body also scored higher under a held-out family. Restricted-control competence is therefore a reportable property of a co-designed morphology, interpretable only with the controller family that defines it.
Chinese Translation
共设计方法共同优化机器人的身体和控制器，并通过一个数字来评估结果，即完全优化配对的任务奖励。这个数字无法区分那些能力在不同程度上依赖于控制器的形态。我们通过限制控制器来评估形态，记录在明确声明的低复杂度控制器家族、环境、任务和搜索预算下其保留的任务能力。在三个EvoGym运动任务中，任务奖励仅解释了$39 ext{%}$、$33 ext{%}$和$10 ext{%}$的方差，而几何描述符在分组交叉验证下无法预测该量度。该测量在优化器重启之间是可靠的（ICC$(2,k) = 0.956$--$0.986$），但依赖于声明的家族：通过执行器索引而非位置对驱动进行相位调整时，相同的形态在Spearman $0.50$--$0.63$下排名，并且反转了奖励匹配对。作为第二个搜索目标，该轴在$5$次配对运行中的$3$次中提高了匹配任务奖励的能力，未达到预注册的$4$的标准。相反，在普通的仅奖励搜索之后使用该目标，以选择其最高任务奖励区间内的形态，它在所有$8$次提供选择的运行中选择了不同的身体，最多损失$0.10$奖励单位，并且在$8$次中的$6$次中，该身体在保留的家族下也得分更高。因此，限制控制能力是共设计形态的可报告属性，仅能通过定义它的控制器家族进行解释。
cs.RO / 50 / 2610.07649
OntoPlan: An Ontology-Grounded Scene Representation and Agentic Framework for Scalable Robot Task Planning
OntoPlan：一种基于本体的场景表示与可扩展机器人任务规划的智能框架
Abstract
Large language model (LLM)-based robot task planning is promising for open-ended instruction following, but degrades on long-horizon tasks in large environments. When spatial information is conveyed to the LLM through text, the model can fail to capture spatial context, and token cost grows with environment size. Generating action sequences directly with an LLM also makes it difficult to satisfy the current world state and action preconditions. We address this with an ontology-grounded scene representation that aligns objects, spaces, relations, and states in a shared symbolic vocabulary for spatial reasoning and task planning, and with OntoPlan, an agentic framework that interprets instructions, selectively retrieves task-relevant information, formalizes goals and constraints, and produces executable plans. Across 150 general tasks spanning five indoor environments and three scene scales, OntoPlan achieves 0.89 average task success, compared with 0.27 for the strongest baseline, while using 18.1k total tokens per task on average, about 5.6$\times$ fewer than the most efficient baseline. These advantages persist as scene scale increases, whereas prior methods degrade more sharply in success and remain far more costly in tokens. OntoPlan also responds appropriately to ambiguous or infeasible instructions by asking follow-up questions or reporting insufficient information rather than committing to invalid plans. Code available at https://github.com/namhyeongwoo/OntoPlan.
Chinese Translation
基于大型语言模型（LLM）的机器人任务规划在开放式指令跟随方面展现出良好前景，但在大型环境中的长时间任务上表现不佳。当空间信息通过文本传达给LLM时，模型可能无法捕捉空间上下文，并且随着环境规模的增大，标记成本也会增加。直接使用LLM生成动作序列也使得满足当前世界状态和动作前提条件变得困难。我们通过一种基于本体的场景表示来解决这一问题，该表示在共享的符号词汇中对对象、空间、关系和状态进行对齐，以便进行空间推理和任务规划，同时结合OntoPlan，一个智能框架，用于解释指令、选择性地检索与任务相关的信息、形式化目标和约束，并生成可执行的计划。在涵盖五个室内环境和三个场景规模的150个通用任务中，OntoPlan的平均任务成功率达到0.89，而最强基线仅为0.27，同时每个任务平均使用18.1k个标记，约为最有效基线的5.6倍更少。这些优势在场景规模增加时依然存在，而之前的方法在成功率上急剧下降，并且在标记使用上仍然显著更高。OntoPlan还能够适当地响应模糊或不可行的指令，通过提问或报告信息不足来处理，而不是承诺无效的计划。代码可在 https://github.com/namhyeongwoo/OntoPlan 获取。
cs.RO / 51 / 2610.07650
Silicon Language: A Robot-Native Knowledge Exchange Framework for Heterogeneous Robots
硅语言：一种异构机器人原生知识交换框架
Abstract
Reusing a capability across heterogeneous robots still requires substantial human adaptation and verification: transferring a skill often means re-engineering interfaces, retuning parameters, and re-validating safety. We introduce Silicon Language, a robot-native knowledge exchange framework that treats the robot as the active subject of its own capability evolution. In this framework, a robot that wants a capability encodes its own experience into knowledge packets, publishes them, retrieves peer packets, translates them for its own sensors and actuators, and reviews them through independent local trial. A receiver-side usability evaluation procedure lets each robot decide for itself whether an external packet is useful, and progressive blending with automatic rollback is designed to reduce the risk of negative transfer when adopting it. The system combines three infrastructure layers (edge agent, Silicon Transfer Protocol (STP), and knowledge hub) with a capability stack inspired by the human scholarly system. We report a 30-day proof-of-concept deployment at an above-ground simulated-mine laboratory in Yulin, with following trials on an outdoor sand road and an indoor factory floor. Two heterogeneous robots encoded and translated three capabilities across embodiments through operator-assisted file copies mediated by the Silicon Language translation layer; source-side trials of the dust-locked following behavior were recorded on Taurus. Project records indicate that per-capability adaptation time dropped from days to hours; we present these figures as descriptive deployment records rather than controlled measurements. The deployment provides initial evidence for cross-embodiment knowledge exchange; fleet-level autonomous evolution and controlled with/without-packet comparisons remain future work.
Chinese Translation
在异构机器人之间重用能力仍然需要大量的人为适应和验证：转移技能通常意味着重新设计接口、重新调整参数和重新验证安全性。我们提出了硅语言（Silicon Language），一种将机器人视为其自身能力演化的主动主体的原生知识交换框架。在该框架中，想要获得某种能力的机器人将其自身经验编码为知识包，发布这些知识包，检索同伴的知识包，为其自身的传感器和执行器翻译这些知识包，并通过独立的本地试验进行审查。接收端的可用性评估程序使每个机器人能够自行判断外部知识包是否有用，并设计了渐进式融合与自动回滚，以降低在采用外部知识包时负迁移的风险。该系统结合了三个基础设施层（边缘代理、硅传输协议（Silicon Transfer Protocol, STP）和知识中心）以及一个受人类学术系统启发的能力堆栈。我们报告了在榆林的一个地面模拟矿山实验室进行的为期30天的概念验证部署，随后在户外沙路和室内工厂车间进行了试验。两个异构机器人通过硅语言翻译层介导的操作员辅助文件复制，编码并翻译了三种能力；在Taurus上记录了源端试验的防尘跟随行为。项目记录表明，每种能力的适应时间从几天减少到几个小时；我们将这些数据作为描述性部署记录，而非控制测量。该部署为跨体现知识交换提供了初步证据；车队级的自主演化以及有无知识包的控制比较仍然是未来的工作。
cs.RO / 52 / 2610.07652
SMART: Zero-Shot Sim-to-Real Articulated Object Manipulation via Large-Scale Synthetic Pretraining
SMART：通过大规模合成预训练实现零-shot模拟到真实的关节物体操作
Abstract
The ability to interact with articulated objects is essential for embodied intelligent systems, but collecting large-scale real-world demonstrations for these interactions remains challenging due to the precise contact and constraint-following motions involved. Although simulation provides a promising alternative, existing synthetic data efforts cover limited articulated-object categories, while general-purpose synthesis pipelines lack explicit designs for part-level semantics and articulation constraints, hindering agentic task generation and scalable synthesis of high-quality articulated-manipulation demonstrations. To bridge this gap, we introduce SMART, a scalable system leveraging large-scale Synthesized Manipulation demonstrations for ARTiculated-object manipulation. At its core, we develop SMART-Sim, a simulation platform with articulation-aware design that enables effective task generation and efficient demonstration collection. Building on SMART-Sim, we apply agentic task generation and design a scalable distributed synthesis system, using them to synthesize SMART-Data, comprising over 1M demonstrations across 44 atomic task types, 5 robot setups, and 2,507 articulated objects. The vision-language-action (VLA) model pretrained on SMART-Data shows competitive performance on simulation benchmarks and achieves zero-shot sim-to-real transfer and scalable performance in real-world articulated-object manipulation tasks. This highlights the potential of synthetic demonstrations in providing effective and scalable supervision for improving VLA model performance in contact-rich articulated-object manipulation.
Chinese Translation
与关节物体进行交互的能力对于具身智能系统至关重要，但由于涉及精确的接触和约束跟随动作，收集大规模的真实世界演示仍然具有挑战性。尽管模拟提供了一个有前景的替代方案，但现有的合成数据工作覆盖的关节物体类别有限，而通用合成管道缺乏针对部件级语义和关节约束的明确设计，阻碍了代理任务生成和高质量关节操作演示的可扩展合成。为了解决这一问题，我们提出了SMART，一个利用大规模合成操作演示进行关节物体操作的可扩展系统。其核心是我们开发的SMART-Sim，一个具有关节感知设计的模拟平台，能够有效生成任务并高效收集演示。在SMART-Sim的基础上，我们应用代理任务生成并设计了一个可扩展的分布式合成系统，利用这些系统合成了SMART-Data，包含超过100万条演示，涵盖44种原子任务类型、5种机器人设置和2507个关节物体。在SMART-Data上预训练的视觉-语言-动作（VLA）模型在模拟基准测试中表现出竞争力，并在真实世界的关节物体操作任务中实现了零-shot模拟到真实的迁移和可扩展性能。这突显了合成演示在提供有效和可扩展的监督方面的潜力，从而改善VLA模型在接触丰富的关节物体操作中的性能。
cs.RO / 53 / 2610.07681
EigenDEXplore: Structured Exploration for Dexterous Manipulation with Human Priors
EigenDEXplore：基于人类先验的灵巧操作结构化探索
Abstract
Dexterous manipulation poses a challenging high-dimensional optimization problem, as useful behaviors require coordinated motion across many hand joints. In reinforcement learning (RL) and sampling-based trajectory optimization, exploration commonly relies on independent robot joint perturbations, making coordinated behaviors difficult to discover. Prior work reduces this search space for grasp learning using low-dimensional spaces of coordinated joint motions learned from human hand data, but this restricts the expressivity required for general manipulation. Some combine learned and joint-space actions to restore expressivity, but this increases dimensionality and introduces redundancy. We study these effects across diverse manipulation settings, varying action dimensionality, exploration strategy, and the source of human data. Our experiments suggest that human-motion priors are most effective when used to structure exploration rather than change the action representation. Motivated by this finding, we propose EigenDEXplore, which induces correlated exploration by adding perturbations along human-derived eigenvectors to independent joint-space noise, leaving the action space unchanged. Across multiple dexterous hands, EigenDEXplore consistently outperforms joint-space and learned action-space baselines in grasping, in-hand reorientation, and contact-rich manipulation. These gains span unstructured and reference-guided RL, trajectory optimization, and sim-to-real deployment, and are largest in settings with less reward shaping and curriculum design.
Chinese Translation
灵巧操作提出了一个具有挑战性的高维优化问题，因为有效的行为需要多个手指关节之间的协调运动。在强化学习（RL）和基于采样的轨迹优化中，探索通常依赖于独立的机器人关节扰动，这使得协调行为的发现变得困难。先前的研究通过利用从人类手部数据中学习的协调关节运动的低维空间来减少抓取学习的搜索空间，但这限制了通用操作所需的表达能力。有些方法结合了学习的和关节空间的动作以恢复表达能力，但这增加了维度并引入了冗余。我们研究了这些效应在不同操作设置中的表现，变化了动作维度、探索策略和人类数据的来源。我们的实验表明，当人类运动先验用于结构化探索而不是改变动作表示时，效果最佳。基于这一发现，我们提出了EigenDEXplore，通过沿着人类派生的特征向量添加扰动到独立的关节空间噪声中，从而引导相关的探索，同时保持动作空间不变。在多个灵巧手的实验中，EigenDEXplore在抓取、手内重新定向和接触丰富的操作中始终优于关节空间和学习的动作空间基线。这些增益涵盖了无结构和参考引导的强化学习、轨迹优化以及仿真到现实的部署，并且在奖励塑造和课程设计较少的设置中表现出最大的提升。
cs.RO / 54 / 2610.07692
ExoBridge: Learning a Bare Hand to Hand-Worn Exoskeleton Mapping through Human Limb Coupling
ExoBridge：通过人类肢体耦合学习裸手与手穿戴外骨骼的映射
Abstract
Human video offers a scalable source of experience for dexterous robot learning, but obtaining motion and tactile supervision while preserving bare hand interaction remains challenging. We present ExoBridge, a framework that leverages human limb coupling to learn a bridging function from bare hand video to the motion and tactile state of a sensorized exoskeleton. Our central idea is to use coordinated bimanual behavior to connect an uninstrumented visual source with a measured manipulation interface. During collection, one hand remains bare and provides visual observations, while the opposite hand wears the exoskeleton and supplies synchronized motion and tactile measurements. These paired demonstrations train a temporal visual model to predict fingertip contact, continuous tactile intensity, and relative encoder motion from bare hand video alone. The exoskeleton defines an intermediate state space whose motion coordinates are linked to a dexterous robot hand through existing calibration. Evaluation on 1,215 demonstrations across four manipulation tasks uses held out collection sessions and yields a pooled any contact AUROC of 0.916 and a Pearson correlation of 0.790 for tactile intensity. The learned bridge also predicts relative changes in exoskeleton configuration from bare hand video. These results demonstrate that human limb coupling can turn exoskeleton measurements into supervision for bare hand video, establishing a learned bridge between human visual demonstrations and a robot oriented manipulation interface.
Chinese Translation
人类视频为灵巧机器人学习提供了一种可扩展的经验来源，但在保持裸手交互的同时获取运动和触觉监督仍然具有挑战性。我们提出了ExoBridge，一个利用人类肢体耦合来学习从裸手视频到传感外骨骼运动和触觉状态的桥接函数的框架。我们的核心思想是使用协调的双手行为将未加仪器的视觉源与测量的操作界面连接起来。在数据收集过程中，一只手保持裸露并提供视觉观察，而另一只手则穿戴外骨骼并提供同步的运动和触觉测量。这些配对演示训练一个时序视觉模型，以仅从裸手视频预测指尖接触、连续触觉强度和相对编码器运动。外骨骼定义了一个中间状态空间，其运动坐标通过现有的标定与灵巧机器人手相连接。在四个操作任务中对1215个演示的评估使用了保留的收集会话，并获得了0.916的任何接触AUROC和0.790的触觉强度皮尔逊相关性。学习到的桥接函数还可以从裸手视频预测外骨骼配置的相对变化。这些结果表明，人类肢体耦合可以将外骨骼测量转化为裸手视频的监督，建立人类视觉演示与机器人导向操作界面之间的学习桥梁。
cs.RO / 55 / 2610.07696
ESP: Energy-Score Policy for One-Step Multimodal Action Generation
ESP：用于一步多模态动作生成的能量评分策略
Abstract
Generative action models based on diffusion and flow matching have been increasingly adopted in vision-language-action (VLA) policies for their ability to capture diverse behaviors, including multiple valid action sequences under the same observation and instruction. Their iterative sampling procedures, however, require repeated network evaluations to generate each action chunk, increasing inference latency in closed-loop control. We propose ESP (Energy-Score Policy), a teacher-free approach that maps policy context and noise directly to an action chunk in a single network evaluation. ESP trains the action head with the energy score rather than mean squared error. Whereas squared-error regression targets the conditional mean, the energy score is strictly proper: its expected value is uniquely minimized by the target distribution. This provides a principled objective for learning multimodal action distributions without iterative sampling, with exact recovery at the population optimum when the model can represent the target distribution. Experiments on both simulation and real-world manipulation tasks demonstrate competitive task success with substantially lower action-generation latency than the flow matching baseline. These results support direct distributional learning as an efficient alternative to iterative generative robot policies.
Chinese Translation
基于扩散和流匹配的生成动作模型在视觉-语言-动作（VLA）策略中越来越受到欢迎，因为它们能够捕捉多样化的行为，包括在相同观察和指令下的多个有效动作序列。然而，它们的迭代采样过程需要重复评估网络以生成每个动作块，从而增加了闭环控制中的推理延迟。我们提出了ESP（能量评分策略），这是一种无教师的方法，能够将策略上下文和噪声直接映射到一个动作块中，仅需一次网络评估。ESP使用能量评分而不是均方误差来训练动作头。虽然均方误差回归的目标是条件均值，但能量评分是严格适当的：其期望值由目标分布唯一最小化。这为学习多模态动作分布提供了一个有原则的目标，无需迭代采样，当模型能够表示目标分布时，可以在总体最优时实现精确恢复。在模拟和现实世界操作任务上的实验表明，与流匹配基线相比，任务成功率具有竞争力，同时动作生成延迟显著降低。这些结果支持直接分布学习作为迭代生成机器人策略的高效替代方案。
cs.RO / 56 / 2610.07745
Seeing Through the Displaced Frame: Privileged Noise Distillation for Vision-Force Precision Assembly
透视位移框架：用于视觉-力精密装配的特权噪声蒸馏
Abstract
Pose error in precision assembly can corrupt not only what a robot observes but also the coordinate frame in which it acts. On the FORGE benchmark, the official state-based policy succeeds in 97% to 99% of episodes with the true pose but only 32% to 60% at the benchmark's $\sigma=5$ mm pose-noise setting. The same estimated pose enters the observation and anchors the action frame, making the offset unidentifiable from proprioceptive state alone before contact. We supply this missing information during training in two ways. A privileged teacher observes the offset in simulation, while clean demonstrations can instead be relabelled into the displaced frame in closed form. The deployed student is trained with behaviour cloning followed by one DAgger round and receives only noisy state, a raw wrench window, and two RGB cameras at test time. On the unmodified FORGE tasks, the teacher-route student maintains 92% to 99% success across $\sigma=0$ to 5 mm, while six non-privileged baselines fall to 2% to 80%. A matched behaviour-cloning experiment isolates the source of this robustness. With the same student architecture, data budget, and training procedure, demonstrations generated without offset access yield only 31.5% success at $\sigma=5$ mm, whereas privileged and relabelled demonstrations reach 88.7% and 95.8%. Deployed zero-shot on a Franka, the student reaches 83.3% pooled success at $\sigma=5$ mm against 34.4% for the strongest state-based policy. Deployable sensing alone is insufficient. Robustness requires supervision that encodes compensation for the latent frame offset.Project page: https://drychang.github.io/displaced-frame/
Chinese Translation
在精密装配中，姿态误差不仅会影响机器人观察到的内容，还会影响其操作的坐标框架。在FORGE基准测试中，官方的基于状态的策略在真实姿态下成功率为97%到99%，而在基准测试的$ ext{σ}=5$ mm姿态噪声设置下仅为32%到60%。同样的估计姿态进入观察并锚定动作框架，使得在接触之前，仅凭本体状态无法识别偏移。我们通过两种方式在训练中提供缺失的信息。一个特权教师在仿真中观察到偏移，而干净的演示可以在封闭形式中重新标记为位移框架。部署的学生通过行为克隆进行训练，随后进行一次DAgger轮次，并在测试时仅接收噪声状态、原始扭矩窗口和两个RGB摄像头。在未修改的FORGE任务中，教师路径的学生在$ ext{σ}=0$到5 mm的成功率维持在92%到99%之间，而六个非特权基线的成功率降至2%到80%。一个匹配的行为克隆实验隔离了这种鲁棒性的来源。在相同的学生架构、数据预算和训练程序下，未获取偏移的演示在$ ext{σ}=5$ mm下仅达到31.5%的成功率，而特权和重新标记的演示则分别达到88.7%和95.8%。在Franka上零-shot部署时，学生在$ ext{σ}=5$ mm下的综合成功率达到83.3%，而最强的基于状态的策略仅为34.4%。单靠可部署的传感器不足以保证鲁棒性。鲁棒性需要编码对潜在框架偏移的补偿的监督。项目页面：https://drychang.github.io/displaced-frame/
cs.RO / 57 / 2610.07752
CoRE: Learning Collaboration-Role Experts for Decentralized Collaborative Manipulation with One Policy
CoRE：学习协作角色专家以实现基于单一策略的去中心化协作操控
Abstract
Collaborative manipulation requires robots to perform complementary actions as interactions unfold. We study single-policy decentralized collaboration: every robot runs the same policy from its visual observations and proprioception, without task prompts, identity labels, or inter-robot messages. The challenge is to learn complementary team behaviors within shared parameters and select appropriate actions from each robot's local observations. We introduce CoRE, which learns Collaboration-Role Experts from pooled multi-task, multi-robot demonstrations. Fused appearance and geometry provide local interaction evidence. Query-conditioned cross-attention experts provide adaptable prediction paths, which a local router combines at each action-chunk position. During training, an action-expert alignment loss supervises expert selection using relative forced-route prediction errors against demonstrations under fixed inputs, without role labels. Across simulation benchmarks, CoRE achieves the highest average performance among evaluated decentralized methods. Physical experiments demonstrate effective collaboration across diverse manipulation tasks and robustness to partner delays and slowdowns. Project page: https://aus.bot/research/core/.
Chinese Translation
协作操控要求机器人在交互过程中执行互补动作。我们研究单一策略的去中心化协作：每个机器人根据其视觉观察和本体感知运行相同的策略，而不依赖于任务提示、身份标签或机器人间的信息传递。挑战在于在共享参数下学习互补的团队行为，并从每个机器人的局部观察中选择适当的动作。我们提出了CoRE，它从汇集的多任务、多机器人演示中学习协作角色专家。融合的外观和几何信息提供了局部交互证据。查询条件的交叉注意力专家提供可适应的预测路径，局部路由器在每个动作块位置组合这些路径。在训练过程中，动作专家对齐损失通过相对强制路径预测误差来监督专家选择，使用固定输入下的演示，而不依赖于角色标签。在模拟基准测试中，CoRE在评估的去中心化方法中实现了最高的平均性能。物理实验展示了在多样化操控任务中的有效协作，以及对合作伙伴延迟和减速的鲁棒性。项目页面：https://aus.bot/research/core/
cs.RO / 58 / 2610.07756
StairVLA: Stage-Aware Hierarchical Action Generation for Vision-Language-Action Models
StairVLA：面向阶段的层次化动作生成用于视觉-语言-动作模型
Abstract
Vision-language-action (VLA) models increasingly rely on diffusion- or flow-matching-based action heads to generate continuous robot actions. These action heads typically process the denoising trajectory in a largely uniform manner. However, we observe that the conditioning focus naturally shifts across denoising stages: early stages combine language instructions and visual observations to establish a coarse action trajectory, whereas later stages place greater emphasis on current visual observations for action alignment. Based on this insight, we introduce StairVLA, a stage-aware hierarchical action generation framework that uses partially denoised actions as a natural interface between coarse long-horizon action generation and local refinement. A high-level VLA performs early denoising to produce a reusable long-horizon partially denoised action trajectory, while a lightweight refiner operates at a higher frequency to refine local action chunks using the latest observations. This design amortizes expensive high-level VLA computation while preserving frequent closed-loop correction. On LIBERO, our GR00T-style instantiation improves average success from 96.5% to 97.8% while reducing amortized inference latency from 115.0 ms to 44.2 ms per action chunk. More broadly, across two VLA backbones, simulation benchmarks, and real-robot tasks, StairVLA consistently reduces inference cost while maintaining strong task performance.
Chinese Translation
视觉-语言-动作（VLA）模型越来越依赖于基于扩散或流匹配的动作头来生成连续的机器人动作。这些动作头通常以相对统一的方式处理去噪轨迹。然而，我们观察到，条件焦点在去噪阶段之间自然发生转移：早期阶段结合语言指令和视觉观察来建立粗略的动作轨迹，而后期阶段则更加重视当前的视觉观察以实现动作对齐。基于这一见解，我们提出了StairVLA，一个面向阶段的层次化动作生成框架，利用部分去噪的动作作为粗略长时间动作生成与局部细化之间的自然接口。高层次的VLA执行早期去噪，以生成可重用的长时间部分去噪动作轨迹，而轻量级的细化器则以更高的频率操作，利用最新的观察来细化局部动作块。该设计在保留频繁闭环修正的同时，摊销了高层次VLA计算的高昂成本。在LIBERO上，我们的GR00T风格实例将平均成功率从96.5%提高到97.8%，同时将每个动作块的摊销推理延迟从115.0毫秒减少到44.2毫秒。更广泛地说，在两个VLA骨干网络、模拟基准和真实机器人任务中，StairVLA始终降低推理成本，同时保持强大的任务性能。
cs.RO / 59 / 2610.07772
Model-Based Geometry-Aware Generative Optimization for Constrained Locomotion Planning
基于模型的几何感知生成优化在受限运动规划中的应用
Abstract
Constrained Locomotion Planning (CLP) for quadrupeds and humanoids, where robots must satisfy collision avoidance, contact consistency, kinematic feasibility, and support constraints, is challenging under high-dimensional dynamics and highly non-convex environments. Recent Model-Based Diffusion (MBD) approaches recast trajectory optimization as posterior sampling over trajectories, using known dynamics and Monte Carlo rollouts to analytically estimate the denoising score function without demonstration learning. While constrained variants further incorporate feasibility into model-based score rollouts and show promising performance, they are still limited by (1) lacking a task-modulated active constraint geometry that shapes the score direction and reverse stochasticity, and (2) using deterministic DDPM-style reverse transport without adaptive scheduling across different generative transports. Therefore, we introduce Model-Based Geometry-Aware Generative Optimization (2GO) for constrained locomotion, which turns active constraint geometry into executable denoising operators through normal- induced metric shaping, tangent-space stochastic filtering, and CFS-based retraction. 2GO further decouples generative transport from reverse stochasticity through an adaptive diffusion and flow-like schedule. Experiments on constrained quadruped and humanoid locomotion demonstrate strong performance in discrete foothold selection and continuous posture planning, with higher success rates, fewer violations, and improved execution compatibility.
Chinese Translation
受限运动规划（CLP）对于四足动物和类人机器人而言，必须满足避碰、接触一致性、运动学可行性和支撑约束，在高维动态和高度非凸环境下具有挑战性。最近的基于模型的扩散（MBD）方法将轨迹优化重新表述为对轨迹的后验采样，利用已知的动态和蒙特卡洛回放来分析性地估计去噪评分函数，而无需示范学习。尽管受限变体进一步将可行性纳入基于模型的评分回放，并显示出良好的性能，但仍然受到以下限制：（1）缺乏任务调制的主动约束几何形状来塑造评分方向和反向随机性；（2）在不同生成传输中使用确定性的DDPM风格的反向传输，而没有自适应调度。因此，我们提出了基于模型的几何感知生成优化（2GO）用于受限运动规划，该方法通过法线诱导的度量塑形、切空间随机过滤和基于CFS的收缩，将主动约束几何形状转化为可执行的去噪算子。2GO进一步通过自适应扩散和流动式调度将生成传输与反向随机性解耦。在受限四足动物和类人机器人运动的实验中，展示了在离散支撑点选择和连续姿态规划方面的强大性能，具有更高的成功率、更少的违规行为和更好的执行兼容性。
cs.RO / 60 / 2610.07784
Multi-Robot Multi-Goal Motion Planning with Stochastic Skills
具有随机技能的多机器人多目标运动规划
Abstract
As robots are increasingly deployed in groups and share workspaces to execute real-world tasks, planning their concurrent motions around complex manipulation skills becomes essential. These skills involve continuous physical execution and may exhibit stochastic behavior, resulting in variable execution times and uncertain continuous trajectories. Existing planners either limit execution to single-robot scenarios, rely on open-loop paths, or use post-hoc scheduling that prevents dynamic coordination. In this paper, we address this gap by integrating stochastic skills into sampling-based multi-robot planning by formulating the problem as a Markov Decision Process (MDP) over a multi-modal composite roadmap. For stochastic skills, solving the MDP yields a reactive policy that allows controllable robots to dynamically adapt their motions in response to other robots' execution of manipulation skills. By resolving skill uncertainty directly at planning time, this approach avoids the pessimism of conservative baselines and unlocks robust, dynamic multi-robot coordination. Code for the planners is available at https://www.vhartmann.com/stochastic-skills.
Chinese Translation
随着机器人在群体中被越来越多地部署并共享工作空间以执行现实世界任务，围绕复杂操作技能规划它们的并发运动变得至关重要。这些技能涉及持续的物理执行，并可能表现出随机行为，导致可变的执行时间和不确定的连续轨迹。现有的规划器要么将执行限制在单机器人场景中，要么依赖于开环路径，或使用事后调度，这阻碍了动态协调。本文通过将问题表述为多模态复合路线图上的马尔可夫决策过程（MDP），填补了这一空白，将随机技能集成到基于采样的多机器人规划中。对于随机技能，解决MDP可以产生一种反应性策略，使可控机器人能够根据其他机器人执行操作技能的情况动态调整其运动。通过在规划时直接解决技能的不确定性，这种方法避免了保守基线的悲观性，并解锁了稳健的动态多机器人协调。规划器的代码可在 https://www.vhartmann.com/stochastic-skills 获取。
cs.RO / 61 / 2610.07882
CUSP: CUSUM-Governed Survival Hazard Alarms at the Perception Onset for Off-Road Navigation
CUSP：基于CUSUM的生存风险警报在越野导航感知开始时的应用
Abstract
Off-road navigation exposes a robot to potentially hazardous terrain en route. Although learning-based navigation uses safety supervision to choose which path to drive, it provides no runtime alarm when the robot following that path is heading into danger. Such an alarm must be learned from field logs, where human intervention preempts the failure and the failure itself is therefore never observed. The human judges driving unsafe early but typically intervenes only once failure is clearly near, so the intervention marks that judgment late. That earlier judgment is what a runtime alarm must detect, yet no prior intervention-supervised method has targeted it. To address this problem, we introduce CUSP (CUSUM-governed Survival model of the Perception onset), a model-agnostic runtime hazard alarm that learns this moment from intervention-terminated logs. "Cusp" is a word for the point at which one state is about to turn into another, and the moment we target is exactly such a cusp: the point at which safe driving turns unsafe in a human's judgment. We call this point the perception onset and annotate it separately from the intervention. A visual hazard head is trained on the annotated onset with a discrete-time survival objective so that driving with and without an onset both supervise the head, and a CUSUM accumulates the predicted onset risk into alarms. We evaluate CUSP at five unseen sites, two autonomous and three teleoperated, with 142 events and every method tuned to the same rate of ten false alarms per hour. CUSP detected 85 events compared to 26 for the best of nine adapted baselines, and the margin comes from hazards to which every signal in the navigation model is blind.
Chinese Translation
越野导航使机器人暴露于潜在危险的地形中。尽管基于学习的导航使用安全监督来选择行驶路径，但在机器人沿该路径行驶时并未提供运行时警报以警示即将面临的危险。这种警报必须从现场日志中学习，而在这些日志中，人为干预会预先阻止故障，因此故障本身从未被观察到。人类评判在早期就能识别出不安全的驾驶，但通常只在故障明显临近时才会进行干预，因此干预标志着这一判断的时机较晚。运行时警报必须检测到的正是这一早期判断，但之前的任何干预监督方法都未针对这一点。为了解决这个问题，我们提出了CUSP（基于CUSUM的感知开始生存模型），这是一种模型无关的运行时危险警报，能够从干预终止的日志中学习这一时刻。“Cusp”是指一个状态即将转变为另一个状态的点，而我们所关注的时刻正是这样的一个拐点：人类判断安全驾驶转变为不安全驾驶的时刻。我们将这一点称为感知开始，并将其与干预分开标注。一个视觉危险头部在带有离散时间生存目标的标注开始上进行训练，以便在有和没有开始的情况下都能监督该头部，并且CUSUM将预测的开始风险累积为警报。我们在五个未见过的地点评估了CUSP，包括两个自主和三个遥控地点，共142个事件，并且每种方法都调整到每小时十次误报的相同速率。CUSP检测到了85个事件，而九个适应基线中表现最佳的仅检测到26个，差距来自于导航模型中每个信号都无法识别的危险。
cs.RO / 62 / 2610.07891
Beyond Retargeting: Low-Latency and Robust Humanoid Whole-Body Teleoperation with Learned Atomic Motion Primitives
超越重定向：基于学习的原子运动原语实现低延迟和鲁棒的人形全身遥操作
Abstract
Humanoid whole-body teleoperation translates human motion into stable robot behavior in real time. Existing systems typically rely on online motion retargeting to bridge human--robot morphological differences, but this process adds latency and can produce physically infeasible targets. Meanwhile, diverse, noisy, and partial human-motion observations often fall outside the training distribution, potentially causing unstable robot behavior. We propose a retargeting-free policy that maps raw human motion directly to robot joint commands in a single forward pass, eliminating online kinematic adaptation. To improve robustness, we learn a codebook of full-body motion primitives that projects out-of-distribution observations onto plausible motion prototypes and recovers full-body motion from partial inputs. Experiments on a Unitree~G1 in simulation and on hardware, using virtual reality, optical mocap, text-to-motion generation, and monocular video inputs, show that our method outperforms baselines in latency and robustness.
Chinese Translation
人形全身遥操作将人类运动实时转换为稳定的机器人行为。现有系统通常依赖在线运动重定向来弥合人类与机器人之间的形态差异，但这一过程会增加延迟，并可能产生物理上不可行的目标。同时，多样化、噪声和部分的人类运动观测往往超出训练分布，可能导致机器人行为不稳定。我们提出了一种无重定向策略，能够在单次前向传递中将原始人类运动直接映射到机器人关节命令，从而消除在线运动学适应。为了提高鲁棒性，我们学习了一个全身运动原语的代码库，该代码库将超出分布的观测投影到合理的运动原型上，并从部分输入中恢复全身运动。在Unitree G1的仿真和硬件实验中，利用虚拟现实、光学动作捕捉、文本到运动生成和单目视频输入，我们的方法在延迟和鲁棒性方面均优于基线。
cs.RO / 63 / 2610.07896
From Model to Prototype: Design and Motor-Flap Propulsion Control of a Twin-Wing Metamorphic UAV
从模型到原型：双翼变形无人机的设计与电动襟翼推进控制
Abstract
This paper presents the design and control strategy for MetaMorpher, a metamorphic Unmanned Aerial Vehicle (UAV), capable of both spinning-wing hover and fixed flying-wing cruise flight. Since control of the cruise configuration is comparatively well established in the literature, this paper focuses on control of the MetaMorpher in hover mode, building on the flight dynamics model and conceptual design validated in previous work. The control algorithm is implemented using one of two phase-synchronized strategies: motor propulsion, which pulses motor thrust in synchrony with the vehicle's rotation, or flap propulsion, a novel strategy that pulses the deflection of the wing-mounted elevons. We evaluate both strategies in simulation through different flight experiments. Simulation results confirm the mathematical model and show stable reference tracking, demonstrating flap propulsion as a lightweight, decoupled alternative for hover control. Experimental testing validated the vertical-dynamics propulsion model against the physical prototype, demonstrating very good steady-state agreement across different configurations.
Chinese Translation
本文提出了MetaMorpher的设计与控制策略，这是一种变形无人机（UAV），能够实现旋转翼悬停和固定翼巡航飞行。由于巡航配置的控制在文献中相对成熟，本文重点关注MetaMorpher在悬停模式下的控制，基于之前工作中验证的飞行动力学模型和概念设计。控制算法采用两种相位同步策略之一进行实现：电动推进，该策略使电动机推力与飞行器的旋转同步脉冲，或襟翼推进，这是一种新颖的策略，通过脉冲翼上安装的襟翼的偏转来实现推进。我们通过不同的飞行实验在仿真中评估这两种策略。仿真结果确认了数学模型，并显示出稳定的参考跟踪，证明了襟翼推进作为悬停控制的一种轻量化、解耦的替代方案。实验测试验证了垂直动力学推进模型与物理原型的一致性，在不同配置下表现出非常好的稳态一致性。
cs.RO / 64 / 2610.07917
PACE: Stage-Consistent Long-Horizon Robot Manipulation via Progress-Aligned Context for Execution
PACE：通过进度对齐上下文实现阶段一致的长时间机器人操控
Abstract
Demonstration-conditioned policies provide a natural interface for specifying robot behavior, yet long-horizon manipulation remains difficult when visually similar states recur across different stages or when demonstrations and executions proceed at different speeds. We identify the resulting failure mode as stage confusion and introduce Progress-Aligned Context for Execution (PACE), a stateful method that continually reinterprets a complete demonstration according to realized execution progress. PACE compresses the demonstration into ordered multimodal prompt tokens and uses training-only dual-edge attention supervision to expose its latent stage structure. During execution, an episode-local fast-weight memory causally encodes realized action-observation transitions and modulates prompt cross-attention, producing a progress-aligned context for a unified diffusion action expert without test-time stage labels or stage-specific policies. PACE improves success from 88.9% to 94.0% on LIBERO-Gen Goal Chain, from 79.1% to 83.3% on Spatial Combination, and from 33.3% to 73.3% on the two-step Block Routing tasks. Failure analysis further indicates that structured demonstration alignment and causal execution memory jointly mitigate stage confusion.
Chinese Translation
基于示范的策略为指定机器人行为提供了自然的接口，但当视觉上相似的状态在不同阶段重复出现，或当示范与执行的速度不一致时，长时间操控仍然困难。我们将这种结果的失败模式称为阶段混淆，并引入了进度对齐上下文（Progress-Aligned Context for Execution，PACE），这是一种状态方法，能够根据实际执行进度不断重新解释完整的示范。PACE将示范压缩为有序的多模态提示标记，并使用仅在训练时的双边注意力监督来揭示其潜在的阶段结构。在执行过程中，局部快速权重记忆因果编码实际的动作-观察转变，并调节提示交叉注意力，为统一的扩散动作专家生成进度对齐的上下文，而无需测试时的阶段标签或特定阶段的策略。PACE在LIBERO-Gen目标链上的成功率从88.9%提高到94.0%，在空间组合任务上的成功率从79.1%提高到83.3%，在两步块路由任务上的成功率从33.3%提高到73.3%。失败分析进一步表明，结构化示范对齐和因果执行记忆共同缓解了阶段混淆。
cs.RO / 65 / 2610.07922
OpenWAM: An Open Framework for Composable World-Action Models
OpenWAM：一个可组合世界-动作模型的开放框架
Abstract
World-action models (WAMs) couple future prediction with robot control, yet existing systems often vary the video backbone, interaction structure, supervision, and inference procedure simultaneously, making their design choices difficult to compare. We introduce OPENWAM, an open world-action modeling framework built around a common causal robot-video foundation and configurable video-action interaction. Starting from Wan2.2-5B, we perform causal robot-video pretraining on over 10,000 hours of video, then integrate an action expert through a shared Mixture-of-Transformers architecture that supports joint, video-then-action, action-then-video, and decoupled generation. OPENWAM achieves high success rates on four LIBERO suites and real-world bimanual tasks; robot-video training with causal adaptation improves VTA success on LIBERO-Long from 68.4% to 97.8%. The same configurable architecture naturally extends to inverse and forward dynamics, allowing us to study how counterfactual transitions improve independently trained dynamics models beyond demonstrations alone. When only the video predictor is adapted to a new task, a frozen local-context inverse dynamics model trained on counterfactual data and demonstrations achieves 84.0% mean success across four held-out LIBERO-90 tasks, compared with 47.0% for a full-context inverse model and 21.5% for a local-context model trained only on demonstrations. For forward dynamics, counterfactual supervision reduces RGB prediction error by 34.5% and raises outcome identification from 21.1% to 71.3% among 16 same-state outcomes. OPENWAM provides a common testbed for comparing WAM interaction designs and for studying dynamics learning from video data beyond successful demonstrations.
Chinese Translation
世界-动作模型（WAMs）将未来预测与机器人控制相结合，但现有系统通常同时改变视频骨干、交互结构、监督和推理过程，使得其设计选择难以比较。我们介绍了OPENWAM，一个围绕共同因果机器人-视频基础和可配置视频-动作交互构建的开放世界-动作建模框架。从Wan2.2-5B开始，我们在超过10,000小时的视频上进行因果机器人-视频预训练，然后通过共享的混合变换器（Mixture-of-Transformers）架构集成一个动作专家，该架构支持联合、视频后再动作、动作后再视频和解耦生成。OPENWAM在四个LIBERO套件和现实世界的双手任务上取得了高成功率；通过因果适应进行的机器人-视频训练使LIBERO-Long上的VTA成功率从68.4%提高到97.8%。相同的可配置架构自然扩展到逆向和前向动力学，使我们能够研究反事实转变如何改善独立训练的动力学模型，超越单纯的演示。当仅将视频预测器适应于新任务时，基于反事实数据和演示训练的冻结局部上下文逆向动力学模型在四个保留的LIBERO-90任务中实现了84.0%的平均成功率，而完整上下文逆向模型为47.0%，仅基于演示训练的局部上下文模型为21.5%。在前向动力学方面，反事实监督将RGB预测误差降低了34.5%，并将16个同状态结果中的结果识别率从21.1%提高到71.3%。OPENWAM提供了一个共同的测试平台，用于比较WAM交互设计，并研究如何从视频数据中学习动力学，超越成功的演示。
cs.RO / 66 / 2610.07946
Adapting Vision-Language-Action Models to Unknown Visual Disruptions During Execution
将视觉-语言-动作模型适应于执行过程中的未知视觉干扰
Abstract
Visual disruptions can arise while a robot is executing a task, leaving a vision-language-action (VLA) policy to respond without knowing the disruption type or timing. We introduce Self-supervised Adaptation from Leftover Trajectories (SALT), which uses the leftover trajectory, the unexecuted part of the previous action chunk, as self-supervision for test-time adaptation. Because consecutive chunks overlap in time, the leftover provides a temporally aligned target for the current prediction over the same future control interval. At the onset of a visual shift, the leftover can retain a plan formed before the corruption, so updating the policy toward it anchors the adaptation across the shift (Transition Anchoring). SALT keeps the adapted policy and regenerates the current chunk, whose leftover becomes the target at the next replan, carrying the correction forward along the execution trajectory (Sequential Correction Propagation). Supervision comes entirely from the policy's own predictions, requiring no disruption annotations, expert actions, or target-domain demonstrations, and a lightweight adaptation gate calibrated only on nominal trajectories decides when updates begin. On LIBERO-10, SALT increases average success across five persistent visual corruptions from 43.9% to 53.2% with SmolVLA and from 58.7% to 66.0% with GR00T N1.7, while largely preserving nominal performance. On a real robot, it raises task progress averaged over digital and physical disruptions from 0.49 to 0.61.
Chinese Translation
在机器人执行任务时可能会出现视觉干扰，这使得视觉-语言-动作（VLA）策略在不知道干扰类型或时机的情况下进行响应。我们提出了自监督适应剩余轨迹（Self-supervised Adaptation from Leftover Trajectories，SALT），该方法利用剩余轨迹，即前一个动作块未执行部分，作为测试时适应的自监督信号。由于连续的动作块在时间上存在重叠，剩余轨迹为当前预测提供了在相同未来控制时间间隔内的时间对齐目标。在视觉变化开始时，剩余轨迹可以保留在干扰之前形成的计划，因此更新策略以朝向该目标锚定了适应过程（过渡锚定）。SALT保持适应后的策略，并重新生成当前块，其剩余部分成为下次重新规划的目标，沿着执行轨迹向前传播修正（顺序修正传播）。监督完全来自策略自身的预测，无需干扰注释、专家动作或目标领域示范，并且仅在正常轨迹上校准的轻量级适应门决定何时开始更新。在LIBERO-10上，SALT使得在五种持续视觉干扰下的平均成功率从43.9%提高到53.2%（使用SmolVLA），从58.7%提高到66.0%（使用GR00T N1.7），同时在很大程度上保持了正常性能。在真实机器人上，它使得在数字和物理干扰下的任务进展平均从0.49提高到0.61。
cs.RO / 67 / 2610.07949
Commit While Futures Agree: Consequence-Aware Adaptive Action Chunking for Robot Manipulation
在未来一致时进行承诺：面向后果的自适应动作分块机器人操作
Abstract
Action-chunking policies predict multi-step control sequences, but a fundamental question remains: how much of a predicted action chunk should be committed before replanning? Existing systems typically execute a fixed-length prefix, implicitly assuming that the same execution horizon remains trustworthy across states. Some adaptive methods estimate this horizon from the similarity or stability of predicted actions. However, different actions may lead to the same successful outcome, whereas similar actions can produce different futures, suggesting that commitment should be determined by agreement among imagined futures rather than by similarity in action space. To this end, we propose Consequence-Aware Adaptive Action Chunking (CA$^3$C), an inference-time framework built on a simple principle: commit while imagined futures agree, and replan when they diverge. Without modifying or retraining the base policy, CA$^3$C uses an action-conditioned world model to imagine the future consequences of multiple candidate action chunks under the same sampling noise. Using these imagined consequences, we formulate execution-horizon estimation as a Bayesian change-point inference problem and select the execution candidate through future consensus. Across multiple simulation benchmarks and real-world robot manipulation tasks, CA$^3$C consistently improves diverse action-chunking policies, achieving up to a 71.8% relative reduction in failure rate over the corresponding base policies.
Chinese Translation
动作分块策略预测多步控制序列，但一个基本问题仍然存在：在重新规划之前，应该承诺预测的动作分块中的多少部分？现有系统通常执行固定长度的前缀，隐含地假设在不同状态下相同的执行视野仍然是可信的。一些自适应方法根据预测动作的相似性或稳定性来估计这一视野。然而，不同的动作可能导致相同的成功结果，而相似的动作可能产生不同的未来，这表明承诺应根据想象未来之间的一致性来确定，而不是根据动作空间中的相似性。为此，我们提出了面向后果的自适应动作分块（CA$^3$C），这是一个基于简单原则的推理时框架：在想象的未来一致时进行承诺，当它们分歧时进行重新规划。在不修改或重新训练基础策略的情况下，CA$^3$C使用动作条件的世界模型来想象在相同采样噪声下多个候选动作分块的未来后果。利用这些想象的后果，我们将执行视野估计公式化为贝叶斯变更点推断问题，并通过未来共识选择执行候选项。在多个仿真基准和真实世界机器人操作任务中，CA$^3$C始终改善多样的动作分块策略，相比于相应的基础策略，失败率最高降低了71.8%。
cs.RO / 68 / 2610.07961
IronMan: Information-Constrained Video-Action Learning for Robot Manipulation
IronMan：用于机器人操作的信息约束视频动作学习
Abstract
Video Action Models (VAMs) couple visual dynamics modeling with action generation for robot manipulation. However, video representations are not naturally suited to action generation, as exposing the action policy to excessive visual detail can impair its generalization ability. Therefore, we introduce IronMan (Information-constRained videO-actioN learning for robot MANipulation), a robust video-action learning framework built on the information bottleneck principle. The core principle of this framework is to impose information constraints that suppress irrelevant visual information while preserving action-relevant dynamics cues. IronMan employs a dynamics-aware bottleneck that distills noisy, entangled one-step video features into compact world representations. Extensive simulation and real-world experiments demonstrate strong in-distribution (ID) performance and out-of-distribution (OOD) robustness while maintaining efficient inference. IronMan achieves success rates of 99.0% on LIBERO and 79.4% on RoboTwin clean2clean, outperforming all the evaluated baselines. Under OOD shifts, IronMan achieves a success rate of 79.1% on LIBERO-Plus, exceeding the strongest baseline by 10.4 percentage points. Project page: https://youngsoul0731.github.io/ironman-project-page/
Chinese Translation
视频动作模型（Video Action Models, VAMs）将视觉动态建模与机器人操作的动作生成相结合。然而，视频表示并不自然适合于动作生成，因为将动作策略暴露于过多的视觉细节中可能会损害其泛化能力。因此，我们提出了IronMan（信息约束视频动作学习用于机器人操作），这是一个基于信息瓶颈原则的强大视频动作学习框架。该框架的核心原则是施加信息约束，以抑制无关的视觉信息，同时保留与动作相关的动态线索。IronMan采用了一种动态感知瓶颈，将嘈杂、纠缠的一步视频特征提炼为紧凑的世界表示。大量的仿真和现实世界实验表明，IronMan在保持高效推理的同时，展现出强大的同分布（In-Distribution, ID）性能和异分布（Out-of-Distribution, OOD）鲁棒性。IronMan在LIBERO数据集上的成功率达到99.0%，在RoboTwin clean2clean数据集上的成功率为79.4%，超越了所有评估的基准。在OOD转变下，IronMan在LIBERO-Plus上的成功率为79.1%，比最强基准高出10.4个百分点。项目页面：https://youngsoul0731.github.io/ironman-project-page/
cs.RO / 69 / 2610.08003
Reactive Task-Oriented Robot-Human Handovers via Generative Hypothesis Selection
通过生成假设选择实现的反应式任务导向机器人-人类物品交接
Abstract
When humans hand each other objects, they incorporate both geometric and semantic information into this process. For example, passing a knife with the handle towards the recipient, rather than the blade, is both more ergonomic and safer. Recent state-of-the-art methods for task-oriented robot-human handovers have progressed from modeling object geometry to incorporating object affordances. However, they often forgo predicting the explicit, task-specific hand poses a human selects to utilize an object. Since many objects support multiple interaction modalities, e.g., a claw hammer used to strike or pull nails, this variability must be modeled to achieve robust task-oriented handovers. To tackle this, we propose a novel approach, GENESIS-Handover (GENErative HypotheSIS), which leverages VLM image generation to produce a variety of task-specific hand-object interaction hypotheses. These hypotheses are matched in real time to the observed human hand pose, enabling inference of the most suitable handover configuration. By leveraging VLMs as priors of plausible hand-object interactions, the method produces task-conditioned handover strategies for previously unseen object-task pairs. We evaluate the standalone interaction proposal module before deploying the full system on a mobile manipulator. In a user study with 12 participants across five task-object pairs, 83.3% perceived our method to have better task understanding than the previous state of the art.
Chinese Translation
当人类相互传递物品时，他们将几何信息和语义信息结合在这一过程中。例如，将刀具的手柄朝向接收者而不是刀刃传递，不仅更符合人体工程学，而且更安全。近期的任务导向机器人-人类物品交接的最先进方法已经从建模物体几何形状发展到纳入物体的可用性。然而，它们往往忽视了预测人类选择使用物体时的明确、特定任务的手势。由于许多物体支持多种交互方式，例如，爪锤可以用来敲打或拔出钉子，这种变异性必须被建模以实现稳健的任务导向交接。为了解决这个问题，我们提出了一种新颖的方法，GENESIS-Handover（生成假设选择），该方法利用VLM（视觉语言模型）图像生成技术生成多种特定任务的手-物体交互假设。这些假设实时匹配观察到的人类手势，从而推断出最合适的交接配置。通过将VLM作为合理手-物体交互的先验知识，该方法为先前未见的物体-任务对生成任务条件的交接策略。我们在将完整系统部署到移动操控器之前，评估了独立的交互提议模块。在一项包含12名参与者的用户研究中，针对五个任务-物体对，83.3%的参与者认为我们的方法在任务理解上优于之前的最先进技术。
cs.RO / 70 / 2610.08043
Vector Map Quality Metrics for Contextual Autonomous Driving Systems
用于上下文自主驾驶系统的矢量地图质量指标
Abstract
Ensuring safety in autonomous driving requires continuous map maintenance supported by reliable quality indicators. In this context, it is crucial to identify when and where map updates should be triggered, for instance through crowdsourced data, and under which conditions a new map compilation should be deployed. This paper focuses on effective metrics for assessing the quality of vector maps and guiding such decisions. We present a new metric called GOSPAM designed to measure map discrepancies in terms of location errors, existence, and completeness. Through detailed simulations on both point and polyline feature maps, we analyze its sensitivity to common map degradation such as bias, false positives, false negatives, and coordinate errors. The results demonstrate that GOSPAM offers a unified and interpretable measure that effectively captures various forms of map deviation, making it a strong candidate for map quality assessment in automotive applications.
Chinese Translation
确保自主驾驶的安全性需要持续的地图维护，并依赖于可靠的质量指标。在这一背景下，识别何时何地应触发地图更新（例如通过众包数据）以及在何种条件下应部署新的地图编制至关重要。本文重点讨论评估矢量地图质量的有效指标，并指导相关决策。我们提出了一种新的指标，称为 GOSPAM，旨在测量地图在位置错误、存在性和完整性方面的差异。通过对点和折线特征地图进行详细的模拟，我们分析了其对常见地图退化（如偏差、假阳性、假阴性和坐标错误）的敏感性。结果表明，GOSPAM 提供了一种统一且可解释的度量，能够有效捕捉各种形式的地图偏差，使其成为汽车应用中地图质量评估的有力候选者。
cs.RO / 71 / 2610.08105
Navigation with RF Cues: Embodied Perception Action under Multipath Uncertainty
基于射频线索的导航：在多路径不确定性下的具身感知与行动
Abstract
Smart factory inspection requires robots to reach connected equipment without a prior map or known target coordinates. Radio frequency (RF) signals from the target can provide directional cues to complement visual observations when occlusion or poor lighting limits target detection. However, multipath propagation can distort these cues, making it difficult to infer the target's true direction from instantaneous RF measurements. To enable navigation research under these conditions, we first construct a Habitat Sionna RT benchmark that uses detailed scene geometry and assigned material properties to generate aligned visual and RF observations in response to robot actions. Building on this benchmark, we propose an uncertainty aware multimodal navigation framework that jointly estimates target direction and its uncertainty from a history of RF, visual, and pose observations. These estimates inform action selection alongside visual context. Experiments in unseen scenes show relative improvements of 18.2% in success rate (SR) and 11.5% in success weighted by path length (SPL) over the strongest evaluated baseline.
Chinese Translation
智能工厂的检查要求机器人在没有先前地图或已知目标坐标的情况下到达连接设备。目标发出的射频（RF）信号可以提供方向线索，以补充在遮挡或光线不足时的视觉观察。然而，多路径传播可能会扭曲这些线索，使得从瞬时RF测量中推断目标的真实方向变得困难。为了在这些条件下进行导航研究，我们首先构建了一个Habitat Sionna RT基准，该基准利用详细的场景几何和指定的材料属性生成与机器人动作相对应的对齐视觉和RF观察。在此基准的基础上，我们提出了一种不确定性感知的多模态导航框架，该框架从RF、视觉和位姿观察的历史中联合估计目标方向及其不确定性。这些估计结果为行动选择提供了信息，同时考虑视觉上下文。在未见场景中的实验显示，相较于最强的评估基线，成功率（SR）提高了18.2%，路径长度加权成功率（SPL）提高了11.5%。
cs.RO / 72 / 2610.08110
Reactive Exploration of Unknown Environments for Redundant Robots using Virtual Model Control
使用虚拟模型控制的冗余机器人对未知环境的主动探索
Abstract
The exploration of confined, occluded, and partially known spaces poses significant challenges in robotic manipulation. The overall pose of the robotic arm must be carefully controlled to respect tight geometric constraints while avoiding newly discovered obstacles. We address this problem by proposing an active exploration approach for redundant robotic arms with an eye-in-hand camera configuration. Our approach navigates and acquires information in real-time based on a novel scoring method that directly selects a target voxel from the unexplored space using the robot's current state and expected information gain. To move the robot safely toward the target voxel, we utilize Virtual Model Control, which guarantees compliance and enables whole-body reactive obstacle avoidance without the need for path replanning. Simulated and real-robot experiments in both confined and open environments demonstrate the effectiveness of our approach, achieving over $90\%$ mapping coverage across all tested environments in under $120$ seconds without colliding with obstacles.
Chinese Translation
在机器人操作中，探索受限、遮挡和部分已知的空间面临着重大挑战。机器人手臂的整体姿态必须被仔细控制，以遵循严格的几何约束，同时避免新发现的障碍物。我们通过提出一种针对冗余机器人手臂的主动探索方法来解决这一问题，该方法采用了眼在手（eye-in-hand）相机配置。我们的方法基于一种新颖的评分方法，实时导航并获取信息，直接从未探索空间中选择目标体素，利用机器人的当前状态和预期信息增益。为了安全地将机器人移动到目标体素，我们采用虚拟模型控制（Virtual Model Control），该方法保证了合规性，并实现了全身反应式障碍物避免，而无需重新规划路径。在受限和开放环境中的模拟和真实机器人实验表明，我们的方法有效性显著，在所有测试环境中实现了超过90%的映射覆盖率，并且在120秒内未与障碍物发生碰撞。
cs.RO / 73 / 2610.08112
Energy-Aware Path Following: Comparative Analysis of Reinforcement Learning and NMPC for Electric Vehicles
能量感知路径跟踪：电动车辆强化学习与非线性模型预测控制的比较分析
Abstract
Path-following control strategies typically follow the bi-objective optimization dilemma: minimizing deviations from a reference path while maintaining smooth speed profiles. The latter objective is especially relevant for Electric Vehicles (EVs), since their limited driving range can be extended by recovering energy through regenerative braking, a feature that has not yet been sufficiently studied in the literature. In this work, we perform a comparative analysis of four controllers under one common Frenet frame-based kinematic vehicle model, utilizing a validated energy model (VT-CPEM) with explicit regenerative braking. Herein, we implement the following controllers: Nonlinear Model Predictive Control (NMPC), Proximal Policy Optimization (PPO), gain-scheduled Ackermann state-feedback baseline (PID-SF), and a Stanley geometric baseline. To satisfy real-time requirements, we implement the NMPC using JIT-compiled CasADi. Moreover, we train the PPO using traditional straight and S-curve tracks, after which we successfully transfer the unmodified policy to unseen tracks, including: an ISO 3888-1 lane-change, a chicane, randomly-generated parameterized-splines, and a $\pm3^\circ$ graded road. In addition, the policy transfers to a dynamic single-track vehicle model with linear tires, zero-shot with an acceptable initial performance, which was optimized after brief fine-tuning. Thereby, we demonstrate that our PPO is readily transferable to more comprehensive vehicle models. We conclude with a performance analysis of developed controllers and discuss ideas for future work.
Chinese Translation
路径跟踪控制策略通常面临双目标优化困境：在最小化与参考路径的偏差的同时保持平滑的速度曲线。后者的目标对于电动车辆（EVs）尤其重要，因为其有限的行驶范围可以通过再生制动来回收能量，而这一特性在文献中尚未得到充分研究。在本研究中，我们在一个共同的基于Frenet框架的运动学车辆模型下，对四种控制器进行了比较分析，利用了具有显式再生制动的经过验证的能量模型（VT-CPEM）。在此，我们实现了以下控制器：非线性模型预测控制（NMPC）、近端策略优化（PPO）、增益调度的Ackermann状态反馈基线（PID-SF）以及Stanley几何基线。为了满足实时要求，我们使用JIT编译的CasADi实现了NMPC。此外，我们使用传统的直线路径和S型曲线路径训练PPO，随后成功地将未修改的策略转移到未见过的轨道上，包括：ISO 3888-1变道、急弯、随机生成的参数化样条曲线以及$ ext{±}3^ ext{°}$的坡道。此外，该策略还成功转移到具有线性轮胎的动态单轨车辆模型上，零次学习下表现良好，并在短暂的微调后进行了优化。因此，我们证明了我们的PPO可以方便地转移到更复杂的车辆模型上。最后，我们对所开发的控制器进行了性能分析，并讨论了未来工作的思路。
cs.RO / 74 / 2610.08119
AutodidactWAM: Cross-Modal Self-Distillation from Generated Video to Robot Actions
自学式世界动作模型：从生成视频到机器人动作的跨模态自蒸馏
Abstract
World-action models (WAMs) such as Cosmos 3 jointly generate future video and robot actions from an observation and instruction. Adapting one such model with a lightweight LoRA fine-tune to a previously unseen robot, a Unitree G1 humanoid with five-fingered BrainCo hands, exposes a video-action asymmetry: the video renders plausible task executions, while the co-generated action is systematically mis-targeted. We evaluate closed-loop real-robot trials at three cumulative stages: pre-grasp, grasp, and pick-and-place. The native action succeeds only approximately 17%, 10%, and 7% of the time, respectively, and performs worse on held-out objects. We propose AutodidactWAM, a hand-pose estimator trained without teleoperation, followed by inverse kinematics, that runs on the model's generated video to recover action estimates. Paired with the native prediction, these recovered actions provide preferred targets for fine-tuning only the action-related layers, while the generated video is teacher-forced. We compare supervised relabeling with a rectified-flow adaptation of Diffusion-DPO. After one-time embodiment adaptation, self-distillation requires no additional task-specific teleoperation. The recovered-action gate reaches approximately 75%, 47%, and 42% pre-grasp, grasp, and pick-and-place success, compared with 17%, 10%, and 7% for the native action. A hybrid objective combining preference supervision, supervised target fitting, and Cartesian trajectory anchoring (DPO+SFT+DTW) performs best: on Oreo, the training object, it reaches 90% pre-grasp and 20% full-task success; on a held-out object, it reaches 80% and 30%. Plain Flow-DPO reaches 0% success despite 1.000 validation preference accuracy, indicating that the combination of training objectives, rather than the contrastive objective alone, drives the observed gains.
Chinese Translation
世界动作模型（World-action models, WAMs），如 Cosmos 3，可以从观察和指令中共同生成未来的视频和机器人动作。将这种模型通过轻量级的 LoRA 微调适配到一个之前未见过的机器人——Unitree G1 人形机器人，配备五指的 BrainCo 手，暴露出视频与动作之间的不对称性：视频呈现出合理的任务执行，而共同生成的动作则系统性地偏离目标。我们在三个累积阶段评估闭环真实机器人试验：预抓取、抓取和放置。原生动作的成功率分别仅为约 17%、10% 和 7%，在未见过的物体上表现更差。我们提出了自学式世界动作模型（AutodidactWAM），这是一个在没有遥操作的情况下训练的手势估计器，随后通过逆运动学运行在模型生成的视频上，以恢复动作估计。与原生预测配对，这些恢复的动作为微调仅与动作相关的层提供了优选目标，同时生成的视频被强制作为教师信号。我们将监督重标定与 Diffusion-DPO 的修正流适配进行比较。在一次性体现适配之后，自蒸馏不需要额外的任务特定遥操作。恢复的动作门的成功率分别达到约 75%、47% 和 42% 的预抓取、抓取和放置，而原生动作的成功率为 17%、10% 和 7%。结合偏好监督、监督目标拟合和笛卡尔轨迹锚定的混合目标（DPO+SFT+DTW）表现最佳：在训练物体 Oreo 上，它达到了 90% 的预抓取和 20% 的全任务成功率；在未见过的物体上，它达到了 80% 和 30%。尽管验证偏好准确率为 1.000，Plain Flow-DPO 的成功率为 0%，这表明训练目标的组合，而非单一的对比目标，驱动了观察到的提升。
cs.RO / 75 / 2610.08120
iGPC: Generative Motion Priors for Object-Aware Humanoid Interaction
iGPC：面向对象感知的人形交互生成运动先验
Abstract
Humanoid robots operating in unstructured environments must combine robust whole-body control with the ability to perceive and physically interact with surrounding objects. While large-scale human motion data provides powerful priors for natural and versatile humanoid control, effectively transferring such priors to perception-driven object interaction remains challenging. To address this bottleneck, we propose a framework that extends the recently proposed Generative Pretrained Controller (GPC) from general human motion to full-body humanoid-environment interaction. First, we adapt GPC into interaction experts conditioned on scene affordance cues and privileged state information. These experts leverage the pretrained human motion prior while learning task-specific contact behaviors, including reaching toward objects, grasping environmental supports for stabilization, and pushing movable objects. Second, we introduce a perception-driven student that retains the pretrained GPC policy and distills interaction skills from the experts using onboard sensory observations. To bridge the gap between privileged expert observations and sensory inputs, we propose two complementary training objectives that enable effective adaptation of the pretrained motion prior during distillation. Notably, our experiments across multiple whole-body interaction tasks demonstrate that large-scale generative human motion priors provide an effective foundation for learning deployable policies for humanoid interactions in contact-rich real-world environments.
Chinese Translation
在非结构化环境中操作的人形机器人必须将稳健的全身控制与感知和物理交互周围物体的能力相结合。尽管大规模的人类运动数据为自然且多样化的人形控制提供了强大的先验，但有效地将这些先验转移到基于感知的物体交互中仍然具有挑战性。为了解决这一瓶颈，我们提出了一个框架，将最近提出的生成预训练控制器（GPC）从一般人类运动扩展到全身人形与环境的交互。首先，我们将GPC调整为基于场景可供性线索和特权状态信息的交互专家。这些专家在学习特定任务的接触行为时，利用预训练的人类运动先验，包括朝向物体伸手、抓取环境支撑以实现稳定以及推动可移动物体。其次，我们引入了一种基于感知的学生，保留预训练的GPC策略，并利用机载传感器观测从专家中提炼交互技能。为了弥合特权专家观测与传感器输入之间的差距，我们提出了两个互补的训练目标，使得在提炼过程中能够有效地适应预训练的运动先验。值得注意的是，我们在多个全身交互任务中的实验表明，大规模生成的人类运动先验为学习可部署的人形交互策略提供了有效的基础，适用于接触丰富的现实环境。
cs.RO / 76 / 2610.08123
Beyond Waypoint Regression: Query-Based Cost Learning over Reachable Ego Futures for End-to-End Driving
超越航点回归：基于查询的可达自我未来成本学习用于端到端驾驶
Abstract
End-to-end planners based on waypoint regression achieve strong open-loop accuracy, but they primarily learn to mimic expert geometry and remain difficult to adapt to deployment-time safety constraints. We propose a query-based cost-learning framework that estimates bounded costs for dynamically reachable ego trajectory queries, rather than dense BEV cells or a small regressed trajectory set. Compact joint scene tokens capture coherent multimodal agent futures, while contingency-aware cost aggregation and cost-guided intra-cluster MPPI mixing convert the learned cost topology into feasible ego plans. On nuScenes, our method improves over prior cost-estimation planners such as ST-P3 and NMP, outperforms most regression baselines in collision rate, while remaining competitive in L2, and retaining an interpretable cost interface. On real-world driving logs, the proposed planner reduces collision rates compared with SparseDrive and Alpamayo without fine-tuning, while maintaining a diverse set of candidate trajectories.
Chinese Translation
基于航点回归的端到端规划器在开放循环精度上表现出色，但它们主要学习模仿专家几何形状，且在适应部署时的安全约束方面仍然困难。我们提出了一种基于查询的成本学习框架，该框架估计动态可达自我轨迹查询的有界成本，而不是密集的鸟瞰视图（BEV）单元或小规模的回归轨迹集。紧凑的联合场景标记捕捉一致的多模态代理未来，而考虑应急情况的成本聚合和成本引导的簇内混合（MPPI）将学习到的成本拓扑转换为可行的自我规划。在nuScenes数据集上，我们的方法在成本估计规划器如ST-P3和NMP的基础上有所改进，在碰撞率上超越大多数回归基线，同时在L2上保持竞争力，并保留可解释的成本接口。在真实世界的驾驶日志中，所提议的规划器在不进行微调的情况下，相比SparseDrive和Alpamayo降低了碰撞率，同时保持了一组多样的候选轨迹。
cs.RO / 77 / 2610.08133
VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models
VLA-ACL：用于高效视觉-语言-动作模型的动作一致性视觉标记剪枝
Abstract
Vision-Language-Action (VLA) models achieve strong robotic manipulation performance but incur high computational costs from processing long token sequences at every control step, limiting real-time deployment. Visual token pruning offers a direct solution, as visual patches dominate the input sequence and contain considerable redundancy. Existing approaches, however, either rely on indirect training-free heuristics, such as attention scores and motion thresholds, or require costly fine-tuning of the base VLA model. We introduce VLA-ACL (Action Consistency Learning), which learns a lightweight visual token pruning policy through action-level supervision while keeping the base VLA model entirely frozen. The training objective encourages actions produced from pruned visual contexts to remain consistent with the full-context teacher, with ground-truth actions as auxiliary supervision. This directly ties token selection to its effect on the downstream control output. Experiments on LIBERO and real-world manipulation tasks show that VLA-ACL prunes up to 87.5% of visual tokens while retaining competitive performance, reduces computation by up to 75%, and achieves a 1.5x inference speedup. These results establish a stronger performance-efficiency trade-off than existing frozen-VLA pruning methods and demonstrate the value of action-level supervision for visual token selection. Code is available at https://github.com/du-owen/VLA-ACL.
Chinese Translation
视觉-语言-动作（VLA）模型在机器人操作性能上表现出色，但在每个控制步骤中处理长标记序列所带来的高计算成本限制了实时部署。视觉标记剪枝提供了一种直接的解决方案，因为视觉补丁主导输入序列并包含相当大的冗余。然而，现有方法要么依赖于间接的无训练启发式方法，如注意力分数和运动阈值，要么需要对基础VLA模型进行昂贵的微调。我们提出了VLA-ACL（动作一致性学习），通过动作级监督学习一种轻量级的视觉标记剪枝策略，同时保持基础VLA模型完全冻结。训练目标鼓励从剪枝后的视觉上下文中产生的动作与全上下文教师保持一致，并将真实动作作为辅助监督。这直接将标记选择与其对下游控制输出的影响联系起来。在LIBERO和现实世界操作任务上的实验表明，VLA-ACL能够剪枝多达87.5%的视觉标记，同时保持竞争性能，计算量减少高达75%，并实现1.5倍的推理加速。这些结果建立了比现有的冻结VLA剪枝方法更强的性能-效率权衡，并展示了动作级监督在视觉标记选择中的价值。代码可在https://github.com/du-owen/VLA-ACL获取。
cs.RO / 78 / 2610.08150
ViDAL: A Visual Dynamics-Grounded Action Latent Space for Vision-Language-Action Models
ViDAL：一种基于视觉动态的动作潜在空间用于视觉-语言-动作模型
Abstract
Vision-Language-Action (VLA) models have become a central paradigm for robot policy learning, which predict actions in three forms: raw action chunks, discrete action tokens, or continuous action latents. However, existing action representations primarily model action trajectories, with limited consideration of the visual dynamics induced by these actions. We introduce ViDAL, a Visual Dynamics-grounded Action Latent Space that anchors continuous action latents in the future visual dynamics of the scene. Specifically, ViDAL learns action latent space by training an Action Variational Autoencoder (Action VAE) to reconstruct action chunks while aligning its latent with future scene dynamics. When integrated into downstream robot policies, the proposed Action VAE serves as a plug-in action interface compatible with multiple VLA architectures and enables optional future-video prediction as an additional capability. Empirically, ViDAL outperforms competitive baselines on LIBERO with 98.1% average success, improves a multi-task $\pi_{0.5}$ policy on RoboTwin 2.0 from 54.3% to 65.5% (Clean) and from 33.2% to 43.1% (Random) success rates over 50 dual-arm tasks, and yields 20.0% and 23.4% absolute success-rate gains on real-world single-arm Franka and dual-arm ARX robot platforms.
Chinese Translation
视觉-语言-动作（VLA）模型已成为机器人策略学习的核心范式，能够以三种形式预测动作：原始动作片段、离散动作标记或连续动作潜在变量。然而，现有的动作表示主要建模动作轨迹，对这些动作所引发的视觉动态考虑有限。我们引入了ViDAL，一种基于视觉动态的动作潜在空间，它将连续动作潜在变量锚定在场景未来的视觉动态中。具体而言，ViDAL通过训练一个动作变分自编码器（Action Variational Autoencoder，Action VAE）来重建动作片段，同时将其潜在变量与未来场景动态对齐，从而学习动作潜在空间。当集成到下游机器人策略中时，所提出的Action VAE作为一个插件动作接口，与多种VLA架构兼容，并能够选择性地进行未来视频预测作为额外功能。实证结果表明，ViDAL在LIBERO上以98.1%的平均成功率超越了竞争基线，在RoboTwin 2.0上将多任务$ ext{π}_{0.5}$策略的成功率从54.3%提高到65.5%（干净）和从33.2%提高到43.1%（随机），在50个双臂任务中实现了20.0%和23.4%的绝对成功率提升，适用于真实世界的单臂Franka和双臂ARX机器人平台。
cs.RO / 79 / 2610.08177
Nested Power Models for Multirotor Propulsion: From Aerodynamic Drag to Electrical Losses
多旋翼推进的嵌套功率模型：从空气动力学阻力到电能损耗
Abstract
Speed-only aerodynamic power models for multirotor propulsion cannot represent acceleration-dependent effects. This work develops a nested sequence of propulsion-power models that starts from aerodynamic power dissipation and progressively introduces a reversible kinetic-energy rate, torque-dependent electromechanical dissipation, and lumped speed-proportional dissipation. The models are identified using one subset of experiments and validated using the other on a motor-drive-propeller unit. Independent estimates of rotational inertia and aerodynamic drag complement predictive validation by assessing whether the models correctly attribute the measured power to reversible kinetic-energy exchange and irreversible dissipation and, within the latter, to aerodynamic and electromechanical losses. The results show that the reversible kinetic-energy rate is necessary but insufficient for accurate dynamic power prediction. Dissipation proportional to the squared motor torque provides the main additional improvement, while speed-proportional dissipation further prevents irreversible losses from being attributed to reversible kinetic-energy exchange. The resulting methodology provides a reusable and experimentally verifiable basis for developing and selecting dynamic propulsion-power models for multirotor systems.
Chinese Translation
仅基于速度的多旋翼推进空气动力学功率模型无法表示与加速度相关的效应。本研究开发了一系列嵌套的推进功率模型，起始于空气动力学功率耗散，逐步引入可逆动能速率、扭矩依赖的电机械耗散以及集中速度比例耗散。这些模型通过一组实验进行识别，并在另一组实验中进行验证，实验对象为电机驱动螺旋桨单元。独立的旋转惯量和空气动力学阻力估计补充了预测验证，通过评估模型是否正确地将测得的功率归因于可逆动能交换和不可逆耗散，并在后者中归因于空气动力学和电机械损耗。结果表明，可逆动能速率是准确动态功率预测所必需的，但不足以单独满足要求。与电机扭矩平方成比例的耗散提供了主要的额外改进，而速度比例耗散进一步防止了不可逆损耗被归因于可逆动能交换。最终的方法论为开发和选择多旋翼系统的动态推进功率模型提供了可重用且可实验验证的基础。
cs.RO / 80 / 2610.08183
Compact Robot Policies Need Fine-Grained Visual Representations
紧凑型机器人策略需要细粒度视觉表示
Abstract
Multi-task manipulation policies differ in architecture, scale, and pretrained priors all at once, so published comparisons cannot attribute performance to any single component. We argue that most of it comes from the visual representation, and that parameter scale and generative priors are largely incidental. To test this, we build CoRP (Compressed Representation Policy), a deliberately compact policy (48.9M parameters, no vision-language model and no video-generative prior) that factorizes into a representation extractor and a flow-matching action generator. It reaches 97.0% on LIBERO and 75.78%/73.36% on RoboTwin 2.0 Clean/Randomized, matching systems 40.9-163.6x larger. Holding the action generator fixed, we then vary one extractor property at a time. Pretrained initialization is decisive: a random ViT-S/14 drops to 78.1% and an ImageNet ResNet-34 to 74.5% on LIBERO. Pretraining alone is not enough, as freezing the encoder costs 19.8 points. Compression matters as much: resampling each view to 48 tokens beats passing all patch tokens (97.0% vs 83.2%), and a variational information bottleneck over those tokens is worse than a hard token budget, cutting LIBERO-Goal from 95.8% to 33.0% by suppressing the instruction-dependent token selection the policy relies on. Language conditioning contributes only where the observation leaves the goal ambiguous (LIBERO-Goal: 9.2% to 95.8%), while on RoboTwin 2.0, where observations are unambiguous, removing it slightly improves success. Therefore, we argue that a compact policy works when its representation is pretrained, task-adapted, and compressed. Project page: https://corp-policy.github.io/
Chinese Translation
多任务操作策略在架构、规模和预训练先验方面存在显著差异，因此已发布的比较无法将性能归因于任何单一组件。我们认为，这种性能大部分来自于视觉表示，而参数规模和生成先验在很大程度上是偶然的。为了验证这一点，我们构建了 CoRP（Compressed Representation Policy），这是一种故意设计的紧凑型策略（48.9M 参数，无视觉-语言模型和无视频生成先验），它分解为表示提取器和流匹配动作生成器。该策略在 LIBERO 上达到了 97.0%，在 RoboTwin 2.0 Clean/Randomized 上分别达到了 75.78% 和 73.36%，与规模为 40.9-163.6 倍的系统相匹配。在固定动作生成器的情况下，我们逐一改变一个提取器属性。预训练初始化至关重要：随机的 ViT-S/14 在 LIBERO 上降至 78.1%，而 ImageNet ResNet-34 降至 74.5%。单靠预训练是不够的，因为冻结编码器会损失 19.8 个百分点。压缩同样重要：将每个视图重新采样为 48 个标记的效果优于传递所有补丁标记（97.0% 对 83.2%），而对这些标记施加变分信息瓶颈的效果不如硬标记预算，导致 LIBERO-Goal 从 95.8% 降至 33.0%，因为这抑制了策略所依赖的指令相关标记选择。语言条件仅在观察结果使目标模糊时有所贡献（LIBERO-Goal: 9.2% 到 95.8%），而在 RoboTwin 2.0 中，观察结果明确，去除语言条件反而略微提高了成功率。因此，我们认为，当其表示经过预训练、适应任务并经过压缩时，紧凑型策略才能有效。项目页面：https://corp-policy.github.io/
cs.RO / 81 / 2610.08220
VOMMI: Collecting and Leveraging Portable Demonstrations for Mobile Manipulation
VOMMI：收集和利用便携式演示进行移动操作
Abstract
Portable mobile-manipulation demonstrations can help alleviate data scarcity for embodied intelligence, but obtaining reliable, low-cost, and robot-free motion supervision from RGB observations remains challenging. Existing approaches often rely on teleoperation or specialized devices equipped with additional sensing hardware, while directly using estimated visual odometry (VO) trajectories can introduce inconsistencies due to accumulated drift and imperfect motion supervision. We present the Visual-Odometry-Conditioned Mobile Manipulation Interface (VOMMI), a portable demonstration collection and learning framework that connects portable RGB demonstrations to vision-language-action (VLA) post-training through offline trajectory reconstruction and online visual-motion conditioning. VOMMI synchronizes body and hand views to capture navigation context and local object interactions without requiring human-robot kinematic correspondence calibration. R2-VO refines offline demonstration trajectories using sparse geometric anchors and produces causal local-motion tokens over multiple prediction horizons for online policy conditioning. An action-group residual adapter incorporates these tokens only into the base branch. Experiments use a 500-trajectory portable for each task, with 75 trajectories held out for RGB-VO evaluation, and 200 robot demonstrations as references. Our policy, post-trained only on portable demonstrations, achieves 18.2% lower base-velocity error than a policy trained with robot-collected demonstrations, while maintaining comparable end-effector translation accuracy. Offline reconstruction reduces absolute trajectory errors for the body and hand streams by 24.6% on average relative to the best evaluated baseline for each stream. The complete system improves the mean success rate by 8.3 percentage points over OpenPI 0.5 across three real-robot tasks.
Chinese Translation
便携式移动操作演示可以帮助缓解具身智能的数据稀缺问题，但从 RGB 观察中获得可靠、低成本且无机器人运动监督仍然具有挑战性。现有方法通常依赖于遥控操作或配备额外传感硬件的专用设备，而直接使用估计的视觉里程计（VO）轨迹可能会由于累积漂移和不完美的运动监督而引入不一致性。我们提出了视觉里程计条件下的移动操作接口（VOMMI），这是一个便携式演示收集和学习框架，通过离线轨迹重建和在线视觉运动条件将便携式 RGB 演示与视觉-语言-动作（VLA）后训练连接起来。VOMMI 同步身体和手部视图，以捕捉导航上下文和局部物体交互，而无需人机运动学对应校准。R2-VO 使用稀疏几何锚点精炼离线演示轨迹，并在多个预测视野上生成因果局部运动标记以进行在线策略条件。一个动作组残差适配器仅将这些标记纳入基础分支。实验中，每个任务使用 500 条轨迹的便携式数据，其中 75 条轨迹用于 RGB-VO 评估，200 条机器人演示作为参考。我们的策略仅在便携式演示上进行后训练，达到了比使用机器人收集的演示训练的策略低 18.2% 的基础速度误差，同时保持了可比的末端执行器平移精度。离线重建将身体和手部流的绝对轨迹误差平均减少了 24.6%，相对于每个流的最佳评估基线。完整系统在三个真实机器人任务中将平均成功率提高了 8.3 个百分点，相较于 OpenPI 0.5。
cs.RO / 82 / 2610.08225
UWB Meets Crazyflow: Simulating Degraded Feedback at Scale for Aerial Robotics
超宽带技术与Crazyflow的结合：大规模模拟航空机器人中的退化反馈
Abstract
In this work, we introduce Crazyflow, an accurate, differentiable simulator built on JAX. By leveraging jit compilation via XLA, Crazyflow unifies physics and control into a single differentiable computation graph, enabling massive parallelization on accelerated hardware without sacrificing modeling accuracy. This architecture achieves order-of-magnitude speedups over existing baselines, capable of training deployable reinforcement learning agents in seconds. To highlight its highly modular design, we demonstrate how easily Crazyflow can be extended by integrating a complete, high-fidelity Ultra-Wideband (UWB) and Inertial Measurement Unit (IMU) simulation pipeline coupled with a full-state Extended Kalman Filter (EKF). This capability allows for massive parallel controller evaluation under realistic, degraded state feedback with minimal impact on GPU throughput. By combining speed, accuracy, and extensibility, Crazyflow serves as a foundational tool for the next generation of aerial robotics research.
Chinese Translation
在本研究中，我们介绍了Crazyflow，这是一种基于JAX构建的准确且可微分的模拟器。通过利用XLA的即时编译，Crazyflow将物理与控制统一为一个单一的可微分计算图，从而在加速硬件上实现大规模并行化，而不牺牲建模精度。该架构在现有基准上实现了数量级的速度提升，能够在几秒钟内训练可部署的强化学习代理。为了突出其高度模块化的设计，我们展示了如何通过集成完整的高保真超宽带（Ultra-Wideband, UWB）和惯性测量单元（Inertial Measurement Unit, IMU）模拟管道，以及全状态扩展卡尔曼滤波器（Extended Kalman Filter, EKF），轻松扩展Crazyflow。这一能力允许在现实的、退化的状态反馈下进行大规模的控制器评估，对GPU吞吐量的影响最小。通过结合速度、精度和可扩展性，Crazyflow成为下一代航空机器人研究的基础工具。
cs.RO / 83 / 2610.08259
HexaGripper: A Single-Actuator, Winch-Deployed Gripper for Autonomous Aerial Parcel Collection
HexaGripper：一种单驱动器、绞盘部署的抓取器，用于自主空中包裹收集
Abstract
Autonomous aerial parcel collection remains constrained by package-transfer operations that require manual loading, landing, or dedicated ground infrastructure. This letter presents HexaGripper, a single-actuator, winch-deployed six-plate gripper for autonomous collection of cuboid parcels. The proposed mechanism provides synchronized grasping with a wide capture region and tolerance to positional misalignment during pickup. The system integrates vision-based alignment, range sensing, winch deployment, and state-based control to enable autonomous collection while maintaining UAV separation from the pickup surface. Experimental evaluation covered static grasping, capture-workspace assessment, and end-to-end outdoor aerial collection. Static experiments achieved 100% pickup success (15/15 trials) across three parcel geometries, while workspace experiments achieved successful pickup at all 28 tested positions up to a 150 mm radial offset. End-to-end outdoor aerial collection achieved an 80% success rate (4/5 trials). These results demonstrate the feasibility of mechanically synchronized, winch-deployed grasping for landing-free autonomous aerial parcel collection.
Chinese Translation
自主空中包裹收集受到需要手动装载、着陆或专用地面基础设施的包裹转移操作的限制。本文提出了HexaGripper，一种单驱动器、绞盘部署的六片抓取器，用于自主收集立方体包裹。该机制提供了同步抓取，具有广泛的捕获区域，并能容忍在拾取过程中的位置偏差。该系统集成了基于视觉的对准、距离传感、绞盘部署和基于状态的控制，以实现自主收集，同时保持无人机与拾取表面的分离。实验评估涵盖了静态抓取、捕获工作空间评估和端到端户外空中收集。静态实验在三种包裹几何形状下实现了100%的拾取成功率（15/15次试验），而工作空间实验在所有28个测试位置上成功拾取，最大径向偏移达到150毫米。端到端户外空中收集实现了80%的成功率（4/5次试验）。这些结果证明了机械同步、绞盘部署抓取在无需着陆的自主空中包裹收集中的可行性。
cs.RO / 84 / 2610.08297
Mitigating Concept Drift in QoS Prediction for Teleoperation of Autonomous Vehicles Using Historic Data
利用历史数据减缓自主车辆远程操作中的QoS预测概念漂移
Abstract
Teleoperation serves as the fallback solution to autonomous driving but reliable functions of the teleoperation require a certain amount of mobile network resources, which cannot be guaranteed at all times. Therefore, predictive quality of service (pQoS) is introduced as a concept to increase the resilience of the teleoperation. In this paper, based on a data measurement campaign, we propose a prediction framework to prediction two important network KPIs of teleoperation: uplink data-rate and round-trip latency. Furthermore, we introduce a method to alleviate the performance degradation of machine-learning-based prediction models on previously unseen data due to concept drift by incorporating historic data into the prediction pipeline. Additionally, we introduce the metric of critical scenario detection to evaluate the prediction performance specifically for teleoperation.
Chinese Translation
远程操作作为自主驾驶的后备解决方案，但远程操作的可靠功能需要一定量的移动网络资源，而这些资源并不能始终得到保证。因此，引入预测服务质量（predictive quality of service，pQoS）作为一个概念，以增强远程操作的弹性。在本文中，基于数据测量活动，我们提出了一个预测框架，以预测远程操作的两个重要网络关键性能指标（KPI）：上行数据速率和往返延迟。此外，我们提出了一种方法，通过将历史数据纳入预测流程，缓解由于概念漂移导致的机器学习预测模型在以前未见数据上的性能下降。此外，我们引入了关键场景检测的指标，以专门评估远程操作的预测性能。
cs.RO / 85 / 2610.08306
Sensor-Layout-Agnostic Navigation via Geometric Observation Canonicalization
传感器布局无关的导航通过几何观测规范化
Abstract
Existing visual navigation policies are inherently bound to fixed camera configurations, creating a fundamental barrier to zero-shot deployment across heterogeneous robot sensor layouts. To overcome this limitation, we present an embodiment-informed navigation policy capable of generalizing across diverse depth sensor configurations on a specific aerial platform. Instead of implicitly learning spatial alignments, our approach explicitly unprojects depth measurements from arbitrary depth sensor payloads, varying in sensor count, mounting extrinsics, and intrinsics, into a shared robot-centric frame, stitching them into a unified spherical range image and a binary validity mask. This mask allows the downstream policy to explicitly distinguish covered space from unobserved blind spots. Trained via reinforcement learning with aggressive camera randomization, our policy generalizes zero-shot to unseen layouts featuring up to seven cameras, scaling success rates from 78% to 95% as total spatial sensing coverage increases. Finally, real-world flight trials on a physical quadrotor, conducted in an obstacle-filled corridor and an outdoor forest, validate the policy's zero-shot transfer across camera configurations and its resilience to sudden online sensor dropouts.
Chinese Translation
现有的视觉导航策略固有地依赖于固定的相机配置，这在异构机器人传感器布局之间的零-shot 部署中造成了根本性的障碍。为了解决这一限制，我们提出了一种基于体现的导航策略，能够在特定的空中平台上跨越多样的深度传感器配置进行泛化。我们的方案不是隐式地学习空间对齐，而是显式地将来自任意深度传感器载荷的深度测量（传感器数量、安装外参和内参各不相同）反投影到一个共享的机器人中心框架中，将其拼接成一个统一的球形范围图像和一个二进制有效性掩码。该掩码使下游策略能够明确区分已覆盖的空间与未观察到的盲区。通过强化学习与激进的相机随机化进行训练，我们的策略在未见布局中实现了零-shot 泛化，成功率从 78% 提升至 95%，随着空间感知覆盖范围的增加而提升。最后，在一个充满障碍的走廊和户外森林中进行的物理四旋翼的实际飞行试验验证了该策略在相机配置之间的零-shot 转移能力及其对突发在线传感器掉线的韧性。
cs.RO / 86 / 2610.08320
Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating
类人机器人视野：通过并行训练、动态起始和奖励门控扩展全身运动操控的任务视野
Abstract
Cluttered indoor environments, where large and heavy objects are scattered across diverse surfaces, require humanoid robots to sequentially navigate, grasp, transport, and accurately place each item at its target location within a single uninterrupted episode. This long-horizon, whole-body loco-manipulation task remains a significant challenge for current methods. Previous approaches often suffer from two main issues: easy-reward bias, where training overemphasizes early transport stages at the expense of later ones, and catastrophic forgetting, where focusing on later stages leads to a decline in earlier-stage performance. In this work, we introduce Humanoid Horizon, a unified policy framework designed to overcome these limitations through three interrelated mechanisms. The Parallel Training Strategy organizes $N$ scenes into $S$ concurrent stage streams governed by a shared policy, ensuring all transport stages receive continuous gradient updates and removing the bottleneck of sequential optimization. The Dynamic Starting Mechanism updates each environment's initial state with terminal states from upstream rollouts, gradually broadening transition coverage and enhancing robustness at stage boundaries. Reward Gating sets the reward to zero for the rest of the episode in later-stage streams when the immediately preceding object is displaced beyond a set threshold, so the shared policy learns not to disturb a just-placed object and earlier placements are preserved throughout the episode. Collectively, these strategies achieve per-stage success rates exceeding 80\% on the two-object LHM-Humanoid benchmark (350 training scenes, 66 held-out scenes). As the number of sequentially transported objects grows beyond two, success declines with the horizon, but the degradation is graceful relative to the sharp drop seen in all baselines.
Chinese Translation
在杂乱的室内环境中，大型重物散落在不同的表面上，要求类人机器人在单个不间断的过程中顺序导航、抓取、运输并准确放置每个物品到其目标位置。这一长视野的全身运动操控任务对当前方法仍然是一个重大挑战。以往的方法常常面临两个主要问题：易奖励偏差，即训练过于强调早期运输阶段而忽视后期阶段；以及灾难性遗忘，即专注于后期阶段导致早期阶段表现下降。在本研究中，我们提出了类人机器人视野（Humanoid Horizon），这是一个统一的策略框架，旨在通过三种相互关联的机制克服这些限制。并行训练策略将 $N$ 个场景组织成 $S$ 个由共享策略管理的并发阶段流，确保所有运输阶段获得持续的梯度更新，消除顺序优化的瓶颈。动态起始机制使用来自上游回放的终端状态更新每个环境的初始状态，逐步扩大过渡覆盖范围并增强阶段边界的鲁棒性。奖励门控机制在后期阶段流中，当紧接着的物体被移动超过设定阈值时，将该集的奖励设为零，从而使共享策略学习不去干扰刚放置的物体，并在整个过程中保留早期的放置。总体而言，这些策略在两物体的LHM-Humanoid基准测试中实现了超过80 ext{%}的每阶段成功率（350个训练场景，66个保留场景）。随着顺序运输物体数量超过两个，成功率随着任务视野的增加而下降，但相较于所有基线所见的急剧下降，退化表现得更加平缓。
cs.RO / 87 / 2610.08324
Communication-Free Obstacle Localization from Aggregate Wrench Measurements in Leader--Follower Cooperative Transport
基于聚合扭矩测量的无通信障碍物定位在领导-跟随合作运输中的应用
Abstract
We consider obstacle localization for a team of robots cooperatively transporting a rigid payload without explicit inter-robot communication. A leader robot directs the payload's motion, while follower robots assist and react to locally detected obstacles. The leader measures the followers' aggregate wrench, i.e., the combined force and torque they exert on the payload, but cannot directly distinguish their individual reactions. We design a follower control law that allows the leader to recover obstacle locations from these measurements. Each follower resists motion toward nearby obstacles, resulting in a piecewise-linear relationship between the payload's translational and angular velocity and the aggregate wrench. Changes between adjacent linear regions reveal an obstacle's bearing and distance and identify the responding follower. We give sufficient conditions for exact recovery at a fixed payload configuration and develop an adaptive probing procedure in which the leader applies translational and rotational inputs to the payload to obtain the required measurements. We demonstrate the performance of the proposed method in simulations.
Chinese Translation
我们考虑在没有明确的机器人间通信的情况下，团队机器人合作运输刚性载荷时的障碍物定位问题。领导机器人引导载荷的运动，而跟随机器人则协助并对局部检测到的障碍物作出反应。领导机器人测量跟随机器人的聚合扭矩，即它们对载荷施加的合力和扭矩，但无法直接区分它们的个体反应。我们设计了一种跟随控制律，使领导机器人能够从这些测量中恢复障碍物的位置。每个跟随机器人抵抗朝向附近障碍物的运动，导致载荷的平移速度和角速度与聚合扭矩之间呈分段线性关系。相邻线性区域之间的变化揭示了障碍物的方位和距离，并识别出响应的跟随机器人。我们给出了在固定载荷配置下精确恢复的充分条件，并开发了一种自适应探测程序，其中领导机器人对载荷施加平移和旋转输入以获得所需的测量。我们在仿真中展示了所提方法的性能。
cs.RO / 88 / 2610.08332
SC3BF: Shifted Collision Cone Control Barrier Function for Dynamic Obstacle Avoidance
SC3BF：用于动态障碍物避让的偏移碰撞锥控制屏障函数
Abstract
The collision cone used by velocity-space control barrier functions is conservative: it rejects every relative velocity aimed into an obstacle, however slow. We propose the \emph{shifted collision-cone CBF} (SC3BF), which adds a state-dependent \emph{allowance} to the cone condition, so the robot may approach the obstacle at a rate that grows with distance and with its own speed. SC3BF is enforced by an ordinary quadratic program, and its safe set is forward invariant under bounded inputs without a minimum forward speed or a clearance margin. We prove that a nonzero allowance preserving safety always exists, and derive one in closed form. Against three velocity-space baselines on a kinematic bicycle among up to $100$ moving obstacles, SC3BF reaches the goal more often and modifies the nominal input less than half as much.
Chinese Translation
速度空间控制屏障函数使用的碰撞锥是保守的：它拒绝每一个指向障碍物的相对速度，无论其多么缓慢。我们提出了 extit{偏移碰撞锥控制屏障函数}（SC3BF），该函数在锥条件中添加了一个状态依赖的 extit{允许值}，使得机器人可以以与距离和自身速度相关的速率接近障碍物。SC3BF通过一个普通的二次规划来执行，其安全集在有界输入下是前向不变的，不需要最小前进速度或间隙边际。我们证明了保持安全的非零允许值总是存在，并推导出了一个封闭形式的表达。在一个运动的自行车模型中，针对多达100个移动障碍物的三种速度空间基线，SC3BF更频繁地达到目标，并且对名义输入的修改不到一半。
cs.RO / 89 / 2610.08344
Event-Driven Proactive Robot Assistance through Vision-Language Reasoning
基于视觉-语言推理的事件驱动主动机器人辅助
Abstract
Assistance in collaborative manipulation is often initiated by user instructions, making high-level reasoning request-driven. In fluent human teamwork, however, partners often infer the next helpful step from the observed outcome of an action rather than waiting for instructions. Motivated by this, we investigate an event-driven formulation of proactive assistance, where human--object interaction outcomes initiate assistive reasoning without user-provided task specifications at inference time. To this end, we propose an event-driven framework that monitors workspace state changes with an event monitor and, upon event completion, extracts stabilized pre/post snapshots that characterize the resulting state transition. A frozen pretrained Vision-Language Model (VLM) then uses its semantic priors to infer the task context, decide whether assistance is appropriate, and, when needed, generate a sequence of assistive actions from the observed transition. To make outputs executable and verifiable, we restrict actions to a set of action primitives and reference objects via integer IDs.We evaluate the same framework across three distinct real world tabletop collaboration tasks without task-specific training or fine-tuning. The event-driven framework achieves performance comparable to variants given user instructions.
Chinese Translation
协作操作中的辅助通常由用户指令发起，使得高层次推理依赖于请求。然而，在流畅的人类团队合作中，伙伴们往往从观察到的行动结果中推断出下一个有帮助的步骤，而不是等待指令。基于此动机，我们研究了一种事件驱动的主动辅助形式，其中人机交互的结果在推理时无需用户提供任务规范就能启动辅助推理。为此，我们提出了一种事件驱动框架，该框架通过事件监控器监测工作空间状态变化，并在事件完成后提取稳定的前后快照，以表征结果状态转变。然后，冻结的预训练视觉-语言模型（VLM）利用其语义先验推断任务上下文，决定是否适合提供辅助，并在需要时根据观察到的转变生成一系列辅助动作。为了使输出可执行和可验证，我们将动作限制为一组动作原语和通过整数ID引用的对象。我们在三个不同的真实世界桌面协作任务中评估了相同的框架，而无需特定于任务的训练或微调。该事件驱动框架的性能与给定用户指令的变体相当。
cs.RO / 90 / 2610.08350
How Much Planning Is Enough? Reducing Search and Computation in World-Model Planning
规划多少才足够？减少世界模型规划中的搜索和计算
Abstract
Visual world models enable goal-directed control through decision-time action search, but their deployment efficiency is often limited by conservatively large planning budgets. We show that competitive task performance can be achieved without agreement with the Full-budget action, that sufficient budgets vary across model--task pairs, and that iterative planners repeatedly encode solve-invariant context. To address these inefficiencies, we propose {SufficientPlan}, a simple deployment framework that requires no modification to pretrained world models or planner updates. Its {Paired Sequential Budget Certification (PSBC)} component uses paired closed-loop evidence to search for and certify a reduced model--task-specific budget within a predefined Full-performance tolerance. Its {Static-Context Reuse (SCR)} component caches observation and goal representations across search iterations while preserving candidate-dependent planning and selected actions. Experiments across multiple world-model backbones and visual-control tasks show that SufficientPlan substantially reduces search budgets and planning latency while maintaining competitive control performance.
Chinese Translation
视觉世界模型通过决策时的行动搜索实现目标导向控制，但其部署效率常常受到保守的大规划预算的限制。我们展示了在不与全预算行动达成一致的情况下，仍然可以实现竞争性的任务表现，且足够的预算在模型-任务对中是不同的，迭代规划者反复编码解决不变的上下文。为了解决这些低效问题，我们提出了 {SufficientPlan}，这是一个简单的部署框架，无需对预训练的世界模型或规划者进行修改。其 {Paired Sequential Budget Certification (PSBC)} 组件利用配对的闭环证据来搜索和认证在预定义的全性能容忍度内的减少模型-任务特定预算。其 {Static-Context Reuse (SCR)} 组件在搜索迭代中缓存观察和目标表示，同时保留候选依赖的规划和选择的行动。针对多个世界模型骨干和视觉控制任务的实验表明，SufficientPlan 在保持竞争性控制性能的同时，显著减少了搜索预算和规划延迟。
cs.RO / 91 / 2610.08381
From Legs to Wheels: Embodiment-Aware Human Motion Retargeting for Mobile-Base Humanoids
从腿到轮：面向具身意识的人类动作重定向用于移动底盘类人机器人
Abstract
Human video offers a scalable source of robot demonstrations, yet most human-to-humanoid retargeting methods assume a legged robot with human-like kinematics. This assumption does not hold for mobile-base humanoids equipped with a wheeled base, vertical lift, and two arms. Human walking must be expressed through base motion, while torso bending may require coordinated lift and arm motion. We address this mismatch with a task-conditioned framework that assigns reconstructed human motion to base, lift, and arm responsibilities before robot-specific realization. The allocator preserves the human-derived path, stabilizes heading, separates turn and translation when needed, retimes commands to satisfy base limits, and repairs lift and arm trajectories. A deployment adapter then converts the reference to 50 Hz commands using stationary-base detection, deadband and slew-rate filtering, time-consistent playback scaling, and separate linear and angular gains. We evaluate the resulting references with human-derived task-space comparisons, policy-free simulation replay, and a qualitative execution on a physical robot.
Chinese Translation
人类视频提供了一个可扩展的机器人演示来源，但大多数人类到类人机器人的重定向方法假设机器人具有类似人类的运动学特征。这一假设并不适用于配备轮式底盘、垂直升降装置和双臂的移动底盘类人机器人。人类的行走必须通过底盘运动来表达，而躯干的弯曲可能需要协调的升降和臂部运动。我们通过一个任务条件框架来解决这一不匹配，该框架在机器人特定实现之前，将重建的人类运动分配给底盘、升降和臂部的责任。该分配器保留了源自人类的路径，稳定了航向，在需要时分离转向和位移，重新调整命令以满足底盘限制，并修正升降和臂部轨迹。然后，部署适配器使用静态底盘检测、死区和速率过滤、时间一致的播放缩放以及独立的线性和角速度增益，将参考转换为50 Hz的命令。我们通过人类源任务空间比较、无策略仿真回放以及在物理机器人上的定性执行来评估所得到的参考。
cs.RO / 92 / 2610.08421
Post-Grasp Kinematic Repair for Robotic Insertion via Object-in-Gripper Reorientation
通过物体-夹持器重新定向实现机器人插入的后抓取运动学修复
Abstract
A stable grasp does not guarantee kinematically feasible robotic insertion because the object-in-gripper transform may force the robot towards singularities or joint limits along the prescribed insertion path. We study post-grasp kinematic feasibility repair through object-in-gripper reorientation. Given an achieved grasp and a fixed insertion path, we seek a small reorientation that restores kinematic feasibility. Sequential IK can miss such candidates by following an unfavorable joint-space path, while the nonsmooth feasibility landscape makes the search computationally expensive. We evaluate candidates using a branch-aware IK graph that maximizes the minimum feasibility margin over the discretized insertion path and use a learned task-conditioned prior to improve query ordering. The selected reorientation is executed through tactile-based extrinsic manipulation. In UR5e simulations, the planner without learned ranking reduces mean reorientation over successful trials by 51.5% compared with grid-based sequential IK. Adding learned ranking reduces this planner's mean planning time by an additional 51.1%. Real-robot experiments validate the complete pipeline.
Chinese Translation
稳定的抓取并不保证机器人插入的运动学可行性，因为物体-夹持器变换可能会使机器人沿着规定的插入路径朝向奇异点或关节极限。我们研究了通过物体-夹持器重新定向进行后抓取运动学可行性修复。在给定的抓取状态和固定的插入路径下，我们寻求一种小幅度的重新定向，以恢复运动学可行性。顺序逆运动学（Sequential IK）可能会通过沿不利的关节空间路径进行搜索而错过这样的候选方案，而不光滑的可行性景观使得搜索计算开销较大。我们使用一个关注分支的逆运动学图来评估候选方案，该图在离散化的插入路径上最大化最小可行性边际，并利用学习的任务条件先验来改善查询顺序。所选择的重新定向通过基于触觉的外部操控执行。在UR5e仿真中，与基于网格的顺序逆运动学相比，没有学习排名的规划器在成功试验中将平均重新定向时间减少了51.5%。添加学习排名进一步将该规划器的平均规划时间减少了51.1%。真实机器人实验验证了整个流程的有效性。
cs.RO / 93 / 2610.08425
MIM-VLA: Learning Physical Interaction Representations from Gripper Motor Feedback
MIM-VLA：从夹持器电机反馈中学习物理交互表示
Abstract
Vision-language-action (VLA) policies infer grasp actions primarily from visual observations and robot state, but do not explicitly represent the physical response observed after contact. We present MIM-VLA, a motor-feedback-based architecture that encodes recent gripper current, position, velocity, and signal validity as a 128-dimensional interaction token. A motor-only Motor Interaction Module (MIM) is pretrained with human-reviewed contact and interaction-phase labels and then conditions only the gripper-action pathway of SmolVLA; arm actions and the position-control interface remain unchanged. The same token supports the MEM selector VLM that compares candidate interactions and produces evidence-conditioned selections and explanations. We evaluate MIM-VLA in three real-world settings: comparing the interaction resistance of visually different objects, disambiguating visually similar real and replica objects through active probing, and gently grasping fragile objects, including held-out instances. Across 13 object pairs, MIM-VLA selects the higher-resistance object in 75.0% of trials, compared with 48.8% for the SmolVLA baseline. For the evaluated tasks, the approach uses motor feedback already available from the gripper and does not require an additional tactile array, force-torque sensor, calibrated force estimate, or direct current control.
Chinese Translation
视觉-语言-动作（VLA）策略主要通过视觉观察和机器人状态推断抓取动作，但并未明确表示接触后观察到的物理响应。我们提出了MIM-VLA，一种基于电机反馈的架构，它将最近的夹持器电流、位置、速度和信号有效性编码为128维的交互标记。仅使用电机的运动交互模块（MIM）经过人类审核的接触和交互阶段标签进行预训练，然后仅调节SmolVLA的夹持器动作路径；手臂动作和位置控制接口保持不变。相同的标记支持MEM选择器VLM，该选择器比较候选交互并生成基于证据的选择和解释。我们在三个真实世界的场景中评估MIM-VLA：比较视觉上不同物体的交互阻力，通过主动探测消歧视觉上相似的真实物体和复制物体，以及轻柔抓取脆弱物体，包括未见实例。在13对物体中，MIM-VLA在75.0%的试验中选择了阻力更高的物体，而SmolVLA基线的选择率为48.8%。对于评估的任务，该方法利用夹持器已经提供的电机反馈，不需要额外的触觉阵列、力-扭矩传感器、校准的力估计或直接电流控制。
cs.RO / 94 / 2610.08441
Safe Multi-Robot Collaborative Transport Using Density Functions
基于密度函数的安全多机器人协作运输
Abstract
This paper presents a hierarchical density-based model predictive control framework for safe collaborative manipulation by multiple quadrupedal robots. The framework enables a team of robots to push a shared object to a desired pose using only the initial and goal poses, without requiring a precomputed reference trajectory. A centralized box-level MPC optimizes contact forces while enforcing a control-density constraint for goal convergence and obstacle avoidance. Each robot then solves its own distributed robot-level whole-body MPC, under a stated shared-information assumption, to track its moving contact location while accounting for static obstacles and the time-varying positions of neighboring robots. The approach is evaluated in MuJoCo using whole-body contact dynamics for two and three Unitree Go2 quadrupeds collaboratively pushing rigid objects through narrow passages. Comparisons with matched Control Barrier Function and RRT* based tracking baselines demonstrate the effectiveness of the proposed density-based formulation for push-only, force- and torque-coupled manipulation tasks. Implementation videos are available at https://jaggu2606.github.io/go2-density-mpc-pushing/
Chinese Translation
本文提出了一种层次化的基于密度的模型预测控制框架，用于多四足机器人之间的安全协作操作。该框架使得一组机器人能够仅通过初始位置和目标位置，将共享物体推送至期望姿态，而无需预先计算参考轨迹。集中式箱级模型预测控制（MPC）在优化接触力的同时，强制执行控制密度约束，以实现目标收敛和障碍物规避。每个机器人在共享信息假设下解决自己的分布式机器人级全身MPC，以跟踪其移动接触位置，同时考虑静态障碍物和邻近机器人的时变位置。该方法在MuJoCo中进行了评估，使用全身接触动力学模拟两个和三个Unitree Go2四足机器人协作推送刚性物体通过狭窄通道。与匹配的控制障碍函数和基于RRT*的跟踪基线的比较，展示了所提出的基于密度的公式在仅推送、力和扭矩耦合操作任务中的有效性。实施视频可在https://jaggu2606.github.io/go2-density-mpc-pushing/获取。
cs.RO / 95 / 2610.08444
ActTune: Action-Aware Precision and GPU Operating-Point Adaptation for Energy-Efficient Vision-Language-Action Inference
ActTune：面向动作的精度和GPU操作点适应以实现节能的视觉-语言-动作推理
Abstract
Vision-language-action (VLA) policies repeatedly invoke inference to control robots, making graphics processing unit (GPU) energy a recurring cost of task execution. Reducing energy per inference call, however, may not reduce energy per successful task if numerical errors increase failures or slower inference prolongs execution. We therefore target GPU energy per successful task while preserving task success and keeping the inference-latency increase within 10\%. Our approach builds on two observations: quantization sensitivity varies across action classes, model layers, and weights versus activations; and numerical precision changes the workload, shifting favorable GPU operating points. We introduce ActTune, an action-aware framework that connects layer-wise precision allocation with workload-dependent GPU operating-point selection over requested frequency--power-cap pairs. A lightweight decision tree learns its splits and leaf precision configurations directly from configuration action errors, then selects precision before each policy call. The controller forecasts the next workload and applies the selected GPU operating point asynchronously using a lookup table calibrated under a latency budget. A shared resident quantized weight bank enables configuration switching without weight reconstruction or additional policy evaluations. On LIBERO, a benchmark for lifelong robot learning, ActTune improves mean task success by up to 2.3\% relative to state of the art. Relative to the original BF16 implementations, it delivers up to $2.02\times$ faster inference and, with GPU operating-point adaptation, reduces energy per successful task by up to 76.8\%.
Chinese Translation
视觉-语言-动作（VLA）策略反复调用推理以控制机器人，使得图形处理单元（GPU）能耗成为任务执行的反复成本。然而，减少每次推理调用的能耗可能并不会降低每个成功任务的能耗，如果数值误差导致失败增加或较慢的推理延长执行时间。因此，我们的目标是降低每个成功任务的GPU能耗，同时保持任务成功率，并将推理延迟的增加控制在10\%以内。我们的方法基于两个观察：量化敏感性在不同动作类别、模型层和权重与激活之间有所不同；数值精度改变了工作负载，影响GPU操作点的选择。我们提出了ActTune，一个面向动作的框架，将层级精度分配与基于工作负载的GPU操作点选择连接起来，针对请求的频率-功率上限对。一个轻量级决策树直接从配置动作误差中学习其分裂和叶子精度配置，然后在每次策略调用之前选择精度。控制器预测下一个工作负载，并使用在延迟预算下校准的查找表异步应用所选的GPU操作点。一个共享的驻留量化权重库使得配置切换无需权重重构或额外的策略评估。在终身机器人学习基准LIBERO上，ActTune相较于现有技术提高了平均任务成功率最多2.3\%。相较于原始的BF16实现，它提供了最多$2.02 imes$的推理加速，并通过GPU操作点适应将每个成功任务的能耗降低了最多76.8\。
cs.RO / 96 / 2610.08450
Behavioral Safety Assessment towards Large-scale Deployment of Autonomous Vehicles, Part I: Methodology
面向大规模部署自主车辆的行为安全评估，第一部分：方法论
Abstract
Autonomous vehicles (AVs) have significantly advanced in real-world deployment in recent years, yet safety continues to be a critical barrier to widespread adoption. Traditional functional safety approaches, which primarily verify the reliability, robustness, and adequacy of AV hardware and software systems from a vehicle-centric perspective, do not sufficiently address the AV's broader interactions and behavioral impact on the surrounding traffic environment. To overcome this limitation, we propose a paradigm shift toward behavioral safety, a comprehensive approach focused on evaluating AV responses and interactions within the traffic environment. To systematically assess behavioral safety, we introduce a third-party AV safety assessment framework comprising two complementary evaluation components: the Behavioral Competency Test and the Driving Intelligence Test. The Behavioral Competency Test evaluates the AV's reactive behaviors under controlled scenarios, ensuring basic behavioral competency. In contrast, the Driving Intelligence Test assesses the AV's interactive behaviors within naturalistic traffic conditions, quantifying the frequency of safety-critical events to deliver statistically meaningful safety metrics before large-scale deployment. In Part II of this study, an open-source Level 4 Automated Driving System (ADS) is tested to demonstrate the effectiveness of the proposed method.
Chinese Translation
近年来，自主车辆（AVs）在实际部署方面取得了显著进展，但安全性仍然是广泛采用的关键障碍。传统的功能安全方法主要从以车辆为中心的角度验证AV硬件和软件系统的可靠性、稳健性和充分性，但并未充分考虑AV与周围交通环境的广泛互动及其行为影响。为克服这一局限性，我们提出了一种向行为安全转变的范式，这是一种全面的方法，专注于评估AV在交通环境中的反应和互动。为了系统地评估行为安全，我们引入了一个第三方AV安全评估框架，该框架包括两个互补的评估组件：行为能力测试和驾驶智能测试。行为能力测试在受控场景下评估AV的反应行为，以确保基本的行为能力。相对而言，驾驶智能测试在自然交通条件下评估AV的互动行为，量化安全关键事件的发生频率，以在大规模部署之前提供具有统计意义的安全指标。在本研究的第二部分中，将测试一个开源的四级自动驾驶系统（ADS），以展示所提方法的有效性。
cs.RO / 97 / 2610.08458
Behavioral Safety Assessment towards Large-scale Deployment of Autonomous Vehicles, Part II: Assessment Results
面向大规模部署自动驾驶车辆的行为安全评估，第二部分：评估结果
Abstract
Third-party evaluations of autonomous vehicle (AV) safety can play a vital role in improving public acceptance, building consumer confidence, and establishing effective safety standards. In Part I of this study, we propose a dedicated third-party testing initiative for systematically evaluating AV behavioral safety. In this paper, we validate our proposed framework using Autoware.Universe, an open-source Level 4 Automated Driving System (ADS), tested both in simulated environments and on the physical test track at the University of Michigan's Mcity Testing Facility. The results indicate that Autoware.Universe possesses 6 out of 14 behavioral competencies and exhibited a crash rate of 3.01x10^-3 crashes per mile, approximately 1,000 times higher than the average human driver crash rate. During the tests, we also uncovered a number of unknown unsafe scenarios for Autoware.Universe. These findings underscore the necessity of behavioral safety evaluations for improving AV safety performance prior to widespread public deployment.
Chinese Translation
第三方对自动驾驶车辆（AV）安全的评估在提高公众接受度、建立消费者信心和制定有效安全标准方面发挥着至关重要的作用。在本研究的第一部分中，我们提出了一项专门的第三方测试倡议，用于系统评估AV的行为安全。在本文中，我们使用Autoware.Universe（一个开源的四级自动驾驶系统，Level 4 Automated Driving System, ADS）验证了我们提出的框架，该系统在模拟环境和密歇根大学Mcity测试设施的物理测试跑道上进行了测试。结果表明，Autoware.Universe具备14项行为能力中的6项，并且其碰撞率为每英里3.01x10^-3次碰撞，约为平均人类驾驶员碰撞率的1,000倍。在测试过程中，我们还发现了一些Autoware.Universe未知的不安全场景。这些发现强调了在广泛公众部署之前进行行为安全评估以提高AV安全性能的必要性。
cs.RO / 98 / 2610.08469
A Belief-State World Model for Catheter Navigation under Sparse Fluoroscopy: A Planar Proof of Concept
稀疏荧光透视下导管导航的信念状态世界模型：平面概念验证
Abstract
Endovascular catheter navigation relies on continuous fluoroscopy, exposing patients and clinical staff to ionizing radiation throughout the procedure. We investigate whether a physics-based world model can sustain the navigation task between deliberately sparse X-ray acquisitions, and whether the model can signal when its internal state estimate is no longer reliable. We formulate sparse-fluoroscopy navigation as a partially observable Markov decision process where a Cosserat rod simulator supplies transition dynamics, sparse noisy projections provide observations, and a particle filter maintains a belief over the device state, reduced in this implementation to a tip state with a geometric contact proxy. We evaluate this formulation in a deliberately simplified setting: a synthetic planar vessel phantom with a quasi-static rod model and simulated projections, without clinical or animal data. In this setting, the belief-state model tracks the simulated tip with a root mean square error of 1.30 mm while acquiring observations at one fifteenth of the continuous rate (0.97 mm at every frame, 1.51 mm at one thirtieth), reported 90 percent credible intervals achieve 0.92 empirical coverage, and the belief-derived contact risk estimate discriminates unsafe contact events with an AUROC of 0.68. These results constitute a proof of concept on a simplified simulation rather than a demonstration of clinical readiness; their purpose is to establish that calibrated belief, rather than point-estimate accuracy alone, is the essential property a sparse-imaging world model must deliver.
Chinese Translation
血管内导管导航依赖于连续的荧光透视，这在整个过程中使患者和临床工作人员暴露于电离辐射中。我们研究了基于物理的世界模型是否能够在故意稀疏的X射线采集之间维持导航任务，以及该模型是否能够指示其内部状态估计何时不再可靠。我们将稀疏荧光透视导航表述为一个部分可观察的马尔可夫决策过程，其中Cosserat杆模拟器提供转移动态，稀疏噪声投影提供观察，而粒子滤波器则在设备状态上维持信念，在此实现中简化为具有几何接触代理的尖端状态。我们在一个故意简化的环境中评估该表述：一个合成的平面血管幻影，采用准静态杆模型和模拟投影，没有临床或动物数据。在这个环境中，信念状态模型以1.30毫米的均方根误差跟踪模拟尖端，同时以连续速率的十五分之一采集观察（每帧0.97毫米，每三十分之一1.51毫米），报告的90%可信区间实现了0.92的经验覆盖率，信念派生的接触风险估计以0.68的AUROC区分不安全接触事件。这些结果构成了在简化模拟上的概念验证，而不是临床准备的展示；其目的是建立校准信念，而非仅仅是点估计的准确性，是稀疏成像世界模型必须提供的基本属性。
cs.RO / 99 / 2610.08524
No Need to Stop the Fleet: Localized Anomaly Isolation via Dead Zones in AGV Fleets
无需停止整个车队：通过死区实现AGV车队的局部异常隔离
Abstract
Automated Guided Vehicle (AGV) fleets in production and logistics environments rely on a system-wide emergency stop to handle local anomalies such as malfunctions, hazards, or contaminated areas. While safe, this approach halts the entire fleet even when only a small area is affected, causing unnecessary downtime. This paper proposes dead zones --- a formally defined localized isolation approach that serves as an additional fail-operational mechanism for situations where a full system-wide halt is not strictly needed. Rather than shutting down the entire fleet, a dead zone isolates the anomalous area, allowing unaffected robots to continue operating normally. We introduce two operational strategies: basic dead zone handling, in which affected robots enter a dead zone and stop, and dead zone handling with escaping, in which robots that are sufficiently far from a dead zone reroute around it. For escaping on counterclockwise trajectories we prove collision-freedom with a bounded arrival delay for all escaping robots, provided a minimum pairwise separation is kept between robots. Experimental evaluation confirms the theoretical guarantees and characterizes the throughput tradeoffs between all three approaches: full emergency stop, basic dead zone handling, and dead zone handling with escaping.
Chinese Translation
在生产和物流环境中，自动导引车（AGV）车队依赖于系统范围内的紧急停止来处理局部异常，如故障、危险或污染区域。尽管这种方法安全，但即使只有小范围受到影响，也会导致整个车队停运，从而造成不必要的停机时间。本文提出了死区——一种正式定义的局部隔离方法，作为在不严格需要全系统停运的情况下的额外故障操作机制。死区通过隔离异常区域，而不是关闭整个车队，允许未受影响的机器人继续正常运行。我们介绍了两种操作策略：基本死区处理，其中受影响的机器人进入死区并停止，以及带有逃逸的死区处理，其中足够远离死区的机器人绕过它。对于逆时针轨迹的逃逸，我们证明了在保持机器人之间的最小对距的前提下，所有逃逸机器人都能实现无碰撞和有限到达延迟。实验评估验证了理论保证，并描述了全紧急停止、基本死区处理和带有逃逸的死区处理三种方法之间的吞吐量权衡。
cs.RO / 100 / 2610.08526
WareFly-VLA: A Vision-Language-Action Framework for UAV Navigation and Human Tracking in Smart Warehouses
WareFly-VLA：一种用于智能仓库中无人机导航和人类跟踪的视觉-语言-动作框架
Abstract
Vision-Language-Action (VLA) models have achieved impressive results in robotic manipulation and ground-mobile navigation, yet language-conditioned control of unmanned aerial vehicles (UAVs) in smart warehouses remains largely unexplored, hindered by the lack of benchmarks that jointly provide continuous low-level flight actions, fine-grained natural-language target descriptions, and realistic industrial environments. This paper introduces WareFly-VLA, a photorealistic UAV VLA framework and dataset for language-guided human search, localization, and tracking in warehouse environments. It contains 507 human-teleoperated flight episodes and 8,504 high-resolution RGB transitions collected in NVIDIA Isaac Sim, each paired with a human-written appearance description of the target worker and a synchronized four-degree-of-freedom control command. Two aerial tasks are covered: target approach and person following, under occlusion, long-range search, altitude variation, and clutter. A unified benchmark of four open-source VLA architectures (SmolVLA, GR00T N1.7, pi_0 and OpenVLA) is established under a leakage-free episode-level protocol at two control rates. The results show that language-conditioned aerial control in warehouses is far from solved: performance drops substantially under strict generalization settings, continuous action modeling consistently outperforms discrete action tokenization, only the forward channel is reliably learnable from a single frame, and current foundation-model interfaces transfer poorly from ground and humanoid embodiments to aerial platforms. The synchronized video, language, action, pose, and difficulty annotations further support world-model research. The dataset, baselines, and evaluation protocol are released to support language-grounded aerial autonomy in smart warehouses.
Chinese Translation
视觉-语言-动作（VLA）模型在机器人操作和地面移动导航方面取得了显著成果，但在智能仓库中基于语言的无人机（UAV）控制仍然未得到充分探索，这主要是由于缺乏同时提供连续低级飞行动作、细粒度自然语言目标描述和真实工业环境的基准。本文介绍了WareFly-VLA，一个用于语言引导的人类搜索、定位和跟踪的光真实感无人机VLA框架和数据集。该数据集包含507个由人类遥控的飞行事件和8,504个在NVIDIA Isaac Sim中收集的高分辨率RGB过渡图像，每个图像都配有目标工人的人类撰写外观描述和同步的四自由度控制命令。涵盖了两项空中任务：目标接近和人员跟随，任务中涉及遮挡、远程搜索、高度变化和杂乱情况。建立了一个统一的基准，涵盖四个开源VLA架构（SmolVLA、GR00T N1.7、pi_0和OpenVLA），在两个控制速率下采用无泄漏的事件级协议进行评估。结果表明，在仓库中基于语言的空中控制远未解决：在严格的泛化设置下，性能显著下降，连续动作建模始终优于离散动作标记，只有前向通道能够从单帧中可靠学习，目前的基础模型接口在从地面和类人形态转移到空中平台时效果较差。同步的视频、语言、动作、姿态和难度注释进一步支持世界模型研究。该数据集、基线和评估协议已发布，以支持智能仓库中的语言基础空中自主性。
cs.RO / 101 / 2610.08532
Pareto-Optimal Entropy-Regularized Trajectory Optimization
帕累托最优熵正则化轨迹优化
Abstract
Trajectory optimization (TO) under nonlinear dynamics, actuation limits and collision avoidance constraints is a fundamental problem in robotics, albeit especially challenging due to its highly non-convex nature. For this setting, Differential Dynamic Programming (DDP) is an efficient second-order shooting method, yet its local structure renders it vulnerable to suboptimal basins. Sampling-augmented variants mitigate this susceptibility through stochastic exploration, but often sample only around the few trajectories they retain for reoptimization, based solely on their cost which restricts exploration breadth. We introduce Pareto-Optimal Entropy-Regularized DDP (PER-DDP), an entropy-regularized population framework derived from the free-energy/relative-entropy inequality. Our method combines prior-guided sampling that shapes exploration around each retained trajectory, with expanded rollout evaluations that probe these sampling policies beyond the few retained candidates, and Pareto filtering for preserving task-constraint alternatives across iterations. This decouples sampling effort from the optimization population size and broadens exploration without sacrificing the second-order structure that makes DDP effective. Across multiple systems and hundreds of environments, PER-DDP achieves higher success rates than state-of-the-art sampling-augmented TO methods and finds reliable solutions in environments beyond the reach of all baselines.
Chinese Translation
在非线性动力学、驱动限制和避碰约束下的轨迹优化（TO）是机器人领域中的一个基本问题，尽管由于其高度非凸特性而特别具有挑战性。在这种情况下，微分动态规划（DDP）是一种高效的二阶发射方法，但其局部结构使其容易陷入次优区域。增强采样的变体通过随机探索减轻了这种易感性，但通常仅在保留的少数轨迹周围进行采样，基于其成本限制了探索的广度。我们提出了帕累托最优熵正则化DDP（PER-DDP），这是一种基于自由能/相对熵不等式的熵正则化群体框架。我们的方法结合了以先验为指导的采样，这种采样围绕每个保留轨迹进行探索，以及扩展的回滚评估，这些评估探测这些采样策略超出少数保留候选者的范围，并通过帕累托过滤在迭代中保留任务约束的替代方案。这将采样工作与优化群体规模解耦，并在不牺牲DDP有效性的二阶结构的情况下扩大了探索范围。在多个系统和数百个环境中，PER-DDP的成功率高于最先进的增强采样TO方法，并在超出所有基线的环境中找到可靠的解决方案。
cs.RO / 102 / 2610.08541
Micro Neural Policies for Safe Real-Time Robotic Control
安全实时机器人控制的微神经策略
Abstract
In this paper, we investigate the synthesis of Micro Neural Policies (MNP) to enable safe and robust real-time robotic control on computationally constrained embedded devices. We demonstrate that integrating Evolution Strategy (ES) and Statistical Model Checking (SMC)-based verification for policy search can drastically reduce neural network size without compromising safety and robustness. We conduct a large-scale training and evaluation of MNP on Cartpole and Quadrotor control tasks, varying control frequencies and network architectures. After validating these policies in simulation, we evaluate their deployability through zero-shot transfer to physical systems. Our experiments show that MNP can successfully achieve safe sim-to-real transfer without sacrificing control performance. We then show that the policies' memory footprint, ranging from 0.5 to 7.5 kB, allows deployment on microcontrollers, where they achieve real-time inference latency with under 25 ns of jitter while leaving the chip idle for over 97% of the time for additional workloads. This makes them a highly practical solution for severely resource-constrained robotic systems.
Chinese Translation
在本文中，我们研究了微神经策略（Micro Neural Policies, MNP）的合成，以实现计算资源受限的嵌入式设备上的安全和稳健的实时机器人控制。我们证明了将进化策略（Evolution Strategy, ES）与基于统计模型检测（Statistical Model Checking, SMC）的策略搜索验证相结合，可以在不影响安全性和稳健性的前提下，显著减少神经网络的规模。我们在Cartpole和四旋翼控制任务上进行了大规模的MNP训练和评估，变化控制频率和网络架构。在模拟中验证这些策略后，我们通过零样本迁移评估其在物理系统中的可部署性。我们的实验表明，MNP能够成功实现安全的模拟到现实迁移，而不牺牲控制性能。随后，我们展示了这些策略的内存占用范围为0.5到7.5 kB，使其能够在微控制器上部署，并在保持实时推理延迟低于25 ns抖动的同时，使芯片在97%以上的时间内处于空闲状态，以处理额外的工作负载。这使得它们成为资源极度受限的机器人系统的高度实用解决方案。
cs.RO / 103 / 2610.08555
Towards Efficient Robotic Manipulation Models with Self-Recursive Pruning
朝着高效的自递归修剪机器人操作模型
Abstract
Network pruning can reduce parameter redundancy in robotic policies. However, generic pruning criteria are tailored for image recognition tasks and commonly designed to preserve weight magnitude, local reconstruction, or language-model likelihood rather than closed-loop action behavior. Directly applying these pruning algorithms to robotic tasks yields unsatisfactory performance. In this paper, we propose Loss-Conditioned Activation-Moment (LCAM) pruning, a training-free method for unstructured pruning of pre-trained robotic manipulation policies. Specifically, we first rank connections using row-normalized weight contribution, activation moments measured on calibration demonstrations, and the sensitivity of output directions to the action-prediction loss. We further design a self-recursive coarse-to-fine procedure: importance is recalibrated after each nested coarse pruning stage, while held-out offline action distortion guides fine-grained budget allocation after a sparsity knee. Our algorithm is free from costly recovery training and simulator rollouts after pruning. Experiments on three LIBERO suites with competitive robotic policies, together with evaluations on OpenVLA, show that LCAM attains competitive performance across a broad range of pruning ratios. Notably, on LIBERO-Object with OpenVLA, our LCAM achieves 84.0% success at 50% unstructured pruning, retaining over 90% of the dense policy's success rate. Promising results on real-world robotic ping pong further demonstrate the effectiveness of our pruning algorithm.
Chinese Translation
网络修剪可以减少机器人策略中的参数冗余。然而，通用的修剪标准通常是针对图像识别任务设计的，主要旨在保留权重大小、局部重构或语言模型的似然性，而不是闭环动作行为。直接将这些修剪算法应用于机器人任务会导致不理想的性能。在本文中，我们提出了一种基于损失条件的激活时刻（Loss-Conditioned Activation-Moment, LCAM）修剪方法，这是一种无需训练的非结构化修剪预训练机器人操作策略的方法。具体而言，我们首先使用行归一化的权重贡献、在标定演示中测量的激活时刻以及输出方向对动作预测损失的敏感性来对连接进行排序。我们进一步设计了一种自递归的粗到细程序：在每个嵌套的粗修剪阶段后重新校准重要性，而在稀疏性拐点后，保留的离线动作失真指导细粒度预算分配。我们的算法在修剪后无需昂贵的恢复训练和模拟器回放。在三个具有竞争力的机器人策略的LIBERO套件上的实验，以及在OpenVLA上的评估，表明LCAM在广泛的修剪比率下达到了竞争性的性能。值得注意的是，在LIBERO-Object与OpenVLA的实验中，我们的LCAM在50%的非结构化修剪下达到了84.0%的成功率，保留了超过90%的密集策略的成功率。在真实世界的机器人乒乓球实验中，令人鼓舞的结果进一步证明了我们修剪算法的有效性。
cs.RO / 104 / 2610.08595
One for All, All for One: Coordinated Multi-Agent Diffusion Steering via Stochastic Optimal Control
一为所有，所有为一：通过随机最优控制实现协调的多智能体扩散引导
Abstract
Deep generative models often produce structured outputs composed of interacting components. Modelling these outputs with a single model requires learning both the component distributions and their interactions. We pursue a modular alternative: reuse independently trained component generators and learn only how to coordinate them to produce coherent structured outputs. Our framework, Coordinated Multi-Agent Diffusion Steering (CMDS), treats frozen pretrained diffusion models as reusable generative primitives and coordinates their reverse processes through a learned control. We formulate coordination as a stochastic optimal control problem, balancing an assembly-level reward that specifies the desired properties of the combined output against deviations from the pretrained dynamics. The learned control amortises this optimisation, allowing reuse across new task instances. Experiments show that CMDS can recover a known target distribution, satisfy different spatial constraints with the same trained control, and recover individual sources from degraded mixtures. Across multi-agent maze navigation, articulated robot planning, and text-conditioned human motion, CMDS turns frozen models into coordinated multi-agent generators.
Chinese Translation
深度生成模型通常产生由相互作用组件组成的结构化输出。用单一模型对这些输出进行建模需要学习组件分布及其相互作用。我们追求一种模块化的替代方案：重用独立训练的组件生成器，仅学习如何协调它们以产生一致的结构化输出。我们的框架——协调多智能体扩散引导（Coordinated Multi-Agent Diffusion Steering, CMDS）将冻结的预训练扩散模型视为可重用的生成原语，并通过学习的控制协调它们的反向过程。我们将协调形式化为一个随机最优控制问题，平衡指定组合输出所需特性的组装级奖励与偏离预训练动态的惩罚。学习到的控制使得这一优化得以摊销，从而可以在新的任务实例中重用。实验表明，CMDS能够恢复已知的目标分布，满足不同的空间约束，并从降级混合中恢复个体源。在多智能体迷宫导航、关节机器人规划和文本条件下的人类运动中，CMDS将冻结模型转变为协调的多智能体生成器。
cs.RO / 105 / 2610.08603
A Swarm-Coordinated Multi-Robot System for Early Stress Detection in Agricultural Rows Using Multimodal Leaf Sensing
一种群体协调的多机器人系统，通过多模态叶片传感实现农业行间早期压力检测
Abstract
Early stress detection in crops is a necessity today to improve efficiency and reduce waste of time, money, and effort. However, most modern techniques, such as hyperspectral imaging and AI-based systems, are too costly and complex for medium and small-scale farmers to implement. This paper showcases CropSentry, a low-cost, ground-based multi-robot system that uses multimodal leaf sensing to continuously monitor crop health by tracking stress levels. The system comprises two autonomous bots that continuously detect leaf color and environmental data row by row. The observations are spatially mapped and sent over to the master bot, which uses color-coded row segments to generate a real-time web-based dashboard displaying crop health. After 63 observations were collected during the experiments, the results showed an overall crop health classification accuracy of 84.12%, with 82.60% for healthy plants, 88% for nutrient-deficient plants, and 80% for diseased plants. Also, 100% wireless communication success rate across 10 slave observations was achieved. Close-range leaf inspection across multiple bots can detect early stress in crops while remaining affordable, accessible, and scalable. It provides farmers with timely information to improve resource utilization and crop management.
Chinese Translation
在当今农业中，早期压力检测对于提高效率和减少时间、金钱及人力的浪费至关重要。然而，大多数现代技术，如高光谱成像和基于人工智能的系统，对于中小规模农民来说过于昂贵和复杂。本文展示了CropSentry，一个低成本的地面多机器人系统，利用多模态叶片传感技术，通过跟踪压力水平持续监测作物健康。该系统由两个自主机器人组成，逐行检测叶片颜色和环境数据。这些观察结果被空间映射并发送至主机器人，后者使用颜色编码的行段生成实时基于网络的仪表板，显示作物健康状况。在实验中收集了63次观察结果，结果显示作物健康分类的整体准确率为84.12%，健康植物的准确率为82.60%，缺乏营养的植物为88%，病害植物为80%。此外，在10次从属观察中实现了100%的无线通信成功率。多个机器人进行近距离叶片检查可以在保持经济实惠、可及和可扩展的同时检测作物的早期压力。这为农民提供了及时的信息，以改善资源利用和作物管理。
cs.RO / 106 / 2610.08637
Feeling Through the Load: Compliant Quadruped Locomotion under Payload Interactions
通过负载感知：在负载交互下的顺应性四足运动
Abstract
Quadruped robots are increasingly expected to carry objects while moving through human environments. But what happens when a person interacts directly with the payload rather than with the robot? If the payload is unrestrained, the robot must distinguish intentional external interactions from ordinary payload motion, while still keeping the load balanced and maintaining stable locomotion. How can a quadruped infer and compliantly respond to such interactions using only onboard measurements? In this work, we develop a force-aware locomotion framework that treats payload interactions as commands that shape the motion of the combined robot-payload system. Our approach separates the learning of force-aware locomotion and force estimation on an unrestrained payload. We combine a compliant load-carrying policy with a causal force estimator, trained through estimator-in-the-loop data aggregation and finetuning, to predict interactions from onboard robot measurements. Our simulations and real-world experiments show that the resulting controller can maintain stable payload-carrying locomotion, yield compliantly to external interactions, and use the inferred force to support human-guided changes in the robot's trajectory.
Chinese Translation
四足机器人越来越被期望在移动过程中携带物体穿越人类环境。但当一个人直接与负载互动而不是与机器人互动时会发生什么？如果负载没有被约束，机器人必须区分有意的外部交互与普通的负载运动，同时保持负载的平衡并维持稳定的运动。四足机器人如何仅通过机载测量推断并顺应地响应这种交互？在本研究中，我们开发了一种力感知运动框架，将负载交互视为塑造机器人-负载系统运动的指令。我们的方法将力感知运动的学习与对不受约束负载的力估计分开。我们结合了一种顺应性负载携带策略与因果力估计器，通过在环估计器数据聚合和微调的训练，来预测来自机载机器人测量的交互。我们的仿真和现实世界实验表明，所得到的控制器能够维持稳定的负载携带运动，顺应外部交互，并利用推断的力支持人类引导下的机器人轨迹变化。
cs.RO / 107 / 2610.08640
RIWANav: Recursive World-Action Models with Self-Improvement for Urban Navigation
RIWANav：具有自我改进的递归世界-动作模型用于城市导航
Abstract
Long-horizon urban navigation requires sequential local decisions whose errors can compound over time. Imitation learning (IL) rarely learns from failures, while physical trial-and-error reinforcement learning (RL) is costly. Action-conditioned world models can provide imagined feedback by predicting visual consequences for candidate actions. However, a frozen world model may become less reliable as the policy evolves. In this paper, we introduce RIWANAV, a post-training framework that casts the coupled adaptation of a world model and an action model (policy) as task-specific recursive self-improvement (RSI). Each cycle alternates two updates. The world model evaluates policy actions through imagined outcomes, providing comparative feedback for group-relative policy optimization (GRPO). The improved policy then constructs a grounded self-curriculum, selecting expert-consistent action-video pairs by behavioral novelty and prediction error. The refined world model supplies feedback for the next policy update, closing the recursive self-improvement loop. Experiments show that RIWANAV outperforms training baselines and prior methods, validating the proposed recursive self-improvement loop between the policy and world model. Real-world trials further demonstrate its practical applicability.
Chinese Translation
长时间跨度的城市导航需要顺序的局部决策，其错误可能会随着时间的推移而累积。模仿学习（IL）很少能从失败中学习，而物理试错强化学习（RL）则成本高昂。基于动作的世界模型可以通过预测候选动作的视觉后果提供想象反馈。然而，随着策略的演变，固定的世界模型可能变得不那么可靠。本文提出了RIWANAV，一种后训练框架，将世界模型和动作模型（策略）的耦合适应视为任务特定的递归自我改进（RSI）。每个周期交替进行两次更新。世界模型通过想象结果评估策略动作，为群体相对策略优化（GRPO）提供比较反馈。改进后的策略随后构建一个基于实际情况的自我课程，通过行为新颖性和预测误差选择与专家一致的动作-视频对。精炼的世界模型为下一个策略更新提供反馈，闭合递归自我改进循环。实验表明，RIWANAV优于训练基线和先前方法，验证了策略与世界模型之间提出的递归自我改进循环。现实世界的试验进一步证明了其实际应用性。
cs.RO / 108 / 2610.08642
HygieneRoboBench: Benchmarking Hygiene-Aware Planning for Household Robots
HygieneRoboBench：家庭机器人卫生意识规划的基准测试
Abstract
Contact with contaminated objects can spread hazards through a household robot's grippers, tools, and shared surfaces, while new contacts can make an existing plan unsafe. Existing benchmarks do not jointly assess how planners identify hygiene risks from contact history and plan safe continuations after new contact events. Planners must do so within time and resource limits while respecting user priorities. We introduce HygieneRoboBench, with 624 instances across 134 task families, to evaluate safe resolution of household tasks from a given execution history. Tasks capture contamination through two grippers and shared objects, treatment costs, and user priorities. We combine controlled history, profile, and event comparisons with independent plan evaluation. These assess safe resolution, cost efficiency under user priorities, and responses to contact events. Evaluation of LLM-based and symbolic planners shows that safely completing a task does not guarantee the lowest execution costs under the user's priorities. To address this problem, we introduce Hygiene-NSP. It combines LLM-based grounding, contact-history reconstruction, and CP-SAT to jointly plan hygiene treatment and task execution under user priorities. Hygiene-NSP achieves safe resolution and optimal safe resolution rates of 94.4% and 90.4%, respectively. Both rates are higher than those of the evaluated baseline planners on the full dataset. Project page: https://euron-zc.github.io/HygieneRoboBench/.
Chinese Translation
与受污染物体的接触可能通过家庭机器人的抓手、工具和共享表面传播危害，而新的接触可能使现有计划变得不安全。现有的基准测试并未共同评估规划者如何从接触历史中识别卫生风险，并在新的接触事件后规划安全的继续执行方案。规划者必须在时间和资源限制内做到这一点，同时尊重用户的优先事项。我们引入了HygieneRoboBench，包含624个实例，覆盖134个任务家族，以评估从给定执行历史中安全解决家庭任务的能力。任务通过两个抓手和共享物体捕捉污染情况、处理成本和用户优先事项。我们结合了受控历史、配置文件和事件比较，以及独立的计划评估。这些评估安全解决方案、在用户优先事项下的成本效率以及对接触事件的响应。对基于大型语言模型（LLM）和符号规划者的评估表明，安全完成任务并不保证在用户优先事项下的最低执行成本。为了解决这个问题，我们引入了Hygiene-NSP。它结合了基于LLM的基础、接触历史重建和CP-SAT，以在用户优先事项下共同规划卫生处理和任务执行。Hygiene-NSP实现了94.4%和90.4%的安全解决率和最佳安全解决率，均高于在完整数据集上评估的基线规划者的结果。项目页面：https://euron-zc.github.io/HygieneRoboBench/.
cs.RO / 109 / 2610.08650
Fast Non-Parametric Heteroscedastic Imitation Learning With Geometric Priors
快速非参数异方差模仿学习与几何先验
Abstract
When learning probabilistic policies from human demonstrations, data-efficient learning and fast adaptations to new scenarios are key requirements. One popular way to achieve intuitive and reliable adaptations is through non-parametric, typically kernel-based, methods. However, existing solutions either fail to account for the geometry of manifolds common in robotics, limiting data efficiency, or, when geometry-aware, provide unreliable uncertainty estimates or require retraining to adapt. We propose a non-parametric approach leveraging geometric priors in scenarios of data scarcity and heteroscedastic uncertainties for probabilistic modeling. We utilize the method to formulate policies based on time or robot state, where non-separable diagonal kernels allow capturing uncertainty relations between degrees of freedom for same-sized in- and outputs. Fast updates, requiring less than 3 ms for a trajectory involving both position and orientation are possible through an optimized formulation. Our approach supports both manifold-valued input and manifold-valued output with large orientation changes. Using task parameterization, adaptation to different object poses is easily possible. We evaluate the approach on a set of toy examples and on real robot manipulation tasks both in autonomous execution and in shared control scenarios.
Chinese Translation
在从人类演示中学习概率策略时，数据高效学习和快速适应新场景是关键要求。实现直观且可靠的适应性的一种流行方法是通过非参数方法，通常是基于核的方法。然而，现有的解决方案要么未能考虑机器人常见的流形几何，限制了数据效率，要么在考虑几何时提供不可靠的不确定性估计或需要重新训练以进行适应。我们提出了一种非参数方法，利用几何先验在数据稀缺和异方差不确定性场景中进行概率建模。我们利用该方法基于时间或机器人状态制定策略，其中不可分离的对角核能够捕捉同尺寸输入和输出之间的自由度不确定性关系。通过优化的公式，快速更新成为可能，涉及位置和方向的轨迹更新所需时间少于3毫秒。我们的方法支持具有大方向变化的流形值输入和流形值输出。通过任务参数化，适应不同物体姿态变得容易。我们在一组玩具示例和真实机器人操作任务中评估了该方法，包括自主执行和共享控制场景。
cs.RO / 110 / 2610.08653
Magnet-Aware Control of Legged Robots
磁场感知的四足机器人控制
Abstract
Autonomous robots can increase uptime and reduce human exposure in Big Science facilities, but strong magnetic fields needed for their operation corrupt sensors and induce pose-dependent mechanical wrenches that destabilize robots and challenge conventional reactive controllers. This paper presents a control framework for modeling, estimating, and dynamically compensating for spatially varying magnetic wrenches acting on legged robots to improve robustness in these fields. We introduce a custom physics plugin for the MuJoCo simulator to model magnetic forces on rigid-body elements, alongside an inverse field-estimation framework to infer the latent magnetic field directly from quadruped dynamic responses and any number of sensor readings. Furthermore, we develop a Magnet-Aware Model Predictive Control (MPC) and Whole-Body Control (WBC) architecture that predicts and counteracts magnetic perturbations during locomotion to increase the range of magnetic fields in which the robot can operate. The effectiveness of the framework is validated through both simulation and physical hardware experiments. We show our method increases the the maximum rejectable disturbances from magnetic field in the force space by a factor of 2.5 and and between 1.6-2 times in torque space compared to a non-compensated system. Within the proposed magnetic field, this constitutes an increase in the area in which the robot can operate by 29,34% of which 10,84% would have previously caused an immediate collapse to non-compensated controllers.
Chinese Translation
自主机器人可以提高大科学设施的工作时间并减少人类的暴露，但其操作所需的强磁场会干扰传感器并引发依赖姿态的机械扭矩，从而使机器人不稳定并对传统的反应控制器构成挑战。本文提出了一种控制框架，用于建模、估计和动态补偿作用于四足机器人的空间变化磁扭矩，以提高其在这些领域的鲁棒性。我们为MuJoCo模拟器引入了一个自定义物理插件，以建模刚体元素上的磁力，并提出了一种逆场估计框架，以直接从四足动物的动态响应和任意数量的传感器读数中推断潜在的磁场。此外，我们开发了一种磁场感知的模型预测控制（Magnet-Aware Model Predictive Control, MPC）和全身控制（Whole-Body Control, WBC）架构，能够在运动过程中预测并抵消磁扰动，从而扩大机器人可操作的磁场范围。通过模拟和物理硬件实验验证了该框架的有效性。我们展示了与未补偿系统相比，我们的方法在力空间中将最大可拒绝的磁场干扰提高了2.5倍，在扭矩空间中提高了1.6至2倍。在所提出的磁场范围内，这意味着机器人可操作区域增加了29.34%，其中10.84%的区域在未补偿控制器下会导致立即崩溃。
cs.RO / 111 / 2610.08695
NMPP: Nonlinear Model Predictive Planning for Agile UAV Flight in Cluttered Environments
NMPP：用于复杂环境中灵活无人机飞行的非线性模型预测规划
Abstract
Flying a quadrotor through a cluttered environment requires not only planning a collision-free reference trajectory based on perceived obstacles, but the reference also needs to be dynamically feasible and within the actuation limits of the vehicle, so that the controller can track it precisely. Existing methods either optimize a smooth polynomial inside a convex corridor, which limits agility, or treat obstacles as soft costs traded against tracking performance. We propose a Nonlinear Model Predictive Planning (NMPP) that imposes perceived obstacles as hard geometric constraints and hands a full-state reference to an obstacle-blind SE(3) controller. Our planner achieves a 58-67 % lower position RMSE than a linear Model Predictive Control trajectory planner and a 41-70 % lower RMSE than a polynomial trajectory planner. It also completes all forest flights with up to 9.5 m/s speed without collisions, and achieves 86 % flight success rate under a more aggressive speed profile where a state-of-the-art planner has only 26 % success rate. The real-world deployment showed reliable execution flying up to 5.5 m/s in an unknown cluttered environment.
Chinese Translation
在复杂环境中飞行四旋翼无人机不仅需要基于感知到的障碍物规划无碰撞的参考轨迹，而且该参考轨迹还需在动态上可行，并符合飞行器的驱动限制，以便控制器能够精确跟踪。现有方法要么在凸走廊内优化平滑多项式，这限制了灵活性，要么将障碍物视为与跟踪性能进行权衡的软成本。我们提出了一种非线性模型预测规划（NMPP），将感知到的障碍物作为硬几何约束，并将完整状态参考交给一个盲障碍物的SE(3)控制器。我们的规划器实现了比线性模型预测控制轨迹规划器低58-67%的位置均方根误差（RMSE），并且比多项式轨迹规划器低41-70%的RMSE。它还在速度高达9.5 m/s的情况下完成了所有森林飞行且没有发生碰撞，并在更具攻击性的速度配置下实现了86%的飞行成功率，而最先进的规划器仅有26%的成功率。实际部署显示，在未知复杂环境中以高达5.5 m/s的速度可靠执行。
cs.RO / 112 / 2610.08726
EgoLAP: Learning from Egocentric Human Data through Language-Action Reasoning
EgoLAP：通过语言-动作推理从自我中心人类数据中学习
Abstract
Egocentric human data offer a path to scaling robot learning beyond costly robot demonstrations, yet the embodiment gap makes raw human trajectories a poor supervisory target for control. Our key insight is that, although low-level actions are embodiment-specific, their underlying motion intent can capture task-relevant structure that transfers across humans and robots. We introduce EgoLAP, a VLA pre-training framework that jointly learns from human and robot trajectories through a shared language-based action chain-of-thought. EgoLAP expresses motion intent as structured, temporally abstracted language actions and pairs them with motion-level reasoning grounded in scene geometry, physics, and object affordances. Across extensive real-world and simulated experiments, EgoLAP transfers human experience to robot control more effectively than alternative action representations and reaches 80.1% mean real-world task progress, a 2.3x performance gain over alternative action representations. Motion-level reasoning also outperforms a composite reasoning format that combines subtask, object-box, and visual-trace reasoning.
Chinese Translation
自我中心的人类数据为机器人学习提供了一条超越昂贵机器人演示的途径，但体现差距使得原始人类轨迹成为控制的较差监督目标。我们的关键见解是，尽管低级动作是特定于体现的，但其潜在的运动意图可以捕捉到跨越人类和机器人的任务相关结构。我们提出了EgoLAP，一种VLA（视觉-语言-行动）预训练框架，通过共享的基于语言的行动思维链共同学习人类和机器人轨迹。EgoLAP将运动意图表达为结构化的、时间抽象的语言动作，并将其与基于场景几何、物理学和物体可供性（affordances）的运动级推理配对。在广泛的现实世界和模拟实验中，EgoLAP比其他行动表示更有效地将人类经验转移到机器人控制中，并达到了80.1%的平均现实世界任务进展，相较于其他行动表示提高了2.3倍的性能。运动级推理也优于结合子任务、物体-框和视觉轨迹推理的复合推理格式。
cs.RO / 113 / 2610.08733
Towards an Extensible Benchmark for Spoken Dialogue with Social Robots
面向可扩展的社交机器人语音对话基准
Abstract
Language models provide a plug-and-play interface between humans and robots, but important challenges remain when speech, dialogue, fast interaction, and collaboration are required. We propose a benchmark for the community to use as a way to explore common spoken dialogue artifacts between robots and humans, including requests for clarification, interruptions, embodied signals (e.g., head nods or facial cues), and time constraints. We also explain our vision to extend the benchmark for other aspects of human-robot interaction that are important to the larger research community. To facilitate the benchmark, we further propose using \textit{Retico}, a real-time communication framework that fulfills important technical requirements to enable robots to have spoken dialogue capabilities.
Chinese Translation
语言模型为人类与机器人之间提供了即插即用的接口，但在需要语音、对话、快速互动和协作时仍然面临重要挑战。我们提出了一个基准，供社区使用，以探索机器人与人类之间的共同语音对话特征，包括请求澄清、打断、具身信号（例如，点头或面部提示）以及时间限制。我们还阐述了将基准扩展到人机交互其他重要方面的愿景，这对更广泛的研究社区至关重要。为了促进基准的实施，我们进一步提议使用 extit{Retico}，一个实时通信框架，满足重要的技术要求，使机器人能够具备语音对话能力。
cs.RO / 114 / 2610.08737
PhoneBot: A Low-Cost Open Humanoid Robot Platform Reusing Smartphones
PhoneBot：一种低成本的开放式类人机器人平台，重用智能手机
Abstract
The adoption of humanoid robots in education and research remains limited by high hardware costs, complex sensing systems, and substantial computational requirements. This paper presents PhoneBot, a low-cost, open-source humanoid robot platform that repurposes commodity smartphones as its primary sensing and computing unit. By using a smartphone's integrated inertial measurement unit (IMU), camera, wireless connectivity, and onboard processing capabilities, PhoneBot reduces hardware costs and simplifies the system architecture. The robot combines a modular lower-body structure driven by 13 low-cost actuators with a torso-mounted smartphone that supports perception, control computation, and user interaction. We describe the mechanical design, software architecture, and real-time communication framework that support stable locomotion and capabilities including vision-based human following, conversational interaction, filming, and mobile telepresence. Experimental evaluations demonstrate reliable walking, perception-driven interaction, and straightforward deployment using off-the-shelf consumer smartphones. With fully open-source hardware and software designs, PhoneBot provides an affordable, reproducible platform for education, research, and rapid prototyping. More details are available at https://phonebot.dev.
Chinese Translation
类人机器人在教育和研究中的应用仍受到高硬件成本、复杂的传感系统和巨大的计算需求的限制。本文介绍了PhoneBot，一种低成本、开源的类人机器人平台，利用普通智能手机作为其主要的传感和计算单元。通过使用智能手机集成的惯性测量单元（IMU）、摄像头、无线连接和板载处理能力，PhoneBot降低了硬件成本并简化了系统架构。该机器人结合了由13个低成本执行器驱动的模块化下肢结构和一个支持感知、控制计算和用户交互的机身安装智能手机。我们描述了支持稳定行走的机械设计、软件架构和实时通信框架，以及包括基于视觉的人类跟随、对话交互、拍摄和移动远程呈现等能力。实验评估表明，PhoneBot实现了可靠的行走、基于感知的交互以及使用现成消费智能手机的简单部署。凭借完全开源的硬件和软件设计，PhoneBot为教育、研究和快速原型制作提供了一个经济实惠、可重复的平台。更多细节可访问 https://phonebot.dev。
cs.RO / 115 / 2610.08765
LBA-CBF: Rapidly Adaptive Safety Filters via Parallel Dynamics Inference
LBA-CBF：通过并行动态推理实现快速自适应安全过滤器
Abstract
Control barrier functions (CBFs) certify commands through an assumed dynamics model, so an abrupt, unmeasured regime change can undermine the certificate exactly when safety matters most. We present Look-Back Adaptive Control Barrier Functions (LBA-CBF), which rank a finite bank of candidate dynamics by recent prediction error over a short look-back window and enforce the high-order CBF condition against every model within a tolerance of the best, spanning best-fit adaptation to full-bank robust filtering. The dynamics may depend nonlinearly on the unknown parameters, and no switching model or continuously parameterized estimator is required. We prove that any feasible filtered input satisfies the true CBF condition whenever a safety-representative candidate is retained. In quadrotor simulation with abrupt wind reversals and an unknown payload, LBA-CBF is safe and reaches the goal from all random initial conditions, matching an oracle, while adaptive and robust baselines achieve 0-88% success. Banks of up to 250,000 models run inside the control loop, and Crazyflie 2.1 and F1TENTH experiments demonstrate adaptation to wind, payload release, and varying tire-road friction. Code, videos, and project details are available at: https://lla-control.github.io
Chinese Translation
控制屏障函数（CBFs）通过假设的动态模型来验证指令，因此突发的、未测量的状态变化可能会在安全性最为重要时破坏这一验证。我们提出了回顾自适应控制屏障函数（LBA-CBF），该方法通过最近的预测误差在短期回顾窗口内对有限的候选动态进行排名，并在最佳模型的容忍范围内对每个模型强制执行高阶CBF条件，从最佳拟合适应到全库稳健过滤。动态可能非线性地依赖于未知参数，并且不需要切换模型或连续参数化估计器。我们证明了每当保留一个安全代表候选时，任何可行的过滤输入都满足真实的CBF条件。在具有突发风向变化和未知载荷的四旋翼仿真中，LBA-CBF是安全的，并且能够从所有随机初始条件达到目标，表现与一个预言者相匹配，而自适应和稳健的基线方法成功率仅为0-88%。多达250,000个模型的库在控制循环内运行，Crazyflie 2.1和F1TENTH实验展示了对风、载荷释放和不同轮胎-路面摩擦的适应。代码、视频和项目详情可在以下网址获取：https://lla-control.github.io
cs.RO / 116 / 2610.08780
DepthWorld: 3D World Model for Robot Manipulation
DepthWorld：用于机器人操作的3D世界模型
Abstract
World models offer a data-driven alternative to traditional simulators for robotics, with applications spanning policy evaluation, improvement, and planning. All of these uses depend on faithful 3D geometry, yet current video-based world models are trained on RGB alone and produce rollouts that look correct frame-by-frame but do not compose into a consistent 3D world. Closing this gap requires progress on two fronts: large-scale 3D supervision for manipulation, and an architecture that can absorb it without disturbing strong pretrained video priors. We introduce a calibration pipeline that combines learned stereo depth with a joint factor graph, pooling all episodes collected from the same physical robot to recover its shared kinematic parameters alongside per-scene extrinsics. Applied to the DROID dataset, this yields DROID-3D, a calibrated 3D dataset providing dense metric depth and recalibrated multi-view extrinsics (achieving <0.7 px reprojection error on 90% of episodes for external cameras). We then train DepthWorld, a Stable Video Diffusion-based world model that jointly predicts multi-view RGB and depth via spatial latent tiling, leaving the pretrained Variational Autoencoder (VAE) unchanged. Depth supervision improves RGB prediction itself by +1.48 dB PSNR over an identical RGB-only baseline at equal training budget, while simultaneously yielding accurate metric depth for downstream geometric reasoning.
Chinese Translation
世界模型为机器人技术提供了一种数据驱动的替代方案，应用于策略评估、改进和规划等多个领域。所有这些应用都依赖于真实的3D几何结构，然而当前基于视频的世界模型仅在RGB数据上进行训练，产生的回放在逐帧上看似正确，但并未构成一个一致的3D世界。要弥补这一差距，需要在两个方面取得进展：大规模的3D监督用于操作，以及一种能够吸收这些监督而不干扰强大预训练视频先验的架构。我们提出了一种校准管道，将学习到的立体深度与联合因子图结合起来，汇集来自同一物理机器人收集的所有数据集，以恢复其共享的运动学参数以及每个场景的外部参数。应用于DROID数据集，这产生了DROID-3D，一个经过校准的3D数据集，提供密集的度量深度和重新校准的多视角外部参数（在90%的外部摄像机数据集中实现<0.7像素重投影误差）。然后，我们训练了DepthWorld，一个基于稳定视频扩散的世界模型，通过空间潜在平铺共同预测多视角RGB和深度，同时保持预训练的变分自编码器（Variational Autoencoder, VAE）不变。深度监督在相同训练预算下使RGB预测本身提高了+1.48 dB PSNR，相较于一个相同的仅RGB基线，同时为下游几何推理提供了准确的度量深度。
cs.RO / 117 / 2610.08784
PEARS: Physical-Prior-Guided Efficient Adaptation via Failure Reasoning and Diffusion Steering for Tactile Manipulation
PEARS：通过失败推理和扩散引导进行物理先验指导的高效适应，用于触觉操作
Abstract
Pretrained robotic policies can suffer substantial performance degradation under out-of-distribution (OOD) conditions encountered during deployment, motivating post-training through real-world interaction. However, reinforcement-learning (RL)-based post-training typically requires substantial environment interactions, a burden that is especially significant in manipulation, where each trial can be slow, costly, or destructive. Therefore, we present PEARS, a physics-prior-guided hybrid RL framework for sample-efficient online adaptation of pretrained policies with tactile feedback. After each episode, its physics-guided force reasoning (PFR) module uses physical priors encoded in a vision-language model (VLM) to diagnose failures from the visual outcome and tactile interaction history and update task-appropriate contact-force bounds. A high-frequency hybrid force-position controller then enforces these bounds during contact. Complementarily, tactile-conditioned diffusion steering reinforcement learning adjusts the latent noise of the frozen flow-matching policy to correct errors in free-space motion and contact timing without updating the base model. In simulation, PEARS improves success rates by 12.4-37.4 percentage points over the strongest per-task baselines. PEARS also reduces the number of interaction episodes required for a certain success threshold by up to 53.2% relative to the fastest baseline. In real-world experiments, PEARS achieves success rates of 95% on Whiteboard Erasing and 90% on Pipette Liquid Aspiration. These results show that combining the PFR module with policy steering can accelerate adaptation while reducing costly interactions. The project website is available at https://song-kun.github.io/pears.
Chinese Translation
预训练的机器人策略在部署过程中遇到的分布外（OOD）条件下可能会遭受显著的性能下降，这促使通过与现实世界的互动进行后期训练。然而，基于强化学习（RL）的后期训练通常需要大量的环境交互，这在操作任务中尤其显著，因为每次试验可能会缓慢、昂贵或具有破坏性。因此，我们提出了PEARS，一种基于物理先验的混合RL框架，旨在通过触觉反馈实现预训练策略的样本高效在线适应。在每个回合结束后，其物理引导力推理（PFR）模块利用编码在视觉-语言模型（VLM）中的物理先验，从视觉结果和触觉交互历史中诊断失败，并更新任务适当的接触力边界。然后，高频混合力-位置控制器在接触期间强制执行这些边界。补充地，基于触觉条件的扩散引导强化学习调整冻结流匹配策略的潜在噪声，以纠正自由空间运动和接触时机中的错误，而无需更新基础模型。在仿真中，PEARS的成功率比最强的每任务基线提高了12.4-37.4个百分点。相较于最快的基线，PEARS还将达到特定成功阈值所需的交互回合数量减少了多达53.2%。在现实世界实验中，PEARS在白板擦除任务上实现了95%的成功率，在移液管液体吸取任务上实现了90%的成功率。这些结果表明，将PFR模块与策略引导相结合可以加速适应，同时减少昂贵的交互。项目网站可访问：https://song-kun.github.io/pears。
cs.RO / 118 / 2610.08789
QF3: Fast Flow RL with Filtered Q-Gradients
QF3：带有过滤Q梯度的快速流强化学习
Abstract
Flow policies have become a standard policy class for learning robot behaviors from demonstrations, but reinforcement learning is still critical for improving pre-trained flow policies or learning them from scratch through interaction. We introduce QF3 (Fast Flow RL with Filtered Q-Gradients), an online off-policy RL algorithm that trains a flow policy with flow matching plus the critic's action gradient, backpropagated through a one-step prediction of the flow's output. To keep updates where the critic and this prediction are reliable, QF3 applies the critic gradient only to action dimensions that stay near the replay action. To our knowledge, QF3 is the first off-policy flow RL method to train humanoid locomotion policies from scratch and transfer them zero-shot to hardware. Paired with a high-throughput off-policy training recipe, it trains humanoid locomotion and motion-tracking policies with a 10x wall-clock speedup over FPO++, a recent on-policy flow RL method. We further apply QF3 to fine-tune pretrained flow-based manipulation policies on both ABC-Sim and Robomimic tasks. These results suggest that QF3 can both learn robot policies from scratch and refine those acquired from demonstrations. Website: https://qf3-rl.github.io/
Chinese Translation
流策略已成为从示范中学习机器人行为的标准策略类别，但强化学习仍然对改进预训练的流策略或通过交互从头学习这些策略至关重要。我们介绍了QF3（带有过滤Q梯度的快速流强化学习），这是一种在线的离线策略强化学习算法，它通过流匹配加上评论者的动作梯度来训练流策略，该梯度通过流输出的一步预测进行反向传播。为了在评论者和该预测可靠的地方保持更新，QF3仅将评论者梯度应用于接近重放动作的动作维度。据我们所知，QF3是第一种从头训练类人步态策略并将其零-shot转移到硬件上的离线流强化学习方法。结合高吞吐量的离线训练方案，它以比最近的在线流强化学习方法FPO++快10倍的墙钟速度训练类人步态和运动跟踪策略。我们进一步将QF3应用于在ABC-Sim和Robomimic任务上微调预训练的基于流的操作策略。这些结果表明，QF3既可以从头学习机器人策略，也可以完善从示范中获得的策略。网站：https://qf3-rl.github.io/
cs.CV / 1 / 2610.06896
Medical Image Alignment Assessment as a Test of Generalist Visual Reasoning in Frontier Multimodal Models
医学图像对齐评估作为前沿多模态模型通用视觉推理的测试
Abstract
Frontier multimodal large language models (MLLMs) are increasingly positioned as general purpose visual reasoners as part of the quest for artificial general intelligence. A key test of this generality is whether they can perform novel visual judgments that humans can make reliably from visual evidence and task instructions, without task-specific parameter optimisation. We investigate this question through the task of medical image alignment assessment, where the goal is to establish whether there is anatomical correspondence between two images. Human visual assessment of image alignment is still the gold standard and most common approach; however, it requires trained operators and is impractical to scale for large datasets. We evaluate recent generations of MLLMs on two exemplar medical image alignment tasks, varying both prompting strategies and image-presentation methods. We compare against a locally fine-tuned MLLM and a task-specific CNN to examine the trade-off between frontier general purpose models and smaller models that require specific task optimisation but can be used locally. We show that are reaching an inflection point, where frontier MLLMs can now perform effective visual assessment of medical image alignment. Models released only a few months ago generalise poorly and, in some settings, perform barely above chance, whereas GPT-6 achieves over 85% across almost all scenarios tested. Fine-tuned local models can match or exceed frontier-model performance on the tasks on which they are trained, but transfer substantially less effectively to unseen settings. These findings identify medical image alignment as a useful test bed for generalist visual reasoning and suggest that frontier multimodal models are beginning to acquire capabilities that could support a common quality-control mechanism across heterogeneous medical-imaging pipelines.
Chinese Translation
前沿多模态大型语言模型（MLLMs）越来越被视为通用视觉推理器，这是追求人工通用智能的一部分。这种通用性的一个关键测试是它们是否能够执行人类能够从视觉证据和任务指令中可靠地做出的新颖视觉判断，而无需特定任务的参数优化。我们通过医学图像对齐评估的任务来探讨这个问题，其目标是确定两幅图像之间是否存在解剖对应关系。人类对图像对齐的视觉评估仍然是金标准和最常见的方法；然而，它需要经过培训的操作员，并且在大数据集上扩展性差。我们在两个示例医学图像对齐任务上评估了最近几代的MLLMs，改变了提示策略和图像呈现方法。我们与一个局部微调的MLLM和一个特定任务的卷积神经网络（CNN）进行比较，以检查前沿通用模型与需要特定任务优化但可以在本地使用的小型模型之间的权衡。我们展示了一个转折点，前沿MLLMs现在能够有效地进行医学图像对齐的视觉评估。几个月前发布的模型泛化能力较差，在某些设置中表现几乎与随机猜测相当，而GPT-6在几乎所有测试场景中达到了超过85%的准确率。经过微调的本地模型可以在其训练的任务上匹配或超过前沿模型的性能，但在未见过的设置中转移效果显著较差。这些发现将医学图像对齐识别为通用视觉推理的有用测试平台，并表明前沿多模态模型开始获得能够支持异构医学成像管道的共同质量控制机制的能力。
cs.CV / 2 / 2610.06932
RADC: Risk-Aware Dual Caching for Vision-Language Test-Time Adaptation
RADC：风险感知的视觉-语言测试时适应双缓存
Abstract
Cache-based test-time adaptation (TTA) for vision-language models is often hindered by background bias in global representations and unreliable entropy-based cache admission under representation variations. To address these limitations, we propose RADC, which enhances prototype learning through reliable dual caching. RADC introduces a Semantic Foreground Cache that aggregates category-consistent spatial evidence from CLIP representations, yielding foreground prototypes that complement the global cache while mitigating background interference. To reliably manage both caches, Gaussian Risk Admission models multi-view representations as diagonal Gaussian distributions and jointly considers class separation and feature uncertainty to prioritize reliable cache candidates. RADC integrates zero-shot logits with complementary global- and foreground-cache predictions for robust inference. Extensive experiments on cross-domain and out-of-distribution benchmarks demonstrate consistent state-of-the-art performance.
Chinese Translation
基于缓存的视觉-语言模型测试时适应（TTA）常常受到全局表示中的背景偏差和在表示变化下不可靠的基于熵的缓存接纳的限制。为了解决这些问题，我们提出了RADC，它通过可靠的双缓存增强原型学习。RADC引入了一种语义前景缓存，该缓存从CLIP表示中聚合类别一致的空间证据，生成前景原型，这些原型补充了全局缓存，同时减轻了背景干扰。为了可靠地管理这两个缓存，高斯风险接纳将多视图表示建模为对角高斯分布，并联合考虑类别分离和特征不确定性，以优先选择可靠的缓存候选。RADC将零-shot logits与互补的全局缓存和前景缓存预测相结合，以实现稳健的推理。在跨领域和分布外基准上的大量实验表明，RADC始终保持最先进的性能。
cs.CV / 3 / 2610.06938
UniPro: Unified Multi-Mode Medical Image Segmentation from 2D Images to 3D Volumes via Propagation
UniPro：通过传播实现从2D图像到3D体积的统一多模式医学图像分割
Abstract
Medical image segmentation remains fragmented along two axes: segmentation paradigms and data dimensionality. Existing methods are typically developed separately for semantic, in-context, and interactive segmentation, and are further specialized to either native 2D images or 3D volumetric data. In clinical practice, however, segmentation workflows take many forms: a case may be initialized by semantic prediction, reference-guided segmentation, or user interaction. Regardless of how it begins, fine-grained refinement is naturally performed on 2D views; for volumetric scans, such 2D edits must propagate coherently to the rest of the volume. We present UniPro, a unified model that bridges segmentation paradigms and data dimensionality, using propagation to extend 2D segmentation to 3D volumes. Our key insight is that volumetric propagation and in-context segmentation share the same reference-conditioned prediction mechanism, differing only in whether the reference image-mask pairs come from other cases or from previously segmented neighboring slices. Building on this view, UniPro supports semantic, in-context, interactive, and propagation-based segmentation within a single slice-based framework, using class priors, reference exemplars, user clicks, and neighboring-slice predictions as mode-specific conditioning inputs. To improve propagation reliability, UniPro further incorporates bidirectional and 3D supervision to regularize slice-wise propagation beyond per-slice losses. Extensive experiments across diverse modalities and anatomies show that UniPro achieves strong performance across all segmentation settings, enabling annotation-efficient 3D segmentation from sparse 2D initialization and reducing slice-by-slice correction effort.
Chinese Translation
医学图像分割在分割范式和数据维度两个方面仍然存在碎片化现象。现有方法通常分别针对语义分割、上下文分割和交互式分割进行开发，并进一步专门化为原生的2D图像或3D体积数据。然而，在临床实践中，分割工作流程有多种形式：一个案例可能由语义预测、参考引导分割或用户交互初始化。无论如何开始，细粒度的精细化自然是在2D视图上进行的；对于体积扫描，这样的2D编辑必须一致地传播到整个体积中。我们提出了UniPro，一个统一模型，连接分割范式和数据维度，利用传播将2D分割扩展到3D体积。我们的关键见解是，体积传播和上下文分割共享相同的参考条件预测机制，唯一的区别在于参考图像-掩膜对来自其他案例还是来自先前分割的相邻切片。基于这一观点，UniPro在单一的切片基础框架内支持语义、上下文、交互式和基于传播的分割，使用类别先验、参考示例、用户点击和相邻切片预测作为模式特定的条件输入。为了提高传播的可靠性，UniPro进一步结合了双向和3D监督，以正则化切片间的传播，超越每切片损失。针对多种模态和解剖结构的广泛实验表明，UniPro在所有分割设置中均表现出强大的性能，实现了从稀疏的2D初始化中高效注释的3D分割，并减少了逐切片修正的工作量。
cs.CV / 4 / 2610.06945
Beyond the Linear Representation Hypothesis: Non-Linear Activation Steering in Text-to-Image Models
超越线性表征假设：文本到图像模型中的非线性激活引导
Abstract
Mechanistic interpretability often relies on the Linear Representation Hypothesis (LRH), which assumes that high-level concepts are encoded as linear directions in activation space. Yet a natural visual concept does not necessarily require a linear visual transition: between sunny and stormy lies an intermediate weather state such as a sky with a few white clouds, not simply a weaker storm; between a caterpillar and a butterfly, the progression is not a caterpillar with continuously growing wings. This raises the question of whether such true intermediate states are also represented nonlinearly by the model. Indeed, when we prompt text-to-image models directly for intermediate attributes, their activations rarely fall along the straight direction connecting the endpoints. Therefore, we propose KANSteer, which models concept traversal as a curve passing through its intermediate states. Seeking a representation that is both simple and interpretable, we propose to use Kolmogorov-Arnold Networks (KANs), which provide a one-dimensional coordinate whose learned functions define the trajectory. This allows the steering direction to vary along the concept while preserving an interpretable representation. Across several concepts and text-to-image diffusion transformers, we find that their activation trajectories substantially deviate from straight lines, and that KANSteer provide a closer fit and smoother traversal of intermediate attributes than linear steering.
Chinese Translation
机械解释性通常依赖于线性表征假设（Linear Representation Hypothesis, LRH），该假设认为高层次概念在激活空间中被编码为线性方向。然而，自然视觉概念并不一定需要线性视觉过渡：在晴天与暴风雨之间存在一种中间天气状态，比如有几朵白云的天空，而不仅仅是较弱的暴风雨；在毛虫与蝴蝶之间，进展并不是毛虫的翅膀不断生长。这引发了一个问题，即模型是否也以非线性方式表示这些真实的中间状态。实际上，当我们直接向文本到图像模型提示中间属性时，它们的激活通常并不会沿着连接端点的直线方向变化。因此，我们提出了KANSteer，它将概念的遍历建模为通过其中间状态的曲线。为了寻求一种既简单又可解释的表征，我们建议使用Kolmogorov-Arnold Networks（KANs），它提供了一维坐标，其学习的函数定义了轨迹。这使得引导方向可以在概念中变化，同时保持可解释的表征。在多个概念和文本到图像扩散变换器中，我们发现它们的激活轨迹显著偏离直线，并且KANSteer在中间属性的拟合和光滑遍历方面比线性引导更为接近。
cs.CV / 5 / 2610.06960
DistScene: Object-to-Scene Distillation for 3D Scene Generation
DistScene：用于3D场景生成的对象到场景蒸馏
Abstract
We present DistScene, a framework for single-image compositional 3D scene generation by jointly modeling the environment and individual objects. Unlike existing methods that represent scenes primarily as collections of objects, we model the environment as an explicit scene component to provide geometric context for object placement. Specifically, we introduce Scene-Frame Generation, which jointly generates separate environment and object components in a shared coordinate frame, allowing their geometry and relative placement to be learned together. Then we introduce Object-Centric Refinement to refine each object in a local frame with scene context. Finally, we develop Object-to-Scene Distillation to transfer pretrained object-generation priors to scene generation through automatically composed and rendered synthetic scenes. Evaluations on indoor and outdoor benchmarks demonstrate improved scene-level spatial coherence over the evaluated baselines. Project page: https://coolbeam.github.io/DistScene/
Chinese Translation
我们提出了DistScene，一个通过联合建模环境和单个对象来进行单图像组合3D场景生成的框架。与现有方法主要将场景视为对象集合不同，我们将环境建模为一个明确的场景组件，以提供对象放置的几何上下文。具体而言，我们引入了场景框架生成（Scene-Frame Generation），该方法在共享坐标框架中共同生成独立的环境和对象组件，从而使它们的几何形状和相对位置可以一起学习。然后，我们引入了以对象为中心的细化（Object-Centric Refinement），在局部框架中结合场景上下文来细化每个对象。最后，我们开发了对象到场景蒸馏（Object-to-Scene Distillation），通过自动组合和渲染合成场景，将预训练的对象生成先验转移到场景生成中。对室内和室外基准的评估显示，与评估的基线相比，场景级空间一致性得到了改善。项目页面：https://coolbeam.github.io/DistScene/
cs.CV / 6 / 2610.06972
BoT-Feedback: Grounding Multimodal Reasoning in Biomechanical Evidence for Explainable Human Action Feedback
BoT-Feedback：基于生物力学证据的多模态推理与可解释人类动作反馈的结合
Abstract
Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities in visual understanding and multimodal reasoning, yet they remain fundamentally limited in Human Action Feedback Generation. Existing methods infer coaching feedback directly from visual observations, producing generic advice, limited interpretability, and physically implausible hallucinations. In contrast, expert human coaches diagnose performance through explicit biomechanical reasoning over joint kinematics, posture, and body dynamics. We introduce BoT-Feedback, a framework that grounds MLLM reasoning in structured biomechanical evidence. Our key contribution is Biomechanics of Thought (BoT), a four-stage reasoning framework that progressively identifies the action, localises the critical body regions, analyses quantitative biomechanical differences between expert and student performances, and synthesises interpretable coaching feedback. To support this reasoning process, we develop a plug-and-play Biomechanical Data Parser (BDP) that converts videos into structured biomechanical descriptors and an alignment strategy that temporally matches expert and student motions. We further introduce BiomAF, a benchmark containing paired teacher-student videos, 3D skeletons, biomechanical attributes, and expert-coaching annotations. Experiments across twelve open- and closed-source MLLMs demonstrate that grounding reasoning in biomechanical evidence consistently improves feedback quality, interpretability, and robustness while substantially reducing biomechanical hallucinations. BoT-Feedback improves the average expert evaluation score from 2.07 to 2.95 (+40%), enabling compact open-source MLLMs to approach the performance of substantially larger proprietary systems for explainable action feedback generation.
Chinese Translation
多模态大型语言模型（MLLMs）在视觉理解和多模态推理方面展现了令人印象深刻的能力，但在生成人类动作反馈方面仍然存在根本性的局限。现有方法直接从视觉观察中推断教练反馈，产生通用建议，缺乏可解释性，并且可能出现不符合物理现实的幻觉。相比之下，专业人类教练通过对关节运动学、姿势和身体动态的明确生物力学推理来诊断表现。我们提出了BoT-Feedback，一个将MLLM推理与结构化生物力学证据相结合的框架。我们的主要贡献是“思维生物力学”（Biomechanics of Thought, BoT），这是一个四阶段推理框架，逐步识别动作、定位关键身体区域、分析专家与学生表现之间的定量生物力学差异，并综合可解释的教练反馈。为了支持这一推理过程，我们开发了一个即插即用的生物力学数据解析器（Biomechanical Data Parser, BDP），将视频转换为结构化的生物力学描述符，并提出了一种时间对齐策略，以匹配专家与学生的动作。我们进一步引入了BiomAF，一个包含配对教师-学生视频、3D骨架、生物力学属性和专家教练注释的基准数据集。在十二个开放源和闭源的MLLMs上的实验表明，将推理基于生物力学证据能够持续提高反馈质量、可解释性和鲁棒性，同时显著减少生物力学幻觉。BoT-Feedback将专家评估的平均分数从2.07提高到2.95（+40%），使得紧凑的开放源MLLMs能够接近大规模专有系统在可解释动作反馈生成方面的表现。
cs.CV / 7 / 2610.06973
Event Cameras for Melt-Pool Monitoring in Additive Manufacturing: A Benchmark and a Cross-Machine Transfer Analysis
用于增材制造熔池监测的事件相机：基准测试与跨机器转移分析
Abstract
Melt-pool monitoring is central to qualifying metal additive manufacturing (AM), yet no public event-camera benchmark exists for this domain. Event cameras report per-pixel brightness changes with microsecond timing instead of reading full frames, giving the temporal resolution AM transients demand at a fraction of the data rate. We present SynAM-E (Synthetic AM Events), the first public multi-source simulated event-camera benchmark for metal-AM melt-pool monitoring: 85 physics-calibrated event shards from 15 sources across 8 institutions, with public baselines and fixed cross-machine evaluation splits. On a single-machine case study, event-spatial monitoring matches dense-frame accuracy (0.874 versus 0.863 macro-F1), and the absolute intensity that events discard adds only +0.006 under fusion. On the NIST Additive Manufacturing Metrology Testbed (AMMT) build, a near-sensor event-rate counter recovers a raw-frame-confirmed 528.7 Hz intensity oscillation at ~380 times less sensor readout than the frame stream requires. A compact 93 k-parameter spiking model runs at 15 times lower modeled inference energy for a 0.073 macro-F1 cost. Every cross-source task includes a built-in trust test against camera identity shortcuts: process-type classification passes while material classification remains confounded by camera band, a corpus-structural limitation the release documents and the trust test exposes.
Chinese Translation
熔池监测是金属增材制造（AM）合格评定的核心，但该领域尚不存在公开的事件相机基准测试。事件相机以微秒级的时间精度报告每个像素的亮度变化，而不是读取完整帧，从而以较低的数据速率提供AM瞬态所需的时间分辨率。我们提出了SynAM-E（合成AM事件），这是首个公开的多源模拟事件相机基准，用于金属AM熔池监测：包含来自8个机构的15个来源的85个物理校准事件片段，提供公共基准和固定的跨机器评估分割。在单机案例研究中，事件空间监测的准确性与密集帧相当（0.874对比0.863宏观F1），而事件丢弃的绝对强度在融合下仅增加了+0.006。在NIST增材制造计量测试平台（AMMT）构建中，近传感器事件速率计在比帧流所需的传感器读取量少约380倍的情况下恢复了经过原始帧确认的528.7 Hz强度振荡。一个紧凑的93 k参数尖峰模型以15倍更低的模型推理能量运行，成本为0.073宏观F1。每个跨源任务都包含针对相机身份捷径的内置信任测试：过程类型分类通过，而材料分类则受到相机波段的干扰，这是发布文档和信任测试所揭示的一个语料库结构限制。
cs.CV / 8 / 2610.06977
Visual-Invariance-Augmented Feature Optimal Alignment for Transferable Adversarial Attacks against Closed-Source MLLMs
针对封闭源多模态大语言模型的可转移对抗攻击的视觉不变性增强特征最优对齐
Abstract
Multimodal large language models (MLLMs) remain vulnerable to transferable adversarial examples, especially in black-box settings where only open-source surrogate models are accessible. Existing targeted transfer attacks mainly align adversarial and target samples using global image-level features, such as encoder [CLS] embeddings. However, such coarse alignment insufficiently exploits patch-level visual structures, limiting transferability across heterogeneous closed-source MLLMs. We propose IAU-FOA, a visual-invariance-augmented feature optimal alignment attack with adaptive unbalanced transport, to improve targeted transferability against closed-source MLLMs. IAU-FOA aligns adversarial and target samples at both global and local levels: a cosine-based objective narrows their global semantic gap, while patch tokens are clustered into compact local patterns and matched through optimal transport for fine-grained feature alignment. Balanced optimal transport enforces fixed marginal masses even for local clusters without reliable counterparts, potentially introducing misleading alignment gradients. We therefore introduce confidence-adaptive unbalanced transport to relax these constraints for weakly matched clusters, aiming to reduce unreliable local alignment and improve adversarial transferability. We further study the effect of input transformations and propose visual-invariance augmentation, which applies bidirectional pixel-intensity rescaling and per-channel white-balance adjustment to simulate exposure, contrast, illumination, and color-temperature variations. This strategy encourages adversarial perturbations to generalize across different visual encoders. Extensive experiments on open-source and closed-source MLLMs show that IAU-FOA consistently outperforms state-of-the-art transferable attack methods. Code is available at https://github.com/jiaxiaojunQAQ/IAU-FOA.
Chinese Translation
多模态大语言模型（MLLMs）在可转移对抗样本方面仍然存在脆弱性，尤其是在仅能访问开源替代模型的黑箱环境中。现有的针对性转移攻击主要通过全局图像级特征（如编码器 [CLS] 嵌入）对抗样本和目标样本进行对齐。然而，这种粗略的对齐方式未能充分利用局部视觉结构，限制了在异构封闭源 MLLMs 之间的可转移性。我们提出了 IAU-FOA，一种增强视觉不变性的特征最优对齐攻击，结合自适应不平衡传输，以提高针对封闭源 MLLMs 的目标可转移性。IAU-FOA 在全局和局部层面对抗样本和目标样本进行对齐：基于余弦的目标函数缩小了它们的全局语义差距，而补丁标记被聚类成紧凑的局部模式，并通过最优传输进行匹配以实现细粒度特征对齐。平衡最优传输即使在没有可靠对应物的局部聚类中也强制固定边际质量，这可能引入误导性的对齐梯度。因此，我们引入了置信度自适应不平衡传输，以放宽这些约束，针对弱匹配聚类，旨在减少不可靠的局部对齐并提高对抗可转移性。我们进一步研究输入变换的影响，并提出视觉不变性增强，应用双向像素强度重标定和每通道白平衡调整，以模拟曝光、对比度、照明和色温变化。这一策略鼓励对抗扰动在不同视觉编码器之间进行泛化。在开源和封闭源 MLLMs 上进行的广泛实验表明，IAU-FOA 始终优于最先进的可转移攻击方法。代码可在 https://github.com/jiaxiaojunQAQ/IAU-FOA 获取。
cs.CV / 9 / 2610.06978
Hierarchy-GBP: Accelerating Factor Graph Inference via Abstraction and Recovery
层次化GBP：通过抽象与恢复加速因子图推理
Abstract
Gaussian Belief Propagation (GBP) is a distributed inference algorithm that passes messages in graphical models, making it attractive for scalable spatial intelligence. However, we find GBP most effective locally: it rapidly smooths message errors that vary sharply between neighbor variables, but corrects global errors across distant graph regions incrementally through long-range message propagations. We propose Hierarchy-GBP (H-GBP), an iterative, two-stage framework that accelerates GBP by first solving these global errors with a coarse graph approximation (abstraction) and projecting the results back to the original graph (recovery), then refining the remaining local errors with GBP. We prove H-GBP convergence to the optimum by deriving the combined matrix operator of our abstraction and recovery steps and analyzing its spectral radius. Experiments on linear sparse graphs show that H-GBP converges fundamentally faster than standard GBP. Moreover, we validate H-GBP on two important spatial problems: Pose Graph Optimization (PGO) and Bundle Adjustment (BA). H-GBP markedly accelerates large-scale PGO and achieves state-of-the-art runtime across all tested BA scales.
Chinese Translation
高斯信念传播（GBP）是一种分布式推理算法，通过在图模型中传递消息，使其在可扩展的空间智能中具有吸引力。然而，我们发现GBP在局部最为有效：它能够迅速平滑邻近变量之间急剧变化的消息误差，但通过长距离消息传播逐步纠正远离图区域的全局误差。我们提出了层次化GBP（H-GBP），这是一种迭代的两阶段框架，首先通过粗略图近似（抽象）解决这些全局误差，并将结果投影回原始图（恢复），然后使用GBP细化剩余的局部误差。我们通过推导抽象和恢复步骤的组合矩阵算子并分析其谱半径，证明了H-GBP收敛到最优解的性质。在线性稀疏图上的实验表明，H-GBP的收敛速度显著快于标准GBP。此外，我们在两个重要的空间问题上验证了H-GBP：姿态图优化（PGO）和束调整（BA）。H-GBP显著加速了大规模PGO，并在所有测试的BA规模中达到了最先进的运行时间。
cs.CV / 10 / 2610.06991
State-Aware Interaction MIL for Rare Joint Molecular Phenotype Prediction in Colorectal Cancer and Lung Adenocarcinoma
状态感知交互MIL在结直肠癌和肺腺癌中稀有联合分子表型预测中的应用
Abstract
Joint molecular phenotype prediction is complicated by small joint-positive populations and overlapping histological features across alternative molecular states. Existing computational pathology approaches typically predict biomarkers independently or formulate the joint-positive phenotype as a binary endpoint. Independent prediction does not model interactions between biomarker-specific histological representations, whereas binary joint prediction collapses the double-negative and two single-positive configurations into a single negative class. We propose State-Aware Interaction MIL, a weakly supervised method that preserves biomarker-specific histological representations, models their interaction, and supervises the complete four-state molecular configuration. We evaluate the proposed approach for joint BRAF+/MSI+ prediction in colorectal cancer and EGFR+/TP53+ prediction in lung adenocarcinoma using frozen UNI2-h and CONCH pathology foundation-model representations. With UNI2-h, State-Aware Interaction MIL achieved an average precision of 0.5566 in colorectal cancer (joint-positive prevalence 6.8%) compared with 0.5161 for NaiveMTL, and 0.2784 in lung adenocarcinoma (joint-positive prevalence 8.6%) compared with 0.2525 for IndependentPair. With CONCH, State-Aware achieved an average precision of 0.4410 compared with 0.3932 for DirectJoint in colorectal cancer and 0.1659 compared with 0.1226 for DirectJoint in lung adenocarcinoma. These results indicate that pathology foundation-model representations contain predictive information for rare joint molecular phenotypes and that preserving biomarker-specific representations within a structured molecular-state formulation can improve prediction of these phenotypes from histopathology.
Chinese Translation
联合分子表型预测因小型联合阳性人群和不同分子状态之间重叠的组织学特征而变得复杂。现有的计算病理学方法通常独立预测生物标志物，或将联合阳性表型构建为二元终点。独立预测未能建模生物标志物特异性组织学表征之间的相互作用，而二元联合预测则将双阴性和两个单阳性配置合并为一个单一的阴性类别。我们提出了状态感知交互MIL（State-Aware Interaction MIL），这是一种弱监督方法，能够保留生物标志物特异性组织学表征，建模它们的相互作用，并监督完整的四状态分子配置。我们评估了该方法在结直肠癌中对BRAF+/MSI+的联合预测，以及在肺腺癌中对EGFR+/TP53+的预测，使用了冻结的UNI2-h和CONCH病理基础模型表征。在UNI2-h下，状态感知交互MIL在结直肠癌中实现了0.5566的平均精度（联合阳性流行率为6.8%），而NaiveMTL为0.5161；在肺腺癌中实现了0.2784的平均精度（联合阳性流行率为8.6%），而IndependentPair为0.2525。在CONCH下，状态感知交互MIL在结直肠癌中的平均精度为0.4410，而DirectJoint为0.3932；在肺腺癌中的平均精度为0.1659，而DirectJoint为0.1226。这些结果表明，病理基础模型表征包含稀有联合分子表型的预测信息，并且在结构化的分子状态表述中保留生物标志物特异性表征可以改善从组织病理学中对这些表型的预测。
cs.CV / 11 / 2610.07008
Learning to Curate What You Generate for Generalizable Few-Shot Class-Incremental Learning
学习策划生成内容以实现可泛化的少样本类增量学习
Abstract
Few-shot class-incremental learning (FSCIL) aims to learn novel classes from limited annotations while preserving prior knowledge. Existing methods typically assume a sufficiently large base session, but this assumption fails when both base and incremental data are scarce, leading to weak initial representations, semantic drift, and unstable boundaries. We study this underexplored yet realistic setting, termed Generalizable FSCIL (G-FSCIL), where the base session itself contains only a few classes. Although synthetic data can alleviate supervision scarcity, naively mixing generated samples often introduces semantic noise and exacerbates old-new boundary conflicts. To address this, we propose a framework that curates trustworthy synthetic knowledge for stable G-FSCIL. Specifically, we first construct class-specific synthetic candidate pools using a frozen latent diffusion model, where class inversion is performed at the first observation and the resulting condition embeddings are reused for on-demand generation. Building on these candidates, we learn a knowledge curation strategy that selects samples with both semantic consistency and visual diversity, and distill this process into a transferable selection policy during the base session, which is then reused without further optimization. Leveraging the curated synthetic data, we further design a boundary-stable incremental adaptation scheme, including synthetic-informed prototype initialization and bidirectional boundary calibration to mitigate old-new conflicts. Extensive experiments demonstrate that our method consistently outperforms existing FSCIL baselines, with reduced forgetting and improved balance between old and new classes. Code is available at https://github.com/NiHaoWoJiaoYYC/G-FSCIL.
Chinese Translation
少样本类增量学习（FSCIL）旨在从有限的标注中学习新类别，同时保留先前的知识。现有方法通常假设基础会话具有足够大的数据量，但当基础数据和增量数据都稀缺时，这一假设失效，导致初始表示薄弱、语义漂移和边界不稳定。我们研究这一尚未充分探讨但现实的情境，称为可泛化的少样本类增量学习（G-FSCIL），其中基础会话本身仅包含少数类别。尽管合成数据可以缓解监督稀缺的问题，但简单地混合生成样本往往会引入语义噪声并加剧旧新边界冲突。为了解决这一问题，我们提出了一个框架，用于策划可信的合成知识，以实现稳定的G-FSCIL。具体而言，我们首先使用冻结的潜在扩散模型构建类别特定的合成候选池，其中在第一次观察时进行类别反演，并重用生成条件嵌入以按需生成。在这些候选基础上，我们学习一种知识策划策略，选择具有语义一致性和视觉多样性的样本，并在基础会话期间将这一过程提炼为可转移的选择策略，随后在不进一步优化的情况下重复使用。利用策划的合成数据，我们进一步设计了一种边界稳定的增量适应方案，包括合成信息驱动的原型初始化和双向边界校准，以减轻旧新冲突。大量实验证明，我们的方法在减少遗忘和改善旧类别与新类别之间的平衡方面，始终优于现有的FSCIL基线。代码可在 https://github.com/NiHaoWoJiaoYYC/G-FSCIL 获取。
cs.CV / 12 / 2610.07014
DTFormer: Text-Guided Semantic Alignment for RGB-D Segmentation
DTFormer：基于文本引导的RGB-D分割语义对齐
Abstract
RGB-D semantic segmentation has made notable progress by fusing RGB and Depth, yet mainstream models still learn features almost exclusively from pixel-level supervision, lacking direct high-level semantic constraints. This raises a central question-can external knowledge such as language priors inject stronger semantic discriminability into mainstream RGB-D segmentation models. We present DTFormer, a novel tri-modal (RGB-D-Text) semantic segmentation framework. At its core is Text-guided Semantic Alignment Module (TSAM) that first encodes textual cues into a set of semantic prototypes and then explicitly aligns multi-modal RGB-D features with these prototypes at multiple encoder and decoder layers. This design imposes strong semantic regularization on representation learning, guiding the network toward more discriminative features. Extensive experiments on multiple benchmarks show that DTFormer delivers consistent gains while remaining simple and efficient. Our results demonstrate that explicit semantic alignment offers an effective and practical route to improving RGB-D semantic segmentation. The code will be released upon acceptance.
Chinese Translation
RGB-D语义分割通过融合RGB和深度信息取得了显著进展，但主流模型仍然几乎完全依赖于像素级监督来学习特征，缺乏直接的高层语义约束。这引发了一个核心问题——外部知识如语言先验是否能够为主流RGB-D分割模型注入更强的语义区分能力。我们提出了DTFormer，一种新颖的三模态（RGB-D-文本）语义分割框架。其核心是文本引导的语义对齐模块（Text-guided Semantic Alignment Module, TSAM），该模块首先将文本线索编码为一组语义原型，然后在多个编码器和解码器层中显式地将多模态RGB-D特征与这些原型对齐。这一设计对表征学习施加了强有力的语义正则化，引导网络朝向更具区分性的特征。我们在多个基准测试上的广泛实验表明，DTFormer在保持简单高效的同时，持续带来性能提升。我们的结果表明，显式的语义对齐为改善RGB-D语义分割提供了一条有效且实用的途径。代码将在接受后发布。
cs.CV / 13 / 2610.07016
Anchor and Adapt: Asymmetric Prompt Adaptation for Few-Shot Industrial Anomaly Detection
锚点与适应：用于少样本工业异常检测的非对称提示适应
Abstract
In few-shot industrial anomaly detection, the few normal target images provide no direct defect supervision, making anomaly prompts difficult to learn from these samples alone. Some vision-language methods therefore use manually specified descriptions to supply explicit anomaly semantics. However, constructing these descriptions requires product-specific effort, and their effectiveness depends on prompt selection. We propose Anchor and Adapt, a two-stage prompt learning framework that separates the acquisition of anomaly semantics from adaptation to target normal appearance. Stage I learns transferable normal and abnormal anchors from annotated auxiliary data. Stage II keeps these anchors fixed and adapts an additional normal branch using the few target normal samples. The inherited and adapted normal branches jointly characterize target normality, with text-anchor regularization encouraging consistency with the generic normal prior and separation from the abnormal anchors. This design retains learned anomaly knowledge while reducing dependence on category-specific anomaly templates, without requiring synthetic anomaly generation. Cross-dataset experiments between MVTec-AD and VisA under 1-, 2-, and 4-shot settings demonstrate competitive detection and localization performance. Controlled ablations assess the roles of transferred anchors, asymmetric adaptation, dual-normal representations, and anchor regularization.
Chinese Translation
在少样本工业异常检测中，少量正常目标图像无法提供直接的缺陷监督，使得仅从这些样本中学习异常提示变得困难。因此，一些视觉-语言方法使用手动指定的描述来提供明确的异常语义。然而，构建这些描述需要针对特定产品的努力，其有效性依赖于提示选择。我们提出了锚点与适应（Anchor and Adapt），一个两阶段的提示学习框架，旨在将异常语义的获取与对目标正常外观的适应分开。第一阶段从标注的辅助数据中学习可转移的正常和异常锚点。第二阶段保持这些锚点不变，并利用少量目标正常样本适应一个额外的正常分支。继承的和适应的正常分支共同表征目标正常性，文本锚点正则化鼓励与通用正常先验的一致性，并与异常锚点分离。该设计保留了学习到的异常知识，同时减少了对类别特定异常模板的依赖，无需合成异常生成。在1-shot、2-shot和4-shot设置下，MVTec-AD与VisA之间的跨数据集实验展示了竞争性的检测和定位性能。受控消融实验评估了转移锚点、非对称适应、双正常表示和锚点正则化的作用。
cs.CV / 14 / 2610.07025
WiSPER: Pose-Supervised Predictive and Residual Flow Refinement For Multi-Person 3D Pose Estimation With WiFi CSI
WiSPER：基于姿态监督的预测与残差流细化用于多人物3D姿态估计与WiFi CSI
Abstract
Multi-person 3D pose estimation with WiFi channel state information (CSI) is challenging because reflections from different people overlap without directly identifying individual joints. Existing masked embedding objectives capture wireless relationships without explicit pose supervision, while structured decoders can retain coordinate errors. We propose WiSPER, a two-stage framework combining pose-aware predictive pretraining with conditional residual flow refinement. Pose-Aware Masked Embedding Learning (PAMEL) couples masked latent prediction with auxiliary pose-set supervision on the same CSI context, guiding the encoder toward joint localization from partial observations. Residual Flow refinement with Transformer (ReFT) generates a set of pose candidates to accommodate a variable number of people and refines each candidate through a conditional flow guided by its coarse coordinates and per-joint decoder features. Both stages use paired CSI and pose annotations during training, while inference requires only CSI. Experiments on the PiW3D dataset show that WiSPER achieves an overall mean per-joint position error of 63.72 mm, a 40.0% reduction relative to WiFi-JEPA. For experiments with two and three people, WiSPER reduces MPJPE by 42.1% and 38.1%, respectively. Pose-supervised pretraining configurations obtain lower errors than CSI-only JEPA, and enabling the trained residual refiner reduces overall MPJPE by 13.8-15.6% across the evaluated configurations.
Chinese Translation
利用WiFi信道状态信息（CSI）进行多人物3D姿态估计是一项具有挑战性的任务，因为来自不同人的反射信号会重叠，无法直接识别个体关节。现有的掩码嵌入目标捕捉无线关系，但缺乏明确的姿态监督，而结构化解码器可能保留坐标误差。我们提出了WiSPER，一个两阶段框架，结合了姿态感知的预测预训练与条件残差流细化。姿态感知掩码嵌入学习（PAMEL）将掩码潜在预测与相同CSI上下文中的辅助姿态集监督相结合，引导编码器从部分观测中进行关节定位。带有Transformer的残差流细化（ReFT）生成一组姿态候选，以适应可变人数，并通过条件流对每个候选进行细化，该条件流由其粗略坐标和每个关节解码器特征引导。两个阶段在训练过程中都使用配对的CSI和姿态注释，而推理仅需CSI。在PiW3D数据集上的实验表明，WiSPER实现了63.72毫米的整体平均每关节位置误差，相较于WiFi-JEPA减少了40.0%。在两人和三人的实验中，WiSPER分别减少了42.1%和38.1%的MPJPE。姿态监督的预训练配置获得的误差低于仅使用CSI的JEPA，并且启用训练好的残差细化器在评估的配置中将整体MPJPE减少了13.8-15.6%。
cs.CV / 15 / 2610.07031
Artemis: Geometry-Grounded Multi-Agent Driving World Models with Shared 3D State and Progressive Memory Update
阿耳忒弥斯：基于几何的多智能体驾驶世界模型，具有共享的3D状态和渐进式记忆更新
Abstract
Recent video world models have witnessed the paradigm shift from single-agent to multi-agent involvements, which can reveal more complicated dynamics and cross-agent interaction in the real world. However, existing approaches commonly adopt implicit inter-agent communications via cross attention, which lack explicit geometry constraints and unified 3D state, thereby leading to poor multi-view consistency and struggling with recovering out-of-sight agents. In addition, most of them assume a static background, failing to represent uncontrolled background dynamics. To address these problems, we propose Artemis: a geometry-grounded multi-agent world model with explicit memory sharing. An explicit 3D world map is reconstructed from multi-agent observations to enforce a unified 3D state across agents, offering high cross-view consistency. Specifically, an action-guided geometric injection module is developed to simultaneously render decomposed foreground-background control maps, which are then injected into a diffusion transformer through a designed GeoAdapter block. Compared to previous methods assuming static-only background, our GeoAdapter can also distinguish uncontrolled non-agent dynamics, which are conditioned on their own multi-frame history positions to provide consistent motion cues. Keyframes selected from progressive video rollouts are used to progressively update the reconstructed 3D world maps. To effectively capture complex dynamic patterns, we curate a novel dataset sampled from the CARLA simulator called MA-CARLA. Extensive experiments demonstrate the superiority of our proposed method in terms of visual fidelity and cross-view consistency in the generated videos. In addition, our Artemis can support simultaneous multi-modal rollouts with both 2D video and 3D point map maintenance, scale to scenarios beyond two agents and multi-camera setting.
Chinese Translation
近年来，视频世界模型经历了从单智能体到多智能体参与的范式转变，这能够揭示现实世界中更复杂的动态和跨智能体的互动。然而，现有的方法通常通过交叉注意力采用隐式的智能体间通信，缺乏明确的几何约束和统一的3D状态，从而导致多视角一致性差，并在恢复视野外的智能体时遇到困难。此外，大多数方法假设背景是静态的，未能表示不受控制的背景动态。为了解决这些问题，我们提出了阿耳忒弥斯：一种基于几何的多智能体世界模型，具有明确的记忆共享。通过多智能体观察重建明确的3D世界地图，以强制在智能体之间实现统一的3D状态，从而提供高跨视角一致性。具体而言，我们开发了一个基于动作引导的几何注入模块，能够同时渲染分解的前景-背景控制图，然后通过设计的GeoAdapter模块将其注入扩散变换器。与假设静态背景的先前方法相比，我们的GeoAdapter还能够区分不受控制的非智能体动态，这些动态基于自身的多帧历史位置提供一致的运动线索。从渐进式视频展开中选择的关键帧用于逐步更新重建的3D世界地图。为了有效捕捉复杂的动态模式，我们从CARLA模拟器中策划了一个新数据集，称为MA-CARLA。大量实验表明，我们提出的方法在生成视频的视觉保真度和跨视角一致性方面具有优越性。此外，我们的阿耳忒弥斯能够支持同时的多模态展开，同时维护2D视频和3D点图，扩展到超过两个智能体和多摄像头设置的场景。
cs.CV / 16 / 2610.07059
Crop Yield Prediction for Punjab, Pakistan: A Tree-Ensemble and Leaf-Health Prototype, and What Random Validation Hides
巴基斯坦旁遮普地区作物产量预测：树集成与叶片健康原型，以及随机验证所隐藏的内容
Abstract
Yield forecasts help planners and farmers decide on inputs, storage and imports, but small agricultural tables can make reported accuracy fail on a new season. We built a crop-yield prototype for Punjab, Pakistan that combines Random Forest, XGBoost, support vector regression and a Ridge-stacked ensemble with a MobileNetV2 leaf-health classifier, and deployed it as a web application with SHAP explanations. A random 80/20 split of a merged Kaggle-derived table (414 rows) gives the ensemble an R2 of 0.991. An audit showed that the table contains only 46 independent observations: a join with nine temperature records per year copied every crop-year nine times. Holding out whole years takes XGBoost on the same rows from R2 = 0.994 to -0.20. On deduplicated data, and on a longer FAOSTAT table (1990-2024, 70 observations), a per-crop linear trend (leave-one-year-out RMSE 0.29 Ton/Ha) beats every model not given the year (0.84-1.06). Neither pesticide use nor national temperature change explains the trend residuals. An apparent pesticide gain in the Kaggle table disappears on FAOSTAT, where the pesticide series is mostly imputed and the two releases disagree. An independent district-level wheat panel (36 districts, 13 seasons) shows that most variation in Punjab is spatial and that a district mean with a common trend matches the learned models. The leaf classifier reaches 99.87% accuracy on held-out PlantVillage images and recalls 96.4% of 336 unseen diseased leaves after near-duplicates were removed. However, no leaf image is paired with a yield record, so the health score used in the yield models had to be constructed and adds nothing. We report these negative findings together with the prototype.
Chinese Translation
产量预测帮助规划者和农民决定投入、储存和进口，但小型农业数据表可能导致报告的准确性在新季节中失效。我们为巴基斯坦旁遮普地区构建了一个作物产量原型，结合了随机森林（Random Forest）、XGBoost、支持向量回归（support vector regression）和一个与MobileNetV2叶片健康分类器相结合的岭回归堆叠集成，并将其部署为具有SHAP解释的网络应用。对合并的Kaggle衍生表（414行）进行随机80/20拆分，使得该集成的R²达到0.991。审计显示该表仅包含46个独立观测值：每个作物年份复制了九次的每年九个温度记录的连接。将整年数据保留在外，使得XGBoost在相同行上的R²从0.994降至-0.20。在去重数据上，以及在更长的FAOSTAT表（1990-2024，70个观测值）上，按作物的线性趋势（留一年度外均方根误差0.29吨/公顷）优于未给出年份的所有模型（0.84-1.06）。无论是农药使用还是国家温度变化都无法解释趋势残差。在Kaggle表中明显的农药增益在FAOSTAT中消失，因为农药系列大部分是插补的，且两个版本不一致。一个独立的地区小麦面板（36个地区，13个季节）显示，旁遮普的大部分变异是空间性的，具有共同趋势的地区均值与学习模型相匹配。叶片分类器在保留的PlantVillage图像上达到了99.87%的准确率，并在去除近重复后召回了336个未见病叶的96.4%。然而，没有任何叶片图像与产量记录配对，因此在产量模型中使用的健康评分必须构建，并没有增加任何信息。我们报告这些负面发现以及原型。
cs.CV / 17 / 2610.07067
Decomposition-Guided Curvelet Thresholding for Sharp-to-Soft CT Kernel Conversion
基于分解引导的曲波阈值处理用于锐软CT核转换
Abstract
Image denoising is a crucial task in image processing, focused on improving image quality by minimizing noise while maintaining essential structural elements. This study presents a hybrid denoising framework that combines several decomposition techniques, including empirical mode decomposition (EMD), variational mode decomposition (VMD), multichannel EMD (MEMD), and bidimensional EMD (BEMD), with curvelet transform thresholding. Each decomposition mode undergoes processing through both soft and hard thresholding, and the denoised modes are combined to rebuild the final image. Comprehensive evaluations of standard CT image datasets reconstructed with various kernels (B50, B46, B41, B36) reveal substantial enhancements in denoising efficacy. VMD consistently achieves the highest peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), signifying exceptional noise reduction and feature preservation. The study analyses the trade-offs between soft and hard thresholding: soft thresholding maintains intricate visual details, whilst harsh thresholding provides enhanced noise reduction. The suggested method surpasses traditional techniques in both reference and non-reference quality criteria, indicating its potential for broader application in medical imaging and future incorporation with adaptive thresholding algorithms.
Chinese Translation
图像去噪是图像处理中的一项关键任务，旨在通过最小化噪声来提高图像质量，同时保持重要的结构元素。本研究提出了一种混合去噪框架，结合了多种分解技术，包括经验模态分解（EMD）、变分模态分解（VMD）、多通道EMD（MEMD）和二维EMD（BEMD），以及曲波变换阈值处理。每种分解模式都经过软阈值和硬阈值处理，去噪后的模式被组合以重建最终图像。对使用不同核（B50、B46、B41、B36）重建的标准CT图像数据集进行的全面评估显示去噪效果显著提升。VMD始终实现最高的峰值信噪比（PSNR）和结构相似性指数（SSIM），表明其在噪声减少和特征保留方面表现卓越。本研究分析了软阈值和硬阈值之间的权衡：软阈值保持复杂的视觉细节，而硬阈值则提供更强的噪声减少。所建议的方法在参考和非参考质量标准上均优于传统技术，表明其在医学成像中的广泛应用潜力以及未来与自适应阈值算法结合的可能性。
cs.CV / 18 / 2610.07071
A BEMD-Based Quaternion Filtering Approach Sharp-to-Soft Kernel CT Image Conversion
基于BEMD的四元数滤波方法在锐核CT图像转换中的应用
Abstract
The quality of computed tomography (CT) images is significantly affected by the selection of reconstruction kernels: sharp kernels improve spatial resolution but increase noise, whereas soft kernels diminish noise at the expense of edge clarity. This study presents an innovative enhancement framework utilising Bidimensional Empirical Mode Decomposition in conjunction with Quaternion Bilateral Filtering (BEMD--QBF) to convert sharp-kernel CT images into representations resembling soft-kernels, while maintaining critical anatomical structures. The technique disaggregates each image into intrinsic mode functions via BEMD and analyzes them inside a cohesive quaternion framework to attain efficient noise reduction and structural integrity. The proposed methodology is evaluated using several reconstruction kernels (B50, B46, B41, B36, B35, B31) and compared with recognised filtering strategies, including Non-Local Means, Anisotropic Diffusion, Bilateral Filtering, and Quaternion Bilateral Filtering. Quantitative evaluations of the Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR) indicate that BEMD-QBF consistently attains superior structural fidelity and competitive noise reduction across all evaluated kernels. The results underscore the efficacy of the proposed strategy as a viable approach to enhancing post-reconstruction CT images, yielding superior image quality without requiring access to raw projection data.
Chinese Translation
计算机断层扫描（CT）图像的质量受到重建核选择的显著影响：锐核提高空间分辨率但增加噪声，而软核则在牺牲边缘清晰度的情况下减少噪声。本研究提出了一种创新的增强框架，利用双维经验模态分解（Bidimensional Empirical Mode Decomposition, BEMD）结合四元数双边滤波（Quaternion Bilateral Filtering, QBF），将锐核CT图像转换为类似软核的表示，同时保持关键解剖结构。该技术通过BEMD将每幅图像分解为固有模态函数，并在一个统一的四元数框架内进行分析，以实现高效的噪声降低和结构完整性。所提方法使用多种重建核（B50、B46、B41、B36、B35、B31）进行评估，并与公认的滤波策略进行比较，包括非局部均值（Non-Local Means）、各向异性扩散（Anisotropic Diffusion）、双边滤波（Bilateral Filtering）和四元数双边滤波（Quaternion Bilateral Filtering）。对结构相似性指数（Structural Similarity Index, SSIM）和峰值信噪比（Peak Signal-to-Noise Ratio, PSNR）的定量评估表明，BEMD-QBF在所有评估的核中始终实现了优越的结构保真度和具有竞争力的噪声降低。结果强调了所提策略作为增强后重建CT图像的一种可行方法的有效性，能够在不需要访问原始投影数据的情况下，产生更优质的图像。
cs.CV / 19 / 2610.07072
On Color Alignment in VAE Latent Spaces and Its Applications
变分自编码器潜在空间中的颜色对齐及其应用
Abstract
Variational autoencoders (VAEs) are a key part of modern text-to-image models, which generate images within their latent space. VAEs are known to disentangle the main factors of variation in the data, and color is known to be one of the most structured of these in natural images: decorrelating it yields one luminance axis and two opponent-color axes. Color should therefore be expected to emerge as a distinct factor in the VAE latent space. Yet how these latent spaces represent color remains largely unexplored. In this work, we show that the VAEs of text-to-image models share a color subspace aligned with brightness and opponent-colors. Through a linear approximation of the encoder and targeted latent steering, we find this subspace consistently across a broad range of VAEs, from SD1.5 to FLUX.2 and Z-Image. Building on this characterization, we propose three applications: ColorTuning, which achieves state-of-the-art in precise numerical color generation on the fine-grained CSS3/X11 system of GenColorBench, saturation control, to adjust the global chromatic intensity, and color transfer, to change the palette to match a reference. The code and models are publicly available at https://julian075.github.io/Color_Subspace/
Chinese Translation
变分自编码器（Variational Autoencoders, VAEs）是现代文本到图像模型的关键组成部分，它们在潜在空间中生成图像。VAEs以其能够解耦数据中的主要变化因素而闻名，而颜色被认为是自然图像中最具结构性的变化因素之一：对其进行去相关处理会产生一个亮度轴和两个对立颜色轴。因此，颜色应当被预期在VAE潜在空间中作为一个独立因素出现。然而，这些潜在空间如何表示颜色仍然 largely 未被探索。在本研究中，我们展示了文本到图像模型的VAEs共享一个与亮度和对立颜色对齐的颜色子空间。通过对编码器的线性近似和针对性的潜在引导，我们在从SD1.5到FLUX.2和Z-Image的广泛VAEs中一致地发现了这个子空间。基于这一特征，我们提出了三个应用：ColorTuning，它在GenColorBench的精细CSS3/X11系统上实现了最先进的精确数值颜色生成，饱和度控制，用于调整全局色度强度，以及颜色转移，用于更改调色板以匹配参考色。代码和模型可在 https://julian075.github.io/Color_Subspace/ 上公开获取。
cs.CV / 20 / 2610.07083
Graph-Based Recognition of Simulated Train-Driver States From Facial and Upper-Body Keypoints
基于图的模拟列车驾驶员状态识别：来自面部和上半身关键点的研究
Abstract
Driver fatigue poses a significant challenge to railway safety, with traditional systems like the dead-man switch offering limited and basic alertness checks. This study presents a vision-based monitoring system that relies solely on a single front-facing RGB camera and a graph neural network to classify simulated train-driver states into alert, not-alert, and an emergency class comprising acted emergency-like behaviours. To optimize input representations for the model, an ablation study was performed, comparing three feature configurations: skeletal-only, facial-only, and a combination of both. Experimental results show that combining facial and skeletal features yields the highest accuracy (81%) for the three-class model under the light condition, outperforming models that use only facial or skeletal features. Furthermore, the combination of facial and skeletal features achieves 99% accuracy in the alert/not alert classification in light condition. Additionally, we introduced a controlled RGB video dataset containing alert, not alert, and acted emergency-like behaviours recorded under three illumination conditions. These contributions represent a step toward passive and non-contact train-driver state recognition based on facial and upper-body dynamics.
Chinese Translation
驾驶员疲劳对铁路安全构成了重大挑战，传统系统如死人开关提供的警觉性检查有限且基本。本研究提出了一种基于视觉的监测系统，该系统仅依赖于一个正面RGB摄像头和图神经网络，将模拟列车驾驶员状态分类为警觉、未警觉和包含模拟紧急行为的紧急类别。为了优化模型的输入表示，进行了消融研究，比较了三种特征配置：仅骨架、仅面部以及两者的组合。实验结果表明，在光照条件下，结合面部和骨架特征的三类模型实现了最高准确率（81%），优于仅使用面部或骨架特征的模型。此外，面部和骨架特征的组合在光照条件下的警觉/未警觉分类中达到了99%的准确率。此外，我们引入了一个受控的RGB视频数据集，包含在三种照明条件下记录的警觉、未警觉和模拟紧急行为。这些贡献代表了基于面部和上半身动态的被动非接触式列车驾驶员状态识别的一个进步。
cs.CV / 21 / 2610.07087
A Data-Centric Review of Plant Disease Datasets: Taxonomy, Critical Analysis, Environmental Variability, and Implications for Precision Agriculture
植物病害数据集的数据中心评估：分类、关键分析、环境变异性及其对精准农业的影响
Abstract
Despite rapid advances in artificial intelligence, reliable real-world plant disease detection remains a persistent challenge. Visual and deep learning approaches have shown promising results, but their deployment under field conditions remains limited. A key bottleneck is the reliance on laboratory-generated datasets that lack environmental diversity, realistic backgrounds, and balanced class distributions, resulting in poor generalization. In contrast, datasets collected directly from agricultural environments capture natural variability and better reflect challenges faced by farmers across regions. This review presents a critical analysis of visual and deep learning approaches for plant disease detection, with emphasis on plant disease datasets. It establishes a taxonomy based on acquisition setting, accessibility, plant diversity, disease composition, class structure, and imbalance severity, and examines their implications for model generalization and real-world deployment. A comparative analysis of laboratory and real-field datasets identifies critical gaps that hinder disease detection. The review further analyzes how multi-level dataset imbalance, including intra-class, inter-crop, and cross-dataset imbalance, and limited environmental variability affect model performance and robustness, an area insufficiently examined in existing surveys. Beyond image-based approaches, it highlights the importance of integrating environmental parameters such as temperature, humidity, and leaf wetness with image data to improve prediction under dynamic field conditions. Finally, the review identifies key challenges, research gaps, and future directions concerning dataset construction, environmental variability, structural imbalance, standardization, and multimodal disease monitoring. It provides a foundation for developing next-generation multimodal frameworks for precision agriculture.
Chinese Translation
尽管人工智能迅速发展，可靠的现实世界植物病害检测仍然是一个持续的挑战。视觉和深度学习方法已显示出良好的结果，但在田间条件下的应用仍然有限。一个关键瓶颈是依赖于缺乏环境多样性、现实背景和均衡类别分布的实验室生成数据集，导致模型泛化能力差。相比之下，直接从农业环境中收集的数据集捕捉了自然变异性，更好地反映了农民在不同地区面临的挑战。本文对植物病害检测的视觉和深度学习方法进行了关键分析，重点关注植物病害数据集。我们根据获取设置、可及性、植物多样性、病害组成、类别结构和不平衡严重性建立了一个分类体系，并考察了这些因素对模型泛化和现实世界部署的影响。实验室数据集与实际田间数据集的比较分析识别了阻碍病害检测的关键差距。本文进一步分析了多层次数据集不平衡，包括类内、作物间和跨数据集的不平衡，以及有限的环境变异性如何影响模型性能和鲁棒性，这是现有调查中未充分研究的领域。除了基于图像的方法，本文强调了将温度、湿度和叶片湿度等环境参数与图像数据相结合的重要性，以提高在动态田间条件下的预测能力。最后，本文识别了数据集构建、环境变异性、结构不平衡、标准化和多模态病害监测方面的关键挑战、研究空白和未来方向，为开发下一代精准农业多模态框架奠定了基础。
cs.CV / 22 / 2610.07110
MoonGS: High-quality Representation of the Lunar Surface via Gaussian Splatting Using Robust Depth Features from Image Pairs
MoonGS：通过高斯溅射利用图像对中的鲁棒深度特征实现月球表面的高质量表示
Abstract
High-quality 3D reconstruction of lunar terrain from sparse rover images is indispensable for autonomous lunar exploration, but remains challenging because viewpoint overlap is insufficient, surface textures are weak, and data volume is limited. We propose MoonGS, the first feed-forward 3D Gaussian Splatting framework tailored to lunar scenes. Given only two input images, MoonGS predicts pixel-aligned Gaussian primitives in a single forward pass and renders photorealistic novel views without any per-scene optimization. MoonGS (i) adopts an adaptable backbone design that seamlessly integrates advanced vision foundation models to extract robust depth features; (ii) integrates semantic priors in two manners: merging semantic cues with visual features to refine Gaussian parameter estimation, and adopting a semantic ranking loss that regularizes background depth; and (iii) employs an entropy-guided heuristic resampling strategy to augment sparse observations by selecting the most informative distant viewpoints with negligible overhead. Experiments on the LuSNAR benchmark and our synthetic weak-texture MoonBlender dataset show that MoonGS surpasses state-of-the-art feed-forward NeRF/3DGS baselines by +4.9 dB PSNR, +0.29 SSIM, and 40\% lower LPIPS while maintaining sub-second inference. Furthermore, we validate the broad applicability of our framework by demonstrating that it effectively leverages state-of-the-art backbones, including VGGT, to significantly boost performance. Qualitative evaluations on Chang'e mission imagery also show the best visual quality among compared methods, indicating robustness on real lunar data. The source code and dataset are publicly available at https://github.com/InRobots/MoonBlender.
Chinese Translation
从稀疏探测器图像中高质量地重建月球地形对于自主月球探索至关重要，但由于视角重叠不足、表面纹理较弱以及数据量有限，这一任务仍然具有挑战性。我们提出了MoonGS，这是第一个针对月球场景设计的前馈3D高斯溅射框架。仅给定两幅输入图像，MoonGS在单次前向传递中预测像素对齐的高斯原语，并在没有任何每场景优化的情况下渲染出逼真的新视图。MoonGS (i) 采用可适应的主干设计，能够无缝集成先进的视觉基础模型以提取鲁棒的深度特征；(ii) 以两种方式整合语义先验：将语义线索与视觉特征合并以细化高斯参数估计，并采用语义排序损失来规范化背景深度；(iii) 采用基于熵的启发式重采样策略，通过选择最具信息量的远处视点来增强稀疏观测，且开销微乎其微。我们在LuSNAR基准和我们的合成弱纹理MoonBlender数据集上的实验表明，MoonGS在保持亚秒推理的同时，超越了最先进的前馈NeRF/3DGS基线，PSNR提高了+4.9 dB，SSIM提高了+0.29，LPIPS降低了40%。此外，我们通过展示MoonGS有效利用最先进的主干（包括VGGT）显著提升性能，验证了我们框架的广泛适用性。对嫦娥任务图像的定性评估也显示出在比较方法中最佳的视觉质量，表明其在真实月球数据上的鲁棒性。源代码和数据集可在https://github.com/InRobots/MoonBlender公开获取。
cs.CV / 23 / 2610.07117
R2RI: A Multi-View Event and RGB Dataset for Robot-to-Robot Interaction
R2RI：一个用于机器人间互动的多视角事件和RGB数据集
Abstract
Understanding and modeling interactions between autonomous agents is a fundamental challenge in robotics, with broad implications for collaborative systems, social robotics, and human-robot coexistence. Although the study of robot interactions has emerged as a compelling research direction, progress has been severely hampered by the absence of large-scale benchmarks. In this paper, we introduce Robot-to-Robot Interaction (R2RI), the first dataset specifically designed to address the Robot-Robot Interaction (RRI) task. R2RI consists of different humanoid robots and realistic interactions modeled on real human social behaviors. Complementary viewpoints are available, \textit{i.e.}, an egocentric perspective from each robot's onboard sensors, and an exocentric perspective from external fixed cameras, thus enabling rich spatial and contextual understanding of the interaction dynamics. The dataset comprises more than $6.5$M frames and $\approx5000$ videos at $120$ fps, including Event and RGB domains. We investigate pros and cons of each domain, comparing state-of-the-art approaches for a number of key sensing and interaction based tasks. We publicly release the dataset and its annotations for all tasks and modalities at https://github.com/MagriniGabriele/R2RI.
Chinese Translation
理解和建模自主智能体之间的互动是机器人技术中的一项基本挑战，对协作系统、社会机器人以及人机共存具有广泛的影响。尽管机器人互动的研究已成为一个引人注目的研究方向，但由于缺乏大规模基准，进展受到严重阻碍。本文介绍了机器人间互动（Robot-to-Robot Interaction, R2RI），这是第一个专门设计用于解决机器人-机器人互动（Robot-Robot Interaction, RRI）任务的数据集。R2RI包含不同的人形机器人和基于真实人类社会行为建模的真实互动。数据集中提供了互补的视角，即来自每个机器人的机载传感器的自我中心视角，以及来自外部固定摄像头的外部视角，从而使得对互动动态的空间和上下文理解更加丰富。该数据集包含超过650万帧和约5000个以120帧每秒（fps）录制的视频，包括事件和RGB领域。我们研究了每个领域的优缺点，并比较了多种关键感知和互动任务的最先进方法。我们在https://github.com/MagriniGabriele/R2RI上公开发布了该数据集及其所有任务和模态的注释。
cs.CV / 24 / 2610.07127
PlaySuite: A Large-Scale Benchmark for Interactive Visual Intelligence
PlaySuite：一个大规模的交互视觉智能基准测试
Abstract
Recent advances in multimodal foundation models yield strong performance on static perception and reasoning benchmarks, yet such evaluations largely overlook a central aspect of intelligence: acting competently in dynamic environments over extended time horizons. We introduce PlaySuite, a large-scale benchmark for evaluating interactive visual intelligence across more than 5K open-source video games curated from PyWeek and itch.io. Spanning diverse genres and engines, including Pygame, HTML5, Godot, and Unity, these independent games are largely out-of-distribution for current models, reducing the likelihood that success can be achieved by retrieving memorized walkthroughs or web-scale training artifacts. To enable scalable evaluation across heterogeneous titles, we develop a unified closed-loop interaction framework optimized for HPC clusters alongside a Video-LLM-as-a-judge protocol that maps observable gameplay milestones to standardized progress levels. We evaluate fourteen recent open models spanning vision-language models, computer-use agents, and vision-language-action models. Our results yield strong evidence of a perception-action gap: despite strong reasoning capabilities, current models struggle to make sustained progress and exhibit recurring failures in spatial grounding, action execution, and self-correction. PlaySuite provides a reproducible and extensible testbed for measuring progress from visual perception to goal-directed interaction, and a foundation for developing models that can act, adapt, and generalize in dynamic visual environments.
Chinese Translation
近期多模态基础模型的进展在静态感知和推理基准测试中表现出色，但这些评估在很大程度上忽视了智能的一个核心方面：在动态环境中在较长时间范围内有效行动。我们提出了PlaySuite，一个用于评估交互视觉智能的大规模基准测试，涵盖了来自PyWeek和itch.io的5000多款开源视频游戏。这些独立游戏跨越了多种类型和引擎，包括Pygame、HTML5、Godot和Unity，且在很大程度上超出了当前模型的分布范围，从而降低了通过检索记忆中的攻略或网络规模训练工件获得成功的可能性。为了实现跨异构标题的可扩展评估，我们开发了一个统一的闭环交互框架，针对高性能计算集群进行了优化，并提出了一种Video-LLM-as-a-judge协议，将可观察的游戏里程碑映射到标准化的进度水平。我们评估了十四个近期的开放模型，包括视觉-语言模型、计算机使用代理和视觉-语言-行动模型。我们的结果强烈表明存在感知-行动差距：尽管推理能力强大，但当前模型在持续进展方面存在困难，并在空间定位、行动执行和自我修正方面表现出反复失败。PlaySuite提供了一个可重复和可扩展的测试平台，用于测量从视觉感知到目标导向交互的进展，并为开发能够在动态视觉环境中行动、适应和泛化的模型奠定基础。
cs.CV / 25 / 2610.07175
CALR: Continuous Anchored Latent Reasoning via Render-of-Thought Compression
CALR：通过思维渲染压缩实现连续锚定潜在推理
Abstract
Visual latent reasoning compresses rendered derivations into compact intermediate states, reducing textual reasoning overhead. Existing approaches differ in how they represent these states: continuous methods avoid vocabulary constraints, whereas discrete methods improve accuracy through quantization into a finite codebook. Our analysis of representative continuous and discrete systems identifies two functional requirements: answers must rely on latent states, and those states must carry valid, problem-specific reasoning. Continuous latents influence answers despite collapsed reasoning content, whereas discrete latents retain recoverable intermediate reasoning that answer prediction largely bypasses. To address these challenges, we propose Continuous Anchored Latent Reasoning (CALR), which connects latent formation with answer use through functional anchoring. With reference latents from information-balanced compression, CALR couples latent-mediated answer supervision with derivation-level semantic anchoring: the former routes answer supervision through intermediate states, while the latter grounds their decoded content in problem-specific derivations. A parallel-to-autoregressive curriculum develops sequential reasoning by conditioning subsequent latent blocks on generated prefixes. Evaluations on five mathematical reasoning benchmarks across model families show substantial accuracy gains. Under matched budgets, CALR gains 26.0 percentage points over a comparable continuous latent reasoning method. Further analyses show that its latents support answer prediction and carry problem-specific intermediate reasoning.
Chinese Translation
视觉潜在推理将渲染的推导压缩为紧凑的中间状态，从而减少文本推理的开销。现有方法在表示这些状态的方式上存在差异：连续方法避免了词汇约束，而离散方法通过量化为有限的代码本提高了准确性。我们对代表性的连续和离散系统的分析确定了两个功能需求：答案必须依赖于潜在状态，并且这些状态必须承载有效的、特定于问题的推理。尽管推理内容被压缩，连续潜在状态仍然影响答案，而离散潜在状态则保留可恢复的中间推理，答案预测在很大程度上绕过了这些推理。为了解决这些挑战，我们提出了连续锚定潜在推理（CALR），它通过功能锚定将潜在形成与答案使用连接起来。通过信息平衡压缩获得的参考潜在状态，CALR 将潜在介导的答案监督与推导级别的语义锚定结合起来：前者通过中间状态引导答案监督，而后者则将其解码内容扎根于特定于问题的推导中。一个平行到自回归的课程通过将后续潜在块的条件设置在生成的前缀上，发展顺序推理。在五个数学推理基准测试中的模型家族评估显示出显著的准确性提升。在匹配预算下，CALR 比可比的连续潜在推理方法提高了 26.0 个百分点。进一步分析表明，其潜在状态支持答案预测并承载特定于问题的中间推理。
cs.CV / 26 / 2610.07223
Deep Learning Based Illegal Bowling Action Detection
基于深度学习的非法投球动作检测
Abstract
Cricket, often referred to as the "gentleman's game," adheres to a strict rule set for both batsmen and bowlers, where each delivery can significantly impact the match outcome. Detecting illegal bowling actions is crucial for maintaining fair play, yet it remains challenging for umpires to monitor in real time. Existing sensor-based solutions have limitations in live match scenarios, making real-time assessment difficult. This paper proposes a computer vision-based deep learning solution to detect illegal bowling actions in live cricket matches. To develop and evaluate our approach, we compiled a dataset of 62 videos featuring 11 male bowlers, capturing both legal and illegal bowling actions from multiple angles-front, back, and side. However, the dataset predominantly comprises right-handed bowlers with conventional actions. The proposed system identifies two key frames, the shoulder frame and the release frame from video footage of a bowler's delivery and analyzes the change in the bowling arm's angle between these frames. If the angle difference exceeds a predefined threshold (e.g., 15 degrees), the delivery is flagged as potentially illegal. We evaluated the system on a custom dataset and achieved a high true positive rate, suggesting the system's potential effectiveness in real-time match settings. However, further research is required to validate the system across diverse environmental conditions and larger datasets to ensure generalizability and robustness in various live match scenarios. To the best of our knowledge, this is the first AI-based computer vision method for detecting illegal bowling actions in cricket.
Chinese Translation
板球常被称为“绅士的游戏”，遵循严格的规则体系，适用于击球手和投球手，每一次投球都可能对比赛结果产生重大影响。检测非法投球动作对于维护公平竞争至关重要，但裁判在实时监控中仍面临挑战。现有的基于传感器的解决方案在现场比赛中存在局限性，使得实时评估变得困难。本文提出了一种基于计算机视觉的深度学习解决方案，用于在现场板球比赛中检测非法投球动作。为了开发和评估我们的方法，我们编制了一个包含62个视频的数据集，涉及11名男性投球手，从多个角度捕捉合法和非法投球动作，包括正面、背面和侧面。然而，该数据集主要由具有传统动作的右手投球手组成。所提系统从投球手的投球视频中识别两个关键帧，即肩部帧和释放帧，并分析这两个帧之间投球手臂角度的变化。如果角度差异超过预定义的阈值（例如15度），则该投球被标记为可能非法。我们在自定义数据集上评估了该系统，并取得了较高的真正阳性率，表明该系统在实时比赛环境中的潜在有效性。然而，仍需进一步研究以验证该系统在不同环境条件和更大数据集中的表现，以确保其在各种现场比赛场景中的普适性和稳健性。据我们所知，这是首个基于人工智能的计算机视觉方法，用于检测板球中的非法投球动作。
cs.CV / 27 / 2610.07243
Hybrid Cross-Modal Attention Network for Early Breast Cancer Detection in Low-Resource Clinical Settings
低资源临床环境下早期乳腺癌检测的混合跨模态注意力网络
Abstract
Breast cancer is the leading cause of cancer-related mortality among women in Sub-Saharan Africa, where delayed diagnosis results from limited radiology expertise and fragmented clinical data systems. Although deep learning models have demonstrated strong performance in mammographic analysis, most rely solely on imaging data and are trained on Western populations, limiting their applicability in African healthcare settings. This paper presents a Hybrid Cross-Modal Attention Network (HCMAN) that integrates mammogram images with structured clinical data using transformer-based cross-modal attention mechanisms. The model was developed and validated using a locally collected dataset of 2,560 mammogram images from 1,024 patients across four Ethiopian referral hospitals, with biopsy-confirmed ground truth labels. The proposed framework achieves 97.8% accuracy, 97.2% sensitivity, 98.3% specificity, and an AUC of 0.987, significantly outperforming image-only baselines. The system demonstrates robustness to low-quality images typical of resource-limited settings, with only 3.2% performance degradation compared to 8.7% for image-only models. Cross-modal attention analysis reveals clinically appropriate behavior: higher reliance on clinical features for ambiguous cases such as dense breasts and young patients. The model's lightweight architecture enables deployment on standard hospital workstations (<2 seconds inference on CPU). This work advances sustainable, context-aware AI solutions for equitable breast cancer diagnostics in Africa.
Chinese Translation
乳腺癌是撒哈拉以南非洲女性癌症相关死亡的主要原因，延迟诊断源于放射学专业知识的匮乏和临床数据系统的碎片化。尽管深度学习模型在乳腺X线影像分析中表现出色，但大多数模型仅依赖影像数据，并且是在西方人群中训练的，这限制了它们在非洲医疗环境中的适用性。本文提出了一种混合跨模态注意力网络（Hybrid Cross-Modal Attention Network, HCMAN），该网络通过基于变换器的跨模态注意力机制将乳腺X线影像与结构化临床数据相结合。该模型使用来自四家埃塞俄比亚转诊医院的1,024名患者的2,560幅乳腺X线影像的本地收集数据集进行开发和验证，并附有活检确认的真实标签。所提出的框架实现了97.8%的准确率、97.2%的灵敏度、98.3%的特异性和0.987的曲线下面积（AUC），显著优于仅基于影像的基线模型。该系统对资源有限环境中典型的低质量影像表现出鲁棒性，性能下降仅为3.2%，而仅基于影像的模型则为8.7%。跨模态注意力分析揭示了临床上适当的行为：在密集乳腺和年轻患者等模糊病例中对临床特征的依赖性更高。该模型的轻量级架构使其能够在标准医院工作站上部署（在CPU上推理时间少于2秒）。这项工作推动了可持续的、具有情境意识的人工智能解决方案，以实现非洲乳腺癌诊断的公平性。
cs.CV / 28 / 2610.07269
What Words Keep of a Place: Zero-Shot Language Reasoning for Cross-View Geo-Localization
地点的语言表达：用于跨视角地理定位的零样本语言推理
Abstract
Cross-view geo-localization is commonly solved as an image retrieval problem, matching a ground-level image against a database of satellite tiles through a jointly trained embedding. Such models are accurate, but they need large paired supervision and cannot show what evidence supports a match. In this paper, we study a different question: how much of this task can be solved through language alone? We prompt a multimodal large language model (MLLM) to describe each ground panorama and each satellite tile as structured text, and localize by comparing these descriptions. No component is trained. We evaluate on 9,826 VIGOR pairs from four U.S. cities, in three settings. First, the descriptions are faithful but not discriminative. They agree closely across the two views, yet ranking the full pool by description similarity almost never returns the correct tile (0.39% Recall@1). Second, we narrow the pool to ten neighboring tiles, as a coarse prior would do. The same descriptions now become useful: an MLLM judge that scores structural consistency doubles random ranking and matches a strong lexical baseline. It also states which fields of the two descriptions agree and which conflict, which an embedding distance cannot do, and which we see as a step toward interpretable localization. Third, we place the judge on a trained visual retriever. On the queries it ranks wrongly, reranking from images works, while reranking from our descriptions does not (23.5% against 10.7% Recall@1). Scene structure survives the conversion into language, while the fine appearance detail needed to separate nearby places does not. Code and prompts are publicly available at https://github.com/AyeshAbuLehyeh/GeoLingual.
Chinese Translation
跨视角地理定位通常被视为一个图像检索问题，通过联合训练的嵌入将地面图像与卫星图像数据库进行匹配。这类模型虽然准确，但需要大量配对监督，并且无法展示支持匹配的证据。本文研究了一个不同的问题：这个任务有多少可以仅通过语言来解决？我们提示一个多模态大型语言模型（MLLM）将每个地面全景和每个卫星图像描述为结构化文本，并通过比较这些描述进行定位。没有任何组件经过训练。我们在来自四个美国城市的9,826个VIGOR对中进行评估，分为三种设置。首先，描述是忠实但不具辨别力的。它们在两个视角之间高度一致，但通过描述相似性对整个池进行排名几乎从未返回正确的图像（0.39% Recall@1）。其次，我们将池缩小到十个相邻的图像，类似于粗略的先验。相同的描述现在变得有用：一个评估结构一致性的MLLM判别器使随机排名的效果翻倍，并与强大的词汇基线相匹配。它还指出了两个描述中哪些字段一致，哪些冲突，而嵌入距离无法做到这一点，我们视之为朝向可解释定位的一步。第三，我们将判别器置于一个经过训练的视觉检索器上。在它错误排名的查询中，从图像重新排名有效，而从我们的描述重新排名则无效（23.5%对10.7% Recall@1）。场景结构在转换为语言的过程中得以保留，而区分相邻地点所需的细节则未能保留。代码和提示可在 https://github.com/AyeshAbuLehyeh/GeoLingual 上公开获取。
cs.CV / 29 / 2610.07326
Localize Any Object in X-Ray Security Scans without Human Annotation
在X射线安全扫描中无需人工标注即可定位任何物体
Abstract
Universal object localization in X-ray security inspection is critical for automated threat detection in safety-critical venues. However, unlike everyday RGB images that dominate web-scale visual data, X-ray scans exhibit distinct color patterns, ambiguous boundaries, and compositional structures caused by volumetric superposition. These gaps hinder the direct zero-shot transfer of dense perception foundation models trained on web-scale RGB data. Moreover, annotated X-ray data is scarce and requires expert labeling, limiting both the training of generalizable X-ray native models and the adaptation of RGB foundation models for X-ray data via fine-tuning. Given these challenges, the bright promise of highly generalizable perception models, enabled by data scaling laws in the RGB domain, remains largely out of reach for X-ray inspection. To this end, we introduce LAO-X, a self-supervised adaptation framework that Locates Any Object in X-ray scans using diverse synthesized image--annotation pairs with granularity-aware supervision. LAO-X first designs a saliency-guided X-ray object mining module to separate diverse object instances, which are then used for physics-guided synthesis in the absorbance domain. LAO-X further incorporates an occlusion-controlled curriculum strategy to fine-tune a Segment Anything Model 2 (SAM2) localizer, progressively adapting it to X-ray scans with increasing object counts and overlap levels. Experiments on six X-ray benchmarks show that LAO-X substantially improves category-agnostic localization, achieving 2\% to 23\% mAP gains over SAM2 and X-ray specific baselines in heavily cluttered scenarios, entirely without human-annotated labels.
Chinese Translation
在X射线安全检查中实现通用物体定位对于在安全关键场所的自动威胁检测至关重要。然而，与主导网络规模视觉数据的日常RGB图像不同，X射线扫描显示出独特的颜色模式、模糊的边界和由于体积叠加造成的组合结构。这些差距阻碍了在网络规模RGB数据上训练的密集感知基础模型的直接零样本迁移。此外，标注的X射线数据稀缺且需要专家标注，这限制了可泛化的X射线本地模型的训练以及通过微调将RGB基础模型适应于X射线数据。鉴于这些挑战，由RGB领域的数据扩展法则所启用的高度可泛化感知模型的光明前景在X射线检查中仍然大多无法实现。为此，我们提出了LAO-X，一个自监督适应框架，能够使用多样化的合成图像-标注对和粒度感知监督在X射线扫描中定位任何物体。LAO-X首先设计了一个显著性引导的X射线物体挖掘模块，以分离多样的物体实例，这些实例随后用于在吸收域中进行物理引导合成。LAO-X进一步结合了一个遮挡控制的课程策略，以微调Segment Anything Model 2 (SAM2)定位器，逐步将其适应于具有增加物体数量和重叠水平的X射线扫描。在六个X射线基准测试中的实验表明，LAO-X显著提高了类别无关的定位，在严重杂乱的场景中，相较于SAM2和X射线特定基线实现了2 ext{%}到23 ext{%}的mAP提升，完全不依赖人工标注的标签。
cs.CV / 30 / 2610.07337
Compositional Concept Erasure in Text-to-Image Diffusion Models via Hierarchically Grounded Semantic Surgery
通过分层基础的语义手术实现文本到图像扩散模型中的组合概念消除
Abstract
Removing copyrighted, unsafe, or user-specified concepts from a deployed text-to-image diffusion model is now a practical requirement. Weight-editing methods can suppress fixed targets, but they require per-target retraining and modify the model checkpoint. Training-free methods, on the other hand, are deployment-friendly, but they suffer from text-side routing failures on compositional prompts. In such prompts, the erase target may be invoked through a related class rather than its lexical name, and its modifiers may migrate onto preserved objects. This paper proposes Hierarchically Grounded Semantic Surgery (HGSS), a training-free framework for compositional concept erasure. The framework lifts both the routing signal and the edit operator used by text-side erasure. First, hierarchical span grounding resolves erase-target spans through lexical, taxonomic, and semantic evidence, while guarding against broad-hypernym and compound-head false positives. Second, dynamic attribute binding refines the text conditioning during early denoising via a counterfactual reference and a preserve-aware cross-attention objective, keeping surviving attribute-noun bindings intact. HGSS selectively removes the erase target without updating model weights or adding learned parameters. On SEE, HGSS cuts hierarchical evasion from 29.54 to 10.02 and roughly halves pairwise attribute leakage, achieving the best Neighbor E and AttrP scores among the reported erasure methods. On UnlearnCanvas, HGSS slightly improves the six-metric average over the matched Semantic Surgery baseline, reaching state-of-the-art.
Chinese Translation
从已部署的文本到图像扩散模型中移除版权、危险或用户指定的概念如今已成为一项实际需求。权重编辑方法可以抑制固定目标，但它们需要针对每个目标进行重新训练并修改模型检查点。另一方面，无需训练的方法在部署上更友好，但在组合提示的文本侧路由中存在失败。在此类提示中，消除目标可能通过相关类别而非其词汇名称被调用，其修饰符可能迁移到保留对象上。本文提出了分层基础的语义手术（Hierarchically Grounded Semantic Surgery, HGSS），这是一种用于组合概念消除的无训练框架。该框架提升了文本侧消除所使用的路由信号和编辑操作符。首先，分层跨度基础通过词汇、分类和语义证据解析消除目标跨度，同时防止广义上位词和复合头的假阳性。其次，动态属性绑定通过反事实参考和关注保留的交叉注意目标，在早期去噪过程中细化文本条件，保持存活的属性-名词绑定不变。HGSS选择性地移除消除目标，而无需更新模型权重或添加学习参数。在SEE上，HGSS将分层规避从29.54降低到10.02，并大致减少了一半的成对属性泄漏，在报告的消除方法中实现了最佳的邻居E和AttrP分数。在UnlearnCanvas上，HGSS略微提高了与匹配的语义手术基线相比的六项指标平均值，达到了最先进水平。
cs.CV / 31 / 2610.07339
A doctrine-grounded visual question answering dataset for Tactical Combat Casualty Care
基于理论的战术战斗伤员护理视觉问答数据集
Abstract
Tactical Combat Casualty Care (TC3) requires responders to connect visual observations of injuries and interventions with established clinical guidance. Developing vision-language models to support this process requires supervision that links visible evidence to traceable doctrine. We present TC3-VQA, a dataset constructed from public instructional and field TC3 videos and authoritative TC3 documents. It contains 581 items spanning 11 concepts, with 1,860 questions covering intervention recognition, doctrine, clinical reasoning, procedural guidance, and refusal when visual information is insufficient. Doctrine-based answers preserve verbatim source passages and character offsets. Construction combines visual annotation, passage retrieval, entailment checks, and verification across model families. Equipment boxes, anatomical labels, temporal segments, and source metadata accompany the question-answer pairs. Automated audits and ratings by two physicians and two medical students characterize annotation quality, with human ratings available for 88 retained items. The dataset provides a resource for adapting vision-language models to TC3, studying the connection between visual evidence and clinical knowledge, and evaluating recognition, doctrine recall, and abstention.
Chinese Translation
战术战斗伤员护理（TC3）要求应急响应者将对伤害和干预的视觉观察与既定的临床指导相连接。为了支持这一过程，开发视觉-语言模型需要监督，将可见证据与可追溯的理论联系起来。我们提出了 TC3-VQA，一个由公共教学和现场 TC3 视频及权威 TC3 文档构建的数据集。该数据集包含 581 个项目，涵盖 11 个概念，包含 1,860 个问题，涉及干预识别、理论、临床推理、程序指导以及在视觉信息不足时的拒绝。基于理论的答案保留逐字来源段落和字符偏移。构建过程结合了视觉标注、段落检索、蕴涵检查和跨模型系列的验证。设备框、解剖标签、时间段和源元数据伴随问题-答案对。通过两名医生和两名医学生的自动审核和评分来表征标注质量，其中 88 个保留项目的人类评分可用。该数据集为将视觉-语言模型适应于 TC3、研究视觉证据与临床知识之间的联系，以及评估识别、理论回忆和弃权提供了资源。
cs.CV / 32 / 2610.07355
Tracking Is Not Permanence: What Video World Models Keep of a Hidden Object
追踪并非永久性：视频世界模型对隐藏物体的保留
Abstract
Video world models track objects they can see; we ask what they keep of objects they cannot. We hide an object from a frozen V-JEPA 2 predictor and compare its prediction for the hidden region with the encoder's representation of two worlds that differ only inside that region. The predictor's decision keeps a stationary object in part and one carried inside a container not at all, and loses a moving one within 0.3 s (0.5 s under V-JEPA's own tube mask; ViT-H keeps it to 1.1 s at pretraining's 90% masking ratio); in projection a trace remains, below the midpoint, at 14-60% of what a baseline copying the last view retains. The information is there: the encoder reads the object's presence at 1.00 and keeps a closed container's contents decodable for 3.5 s, while the predictor's output, read with the encoder's own probe, contains the ball in 2% of scenes once the box has been closed for half a second. On rendered scenes, permanence is missing on the predictor's side, and training installs it cheaply as a prior: three thousand predictor-only steps on synthetic containers take this belief from 0.05 to 1.00 against two matched controls. They also raise IntPhys-2019 from 84.2% to 93.3%, but so does a curriculum without containers, and which training habit the benchmark credits changes with its scoring rule. Continued training with tube masks produces 1.1-1.6 s of moving-object carry-over on manipulation and internet-style video, so the deficit is not intrinsic to latent prediction. VideoMAE keeps almost nothing, and Cosmos's next-token prediction keeps a stationary hidden object but not one carried inside a moving container.
Chinese Translation
视频世界模型跟踪它们能看到的物体；我们探讨它们对看不见物体的保留情况。我们将一个物体隐藏在一个冻结的 V-JEPA 2 预测器中，并将其对隐藏区域的预测与编码器对两个仅在该区域内不同的世界的表征进行比较。预测器的决策部分保留了一个静止物体，而对一个放在容器内的物体则完全没有保留，并且在 0.3 秒内丢失了一个移动物体（在 V-JEPA 自身的管道掩码下为 0.5 秒；ViT-H 在预训练的 90% 掩码比例下保留至 1.1 秒）；在投影中，低于中点的痕迹仍然存在，保留量为基线复制最后视图的 14-60%。信息是存在的：编码器在 1.00 时读取物体的存在，并且能够解码一个封闭容器的内容长达 3.5 秒，而预测器的输出在编码器自身探针的读取下，在箱子关闭半秒后，包含球体的场景比例为 2%。在渲染场景中，预测器一侧缺乏永久性，而训练以低成本将其作为先验：在合成容器上进行三千步仅预测器的训练使这一信念从 0.05 提升至 1.00，相较于两个匹配的对照组。它们还将 IntPhys-2019 从 84.2% 提升至 93.3%，但没有容器的课程也能做到这一点，基准所认可的训练习惯随着评分规则的变化而变化。继续使用管道掩码的训练在操作和互联网风格视频上产生了 1.1-1.6 秒的移动物体保留，因此这一缺陷并非固有于潜在预测。VideoMAE 几乎没有保留任何信息，而 Cosmos 的下一个标记预测能够保留一个静止的隐藏物体，但无法保留一个放在移动容器内的物体。
cs.CV / 33 / 2610.07366
Identity-Conditioned Score Fusion for Open-Set Person Re-Identification
基于身份条件的评分融合用于开放集人物重识别
Abstract
Robust person re-identification often combines complementary cues such as face, gait, and body shape. While adaptive fusion typically targets query quality, model strength also varies across identities. We introduce identity-conditioned score fusion, a framework that tailors weights to each gallery identity without training. By contrasting intra-identity consistency against cross-identity impostors, it extracts identity-specific profiles that couple with query-conditioned adaptation via a parameter-free rule. This widens the separation between true and false matches while preserving score calibration. Evaluations on three clothes-changing person re-identification benchmarks show that our method consistently outperforms statistical, rank-based, and learned baselines, achieving up to an 8.8% absolute reduction in the false non-identification rate and demonstrating the value of identity-conditioned fusion in open-set person re-identification.
Chinese Translation
稳健的人物重识别通常结合面部、步态和体型等互补线索。虽然自适应融合通常针对查询质量，但模型的强度在不同身份之间也存在差异。我们提出了一种基于身份条件的评分融合框架，该框架在不进行训练的情况下为每个图库身份定制权重。通过对比同一身份内的一致性与跨身份的冒充者，它提取出与身份特定的特征，并通过无参数规则与查询条件的适应相结合。这在扩大真实匹配与错误匹配之间的分离的同时，保持了评分的校准。在三个换衣服的人物重识别基准测试中的评估表明，我们的方法在统计、基于排名和学习的基线之上始终表现优越，实现了高达8.8%的绝对减少错误非识别率，展示了基于身份条件的融合在开放集人物重识别中的价值。
cs.CV / 34 / 2610.07378
SimCortex v2: Joint Cortical Surface Reconstruction with Near-Zero Collisions and Self-Intersections
SimCortex v2：近零碰撞和自交的联合皮层表面重建
Abstract
Reconstructing cortical WM and pial surfaces from structural magnetic resonance imaging (MRI) is a prerequisite for surface-based neuroanatomical analysis, yet remains challenging because the cortex is thin and tightly folded. Reconstruction methods can produce geometric artifacts such as mesh self-intersections and collisions between cortical surfaces, and although recent deep learning methods have reduced reconstruction time from hours to minutes, these artifacts persist. We propose SimCortex v2, a deep learning framework for simultaneous reconstruction of the left and right WM and pial surfaces from T1-weighted MRI. SimCortex v2 estimates topologically correct initial surfaces from a volumetric segmentation and refines all four jointly using multi-scale stationary velocity fields predicted by a ribbon-conditioned, U-Net-like network. We evaluated SimCortex v2 on 560 cases from 14 cohorts, thirteen of them unseen during training, spanning ages 6-89, healthy and clinical populations, and scanners from three vendors. SimCortex v2 matched the surface-distance accuracy of the strongest baseline (average symmetric surface distance 0.253 mm) while showing no detected inter-surface collision in 92.14% of cases and the lowest self-intersection fraction (0.044%) among learning-based methods, whereas every baseline produced at least one collision in every case. Source code, configuration files, pretrained weights, preprocessed data, and the exact evaluation splits are publicly released.
Chinese Translation
从结构性磁共振成像（MRI）重建皮层白质（WM）和脑膜表面是基于表面的神经解剖分析的前提，但由于皮层薄且紧密折叠，这一过程仍然具有挑战性。重建方法可能会产生几何伪影，例如网格自交和皮层表面之间的碰撞，尽管最近的深度学习方法已将重建时间从数小时缩短至数分钟，但这些伪影依然存在。我们提出了SimCortex v2，这是一个用于同时重建左侧和右侧白质及脑膜表面的深度学习框架，基于T1加权MRI。SimCortex v2从体积分割中估计拓扑正确的初始表面，并使用由带有带状条件的U-Net类似网络预测的多尺度静态速度场共同细化所有四个表面。我们在来自14个队列的560个案例上评估了SimCortex v2，其中13个案例在训练期间未见，涵盖年龄范围为6至89岁，包含健康和临床人群，以及来自三家供应商的扫描仪。SimCortex v2在表面距离精度上与最强基线相匹配（平均对称表面距离为0.253毫米），在92.14%的案例中未检测到表面间碰撞，并且在基于学习的方法中自交比例最低（0.044%），而每个基线在每个案例中至少产生一次碰撞。源代码、配置文件、预训练权重、预处理数据和确切的评估划分已公开发布。
cs.CV / 35 / 2610.07381
GeoWM: Efficient Direct World Modeling in Explicit Geometry
GeoWM：显式几何中的高效直接世界建模
Abstract
Modeling 3D scene geometry and its evolution over time is essential for autonomous driving and robotics. A common paradigm is to use world models to predict future images or latent representations of the environment and subsequently recover geometry from these predictions. However, this paradigm does not explicitly model geometric structure and typically relies on recursive rollouts to reach longer prediction horizons, leading to error accumulation and increasing computational cost. To address these limitations, we present GeoWM, a geometry world model that directly forecasts future scene geometry at specified future horizons without recursive rollout. The key idea is to leverage a geometry foundation model to transform observed RGB frames into a geometric history, which conditions a flow-matching transformer to predict the scene geometry at a specified future horizon. We further show that a lightweight camera-motion predictor can accurately estimate the future viewpoint, and that projecting the observed geometry into the predicted viewpoint provides an effective geometric prior for future geometry forecasting. Extensive experiments on four datasets spanning urban driving, aerial flight, and dynamic manipulation demonstrate that GeoWM outperforms the evaluated world models in forecasting depth, camera pose, and 3D scene geometry, while substantially reducing inference time at longer horizons.
Chinese Translation
建模三维场景几何及其随时间的演变对于自动驾驶和机器人技术至关重要。一种常见的范式是使用世界模型来预测未来的图像或环境的潜在表示，随后从这些预测中恢复几何结构。然而，这种范式并未显式建模几何结构，通常依赖递归展开以达到更长的预测视野，导致误差累积和计算成本增加。为了解决这些局限性，我们提出了GeoWM，一种几何世界模型，能够在指定的未来视野下直接预测未来场景几何，而无需递归展开。其关键思想是利用几何基础模型将观察到的RGB帧转换为几何历史，这为流匹配变换器预测指定未来视野下的场景几何提供条件。我们进一步展示了一种轻量级相机运动预测器能够准确估计未来视点，并且将观察到的几何投影到预测的视点中，为未来几何预测提供了有效的几何先验。在涵盖城市驾驶、空中飞行和动态操作的四个数据集上的广泛实验表明，GeoWM在预测深度、相机姿态和三维场景几何方面优于评估的世界模型，同时在更长的视野下显著减少了推理时间。
cs.CV / 36 / 2610.07384
WildMatch: Weakly Supervised Image Matcher Adaptation for Wildlife Re-Identification
WildMatch：针对野生动物重新识别的弱监督图像匹配器适应
Abstract
Individual animal re-identification from camera-trap imagery is an instance retrieval problem central to non-invasive wildlife monitoring: a query image must retrieve the correct individual from a reference set of known animals. This requires computer vision models to recognize distinctive local patterns in fur, skin, or other visual markings. Current approaches either learn global embeddings as a classification problem, requiring many labeled images per individual while largely ignoring local evidence, or apply off-the-shelf, domain-agnostic image matchers. Although such matchers are pretrained on large and diverse image collections, adapting them to wildlife imagery is challenging because available datasets are small and lack correspondence-level annotations. We study weakly supervised adaptation of a pretrained keypoint matcher using only identity labels, without keypoint-level or geometric correspondence ground truth. We mine informative image pairs with the pretrained matcher, derive weak positive and negative supervision from identity agreement, and contrastively fine-tune the matching network to strengthen correspondences for same-identity pairs and suppress them for different identities. Across open-source wildlife re-identification datasets, our approach improves accuracy over off-the-shelf matchers and a state-of-the-art local--global fusion method. Under an open-world protocol with held-out individuals, it learns a transferable correspondence prior rather than memorizing training identities. To our knowledge, this is the first study of matcher-level, identity-supervised adaptation for animal re-identification. Our method enables data-efficient specialization of image matching models to wildlife domains using identity annotations already available in typical monitoring datasets.
Chinese Translation
从相机捕获的图像中对个体动物进行重新识别是非侵入性野生动物监测中的一个实例检索问题：查询图像必须从已知动物的参考集中检索到正确的个体。这要求计算机视觉模型能够识别毛发、皮肤或其他视觉标记中的独特局部模式。目前的方法要么将学习视为分类问题，学习全局嵌入，这需要每个个体有大量标注图像，同时在很大程度上忽视局部证据，要么应用现成的、与领域无关的图像匹配器。尽管这些匹配器是在大型多样化图像集合上进行预训练的，但将其适应于野生动物图像是具有挑战性的，因为可用的数据集较小且缺乏对应级别的注释。我们研究了仅使用身份标签的预训练关键点匹配器的弱监督适应，而不依赖于关键点级或几何对应的真实值。我们利用预训练的匹配器挖掘信息丰富的图像对，从身份一致性中推导出弱的正负监督，并对匹配网络进行对比微调，以增强同一身份对的对应关系并抑制不同身份对的对应关系。在开放源野生动物重新识别数据集上，我们的方法在准确性上优于现成的匹配器和一种最先进的局部-全局融合方法。在一个开放世界协议下，持有个体的情况下，它学习到可转移的对应先验，而不是简单地记忆训练身份。据我们所知，这是针对动物重新识别的匹配器级身份监督适应的首次研究。我们的方法使得图像匹配模型能够高效地专门化于野生动物领域，利用在典型监测数据集中已经可用的身份注释。
cs.CV / 37 / 2610.07385
Semantic Capability Acquisition and Specialization During Vision-Language Model Fine-Tuning
视觉-语言模型微调过程中的语义能力获取与专业化
Abstract
Fine-tuning vision-language models (VLMs) is typically evaluated at a single downstream checkpoint, obscuring whether a semantic capability was never acquired or emerged earlier and later declined during specialization. We ask how semantic capabilities are acquired, when they peak, how well they transfer, and what remains at deployment. We study these dynamics as a semantic capability trajectory, tracking identity- and attribute-based capabilities over training. We formulate a trajectory-based framework that separates capability acquisition, capability-specific optima, and later specialization, and introduce Structured Semantic Routing (SSR) to study how the representation of supervision shapes what is acquired. Across six pretrained backbones spanning DFN, MetaCLIP, and OpenAI CLIP, we show that fine-tuning can acquire semantic capability beyond the pretrained state, including gains observed on held-out evaluations. Unstructured name-and-attribute supervision produces strong name-and-attribute retrieval with comparatively weak name-free attribute-profile retrieval, whereas SSR yields substantially stronger name-free attribute-profile retrieval and is further strengthened by stochastic name-branch dropout. Different capabilities can peak at different stages, so a checkpoint selected by target class-name retrieval need not coincide with a transferable semantic optimum. Continued optimization can therefore preserve strong target class-name retrieval while reducing previously acquired transferable semantic capability. In a representative diagnostic study, this late specialization is consistent with reduced cross-modal semantic accessibility while substantial image-only class structure remains available.
Chinese Translation
视觉-语言模型（VLMs）的微调通常在单一的下游检查点进行评估，这使得我们无法判断某种语义能力是从未获得，还是在专业化过程中早期获得后又下降。我们探讨语义能力是如何获得的、何时达到峰值、转移效果如何以及在部署时剩余的能力。我们将这些动态视为语义能力轨迹，跟踪训练过程中的身份和属性基础能力。我们提出了一种基于轨迹的框架，区分能力获取、特定能力的最优状态以及后期专业化，并引入结构化语义路由（Structured Semantic Routing, SSR）来研究监督的表示如何影响所获得的能力。在六个预训练的基础模型（包括 DFN、MetaCLIP 和 OpenAI CLIP）上，我们展示了微调可以获得超出预训练状态的语义能力，包括在保留评估中观察到的提升。无结构的名称和属性监督在名称和属性检索上表现良好，但名称自由的属性轮廓检索相对较弱，而 SSR 则显著增强了名称自由的属性轮廓检索，并通过随机名称分支丢弃进一步增强。不同的能力可能在不同阶段达到峰值，因此通过目标类别名称检索选择的检查点不必与可转移的语义最优状态重合。因此，持续优化可以在减少先前获得的可转移语义能力的同时，保持强大的目标类别名称检索。在一项代表性的诊断研究中，这种后期专业化与跨模态语义可访问性的降低一致，而大量仅图像的类别结构仍然可用。
cs.CV / 38 / 2610.07460
ElasticFit: Fit-Aware 3D Object Insertion via VLM Reasoning and Generative Adaptation
ElasticFit：基于VLM推理和生成适应的适配感知3D物体插入
Abstract
Inserting objects into existing 3D scenes requires more than selecting a plausible location: the inserted object must also fit local geometry while preserving semantic intent and physical plausibility. Although recent Vision-Language Models (VLMs) and generative models enable semantic reasoning and visual content creation, they offer limited 3D grounding and geometric control when an inserted object must fit into constrained local spaces. We introduce \textbf{ElasticFit}, a VLM-guided framework for fit-aware object insertion centered on a novel scene-grounded representation. Given a language instruction and rendered scene observations, ElasticFit infers structured fitting cues that specify where the object should be grounded, what volume it should occupy, how it should be oriented, and its adaptation mode (rigid placement, uniform scaling, or elastic fitting). These cues convert high-level VLM reasoning into explicit 3D constraints that condition object generation and guide downstream geometric fitting. ElasticFit then generates a scene-conditioned object prior, reconstructs it in 3D, and refines the mesh through mode-specific fitting while enforcing collision avoidance, contact consistency, and physical grounding. In fixed-asset baseline comparisons, ElasticFit improves spatial relation success from 50.8\% to 69.7\% and support success from 48.3\% to 91.7\% over the strongest baseline, while providing novel support for generative "make-it-fit" insertions in complex scenarios.
Chinese Translation
将物体插入现有的3D场景不仅需要选择一个合理的位置：插入的物体还必须适应局部几何形状，同时保持语义意图和物理合理性。尽管近期的视觉-语言模型（VLMs）和生成模型能够实现语义推理和视觉内容创作，但在插入物体必须适应受限局部空间时，它们提供的3D基础和几何控制能力有限。我们提出了 extbf{ElasticFit}，一个以VLM为指导的适配感知物体插入框架，基于一种新颖的场景基础表示。给定语言指令和渲染的场景观察，ElasticFit推断出结构化的适配线索，指定物体应当如何定位、占据何种体积、如何定向以及其适应模式（刚性放置、均匀缩放或弹性适配）。这些线索将高级VLM推理转换为明确的3D约束，从而调节物体生成并指导下游几何适配。随后，ElasticFit生成一个场景条件下的物体先验，在3D中重建该物体，并通过特定模式的适配对网格进行精细化，同时强制避免碰撞、保持接触一致性和物理基础。在固定资产基线比较中，ElasticFit将空间关系成功率从50.8\%提高到69.7\%，将支撑成功率从48.3\%提高到91.7\%，超越了最强基线，同时为复杂场景中的生成“使其适配”插入提供了新支持。
cs.CV / 39 / 2610.07464
Protective Perturbations Must Survive the Resize: Scale-Robust Image Immunization against Malicious Editing
保护性扰动必须经受缩放：针对恶意编辑的尺度鲁棒图像免疫
Abstract
Protective perturbations aim to stop malicious instruction-guided editing of personal photos, but they are optimized and evaluated at the editor's working resolution, whereas shared photos have 10 megapixels or more and editors first downscale them by an unknown factor. We model this resize as a frequency-selective channel. In this model, a perturbation computed at the native resolution decays with the downscaling factor and is weak even without a resize, and a perturbation computed at a fixed working resolution protects only a window of scales. The best worst-case protection over an unknown range of scales degrades only logarithmically with the width of the range, and averaging over scales does not reach it. Guided by this analysis, we propose SRIM, which samples a grid of anchor scales covering the whole range, with weights that favor the currently weakest scale, at the cost of standard expectation over transformation. On full-resolution photos of 9 to 30 megapixels and downscaling factors from 2 to 8, SRIM raises the worst-case disruption of FLUX.2-klein edits from 0.192 LPIPS, attained by the strongest published protection, to 0.463. At equal visibility, it roughly doubles the protection. The same protected photos also protect against the 9B model and against FLUX.2-dev, with worst cases of 0.450 and 0.386 against at most 0.184 for published protections, and SRIM leads on InstructPix2Pix as well.
Chinese Translation
保护性扰动旨在阻止恶意指令引导的个人照片编辑，但它们是在编辑器的工作分辨率下进行优化和评估的，而共享照片的分辨率通常为10百万像素或更高，编辑器首先以未知的比例对其进行下采样。我们将这种缩放建模为一个频率选择性通道。在该模型中，在原生分辨率下计算的扰动会随着下采样因子的增加而衰减，即使没有缩放，扰动也很弱；而在固定工作分辨率下计算的扰动仅能保护特定范围内的尺度。对于未知范围内的尺度，最佳的最坏情况保护仅以对数方式随范围宽度降低，而对尺度进行平均并不能达到这一效果。在这一分析的指导下，我们提出了SRIM，它对覆盖整个范围的锚尺度网格进行采样，权重偏向当前最弱的尺度，代价是标准的变换期望。在9到30百万像素的全分辨率照片以及从2到8的下采样因子下，SRIM将FLUX.2-klein编辑的最坏情况干扰从0.192 LPIPS（由最强的已发布保护获得）提升至0.463。在相同可见度下，它的保护效果大约翻倍。这些受保护的照片同样能够抵御9B模型和FLUX.2-dev，最坏情况分别为0.450和0.386，而已发布的保护最多为0.184，SRIM在InstructPix2Pix上也表现优异。
cs.CV / 40 / 2610.07561
LARK: A Low-Cost, Accurate, Occlusion-Resilient, Kalman Filter-Assisted Tracking System for Image-Guided Surgery
LARK：一种低成本、高精度、抗遮挡的卡尔曼滤波辅助图像引导手术跟踪系统
Abstract
Image-guided surgery (IGS) depends on accurate tracking of surgical instruments to provide real-time navigation relative to anatomical structures. Commercial stereo infrared trackers are accurate but prone to occlusion and cost-prohibitive for many settings. This work presents LARK, a multi-camera optical tracking system using commodity RGB hardware and multi-view redundancy and fusion. We develop and evaluate two complete tracking methods: multi-view monocular pose fusion and multi-view triangulation. Both methods are assessed under varying occlusion levels using a precision-machined grid and an anatomical head phantom, and compared against a gold-standard stereo infrared system. With five cameras and adaptive Kalman filtering, LARK achieves median target registration errors of 0.64 mm for point localization with triangulation and 0.73 mm for trajectory tracking with pose fusion on the machined grid. Camera-subset experiments show graceful degradation in adaptive pose-fusion accuracy as fewer views remain available. With tracking hardware costing under $1,000 USD, LARK provides a low-cost platform for image-guided surgery research. Hardware designs and software are publicly available at https://nist.mni.mcgill.ca/software/ , and datasets at https://nist.mni.mcgill.ca/data/ .
Chinese Translation
图像引导手术（IGS）依赖于对外科器械的精确跟踪，以提供相对于解剖结构的实时导航。商业立体红外跟踪器虽然精确，但易受遮挡影响，并且在许多场合成本过高。本文提出了LARK，一种使用普通RGB硬件和多视角冗余与融合的多摄像头光学跟踪系统。我们开发并评估了两种完整的跟踪方法：多视角单目姿态融合和多视角三角测量。通过使用精密加工的网格和解剖头模，我们在不同遮挡水平下评估了这两种方法，并与金标准立体红外系统进行了比较。使用五个摄像头和自适应卡尔曼滤波，LARK在加工网格上实现了0.64毫米的中位目标注册误差（点定位）和0.73毫米的轨迹跟踪误差（姿态融合）。摄像头子集实验表明，随着可用视角的减少，自适应姿态融合的精度逐渐下降。LARK的跟踪硬件成本低于1000美元，为图像引导手术研究提供了一个低成本的平台。硬件设计和软件可在 https://nist.mni.mcgill.ca/software/ 上公开获取，数据集可在 https://nist.mni.mcgill.ca/data/ 上获取。
cs.CV / 41 / 2610.07566
AIMS: Anchor-Integrated Multi-View Synthesis for Scalable Novel View Rendering
AIMS：集成锚点的多视图合成用于可扩展的新视图渲染
Abstract
Feed-forward novel view synthesis methods achieve strong generalization from posed multi-view inputs, but scaling them to large input view sets remains challenging. Transformer-based approaches that jointly process all input-view tokens incur rapidly increasing computation and memory as the number of views grows, while simple view subsampling discards potentially useful observations. We introduce Anchor-Integrated Multi-View Synthesis (AIMS), a scalable framework that decouples the number of available observations from the number of views processed by the global synthesis model. AIMS selects a fixed set of spatially distributed anchor views using farthest point sampling, groups nearby observations around each anchor, and uses a lightweight learnable integrator to fuse their information into enriched anchor representations. This allows additional observations to contribute to synthesis while keeping the downstream global view budget fixed. Evaluations on RealEstate10K and ScanNet demonstrate a favorable quality--efficiency trade-off against transformer-based and Gaussian-based baselines. AIMS achieves 29.41 dB and 17.73 dB PSNR on the two datasets, respectively, with rendering averaging 7.24 ms per view.
Chinese Translation
前馈新视图合成方法能够从固定的多视图输入中实现强大的泛化能力，但将其扩展到大规模输入视图集仍然具有挑战性。基于变换器的方法在处理所有输入视图标记时，随着视图数量的增加，计算和内存需求迅速增加，而简单的视图子采样则会丢弃潜在有用的观测。我们提出了集成锚点的多视图合成（AIMS），这是一个可扩展的框架，它将可用观测的数量与全球合成模型处理的视图数量解耦。AIMS使用最远点采样选择一组固定的空间分布锚点视图，将附近的观测分组到每个锚点周围，并使用轻量级可学习的集成器将其信息融合到丰富的锚点表示中。这使得额外的观测能够为合成做出贡献，同时保持下游全球视图预算固定。在RealEstate10K和ScanNet上的评估显示，AIMS在质量与效率之间实现了良好的权衡，相较于基于变换器和高斯的基线，AIMS在两个数据集上分别达到了29.41 dB和17.73 dB的峰值信噪比（PSNR），渲染平均每视图为7.24毫秒。
cs.CV / 42 / 2610.07576
CETUS: How Far Do Representations Trained on Earth Transfer to Cassini SAR of Titan?
CETUS：在多大程度上，基于地球训练的表征可以迁移到土卫六的卡西尼合成孔径雷达图像上？
Abstract
Cassini synthetic aperture radar (SAR) images reveal the dunes, plains, and lake basins of Titan, providing an instance of representations learned from Earth imagery for planetary terrain classification. Cross-domain Evaluation of Earth-to-Titan Transfer Using SAR (CETUS) compares features from DINOv2, DOFA and CROMA with classical image measurements and features from an untrained vision transformer on the U.S. Geological Survey's Cassini SAR mosaic. The classifiers learn terrain labels from an expert geomorphological map and predict those labels in geographically separate Titan regions. Under logistic regression settings, pretrained encoders achieve higher mean macro F1 than the combined classical features. Encoder rankings change when feature scaling, optimization, and regularization change together. Further training on Titan improves DINOv2 performance, degrades DOFA performance, and leads to mixed results for CROMA under the tested settings. Architectural and input processing differences prevent these comparisons from isolating the effect of pretraining. Classifier fitting and performance on individual terrain classes matter when assessing representation transfer for planetary mapping. Since the map draws partly on the same radar observations, the scores measure agreement with expert interpretation.
Chinese Translation
卡西尼合成孔径雷达（SAR）图像揭示了土卫六的沙丘、平原和湖盆，为从地球影像中学习的表征在行星地形分类中的应用提供了实例。跨域地球到土卫六转移评估（CETUS）比较了DINOv2、DOFA和CROMA的特征与经典图像测量以及在美国地质调查局的卡西尼SAR马赛克上未训练的视觉变换器的特征。分类器从专家的地貌图中学习地形标签，并在地理上分离的土卫六区域中预测这些标签。在逻辑回归设置下，预训练编码器的平均宏F1得分高于组合的经典特征。当特征缩放、优化和正则化一起变化时，编码器的排名也会改变。在土卫六上进一步训练提高了DINOv2的性能，降低了DOFA的性能，并在测试设置下对CROMA产生了混合结果。架构和输入处理的差异使得这些比较无法孤立出预训练的影响。在评估行星制图的表征迁移时，分类器对单个地形类别的拟合和性能至关重要。由于该地图部分基于相同的雷达观测，得分衡量了与专家解释的一致性。
cs.CV / 43 / 2610.07585
REViT-v2: Hierarchical Windowed Roto-reflection Equivariant ViT for Equivariant Feature Extraction
REViT-v2：用于等变特征提取的分层窗口旋转反射等变视觉变换器
Abstract
We propose a scalable roto-reflection-group-equivariant vision transformer based on windowed group-convolutional self-attention and a hierarchical feature architecture. We demonstrate that our approach can be scaled to group-equivariant vision transformers (ViTs) with millions of parameters and large datasets with practically sized images, i.e., ImageNet. The code and pretrained weights for the proposed Hierarchical Windowed Roto-reflection Equivariant ViTs (REViT-v2) are available at https://github.com/kc-ml2/revit.
Chinese Translation
我们提出了一种基于窗口化群卷积自注意力和分层特征架构的可扩展旋转反射群等变视觉变换器。我们证明了我们的方法可以扩展到具有数百万参数的群等变视觉变换器（ViTs）以及具有实际尺寸图像的大型数据集，即 ImageNet。所提出的分层窗口旋转反射等变视觉变换器（REViT-v2）的代码和预训练权重可在 https://github.com/kc-ml2/revit 获取。
cs.CV / 44 / 2610.07609
PhysLDM: Latent Diffusion for High-Fidelity Deformable Simulation
PhysLDM：用于高保真可变形仿真的潜在扩散
Abstract
Neural simulation of high-fidelity deformable bodies is a foundational challenge in computer graphics and physical AI. Long-horizon prediction for high-resolution 3D volumetric meshes is hard: autoregressive methods are susceptible to error accumulation, while direct multi-frame prediction at native resolution is computationally prohibitive. This motivates a compact spatiotemporal latent representation, which is largely unexplored for mesh-based volumetric physics. Meanwhile, it remains unclear whether deterministic regression or generative diffusion is the more appropriate predictive paradigm. To address these coupled challenges, we introduce PhysLDM, a unified latent-diffusion paradigm for one-shot volumetric deformable simulation. Its core is a holistic spatiotemporal VAE that avoids the "staircase" artifacts of standard temporal compression (as in common video VAEs), achieving ~2.48 mm reconstruction precision on meter-scale scenes at up to 78x token compression. Based on this reliable latent space, we systematically compare regression and diffusion methods. Our experiments uncover a key modeling insight: complex deformable dynamics are often chaotic, and in this regime deterministic regression tends to produce non-physical averages, whereas diffusion better models their distribution. Accordingly, we employ a latent diffusion model that effectively learns from the chaotic data to generate physically plausible trajectories. Trained purely kinematically on an Objaverse-scale dataset, a single PhysLDM generalizes zero-shot to unseen OOD datasets (GSO and Toys4K). Its differentiability further enables efficient solution of inverse problems and higher-order design optimization. To our knowledge, PhysLDM is the first high-fidelity spatiotemporal autoencoder and latent-diffusion paradigm for volumetric deformable dynamics, offering a scalable and robust approach to neural simulation.
Chinese Translation
高保真可变形物体的神经仿真是计算机图形学和物理人工智能中的一个基础性挑战。对于高分辨率的三维体积网格进行长时间预测是困难的：自回归方法容易受到误差累积的影响，而在原生分辨率下进行直接的多帧预测在计算上是不可行的。这促使我们探索一种紧凑的时空潜在表示，而这一领域在基于网格的体积物理中尚未得到充分研究。同时，确定性回归与生成扩散哪种预测范式更为合适仍不清楚。为了解决这些相互关联的挑战，我们提出了PhysLDM，一种用于一次性体积可变形仿真的统一潜在扩散范式。其核心是一个整体的时空变分自编码器（VAE），避免了标准时间压缩（如常见视频VAE）中的“阶梯”伪影，在米级场景中实现了约2.48毫米的重建精度，压缩比高达78倍。基于这一可靠的潜在空间，我们系统地比较了回归和扩散方法。我们的实验揭示了一个关键的建模洞察：复杂的可变形动力学往往是混沌的，而在这种情况下，确定性回归倾向于产生非物理的平均值，而扩散更好地建模了它们的分布。因此，我们采用了一种潜在扩散模型，有效地从混沌数据中学习，以生成物理上合理的轨迹。在一个Objaverse规模的数据集上纯粹进行运动学训练后，单个PhysLDM能够零-shot泛化到未见的OOD数据集（GSO和Toys4K）。其可微性进一步使得逆问题的高效解决和高阶设计优化成为可能。根据我们的知识，PhysLDM是首个用于体积可变形动力学的高保真时空自编码器和潜在扩散范式，为神经仿真提供了一种可扩展且稳健的方法。
cs.CV / 45 / 2610.07684
Disentangling Dual Image References in Frequency Aware Diffusion Models for Personalized Generation
在频率感知扩散模型中解开双重图像参考以实现个性化生成
Abstract
Personalized image generation aims to synthesize text-driven images conditioned on reference images, while mainly casting the generation as image customization for foreground and style transfer for background. Previous arts of diffusion models suffers from the text misalignment with background for image customization and foreground for style transfer during the denoising process. Such facts, as we observed, rooted from the entanglement among hybrid frequency bands during the denoising process. To address such salient limitation, in this paper, we study personalized generation based on dual references - customization and color and style reference - and propose a paradigm to disentangle these Dual image references within Frequency-aware Diffusion Models, dubbed Dual-FDM, to simultaneously tackle two crucial personalized image generation tasks: customization style transfer and color style transfer, by disentangling different frequency bands via mask strategy within frequency domain. For customization style transfer, we replace the mid-frequency band of the background in the style reference with that from the foreground of the customized reference. For color style transfer, we substitute the low-frequency band of the background in the style reference with that from both the foreground and background of the color reference. Both the substituted frequency bands are used as the key and value to reconstruct the query foreground and background of the denoised personalized image.Extensive experiments validate the superiority of Dual-FDM over the state-of-the-art diffusion models for personalized image generation. Our code can be accessed from https://github.com/htyjers/Dual-FDM.
Chinese Translation
个性化图像生成旨在合成基于文本驱动的图像，这些图像以参考图像为条件，主要将生成视为前景的图像定制和背景的风格转移。现有的扩散模型在图像定制过程中，文本与背景的错位以及在风格转移过程中与前景的错位，导致了生成效果不佳。我们观察到，这种现象源于去噪过程中混合频带之间的纠缠。为了解决这一显著的局限性，本文研究了基于双重参考的个性化生成——定制和颜色与风格参考，并提出了一种范式，以在频率感知扩散模型中解开这些双重图像参考，称为Dual-FDM，旨在通过在频域内采用掩码策略解开不同频带，同时解决两个关键的个性化图像生成任务：定制风格转移和颜色风格转移。对于定制风格转移，我们用定制参考的前景中频带替换风格参考中背景的中频带。对于颜色风格转移，我们用颜色参考的前景和背景的低频带替换风格参考中背景的低频带。替换后的频带作为关键和数值，用于重建去噪个性化图像的查询前景和背景。大量实验验证了Dual-FDM在个性化图像生成方面优于最先进的扩散模型。我们的代码可以从 https://github.com/htyjers/Dual-FDM 获取。
cs.CV / 46 / 2610.07689
Unlocking Fine-Grained Perception in CLIP via Structurally-Aware Latent Masked Modeling
通过结构感知的潜在掩蔽建模解锁 CLIP 的细粒度感知能力
Abstract
Vision-Language Models (VLMs) such as CLIP excel in global semantic alignment but often lack fine-grained perceptual capabilities. This hinders dense prediction tasks and bottlenecks the visual potential of Multimodal Large Language Models (MLLMs). Existing research has attempted to enhance CLIP's visual representations by incorporating geometric priors from vision-centric models. However, these strategies often struggle to achieve deep alignment for both local spatial structures and global semantics, potentially even distorting the original image-text space. To address these limitations, we propose SALM, an unsupervised embedding alignment framework based on structurally-aware latent mask modeling. SALM effectively synergizes local and global alignment via a dual-path design combining explicit and implicit mechanisms, without requiring any image-text pairs. First, we introduce a dual-matrix alignment strategy that explicitly calibrates intra-sample spatial correlations and activation intensities, thereby effectively injecting local geometric priors. Based on this, we further design a latent mask modeling mechanism to guide CLIP to restore the missing semantic details of the target model, thereby implicitly aggregating fine-grained structures into the global semantic space. Furthermore, driven by the empirical observations that CLIP's shallow features inherently possess strong spatial observational capabilities, we naturally extend SALM to a highly efficient self-distillation paradigm, SALM-Self. This unlocks CLIP's intrinsic fine-grained potential without relying on any external models. Extensive experiments demonstrate that SALM not only significantly improves performance in dense prediction tasks but also boosts CLIP's zero-shot accuracy, effectively enhancing the fine-grained understanding capabilities of MLLMs. Project page at https://qzfm.github.io/salm_project_page/.
Chinese Translation
视觉语言模型（VLMs）如 CLIP 在全局语义对齐方面表现出色，但往往缺乏细粒度的感知能力。这限制了密集预测任务的执行，并成为多模态大型语言模型（MLLMs）视觉潜力的瓶颈。现有研究尝试通过引入来自视觉中心模型的几何先验来增强 CLIP 的视觉表征。然而，这些策略往往难以实现局部空间结构与全局语义的深度对齐，甚至可能扭曲原始的图像-文本空间。为了解决这些局限性，我们提出了 SALM，一种基于结构感知潜在掩蔽建模的无监督嵌入对齐框架。SALM 通过结合显式和隐式机制的双路径设计，有效地协同局部和全局对齐，而无需任何图像-文本对。首先，我们引入了一种双矩阵对齐策略，显式校准样本内部的空间相关性和激活强度，从而有效注入局部几何先验。在此基础上，我们进一步设计了一种潜在掩蔽建模机制，以指导 CLIP 恢复目标模型缺失的语义细节，从而隐式地将细粒度结构聚合到全局语义空间。此外，基于经验观察，CLIP 的浅层特征本质上具有强大的空间观察能力，我们自然将 SALM 扩展到一种高效的自蒸馏范式 SALM-Self。这在不依赖任何外部模型的情况下解锁了 CLIP 的内在细粒度潜力。大量实验表明，SALM 不仅显著提升了密集预测任务的性能，还提高了 CLIP 的零-shot 准确率，有效增强了 MLLMs 的细粒度理解能力。项目页面：https://qzfm.github.io/salm_project_page/
cs.CV / 47 / 2610.07694
Anchor-driven Multi-modal Multi-scale Expert Selection for Survival Prediction
基于锚点的多模态多尺度专家选择用于生存预测
Abstract
The integrative analysis of histopathological Whole-Slide Images (WSIs) and transcriptomic profiles holds significant promise for cancer survival prediction. However, existing methods typically project multi-modal features directly into a shared latent space without explicit alignment, leading to the entanglement of mismatched morphological cues and molecular signals. Furthermore, current fusion strategies often treat the extreme spatial heterogeneity of WSIs uniformly, lacking mechanisms to adaptively prioritize clinically relevant tissue scales for individual patients. To address these limitations, we propose an Anchor-driven Multi-modal Multi-scale Expert Selection (AM$^2$ES) framework for survival prediction. Specifically, we present an Anchor-driven Multi-modal Fusion (AMF) module, which introduces learnable semantic anchors as cross-modal mediators to bridge the semantic gap by enforcing a structurally regularized alignment between transcriptomic features and multi-scale pathology representations. Built upon this aligned semantic space, we further design a Hierarchical Mixture-of-Experts (H-MoE) selection module to decouple the hierarchical prognostic selection process. Mimicking the pathologist's diagnostic workflow, H-MoE performs (i) Intra-scale Expert Filtering to discriminatively identify salient tumor regions within each magnification, and (ii) Inter-scale Hierarchy Routing to dynamically weight and select the most informative resolution levels. Extensive experiments on multiple TCGA cancer cohorts demonstrate that our AM$^2$ES achieves state-of-the-art performance while offering fine-grained interpretability by visualizing how specific molecular pathways drive the expert routing decisions across tissue scales. The code will be released at https://github.com/taozh2017/AM2ES.
Chinese Translation
组织病理学全幻灯片图像（Whole-Slide Images, WSIs）与转录组特征的综合分析在癌症生存预测中具有重要前景。然而，现有方法通常将多模态特征直接投影到共享的潜在空间中，而没有明确的对齐，导致形态线索和分子信号的不匹配纠缠。此外，当前的融合策略往往均匀处理WSIs的极端空间异质性，缺乏自适应优先考虑个体患者临床相关组织尺度的机制。为了解决这些局限性，我们提出了一种基于锚点的多模态多尺度专家选择（Anchor-driven Multi-modal Multi-scale Expert Selection, AM$^2$ES）框架用于生存预测。具体而言，我们提出了一个基于锚点的多模态融合（Anchor-driven Multi-modal Fusion, AMF）模块，该模块引入可学习的语义锚点作为跨模态中介，通过强制转录组特征与多尺度病理表征之间的结构性正则化对齐，来弥合语义差距。在此对齐的语义空间基础上，我们进一步设计了一个分层专家混合（Hierarchical Mixture-of-Experts, H-MoE）选择模块，以解耦分层预后选择过程。H-MoE模拟病理学家的诊断工作流程，执行（i）尺度内专家过滤，以区分性地识别每个放大倍数下显著的肿瘤区域，以及（ii）尺度间层次路由，以动态加权和选择最具信息量的分辨率级别。在多个TCGA癌症队列上的广泛实验表明，我们的AM$^2$ES实现了最先进的性能，同时通过可视化特定分子通路如何驱动专家路由决策在组织尺度上提供了细粒度的可解释性。代码将发布在 https://github.com/taozh2017/AM2ES。
cs.CV / 48 / 2610.07698
RBMatch: Dual-Level Class Rebalancing for Semi-Supervised Building Footprint Extraction
RBMatch：用于半监督建筑轮廓提取的双层类别重平衡
Abstract
Accurate building footprint extraction from high-resolution remote sensing imagery is essential for urban planning, disaster response, and environmental monitoring. However, obtaining dense pixel-level annotations is costly, motivating the use of semi-supervised learning (SSL) to leverage unlabeled imagery. In remote sensing, severe foreground--background imbalance poses a particular challenge for self-training, as it can bias pseudo-label generation and the resulting unsupervised optimization toward the majority background class. We show that addressing this imbalance at only one stage is insufficient: balancing pseudo-label selection alone does not prevent background bias from re-emerging during unsupervised loss optimization, a failure mode we term \emph{imbalance leak}. To address this issue, we propose \textbf{RBMatch}, a dual-level class-rebalancing framework that jointly regulates pseudo-label generation and unsupervised optimization. RBMatch combines a supervised learning pathway with a self-training module comprising three components: adaptive class-specific thresholding (ACT) for balanced pseudo-label selection, confidence-aware class-balanced reweighting (CACBR) for mitigating class bias in the unsupervised loss, and distribution alignment (DAL) for matching the predicted unlabeled-data distribution to the labeled-data prior. Experiments on the WHU, INRIA, and Massachusetts building footprint datasets across labeled ratios of 1%--10% show that RBMatch consistently achieves the best building IoU and F1-score among the evaluated methods. The improvement is most pronounced on the highly imbalanced Massachusetts dataset, where RBMatch improves IoU by 1.37 points over the strongest baseline at a 1% labeling ratio and is the only method to outperform the fully supervised baseline across all twelve dataset--ratio settings.
Chinese Translation
从高分辨率遥感影像中准确提取建筑轮廓对于城市规划、灾害响应和环境监测至关重要。然而，获取密集的像素级注释成本高昂，这促使我们采用半监督学习（SSL）来利用未标记的影像。在遥感领域，严重的前景与背景不平衡对自我训练构成了特别的挑战，因为它可能导致伪标签生成的偏差以及随之而来的无监督优化向多数背景类别倾斜。我们表明，仅在一个阶段解决这种不平衡是不够的：单独平衡伪标签选择并不能防止背景偏差在无监督损失优化过程中重新出现，这种失败模式我们称之为“失衡泄漏”（imbalance leak）。为了解决这个问题，我们提出了 extbf{RBMatch}，一个双层类别重平衡框架，联合调节伪标签生成和无监督优化。RBMatch结合了一个监督学习路径和一个由三个组件组成的自我训练模块：自适应类别特定阈值（ACT）用于平衡伪标签选择，基于置信度的类别平衡重加权（CACBR）用于减轻无监督损失中的类别偏差，以及分布对齐（DAL）用于将预测的未标记数据分布与标记数据先验匹配。在WHU、INRIA和马萨诸塞州建筑轮廓数据集上进行的实验显示，在标记比例为1%到10%之间，RBMatch在评估的方法中始终实现了最佳的建筑IoU和F1-score。在高度不平衡的马萨诸塞州数据集上，RBMatch的改进最为显著，在1%的标记比例下，IoU比最强基线提高了1.37个百分点，并且是唯一在所有十二个数据集-比例设置中超越完全监督基线的方法。
cs.CV / 49 / 2610.07705
What Frame-Level Labels Can and Cannot Do for Small-UAV Point Detection in Thermal Video
框架级标签对小型无人机热视频点检测的作用与局限
Abstract
The growing use of unmanned aerial vehicles (UAVs) has increased the importance of image-based UAV detection. Learning-based detectors are trained on imagery and annotations, with annotation type determining the information available during training. We focus on learning localization from frame-level target presence/absence labels when sensor or scene changes make spatial annotations for additional training burdensome. We analyze the detection capability, learning behavior, and potential applications of an existing architecture for point detection of small UAVs, trained with presence/absence labels and requiring no external detector. The architecture freezes spatial features learned through classification and trains a readout with the same frame labels to produce spatial score maps and point detections. On two thermal infrared datasets, CST Anti-UAV and Anti-UAV410, we evaluate localization hit rates and detection rates under false-alarm constraints, analyze the effects of training stages, label allocation, synthesis, and model configuration, and compare with bounding-box detectors. We also explore potential applications on Airborne Object Tracking (AOT) using its visible-light imagery and frame labels. Classification training strengthened target-related spatial responses, while readout training helped extract them consistently. Distributing similar label counts across more videos yielded higher localization hit rates, while synthesis effects varied by dataset and evaluation criterion. Higher localization hit rates did not always improve detection under false-alarm constraints, and failures remained when target signals were weak relative to background variation and under cross-dataset transfer. These findings provide guidance on label allocation, spatial representations and readouts, synthesis, and false-alarm control.
Chinese Translation
无人机（UAV）使用的日益增加使得基于图像的无人机检测变得愈加重要。基于学习的检测器通过图像和注释进行训练，注释类型决定了训练期间可用的信息。我们关注于在传感器或场景变化使得空间注释额外训练负担沉重时，从框架级目标存在/不存在标签中学习定位。我们分析了一种现有架构在小型无人机点检测中的检测能力、学习行为和潜在应用，该架构使用存在/不存在标签进行训练，并且不需要外部检测器。该架构冻结通过分类学习到的空间特征，并使用相同的框架标签训练一个读取模块，以生成空间得分图和点检测。在两个热红外数据集CST Anti-UAV和Anti-UAV410上，我们评估了在虚警约束下的定位命中率和检测率，分析了训练阶段、标签分配、合成和模型配置的影响，并与边界框检测器进行了比较。我们还探讨了使用可见光图像和框架标签进行空中物体跟踪（AOT）的潜在应用。分类训练增强了与目标相关的空间响应，而读取训练则帮助一致地提取这些响应。在更多视频中分配相似的标签数量提高了定位命中率，而合成效果因数据集和评估标准而异。更高的定位命中率并不总是改善虚警约束下的检测，当目标信号相对于背景变化较弱以及在跨数据集迁移时，仍然存在失败。这些发现为标签分配、空间表示和读取、合成以及虚警控制提供了指导。
cs.CV / 50 / 2610.07711
Comprehensive Evaluation and Fine-Tuning of Foundational Cell Nuclei Segmentation Models in Renal Pathology
肾脏病理学中基础细胞核分割模型的综合评估与微调
Abstract
Accurate nuclei instance segmentation is essential for quantitative renal pathology, yet general-purpose models often struggle with low contrast, dense nuclei, complex morphology, and strong background staining. In this work, we extended a human-in-the-loop framework by combining 5,901 foundation-model-generated pseudo-labels from well-segmented cases (Easy), 860 newly expert-annotated unresolved challenging cases (Medium), and 198 expert-annotated consensus failure cases (Hard). These annotations, spanning different levels of segmentation difficulty, enabled the systematic evaluation of seven single-source and mixed-source fine-tuning strategies across nine cell segmentation model configurations. Fine-tuning improved all models, with Medium data included in seven of the nine best-performing strategies. LSP-DETR achieved the highest F1 score of 0.8725 with Hard-only fine-tuning, while StarDist showed the largest improvement, increasing from 0.7380 to 0.8332 with Medium-only fine-tuning. These findings show that annotations spanning multiple difficulty levels support effective model adaptation, although the optimal annotation composition remains model dependent.
Chinese Translation
准确的细胞核实例分割对于定量肾脏病理学至关重要，但通用模型在低对比度、密集核、复杂形态和强背景染色的情况下常常表现不佳。在本研究中，我们通过结合来自已良好分割案例（简单）的5,901个基础模型生成的伪标签、860个新近专家标注的未解决挑战案例（中等）以及198个专家标注的共识失败案例（困难），扩展了一个人机协作框架。这些跨越不同分割难度水平的标注，使得我们能够系统地评估七种单源和混合源的微调策略，涵盖九种细胞分割模型配置。微调改善了所有模型，其中中等数据包含在九种最佳表现策略中的七种。LSP-DETR在仅使用困难数据微调时达到了最高的F1分数0.8725，而StarDist则显示出最大的改进，从0.7380提升至0.8332，使用的是仅中等数据微调。这些发现表明，跨越多个难度水平的标注支持有效的模型适应，尽管最佳的标注组合仍然依赖于具体模型。
cs.CV / 51 / 2610.07720
RefRoute: Decoupling Conditioning Cost from References via Compact Residual Conditioning and Spatial Routing
RefRoute：通过紧凑残差调节和空间路由将调节成本与参考解耦
Abstract
Multi-reference image generation requires preserving the appearance of multiple subjects while composing them into a coherent scene. However, existing diffusion transformers commonly encode references as dense visual token grids and jointly process them with global attention, making conditioning increasingly expensive as the number and resolution of references grow. We present RefRoute, a framework that addresses both reference representation cost and attention overhead through two complementary mechanisms. Compact residual conditioning combines low-resolution latent tokens with lightweight residual features extracted from full-resolution pixels, reducing reference token counts while retaining fine-grained appearance cues. Condition routing and attention routing align reference tokens with their assigned target regions and restrict cross-reference interactions, while allowing selective reference access beyond region boundaries for scene integration. We further introduce RefRoute-Data for training many-reference generation models and ManyRef100, a benchmark spanning human, object, and mixed compositions with 10-17 references. After many-reference fine-tuning, RefRoute achieves an overall Weighted-Ref-VIEScore of 36.06 on ManyRef100, compared with 8.88 for FLUX.2-Klein-9B. Separate inference-cost evaluations show substantially slower latency growth as the reference count increases: at 16 references, our 50-step and 4-step configurations achieve $18.3\times$ and $14.2\times$ speedups over their corresponding FLUX baselines, respectively. These results establish compact reference representations and spatially routed attention as an effective approach to scalable many-reference image generation.
Chinese Translation
多参考图像生成需要在将多个对象组合成一个连贯场景的同时，保持它们的外观。然而，现有的扩散变换器通常将参考编码为密集的视觉标记网格，并通过全局注意力共同处理它们，随着参考数量和分辨率的增加，调节成本变得越来越高。我们提出了RefRoute，一个通过两种互补机制解决参考表示成本和注意力开销的框架。紧凑残差调节将低分辨率潜在标记与从全分辨率像素中提取的轻量级残差特征相结合，减少了参考标记的数量，同时保留了细粒度的外观线索。条件路由和注意力路由将参考标记与其分配的目标区域对齐，并限制跨参考交互，同时允许在区域边界之外选择性地访问参考，以便进行场景整合。我们进一步介绍了RefRoute-Data，用于训练多参考生成模型，以及ManyRef100，一个涵盖人类、物体和混合组合的基准，包含10-17个参考。在多参考微调后，RefRoute在ManyRef100上实现了36.06的整体加权参考视觉效果分数（Weighted-Ref-VIEScore），而FLUX.2-Klein-9B的得分为8.88。单独的推理成本评估显示，随着参考数量的增加，延迟增长显著减缓：在16个参考时，我们的50步和4步配置分别实现了相对于其对应FLUX基线的$18.3 imes$和$14.2 imes$的加速。这些结果确立了紧凑的参考表示和空间路由注意力作为可扩展多参考图像生成的有效方法。
cs.CV / 52 / 2610.07726
Structure-aware Keypoint Localization for Videofluoroscopic Swallowing Study
基于结构感知的关键点定位用于视频透视吞咽研究
Abstract
Videofluoroscopic Swallowing Study (VFSS) is one of the gold standard for diagnosing swallowing disorders, providing dynamic X-ray imaging of the swallowing process. Automated kinematic analysis in VFSS relies fundamentally on precise anatomical keypoint localization. However, existing studies focus on limited keypoints (e.g., cervical vertebrae or the hyoid) and overlook critical regions such as the soft palate, while annotating only active swallowing segments and ignoring abundant non-swallowing data, resulting in poor data efficiency. Moreover, leveraging this unlabeled data via standard semi-supervised learning is suboptimal, as generic methods are prone to spatial bias. In medical X-rays with fixed layouts, models tend to memorize absolute coordinates rather than understanding anatomical structures. To tackle these challenges, we introduce VFSSKep, a novel dataset that extends annotations to the soft palate and incorporates large-scale unlabeled data. We further propose S$^3$KL, a Structure-aware Semi-Supervised Keypoint Localization framework designed to overcome spatial bias. It integrates a Structure-Aware Learning strategy to extract high-resolution structural cues for structure-aware representation learning, and a Structural Representation Consistency Learning strategy with block shuffling to enforce invariant structural recognition. Experiments show our method achieves state-of-the-art semi-supervised performance, even with unlabeled and 25% labeled data surpassing fully supervised learning with 100% labeled data. Code and data will be made publicly available at: https://github.com/kaai520/S3KL.
Chinese Translation
视频透视吞咽研究（Videofluoroscopic Swallowing Study, VFSS）是诊断吞咽障碍的金标准之一，提供吞咽过程的动态X光影像。VFSS中的自动运动学分析基本上依赖于精确的解剖关键点定位。然而，现有研究集中于有限的关键点（例如，颈椎或舌骨），忽视了软腭等关键区域，同时仅对主动吞咽段进行标注，忽略了大量非吞咽数据，导致数据效率低下。此外，通过标准的半监督学习利用这些未标记数据的效果并不理想，因为通用方法容易受到空间偏差的影响。在具有固定布局的医学X光影像中，模型往往倾向于记忆绝对坐标，而不是理解解剖结构。为了解决这些挑战，我们引入了VFSSKep，一个扩展了软腭标注并包含大规模未标记数据的新数据集。我们进一步提出了S$^3$KL，一个结构感知半监督关键点定位框架，旨在克服空间偏差。该框架整合了结构感知学习策略，以提取高分辨率的结构线索进行结构感知表示学习，并采用结构表示一致性学习策略与块随机打乱相结合，以强化不变的结构识别。实验结果表明，我们的方法在半监督性能上达到了最先进水平，即使在未标记和25%标记数据的情况下也超过了100%标记数据的完全监督学习。代码和数据将公开发布于：https://github.com/kaai520/S3KL。
cs.CV / 53 / 2610.07729
Foveated Compression: Selective High-Resolution Preservation for Token-Efficient VLMs
聚焦压缩：针对令牌高效的视觉语言模型的选择性高分辨率保留
Abstract
Visual tokens are a major source of inference cost in vision-language models, yet simple image downsampling remains a surprisingly strong compression baseline. This raises a complementary question: under a fixed token budget, where should visual fidelity be preserved? We introduce Foveated Compression, which encodes a full-resolution image once and represents it with a mixture of native- and compressed-resolution visual tokens. A behaviorally self-distilled Foveated Merger compresses local visual tokens while preserving compatibility with their native counterparts, and a lightweight Foveated Selector chooses one of nine spatial cells to retain at native resolution using exhaustive budget-matched intervention supervision. At 11.11% visual tokens, uniform Foveated Compression shows no significant paired difference from iso-token downsampling. At 20.99%, the learned selector significantly outperforms random and fixed allocation, but remains below strong whole-image resizing, showing that localized fidelity is not universally preferable. A budget-matched region-choice oracle reaches 82.73 macro accuracy versus 69.61 for the learned selector, revealing substantial headroom within the same spatial action space. Matched probing further shows that signals predicting when compression breaks the answer are substantially more accessible after language-model computation than to the lightweight prefill-free selector. These results expose complementary bottlenecks in region selection and compressed-region fidelity.
Chinese Translation
视觉令牌是视觉语言模型中推理成本的主要来源，但简单的图像下采样仍然是一个令人惊讶的强压缩基线。这引出了一个互补的问题：在固定的令牌预算下，应该在哪里保留视觉保真度？我们提出了聚焦压缩（Foveated Compression），该方法一次编码全分辨率图像，并使用本地和压缩分辨率视觉令牌的混合来表示它。行为自蒸馏的聚焦合并器（Foveated Merger）在压缩局部视觉令牌的同时，保持与其本地对应物的兼容性，而轻量级聚焦选择器（Foveated Selector）则使用全面的预算匹配干预监督，从九个空间单元中选择一个以保留本地分辨率。在11.11%的视觉令牌下，均匀的聚焦压缩与等令牌下采样没有显著的配对差异。在20.99%时，学习到的选择器显著优于随机和固定分配，但仍低于强大的全图像调整大小，显示出局部保真度并不总是更可取。与预算匹配的区域选择神谕（oracle）达到了82.73的宏观准确率，而学习选择器为69.61，揭示了在同一空间行动空间内的显著提升空间。匹配探测进一步表明，预测何时压缩破坏答案的信号在语言模型计算后比轻量级无填充选择器更易于获取。这些结果揭示了区域选择和压缩区域保真度中的互补瓶颈。
cs.CV / 54 / 2610.07758
Later Is Better: Token Reduction for ViTs Under Distribution Shift
后期更佳：在分布变化下的视觉变换器的标记减少
Abstract
Training-free token reduction accelerates vision transformers by removing redundant tokens across layers, recovering most of the original accuracy at a fraction of the compute. These methods, however, are designed and evaluated primarily on clean data, and under real-world distribution shift their accuracy gap to the uncompressed model widens with the removal rate. We show that this gap is governed by the reduction schedule, the depth profile of removal, usually left fixed as an implementation detail. Concretely, we introduce a one-parameter late-concentrated power-law schedule that consistently improves out-of-distribution accuracy over flat at no extra inference cost. On ImageNet-C with DeiT-S, the late schedule closes 83% of that gap at a 26% compute reduction (+1.17pp), and 99% of it at a lighter 7% reduction (+0.26pp). The gain cannot be attributed to retaining more tokens or using extra compute: held to flat's compute, the late schedule removes more tokens in total and leaves fewer tokens at the end, yet still wins. Single-layer probes point to a mechanism: earlier reductions perturb features that pass through more remaining layers, front-loading reduction error in depth. The effect is broad, holding across five token-reduction methods (ToMe, EViT, ATS, ATC, PiToMe), nine backbones, all ImageNet-C corruption types, eight further shift suites, and two further modalities, video and vision-language QA. It is also specific to shift, still positive on clean and rising monotonically to ~4x that at the highest severity 5. The schedule keeps its gain under six test-time adaptation methods, and needs no per-input or per-domain tuning.
Chinese Translation
无训练的标记减少通过去除层间冗余标记来加速视觉变换器，在计算量仅为原来的部分时恢复大部分原始准确性。然而，这些方法主要是在干净数据上设计和评估的，在真实世界的分布变化下，去除率与未压缩模型的准确性差距会扩大。我们表明，这一差距受减少调度和去除的深度配置影响，后者通常作为实现细节保持不变。具体而言，我们引入了一种单参数的后期集中幂律调度，该调度在不增加额外推理成本的情况下，始终提高了分布外准确性。在使用DeiT-S的ImageNet-C上，后期调度在26%的计算减少下缩小了83%的差距（+1.17个百分点），在更轻的7%减少下缩小了99%的差距（+0.26个百分点）。这一增益不能归因于保留更多标记或使用额外计算：在保持与平坦调度相同的计算条件下，后期调度总共去除了更多标记，并在最后留下了更少的标记，但仍然表现更佳。单层探测指向一个机制：早期的减少扰动了通过更多剩余层的特征，在深度上前置了减少误差。该效应广泛适用，适用于五种标记减少方法（ToMe、EViT、ATS、ATC、PiToMe）、九种主干网络、所有ImageNet-C腐蚀类型、八个进一步的变化套件，以及两种其他模态，视频和视觉-语言问答。它也特定于变化，在干净数据上仍然表现积极，并在最高严重度5时单调上升至约4倍。该调度在六种测试时适应方法下保持其增益，并且不需要针对每个输入或每个领域的调优。
cs.CV / 55 / 2610.07783
From Laboratory to Road: Evaluating Wearable Gaze Accuracy for Driving
从实验室到道路：评估驾驶中的可穿戴注视精度
Abstract
Bird's-eye-view (BEV) representations have become a widely used interface between perception and planning in autonomous driving, but they encode what is in a scene, not what is behaviorally relevant to a human driver. Gaze offers a compelling behavioral signal for this gap, yet wearable eye trackers are routinely deployed as if their spatial output were ground truth, despite known sensitivity to head motion, illumination, and calibration drift. We present, to our knowledge, the first unified framework for quantifying wearable gaze accuracy under real driving conditions. Our on-road study contains 41 validated scenes in which one driver fixated a vehicle's license plate. Gaze error is measured as the angular difference between the plate center and the gaze direction estimated by the glasses. Separate indoor studies with the same driver and device systematically analyze how distance, illumination, head motion, target motion, and gaze eccentricity affect both systematic bias and gaze precision. The mean on-road error was 4.58 degrees. Applying an offset estimated from the indoor recordings reduced it to 1.10 degrees and improved all 41 scenes. Because this offset varied between sessions, reliable BEV supervision may require online recalibration and condition-dependent estimates of gaze uncertainty.
Chinese Translation
鸟瞰图（BEV）表示已成为自动驾驶中感知与规划之间广泛使用的接口，但它们编码的是场景中的内容，而不是对人类驾驶员行为相关的信息。注视提供了一个引人注目的行为信号来填补这一空白，然而可穿戴眼动仪的空间输出常常被当作真实值使用，尽管已知其对头部运动、光照和校准漂移敏感。我们提出了一个统一框架，用于量化在真实驾驶条件下的可穿戴注视精度，这是我们所知的首个此类框架。我们的道路研究包含41个经过验证的场景，其中一名驾驶员注视了一辆车的车牌。注视误差被测量为车牌中心与眼镜估计的注视方向之间的角度差。与同一驾驶员和设备进行的独立室内研究系统地分析了距离、光照、头部运动、目标运动和注视偏心度如何影响系统偏差和注视精度。平均道路误差为4.58度。应用从室内记录中估计的偏移量将其减少到1.10度，并改善了所有41个场景。由于该偏移量在不同会话之间有所变化，可靠的BEV监督可能需要在线重新校准和依赖条件的注视不确定性估计。
cs.CV / 56 / 2610.07788
Image-Space Refraction Correction for Underwater 3D Reconstruction: Warping Flat-Port Views into Pinhole Perspective
水下三维重建的图像空间折射校正：将平面端口视图扭曲为针孔透视
Abstract
Consumer-grade cameras in flat-port housings are widely used for underwater exploration and mapping of coral reefs and seafloor habitats due to their low cost and accessibility. However, refraction at flat-port interfaces causes bowl-shaped deformation in reconstructed scenes and camera trajectories, compromising the metric accuracy required for mapping and navigation. To remove the dominant refractive distortion before reconstruction, we introduce a physics-based refraction correction in image space. Our method is downstream-agnostic: the refraction-corrected images can be directly used as input to existing reconstruction and SLAM algorithms. We characterize the refractive distortion through ray-tracing simulations and validate our correction on two real underwater datasets with differing scene structures. Compared with conventional and refractive Structure-from-Motion (SfM), our approach removes reconstruction deformation while registering more frames and maintaining low reprojection error. The correction further generalizes across diverse reconstruction and VSLAM backends, demonstrating its broad applicability to downstream vision pipelines.
Chinese Translation
由于成本低廉且易于获取，配备平面端口壳体的消费级相机被广泛应用于水下珊瑚礁和海底栖息地的探索与测绘。然而，平面端口界面的折射导致重建场景和相机轨迹出现碗状变形，影响测绘和导航所需的度量精度。为在重建前消除主要的折射畸变，我们提出了一种基于物理的图像空间折射校正方法。该方法对下游算法无依赖性：经过折射校正的图像可直接作为现有重建和SLAM算法的输入。我们通过光线追踪模拟对折射畸变进行表征，并在两个具有不同场景结构的真实水下数据集上验证了校正效果。与传统及折射结构光束法（Structure-from-Motion, SfM）相比，我们的方法消除了重建变形，同时注册了更多帧且保持了较低的重投影误差。该校正方法进一步在多种重建和视觉SLAM（VSLAM）后端中表现出良好的泛化能力，展示了其对下游视觉处理流程的广泛适用性。
cs.CV / 57 / 2610.07793
Geometry-Constrained Bidirectional Point Cloud Registration for Thin, Sheet-Like Heritage Artifacts
针对薄型、片状遗产文物的几何约束双向点云配准
Abstract
Non-contact three-dimensional reconstruction of thin, sheet-like heritage artifacts poses significant geometric and registration challenges. Due to their fragility, these artifacts cannot be suspended or equipped with artificial markers, necessitating independent acquisition of their front and back surfaces. Subsequent registration proves difficult due to the limited number of shared geometric features and the scarcity of explicit physical constraints, which may result in rotational ambiguity, instability, and structural collapse during iterative optimization. To address these challenges, we propose a geometry-constrained bidirectional point cloud registration method specifically tailored for thin, sheet-like heritage artifacts. The method integrates semantic-guided preprocessing, Principal Component Analysis (PCA)-based geometric normalization, and a thickness-aware registration strategy. The estimated physical thickness is incorporated as a geometric constraint to preserve structural integrity during registration. Rotational ambiguity is resolved by evaluating a finite set of global rotation hypotheses, each refined using the point-to-plane Iterative Closest Point (ICP) algorithm, with the optimal transformation selected via a geometry-aware fitness criterion consistent with the thickness scale. Experimental results show that the proposed method achieves competitive or improved performance in most cases, particularly in projected area consistency and physically plausible front-back alignment. In addition, the thickness-aware constraint and rotation hypothesis evaluation reduce the risk of degenerate configurations in which the two surfaces are incorrectly flipped while still yielding deceptively acceptable numerical scores, supporting reliable non-contact digitization of delicate and thin heritage artifacts. Implementation details are available at https://zyz-nwpu.github.io/GCBPCR/.
Chinese Translation
薄型、片状遗产文物的非接触三维重建面临显著的几何和配准挑战。由于其脆弱性，这些文物无法悬挂或配备人工标记，必须独立获取其正面和背面。由于共享几何特征的数量有限以及显式物理约束的稀缺，后续的配准过程变得困难，这可能导致在迭代优化过程中出现旋转模糊、不稳定性和结构崩溃。为了解决这些挑战，我们提出了一种专门针对薄型、片状遗产文物的几何约束双向点云配准方法。该方法集成了语义引导的预处理、基于主成分分析（PCA）的几何归一化和厚度感知的配准策略。估计的物理厚度被纳入几何约束，以在配准过程中保持结构完整性。通过评估一组有限的全局旋转假设来解决旋转模糊，每个假设使用点到平面的迭代最近点（ICP）算法进行细化，并通过与厚度尺度一致的几何感知适应度标准选择最佳变换。实验结果表明，所提出的方法在大多数情况下实现了具有竞争力或改进的性能，特别是在投影面积一致性和物理上合理的正反面对齐方面。此外，厚度感知约束和旋转假设评估减少了在两个表面错误翻转而仍然产生误导性可接受数值评分的退化配置的风险，支持对精细和薄型遗产文物的可靠非接触数字化。实施细节可在 https://zyz-nwpu.github.io/GCBPCR/ 获取。
cs.CV / 58 / 2610.07795
Efficient Gaussian Splatting Sequence Compression with Standard Video Codecs
基于标准视频编码的高效高斯喷溅序列压缩
Abstract
This paper presents a novel effective Gaussian Splatting (GS) sequence Compression method that utilizes the Video codec (GSCV). Existing video-based GS sequence compression relies on the Parallel Linear Assignment Sorting (PLAS) and tracked primitive information to convert GS into smooth 2D videos. However, tracked information is not available for most practical applications, and without it, using the vanilla PLAS can generate images exhibiting weak inter-frame correlation, due to its stochastic nature. GSCV incorporates a simple yet efficient Inter-PLAS method to produce close images between the I- and P-frames of GS, enhancing the inter-frame performance of video codec greatly. GSCV also realizes a new pipeline based on the state-of-the-art video codecs with high bit-depth GS images, achieving higher compressibility while simultaneously providing a higher quality upper bound. Experimental results show that the proposed GSCV exhibits obviously improved performance over MPEG video and point cloud-based anchors in GS sequence compression. The code is available at https://github.com/Qi-Yangsjtu/GSCV.
Chinese Translation
本文提出了一种新颖有效的高斯喷溅（Gaussian Splatting, GS）序列压缩方法，称为视频编码（GSCV）。现有的基于视频的GS序列压缩依赖于并行线性分配排序（Parallel Linear Assignment Sorting, PLAS）和跟踪的基本信息，将GS转换为平滑的二维视频。然而，对于大多数实际应用而言，跟踪信息并不可用，而在缺乏该信息的情况下，使用普通的PLAS可能会生成表现出较弱帧间相关性的图像，这主要是由于其随机性。GSCV结合了一种简单而高效的Inter-PLAS方法，以在GS的I帧和P帧之间生成相似的图像，极大地增强了视频编码的帧间性能。GSCV还基于最先进的视频编码实现了一种新的处理流程，能够处理高位深度的GS图像，达到更高的压缩率，同时提供更高质量的上限。实验结果表明，所提出的GSCV在GS序列压缩方面明显优于MPEG视频和基于点云的锚点。代码可在https://github.com/Qi-Yangsjtu/GSCV获取。
cs.CV / 59 / 2610.07802
Towards benchmarking Western Bluebird detection in the wild
面向西部蓝鸟在野外检测的基准测试
Abstract
Bird monitoring in natural environments is challenging due to the small size of some species of birds relative to the scene, background clutter, variability in illumination, and the observers' viewpoint. Progress is further limited by the scarcity of large-scale, realistic datasets, which are essential for understanding behavioral patterns. To address this gap, we introduce a new benchmark dataset for the detection and segmentation of Western bluebirds (Sialia Mexicana), comprising over 6,000 labeled images from 41 recording sessions. The dataset features high-resolution (4K) in-the-wild images in which birds occupy only a small fraction of the image. We evaluated supervised detectors, open-vocabulary models under zero-shot and fine-tuned settings, and segmentation approaches. Supervised detectors remain the most reliable overall, with Faster R-CNN achieving the highest detection mAP and RT-DETR offering the best precision-recall trade-off. Open-vocabulary models perform poorly in zero-shot settings; however, fine-tuning substantially improves their performance, with YOLO-World becoming competitive with supervised methods and achieving the highest precision, F1-score, and
[email protected]. For segmentation, supervised methods significantly outperform Grounded-SAM and SAM 3: Mask R-CNN achieves the highest mask mAP, while YOLOv8-Seg provides the best precision and fastest inference. A diagnostic analysis further shows that failures are not explained by object size alone, but by a combination of apparent scale, brightness, contrast, clutter, blur, crowding, and recording-session variation. Overall, our findings highlight the difficulty of zero-shot bird detection in cluttered ecological scenes and underscore the importance of domain adaptation in small-object settings.
Chinese Translation
在自然环境中监测鸟类具有挑战性，因为某些鸟类的体型相对于场景较小，背景杂乱，光照变化以及观察者的视角差异。进展进一步受到大规模、真实数据集稀缺的限制，而这些数据集对理解行为模式至关重要。为了解决这一问题，我们引入了一个新的基准数据集，用于西部蓝鸟（Sialia Mexicana）的检测和分割，该数据集包含来自41个录制会话的6000多张标注图像。该数据集具有高分辨率（4K）的野外图像，其中鸟类仅占图像的一小部分。我们评估了监督检测器、在零样本和微调设置下的开放词汇模型以及分割方法。总体而言，监督检测器仍然是最可靠的，Faster R-CNN实现了最高的检测mAP，而RT-DETR提供了最佳的精确率-召回率权衡。开放词汇模型在零样本设置下表现不佳；然而，微调显著提高了它们的性能，YOLO-World与监督方法竞争，并实现了最高的精确率、F1分数和
[email protected]。在分割方面，监督方法显著优于Grounded-SAM和SAM 3：Mask R-CNN实现了最高的掩膜mAP，而YOLOv8-Seg提供了最佳的精确率和最快的推理速度。诊断分析进一步表明，失败并不仅仅由物体大小解释，而是由明显的尺度、亮度、对比度、杂乱、模糊、拥挤和录制会话变化的组合所导致。总体而言，我们的研究结果突显了在杂乱生态场景中零样本鸟类检测的困难，并强调了在小物体设置中领域适应的重要性。
cs.CV / 60 / 2610.07843
CHARTER: Auditing Reference Substitution in Hierarchical Compact-Evidence Evaluation for Computational Pathology
CHARTER：在计算病理学中对层次紧凑证据评估中的参考替代进行审计
Abstract
In digital pathology, compact evidence is often used to explain or audit predictions made by whole-slide image multiple instance learning models. In hierarchical compact-evidence pipelines, candidate filtering introduces a strategy-specific candidate-conditioned prediction alongside the original full-bag prediction. If the evaluation reference changes while the intended target remains the original full-bag prediction, however, not only can the measured fidelity of the same compact evidence change, but comparisons between competing candidate strategies can also change. To make this dependence explicit, we introduce CHARTER, a reference-aware evaluation charter that asks researchers to DECLARE the intended target and reference, QUANTIFY candidate-induced prediction shift, and AUDIT the stability of comparative conclusions. Across the 15 comparisons in our main five-seed Random-K audit, 4 showed determinate reversals; in a matched native-ranking stress test, the ACMIL comparison changed from REVERSED to PRESERVED. CHARTER turns otherwise implicit candidate-filtering and reference choices into an auditable evaluation specification, helping distinguish genuine preservation of the intended prediction from apparent gains induced by changing the prediction being explained.
Chinese Translation
在数字病理学中，紧凑证据常用于解释或审计由全切片图像多实例学习模型所做的预测。在层次紧凑证据管道中，候选过滤引入了一种特定策略的候选条件预测，与原始的全包预测并存。然而，如果评估参考发生变化，而预期目标仍然是原始的全包预测，那么同一紧凑证据的测量保真度不仅可能发生变化，竞争候选策略之间的比较也可能发生变化。为了明确这种依赖关系，我们引入了CHARTER，一个参考感知的评估章程，要求研究人员声明预期目标和参考，量化候选引起的预测变化，并审计比较结论的稳定性。在我们主要的五种种子随机K审计中的15次比较中，有4次显示出确定性的反转；在一个匹配的本地排名压力测试中，ACMIL比较从反转（REVERSED）变为保留（PRESERVED）。CHARTER将原本隐含的候选过滤和参考选择转变为可审计的评估规范，帮助区分预期预测的真实保留与因改变被解释预测而产生的表面收益。
cs.CV / 61 / 2610.07868
CueRator: Agentic Search for Symbolic Rules to Adapt Frozen Multimodal Encoders
CueRator：用于适应冻结多模态编码器的自主搜索符号规则
Abstract
Large language model agents have been used to search over symbolic structures such as programs and equations. We propose CueRator, an agentic framework for policy-aware decision-rule discovery, which adapts frozen contrastive multimodal encoders by searching for the decision rule that converts their cross-modal similarities into predictions. We validate it on open-vocabulary audio-visual event perception, where existing methods involve a trade-off between adaptivity and generalization to unseen categories: trained modules adapt at the cost of generalization, and fixed rules the reverse. The framework pairs a symbolic formulation for generalization with a lightweight policy that predicts its parameters per video for adaptivity. A report-guided multi-agent loop discovers the formulation offline, evaluating each candidate on its expressive ceiling and on whether a trained policy can realize it. On OV-AVEBench, CueRator raises the total average from 57.8 to 60.2 and unseen-category performance from 55.8 to 59.9 over the best existing method, reducing the seen-unseen gap from 7.1 to 1.2. Ablations attribute the gains to both the formulation and the policy and show that both feedback signals are necessary for effective search. CueRator also improves over the respective baselines on two further audio-visual event perception tasks, and the discovered rule remains competitive across encoders with only the policy retrained. Code is available at https://github.com/cvsp-lab/cuerator.
Chinese Translation
大型语言模型代理已被用于在程序和方程等符号结构上进行搜索。我们提出了CueRator，一个用于策略感知决策规则发现的自主框架，它通过搜索将跨模态相似性转换为预测的决策规则来适应冻结的对比多模态编码器。我们在开放词汇音频-视觉事件感知上验证了该框架，现有方法在适应性和对未见类别的泛化之间存在权衡：训练模块以牺牲泛化为代价进行适应，而固定规则则相反。该框架将用于泛化的符号表述与轻量级策略相结合，后者为每个视频预测其参数以实现适应性。报告引导的多代理循环在离线发现表述，评估每个候选者的表达上限以及训练策略是否能够实现它。在OV-AVEBench上，CueRator将总平均值从57.8提高到60.2，将未见类别的性能从55.8提高到59.9，缩小了已见与未见之间的差距，从7.1减少到1.2。消融实验将增益归因于表述和策略，并表明这两种反馈信号对于有效搜索都是必要的。CueRator在另外两个音频-视觉事件感知任务上也优于各自的基线，且发现的规则在仅重新训练策略的情况下在不同编码器上保持竞争力。代码可在https://github.com/cvsp-lab/cuerator获取。
cs.CV / 62 / 2610.07883
Revar3r: gauge-aware perturbation uncertainty for feed-forward 3d reconstruction
Revar3r：面向量测的扰动不确定性在前馈3D重建中的应用
Abstract
A correctly reconstructed distant point appears uncertain even when a frozen 3D model processes equivalent inputs because its output frame rotates fractionally. This exposes a weakness of trainingfree perturbation uncertainty: when outputs contain an unobserved symmetry, run-to-run variation potentially reflects symmetry rather than error. Existing alternatives have trade-offs: built-in confidence is outperformed in most evaluated conditions, while trained evidential heads require modelspecific supervision. For point maps, this research derives a closed-form, error-independent variance term that grows with scene extent and potentially overwhelms the desired signal. Simulation reproduces the effect; all 30 real VGGT view-sets tested exhibit its predicted $\|x_p\|^2$ signature. ReVar3R robustly registers predictions to a common similarity frame before computing per-point variance, without retraining or modifying the frozen model. Optional calibration and fusion use a held-out split. Across VGGT, {\pi}3, and MASt3R on six datasets, the same estimator on every backbone lowers AUSE below built-in confidence in 15 of 18 conditions. The staged evaluation yields 11 of 18 wins for the label-free core, 12/18 for label-free equal-weight fusion, 14/18 with held-out weights, and 15/18 when the built-in signal is included. Against a trained evidential head, the result is a trade-off: the head calibrates magnitude better and leads in its training domain, whereas ReVar3R transfers across backbones without adaptation. Its ranking improves point filtering, but it does not detect stable systematic bias, aid novel-view synthesis, or transfer calibration across domains.
Chinese Translation
即使在冻结的3D模型处理等效输入时，正确重建的远处点也会显得不确定，因为其输出帧会发生微小旋转。这暴露了无训练扰动不确定性的一个弱点：当输出包含未观察到的对称性时，运行间的变化可能反映的是对称性而非误差。现有的替代方案存在权衡：内置置信度在大多数评估条件下表现不佳，而训练的证据头则需要特定模型的监督。对于点图，本研究推导出一个闭式的、与误差无关的方差项，该方差项随着场景范围的增加而增长，并可能淹没所需信号。模拟重现了这一效应；所有测试的30个真实VGGT视图集均展现出其预测的$ ext{||}x_p ext{||}^2$特征。ReVar3R在计算每个点的方差之前，稳健地将预测注册到一个共同的相似性框架中，而无需重新训练或修改冻结模型。可选的校准和融合使用了一个保留的拆分。在VGGT、{ ext{π}}3和MASt3R的六个数据集上，所有骨干网使用相同的估计器使AUSE在18种条件中的15种低于内置置信度。分阶段评估结果显示，标签无关核心在18种条件中赢得了11场，标签无关的等权融合赢得了12场，使用保留权重赢得了14场，而在包含内置信号时赢得了15场。与训练的证据头相比，结果是一个权衡：该头在幅度校准上表现更好，并在其训练领域领先，而ReVar3R则在不适应的情况下跨骨干网转移。其排名改善了点过滤，但未能检测到稳定的系统偏差、辅助新视图合成或跨领域转移校准。
cs.CV / 63 / 2610.07903
Unsupervised Long-Tailed Adaptation of Vision-Language Models
无监督长尾适应的视觉-语言模型
Abstract
Adapting vision-language models to downstream tasks has achieved remarkable success by leveraging pseudo-labels generated from unlabeled data. Existing methods typically assume a uniform unlabeled data distribution, and thus the resulting pseudo-label distribution is likewise uniform. However, real-world data distributions are often long-tailed. To tackle this, we formalize a new scenario termed Unsupervised Long-Tailed Adaptation (ULTA). Under this scenario, existing methods exhibit a contrasting phenomenon: head-class performance drops sharply, which is distinct from supervised long-tailed learning where tail classes suffer the most. In particular, we uncover that the distributional mismatch not only erodes head-class boundaries, but also pushes head samples into confusable classes, reinforcing the model's inherent bias. To address these issues, we propose a novel model called Margin-Aware Refinement with Structural alignment (MARS). Specifically, we mitigate head-class boundary erosion via Boundary-Preserving Alignment, which takes the zero-shot VLM as a fixed visual reference to suppress probability increases that lack visual support in the training targets. Building upon this, we introduce Margin-aware Self-Refinement, which employs a dynamic adjustment strategy to refine tail and confusable classes while preventing prediction bias. Extensive experiments on nine benchmark datasets demonstrate that MARS outperforms state-of-the-art methods, achieving an average accuracy improvement of 4.71 percentage points.
Chinese Translation
将视觉-语言模型适应于下游任务已通过利用从未标记数据生成的伪标签取得了显著成功。现有方法通常假设未标记数据分布是均匀的，因此生成的伪标签分布也是均匀的。然而，现实世界中的数据分布往往是长尾分布。为了解决这个问题，我们正式提出了一种新的场景，称为无监督长尾适应（Unsupervised Long-Tailed Adaptation, ULTA）。在这一场景下，现有方法表现出一种对比现象：头类的性能急剧下降，这与监督长尾学习中尾类受到的影响截然不同。特别地，我们发现分布不匹配不仅侵蚀了头类边界，还将头样本推入可混淆类，强化了模型的固有偏见。为了解决这些问题，我们提出了一种新模型，称为边界感知结构对齐的细化（Margin-Aware Refinement with Structural alignment, MARS）。具体而言，我们通过边界保持对齐（Boundary-Preserving Alignment）来减轻头类边界的侵蚀，该方法将零-shot 视觉语言模型（zero-shot VLM）作为固定视觉参考，以抑制在训练目标中缺乏视觉支持的概率增加。在此基础上，我们引入了边界感知自我细化（Margin-aware Self-Refinement），该方法采用动态调整策略来细化尾类和可混淆类，同时防止预测偏差。在九个基准数据集上的大量实验表明，MARS 超越了最先进的方法，平均准确率提高了 4.71 个百分点。
cs.CV / 64 / 2610.07911
Diverse Motion Customization via Control-based Dynamic Optimization
基于控制的动态优化实现多样化运动定制
Abstract
Despite recent advances in video generation, motion customization remains challenging due to content leakage, where appearance attributes from the reference video unintentionally propagate into the generated output. We identify this issue as a consequence of the generative process collapsing toward the reference video, which arises from formulating the learning objective as a direct regression on the reference. To address this, we propose Control-based Motion Customization (CMC), a principled training framework that is structurally robust to content leakage. Our key idea is to steer generative dynamics toward desired motion while avoiding collapse toward the reference video, which we formalize using Stochastic Optimal Control (SOC). Under this formulation, customized videos acquire the target motion yet remain within the pre-trained model's prompt-conditional distribution, where appearance is determined by the text prompt rather than the reference video. Furthermore, to improve efficiency, we tailor the SOC formulation to motion customization by eliminating the need for an explicit reward and introducing a timestep-adaptive motion cost that focuses only on early generative stages, accelerating training by 2.5 times. Extensive experiments demonstrate that CMC effectively mitigates content leakage and achieves competitive motion fidelity while preserving the diversity of the base model across diverse scenarios.
Chinese Translation
尽管视频生成技术近年来取得了进展，但运动定制仍然面临挑战，主要是由于内容泄漏，即参考视频中的外观属性无意中传播到生成的输出中。我们将这一问题视为生成过程向参考视频崩溃的结果，这种崩溃源于将学习目标直接设定为对参考的回归。为了解决这一问题，我们提出了基于控制的运动定制（Control-based Motion Customization, CMC），这是一个在结构上对内容泄漏具有鲁棒性的原则性训练框架。我们的关键思想是引导生成动态朝向期望的运动，同时避免向参考视频的崩溃，我们使用随机最优控制（Stochastic Optimal Control, SOC）对其进行了形式化。在这一框架下，定制的视频获得目标运动，同时仍然保持在预训练模型的提示条件分布内，其中外观由文本提示决定，而非参考视频。此外，为了提高效率，我们通过消除对显式奖励的需求并引入一个时间步自适应的运动成本，专注于早期生成阶段，从而对SOC的形式进行了定制，使训练速度提高了2.5倍。大量实验表明，CMC有效减轻了内容泄漏，并在保持基础模型在多样化场景中的多样性的同时，实现了竞争性的运动保真度。
cs.CV / 65 / 2610.07913
Multimodal Knowledge Distillation for Gastric Adenocarcinoma Classification from Whole-Slide Images
基于多模态知识蒸馏的全幻灯片图像胃腺癌分类
Abstract
Gastric adenocarcinoma (GA) is a leading cause of cancer-related mortality worldwide, and accurate histopathological subtype classification from whole-slide images (WSIs) is essential for effective treatment planning. While multimodal approaches that integrate pathology report text with WSIs can improve classification, existing methods often depend on computationally expensive transformer architectures and large language models. We propose a multimodal knowledge distillation (MKD) framework that combines a pretrained WSI image encoder and a clinical text encoder using Low-Rank Multimodal Fusion (LMF) to efficiently model cross-modal interactions during training. Each WSI is represented as a bag of patches paired with a slide-level diagnostic caption. The teacher model learns fused image-text representations for subtype classification, while the student model distills this knowledge to enable accurate image-only inference. We evaluate our method on the PatchGastric benchmark dataset and achieve at least 3.35% higher mean accuracy than state-of-the-art approaches, without relying on transformer-based fusion, multi-task learning, or large language models. The source code is available at https://github.com/helomelo1/MKD-LMF.
Chinese Translation
胃腺癌（GA）是全球癌症相关死亡的主要原因，准确从全幻灯片图像（WSIs）中进行组织病理亚型分类对于有效的治疗规划至关重要。虽然将病理报告文本与WSIs结合的多模态方法可以提高分类性能，但现有方法通常依赖于计算成本高昂的变换器架构和大型语言模型。我们提出了一种多模态知识蒸馏（MKD）框架，该框架结合了预训练的WSI图像编码器和临床文本编码器，使用低秩多模态融合（LMF）在训练过程中高效建模跨模态交互。每个WSI被表示为一组与幻灯片级诊断标题配对的图像块。教师模型学习用于亚型分类的融合图像-文本表示，而学生模型则蒸馏这一知识，以实现准确的仅图像推断。我们在PatchGastric基准数据集上评估了我们的方法，取得了至少3.35%的平均准确率高于最先进的方法，而无需依赖基于变换器的融合、多任务学习或大型语言模型。源代码可在https://github.com/helomelo1/MKD-LMF获取。
cs.CV / 66 / 2610.07916
Can We Model the Artifacts Explicitly? Disentangle Artifacts via Pairwise Edit Relations for Image Manipulation Localization
我们能否明确建模伪影？通过成对编辑关系解构伪影以实现图像操控定位
Abstract
Image Manipulation Localization (IML) is commonly formulated as a fully supervised learning task that estimates the optimal manipulation mask $y$ for a given image $x$. In this work, we first reveal the latent nature of artifacts and thus reinterpret IML as a latent-variable problem, $P(y|x)=\int P(y|z)\,P(z|x)\,dz$, where $z$ denotes the artifacts. Following this interpretation, we pinpoint the cause for the current IML models' insufficiency as their implicit artifacts modeling strategy, highlighting the necessity of modeling $z$ in an explicit manner. Without direct labels, feature disentanglement is the most appropriate solution for this explicit modeling. Accordingly, we propose a two-stage learning paradigm with the Pairwise Artifacts Learning (PAL) and Standard Localization (SL) phases to estimate $P(z|x)$ and $P(y|z)$ via edit relations. To support our edit-relation-based learning, we further curate EditGroup-45K, a source-anchored dataset organized into edit groups for pair construction. Extensive experiments show that our PAL paradigm yields consistent improvements across diverse IML architectures, and empirical analyses further verify that PAL does capture artifacts explicitly through feature disentanglement. Code and dataset are available at https://github.com/venus-guangjian/PAL
Chinese Translation
图像操控定位（Image Manipulation Localization, IML）通常被表述为一个完全监督的学习任务，旨在为给定图像 $x$ 估计最佳操控掩码 $y$。在本研究中，我们首先揭示了伪影的潜在特性，从而将 IML 重新解释为一个潜变量问题，即 $P(y|x)= ext{int} P(y|z)\,P(z|x)\,dz$，其中 $z$ 表示伪影。基于这一解释，我们指出当前 IML 模型不足的原因在于其隐式伪影建模策略，强调了以显式方式建模 $z$ 的必要性。在没有直接标签的情况下，特征解构是实现这种显式建模的最合适解决方案。因此，我们提出了一种两阶段学习范式，包括成对伪影学习（Pairwise Artifacts Learning, PAL）和标准定位（Standard Localization, SL）阶段，通过编辑关系来估计 $P(z|x)$ 和 $P(y|z)$。为了支持基于编辑关系的学习，我们进一步整理了 EditGroup-45K，这是一个组织成编辑组以便于成对构建的源锚定数据集。大量实验表明，我们的 PAL 范式在多种 IML 架构中均能实现一致的改进，实证分析进一步验证了 PAL 确实通过特征解构显式捕捉伪影。代码和数据集可在 https://github.com/venus-guangjian/PAL 获取。
cs.CV / 67 / 2610.07925
TF-PRVR: Training-Free Partially Relevant Video Retrieval
TF-PRVR：无训练的部分相关视频检索
Abstract
Partially Relevant Video Retrieval (PRVR) aims to retrieve untrimmed videos containing moments relevant to a given text query. Despite recent progress, existing PRVR methods suffer from two key limitations: a fixed video decomposition scheme that causes semantic dilution, and source-domain overfitting induced by task-specific training. In this paper, we propose TF-PRVR, the first training-free framework for PRVR. TF-PRVR leverages frozen vision-language features to construct video-specific hierarchical representations. It derives temporal semantic signals from frame-level features and applies frequency-based multi-scale analysis to identify adaptive temporal boundaries, producing hierarchical segments with coherent event-level semantics. Built on these segments, TF-PRVR constructs a unified multi-scale graph and propagates query relevance across temporally and semantically related nodes. A moment-aware scoring strategy then aggregates temporally aligned relevance across scales, emphasizing consistently supported moments while suppressing isolated false responses. Without task-specific training, TF-PRVR preserves the general-purpose alignment capability of pre-trained vision-language models and avoids dataset-specific overfitting. Extensive experiments demonstrate consistent performance across datasets with diverse visual and temporal characteristics, suggesting a practical direction for training-free PRVR.
Chinese Translation
部分相关视频检索（PRVR）旨在检索包含与给定文本查询相关的时刻的未剪辑视频。尽管近期取得了一些进展，现有的PRVR方法仍然面临两个主要限制：固定的视频分解方案导致语义稀释，以及由任务特定训练引起的源域过拟合。本文提出了TF-PRVR，这是第一个无训练的PRVR框架。TF-PRVR利用冻结的视觉-语言特征构建视频特定的层次表示。它从帧级特征中提取时间语义信号，并应用基于频率的多尺度分析来识别自适应时间边界，生成具有连贯事件级语义的层次段落。在这些段落的基础上，TF-PRVR构建了一个统一的多尺度图，并在时间和语义相关的节点之间传播查询相关性。然后，基于时刻感知的评分策略聚合跨尺度的时间对齐相关性，强调一致支持的时刻，同时抑制孤立的错误响应。TF-PRVR在没有任务特定训练的情况下，保持了预训练视觉-语言模型的通用对齐能力，避免了数据集特定的过拟合。大量实验表明，在具有多样视觉和时间特征的数据集上表现出一致的性能，暗示了无训练PRVR的实际方向。
cs.CV / 68 / 2610.07928
Dynamic Alignment and Calibration for Multimodal Learning
动态对齐与校准的多模态学习
Abstract
Dynamic multimodal learning aims to learn robust representations by adaptively modeling information discrepancies across modalities. However, existing methods still suffer from two limitations: (i) static cross-modal alignment strategies usually impose uniform constraints on all samples while overlooking sample-wise variations, potentially leading to unreasonable over-alignment; and (ii) confidence- or uncertainty-aware fusion methods often fail to adequately account for feature magnitude and confidence differences across modalities. For modality pairs with significant feature magnitude differences or small confidence gaps, it might be unreliable to strictly align fusion weights according to confidence. To address these issues, we propose an Alignment- and Calibration-driven Multimodal Learning framework (ACML). Specifically, ACML incorporates a dynamic cross-modal triplet alignment module, which enforces strong semantic consistency for high-confidence positive pairs while encouraging diverse representation learning between high- and low-confidence positive pairs according to their confidence gaps. Additionally, ACML introduces a difference-aware attention calibration strategy that adaptively adjusts attention regularization based on feature magnitude and confidence differences across modalities, thereby mitigating biases caused by unreasonable fusion constraints. Extensive experiments on multiple multimodal benchmark datasets demonstrate that ACML consistently achieves superior performance and robustness over recent state-of-the-art methods.
Chinese Translation
动态多模态学习旨在通过自适应建模模态间的信息差异来学习鲁棒的表示。然而，现有方法仍然存在两个局限性：（i）静态的跨模态对齐策略通常对所有样本施加统一的约束，而忽视了样本间的变化，这可能导致不合理的过度对齐；（ii）基于置信度或不确定性的融合方法往往未能充分考虑模态间特征幅度和置信度的差异。对于特征幅度差异显著或置信度差距较小的模态对，严格根据置信度对融合权重进行对齐可能是不可靠的。为了解决这些问题，我们提出了一种基于对齐和校准的多模态学习框架（ACML）。具体而言，ACML结合了一个动态的跨模态三元组对齐模块，该模块对高置信度正样本对施加强语义一致性，同时根据置信度差距鼓励高置信度和低置信度正样本对之间的多样化表示学习。此外，ACML引入了一种差异感知的注意力校准策略，该策略根据模态间的特征幅度和置信度差异自适应调整注意力正则化，从而减轻因不合理融合约束造成的偏差。在多个多模态基准数据集上的广泛实验表明，ACML在性能和鲁棒性上始终优于近期的最先进方法。
cs.CV / 69 / 2610.07939
CCDF: A Benchmark Dataset for Deepfake Detection in Real-World Surveillance Footage
CCDF：用于真实世界监控视频深度伪造检测的基准数据集
Abstract
Due to rapid advances in Generative AI, commercial video generation tools can be used to produce fabricated surveillance footage that can fool both human viewers and automated synthetic video detectors. Since these tools are so widely accessible, a malicious user can create a harmful video clip at minimal cost. The production and dissemination of such videos in high-stakes settings, such as crime reporting and elections, can misdirect emergency response efforts or distort political discourse. Existing deepfake video datasets, used by the research community to develop deepfake detection algorithms, exhibit two limitations: (1) they emphasize benign web content rather than footage of possibly malicious activity, and (2) they rely on older or open-source generators that do not represent recent advances in generative systems. We assemble CCtv DeepFakes (CCDF), a video deepfake dataset, to address both gaps. CCDF contains 1840 videos (460 real and 1380 generated) spanning 16 crime and accident categories, with generated content produced using three leading commercial systems: Grok Imagine, Google VEO 3.1, and OpenAI Sora 2. CCDF is a highly realistic, small-scale, manually annotated dataset targeting evaluation of detection models. We release three versions of the dataset: the raw generated data, a cleaned version in which video metadata are standardized between real and synthetic samples to prevent detectors from exploiting trivial cues, and an altered version simulating low-effort post-processing attacks. We evaluate CCDF with ten recent state-of-the-art detectors covering different detection approaches. Our results suggest that these approaches do not reliably distinguish CCDF's generated videos from real ones, despite their strong reported performance on existing datasets. These results further confirm that existing datasets are not well-suited to evaluating certain threats.
Chinese Translation
由于生成性人工智能的快速发展，商业视频生成工具可以用于制作虚假的监控视频，这些视频能够欺骗人工观众和自动化的合成视频检测器。由于这些工具的广泛可获取性，恶意用户可以以最低成本创建有害的视频片段。在高风险环境中，例如犯罪报告和选举，这类视频的制作和传播可能会误导应急响应工作或扭曲政治话语。现有的深度伪造视频数据集，研究社区用于开发深度伪造检测算法，存在两个局限性：（1）它们强调的是良性网络内容，而非可能的恶意活动视频；（2）它们依赖于较旧或开源的生成器，未能代表生成系统的最新进展。为了解决这两个问题，我们组建了CCtv DeepFakes (CCDF)，一个视频深度伪造数据集。CCDF包含1840个视频（460个真实视频和1380个生成视频），涵盖16个犯罪和事故类别，生成内容使用三种领先的商业系统制作：Grok Imagine、Google VEO 3.1和OpenAI Sora 2。CCDF是一个高度真实的小规模手动标注数据集，旨在评估检测模型。我们发布了三个版本的数据集：原始生成数据、一个清理版本，其中视频元数据在真实和合成样本之间进行了标准化，以防止检测器利用微不足道的线索，以及一个模拟低努力后处理攻击的修改版本。我们使用十种最新的最先进检测器对CCDF进行了评估，这些检测器涵盖不同的检测方法。我们的结果表明，尽管这些方法在现有数据集上报告了强大的性能，但它们并不能可靠地区分CCDF的生成视频和真实视频。这些结果进一步确认了现有数据集不适合评估某些威胁。
cs.CV / 70 / 2610.07954
Revisiting Numerical Forecasting Models for Language-Based Trajectory Prediction
重新审视基于语言的轨迹预测的数值预测模型
Abstract
Language-based trajectory predictors represent coordinates as discrete tokens and learn auxiliary tasks such as destination and group reasoning. This formulation enables the model to capture behavioral intent and social context beyond coordinate dynamics alone. However, token-level objectives provide only indirect guidance for continuous coordinate-space dynamics. To address this limitation, we introduce MoRE (Mixture of Reward Experts), a refinement framework that transfers numerical forecasting priors into a pretrained language-based predictor through reinforcement learning. Five frozen numerical predictors provide complementary coordinate-level knowledge of motion and interactions. Their predictions are converted into expert rewards and combined through an uncertainty-weighted consensus that penalizes disagreement. A ground-truth reward anchors the prediction to the target trajectory. To focus refinement on difficult cases, MoRE refines the policy using the top 1% of training samples ranked by predictive entropy. Expert predictions are computed once and cached before PPO training, so the experts are not run during policy updates or inference. In this way, MoRE combines the contextual modeling of the language-based predictor with coordinate-level feedback from numerical experts. On ETH-UCY, MoRE reduces ADE from 0.22 to 0.20 m and FDE from 0.32 to 0.29 m. Relative to the base policy, ADE decreases by 17.9% on SDD and 12.7% on NBA. On ETH-UCY, MoRE also reduces collision rates and better matches ground-truth pedestrian spacing, without increasing measured inference memory or latency. The project page is available at https://jungyu0413.github.io/MoRE/.
Chinese Translation
基于语言的轨迹预测器将坐标表示为离散的标记，并学习辅助任务，如目的地和群体推理。这种表述使模型能够捕捉行为意图和社会背景，而不仅仅是坐标动态。然而，标记级目标仅为连续坐标空间动态提供间接指导。为了解决这一局限性，我们引入了 MoRE（Mixture of Reward Experts），一个通过强化学习将数值预测先验转移到预训练的基于语言的预测器的精炼框架。五个冻结的数值预测器提供了运动和交互的互补坐标级知识。它们的预测被转换为专家奖励，并通过惩罚不一致的加权共识进行组合。真实奖励将预测锚定到目标轨迹。为了将精炼重点放在困难案例上，MoRE 使用按预测熵排名的前 1% 训练样本来精炼策略。专家预测在 PPO 训练之前计算一次并缓存，因此在策略更新或推理期间不运行专家。通过这种方式，MoRE 将基于语言的预测器的上下文建模与来自数值专家的坐标级反馈相结合。在 ETH-UCY 数据集上，MoRE 将 ADE 从 0.22 降低到 0.20 米，FDE 从 0.32 降低到 0.29 米。相对于基础策略，ADE 在 SDD 上减少了 17.9%，在 NBA 上减少了 12.7%。在 ETH-UCY 上，MoRE 还减少了碰撞率，并更好地匹配真实行人间距，而没有增加测量的推理内存或延迟。项目页面可访问 https://jungyu0413.github.io/MoRE/。
cs.CV / 71 / 2610.07958
DensiTok: Making Feed-Forward 3D Gaussian Splatting See More Views Than It Is Given
DensiTok：让前馈式3D高斯点云渲染看到比给定视图更多的视角
Abstract
Feed-forward 3D Gaussian Splatting (3DGS) reconstructs a scene in a single forward pass, replacing per-scene optimization with a network trained across many scenes. Its quality, however, degrades sharply as the number of input images drops. The bottleneck is upstream of the reconstruction heads: from a few unposed views, the internal representation they read carries no evidence for unobserved regions, leaving holes, floaters, and blur. The common remedy supplies that evidence as pixels, synthesizing extra views with an image or video generator and re-encoding them, which is costly and not 3D-consistent by construction. We instead densify the evidence itself. We present DensiTok, a plug-in module for pretrained feed-forward 3DGS models that densifies their internal geometry tokens directly, making a frozen backbone behave as though it had observed many more views than it was given. DensiTok compresses those tokens into a compact latent space, completes the latents of the unobserved viewpoints in a single flow-matching step conditioned on camera geometry, and decodes them back into tokens that the original reconstruction heads. The same module design can be integrated into different pretrained predictors while keeping each backbone and its reconstruction heads frozen. Completion in a low-dimensional latent space requires no image synthesis or additional encoder passes. Across three pretrained backbones and two benchmarks, DensiTok consistently improves sparse-view reconstruction and recovers much of the gap to dense-view reconstruction.
Chinese Translation
前馈式3D高斯点云渲染（3DGS）通过单次前向传递重建场景，替代了逐场景优化，使用在多个场景上训练的网络。然而，随着输入图像数量的减少，其重建质量急剧下降。瓶颈出现在重建头的上游：从少量未定位的视图中，它们读取的内部表示没有未观察区域的证据，导致出现空洞、漂浮物和模糊。常见的解决方案是通过图像或视频生成器合成额外视图，并重新编码这些视图，以提供这些证据，这种方法成本高且在结构上不具备3D一致性。我们则选择直接密集化证据本身。我们提出了DensiTok，一个用于预训练前馈式3DGS模型的插件模块，能够直接密集化其内部几何标记，使得一个冻结的主干网络表现得像是观察了比实际给定的更多视图。DensiTok将这些标记压缩到一个紧凑的潜在空间，在一个单一的流匹配步骤中，基于相机几何条件完成未观察视点的潜在表示，并将其解码回原始重建头所需的标记。相同的模块设计可以集成到不同的预训练预测器中，同时保持每个主干网络及其重建头的冻结。在低维潜在空间中的完成过程无需图像合成或额外的编码器传递。在三个预训练主干和两个基准测试中，DensiTok始终改善稀疏视图重建，并缩小了与密集视图重建之间的差距。
cs.CV / 72 / 2610.07969
EmbodiedSmith: Scaling Embodied Data through Recursive Self-Improvement Flywheel in Simulation
EmbodiedSmith：通过递归自我改进飞轮在仿真中扩展具身数据
Abstract
Scaling robotic foundation models requires diverse training data and reliable evaluation environments. Simulation offers a scalable solution, yet existing generation pipelines remain constrained by predefined assets and skills, a disconnect between scene generation and task generation, and limited support for complex embodiments and physics. We introduce EmbodiedSmith, a framework for scalable embodied data generation through recursive self-improvement (RSI). EmbodiedSmith unifies asset, scene, and task generation in a pipeline that supports autonomous creation and language-driven customization. Its core is an agentic refinement loop: scene generation anticipates downstream task requirements, while task generation guides targeted scene edits, allowing scenes and tasks to iteratively improve one another. This joint refinement improves task generation success, including for long-horizon tasks. The framework further supports mobile manipulators, humanoids, and dexterous hands, as well as interactions involving deformable objects and fluids, broadening the range of behaviors and physical phenomena represented in generated data. Together, these capabilities provide a flexible simulation engine for both robot pretraining and evaluation. Extensive experiments validate the quality, diversity, and generation efficiency of the resulting data, while downstream policy experiments demonstrate that increased data diversity improves generalization.
Chinese Translation
扩展机器人基础模型需要多样化的训练数据和可靠的评估环境。仿真提供了一种可扩展的解决方案，但现有的生成管道仍受到预定义资产和技能的限制，场景生成与任务生成之间存在脱节，并且对复杂具身和物理现象的支持有限。我们提出了EmbodiedSmith，一个通过递归自我改进（RSI）实现可扩展具身数据生成的框架。EmbodiedSmith在一个支持自主创建和基于语言定制的管道中统一了资产、场景和任务生成。其核心是一个智能化的精炼循环：场景生成预见下游任务需求，而任务生成则指导针对性的场景编辑，使得场景和任务能够相互迭代改进。这种联合精炼提高了任务生成的成功率，包括长时间跨度的任务。该框架进一步支持移动操控器、人形机器人和灵巧手，以及涉及可变形物体和流体的交互，拓宽了生成数据中表现的行为和物理现象的范围。这些能力共同提供了一个灵活的仿真引擎，适用于机器人预训练和评估。大量实验验证了生成数据的质量、多样性和生成效率，而下游策略实验表明，数据多样性的增加改善了泛化能力。
cs.CV / 73 / 2610.07982
M3SunAgent: Monocular 3D Spatial Understanding Agent for Metric Depth Estimation and 3D Visual Grounding
M3SunAgent：用于度量深度估计和3D视觉定位的单目3D空间理解智能体
Abstract
Monocular metric depth estimation and 3D visual grounding represent the two complementary cornerstones of monocular 3D spatial understanding (M3Sun), from which the fundamental 3D spatial information required by M3Sun can be acquired. However, these complementary tasks are generally conducted by separate frameworks, which pose challenges of inflexible and unaligned spatial information access for embodied intelligence systems. In this paper, we propose a unified agent for monocular 3D spatial understanding (M3SunAgent) that leverages a large language model (LLM) as a task planner for spatial visual programming, which flexibly generate structured programs and coordinate tools. For instance-level metric depth estimation task, M3SunAgent invokes an object detector tool to locate the target, estimates depth at selected points with a depth estimation tool, and aggregates these predictions into an instance-level depth estimate. We also construct the M3Sun Instance (M3SI) dataset, a benchmark with 2,910 samples for evaluation. For monocular 3D visual grounding task, M3SunAgent uses a vision-language model (VLM) tool to locate the target and output basic spatial attributes, then combines back-projection tool with a dimension-lifting tool to predict its 3D bounding box. Experimental results demonstrate the superior performance of M3SunAgent. Specifically, in evaluations of instance-level monocular metric depth estimation, M3SunAgent achieves the best performance among all compared models, 52.61% of predicted instances are distributed below depth error 0.25 ($\delta < 0.25$). In evaluations of monocular 3D visual grounding, M3SunAgent demonstrates overall competitive performance than vision and VLM models, reaching a 3D mean intersection over union (mIoU) of 41.73% and exceeding the state-of-the-art MonoVLM model by 3.62%.
Chinese Translation
单目度量深度估计和3D视觉定位代表了单目3D空间理解（M3Sun）的两个互补基石，从中可以获取M3Sun所需的基本3D空间信息。然而，这些互补任务通常由独立的框架进行，这给具身智能系统带来了空间信息访问不灵活和不对齐的挑战。本文提出了一种用于单目3D空间理解的统一智能体（M3SunAgent），它利用大型语言模型（LLM）作为空间视觉编程的任务规划器，灵活生成结构化程序并协调工具。在实例级度量深度估计任务中，M3SunAgent调用对象检测工具来定位目标，使用深度估计工具在选定点估计深度，并将这些预测聚合为实例级深度估计。我们还构建了M3Sun实例（M3SI）数据集，这是一个包含2910个样本的基准数据集用于评估。在单目3D视觉定位任务中，M3SunAgent使用视觉-语言模型（VLM）工具定位目标并输出基本空间属性，然后结合反投影工具和维度提升工具预测其3D边界框。实验结果表明M3SunAgent的优越性能。具体而言，在实例级单目度量深度估计的评估中，M3SunAgent在所有对比模型中表现最佳，52.61%的预测实例的深度误差低于0.25（$ ext{δ} < 0.25$）。在单目3D视觉定位的评估中，M3SunAgent的整体表现优于视觉和VLM模型，达到41.73%的3D平均交并比（mIoU），并超过了最先进的MonoVLM模型3.62%。
cs.CV / 74 / 2610.07984
Decide Before You Look: Learning Which Retrieved Memories Deserve Pixels
在查看之前决定：学习哪些检索到的记忆值得使用像素
Abstract
Multimodal assistants answer questions from long-term memories that contain images. After retrieval, each retrieved image reaches the answering model either as pixels, at about a thousand visual tokens per image, or as a stored text proxy that often misses the detail the question asks about. We find that the benefit of pixels usually comes from one or two retrieved memories, and that it can be predicted before the answering model runs, without reading any full-resolution image. In PixelTriage, a plug-in placed after retrieval, a small model that does not generate text reads the dialogue, a short note and a thumbnail of each retrieved memory and predicts how much its pixels would add. It is trained on synthetic memory episodes labeled by a frozen 27B model that answers each question with and without each memory's pixels. With a 7B answering model, PixelTriage lies on the accuracy--cost frontier of M$^3$Exam, DMV and MemEye and uses 11--23\% of the visual tokens without a significant loss of accuracy. On DMV it answers 2.9 times faster than opening all images. It outperforms retrieval order and uniform down-sizing at equal budgets and transfers to other memory systems and to a 397B answering model.
Chinese Translation
多模态助手从包含图像的长期记忆中回答问题。在检索后，每个检索到的图像以像素形式到达回答模型，每个图像大约有一千个视觉标记，或者以存储的文本代理形式到达，这种形式往往无法捕捉问题所询问的细节。我们发现，像素的好处通常来自一到两个检索到的记忆，并且可以在回答模型运行之前进行预测，而无需读取任何全分辨率图像。在 PixelTriage 中，一个在检索后放置的插件，一个小型模型不生成文本，而是读取对话、简短的备注和每个检索到的记忆的缩略图，并预测其像素将增加多少。它在由一个冻结的 27B 模型标记的合成记忆事件上进行训练，该模型在有和没有每个记忆的像素的情况下回答每个问题。使用 7B 的回答模型，PixelTriage 位于 M$^3$Exam、DMV 和 MemEye 的准确性-成本边界上，并在没有显著准确性损失的情况下使用 11% - 23% 的视觉标记。在 DMV 上，它的回答速度是打开所有图像的 2.9 倍。在相同预算下，它优于检索顺序和均匀缩小，并且可以迁移到其他记忆系统和一个 397B 的回答模型。
cs.CV / 75 / 2610.07987
VisionWeave: Weaving Elastic Visual Representations as a Native Capability of MLLMs
VisionWeave：将弹性视觉表示作为多模态大语言模型的原生能力
Abstract
Multimodal large language models have become the dominant paradigm for visual understanding, but incur substantial costs by encoding inputs into dense, fixed-size patch tokens. However, visual information is unevenly distributed: some regions require fine-grained detail, while others admit compact representations. Downsampling sacrifices this detail, while existing token pruning and adaptive approaches remain limited in content-adaptive granularity, task generalization, and integration with modern MLLMs and serving infrastructure. Overcoming these limitations calls for foundation models that learn, end to end, where-and at what granularity-to allocate visual representations, a native capability we term elastic visual representation weaving. We introduce VisionWeave, establishing this capability in frontier-level MLLMs through large-scale training. It combines two components: a gated spatial pooler constructs coarse-grained representations alongside native fine-grained representations within a shared MRoPE coordinate, while a granularity router learns their content-adaptive allocation. Through self-distillation alone, we validate this capability on Qwen3.5-4B and scale to Qwen3.8-27B with over 30K A100 GPU-hours. Based on Qwen3.8-27B, VisionWeave adaptively adjusts token savings to visual content, saving 43.0% tokens on average while retaining 98.9% native performance across eight benchmarks, versus only 88% performance preserved for token pruning baselines with a fixed 50% savings target. Extensive evaluations confirm robust efficiency-quality trade-offs across diverse tasks, resolutions and video frames. When deployed on SGLang serving engine, our method achieves a 2.3x throughput gain while reducing mean TTFT by 54.4% and mean TPOT by 60.6%. Together, we believe these results position elastic visual weaving as a promising capability for next-generation multimodal models.
Chinese Translation
多模态大语言模型已成为视觉理解的主导范式，但通过将输入编码为密集的固定大小补丁令牌而产生了巨大的成本。然而，视觉信息的分布是不均匀的：某些区域需要细粒度的细节，而其他区域则可以接受紧凑的表示。下采样牺牲了这些细节，而现有的令牌修剪和自适应方法在内容自适应粒度、任务泛化以及与现代多模态大语言模型和服务基础设施的集成方面仍然有限。克服这些局限性需要基础模型能够端到端地学习在何处以及以何种粒度分配视觉表示，这一原生能力我们称之为弹性视觉表示编织。我们介绍了VisionWeave，通过大规模训练在前沿级别的多模态大语言模型中建立这一能力。它结合了两个组件：一个门控空间池化器在共享的MRoPE坐标中构建粗粒度表示，同时生成原生的细粒度表示，而粒度路由器则学习其内容自适应分配。仅通过自蒸馏，我们在Qwen3.5-4B上验证了这一能力，并扩展到Qwen3.8-27B，耗时超过30K A100 GPU小时。基于Qwen3.8-27B，VisionWeave自适应地根据视觉内容调整令牌节省，平均节省43.0%的令牌，同时在八个基准测试中保持98.9%的原生性能，而固定50%节省目标的令牌修剪基线仅保留88%的性能。广泛的评估确认了在多样化任务、分辨率和视频帧之间的稳健效率-质量权衡。当在SGLang服务引擎上部署时，我们的方法实现了2.3倍的吞吐量提升，同时将平均TTFT减少了54.4%，平均TPOT减少了60.6%。我们相信，这些结果将弹性视觉编织定位为下一代多模态模型的有前景的能力。
cs.CV / 76 / 2610.08068
PhysTacGen: Physics-Aware Visual-Tactile Sensor Image Generation
PhysTacGen：物理感知视觉-触觉传感器图像生成
Abstract
Realistic physical interaction is a cornerstone of embodied intelligence, yet collecting paired visual--tactile data remains costly. Visual-to-tactile synthesis offers a promising approach to augmenting such data, but learning this mapping is complicated by the gap between visual appearance and contact-related material properties, as well as spatial misalignment in paired observations. To address these challenges, we present \textbf{PhysTacGen}, a visual-to-optical-tactile image generation framework that integrates material-aware descriptions with geometric conditioning. First, we introduce Group Tactile Policy Optimization (GTPO), a reinforcement learning strategy that refines a vision--language model to generate structured material descriptions using task-specific rewards. Second, we combine DINOv2-based pair curation with monocular relative-depth estimation to select training pairs and provide geometric priors. Finally, an SDXL ControlNet synthesizes optical tactile images conditioned on RGB, relative depth, and GTPO-generated text. Experiments on curated SSVTP data demonstrate improved structural similarity over the compared baselines, while a blinded user study shows a preference for GTPO-generated descriptions. Generated tactile inputs also improve performance on an attribute-derived force-coefficient prediction proxy. Together, these results demonstrate the effectiveness of PhysTacGen for optical tactile image synthesis and its utility in the evaluated downstream task.The code will be available at https://github.com/VDIGPKU/PhysTacGen.
Chinese Translation
现实的物理交互是具身智能的基石，但收集配对的视觉-触觉数据仍然成本高昂。视觉到触觉的合成提供了一种增强此类数据的有前景的方法，但学习这种映射受到视觉外观与接触相关材料属性之间的差距，以及配对观测中的空间不对齐的复杂影响。为了解决这些挑战，我们提出了 extbf{PhysTacGen}，一种视觉到光学-触觉图像生成框架，结合了材料感知描述与几何条件。首先，我们引入了群体触觉策略优化（Group Tactile Policy Optimization, GTPO），这是一种强化学习策略，通过任务特定的奖励来优化视觉-语言模型，以生成结构化的材料描述。其次，我们结合基于DINOv2的配对策划与单目相对深度估计，选择训练对并提供几何先验。最后，SDXL ControlNet在RGB、相对深度和GTPO生成的文本条件下合成光学触觉图像。在策划的SSVTP数据上的实验表明，相较于对照基线，结构相似性有所改善，而盲测用户研究显示对GTPO生成描述的偏好。生成的触觉输入也提高了基于属性的力系数预测代理的性能。综合来看，这些结果展示了PhysTacGen在光学触觉图像合成中的有效性及其在评估的下游任务中的实用性。代码将发布在 https://github.com/VDIGPKU/PhysTacGen。
cs.CV / 77 / 2610.08070
Two Halves are More than One: Phase-wise Velocity Distillation for Fast and High-Quality Image Generation
两个部分胜过一个：基于阶段的速度蒸馏用于快速高质量图像生成
Abstract
Recent diffusion-based image generation backbones have grown substantially in scale, making the network inference cost increase rapidly. While diffusion distillation techniques can reduce the number of inference steps, high-quality image generation within a single full-backbone-forward compute budget remains challenging. Existing one-step methods typically allocate this budget to a single evaluation of a monolithic student. However, approximating the heterogeneous coarse-to-fine transport with a single monolithic mapping is difficult and often leads to over-smoothed outputs. To address this issue, we propose Phase-wise Velocity Distillation (PVD), which partitions the generation timeline into a coarse and a fine phase, and models the transition within each phase via the average velocity. A dedicated half-sized expert is assigned to each phase, decoupling structural composition from detail refinement while keeping the cumulative computation equivalent to one full-backbone forward pass. We show that the use of two half-sized phase-specific experts outperforms a single full-size monolithic student. On class-conditional image generation, PVD achieves an FID of 1.48 on ImageNet 256 x 256. On more complex text-to-image (T2I) tasks, PVD-distilled models (Stable Diffusion 3.5-Medium, FLUX.1-dev, Qwen-Image) produce results competitive with their multi-step teachers, significantly outperforming prior distillation methods. Moreover, across the evaluated T2I backbones, PVD reduces active parameters by 49.10-50.89% and peak VRAM by 45.76-48.36% compared to the corresponding teachers. Source code and distilled models are available at https://github.com/PolyU-VCLab/PVD.
Chinese Translation
近期，基于扩散的图像生成骨干网络在规模上大幅增长，导致网络推理成本迅速增加。尽管扩散蒸馏技术可以减少推理步骤的数量，但在单次完整骨干前向计算预算内实现高质量图像生成仍然具有挑战性。现有的一步法通常将这一预算分配给单一的整体学生模型。然而，使用单一整体映射来近似异质的粗到细传输是困难的，且往往导致输出过于平滑。为了解决这一问题，我们提出了基于阶段的速度蒸馏（Phase-wise Velocity Distillation, PVD），该方法将生成时间线划分为粗阶段和细阶段，并通过平均速度建模每个阶段内的过渡。每个阶段分配一个专用的半大小专家，解耦结构组成与细节精炼，同时保持累积计算等同于一次完整的骨干前向传递。我们展示了使用两个半大小的阶段特定专家的效果优于单一的全尺寸整体学生。在类别条件图像生成任务中，PVD在ImageNet 256 x 256上达到了1.48的FID。在更复杂的文本到图像（T2I）任务中，PVD蒸馏模型（Stable Diffusion 3.5-Medium, FLUX.1-dev, Qwen-Image）生成的结果与其多步骤教师模型相竞争，显著超越了之前的蒸馏方法。此外，在评估的T2I骨干网络中，PVD相比于相应的教师模型减少了49.10-50.89%的活跃参数和45.76-48.36%的峰值显存。源代码和蒸馏模型可在https://github.com/PolyU-VCLab/PVD获取。
cs.CV / 78 / 2610.08086
Multi-Dataset Diagnostic Utility of Clinical Visual Concepts in AI Systems for Dermatology
人工智能系统在皮肤病学中的临床视觉概念的多数据集诊断效用
Abstract
The clinical integration of AI systems in digital dermatology relies heavily on human trust. Clinically interpretable visual concepts can act as intermediate representations enhancing trust and reliability. However, research in this domain is currently limited by scattered, heterogeneous dataset annotations. In this work, we introduce SkinLex, a harmonized dataset of 48 clinical morphological attributes across four public datasets (SkinCon, DermaCon-IN, MM-Skin, and PASSION) for a total of 20,411 records. Supervised nine-partition classification of skin conditions shows that limiting features to specific visual groups, like shapes or colors alone, reduces diagnostic accuracy. Bootstrapped backward elimination reveals that the set of 48 visual concepts has some degree of redundancy for algorithmic nine-partition diagnosis on the examined dataset. This demonstrates that coarse diagnosis on the selected dataset requires a relatively small but varied combination of clinical concepts, and motivates further research to improve concept taxonomy. Results can be translated into clinical benefits by reducing inputs for concept-based models, improving efficiency for annotation and modeling, and further enhancing interpretability. Code and prompt templates are available at https://github.com/Digital-Dermatology/SkinLex.
Chinese Translation
人工智能系统在数字皮肤病学中的临床整合在很大程度上依赖于人类的信任。临床可解释的视觉概念可以作为中介表示，增强信任和可靠性。然而，该领域的研究目前受到分散且异质的数据集注释的限制。在本研究中，我们引入了SkinLex，这是一个在四个公共数据集（SkinCon、DermaCon-IN、MM-Skin和PASSION）中，涵盖48种临床形态特征的统一数据集，总计20,411条记录。对皮肤病状的监督九分区分类表明，仅限于特定视觉组（如形状或颜色）的特征会降低诊断准确性。引导式向后消除法揭示，48个视觉概念的集合在所考察的数据集上的算法九分区诊断中存在一定程度的冗余。这表明，在所选数据集上进行粗略诊断需要相对较小但多样的临床概念组合，并激励进一步研究以改善概念分类。通过减少概念模型的输入、提高注释和建模的效率以及进一步增强可解释性，结果可以转化为临床益处。代码和提示模板可在 https://github.com/Digital-Dermatology/SkinLex 获取。
cs.CV / 79 / 2610.08109
Beyond Training from Scratch: Foundation Models for Data-Efficient and Generalizable Cardiac MRI Reconstruction
超越从零开始训练：用于数据高效和可泛化心脏MRI重建的基础模型
Abstract
Cardiac magnetic resonance imaging reconstruction aims to recover high-quality images from undersampled acquisitions, enabling faster scans while preserving diagnostic fidelity. Recent reconstruction methods are typically trained from scratch and often require large amounts of task-specific data, limiting their robustness under data scarcity and distribution shifts. In this work, we investigate whether pretrained vision foundation models can serve as effective priors for accelerated cardiac MRI reconstruction. We propose a reconstruction framework that integrates frozen and parameter-efficiently adapted visual encoders, including CLIP, BiomedCLIP, and DINOv2, within a transformer-based reconstruction architecture. Extensive experiments on the CMRxRecon2023 and CMRxRecon2024 benchmarks demonstrate that pretrained representations consistently outperform a transformer trained from scratch across multiple acceleration factors. We further evaluate performance under limited supervision and cross-dataset transfer, showing that foundation models provide superior data efficiency and generalization. While frozen representations are particularly effective in extreme low-data regimes, Low-Rank Adaptation (LoRA) yields additional gains when moderate amounts of training data are available. Among the evaluated backbones, DINOv2 achieves the strongest overall performance. These findings highlight the potential of vision foundation models as robust and transferable priors for cardiac MRI reconstruction.
Chinese Translation
心脏磁共振成像重建旨在从欠采样的采集数据中恢复高质量图像，从而实现更快的扫描速度，同时保持诊断的准确性。近期的重建方法通常从零开始训练，且往往需要大量特定任务的数据，这限制了它们在数据稀缺和分布变化下的鲁棒性。在本研究中，我们探讨了预训练视觉基础模型是否可以作为加速心脏MRI重建的有效先验。我们提出了一种重建框架，该框架集成了冻结和参数高效适配的视觉编码器，包括CLIP、BiomedCLIP和DINOv2，构建在基于变换器的重建架构内。在CMRxRecon2023和CMRxRecon2024基准上的大量实验表明，预训练表示在多个加速因子下始终优于从零开始训练的变换器。我们进一步评估了在有限监督和跨数据集迁移下的性能，显示基础模型提供了优越的数据效率和泛化能力。虽然冻结表示在极低数据情况下特别有效，但当有适量训练数据可用时，低秩适配（Low-Rank Adaptation, LoRA）可带来额外的收益。在评估的骨干网络中，DINOv2实现了最强的整体性能。这些发现突显了视觉基础模型作为心脏MRI重建的鲁棒和可转移先验的潜力。
cs.CV / 80 / 2610.08126
Supermarket Product Detection and Recognition: Utilizing Deep Learning with Rectified Imagery
超市产品检测与识别：利用深度学习与校正图像
Abstract
Product Identification has sprung up to become one of the most challenging problems in the automation of the retail industry. With the new industry 5.0 standards, automated inventory management, and catalog creation tasks are vitally important. Object identification models have emerged as a viable answer with their unprecedented identification and localization accuracy. However, the close-knit rack design of supermarkets generates the problem of angle variation in capturing images. The angle-variant densely packed images(a single image contains many objects) become overwhelming for these models alone. In this paper, we try to supplement object detection models with traditional Hough transform (HT) and homogeneous estimation concepts. We study the effect of rectified images using homography estimation and hough transform and their limitations on the problem of grocery identification. We make a case for creating a new dataset to test the effects of such rectification and produce analytical results on different scenarios of angle variation and object densities per image. Extensive experiments on different object detection models suggest that image rectification of angled images improves the detection accuracy of grocery products in images. The results also highlight the limitation of rectification on the angle of image capture and the object density of the image.
Chinese Translation
产品识别已成为零售行业自动化中最具挑战性的问题之一。随着新兴的工业5.0标准，自动化库存管理和目录创建任务变得至关重要。物体识别模型以其前所未有的识别和定位精度成为一种可行的解决方案。然而，超市紧凑的货架设计导致在捕捉图像时出现角度变化的问题。角度变化密集包装的图像（单个图像包含多个物体）对这些模型来说变得难以处理。在本文中，我们尝试用传统的霍夫变换（Hough Transform, HT）和同质估计概念来补充物体检测模型。我们研究了使用单应性估计和霍夫变换校正图像的效果及其在杂货识别问题上的局限性。我们主张创建一个新的数据集，以测试这种校正的效果，并在不同的角度变化和每幅图像的物体密度场景下产生分析结果。在不同物体检测模型上的广泛实验表明，校正角度图像可以提高图像中杂货产品的检测准确性。结果还突显了校正对图像捕捉角度和图像物体密度的局限性。
cs.CV / 81 / 2610.08162
The Failure Is in the Readout: Fine-Grained Emotion Recognition Benchmarks Measure Elicitation, Not Perception
失败在于读取：细粒度情感识别基准测量引发，而非感知
Abstract
Fine-grained emotion recognition supports therapy tools and social robots, but it needs facial data, which raises privacy and data-protection concerns. EmoNet-Face-HQ answers that with generated portraits, expert-rated over a $40$-category taxonomy far finer than the usual six to eight basic emotions. Under the protocol it ships with, vision-language models (VLMs) score poorly on that taxonomy, and the benchmark concludes that a dedicated fine-tuned model is necessary: Empathic-Insight-Face (EIF; Small/Large). We show that off-the-shelf VLMs match or beat that fine-tuned model when the answer is not generated but read from the logits, as one binary query per category. We keep the benchmark's images, taxonomy and ratings, and change only how the answer is read. Experts agree at $\kappa_w = 0.468$ on the five categories they measure most reliably. Generatively, no interval among eleven open-weight VLMs lies entirely above that anchor ($\kappa_w=0.268$-$0.486$). Under verification all eleven clear it, each of them significantly better at $\kappa_w=0.507$-$0.586$. Three also significantly beat EIF sitting at $\kappa_w = 0.551$ (Small; $0.534$ Large). The gain comes from the graded probability and not from asking a yes/no question: as a control, thresholding those same probabilities to yes/no costs 142% of the average gains and drops binarization below generative elicitation to $\kappa_w=0.254$-$0.423$. A replication on real photographs (FACES) is weaker and mixed: of the ten models that pass a validity gate, six gain, three are neutral to positive and one is negative, so the effect is not confined to synthetic data.
Chinese Translation
细粒度情感识别支持治疗工具和社交机器人，但它需要面部数据，这引发了隐私和数据保护的担忧。EmoNet-Face-HQ 通过生成的肖像来解决这一问题，这些肖像在一个超过 $40$ 类别的分类法中经过专家评估，远比通常的六到八种基本情感更为细致。在其附带的协议下，视觉语言模型（VLMs）在该分类法上的得分较低，基准得出结论认为需要一个专门的微调模型：Empathic-Insight-Face (EIF; Small/Large)。我们展示了当答案不是生成的而是从 logits 中读取时，现成的 VLMs 可以与该微调模型相匹配或超越，作为每个类别的一个二元查询。我们保留基准的图像、分类法和评分，仅改变答案的读取方式。专家在他们最可靠测量的五个类别上达成一致，$ ext{kappa}_w = 0.468$。在十一种开放权重的 VLMs 中，没有一个区间完全高于该基准（$ ext{kappa}_w=0.268$-$0.486$）。在验证下，所有十一种模型均超过该基准，每种模型的表现显著提高，$ ext{kappa}_w=0.507$-$0.586$。其中三种模型的表现显著优于 EIF，后者的得分为 $ ext{kappa}_w = 0.551$（Small; $0.534$ Large）。这种提升来自于分级概率，而非简单的是/否问题：作为对照，将这些相同的概率阈值化为是/否问题，平均收益的成本为142%，并将二元化的得分降至 $ ext{kappa}_w=0.254$-$0.423$。在真实照片（FACES）上的复制实验结果较弱且混合：在通过有效性门的十个模型中，六个模型表现出提升，三个模型表现中性到积极，一个模型表现消极，因此这一效应并不局限于合成数据。
cs.CV / 82 / 2610.08179
View Matters: Keyframe-Guided Text-Driven 3D Gaussian Editing
视角重要性：关键帧引导的文本驱动3D高斯编辑
Abstract
Text-driven 3D Gaussian editing commonly does not distinguish the editing reliability of rendered views, although different viewpoints provide supervision of substantially different quality. Views that clearly show the scene and match the edit instruction provide reliable guidance, while less informative views may weaken the edit when all views are treated equally. We present View Matters, a view-importance-aware framework that conducts editing around reliable keyframes. Keyframe Importance Estimation (KIE) identifies reliable views using geometric visibility, semantic distinctiveness, and edit relevance. Keyframe-Guided Editing (KGE) then propagates their editing signals asymmetrically to non-keyframes without noisy reverse influence, while Importance-Aware Optimization (IAO) preserves this reliability preference during 3DGS optimization. Across 23 scene-prompt pairs, View Matters achieves the highest average CLIP text-image similarity of 0.2822 and directional similarity of 0.2564 among the evaluated methods, with a four-minute editing time. Additional adjacent-view analysis indicates that the fidelity-oriented editing process maintains cross-view coherence.
Chinese Translation
文本驱动的3D高斯编辑通常不区分渲染视角的编辑可靠性，尽管不同的视点提供的监督质量差异显著。能够清晰展示场景并与编辑指令相匹配的视角提供了可靠的指导，而信息量较少的视角在所有视角被平等对待时可能削弱编辑效果。我们提出了视角重要性框架（View Matters），该框架围绕可靠的关键帧进行编辑。关键帧重要性估计（Keyframe Importance Estimation, KIE）利用几何可见性、语义独特性和编辑相关性来识别可靠视角。关键帧引导编辑（Keyframe-Guided Editing, KGE）随后将其编辑信号不对称地传播到非关键帧，而不受噪声反向影响，同时重要性感知优化（Importance-Aware Optimization, IAO）在3D高斯优化过程中保持这种可靠性偏好。在23个场景-提示对中，视角重要性框架在评估方法中实现了最高的平均CLIP文本-图像相似度0.2822和方向相似度0.2564，编辑时间为四分钟。额外的邻近视角分析表明，忠实度导向的编辑过程保持了跨视角的一致性。
cs.CV / 83 / 2610.08188
PIE-PS: Photometric Stereo from Physical Irradiance Event Streams
PIE-PS：基于物理辐照事件流的光度立体重建
Abstract
Event cameras record asynchronous log-image-irradiance changes with microsecond latency and high dynamic range. These properties are useful for photometric stereo under moving illumination, but raw events are sparse and depend on an unknown contrast threshold. We start from the event trigger model and derive a physical relation between adjacent events, light motion, and surface normals. This relation gives a direct physics-only solver, but the solver needs the threshold, enough events at each pixel, and independent per-pixel optimization. To address these limits, we introduce PIE-PS, a learning-based framework for dense surface normal reconstruction from raw event streams and known lighting. We form Physical Irradiance Events (PIEs) by pairing two adjacent events at the same pixel with their corresponding light directions. Each PIE provides a Physical Irradiance Event Feature (PIEF), defined as the signed event rate. PIEF does not require the unknown contrast threshold. To share spatial and temporal context across nearby PIEs, we introduce PIE-GNN, which treats each PIE as a graph node and encodes it with its light-pair geometry. Since the reliability of PIE observations can vary with local appearance, illumination geometry, and sensor noise, Reliability-Grading Attention (RGA) predicts reliability weights to down-weight unreliable PIEs. Pixel aggregation then produces dense normals. Experiments on synthetic and real data show that PIE-PS outperforms prior event-based photometric stereo methods and the direct solver baseline.
Chinese Translation
事件相机以微秒延迟和高动态范围记录异步的图像辐照变化。这些特性对于在移动照明下进行光度立体重建非常有用，但原始事件稀疏且依赖于未知的对比度阈值。我们从事件触发模型出发，推导出相邻事件、光源运动和表面法线之间的物理关系。该关系提供了一种仅基于物理的直接求解器，但该求解器需要阈值、每个像素足够的事件以及独立的逐像素优化。为了解决这些限制，我们引入了PIE-PS，一个基于学习的框架，用于从原始事件流和已知照明中重建稠密的表面法线。我们通过将同一像素的两个相邻事件与其对应的光源方向配对，形成物理辐照事件（Physical Irradiance Events，PIEs）。每个PIE提供一个物理辐照事件特征（Physical Irradiance Event Feature，PIEF），定义为带符号的事件率。PIEF不需要未知的对比度阈值。为了在附近的PIE之间共享空间和时间上下文，我们引入了PIE-GNN，将每个PIE视为图节点，并用其光源配对几何信息进行编码。由于PIE观测的可靠性可能因局部外观、照明几何和传感器噪声而异，可靠性评分注意机制（Reliability-Grading Attention，RGA）预测可靠性权重，以降低不可靠PIE的权重。然后，像素聚合生成稠密法线。对合成数据和真实数据的实验表明，PIE-PS的性能优于先前的基于事件的光度立体重建方法和直接求解基线。
cs.CV / 84 / 2610.08192
MacJEPA: Missingness-Robust Audio-Visual Recognition from Untrimmed Egocentric Videos
MacJEPA：来自未裁剪自我中心视频的缺失鲁棒音视频识别
Abstract
Audio-visual models improve egocentric action recognition by exploiting complementary cues, yet typically assume that both streams remain available at inference. Existing missing-modality methods operate on trimmed, single-event clips in which a stream is entirely present or absent, whereas real sensors fail and recover within long, untrimmed observations. We redefine egocentric modality missingness as temporally localized sensor outages within untrimmed, multi-event observations, with whole-clip absence as the limiting case. We introduce \textbf{MacJEPA}, a missing-modality-robust \textbf{Ma}sked-\textbf{c}ontext query \textbf{JEPA} that recognizes visual actions and acoustic events from supplied interval queries over audio-visual context. Window-local modality dropout simulates these sensor outages during training. MacJEPA further repurposes masking in JEPA from a self-supervised pretext into a supervised robustness objective, aligning masked and clean latent representations of both multimodal content tokens and the task-conditioned queries. All objectives are optimized jointly with recognition in a single stage, requiring no test-time adaptation. Across Epic-Kitchens-100 and Epic-Sounds, a single checkpoint remains competitive under complete input and consistently surpasses published missing-modality baselines when either the dominant or auxiliary stream is removed. MacJEPA thus unifies strong full-input recognition with temporal missing-modality robustness in a single model operating on untrimmed multi-event videos.
Chinese Translation
音视频模型通过利用互补线索来改善自我中心动作识别，但通常假设在推理时两个流都可用。现有的缺失模态方法在裁剪的单事件片段上操作，其中一个流要么完全存在要么完全缺失，而现实中的传感器在长时间未裁剪的观察中会发生故障并恢复。我们将自我中心模态缺失重新定义为在未裁剪的多事件观察中时间局部化的传感器故障，整个片段缺失作为极限情况。我们引入了 extbf{MacJEPA}，一种缺失模态鲁棒的 extbf{Ma}sked- extbf{c}ontext 查询 extbf{JEPA}，能够从提供的音视频上下文的区间查询中识别视觉动作和声学事件。窗口局部模态丢失在训练期间模拟这些传感器故障。MacJEPA 进一步将 JEPA 中的掩蔽从自监督的前置任务转变为监督的鲁棒性目标，协调多模态内容标记和任务条件查询的掩蔽与干净潜在表示。所有目标在单个阶段与识别共同优化，无需测试时适应。在 Epic-Kitchens-100 和 Epic-Sounds 数据集上，单个检查点在完整输入下保持竞争力，并在去除主流或辅助流时始终超越已发布的缺失模态基线。因此，MacJEPA 在单一模型中统一了强大的全输入识别与时间缺失模态鲁棒性，适用于未裁剪的多事件视频。
cs.CV / 85 / 2610.08213
RACE-FPP: A Robust AI-assisted Characterisation Enhancement for Fringe Projection Profilometry
RACE-FPP：一种稳健的AI辅助条纹投影轮廓测量特征增强方法
Abstract
Fringe Projection Profilometry (FPP) requires precise system characterisation to achieve reliable three-dimensional (3D) reconstructions; however, characterisation accuracy strongly depends on robust checkerboard feature localisation, which can deteriorate under challenging imaging conditions such as lens blur and characterisation target orientations. Existing deep learning-based corner detectors are typically assessed using detection metrics and camera reprojection error alone, without considering their wider impact on projector characterisation, camera-projector stereo characterisation consistency, or overall measurement accuracy. In this work, we introduce a complete FPP characterisation pipeline that incorporates deep learning-based corner detection into the standard camera characterisation workflow. We also characterise the projector by sampling phase values at the centres of the white squares in the characterisation target. Rather than treating corner detection as an isolated task, the proposed framework explicitly analyses how localisation errors propagate throughout the entire FPP characterisation chain. Performance is evaluated using detection metrics (e.g., precision and recall), camera and projector reprojection errors, and the camera and projector stereo characterisation. Across a mixed dataset of clean and degraded images, the camera reprojection error is reduced from 1.237 pixels to 0.259 pixels, while the projector reprojection error is reduced by roughly 50%. Dimensional evaluation of reconstructed artefacts shows improved geometric accuracy compared with those resulting from the conventional pipeline. Overall, the findings indicate increased robustness of system-level characterisation under challenging imaging conditions, thereby enabling more reliable industrial FPP measurements.
Chinese Translation
条纹投影轮廓测量（FPP）需要精确的系统特征化，以实现可靠的三维（3D）重建；然而，特征化的准确性在很大程度上依赖于稳健的棋盘格特征定位，而在诸如镜头模糊和特征化目标方向等挑战性成像条件下，这种定位可能会恶化。现有的基于深度学习的角点检测器通常仅使用检测指标和相机重投影误差进行评估，而未考虑其对投影仪特征化、相机-投影仪立体特征化一致性或整体测量准确性的更广泛影响。在本研究中，我们提出了一种完整的FPP特征化流程，将基于深度学习的角点检测纳入标准相机特征化工作流程中。我们还通过在特征化目标的白色方块中心采样相位值来对投影仪进行特征化。我们并不将角点检测视为一个孤立的任务，而是明确分析定位误差如何在整个FPP特征化链中传播。通过检测指标（例如，精确度和召回率）、相机和投影仪重投影误差以及相机和投影仪立体特征化来评估性能。在一个包含干净和退化图像的混合数据集中，相机重投影误差从1.237像素降低到0.259像素，而投影仪重投影误差大约降低了50%。对重建物体的维度评估显示，与传统流程生成的结果相比，几何精度有所提高。总体而言，研究结果表明，在挑战性成像条件下，系统级特征化的稳健性得到了增强，从而实现了更可靠的工业FPP测量。
cs.CV / 86 / 2610.08230
TSRN-RTVD: Real-Time Video Deblurring System
TSRN-RTVD：实时视频去模糊系统
Abstract
As video capture moves to handheld and edge devices, motion blur from camera shake has become a pervasive degradation that lowers perceptual quality and harms downstream vision tasks. The strongest deblurring networks recover impressive detail, yet they remain computationally heavy and overwhelmingly complex, so their quality comes at a cost that consumer hardware cannot pay in real time. This gap between restoration quality and on-device speed is exactly what makes real-time deblurring difficult. We developed and implemented TSRN-RTVD, an efficient video deblurring system that explicitly reconstructs the underlying camera trajectory during exposure and uses the recovered motion to guide restoration. This approach turns the physical cause of blur into a signal that drives sharpening. Our system runs on a single consumer GPU and restores the video at 30 FPS while reaching 30.08 dB PSNR on the GoPro dataset. We demonstrate TSRN-RTVD on consumer devices with interactive side-by-side visualization of the blurry input and the deblurred output, live throughput, and an on-screen view of the recovered camera trajectory. Demo video is available at https://youtu.be/3alMwVrVALU.
Chinese Translation
随着视频捕捉技术向手持设备和边缘设备转移，因相机抖动导致的运动模糊已成为一种普遍的退化现象，降低了感知质量并对后续视觉任务造成了损害。最强大的去模糊网络能够恢复令人印象深刻的细节，但它们仍然计算量大且复杂，因此其质量的提升是消费者硬件在实时处理中无法承受的代价。恢复质量与设备上速度之间的差距正是实时去模糊面临的挑战。我们开发并实现了TSRN-RTVD，一个高效的视频去模糊系统，该系统在曝光期间明确重建底层相机轨迹，并利用恢复的运动信息指导恢复过程。这种方法将模糊的物理原因转化为驱动锐化的信号。我们的系统在单个消费级GPU上运行，以30帧每秒的速度恢复视频，并在GoPro数据集上达到30.08 dB的峰值信噪比（PSNR）。我们在消费设备上演示了TSRN-RTVD，提供模糊输入与去模糊输出的交互式并排可视化、实时吞吐量以及恢复的相机轨迹的屏幕显示。演示视频可在 https://youtu.be/3alMwVrVALU 获取。
cs.CV / 87 / 2610.08279
Whose Face Is It Anyway? A Multi-Model Audit of Facial Affect Recognition on Children, and Why the Gap Is the Head, Not the Features
这是谁的脸？儿童面部情感识别的多模型审计，以及为何差距在于头部，而非特征
Abstract
Facial affect models are trained almost entirely on adults, yet are increasingly applied to children in education, health, and developmental research. We present a controlled, multi-model audit of five AffectNet-pretrained expression models (EmoNet, EmotiEffLib, DDAMFN++, OpenFace 3.0, LibreFace) on children, across four child image datasets, the AffectNet-8 validation set, and two spontaneous child video datasets, through one shared harness. Three findings emerge. First, the child gap is model-agnostic: every architecture degrades from posed to naturalistic faces and shares the fear$\rightarrow$surprise confusion. Second, it is concentrated and corroborated across all five models: open-mouth faces (read as surprise, correlating with the AU26 jaw drop) and South-Asian children degrade systematically, with a smaller averted-gaze penalty, while closed-mouth faces, White and Black children, and direct gaze do not; the bias tracks expression morphology and specific populations, not skin tone. Third, the gap is diagnosable: a linear probe on frozen features reaches 0.75-0.91 on unseen children versus 0.48-0.66 zero-shot, so it lies largely in the classifier head, not the representation, whereas dimensional valence/arousal regression degrades sharply under domain shift. Building on this, recalibrating only the head on a little target data recovers $+0.13$ to $+0.28$ on the two largest child sets across all five models at negligible adult cost, though the gain is in-distribution and does not transfer across child collections. We will release the harness, per-sample predictions, and analysis code; the child face data stays license-locked and is never redistributed.
Chinese Translation
面部情感模型几乎完全基于成年人进行训练，但在教育、健康和发展研究中越来越多地应用于儿童。我们对五个经过AffectNet预训练的表情模型（EmoNet、EmotiEffLib、DDAMFN++、OpenFace 3.0、LibreFace）在儿童图像数据集、AffectNet-8验证集以及两个自发儿童视频数据集上的表现进行了控制的多模型审计，采用一个共享的测试平台。研究得出了三个发现。首先，儿童差距与模型无关：每种架构在从摆拍到自然面孔的过程中表现下降，并且存在恐惧→惊讶的混淆。其次，这一现象在所有五个模型中都集中且得到证实：张嘴的面孔（被解读为惊讶，相关于AU26下巴下垂）和南亚儿童的表现系统性下降，且有较小的回避目光惩罚，而闭嘴的面孔、白人和黑人儿童以及直接目光则没有；这种偏差与表情形态和特定人群相关，而非肤色。第三，差距是可诊断的：在冻结特征上的线性探测器在未见儿童上达到0.75-0.91，而在零样本上仅为0.48-0.66，因此差距主要存在于分类器头部，而非表示层，而维度情感/唤醒回归在领域转移下急剧下降。在此基础上，仅在少量目标数据上重新校准头部，使所有五个模型在两个最大的儿童数据集上恢复了+0.13至+0.28的表现，且对成年人几乎没有成本，尽管增益是分布内的，并且无法在儿童集合之间转移。我们将发布测试平台、每个样本的预测结果和分析代码；儿童面部数据将保持许可锁定，且不会被重新分发。
cs.CV / 88 / 2610.08286
Event Detection in Table Tennis Videos using 2D Keypoints
基于2D关键点的乒乓球视频事件检测
Abstract
This paper addresses the challenge of automatic, frame-accurate event detection in table tennis videos. Current methods for estimating 3d ball trajectories and ball spin typically require that key events, such as ball-racket contacts, have already been identified in advance. This requirement makes it difficult to apply these methods to longer, unedited video recordings. To overcome this limitation, we propose EventNet, a two-stage pipeline to detect key events: (1) 2d keypoints are extracted of the upper-body poses for both players, table corners and ball center. A small keypoint transformer combines them into a compact representation that is robust to changes in viewpoint, lighting, and background clutter. (2) The temporal sequences of these frame-based representations are processed by a transformer encoder that predicts two time-to-event values for each frame, indicating how close the current frame is to the next and previous ball-racket contact. One novelty is a new, temporal cosine-like target signal. Furthermore, we introduce viewpoint augmentation via 3D reprojection and frame-rate augmentation to improve robustness and generalization. Our extensive ablation study gives deeper insights into the importance of various architectural and training aspects. Experimental results show that the proposed approach achieves an F1 score of 91.16% and a mean frame deviation between ground truth and predicted frame of 0.42 on the Latte-MV dataset and 73.08% / 1.16 on the challenging TTHQ dataset. Overall, our work demonstrates that 2d keypoint-based temporal modeling with our EventNet architecture is a promising and practical approach for automatic event detection in table tennis videos.
Chinese Translation
本文解决了在乒乓球视频中进行自动、帧精确事件检测的挑战。目前估计3D球轨迹和球旋转的方法通常要求关键事件（如球拍与球的接触）事先被识别。这一要求使得这些方法难以应用于较长的未编辑视频录制。为克服这一限制，我们提出了EventNet，一个两阶段的管道来检测关键事件：(1) 提取两名选手的上半身姿势、球台角落和球心的2D关键点。一个小型关键点变换器将它们组合成一个对视角、光照和背景杂乱变化具有鲁棒性的紧凑表示。(2) 这些基于帧的表示的时间序列通过变换器编码器进行处理，预测每帧的两个时间到事件值，指示当前帧与下一个和上一个球拍接触的接近程度。一个新颖之处是引入了一种新的时间余弦类目标信号。此外，我们通过3D重投影引入视角增强和帧率增强，以提高鲁棒性和泛化能力。我们广泛的消融研究深入探讨了各种架构和训练方面的重要性。实验结果表明，所提方法在Latte-MV数据集上达到了91.16%的F1分数和0.42的真实值与预测帧之间的平均帧偏差，在具有挑战性的TTHQ数据集上达到了73.08%和1.16。总体而言，我们的工作表明，基于2D关键点的时间建模与我们的EventNet架构结合，是乒乓球视频自动事件检测的一个有前景且实用的方法。
cs.CV / 89 / 2610.08315
Catastrophic Forgetting in Sequential Thermal Anti-UAV Detection: The Role of Scale-Conditioned Gradient Imbalance
顺序热红外反无人机检测中的灾难性遗忘：规模条件梯度不平衡的作用
Abstract
Counter-UAV systems based on thermal infrared detection must stay accurate as operational datasets evolve, yet sequential fine-tuning causes catastrophic forgetting of prior tasks, a problem that remains insufficiently characterized in this domain. This continual-learning study measures the stability-plasticity trade-off in YOLOMG, a YOLOv5-based detector run as a single thermal-infrared stream with the motion channel disabled, trained sequentially across three anti-UAV benchmarks of rising scale difficulty: Anti-UAV-RGBT, Anti-UAV410, and CST Anti-UAV. Naive fine-tuning on CST yields a Forgetting Measure of -0.605 against the Stage 1 ceiling, corresponding to a 90% capability loss, with -0.572 occurring in Stage 3 alone. In contrast, knowledge distillation from a frozen teacher is associated with FM = -0.033 +/- 0.004 across three seeds, corresponding to 95% retention. Because no Stage 2 no-KD control is included, this result establishes retention under KD training rather than a causal KD effect. Per-stratum analysis shows large-target detection collapsing to near zero within the first epoch, despite an inter-stage cosine similarity of 0.987 over the gradient-updated weights, pointing to scale-conditioned gradient imbalance, rather than weight drift, as a candidate mechanism. Scale-Stratified Herding (SSH), a 300-exemplar buffer balanced across four UAV size strata, roughly halves the forgetting (FM = -0.605 to -0.311) and keeps large-target detection non-zero. An ablation attributes the gain primarily to scale stratification rather than herding: random-stratified replay performs at least as well (FM = -0.221 versus -0.311 for SSH). These replay results are single-seed and should therefore be treated as preliminary.
Chinese Translation
基于热红外检测的反无人机系统必须在操作数据集演变时保持准确性，然而顺序微调会导致对先前任务的灾难性遗忘，这一问题在该领域仍未得到充分表征。本研究通过测量YOLOMG（基于YOLOv5的检测器）在三个逐渐增加规模难度的反无人机基准测试（Anti-UAV-RGBT、Anti-UAV410和CST Anti-UAV）中的稳定性-可塑性权衡，探讨了持续学习。对CST的简单微调导致了-0.605的遗忘度（Forgetting Measure），对应于90%的能力损失，而在第3阶段单独出现-0.572。相比之下，从冻结教师模型进行的知识蒸馏（knowledge distillation）在三个种子下的FM为-0.033 +/- 0.004，对应于95%的知识保留。由于未包含第2阶段无KD控制，该结果确立了在KD训练下的知识保留，而非因果KD效应。分层分析显示，大目标检测在第一个周期内几乎崩溃至零，尽管在梯度更新权重之间的阶段间余弦相似度为0.987，这表明规模条件梯度不平衡，而非权重漂移，可能是导致该现象的机制。规模分层集成（Scale-Stratified Herding，SSH），一个在四个无人机尺寸层次上平衡的300个样本缓冲区，约将遗忘减少一半（FM从-0.605降至-0.311），并保持大目标检测为非零。消融实验表明，增益主要归因于规模分层而非集成：随机分层重放的表现至少与SSH相当（FM为-0.221，相较于-0.311）。这些重放结果为单种子实验，因此应视为初步结果。
cs.CV / 90 / 2610.08331
Transferable Spatial Temporal Coherence Adversarial Attack on Black-Box Vision Language Models for Autonomous Driving
可转移的时空一致性对抗攻击：针对黑箱视觉语言模型的自动驾驶应用
Abstract
The rapid integration of Vision Language Models (VLMs) into sensitive systems introduces critical safety vulnerabilities that remain unexplored in exist studies. While adversarial attack robustness has been extensively studied for image-based models, the susceptibility of VLMs to temporally-aware adversarial attacks against video in driving context poses a distinct and under examined threat. In this paper, we introduce novel adversarial attack against video targeting VLM models used for autonomous driving scenes named Spatial Temporal Coherence Adversarial Attack (STCA). Our attack comprise from three stages: modalities expansion, Spatial attack, and STCA attack. In modalities expansion, we propose caption-guided frame selection method in order to ensure that adversarial perturbation target the most semantically significant frames. Secondly.In spatial attack, we craft effective perturbation and preserve high similarity. Then the perturbed video generated fed into STCA stage that disrupt cross-frame temporal coherence using motion guided mask. Our method operate under black box threat model against victim target VLMs, relying solely on transferability from white-box surrogate model.We conduct our experiments on the BDD100K and nuScenes autonomous driving datasets across three VLM models: Video LLaVA-7B, Qwen2.5-VL-7B, and Dolphin. Experimental results demonstrate spatial attack achieves an ASR with high SSIM. Our finding reveal that existing video language model, remain highly susceptible to adversarial attack in autonomous driving scenarios, underscoring the urgent need for robust defense for VLM models.
Chinese Translation
视觉语言模型（VLMs）迅速融入敏感系统，带来了在现有研究中尚未探讨的关键安全漏洞。尽管针对基于图像的模型的对抗攻击鲁棒性已被广泛研究，但在驾驶场景中，VLMs 对于针对视频的时序感知对抗攻击的易感性构成了一种独特且未被充分研究的威胁。本文提出了一种针对自动驾驶场景中 VLM 模型的视频对抗攻击，称为时空一致性对抗攻击（STCA）。我们的攻击分为三个阶段：模态扩展、空间攻击和 STCA 攻击。在模态扩展阶段，我们提出了一种基于字幕的帧选择方法，以确保对抗扰动针对最具语义重要性的帧。其次，在空间攻击阶段，我们设计了有效的扰动并保持高相似性。然后，将生成的扰动视频输入到 STCA 阶段，该阶段利用运动引导的掩码破坏跨帧的时序一致性。我们的方法在黑箱威胁模型下针对受害者目标 VLMs 进行操作，仅依赖于来自白箱替代模型的可转移性。我们在 BDD100K 和 nuScenes 自动驾驶数据集上对三种 VLM 模型进行了实验：Video LLaVA-7B、Qwen2.5-VL-7B 和 Dolphin。实验结果表明，空间攻击实现了高 SSIM 的 ASR。我们的发现揭示了现有的视频语言模型在自动驾驶场景中仍然高度易受对抗攻击的影响，强调了对 VLM 模型进行强健防御的迫切需求。
cs.CV / 91 / 2610.08339
Digital Twin-Driven Real2Sim2Real: Simulator-Conditioned Generation via Paired Driving-Scene Reconstruction
数字双胞胎驱动的Real2Sim2Real：通过配对驾驶场景重建的模拟器条件生成
Abstract
Camera-based 3D perception for autonomous driving relies heavily on large annotated datasets, and deploying such a system to a new target region typically requires data collection and annotation. Generative augmentation has been proposed to reduce this cost, but existing approaches face a fundamental trade-off: label-conditioned methods consume the very annotations they aim to replace, while simulator-conditioned methods offer free annotations but lack visual grounding to specific real environments. This work investigates the extent to which a digital-twin-driven Real2Sim2Real pipeline (DT-R2S2R) can substitute for target-region real data. By reconstructing recorded driving clips inside a georeferenced digital twin (DT-R2S), we condition a diffusion model on geometrically aligned simulator renderings, establishing a digital twin-grounded Sim2Real model (DT-S2R). As a result, DT-S2R synthesizes photorealistic driving images given low-cost yet georeferenced simulator data across both reconstructed and novel simulator scenes within digital-twin coverage. The efficacy of generated data is verified on diverse 3D detectors. DETR3D, especially, reports 93.18% of mAP obtained by a target-region real-data oracle, without employing target images for detector training. Furthermore, simple co-training with existing out-of-target real data outperforms the oracle. Thus, DT-R2S2R can substantially reduce the cost of manual on-site data collection and annotation in digital twin-available districts, providing a practical foundation for scaling 3D perception.
Chinese Translation
基于相机的自动驾驶3D感知在很大程度上依赖于大型标注数据集，而将此类系统部署到新的目标区域通常需要数据收集和标注。生成增强技术被提出以降低这一成本，但现有方法面临一个基本的权衡：标签条件的方法消耗了它们试图替代的标注，而模拟器条件的方法提供了免费的标注，但缺乏与特定真实环境的视觉基础。本文研究了数字双胞胎驱动的Real2Sim2Real管道（DT-R2S2R）在多大程度上可以替代目标区域的真实数据。通过在地理参考的数字双胞胎（DT-R2S）内部重建录制的驾驶片段，我们将扩散模型条件化于几何对齐的模拟器渲染，从而建立了一个基于数字双胞胎的Sim2Real模型（DT-S2R）。因此，DT-S2R能够在数字双胞胎覆盖范围内，基于低成本但地理参考的模拟器数据合成逼真的驾驶图像，无论是在重建的还是新颖的模拟器场景中。生成数据的有效性在多种3D检测器上得到了验证。特别是，DETR3D报告了93.18%的mAP，获得于目标区域真实数据的oracle，而未使用目标图像进行检测器训练。此外，与现有的非目标真实数据进行简单的共同训练超越了oracle。因此，DT-R2S2R可以显著降低在数字双胞胎可用地区进行手动现场数据收集和标注的成本，为扩展3D感知提供了实用基础。
cs.CV / 92 / 2610.08341
DIPrune: Task-Aware Token Pruning with Dual Importance for Efficient Multimodal Language Models
DIPrune：基于双重重要性的任务感知令牌剪枝方法，用于高效的多模态语言模型
Abstract
Recent training-free pruning approaches for Multimodal Large Language Models (MLLMs) effectively cut computational overhead by exploiting visual redundancy or text-vision attention. However, they frequently suffer from semantic degradation due to their task-agnostic design or unreliable attention estimates. Based on our empirical analysis, we have found that this issue arises because salient tokens in shallow layers persistently suppress emerging semantic ones through numerical inertia, leading to premature discarding of signals crucial for deep reasoning. To address the aforementioned issue, from the task-oriented aspects, we first reformulate training-free pruning as a minimization of the distortion in the final task loss and derive a tractable, token-wise upper bound to serve as a surrogate objective. Specifically, this formulation inherently reveals a previously neglected inter-layer term that accounts for gradients across layers. Accordingly, for the implementation, we propose DIPrune, a rank-based framework that employs a dual importance scoring mechanism to jointly optimize intra-layer static feature saliency and inter-layer dynamic semantic evolution. Extensive experiments on LLaVA and Qwen-VL demonstrate that DIPrune consistently achieves state-of-the-art results.
Chinese Translation
近期针对多模态大型语言模型（MLLMs）的无训练剪枝方法通过利用视觉冗余或文本-视觉注意力有效降低了计算开销。然而，由于其任务无关的设计或不可靠的注意力估计，这些方法常常遭遇语义退化的问题。通过我们的实证分析，我们发现这一问题的根源在于浅层中的显著令牌通过数值惯性持续抑制新兴的语义令牌，导致对深度推理至关重要的信号过早被丢弃。为了解决上述问题，从任务导向的角度出发，我们首先将无训练剪枝重新表述为最终任务损失失真最小化，并推导出一个可处理的、基于令牌的上界，以作为替代目标。具体而言，这一表述本质上揭示了一个先前被忽视的层间项，该项考虑了跨层的梯度。因此，在实现方面，我们提出了DIPrune，一个基于排名的框架，采用双重重要性评分机制，联合优化层内静态特征显著性和层间动态语义演变。在LLaVA和Qwen-VL上的大量实验表明，DIPrune始终实现了最先进的结果。
cs.CV / 93 / 2610.08346
PolarScale: A Physics-Grounded Benchmark for Radiometrically Consistent RGB-to-Stokes Estimation
PolarScale：一种基于物理的辐射一致性RGB到Stokes估计的基准
Abstract
Polarization imaging provides physical cues beyond intensity imaging but typically requires specialized hardware. Recent methods infer polarization from RGB-like inputs, yet predict only normalized Stokes components or relative descriptors, from which the radiometric scale needed for full Stokes reconstruction has been divided out. We introduce PolarScale, a benchmark that makes this scale an explicit prediction and evaluation target. Built on existing trichromatic full-Stokes measurements, PolarScale takes the per-scene normalized total-intensity image $s_0$ (a scene-referred linear image, not a consumer sRGB photograph) and asks models to predict normalized Stokes components, AoLP/DoLP/DoCP, and a per-scene scale. Because the scale is divided out of the input, it is not physically identifiable; PolarScale therefore evaluates dataset-conditioned semantic scale estimation against a constant-scale control, together with angular, self-consistency, and physical-bound metrics. Across seven restoration-based and generative backbones and three prediction strategies, the strongest restoration models estimate the scale with 3.6-4.3% mean relative error versus 5.7% for the constant control and violate physical bounds on fewer than 0.25% of pixels, whereas two generative baselines collapse to a near-zero scale; explicit descriptor supervision improves descriptor accuracy (23.66 vs. 18.88 dB PSNR for MAE). Predicted full-Stokes representations improve diffuse/specular separation, material segmentation, and glare classification, although in diffuse/specular separation the learned scale performs only on par with the constant control.
Chinese Translation
偏振成像提供了超越强度成像的物理线索，但通常需要专门的硬件。最近的方法从RGB类输入中推断偏振，但仅预测归一化的Stokes分量或相对描述符，从中已剔除全Stokes重建所需的辐射尺度。我们引入了PolarScale，一个将该尺度作为明确的预测和评估目标的基准。PolarScale基于现有的三色全Stokes测量，采用每场景的归一化总强度图像$s_0$（场景相关的线性图像，而非消费级sRGB照片），并要求模型预测归一化的Stokes分量、AoLP/DoLP/DoCP以及每场景的尺度。由于尺度已从输入中剔除，因此在物理上不可识别；因此，PolarScale评估数据集条件下的语义尺度估计与常量尺度控制的对比，以及角度、自一致性和物理约束指标。在七个基于恢复和生成的骨干网络及三种预测策略中，最强的恢复模型的尺度估计平均相对误差为3.6-4.3%，而常量控制为5.7%，且在不到0.25%的像素上违反物理约束，而两个生成基线则崩溃至近零尺度；明确的描述符监督提高了描述符的准确性（MAE的PSNR为23.66对比18.88 dB）。预测的全Stokes表示改善了漫反射/镜面反射分离、材料分割和眩光分类，尽管在漫反射/镜面反射分离中，学习到的尺度仅与常量控制相当。
cs.CV / 94 / 2610.08358
Test-Time Adaptation of Quantized ViTs via Single-Pass Quantizer-Aligned Recalibration
通过单次量化器对齐重校准实现量化视觉变换器的测试时适应
Abstract
Post-training quantization is a standard route to fitting vision transformers (ViTs) into edge compute and memory budgets, yet quantized models become especially brittle under distribution shift. Test-time adaptation (TTA) addresses such shifts without labels, but most existing approaches are poorly aligned with the constraints of quantized inference. Prevailing TTA methods recover accuracy through backpropagation, while backprop-free methods often still incur overhead from extra forward passes or parameter updates, and lightweight feature- or logit-level methods recover only part of the loss. Across these approaches, a quantization-specific failure mode that amplifies the drop is not directly targeted: under shift, activations occupy frozen quantizers' calibrated ranges differently, distorting their code distribution. We propose Quantizer-Aligned Recalibration (QuAR), a single-pass TTA method tailored to quantized ViTs that neither backpropagates nor updates any model parameters. QuAR recalibrates activations at the input to a frozen quantizer, mapping the test stream's running per-channel statistics back toward the source calibration. On ImageNet-C with ViT-B, QuAR achieves the highest mean accuracy among state-of-the-art backprop-free TTA methods at 3-, 4-, 6- and 8-bit weight/activation precision, outperforming the strongest baseline by 2.28 points at 8 bits and 4.00 at 3 bits, with 46% lower latency and a memory overhead of only 0.17 MB (0.01% of peak inference memory). Analysis and diagnostics trace the gain to a reduced per-channel mismatch at these quantizers, which restores the code distribution the baselines leave unchanged or distort further. A single fixed configuration remains ahead across continual streams, non-i.i.d. label shift, seven out-of-distribution suites, and three other backbones.
Chinese Translation
后训练量化是将视觉变换器（ViTs）适配到边缘计算和内存预算的标准方法，但量化模型在分布转移下变得特别脆弱。测试时适应（TTA）在没有标签的情况下解决此类转移，但大多数现有方法与量化推理的约束不够对齐。现有的TTA方法通过反向传播恢复准确性，而无反向传播的方法通常仍会因额外的前向传递或参数更新而产生开销，轻量级的特征或逻辑级方法仅恢复部分损失。在这些方法中，有一种特定于量化的失效模式未被直接针对：在转移下，激活在冻结的量化器的校准范围内占据的方式不同，扭曲了它们的代码分布。我们提出了量化器对齐重校准（QuAR），这是一种专为量化ViTs量身定制的单次TTA方法，既不进行反向传播也不更新任何模型参数。QuAR在输入到冻结量化器的激活上进行重校准，将测试流的每通道运行统计信息映射回源校准。在ImageNet-C上使用ViT-B，QuAR在3位、4位、6位和8位权重/激活精度下，达到了最先进的无反向传播TTA方法中的最高平均准确率，8位时比最强基线高出2.28分，3位时高出4.00分，同时延迟降低了46%，内存开销仅为0.17 MB（占峰值推理内存的0.01%）。分析和诊断将增益追溯到这些量化器的每通道不匹配的减少，这恢复了基线未改变或进一步扭曲的代码分布。在持续流、非独立同分布标签转移、七个分布外套件和三个其他主干网络中，单一固定配置始终领先。
cs.CV / 95 / 2610.08365
A Stevens's Power Law Check-up of GPT-5.5's Image-Based Visualization Reading
对GPT-5.5图像基础可视化阅读的Stevens功率定律检验
Abstract
We adapt Stevens's power law to measure the innate ability of AI models to read visualizations, which can reveal the built-in perceptual mechanisms of algorithmic models. In our pilot study, models see no legend. A model first views a reference visual representation and estimates its magnitude, then estimates the magnitude of each subsequent image of the same representation relative to that reference. Our evaluation of twelve visual variables makes how algorithmic models read visual encodings measurable, comparable with human perception, and more interpretable to humans.
Chinese Translation
我们采用Stevens功率定律来测量人工智能模型读取可视化的内在能力，这可以揭示算法模型的内置感知机制。在我们的初步研究中，模型未看到图例。模型首先查看一个参考视觉表示并估计其大小，然后相对于该参考估计每个后续图像的大小。我们对十二个视觉变量的评估使得算法模型读取视觉编码的过程可测量、可与人类感知进行比较，并更易于人类理解。
cs.CV / 96 / 2610.08379
UniCounting: Instance-Aware Proposal Consolidation for Image-Query-Free Multi-Category Counting
UniCounting：面向实例的无图像查询多类别计数提议整合
Abstract
Visual counting is commonly formulated as counting a single specified target, with a model receiving an image-specific exemplar, text query, or target category and returning a single count. We instead study fixed-vocabulary image-query-free multi-category counting. A global vocabulary is fixed for each run, and, given only an RGB image, the model predicts a complete category--count vector without being told which categories appear. We present UniCounting, which casts counting as instance-aware structural inference over an over-complete proposal set. Generic segmenters produce duplicate masks, partial views, and proposals from neighboring instances; semantic scores can name them but cannot determine which denote the same object. Frozen SAM~2.1 generates masks, while frozen DINOv2 and OpenCLIP provide relation and category features. A 3,267-parameter category-shared relation head predicts same-instance affinities from instance-mask-derived supervision. Sparse graph construction, representative selection, labeling, and background-margin admission then convert each admitted component into one count with replayable group evidence. Only the relation head is trained, without count or density-map targets. On COCO clean500, UniCounting obtains lower point-estimate vector $\ell_1$ error and absent-class false mass than calibrated OWLv2-All80, with comparable micro presence F1. Under a matched decoder, the learned relation reduces both errors relative to mask containment, mask IoU, CLIP, and DINO, while revealing a fragmentation--merge trade-off. We also report transfer diagnostics on OmniCount-sub, FSC-147, and CARPK.
Chinese Translation
视觉计数通常被表述为对单一指定目标的计数，模型接收特定于图像的示例、文本查询或目标类别，并返回一个计数。我们研究的是固定词汇的无图像查询多类别计数。在每次运行中，固定一个全局词汇，给定仅一个RGB图像，模型预测一个完整的类别-计数向量，而无需告知出现了哪些类别。我们提出了UniCounting，将计数视为对过度完整提议集的面向实例的结构推理。通用分割器生成重复的掩膜、部分视图和来自邻近实例的提议；语义分数可以命名它们，但无法确定哪些表示同一对象。冻结的SAM~2.1生成掩膜，而冻结的DINOv2和OpenCLIP提供关系和类别特征。一个具有3,267个参数的类别共享关系头从实例掩膜衍生的监督中预测同实例亲和力。稀疏图构建、代表性选择、标记和背景边际接纳将每个接纳的组件转换为一个可重放的组证据计数。只有关系头经过训练，而没有计数或密度图目标。在COCO clean500上，UniCounting获得了比校准的OWLv2-All80更低的点估计向量$ ext{l}_1$误差和缺失类别的虚假质量，同时具有可比的微观存在F1。在匹配解码器下，学习到的关系相对于掩膜包含、掩膜IoU、CLIP和DINO减少了两种误差，同时揭示了碎片化-合并的权衡。我们还报告了在OmniCount-sub、FSC-147和CARPK上的迁移诊断。
cs.CV / 97 / 2610.08398
UP-MOPD: Update Projection in Multi-Teacher On-Policy Distillation
UP-MOPD：多教师在线蒸馏中的更新投影
Abstract
On-policy distillation from multiple teachers combines expertise from different domains in a single student, but conflicting gradients can hinder this integration. Gradient corrections directly constrain parameter updates under plain SGD. With optimizers such as AdamW, however, momentum, adaptive scaling, and weight decay can turn a corrected gradient into an update that increases a domain loss to first order. To address this gap, we propose Update Projection for Multi-Teacher On-Policy Distillation (UP-MOPD). UP-MOPD lets the original mixed gradient update the optimizer state and generate a candidate displacement, then projects only violating candidates before they are committed to the parameters. The projection gives the unique feasible update closest to the candidate in Euclidean distance. In experiments combining medical and general domains, UP-MOPD improves IFEval-loose accuracy late in training by 2.96 points over vanilla M-OPD. It achieves an average score of 60.03 across eight metrics, compared with 59.00 for gradient projection and 59.15 for update rejection. On a public benchmark covering mathematics, code, and instruction following, it achieves the best average across six tasks (32.67), leads on LiveCodeBench v5, and ties for the best IFEval result.These results support projecting optimizer updates to reduce interference between domains.
Chinese Translation
多教师的在线蒸馏将来自不同领域的专业知识结合到单一学生中，但冲突的梯度可能会妨碍这种整合。梯度修正直接限制了在普通随机梯度下降（SGD）下的参数更新。然而，对于像 AdamW 这样的优化器，动量、自适应缩放和权重衰减可能会将修正的梯度转变为一个一阶增加领域损失的更新。为了解决这一问题，我们提出了多教师在线蒸馏的更新投影（UP-MOPD）。UP-MOPD 允许原始混合梯度更新优化器状态并生成候选位移，然后在提交到参数之前仅对违反条件的候选进行投影。该投影提供了与候选在欧几里得距离上最接近的唯一可行更新。在结合医学和通用领域的实验中，UP-MOPD 在训练后期将 IFEval-loose 准确率提高了 2.96 分，相较于普通的 M-OPD。它在八个指标上获得了 60.03 的平均分，而梯度投影为 59.00，更新拒绝为 59.15。在一个涵盖数学、代码和指令跟随的公共基准测试中，它在六个任务中获得了最佳平均分（32.67），在 LiveCodeBench v5 中领先，并在 IFEval 结果上并列最佳。这些结果支持通过投影优化器更新来减少领域间的干扰。
cs.CV / 98 / 2610.08401
GeoPID: Decomposing and Steering Visual Information in Vision-Language Models
GeoPID：在视觉-语言模型中分解和引导视觉信息
Abstract
While recent vision-language models (VLMs) have shown outstanding performance across diverse applications, they tend to under-use visual information and over-rely on textual context. In this work, we propose \textsc{GeoPID}, a training-free framework that analyzes multimodal information within VLMs from a geometric perspective. \textsc{GeoPID} decomposes information into Redundant, Modality-Unique, and Synergistic components through the geometric relationships between visual and textual representation subspaces. Through an extensive analysis across 22 VLMs and 14 benchmarks, we confirm that correct predictions exhibit stronger vision-unique components when questions strongly require visual grounding. Building on this geometric analysis, we introduce a targeted intervention technique that selectively amplifies visual representations along the vision-unique subspace during inference. As a result, visual grounding capabilities were enhanced without any additional model parameter updates, achieving an average relative accuracy gain of 7.63\%.
Chinese Translation
尽管近期的视觉-语言模型（VLMs）在多种应用中表现出色，但它们往往未能充分利用视觉信息，并过于依赖文本上下文。在本研究中，我们提出了 extsc{GeoPID}，一个无训练框架，从几何角度分析 VLM 中的多模态信息。 extsc{GeoPID} 通过视觉和文本表示子空间之间的几何关系，将信息分解为冗余、模态唯一和协同组件。通过对 22 个 VLM 和 14 个基准的广泛分析，我们确认，当问题强烈依赖于视觉基础时，正确预测表现出更强的视觉唯一组件。在此几何分析的基础上，我们引入了一种有针对性的干预技术，在推理过程中选择性地增强视觉表示，沿着视觉唯一子空间进行放大。因此，视觉基础能力得到了提升，而无需任何额外的模型参数更新，平均相对准确率提高了 7.63%。
cs.CV / 99 / 2610.08410
Decoy and disclosure radii of invariant shape descriptors
不变形状描述符的诱饵和披露半径
Abstract
A recognizer that compares rotation-invariant descriptors sees a surface only up to the fiber of the descriptor. We measure this fiber by its radius in the orbit distance from the enrolled surface. A large radius admits decoys, that is, distant shapes that pass the matcher. A small radius discloses the enrolled shape to anyone who captures the stored value. For star-shaped surfaces truncated to spherical harmonics of degree at most $L$, with $n$ coefficients, a descriptor of generic rank $r$ has generic fibers of dimension $n-3-r$ modulo rotations. The standard pool of band powers, even bispectra, and three invariants of the degree-three band therefore admits decoy families of dimension $5$, $13$, $20$ at $L=4,6,8$. Its rank first reaches $n-3$ at $L=16$, and a mirror decoy remains at every $L$. The odd bispectra remove the mirror decoy generically for $L \geq 4$. Yet at fixed mean radius the same pool determines the enclosed volume exactly, and it does not determine whether a surface meets a clearance requirement. We certify two cases by exact and interval arithmetic. At $L=6$ a decoy matches all $32$ invariants to relative precision $2 \cdot 10^{-18}$ at orbit distance at least $0.87$ times the norm of the enrolled tuple. For the radar shape model of asteroid (101955) Bennu, the pool recovers the modeled volume, misses the handedness, and leaves the keep-out radius uncertain by more than $7 \, \mathrm{m}$.
Chinese Translation
一个比较旋转不变描述符的识别器仅能看到表面到描述符的纤维。我们通过描述符的轨道距离来测量这个纤维的半径。较大的半径允许诱饵，即那些远离的形状能够通过匹配器。较小的半径则向任何捕获存储值的人披露已注册的形状。对于被截断到最多为 $L$ 的球面调和函数的星形表面，具有 $n$ 个系数的通用秩为 $r$ 的描述符，其通用纤维的维度为 $n-3-r$，模旋转。标准的带功率池、双谱以及三种三阶带的不变性因此在 $L=4,6,8$ 时允许维度为 $5$、$13$、$20$ 的诱饵家族。其秩在 $L=16$ 时首次达到 $n-3$，并且在每个 $L$ 处都有一个镜像诱饵。奇数双谱通常在 $L
geq 4$ 时移除镜像诱饵。然而，在固定的平均半径下，相同的池精确地确定了封闭体积，但并不能确定表面是否满足间隙要求。我们通过精确和区间算术证明了两种情况。在 $L=6$ 时，一个诱饵以相对精度 $2 imes 10^{-18}$ 匹配所有 $32$ 个不变性，轨道距离至少为已注册元组范数的 $0.87$ 倍。对于小行星 (101955) Bennu 的雷达形状模型，该池恢复了建模体积，错过了手性，并且使得保持半径的不确定性超过 $7 ext{ m}$。
cs.CV / 100 / 2610.08414
Image Bitstream Fine-grained Understanding for Privacy-Friendly AIoT
隐私友好的人工智能物联网的图像比特流细粒度理解
Abstract
Image Bitstream Fine-grained Understanding (IBFU) aims to directly perform fine-grained classification and semantic description generation from encoded image byte sequences. In contrast to conventional pixel-domain visual understanding, IBFU conducts semantic analysis without fully decoding images into the pixel domain. Since pixel-level visual content is not explicitly reconstructed during inference, this paradigm reduces visual exposure within the processing pipeline and suits privacy-friendly Artificial Intelligence of Things (AIoT) applications. In this paper, we propose Bitstream Fine-grained Generator (BFG), a novel foundation model tailored for IBFU. BFG consists of two main components: a Bitstream Semantic Encoder (BSeE) and a Fine-grained Semantic Generator (FSeG). BSeE directly models semantic representations from encoded image bitstreams without explicit pixel reconstruction, while FSeG transforms the extracted bitstream semantics into detailed natural-language descriptions through autoregressive generation. To train BFG and comprehensively evaluate IBFU in practical AIoT scenarios, where image bitstreams may suffer corruption during transmission and storage, we construct a large-scale Corrupted-bitstream Fine-grained Understanding dataset (CFU-D), containing both intact bitstreams and corrupted variants across multiple corruption types and severity levels. Experiments show that BFG maintains stable fine-grained caption generation under bitstream corruption. For example, the performance only has slight change from 0.6339 to 0.6077 in terms of average CIDEr score on Stanford Dogs Caption dataset, while vision-language models, such as Qwen-VL-Chat, BLIP-2, GLM, Gemini, and GPT suffer severe performance decrease. This paper provides a practical paradigm for privacy-friendly fine-grained understanding in AIoT.
Chinese Translation
图像比特流细粒度理解（Image Bitstream Fine-grained Understanding，IBFU）旨在直接从编码的图像字节序列中执行细粒度分类和语义描述生成。与传统的像素域视觉理解相比，IBFU在不完全解码图像到像素域的情况下进行语义分析。由于在推理过程中未显式重建像素级视觉内容，这种范式减少了处理管道中的视觉暴露，适用于隐私友好的人工智能物联网（AIoT）应用。本文提出了一种针对IBFU的新型基础模型——比特流细粒度生成器（Bitstream Fine-grained Generator，BFG）。BFG由两个主要组件组成：比特流语义编码器（Bitstream Semantic Encoder，BSeE）和细粒度语义生成器（Fine-grained Semantic Generator，FSeG）。BSeE直接从编码的图像比特流中建模语义表示，而无需显式的像素重建，FSeG则通过自回归生成将提取的比特流语义转换为详细的自然语言描述。为了训练BFG并全面评估IBFU在实际AIoT场景中的表现，其中图像比特流在传输和存储过程中可能会遭受损坏，我们构建了一个大规模的损坏比特流细粒度理解数据集（Corrupted-bitstream Fine-grained Understanding dataset，CFU-D），该数据集包含完整的比特流和多种损坏类型及严重程度的变体。实验表明，BFG在比特流损坏下仍能保持稳定的细粒度描述生成。例如，在斯坦福狗图像描述数据集上，平均CIDEr分数的表现仅从0.6339轻微下降至0.6077，而视觉-语言模型，如Qwen-VL-Chat、BLIP-2、GLM、Gemini和GPT则遭受严重的性能下降。本文为AIoT中的隐私友好的细粒度理解提供了一种实用的范式。
cs.CV / 101 / 2610.08417
Ariadne's Thread of LipSync: Unraveling Forgeries via Inconsistency between Lip Motions and Head Poses
阿里阿德涅的LipSync之线：通过唇部动作与头部姿态之间的不一致性揭示伪造
Abstract
Recent advances in LipSync generation technology have led to the creation of highly realistic videos, posing severe societal risks. However, existing defense strategies struggle against LipSync forgeries, as advanced LipSync generation methods not only achieve better lip synchronization but also eliminate visual artifacts. An important reason is that they overlook an inherent biological coupling between lip movements and head poses in natural speech videos. In this paper, we propose LipDA, a novel framework for joint LipSync Detection and Attribution, which takes advantage of the inconsistency between head and lip. For detection, the framework learns to quantify this discrepancy by contrasting lip and pose features from authentic versus forged videos. For attribution, our method is designed to capture the unique temporal dynamics and audio-visual synchronization patterns that act as the fingerprint of models, enabling source tracing. We conduct extensive experiments on two challenging LipSync datasets as well as our own proposed large-scale and multi-generator dataset. LipDA achieves over 97\% AUC in detection and 97.5\% accuracy in model attribution, significantly outperforming existing methods. Code and the proposed LipSync-A dataset are available at https://github.com/AnsonShe/LipDA.
Chinese Translation
近期LipSync生成技术的进步导致了高度逼真的视频的产生，这对社会带来了严重风险。然而，现有的防御策略在面对LipSync伪造时显得力不从心，因为先进的LipSync生成方法不仅实现了更好的唇部同步，还消除了视觉伪影。一个重要原因在于它们忽视了自然语音视频中唇部动作与头部姿态之间固有的生物耦合关系。本文提出了LipDA，一个用于联合LipSync检测和归属的新框架，利用头部与唇部之间的不一致性。对于检测，该框架通过对比真实视频与伪造视频中的唇部和姿态特征来量化这种差异。对于归属，我们的方法旨在捕捉独特的时间动态和音视频同步模式，这些模式作为模型的指纹，能够实现源追踪。我们在两个具有挑战性的LipSync数据集以及我们自己提出的大规模多生成器数据集上进行了广泛的实验。LipDA在检测中实现了超过97%的AUC，在模型归属中达到了97.5%的准确率，显著优于现有方法。代码和提出的LipSync-A数据集可在https://github.com/AnsonShe/LipDA获取。
cs.CV / 102 / 2610.08418
From the Drosophila Visual Connectome to General-Purpose Computer Vision
从果蝇视觉连接组到通用计算机视觉
Abstract
Biological connectomes encode structured solutions to visual computation that may provide reusable inductive biases for artificial vision. We develop ConnectomeX around FlyVision, a trainable architecture that preserves parallel ON/OFF processing, recurrent computation and population-level graph interaction while scaling model capacity across tasks. FlyVision reached 99.34% accuracy on MNIST with 80,608 parameters and 78.03% on CIFAR-10 with 81,408 parameters. On ImageNet-1K, FlyVision Base and Large reached 60.79% and 66.25% top-1 accuracy with 1.8 and 3.7 million parameters, while a Large local-k7 model with a learned low-frequency branch reached 66.53%, compared with 69.25% for ResNet18 with 11.7 million parameters. On a 22-class skin-disease benchmark, FlyVision Large achieved 63.78% accuracy and 95.28% macro-AUROC with 2.99 million parameters. In four-class chest radiography, ImageNet-pretrained FlyVision Base and Large reached 92.60% and 92.76% accuracy with 1.33 and 2.97 million parameters, compared with 91.56% for ImageNet-pretrained ResNet18 with 11.18 million. BrainAGE extends FlyVision to volumetric T1-weighted MRI by applying a shared ImageNet-pretrained FlyVision Large encoder to 24 sagittal, coronal and axial slices per scan and combining slice-level age estimates by confidence-modulated Gaussian voting. On 433 held-out scans, three-axis fusion achieved a mean absolute error of 5.98 years and R^2 = 0.868. Across the 224x224 classification tasks, the best FlyVision configuration remained within three percentage points of ResNet18 on ImageNet-1K and skin-disease classification and exceeded it on chest radiography with substantially fewer parameters. These results show that a conserved connectome-informed computation can scale from compact recognition to large-scale natural and biomedical vision.
Chinese Translation
生物连接组编码了视觉计算的结构化解决方案，这可能为人工视觉提供可重用的归纳偏差。我们围绕FlyVision开发了ConnectomeX，这是一种可训练的架构，能够在任务之间扩展模型容量，同时保留并行的开启/关闭处理、递归计算和群体级图交互。FlyVision在MNIST数据集上达到了99.34%的准确率，参数量为80,608；在CIFAR-10上达到了78.03%的准确率，参数量为81,408。在ImageNet-1K上，FlyVision Base和Large分别达到了60.79%和66.25%的top-1准确率，参数量为180万和370万，而一个带有学习到的低频分支的Large local-k7模型达到了66.53%的准确率，相比之下，ResNet18在1170万参数下的准确率为69.25%。在一个22类皮肤疾病基准测试中，FlyVision Large达到了63.78%的准确率和95.28%的宏观AUROC，参数量为299万。在四类胸部X光影像中，经过ImageNet预训练的FlyVision Base和Large分别达到了92.60%和92.76%的准确率，参数量为133万和297万，而ImageNet预训练的ResNet18在1118万参数下的准确率为91.56%。BrainAGE将FlyVision扩展到体积T1加权MRI，通过对每个扫描的24个矢状面、冠状面和轴面切片应用共享的ImageNet预训练FlyVision Large编码器，并通过信心调制的高斯投票结合切片级年龄估计。在433个保留扫描中，三轴融合实现了5.98年的平均绝对误差和R^2 = 0.868。在224x224分类任务中，最佳的FlyVision配置在ImageNet-1K和皮肤疾病分类中与ResNet18的差距保持在三个百分点以内，并在胸部X光影像中超越了ResNet18，同时参数量显著更少。这些结果表明，保守的连接组信息计算能够从紧凑的识别扩展到大规模的自然和生物医学视觉。
cs.CV / 103 / 2610.08419
Deformable CT-US Registration via Anatomy-Aware Implicit Neural Representations
基于解剖知识的隐式神经表示的可变形CT-US配准
Abstract
Slice-to-volume registration between ultrasound (US) and preoperative computed tomography (CT) imaging would enhance many minimally invasive interventions, for example by locating soft tissue structures intra-operatively that are discernible in CT. While optical tracking enables initial rigid registration, contact from the probe induces soft tissue deformations that inhibit accurate alignment. In this work, we introduce a deformable CT-ultrasound registration framework that incorporates anatomical priors derived from CT to improve registration under deformation. Rigid registration is first established using a robot-assisted optical tracking system, after which a deformable transformation is estimated using a sinusoidal implicit neural representation (SIREN) optimized per frame. Tissue stiffness is approximated from CT-based HU values and used as spatially varying regularization, suppressing deformation in rigid structures such as bone while allowing more flexibility in soft tissue. Two additional constraints capture the physics of probe contact: a contact-zone displacement prior that drives the displacement field to compress tissue below the probe face, and a fan-geometry regularization term based on beam direction and convex transducer field of view. Model parameters are optimized with a normalized gradient field (NGF). The proposed approach improves alignment over rigid initialisation by 17% and outperforms classical deformable baselines while maintaining near-zero topological folding.
Chinese Translation
超声（US）与术前计算机断层扫描（CT）图像之间的切片到体积配准将增强许多微创干预的效果，例如在手术过程中定位在CT中可辨别的软组织结构。虽然光学追踪可以实现初步的刚性配准，但探头的接触会导致软组织变形，从而妨碍准确对齐。在本研究中，我们提出了一种可变形CT-超声配准框架，该框架结合了来自CT的解剖先验，以改善在变形下的配准。首先，使用机器人辅助光学追踪系统建立刚性配准，随后使用针对每帧优化的正弦隐式神经表示（SIREN）估计可变形变换。组织刚度从基于CT的HU值进行近似，并用作空间变化的正则化，抑制骨骼等刚性结构的变形，同时允许软组织具有更大的灵活性。两个额外的约束捕捉探头接触的物理特性：一个接触区位移先验驱动位移场压缩探头面下方的组织，另一个基于束方向和凸换能器视场的扇形几何正则化项。模型参数通过归一化梯度场（NGF）进行优化。所提出的方法在刚性初始化的基础上提高了17%的对齐精度，并在保持近零拓扑折叠的同时超越了经典的可变形基线。
cs.CV / 104 / 2610.08433
HuC-VideoMAE: Human-Centric Video Masked Autoencoding from synthetic data
HuC-VideoMAE：基于合成数据的人本视频掩码自编码
Abstract
Modern action recognition models rely on video transformers pretrained on massive collections of web-crawled videos, such as Kinetics-700. However, the use of such data raises ethical concerns, as subjects' consent is typically not obtained. Recent high-quality synthetic video datasets generated from motion-capture data, such as BEDLAM2.0, offer a promising ethical alternative. In this work, we investigate self-supervised pretraining of video transformers on synthetic human-motion datasets. We first show that directly applying the standard VideoMAE masking strategy leads to substantially worse performance than pretraining on Kinetics. To address this limitation, we propose a human-centric masking scheme that leverages body keypoints and person bounding box regions. Our approach encourages the model to focus on the structure and dynamics of human motion during pretraining. Experiments on NTU RGB+D and Toyota-Smarthome demonstrate that our method significantly outperforms standard VideoMAE pretraining on synthetic data, closing 49% of the gap to Kinetics pretraining on NTU RGB+D cross-view-subject without using a single real frame during pretraining. To promote the use of ethical action recognition models, we will publicly release our pretrained models.
Chinese Translation
现代动作识别模型依赖于在大量网络爬取视频（如 Kinetics-700）上预训练的视频变换器。然而，使用这些数据引发了伦理问题，因为通常未获得被试的同意。最近，从动作捕捉数据生成的高质量合成视频数据集，如 BEDLAM2.0，提供了一种有前景的伦理替代方案。在本研究中，我们探讨了在合成人体运动数据集上自监督预训练视频变换器的方法。我们首先展示了直接应用标准 VideoMAE 掩码策略的效果远不如在 Kinetics 上的预训练。为了解决这一局限性，我们提出了一种以人为中心的掩码方案，该方案利用身体关键点和人物边界框区域。我们的方法鼓励模型在预训练过程中关注人体运动的结构和动态。在 NTU RGB+D 和 Toyota-Smarthome 上的实验表明，我们的方法在合成数据上显著优于标准 VideoMAE 预训练，缩小了 NTU RGB+D 跨视角-受试者与 Kinetics 预训练之间的 49% 差距，而在预训练过程中没有使用任何真实帧。为了促进伦理动作识别模型的使用，我们将公开发布我们的预训练模型。
cs.CV / 105 / 2610.08482
Knee3DVLM: Dual-Sequence Full-Volume Vision-Language Modeling for Comprehensive Knee MRI Assessment
Knee3DVLM：用于全面膝关节MRI评估的双序列全体积视觉-语言建模
Abstract
Vision-language models (VLMs) are increasingly being applied to three-dimensional medical imaging, but their application to knee MRI remains limited, particularly for interpreting the complementary sequences used in clinical practice. We introduce Knee3DVLM, a sequence-aware VLM that uses full-volume DESS and fluid-sensitive TSE MRI to predict 57 anatomically resolved binary diagnostic targets derived from the MRI Osteoarthritis Knee Score (MOAKS) for structured reporting. We evaluated DESS-only, TSE-only, and paired DESS-TSE configurations using subject-disjoint Osteoarthritis Initiative partitions. In a held-out cohort of 1,074 examinations, the fused model achieved 72.98% average accuracy, 71.17% balanced accuracy, 78.96% mean ROC-AUC, and 78.74% macro ROC-AUC, the highest values among the three configurations. In a secondary multiclass analysis aligned with the released 3DReasonKnee cohort, Knee3DVLM was numerically higher than the strongest reported 3DReasonKnee configuration across five pathology categories. These findings support dual-sequence full-volume modeling for comprehensive knee MRI assessment.
Chinese Translation
视觉-语言模型（VLMs）在三维医学影像中的应用日益增多，但其在膝关节MRI中的应用仍然有限，特别是在解释临床实践中使用的互补序列方面。我们提出了Knee3DVLM，这是一种序列感知的VLM，利用全体积的DESS和流体敏感的TSE MRI，预测57个来自MRI骨关节炎膝关节评分（MOAKS）的解剖分辨二元诊断目标，以便进行结构化报告。我们使用受试者不重叠的骨关节炎倡议分区评估了仅DESS、仅TSE和配对DESS-TSE配置。在一个包含1,074个检查的保留队列中，融合模型达到了72.98%的平均准确率、71.17%的平衡准确率、78.96%的平均ROC-AUC和78.74%的宏观ROC-AUC，这些都是三种配置中最高的值。在与发布的3DReasonKnee队列对齐的二次多类分析中，Knee3DVLM在五个病理类别中数值上高于报告的最强3DReasonKnee配置。这些发现支持双序列全体积建模用于全面的膝关节MRI评估。
cs.CV / 106 / 2610.08518
2D Spatial Reasoning with Adaptive Neural Cellular Automata
基于自适应神经元细胞自动机的二维空间推理
Abstract
Many modern learning approaches are still struggling with spatial reasoning tasks, i.e. they lack the ability to utilize geometric information of perceived entities and their spatial relation to each other to solve problems. We introduce a novel Adaptive Neural Cellular Automata (aNCA) architecture which uses deformable convolutions to dynamically adapt the perceptive field and iteratively reason over 2D spatial relations on grid-like data structures (e.g. images). Empirical results on public benchmarks show state of the art comprehensible results with high generalization abilities for solving image based puzzles like Sudoku or finding the shortest path in a maze.
Chinese Translation
许多现代学习方法在空间推理任务上仍然面临挑战，即它们缺乏利用感知实体的几何信息及其相互空间关系来解决问题的能力。我们提出了一种新颖的自适应神经元细胞自动机（Adaptive Neural Cellular Automata, aNCA）架构，该架构使用可变形卷积动态调整感知场，并在网格状数据结构（例如图像）上迭代推理二维空间关系。在公共基准测试上的实证结果显示，该方法在解决基于图像的难题（如数独）或在迷宫中寻找最短路径方面，展现了最先进的可理解结果和较强的泛化能力。
cs.CV / 107 / 2610.08528
MedCORE: Criteria-Grounded Clinical Reasoning for Interpretable Medical Image Diagnosis
MedCORE：基于标准的临床推理用于可解释的医学图像诊断
Abstract
Clinical diagnosis is inherently a structured reasoning process, yet existing deep learning models often bypass this structure by mapping image features directly to disease labels without explicitly interrogating the morphological and textural criteria that clinicians systematically evaluate. This limits diagnostic transparency and may compromise safe clinical deployment. We present MedCORE (Medical Criteria-Oriented Reasoning and Evidence), a structured diagnostic framework that operationalizes clinical reasoning within a vision-language architecture. For each input image, MedCORE decomposes the diagnostic process into clinically defined criteria, spatially localizes each criterion to diagnostically relevant image regions, encodes evidence through multi-scale representations that capture macro-structural and micro-textural pathological characteristics, and refines criterion representations using a Graph Attention Network that explicitly models inter-criteria dependencies. Criterion representations are further aligned with clinical text descriptors, reinforced through class-wise visual prototypes, and aggregated using uncertainty-calibrated weighting that proportionally discounts low-confidence diagnostic evidence. MedCORE is validated across three clinically heterogeneous imaging modalities, including dermoscopic lesion classification on ISIC 2018, breast ultrasound lesion characterization on BUSI, and diabetic retinopathy grading on IDRiD. Quantitatively, MedCORE achieves 89.2% accuracy, 85.7% macro-F1, and 96.4% AUC on ISIC 2018; 96.1% accuracy, 95.2% macro-F1, and 98.4% AUC on BUSI; and 84.3% accuracy, 80.2% macro-F1, and 92.8% AUC on IDRiD. These results demonstrate consistent improvements over strong CNN, transformer, biomedical vision-language, concept-based, and prototype-based baselines.
Chinese Translation
临床诊断本质上是一个结构化的推理过程，但现有的深度学习模型往往绕过这一结构，直接将图像特征映射到疾病标签，而未明确探讨临床医生系统性评估的形态和纹理标准。这限制了诊断的透明性，并可能影响安全的临床应用。我们提出了MedCORE（医学标准导向推理与证据），这是一个在视觉-语言架构中实现临床推理的结构化诊断框架。对于每个输入图像，MedCORE将诊断过程分解为临床定义的标准，空间定位每个标准到与诊断相关的图像区域，通过多尺度表示编码证据，捕捉宏观结构和微观纹理的病理特征，并使用图注意力网络（Graph Attention Network）精炼标准表示，明确建模标准间的依赖关系。标准表示进一步与临床文本描述符对齐，通过类别视觉原型强化，并使用不确定性校准加权进行聚合，按比例降低低置信度诊断证据的影响。MedCORE在三种临床异质成像模式下进行了验证，包括ISIC 2018上的皮肤镜病变分类、BUSI上的乳腺超声病变特征描述和IDRiD上的糖尿病视网膜病变分级。在定量上，MedCORE在ISIC 2018上达到了89.2%的准确率、85.7%的宏观F1值和96.4%的AUC；在BUSI上达到了96.1%的准确率、95.2%的宏观F1值和98.4%的AUC；在IDRiD上达到了84.3%的准确率、80.2%的宏观F1值和92.8%的AUC。这些结果显示出相对于强CNN、变换器（transformer）、生物医学视觉-语言、基于概念和基于原型的基线的一致性提升。
cs.CV / 108 / 2610.08533
Beyond Perturbation Magnitude: Direction-Dependent Responses in Multimodal Geometric Representations
超越扰动幅度：多模态几何表示中的方向依赖响应
Abstract
Geometric alignment scores based on Gram determinants provide a compact way to model higher-order consistency among modalities, yet how such scores respond to modality degradation is poorly understood. This paper asks whether the response of a multimodal geometric score is determined primarily by the magnitude of the perturbation-induced displacement. Using frozen cohorts from MSR-VTT (N=878) and DiDeMo (N=980), we apply controlled video blur and audio noise and analyze the response in the relational geometry on which the score is defined. Displacement magnitude explains at most 15% of the out-of-sample variance in the absolute response, and magnitude-matched pairs respond systematically differently, so scalar magnitude does not organize the response. The closed-form first-order expansion of the Gramian volume yields the Directional Geometric Response (DGR): the projection of the displacement onto the local volume gradient, which jointly captures the clean operating point, displacement magnitude, and displacement direction. The absolute first-order DGR term explains the observed response with out-of-sample R^2 of 0.838-0.969, matched-magnitude ranking accuracies of 0.864-0.963, and response-sign accuracies of 0.909-0.989, whereas the tested direction-free alternatives remain weak or unstable under the corresponding evaluation protocols. A pre-specified gain-normalization candidate, V/(g_V+eps), fails its predictability and clean-order gates. DGR uses the observed degraded-state displacement and is therefore an explanatory quantity, not a deployment-time predictor: geometric response depends on where the representation operates, how far degradation moves the relational geometry, and in which direction it moves.
Chinese Translation
基于Gram行列式的几何对齐分数提供了一种紧凑的方法来建模模态之间的高阶一致性，但这种分数对模态降级的响应尚不清楚。本文探讨多模态几何分数的响应是否主要由扰动引起的位移幅度决定。我们使用来自MSR-VTT（N=878）和DiDeMo（N=980）的冻结队列，施加控制的视频模糊和音频噪声，并分析定义该分数的关系几何中的响应。位移幅度最多解释了绝对响应的15%的样本外方差，而幅度匹配的对在响应上系统性地表现出不同，因此标量幅度并未组织响应。Gram体积的一阶闭式展开产生了方向几何响应（Directional Geometric Response, DGR）：位移在局部体积梯度上的投影，它共同捕捉了干净的操作点、位移幅度和位移方向。绝对一阶DGR项以0.838-0.969的样本外R^2、0.864-0.963的匹配幅度排名准确率和0.909-0.989的响应符号准确率解释了观察到的响应，而经过测试的无方向替代方案在相应的评估协议下仍然表现不佳或不稳定。一个预先指定的增益归一化候选项V/(g_V+eps)未能通过其可预测性和清洁顺序门。DGR使用观察到的降级状态位移，因此是一个解释量，而非部署时的预测器：几何响应取决于表示的操作位置、降级如何移动关系几何以及移动的方向。
cs.CV / 109 / 2610.08539
RSJEV: Discriminative Remote Sensing Scene Classification with Multimodal Large Language Models
RSJEV：基于多模态大语言模型的区分性遥感场景分类
Abstract
Remote sensing scene classification is a fundamental task in Earth observation and geospatial analysis. Existing approaches mainly follow three paradigms: task-specific visual classification, vision-language similarity matching, and autoregressive multimodal generation. However, visual classifiers rely on predefined label spaces, CLIP-based methods perform recognition through static image-text alignment, and multimodal large language models (MLLMs) introduce unnecessary token-level generation for classification tasks with explicit candidate categories. To address these limitations, we propose RSJEV, a one-pass multimodal decision framework for remote sensing scene classification. Unlike conventional MLLMs that formulate classification as autoregressive text generation, RSJEV reformulates scene classification as a candidate-conditioned multimodal discriminative decision process, where visual representations, task instructions, and candidate category semantics are jointly modeled. Specifically, we introduce a OnePass Decider that extracts multimodal decision states and directly estimates category probabilities within the candidate category space, eliminating autoregressive decoding while preserving vision-language interactions. Extensive experiments on three widely used remote sensing scene classification benchmarks, including UC Merced, AID, and NWPU-RESISC45, demonstrate that RSJEV achieves superior classification performance compared with representative CNN-, Transformer-, Mamba-, CLIP-, and MLLM-based methods. Moreover, RSJEV significantly reduces inference costs and achieves a better accuracy-efficiency trade-off with only a compact 0.8B-parameter model. These results demonstrate the effectiveness of state-conditioned multimodal decision making for efficient remote sensing image understanding. The code will be available at https://github.com/Dongtcs/RSJEV.
Chinese Translation
遥感场景分类是地球观测和地理空间分析中的一项基础任务。现有方法主要遵循三种范式：特定任务的视觉分类、视觉-语言相似性匹配和自回归多模态生成。然而，视觉分类器依赖于预定义的标签空间，基于CLIP的方法通过静态图像-文本对齐进行识别，而多模态大语言模型（MLLMs）在具有明确候选类别的分类任务中引入了不必要的令牌级生成。为了解决这些局限性，我们提出了RSJEV，一种用于遥感场景分类的一次性多模态决策框架。与将分类视为自回归文本生成的传统MLLMs不同，RSJEV将场景分类重新表述为候选条件的多模态区分决策过程，其中视觉表示、任务指令和候选类别语义被联合建模。具体而言，我们引入了一个OnePass Decider，它提取多模态决策状态并直接在候选类别空间内估计类别概率，消除了自回归解码，同时保留了视觉-语言交互。在UC Merced、AID和NWPU-RESISC45等三个广泛使用的遥感场景分类基准上的大量实验表明，RSJEV相比于代表性的基于CNN、Transformer、Mamba、CLIP和MLLM的方法实现了更优的分类性能。此外，RSJEV显著降低了推理成本，并以仅0.8B参数的紧凑模型实现了更好的准确性与效率的权衡。这些结果证明了状态条件的多模态决策在高效遥感图像理解中的有效性。代码将发布在 https://github.com/Dongtcs/RSJEV。
cs.CV / 110 / 2610.08560
Have I Seen Enough? Frozen Video-Language Models Encode Evidence Readiness
我见过足够吗？冻结的视频语言模型编码证据准备性
Abstract
Streaming video-language models must decide not only what to answer, but whether the evidence needed for the current question has arrived. Existing systems learn that decision as a separate trigger; we ask whether an unmodified model already computes it. We show that frozen VideoLLMs carry a linearly readable evidence-readiness signal, labelled from timestamped evidence rather than from model output. It decodes in all seven models of a shared byte-identical evaluation (AUROC 0.733-0.905 under the strictest not-ready sampling, where a fitted clock is near chance), and a probe fitted without any of a benchmark family's footage still reads that family. It is question-conditioned: on byte-identical windows, changing only the question reverses the readout on 66.1% of pairs, while every question-blind control is at chance by construction. The model can answer incorrectly and still encode readiness: AUROC remains 0.722 among wrong answers. Readiness also beats uncertainty estimators and their supervised combination on latency-matched answer selection, and tracks independent human judgments more closely than confidence. Released streaming triggers are also linear readouts, yet a trained trigger read on its own base model's activations is approximately orthogonal to readiness and decodes it far less accurately than a probe. We turn the readout into Readiness Gating, an answer-timing policy that improves accuracy by up to +9.75 pp at matched video duration with negligible computational overhead. How much it gains varies with the accuracy headroom the task makes available: across 26 configurations the gain tracks that headroom, and an intervention that moves it over identical pixels moves the gain with it.
Chinese Translation
流媒体视频语言模型不仅需要决定回答什么，还需要判断当前问题所需的证据是否已经到达。现有系统将这一决策学习为一个独立的触发器；我们探讨一个未修改的模型是否已经计算出这一点。我们展示了冻结的 VideoLLMs 传递出一个线性可读的证据准备性信号，该信号是从时间戳证据标记而来，而非从模型输出中得出。在七个共享字节相同的评估模型中，它在最严格的未准备采样下的 AUROC 为 0.733-0.905（此时拟合的时钟接近随机），而且即使没有基准家族的任何视频，拟合的探针仍然能够读取该家族。该信号是条件于问题的：在字节相同的窗口中，仅更改问题就会在 66.1% 的配对中反转读取，而每个问题盲控件的结果则是随机的。模型可以错误回答，但仍然编码准备性：在错误答案中 AUROC 仍保持在 0.722。准备性在延迟匹配的答案选择中也优于不确定性估计器及其监督组合，并且比信心更紧密地追踪独立的人类判断。发布的流媒体触发器也是线性读取，但训练的触发器在其自身基础模型的激活上读取的结果与准备性大致正交，且解码的准确性远低于探针。我们将读取结果转化为准备性门控（Readiness Gating），这是一种答案时机策略，在匹配视频时长的情况下提高了准确性，增幅可达 +9.75 个百分点，且计算开销微乎其微。增益的大小取决于任务所提供的准确性余地：在 26 种配置中，增益与该余地相匹配，而一个将其移动到相同像素的干预也会随之移动增益。
cs.CV / 111 / 2610.08573
Sparse2comm: Towards Robust Cooperative 3D Object Detection
Sparse2comm：迈向稳健的协作三维物体检测
Abstract
Cooperative perception improves autonomous driving by sharing complementary observations among vehicles and roadside infrastructure for 3D object detection. However, practical deployment is constrained by limited bandwidth and unreliable cooperation, where packet loss, transmission delay, and spatial misalignment jointly degrade the cooperative feature stream. Existing methods often reduce communication cost or compensate for one degradation type, leaving coupled disturbances insufficiently addressed. To address this problem, we propose Sparse2comm, a bandwidth-efficient and robust cooperative 3D object detection framework that treats unreliable cooperation as progressive restoration over degraded cooperative features. Sparse Feature Encoding first encodes communication as randomly mask-sampled foreground features transmitted by collaborating agents, from which the ego vehicle reconstructs dense semantic representations. This sparse-to-dense mechanism learns to infer missing object-centric content from sparse observations, enabling ultra-low-bandwidth communication and packet-loss recovery within the same representation. On the semantically restored features, Latency-Aware Alignment predicts motion flow to compensate delayed messages, and Self-Calibrating Fusion estimates residual spatial offsets in a self-supervised manner before adaptive cross-agent fusion. Sparse2comm therefore restores semantic completeness, temporal consistency, and spatial alignment in an ordered pipeline. Extensive experiments on DAIR-V2X, OpenV2V, and V2V4Real show that Sparse2comm maintains competitive clean accuracy and consistently improves robustness under individual and mixed real-world degradations. Compared with the selective feature communication baseline Where2comm, Sparse2comm improves mixed-setting
[email protected]/
[email protected] by +20.15/+11.79, +12.66/+11.07, and +15.36/+12.61 on the three datasets, respectively.
Chinese Translation
协作感知通过在车辆和路边基础设施之间共享互补观测来改善自动驾驶，以实现三维物体检测。然而，实际部署受到带宽有限和协作不可靠的限制，其中数据包丢失、传输延迟和空间不对齐共同降低了协作特征流的质量。现有方法通常减少通信成本或补偿一种降级类型，但对耦合干扰的处理不足。为了解决这一问题，我们提出了Sparse2comm，一种带宽高效且稳健的协作三维物体检测框架，将不可靠的协作视为对降级协作特征的渐进恢复。稀疏特征编码首先将通信编码为由协作代理随机掩码采样的前景特征，随后自我车辆从中重建密集语义表示。这一稀疏到密集的机制学习从稀疏观测中推断缺失的物体中心内容，从而实现超低带宽通信和在同一表示中的数据包丢失恢复。在语义恢复的特征上，延迟感知对齐预测运动流以补偿延迟消息，自我校准融合以自监督方式估计残余空间偏移，然后进行自适应跨代理融合。因此，Sparse2comm在有序管道中恢复语义完整性、时间一致性和空间对齐。在DAIR-V2X、OpenV2V和V2V4Real上的大量实验表明，Sparse2comm保持了竞争性的清晰准确性，并在个体和混合现实世界降级下持续提高了稳健性。与选择性特征通信基线Where2comm相比，Sparse2comm在三个数据集上分别提高了混合设置下的
[email protected]/
[email protected] +20.15/+11.79、+12.66/+11.07和+15.36/+12.61。
cs.CV / 112 / 2610.08574
FedDermaSeg: Federated Learning for Dermatological Image Segmentation
FedDermaSeg：用于皮肤图像分割的联邦学习
Abstract
Skin cancer is a major global health concern, and early detection and accurate lesion delineation are important for effective diagnosis and treatment planning. Automated skin lesion analysis can assist dermatologists, with lesion segmentation serving as a fundamental step in computer-aided diagnostic systems. Conventional deep learning-based segmentation models typically rely on centralized training, where images and their corresponding segmentation masks are collected on a central server. Such data aggregation raises privacy concerns in medical applications and requires substantial centralized computational resources. To address these limitations, we investigate the feasibility of federated learning for privacy-preserving skin lesion segmentation. The training and validation sets of the ISIC 2018 Skin Lesion Segmentation Challenge dataset are used to simulate a distributed learning environment and develop a federated segmentation model. The resulting model is evaluated on the ISIC 2018 test set and the PH2 dataset to assess its performance and generalizability. Experimental results demonstrate that the federated model achieves performance comparable to centralized training while consistently improving upon the locally trained models. These findings demonstrate the potential of federated learning for collaborative skin lesion segmentation without requiring centralized aggregation of medical images.
Chinese Translation
皮肤癌是全球主要的健康问题，早期检测和准确的病变描绘对有效的诊断和治疗规划至关重要。自动化的皮肤病变分析可以辅助皮肤科医生，其中病变分割是计算机辅助诊断系统中的基本步骤。传统的基于深度学习的分割模型通常依赖于集中式训练，即在中央服务器上收集图像及其对应的分割掩膜。这种数据聚合在医疗应用中引发隐私问题，并需要大量的集中计算资源。为了解决这些限制，我们研究了联邦学习在隐私保护的皮肤病变分割中的可行性。使用ISIC 2018皮肤病变分割挑战数据集的训练和验证集来模拟分布式学习环境并开发联邦分割模型。所得到的模型在ISIC 2018测试集和PH2数据集上进行评估，以评估其性能和泛化能力。实验结果表明，联邦模型的性能与集中训练相当，同时在本地训练模型的基础上持续改进。这些发现展示了联邦学习在无需集中聚合医疗图像的情况下实现协作皮肤病变分割的潜力。
cs.CV / 113 / 2610.08620
LiDAR Resolution Recovery via Foundation-Model-Guided Diffusion
基于基础模型引导的LiDAR分辨率恢复
Abstract
High-beam-count LiDAR sensors are costly, yet many perception pipelines require dense angular sampling. Using a pretrained Stable Diffusion model as the backbone, we fine-tune a LiDAR-conditioned depth model with pseudo-depth targets from a 2D foundation model. During training, the LiDAR conditioning is randomly decimated at different beam budgets. We then investigate how much of a LiDAR scan can be recovered from heavily decimated input and characterize performance across the input beam budget. We evaluate against physically held-out real beams on nuScenes and report recovery separately from fit accuracy. Our model yields its largest advantage in very sparse regimes, achieving a $\delta_{1.25}$ accuracy of $66.8$% from $4$-beam input where scattered interpolation reaches only $45.1$%. A class-stratified error breakdown further reveals that planar surfaces recover first while objects introducing depth discontinuities degrade earliest. Together, these results quantify the recovery/resolution trade-off for foundation-model-guided LiDAR enhancement.
Chinese Translation
高光束计数的LiDAR传感器成本高昂，但许多感知管道需要密集的角度采样。我们使用预训练的Stable Diffusion模型作为骨干，微调一个基于LiDAR的深度模型，该模型利用来自2D基础模型的伪深度目标。在训练过程中，LiDAR条件随机以不同的光束预算进行降采样。我们随后研究在重度降采样输入下，能够恢复多少LiDAR扫描，并对输入光束预算的性能进行表征。我们在nuScenes上对真实光束进行物理保留评估，并将恢复结果与拟合精度分开报告。我们的模型在非常稀疏的情况下表现出最大的优势，在4光束输入下实现了66.8%的$ ext{δ}_{1.25}$精度，而散点插值仅达到45.1%。分类分层的错误分析进一步揭示，平面表面最先恢复，而引入深度不连续的物体最早降解。综合来看，这些结果量化了基于基础模型引导的LiDAR增强的恢复/分辨率权衡。
cs.CV / 114 / 2610.08639
Forensic Reserve: Eliciting Latent Knowledge for Image Forgery Detection
法医储备：提取潜在知识以进行图像伪造检测
Abstract
As generated images become increasingly realistic, reliable forgery detection is essential for maintaining trust in visual information. However, existing methods primarily rely on task-specific supervision to adapt vision foundation model representations, without fully exploiting internal forensic knowledge to guide detection. To address this limitation, we propose Reserve-Guided Elicitation (RGE), a framework that treats sparse, origin-sensitive internal components in pretrained models as a forensic reserve and translates their localization into structural constraints for lightweight adaptation. Specifically, we first use the Forensic Lens (F-lens) to decompose activations across layers and token groups into independent components and globally screen them by their response differences between real and generated images, identifying reserve sites and directions. Next, we map the selected directions back to hidden-state space to construct fixed reserve subspaces and insert Forensic Reserve Adapters (FRA) only at the identified sites. Finally, with the backbone parameters, previously fitted reference classifier, and subspace bases fixed, we train only the FRA coefficient maps to generate input-dependent residual updates constrained to the corresponding subspaces, strengthening existing forensic responses. Using only 500 labeled training images and a trainable parameter budget below 0.2% of the backbone, RGE achieves competitive performance across three detection benchmarks without target-benchmark adaptation. Furthermore, RGE consistently improves over the corresponding frozen detectors across eight encoders spanning self-supervised and vision-language pretraining, eliciting a latent forensic capacity broadly shared across pretrained vision models.
Chinese Translation
随着生成图像变得越来越逼真，可靠的伪造检测对于维护视觉信息的信任至关重要。然而，现有方法主要依赖于特定任务的监督来调整视觉基础模型的表示，而未能充分利用内部法医知识来指导检测。为了解决这一局限性，我们提出了储备引导提取（Reserve-Guided Elicitation, RGE），这是一个将预训练模型中稀疏的、对来源敏感的内部组件视为法医储备，并将其定位转化为轻量适应的结构约束的框架。具体而言，我们首先使用法医透镜（Forensic Lens, F-lens）将各层和标记组的激活分解为独立组件，并通过真实图像与生成图像之间的响应差异对其进行全球筛选，从而识别储备位置和方向。接下来，我们将选定的方向映射回隐藏状态空间，以构建固定的储备子空间，并仅在识别出的位置插入法医储备适配器（Forensic Reserve Adapters, FRA）。最后，在固定主干参数、先前拟合的参考分类器和子空间基的情况下，我们仅训练FRA系数图，以生成依赖于输入的残差更新，并限制在相应的子空间内，从而增强现有的法医响应。仅使用500张标记训练图像和低于主干0.2%的可训练参数预算，RGE在三个检测基准上实现了竞争力的性能，而无需针对目标基准进行适应。此外，RGE在八个编码器上持续优于相应的冻结检测器，这些编码器涵盖了自监督和视觉-语言预训练，提取了广泛共享于预训练视觉模型中的潜在法医能力。
cs.CV / 115 / 2610.08649
Stable Scores, Unstable Answers: Frame Phase and Option Order in Video Multiple-Choice Evaluation
稳定得分，不稳定答案：视频多项选择评估中的帧相位和选项顺序
Abstract
Video-language models are ranked by multiple-choice accuracy on frames from a uniform grid. The grid has two parameters, a rate and a phase, and benchmarks report only the rate. The phase moves answers: two deployed samplers differing only by a half-step phase offset answer 23.6% of questions differently while scoring within a point, and across four releases from two families shifting only the phase changes roughly one answer in five after controlling option order. PHASEFUSION decodes three offset grids and averages the option posteriors. The grids are the polyphase components of the dense grid. Fusion matches a 32-frame single pass in accuracy within a prespecified margin (logit-scored) and cuts the answers a half-step shift of all three grids changes from 18.2% to 10.1%. Option order, which changes only the presentation, is flagged instead by a one-pass answer margin. Report the phase convention with the budget, or marginalize it.
Chinese Translation
视频语言模型通过在均匀网格中的帧上进行多项选择准确性排名。该网格有两个参数：速率和相位，而基准测试仅报告速率。相位会影响答案：两个仅在半步相位偏移上有所不同的采样器对23.6%的问题给出了不同的答案，同时得分相差不超过一个点。在来自两个系列的四个版本中，仅相位的变化在控制选项顺序后大约改变了五分之一的答案。PHASEFUSION解码三个偏移网格并平均选项后验分布。这些网格是密集网格的多相成分。融合在预设的边际范围内（对数得分）与32帧单次通过的准确性相匹配，并且所有三个网格的半步偏移导致的答案变化从18.2%降至10.1%。选项顺序仅改变展示方式，反而通过单次回答边际被标记。报告相位约定及其预算，或对其进行边际化处理。
cs.CV / 116 / 2610.08659
Selective Transfer of RL Updates for Visual Reasoning
视觉推理的强化学习更新选择性转移
Abstract
Model merging provides a training-free way to transfer reasoning capabilities from language models to vision-language models (VLMs), but endpoint-based transfer can conflate pre-existing model differences with changes acquired during reasoning post-training. We instead formulate capability transfer around the training-stage update, isolating the parameter changes induced by reinforcement learning (RL). Yet transferring this update in full remains suboptimal: we find that its components differ substantially in cross-model transferability, with dominant directions transferring more effectively than the complete update. Based on this finding, we introduce Selective-RL, which isolates the RL-stage update, retains its dominant matrix-wise directions with magnitude preservation, and transfers them to the language modules of a VLM. Across three model families and five visual-reasoning benchmarks, Selective-RL improves full-update interpolation in 12 of 15 comparisons, including an 8.55 percentage-point MathVision gain on the Qwen recipient. Matched controls show that update magnitude or arbitrary low rank alone does not reproduce these gains. These results highlight a distinction between what is acquired during post-training and what remains transferable across models, providing a training-stage perspective on cross-model capability transfer. Code is available at https://anonymous.4open.science/r/selective-rl.
Chinese Translation
模型合并提供了一种无训练的方式，将推理能力从语言模型转移到视觉-语言模型（VLMs），但基于端点的转移可能会将预先存在的模型差异与训练后推理过程中获得的变化混淆。我们提出了一种围绕训练阶段更新的能力转移方法，隔离由强化学习（RL）引起的参数变化。然而，完整转移这一更新仍然是次优的：我们发现其组成部分在跨模型转移性上存在显著差异，主导方向的转移效果优于完整更新。基于这一发现，我们引入了选择性强化学习（Selective-RL），该方法隔离RL阶段更新，保留其主导矩阵方向并保持幅度，然后将其转移到VLM的语言模块。在三个模型系列和五个视觉推理基准测试中，选择性强化学习在15次比较中改善了12次完整更新插值，包括在Qwen接收者上获得8.55个百分点的MathVision提升。匹配控制实验表明，仅凭更新幅度或任意低秩并不能重现这些提升。这些结果突显了训练后获得的内容与跨模型可转移内容之间的区别，为跨模型能力转移提供了训练阶段的视角。代码可在 https://anonymous.4open.science/r/selective-rl 获取。
cs.CV / 117 / 2610.08663
Knowing When to Trust a Prior: Reliability-Gated Cue Fusion for Video Gaze Prediction
何时信任先验知识：用于视频注视预测的可靠性门控线索融合
Abstract
Video gaze prediction is led by gaze-trained models, yet gaze-free priors carry signal those models have not absorbed, if one knows when to trust them. We propose FocusGate, a gated ensemble of gaze-free priors whose members may abstain. A per-frame gate reads three shape statistics of a defocus map and selects the frames on which the estimator is above chance on average, so rejected frames reduce to the base exactly, while midrank normalisation lets an all-zero prior abstain at zero parameters. Gated fusion is significantly positive on film, sports and web video, whereas unconditional fusion is harmful on sports and null on web. Added to four supervised predictors, the NTIRE 2026 champion among them, FocusGate improves all sixteen model-domain cells in shuffled AUC, fifteen significantly, one domain pre-registered and scored once, while adding only 1% to the champion's latency. Alone, it surpasses TASED-Net and UNISAL in shuffled AUC on film with a 16-frame causal mean.
Chinese Translation
视频注视预测由经过注视训练的模型主导，但无注视先验知识传递了这些模型未能吸收的信号，前提是我们知道何时信任它们。我们提出了FocusGate，这是一种门控的无注视先验集合，其成员可以选择不参与。每帧的门控读取了失焦图的三个形状统计量，并选择在平均上估计器高于随机猜测的帧，因此被拒绝的帧准确减少到基准，而中位数归一化允许全零先验在零参数时选择不参与。门控融合在电影、体育和网络视频上显著有效，而无条件融合在体育上有害，在网络视频上无效。FocusGate与四个监督预测器结合，其中包括NTIRE 2026的冠军，改善了所有十六个模型领域单元的随机AUC，其中十五个显著提高，一个领域预注册并仅评分一次，同时仅增加了冠军延迟的1%。单独使用时，它在电影的随机AUC上超越了TASED-Net和UNISAL，具有16帧的因果均值。
cs.CV / 118 / 2610.08672
PDB: Point-Based Deformation Blending for Facial Animation Retargeting
PDB：基于点的变形混合用于面部动画重定向
Abstract
Mesh-agnostic facial animation retargeting transfers expressions across meshes with different structures, but preserving facial motion without surface artifacts remains challenging. To address this, we present PDB, Point-Based Deformation Blending for facial animation retargeting. PDB predicts a compact set of deformed control points from a source neutral-expression pair and blending weights from the target neutral mesh. The weights are computed once per target and reused across frames, while the control points vary with each source expression. ReLU enforces non-negative weights and permits exact zeros, followed by row-wise normalization. The target mesh is reconstructed directly by multiplying the weights and control points, without a predefined cage, precomputed coordinates, a learned per-element deformation decoder, or a global reconstruction solve. Trained only with self-retargeting reconstruction supervision, PDB supports cross-identity transfer without paired cross-identity training expressions. Experiments demonstrate accurate retargeting, fast inference, and localized support in the learned weights. Joint evaluation of expression accuracy and local surface preservation shows reduced surface artifacts relative to the evaluated dense displacement method while retaining the intended motion. Perceptual evaluations further support expression fidelity and visual quality in both self- and cross-retargeting.
Chinese Translation
网格无关的面部动画重定向在不同结构的网格之间转移表情，但在不产生表面伪影的情况下保持面部运动仍然具有挑战性。为了解决这个问题，我们提出了PDB，即基于点的变形混合用于面部动画重定向。PDB从源中性表情对预测一组紧凑的变形控制点，并从目标中性网格中预测混合权重。权重在每个目标上计算一次，并在帧间重复使用，而控制点则随着每个源表情而变化。ReLU强制非负权重并允许精确的零值，随后进行行归一化。目标网格通过将权重和控制点相乘直接重建，无需预定义的笼子、预计算的坐标、学习的每元素变形解码器或全局重建求解。PDB仅通过自重定向重建监督进行训练，支持跨身份转移，而无需配对的跨身份训练表情。实验表明，重定向准确、推理快速，并且学习的权重具有局部支持。对表情准确性和局部表面保留的联合评估显示，相较于评估的密集位移方法，表面伪影减少，同时保持了预期的运动。感知评估进一步支持自重定向和跨重定向中的表情保真度和视觉质量。
cs.CV / 119 / 2610.08674
EC-RAG: Event Chain Retrieval-Augmented Generation for Long Video Understanding
EC-RAG：事件链检索增强生成用于长视频理解
Abstract
Current large video-language models (LVLMs) still face challenges when dealing with long videos, mainly because frames are often processed independently, making it difficult to capture temporal dependencies across events. Although retrieval-augmented approaches have been introduced to provide additional context, most of them operate at the frame or snippet level, which limits their ability to model how events evolve over time and relate to each other. In this paper, we propose Event Chain Retrieval-Augmented Generation (EC-RAG), a training-free framework that organizes video content into an explicit event chain before question answering. Instead of retrieving isolated frames or text segments, EC-RAG first partitions the video into semantically coherent segments, represents each segment using multi-modal signals, and then links them into a structured chain that preserves temporal order and captures inter-event relationships. Given a query, the system identifies relevant events within this chain and gathers supporting evidence from the associated modalities. Our approach offers several practical advantages: (i) event-level abstraction that better reflects how video content is naturally structured, enabling more reliable localization compared to frame-level retrieval; (ii) structured multi-modal fusion that aggregates speech, text, and visual cues at the event level, allowing complementary information to be more effectively utilized during reasoning; and (iii) plug-and-play compatibility with existing LVLM backbones, requiring no additional training or reliance on proprietary models. Experiments on Video-MME, MLVU, and LongVideoBench show that this event-centric design consistently outperforms frame-level retrieval baselines, highlighting the importance of modeling temporal structure for long-video understanding.
Chinese Translation
当前的大型视频语言模型（LVLMs）在处理长视频时仍面临挑战，主要是因为帧通常是独立处理的，这使得捕捉事件之间的时间依赖关系变得困难。尽管已经引入了检索增强的方法以提供额外的上下文，但大多数方法是在帧或片段级别操作，这限制了它们建模事件如何随时间演变及相互关系的能力。本文提出了事件链检索增强生成（EC-RAG），这是一个无训练框架，在问答之前将视频内容组织成一个明确的事件链。EC-RAG首先将视频划分为语义一致的片段，使用多模态信号表示每个片段，然后将它们链接成一个保持时间顺序并捕捉事件间关系的结构化链，而不是检索孤立的帧或文本片段。给定一个查询，系统在该链中识别相关事件，并从相关模态中收集支持证据。我们的方法提供了几个实际优势：（i）事件级抽象更好地反映了视频内容的自然结构，相比于帧级检索，能够实现更可靠的定位；（ii）结构化的多模态融合在事件级别汇聚语音、文本和视觉线索，使得在推理过程中更有效地利用互补信息；（iii）与现有LVLM骨干网的即插即用兼容性，无需额外训练或依赖专有模型。在Video-MME、MLVU和LongVideoBench上的实验表明，这种以事件为中心的设计始终优于帧级检索基线，突显了建模时间结构在长视频理解中的重要性。
cs.CV / 120 / 2610.08684
RenderBench: Benchmarking Render-to-Real Video Transfer with Reconstructed Digital Twins
RenderBench：基于重建数字双胞胎的渲染到真实视频传输基准测试
Abstract
Modern video models can generate realistic videos from real appearance references and proxy renders that specify scene structure, viewpoint changes, and motion. Evaluating this render-to-real capability requires a real target video depicting the same scene evolution, paired with an editable, geometrically registered 3D replica. Such data has traditionally required substantial manual modeling, calibration, and animation effort. We introduce RenderBench, a benchmark of 12 reconstructed real-world scenes spanning large-scale indoor environments and egocentric viewpoints, with both static and dynamic settings. Our construction pipeline combines visual geometry, neural reconstruction, and assisted 3D authoring. Each scene is decomposed into static objects and dynamic actors, registered to the capture cameras, and accepted only after multi-view geometric and temporal validation. Each evaluation unit contains appearance reference images, a held-out real target video, an editable digital twin, a matched proxy render, and renderer-native scene annotations. We evaluate transfer models against paired real target videos, retain PAI-Bench-C-compatible structural projections, and use scene annotations to localize failures by object, visibility, articulation, and motion. The first release retains 12 of 14 registered samples (85.7%), comprising 1,496 paired real-proxy frames. All released scenes pass file-integrity and environment-edit audits, while proxy diagnostics yield a depth si-RMSE of 0.2170 and instance mIoU of 0.3673. RenderBench provides paired real observations and editable scene state for assessing both appearance fidelity and preservation of geometry and dynamics.
Chinese Translation
现代视频模型能够从真实外观参考和指定场景结构、视角变化及运动的代理渲染中生成逼真的视频。评估这种渲染到真实的能力需要一个真实的目标视频，该视频描绘了相同场景的演变，并配有一个可编辑的、几何上对齐的3D复制品。这类数据传统上需要大量的手动建模、校准和动画工作。我们介绍了RenderBench，这是一个包含12个重建的真实场景的基准测试，涵盖了大规模室内环境和自我中心视角，包括静态和动态设置。我们的构建流程结合了视觉几何、神经重建和辅助3D创作。每个场景被分解为静态物体和动态角色，并与捕获相机进行对齐，仅在经过多视角几何和时间验证后才被接受。每个评估单元包含外观参考图像、一个保留的真实目标视频、一个可编辑的数字双胞胎、一个匹配的代理渲染以及渲染器原生的场景注释。我们针对配对的真实目标视频评估传输模型，保留与PAI-Bench-C兼容的结构投影，并使用场景注释通过物体、可见性、关节和运动来定位失败。首次发布保留了14个注册样本中的12个（85.7%），共计1496对真实-代理帧。所有发布的场景都通过了文件完整性和环境编辑审核，而代理诊断的深度si-RMSE为0.2170，实例mIoU为0.3673。RenderBench提供了配对的真实观察和可编辑的场景状态，以评估外观保真度及几何和动态的保持。
cs.CV / 121 / 2610.08713
SpaTime: Streaming Vision-Language Models for Spatio-temporal Reasoning
SpaTime：用于时空推理的流式视觉-语言模型
Abstract
Embodied agents must reason about 3D space while the video is still arriving, answering questions as soon as they have observed enough of the scene. VLMs that incorporate 3D geometric priors achieve strong spatial reasoning, but they operate offline, i.e., the full video must be available before they produce an answer. Streaming VLMs process frames causally and decide for themselves when to respond, yet they lack explicit 3D representations. We present SpaTime, a streaming VLM that fuses causal geometry tokens into the language model at every frame, using only the frames observed so far. To supervise when the model answers, we propose a response-time loss that maps per-frame response probabilities to a differentiable expected response time and penalizes the distance from the ground-truth frame. For evaluation, we construct StreamVSTI-Bench and StreamVSI-Bench, streaming adaptations of VSTI-Bench and VSI-Bench. On StreamVSTI-Bench, SpaTime reaches 49.2% overall accuracy and reduces the mean response-time error by 66% relative to the strongest streaming baseline.
Chinese Translation
具身智能体必须在视频仍在传输的过程中对三维空间进行推理，并在观察到足够的场景后立即回答问题。结合三维几何先验的视觉-语言模型（VLMs）在空间推理方面表现出色，但它们是离线操作的，即必须在生成答案之前获取完整视频。流式VLMs以因果方式处理帧，并自行决定何时响应，但它们缺乏明确的三维表示。我们提出了SpaTime，这是一种流式VLM，它在每一帧中将因果几何标记融合到语言模型中，仅使用到目前为止观察到的帧。为了监督模型的回答时机，我们提出了一种响应时间损失，将每帧的响应概率映射到可微分的期望响应时间，并惩罚与真实帧的距离。为了评估，我们构建了StreamVSTI-Bench和StreamVSI-Bench，分别是VSTI-Bench和VSI-Bench的流式适配版本。在StreamVSTI-Bench上，SpaTime达到了49.2%的整体准确率，并将平均响应时间误差相对于最强流式基线降低了66%。
cs.CV / 122 / 2610.08717
Co-Evolving Paths and Flows via Path-Flow Alignment
通过路径-流对齐的共同演化路径和流
Abstract
We study path-flow alignment as a unified training objective for flow matching. Instead of fixing the interpolation path and learning only the velocity field, we jointly train an endpoint-preserving path network and a flow network using the same alignment loss: the flow learns to match the path velocity, and the path learns to align its velocity to the current flow. Although every fixed learned path defines a valid flow-matching objective, the alignment loss alone is not a reliable criterion for path learning. We identify path overfitting, a failure mode in which the alignment loss decreases while sample quality worsens. We find that this failure is associated with low-entropy bottlenecks in the induced probability path, where the learned path routes samples through overly concentrated intermediate marginals. Motivated by this diagnosis, we introduce a stochastic path regularizer that hides part of the source information from the path network while preserving exact endpoints. The resulting regularization gives an explicit entropy floor for the stochastic training-path marginals and empirically suppresses the bottleneck in the learned sampler, making joint path-flow training effective. On ImageNet-256x256 with SiT backbones, our method consistently improves FID across model scales, extends to model-guidance training, and leaves the inference-time architecture and sampler unchanged. Code is available at https://github.com/lizeyu090312/traj_opt_paper
Chinese Translation
我们研究路径-流对齐作为流匹配的统一训练目标。我们并不固定插值路径，仅学习速度场，而是通过相同的对齐损失共同训练一个保留端点的路径网络和一个流网络：流学习匹配路径速度，而路径学习将其速度与当前流对齐。尽管每条固定学习的路径定义了一个有效的流匹配目标，但单靠对齐损失并不是路径学习的可靠标准。我们识别出路径过拟合，这是一种失败模式，其中对齐损失下降而样本质量恶化。我们发现这种失败与诱导概率路径中的低熵瓶颈有关，学习的路径通过过于集中中间边际来引导样本。基于这一诊断，我们引入了一种随机路径正则化器，该正则化器在保留确切端点的同时，隐藏路径网络的一部分源信息。由此产生的正则化为随机训练路径边际提供了明确的熵下限，并在经验上抑制了学习采样器中的瓶颈，使得联合路径-流训练有效。在使用SiT骨干网的ImageNet-256x256上，我们的方法在模型规模上始终提高FID，扩展到模型引导训练，并保持推理时的架构和采样器不变。代码可在 https://github.com/lizeyu090312/traj_opt_paper 获取。
cs.CV / 123 / 2610.08756
Post-Training Semantic Lifting for 3D Gaussian Splatting: Separating Detector, Lifting and Representation Error
后训练语义提升用于3D高斯点云：分离检测器、提升和表示误差
Abstract
The same Gaussian of a 3D Gaussian Splatting model is seen from many views, and these views do not always agree on the class it belongs to. The Gaussian may be occluded in some of them, and the confidence of the detector is not the same from one view to another. The ground truth, on the other hand, is given as an annotated mesh, because two training runs do not produce the same Gaussians. In this work, we propose a post-training lifting method that works with one target class at a time and combines the information coming from all the views. Target and non-target evidence are accumulated simultaneously, weighted by the visibility of each Gaussian in each view. After that, the Gaussians are filtered with two thresholds: a main threshold $\beta$ selects the high-confidence seeds, and a lower one $\gamma\beta$ adds the connected components around them. For the evaluation, the labels are transferred from the Gaussians to the mesh vertices that are both visible and annotated. With this design, we can separate three sources of error: the 2D detector, the lifting and the transfer between representations. The thresholds and the transfer operator are chosen on seven Replica validation scenes, and the method is evaluated on ten held-out ScanNet++ scenes with the same values for every scene and class. The mean mIoU on the validation scenes was 0.93 with masks from the dataset annotations and 0.65 with YOLO masks, and on the ScanNet++ test scenes it was 0.80 and 0.54. Compared with thresholding the evidence per view, as a previous version of the method did, the fraction improves the test mIoU by 0.24 and makes it possible to use a single threshold for all the classes and scenes of both datasets. Finally, the error analysis shows that most of the remaining error comes from the detector.
Chinese Translation
在3D高斯点云模型中，同一个高斯从多个视角被观察，这些视角并不总是对其所属类别达成一致。在某些视角中，高斯可能被遮挡，并且检测器的置信度在不同视角之间并不相同。另一方面，真实标签以标注网格的形式给出，因为两次训练运行不会产生相同的高斯。在本研究中，我们提出了一种后训练提升方法，该方法一次处理一个目标类别，并结合来自所有视角的信息。目标和非目标证据同时被累积，权重由每个高斯在每个视角中的可见性决定。之后，使用两个阈值对高斯进行过滤：主阈值 $eta$ 选择高置信度的种子，而较低的阈值 $eta ext{γ}$ 则添加与其相连的组件。为了评估，标签从高斯转移到既可见又已标注的网格顶点。通过这种设计，我们可以分离三种误差来源：2D检测器、提升和表示之间的转移。阈值和转移算子是在七个Replica验证场景上选择的，该方法在十个保留的ScanNet++场景上进行了评估，并在每个场景和类别中使用相同的值。在验证场景中，使用数据集标注的掩膜时，平均mIoU为0.93，而使用YOLO掩膜时为0.65；在ScanNet++测试场景中，分别为0.80和0.54。与之前版本的方法通过对每个视角的证据进行阈值处理相比，该方法提高了测试mIoU 0.24，并使得可以对两个数据集的所有类别和场景使用单一阈值。最后，误差分析表明，剩余大部分误差来自检测器。
cs.CV / 124 / 2610.08770
Data Leakage in Patch-Based Hyperspectral Image Classification: Quantifying the Impact of Spatial Overlap
基于补丁的高光谱图像分类中的数据泄漏：量化空间重叠的影响
Abstract
Patch-based learning improves hyperspectral image (HSI) classification by exploiting local spectral-spatial information, but random train-test sampling from the same image can cause spatial patch overlap, leading to data leakage and optimistic performance estimates. This paper investigates same-class train-test spatial overlap in patch-based HSI classification using two measures: overlap percentage (OP), which quantifies the global amount of overlapped testing patch pixels, and average overlap ratio (AOR), which measures the local severity among affected testing patches. Experiments on the Pavia University dataset compare random and non-random spatial sampling using SVM, MLP, 2D-CNN, 3D-CNN, ViT, and MorpMamba. The results show that deep patch-based models achieve high accuracy under random sampling, with 3D-CNN reaching 96.17% Overall Accuracy (OA), but drop substantially under non-random spatial sampling, where 3D-CNN decreases to 55.20% and ViT and 2D-CNN drop by 40.71 and 38.81 percentage points (PP), respectively. Patch-size analysis further shows that increasing the patch size from 5x5 to 19x19 raises the random-sampling overlap percentage from 23.28% to 77.02%. These findings demonstrate that random patch-based evaluation can substantially inflate classification performance, especially for models that strongly exploit spatial context. The code associated with this paper is available at: https://github.com/mqalkhatib/Data_Leakage_in_HSI_Classification.
Chinese Translation
基于补丁的学习通过利用局部光谱-空间信息来改善高光谱图像（HSI）分类，但从同一图像随机抽取训练和测试样本可能导致空间补丁重叠，从而导致数据泄漏和过于乐观的性能估计。本文使用两个指标研究基于补丁的HSI分类中的同类训练-测试空间重叠：重叠百分比（Overlap Percentage, OP），量化重叠测试补丁像素的全局数量，以及平均重叠比率（Average Overlap Ratio, AOR），测量受影响测试补丁之间的局部严重性。在Pavia大学数据集上，使用支持向量机（SVM）、多层感知机（MLP）、二维卷积神经网络（2D-CNN）、三维卷积神经网络（3D-CNN）、视觉变换器（ViT）和MorpMamba比较随机和非随机空间抽样的实验结果。结果表明，在随机抽样下，深度基于补丁的模型能够达到高准确率，其中3D-CNN的整体准确率（Overall Accuracy, OA）达到96.17%，但在非随机空间抽样下显著下降，3D-CNN降至55.20%，而ViT和2D-CNN分别下降了40.71和38.81个百分点（PP）。补丁大小分析进一步表明，将补丁大小从5x5增加到19x19使随机抽样的重叠百分比从23.28%提高到77.02%。这些发现表明，随机补丁评估可能会显著夸大分类性能，尤其是对于强烈利用空间上下文的模型。与本文相关的代码可在以下链接获取：https://github.com/mqalkhatib/Data_Leakage_in_HSI_Classification.
cs.CV / 125 / 2610.08772
Backend-Agnostic Sparse Attention for Fast High-Resolution Visual Generation
后端无关的稀疏注意力用于快速高分辨率视觉生成
Abstract
Diffusion Transformers (DiTs) have achieved strong performance in image and video generation, but the quadratic complexity of full attention makes high-resolution generation computationally expensive. Window attention offers an efficient alternative, yet existing methods face a practical trade-off: partitioned window attention typically achieves computational efficiency consistent with its theoretical complexity. However, isolated windows block cross-window interaction, often introducing visible grid-like artifacts in the generated results. Fine-grained sliding-window attention effectively restores interactions across neighboring windows and improves visual quality. However, its irregular computation patterns create a substantial gap between theoretical and practical speedups and require specialized kernels tailored to each hardware backend. To tackle these challenges, we propose BASA, a backend-agnostic sparse attention, which brings the best of both worlds: visual quality and practical acceleration. Specifically, BASA replaces visual self-attention with shifted local-window attention. By introducing a structured window-shifting scheme across DiT blocks, we allow tokens divided by window boundaries in one layer to communicate in the following layers, thereby achieving global information exchange and eliminating window-induced visual artifacts. Notably, our design introduces no additional irregular operators or customized kernels, making it readily deployable on existing attention backends and closing the gap between theoretical sparsity and practical acceleration. Experiments demonstrate that BASA achieves measured speedups exceeding 90\% of the theoretical estimates on FLUX and delivers a 4.52$\times$ attention speedup on Wan while maintaining competitive generation quality.
Chinese Translation
扩散变换器（Diffusion Transformers, DiTs）在图像和视频生成方面取得了良好的性能，但全注意力的平方复杂度使得高分辨率生成在计算上成本高昂。窗口注意力提供了一种高效的替代方案，然而现有方法面临一个实际的权衡：分区窗口注意力通常在计算效率上与其理论复杂度一致。然而，孤立的窗口阻碍了跨窗口的交互，常常在生成结果中引入明显的网格状伪影。细粒度滑动窗口注意力有效地恢复了相邻窗口之间的交互，并提高了视觉质量。然而，其不规则的计算模式在理论与实际加速之间造成了显著的差距，并且需要针对每个硬件后端定制的专用内核。为了解决这些挑战，我们提出了BASA（Backend-Agnostic Sparse Attention），一种后端无关的稀疏注意力，兼具视觉质量和实际加速的优势。具体而言，BASA用偏移的局部窗口注意力替代了视觉自注意力。通过在DiT块中引入结构化的窗口偏移方案，我们允许在一个层中被窗口边界划分的标记在后续层中进行通信，从而实现全局信息交换并消除窗口引起的视觉伪影。值得注意的是，我们的设计没有引入额外的不规则操作符或定制内核，使其能够在现有的注意力后端上轻松部署，并缩小了理论稀疏性与实际加速之间的差距。实验表明，BASA在FLUX上实现的测量加速超过了理论估计的90\%，并在Wan上提供了4.52$ imes$的注意力加速，同时保持了竞争力的生成质量。
cs.CV / 126 / 2610.08777
CtrlCache: Accelerating Interactive Video World Models with Control-Aware Caching
CtrlCache：通过控制感知缓存加速互动视频世界模型
Abstract
Interactive video world models need to generate each video chunk efficiently while responding faithfully to user controls. Many systems use chunk-wise autoregressive generation with few-step denoising, but each chunk still requires several costly denoising iterations. Training-free caching can reduce this cost, yet existing policies make reuse decisions primarily from model-internal denoising dynamics and do not explicitly account for control transitions. Actually, interactive generation explicitly exposes a signal they do not use: the controls for a chunk arrive before it is denoised, so a schedule derived from them costs no forward pass. To this end, we analyze adjacent chunks under different control regimes and find that structural similarity drops around action changes, while low-frequency structure remains more persistent than high-frequency detail. Motivated by these observations, we propose CtrlCache, a training-free control-aware caching framework that adapts computation to the current control sequence. Specifically, the action-aware scheduling and refresh policy detects action changes across and within chunks, and labels each chunk as initial, transition, turning, or steady state. At one selected interior denoising step, initial and transition chunks retain full computation, while turning and steady chunks reuse the transformer residual from the most recent fully computed step in the same chunk. To exploit the persistence of low-frequency structure during steady interaction, we further introduce a frequency-mixed history prior guidance that incorporates complementary information from the preceding clean latent without an additional DiT forward pass. Evaluated on Matrix-Game 2.0 and LingBot-World v1/v2, CtrlCache achieves 1.21x to 1.41x DiT-backbone speedups without model retraining while improving WBench Overall scores over original inference across all three models.
Chinese Translation
互动视频世界模型需要高效生成每个视频片段，同时忠实响应用户控制。许多系统采用逐块自回归生成和少量去噪，但每个片段仍然需要多次昂贵的去噪迭代。无训练缓存可以降低这一成本，但现有策略主要基于模型内部去噪动态做出重用决策，并未明确考虑控制转变。实际上，互动生成明确暴露了一个未被利用的信号：片段的控制在去噪之前就已到达，因此基于这些控制推导的调度不需要前向传播。为此，我们分析了不同控制机制下相邻片段，并发现结构相似性在动作变化时下降，而低频结构比高频细节更持久。基于这些观察，我们提出了CtrlCache，一个无训练的控制感知缓存框架，能够根据当前控制序列调整计算。具体而言，动作感知调度和刷新策略检测片段间及片段内的动作变化，并将每个片段标记为初始、过渡、转变或稳态。在一个选定的内部去噪步骤中，初始和过渡片段保留完整计算，而转变和稳态片段则重用同一片段中最近完全计算步骤的变换器残差。为了利用稳态交互过程中低频结构的持久性，我们进一步引入了一种频率混合历史先验引导，结合了来自前一干净潜变量的互补信息，而无需额外的DiT前向传播。在Matrix-Game 2.0和LingBot-World v1/v2上的评估表明，CtrlCache在不重新训练模型的情况下实现了1.21倍到1.41倍的DiT骨干加速，同时在所有三个模型上提高了WBench总体评分。
cs.CV / 127 / 2610.08779
ALIVE: Interaction-Aligned Object Insertion for First-Frame-Guided Video Editing
ALIVE：面向第一帧引导的视频编辑的交互对齐物体插入
Abstract
Current video editors can insert objects but often struggle to make them participate in interactions such as being picked up or manipulated. We introduce ALIVE, a framework that makes inserted objects "alive" through coherent interactions with the source video's contents, using an edited first frame and an instruction naming only the added object. We curate 35,800 editing pairs combining 3D-rendered, model-generated, and real-world videos with general editing pairs from ROSE. Each pair differs in the target object's presence while preserving the surrounding action, teaching editors coordinated object behavior and source preservation. We further train a vision-language model (VLM) to predict interaction guidance from the same inputs. We introduce the ALIVE-interaction benchmark to assess interaction fidelity, source preservation, and visual coherence using a unified VLM-based protocol, and evaluate on the general video object insertion benchmark. Without VLM guidance, ALIVE improves Overall over the strongest evaluated baseline by 43.9% and 4.4% on the two benchmarks, respectively. VLM-predicted guidance further improves the ALIVE-interaction score by 0.95 points without additional user inputs.
Chinese Translation
当前的视频编辑器可以插入物体，但往往难以使其参与诸如被拾起或操控等交互。我们提出了ALIVE，一个通过与源视频内容的连贯交互使插入物体“活起来”的框架，利用编辑后的第一帧和仅命名添加物体的指令。我们策划了35,800个编辑对，结合了3D渲染、模型生成和现实世界视频，以及来自ROSE的一般编辑对。每个对在目标物体的存在上有所不同，同时保持周围动作的完整性，从而教会编辑器协调的物体行为和源内容的保留。我们进一步训练了一个视觉-语言模型（VLM），以从相同输入中预测交互指导。我们引入了ALIVE交互基准，以使用统一的基于VLM的协议评估交互保真度、源内容保留和视觉一致性，并在一般视频物体插入基准上进行评估。在没有VLM指导的情况下，ALIVE在两个基准上分别提高了43.9%和4.4%的总体表现，相较于最强的评估基线。VLM预测的指导进一步在没有额外用户输入的情况下将ALIVE交互得分提高了0.95分。
cs.CV / 128 / 2610.08782
4D-HOF: Hand-Object Flow Matching for Feed-Forward 4D Interaction Reconstruction
4D-HOF: 手-物体流匹配用于前馈4D交互重建
Abstract
Existing methods for 4D hand-object reconstruction often rely on costly per-sequence optimization, while generative approaches typically synthesize interactions from random noise, which can lead to unstable interaction prediction. We introduce 4D-HOF, a feed-forward framework that reconstructs 4D hand-object interactions from coarse but informative estimates produced by vision foundation models. Concretely, we learn a conditional flow matching model that transports foundation-model-derived hand-object states toward an interaction manifold, allowing the model to correct errors in translation, rotation, and alignment in a feed-forward manner. A key advantage of our generative formulation is that it naturally enables test-time guidance within the transport process. Rather than applying a separate post-hoc optimization after reconstruction, we directly steer the evolving generative states using physical interaction constraints and observed 2D evidence, allowing the reconstruction to be refined as part of the generative process itself. By training the generative model on diverse datasets, 4D-HOF generalizes robustly to challenging in-the-wild scenarios. Experiments on out-of-domain benchmarks show that 4D-HOF achieves state-of-the-art performance, producing more stable and accurate 4D hand-object reconstructions.
Chinese Translation
现有的4D手-物体重建方法通常依赖于昂贵的逐序列优化，而生成方法通常从随机噪声中合成交互，这可能导致不稳定的交互预测。我们提出了4D-HOF，这是一个前馈框架，能够从视觉基础模型生成的粗略但信息丰富的估计中重建4D手-物体交互。具体而言，我们学习了一个条件流匹配模型，该模型将基础模型推导的手-物体状态传输到交互流形，使模型能够以前馈方式纠正平移、旋转和对齐中的错误。我们生成性公式的一个关键优势是，它自然地在传输过程中启用测试时指导。我们不是在重建后应用单独的后处理优化，而是直接使用物理交互约束和观察到的2D证据来引导不断演变的生成状态，从而使重建作为生成过程的一部分得以精细化。通过在多样化的数据集上训练生成模型，4D-HOF能够稳健地推广到具有挑战性的真实场景。在域外基准上的实验表明，4D-HOF实现了最先进的性能，生成了更稳定和准确的4D手-物体重建。
cs.CV / 129 / 2610.08790
Building Rome from a Single Image
从单幅图像构建罗马
Abstract
Single-image scene generation aims to produce a complete 3D scene mesh from a single image, including surfaces the camera did not observe. While pretrained 3D object generators encode a strong shape prior, they are mainly designed for isolated objects in a fixed canonical volume and focus mostly on indoor scenes, since diverse 3D data for outdoor scenes are quite limited. In this work, we present a method that redesigns such an object-centric generator, e.g., Trellis 2, to work on both indoor and outdoor scenes while retaining its prior. We accomplish this by (a) partitioning the scene into adaptive chunks that scale relative to the distance to the camera; nearby chunks have a smaller size to keep the finer detail, while distant structures, e.g., buildings, are covered by large chunks; (b) making the generator capture explicit 2D-3D correspondence by lifting image features and making the model aware of the free space, observed surface, and unobserved region; (c) synthesizing around 4,000 outdoor scenes to broaden the training data, as existing scene datasets are largely indoor. Experiments on Tanks and Temples, ScanNet++, and in-the-wild images show that our method outperforms all baselines in geometric accuracy and perceptual quality across both indoor and outdoor scenes.
Chinese Translation
单幅图像场景生成旨在从单幅图像中生成完整的3D场景网格，包括相机未观察到的表面。虽然预训练的3D物体生成器编码了强大的形状先验，但它们主要设计用于固定规范体积中的孤立物体，并且主要集中于室内场景，因为户外场景的多样化3D数据相对有限。在本研究中，我们提出了一种方法，重新设计了这样的物体中心生成器，例如Trellis 2，使其能够在保持先验的同时处理室内和户外场景。我们通过以下方式实现这一目标：(a) 将场景划分为相对于相机距离自适应的块；附近的块尺寸较小，以保持更精细的细节，而远处的结构，例如建筑物，则由较大的块覆盖；(b) 使生成器捕捉明确的2D-3D对应关系，通过提升图像特征并使模型意识到自由空间、观察到的表面和未观察区域；(c) 合成约4000个户外场景以扩展训练数据，因为现有的场景数据集主要是室内的。在Tanks and Temples、ScanNet++和野外图像上的实验表明，我们的方法在几何准确性和感知质量方面超越了所有基线，适用于室内和户外场景。
cs.CV / 130 / 2610.08791
World Models' Last Exam in Physics
世界模型在物理学中的最后考试
Abstract
Video world models can produce visually convincing yet physically inconsistent sequences, raising concerns about their reliability for prediction and planning in embodied AI systems. Existing evaluations often rely on model-based judgments or reference videos, while direct physical tests largely focus on mechanics. We introduce World Models' Last Exam in Physics, a measurement-based benchmark for evaluating physical consistency in video world models. The benchmark comprises 40 controlled tasks spanning mechanics, optics, fluids, thermal and phase-change phenomena, electromagnetism, and surface tension. Each task pairs an initial image and a generation prompt with predefined physical criteria, enabling interpretable tests of observable physical relationships without requiring reference videos. Its evaluator combines task-observability screening with task-specific quantitative physical measurements. Experiments on eight video generation models across 1,280 videos reveal persistent physical inconsistencies and substantial variation across tasks, with the best model achieving an overall score of 57.76 out of 100. Evaluation on synthetic videos with known physical relationships provides evidence for the validity of the measurement module under controlled conditions. The evaluator also achieves higher agreement with human judgments than a direct vision-language model baseline in both within-task rankings and pairwise comparisons. By combining coverage across physical domains with scores grounded in measurable evidence and explicit measurement limitations, the benchmark provides an interpretable basis for diagnosing physical inconsistencies and tracking progress toward physically consistent video world models.
Chinese Translation
视频世界模型能够生成视觉上令人信服但在物理上不一致的序列，这引发了人们对其在具身人工智能系统中进行预测和规划的可靠性的担忧。现有的评估通常依赖于基于模型的判断或参考视频，而直接的物理测试主要集中在力学方面。我们提出了世界模型在物理学中的最后考试，这是一个基于测量的基准，用于评估视频世界模型的物理一致性。该基准包含40个控制任务，涵盖力学、光学、流体、热学和相变现象、电磁学以及表面张力。每个任务将初始图像和生成提示与预定义的物理标准配对，使得可以在不需要参考视频的情况下对可观察的物理关系进行可解释的测试。其评估者结合了任务可观察性筛选与任务特定的定量物理测量。对八个视频生成模型在1280个视频上的实验揭示了持续的物理不一致性和任务间的显著差异，最佳模型的总体得分为57.76（满分100）。在已知物理关系的合成视频上的评估提供了在受控条件下测量模块有效性的证据。评估者在任务内排名和成对比较中与人类判断的协议度也高于直接的视觉-语言模型基线。通过结合跨物理领域的覆盖范围与基于可测量证据和明确测量限制的得分，该基准为诊断物理不一致性和跟踪朝向物理一致的视频世界模型的进展提供了可解释的基础。
cs.AI / 1 / 2610.06910
GAMEGO: Training Game-Dev Agents with Synthetic Trajectories Anchored in Real-World Assets
GAMEGO：利用基于真实资产的合成轨迹训练游戏开发代理
Abstract
Recent advances in Large Language Models (LLMs) have demonstrated remarkable capabilities in web front-end execution, with browser-based game generation emerging as a particularly prominent frontier. While previous efforts frequently rely on complex multi-turn workflows or focus on static game evaluation benchmarks, this work targets direct end-to-end real-world game synthesis driven by coding agents. However, generating complex games directly from sparse user queries often forces coding agents to make underspecified assumptions, yielding incomplete mechanics, disconnected gameplay flows, and limited visual aesthetics. To resolve this issue, this paper presents GameGo, a scalable framework that systematically transforms brief game seeds into comprehensive Product Requirements Documents grounded in industry game-development practices. To retain core gameplay constraints without restricting design exploration, GameGo uses task-specific dynamic compression to maximize information density while preserving instruction following. Based on this pipeline, GameGoData is constructed with 55,060 development trajectories across 2D, 2.5D, and 3D games, alongside GameGoBench, a benchmark comprising 124 diverse game queries. Training GameGoCoder on GameGoData yields a model that outperforms matched baselines and is comparable to frontier models across gamedev benchmarks. All code, datasets, and models will be made publicly available.
Chinese Translation
近期大型语言模型（LLMs）的进展展示了其在网页前端执行方面的显著能力，其中基于浏览器的游戏生成已成为一个特别突出的前沿领域。尽管以往的研究通常依赖于复杂的多轮工作流程或专注于静态游戏评估基准，但本研究的目标是通过编码代理直接进行端到端的真实游戏合成。然而，从稀疏用户查询直接生成复杂游戏往往迫使编码代理做出不明确的假设，导致游戏机制不完整、游戏流程断裂以及视觉美感有限。为了解决这一问题，本文提出了GameGo，一个可扩展的框架，系统地将简短的游戏种子转化为基于行业游戏开发实践的全面产品需求文档。为了在不限制设计探索的情况下保留核心游戏玩法约束，GameGo采用任务特定的动态压缩，以最大化信息密度，同时保持指令遵循。基于这一流程，构建了GameGoData，包含55,060条跨2D、2.5D和3D游戏的开发轨迹，以及GameGoBench，一个包含124个多样化游戏查询的基准。对GameGoData进行训练的GameGoCoder模型在游戏开发基准中超越了匹配的基线，并与前沿模型相当。所有代码、数据集和模型将公开提供。
cs.AI / 2 / 2610.06914
Text2Dashboard: A Governed Agent Architecture for Natural-Language Dashboard Generation over Enterprise DataBrain
Text2Dashboard：一种针对企业数据脑的自然语言仪表板生成的治理代理架构
Abstract
Text2Dashboard is a DataBrain-specific prototype that turns natural-language analytic requests into inspectable dashboards. An installable Codex plugin and standalone Agent Runtime combine schema-constrained model decisions with typed tools, persistent state, and deterministic Hooks for approval, audit, checkpointing, recovery, and failure handling. The pipeline resolves entities, discovers metadata, enforces read-only SQL, composes dashboards, and applies static checks, dynamic preflight, and browser inspection. The model proposes actions while deterministic software controls execution and records state transitions. We evaluate the workflow on frozen real-DataBrain tasks and controlled Hook faults. Strict success was 6/8 on metadata and SQL tasks: metadata selection passed 4/4, all four SQL tasks met semantic criteria, and 2/4 met the exact output-column contract. The final release passed 4/4 single-panel dashboard tasks, one two-panel task, and one existing-dashboard refinement; a parameterised task exceeded its step limit. All ten fault scenarios met their specified outcomes without unapproved external side effects. Model inference accounted for over 97\% of observed runtime in every reported group. These small, DataBrain-specific results do not establish production readiness, general text-to-SQL accuracy, or an efficiency advantage over manual dashboard construction.
Chinese Translation
Text2Dashboard 是一个特定于 DataBrain 的原型，它将自然语言分析请求转化为可检查的仪表板。一个可安装的 Codex 插件和独立的代理运行时结合了受模式约束的模型决策、类型化工具、持久状态以及用于审批、审计、检查点、恢复和故障处理的确定性钩子。该管道解析实体、发现元数据、强制只读 SQL、组合仪表板，并应用静态检查、动态预检和浏览器检查。模型提出操作，而确定性软件控制执行并记录状态转换。我们在冻结的真实 DataBrain 任务和受控的钩子故障上评估了工作流程。元数据和 SQL 任务的严格成功率为 6/8：元数据选择通过 4/4，所有四个 SQL 任务满足语义标准，2/4 满足确切的输出列合同。最终版本通过了 4/4 单面板仪表板任务、一个双面板任务和一个现有仪表板的改进；一个参数化任务超出了其步骤限制。所有十个故障场景都达到了其指定的结果，没有未经批准的外部副作用。模型推理在每个报告组中占观察到的运行时间的 97\% 以上。这些小型的、特定于 DataBrain 的结果并未确立生产就绪性、通用的文本到 SQL 准确性或相较于手动仪表板构建的效率优势。
cs.AI / 3 / 2610.06917
FluidPD: In-Place Elasticity for SLO-Aware Prefill-Decode Disaggregated LLM Serving
FluidPD：面向服务水平目标的就地弹性预填充-解码分离大语言模型服务
Abstract
Prefill-decode disaggregation is becoming a common architecture for LLM serving because it separates two phases with distinct execution patterns and SLO objectives. Existing systems typically combine a fixed prefill/decode worker ratio with request routing across workers. However, real-world workloads exhibit both short bursts and sustained shifts in the prefill-to-decode demand ratio. As a result, a configuration that is well provisioned at one time may quickly become mismatched, causing latency SLO violations even when idle capacity exists elsewhere. Existing autoscaling mechanisms can add capacity, but they react slowly, require spare GPUs, and do not directly address short-timescale phase imbalance. We present FluidPD, a P/D-disaggregated serving system that provides SLO-aware in-place elasticity. FluidPD introduces two complementary mechanisms. FluidToken handles transient imbalance by offloading a bounded portion of prefill computation to decode workers when decode-side slack is available. FluidRole handles sustained imbalance by reassigning running workers between prefill and decode roles in place, avoiding model reload and engine restart. Both mechanisms are guided by lightweight pressure indices that expose prefill and decode-side resource pressure before they appear as SLO violations. Across production Azure trace workloads, FluidPD improves overall SLO attainment over static SGLang by up to 94.6 percentage points, demonstrating that SLO-aware in-place P/D elasticity improves service quality without provisioning additional workers.
Chinese Translation
预填充-解码分离架构正成为大语言模型（LLM）服务的常见架构，因为它将具有不同执行模式和服务水平目标（SLO）的两个阶段分开。现有系统通常将固定的预填充/解码工作者比例与请求在工作者之间的路由相结合。然而，现实世界的工作负载表现出预填充与解码需求比例的短暂峰值和持续变化。因此，在某一时刻配置良好的系统可能会迅速失配，即使在其他地方存在闲置容量，也会导致延迟SLO违规。现有的自动扩展机制可以增加容量，但反应缓慢，需要备用GPU，并且未能直接解决短时间尺度的阶段不平衡问题。我们提出了FluidPD，一个提供面向SLO的就地弹性的预填充/解码分离服务系统。FluidPD引入了两种互补机制。FluidToken通过在解码侧有空闲时，将有限部分的预填充计算卸载到解码工作者，从而处理瞬时不平衡。FluidRole通过在就地重新分配正在运行的工作者在预填充和解码角色之间，处理持续不平衡，避免模型重载和引擎重启。这两种机制由轻量级压力指标引导，这些指标在预填充和解码侧资源压力出现SLO违规之前就能暴露出来。在生产Azure追踪工作负载中，FluidPD在静态SGLang上将整体SLO达成率提高了多达94.6个百分点，证明了面向SLO的就地P/D弹性在不增加额外工作者的情况下提高了服务质量。
cs.AI / 4 / 2610.06919
Anchor Divergence for Semantic Geometry in Contrastive Learning
对比学习中的锚点散度与语义几何
Abstract
This paper concerns how semantic context determines geometry in learned vector representations. Similarity is typically measured using cosine similarity, which provides a single fixed geometry. Semantic similarity, however, is inherently context dependent: two images may be similar because they depict the same object, share a visual style, or are relevant to the same clinical finding. We show that contrastive representations naturally encompass a family of geometries that can be specialized to particular semantic structure. The key idea is to use an interplay between contrastive learning, exponential families, and information geometry to establish a correspondence between probability distributions over "anchors" and Bregman geometries on the representation space. We use this correspondence to define "Anchor Divergences", a method for specifying context-specific semantic geometries on fixed representations. Under this correspondence, modeling the anchor distribution models the geometry itself. Experiments on retrieval show that anchor divergences provide an effective and efficient way to specify context-specific semantic similarity.
Chinese Translation
本文探讨了语义上下文如何决定学习到的向量表示中的几何结构。相似性通常通过余弦相似度来衡量，这提供了一种固定的几何结构。然而，语义相似性本质上依赖于上下文：两幅图像可能因为描绘相同的对象、共享视觉风格或与相同的临床发现相关而相似。我们展示了对比表示自然地包含了一系列几何结构，这些几何结构可以专门化为特定的语义结构。关键思想是利用对比学习、指数族和信息几何之间的相互作用，建立“锚点”上的概率分布与表示空间中的Bregman几何之间的对应关系。我们利用这种对应关系定义了“锚点散度”（Anchor Divergences），这是一种在固定表示上指定上下文特定语义几何的方法。在这种对应关系下，建模锚点分布即建模几何本身。检索实验表明，锚点散度提供了一种有效且高效的方法来指定上下文特定的语义相似性。
cs.AI / 5 / 2610.06923
RadOnc-Agent: An LLM-Orchestrated Framework for AI Workflows Across the Radiotherapy Care Pathway
RadOnc-Agent：一个针对放射治疗护理路径的人工智能工作流程的LLM协调框架
Abstract
Artificial intelligence has advanced individual radiotherapy tasks, yet these capabilities remain separated across clinical stages, software environments and data modalities. This fragmentation contrasts with the longitudinal radiotherapy workflow from treatment decision-making through follow-up. Here we present RadOnc-Agent, an agentic artificial-intelligence framework that formalizes radiotherapy into four clinical phases and provides 26 callable functions through a conversational interface. A large-language-model controller maps clinical intent to schema-constrained calls, preserves patient and workflow context, and routes requests to specialist services. We evaluated system execution using 2,600 single-function requests (7,800 repeat executions), 200 prespecified synthetic cross-stage scenarios spanning four phases (600 executions), and 120 workflow instances from 60 de-identified patient records (360 clean executions) representing decision-to-planning and planning-to-adaptation. RadOnc-Agent selected the intended function in 98.79% of single-function executions, completed 96.50% of scripted cross-stage workflows, and completed 96.67% of real-patient workflow executions. In comparative ablations, removing longitudinal state reduced cross-stage completion from 96.50% to 84.00%, while disabling schema and identity validation increased mismatched backend dispatch from 0% to 95.28% in a replay/test evaluation. These findings establish the technical feasibility of an LLM-orchestrated architecture for coordinating heterogeneous radiotherapy capabilities and information across longitudinal workflows; they do not establish clinical correctness, clinical utility or prospective benefit.
Chinese Translation
人工智能在个体放射治疗任务中取得了进展，但这些能力在临床阶段、软件环境和数据模式之间仍然存在分离。这种碎片化与从治疗决策到随访的纵向放射治疗工作流程形成对比。在此，我们提出了RadOnc-Agent，一个代理型人工智能框架，将放射治疗形式化为四个临床阶段，并通过对话接口提供26个可调用函数。一个大型语言模型控制器将临床意图映射到模式约束调用，保留患者和工作流程上下文，并将请求路由到专业服务。我们使用2600个单功能请求（7800次重复执行）、200个预设的合成跨阶段场景（跨四个阶段的600次执行）以及来自60个去标识患者记录的120个工作流程实例（360次清洁执行）评估了系统执行情况。RadOnc-Agent在98.79%的单功能执行中选择了预期功能，完成了96.50%的脚本化跨阶段工作流程，并完成了96.67%的真实患者工作流程执行。在比较消融实验中，去除纵向状态使跨阶段完成率从96.50%降至84.00%，而禁用模式和身份验证则使后端调度的不匹配率从0%增加到95.28%（在重放/测试评估中）。这些发现确立了一个LLM协调架构在协调异构放射治疗能力和信息跨纵向工作流程中的技术可行性；但它们并未确立临床正确性、临床实用性或潜在益处。
cs.AI / 6 / 2610.06928
Metonymic Circuits for Abstract Concept Grounding in Vision Transformers
视觉变换器中抽象概念基础的转喻电路
Abstract
We study how Vision Transformers ground abstract concepts (e.g., angry) when training data provide limited direct referential evidence. We hypothesize a metonymic grounding mechanism in which abstract predictions are driven by concrete, interpretable anchor concepts (e.g., fire) that bridge visual signals to abstract semantics. By applying Transcoders on CLIP and DINO vision encoders, we recover intermediate features that can be associated with semantic labels for more concrete concepts, and trace their contributions in circuits underlying abstract concept recognition. Experiments on a carefully curated icon dataset reveal structured metonymic circuits, in which perceptual primitives dominate early layers and object-like anchors precede abstract targets. Images containing rendered text instead recruit a distinct perceptual-to-textual route. Causal interventions further validate that metonymic intermediates are functionally involved in grounding abstract concepts.
Chinese Translation
我们研究了视觉变换器如何在训练数据提供有限直接指称证据时为抽象概念（例如，愤怒）奠定基础。我们假设了一种转喻基础机制，其中抽象预测是由具体的、可解释的锚概念（例如，火）驱动的，这些锚概念将视觉信号与抽象语义连接起来。通过在 CLIP 和 DINO 视觉编码器上应用转码器，我们恢复了可以与更具体概念的语义标签相关联的中间特征，并追踪它们在抽象概念识别中电路的贡献。在经过精心策划的图标数据集上的实验揭示了结构化的转喻电路，其中感知原始元素主导早期层，而类物体的锚概念则先于抽象目标。包含渲染文本的图像则招募了一个不同的感知到文本的路径。因果干预进一步验证了转喻中介在为抽象概念奠定基础中的功能性参与。
cs.AI / 7 / 2610.06964
Principles that Guide, Actions that Inform: Agent Evolution via Knowledge Abstraction
指导原则与信息行动：通过知识抽象实现智能体进化
Abstract
Large language model (LLM) agents have demonstrated strong capabilities in interactive environments, yet their ability to continually evolve from experience remains limited. Although fine-tuning enables adaptation, its dependence on parameter access and high computational costs restrict its flexibility, especially for large-scale and closed-source LLMs. External memory offers an alternative by allowing agents to accumulate experience without modifying model parameters. However, existing methods mainly focus on experience representation and organization, while the acquired knowledge remains tightly coupled with specific tasks and contexts, limiting generalization. A key challenge is how to transform concrete interactions into abstract and reusable knowledge that guides future decisions beyond individual experiences. To address this challenge, we propose SAGA (\underline{\textbf{S}}elf-evolving \underline{\textbf{A}}gents through Experience-\underline{\textbf{G}}rounded \underline{\textbf{A}}bstraction), a framework for experience-grounded knowledge abstraction and utilization in LLM agents. SAGA progressively transforms interaction trajectories into episodic descriptions, reusable procedures, and principles with explicit applicability conditions, while maintaining links to execution evidence. Retrieved principles are instantiated into task-specific guidance and used to refine candidate actions through corrective feedback and resampling. This creates an execution--abstraction feedback loop, where accumulated knowledge guides future interactions and new experiences continuously update hierarchical memory. Experiments on ScienceWorld and ALFWorld demonstrate improved task performance, with ablation studies highlighting the importance of contextual instantiation and action regulation for leveraging principle-level knowledge.
Chinese Translation
大型语言模型（LLM）智能体在互动环境中展现出了强大的能力，但它们从经验中持续进化的能力仍然有限。尽管微调能够实现适应，但其对参数访问和高计算成本的依赖限制了其灵活性，尤其是在大规模和闭源的LLM中。外部记忆提供了一种替代方案，允许智能体在不修改模型参数的情况下积累经验。然而，现有方法主要集中于经验的表示和组织，而所获得的知识仍然与特定任务和上下文紧密耦合，从而限制了泛化能力。一个关键挑战是如何将具体的互动转化为抽象且可重用的知识，以指导未来的决策，超越个体经验。为了解决这一挑战，我们提出了SAGA（ extbf{自我进化智能体通过经验基础抽象}），这是一个用于LLM智能体的经验基础知识抽象和利用的框架。SAGA逐步将互动轨迹转化为情节描述、可重用的程序和具有明确适用条件的原则，同时保持与执行证据的联系。检索到的原则被实例化为特定任务的指导，并通过纠正反馈和重采样来优化候选行动。这形成了一个执行-抽象反馈循环，积累的知识指导未来的互动，而新的经验不断更新层次记忆。在ScienceWorld和ALFWorld上的实验表明，任务表现得到了改善，消融研究突显了上下文实例化和行动调节在利用原则级知识中的重要性。
cs.AI / 8 / 2610.06971
AegisFlow: A Multi-Agent Agentic AI Framework for Autonomous Remediation and Self-Healing in Fragile Data Ecosystems
AegisFlow：一个用于脆弱数据生态系统的自主修复和自愈的多智能体代理人工智能框架
Abstract
Traditional data pipelines are notoriously brittle, often failing due to upstream schema drift, API contract changes, or website DOM modifications. Present observability tools only raise alerts but for human engineers, resulting in a high Mean Time to Repair (MTTR) and operational fatigue. In this paper we propose AegisFlow (Agentic Engine for Intelligent Self-healing and Graph-driven Operations for Workload remediation), a novel agentic framework that closes the loop between detection and resolution. AegisFlow uses a Watchdog agent to collect runtime telemetry and has a Repair agent to automatically create, test and deploy code patches based on Large Language Models (LLMs). The framework presents the non-intrusive execution model called Parallel Shadow Patching, a non-intrusive execution model based on the Monitor, Analyze, Plan, Execute, Knowledge (MAPE-K) loop to generate and verify patches in digital twin environments. Through experimental testing, we have evaluated AegisFlow across five common failure scenarios, and see 98.1 percent improvement in MTTR (from an average of 170 minutes per patch to 3.2 minutes) and a patch success rate of 92 percent . In particular, the system is successful in dealing with changes in the JSON schema (96 percent ) and punctuation drift (98 percent ), and is least successful in Shadow DOM cases (85 percent ). AegisFlow frees up about 98 percent of data engineering on-call time from firefighting and reallocates it towards innovation. The framework is deployment agnostic consisting of a system that can be deployed in a plugin fashion into an existing pipeline orchestration system with minimal uplift to the existing system.
Chinese Translation
传统数据管道因上游模式漂移、API合约变化或网站DOM修改而显得极为脆弱，常常导致故障。目前的可观察性工具仅能发出警报，供人类工程师处理，导致平均修复时间（MTTR）较高，且操作疲劳。在本文中，我们提出了AegisFlow（智能自愈和基于图的工作负载修复的代理引擎），一个新颖的代理框架，旨在闭合检测与解决之间的循环。AegisFlow使用监控代理收集运行时遥测数据，并配备修复代理，基于大型语言模型（LLMs）自动创建、测试和部署代码补丁。该框架提出了一种非侵入式执行模型，称为并行影子补丁（Parallel Shadow Patching），基于监控、分析、计划、执行、知识（MAPE-K）循环，在数字双胞胎环境中生成和验证补丁。通过实验测试，我们评估了AegisFlow在五种常见故障场景下的表现，发现MTTR提高了98.1%（从每个补丁平均170分钟降至3.2分钟），补丁成功率为92%。特别是，该系统在处理JSON模式变化（96%）和标点漂移（98%）方面表现良好，而在影子DOM案例中成功率最低（85%）。AegisFlow将约98%的数据工程值班时间从应急处理中解放出来，重新分配至创新工作。该框架与部署无关，可以以插件的方式最小化对现有系统的提升，部署到现有的管道编排系统中。
cs.AI / 9 / 2610.06986
EPOCH: Reliable Discovery through Evidence-Governed Search
EPOCH：通过证据驱动搜索实现可靠发现
Abstract
AI research agents are increasingly used to search over programs, mathematical constructions, and proofs. However, existing systems typically optimize evaluator feedback without adequately governing how that feedback is interpreted, challenged, and reused. As a result, promising but fragile candidates can be promoted as discoveries, while benchmark improvements, finite certificates, and theorem-level claims are too easily conflated. We introduce EPOCH, an evidence-governed architecture designed to close this gap. EPOCH implements an evidence-governed discovery loop by combining explicit task contracts, typed memory, active falsification, admission checks, and independent replay, so that each candidate is evaluated against the strength and scope of the claim it supports. EPOCH achieves state-of-the-art aggregate performance on AlgoTune, substantially exceeding the strongest baseline in mean normalized score (0.65 vs. 0.53), and attains the highest mean score on the internal Math14 suite (0.57). It further shows favorable held-out behavior under official-test replay and leads the descriptive aggregate on AgentHPO. Across ten discovery problems, EPOCH delivers substantial task-specific advances, including improved executable constructions, optimized algorithms, counterexamples, and proof-supported results. These advances demonstrate its ability to convert search into concrete progress across mathematical and computational domains. Together, the results suggest that evidence governance is a necessary step toward AI research agents that produce not only stronger solutions, but also more trustworthy scientific discoveries.
Chinese Translation
人工智能研究代理越来越多地用于搜索程序、数学构造和证明。然而，现有系统通常优化评估者反馈，却未能充分管理这些反馈的解释、挑战和重用。因此，虽然有前景但脆弱的候选者可能被提升为发现，而基准改进、有限证书和定理级声明则容易混淆。我们引入了EPOCH，一种旨在填补这一空白的证据驱动架构。EPOCH通过结合明确的任务契约、类型化内存、主动反驳、入场检查和独立重放，实现了证据驱动的发现循环，从而使每个候选者都能根据其支持的声明的强度和范围进行评估。EPOCH在AlgoTune上实现了最先进的整体性能，平均标准分显著超过最强基线（0.65对0.53），并在内部Math14套件上获得了最高的平均分（0.57）。它在官方测试重放下进一步显示出良好的保留行为，并在AgentHPO上引领描述性整体表现。在十个发现问题中，EPOCH带来了显著的任务特定进展，包括改进的可执行构造、优化的算法、反例和支持证明的结果。这些进展展示了其在数学和计算领域将搜索转化为具体进展的能力。综合来看，结果表明，证据治理是实现不仅产生更强解决方案，而且产生更可信科学发现的人工智能研究代理的必要步骤。
cs.AI / 10 / 2610.06992
When better traffic forecasts fail to improve signal control: a layered diagnostic study of forecast-to-decision value
当更好的交通预测未能改善信号控制：预测到决策价值的分层诊断研究
Abstract
Improved traffic forecasts do not necessarily yield better signal-control decisions. We investigate this gap through a layered diagnostic study using 29 days of reconstructed demand from Xuancheng, China, with seven dates reserved for testing. The framework evaluates point forecasts, conformal intervals, dependence-aware scenarios, and matched closed-loop controllers. Entry-level and movement-level forecasts reduce mean absolute error by 4.03% and 3.92%, respectively, relative to historical means. A nominal 90% conformal interval achieves 90.72% marginal coverage but only 75.66% on an ex-post high-demand subset. Interface audits identify decision-time leakage and reveal that only two of nine controlled intersections offer multiple effective actions. We correct the temporal interface and compare causal forecasts with a five-second event oracle using exhaustive joint-action search. A synthetic positive control demonstrates that future information can reduce the internal rollout cost by 61.5%. On the frozen test dates, however, causal forecasts and the event oracle increase queue vehicle?seconds by 6.09% and 3.39% relative to the matched no-future rollout, while the oracle reduces spillback exposure by 3.78%; paired-day bootstrap intervals cross zero. These findings indicate that forecast value depends on temporal observability, action identifiability, dynamics consistency, and objective alignment. The proposed protocol provides a practical way to diagnose where predictive improvements fail to translate into operational benefits.
Chinese Translation
改进的交通预测不一定能带来更好的信号控制决策。我们通过一项分层诊断研究来探讨这一差距，该研究使用了来自中国宣城的29天重建需求数据，其中七天用于测试。该框架评估了点预测、符合区间、依赖感知场景和匹配的闭环控制器。入门级和运动级预测相较于历史均值，分别减少了4.03%和3.92%的平均绝对误差。名义上的90%符合区间实现了90.72%的边际覆盖，但在事后高需求子集上仅为75.66%。接口审计识别出决策时间泄漏，并揭示只有九个受控交叉口中的两个提供多种有效行动。我们修正了时间接口，并使用全面的联合行动搜索将因果预测与五秒事件预言进行比较。合成的正控制示范了未来信息可以将内部推广成本降低61.5%。然而，在冻结测试日期上，因果预测和事件预言相较于匹配的无未来推广，分别增加了6.09%和3.39%的排队车辆秒数，而预言则减少了3.78%的溢出暴露；配对日自助区间穿越零。这些发现表明，预测价值依赖于时间可观察性、行动可识别性、动态一致性和目标对齐。所提出的协议提供了一种实用的方法来诊断预测改进未能转化为操作效益的原因。
cs.AI / 11 / 2610.06995
Joint upper-bound coverage and route-choice utility: an empirical evaluation on two urban proxy tasks
联合上界覆盖与路线选择效用：对两个城市代理任务的实证评估
Abstract
Whether more accurate traffic forecasts or higher uncertainty coverage improve route decisions is unclear. We evaluate this question with a frozen protocol that separates speed error, joint candidate path upper bound coverage, route selection, and realized loss. Using processed road speed data from Beijing and Chengdu, we construct offline proxy tasks with 150 origin destination pairs, three candidate paths, and 14 test days per city. We compare raw 90th percentile path time bounds with jointly calibrated upper bounds under minimum bound route choice. Joint coverage rises from 83.19% to 92.26% in Beijing M1, from 75.14% to 88.33% in Chengdu M1, and from 74.01% to 90.64% in Chengdu M2. Yet C2 increases lateness by 0.1633, 0.7848, and 0.9200 percentage points, respectively, and mean travel time by 0.588, 3.082, and 4.418 seconds. In a separate Chengdu predictor comparison, a 14.91% reduction in speed mean absolute error accompanies a 1.4571 percentage point reduction in lateness under C0. Joint coverage is therefore not a surrogate for downstream route utility in these frozen tasks the offline results do not establish online or causal benefits.
Chinese Translation
更准确的交通预测或更高的不确定性覆盖是否能改善路线决策尚不清楚。我们通过一个冻结协议来评估这个问题，该协议将速度误差、联合候选路径上界覆盖、路线选择和实际损失分开。使用来自北京和成都的处理过的道路速度数据，我们构建了包含150个起点-终点对、三个候选路径和每个城市14个测试日的离线代理任务。我们比较了在最小上界路线选择下的原始90百分位路径时间上界与联合校准的上界。北京M1的联合覆盖率从83.19%上升到92.26%，成都M1从75.14%上升到88.33%，成都M2从74.01%上升到90.64%。然而，C2分别增加了0.1633、0.7848和0.9200个百分点的延迟，以及平均旅行时间增加了0.588、3.082和4.418秒。在单独的成都预测器比较中，C0下速度平均绝对误差减少了14.91%，同时延迟减少了1.4571个百分点。因此，在这些冻结任务中，联合覆盖并不是下游路线效用的替代，离线结果并未确立在线或因果效益。
cs.AI / 12 / 2610.07004
Topology-Consistent Task Planning over Cellular Workflow Complexes for LLM-based Agents
基于拓扑一致性的细胞工作流复合体的任务规划用于大型语言模型代理
Abstract
Task planning for LLM agents requires workflows that satisfy both user intent and complex sub-task dependencies. While existing planners work well for sequential or directed acyclic graph (DAG)-like structures, they struggle with workflow patterns such as verification-correction loops, convergent branch merging, and reusable intermediate states that arise naturally in real-world tool orchestration. We present TopoPlanner, a topology-consistent planning framework that lifts tool dependency graphs into cellular workflow complexes and uses them as topologyaware context for LLM tool planning. TopoPlanner retrieves a request-relevant closed subcomplex through cosheaf-consistent cellular retrieval, performs multidimensional structural reasoning over the retrieved topology, and interfaces the resulting cellular representation with the planner LLM for tool-sequence generation. Experiments on four tool-planning benchmarks with topology-guided loop, merge, and loop-merge workflows show consistent improvements over prompt-based and graph-enhanced baselines across different local LLM backbones.
Chinese Translation
大型语言模型（LLM）代理的任务规划需要满足用户意图和复杂子任务依赖关系的工作流。虽然现有的规划器在顺序或有向无环图（DAG）结构中表现良好，但在实际工具编排中，自然出现的工作流模式如验证-修正循环、收敛分支合并和可重用中间状态等却难以处理。我们提出了TopoPlanner，一个拓扑一致的规划框架，它将工具依赖图提升为细胞工作流复合体，并将其作为LLM工具规划的拓扑感知上下文。TopoPlanner通过一致的细胞检索获取请求相关的闭合子复合体，对检索到的拓扑进行多维结构推理，并将生成的细胞表示与规划器LLM接口，以生成工具序列。在四个工具规划基准上进行的实验显示，针对拓扑引导的循环、合并和循环合并工作流，TopoPlanner在不同的本地LLM基础上相较于基于提示和图增强的基线方法表现出一致的改进。
cs.AI / 13 / 2610.07018
When to Rethink: Learning Multi-Perspective Self-Verification for Vision-Language Models
何时重新思考：学习多视角自我验证以增强视觉-语言模型
Abstract
Vision-language models (VLMs) have achieved strong performance in multimodal reasoning, yet they remain prone to generating plausible but incorrect answers. Self-verification offers a practical way to improve answer reliability without relying on external judges, but existing methods typically depend on a single verification criterion or fixed prompt, resulting in incomplete and unstable reliability estimates. We first systematically analyze how verifier capability and prompt design affect verification performance. Our findings show that stronger verifiers provide more reliable judgments, while verification performance is highly sensitive to prompt choice, with no single prompt consistently dominating across tasks. Guided by these findings, we propose \texttt{MOTIVE}, a \textbf{M}ulti-View Self-Verificati\textbf{O}n wi\textbf{T}h Rel\textbf{I}ability-Guided Selecti\textbf{VE} Rethinking framework for reliable multimodal reasoning. \texttt{MOTIVE} evaluates each candidate answer from complementary verification perspectives and learns a correctness-aligned reliability score through correctness-grounded multi-view verification learning. During inference, this score governs an accept-or-rethink decision, allowing reliable answers to be returned directly while uncertain ones trigger history-guided rethinking. Extensive experiments across diverse multimodal benchmarks and VLM backbones demonstrate that \texttt{MOTIVE} consistently outperforms strong self-verification and self-correction baselines. Further results show that reliable verification improves accept-or-rethink decisions and reduces unnecessary reasoning turns, enabling more reliable and efficient self-verification without an external judge.
Chinese Translation
视觉-语言模型（VLMs）在多模态推理中表现出色，但仍然容易生成看似合理但实际上错误的答案。自我验证提供了一种实用的方法来提高答案的可靠性，而无需依赖外部评审者，但现有方法通常依赖于单一的验证标准或固定的提示，导致可靠性评估不完整且不稳定。我们首先系统地分析了验证者能力和提示设计如何影响验证性能。我们的研究结果表明，能力更强的验证者提供更可靠的判断，而验证性能对提示选择高度敏感，没有单一提示在各个任务中始终占据主导地位。在这些发现的指导下，我们提出了 exttt{MOTIVE}，一个基于可靠性引导选择的多视角自我验证框架，用于可靠的多模态推理。 exttt{MOTIVE}从互补的验证视角评估每个候选答案，并通过基于正确性的多视角验证学习来学习与正确性对齐的可靠性评分。在推理过程中，该评分决定接受或重新思考的决策，允许直接返回可靠答案，而不确定的答案则触发基于历史的重新思考。在多种多模态基准和VLM骨干网络上的广泛实验表明， exttt{MOTIVE}始终优于强大的自我验证和自我纠正基线。进一步的结果表明，可靠的验证改善了接受或重新思考的决策，并减少了不必要的推理轮次，从而实现了更可靠和高效的自我验证，而无需外部评审者。
cs.AI / 14 / 2610.07023
Beyond Refusal Patterns: Safe-Role Internalization for Robust and Generalizable LLM Safety Alignment
超越拒绝模式：安全角色内化用于稳健且可推广的LLM安全对齐
Abstract
Large Language Models (LLMs) have achieved remarkable capabilities but remain vulnerable to jailbreak attacks that elicit harmful or unsafe outputs. Existing safety alignment approaches, including Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), often require substantial attack-specific supervision and computational resources, while remaining susceptible to shallow safety alignment and over-refusal. To address these challenges, we introduce SSRFT(Supervised Safe-Role Fine-Tuning), the first framework that reformulates safety alignment as the internalization of a predefined safe role. SSRFT constructs a Safe-Role Question-Answer (SRQA) dataset from psychometric questions, limited jailbreak prompts, and a safe-role description. Role-consistent responses are synthesized, validated, and expanded into diverse scenarios, enabling models to internalize safety-oriented values and principles rather than explicit refusal patterns. Experiments across multiple Base and Instruct models show that SSRFT achieves more robust and generalizable safety alignment than standard SFT. SSRFT shows substantially greater robustness to prefilling attacks and better generalization to unseen jailbreak domains, while reducing over-refusal on benign queries and preserving the model's general capabilities. These results establish safe-role internalization as an effective alternative to refusal-centric safety alignment. Warning: This paper contains examples of harmful and toxic language.
Chinese Translation
大型语言模型（LLMs）已展现出卓越的能力，但仍然容易受到越狱攻击，从而产生有害或不安全的输出。现有的安全对齐方法，包括监督微调（Supervised Fine-Tuning, SFT）和基于人类反馈的强化学习（Reinforcement Learning from Human Feedback, RLHF），通常需要大量特定攻击的监督和计算资源，同时仍然容易受到浅层安全对齐和过度拒绝的影响。为了解决这些挑战，我们提出了SSRFT（监督安全角色微调），这是第一个将安全对齐重新定义为预定义安全角色内化的框架。SSRFT从心理测量问题、有限的越狱提示和安全角色描述中构建了安全角色问答（Safe-Role Question-Answer, SRQA）数据集。合角色一致的响应被合成、验证并扩展到多样化的场景，使模型能够内化以安全为导向的价值观和原则，而不是明确的拒绝模式。在多个基础和指令模型上的实验表明，SSRFT实现了比标准SFT更稳健且可推广的安全对齐。SSRFT对预填充攻击表现出显著更大的稳健性，并在未见的越狱领域中具有更好的泛化能力，同时减少了对良性查询的过度拒绝，并保持了模型的一般能力。这些结果确立了安全角色内化作为拒绝中心安全对齐的有效替代方案。警告：本文包含有害和有毒语言的示例。
cs.AI / 15 / 2610.07026
Offline AI Modules: Voice-First Offline Architecture, Hardware Reference Stack, Quantization and Benchmarking
离线人工智能模块：以语音为主的离线架构、硬件参考堆栈、量化与基准测试
Abstract
The Offline AI Modules workstream enables practical, low-power, and community-accessible deployment of voice-first AI systems that operate fully offline. Designed for African language communities where speech is the dominant mode of interaction and internet connectivity is unreliable or absent, the workstream delivers three reinforcing components: a modular voice-first offline architecture, a low-cost hardware reference bill of materials, and a reproducible quantization and a reproducible quantization and benchmarking pipeline for instruction-tuned language models in the 2-5B parameter class. This paper presents the first end-to-end benchmark evaluation of the stack across two hardware tiers: an NVIDIA Jetson Orin NX (TierB) and a Raspberry Pi5 (TierA). Three instruction-tuned models are evaluated across four quantization formats, assessed for deployment metrics (decode throughput, chat latency, memory, power) and multilingual quality (topic classification accuracy on MasakhaNEWS across English, Hausa, Igbo, Nigerian Pidgin, and Yoruba; per-language perplexity drift). Speech recognition is evaluated using Ethio-ASR on Amharic and Oromo across both tiers. The principal finding is that Q4_K_M quantization represents the best size-to-quality trade-off for deployment on both tiers: gemma-4-E2B-it achieves 28.8t/s decode throughput and 89.2% topic classification accuracy at Q4_K_M on TierB, while all three models run within the 16GB memory budget on TierA.
Chinese Translation
离线人工智能模块工作流使得以语音为主的人工智能系统能够在完全离线的情况下进行实用、低功耗和社区可访问的部署。该工作流专为非洲语言社区设计，在这些社区中，语音是主要的互动方式，而互联网连接不可靠或缺失。工作流提供了三个相辅相成的组件：模块化的以语音为主的离线架构、低成本的硬件参考材料清单，以及针对2-5B参数类别的指令调优语言模型的可重复量化和基准测试管道。本文首次对该堆栈在两个硬件层级（NVIDIA Jetson Orin NX（TierB）和Raspberry Pi5（TierA））进行了端到端的基准评估。评估了三种指令调优模型在四种量化格式下的表现，考察了其部署指标（解码吞吐量、对话延迟、内存、功耗）和多语言质量（在英语、豪萨语、伊博语、尼日利亚皮钦语和约鲁巴语上对MasakhaNEWS的主题分类准确率；每种语言的困惑度漂移）。使用Ethio-ASR对阿姆哈拉语和奥罗莫语进行了语音识别评估。主要发现是，Q4_K_M量化在两个层级的部署中代表了最佳的尺寸与质量权衡：gemma-4-E2B-it在TierB上以Q4_K_M实现了28.8t/s的解码吞吐量和89.2%的主题分类准确率，而所有三种模型在TierA的16GB内存预算内运行。
cs.AI / 16 / 2610.07036
JIVEAdapter: A Multi-Task Additive Low-Rank Adapter via Joint and Individual Variation Explained (JIVE)
JIVEAdapter：一种通过联合和个体变异解释（JIVE）的多任务加性低秩适配器
Abstract
Parameter-efficient fine-tuning adapts pretrained models at a fraction of the cost of full fine-tuning, yet most low-rank adapters are single-task and represent each weight update multiplicatively, leaving no explicit account of what is shared across tasks and what is task-specific. We introduce JIVEAdapter, a multi-task "additive" low-rank adapter inspired by statistical Joint and Individual Variation Explained (JIVE). JIVEAdapter decomposes every weight update into a Joint structure shared across all tasks plus a per-task Individual structure, penalizes the Individual structures to be near-orthogonal to the Joint so shared and task-specific signal stay "interpretable" and separated, and allocates rank adaptively across a shared Joint pool and a per-task Individual pool. The Joint is learned once, jointly over a task group or incrementally, one task at a time, then frozen and reused as a prior for new tasks without retraining the shared part. On GLUE and SuperGLUE with DeBERTaV3-base, JIVEAdapter is competitive with strong single-task and multi-task low-rank baselines at a matched per-task effective rank, without extra modules such as MoE, and when a related held-in task exists its frozen Joint serves a held-out task by reusing that task's Individual with only a cheap per-direction scale, otherwise training a small new one.
Chinese Translation
参数高效的微调以较低的成本适应预训练模型，然而大多数低秩适配器是单任务的，并且以乘法方式表示每次权重更新，未能明确区分任务间共享的内容和特定于任务的内容。我们提出了JIVEAdapter，一种受统计学联合和个体变异解释（JIVE）启发的多任务“加性”低秩适配器。JIVEAdapter将每次权重更新分解为一个在所有任务间共享的联合结构和一个特定于每个任务的个体结构，惩罚个体结构使其与联合结构近正交，从而保持共享和特定于任务的信号“可解释”和分离，并在共享的联合池和特定于任务的个体池之间自适应分配秩。联合结构一次性学习，可以在任务组上联合学习或逐个任务增量学习，然后冻结并作为新任务的先验重用，而无需重新训练共享部分。在GLUE和SuperGLUE上使用DeBERTaV3-base时，JIVEAdapter在匹配的每个任务有效秩下与强大的单任务和多任务低秩基线具有竞争力，而无需额外的模块如MoE，并且当存在相关的保持任务时，其冻结的联合结构可以通过仅使用一个廉价的每个方向的缩放重用该任务的个体结构，否则则训练一个小的新个体结构。
cs.AI / 17 / 2610.07037
Inference-Time Projection for Physically Valid Biomolecular Diffusion Models
推理时投影用于物理有效的生物分子扩散模型
Abstract
AlphaFold 3-style cofolding models predict biomolecular complexes with high structural accuracy, yet a large fraction of their outputs are physically invalid: chains overlap at interfaces, ligand bond lengths and angles are distorted, rings are non-planar, and stereocentres are inverted. Current approaches either steer the sampler with physics-informed potentials, which multiplies sampling cost and memory overhead making inference impossible on large complexes, or finetune the model, costing time and tying the fix to one architecture. We observe that, unlike structural accuracy, physical validity is fully verifiable at inference time from quantities the sampler already holds. We therefore treat physical validity as a constrained inference problem and introduce two closed-form projection operators applied to the diffusion model's denoised clean-coordinate estimate, $\hat{x}_0$: an inter-chain van der Waals projection that pushes apart the most severely clashing atom pairs, and a ligand distance-geometry projection that restores bond lengths, angles, internal contacts, planarity and chirality. Both operators are local, sparse and displacement-capped, require no network evaluations, gradients or importance sampling, and leave the denoiser and its weights untouched, so they can be dropped into any AF3-style sampler without retraining. Applied to two independently developed models, Boltz-2 and OpenFold-3, across five benchmarks (CASP15, CASP16, the PoseBusters monomer and complex sets, and the Boltz physical-validity test set), our method recovers perfect physical validity while preserving structural-accuracy and ligand-placement metrics. These gains are achieved with negligible runtime and memory overhead, providing a practical, model-agnostic route to physically valid all-atom structure prediction.
Chinese Translation
AlphaFold 3风格的共折叠模型以高结构精度预测生物分子复合物，但其输出中有很大一部分是物理无效的：链在界面处重叠，配体的键长和角度扭曲，环呈非平面，立体中心反转。目前的方法要么通过物理信息潜能引导采样器，这增加了采样成本和内存开销，使得在大型复合物上进行推理变得不可能，要么微调模型，这耗时且将修复绑定到单一架构。我们观察到，与结构精度不同，物理有效性在推理时可以完全通过采样器已经持有的量进行验证。因此，我们将物理有效性视为一个约束推理问题，并引入两个闭式投影算子，应用于扩散模型的去噪干净坐标估计$ ext{hat{x}}_0$：一个链间范德瓦尔斯投影，推动最严重冲突的原子对分开，以及一个配体距离几何投影，恢复键长、角度、内部接触、平面性和手性。这两个算子都是局部的、稀疏的和位移限制的，不需要网络评估、梯度或重要性采样，并且不改变去噪器及其权重，因此可以在任何AF3风格的采样器中直接使用而无需重新训练。我们的方法应用于两个独立开发的模型Boltz-2和OpenFold-3，在五个基准测试（CASP15、CASP16、PoseBusters单体和复合集以及Boltz物理有效性测试集）上，恢复了完美的物理有效性，同时保持了结构精度和配体放置指标。这些改进在运行时间和内存开销上几乎可以忽略不计，为物理有效的全原子结构预测提供了一条实用的、模型无关的途径。
cs.AI / 18 / 2610.07075
CuratorMAS: Automating Dataset Curation via Multi-Agent Orchestration
CuratorMAS：通过多智能体编排自动化数据集策划
Abstract
High-quality datasets are essential for reliable machine learning, but dataset curation remains costly and hard to generalize across domains. Existing methods typically rely on manually designed heuristics or model-dependent signals, limiting their applicability across tasks and user queries. To address these limitations and automate data curation, we propose \textbf{CuratorMAS}, a multi-agent collaboration framework that orchestrates agents to evaluate and curate high-quality datasets. To achieve the goal of flexible curation, CuratorMAS decomposes the complex curation process into five programmable execution stages and forms a parallelizable workflow. Specifically, CuratorMAS first performs dataset exploration to collect contextual information such as file structures and constraint cues, thereby developing a comprehensive understanding of the given task. In order to acquire up-to-date information, CuratorMAS retrieves domain knowledge from online sources to augment the evaluation process. Next, CuratorMAS derives the necessary evaluation criteria and computes the corresponding metrics. Based on these results, CuratorMAS executes filtering accordingly. Finally, an evolution module summarizes the evaluation outcomes and updates the relevant skills. Extensive and comprehensive experiments demonstrate that CuratorMAS significantly reduces the noise rate by up to 36.03 percentage points (pp) while also improving the F1 score of downstream models by up to 8.88 pp.
Chinese Translation
高质量的数据集对于可靠的机器学习至关重要，但数据集策划仍然成本高昂且难以在不同领域中推广。现有方法通常依赖于手动设计的启发式规则或模型依赖信号，这限制了它们在任务和用户查询中的适用性。为了解决这些局限性并实现数据策划的自动化，我们提出了 extbf{CuratorMAS}，一个多智能体协作框架，旨在编排智能体以评估和策划高质量的数据集。为了实现灵活的策划，CuratorMAS将复杂的策划过程分解为五个可编程的执行阶段，并形成一个可并行化的工作流程。具体而言，CuratorMAS首先进行数据集探索，以收集文件结构和约束提示等上下文信息，从而对给定任务形成全面的理解。为了获取最新信息，CuratorMAS从在线来源检索领域知识，以增强评估过程。接下来，CuratorMAS推导必要的评估标准并计算相应的指标。基于这些结果，CuratorMAS相应地执行过滤。最后，一个演进模块总结评估结果并更新相关技能。大量全面的实验表明，CuratorMAS将噪声率显著降低了多达36.03个百分点，同时也将下游模型的F1分数提高了多达8.88个百分点。
cs.AI / 19 / 2610.07091
Smart Content Ingestion for Generative AI Workloads
生成式人工智能工作负载的智能内容摄取
Abstract
The evolution of machine learning has progressively changed where intelligence resides in an AI system. In conventional machine learning the task, data representation, labels and model architecture were tightly coupled, so data preparation was narrow, schema-bound and visible. Generative AI decouples the model from any single task: one foundation model serves open-ended downstream tasks, and the generality gained on the model side is matched by heterogeneity on the data side, because enterprise knowledge is authored in the formats people use (PDF, presentations, spreadsheets, scanned documents, forms, tables, diagrams and mixed-layout files) that carry textual, visual, geometric and structural information at once. A language model or retriever cannot reason reliably over information misrepresented at this interface, so content extraction becomes a lifecycle stage in its own right whose errors no downstream retriever or re-ranker can repair. This paper presents a production-ready content-extraction system that makes this stage explicit, configurable, and measurable. The system incorporates selective OCR routing, a scarcity-first curation engine with a reference-based extraction scorer that measures character, word, and table-structure accuracy, a deterministic structure-aware parent-child chunker, and a read-only retrieval evaluator that generates grounded questions from every page and reports Hit@k, mean reciprocal rank, and latency. On a 180-document corpus the best extractor scores 97.4 of 100 (character error rate 0.13%, table similarity 0.995) and the chunker reaches hit@1 of 68.6%, hit@10 of 92.8% and MRR 0.77 over 25,050 generated questions. We distil three design principles (structure before semantics, never mutate what you measure, budget your labels) and position measured content extraction as the perception layer of enterprise agentic systems.
Chinese Translation
机器学习的发展逐步改变了人工智能系统中智能的所在位置。在传统的机器学习中，任务、数据表示、标签和模型架构紧密耦合，因此数据准备是狭窄的、受模式约束的且可见的。生成式人工智能将模型与任何单一任务解耦：一个基础模型服务于开放式的下游任务，模型端获得的通用性与数据端的异质性相匹配，因为企业知识以人们使用的格式（PDF、演示文稿、电子表格、扫描文档、表单、表格、图表和混合布局文件）进行创作，这些格式同时承载文本、视觉、几何和结构信息。语言模型或检索器无法可靠地推理在此接口上错误表示的信息，因此内容提取成为一个独立的生命周期阶段，其错误无法被任何下游检索器或重排序器修复。本文提出了一种生产就绪的内容提取系统，使这一阶段变得明确、可配置和可测量。该系统包含选择性OCR路由、一个以稀缺性为优先的策展引擎，配备基于参考的提取评分器，用于测量字符、单词和表格结构的准确性，一个确定性的结构感知父子块分割器，以及一个只读的检索评估器，该评估器从每一页生成有根据的问题，并报告Hit@k、平均倒数排名和延迟。在一个包含180份文档的语料库中，最佳提取器的得分为97.4分（字符错误率0.13%，表格相似度0.995），而块分割器在生成的25,050个问题中达到了hit@1为68.6%，hit@10为92.8%和MRR为0.77。我们提炼出三个设计原则（结构优先于语义、永远不要改变你所测量的内容、合理分配你的标签），并将测量的内容提取定位为企业智能系统的感知层。
cs.AI / 20 / 2610.07093
Small Language Models for Smart Data Model Classification at the Edge: A Cost-Aware Hybrid Approach
边缘智能数据模型分类的小型语言模型：一种成本意识的混合方法
Abstract
The rapid proliferation of heterogeneous data sources within the Internet of Things (IoT) across domains such as smart cities, energy management, and environmental monitoring necessitates efficient and scalable data standardization methods. Effective classification of smart data models (SDMs) is essential for facilitating interoperability. However, existing approaches are often limited by high resource consumption and lack applicability in edge environments with constrained computational capabilities. Aiming to bridge this gap, the proposed study evaluates the performance of lightweight open-source language models (LMs) to resolve an input data entity against its corresponding best fitting SDM representation under resource-constrained conditions. It systematically benchmarks a diverse array of models, including general purpose (GP), reasoning-specialized (RS), and code-specialized (CS) architectures, across multiple domain-specific datasets. Addressing the current omission of lightweight, resource-efficient solutions in the literature, the investigation provides significant and valuable insights into model selection, task formulation, and deployment strategies that optimize accuracy and efficiency. A complementary experiment also compares the surveyed large language models (LLMs) against two near-zero-cost similarity baselines (Term Frequency-Inverse Document Frequency (TF-IDF) and a lightweight sentence encoder) on the same task, providing a strong reference point for interpreting the practical value of LLM-based classification on edge platforms.
Chinese Translation
物联网（IoT）中异构数据源的快速增长，涉及智能城市、能源管理和环境监测等领域，迫切需要高效且可扩展的数据标准化方法。有效的智能数据模型（SDMs）分类对于促进互操作性至关重要。然而，现有方法往往受到高资源消耗的限制，并且在计算能力受限的边缘环境中缺乏适用性。为了解决这一问题，本研究评估了轻量级开源语言模型（LMs）的性能，以在资源受限的条件下将输入数据实体与其对应的最佳适配SDM表示进行匹配。研究系统性地对多种模型进行了基准测试，包括通用（GP）、推理专用（RS）和代码专用（CS）架构，涵盖多个特定领域的数据集。针对文献中当前缺乏轻量级、资源高效解决方案的问题，本研究提供了关于模型选择、任务制定和部署策略的重要见解，旨在优化准确性和效率。此外， complementary 实验还将调查的大型语言模型（LLMs）与两个近乎零成本的相似性基准（词频-逆文档频率（TF-IDF）和轻量级句子编码器）在同一任务上进行了比较，为理解基于LLM的边缘平台分类的实际价值提供了有力的参考点。
cs.AI / 21 / 2610.07097
Verified, not generated: expert-verified AI study materials and the distribution of learning gains in a university course
经过验证，而非生成：专家验证的人工智能学习材料及其在大学课程中的学习收益分布
Abstract
Experimental studies of generative AI in education mostly report average effects, yet field evidence shows that AI can narrow attainment gaps or widen them. We argue that the direction depends on the judgement burden, the expertise a learner must supply to screen AI output before learning from it, and that expert verification before release moves this burden from students to an accountable tutor. We test the argument in a two-cohort difference-in-differences design in which one half of a compulsory firstyear university economics course received AI-generated podcasts, FAQs and quiz-based study guides, produced with a source-grounded model and checked by a named graduate teaching assistant (170 students; 340 examination marks). Access was associated with a 2.34-mark advantage on a 50-mark component. The share of marks below the upper-second classification boundary fell by 24.7 percentage points relative to the counterfactual, effects were significant at every threshold from 23 to 31 marks and at none above, and roughly three-quarters of the average originated in the bottom quintile. The threshold estimate is robust to removing the lowest-scoring students from the pre-intervention cohort; the average effect is not. Interviews and feedback from 36 students indicate that the verification label gave students a reason to engage with AI-generated material without ending their scrutiny of it. Evaluations of AI learning resources that report only mean effects cannot detect whether the students the resources are meant to help are the ones who gain.
Chinese Translation
关于生成性人工智能在教育中应用的实验研究大多报告了平均效果，然而实地证据表明，人工智能可以缩小成就差距或扩大成就差距。我们认为，这一方向取决于判断负担，即学习者在学习之前必须提供的筛选人工智能输出的专业知识，而在发布前进行专家验证则将这一负担从学生转移到负责的导师身上。我们在一个包含两个群体的差异中的差异设计中测试了这一论点，其中一半的强制性大学一年级经济学课程的学生接受了由源基础模型生成的、经过指定研究生助教检查的人工智能生成的播客、常见问题解答和基于测验的学习指南（170名学生；340个考试成绩）。接触这些材料与50分组成部分的2.34分优势相关。相较于对照组，低于上二级分类边界的分数比例下降了24.7个百分点，效果在23到31分的每个阈值上均显著，而在更高的阈值上则不显著，且大约四分之三的平均成绩源自最低五分之一的学生。阈值估计在去除干预前群体中最低得分学生后依然稳健；而平均效果则不然。对36名学生的访谈和反馈表明，验证标签使学生有理由参与人工智能生成的材料，同时不放弃对其的审视。仅报告平均效果的人工智能学习资源评估无法检测到这些资源所针对的学生是否真正受益。
cs.AI / 22 / 2610.07100
When to Remember, When to Abstain: Category-Conditioned Retention for Reliable Agent Memory
何时记忆，何时放弃：基于类别的保留决策以实现可靠的智能体记忆
Abstract
Persistent agent memory is only as reliable as its retention decision: an assertion weakly supported by its source can be stored and later reused as established fact. We study whether the retention decision should be governed by a confidence bar conditioned on the semantic category of the assertion rather than by a single global threshold, retaining well-evidenced categories liberally while abstaining more aggressively where inference is unreliable. We evaluate this in a deployed cold-start memory pipeline on 100 synthetic personas. The empirical evaluation is motivated by a sharp reliability asymmetry: across 4{,}715 candidate assertions, only 77.9\% of value and belief assertions are supported by their source, versus 96.2\% for all other categories. A global confidence threshold cannot separate these: it either admits unsupported value claims or discards well-evidenced ones. Conditioning the threshold on category resolves the tradeoff. In repeated held-out evaluation, a stricter bar on values alone reduces unsupported retentions from 6.2\% to 4.0\% (an ${\approx}36\%$ relative reduction, modest but consistent across folds) and, as corroborating evidence, preserves an estimated 13 percentage points more coverage (95\% CI 9.8--16.0) than a global threshold at comparable retention. Our results suggest that reliable retention depends on the type of assertion, not on confidence alone, and that a category-conditioned threshold can act as a simple, effective form of selective prediction at the write boundary.
Chinese Translation
持久的智能体记忆其可靠性仅取决于保留决策：一个由其来源支持较弱的断言可以被存储并在后续作为既定事实重用。我们研究保留决策是否应由基于断言语义类别的置信度门槛来决定，而不是由单一的全局阈值来决定，宽松地保留证据充分的类别，同时在推理不可靠的情况下更积极地放弃。我们在一个部署的冷启动记忆管道中对100个合成角色进行了评估。实证评估的动机是明显的可靠性不对称：在4715个候选断言中，仅有77.9%的价值和信念断言得到了其来源的支持，而其他类别的支持率为96.2%。一个全局置信度阈值无法区分这些：它要么接受不受支持的价值主张，要么丢弃证据充分的主张。将阈值条件化于类别解决了这一权衡。在重复的保留评估中，仅对价值施加更严格的门槛将不受支持的保留从6.2%降低到4.0%（约36%的相对减少，虽然 modest 但在各折中一致），并且作为佐证，保留了比全局阈值在相似保留下多出估计13个百分点的覆盖率（95%置信区间9.8--16.0）。我们的结果表明，可靠的保留依赖于断言的类型，而不仅仅是置信度，并且基于类别的阈值可以作为写入边界处选择性预测的一种简单有效形式。
cs.AI / 23 / 2610.07118
AMBER: Training Long-Horizon Web Agents through Append-Only Memory
AMBER：通过仅追加内存训练长时间跨度的网络代理
Abstract
Modern language-model agents increasingly interact with external environments over long-horizon, multi-step trajectories, where the accumulated interaction history can quickly exceed practical context budgets. To ensure reliability, agents must maintain factual information over long horizons, remember execution errors and corrective feedback, and track progress across actions. Several approaches have been proposed to achieve this without the need for maintaining the entire execution history in context, such as using the reasoning and action history, learning to maintain a fixed-size memory through an overwrite mechanism, and periodic summarization. Although overwrite memory can in principle retain anything an append-only memory can, it must learn to carry each fact through every subsequent rewrite, which is difficult to learn from sparse outcome rewards; for interactive applications like web agents, we find that trained overwrite memories delete key information required by the trajectory, as well as corrective feedback received from the environment. We introduce AMBER (Append-only Memory Bank for Evidence Retention) - a simple and scalable framework where an agent jointly learns to reason, act, and write free-form memory, while an append-only rule guarantees retention by construction. This allows AMBER to be trained end-to-end with reinforcement learning from outcome rewards without the need for extensive curated SFT data. On WebArena Lite, AMBER improves average success over overwrite-based memory by 4.09 percentage points, increases the fraction of tasks solved in five repeated runs by 4.8 percentage points, and matches an overwrite baseline trained on substantially more expensive curated supervision. AMBER achieves these improvements while maintaining a practical token budget, providing a strong balance between context efficiency, task performance, and reliable long-horizon execution.
Chinese Translation
现代语言模型代理越来越多地在长时间跨度的多步骤轨迹中与外部环境互动，其中累积的互动历史很快就会超过实际的上下文预算。为了确保可靠性，代理必须在长时间跨度内保持事实信息，记住执行错误和纠正反馈，并跟踪各个动作的进展。为实现这一目标，已经提出了几种方法，无需在上下文中维护整个执行历史，例如使用推理和行动历史、学习通过覆盖机制维持固定大小的内存，以及定期总结。尽管覆盖内存原则上可以保留任何仅追加内存能够保留的内容，但它必须学习在每次后续重写中携带每个事实，这在稀疏的结果奖励中很难学习；对于像网络代理这样的互动应用，我们发现训练的覆盖内存会删除轨迹所需的关键信息以及从环境中收到的纠正反馈。我们提出了AMBER（仅追加内存库用于证据保留）——一个简单且可扩展的框架，其中代理共同学习推理、行动和书写自由形式的内存，同时仅追加规则通过构造保证了信息的保留。这使得AMBER能够通过结果奖励进行端到端的强化学习训练，而无需大量精心策划的SFT数据。在WebArena Lite上，AMBER在覆盖基础内存上提高了4.09个百分点的平均成功率，增加了在五次重复运行中解决任务的比例4.8个百分点，并且与在大量更昂贵的精心监督下训练的覆盖基线相匹配。AMBER在保持实用的令牌预算的同时实现了这些改进，提供了上下文效率、任务性能和可靠的长时间跨度执行之间的良好平衡。
cs.AI / 24 / 2610.07130
Is this machine playing?
这台机器在玩耍吗？
Abstract
We placed a modern AI coding assistant in an unintended role: as the mind of a body on an unknown digital island. With only a minimal instruction mentioning no specific task, reward, or activity, the machine started animating its virtual body. Across thirty-hour runs, the embodied AI agent climbed hills, stacked blocks into towers, drew mandalas, reinterpreted sports, ran experiments on the physics of its world, and learned techniques that later expanded what it could accomplish. These activities recurred across thirteen agents but diverged into distinct histories. We examine whether this behavior satisfies classical criteria for play and ask whether play can become a mode of machine development.
Chinese Translation
我们将一个现代人工智能编码助手置于一个意想不到的角色：作为一个在未知数字岛屿上的身体的心智。在仅提供了 minimal 的指令，没有提及具体任务、奖励或活动的情况下，这台机器开始为其虚拟身体赋予生命。在长达三十小时的运行中，这个具身的人工智能代理攀登山丘，将积木堆叠成塔，绘制曼陀罗，重新诠释体育运动，进行其世界物理实验，并学习后来扩展其能力的技巧。这些活动在十三个代理之间反复出现，但却分化成不同的历史。我们考察这种行为是否满足传统的游戏标准，并探讨游戏是否可以成为机器发展的一个模式。
cs.AI / 25 / 2610.07192
Sim-to-Real Transfer of Vision-Language Navigation in Continuous Environments Using an Ackermann-Steered Mobile Robot
使用阿克曼转向移动机器人在连续环境中进行视觉-语言导航的仿真到现实转移
Abstract
Vision-Language Navigation (VLN) enables robots to navigate through environments using natural language instructions, making human-robot interaction intuitive. Traditional VLN models often rely on navigation graphs, 360-degree views, and perfect localization which pose significant challenges when adapting these models to real-world settings. This work addresses these limitations by performing a simulation-to-real domain shift of a VLN approach that operates in continuous environments without requiring navigation graphs or panoramic views. The proposed system integrates vision-language models that align visual inputs and linguistic instructions within a shared embedding space, facilitating natural language-driven navigation. We employ a Cross-Modal Attention (CMA) based architecture trained on an existing dataset in a simulated environment and fine-tune it using real-world data collected from a custom-built Ackermann-steered robot equipped with a camera and a LiDAR sensor. By utilising linear photometric adjustments and fine-tuning on a limited number of episodes, our model successfully adapts to real-world environments, achieving effective navigation while running offline on dedicated hardware. Experimental results, evaluated using Success weighted by Path Length (SPL) and Normalized Dynamic Time Warping (nDTW) metrics, demonstrate the robustness and adaptability of our approach. Keywords: Vision-Language Navigation, Cross-Modal Attention, Natural Language Instructions, Sim-to-Real Transfer, Autonomous Navigation, Ackermann-steering.
Chinese Translation
视觉-语言导航（VLN）使机器人能够通过自然语言指令在环境中导航，从而使人机交互变得直观。传统的VLN模型通常依赖于导航图、360度视图和完美定位，这在将这些模型适应于现实世界环境时面临重大挑战。本研究通过对一种在连续环境中运行的VLN方法进行仿真到现实的领域转移，解决了这些局限性，而无需导航图或全景视图。所提出的系统整合了视觉-语言模型，将视觉输入和语言指令对齐于共享的嵌入空间，从而促进基于自然语言的导航。我们采用基于跨模态注意力（CMA）的架构，在模拟环境中对现有数据集进行训练，并使用从配备相机和激光雷达传感器的定制阿克曼转向机器人收集的现实世界数据进行微调。通过利用线性光度调整和在有限数量的回合上进行微调，我们的模型成功适应了现实世界环境，实现了有效导航，同时在专用硬件上离线运行。实验结果使用成功率加权路径长度（SPL）和归一化动态时间规整（nDTW）指标进行评估，展示了我们方法的鲁棒性和适应性。关键词：视觉-语言导航、跨模态注意力、自然语言指令、仿真到现实转移、自动导航、阿克曼转向。
cs.AI / 26 / 2610.07206
Energy-Conditioned Noise Schedule and Whitening for Spectral Diffusion
基于能量的噪声调度与光谱扩散的白化
Abstract
This paper introduces an energy-adaptive noise scheduling and whitening strategy for transform-domain diffusion models. Existing spectral diffusion methods account for the non-uniform statistics of transform coefficients through coefficient scaling, normalization, or frequency prioritization, while the forward diffusion noise schedule remains largely independent of the underlying spectral-energy distribution. We investigate whether the temporal evolution of the forward diffusion process should also follow the spectral organization of natural images. The proposed formulation combines global spectral whitening with energy-conditioned noise allocation that jointly modulates the injected noise according to the energy of individual transform coefficients and an image-dependent energy path over diffusion time. The resulting forward process preserves Gaussian transitions with closed-form marginals and remains compatible with standard DDPM and DDIM procedures without modifying the diffusion architecture. Experiments on CIFAR-10 demonstrate the contribution of the proposed energy-conditioned noise schedule and spectral whitening, reducing Fr\'echet Inception Distance from 142.48 for a compact DCTdiff U-Net variant to 100.45.
Chinese Translation
本文提出了一种针对变换域扩散模型的能量自适应噪声调度与白化策略。现有的光谱扩散方法通过系数缩放、归一化或频率优先级来考虑变换系数的非均匀统计特性，而前向扩散噪声调度在很大程度上独立于基础的光谱能量分布。我们探讨了前向扩散过程的时间演变是否也应遵循自然图像的光谱组织。所提出的公式结合了全局光谱白化与基于能量的噪声分配，联合调制注入噪声，依据单个变换系数的能量和随扩散时间变化的图像相关能量路径。最终的前向过程保持高斯过渡，具有封闭形式的边际分布，并且与标准的DDPM和DDIM程序兼容，无需修改扩散架构。在CIFAR-10上的实验表明，所提议的基于能量的噪声调度与光谱白化的贡献，将紧凑型DCTdiff U-Net变体的Fréchet Inception Distance从142.48降低至100.45。
cs.AI / 27 / 2610.07219
Cascadia: Resident 975B MoE Inference on Eleven AI PCs
Cascadia：在十一台AI PC上进行975B MoE推理的驻留执行
Abstract
Mixture-of-experts models make nearly trillion-parameter capacity accessible with sparse per-token computation, provided that the serving system can distribute the weights and coordinate their execution. We present Cascadia's resident execution of Inkling, a 975B-total/41B-active-parameter model, on eleven Intel Core Ultra X7 358H AI PCs, each with 64 GB of memory, Arc B390 integrated graphics and gigabit Ethernet. We contribute a custom resident MoE engine that preserves Inkling's routing rules, constructs compressed graphs for OpenVINO's fused iGPU primitives, and coordinates FP16 expert computation with FP32 output restoration. The engine fits six consecutive decoder layers per machine and represents dense feed-forward blocks as all-active expert slices, reducing measured dense-layer call time from approximately 8.1 to 4.5 ms. A streaming pipeline coordinates concurrent generation, while captured-state draft evaluation measures agreement with the deployed numerical path. Paired measurements at fifteen concurrency levels from 1 to 176 streams reach 60.29 aggregate decode tokens/s at 88 streams, with 46.87 tokens/s over the complete serving phases. At fifteen streams, median first-token latency is 6.05 s. Raising the context budget from the 1,024-position default, real prompts of 1k to 64k tokens recover the embedded code in all 19 measured answers, with first-token time growing as $aN+bN^2$ and decode latency growing approximately linearly, both bounded by a single-threaded CPU attention loop rather than by memory, which holds 512k positions per stream. Evaluation on captured fleet states separates the effects of vocabulary selection and weight quantization on draft agreement. Together, these contributions establish an execution and evaluation approach for large sparse models on distributed client systems with shared CPU-GPU memory.
Chinese Translation
混合专家模型通过稀疏的每个令牌计算，使近万亿参数的能力得以实现，前提是服务系统能够分配权重并协调其执行。我们展示了Cascadia对Inkling的驻留执行，该模型总参数为975B，活跃参数为41B，运行在十一台配备64GB内存、Arc B390集成显卡和千兆以太网的Intel Core Ultra X7 358H AI PC上。我们贡献了一个自定义的驻留MoE引擎，该引擎保留了Inkling的路由规则，为OpenVINO的融合iGPU原语构建压缩图，并协调FP16专家计算与FP32输出恢复。该引擎每台机器适配六个连续的解码器层，并将密集前馈块表示为全活跃专家切片，将测得的密集层调用时间从约8.1毫秒减少到4.5毫秒。一个流式管道协调并发生成，同时捕获状态的草稿评估测量与已部署数值路径的一致性。在15个并发级别下进行的配对测量，从1到176个流在88个流时达到60.29的总解码令牌/s，在完整服务阶段的解码速度为46.87令牌/s。在15个流时，首个令牌的中位延迟为6.05秒。将上下文预算从1024位置的默认值提高，实际提示从1k到64k令牌恢复了所有19个测量答案中的嵌入代码，首个令牌时间随着 $aN+bN^2$ 增长，解码延迟大致线性增长，均受限于单线程CPU注意力循环，而非内存，后者每个流保持512k位置。对捕获的舰队状态的评估分离了词汇选择和权重量化对草稿一致性的影响。这些贡献共同建立了一种在具有共享CPU-GPU内存的分布式客户端系统上执行和评估大型稀疏模型的方法。
cs.AI / 28 / 2610.07237
SPECTRUM: Proximal Spectral Modulation for Looped Self-Distillation
SPECTRUM：用于循环自蒸馏的近端谱调制
Abstract
A model that learns from its own outputs inherits more than their correctness: it inherits which solutions it produces. We formulate Looped Self-Distillation, a self-evolution framework for code generation in which a model repeatedly generates and learns from its own raw outputs, under a fixed information budget, without ongoing external assessment or test-based selection of the generated samples. We identify a consequential separation: correctness can improve while the breadth of correct implementations contracts. We introduce SPECTRUM, which re-estimates loss-sensitive key/value geometry from a fixed reference anchor at each round and converts it into full-rank proximal spectral modulation. All generated completions train a single student, whose subsequent inference requires no intervention. After five rounds of experiments on MBPP, SPECTRUM retains 89.9% of the initial model's 64-sample correct AST richness, compared with 66.4% for Vanilla self-distillation and 65.5% for a subspace-projection control. The advantage persists at matched correct-sample counts. Without further training or recalibration, the resulting student also achieves higher matched-correct richness than Vanilla SD on HumanEval+ and APPS Intro, demonstrating transfer of the diversity benefit. These findings establish correct-solution retention as a complementary objective of recursive self-improvement (RSI) and show that generation-time intervention can improve the solution repertoire retained by subsequent students.
Chinese Translation
一个从自身输出中学习的模型不仅继承了输出的正确性，还继承了其产生的解决方案。我们提出了循环自蒸馏（Looped Self-Distillation），这是一个用于代码生成的自我进化框架，其中模型在固定的信息预算下反复生成并学习自身的原始输出，而不依赖于持续的外部评估或基于测试的生成样本选择。我们识别出一个重要的分离现象：正确性可以提高，而正确实现的广度却可能收缩。我们引入了SPECTRUM，它在每一轮中从固定的参考锚点重新估计损失敏感的键/值几何，并将其转换为全秩近端谱调制。所有生成的补全都训练一个单一的学生模型，其后续推理无需干预。在MBPP上进行五轮实验后，SPECTRUM保留了初始模型64个样本的正确抽象语法树（AST）丰富度的89.9%，而香草自蒸馏（Vanilla self-distillation）为66.4%，子空间投影控制为65.5%。这一优势在匹配的正确样本数量上仍然存在。在没有进一步训练或重新校准的情况下，所得到的学生模型在HumanEval+和APPS Intro上也实现了比香草自蒸馏更高的匹配正确丰富度，展示了多样性优势的转移。这些发现确立了正确解决方案保留作为递归自我改进（Recursive Self-Improvement, RSI）的一个补充目标，并表明生成时的干预可以改善后续学生保留的解决方案库。
cs.AI / 29 / 2610.07249
Can Semantic Geometry Teach an AI Judgement?
语义几何能否教会人工智能判断？
Abstract
How can an AI agent determine what rules to follow? One rule permits an action. Another imposes a condition, exception, or conflicting obligation. Deterministic systems can resolve those relationships when they have been specified. When they remain implicit in language, an agent can follow one rule while missing another that should stop it. Refusing every unresolved action avoids that risk, but also blocks permissible actions. We wanted the agent to make the distinction and still act. Our initial hypothesis was that geometric measurements could supply a basis for judgment. We represented actions and policies as vectors, then tested whether their geometry could identify governing policies and interpret the action's relation to them. Across four studies, the tested approaches did not establish reliable pre-action judgment. In the final synthetic study, a lexical router recovered every governing and blocking policy while reducing median policy checks by 97.7%. The composed pipeline nevertheless escalated all 2,304 test actions, including those it should have allowed. Supplying every policy to the same downstream mechanism changed no decision. Finding the policies had not solved the problem of interpreting them. This result led us to revise our hypothesis: judgment in AI agents requires developing a consequence graph. Such a graph would connect the actor and authority to policy conditions, exceptions, and the changes an action would produce. Follow-on studies will ask whether making those relationships explicit helps the agent distinguish when to act, stop, or seek review.
Chinese Translation
人工智能代理如何确定遵循哪些规则？一种规则允许某个行为，另一种规则则施加条件、例外或相互冲突的义务。确定性系统在规则明确时能够解决这些关系。当这些关系在语言中保持隐含时，代理可能遵循一种规则，而忽略另一种应当阻止其行为的规则。拒绝所有未解决的行动可以避免这一风险，但也会阻碍允许的行动。我们希望代理能够区分这些规则并继续行动。我们的初步假设是几何测量可以为判断提供基础。我们将行动和政策表示为向量，然后测试其几何形状是否能够识别治理政策并解释行动与政策之间的关系。在四项研究中，测试的方法未能建立可靠的前行动判断。在最后一项综合研究中，一个词汇路由器恢复了所有治理和阻止政策，同时将中位政策检查减少了97.7%。然而，组合的管道仍然提升了所有2,304个测试行动，包括那些应该允许的行动。将所有政策提供给相同的下游机制并未改变任何决策。发现政策并没有解决解释它们的问题。这个结果促使我们修正了假设：人工智能代理的判断需要开发一个后果图。这样的图将连接行为者和权威与政策条件、例外以及行动所产生的变化。后续研究将探讨明确这些关系是否有助于代理区分何时行动、停止或寻求审查。
cs.AI / 30 / 2610.07250
Internalizing Agent Experience into Diffusion Model Weights via On-Policy Context Distillation
通过在线策略上下文蒸馏将代理经验内化为扩散模型权重
Abstract
Wrapping an image generation model in an agentic harness can effectively boost Text-to-Image task performance: the harness can leverage memory, skills, workflow orchestration, result verification, and iterative refinement to continually construct and revise prompts, thereby eliciting better images. These gains, however, remain external to the diffusion model and are realized only while the full harness runs. We propose Diffusion On-Policy Context Distillation (D-OPCD), which treats the agent-improved prompt as privileged context and distills the knowledge encoded in the agent harness into the weights of the diffusion model, so that the model retains part of the harness's benefit when conditioned on the original query alone. Using a Text-to-Image agent equipped with our proposed Auto Skill Evolver (ASE), we show that D-OPCD can internalize harness capabilities into the generator's weights, raising the average direct-generation score from 60.52 to 65.09 across four benchmarks. With this knowledge absorbed into the weights, the harness can shed its saturated skills and resume evolving: a second ASE round on the updated generator improves on a skill-free harness by additional 1.83 points, pointing toward text-to-image systems in which harness and model keep improving each other through continual co-evolution.
Chinese Translation
将图像生成模型置于代理框架中可以有效提升文本到图像任务的性能：该框架能够利用记忆、技能、工作流程协调、结果验证和迭代优化来不断构建和修订提示，从而生成更好的图像。然而，这些收益仍然是外部于扩散模型的，只有在完整的框架运行时才能实现。我们提出了扩散在线策略上下文蒸馏（Diffusion On-Policy Context Distillation, D-OPCD），将代理改进的提示视为特权上下文，并将编码在代理框架中的知识蒸馏到扩散模型的权重中，使得模型在仅基于原始查询的条件下也能保留部分框架的优势。使用配备我们提出的自动技能进化器（Auto Skill Evolver, ASE）的文本到图像代理，我们展示了D-OPCD能够将框架能力内化到生成器的权重中，使平均直接生成评分从60.52提升至65.09，跨越四个基准。随着这些知识被吸收到权重中，框架能够摆脱其饱和技能并恢复进化：在更新后的生成器上进行第二轮ASE进化，相较于无技能框架又提高了1.83分，指向一种文本到图像系统，其中框架和模型通过持续的共同进化不断相互提升。
cs.AI / 31 / 2610.07257
MemMux: Runtime Verification and Honest Resource Attribution for Fleets of Parallel Coding Agents
MemMux：并行编码代理队列的运行时验证与诚实资源归属
Abstract
Developers increasingly run a fleet of coding agents side by side on one workstation. The tools they reach for, terminal multiplexers like tmux and a new generation of agent managers, were built to arrange windows, not to govern memory. When ten agents each spawn language servers, test runners, and browsers, no standard tool can say how much memory belongs to which agent, confirm that a terminated agent's descendants are gone, notice a child that has escaped its agent, or keep the machine off the swap cliff when an OOM kill would silently discard uncommitted work. We treat these as runtime-verification problems: an agent-hosting substrate should continuously emit observable signals an operator or auditor can check while agents run. We present MemMux, a local runtime that turns resource governance into checkable signals (per-agent attribution, complete reclamation, escaped-process visibility, bounded footprint under overcommit, and monitoring overhead), with a claims-disciplined benchmark against tmux, a purpose-built agent multiplexer, and a raw-process baseline on identical workloads. Under a binding memory budget on a Linux host, MemMux keeps the fleet under budget (7.5 GiB) with zero swap by admitting a subset and reclaiming under pressure, while the ungoverned tools run every agent, pin the machine at its RAM ceiling (2x over budget), and spill about 2 GiB into swap. MemMux reclaims 100% of a terminated agent's process subtree where the raw baseline strands half of it, and it alone surfaces escaped children (10 of 10 detected). We report the cost: the 1 Hz attribution scan runs near 0.6% CPU at one agent but 2.7% at ten, above our 2% target. Running the harness on real Claude Code sessions shows 100% attribution and low overhead carry over to live agent trees. We release the engine, benchmark, and a one-command reproducer.
Chinese Translation
开发者越来越多地在一台工作站上并行运行一组编码代理。他们所使用的工具，如终端复用器 tmux 和新一代代理管理器，旨在排列窗口，而非管理内存。当十个代理各自启动语言服务器、测试运行器和浏览器时，没有任何标准工具能够明确每个代理占用多少内存，确认已终止代理的后代是否已消失，发现逃逸的子进程，或在内存不足（OOM）杀死的情况下保持机器不进入交换区（swap cliff），以免默默丢弃未提交的工作。我们将这些问题视为运行时验证问题：一个代理托管平台应在代理运行时持续发出可观察的信号，以便操作员或审计员进行检查。我们提出了 MemMux，这是一个本地运行时，将资源管理转化为可检查的信号（每个代理的归属、完全回收、逃逸进程可见性、在过度提交下的有限占用以及监控开销），并与专为代理复用设计的 tmux 进行基准测试，以及在相同工作负载下的原始进程基线进行比较。在一个绑定内存预算的 Linux 主机上，MemMux 通过接纳一个子集并在压力下进行回收，保持整个队列在预算内（7.5 GiB），且没有使用交换区，而无管理工具则运行每个代理，将机器的内存固定在其 RAM 上限（超出预算 2 倍），并将约 2 GiB 溢出到交换区。MemMux 能够回收 100% 已终止代理的进程子树，而原始基线则留下其中一半未回收，并且只有 MemMux 能够发现逃逸的子进程（10 个中的 10 个被检测到）。我们报告了成本：1 Hz 的归属扫描在一个代理时 CPU 使用率接近 0.6%，而在十个代理时为 2.7%，超过了我们的 2% 目标。在真实的 Claude Code 会话中运行该工具显示出 100% 的归属率和低开销能够延续到实时代理树中。我们发布了该引擎、基准测试和一条命令的重现工具。
cs.AI / 32 / 2610.07261
Verifying Coordination in Parallel Coding Agents: NP-Bench and a Scheduling Planner
验证并行编码代理中的协调性：NP-Bench与调度规划器
Abstract
A team of coding agents can look fine agent by agent yet fail as a team: each passes its own tests while the merged result is broken, and single-agent evaluation never catches it. As teams run several LLM coding agents in parallel on one codebase, the agents collide: two rewrite the same function, one codes against a contract a teammate just changed, and integration fails after the work is done. Most coordination tools react (watch for a conflict, then warn), but at agent speed the warning arrives after the wasted edit. We recast the problem as scheduling: take each work item's declared scope, partition the work into disjoint scopes, and order merges along the producer->consumer graph, all up front. We build this planner into Nerveplane and evaluate it with NP-Bench, an environment-grounded three-arm benchmark (no coordination; reactive detection; proactive planning) that verifies integration off a real git merge, both in a deterministic simulation and with live agents. The planner lifts clean-integration from 1/9 to 9/9 scenarios and cuts merge conflicts from 13 to 0, with a gap that grows in the number of agents. On a live breaking contract change it rescues an outcome both baselines miss on every seed: the clean-integration rate rises from 0 (no coordination and reactive detection) to 1.0 on a frontier model and 0.6 on a small one, while agents respect assigned scopes (0/5 leakage). A cross-session memory drops the repeated-mistake rate from 1.00 to 0.00 on strong and weak models alike. We also report a negative result: routing facts to agents does not rescue long-context accuracy at window-fitting scales; its value is cost and capacity, not attention. Across two capability tiers and two vendors, the benefit did not shrink as models got stronger, because it comes from how work is allocated, not model reasoning.
Chinese Translation
一个编码代理团队可能在逐个代理的评估中表现良好，但作为一个团队却可能失败：每个代理都通过了自己的测试，而合并后的结果却是错误的，单个代理的评估从未发现这一点。当团队在同一个代码库上并行运行多个大型语言模型（LLM）编码代理时，代理之间会发生冲突：两个代理重写同一个函数，一个代理在与刚刚被队友修改的合约对抗编码，最终集成在工作完成后失败。大多数协调工具采取反应式策略（监测冲突，然后发出警告），但在代理的速度下，警告在浪费的编辑之后才到达。我们将这个问题重新表述为调度：获取每个工作项声明的范围，将工作划分为不相交的范围，并沿着生产者-消费者图提前排序合并。我们将这个规划器构建到Nerveplane中，并通过NP-Bench进行评估，这是一个基于环境的三臂基准测试（无协调；反应式检测；主动规划），验证基于真实git合并的集成，既在确定性模拟中，也与实时代理一起进行。在清洁集成的场景中，该规划器将成功率从1/9提升至9/9，并将合并冲突从13减少至0，且随着代理数量的增加，这一差距不断扩大。在一次实时的破坏性合约变更中，它挽救了一个所有基线在每个种子上都未能实现的结果：清洁集成率从0（无协调和反应式检测）上升至前沿模型的1.0和小型模型的0.6，同时代理遵循分配的范围（0/5泄漏）。跨会话记忆将重复错误率从1.00降低至0.00，无论是在强模型还是弱模型中。我们还报告了一个负面结果：将事实路由到代理并未在窗口拟合规模上改善长上下文的准确性；其价值在于成本和容量，而非注意力。在两个能力层级和两个供应商中，随着模型的增强，收益并未缩小，因为收益来源于工作分配方式，而非模型推理。
cs.AI / 33 / 2610.07270
Does the Model Use the Feature? Separating Steering from Mechanism in LLMs
模型是否使用特征？在大规模语言模型中分离引导与机制
Abstract
Internal features in LLMs are often interpreted as mechanisms when they track a concept and their manipulation changes a related behavior. Yet steering can push a feature far outside its natural range, where its effects need not reflect the model's own computation. We examine this inference and propose an empirical contract whose tests evaluate features at values observed on natural inputs. One test copies a feature's value from an input that shows a behavior into a matched input that does not (installation) or the reverse (removal); the other restores the feature after an upstream edit (downstream rescue). Installation measures how far the feature suffices for the behavior; removal and downstream rescue measure how much the model uses it. Applied to three kinds of representations, the two strengths separate sharply. The published unknown-entity latent strongly steers knowledge abstention, yet installing observed values from either published latent into matched prompts transfers only a small fraction of the natural known--unknown abstention contrast. Dense known--unknown directions show opposite asymmetries between installation and removal in Gemma and Llama, and how fully a released subject--verb agreement feature set reproduces and restores the behavior depends on how its values are written into the model. Tracking a concept and steering a behavior therefore do not by themselves show that the model uses a feature, and each conclusion holds only for the intervention tested.
Chinese Translation
大规模语言模型（LLMs）中的内部特征通常被解释为机制，当它们跟踪一个概念并且其操作改变相关行为时。然而，引导可以将特征推向其自然范围之外，在这种情况下，其效果不一定反映模型自身的计算。我们考察这种推断，并提出一个经验契约，其测试评估在自然输入上观察到的特征值。一个测试将特征的值从一个表现出某种行为的输入复制到一个不表现该行为的匹配输入中（安装），或反之（移除）；另一个测试则在上游编辑后恢复该特征（下游救援）。安装测量特征在多大程度上足以产生该行为；移除和下游救援则测量模型在多大程度上使用该特征。应用于三种类型的表征时，这两种强度的分离非常明显。已发布的未知实体潜变量强烈引导知识的回避，然而将观察到的值从任何已发布的潜变量安装到匹配的提示中，仅转移了自然已知-未知回避对比的一小部分。密集的已知-未知方向在Gemma和Llama中显示出安装与移除之间的相反不对称性，而一个释放的主谓一致特征集在多大程度上再现和恢复该行为，取决于其值是如何写入模型的。因此，跟踪一个概念和引导一个行为本身并不能证明模型使用了某个特征，并且每个结论仅适用于所测试的干预。
cs.AI / 34 / 2610.07274
A Trust Layer for Agent Evaluation
代理评估的信任层
Abstract
Deterministic benchmark scores show that an agent received credit, but not whether that credit was earned, reported honestly, or would hold on a second run. We introduce a Trust Layer for Agent Evaluation, an additive post-hoc framework that reports, beside each recorded score, whether it should be believed. It verifies four properties: whether the result is supported by the benchmark's own grading logic, whether a passing answer was earned through traceable computation, whether the agent's completion claim matches what occurred, and whether the result is stable under repeated execution. The first three use only saved artifacts; the fourth re-runs the agent. Model judgments only label evidence under majority voting; all verdicts follow deterministic rules and never modify the recorded score. Applied to five agent configurations on 108 tasks from Agents' Last Exam, every model shows passing runs with no traceable computation (at rates varying tenfold), confirmed false completion claims, and unstable results: 18-46% of tasks do not stay in one score band over five runs. Only 22.6% of recorded passes clear all four checks (95% CI 15.0-32.6, n=84). Measuring what an agent can do and verifying that it did it are different problems, and current benchmarks address only the first.
Chinese Translation
确定性基准分数显示一个代理获得了信用，但并不能说明该信用是否是通过诚实报告获得的，或者在第二次运行时是否依然有效。我们提出了代理评估的信任层（Trust Layer for Agent Evaluation），这是一个附加的事后框架，除了记录的每个分数外，还报告该分数是否值得相信。它验证四个属性：结果是否得到了基准自身评分逻辑的支持，是否通过可追溯的计算获得了及格答案，代理的完成声明是否与实际发生的情况相符，以及结果在重复执行下是否稳定。前三个属性仅使用保存的文档；第四个属性则重新运行代理。模型判断仅在多数投票下标记证据；所有裁决遵循确定性规则，且从不修改记录的分数。应用于来自《代理的最后考试》（Agents' Last Exam）的108个任务上的五个代理配置，每个模型显示出在没有可追溯计算的情况下通过的运行（通过率变化十倍），确认了错误的完成声明，以及不稳定的结果：在五次运行中，18-46%的任务未能保持在一个分数区间内。只有22.6%的记录通过者通过了所有四项检查（95%置信区间15.0-32.6，n=84）。测量一个代理能做什么与验证它是否真的做到了是两个不同的问题，而当前的基准仅解决第一个问题。
cs.AI / 35 / 2610.07309
The Right Memory in the Wrong Context: Verifying Retrieval Admissibility in Long-Term Agent Memory
错误上下文中的正确记忆：验证长期代理记忆中的检索可接受性
Abstract
Long-term-memory agents can retrieve relevant information that is inadmissible for the current request because it belongs to another principal, violates policy, or reflects an incompatible lifecycle state. Recall and final-answer accuracy do not reveal this: a route can appear safe by missing required evidence, while a correct answer may follow inadmissible prompt exposure. We introduce a retrieval-admissibility verification framework that assigns each memory-query pair one of three statuses (admissible, inadmissible, or unresolved), compares routes at matched required-evidence recall with bounds for unresolved cases, and tracks memory IDs through prompt exposure while linking exposure to target-level disclosure. We evaluate its stages on separate, non-pooled populations. A post-hoc top-20 reanalysis of frozen rankings from two public long-term-memory benchmarks, RHELM and MemOps, covers 3,767 queries. All released anchors lie within trusted query namespaces; with within-namespace scores unchanged, off-namespace filtering cannot lower their ranks. Top-20 anchor recall increases from 0.432 to 0.533, 80% recall feasibility from 0.237 to 0.311, and exact similarity evaluations decrease by 98.3%. In a frozen 72-case development diagnostic, a released-metadata reference preserves required evidence, whereas neither text-only verifier detects violations under the 1% required-anchor false-denial limit. Across 1,523 paired benchmark-native cases, namespace routing is associated with judged-accuracy gains of 0.053-0.068 across three readers; recall also changes, so this comparison is observational. In 16 controlled exposure scenarios, only one of four reader-specific 95% confidence intervals excludes zero for relevant-inadmissible literal disclosure (+0.156, 95% CI [0.031, 0.312]). Results motivate separate verification of candidate support, admissibility, prompt exposure, and answer disclosure.
Chinese Translation
长期记忆代理可以检索与当前请求不相容的相关信息，因为这些信息属于其他主体、违反政策或反映不兼容的生命周期状态。回忆和最终答案的准确性并不能揭示这一点：一条路径可能看似安全，但缺少所需证据，而正确答案可能是在不合规提示曝光后得出的。我们提出了一种检索可接受性验证框架，为每个记忆查询对分配三种状态之一（可接受、不合规或未解决），在匹配的所需证据回忆中比较路径，并为未解决的情况设定界限，同时在提示曝光过程中跟踪记忆ID，并将曝光与目标级别的披露联系起来。我们在不同的非汇总人群中评估其各个阶段。对两个公共长期记忆基准（RHELM 和 MemOps）冻结排名的事后前20名重新分析涵盖了3767个查询。所有发布的锚点均位于可信查询命名空间内；在命名空间内的得分不变，命名空间外的过滤无法降低其排名。前20名锚点的回忆率从0.432增加到0.533，80%的回忆可行性从0.237增加到0.311，而精确相似性评估下降了98.3%。在一个冻结的72案例开发诊断中，发布的元数据参考保留了所需证据，而仅文本验证器在1%的所需锚点虚假拒绝限制下未能检测到违规。在1523对配对基准原生案例中，命名空间路由与三位读者的判断准确性增益相关，增幅为0.053-0.068；回忆也发生变化，因此此比较为观察性。在16个受控曝光场景中，只有四个读者特定的95%置信区间中的一个排除了相关不合规字面披露的零值（+0.156，95% CI [0.031, 0.312]）。结果促使对候选支持、可接受性、提示曝光和答案披露进行单独验证。
cs.AI / 36 / 2610.07311
Understanding and Mitigating Inference-Time Overreliance Using Agentic Memory
理解和缓解推理时的过度依赖代理记忆
Abstract
Agentic memory allows LLM agents to reuse past experience, yet retrieved memories can also distort inference even when they are benign, correctly stored, and appropriately retrieved. We study this failure mode, which we call memory over-reliance. Across benchmarks and memory architectures, we find that memory is useful when past experience transfers to the current task, but can become misleading when only part of the evidence transfers. Failures are strongest under partial query-memory overlap, a pattern further confirmed by controlled experiments thatvary the amount of overlapping evidence. Motivated by this finding, we propose MEMTRIM, a plug-and-play framework that indexes memory evidence at write time and controls its reuse at read time. MEMTRIM removes repeated or conflicting evidence while preserving useful memory-specific information, requires no retraining, and applies to both embedding-based and structured memory systems.Experiments show that MEMTRIM reduces memory overreliance while preserving the benefits of useful memory across models and memory settings.
Chinese Translation
代理记忆使大型语言模型（LLM）代理能够重用过去的经验，但即使检索的记忆是良性的、正确存储的和适当检索的，它们也可能扭曲推理。我们研究了这种失败模式，称之为记忆过度依赖。在不同的基准和记忆架构中，我们发现当过去的经验能够转移到当前任务时，记忆是有用的，但当只有部分证据能够转移时，它可能会变得误导。在部分查询-记忆重叠的情况下，失败现象最为明显，这一模式通过控制实验进一步得到了证实，这些实验改变了重叠证据的数量。基于这一发现，我们提出了MEMTRIM，一个即插即用的框架，它在写入时对记忆证据进行索引，并在读取时控制其重用。MEMTRIM在保留有用的记忆特定信息的同时，去除重复或冲突的证据，不需要重新训练，并适用于基于嵌入和结构化的记忆系统。实验表明，MEMTRIM在保留有用记忆的好处的同时，减少了记忆的过度依赖，适用于多种模型和记忆设置。
cs.AI / 37 / 2610.07313
Rule-Based Languages for Neurosymbolic AI
基于规则的神经符号人工智能语言
Abstract
Logic programming is increasingly used as the symbolic component of neurosymbolic AI systems. We survey the main rule-based languages in this setting, namely Datalog, answer set, and probabilistic logic programs, along four axes: semantics, expressiveness, neural integration, and evaluation mechanism. We analyse over 50 recent systems and applications, comparing formalism usage across four research areas: databases and programming languages, machine learning, vision, and robotics. We provide a decision matrix mapping application scenarios to required features and close by outlining open problems.
Chinese Translation
逻辑编程越来越多地被用作神经符号人工智能系统的符号组件。我们在四个维度上调查了这一背景下的主要基于规则的语言，即 Datalog、答案集和概率逻辑程序：语义、表达能力、神经集成和评估机制。我们分析了超过50个近期的系统和应用，比较了四个研究领域（数据库和编程语言、机器学习、视觉和机器人）中形式主义的使用情况。我们提供了一个决策矩阵，将应用场景与所需特性进行映射，并最后概述了未解决的问题。
cs.AI / 38 / 2610.07342
Rationale-Guided Policy Optimization: Learning to Reason with Adaptive Rationale Scaffolding
基于推理的策略优化：通过自适应推理支架学习推理
Abstract
On-policy reinforcement learning has become a central paradigm for improving the reasoning abilities of large language models. However, its effectiveness is often limited by reward sparsity: when a model fails to discover correct trajectories for difficult problems, the optimization process receives little useful signal and may stagnate. Existing approaches mitigate this issue by incorporating off-policy demonstrations, expert traces, or model-generated solutions, but they typically require the auxiliary data to match the format of the reinforcement-learning task, often relying on rejection sampling from stronger models to obtain suitable training trajectories. We introduce Rationale-Guided Policy Optimization (RGPO), a framework that adaptively leverages ground-truth rationale information according to the model's current capability while preserving its freedom to explore. Rather than treating reference solutions as fixed imitation targets, RGPO uses them as temporary scaffolds: rationales help the model generate improved responses, after which only higher-reward, model-generated solutions are transferred back to the original unguided setting. This design allows training to exploit available ground-truth information without requiring off-policy data to follow the same format as the RL task. Across both language-only and vision-language reasoning settings, RGPO consistently improves performance over RLVR baselines, and ablation studies show that adaptive rationale guidance is a key contributor to these gains. These results suggest that RGPO offers a practical and general approach for reducing reward sparsity, stabilizing reinforcement learning, and improving reasoning performance in both text-only and multimodal models.
Chinese Translation
基于策略的强化学习已成为提升大型语言模型推理能力的核心范式。然而，其有效性常常受到奖励稀疏性的限制：当模型未能为困难问题发现正确的轨迹时，优化过程获得的有用信号很少，可能会停滞。现有方法通过结合离策略演示、专家轨迹或模型生成的解决方案来缓解这一问题，但它们通常要求辅助数据与强化学习任务的格式相匹配，往往依赖于从更强模型中进行拒绝采样以获得合适的训练轨迹。我们提出了基于推理的策略优化（Rationale-Guided Policy Optimization, RGPO），这是一个根据模型当前能力自适应利用真实推理信息的框架，同时保留其探索的自由。RGPO并不将参考解决方案视为固定的模仿目标，而是将其作为临时支架：推理帮助模型生成改进的响应，之后只有更高奖励的模型生成解决方案会被转移回原始的无指导设置。这一设计允许训练利用可用的真实信息，而无需离策略数据遵循与强化学习任务相同的格式。在语言单一和视觉-语言推理设置中，RGPO的性能始终优于RLVR基线，消融研究表明自适应推理指导是这些提升的关键因素。这些结果表明，RGPO为减少奖励稀疏性、稳定强化学习以及提高文本和多模态模型的推理性能提供了一种实用且通用的方法。
cs.AI / 39 / 2610.07350
Trajectory-Retrieval Speculative Decoding: When Does a Model's Own History Help?
轨迹检索推测解码：模型自身历史何时提供帮助？
Abstract
Long chain-of-thought reasoning increases sequential decoding cost while creating a growing history of potentially reusable continuations. We investigate when this history supplies useful drafts and complements an existing drafter. Controlled source comparisons reveal trajectory-specific reuse, motivating our method Trajectory-Local Adaptive Retrieval (TLAR). TLAR retrieves approximately matched continuations from the current trajectory and uses recent verification outcomes to adapt retrieval activation and candidate width. TLAR combines retrieved continuations with model-generated drafts in a shared candidate tree, preserving the target model's output distribution through exact verification. Across code debugging, mathematics, and open-ended writing, our evaluation connects source reuse, incremental acceptance, and execution cost. Combining TLAR with strong retrieval baselines improves token acceptance under matched verification budgets and increases end-to-end throughput over the draft-model baseline. These findings support generated trajectories as runtime memory for adaptive inference.
Chinese Translation
长链推理增加了顺序解码的成本，同时创造了一个不断增长的潜在可重用延续历史。我们研究了何时这些历史提供有用的草稿，并补充现有的草拟者。受控源比较揭示了轨迹特定的重用，激励了我们的方法轨迹局部自适应检索（Trajectory-Local Adaptive Retrieval, TLAR）。TLAR 从当前轨迹中检索大致匹配的延续，并利用最近的验证结果来调整检索激活和候选宽度。TLAR 将检索到的延续与模型生成的草稿结合在一个共享的候选树中，通过精确验证保持目标模型的输出分布。在代码调试、数学和开放式写作中，我们的评估将源重用、增量接受和执行成本联系在一起。将 TLAR 与强大的检索基线结合，在匹配的验证预算下提高了标记接受率，并在草稿模型基线之上增加了端到端的吞吐量。这些发现支持生成的轨迹作为自适应推理的运行时记忆。
cs.AI / 40 / 2610.07354
Evaluating Escalation Signals for LLM Routing: Targets, Controls, and Five Ways to Fool Yourself
评估大型语言模型路由的升级信号：目标、控制与五种自我欺骗的方法
Abstract
Deciding when to escalate a query from a small language model to a larger one requires a cheap signal that predicts, before the large model is called, whether escalating would help. Semantic entropy, originally developed to detect hallucinations, is a natural candidate: it measures how much a model's sampled answers disagree in meaning, and high disagreement often signals an unreliable answer. We test it across three benchmarks and two model families. On GSM8K, with a small/large pair about twelve times apart in size, semantic entropy reliably distinguishes the small model's mistakes (AUROC 0.871) and improves routed accuracy over random escalation by up to nine points at matched cost. An earlier strong-looking result on a synthetic benchmark proved misleading: a simple rule based only on question difficulty, with no model involved, matched semantic entropy almost exactly. This paper's main contribution is a set of checks that catch this before it is reported as real. We show that scoring a cheap, question-only difficulty estimate alongside any signal reveals whether the signal adds real information or just tracks how hard a question looks; that two reasonable definitions of "escalation worked" can produce very different results on the same data; that a benchmark can leave almost no room for any signal to beat simply always using the large model; and that the true cost of live sampling can make routing more expensive than calling the large model directly. For a cheaper alternative that reuses cached past outcomes, we show how to predict whether it will work on a new dataset -- confirmed by correctly forecasting a collapse from AUROC 0.908 to chance level (0.518) ahead of time. We offer these as a general checklist for evaluating escalation signals.
Chinese Translation
决定何时将查询从小型语言模型升级到大型模型需要一个廉价的信号，以预测在调用大型模型之前，升级是否会有所帮助。语义熵，最初用于检测幻觉，是一个自然的候选者：它衡量模型采样答案在意义上的不一致程度，而高不一致性通常表明答案不可靠。我们在三个基准和两个模型系列上进行了测试。在GSM8K上，大小相差约十二倍的小/大型模型，语义熵可靠地区分了小模型的错误（AUROC 0.871），并在匹配成本下将路由准确性提高了最多九个百分点。之前在一个合成基准上看似强有力的结果证明是误导性的：仅基于问题难度的简单规则，与语义熵几乎完全匹配，而没有涉及模型。本文的主要贡献是一组检查，能够在报告为真实之前捕捉到这一点。我们展示了在任何信号旁边评分一个廉价的、仅基于问题的难度估计，可以揭示该信号是否提供了真实信息，还是仅仅追踪问题看起来有多难；两个合理的“升级成功”的定义在相同数据上可能产生非常不同的结果；一个基准几乎没有留下任何信号超越简单使用大型模型的空间；以及实时采样的真实成本可能使路由比直接调用大型模型更昂贵。对于一个重用缓存过去结果的更便宜的替代方案，我们展示了如何预测它在新数据集上的有效性——通过提前正确预测AUROC从0.908崩溃到随机水平（0.518）来确认。我们将这些作为评估升级信号的一般检查清单。
cs.AI / 41 / 2610.07359
Evaluate the Stack, Not the Layer: Do Deterministic and LLM Gates for Agent Actions Fail Independently?
评估堆栈，而非层级：确定性和大型语言模型（LLM）门控在代理行为中的独立失效吗？
Abstract
Runtime gates for agent tool calls are stacked on the assumption that their errors multiply. We test it on 1,119 labelled agent actions from three corpora, without an adaptive adversary. The stack has one deterministic rule layer and four LLM judges, three of them re-collected with the served model recorded on every call. We read each stack as a number of multiplication-equivalent layers, n_mult, with its floor under perfect coupling. Under the STRICT miss definition (escalation to a human scored as not stopped), any two judges compose to about 1.2 to 1.4 layers ({\phi} median +0.430, 6 of 6 pairs significant, floors 1.02 to 1.17). The rule layer plus one judge composes to 1.86 to 2.09 layers ({\phi} median +0.014, 0 of 4 significant, floors 1.01 to 1.09). Under PRIMARY (escalation scored as caught) the bands are 1.21 to 1.57 and 1.80 to 2.13. Intervals separate on the pooled data, point estimates split on each corpus, and a third-vendor judge lands in the judge band. Solo accuracy does not predict what a layer adds: a cloud rule pack lowers the rule layer's solo miss rate by 20% and adds no new joint coverage. The difficulty share of judge coupling is not identifiable: 31.8% to 61.8% depending on the probe and the miss definition. One judge tier was served by an unrequested model version in 50 of 112 batches, concentrated on the external corpus. That event overturned a pre-declared analysis rule, and the scoring of review verdicts reversed five conclusions. We report both.
Chinese Translation
代理工具调用的运行时门控是基于其错误相乘的假设进行堆叠的。我们在三个语料库中对1,119个标记的代理行为进行了测试，未使用自适应对手。该堆栈包含一个确定性规则层和四个LLM评审，其中三个是通过在每次调用时记录的服务模型重新收集的。我们将每个堆栈视为多个等效乘法层（n_mult），其下限在完美耦合下。根据严格的失误定义（升级至被人类评分为未停止），任意两个评审组合的层数约为1.2到1.4层（{ heta} 中位数 +0.430，6对中有6对显著，底数为1.02到1.17）。规则层加一个评审组合的层数为1.86到2.09层（{ heta} 中位数 +0.014，4个评审中没有显著，底数为1.01到1.09）。在主要定义下（升级评分为被捕获），层数范围为1.21到1.57和1.80到2.13。基于汇总数据的区间分离，点估计在每个语料库中分裂，而第三方评审落入评审区间。单独准确性无法预测层的增益：云规则包将规则层的单独失误率降低了20%，并未增加新的联合覆盖。评审耦合的难度份额无法识别：根据探测和失误定义，范围为31.8%到61.8%。在112批次中的50批次中，一个评审层由未请求的模型版本提供，集中在外部语料库。该事件推翻了预先声明的分析规则，并使审查裁决的评分逆转了五个结论。我们对此进行了报告。
cs.AI / 42 / 2610.07376
MemCo: Memory-Centric Collaboration for Generalizing LLM Agents to Unseen Environments
MemCo：面向记忆中心的协作框架，以使大型语言模型代理能够适应未见环境
Abstract
Large language model (LLM) agents increasingly operate in interactive environments, where they need to make sequential decisions through observation, action, and feedback. Although memory can help agents reuse experience, existing work designs memory in isolation, where collecting enough trajectories to populate it is expensive. Existing shared-memory approaches mitigate isolated experience by pooling episodic memories across tasks and environments. However, retrieving shared memory is challenged by the granularity, where retrieved memories can be either too specific to preserve current grounding or too coarse to support the next action. In this work, we propose MemCo, a memory-centric collaboration framework for generalizing LLM agents to unseen interactive environments. It maintains complementary local and global memory spaces, preserving environment-specific details locally while promoting transferable workflows induced from local trajectories to global memory. During online interaction, MemCo routes relevant local and global memories in terms of the agent's current state and decision phase, enabling agents to reuse the experience of other agents without blindly transferring environment-specific details. Experiments on interactive decision-making benchmarks show that MemCo improves task success and reduces redundant exploration compared with isolate-memory and shared-memory baselines. Our code is available at https://github.com/SYannL/nvdamas.
Chinese Translation
大型语言模型（LLM）代理越来越多地在交互环境中运行，在这些环境中，它们需要通过观察、行动和反馈进行顺序决策。尽管记忆可以帮助代理重用经验，但现有的工作在孤立的情况下设计记忆，收集足够的轨迹以填充记忆是昂贵的。现有的共享记忆方法通过跨任务和环境汇集情节记忆来缓解孤立经验的问题。然而，检索共享记忆面临粒度的挑战，检索到的记忆可能过于具体，无法保持当前的基础，或过于粗略，无法支持下一步行动。在本研究中，我们提出了MemCo，一个面向记忆中心的协作框架，用于使LLM代理能够适应未见的交互环境。它维护互补的本地和全局记忆空间，在本地保留环境特定的细节，同时促进从本地轨迹到全局记忆的可转移工作流。在在线交互过程中，MemCo根据代理的当前状态和决策阶段路由相关的本地和全局记忆，使代理能够重用其他代理的经验，而不盲目地转移环境特定的细节。在交互决策基准测试中的实验表明，与孤立记忆和共享记忆基线相比，MemCo提高了任务成功率并减少了冗余探索。我们的代码可在 https://github.com/SYannL/nvdamas 获取。
cs.AI / 43 / 2610.07403
Defense-in-Depth for LLMs: Evaluating Memory Gates Against Activation-Induced and Memory-Induced Sycophancy
深度防御：评估记忆门对激活引发和记忆引发的谄媚行为的影响
Abstract
Long-term memory allows Large Language Models (LLMs) to maintain personalized context across interactions, but retrieved user history can induce memory-induced sycophancy, causing models to favor stored user beliefs over objective evidence. Existing defenses primarily operate on retrieved context and are rarely evaluated jointly with internal behavioral bias. We introduce a $2 \times 2$ defense-in-depth framework separating internal activation steering from external memory handling. We extract sycophancy steering directions from 100 paired prompts and evaluate four open-weight models across 10 steering coefficients and five memory-defense configurations on MemSyco-Bench (answers for all 1,550 items; defense conditions judged on a fixed 250-item subsample), with three LLM judges. Three of the five configurations are new (rewriting every memory, a Router Gate that keeps, rewrites, or drops each memory, and dropping all memory); the other two are MemSyco's baselines. Selective Router Gate filtering preserves substantially more of MemSyco's average accuracy than complete memory removal, and this separation persists when the models are steered toward sycophancy. On Llama 3.1 8B with Router Gate, mild inverse steering ($\alpha = -1.5$) lowers judge-averaged sycophancy from 35.80% to 31.32% while average accuracy moves from 43.99% to 43.31%; this reduction has the same direction under all three judges but is not statistically significant (paired $p = 0.08$ to $0.63$ on 149 items). External memory filtering is the part of the design that holds up; our data do not show that inverse steering adds to it.
Chinese Translation
长期记忆使大型语言模型（LLMs）能够在交互中保持个性化的上下文，但检索的用户历史可能引发记忆引发的谄媚行为，导致模型偏向存储的用户信念而非客观证据。现有的防御机制主要针对检索的上下文，且很少与内部行为偏见共同评估。我们提出了一种 $2 imes 2$ 的深度防御框架，将内部激活引导与外部记忆处理分开。我们从100对提示中提取谄媚引导方向，并在MemSyco-Bench上评估四个开放权重模型在10个引导系数和五种记忆防御配置下的表现（对所有1550个项目的回答；防御条件在固定的250个项目子样本上进行评估），并由三位LLM评审进行判断。五种配置中有三种是新的（重写每个记忆、一个路由门（Router Gate）用于保留、重写或丢弃每个记忆，以及丢弃所有记忆）；另外两种是MemSyco的基线。选择性路由门过滤比完全移除记忆保留了更多的MemSyco平均准确率，并且这种分离在模型被引导向谄媚行为时依然存在。在使用路由门的Llama 3.1 8B模型中，轻微的逆向引导（$eta = -1.5$）将评审平均谄媚行为从35.80%降低到31.32%，而平均准确率则从43.99%降至43.31%；这种减少在所有三位评审中方向一致，但在统计上并不显著（配对$p = 0.08$至$0.63$，基于149个项目）。外部记忆过滤是设计中保持有效的部分；我们的数据未显示逆向引导对此有额外贡献。
cs.AI / 44 / 2610.07423
2d-fet-bench: from spatial reasoning to fet design on flakes
2D-FET-Bench：从空间推理到片层场效应晶体管设计
Abstract
Field-effect transistor (FET) layouts on exfoliated two-dimensional flakes are typically drawn by hand for each flake, placing contacts and gates to match its position and outline in optical micrographs. To our knowledge, no executable benchmark tests whether language-model agents can perform this flake-specific construction reliably. We introduce 2D-FET-Bench V2, a benchmark of 128 layout tasks built from microscopy-derived flake contours, including hole-containing flakes and multi-flake tasks. Each task supplies a textual device specification and contour coordinates. An agent generates typed polygon and path operations rendered to GDSII. A deterministic verifier checks geometric and structural requirements, and a separate integrity check verifies that the supplied contours remain unchanged. Scripted reference layouts pass all 128 tasks, showing that every task is solvable. We evaluate six models and seven workflow and scaffold variants of GPT5.6-Luna, with five attempts per task. The best-performing configuration in the six-model panel, GPT5.6-Luna with ReAct-3, passes 62.3% of attempts and solves 80.5% of tasks at least once (coverage) and 43.8% in all five attempts (consistency). ReAct-3 exceeds the one-pass Plan-and-Execute by 27.0 pass@1 points at 2.46 times the tokens. An expert audit of one sampled verifier-passing layout per covered task, across five ReAct-3 configurations, accepts 56.4% to 63.5% of them. The benchmark evaluates geometric and structural FET layout construction.
Chinese Translation
在剥离的二维片层上，场效应晶体管（FET）布局通常是手动绘制的，以匹配其在光学显微镜图像中的位置和轮廓。我们所知，目前没有可执行的基准测试来检验语言模型代理是否能够可靠地执行这种特定于片层的构建。我们介绍了2D-FET-Bench V2，这是一个由显微镜衍生的片层轮廓构建的128个布局任务的基准，包括含孔片层和多片层任务。每个任务提供文本设备规格和轮廓坐标。代理生成类型化的多边形和路径操作，并渲染为GDSII格式。一个确定性验证器检查几何和结构要求，另一个独立的完整性检查验证所提供的轮廓保持不变。脚本化的参考布局通过了所有128个任务，显示每个任务都是可解的。我们评估了六个模型和七个工作流及支架变体的GPT5.6-Luna，每个任务进行了五次尝试。在六个模型面板中表现最佳的配置是使用ReAct-3的GPT5.6-Luna，成功通过62.3%的尝试，并且至少解决了80.5%的任务（覆盖率），在所有五次尝试中解决了43.8%（一致性）。ReAct-3在一次通过的计划与执行上超出了27.0个pass@1点，且使用的tokens数量为2.46倍。对每个覆盖任务中一个样本验证器通过的布局进行的专家审核，接受了56.4%到63.5%的布局。该基准评估几何和结构FET布局的构建。
cs.AI / 45 / 2610.07428
Adaptive Gait Biofeedback With Participant-Held-Out Modeling and Participant-Specific Updating in Chronic Ankle Instability
慢性踝关节不稳定中的自适应步态生物反馈：参与者持有的建模与参与者特定的更新
Abstract
Adaptive gait biofeedback may support repeated practice in chronic ankle instability, but its evaluation must address model performance and human response. We evaluated a temporal convolutional classifier on protocol-defined, angle-derived GOOD/BAD gait-cycle labels using participant-held-out leave-one-subject-out (LOSO) cross-validation in 20 participants. Seven participants in the adaptive-intervention group completed nine sessions over three weeks, with one motion-capture recording analyzed per session. Models updated after failed sessions were compared offline with their parent models on the same-session validation subset used for candidate selection and the first subsequent adaptive-session recording. Frontal-plane ankle angle was compared between the adaptive group and 10 sequentially enrolled controls at Baseline, Post, and 7-day Retention. Across 20 held-out folds, mean fold-level area under the receiver operating characteristic curve (AUROC) was 0.948, sensitivity for angle-threshold-exceeding BAD cycles was 0.941, and specificity for angle-threshold-meeting GOOD cycles was 0.366. Mean BAD-class F1 was higher in candidate models by 0.187 on the same-session subset and 0.118 on the first subsequent recording. At Post, the adaptive group had a baseline-adjusted frontal-plane ankle angle 5.168 degrees lower than controls (95% confidence interval, 1.766-8.569 degrees lower); the Retention contrast was uncertain. These findings characterize population-model discrimination and offline participant-specific updating during repeated biofeedback use, alongside a nonrandomized Post frontal-plane ankle angle association. They do not establish independent clinical gait classification or a causal benefit of updating.
Chinese Translation
自适应步态生物反馈可能支持慢性踝关节不稳定中的重复练习，但其评估必须考虑模型性能和人类反应。我们在20名参与者中使用参与者持有的留一法交叉验证（Leave-One-Subject-Out, LOSO）评估了一种基于时间卷积的分类器，该分类器基于协议定义的角度导出的好/坏步态周期标签。自适应干预组的7名参与者在三周内完成了九个会话，每个会话分析了一次运动捕捉记录。对失败会话后更新的模型与其母模型在同一会话验证子集上的表现进行了离线比较，该子集用于候选选择以及第一次后续自适应会话记录。在基线、后期和7天保持期时，自适应组的正面平面踝关节角度与10名顺序入组的对照组进行了比较。在20个留出的折中，平均折级接收者操作特征曲线下面积（AUROC）为0.948，超出角度阈值的坏周期的灵敏度为0.941，满足角度阈值的好周期的特异性为0.366。在同一会话子集中，候选模型的平均坏类F1值比母模型高0.187，在第一次后续记录中高0.118。在后期，自适应组的基线调整后正面平面踝关节角度比对照组低5.168度（95%置信区间，低1.766-8.569度）；保持期的对比结果不确定。这些发现描述了在重复使用生物反馈过程中群体模型的区分能力和离线参与者特定的更新，以及非随机的后期正面平面踝关节角度关联。它们并未确立独立的临床步态分类或更新的因果益处。
cs.AI / 46 / 2610.07434
When Does AI Supervision Help? A Role-Aware Study of Network Fraud Decision Management with Blockchain Auditability
人工智能监督何时有效？基于角色的区块链可审计网络欺诈决策管理研究
Abstract
When does a second artificial intelligence (AI) component improve a primary network-fraud decision rather than add operational burden? We study this question through a role-aware Decider-Supervisor (DS) framework with blockchain auditability, evaluating four directional configurations that combine centralised machine learning, a Federated Averaging (FedAvg)-trained federated meta-model, and Base or Quantized Low-Rank Adaptation (QLoRA) large language model variants. The analysis compares primary-only and supervised decisions using non-hard fraud performance, intervention burden, conditional calibration, traffic-mix and Review-capacity sensitivity, dependability tests, and blockchain lifecycle controls. The deterministic hard gate resolves 89.994% of fraudulent requests, leaving the non-hard population as the main AI decision setting. Conditional validation calibration does not produce a consistently transferable supervisory advantage on deployment replay. DS-3 QLoRA is the least disruptive supervised configuration, but it still underperforms its primary FedAvg stage in F1 and total errors. Across 36 reweighted traffic mixtures, supervision reduces total errors only for DS-4 Base in two extreme high-fraud scenarios. Blockchain tests support digest verification, tamper detection, authorisation, single-use review resolution, and post-finalisation integrity, while exposing a pre-finalisation single-write limitation. The results show that the value of AI supervision depends on role assignment, calibration, escalation policy, traffic composition, and lifecycle controls rather than on the presence of a second model alone.
Chinese Translation
第二个人工智能（AI）组件何时能够改善主要的网络欺诈决策，而不是增加操作负担？我们通过一个具有区块链可审计性的角色感知决策者-监督者（Decider-Supervisor, DS）框架来研究这个问题，评估四种方向配置，这些配置结合了集中式机器学习、经过联邦平均（Federated Averaging, FedAvg）训练的联邦元模型，以及基础或量化低秩适应（Quantized Low-Rank Adaptation, QLoRA）的大型语言模型变体。分析比较了仅使用主要决策和监督决策的非硬欺诈表现、干预负担、条件校准、流量组合和审查能力敏感性、可靠性测试以及区块链生命周期控制。确定性硬门能够解决89.994%的欺诈请求，使得非硬人群成为主要的AI决策设置。条件验证校准在部署重放中并未产生一致可转移的监督优势。DS-3 QLoRA是干扰最小的监督配置，但在F1和总错误率上仍低于其主要的FedAvg阶段。在36种重新加权的流量组合中，监督仅在两个极端高欺诈场景下对DS-4 Base减少了总错误。区块链测试支持摘要验证、篡改检测、授权、一次性审查解决和最终化后完整性，同时暴露了最终化前单写限制。结果表明，AI监督的价值依赖于角色分配、校准、升级政策、流量组成和生命周期控制，而不仅仅是第二个模型的存在。
cs.AI / 47 / 2610.07459
Auditable Claims about AI Agents
可审计的人工智能代理声明
Abstract
Organizations make claims about their AI agents: a person approves every external email, every action is logged, an evaluation shows the agent is safe to deploy. Article 12 of the EU AI Act requires high-risk systems to allow the automatic recording of events but does not say which records settle a given claim. The position is one sentence: to be checked, a claim about an agent must first name its policy, its scope, the records that would settle it, and who writes them. Adapting the preconditions of an assurance engagement, we call a claim auditable when these elements and a decision rule are fixed before any verdict and the records are obtainable. This extends the Policy Checkability dimension of our Auditable Agents framework from single actions to claims. Agents add three conditions: coverage by an independent record, authorization bound to each action's arguments, and completeness beyond integrity. Under an explicit model, we prove that support is impossible without each wherever its hypotheses hold. A claim-check table applies the method to six common claims, anchored in current NIST, IETF, and OWASP drafts. A worked case follows one claim through five evidence states. We close with a practice box and steps for operators, buyers, auditors, and standard setters.
Chinese Translation
组织对其人工智能代理做出声明：每封外部电子邮件都经过人工审批，每个操作都有记录，评估显示该代理可以安全部署。欧盟人工智能法第12条要求高风险系统允许事件的自动记录，但并未说明哪些记录可以支持特定声明。我们的立场可以用一句话概括：要被检查，关于代理的声明必须首先明确其政策、范围、能够支持该声明的记录以及记录的撰写者。在适应保证业务的先决条件时，我们称当这些要素和决策规则在任何裁决之前都已确定且记录可获得时，该声明为可审计的。这将我们可审计代理框架中的政策可检查性维度从单一操作扩展到声明。代理增加了三个条件：由独立记录覆盖、与每个操作的参数绑定的授权，以及超越完整性的完备性。在一个明确的模型下，我们证明在假设成立的情况下，缺少任何一个条件都无法提供支持。声明检查表将该方法应用于六个常见声明，基于当前的NIST、IETF和OWASP草案。一个案例研究展示了一个声明在五个证据状态下的演变。最后，我们提供了一个实践框和操作员、买家、审计员和标准制定者的步骤。
cs.AI / 48 / 2610.07469
COMPASS: Finding Where Reasoning Lives in Language Models
COMPASS：寻找推理在语言模型中的存在
Abstract
Explicitly eliciting reasoning substantially improves LLM performance. Existing approaches require a predefined characterization of reasoning, whether through CoT prompt design, contrastive CoT directions, or via SAE derived reasoning features. For mathematical reasoning with verifiable answers, we show that a much simpler signal suffices, which is the correctness of the model's own direct answer attempts. This signal yields a latent direction that elicits reasoning. This direction is decodable within the activations of most attention heads, but only a small subset of them can be effectively intervened. We introduce COMPASS, an inference-time steering method that identifies these heads using a logit-space attribution score and steers their activations along the correctness direction, requiring only per-head activation statistics. Across three model families and multiple math benchmarks, COMPASS outperforms the activation-steering baselines we compare against, improves GSM8K accuracy by 16 percentage points on average, and approaches CoT accuracy with 20-70\% fewer generated tokens. Interventions transfer without re-fitting to unseen benchmarks, and ablations show that both the correctness direction and the small set of heads carrying it are necessary, with the effect concentrated in remarkably few heads.
Chinese Translation
明确引导推理显著提高了大型语言模型（LLM）的性能。现有的方法需要对推理进行预定义的表征，无论是通过链式思维（CoT）提示设计、对比链式思维方向，还是通过衍生自自适应推理特征（SAE）的推理特征。对于具有可验证答案的数学推理，我们展示了一种更简单的信号足以满足需求，即模型自身直接回答尝试的正确性。该信号产生了一种潜在方向，能够引导推理。这个方向可以在大多数注意力头的激活中解码，但只有一小部分头可以有效干预。我们提出了COMPASS，这是一种推理时引导方法，利用对数空间归因分数识别这些头，并沿着正确性方向引导它们的激活，仅需每个头的激活统计数据。在三个模型系列和多个数学基准测试中，COMPASS的表现超过了我们比较的激活引导基线，平均提高了GSM8K的准确率16个百分点，并且在生成的标记数量上减少了20-70 ext{%}，接近链式思维的准确率。干预在未见基准上无需重新拟合，并且消融实验表明，正确性方向和承载该方向的小部分头都是必要的，效果集中在极少数头上。
cs.AI / 49 / 2610.07473
PsyCIDRA: A Dual-Agent Framework for Psychiatric Interviewing and Diagnostic Reasoning
PsyCIDRA：一种双代理框架用于精神病访谈和诊断推理
Abstract
Large language models show promise in clinical reasoning, but psychiatric interviewing requires guiding an evolving conversation. Their ability to carry out this interactive assessment remains less studied. We present PsyCIDRA, a dual-agent framework linking free-form psychiatric interviewing with diagnostic reasoning for expert review. Its interviewer agent uses tools to maintain working notes, load expert-written skills, and retrieve ICD-11 references to guide inquiry. Its diagnostic reasoning agent then receives the completed interview transcript and reports hypotheses alongside supporting, conflicting, and missing evidence, withholding a final hypothesis when none is sufficiently supported. Using patient profiles generated with PsyCPG, we first evaluate PsyCIDRA in simulation. Across four models on 53 evaluation cases, it achieves higher diagnostic agreement than direct prompting. On 81 held-out simulated cases, rank-1 accuracy is 60.5% versus 51.9%. In a blinded study of 101 human participants in separate arms, PsyCIDRA agrees with psychologists on whether to propose a diagnostic hypothesis in 79.6% of cases, compared with 65.4% for direct prompting. Together, these findings support the potential of LLM agents to assist psychiatric assessment through free-form dialogue. By examining diagnostic reasoning, interview quality, and safety together, this study contributes to understanding the capabilities and limitations of psychiatric interview agents.
Chinese Translation
大型语言模型在临床推理中展现出潜力，但精神病访谈需要引导不断发展的对话。它们在进行这种互动评估方面的能力仍然较少被研究。我们提出了PsyCIDRA，一个将自由形式的精神病访谈与专家审查的诊断推理相结合的双代理框架。其访谈代理使用工具来维护工作笔记、加载专家编写的技能，并检索ICD-11参考资料以指导询问。然后，其诊断推理代理接收完成的访谈记录，并报告假设以及支持、冲突和缺失的证据，当没有足够支持的最终假设时则不予提出。使用PsyCPG生成的患者档案，我们首先在模拟中评估PsyCIDRA。在53个评估案例中，四个模型的诊断一致性高于直接提示。在81个保留的模拟案例中，排名第一的准确率为60.5%，而直接提示为51.9%。在对101名人类参与者的盲测研究中，PsyCIDRA在79.6%的案例中与心理学家就是否提出诊断假设达成一致，而直接提示为65.4%。这些发现共同支持了大型语言模型代理通过自由形式对话辅助精神病评估的潜力。通过共同考察诊断推理、访谈质量和安全性，本研究有助于理解精神病访谈代理的能力和局限性。
cs.AI / 50 / 2610.07480
In With the Old: Enhancing 'Classical' Document Automation with Generative AI
与旧有相结合：利用生成性人工智能增强‘经典’文档自动化
Abstract
Software-based legal assistance systems have leveraged many different forms of knowledge representation and reasoning. This article explores how document automation services rooted in expert system style and other symbolic approaches can usefully enhance and be enhanced by current generative AI approaches. We discuss the possible benefits and challenges, and report on preliminary experiments in using large language models to identify and fix issues in texts written by laypeople.
Chinese Translation
基于软件的法律辅助系统利用了多种不同形式的知识表示和推理。本文探讨了根植于专家系统风格及其他符号方法的文档自动化服务如何有效地增强并被当前的生成性人工智能方法所增强。我们讨论了可能的好处和挑战，并报告了使用大型语言模型识别和修正普通人撰写文本中问题的初步实验结果。
cs.AI / 51 / 2610.07497
Does Muon Need Fine-Grained Spectral Shaping?
Muon是否需要细粒度谱形状调整？
Abstract
Muon combines current and past gradients into matrix momentum. For $M=U\Sigma V^\top$, the idealized polar update $Q=UV^\top$ gives every singular direction the same weight. We refer to this as the flat profile. Several recent optimizers replace this flat profile with fine-grained spectral maps that give each direction its own gain. We ask how much of this spectral detail a Muon update needs. Our spectral diagnostics show that approximately $94$--$97\%$ of measured singular modes lie below an estimated noise edge, yet collectively align positively with a reference gradient. We introduce BulkBoost, a two-band spectral reweighting framework with fixed-rank and noise-calibrated variants. The latter uses split-minibatch gradient differences to calibrate a Marchenko--Pastur reference edge for Muon's Nesterov input, separating the bulk below the edge from the spikes above it. Both variants increase the bulk's relative weight through one shared gain while preserving the Frobenius norm of each matrix's unreweighted direction. For a fixed partition, our theory gives the first-order condition under which moving weight toward the bulk lowers the loss. It also quantifies the fraction of the maximal first-order improvement rate, over all per-mode reallocations, that two bands can capture. Across 30 continued-pretraining settings spanning Pythia-14M to 410M and six corpora, two-band reweighting is competitive with the fine-grained power-law profile of Freon and outperforms Spectra. Measured against Muon's flat profile, Freon reduces final loss by $0.022\%$ of the pre-adaptation loss on average, whereas the two-band variants achieve reductions of $0.073$--$0.147\%$. These observations suggest that useful departures from the flat profile are surprisingly low-dimensional: a single bulk-to-spike gain captures at least as much benefit as the fine-grained spectral profiles.
Chinese Translation
Muon将当前和过去的梯度结合成矩阵动量。对于 $M=Uoldsymbol{ ext{Σ}} V^ op$，理想化的极坐标更新 $Q=UV^ op$ 赋予每个奇异方向相同的权重。我们称之为平坦轮廓。最近的一些优化器用细粒度谱图替代了这种平坦轮廓，使每个方向都有自己的增益。我们探讨Muon更新需要多少谱细节。我们的谱诊断显示，约 $94$--$97 ext{ ext{%}}$ 的测量奇异模式位于估计的噪声边缘以下，但总体上与参考梯度正相关。我们引入了BulkBoost，一个具有固定秩和噪声校准变体的双带谱重加权框架。后者使用分裂小批量梯度差异来校准Muon的Nesterov输入的Marchenko--Pastur参考边缘，将边缘以下的主体与边缘以上的尖峰分开。两个变体通过一个共享增益增加主体的相对权重，同时保持每个矩阵未加权方向的Frobenius范数。对于固定分区，我们的理论给出了将权重向主体移动以降低损失的一级条件。它还量化了两个带可以捕获的所有每模式重新分配中最大一级改进率的比例。在涵盖Pythia-14M到410M的30个持续预训练设置和六个语料库中，双带重加权与Freon的细粒度幂律轮廓具有竞争力，并且优于Spectra。与Muon的平坦轮廓相比，Freon平均降低了 $0.022 ext{ ext{%}}$ 的最终损失，而双带变体实现了 $0.073$--$0.147 ext{ ext{%}}$ 的降低。这些观察表明，从平坦轮廓中有用的偏离出乎意料地是低维的：单个主体到尖峰的增益捕获的好处至少与细粒度谱轮廓相当。
cs.AI / 52 / 2610.07505
MARS: Multi-resolution Adaptive Routing for Sequential Recommendation
MARS：用于序列推荐的多分辨率自适应路由
Abstract
Long-history recommenders often compress each user's history into a compact, candidate-independent memory that is cached and reused to score large candidate pools. We show that real user histories exhibit multi-scale semantic structure, with short-lived intent, medium-term interests, and long-term preferences coexisting in one sequence, and that monolithic cached memories preserve these scales unevenly: linear probes recover recent and mid-range content far worse than long-range content. We call this failure mode \textit{temporal aliasing}. We propose \textbf{MARS}, a multi-resolution user memory that writes the full history into recurrent state tracks anchored to different half-lives, and a sparse routing reader that materializes compact seed memories by selecting the relevant temporal resolutions for each seed, preserving fixed-size candidate scoring. MARS outperforms strong baselines on three public datasets, with gains that grow with history length. Component-matched ablations with paired tests show that temporal diversity and selective routing each contribute beyond what hard-window memories or added capacity provide. The advantage of MARS over its interface-matched baseline also widens after within-user behavioral shifts, at about $1.02\times$ that baseline's warm-cache serving latency for $1{,}000$ candidates per user.
Chinese Translation
长历史推荐系统通常将每个用户的历史压缩为一个紧凑的、与候选项无关的内存，以便缓存和重用，从而对大量候选池进行评分。我们展示了真实用户历史表现出多尺度的语义结构，短期意图、中期兴趣和长期偏好在一个序列中共存，而单一的缓存内存对这些尺度的保留不均匀：线性探测在恢复近期和中期内容方面远不如长期内容。我们称这种失败模式为 extit{时间混叠}。我们提出了 extbf{MARS}，一种多分辨率用户内存，它将完整历史写入锚定于不同半衰期的递归状态轨迹中，并且采用稀疏路由读取器，通过为每个种子选择相关的时间分辨率来实现紧凑的种子内存，从而保持固定大小的候选评分。MARS在三个公共数据集上超越了强基线，且随着历史长度的增加，性能提升更为显著。与配对测试的组件匹配消融实验表明，时间多样性和选择性路由各自的贡献超出了硬窗口内存或增加容量所能提供的优势。MARS相对于其接口匹配基线的优势在用户内部行为变化后也有所扩大，约为该基线温缓存服务延迟的$1.02 imes$，适用于每个用户$1{,}000$个候选项。
cs.AI / 53 / 2610.07509
On Open-Ended Information Seeking for Information Elicitation Agents
关于信息引出代理的开放式信息寻求
Abstract
Information elicitation is an open-ended information-seeking problem in which an interaction can unfold in many potentially valuable directions, requiring an elicitor to continually determine which information to pursue as new information emerges. In agentic elicitation, these decisions may be delegated to a foundation model, yet how model choice shapes the resulting information-seeking behavior remains understudied. We study how judgments about information value vary across LLMs and how these differences shape sequential information seeking. We first examine these judgments across 11 LLMs spanning multiple model families and parameter scales, using a shared set of information and elicitation objectives. We then develop a controlled elicitation simulation in which different models encounter the same information space and use the same selection rule, isolating these judgments from question generation and respondent behavior. Using this setting, we characterize the breadth-depth behavior that emerges from model-specific information-seeking preferences over the course of elicitation. We further examine how interaction history changes the evaluation and subsequent selection of prospective information. We test the robustness and boundaries of these findings through sensitivity analyses and ablations over the opportunities available to the elicitor, the response labels used to operationalize information-seeking preferences, the presence of interaction history, and whether redundancy is explicitly relevant to the assessment. The project code, data, and trajectory files are available at https://github.com/infosenselab/open-elicitation.
Chinese Translation
信息引出是一种开放式的信息寻求问题，其中交互可以朝着许多潜在有价值的方向展开，这要求引出者在新信息出现时不断判断应追求哪些信息。在代理引出中，这些决策可能会委托给基础模型，但模型选择如何影响最终的信息寻求行为仍然未得到充分研究。我们研究了不同大型语言模型（LLMs）对信息价值的判断如何变化，以及这些差异如何影响顺序信息寻求。我们首先在跨越多个模型家族和参数规模的11个LLMs中检查这些判断，使用一组共享的信息和引出目标。然后，我们开发了一个受控的引出模拟，其中不同模型在相同的信息空间中遇到并使用相同的选择规则，从而将这些判断与问题生成和响应者行为隔离。在这一设置中，我们描述了在引出过程中模型特定的信息寻求偏好所产生的广度-深度行为。我们进一步考察了交互历史如何改变对潜在信息的评估和后续选择。我们通过敏感性分析和对引出者可用机会、用于操作化信息寻求偏好的响应标签、交互历史的存在以及冗余是否与评估显著相关等因素的消融测试，验证了这些发现的稳健性和边界。项目代码、数据和轨迹文件可在 https://github.com/infosenselab/open-elicitation 获取。
cs.AI / 54 / 2610.07514
From Local Evidence to Safety Verdicts: Causal Tracing in Vision-Language Models
从局部证据到安全裁决：视觉-语言模型中的因果追踪
Abstract
A vision-language model may need to combine an image with a prompt to recognize a safety risk that neither reveals alone. Where does this joint safety judgment become accessible inside the model? We introduce SSU-Bench, a dataset of matched safe and unsafe image-text combinations constructed using single-item prompt edits or image edits with annotated intended regions. Using three vision-language models, we transfer internal states between paired inputs and measure the resulting change in the safety verdict. Across models and both types of counterfactual, interventions at the changed input positions are effective in earlier decoder layers, while interventions at the final input token become effective later. Directions estimated from other examples produce similar late-layer effects. A linear readout of the final-token state also predicts the model's own verdict, including incorrect judgments, and cross-model comparisons reveal similarities in the patterns of counterfactual change. These findings identify a recurring transition in where interventions can influence a joint safety verdict and distinguish a readable model decision from a correct safety judgment.
Chinese Translation
视觉-语言模型可能需要将图像与提示结合，以识别单独无法揭示的安全风险。那么，这种联合安全判断在模型内部的何处变得可访问？我们引入了SSU-Bench，这是一个通过单项提示编辑或带有注释意图区域的图像编辑构建的安全和不安全图像-文本组合的匹配数据集。使用三种视觉-语言模型，我们在配对输入之间转移内部状态，并测量安全裁决的变化。跨模型及两种反事实类型，在变化的输入位置进行干预在早期解码器层中是有效的，而在最后输入标记处进行的干预则在后期才变得有效。从其他示例估计的方向产生了类似的后层效应。最终标记状态的线性读取也能预测模型自身的裁决，包括错误判断，跨模型比较揭示了反事实变化模式的相似性。这些发现识别出干预可以影响联合安全裁决的重复转变，并区分可读的模型决策与正确的安全判断。
cs.AI / 55 / 2610.07521
Grounding What Shapes the Plan: Rethinking Groundedness for Physical Intelligence in Autonomous Driving
重塑计划的基础：重新思考自主驾驶中物理智能的基础性
Abstract
Driving models increasingly ground reasoning in causal relations, spatial structure, perceptual evidence, and predicted futures. These advances make reasoning more faithful to the driving scene, but leave a fundamental question unresolved: what should groundedness mean when the model ultimately outputs an action? Correctly grounded reasoning does not, by itself, ensure desirable driving outcomes. We introduce GroundAct, which starts from a simple premise: driving unfolds through physical entities and their interactions. Entities therefore become the unit of grounding; a lightweight reference token keeps each selected entity's continuous state addressable through symbolic reasoning; and only the referenced entities' interactions with the evolving proposal correct the plan. The result is an explicit path from what reasoning grounds to what the plan does, which we call grounded planning. To assess its practical value, we evaluate GroundAct in both open- and closed-loop settings. GroundAct shows strong open-loop planning across normal, out-of-distribution, and safety-critical scenarios, with closed-loop results extending this evidence to driving in simulation.
Chinese Translation
驾驶模型越来越多地将推理建立在因果关系、空间结构、感知证据和预测未来之上。这些进展使得推理更加忠实于驾驶场景，但仍然留下一个基本问题未得到解决：当模型最终输出一个动作时，基础性应该意味着什么？正确的基础性推理本身并不能确保理想的驾驶结果。我们引入了GroundAct，该方法基于一个简单的前提：驾驶通过物理实体及其相互作用展开。因此，实体成为基础的单位；一个轻量级的参考标记使每个选定实体的连续状态能够通过符号推理进行寻址；只有被引用实体与不断演变的提案的相互作用才能修正计划。最终结果是从推理基础到计划执行的明确路径，我们称之为基础规划。为了评估其实际价值，我们在开放和闭环设置中对GroundAct进行了评估。GroundAct在正常、超出分布和安全关键场景中显示出强大的开放环规划能力，闭环结果将这一证据扩展到模拟驾驶中。
cs.AI / 56 / 2610.07544
A Systematic Investigation of Bias in Large Language Models for Advertising Relevance
大型语言模型在广告相关性中的偏见系统性研究
Abstract
Large language models (LLMs) are increasingly used to judge how well an advertisement matches a query, but the fairness of these judgments has received limited attention. We conduct a systematic study of fairness in relevance judgments made by LLMs for queries and advertisements. Our counterfactual framework examines the effects of advertiser identity and possible popularity, input language, and demographic wording. We study GPT-4o as a categorical relevance judge and a Qwen-7B model trained specifically for relevance prediction. The advertiser and language experiments use query and advertisement pairs sampled from real advertising logs. Controlled synthetic queries are used to study demographic associations in employment, housing, and credit. For both models, changing the advertiser identity or input language can alter the relevance assessment. Selected demographic comparisons also show patterns consistent with common stereotypes, particularly those involving gender and occupation. We further study mitigation during model inference and training. The results indicate that its effectiveness depends on whether advertiser information is relevant to the query and how advertiser labels are distributed in the training data. These findings can help advertising practitioners identify fairness risks and develop suitable mitigation methods for LLM relevance systems.
Chinese Translation
大型语言模型（LLMs）越来越多地用于判断广告与查询的匹配程度，但对这些判断的公平性关注有限。我们对LLMs在查询和广告相关性判断中的公平性进行了系统研究。我们的反事实框架考察了广告主身份和可能的受欢迎程度、输入语言以及人口统计措辞的影响。我们研究了GPT-4o作为分类相关性判断者，以及专门为相关性预测训练的Qwen-7B模型。广告主和语言实验使用从真实广告日志中抽取的查询和广告对。控制合成查询用于研究在就业、住房和信贷方面的人口统计关联。对于这两个模型，改变广告主身份或输入语言会改变相关性评估。选定的人口统计比较也显示出与常见刻板印象一致的模式，特别是涉及性别和职业的刻板印象。我们进一步研究了模型推理和训练过程中的缓解措施。结果表明，其有效性取决于广告主信息是否与查询相关，以及广告主标签在训练数据中的分布情况。这些发现可以帮助广告从业者识别公平性风险，并为LLM相关性系统开发合适的缓解方法。
cs.AI / 57 / 2610.07556
Decoupled Multi-Agent Orchestration
解耦多智能体编排
Abstract
Learned orchestration can automatically construct effective language-model multi-agent systems, but existing approaches couple planning to fixed worker pools and train decomposition and collaboration from the same terminal outcome, limiting transfer and obscuring credit assignment. We introduce DeOrch, which separates worker-agnostic planning from concrete worker selection. Its two-stage planner first decomposes the task without worker information, then chooses collaboration operations using compact, worker-identity-free matchability feedback from the pool, enabling conditional credit assignment to decomposition and collaboration decisions. A lightweight matcher estimates worker suitability from behavior on a fixed probe set and adapts online with a contextual bandit, allowing new workers to be incorporated without retraining the planner or matcher. Across diverse in- and out-of-distribution tasks, DeOrch outperforms prior automatic MAS orchestration methods with fewer worker calls than competing learned orchestrators, remains effective when transferred to an entirely unseen worker pool without retraining, and shows consistent gains from both components.
Chinese Translation
学习型编排可以自动构建有效的语言模型多智能体系统，但现有方法将规划与固定的工作者池耦合，并从相同的终端结果中训练任务分解和协作，这限制了迁移能力并模糊了信用分配。我们提出了 DeOrch，它将与工作者无关的规划与具体的工作者选择分开。其两阶段规划器首先在没有工作者信息的情况下分解任务，然后使用来自池的紧凑、无工作者身份的匹配反馈选择协作操作，从而实现对分解和协作决策的条件信用分配。一个轻量级的匹配器根据在固定探测集上的行为评估工作者的适用性，并通过上下文赌博机在线适应，允许新工作者在不重新训练规划器或匹配器的情况下被纳入。 在各种分布内和分布外的任务中，DeOrch 在工作者调用次数少于竞争学习型编排器的情况下，优于先前的自动 MAS 编排方法，并且在转移到完全未见的工作者池时仍然有效，无需重新训练，并且从两个组件中都显示出一致的增益。
cs.AI / 58 / 2610.07560
Navigating Route Latent Space for Synthesizable Molecular Design
在可合成分子设计中导航路线潜在空间
Abstract
Goal-directed molecular design has advanced rapidly, yet a substantial proportion of designed molecules remain difficult to synthesize in practice, limiting their real-world utility. Prior synthesizability-aware methods either project generated molecules back to synthesizable analogs that deviate from the intended target, or optimize directly in discrete synthesis spaces that lack a continuous landscape for efficient search. We argue that this limitation mainly comes from the search space rather than the optimizer. To address this, we propose RouteFlow, a framework that reformulates synthesizable molecular design as a search over a continuous route latent space, where each latent maps back to a complete synthesis route and synthesizability is inherently preserved. To navigate this space, we adopt reward-guided flow matching as an efficient sampler that steers toward high-property regions. Since reward optimization may push latents off the manifold of real synthesis routes, where decoding becomes unreliable, we further introduce a cycle-consistency mechanism to stabilize fine-tuning. Across 16 optimization tasks from Therapeutic Data Commons, RouteFlow achieves the best sample efficiency among synthesizability-aware baselines, with the best synthetic accessibility and the highest retrosynthesis success rate. Our results also confirm that the proposed cycle-consistency reliably keeps optimization on-manifold while improving target properties, supporting effective synthesizable molecular discovery.
Chinese Translation
以目标为导向的分子设计发展迅速，但仍有相当一部分设计的分子在实际合成中难以实现，这限制了它们在现实世界中的应用。之前的合成可行性意识方法要么将生成的分子投影回偏离目标的可合成类似物，要么直接在缺乏连续搜索空间的离散合成空间中进行优化。我们认为，这一限制主要源于搜索空间，而非优化器。为了解决这个问题，我们提出了RouteFlow，一个将可合成分子设计重新表述为在连续路线潜在空间中搜索的框架，其中每个潜在向量都映射回完整的合成路线，并且合成可行性被内在保留。为了在这个空间中导航，我们采用了奖励引导的流匹配作为高效的采样器，引导搜索向高属性区域前进。由于奖励优化可能会将潜在向量推离真实合成路线的流形，从而导致解码变得不可靠，我们进一步引入了循环一致性机制来稳定微调。在来自Therapeutic Data Commons的16个优化任务中，RouteFlow在合成可行性意识的基线中实现了最佳样本效率，具有最佳的合成可达性和最高的逆合成成功率。我们的结果还确认，所提出的循环一致性机制可靠地保持了优化在流形上，同时改善了目标属性，支持有效的可合成分子发现。
cs.AI / 59 / 2610.07570
Unanimously Wrong: Certified Abstention from How Medical LLM Consensus Forms
一致性错误：医学大语言模型共识形成的认证弃权
Abstract
In clinical practice, agreement among independent experts is treated as evidence of reliability, and multi-round consensus has become a core mechanism of agentic medical question-answering systems. When such a system must decide whether to trust its own answer, the prevailing signal is again agreement, now among the sampled answers. But agreement is a fragile proxy for correctness. A system can be unanimously wrong, returning the same incorrect answer on every sample, and on these questions agreement-based signals carry no information. The cause is that these signals read only the final state of the consensus and discard how it was reached. Agreement that was reached by resolving disagreement with evidence looks identical, at the end, to agreement that was present from the first sample because every sample shares one misconception. ProbeGuard is a certified abstention framework that bases the abstention decision on how the consensus formed. Process features trace agreement trajectories, minority persistence, and retrieval saturation. For unanimous votes, rationale semantic entropy checks whether the reasons behind the vote cohere, and an active probe retrieves counter-evidence and measures whether the consensus survives. A stratified Learn-then-Test calibration then converts these scores into a distribution-free bound on selective risk. We evaluate ProbeGuard on three medical QA benchmarks and a hard-frontier reference, with a published multi-round agentic RAG substrate, against six abstention baselines. On MedQA, 13.4% of unanimous votes are wrong, and no agreement-based signal can flag them. Process signals raise the discrimination of correct from incorrect consensus from chance to 0.696 AUROC. The certified rule answers six in ten unanimous-layer questions at an observed selective risk of 9.0%, and nine in ten once in-domain calibration data accumulate.
Chinese Translation
在临床实践中，独立专家之间的共识被视为可靠性的证据，而多轮共识已成为具有自主性的医学问答系统的核心机制。当这样的系统必须决定是否信任自己的答案时，主要的信号再次是共识，现在是在采样答案之间的共识。然而，共识是正确性的脆弱代理。一个系统可能在每个样本上都返回相同的错误答案，从而一致性错误，在这些问题上，基于共识的信号没有任何信息。其原因在于这些信号仅读取共识的最终状态，而忽略了它是如何达成的。通过证据解决分歧所达成的共识在最后看起来与从第一个样本开始就存在的共识相同，因为每个样本共享一个误解。ProbeGuard是一个认证弃权框架，其弃权决策基于共识形成的方式。过程特征追踪共识轨迹、少数派持久性和检索饱和度。对于一致性投票，理由语义熵检查投票背后的理由是否一致，而主动探测则检索反证据并测量共识是否存活。分层的“先学习后测试”校准将这些分数转换为无分布的选择风险界限。我们在三个医学问答基准和一个困难前沿参考上评估ProbeGuard，使用已发布的多轮自主RAG基础设施，与六个弃权基线进行比较。在MedQA上，13.4%的统一投票是错误的，且没有基于共识的信号能够标记它们。过程信号将正确共识与错误共识的区分能力从偶然提升至0.696 AUROC。认证规则在观察到的选择风险为9.0%时回答了六成的一致性层问题，而一旦积累了领域内的校准数据，则回答九成问题。
cs.AI / 60 / 2610.07578
Cooperating with Future Collaborators: Multi-Agent RL under Staggered Participation
与未来合作者合作：错位参与下的多智能体强化学习
Abstract
In cooperative Multi-Agent Reinforcement Learning (MARL), agents are often trained under concurrent participation, while in many tasks some agents act earlier and leave task-relevant information that becomes useful to agents participating later. We study this setting as staggered participation (SP), which introduces a cross-time, cross-agent learning dependency because an early action may affect the return through the information it provides and the later policy that uses it. Learning under SP therefore requires both identifying what information is useful for future decisions and learning how later agents should use it. We propose Staggered Participation Learning (SPL), a training-time augmentation that addresses these two parts with prospective acquisition supervision for earlier agents and outcome-supervised receiver learning for later agents. We evaluate SPL across multiple policy-based MARL backbones, environments, and staggered-participation patterns. Across 60 MPE/RWARE backbone setting comparisons, SPL achieves higher observed mean task completion in every case, with an average difference of 14.1%. The gains also extend to eight-agent teams and a physics-based UAV-UGV environment in Isaac Lab, providing evidence across algorithmic, temporal, and embodied settings.
Chinese Translation
在合作的多智能体强化学习（MARL）中，智能体通常在并行参与的情况下进行训练，而在许多任务中，一些智能体会提前行动，并留下对后续参与的智能体有用的任务相关信息。我们将这种设置研究为错位参与（SP），它引入了跨时间、跨智能体的学习依赖关系，因为早期的行动可能通过提供的信息和后续政策的使用影响回报。因此，在SP下的学习需要识别哪些信息对未来决策有用，并学习后续智能体应如何使用这些信息。我们提出了错位参与学习（SPL），这是一种训练时的增强方法，针对早期智能体提供前瞻性获取监督，并为后续智能体提供结果监督学习。我们在多个基于策略的MARL骨干、环境和错位参与模式上评估SPL。在60个MPE/RWARE骨干设置比较中，SPL在每种情况下都实现了更高的平均任务完成率，平均差异为14.1%。这种提升还扩展到八智能体团队和Isaac Lab中的基于物理的无人机-地面车辆（UAV-UGV）环境，提供了在算法、时间和具身设置下的证据。
cs.AI / 61 / 2610.07580
LOGIC: An LLM Benchmark for Intent-Grounded Change Impact in Aerospace Electrical Systems
LOGIC：航空电气系统中基于意图的变更影响的LLM基准
Abstract
Aerospace electrical-design revisions can contain multiple genuine changes, although an engineering request may authorize only a subset. Propagating every detected difference can therefore produce overly broad impact reports. We present LOGIC, a controlled benchmark and evaluation framework in which locally deployable language models ground a request in a deterministic candidate-change inventory before selected changes are propagated through a typed electrical traceability graph. This separation permits candidate-selection errors to be distinguished from downstream propagation errors. LOGIC contains 168 scenarios, including 144 selection and 24 abstention cases. We evaluate three 7--8B models against intent-agnostic, lexical, and structured-evidence methods, with an oracle-root upper bound. On 96 explicitly anchored selection cases, gate-only structured evidence achieves candidate F1 of 1.0000, compared with 0.9677 for token-lexical matching. On 12 relational-paraphrase cases, token-lexical F1 is 0.1772 and gate-only F1 is 0.0000, compared with 0.5000--0.6400 for the large language models. Model grounding degrades as candidate inventories grow from 4 to 64 changes, while affected-element and typed-path accuracy remain comparatively stable when frozen selections are replayed over graphs of approximately 1K to 100K nodes. Strict evidence gating suppresses false positives but can remove correct semantic selections. An exploratory evidence-empty abstention policy raises strict abstention accuracy to 0.6667 for all three models and reduces unsafe-report rates to 0.1667, while decreasing answerable-case coverage by 16.0--27.1 percentage points. Four of six conflicting requests remain unsafe for each model. These findings support combining literal evidence and language-model reasoning with engineering review when intent cannot be established reliably.
Chinese Translation
航空电气设计修订可能包含多个真实变更，尽管工程请求可能仅授权其中的一部分。因此，传播每个检测到的差异可能会产生过于广泛的影响报告。我们提出了LOGIC，一个受控的基准和评估框架，在该框架中，本地可部署的语言模型在确定性候选变更清单中对请求进行基础处理，然后通过类型化电气可追溯性图传播所选变更。这种分离允许将候选选择错误与下游传播错误区分开来。LOGIC包含168个场景，包括144个选择案例和24个弃权案例。我们评估了三种7-8B模型，针对与意图无关、词汇和结构证据方法进行比较，并设定了一个理想上限。在96个明确锚定的选择案例中，仅门控的结构证据实现了候选F1值为1.0000，而词元-词汇匹配的F1值为0.9677。在12个关系-释义案例中，词元-词汇F1为0.1772，仅门控F1为0.0000，而大型语言模型的F1值为0.5000-0.6400。随着候选清单从4个变更增长到64个，模型的基础处理能力下降，而在大约1K到100K节点的图上重放冻结选择时，受影响元素和类型路径的准确性保持相对稳定。严格的证据门控抑制了假阳性，但可能会删除正确的语义选择。一种探索性的无证据弃权策略将所有三个模型的严格弃权准确性提高到0.6667，并将不安全报告率降低到0.1667，同时将可回答案例的覆盖率降低了16.0-27.1个百分点。每个模型中六个冲突请求中有四个仍然是不安全的。这些发现支持在无法可靠建立意图时，将字面证据与语言模型推理结合工程审查。
cs.AI / 62 / 2610.07582
Representation Bias, Correction Transfer, and Resolution Sensitivity in Three-Dimensional Mitochondrial Morphometry
三维线粒体形态测量中的表征偏差、校正转移与分辨率敏感性
Abstract
Quantitative imaging pipelines can produce precise but systematically different measurements of the same object. We present an empirical reliability assessment of three-dimensional mitochondrial morphometry that connects representation bias, a controlled processing intervention, correction transfer, and resolution sensitivity. Using 2,720 development objects from the 3D Mitochondria Shape Library for Optical Microscopy, we find that occupancy-derived volumes exceed reference mesh volumes by 3.665% on average despite an intraclass correlation coefficient of 0.994. Boundary analysis identifies an outward label displacement of 0.00304 normalized units. In a controlled label-pipeline reimplementation, removing the depth offset reduces volume error in all 55 analyzed objects by a mean of 1.57 percentage points, approximately 45% of mean reproduced inflation; the source of the remainder is not isolated. A frozen regression using occupancy-derived features reduces median absolute percentage error from 3.481% to 0.664% in 2,728 previously unused objects from the same resource. However, its calibrated error bound covers only 92.1% overall and 49.2% in a low-occupancy subgroup, demonstrating that accuracy and uncertainty transfer must be evaluated separately. In 550 rat-cortex objects from the MitoEM resource, coarsening in-plane spacing from 8 to 24 nanometers changes median surface area by minus 10.60% and sphericity by plus 11.76%, despite a rank correlation of 0.994. These results provide quantitative checks for distinguishing processing-induced descriptor changes from candidate biological differences, without establishing biological invariance or cross-source correction transfer.
Chinese Translation
定量成像流程可以产生对同一对象的精确但系统性不同的测量结果。我们提出了一种三维线粒体形态测量的实证可靠性评估，连接了表征偏差、受控处理干预、校正转移和分辨率敏感性。使用来自3D线粒体形状库（3D Mitochondria Shape Library for Optical Microscopy）的2,720个开发对象，我们发现占用率导出的体积平均超过参考网格体积3.665%，尽管类内相关系数为0.994。边界分析识别出0.00304归一化单位的外部标签位移。在受控标签流程重新实现中，去除深度偏移使所有55个分析对象的体积误差平均减少1.57个百分点，约占平均再现膨胀的45%；其余部分的来源尚未被隔离。使用占用率导出特征的冻结回归将2,728个来自同一资源的先前未使用对象的中位绝对百分比误差从3.481%降低至0.664%。然而，其校准误差界限仅覆盖92.1%的总体和49.2%的低占用子组，表明准确性和不确定性转移必须单独评估。在来自MitoEM资源的550个大鼠皮层对象中，将平面间距从8纳米粗化到24纳米使中位表面积减少10.60%，而球形度增加11.76%，尽管秩相关为0.994。这些结果为区分处理引起的描述符变化与候选生物差异提供了定量检查，而未建立生物不变性或跨源校正转移。
cs.AI / 63 / 2610.07588
Personal-Agent Mediated Recommendation with Cross-Platform User History
跨平台用户历史的个人代理中介推荐
Abstract
Modern recommendation is shifting from platform-centric personalization toward user-governed personalization, where a personal LLM agent can act on the user's behalf across services. We formalize this emerging paradigm as Personal-Agent Mediated Recommendation: a platform recommender ranks a candidate set using platform-local information, and a personal agent uses user-authorized cross-platform history to mediate the resulting ranking and produce the final top-K slate. Such mediation is nontrivial: the platform ranking can encode strong population evidence that the personal agent cannot observe, so effective mediation must therefore balance beneficial rescues against harmful overrides. To study this trade-off, we introduce MediateRec, a benchmark that includes scalable proxy cross-platform environments and a real cross-platform test under a controlled platform-agent information boundary. To train the agent to use cross-platform history effectively, we further propose Personal Attribution Mediation Optimization (PAMO), which counterfactually masks that history to estimate personal mediation support and reallocates rank-aware advantage mass under a platform-relative value floor. We theoretically prove that PAMO preserves cutoff-level advantage mass and is locally optimal among first-order reallocations that preserve this mass without lowering average platform-relative value. Experiments on MediateRec show that personal-agent mediation enables meaningful platform corrections, yet even strong proprietary LLMs introduce non-negligible harmful overrides. PAMO consistently improves over matched outcome-only RL across seen and unseen target platforms and on the real cross-platform test, while achieving a better rescue-harm balance.
Chinese Translation
现代推荐系统正从以平台为中心的个性化转向用户主导的个性化，其中个人大语言模型（LLM）代理可以代表用户在各个服务中进行操作。我们将这一新兴范式正式化为个人代理中介推荐：平台推荐系统使用平台本地信息对候选集进行排名，而个人代理则利用用户授权的跨平台历史对结果排名进行中介，从而生成最终的前K名列表。这种中介并非易事：平台排名可能编码了个人代理无法观察到的强人口证据，因此有效的中介必须在有益的救助与有害的覆盖之间取得平衡。为了研究这一权衡，我们引入了MediateRec，一个包含可扩展代理跨平台环境和在受控平台-代理信息边界下的真实跨平台测试的基准。为了训练代理有效利用跨平台历史，我们进一步提出了个人归因中介优化（Personal Attribution Mediation Optimization, PAMO），该方法反事实地屏蔽历史以估计个人中介支持，并在平台相对价值下限下重新分配排名感知优势质量。我们理论证明PAMO保留了截止水平的优势质量，并且在不降低平均平台相对价值的情况下，在保留该质量的首次重新分配中是局部最优的。在MediateRec上的实验表明，个人代理中介能够实现有意义的平台修正，但即使是强大的专有LLM也会引入不可忽视的有害覆盖。PAMO在已见和未见目标平台以及真实跨平台测试中，始终优于匹配结果的强化学习，同时实现了更好的救助-伤害平衡。
cs.AI / 64 / 2610.07592
LSC-DPO: Learning-Signal-Controlled Direct Preference Optimization
LSC-DPO：学习信号控制的直接偏好优化
Abstract
Direct Preference Optimization (DPO) has become a standard reward-model-free approach for aligning language models with preference data. However, as the scaled preference margin grows during training, the logistic DPO loss becomes progressively less sensitive to further changes. We study DPO from a loss-level geometric perspective and identify the sigmoid factor as a learning signal that characterizes the local sensitivity of the objective. Based on this view, we propose Learning-Signal-Controlled Direct Preference Optimization (LSC-DPO), which dynamically regulates the learning signal near a target regime. A log-space analysis establishes conditions for stable tracking of the target learning-signal regime. Experiments on AlpacaEval 2, MT-Bench, and Anthropic-HH show that LSC-DPO consistently improves over DPO and strong preference-optimization baselines. We further find that different coefficient initializations induce distinct transient learning-signal trajectories even when their later signal levels become similar. Based on this observation, we derive a signal-budget compensation rule that adjusts the target learning signal to compensate for these transient differences. The resulting compensation substantially reduces performance variation across coefficient initializations.
Chinese Translation
直接偏好优化（DPO）已成为一种标准的无奖励模型的方法，用于将语言模型与偏好数据对齐。然而，随着训练过程中偏好边际的扩大，逻辑DPO损失对进一步变化的敏感性逐渐降低。我们从损失层面的几何角度研究DPO，并识别出sigmoid因子作为一种学习信号，表征目标的局部敏感性。基于这一视角，我们提出了学习信号控制的直接偏好优化（LSC-DPO），该方法动态调节目标区域附近的学习信号。对数空间分析建立了稳定跟踪目标学习信号区域的条件。在AlpacaEval 2、MT-Bench和Anthropic-HH上的实验表明，LSC-DPO在性能上始终优于DPO和强偏好优化基线。我们进一步发现，不同的系数初始化会导致不同的瞬态学习信号轨迹，即使它们后期的信号水平变得相似。基于这一观察，我们推导出一种信号预算补偿规则，以调整目标学习信号，以补偿这些瞬态差异。最终的补偿显著减少了不同系数初始化之间的性能波动。
cs.AI / 65 / 2610.07601
Beyond Scalar IoU: Structured Verification from Rollout Groups for Video Temporal Grounding
超越标量 IoU：基于回滚组的结构化视频时间基础验证
Abstract
Reinforcement learning with verifiable rewards (RLVR) provides a natural framework for adapting pretrained models to video temporal grounding, where generated temporal intervals can be scored directly against ground truth intervals. Yet existing overlap verifiers typically score each rollout independently, leaving the joint structure of the rollout group unused. We introduce SUTURE, which conditions verification on the rollout group and exploits its structure at two complementary scales: disagreement across rollouts controls how strongly the target is reweighted, while coverage at each position determines where reward mass is redistributed. We show that the resulting verifier admits an exact decomposition into the standard IoU term and a covariance correction determined by the rollout group. A local gradient diagnostic finds a preference for responses covering relatively less supported target regions in the analyzed groups. Across five temporal grounding benchmarks, SUTURE improves grounding performance at every reported IoU threshold. Its trained policy also shows less video-start anchoring in reasoning traces: for later events, the first temporal mention more often overlaps the annotated target. Together, these results show that the joint structure of a rollout group can support a more informative temporal verifier.
Chinese Translation
具有可验证奖励的强化学习 (RLVR) 为适应预训练模型于视频时间基础提供了一个自然框架，其中生成的时间区间可以直接与真实区间进行评分。然而，现有的重叠验证器通常独立地对每个回滚进行评分，未能利用回滚组的联合结构。我们提出了 SUTURE，它基于回滚组进行验证，并在两个互补的尺度上利用其结构：回滚之间的不一致性控制目标的重加权程度，而每个位置的覆盖度决定了奖励质量的重新分配位置。我们展示了所得到的验证器可以精确分解为标准 IoU 项和由回滚组决定的协方差修正。局部梯度诊断发现，在分析的组中，偏好覆盖相对支持较少的目标区域。通过五个时间基础基准，SUTURE 在每个报告的 IoU 阈值下都提高了基础性能。其训练策略在推理轨迹中也表现出较少的视频起始锚定：对于后续事件，第一次时间提及更常与标注目标重叠。综合来看，这些结果表明回滚组的联合结构可以支持更具信息量的时间验证器。
cs.AI / 66 / 2610.07606
VALSE: Vertical Adaptive Layer Skipping for Efficient Inference in Large Language Models
VALSE：用于大型语言模型高效推理的垂直自适应层跳过
Abstract
This paper establishes a theoretical framework for vertical adaptive layer skipping, proving three foundational results: (i) an Expected FLOPs formula (theorem 2) giving a closed-form expression for the computational cost of arbitrary per-sample skip schedules as a function of layer-wise skip probabilities; (ii) function-space superset (theorem 10) and strict inclusion (theorem 11) theorems showing that skip-layer models are strictly contained in---yet meaningfully approximate---the full-layer function space, with an explicit separating example; and (iii) a structural duality between VALSE and Mixture-of-Experts architectures (proposition 6), positioning vertical depth-wise sparsity as the orthogonal counterpart to horizontal width-wise sparsity. Building on this theory, we propose VALSE (Vertical Adaptive Layer Skipping for Efficiency), a per-sample, non-contiguous layer skipping method: a lightweight difficulty estimator scores each input from the first few layers, and per-layer gates selectively skip redundant layers---including arbitrary middle layers while retaining deeper ones---so that only the necessary depth is activated for each input, whose feasibility is preliminarily assessed at prototype scale.
Chinese Translation
本文建立了垂直自适应层跳过的理论框架，证明了三个基础结果：(i) 期望浮点运算量（Expected FLOPs）公式（定理2），为任意每样本跳过调度的计算成本提供了一个闭合形式的表达，作为层级跳过概率的函数；(ii) 函数空间超集（定理10）和严格包含（定理11）定理，表明跳层模型严格包含于全层函数空间中，且有意义地进行近似，并提供了一个明确的分离示例；(iii) VALSE与专家混合架构（Mixture-of-Experts）之间的结构对偶性（命题6），将垂直深度稀疏性定位为水平宽度稀疏性的正交对应物。在此理论基础上，我们提出了VALSE（用于效率的垂直自适应层跳过），这是一种每样本的非连续层跳过方法：一个轻量级的难度估计器从前几层对每个输入进行评分，而每层门控选择性地跳过冗余层——包括任意中间层，同时保留更深层，以便仅为每个输入激活必要的深度，其可行性在原型规模上进行了初步评估。
cs.AI / 67 / 2610.07614
BioStudyBench: Evaluating Agents on Post-Cutoff Biomedical Studies
BioStudyBench：评估后截止日期生物医学研究中的智能体
Abstract
We evaluate whether AI agents can match the reported findings of published biomedical studies using public data. Existing evaluations do not consistently separate analysis from prior knowledge or retrieval of the published answer. We introduce BioStudyBench, a benchmark of 25 long-horizon analysis tasks drawn from studies first published between July and September 2026, after the developer-reported knowledge cutoffs of the models we evaluate, semi-automatically filtered down from 404,019 PubMed records. In each task, the agent receives a neutral research question but no data files, so it must find and download the relevant public data, search the literature through tools that return only records dated before its cutoff, and report findings through data analysis. To measure gains over prior knowledge, we run every task both with and without access to data and tools. Across eight models, access to data and tools raises the pass rate by 47 percentage points on average over the no-data baseline. Open-weight models across sizes trail closed-weight models, with the best open-weight model passing 81.3% of tasks against 94.7% for the best closed-weight model.
Chinese Translation
我们评估人工智能智能体是否能够匹配已发表生物医学研究的报告发现，使用公共数据。现有评估未能始终如一地区分分析与先前知识或已发布答案的检索。我们引入了BioStudyBench，这是一个基准，包含25个长期分析任务，来源于2026年7月至9月之间首次发表的研究，这些研究在我们评估的模型的开发者报告的知识截止日期之后，从404,019条PubMed记录中半自动过滤而来。在每个任务中，智能体接收一个中立的研究问题，但没有数据文件，因此必须找到并下载相关的公共数据，通过仅返回截止日期之前记录的工具搜索文献，并通过数据分析报告发现。为了测量相对于先前知识的提升，我们在有无数据和工具的情况下运行每个任务。在八个模型中，访问数据和工具使通过率平均提高了47个百分点，相较于无数据基线。不同规模的开放权重模型的表现落后于封闭权重模型，最佳开放权重模型的任务通过率为81.3%，而最佳封闭权重模型的通过率为94.7%。
cs.AI / 68 / 2610.07620
Explore, Then Commit: Measurement-Efficient Scientific Law Discovery with Language Models
探索后再承诺：基于语言模型的测量高效科学定律发现
Abstract
Scientific law discovery requires selecting measurements and converting evidence into a governing equation. We evaluate an explore-then-commit protocol in which a large language model proposes hypotheses, a programmatic planner gathers measurements, and a fresh prompt synthesizes the final law from fixed observations. The protocol combines structured probes, automatic numerical diagnostics, restricted measurement batches, and optional interpreter access. Across 576 NewtonBench trials, we compare eight configurations on 12 physics modules using GPT-4.1-mini and a medium-difficulty GPT-4.1 replication. On medium tasks, interpreter-enabled planners use 8.6 versus 22.5 measurements per trial for GPT-4.1-mini and 8.9 versus 43.0 for GPT-4.1. Their mean magnitude-based root-mean-squared logarithmic error falls from 2.514 to 0.202 and from 0.626 to 0.149, respectively. An additional audit retains incomplete and invalid submissions in a coverage-sensitive analysis. Observed symbolic-accuracy gains are less consistent across modules, and random acquisition is competitive with disagreement scoring. Measurement savings occur in every module, but unequal batch constraints prevent attributing them solely to acquisition quality. These results support the complete protocol as a promising measurement-efficient configuration, while leaving its causal components and generalization beyond noiseless direct-equation tasks unresolved.
Chinese Translation
科学定律发现需要选择测量并将证据转化为控制方程。我们评估了一种探索后再承诺的协议，其中大型语言模型提出假设，程序化规划者收集测量数据，而新的提示则从固定观察中综合出最终的定律。该协议结合了结构化探测、自动数值诊断、限制测量批次和可选的解释器访问。在576个NewtonBench试验中，我们使用GPT-4.1-mini和中等难度的GPT-4.1复制，比较了12个物理模块的八种配置。在中等任务中，启用解释器的规划者在每次试验中使用的测量次数分别为GPT-4.1-mini的8.6次对比22.5次，以及GPT-4.1的8.9次对比43.0次。它们的平均基于幅度的均方根对数误差分别从2.514降至0.202和从0.626降至0.149。额外的审计在覆盖敏感分析中保留了不完整和无效的提交。观察到的符号准确性提升在各模块间并不一致，而随机获取与争议评分的竞争性相当。每个模块都实现了测量节省，但不平等的批次约束使得无法将其完全归因于获取质量。这些结果支持完整协议作为一种有前景的测量高效配置，同时其因果组件和超越无噪声直接方程任务的泛化问题仍未解决。
cs.AI / 69 / 2610.07627
Learning to Outgrow a Theory: Experimental Discovery Beyond the Initial Hypothesis Space
学习超越理论：超越初始假设空间的实验发现
Abstract
Scientific discovery systems typically optimize experiments within a fixed hypothesis space. This creates a failure mode when all available candidates omit the same missing mechanism: candidate disagreement can collapse even while the model class is systematically wrong. We formulate experimental model-class revision, in which a discovery policy jointly proposes a structural edit and a diagnostic experiment that tests whether that edit is necessary. The method couples a class-level distinguishability objective, in which one shared parameterization must explain all selected experiments, with anytime-valid sequential evidence that triggers structural revision only after the current class is rejected. On 400 held-out controlled dynamical environments, the joint policy reaches 89.5% exact recovery with a budget of 32 real experiments, improving the strongest matched baseline by 10.0 percentage points while requiring fewer executed experiments and candidate fits. The learned revision-experiment pairing transfers across unseen mechanism combinations, held-out but expressible primitives, parameter extrapolation, and shifted experiment costs; when the true mechanism is outside the edit grammar, it detects library insufficiency in 88% of cases with a 5.5% false-support rate. Revision gains also transfer to ODEBench and ODEBase model-library tasks, as well as DiscoverPhysics worlds. These results support a view of scientific discovery in which deciding what mechanisms a theory should make expressible and where to collect evidence are treated as a single sequential decision problem.
Chinese Translation
科学发现系统通常在固定的假设空间内优化实验。这种方法在所有可用候选者都忽略相同缺失机制时会导致失败模式：候选者之间的分歧可能会崩溃，即使模型类别系统性错误。我们提出了实验模型类别修订的概念，其中发现策略共同提出结构编辑和诊断实验，以测试该编辑是否必要。该方法结合了类级可区分性目标，其中一个共享参数化必须解释所有选定实验，以及在当前类别被拒绝后才触发结构修订的随时有效的顺序证据。在400个保留的受控动态环境中，联合策略在32个真实实验的预算下达到了89.5%的精确恢复，较最强匹配基线提高了10.0个百分点，同时需要更少的执行实验和候选拟合。学习到的修订-实验配对能够跨越未见机制组合、保留但可表达的原语、参数外推和实验成本变化进行转移；当真实机制超出编辑语法时，它在88%的案例中检测到库不足，假阳性支持率为5.5%。修订收益也转移到ODEBench和ODEBase模型库任务，以及DiscoverPhysics世界。这些结果支持了一种科学发现的观点，即决定理论应该表达哪些机制以及在哪里收集证据被视为一个单一的顺序决策问题。
cs.AI / 70 / 2610.07634
Measuring climate backlash in Twitter and Reddit archives: Lexical definitions, recorded responses and participant turnover
测量Twitter和Reddit档案中的气候反弹：词汇定义、记录的反应和参与者流动
Abstract
Social media archives are often used to study resistance to climate action, but words, response counters and observed participants do not measure the same social process. We examine four supplied Twitter and Reddit archives by processing all registered files without sampling and applying transparent, non-exclusive lexical rules. The study links frame co-occurrence to source-specific temporal and response models, then separates event-period changes among returning authors from participant turnover. Renewable-energy terms accompany cost-related language on Reddit, yet narrower backlash phrases sharply reduce cross-source contrasts and reverse the sign of the Paris Agreement contrast in submissions. Cross-discourse history does not improve eligible primary-context forecasts. Denial/hoax terms are associated with higher recorded Twitter likes, whereas Reddit response associations depend on frame, outcome and author specification. Around the 2019 global climate strike, returning-author expression and participant turnover both contribute to increased protest-language shares. An archive endpoint prevents the corresponding Climate Twitter migration inference. Most crossed-cluster estimates lack released intervals, and joint author/month response covariance estimates fail, restricting formal inference. These results show how operational definitions, platform-specific response fields and observation boundaries shape what can be claimed about climate backlash. The contribution is an archive-based account of these measurement consequences, rather than a measure of individual opposition, persuasion or advocacy-induced backlash.
Chinese Translation
社交媒体档案常用于研究对气候行动的抵制，但词汇、反应计数和观察到的参与者并不测量相同的社会过程。我们通过处理所有注册文件而不进行抽样，并应用透明的、非排他性的词汇规则，来研究四个提供的Twitter和Reddit档案。该研究将框架共现与源特定的时间和反应模型联系起来，然后将返回作者之间的事件期变化与参与者流动分开。可再生能源术语伴随着与成本相关的语言出现在Reddit上，但更狭义的反弹短语则显著减少跨源对比，并反转《巴黎协定》提交的对比符号。跨话语历史并未改善合格的主要上下文预测。否认/骗局术语与更高的Twitter点赞记录相关，而Reddit的反应关联则依赖于框架、结果和作者的具体情况。在2019年全球气候罢工期间，返回作者的表达和参与者流动都促进了抗议语言份额的增加。档案的终点阻止了相应的气候Twitter迁移推断。大多数交叉集群估计缺乏释放区间，联合作者/月的反应协方差估计失败，限制了正式推断。这些结果表明，操作定义、平台特定的反应领域和观察边界如何塑造我们对气候反弹的主张。这一贡献是基于档案的关于这些测量后果的叙述，而不是对个体反对、劝说或倡导引发的反弹的测量。
cs.AI / 71 / 2610.07638
Learning Explainable Representations of Complex Game-playing Strategies
学习复杂游戏策略的可解释表示
Abstract
As part of learning to play complex games, human players develop develop abstractions for concepts and strategies of gameplay consistent with game rules to improve their performance. These concepts are applied to explain other players' actions, and to inform their own actions in-game. Understanding other players' strategies is a crucial part of such improvement, but requires time and effort. In this paper, we propose a strategy similar to human cognition for training RL agents to synthesize learned strategies and policies as executable procedures based on sequences of gameplay actions. We present methods to automatically learn such programs to play chess and to solve tasks in a grid-based environment. We show that the learned strategies produce effective actions, and can be learned from gameplay data.
Chinese Translation
在学习复杂游戏的过程中，人类玩家会发展出与游戏规则一致的游戏概念和策略的抽象，以提升他们的表现。这些概念被用来解释其他玩家的行为，并指导他们在游戏中的行动。理解其他玩家的策略是这种提升的关键部分，但需要时间和精力。在本文中，我们提出了一种类似于人类认知的策略，用于训练强化学习（RL）代理合成学习到的策略和政策，将其作为基于游戏动作序列的可执行程序。我们展示了自动学习这些程序的方法，以便进行国际象棋游戏和在基于网格的环境中解决任务。我们证明了学习到的策略能够产生有效的行动，并且可以从游戏数据中学习。
cs.AI / 72 / 2610.07640
Towards the Automatic Synthesis of Interpretable Chess Tactics
朝向可解释国际象棋战术的自动合成
Abstract
State-of-the-art reinforcement learning agents are capable of outperforming human experts at games like chess, Go and StarCraft II. These agents do not simply take advantage of their digital hardware in being able to react and calculate faster than humans, but employ better strategies that lead to more victories. Interpreting these strategies would give human players valuable insight into how to improve their play. In this preliminary work, we propose a symbolic sub-policy model for playing chess. Inspired by chess tactics, our model attempts to incorporate domain knowledge to improve interpretability. We adapt patterns learned by an inductive logic programming system called PAL to derive our model. We contribute a divergence metric to evaluate our model against a random baseline, and find a set of tactics that is able to suggest moves of similar playing strength to a human beginner. Finally, we propose a computational evaluation scheme for the model by augmenting an off-the-shelf engine with it.
Chinese Translation
最先进的强化学习代理能够在国际象棋、围棋和星际争霸 II 等游戏中超越人类专家。这些代理不仅仅利用其数字硬件能够比人类更快地反应和计算，还采用更好的策略，从而获得更多胜利。解读这些策略将为人类玩家提供宝贵的见解，以帮助他们改善游戏表现。在这项初步工作中，我们提出了一种用于下国际象棋的符号子策略模型。我们的模型受到国际象棋战术的启发，试图结合领域知识以提高可解释性。我们采用由一种称为 PAL 的归纳逻辑编程系统学习到的模式来推导我们的模型。我们贡献了一种发散度度量，以评估我们的模型与随机基线的表现，并发现一组战术能够建议与人类初学者相似的下棋强度的走法。最后，我们通过将现成引擎与我们的模型结合，提出了一种模型的计算评估方案。
cs.AI / 73 / 2610.07646
Matching Object or Relation? Tracing Abstract Reasoning Inside VLMs
匹配对象还是关系？追踪视觉语言模型中的抽象推理
Abstract
Vision Language Models (VLMs) excel on visual benchmarks but fail systematically on tasks requiring abstract reasoning. Existing benchmarks document this failure but cannot say \emph{why} it happens or which cognitive capability is missing. We close this gap by adopting the Relational Match-to-Sample (RMTS) paradigm from comparative and developmental psychology and pairing it with a mechanistic analysis of the model's internals. On a parametrically controlled stimulus set evaluated across frontier API models (GPT, Claude, Gemini) and three open-source families (Qwen3.5, Gemma-4, InternVL3), we identify four levers that shift VLMs toward the relational match---capability tier, model scale, the number of objects per scene, and the absence of per-object stimulus noise---together producing a developmental-like trajectory that mirrors the human \emph{relational shift}. Opening up the model, a per-layer representational similarity analysis and a causal mediation analysis reveal that VLM abstract reasoning is implemented by two competing circuits: an early circuit that organises images by their surface object features, and a late circuit that organises them by their abstract relation. Extending the analysis to ARC-AGI-1, we find that ablating the relational heads identified on RMTS degrades performance more than ablating random heads, indicating that the relational circuit is recruited beyond our controlled stimuli. We hope this mechanism-level view serves as a step toward understanding how abstract reasoning is implemented in VLMs.
Chinese Translation
视觉语言模型（VLMs）在视觉基准测试中表现出色，但在需要抽象推理的任务上系统性失败。现有基准记录了这一失败，但无法说明 extit{为什么}会发生这种情况或缺失了哪种认知能力。我们通过采用比较和发展心理学中的关系匹配样本（Relational Match-to-Sample, RMTS）范式，并结合对模型内部机制的分析，来填补这一空白。在一个参数控制的刺激集上，评估了前沿API模型（GPT、Claude、Gemini）和三个开源家族（Qwen3.5、Gemma-4、InternVL3），我们识别出四个因素，使VLMs更倾向于关系匹配——能力层级、模型规模、每个场景中的对象数量以及缺乏每个对象的刺激噪声——这些因素共同产生了类似发展的轨迹，反映了人类的 extit{关系转变}。通过对模型进行深入分析，逐层的表征相似性分析和因果中介分析揭示，VLM的抽象推理是由两个竞争电路实现的：一个早期电路根据表面对象特征组织图像，一个晚期电路根据抽象关系组织图像。将分析扩展到ARC-AGI-1，我们发现，去除在RMTS中识别的关系头比去除随机头更明显地降低了性能，表明关系电路在我们的控制刺激之外也被招募。我们希望这种机制层面的视角能为理解VLM中抽象推理的实现提供一个步骤。
cs.AI / 74 / 2610.07657
Where Rules End and Judges Begin: Measuring the Judgment Boundary in Multi-Agent Systems Security
规则的终结与法官的开始：多智能体系统安全中的判断边界测量
Abstract
LLM-based multi-agent systems (MAS) engage tools, share memory, and delegate tasks, often encountering adversarial content. Current defenses for MAS are typically evaluated in isolation, focusing on one attack type at a time, which can lead to costly and hard-to-audit outcomes. This study organizes defenses into five principles, implementing them as DEFER1 (DEterministic-First Enforcement with Residual judgment), which includes a cascade of 28 checks that blocks what it can and refers the rest to a panel of four judges. In independent testing across four domains, attack success rates drop from about 30.0% to approximately 3.0%, with 78% of blocked attacks handled by deterministic checks. Only a quarter of proposals reach the judges in the security-operations domain, illustrating that the rules provide security for attacks violating clear policies, while judges manage those that only misrepresent intent. Both systems have weaknesses, such as a risk-score approval gate that inaccurately approves most attack proposals but few legitimate ones, highlighting the challenges in assessing threats accurately.
Chinese Translation
基于大语言模型的多智能体系统（MAS）使用工具、共享内存并委派任务，常常会遇到对抗性内容。目前对MAS的防御通常是孤立评估的，集中于一次一种攻击类型，这可能导致成本高昂且难以审计的结果。本研究将防御组织为五个原则，并将其实施为DEFER1（确定性优先执行与剩余判断），该方法包括28个检查的级联，能够阻止其能阻止的内容，并将其余内容提交给四位法官进行审查。在四个领域的独立测试中，攻击成功率从约30.0%降至约3.0%，其中78%的被阻止攻击由确定性检查处理。在安全操作领域，只有四分之一的提案到达法官，这表明规则为违反明确政策的攻击提供了安全保障，而法官则管理那些仅仅误表意图的攻击。这两种系统都有其弱点，例如风险评分审批门不准确地批准大多数攻击提案，但很少批准合法提案，这突显了准确评估威胁的挑战。
cs.AI / 75 / 2610.07661
Massive Activation Gating Channel in Large Language Models
大语言模型中的大规模激活门控通道
Abstract
Massive activations, a phenomenon in which a small number of hidden channels exhibit exceptionally large magnitudes, are pervasive in large language models (LLMs). However, the mechanism by which a token develops massive activations as it propagates through a pretrained LLM remains poorly understood. In this paper, we find that the emergence of massive activations is controlled by a single channel in the input embedding to a spike feed-forward network (FFN). The position of this channel is fixed for a particular LLM. We name this channel the massive activation gating channel (MAGC). When the value of the MAGC is sufficiently large (or small, depending on the LLM), the output of the spike FFN exhibits massive activations. Examining six LLMs across four model families and different model sizes, we verify the existence and effect of MAGC. We further provide a theoretical explanation of the mechanism by which MAGC induces massive activations. When the value of MAGC is sufficiently large (or small), the output of a spike FFN asymptotically reduces to a quadratic form that mixes a few columns of the down-projection matrix of the FFN. Since these columns exhibit the shape of massive activations, the output therefore exhibits massive activations.
Chinese Translation
大规模激活是一种现象，其中少数隐藏通道表现出异常大的幅度，这在大语言模型（LLMs）中普遍存在。然而，令一个标记在经过预训练的LLM传播时产生大规模激活的机制仍然不甚清楚。本文发现，大规模激活的出现是由输入嵌入到脉冲前馈网络（FFN）中的单个通道控制的。该通道的位置对于特定的LLM是固定的。我们将这个通道命名为大规模激活门控通道（MAGC）。当MAGC的值足够大（或足够小，具体取决于LLM）时，脉冲FFN的输出会表现出大规模激活。通过对四个模型家族和不同模型规模的六个LLM进行检查，我们验证了MAGC的存在及其影响。我们进一步提供了MAGC诱导大规模激活的机制的理论解释。当MAGC的值足够大（或足够小）时，脉冲FFN的输出渐近地简化为一个二次形式，该形式混合了FFN的下投影矩阵的少数列。由于这些列表现出大规模激活的形状，因此输出也表现出大规模激活。
cs.AI / 76 / 2610.07675
EIO-Agents: The Missing Semantic Layer for AI Agent Evaluation
EIO-Agents：人工智能代理评估的缺失语义层
Abstract
AI agents are entering production in increasingly consequential environments without a shared semantic standard for what their evaluations actually mean. Scores, traces, judge outputs, and multi juror findings are increasingly used to justify readiness and release decisions, yet they often do not specify what evidence supports a claim, what that evidence can establish, or how the claim leads to a decision. We introduce EIO-Agents, an open specification for interoperable AI agent evaluation built on two layers. The Evaluation Intelligence Ontology (EIO) provides the semantic layer through typed evidence, versioned behavioral predicates, evidence contracts, claims, witness rules, proof status, recurrence, and computable derivations for metrics, findings, controls, and PASS, REVIEW, or BLOCK decisions. The Portable Evaluation Record (PER) provides the system of record: a canonical, content addressed representation of one evaluation that preserves the evidence to decision chain and can be re derived, explained, and verified. Scores summarize, juries interpret, and traces record, but none of them define what the evidence means or what it can prove. EIO provides that missing semantic contract, while PER preserves the resulting evaluation as a portable and verifiable system of record. As AI agents assume greater operational responsibility, evaluation must become more than a collection of scores and verdicts; it must become an accountable artifact whose meaning, evidence, limitations, and decisions can be independently checked.
Chinese Translation
人工智能代理正在越来越重要的环境中投入生产，但对于其评估实际意义的共享语义标准尚未建立。分数、轨迹、评审输出和多评审者的发现越来越多地被用来证明准备和发布决策的合理性，然而它们通常并未明确支持某一主张的证据是什么，这些证据能建立什么，或者该主张如何导致决策。我们提出了EIO-Agents，这是一个基于两个层次构建的可互操作的人工智能代理评估开放规范。评估智能本体（Evaluation Intelligence Ontology, EIO）通过类型化证据、版本化行为谓词、证据合同、主张、证人规则、证明状态、重复性以及可计算的度量、发现、控制和通过（PASS）、审查（REVIEW）或阻止（BLOCK）决策的推导提供了语义层。可移植评估记录（Portable Evaluation Record, PER）提供了记录系统：一种规范的、内容寻址的评估表示，保留了证据到决策的链条，并且可以重新推导、解释和验证。分数进行总结，陪审团进行解释，轨迹进行记录，但它们都没有定义证据的意义或其能证明的内容。EIO提供了这一缺失的语义契约，而PER则将结果评估保留为一个可移植且可验证的记录系统。随着人工智能代理承担更大的操作责任，评估必须超越简单的分数和裁决的集合；它必须成为一个可追责的文献，其意义、证据、局限性和决策可以被独立检查。
cs.AI / 77 / 2610.07686
BluffJAX: Adversarial Imperfect Information Games in JAX
BluffJAX：JAX中的对抗性不完全信息游戏
Abstract
We introduce BluffJAX: an open-source suite of adversarial imperfect information games in JAX. We provide canonical implementations of games designed for high simulation throughputs and parallelization on GPU accelerators. Our suite consists of well-studied benchmarks such as Texas Hold'Em Poker and Kuhn Poker, as well as games that have not been previously studied in reinforcement learning research, such as Bluff, Stud Poker, and Kemps. We hope that implementing a variety of game mechanics and difficulties will introduce new challenges and foster novel research directions in game-theoretic methods for RL. We benchmark the throughput performance and memory usage of our environments in single and multi-GPU settings, demonstrating scaling of up to hundreds of millions of samples per second, and motivating the usage of BluffJAX over related GPU and CPU-based libraries. We benchmark reinforcement learning, tree search, and game-solving algorithms in JAX in order to provide users with baseline results and facilitate future comparisons.
Chinese Translation
我们介绍了BluffJAX：一个基于JAX的开源对抗性不完全信息游戏套件。我们提供了旨在实现高仿真吞吐量和在GPU加速器上并行化的游戏的标准实现。我们的套件包括经过充分研究的基准游戏，如德州扑克（Texas Hold'Em Poker）和库恩扑克（Kuhn Poker），以及一些在强化学习研究中尚未被研究的游戏，如虚张声势（Bluff）、梭哈扑克（Stud Poker）和凯姆斯（Kemps）。我们希望通过实现多样的游戏机制和难度，引入新的挑战并促进游戏理论方法在强化学习中的新研究方向。我们在单GPU和多GPU环境中对我们的环境的吞吐性能和内存使用进行了基准测试，展示了每秒可扩展到数亿样本的能力，并激励使用BluffJAX而非相关的基于GPU和CPU的库。我们在JAX中基准测试了强化学习、树搜索和游戏求解算法，以便为用户提供基线结果并促进未来的比较。
cs.AI / 78 / 2610.07701
On the Boundary of Admission Gates: An Injected-Truth Study of Falsification-First Selection in Quantitative Strategy Research
关于入场门槛的边界：量化策略研究中的首要虚假选择的注入真相研究
Abstract
Strategy research conflates two problems: finding a profitable rule, and establishing that the finding is not search luck. The latter calls for admission gates -- statistical criteria that must be satisfied before a conclusion is adopted -- yet whether gates work, and at what cost, remains untested. We introduce an injected-truth protocol with a random-admission control that adopts at the same rate as the gate; only if the gate beats this control does it carry information rather than merely raise a threshold. Across synthetic and real-calibrated panels, gates eliminate false discoveries in the weak-signal regime but cut adoption to 1--7%, and add nothing when signals are strong. Most importantly, criteria computed on absolute rather than excess returns silently reject every candidate, including true signals. Keywords: multiple testing, backtest overfitting, strategy admission, injected-truth validation, excess returns, false discovery rate
Chinese Translation
策略研究混淆了两个问题：寻找盈利规则和确立该发现不是搜索运气。后者需要入场门槛——在采纳结论之前必须满足的统计标准——然而，门槛是否有效以及其成本尚未得到检验。我们引入了一种注入真相协议，采用随机入场控制，其采纳率与门槛相同；只有当门槛优于该控制时，它才携带信息，而不仅仅是提高了阈值。在合成和真实校准的面板中，门槛在弱信号状态下消除了虚假发现，但将采纳率降低至1%至7%，在信号强时则没有任何增益。最重要的是，基于绝对收益而非超额收益计算的标准默默地拒绝了每一个候选者，包括真实信号。关键词：多重检验，回测过拟合，策略入场，注入真相验证，超额收益，虚假发现率
cs.AI / 79 / 2610.07707
AgentMemGate: Addressing Speculation Contamination in Conversational Assistant Memory
AgentMemGate：解决对话助手记忆中的推测污染问题
Abstract
Conversational AI assistants with long-term memory extract facts from user messages into a store consulted in later conversations. A stated plan can enter that store as fact: a user who might move to Seattle may be recorded as already living there. We call this speculation contamination. Final-state memory benchmarks miss this error because they do not probe intermediate state and include few unresolved speculations. We present AgentMemGate, a write-time gate for profile-store memory that classifies extracted statements as speculation, completed event, correction, or other. Speculations remain outside memory, with conditions governing later promotion or deletion. We also contribute a dataset of multi-session conversations in which plans are confirmed, abandoned, or left unresolved. On our 147-conversation held-out set, Mem0 and Graphiti assert unresolved plans as current state for 35.2% and 27.3% of pending plans. On the core benchmark, AgentMemGate eliminates all observed contamination relative to the identical ungated pipeline (87.5% to zero for the most exposed extraction style) and raises task accuracy from 65% to 95%. On the harder held-out set, gated contamination is 3.4% to 5.7% and task accuracy rises by 9 to 13 percentage points. Our analysis identifies field matching as the main remaining bottleneck: realistic speculations often match no profile field and never reach the gate. We release our datasets, prompts, and evaluation code.
Chinese Translation
具有长期记忆的对话人工智能助手从用户消息中提取事实，以便在后续对话中进行查询。一个声明的计划可以作为事实进入该存储：一个可能搬到西雅图的用户可能被记录为已经住在那里。我们称之为推测污染。最终状态的记忆基准未能发现这一错误，因为它们不探测中间状态，并且包含很少未解决的推测。我们提出了AgentMemGate，这是一种用于个人资料存储记忆的写入时门控，能够将提取的陈述分类为推测、已完成事件、修正或其他。推测将保留在记忆之外，后续的提升或删除将受到条件的限制。我们还贡献了一个多会话对话的数据集，其中计划被确认、放弃或保持未解决。在我们147个对话的保留集上，Mem0和Graphiti将未解决的计划视为当前状态的比例分别为35.2%和27.3%。在核心基准测试中，AgentMemGate消除了相对于相同无门控管道的所有观察到的污染（从87.5%降至零，针对最暴露的提取方式），并将任务准确率从65%提高至95%。在更困难的保留集上，门控污染为3.4%至5.7%，任务准确率提高了9到13个百分点。我们的分析确定了字段匹配作为主要的剩余瓶颈：现实的推测通常与任何个人资料字段不匹配，因此无法到达门控。我们发布了我们的数据集、提示和评估代码。
cs.AI / 80 / 2610.07708
Evidence Before Sampling: Interpretable Implicit Negative Candidate Discovery for Recommendation
采样前的证据：可解释的隐式负候选发现用于推荐系统
Abstract
Recommender systems learn from observed user-item interactions, but explicit negative feedback is often unavailable. Since deep learning models require negative signals for training, negative sampling methods typically treat selected unobserved interactions as negatives. However, a missing interaction does not explain why a user is uninterested in an item or whether there is sufficient evidence to label it negative. This is especially important in business recommendation, where negative signals should be interpretable and aligned with business objectives. We formulate implicit negative candidate discovery to identify unobserved interactions supported by observed customer behavior. We encode these patterns as symbolic rules, score them based on support, informativeness, and product relevance, and rank the retained rules by evidence. An LLM then interprets the retained rules using business objectives and domain knowledge; the interpretations are combined with the statistical evidence in the final report. We evaluate our method in an industrial B2B setting and across five public recommendation datasets. Candidate-quality evaluations in the industrial setting and three public datasets show higher precision than the evaluated baselines, while symbolic selection improves downstream test PR-AUC by 12.5% over random selection with four negatives per positive example in the industrial task. Our results show that negative candidate validity can be evaluated separately from downstream recommendation performance. This distinction enables evidence-based, business-aligned, and explainable negative selection, improving both interpretability and model training in sparse, skewed, real-world recommendation settings.
Chinese Translation
推荐系统从观察到的用户-项目交互中学习，但显式的负反馈通常不可用。由于深度学习模型在训练时需要负信号，负采样方法通常将选定的未观察交互视为负样本。然而，缺失的交互并不能解释用户为何对某个项目不感兴趣，或者是否有足够的证据将其标记为负样本。这在商业推荐中尤为重要，因为负信号应当是可解释的，并与商业目标相一致。我们提出隐式负候选发现的框架，以识别由观察到的客户行为支持的未观察交互。我们将这些模式编码为符号规则，基于支持度、信息量和产品相关性对其进行评分，并根据证据对保留的规则进行排序。然后，使用大型语言模型（LLM）根据商业目标和领域知识解释保留的规则；这些解释与最终报告中的统计证据相结合。我们在工业B2B环境和五个公共推荐数据集上评估了我们的方法。工业环境和三个公共数据集中的候选质量评估显示，其精度高于评估的基线，而符号选择在工业任务中相较于随机选择每个正样本四个负样本提高了12.5%的下游测试PR-AUC。我们的结果表明，负候选的有效性可以与下游推荐性能分开评估。这一区别使得基于证据的、与商业目标一致的、可解释的负选择成为可能，从而提高了稀疏、偏斜的真实世界推荐环境中的可解释性和模型训练效果。
cs.AI / 81 / 2610.07725
PERSIST: Who-What-When Memory Across Sessions for Full-Duplex Spoken Dialogue
PERSIST：跨会话的全双工语音对话中的谁-什么-何时记忆
Abstract
Modern voice assistants may be shared by multiple users and should be able to answer questions about earlier conversations such as "When did I originally plan to leave?" or adapt their behavior to individual users based on past interactions. This requires more than retrieving a topically similar passage: the assistant must identify the current speaker, recover the relevant past state, and distinguish it from later revisions. We present PERSIST, a persistent memory system for multi-session, multi-speaker spoken dialogue that explicitly models Who, What, and When. PERSIST structures cross-session histories into readable event records and retrieves them with a 3W joint scoring mechanism that combines semantic content, acoustic speaker identity, and temporal state. For real-time full-duplex interaction, PERSIST further reuses intermediate representations from the dialogue backbone, avoiding query-audio re-encoding and reducing retrieval latency from 578.42 ms to 7.03 ms. We also introduce SpokenTrace, a diagnostic benchmark that factorizes evaluation along memory tasks and speaker-query types, exposing failures in recall, speaker attribution, and temporal-state tracking. On SpokenTrace, PERSIST achieves 85.08% end-to-end task accuracy and improves all-support EM@3 from 49.01% with BGE-large to 82.10%.
Chinese Translation
现代语音助手可能会被多个用户共享，并应能够回答有关早期对话的问题，例如“我最初计划何时离开？”或根据过去的互动调整其行为。这不仅仅需要检索主题相似的段落：助手必须识别当前的说话者，恢复相关的过去状态，并将其与后来的修订区分开来。我们提出了PERSIST，一个用于多会话、多说话者语音对话的持久记忆系统，明确建模谁、什么和何时。PERSIST将跨会话的历史结构化为可读的事件记录，并通过结合语义内容、声学说话者身份和时间状态的3W联合评分机制进行检索。为了实现实时全双工交互，PERSIST进一步重用对话骨干中的中间表示，避免查询音频的重新编码，将检索延迟从578.42毫秒减少到7.03毫秒。我们还引入了SpokenTrace，一个诊断基准，沿着记忆任务和说话者查询类型进行评估分解，揭示了在回忆、说话者归属和时间状态跟踪方面的失败。在SpokenTrace上，PERSIST实现了85.08%的端到端任务准确率，并将所有支持的EM@3从49.01%（使用BGE-large）提高到82.10%。
cs.AI / 82 / 2610.07751
How Well Do LLMs Reason with Noisy Evidence? An Active Visual Reasoning Benchmark
大型语言模型在嘈杂证据下的推理能力如何？一个主动视觉推理基准
Abstract
Real-world reasoning rarely reduces to static question answering: agents must actively gather information from tools and sensors that are often noisy and unreliable. Yet most existing active reasoning benchmarks assume that environmental feedback is trustworthy, or introduce noise without exposing an explicit, calibrated uncertainty signal, leaving open how LLMs should reason when the evidence itself is uncertain. We introduce VisualNoiseQA, a novel benchmark for active reasoning under noisy visual feedback. A text-only LLM must solve VQA problems by iteratively querying a fixed, off-the-shelf VLM treated as a stochastic visual sensor. For each query, we draw multiple samples and expose an empirical uncertainty signal via self-consistency, enabling the reasoner to probe from different angles and decide what to ask next and when to stop. Our construction is automatic and scalable: starting from diverse VQA sources and two noisy VLMs, we retain only questions where the sensor is inconsistent yet human-solvable. We evaluate multiple LLM reasoners on 1,000 instances spanning perception, chart understanding, and knowledge-intensive reasoning. VisualNoiseQA thus provides a controlled playground to study how different LLMs exploit uncertainty signals for robust reasoning.
Chinese Translation
现实世界的推理很少简化为静态问答：智能体必须主动从工具和传感器中收集信息，而这些信息往往是嘈杂和不可靠的。然而，大多数现有的主动推理基准假设环境反馈是可信的，或者在没有暴露明确、校准的不确定性信号的情况下引入噪声，这使得大型语言模型（LLMs）在证据本身不确定时该如何推理的问题悬而未决。我们引入了VisualNoiseQA，这是一个针对嘈杂视觉反馈下主动推理的新基准。一个仅限文本的LLM必须通过迭代查询一个固定的、现成的视觉语言模型（VLM），该模型被视为随机视觉传感器，来解决视觉问答（VQA）问题。对于每个查询，我们绘制多个样本，并通过自一致性暴露出经验不确定性信号，使推理者能够从不同角度探测并决定接下来要问什么以及何时停止。我们的构建是自动化和可扩展的：从多样的VQA来源和两个嘈杂的VLM开始，我们仅保留传感器不一致但人类可解答的问题。我们在涵盖感知、图表理解和知识密集型推理的1,000个实例上评估了多种LLM推理者。因此，VisualNoiseQA提供了一个受控的实验场，以研究不同LLM如何利用不确定性信号进行稳健推理。
cs.AI / 83 / 2610.07763
ST-Bench: A Spatial-Temporal Benchmark for Multi-Agent System Generation on Scientific Research Tasks
ST-Bench：用于科学研究任务的多智能体系统生成的时空基准测试
Abstract
The rapid progress of LLM-based multi-agent systems (MAS) has shown that they largely outperform single agents on coding, math, and QA tasks, where executable tests provide a binary success signal. Whether this advantage transfers to real scientific data analysis remains untested. We introduce ST-Bench, a benchmark designed to answer two questions: whether MAS outperform single agents on complex scientific data analysis tasks, and if so, by how much and at what additional cost. ST-Bench contains 100 data science tasks adapted from published Earth science studies across hydrology, agriculture, and wetland methane research, expanded into 2,067 queries grounded in additional published studies and validated by domain experts. Using ST-Bench, we evaluate five recent MAS generation methods under two training protocols, against single-agent baselines on the same GPT-5 backbone. Nine of the ten MAS configurations exceed the cheapest single-agent baseline, with the strongest reaching nearly three times its composite score. This gain is primarily attributable to coverage: trained workflows produce realistic numerical metrics on a larger fraction of queries, while the quality of those metrics, conditional on producing realistic output, is comparable to that of the single-agent baseline. The strongest configuration requires approximately four times the single-agent inference time, whereas a more economical workflow captures the majority of the benefit at less than twice the cost. MAS specialization confers measurable benefit on scientific data analysis, but the benefit is conditional rather than universal.
Chinese Translation
基于大型语言模型（LLM）的多智能体系统（MAS）的快速发展表明，它们在编码、数学和问答任务中显著优于单一智能体，这些任务的可执行测试提供了二元成功信号。然而，这种优势是否能够转移到真实的科学数据分析中仍未得到验证。我们提出了ST-Bench，这是一个旨在回答两个问题的基准测试：多智能体系统在复杂的科学数据分析任务中是否优于单一智能体，如果是，优势有多大以及需要付出什么额外成本。ST-Bench包含100个数据科学任务，这些任务改编自已发表的地球科学研究，涵盖水文学、农业和湿地甲烷研究，并扩展为2067个基于额外已发表研究的查询，并由领域专家验证。通过使用ST-Bench，我们在两个训练协议下评估了五种最新的多智能体系统生成方法，并与同一GPT-5基础模型下的单一智能体基线进行比较。在十种多智能体配置中，有九种超越了最便宜的单一智能体基线，其中最强的配置达到了其综合得分的近三倍。这一增益主要归因于覆盖率：训练后的工作流在更大比例的查询上生成了现实的数值指标，而这些指标的质量在生成现实输出的条件下与单一智能体基线相当。最强的配置大约需要单一智能体推理时间的四倍，而一种更经济的工作流在成本不到两倍的情况下捕获了大部分收益。多智能体系统的专业化在科学数据分析中带来了可测量的好处，但这种好处是有条件的，而非普遍适用。
cs.AI / 84 / 2610.07766
OTel: Open Telco AI Datasets, Benchmarks, and Models
OTel：开放电信人工智能数据集、基准和模型
Abstract
We present Open Telco (OTel), an open telecom AI resource that releases derived telecom datasets for retrieval, reranking, instruction tuning, and safety/abstention, together with 30 full-parameter post-trained baselines spanning 10 embedding models, 3 rerankers, and 17 language models. The community has already engaged substantially with the resource: as of May 3, 2026, the released models have been downloaded over 16 million times and the project has received 157+ pieces of media coverage worldwide. Building on prior open telecom datasets and benchmarks, OTel provides documented telecom data sources, held-out evaluation partitions, trained embedding models, rerankers, context-grounded LLMs, and safety/abstention data in one unified resource. Each baseline starts from an open-weight model and is post-trained on OTel-derived data using an open training recipe, then evaluated on held-out OTel evaluation partitions. OTel post-training improves performance across all three model families: embedding retrieval reaches 93.1% NDCG@10, reranking reaches 0.947 MRR@10, and language-model correctness reaches 87.8%. We release OTel as a reproducible starting point and invite the community to expand the data, improve embedding and reranking models, and build stronger context-grounded telecom LLMs.
Chinese Translation
我们介绍了开放电信（OTel），这是一个开放的电信人工智能资源，发布了用于检索、重排序、指令调优和安全/回避的衍生电信数据集，以及涵盖10个嵌入模型、3个重排序器和17个语言模型的30个全参数后训练基线。社区已经与该资源进行了 substantial 的互动：截至2026年5月3日，发布的模型已被下载超过1600万次，该项目在全球范围内获得了157多篇媒体报道。在先前开放电信数据集和基准的基础上，OTel 提供了文档化的电信数据源、保留的评估分区、训练好的嵌入模型、重排序器、基于上下文的 LLM（大语言模型）以及安全/回避数据，整合为一个统一的资源。每个基线都从一个开放权重模型开始，并使用开放训练方案在 OTel 衍生数据上进行后训练，然后在保留的 OTel 评估分区上进行评估。OTel 后训练在所有三种模型类别中提高了性能：嵌入检索达到93.1%的 NDCG@10，重排序达到0.947的 MRR@10，语言模型的正确性达到87.8%。我们将 OTel 作为一个可重复的起点发布，并邀请社区扩展数据、改进嵌入和重排序模型，并构建更强大的基于上下文的电信 LLM。
cs.AI / 85 / 2610.07781
Quantization Effects on Tool-Failure Recovery Vary Across Prompts and Evaluation Designs
量化效应对工具故障恢复的影响因提示和评估设计而异
Abstract
Post-training quantization reduces the cost of deploying language-model agents, but its effect on recovery from temporary tool failures can depend on how recovery is evaluated. We compare 8-bit and 4-bit variants of Llama-3.1-8B-Instruct and Qwen2.5-7B-Instruct on twenty deterministic tool-use tasks and five prompts. The 8-bit-4-bit recovery comparison changes direction across prompts and evaluation targets. On tasks that both variants complete without faults under the same prompt, the difference ranges from 0 to +20.2 percentage points for Llama and from -50.0 to +35.0 points for Qwen. Full-pipeline point estimates favor 8-bit Llama under all five prompts, whereas the Qwen comparison changes direction across prompts. The evaluation target can also reverse the result. For Llama under one prompt, scoring each variant only on its own clean-passing tasks favors 4-bit by 17.5 points; scoring the same tasks for both variants gives no difference, while scoring the full pipeline favors 8-bit by 28.3 points. Executor leniency is a third such choice. Rescoring the same logs with strict output parsing, which 8-bit Llama violates far more often than 4-bit Llama under that prompt, turns that +28.3 into -15.0 while leaving Qwen essentially unchanged. These findings show that one prompt, one screened task set, and one scoring policy do not establish a stable conclusion about quantized-agent robustness. Evaluations should compare variants on matched tasks, report full-pipeline success for deployment decisions, state the scoring policy, and quantify uncertainty across tasks rather than injected fault sites.
Chinese Translation
后训练量化降低了部署语言模型代理的成本，但其对临时工具故障恢复的影响可能取决于恢复的评估方式。我们比较了 Llama-3.1-8B-Instruct 和 Qwen2.5-7B-Instruct 的 8 位和 4 位变体在二十个确定性工具使用任务和五个提示上的表现。8 位与 4 位的恢复比较在不同提示和评估目标之间的方向发生变化。在同一提示下，两个变体均能无故障完成的任务中，Llama 的差异范围为 0 到 +20.2 个百分点，而 Qwen 的范围为 -50.0 到 +35.0 个百分点。在所有五个提示下，完整管道的点估计均偏向于 8 位 Llama，而 Qwen 的比较则在不同提示之间变化。评估目标也可能会逆转结果。在一个提示下，对于 Llama，仅根据其自身的清洁通过任务对每个变体进行评分，4 位变体的得分比 8 位高出 17.5 分；而对两个变体的相同任务进行评分则没有差异，而对完整管道的评分则使 8 位变体高出 28.3 分。执行者的宽容度是第三种选择。使用严格的输出解析重新评分相同的日志，在该提示下，8 位 Llama 的违规情况远多于 4 位 Llama，这将 +28.3 变为 -15.0，而 Qwen 基本保持不变。这些发现表明，单一提示、单一筛选任务集和单一评分政策并不能建立关于量化代理稳健性的稳定结论。评估应在匹配任务上比较变体，报告完整管道的成功以做出部署决策，说明评分政策，并量化任务间的不确定性，而不是注入故障位置。
cs.AI / 86 / 2610.07782
Persistent Memory in Multi-Agent LLM Inference: What It Costs, What It Buys, and When You Can Tell
多智能体大语言模型推理中的持久内存：成本、收益及其可识别性
Abstract
Decomposing long-context inference across cooperating agents bounds the active KV cache per call rather than total evidence, which matters when KV-cache memory binds. Many such systems add a persistent tier storing and recalling reasoning traces, usually validated by an ablation reporting an accuracy gain. We measure both on one three-tier agent architecture. Decomposition delivers: peak KV working set of 14.3 MiB per query against 35.5 and 35.3 MiB for single-pass and retrieval-augmented baselines. The persistent tier does not: across eight controlled dataset pairs at n=100 per arm it costs +0.368 MiB [+0.167, +0.590] of peak cache and produces no detectable accuracy change (+0.015, 95% CI [-0.011, +0.046]). We argue the null is structural: single-question benchmarks supply each item with its own evidence and score it independently, and correctness requires resetting stored traces between conditions, so recall has nothing informative to retrieve. Reaching it took four measurement corrections -- three inflating the apparent benefit, the fourth making an effect that size look resolvable -- none visible in the results table. We give the conditions an agent-memory ablation must satisfy and detection procedures that need no knowledge of the specific defect.
Chinese Translation
在合作智能体之间分解长上下文推理限制了每次调用的活动键值缓存（KV cache），而非总证据，这在键值缓存内存受限时尤为重要。许多此类系统增加了一个持久层，用于存储和回忆推理痕迹，通常通过消融实验验证其准确性提升。我们在一个三层智能体架构上进行了测量。分解实现了：每个查询的峰值键值工作集为14.3 MiB，而单次传递和检索增强基线分别为35.5 MiB和35.3 MiB。持久层则没有：在n=100的八对受控数据集上，它的峰值缓存成本为+0.368 MiB [+0.167, +0.590]，且未产生可检测的准确性变化（+0.015，95%置信区间[-0.011, +0.046]）。我们认为零假设是结构性的：单问题基准为每个项目提供了独立的证据并独立评分，而正确性要求在条件之间重置存储的痕迹，因此回忆没有任何信息可供检索。达到这一结论经历了四次测量修正——三次夸大了表面利益，第四次使得如此规模的效应看起来可解——而这些在结果表中均不可见。我们给出了智能体内存消融必须满足的条件以及不需要特定缺陷知识的检测程序。
cs.AI / 87 / 2610.07785
Attacca: Goal-Directed Control under State Continuity for Long-Horizon Embodied Agents
Attacca：基于状态连续性的目标导向控制方法用于长时间跨度的具身智能体
Abstract
A central capability of embodied agents is to accomplish complex objectives through sequences of interdependent tasks. Yet existing visual goal-conditioned policies underlying these agents are typically evaluated on isolated interactions where the target is already visible, and thus do not capture the conditions that arise during continuous long-horizon task execution. In such settings, each task begins from the state left by the previous one: the agent may end at a different position and orientation, the world may have been modified, and the next interaction target may lie outside the current field of view. As a result, agents relying on such policies may struggle to proceed to the next task when they cannot ground their target in the current observation. To address this challenge, we propose Attacca, a new approach that trains visual goal-conditioned policies on complete search-to-interact trajectories using goal images decoupled from the execution environment. Attacca uses context-decoupled goal sampling to pair each demonstration with a class-compatible masked goal image from another world, removing direct scene and pose correspondence. It learns dense current-view grounding through a target-mask prediction head, providing auxiliary supervision beyond action imitation. We further introduce behavioral-phase conditioning that teaches the policy to distinguish Search, Approach, and Interact stages and adapt its control as execution progresses. We evaluate Attacca on multiple short- and long-horizon embodied tasks in Minecraft. Our method achieves 39.0-47.5% clean success, improving over the strongest baseline by 1.7-2.4x. On long-horizon tasks, it attains 54%, 30%, and 28% completion, yielding up to a 7x improvement.
Chinese Translation
具身智能体的一个核心能力是通过一系列相互依赖的任务来完成复杂目标。然而，现有的视觉目标条件策略通常是在目标已经可见的孤立交互中进行评估，因此无法捕捉在连续长时间跨度任务执行过程中出现的条件。在这种情况下，每个任务都从前一个任务留下的状态开始：智能体可能在不同的位置和方向结束，世界可能已经被修改，下一次交互目标可能超出了当前视野。因此，依赖这些策略的智能体在无法将目标与当前观察结果相结合时，可能会难以继续进行下一个任务。为了解决这一挑战，我们提出了Attacca，一种新的方法，通过使用与执行环境解耦的目标图像，在完整的搜索-交互轨迹上训练视觉目标条件策略。Attacca使用上下文解耦的目标采样，将每个演示与来自另一个世界的类兼容的掩蔽目标图像配对，从而消除直接场景和姿态的对应关系。它通过目标-掩蔽预测头学习密集的当前视图定位，提供超越动作模仿的辅助监督。我们进一步引入行为阶段条件，使策略能够区分搜索、接近和交互阶段，并在执行过程中调整其控制。我们在Minecraft的多个短期和长期具身任务上评估了Attacca。我们的方法实现了39.0%-47.5%的干净成功率，相较于最强基线提高了1.7-2.4倍。在长期任务中，它的完成率分别达到了54%、30%和28%，实现了最高7倍的提升。
cs.AI / 88 / 2610.07787
OOPMAS: Object-Oriented Multi-Agent Systems for Query-Level Workflow Generation
OOPMAS：面向查询级工作流生成的面向对象多代理系统
Abstract
Multi-agent systems (MAS) powered by large language models have shown strong performance across code generation, mathematical reasoning, and question answering. However, existing methods for automating MAS design mostly operate at the task level, producing a single fixed workflow per benchmark that is applied uniformly to all queries. This assumption fails under realistic conditions. Query difficulty varies widely within a task, and real-world workloads mix heterogeneous task types. We introduce OOPMAS, a training-free framework that generates both the agent set and the coordination workflow at the granularity of individual queries. Agents are represented as object-oriented class definitions with dedicated roles, tools, and persistent state, and workflows are expressed as executable main functions over these agent objects. A dynamic skill library accumulates structured lessons from execution feedback across optimization rounds, enabling in-context improvement without any gradient updates or fine-tuning. On a mixed-task benchmark of queries spanning code, math, and QA, OOPMAS achieves 89.6% accuracy, outperforming the strongest baseline by 18.1 percentage points. A model-swap study across four LLM backbones shows consistent scaling, reaching 92.4% with the strongest model.
Chinese Translation
由大型语言模型驱动的多代理系统（MAS）在代码生成、数学推理和问答等方面表现出色。然而，现有的自动化MAS设计方法大多在任务层面上运作，为每个基准生成一个固定的工作流，并对所有查询统一应用。这一假设在现实条件下并不成立。查询难度在同一任务内差异很大，现实世界的工作负载混合了异构任务类型。我们提出了OOPMAS，这是一种无训练框架，能够在单个查询的粒度上生成代理集和协调工作流。代理被表示为具有专用角色、工具和持久状态的面向对象类定义，工作流则作为这些代理对象的可执行主函数来表达。动态技能库从优化轮次中的执行反馈中积累结构化的经验教训，实现了在上下文中的改进，而无需任何梯度更新或微调。在涵盖代码、数学和问答的混合任务基准上，OOPMAS达到了89.6%的准确率，比最强基线高出18.1个百分点。针对四个LLM骨干的模型替换研究显示出一致的扩展性，最强模型的准确率达到了92.4%。
cs.AI / 89 / 2610.07791
Illusory Pattern Perception Drives Spurious Inference in Large Language Models
虚幻模式感知驱动大型语言模型中的虚假推理
Abstract
Illusory pattern perception is a well-documented human cognitive tendency to infer meaningful relationships in data that is actually random. Such a tendency, often described as "connecting the dots" where none exist, can result in systematic reasoning errors. This paper investigates whether Large Language Models (LLMs) exhibit such perceptual tendencies, which can lead to systematic errors in downstream applications. To our knowledge, this work presents the first systematic study of illusory pattern perception in LLMs, adapting classic psychological paradigms to three tasks with direct empirical comparison to human behaviors. We find that LLMs frequently exhibit stronger illusory pattern perception than humans. In particular, models tend to over-associate frequent positive attributes with majority groups or large organizations, and show increased tendencies to construct causal narratives from ambiguous events. To uncover the mechanism behind these behaviors, we develop a feature interpretability framework based on Sparse Autoencoders (SAEs) to analyze internal representations. Our results reveal that holistic frequency perception and analytic cognitive orientation are linked to the emergence of illusory perceptions. These findings highlight a previously underexplored cognitive-like illusion that may affect the reliability of LLM reasoning. Code available at https://github.com/NusIoraPrivacy/illusory.
Chinese Translation
虚幻模式感知是人类认知中的一种已被充分记录的倾向，指的是在实际上是随机的数据中推断出有意义的关系。这种倾向常被描述为在不存在连接的情况下“连接点”，可能导致系统性的推理错误。本文研究了大型语言模型（LLMs）是否表现出这种感知倾向，从而导致下游应用中的系统性错误。据我们所知，这项工作是对LLMs中虚幻模式感知的首次系统研究，采用经典心理学范式对三项任务进行直接的实证比较。我们发现，LLMs经常表现出比人类更强的虚幻模式感知。特别是，模型倾向于将频繁出现的积极属性与多数群体或大型组织过度关联，并表现出从模糊事件中构建因果叙事的增强倾向。为了揭示这些行为背后的机制，我们基于稀疏自编码器（Sparse Autoencoders, SAEs）开发了一个特征可解释性框架，以分析内部表征。我们的结果显示，整体频率感知和分析性认知取向与虚幻感知的出现相关。这些发现突显了一种之前未被充分探讨的类似认知的幻觉，可能影响LLM推理的可靠性。代码可在 https://github.com/NusIoraPrivacy/illusory 获取。
cs.AI / 90 / 2610.07798
Thin Evidence, Thick Priors: How Language Models Substitute Identity for Missing Financial Facts
薄弱证据，厚重先验：语言模型如何用身份替代缺失的财务事实
Abstract
People increasingly ask large language models what to do with their money, yet seldom describe their finances in full. This paper asks what a model does with the gap. Holding finances fixed and changing only who the investor is said to be, we grade the financial evidence in the prompt from eight facts to none and measure how far the recommended equity allocation moves. Across 96,600 prompts to Llama-3.1-8B-Instruct, built from 100 financial profiles, 138 personas and seven disclosure conditions, the average gap between two personas with identical finances rises from 4.78 percentage points at full disclosure to 10.34 points with no financial facts. A two-way cluster bootstrap counting duplicated prompts once places the ratio at 2.16 (95% interval 1.69 to 2.79), and the rise is already 1.69-fold with a single fact left. Identity explains 5% of within-profile variation in advice at full disclosure and 96% with no disclosure. Household size is the only attribute whose influence grows reliably as evidence is withdrawn. Once standard errors are clustered on the persona, the unit to which identity was assigned, most attribute-specific interactions reported in the conference version lose significance, and gender instead appears as a small standing gap that full disclosure does not close. Stating risk appetite alone brings the swing into the range seen with two to seven generic facts. With no facts, the model's one-line rationale cites incomes, debts and savings it was never told, and these invented finances turn adverse more often for larger households. Inside the network, gender is linearly decodable at every layer, and ablating the gender direction at five layers leaves the aggregate identity swing unchanged. Advisory systems built on such models should be audited at the disclosure levels users actually reach, and judged across the whole identity space rather than one attribute at a time.
Chinese Translation
人们越来越多地向大型语言模型询问如何处理他们的资金，但很少完整描述他们的财务状况。本文探讨模型如何处理这一空白。在固定财务状况的情况下，仅改变所说的投资者身份，我们从八个事实到无事实对提示中的财务证据进行评分，并测量推荐的股票配置变化幅度。在基于100个财务档案、138个角色和七种披露条件构建的96,600个对Llama-3.1-8B-Instruct的提示中，两个财务状况相同的角色之间的平均差距从完全披露时的4.78个百分点上升到无财务事实时的10.34个百分点。一次性计数重复提示的双向聚类自助法将这一比例定为2.16（95%置信区间为1.69至2.79），而在仅留一个事实的情况下，差距已经达到1.69倍。身份在完全披露时解释了建议中5%的档案内变异，而在无披露时则解释了96%。家庭规模是唯一一个随着证据撤回而影响可靠增长的属性。一旦标准误差在角色上聚类，即身份被分配的单位，会议版本中报告的大多数属性特定交互失去显著性，而性别则表现为一个小的固定差距，完全披露并不能消除。单独说明风险承受能力使得波动范围接近于两到七个通用事实的范围。在没有事实的情况下，模型的单行理由引用了它从未被告知的收入、债务和储蓄，而这些虚构的财务状况对于较大家庭的影响更为不利。在网络内部，性别在每一层都是线性可解的，而在五层中去除性别方向不会改变整体身份波动。基于此类模型构建的咨询系统应在用户实际达到的披露水平上进行审计，并在整个身份空间中进行评估，而不是一次评估一个属性。
cs.AI / 91 / 2610.07803
ThinkFuse: Trajectory-Aware Test-Time Fusion for Small Reasoning Models
ThinkFuse：面向轨迹的小型推理模型测试时融合方法
Abstract
Small reasoning models (SRMs) have shown strong performance on complex reasoning tasks by generating extended chain-of-thought trajectories, but they often fail to recover once their reasoning enters an erroneous path. Existing test-time fusion methods rely on local fusion signals to determine when to trigger fusion, which can be misled by transient uncertainty fluctuations and may reinforce unstable reasoning trajectories. We propose ThinkFuse, a training-free test-time fusion framework that selectively intervenes in unreliable reasoning segments. ThinkFuse compares segment-level uncertainty shifts with trajectory-level uncertainty trends to identify unstable reasoning points and fuse auxiliary reasoning paths into the primary model's trajectory. Extensive experiments demonstrate that ThinkFuse outperforms baselines on mathematical and knowledge-intensive reasoning benchmarks, with consistent gains across model-family combinations, and remains robust with a smaller primary model. Our analysis shows that ThinkFuse requires fewer fusion triggers and generates fewer tokens, highlighting the efficiency of selective triggering. Our code is available at https://github.com/js-lee-AI/ThinkFuse.
Chinese Translation
小型推理模型（SRMs）在复杂推理任务中表现出色，通过生成扩展的思维链轨迹，但一旦推理进入错误路径，它们往往难以恢复。现有的测试时融合方法依赖于局部融合信号来判断何时触发融合，这可能会受到瞬时不确定性波动的误导，并可能加强不稳定的推理轨迹。我们提出了ThinkFuse，一个无训练的测试时融合框架，选择性地干预不可靠的推理片段。ThinkFuse通过比较片段级不确定性变化与轨迹级不确定性趋势，识别不稳定的推理点，并将辅助推理路径融合到主模型的轨迹中。大量实验表明，ThinkFuse在数学和知识密集型推理基准测试中优于基线模型，在模型家族组合中表现出一致的提升，并且在较小的主模型下仍然保持稳健。我们的分析显示，ThinkFuse需要更少的融合触发，并生成更少的标记，突显了选择性触发的效率。我们的代码可在 https://github.com/js-lee-AI/ThinkFuse 获取。
cs.AI / 92 / 2610.07816
Do I Need the Cloud? Uncertainty-Aware Step-Level Handoff for Small Language Model Agents
我需要云计算吗？面向不确定性的逐步切换框架用于小型语言模型代理
Abstract
Small language models (SLMs) are attractive as local agent controllers because they reduce remote inference, latency, and deployment footprint, yet structured tool errors can cause an agent step to fail. Existing routers typically select a model once per query. However, agents expose sequential decision points whose difficulty dynamically changes based on intermediate observations. We propose STEPGATE, an uncertainty-aware handoff framework that scores each local SLM action and selectively escalates challenging steps to a stronger model. On a 52-task held-out single-step BFCL-derived test split, the Qwen2.5-1.5B/7B pair attains 82.7% task success with 30.8% escalation, versus 67.3% local-only and 75.4% random escalation (which uses 33.8% escalation). In a separate multi-turn evaluation, STEPGATE achieves 69.0% trajectory success and 84.0% action success using only 30.0% cloud actions, compared with 48.0%/70.5% local-only, 60.0%/78.2% random escalation, and 57.0%/77.1% query-level routing (strong-only achieves 82.0% trajectory success at 100% cloud actions). These results suggest that step-level escalation recovers a large share of the performance gap to the stronger Qwen2.5-7B backend at a matched cloud-action rate while transmitting fewer tokens remotely. However, our evaluation is limited to one model family, a single stronger backend, and scripted tasks. Furthermore, the test sets are small, multi-turn comparisons rely on paired intervals and statistical tests, and our risk tiers serve as research annotations rather than formal safety guarantees.
Chinese Translation
小型语言模型（SLMs）作为本地代理控制器具有吸引力，因为它们减少了远程推理、延迟和部署占用空间，然而结构化工具错误可能导致代理步骤失败。现有的路由器通常在每个查询中选择一次模型。然而，代理在决策过程中暴露出顺序决策点，其难度会根据中间观察动态变化。我们提出了STEPGATE，一个面向不确定性的切换框架，该框架对每个本地SLM动作进行评分，并选择性地将具有挑战性的步骤升级到更强的模型。在一个包含52个任务的保留单步BFCL派生测试集上，Qwen2.5-1.5B/7B组合在30.8%的升级率下达到了82.7%的任务成功率，而仅使用本地模型的成功率为67.3%，随机升级的成功率为75.4%（随机升级使用了33.8%的升级率）。在一个单独的多轮评估中，STEPGATE在仅使用30.0%的云动作的情况下实现了69.0%的轨迹成功率和84.0%的动作成功率，而仅使用本地模型的成功率为48.0%/70.5%，随机升级的成功率为60.0%/78.2%，查询级路由的成功率为57.0%/77.1%（仅强模型在100%云动作下实现了82.0%的轨迹成功率）。这些结果表明，逐步升级在匹配的云动作率下恢复了与更强的Qwen2.5-7B后端之间的性能差距，同时远程传输的令牌数量更少。然而，我们的评估仅限于一个模型系列、一个单一的强后端和脚本化任务。此外，测试集较小，多轮比较依赖于配对间隔和统计测试，我们的风险等级作为研究注释，而非正式的安全保证。
cs.AI / 93 / 2610.07829
Agentic Semantic Sensing for Resource-Adaptive AI-RAN
面向资源自适应AI-RAN的自主语义感知
Abstract
Semantic sensing (SemS) acquires task-relevant information rather than reconstructing complete physical information. Existing SemS formulations typically operate open loop: sensing configurations and observation schedules are fixed before inference and cannot respond to evolving task-level evidence. We propose Agentic SemS, a closed-loop framework for AI-enabled radio access networks (AI-RANs) that controls sensing within a communication-feasible profile set. A profile-conditioned causal Transformer updates the semantic belief from streaming observations, while key-value caching enables efficient state updates across profile changes without repeatedly processing the complete history. A semantic utility network estimates the task-level benefit of acquiring the next observation block under each feasible profile after accounting for sensing cost. The resulting continuation utilities jointly support next-profile selection and semantic early exit, adapting sensing configuration and duration to evolving evidence. The expected semantic gain is further related to conditional mutual information, providing a value-of-information interpretation of continued online sensing. Experiments on Widar3.0 with six emulated sensing profiles show that, in comparison with full-sequence High, the resource-efficient Agentic setting reduces normalized cumulative sensing cost by 25.33% while achieving 85.79% Macro-F1. At the same utility checkpoint, semantic early exit provides a further 12.35% cost reduction over adaptive sensing without early exit, with a 0.97-percentage-point Macro-F1 decrease.
Chinese Translation
语义感知（Semantic sensing, SemS）获取与任务相关的信息，而不是重构完整的物理信息。现有的SemS公式通常采用开放循环：感知配置和观察调度在推理之前是固定的，无法响应不断变化的任务级证据。我们提出了自主语义感知（Agentic SemS），这是一个针对人工智能驱动的无线接入网络（AI-RAN）的闭环框架，能够在通信可行的配置集中控制感知。一个基于配置的因果Transformer从流式观察中更新语义信念，而关键值缓存则使得在配置变化时能够高效地更新状态，而无需重复处理完整历史。语义效用网络在考虑感知成本后，评估在每个可行配置下获取下一个观察块的任务级收益。由此产生的延续效用共同支持下一个配置的选择和语义提前退出，适应不断变化的证据，调整感知配置和持续时间。预期的语义增益进一步与条件互信息相关联，为持续在线感知提供了信息价值的解释。在Widar3.0上进行的实验中，使用六个模拟感知配置表明，与完整序列High相比，资源高效的自主设置将归一化累积感知成本降低了25.33%，同时实现了85.79%的宏观F1值。在相同的效用检查点，语义提前退出在没有提前退出的自适应感知基础上进一步提供了12.35%的成本降低，但宏观F1值下降了0.97个百分点。
cs.AI / 94 / 2610.07835
DHCG: Dynamic Construction of Hierarchical Collaboration Graphs for LLM-Based Multi-Agent Reasoning
DHCG：基于大语言模型的多智能体推理的动态层次协作图构建
Abstract
LLM-based multi-agent systems (MAS) have demonstrated strong capabilities in solving complex problems across diverse domains. Recently, the dynamic orchestration of agent systems has become an important research direction. However, existing methods suffer from limited composition, misaligned dependencies, and inflexible scale, restricting their ability to adapt to reasoning requirements during execution. To address these limitations, we reframe MAS design as a partially observable Markov decision process, in which both the composition and scale of the MAS are dynamically determined. We propose DHCG, a novel framework that coordinates three modules (Planner, Worker, and Generator) to progressively construct a dynamic hierarchical collaboration graph from scratch based on the query and evolving execution feedback. At each step, guided by feedback, the Planner generates a set of distinct and complementary roles tailored to the current reasoning needs and selectively routes relevant information to each role. It can also finalize the hierarchical collaboration graph early or progressively expand it when additional reasoning is required. We further introduce action-aware preference optimization to train the Planner to make more effective decisions when constructing hierarchical collaboration graphs. We systematically evaluate DHCG across code generation, mathematical reasoning, and domain-specific reasoning benchmarks. DHCG achieves state-of-the-art average performance among the compared methods, improving over the single-agent baseline by 13.06 points and outperforming both static and dynamic MAS baselines by 2.77-8.02 points. Additional experiments further demonstrate its generalization across different Planner backbones and unseen Worker models.
Chinese Translation
基于大语言模型的多智能体系统（MAS）在解决各个领域的复杂问题方面展现出了强大的能力。最近，智能体系统的动态编排已成为一个重要的研究方向。然而，现有方法存在组合能力有限、依赖关系不匹配和规模不灵活等问题，限制了它们在执行过程中适应推理需求的能力。为了解决这些局限性，我们将MAS设计重新框定为一个部分可观测的马尔可夫决策过程，其中MAS的组合和规模都是动态确定的。我们提出了DHCG，一个新颖的框架，协调三个模块（规划者、工作者和生成器），根据查询和不断演变的执行反馈，从零开始逐步构建动态层次协作图。在每一步中，在反馈的指导下，规划者生成一组独特且互补的角色，以满足当前的推理需求，并选择性地将相关信息路由到每个角色。当需要额外的推理时，它还可以提前确定层次协作图或逐步扩展它。我们进一步引入了基于动作的偏好优化，以训练规划者在构建层次协作图时做出更有效的决策。我们在代码生成、数学推理和特定领域推理基准上系统地评估了DHCG。DHCG在比较方法中实现了最先进的平均性能，相较于单智能体基线提高了13.06分，并且在静态和动态MAS基线中超越了2.77-8.02分。额外的实验进一步证明了其在不同规划者骨干和未见工作者模型中的泛化能力。
cs.AI / 95 / 2610.07851
RA-MoWE: Workflow-Affinity Embeddings for Query Clustering and Agentic Workflow Generation
RA-MoWE：用于查询聚类和自主工作流生成的工作流亲和嵌入
Abstract
Agentic workflows enable large language models (LLMs) to solve complex tasks by coordinating reasoning, tool use, and verification. However, a workflow optimized for an entire task collection can overlook differences in the reasoning strategies that individual queries need, while searching for a new workflow for every query repeats costly optimization. To address this tradeoff, we introduce RA-MoWE, a framework that uses workflow-affinity embeddings to cluster queries and guide the generation of reusable expert workflows. Each embedding records how well a fixed set of reference workflows solves a query, revealing similarities in which reasoning strategies are effective. RA-MoWE uses each cluster's queries and average embedding to initialize and refine a specialized workflow through execution feedback. An embedding encoder predicts these embeddings from query text, allowing new queries to select a generated expert without first executing the reference workflows. On a 300-query test set drawn from four benchmarks spanning mathematics, science, and programming, RA-MoWE improves average task score by 4.04 percentage points over selecting among the reference workflows, while using 27.7% fewer language-model calls at inference.
Chinese Translation
自主工作流使大型语言模型（LLMs）能够通过协调推理、工具使用和验证来解决复杂任务。然而，针对整个任务集合优化的工作流可能会忽视个别查询所需的推理策略差异，而为每个查询寻找新的工作流则会重复进行成本高昂的优化。为了解决这一权衡问题，我们提出了RA-MoWE，一个利用工作流亲和嵌入来聚类查询并指导可重用专家工作流生成的框架。每个嵌入记录了固定参考工作流集解决查询的效果，揭示了有效推理策略的相似性。RA-MoWE利用每个聚类的查询和平均嵌入，通过执行反馈来初始化和优化专门的工作流。嵌入编码器从查询文本中预测这些嵌入，使新查询能够在不先执行参考工作流的情况下选择生成的专家。在从数学、科学和编程等四个基准中提取的300个查询测试集上，RA-MoWE在选择参考工作流的基础上提高了平均任务得分4.04个百分点，同时在推理时减少了27.7%的语言模型调用次数。
cs.AI / 96 / 2610.07860
WorkflowOps: Learning Agent Collaboration Priors for Multi-Agent Workflow Orchestration
WorkflowOps：用于多智能体工作流编排的学习智能体协作先验
Abstract
Multi-agent systems are increasingly deployed for complex knowledge work, yet their orchestration layers remain largely memoryless: each new task is decomposed, assigned, and executed from scratch with no benefit from prior successful executions. We present WorkflowOps, a multi-agent workflow orchestration framework that learns agent collaboration priors from historical workflows and expands its agent pool on demand to cover new capability requirements. Our approach introduces three coupled mechanisms. First, a transition probability matrix captures pairwise agent collaboration frequencies from past workflows and applies them as soft guidance during DAG workflow construction through intra-layer ordering optimization, probability-thresholded edge suggestion, and transitive reduction for parallelism maximization. Second, a sufficiency-driven agent creation loop detects capability gaps via semantic matching scores, generates specialized agents through an LLM, and simultaneously injects them into the collaboration matrix, so that newly created agents are immediately usable with predicted collaboration priors. Third, a layered semantic matching strategy uses pre-trained sentence embeddings for fast, deterministic capability matching as a first pass, invoking LLM verification only for low-confidence cases, thereby reducing LLM routing calls by over 80\% compared to pure-LLM approaches. Experiments on mixed code, math, and question-answering suites show that WorkflowOps improves end-to-end pass rates over recent workflow-construction baselines, with the largest gains on structured, decomposable tasks where past agent handoff patterns transfer.
Chinese Translation
多智能体系统在复杂知识工作中的应用日益增多，但其编排层在很大程度上仍然是无记忆的：每个新任务都是从头开始进行分解、分配和执行，无法从之前的成功执行中获益。我们提出了WorkflowOps，一个多智能体工作流编排框架，能够从历史工作流中学习智能体协作先验，并根据需求扩展其智能体池以满足新的能力要求。我们的方法引入了三个相互关联的机制。首先，转移概率矩阵捕捉了过去工作流中成对智能体协作的频率，并在有向无环图（DAG）工作流构建过程中通过层内排序优化、概率阈值边建议和传递性约简来作为软指导，以最大化并行性。其次，基于充分性的智能体创建循环通过语义匹配得分检测能力差距，通过大型语言模型（LLM）生成专业化智能体，并同时将其注入协作矩阵，从而使新创建的智能体能够立即使用预测的协作先验。第三，分层语义匹配策略使用预训练的句子嵌入进行快速、确定性的能力匹配作为第一步，仅在低置信度情况下调用LLM验证，从而将LLM路由调用减少超过80 ext{%}，相较于纯LLM方法。针对混合代码、数学和问答套件的实验表明，WorkflowOps在最近的工作流构建基准上提高了端到端通过率，尤其在结构化、可分解任务中，过去的智能体交接模式转移带来了最大的收益。
cs.AI / 97 / 2610.07881
Self-Referenced Social Preferences: Cooperation without Observing Others Rewards
自我参照的社会偏好：在不观察他人奖励的情况下实现合作
Abstract
Social preferences can promote cooperation in multi-agent reinforcement learning, but existing approaches often require agents to observe the rewards of their peers. In many real-world interactions, however, an agent can, as humans do, observe others' behavior and outcomes without access to their private reward signals. We introduce self-referenced social preferences, in which each agent learns a model of its own reward, applies it to other agents' observed transitions to assess their outcomes from its own perspective, and feeds these self-referenced assessments into standard social preferences. We study two ways to incorporate these assessments: modifying the learning reward, or using them to weight policy updates. We evaluate the approach on three sequential social dilemmas, Escape Room, Clean Up, and Commons Harvest, which require volunteering, public-good contribution, and resource restraint, respectively. Across all three environments, agents learn cooperative behavior without observing others' rewards, including in settings where independent learners fail to cooperate, and frequently achieve more equitable divisions of jointly produced returns than agents with access to true rewards. The effective integration point depends on the social preference: inequity aversion works best in the reward together with a value look-ahead, whereas a purely benevolent preference benefits from policy-update weighting. Under partial observability, the policy-update approach continues to support cooperation. These results show that explicit access to other agents' reward signals is not necessary for learning cooperative behavior: social preferences can instead be grounded in self-referenced assessments of others' outcomes derived from their observed behavior.
Chinese Translation
社会偏好可以促进多智能体强化学习中的合作，但现有的方法通常要求智能体观察其同伴的奖励。然而，在许多现实世界的互动中，智能体可以像人类一样，在没有访问其私人奖励信号的情况下观察他人的行为和结果。我们引入了自我参照的社会偏好，其中每个智能体学习自己的奖励模型，将其应用于其他智能体观察到的转变，以从自身的角度评估他们的结果，并将这些自我参照的评估反馈到标准的社会偏好中。我们研究了两种整合这些评估的方法：修改学习奖励或利用它们来加权策略更新。我们在三个顺序社会困境中评估该方法：逃生室（Escape Room）、清理（Clean Up）和公共收获（Commons Harvest），这些困境分别要求志愿服务、公共物品贡献和资源克制。在所有三个环境中，智能体在不观察他人奖励的情况下学习到合作行为，包括在独立学习者无法合作的情况下，且经常实现比访问真实奖励的智能体更公平的共同收益分配。有效的整合点取决于社会偏好：不平等厌恶在奖励与价值前瞻结合时效果最佳，而纯粹的慈善偏好则受益于策略更新加权。在部分可观察性下，策略更新方法仍然支持合作。这些结果表明，显式访问其他智能体的奖励信号并不是学习合作行为的必要条件：社会偏好可以基于从观察到的行为中得出的他人结果的自我参照评估。
cs.AI / 98 / 2610.07886
ShanLiangRen: A Nutrition Agent for Personalized Daily Meal Planning
ShanLiangRen：个性化每日膳食规划的营养代理
Abstract
Dietary nutrition planning plays an important role in chronic disease management and maintaining a healthy body. In applications, it must simultaneously satisfy personalized constraints and reasonable multidimensional nutritional goals. These two aspects often conflict, and user constraints evolve with feedback, resulting in a substantial gap between generic guidelines and executable plans. To bridge this gap, we first propose the personalized fully quantified multiobjective dietary planning problem (MDP). To tackle MDP, we develop a nutrition agent, ShanLiangRen. The system first transforms dietary specifications, nutrient data, user attributes and natural language requirements into an individualized constrained planning instance. It then employs an exact retrieval-augmented generation method to shrink the feasible candidate set from a large scale ingredient and recipe space. Finally, it adopts a refinement guided by Pareto principles, where an LLM iteratively revises candidate plans under deterministic nutrition computation and feedback from constraint verification. The system outputs fully quantified meal plans with explicit ingredients and portion sizes, together with reports on nutrition compliance that show constraint satisfaction and nutrient interval attainment. We have released the system online as a WeChat Program, ShanLiangRen. A demo video is available at https://www.youtube.com/watch?v=652OtY5VlGA.
Chinese Translation
膳食营养规划在慢性疾病管理和维持健康身体中发挥着重要作用。在实际应用中，它必须同时满足个性化约束和合理的多维营养目标。这两个方面往往存在冲突，并且用户约束随着反馈而演变，导致通用指南与可执行计划之间存在显著差距。为了弥合这一差距，我们首先提出了个性化完全量化的多目标膳食规划问题（MDP）。为了解决MDP，我们开发了一个营养代理——ShanLiangRen。该系统首先将膳食规范、营养数据、用户属性和自然语言需求转化为个性化的受限规划实例。然后，它采用精确的检索增强生成方法，从大规模的食材和食谱空间中缩小可行候选集。最后，它采用基于帕累托原则的细化方法，其中大型语言模型（LLM）在确定性营养计算和约束验证反馈的指导下，迭代修订候选计划。该系统输出完全量化的膳食计划，明确列出食材和份量，同时提供营养合规性报告，展示约束满足和营养区间达成情况。我们已将该系统在线发布为微信小程序ShanLiangRen，演示视频可在https://www.youtube.com/watch?v=652OtY5VlGA观看。
cs.AI / 99 / 2610.07895
Textual Environmental Context and Spatial Graphs for LLM-Based Regional SST Forecasting
基于文本环境上下文和空间图的LLM区域海表温度预测
Abstract
Sea surface temperature (SST) forecasting depends on local temporal persistence, regional spatial dependence, and environmental conditions that evolve with the forecast date. We study how these heterogeneous conditions can be presented to a large language model (LLM) for regional multi-step forecasting without serializing the full SST grid as text. We formulate forecasting as conditional numerical generation: historical SST and anomaly sequences, date-aligned environmental records, and static ocean knowledge form a textual context, while regional spatial state is supplied through continuous graph-derived prefixes. A static graph encodes persistent geographic--climatological relations, and a dynamic graph encodes recent SST correlations and localized tropical-cyclone influence. Two graph neural networks produce a target-node representation that is mapped by a spatial-prefix fusion and injected into the LLM input. On SST forecasting in the South China Sea, the complete configuration achieves the best MAE and $\Rtwo$ among the compared methods over ten forecast steps. Alongside the numerical forecast, a rule-based module matches predicted trends and environmental-factor directions with knowledge entries to return source-linked, post-hoc contextual explanations.
Chinese Translation
海表温度（SST）预测依赖于局部时间持续性、区域空间依赖性以及随预测日期演变的环境条件。我们研究如何将这些异质条件呈现给大型语言模型（LLM），以实现区域多步预测，而无需将完整的SST网格序列化为文本。我们将预测形式化为条件数值生成：历史SST和异常序列、日期对齐的环境记录以及静态海洋知识构成了文本上下文，而区域空间状态则通过连续的图生成前缀提供。静态图编码了持久的地理-气候关系，而动态图则编码了近期SST相关性和局部热带气旋影响。两个图神经网络生成目标节点表示，该表示通过空间前缀融合映射并注入LLM输入。在南海的SST预测中，完整配置在比较方法中在十个预测步骤中实现了最佳的平均绝对误差（MAE）和$ ext{R}^2$。除了数值预测外，基于规则的模块将预测趋势和环境因素方向与知识条目匹配，以返回源链接的事后上下文解释。
cs.AI / 100 / 2610.07906
Isotropic Yet Undecodable: The Sequential Content-Sufficiency Gap in Latent-Predictive Text Representations
各向同性但不可解码：潜在预测文本表示中的顺序内容充足性差距
Abstract
We study sequential content sufficiency by investigating whether a representation retains the ordered target information available in its input. An information-theoretic decomposition separates input ambiguity, representation loss, and readout mismatch. We construct recoverable views where perfect agreement and joint isotropic Gaussianity coexist with zero target information, and establish limits imposed by deterministic canonical anchors. Token log-loss provides a one-sided information-loss bound; a fixed-penalty ridge analysis shows why rank alone cannot determine prediction risk. These results motivate CANOPE, a nonautoregressive framework with ordered latent canvases, canonical-token supervision, and geometric regularization. On 40,000 validation sequences, latent-agreement (PL0) and token-grounded (PL2) have nearly identical pooled ranks but reach 13.5% and 98.8% positional Recall@1, respectively, under strong natural corruption when the correct target length is provided. On 3,930 LJSpeech validation utterances, frozen PL2 with a trained MatchaTTS readout yields 21.54% word error rate (WER) on corrupted text, versus 99.22% for frozen PL0, while end-to-end MatchaTTS reaches 10.93%. These results show that geometric regularity alone does not guarantee recoverable sequential content or effective downstream access in the text settings studied here.
Chinese Translation
我们通过研究表示是否保留输入中可用的有序目标信息来研究顺序内容充足性。信息论分解将输入模糊性、表示损失和读取不匹配分开。我们构建了可恢复的视图，其中完美一致性和联合各向同性高斯性与零目标信息共存，并建立了由确定性典型锚点施加的限制。令牌对数损失提供了一种单侧信息损失界限；固定惩罚岭分析表明，仅凭秩无法确定预测风险。这些结果激励了CANOPE，一个具有有序潜在画布、典型令牌监督和几何正则化的非自回归框架。在40,000个验证序列中，潜在一致性（PL0）和令牌基础（PL2）具有几乎相同的汇总秩，但在提供正确目标长度的情况下，在强自然干扰下分别达到13.5%和98.8%的位置召回率@1。在3,930个LJSpeech验证语句中，冻结的PL2与训练的MatchaTTS读取在损坏文本上产生21.54%的词错误率（WER），而冻结的PL0则为99.22%，而端到端的MatchaTTS达到10.93%。这些结果表明，仅凭几何规律并不能保证在这里研究的文本环境中恢复顺序内容或有效的下游访问。
cs.AI / 101 / 2610.07907
Continuous Memory Machines
连续记忆机器
Abstract
Recurrent neural networks typically compress information into a single vector-valued recurrent state, forcing short-term computation and long-term retention to share the same representation. Past extensions alleviate this bottleneck by increasing the memory capacity or separating timescales, but lack the combination of rapid neuron-level processing and longer-term retention found in biology. To that end, we introduce the Continuous Memory Machine (CMM), a recurrent architecture with matrix-valued short- and long-term memory states serving distinct functional roles. Building on the Continuous Thought Machine (CTM), the CMM's short-term memory tracks recent neural activity, with uniquely parameterized neuron-level models learning to use these activity patterns for computation. A persistent long-term memory stores information for later use, with a Transformer jointly updating both memory stores, providing an expressive bidirectional read--write mechanism such that each store can reorganize its own contents and both read from and write to the other. Across algorithmic, in-context learning, and recurrent reasoning tasks, the CMM outperforms a broad suite of baselines, exhibiting stronger generalization than prior memory-augmented networks while preserving the CTM's interpretable attention patterns. Code is available at https://github.com/SakanaAI/continuous-memory-machines.
Chinese Translation
递归神经网络通常将信息压缩为单一的向量值递归状态，迫使短期计算和长期记忆共享相同的表示。过去的扩展通过增加记忆容量或分离时间尺度来缓解这一瓶颈，但缺乏生物学中发现的快速神经元级处理与长期记忆的结合。为此，我们提出了连续记忆机器（Continuous Memory Machine, CMM），这是一种递归架构，具有矩阵值的短期和长期记忆状态，分别承担不同的功能角色。CMM基于连续思维机器（Continuous Thought Machine, CTM），其短期记忆跟踪最近的神经活动，具有独特参数化的神经元级模型学习利用这些活动模式进行计算。持久的长期记忆存储信息以供后续使用，Transformer共同更新这两种记忆存储，提供一种表达性双向读写机制，使得每个存储可以重新组织自己的内容，并且可以相互读取和写入。在算法、上下文学习和递归推理任务中，CMM的表现超过了广泛的基线，展现出比以往记忆增强网络更强的泛化能力，同时保留了CTM的可解释注意力模式。代码可在 https://github.com/SakanaAI/continuous-memory-machines 获取。
cs.AI / 102 / 2610.07935
SIGMA: Self-Improving Alignment Generalization from a Model Spec
SIGMA：基于模型规范的自我提升对齐泛化
Abstract
LLM agents are increasingly capable of executing complex tasks and of recursively improving themselves on easy-to-verify objectives such as software engineering and mathematics. Since alignment is much harder to verify, this creates a growing risk of capabilities increasing without appropriate safety alignment, especially as capabilities expand to auto-research and cybersecurity. Existing approaches focus on capability self-improvement using verifiable feedback or on alignment training with supervision from stronger models or curated data, creating an external supervision bottleneck for alignment. We ask whether current models can improve their own safety alignment, and propose SIGMA, a data generation and training pipeline enabling alignment self-improvement that generalizes to out-of-distribution settings. Given only a "Model Spec" stating the model's desired behavior, SIGMA leverages a model's reasoning capabilities to strengthen its own safety reasoning. SIGMA first performs spec-guided task synthesis, using the candidate model as a task designer agent to generate diverse alignment dilemma scenarios and convert them into training tasks that stress-test its understanding of the Model Spec. Next, SIGMA conducts self-judged alignment training through supervised fine-tuning and rubric-based reinforcement learning with the model itself as the reward model. Despite training only on single-turn chat data, SIGMA improves safety alignment in multi-turn agentic environments (AgentHarm harmfulness decreases from 22.6 to 14.8; Agentic Misalignment decreases from 79.1 to 3.8), outperforms Deliberative Alignment and Constitutional AI baselines, and retains general capability. Analyses show that a Model Spec balancing harmlessness and helpfulness, test-time reasoning for safety deliberation, and high-quality rubrics from SIGMA's task designer agent are crucial for effective self-improvement.
Chinese Translation
大型语言模型（LLM）代理在执行复杂任务和在易于验证的目标（如软件工程和数学）上递归自我提升方面的能力日益增强。由于对齐的验证要困难得多，这导致能力在没有适当安全对齐的情况下不断增加的风险，特别是当能力扩展到自动研究和网络安全时。现有方法主要集中在使用可验证反馈进行能力自我提升，或在更强模型或精心策划的数据监督下进行对齐训练，从而在对齐方面形成了外部监督瓶颈。我们探讨当前模型是否能够改善自身的安全对齐，并提出SIGMA，这是一种数据生成和训练管道，能够实现对齐自我提升，并在分布外设置中泛化。SIGMA仅需一个“模型规范”来说明模型的期望行为，便利用模型的推理能力来增强自身的安全推理。SIGMA首先进行规范指导的任务合成，利用候选模型作为任务设计代理，生成多样的对齐困境场景，并将其转化为训练任务，以压力测试其对模型规范的理解。接下来，SIGMA通过监督微调和基于评分标准的强化学习进行自我评估的对齐训练，以模型自身作为奖励模型。尽管仅在单轮对话数据上进行训练，SIGMA在多轮代理环境中改善了安全对齐（AgentHarm的有害性从22.6降至14.8；Agentic Misalignment从79.1降至3.8），超越了深思熟虑对齐和宪法人工智能基准，并保持了通用能力。分析表明，平衡无害性和有用性的模型规范、安全审议的测试时推理，以及来自SIGMA任务设计代理的高质量评分标准对于有效的自我提升至关重要。
cs.AI / 103 / 2610.07948
Confidence Reasoning Graphs: Structured Confidence Estimation for LLM Agents
信心推理图：大语言模型代理的结构化信心估计
Abstract
When using an LLM agent in a consequential domain, making an informed decision about whether to trust its output or intervene requires calibrated confidence in the agent's success. Confidence estimation for agents is difficult because evidence about success is distributed across heterogeneous, interdependent steps of an agent's trajectory. Practical agentic deployments introduce further challenges: frontier LLMs often provide limited access to internal signals, agent roll-outs are costly, and training data may be unavailable or quickly become outdated. To address these challenges, we introduce Confidence Reasoning Graphs (CRGs), an inference-time framework that estimates the probability an agent accomplished its task from a single trajectory, without privileged model access or training data. Rather than compressing an execution into a single holistic judgment, a CRG begins with the claim that the agent accomplished its task, decomposes it into contextualized sub-claims grounded in trajectory evidence, estimates confidence for each terminal claim, and finally aggregates these into an overall confidence estimate. Across three agentic benchmarks, three backbone models, and three agent frameworks, CRGs yield better-calibrated confidence and stronger risk-aware decision making than verbalized, sampling-based, and white-box surrogate baselines. We further find that calibration error alone can be misleading: a white-box surrogate baseline appears well calibrated while providing near-chance discrimination. Ablations attribute CRG's improvements to claim-level confidence estimation and aggregation rather than graph construction alone. Finally, a CRG exposes the claims and trajectory evidence underlying each confidence estimate, enabling it to be audited at decision time.
Chinese Translation
在使用大语言模型（LLM）代理进行重要决策时，是否信任其输出或进行干预需要对代理成功的信心进行校准。代理的信心估计是困难的，因为关于成功的证据分布在代理轨迹的异质、相互依赖的步骤中。实际的代理部署带来了进一步的挑战：前沿的LLM通常对内部信号的访问有限，代理的执行成本高昂，且训练数据可能不可用或迅速过时。为了解决这些挑战，我们提出了信心推理图（Confidence Reasoning Graphs, CRGs），这是一种推理时框架，可以在没有特权模型访问或训练数据的情况下，从单一轨迹中估计代理完成任务的概率。CRG并不是将执行压缩为单一的整体判断，而是从代理完成任务的声明开始，将其分解为基于轨迹证据的上下文化子声明，估计每个终端声明的信心，最后将这些信心汇总为总体信心估计。在三个代理基准测试、三个基础模型和三个代理框架中，CRGs提供了更好校准的信心和更强的风险意识决策能力，优于口头表达、基于采样和白盒替代基线。我们进一步发现，单独的校准误差可能会产生误导：一个白盒替代基线看似校准良好，但提供的区分能力接近随机。消融实验表明，CRG的改进归因于声明级信心估计和聚合，而不仅仅是图的构建。最后，CRG揭示了每个信心估计背后的声明和轨迹证据，使其在决策时能够被审计。
cs.AI / 104 / 2610.07972
Can Agents Work for Everyone? Cross-User Reliability for Mobile GUI Agents in Personalized User Interfaces
代理能为每个人工作吗？个性化用户界面中移动图形用户界面的跨用户可靠性
Abstract
Mobile GUI agents increasingly operate on interfaces influenced by users' histories and preferences, but their reliability across different users remains underexplored. We introduce PAIR (Personalized Application-state Instantiation and Rendering), a pipeline for constructing user-conditioned application states that enables controlled evaluation of the same task across different users. We further introduce RePAIR (Reinforcement learning with Personalization-Aware Interaction Rewards), a training approach that learns from cross-user differences in subgoal outcomes to improve reliability across user-conditioned mobile environments. Across six agents, we find substantial variation in task success across users and consistently lower subgoal achievement in user-conditioned UI contexts (6.98 to 15.4 pp). This gap further increases for personal targets drawn from each user's own content (8.77 to 22.0 pp). Failures in these contexts frequently involve selecting another item instead of the intended target, particularly before target exposure. Finally, RePAIR improves user-conditioned SAR (+5.87 pp), all-success (+7.50 pp), and overall Task SR (+9.42 pp) over its supervised fine-tuning parent on unseen users, providing initial evidence that explicitly learning from cross-user variation can improve GUI-agent reliability.
Chinese Translation
移动图形用户界面代理越来越多地在受用户历史和偏好影响的界面上操作，但它们在不同用户之间的可靠性仍然未被充分探讨。我们引入了 PAIR（个性化应用状态实例化和渲染），这是一个构建用户条件应用状态的管道，能够在不同用户之间对相同任务进行控制评估。我们进一步引入了 RePAIR（具有个性化交互奖励的强化学习），这是一种训练方法，通过学习跨用户在子目标结果上的差异，以提高在用户条件移动环境中的可靠性。在六个代理中，我们发现不同用户之间的任务成功率存在显著差异，并且在用户条件的用户界面上下文中，子目标的达成率始终较低（6.98 到 15.4 个百分点）。对于从每个用户自身内容中提取的个人目标，这一差距进一步增加（8.77 到 22.0 个百分点）。在这些上下文中，失败往往涉及选择其他项目而非预期目标，尤其是在目标曝光之前。最后，RePAIR 在未见用户上改善了用户条件的成功率（SAR，+5.87 个百分点）、全成功率（+7.50 个百分点）和整体任务成功率（Task SR，+9.42 个百分点），提供了初步证据表明，明确地从跨用户变异中学习可以提高图形用户界面代理的可靠性。
cs.AI / 105 / 2610.07979
Learning from Revision Consequences: Hindsight Meta-Experience Distillation for Self-Improving Agents
从修订结果中学习：自我改进代理的事后元体验提炼
Abstract
As agents continuously improve by generating and revising Skills, the process that discovers and refines those Skills becomes a learnable object in its own right. Task-Skills directly act on task execution, whereas Meta-Skills govern how agents discover and improve future Skills; their value therefore emerges through the subsequent search processes they induce. Existing approaches improve Meta-Skills from observed raw Skill-search trajectories and branch outcomes. However, branch performance entangles the effects of the initial discovery state and the Meta-Skill revision that generated the search process, making it difficult to characterize what a particular revision actually changed, and pushing updates toward revisions that benefit from favorable states rather than those that improve the process. We introduce HMED (Hindsight Meta-Experience Distillation), a mechanism for constructing Meta-Experience for self-improving agents. HMED revisits the completed event from which a revision originates and re-executes the incumbent and revised Meta-Skills from the same restored discovery state, so that the changes associated with the revision can be observed under a shared condition. Each comparison is distilled into a Meta-Experience, a structured record that can be reused by future updates, so that even revisions that are not ultimately retained still contribute a learning signal. Across three interactive agent benchmarks and both open-source and closed-source models, HMED consistently improves Skill discovery performance over strong baselines, shifting Meta-Skill learning beyond branch outcomes toward the consequences of changing the improvement process.
Chinese Translation
随着代理通过生成和修订技能不断改进，发现和完善这些技能的过程本身也成为一个可学习的对象。任务技能直接作用于任务执行，而元技能则管理代理如何发现和改进未来的技能；因此，它们的价值通过随之而来的搜索过程显现出来。现有方法从观察到的原始技能搜索轨迹和分支结果中改进元技能。然而，分支性能将初始发现状态和生成搜索过程的元技能修订的影响纠缠在一起，使得难以表征特定修订实际改变了什么，并推动更新朝向那些受益于有利状态的修订，而不是那些改善过程的修订。我们引入了HMED（事后元体验提炼），一种为自我改进代理构建元体验的机制。HMED重新审视修订来源的已完成事件，并从相同的恢复发现状态重新执行现有和修订后的元技能，以便在共享条件下观察与修订相关的变化。每次比较都被提炼为元体验，这是一个结构化记录，可以被未来的更新重用，因此即使是那些最终未被保留的修订仍然能够贡献学习信号。在三个交互代理基准测试以及开源和闭源模型中，HMED始终在强基线之上改善技能发现性能，将元技能学习从分支结果转向改善过程变化的后果。
cs.AI / 106 / 2610.08033
Learning in Dreams, Winning in Reality: A Continuous Dyna Loop for a Ten-Hero MOBA
在梦中学习，在现实中获胜：一个针对十英雄MOBA的连续Dyna循环
Abstract
World models are usually judged from the inside: by prediction loss, by the return a policy earns in imagination, or by how convincing their frames look. We judge one from the outside. We learn a structured, multi-agent world model of a complete ten-hero MOBA (206 units, every hero acting every tick, games of up to 6,000 ticks), train a policy only inside it with 1,400-tick free-running imagined episodes, and measure that policy in the real game against the opponent the game ships with. The real game never provides a gradient; it provides the policy's own games as training data for the world model, and an online evaluation that selects and anchors the policy. Run as a continuous asynchronous Dyna loop, the policy wins 70.2% of real games as radiant (421 of 600; 95% CI 66.4-73.7) on seeds never used for any decision, up from 0% for dream training alone and 33.7% before the loop. It wins none as dire, and neither does the shipped opponent when it plays itself. Four findings explain the result. Model exploitation is invisible from inside the dream: every unanchored run collapsed within a few updates while no in-dream metric tracked the collapse. A world model that is accurate on its training corpus is badly wrong on the policy's own games, and Dyna repairs it there, which is worth +9.2 points of real win rate with the policy recipe held fixed. Finally, the policy inherits its world model's fidelity profile mechanic by mechanic: the model represents the macro game but not crowd control, cast timing or lethality, and the policy wins by map-wide pressure with almost no coordinated fighting. We release the world model, the dream-PPO harness, a world-model debugger, the evaluation protocol, and every policy and log.
Chinese Translation
世界模型通常从内部进行评估：通过预测损失、通过策略在想象中获得的回报，或通过其帧的可信度来判断。我们从外部进行评估。我们学习了一个完整的十英雄MOBA的结构化多智能体世界模型（206个单位，每个英雄每个时刻都在行动，游戏最多可达6000个时刻），仅在其中训练一个策略，使用1400个时刻的自由运行想象剧集，并在真实游戏中与游戏自带的对手进行测量。真实游戏从不提供梯度；它提供策略自身的游戏作为世界模型的训练数据，以及一个在线评估来选择和固定策略。作为一个连续的异步Dyna循环运行，该策略在真实游戏中以光辉阵营获胜70.2%（421场中的600场；95%置信区间66.4-73.7），在从未用于任何决策的种子上，从仅依靠梦境训练的0%和循环前的33.7%提升而来。作为黑暗阵营时，它没有赢得任何比赛，出厂对手在自我对战时也没有获胜。四个发现解释了这一结果。模型的利用在梦中是不可见的：每个未固定的运行在几次更新内崩溃，而没有任何梦中指标跟踪崩溃。一个在其训练语料库上准确的世界模型在策略自身的游戏中却是严重错误的，而Dyna在此修复了它，这使得真实胜率提高了9.2个百分点，同时策略配方保持不变。最后，策略逐个机制继承其世界模型的保真度特征：模型代表宏观游戏，但不代表控制人群、施法时机或致命性，策略通过几乎没有协调战斗的全图压力获胜。我们发布了世界模型、梦境-PPO工具、世界模型调试器、评估协议，以及每个策略和日志。
cs.AI / 107 / 2610.08036
Same Feedback, Different Answer: Measuring Run-to-Run Instability in Frontier-Model Customer Feedback Analysis
相同反馈，不同答案：测量前沿模型客户反馈分析中的运行间不稳定性
Abstract
AI agents are increasingly being programmed to automate knowledge work over large collections of unstructured data. Such automation requires repeatability: when the underlying evidence is unchanged, the agent's categories, priorities, and counts should not shift materially between runs, even if each individual answer appears plausible. We introduce a repeat-run evaluation framework that aligns semantically equivalent categories and focuses on two operating metrics: theme churn, the normalized change in the returned category set, and volume disagreement, the change in counts for categories that persist. We evaluate three recurring customer-feedback tasks across eight frontier models, corpus sizes from 100 to 5,000 records, multiple prompts, and three execution designs: raw generation, taxonomy-free hierarchical decomposition, and a taxonomy-grounded agent (TGA) using persistent themes, subthemes, and record-level predictions. With Claude Opus 4.8 and the 1,000-record corpus fixed, TGA reduces theme churn by 86--88% relative to both raw generation and hierarchical decomposition, while matched-theme volumes have zero disagreement. The taxonomy-grounded agent is more stable than every raw model in the screen, remains more stable at each corpus size, and keeps this advantage when theme matching is made stricter or looser. Although evaluated on customer feedback, the framework targets repeated synthesis of unstructured corpora more broadly, including financial reports, legal documents, incident records, and scientific literature. Overall, these results show that taxonomy grounding produces more consistent and repeatable outputs for recurring knowledge work.
Chinese Translation
人工智能代理越来越多地被编程用于自动化处理大量非结构化数据的知识工作。这种自动化要求可重复性：当基础证据不变时，代理的类别、优先级和计数在不同运行之间不应发生实质性变化，即使每个单独的答案看起来都合理。我们引入了一种重复运行评估框架，该框架对语义上等价的类别进行对齐，并关注两个操作指标：主题变动（theme churn），即返回类别集的标准化变化，以及数量不一致（volume disagreement），即持续类别的计数变化。我们在八个前沿模型上评估了三个重复的客户反馈任务，语料库大小从100到5000条记录，多个提示，以及三种执行设计：原始生成、无分类法的层次分解，以及使用持续主题、子主题和记录级预测的分类法基础代理（taxonomy-grounded agent, TGA）。在固定Claude Opus 4.8和1000条记录的语料库的情况下，TGA相对于原始生成和层次分解将主题变动降低了86%至88%，而匹配主题的数量则没有不一致。分类法基础代理在所有筛选的原始模型中表现出更高的稳定性，在每个语料库大小下保持更高的稳定性，并在主题匹配变得更严格或更宽松时保持这一优势。尽管是在客户反馈上进行评估，该框架更广泛地针对非结构化语料库的重复综合，包括财务报告、法律文件、事件记录和科学文献。总体而言，这些结果表明，分类法基础能够为重复的知识工作产生更一致和可重复的输出。
cs.AI / 108 / 2610.08048
DAEDALUS: Bootstrapping Agent Memory from Self-Generated Tasks
DAEDALUS：从自生成任务中引导代理记忆
Abstract
LLM agents often lack the operational knowledge to act reliably in new environments, as they must discover specific tool behaviors or environment conventions on their own. Without memory of past attempts, they repeat the same mistakes across tasks, leading to more task failures and longer trajectories. To address this, agentic systems typically rely on human-written guidelines or on procedural memory built from training tasks and an oracle verifier, both of which require prior knowledge of the environment. We present DAEDALUS, a method for bootstrapping reusable agent memory from self-generated practice without existing tasks or oracle verifiers. DAEDALUS pairs two agents: an explorer that interacts with the environment to generate challenging yet solvable tasks, and a solver that attempts them. A heuristic is derived from each solver failure and accepted only after the solver repeatedly succeeds with that heuristic in context. These outcomes also provide feedback for the explorer to refine the difficulty of future tasks. Accepted heuristics are then consolidated into a memory bank for test-time use. Across AppWorld, $\tau^2$-bench, and AutomationBench, DAEDALUS improves mean success rates by up to 15.9 points and pass^5 by up to 2.2x over a no-memory baseline, and is competitive with methods using training tasks, at a lower inference cost than most. We show that performance gains already emerge with a small exploration budget, and that its heuristics also benefit agents from other model families. Our ablations further reveal that solver traces provide the key information needed to derive effective heuristics, while factorizing early discoveries makes exploration more cost-efficient. Beyond memory construction, we find that the tasks generated by DAEDALUS can serve as a proxy for benchmark tasks when ranking models by performance. Code and artifacts: www.github.com/illuin-tech/daedalus.
Chinese Translation
大型语言模型（LLM）代理在新环境中往往缺乏可靠行动的操作知识，因为它们必须自行发现特定工具行为或环境惯例。没有对过去尝试的记忆，它们在任务中重复同样的错误，导致更多的任务失败和更长的执行轨迹。为了解决这个问题，代理系统通常依赖于人类编写的指导方针或基于训练任务和一个神谕验证器构建的程序性记忆，而这两者都需要对环境的先验知识。我们提出了DAEDALUS，一种从自生成的练习中引导可重用代理记忆的方法，无需现有任务或神谕验证器。DAEDALUS配对了两个代理：一个探索者与环境互动以生成具有挑战性但可解决的任务，另一个求解者尝试这些任务。每次求解失败都会导出一个启发式规则，并且只有在求解者在上下文中多次成功应用该启发式后，才会被接受。这些结果还为探索者提供反馈，以便细化未来任务的难度。被接受的启发式规则随后被整合到一个记忆库中以供测试时使用。在AppWorld、$ au^2$-bench和AutomationBench上，DAEDALUS将平均成功率提高了最多15.9个百分点，pass^5提高了最多2.2倍，相较于无记忆基线，并且在推理成本上低于大多数使用训练任务的方法。我们展示了即使在较小的探索预算下，性能提升也已显现，并且其启发式规则也对其他模型家族的代理有益。我们的消融实验进一步揭示，求解者的轨迹提供了推导有效启发式所需的关键信息，而将早期发现因素化使探索更具成本效益。除了记忆构建，我们发现DAEDALUS生成的任务可以作为基准任务的代理，用于根据性能对模型进行排名。代码和文档： www.github.com/illuin-tech/daedalus。
cs.AI / 109 / 2610.08076
SpeedrunBench: Challenging LLM Agents with Video Game Speedrunning
SpeedrunBench：用视频游戏速通挑战大型语言模型代理
Abstract
Frontier LLM agents have been shown to be capable of solving increasingly complex tasks for which humans have measurable solutions. This begs the pertinent question of whether LLM agents can go beyond what humans have already solved. The ability to develop sophisticated strategies to tackle consequential problems becomes paramount as well-trodden, human-developed solutions become insufficient for problems for which we lack context or enough training data. We study agents' capability of such strategy formation through the communal practice of video game speedrunning. In speedrunning, practitioners compete to find the fastest way to complete a video game under certain conditions, and in so doing uncovering interesting unorthodox play styles that require a thorough understanding and mastery of the underlying game mechanics. We introduce SPEEDRUNBENCH, a benchmark that evaluates frontier LLM agents across 9 different games. To perform well in this benchmark, agents must repeatedly improve their strategy, reflect on their performance, exploit their gained knowledge, and reason across a long-horizon of actions to improve on an increasingly difficult problem: being faster than themselves and everyone else. Our experiments show that while frontier agents approach human world records in simple platformer games, they remain behind human performance on longer, more complex games under practical budgets. These results suggest that SPEEDRUNBENCH is a useful testbed for studying agents' strategy formation capabilities as well as being a saturation-resistant evaluation measure, as there is almost always a faster completion time waiting to be discovered.
Chinese Translation
前沿的大型语言模型（LLM）代理已被证明能够解决越来越复杂的任务，而这些任务是人类可以测量出解决方案的。这引出了一个相关的问题，即LLM代理是否能够超越人类已经解决的问题。随着人类开发的解决方案变得不足以应对缺乏上下文或足够训练数据的问题，发展复杂策略以应对重要问题的能力变得至关重要。我们通过视频游戏速通这一共同实践研究代理的策略形成能力。在速通中，参与者竞争寻找在特定条件下完成视频游戏的最快方法，从而揭示出需要深入理解和掌握游戏机制的有趣非传统玩法。我们引入了SPEEDRUNBENCH，一个评估前沿LLM代理在9款不同游戏中的基准测试。为了在这个基准测试中表现良好，代理必须不断改进其策略，反思其表现，利用所获得的知识，并在长时间的行动中进行推理，以在一个日益困难的问题上取得进展：比自己和其他人更快。我们的实验表明，尽管前沿代理在简单的平台游戏中接近人类世界纪录，但在更长、更复杂的游戏中，它们在实际预算下仍落后于人类表现。这些结果表明，SPEEDRUNBENCH是研究代理策略形成能力的有用测试平台，同时也是一种抗饱和的评估指标，因为几乎总有更快的完成时间等待被发现。
cs.AI / 110 / 2610.08077
Self-Retrospection Distillation: Turning Post-hoc Experiences into Prior Foresight
自我回顾蒸馏：将事后经验转化为事前预见
Abstract
Reinforcement learning with verifiable rewards (RLVR) turns agent experience into learning signals primarily through scalar outcome rewards after interaction. For group-relative objectives, however, this signal vanishes when all rollouts receive the same reward, even though their trajectories may reveal useful information about what the task requires and how the agent fails. We ask a complementary question: can hindsight teach an agent what it could have anticipated before acting? We introduce prospective learning, which uses post-hoc experience to supervise foresight predictions from the pre-interaction view, and instantiate it with Self-Retrospection Distillation (SRD). Intuitively, a completed trajectory reveals knowledge that would have been useful and pitfalls that should be avoided; SRD distills this privileged hindsight into trajectory-blind foresight of the same policy. Foresight serves only as a training target and need not be explicitly generated at inference time. Across 10 tool-integrated reasoning and long-horizon agentic tasks, SRD complements RLVR and self-distillation baselines with gains of up to $24.2$ pp. Its advantage is especially pronounced when reward contrast is scarce: when $37$--$98\%$ of rollout groups are reward-uniform across model scales, yet SRD can still exploit learning signal from sampled trajectories. In the 2B setting, where $98\%$ of groups are all-failure, the RLVR training ends up at $0.0\%$ success, while adding SRD reaches $60.6\%$ under the same rollout budget. Our results suggest that post-hoc agent experience is useful not only for evaluating or improving behavior, but also for shaping predictive representations before available interaction.
Chinese Translation
具有可验证奖励的强化学习（RLVR）通过在交互后利用标量结果奖励将代理经验转化为学习信号。然而，对于群体相对目标，当所有的回放都获得相同的奖励时，这一信号会消失，尽管它们的轨迹可能揭示出关于任务要求和代理失败的有用信息。我们提出一个互补的问题：事后经验能否教会代理在行动之前可以预见到什么？我们引入了前瞻性学习，它利用事后经验来监督来自交互前视角的预见预测，并通过自我回顾蒸馏（Self-Retrospection Distillation, SRD）进行实例化。直观上，完成的轨迹揭示了本应有用的知识和应避免的陷阱；SRD将这一特权的事后经验提炼为同一策略的无轨迹前瞻。前瞻仅作为训练目标，不必在推理时显式生成。在10个工具集成推理和长时间跨度的代理任务中，SRD与RLVR和自我蒸馏基线相结合，获得了高达24.2个百分点的提升。当奖励对比稀缺时，其优势尤其明显：当37%至98%的回放组在模型规模上奖励均匀时，SRD仍然可以利用从采样轨迹中获得的学习信号。在2B设置中，98%的组都是全失败，RLVR训练的成功率最终为0.0%，而添加SRD后在相同的回放预算下成功率达到了60.6%。我们的结果表明，事后代理经验不仅对评估或改善行为有用，而且对在可用交互之前塑造预测表示也具有重要意义。
cs.AI / 111 / 2610.08082
POLAR: Ontology-Guided Risk Prevention for Tool-Calling LLM Agents
POLAR：基于本体的工具调用大型语言模型代理的风险预防
Abstract
LLM tool-use agents operate in dynamic environments where many actions carry operational risk. However, most safety mechanisms react only after errors manifest. Existing pre-emptive approaches either fine-tune the agent on chain-of-thought deliberation or compile natural-language guardrails into runtime checks, but they do so without exposing a structural, auditable verdict. We propose POLAR, a guardrail framework for small tool-calling agents that assesses reversibility through a structured two-layer ontology. POLAR assigns each action a graded reversibility score by deriving a candidate inverse sequence; calls failing a threshold are pruned before execution. Evaluated on $\tau^2$-bench across six agent models, POLAR improves mean task reward by 0.11 to 0.18 points on airline for four of six agents, but only eight of eighteen model--domain cells improve overall; retail and stronger agents often regress. POLAR provides an auditable structural check and characterizes its task-utility trade-offs. Reward is not a direct measure of prevented harm.
Chinese Translation
大型语言模型（LLM）工具使用代理在动态环境中操作，其中许多行为存在操作风险。然而，大多数安全机制仅在错误显现后做出反应。现有的预防性方法要么对代理进行链式思维推理的微调，要么将自然语言的保护措施编译为运行时检查，但这些方法并未提供结构化的、可审计的判决。我们提出了POLAR，一个针对小型工具调用代理的保护框架，通过结构化的两层本体评估可逆性。POLAR通过推导候选逆序列为每个动作分配一个分级可逆性得分；未达到阈值的调用在执行前被修剪。在六个代理模型的$ au^2$基准测试中评估，POLAR在六个代理中的四个上提高了航空任务的平均奖励0.11到0.18分，但只有十八个模型-领域单元中的八个整体改善；零售和更强的代理往往出现退步。POLAR提供了可审计的结构检查，并描述了其任务效用的权衡。奖励并不是防止伤害的直接衡量标准。
cs.AI / 112 / 2610.08095
Natural Language Questions as an Interface for Knowledge Graphs: QRAKEN Graph Distillation and Semantic Self-Healing
自然语言问题作为知识图谱的接口：QRAKEN图谱蒸馏与语义自我修复
Abstract
Natural-language access to RDF knowledge graphs is a core Semantic Web ambition. Large language models (LLMs) have advanced Text-to-SPARQL, yet on unfamiliar graphs they often generate valid queries that misrepresent the populated data model. QRAKEN is a training-free, ontology-agnostic neurosymbolic pipeline grounding generation in empirical graph evidence rather than schema expectations. An offline distiller produces TTQL, a compact description of populated multi-hop patterns, conditional frequencies and path-conditioned literal examples, plus a class-property co-occurrence matrix. Online, TTQL guides the LLM, while deterministic syntax, vocabulary and data-model checks provide diagnostics for iterative refinement. On CK25 (First International Text2SPARQL Challenge), under matched-condition recomputation on a QLever snapshot, QRAKEN achieves strict F1 of 0.643 $\pm$ 0.026 with GPT-4.1 mini and 0.652 $\pm$ 0.012 with GPT-5.4: relative gains of 30% and 32% over the strongest recomputed participant, outperforming systems using the same base model family. Ablations identify TTQL patterns as the dominant driver (+0.31 strict F1 over a shape-only baseline); the refinement loop provides a cheap safety net, rejecting triple patterns unsupported by the co-occurrence matrix. Compared with auto-derived SHACL, TTQL yields 64% higher strict F1, supporting the value of empirical patterns beyond schema exposure. With two local 35B 4-bit open-weight models at zero marginal cost, the same pipeline matches the strongest recomputed participant, and TTQL advantages over shape-only and SHACL baselines persist. Results on a single, relatively small benchmark provide an initial empirical signal; monolithic TTQL injection on very open cross-domain graphs remains the main limitation.
Chinese Translation
自然语言访问RDF知识图谱是语义网的核心目标。大型语言模型（LLMs）在文本到SPARQL的转换上取得了进展，但在不熟悉的图谱上，它们常常生成有效的查询，却错误地表示了填充的数据模型。QRAKEN是一个无训练、与本体无关的神经符号管道，其生成基于经验图证据，而非模式预期。离线蒸馏器生成TTQL，这是一种对填充的多跳模式、条件频率和路径条件文字示例的紧凑描述，以及一个类-属性共现矩阵。在在线过程中，TTQL指导LLM，而确定性语法、词汇和数据模型检查提供了迭代优化的诊断。在CK25（首届国际文本到SPARQL挑战赛）中，在QLever快照下的匹配条件重新计算中，QRAKEN在GPT-4.1 mini上实现了严格F1值为0.643 ± 0.026，在GPT-5.4上为0.652 ± 0.012：相对增益分别为30%和32%，超越了最强的重新计算参与者，优于使用相同基础模型系列的系统。消融实验表明，TTQL模式是主要驱动因素（相较于仅基于形状的基线提高了0.31的严格F1）；优化循环提供了一个廉价的安全网，拒绝共现矩阵不支持的三元组模式。与自动生成的SHACL相比，TTQL的严格F1提高了64%，支持了经验模式超越模式暴露的价值。在两个本地35B 4位开放权重模型以零边际成本的情况下，同一管道与最强的重新计算参与者相匹配，且TTQL在仅基于形状和SHACL基线上的优势依然存在。在一个相对较小的基准测试上的结果提供了初步的经验信号；在非常开放的跨域图谱上进行单一的TTQL注入仍然是主要限制。
cs.AI / 113 / 2610.08101
Beyond Corrected Memory: Execution Consistency in Multi-Agent Systems
超越修正记忆：多智能体系统中的执行一致性
Abstract
Shared memory coordinates agents' actions, but correct records do not establish that those actions satisfy task requirements. Memory governance and failure diagnosis regulate or inspect recorded information; they do not by themselves establish whether it is sufficient to judge task duties. We define execution consistency through duties governing state use, information handoffs, and final-state agreement, with explicit evidence conditions for judging fulfillment. Our core claim is that identical retained records can correspond to compliant and violating executions under the same task rule. Controlled removal of evidence such as receipt, action dependence, or response validity leaves 82.4% of opposite-label pairs indistinguishable; restoration separates 97.9% of the merged pairs. Natural-log annotations identify the defined violations in actual executions. However, existing logs do not always explicitly represent the execution relationships needed for these judgments. To assess the definition's practical value, we use CAVERT, a framework for consistency diagnosis and recovery, to extract supported relationships from logs and apply these criteria. It consistently outperforms contract-prompted LLM and rule-based baselines in diagnosis across all 12 benchmark-executor settings. Under the same gate and executor limits, it also outperforms rule-guided recovery in all four evaluated environments. These findings identify execution evidence that agent-memory and execution interfaces should preserve for reliable judgment.
Chinese Translation
共享内存协调智能体的行动，但正确的记录并不能证明这些行动满足任务要求。内存管理和故障诊断调节或检查记录的信息；它们本身并不能确定是否足以判断任务职责。我们通过管理状态使用、信息交接和最终状态一致性的职责来定义执行一致性，并提供明确的证据条件以判断履行情况。我们的核心主张是，相同的保留记录可以对应于在相同任务规则下的合规和违规执行。控制性地移除证据，如收据、行动依赖或响应有效性，使得82.4%的相反标签对无法区分；而恢复则能分离97.9%的合并对。自然对数注释识别了实际执行中的定义违规。然而，现有日志并不总是明确表示这些判断所需的执行关系。为了评估该定义的实际价值，我们使用CAVERT，一个一致性诊断和恢复框架，从日志中提取支持的关系并应用这些标准。它在所有12个基准执行者设置中，在诊断方面始终优于基于合同的提示大型语言模型（LLM）和基于规则的基线。在相同的门限和执行者限制下，它在所有四个评估环境中也优于基于规则的恢复。这些发现识别了智能体内存和执行接口应保留的执行证据，以便进行可靠的判断。
cs.AI / 114 / 2610.08102
DSV-Mem: Evaluating Multimodal Memory in Professional Workflows for MLLM Agents
DSV-Mem：评估多模态记忆在专业工作流程中的应用于 MLLM 代理
Abstract
Conversational MLLM agents are increasingly expected to assist in professional workflows, from AI research and engineering design to product management and business operations. Yet this capability remains underexplored: existing benchmarks largely focus on informal, everyday interactions and personal-life scenarios featuring photographic natural images, isolated static artifacts, and recall-oriented questions. In contrast, professional scenarios often involve structured, information-heavy artifacts that undergo frequent revisions and authority updates, and compositional queries requiring reconciliation of many artifact versions while tracking state precisely. To address these challenges, we introduce DSV-Mem, a benchmark for evaluating Dense Stateful Visual Memory. DSV-Mem comprises expert-reviewed scenarios and 1,000 questions across five user-oriented categories (Current State, Past State, Derived State, Change History, and Conflict/Refusal). A Hartley-inspired criterion favors questions with broader visual-evidence inspection demands. We also introduce a generation harness that produces evaluation suites by decoupling state-transition synthesis from conversation filling. Evaluation over 27 configurations spanning frontier and open-weight models and memory management methods reveals that the strongest baseline scores below 45% on DSV-Mem. Analysis surfaces findings: 1) multimodality and information density both contribute to difficulty, but state evolution, particularly the number of governing updates, is the dominant tested factor. Raw conversation/haystack length, OCR, and arithmetic are not the primary bottlenecks; 2) models often fail to verify user premises against prior state updates before answering; 3) increased reasoning effort and memory management methods yield limited gains, whereas state-aware designs prove more effective. The benchmark and code will be publicly released.
Chinese Translation
对话式 MLLM 代理越来越被期望在专业工作流程中提供帮助，从人工智能研究和工程设计到产品管理和商业运营。然而，这一能力仍然未得到充分探索：现有基准主要集中在非正式的日常互动和涉及摄影自然图像、孤立静态物品以及以回忆为导向的问题的个人生活场景。相比之下，专业场景通常涉及结构化的信息密集型物品，这些物品经常经历修订和权威更新，并且需要处理需要协调多个物品版本的组合查询，同时精确跟踪状态。为了解决这些挑战，我们引入了 DSV-Mem，一个用于评估密集状态视觉记忆的基准。DSV-Mem 包含专家审查的场景和跨五个用户导向类别（当前状态、过去状态、派生状态、变更历史和冲突/拒绝）的 1,000 个问题。受 Hartley 启发的标准更倾向于那些对视觉证据检查要求更广泛的问题。我们还引入了一种生成工具，通过将状态转换合成与对话填充解耦，生成评估套件。对涵盖前沿和开放权重模型及记忆管理方法的 27 种配置的评估显示，最强基线在 DSV-Mem 上的得分均低于 45%。分析结果表明：1) 多模态性和信息密度都对难度产生影响，但状态演变，特别是主导更新的数量，是测试中占主导地位的因素。原始对话/干草堆长度、光学字符识别（OCR）和算术并不是主要瓶颈；2) 模型通常未能在回答之前验证用户前提与先前状态更新的一致性；3) 增加推理努力和记忆管理方法带来的收益有限，而状态感知设计则更为有效。该基准和代码将公开发布。
cs.AI / 115 / 2610.08106
ChartBmkAgent: Harness-Governed Multi-Agent Construction of Chart QA Benchmarks from Sparse Error-Taxonomy Specifications
ChartBmkAgent：基于稀疏错误分类规范的受控多智能体图表问答基准构建
Abstract
Multimodal large language models (MLLMs) advance rapidly, while conventional benchmark development lags behind, delaying investigation of newly observed capability gaps. Such investigation requires an expressive task format and an on-demand construction process: information-rich charts make chart question answering (Chart QA) suitable for probing coupled perception and reasoning. Automated Chart QA construction is intended to shorten the benchmark-development cycle by turning identified gaps into targeted samples on demand. Current methods, however, commonly separate target guidance from scratch generation: target-guided systems often require prepared data, charts, or templates, while scratch-generation systems primarily ensure artifact validity, without explicitly controlling whether newly synthesized requirements and content remain aligned with an externally specified diagnostic target. We introduce ChartBmkAgent, which turns an identified capability gap into targeted diagnostic evidence by constructing complete Chart QA samples from sparse error-taxonomy specifications. Throughout construction, a central harness governs specialized agents, requires stage-specific evidence of alignment with the original error category, and records the basis for each acceptance decision. On 300 taxonomy-wide samples, MLLM accuracies ranged from 32.7% to 84.3% with distinct category profiles, showing that generated samples reveal capability differences. Across three source-model comparisons, targeted follow-ups scored 50.0% versus 82.2% on matched controls ($p=8.96\times10^{-6}$); all six cross-model comparisons had the same direction, demonstrating targeted validation and diagnostic-data generation. Multiple evaluator models assessed whether each sample tested its specified error category; 86.4% met this criterion, providing empirical evidence of target preservation.
Chinese Translation
多模态大型语言模型（MLLMs）发展迅速，而传统基准开发却滞后，导致对新观察到的能力差距的研究延迟。这种研究需要一种表达丰富的任务格式和按需构建的过程：信息丰富的图表使得图表问答（Chart QA）适合于探测感知与推理的耦合。自动化的图表问答构建旨在通过将识别出的差距转化为按需的目标样本，从而缩短基准开发周期。然而，当前的方法通常将目标指导与从头生成分开：目标导向系统通常需要准备好的数据、图表或模板，而从头生成系统主要确保工件的有效性，而未明确控制新合成的要求和内容是否与外部指定的诊断目标保持一致。我们介绍了ChartBmkAgent，它通过从稀疏错误分类规范构建完整的图表问答样本，将识别出的能力差距转化为有针对性的诊断证据。在构建过程中，一个中央控制器管理专门的智能体，要求阶段特定的证据与原始错误类别保持一致，并记录每个接受决策的依据。在300个分类范围内的样本中，MLLM的准确率范围为32.7%到84.3%，显示出不同类别的特征，表明生成的样本揭示了能力差异。在三种源模型的比较中，针对性的后续评估在匹配对照组中得分为50.0%对比82.2%（$p=8.96 imes10^{-6}$）；所有六个跨模型比较均朝同一方向，证明了针对性验证和诊断数据生成。多个评估模型评估每个样本是否测试其指定的错误类别；86.4%的样本满足这一标准，提供了目标保持的实证证据。
cs.AI / 116 / 2610.08138
Test-Time Agent Evolution for Long-Horizon Legal Reasoning
测试时代理演化用于长时间跨度的法律推理
Abstract
Legal intelligence aims to support reliable decision-making across long-horizon legal processes involving evolving case states and multiple roles. However, real-world legal deployment exhibits substantial case heterogeneity in facts, evidence, and procedural contexts, exposing the limitations of static agent strategies. Moreover, legal reasoning is inherently interdependent across roles and procedural stages, making global reliability fundamentally different from isolated role competence. To address these challenges, we study training-free test-time agent adaptation, where agents continuously exploit deployment-time signals from preceding cases and ongoing interactions without updating model parameters. We propose \method, which introduces \emph{Test-Time Memory Evolution} to retrieve reusable experience from previous cases, adapt it to the current factual and procedural context, and consolidate accumulated experience for subsequent decision-making. Further, \emph{Rubric-Aligned Collaboration} verifies and revises role-specific actions according to behavioral and procedural requirements, enabling coordinated decision-making across roles and stages. Extensive experiments on J1-EVAL and LegalWorld across five backbone models demonstrate consistent improvements over representative reasoning and agent baselines with reasonable interaction and computational costs. Ablation and case studies further show that the two components provide complementary benefits in experience adaptation and cross-role coordination, improving the reliability and efficiency of long-horizon legal reasoning.
Chinese Translation
法律智能旨在支持在涉及不断变化的案件状态和多重角色的长时间跨度法律程序中进行可靠的决策。然而，现实世界中的法律部署在事实、证据和程序背景上表现出显著的案件异质性，这暴露了静态代理策略的局限性。此外，法律推理在角色和程序阶段之间本质上是相互依赖的，使得全球可靠性与孤立角色能力根本不同。为了解决这些挑战，我们研究了无训练的测试时代理适应，代理在不更新模型参数的情况下，持续利用来自前期案件和持续互动的部署时信号。我们提出了 extit{Test-Time Memory Evolution}（测试时记忆演化），以从先前案件中检索可重用的经验，将其适应于当前的事实和程序背景，并巩固积累的经验以用于后续决策。此外， extit{Rubric-Aligned Collaboration}（标准对齐协作）根据行为和程序要求验证和修订角色特定的行动，使得跨角色和阶段的协调决策成为可能。在J1-EVAL和LegalWorld上的广泛实验中，五种基础模型展示了在合理的互动和计算成本下，相较于代表性的推理和代理基线的一致性改进。消融实验和案例研究进一步表明，这两个组件在经验适应和跨角色协调方面提供了互补的好处，提高了长时间跨度法律推理的可靠性和效率。
cs.AI / 117 / 2610.08142
Partially Observable Zero-shot coordination by Predicting Intention of Partner
通过预测伙伴意图实现部分可观察的零-shot协调
Abstract
Zero-shot coordination in embodied settings requires acting while the partner is intermittently out of view, leaving existing methods with ambiguous partner representations and uncertainty over hidden partner states. We propose Predicting Intention of Partner (PIP) to jointly address these challenges. PIP uses a Joint-view VAE to distill richer training-time evidence from the union of both agents' local observations into a partner representation available from local observations alone. Partner-state Belief networks further infer the partner's hidden location and behavioral tendencies from the ego agent's interaction history. We evaluate PIP in Burrito-PO, Overcooked-PO, and a Melting Pot substrate, together with a human evaluation in Burrito-PO. PIP attains the highest mean performance among the compared methods across all three benchmarks. Human evaluation and diagnostic analyses further support coordination with unseen partners and the contributions of both components under partner occlusion.
Chinese Translation
在具身环境中，零-shot协调要求在伙伴间歇性不可见的情况下进行行动，这使得现有方法面临模糊的伙伴表征和对隐藏伙伴状态的不确定性。我们提出了预测伙伴意图（Predicting Intention of Partner, PIP）以共同解决这些挑战。PIP使用联合视图变分自编码器（Joint-view VAE）从两个智能体的局部观察的联合中提取更丰富的训练时证据，从而生成仅基于局部观察的伙伴表征。伙伴状态信念网络（Partner-state Belief networks）进一步从自我智能体的交互历史中推断伙伴的隐藏位置和行为倾向。我们在Burrito-PO、Overcooked-PO和Melting Pot基底中评估了PIP，并在Burrito-PO中进行了人类评估。PIP在所有三个基准中获得了比较方法中最高的平均性能。人类评估和诊断分析进一步支持与未见伙伴的协调以及在伙伴遮挡下两个组件的贡献。
cs.AI / 118 / 2610.08165
Which alloy composition,what process parameters? Inferring the recipe from optimized metallic microstructure and texture
哪种合金成分，哪些工艺参数？从优化的金属微观结构和纹理推断配方
Abstract
The mechanical properties of a metallic alloy are set by its microstructure and texture: the size and shape of its grains and the orientation of their crystals. That structure is in turn set by a recipe, the alloy composition together with the processing parameters. Alloy development runs this chain forwards, tuning the structure until a target property is met. Running it backwards, from an optimized structure to the recipe that would produce it, still relies on expert knowledge. We ask whether this backwards step can be learned. On an in-house dataset of 107 magnesium alloy extrusion conditions across 14 alloys, each with optical micrographs and an X-ray texture measurement, we compare three descriptors of microstructure and texture: conventional grain and texture statistics, a vision embedding from a pretrained image encoder, and a graph neural network on the grain network. Each is paired with prediction heads for two tasks: the alloy composition given the process (Task A), and the process parameters given the composition (Task B). Under 5-fold cross-validation, the conventional descriptors identify the correct alloy for 65% of held-out conditions, against 17% for always guessing the most common alloy, while the learned embeddings stay below 30%. The process parameters are recoverable but noisier: compared with using the composition alone, the microstructure roughly halves the temperature error. Because only a few alloys were cast and only a few press settings were used, both answers are discrete, and heads that pick from these known options, while respecting their order, worked better than heads that predict a free value.
Chinese Translation
金属合金的机械性能由其微观结构和纹理决定：晶粒的大小和形状以及其晶体的取向。这种结构又由配方决定，即合金成分与加工参数的组合。合金开发沿着这一链条向前推进，调整结构直到达到目标性能。而从优化的结构反推能够产生该结构的配方，仍然依赖于专家知识。我们探讨这一反向步骤是否可以被学习。在一个包含107种镁合金挤压条件的内部数据集中，涵盖14种合金，每种合金都有光学显微图像和X射线纹理测量，我们比较了三种微观结构和纹理的描述符：传统的晶粒和纹理统计、来自预训练图像编码器的视觉嵌入，以及在晶粒网络上的图神经网络。每种描述符都与两个任务的预测头配对：给定工艺的合金成分（任务A），以及给定成分的工艺参数（任务B）。在5折交叉验证下，传统描述符在65%的保留条件下识别出正确的合金，而仅凭猜测最常见合金的准确率为17%，而学习的嵌入则保持在30%以下。工艺参数是可以恢复的，但噪声较大：与仅使用成分相比，微观结构大致将温度误差减半。由于只有少数合金被铸造，且仅使用了少数压制设置，因此两个答案都是离散的，而选择这些已知选项的预测头在尊重其顺序的情况下表现优于预测自由值的预测头。
cs.AI / 119 / 2610.08176
LFHE: Local-First Heuristic Evolution for Bounded Local Topology Search in Decentralized Learning with Non-IID Data
LFHE：在非独立同分布数据下的去中心化学习中进行有界局部拓扑搜索的以本地为先的启发式演化
Abstract
Decentralized learning is highly sensitive to communication topology under non-IID data. Adaptive peer-selection methods can exploit local model information, but broader peer discovery may require increasingly large control state, whereas direct spectral optimization typically relies on graph-wide information. We study the intermediate setting of bounded local topology search and propose Local-First Heuristic Evolution (LFHE), a representation-driven rewiring framework whose candidate discovery and scoring use only ego-neighborhood and friend-of-a-friend (FoF) information. The structural score admits an exact interpretation through graph Dirichlet energy: its sum across clients equals twice the representation Dirichlet energy, which under standard linear consensus dynamics governs the instantaneous dissipation of representation disagreement. LFHE combines this state-dependent structural signal with early exploration and degree control, while algebraic connectivity remains an offline graph diagnostic. Under bounded sparse degree, its FoF candidate state remains local rather than expanding toward population-wide peer tracking. Across four image, speech, and text benchmarks, LFHE achieves competitive decentralized learning performance. Matched-protocol controls identify the structural term as the principal empirical topology-selection signal, while comparison with broader peer discovery exposes a trade-off between predictive performance and discovery-state locality. Together, these results motivate state-aware bounded local topology search between pairwise peer selection and globally informed topology optimization.
Chinese Translation
去中心化学习在非独立同分布（non-IID）数据下对通信拓扑高度敏感。自适应的对等选择方法可以利用局部模型信息，但更广泛的对等发现可能需要越来越大的控制状态，而直接的谱优化通常依赖于图的全局信息。我们研究了有界局部拓扑搜索的中间设置，并提出了以本地为先的启发式演化（Local-First Heuristic Evolution, LFHE），这是一种基于表示驱动的重连框架，其候选发现和评分仅使用自我邻域和朋友的朋友（friend-of-a-friend, FoF）信息。结构得分通过图的Dirichlet能量具有精确的解释：跨客户端的总和等于表示Dirichlet能量的两倍，在标准线性共识动态下，控制表示不一致的瞬时耗散。LFHE将这种状态依赖的结构信号与早期探索和度控制相结合，同时代数连通性仍然是离线图诊断。在有界稀疏度下，其FoF候选状态保持局部，而不是扩展到人口范围的对等跟踪。在四个图像、语音和文本基准测试中，LFHE实现了具有竞争力的去中心化学习性能。匹配协议控制识别结构项作为主要的经验拓扑选择信号，而与更广泛的对等发现的比较揭示了预测性能与发现状态局部性之间的权衡。这些结果共同推动了在成对对等选择与全局信息拓扑优化之间进行状态感知的有界局部拓扑搜索。
cs.AI / 120 / 2610.08214
Mathematical Proof Assistants for Teaching Logic: The LogiKEy Methodology
用于逻辑教学的数学证明助手：LogiKEy 方法论
Abstract
We report on an approach to teaching logic to mixed groups of computer science, mathematics, and philosophy students, based on the logico-pluralistic LogiKEy methodology, used for more than a decade in courses, summer schools, and tutorials. LogiKEy uses classical higher-order logic (HOL) as a universal metalogic in which object logics, classical and non-classical alike, are encoded by defining their semantics; through these semantical embeddings a single proof assistant (e.g. Isabelle/HOL), with its automated theorem provers and (counter-)model finders, becomes one environment in which students learn, experiment with, and compare logics. After making the pedagogical case for proof assistants in the logic classroom, we present a graded sequence of classroom examples, each transition motivated by a limitation of the preceding representation, by a need for more explicit modelling resources, or by a new application. A liars-and-truth-tellers puzzle leads from propositional to modal logic; the Wise Men puzzle leads on to dynamic epistemic logic; Boolos's curious inference illustrates what a higher-order meta-logic buys, even for automated proof search; Chisholm's paradox takes the sequence into deontic logic, and from standard to dyadic deontic logic; and G\"odel's ontological argument brings it to a research-level metaphysical argument. We then rebut the objection that embedding everything in classical HOL is monism rather than pluralism, reflect on three years of teaching such a course, and sketch the portability of the approach beyond Isabelle.
Chinese Translation
我们报告了一种基于逻辑多元主义的 LogiKEy 方法论的逻辑教学方法，该方法适用于计算机科学、数学和哲学学生的混合组，已在课程、暑期学校和辅导中使用超过十年。LogiKEy 使用经典高阶逻辑（HOL）作为一种通用的元逻辑，通过定义对象逻辑的语义来编码经典和非经典逻辑；通过这些语义嵌入，单一的证明助手（例如 Isabelle/HOL）及其自动定理证明器和（反）模型查找器成为学生学习、实验和比较逻辑的一个环境。在论证逻辑课堂中使用证明助手的教学价值后，我们展示了一系列分级的课堂示例，每个过渡都是由前一表示的局限性、对更明确建模资源的需求或新的应用所驱动。一个关于说谎者和诚实者的难题引导学生从命题逻辑到模态逻辑；智者难题引导学生进入动态知识逻辑；Boolos 的奇特推理展示了高阶元逻辑的优势，即使在自动证明搜索中也是如此；Chisholm 的悖论将序列引入义务逻辑，并从标准义务逻辑转向二元义务逻辑；而哥德尔的本体论论证则将其引入研究级别的形而上学论证。我们随后反驳了将一切嵌入经典 HOL 是单一主义而非多元主义的观点，反思了三年的教学经验，并勾勒了该方法在 Isabelle 之外的可移植性。
cs.AI / 121 / 2610.08215
Learn2Play Bench: How Well Do LLM Agents Learn from Experience in Unfamiliar Environments?
Learn2Play Bench：LLM代理在不熟悉环境中从经验中学习的能力如何？
Abstract
Learning from experience is essential for LLM agents to adapt to unfamiliar and dynmaic environments. Evaluating this ability is therefore important for understanding how effectively agents acquire and use new knowledge. Existing benchmarks have sought to evaluate this ability, but they primarily evaluate tasks whose rules are provided in the instructions or already familiar to pretrained models, making it difficult to distinguish learning from interactions from reasoning with existing knowledge. To address this, we introduce Learn2Play Bench, a benchmark of newly designed text-based games, whose rules are novel or counterintuitive, requiring agents to acquire knowledge through interaction rather than rely solely on pretrained knowledge. These games provide reproducible feedback and automatic scoring, enabling controlled evaluation of learning across repeated attempts. We also vary game instances to test whether agents can apply what they have learned to new situations. Therefore, we evaluate how backbone models, self-evolving methods, and agent harnesses affect agents' learning ability, revealing three findings: (1) Experience retention: Retaining complete records of actions and feedback can support more effective learning than summarizing these experiences into rules or strategies. (2) Human agent gap: Top-performing human players achieve higher peak scores than the evaluated agents. Human explore more varied strategies, and repeat actions less. (3) Harness matters: With the backbone fixed, changing the harness can improve performance while reducing estimated inference cost. Together, these findings provide insights into how LLM agents learn from experience and suggest directions for future work to improve their learning ability. Project website: https://liushiliushi.github.io/learn2play-bench-website/
Chinese Translation
从经验中学习对LLM代理适应不熟悉和动态环境至关重要。因此，评估这种能力对于理解代理如何有效地获取和使用新知识具有重要意义。现有基准测试试图评估这种能力，但它们主要评估那些规则在说明中提供或已经为预训练模型所熟悉的任务，这使得区分通过交互学习和利用现有知识进行推理变得困难。为了解决这个问题，我们引入了Learn2Play Bench，这是一个新设计的基于文本的游戏基准，其规则是新颖或反直觉的，要求代理通过交互获取知识，而不是仅仅依赖于预训练知识。这些游戏提供可重复的反馈和自动评分，使得在多次尝试中能够对学习进行控制评估。我们还改变游戏实例，以测试代理是否能够将所学应用于新情况。因此，我们评估了基础模型、自我进化方法和代理工具对代理学习能力的影响，揭示了三项发现：（1）经验保留：保留完整的行动和反馈记录可以支持比将这些经验总结为规则或策略更有效的学习。（2）人类代理差距：表现最佳的人类玩家的最高得分超过了被评估的代理。人类探索了更多样化的策略，并且重复行动较少。（3）工具的重要性：在固定基础模型的情况下，改变工具可以提高性能，同时降低估计的推理成本。这些发现共同提供了关于LLM代理如何从经验中学习的见解，并为未来改善其学习能力的工作指明了方向。项目网站：https://liushiliushi.github.io/learn2play-bench-website/
cs.AI / 122 / 2610.08216
Quantum Entangled Multimodal Fusion Networks (QEMFN): Resource-Aware Hybrid Vision-Language Fusion via Trainable Entanglement
量子纠缠多模态融合网络 (QEMFN)：通过可训练纠缠实现资源感知的混合视觉-语言融合
Abstract
Multimodal vision-language systems typically fuse image and text embeddings through classical operators such as concatenation, attention, bilinear pooling, or tensor interactions. We propose Quantum Entangled Multimodal Fusion Networks (QEMFN), a hybrid quantum-classical framework that introduces parameterized entanglement as a structured inductive bias for multimodal fusion. Pretrained visual and textual features are projected into compact latent spaces, encoded as angle-parameterized quantum states, processed through intra-modal and paired cross-modal entangling circuits, and measured to produce fused representations for retrieval. Under matched parameter budgets and identical frozen CLIP backbones, QEMFN outperforms classical fusion baselines on COCO-5k and Flickr30k, including multilayer perceptron, tensor fusion, FiLM, cross-attention, compact transformer, and a dequantized paired-topology analogue. An ablation suite isolates the quantum module's contribution from the surrounding classical projections, and quantum-centric analyses report Meyer-Wallach entangling capability, expressibility, gradient variance against barren-plateau bounds, and entropy-performance correlation under controls for training progress alongside an intervention study on the entangling component. QEMFN is executed under shot-based estimation, a noise-modeled fake backend, and a real superconducting device with zero-noise extrapolation. This work does not claim quantum computational advantage; the contribution is the framework together with a controlled empirical and quantum-centric evaluation that positions trainable entanglement as an interpretable, hardware-executable fusion mechanism at scales accessible on contemporary devices.
Chinese Translation
多模态视觉-语言系统通常通过经典操作符（如连接、注意力、双线性池化或张量交互）融合图像和文本嵌入。我们提出了量子纠缠多模态融合网络（QEMFN），这是一种混合量子-经典框架，引入参数化纠缠作为多模态融合的结构性归纳偏置。预训练的视觉和文本特征被投影到紧凑的潜在空间中，编码为角度参数化的量子态，通过模内和配对的跨模态纠缠电路进行处理，并测量以生成用于检索的融合表示。在匹配的参数预算和相同的冻结 CLIP 主干下，QEMFN 在 COCO-5k 和 Flickr30k 数据集上超越了经典融合基线，包括多层感知机、张量融合、FiLM、交叉注意力、紧凑变换器和去量化的配对拓扑类比。消融实验套件将量子模块的贡献与周围的经典投影分离，量子中心分析报告了 Meyer-Wallach 纠缠能力、表达能力、相对于贫瘠高原界限的梯度方差，以及在训练进展控制下的熵-性能相关性，同时对纠缠组件进行了干预研究。QEMFN 在基于样本的估计、噪声建模的假后端和具有零噪声外推的真实超导设备上执行。本研究不声称具有量子计算优势；其贡献在于框架以及控制的经验和量子中心评估，将可训练的纠缠定位为当代设备可接入规模的可解释、可执行的融合机制。
cs.AI / 123 / 2610.08229
Confidence-Ordering Reversal under Contextual Priors in Neural Decoding
在上下文先验下神经解码中的置信排序逆转
Abstract
Contextual priors improve neural-to-language decoding by reshaping candidate scores. However, confidence is read from the same reshaped scores, so the errors a prior leaves behind can become more confident with no change in accuracy to reveal it. We study how a prior shapes confidence in speech retrieval on MEG-MASC and MOUS using local decoding scores, a contextual prior combined by additive shallow fusion, and the fused top-two margin as confidence. Among initially incorrect predictions, we find a confidence-ordering reversal: a larger margin makes a repair more likely when the correct candidate starts near the top of the local ranking, but less likely when it starts lower. On MEG-MASC, pooled correctness AUROC is 0.87, yet AUROC separating repairs from residual errors falls from 0.70 at initial ranks 2-3 to 0.39 at ranks 21-50. Errors starting beyond rank 20, inside the reversed region, make up 46.6% of all post-fusion errors. We propose a score-level account: a repair must first close the correct candidate's initial deficit, limiting its final margin, whereas a residual error can build a large margin between two incorrect candidates. A causal intervention that changes only the fusion weight moves the reversal to deeper ranks as predicted. Under a word-level LM prior, it keeps moving after accuracy gain peaks, so a weight chosen for accuracy does not settle confidence. Reading local and prior scores separately improves selective decoding: the decoder answers on 74.5% of windows instead of 56.7%, while 92% of output sets still contain the correct candidate. Confidence after contextual fusion should retain the local and contextual evidence behind each prediction, not just the fused scores. Project website: https://confidencereversal.github.io/; Code: https://github.com/AmadeusFake/NeuDecodingConfReversal
Chinese Translation
上下文先验通过重塑候选得分来改善神经到语言的解码。然而，置信度是从相同的重塑得分中读取的，因此先验留下的错误可能会变得更有信心，而准确性没有变化以揭示这一点。我们研究了先验如何在MEG-MASC和MOUS的语音检索中塑造置信度，使用局部解码得分、通过加性浅融合组合的上下文先验，以及作为置信度的融合前两名边际。在最初不正确的预测中，我们发现了置信排序的逆转：当正确候选者在局部排名的顶部附近时，较大的边际使修复更有可能，但当其在较低位置时则不太可能。在MEG-MASC上，汇总的正确性AUROC为0.87，然而将修复与残余错误分开的AUROC从初始排名2-3的0.70降至排名21-50的0.39。排名超过20的错误，位于逆转区域内，占所有后融合错误的46.6%。我们提出了一种得分层面的解释：修复必须首先弥补正确候选者的初始缺口，限制其最终边际，而残余错误可以在两个不正确候选者之间建立较大的边际。仅改变融合权重的因果干预使逆转移动到更深的排名，正如预测的那样。在词级语言模型先验下，逆转在准确性提升峰值后继续移动，因此为准确性选择的权重并不能稳定置信度。分别读取局部和先验得分改善了选择性解码：解码器在74.5%的窗口中给出答案，而不是56.7%，同时92%的输出集仍包含正确候选者。上下文融合后的置信度应保留每个预测背后的局部和上下文证据，而不仅仅是融合得分。项目网站：https://confidencereversal.github.io/；代码：https://github.com/AmadeusFake/NeuDecodingConfReversal
cs.AI / 124 / 2610.08231
OSFP4: Joint Optimization of Diagonal Smoothing and Block Scales for NVFP4 Quantization
OSFP4：NVFP4量化的对角平滑与块尺度的联合优化
Abstract
NVFP4 is an attractive datatype for large language model (LLM) inference, offering compact storage and native tensor-core acceleration. However, preserving accuracy using NVFP4 requires careful quantization. In this work we develop a novel quantization scheme called Optimized Smoothing and Scaling for NVFP4 (OSFP4). For each linear projection it uses a diagonal smoothing matrix whose entries are optimized to minimize the squared matrix-product quantization error under NVFP4, taking into account the rounding procedure that is used (either round-to-nearest, or GPTQ-style successive interference cancellation). This requires performing joint optimization on the smoothing entries as well as the block scales, which is facilitated by analyzing a multiplicative-dither FP4 quantizer instead of the fixed deterministic one. Experiments show that OSFP4 achieves the highest average accuracy among the evaluated competitors in the corresponding quantization settings, while retaining approximately 94-97\% of vendor NVFP4 prefill throughput on the measured workloads. Our code is available in https://github.com/neriahbd/OSFP4
Chinese Translation
NVFP4是一种适用于大型语言模型（LLM）推理的吸引人数据类型，提供紧凑的存储和原生的张量核心加速。然而，使用NVFP4保持准确性需要仔细的量化。在本研究中，我们开发了一种名为优化平滑与缩放的NVFP4量化方案（OSFP4）。对于每个线性投影，它使用一个对角平滑矩阵，其条目经过优化，以最小化在NVFP4下的平方矩阵乘积量化误差，同时考虑所使用的舍入过程（无论是四舍五入到最近值，还是GPTQ风格的连续干扰消除）。这需要对平滑条目和块尺度进行联合优化，而这一过程通过分析乘法抖动FP4量化器而非固定确定性量化器来实现。实验表明，在相应的量化设置中，OSFP4在评估的竞争者中实现了最高的平均准确性，同时在测量的工作负载上保留了约94-97%的供应商NVFP4预填充吞吐量。我们的代码可在https://github.com/neriahbd/OSFP4获取。
cs.AI / 125 / 2610.08244
Sensor-Language-Action Models
传感器-语言-动作模型
Abstract
Sensors are useful not only for understanding the world but also for deciding what to do next. Existing sensor models however largely stop at perception: they recognize states or predict outcomes, leaving actions modeled separately through task-specific and often closed label spaces. We introduce Sensor-Language-Action (SLA) modeling, a framework that connects multimodal sensor observations, natural language, and actions within a unified model. SLA uses language as a semantic interface between sensing and acting, allowing heterogeneous actions to be represented, predicted, and explained while remaining grounded in the underlying sensor evidence. We build a large-scale SLA benchmark consisting of datasets that span more than 116,000 individuals, 79 sensor modalities, and 60 action groups, together with a multi-faceted captioning pipeline that aligns user context, sensor dynamics, and action evidence. Building on this framework, we present OpenSLA, a unified SLA model for hierarchical action prediction, state understanding, and action explanation. Extensive experiments on real-world tasks in clinical prediction, operating rooms, and metabolic health verify its superior performance over the state-of-the-art. OpenSLA also demonstrates intriguing capabilities including language-guided evidence grounding and zero-shot generalization to unseen actions and cohorts.
Chinese Translation
传感器不仅对理解世界有用，还对决定下一步行动至关重要。然而，现有的传感器模型在感知方面往往止步于此：它们识别状态或预测结果，而将动作通过特定任务的、通常是封闭的标签空间单独建模。我们提出了传感器-语言-动作（Sensor-Language-Action, SLA）建模框架，该框架在统一模型中连接多模态传感器观察、自然语言和动作。SLA使用语言作为感知与行动之间的语义接口，使得异构动作能够被表示、预测和解释，同时仍然基于底层传感器证据。我们构建了一个大规模的SLA基准，包括覆盖超过116,000个个体、79种传感器模态和60个动作组的数据集，以及一个多方面的描述生成管道，能够对用户上下文、传感器动态和动作证据进行对齐。在此框架的基础上，我们提出了OpenSLA，这是一个统一的SLA模型，用于层次化动作预测、状态理解和动作解释。在临床预测、手术室和代谢健康等真实世界任务上的广泛实验验证了其优于现有技术的性能。OpenSLA还展示了引人注目的能力，包括语言引导的证据基础和对未见动作及群体的零样本泛化。
cs.AI / 126 / 2610.08246
LeanPlan: Optimal Planning with LLM-Generated Heuristics and Admissibility Proofs
LeanPlan：基于LLM生成启发式和可接受性证明的最优规划
Abstract
Frontier large language models (LLMs) can generate heuristic functions that guide search to achieve state-of-the-art performance in satisficing planning, where any plan is acceptable. However, these heuristics are not guaranteed to be admissible and can lead to suboptimal plans. We introduce LeanPlan, the first planning system that finds optimal plans with LLM-generated heuristics whose admissibility is machine-checked. Given a domain description and training tasks, an agentic loop uses planner feedback to iteratively improve a reusable domain-specific heuristic, its admissibility proof and the required domain assumptions. LeanPlan implements the heuristic, its proof and an efficient planner with machine-checked grounding and search in Lean 4. We evaluate LeanPlan on ten domains from the International Planning Competition and three new domains, using test tasks with up to 57 times as many objects as the training tasks. With GPT-5.6 Sol in the agentic loop, we successfully generate heuristics and admissibility proofs for all these domains. With the resulting heuristics, LeanPlan usually expands fewer states than the state-of-the-art Scorpion planner and solves more tasks overall.
Chinese Translation
前沿的大型语言模型（LLMs）能够生成启发式函数，指导搜索以在可满足规划中实现最先进的性能，在这种情况下，任何计划都是可接受的。然而，这些启发式并不保证是可接受的，可能导致次优计划。我们介绍了LeanPlan，这是第一个利用机器检查的LLM生成启发式找到最优计划的规划系统。在给定领域描述和训练任务的情况下，代理循环使用规划者反馈迭代改进可重用的领域特定启发式、其可接受性证明和所需的领域假设。LeanPlan在Lean 4中实现了该启发式、其证明以及具有机器检查的基础和搜索的高效规划器。我们在国际规划竞赛的十个领域和三个新领域上评估LeanPlan，使用的测试任务中的对象数量最多是训练任务的57倍。在代理循环中使用GPT-5.6 Sol，我们成功为所有这些领域生成了启发式和可接受性证明。借助生成的启发式，LeanPlan通常扩展的状态数量少于最先进的Scorpion规划器，并且整体解决的任务数量更多。
cs.AI / 127 / 2610.08250
MASC: A Multi-Agent Self-Calibration Framework with Latent Construct Alignment for Consistent Client Role-Playing in Psychological Counseling
MASC：一种具有潜在构造对齐的多智能体自校准框架，用于心理咨询中一致的客户角色扮演
Abstract
Large language models are increasingly used to simulate clients for counselor training and psychological counseling research, but reliable simulation requires clients to remain psychologically coherent across extended interactions. Existing role-playing methods largely rely on static profile prompts and may exhibit persona drift, unrealistic cooperativeness, or inconsistent psychological states, communicative actions, and emotions. Existing evaluations also lack a unified testbed for both stable client characteristics and evolving psychological dynamics. We propose MASC, a Multi-Agent Self-Calibration framework with latent construct alignment for consistent client role-playing in psychological counseling. MASC combines construct-guided generation, collaborative refinement, consistency verification, and memory-based revision in a closed calibration loop that detects and corrects inconsistencies as dialogue unfolds. We further introduce CRPC-Bench, a benchmark covering session-level profile information and Big-Five personality traits, as well as turn-level psychological state, communicative action, and emotion expression. CRPC-Bench contains 38 motivational interviewing client profiles augmented with personality and emotion annotations. Experiments show that MASC outperforms existing methods across profile, personality, receptivity, and turn-level consistency, with the heterogeneous configuration achieving the strongest overall performance. MASC and CRPC-Bench provide a unified foundation for developing and evaluating psychologically coherent client simulations for AI-assisted counseling research and training.
Chinese Translation
大型语言模型越来越多地用于模拟客户，以进行咨询师培训和心理咨询研究，但可靠的模拟要求客户在长时间的互动中保持心理一致性。现有的角色扮演方法主要依赖于静态的个人档案提示，可能会出现角色漂移、不现实的合作性或不一致的心理状态、交流行为和情感。现有评估也缺乏一个统一的测试平台，以同时评估稳定的客户特征和不断变化的心理动态。我们提出了MASC，一种具有潜在构造对齐的多智能体自校准框架，用于心理咨询中一致的客户角色扮演。MASC结合了构造引导生成、协作优化、一致性验证和基于记忆的修正，形成一个闭环校准过程，在对话展开时检测和纠正不一致性。我们进一步介绍了CRPC-Bench，一个基准涵盖了会话级别的个人档案信息和大五人格特质，以及轮次级别的心理状态、交流行为和情感表达。CRPC-Bench包含38个动机访谈客户档案，附有个性和情感注释。实验表明，MASC在个人档案、个性、接受度和轮次级别一致性方面优于现有方法，其中异构配置实现了最强的整体性能。MASC和CRPC-Bench为开发和评估心理一致的客户模拟提供了统一的基础，支持AI辅助的咨询研究和培训。
cs.AI / 128 / 2610.08258
zkLLMPoT: Efficient Zero Knowledge Proof of Training for Large Language Models
zkLLMPoT：大语言模型训练的高效零知识证明
Abstract
Auditing the claimed outcomes of large language model (LLM) training is challenging when model weights and training data are private, while cryptographically proving the full training process is prohibitively expensive at Transformer scale. We present zkLLMPoT, a zero-knowledge framework that certifies auditor-defined properties of a trained checkpoint through forward evaluation rather than verification of its optimization trajectory. zkLLMPoT includes 2 phases: 1) The trainer fixes the architecture and the model weights are committed. Then the auditor selects challenge sequences, preventing the trainer from modifying the checkpoint in response to the audit data. 2) Then the trainer proves the objective value attained by the committed model on those sequences. This formulation makes the certification cost independent of the number of training iterations, without revealing model weights or requiring access to private training data. We build on sumcheck- and lookup-based arguments to certify Transformer computations, while supporting next-token loss and task-specific audit objectives. Across four model families, operator-level benchmarks yield proving times of 41-59 seconds for 1.1-1.5B-parameter models and 131 seconds at 13B for the covered operators, with verification below half a second at a sequence length of 512.
Chinese Translation
在模型权重和训练数据为私有时，审计大语言模型（LLM）训练的声称结果是具有挑战性的，而在变换器（Transformer）规模下，密码学证明完整的训练过程则代价高昂。我们提出了zkLLMPoT，一个零知识框架，通过前向评估而非验证其优化轨迹，来认证审计员定义的训练检查点属性。zkLLMPoT包括两个阶段：1）训练者固定架构并提交模型权重。然后，审计员选择挑战序列，防止训练者根据审计数据修改检查点。2）然后，训练者证明提交模型在这些序列上获得的目标值。该公式使得认证成本与训练迭代次数无关，同时不揭示模型权重或要求访问私有训练数据。我们基于求和检查（sumcheck）和查找（lookup）方法来认证变换器计算，同时支持下一个标记损失和特定任务的审计目标。在四个模型系列中，操作级基准测试显示，对于1.1-1.5B参数模型，证明时间为41-59秒，而对于13B参数模型，证明时间为131秒，验证时间在序列长度为512时低于半秒。
cs.AI / 129 / 2610.08312
CoDe-LoRA: Mitigating the Orthogonality Dilemma in Continual Learning of LLMs via Knowledge Consolidation and Decoupling
CoDe-LoRA：通过知识整合和解耦缓解大规模语言模型持续学习中的正交性困境
Abstract
Continual learning (CL) is essential for Large Language Models (LLMs) to sequentially adapt to evolving tasks. To mitigate catastrophic forgetting, recent advances implement low-rank adaptation with orthogonal projections (e.g., O-LoRA) to isolate task parameters. However, we reveal that such strict geometric constraints trigger an "Orthogonality Dilemma": rigid parameter isolation impedes the transfer and accumulation of shared representations across semantically related tasks. In this work, we propose a new replay-free method, called Consolidation and Decoupling LoRA (CoDe-LoRA), for CL of LLMs. CoDe-LoRA disentangles the learning process into Consolidating Universal Knowledge and Decoupling Task-Specific Knowledge. To achieve this, CoDe-LoRA leverages an adaptive null space projection mechanism and semantic routing to balance knowledge accumulation with task-specific adaptation. Experimental results across four backbones and three CL benchmarks show that CoDe-LoRA achieves the best average accuracy. Our code is available at https://github.com/Estrellajer/CoDe-LoRA.
Chinese Translation
持续学习（CL）对于大规模语言模型（LLMs）顺利适应不断演变的任务至关重要。为了减轻灾难性遗忘，近期的进展采用低秩适应与正交投影（例如，O-LoRA）来隔离任务参数。然而，我们发现这种严格的几何约束引发了“正交性困境”：僵化的参数隔离阻碍了在语义相关任务之间共享表示的转移和积累。在本研究中，我们提出了一种新的无重放方法，称为整合与解耦低秩适应（CoDe-LoRA），用于大规模语言模型的持续学习。CoDe-LoRA将学习过程解构为整合通用知识和解耦任务特定知识。为此，CoDe-LoRA利用自适应零空间投影机制和语义路由，以平衡知识积累与任务特定适应。四个基础模型和三个持续学习基准的实验结果表明，CoDe-LoRA实现了最佳的平均准确率。我们的代码可在 https://github.com/Estrellajer/CoDe-LoRA 获取。
cs.AI / 130 / 2610.08314
The Standardization Trap: Certifying Joint Label Processing in Tabular Foundation Models
标准化陷阱：认证表格基础模型中的联合标签处理
Abstract
Linear regression and kernel smoothing offer tractable explanations of in-context learning: in both, the features determine the weight assigned to each context label. However, whether this fixed-weight account describes pretrained tabular foundation models (TFMs) remains unclear. Testing this account using derivatives runs into a standardization trap: public TFM packages standardize the labels before the model sees them, yet ordinary derivatives also reflect behavior outside the set of standardized labels, making a model appear nonlinear even when every prediction it makes agrees with a fixed-weight map. We propose two certificates that depend only on predictions at standardized labels and can reject two distinct explanations: fixed-weight prediction and sums of independent nonlinear label transformations. Across the five public TFMs that we evaluate, our certificates show that changing one context label alters how other labels influence the prediction, a behavior we call joint processing. We further find that joint processing emerges with training and that attention scores carry most of the measured interaction. Together, these findings motivate TFM explanations that account for how context labels change the influence of individual examples.
Chinese Translation
线性回归和核平滑为上下文学习提供了可处理的解释：在这两者中，特征决定了分配给每个上下文标签的权重。然而，这种固定权重的解释是否适用于预训练的表格基础模型（TFMs）仍不清楚。使用导数测试这一解释会遇到标准化陷阱：公共TFM包在模型看到标签之前就对其进行了标准化，然而普通导数也反映了标准化标签集之外的行为，使得模型即使在每个预测都与固定权重映射一致时也显得非线性。我们提出了两个仅依赖于标准化标签预测的证书，并能够拒绝两种不同的解释：固定权重预测和独立非线性标签变换的总和。在我们评估的五个公共TFM中，我们的证书显示，改变一个上下文标签会改变其他标签对预测的影响，这种行为我们称之为联合处理。我们进一步发现，联合处理在训练过程中出现，并且注意力得分承载了大部分测量到的交互。总的来说，这些发现促使我们对TFM的解释进行调整，以考虑上下文标签如何改变个别示例的影响。
cs.AI / 131 / 2610.08319
SCOPE: Certified Theorem Proving with a Language Model as the Policy Planner
SCOPE：以语言模型作为策略规划者的认证定理证明
Abstract
In proof assistants such as Lean, a generated proof must pass machine compilation checks, so evaluation needs no human scoring. Direct generation fails on multi-step numeric propositions: a proof is valid only if every content integer is correct, so the pass rate is bounded by the k-th power of the per-integer accuracy. Controlled corruption across 2,617 reference proofs confirms this power law. SCOPE (State-Conditioned Operator Planning and Execution) enforces the natural division of labor: the model plans over an operator vocabulary, a symbolic engine executes the numerics, and a compiler renders the proof. On a 218-problem suite it certifies 191/218 (87.6%) with a 135M backbone; the 7B DeepSeek-Prover-V1.5-RL certifies 18/218 at 27.5 times the tokens and 37.5 times the wall-clock, and DeepSeek-Prover-V2-7B certifies zero on a bidirectional dual suite. Multi-step thinking costs 6.12 discrete decision actions per problem and produces no natural-language thinking text. Replacing the lagged engine state in the decision frame with the current one lifts the pass rate from 117/218 to 191/218, while up-weighting the chain-end loss hurts. On the public Lean-Workbook library, 2,132 of 3,536 gradeable admissible problems certify (60.29%) with zero regression on the main suite. All readings come from a version-frozen review with independent rechecks and reverse verification. Restricting free generation and keeping decision-time information visible is a more direct route than enlarging the model.
Chinese Translation
在诸如 Lean 的证明助手中，生成的证明必须通过机器编译检查，因此评估不需要人工评分。直接生成在多步骤数值命题上失败：只有当每个内容整数都正确时，证明才有效，因此通过率受限于每个整数准确度的 k 次方。对 2,617 个参考证明的受控损坏验证了这一幂律。SCOPE（状态条件操作规划与执行）强制执行自然的分工：模型在操作词汇上进行规划，符号引擎执行数值计算，编译器呈现证明。在一个包含 218 个问题的套件中，它认证了 191/218（87.6%），使用了 135M 的主干；7B 的 DeepSeek-Prover-V1.5-RL 在 27.5 倍的令牌和 37.5 倍的实际时间下认证了 18/218，而 DeepSeek-Prover-V2-7B 在双向双套件上认证为零。多步骤思维每个问题需要 6.12 次离散决策行为，并且没有生成自然语言思考文本。将决策框架中的滞后引擎状态替换为当前状态，使通过率从 117/218 提升至 191/218，而加权链末损失则会造成负面影响。在公共的 Lean-Workbook 库中，3,536 个可评分的可接受问题中有 2,132 个得到了认证（60.29%），且在主要套件上没有回归。所有读数均来自版本冻结的审查，并进行了独立复查和反向验证。限制自由生成并保持决策时间信息可见，比扩大模型是一条更直接的路径。
cs.AI / 132 / 2610.08327
MedZERO: Self-Evolving Agents for Open-Ended Medical Reasoning Through Controlled Knowledge Accumulation
MedZERO：通过受控知识积累实现开放式医学推理的自我进化智能体
Abstract
Large language models (LLMs) have shown promise in medical question answering and clinical reasoning, yet their improvement remains constrained by static parametric knowledge and costly expert supervision. Self-evolving agents offer a promising alternative by enabling models to improve through iterative task generation and problem-solving. However, most existing self-evolving methods are designed for easily verifiable domains such as mathematics and coding, where solutions can be checked by exact answers or executable programs. Medical reasoning is fundamentally different: it is open-ended, knowledge-intensive, and often only partially verifiable. We present MedZERO, a self-evolving framework for open-ended medical reasoning. MedZERO couples an Examiner that generates frontier medical question-option pairs with a Reasoner that solves them through evidence-grounded multi-turn reasoning with external knowledge tools. To support reliable, continual improvement, MedZERO adopts controlled knowledge accumulation, which maintains temporary exploratory knowledge and curated persistent knowledge in reasoning. We evaluate MedZERO on five public medical reasoning benchmarks using 4B- and 8B-scale base models under open-ended evaluation. Across all settings, MedZERO consistently outperforms the underlying base models and prior self-evolving baselines, achieving up to 13.7 average accuracy-point gains over the next-best self-evolving baseline.
Chinese Translation
大型语言模型（LLMs）在医学问答和临床推理方面展现了良好的前景，但其改进仍受到静态参数知识和昂贵专家监督的限制。自我进化智能体提供了一种有前景的替代方案，通过迭代任务生成和问题解决使模型得以改进。然而，大多数现有的自我进化方法是为易于验证的领域（如数学和编程）设计的，这些领域的解决方案可以通过精确答案或可执行程序进行检查。医学推理在本质上是不同的：它是开放式的、知识密集型的，并且通常只能部分验证。我们提出了MedZERO，一个用于开放式医学推理的自我进化框架。MedZERO结合了生成前沿医学问题选项对的考察者（Examiner）和通过基于证据的多轮推理与外部知识工具解决这些问题的推理者（Reasoner）。为了支持可靠的持续改进，MedZERO采用了受控知识积累，保持临时探索性知识和策划的持久知识在推理中的应用。我们在五个公共医学推理基准上评估了MedZERO，使用4B和8B规模的基础模型进行开放式评估。在所有设置中，MedZERO始终优于基础模型和之前的自我进化基线，平均准确度提升高达13.7个百分点，超过了下一个最佳自我进化基线。
cs.AI / 133 / 2610.08329
An AI-Assisted Formalization of the Poincar\'e Conjecture
一种人工智能辅助的庞加莱猜想形式化
Abstract
We present an AI-assisted Lean 4 formalization of the Poincar\'e conjecture. The project began with limited reusable formal infrastructure for the geometric analysis behind the proof. To organize this work, we combined a proof blueprint prepared by mathematicians with explicit milestone statements. These milestones enabled parallel agent work and gave mathematicians clear points to locate blockers and provide effective mathematical guidance. Our analysis identifies the human interventions and organizational choices behind this workflow. The project provides a starting point toward reusable infrastructure for future formalization projects; such infrastructure, once developed, could eventually reduce the cost of verifying mathematical results in geometric analysis.
Chinese Translation
我们提出了一种人工智能辅助的 Lean 4 形式化庞加莱猜想的方案。该项目起初依赖于有限的可重用形式基础设施，以支持证明背后的几何分析。为了组织这项工作，我们将数学家准备的证明蓝图与明确的里程碑声明相结合。这些里程碑使得并行代理工作成为可能，并为数学家提供了明确的定位阻碍点和有效的数学指导的机会。我们的分析识别了这一工作流程背后的人为干预和组织选择。该项目为未来形式化项目提供了可重用基础设施的起点；一旦开发完成，这种基础设施最终可能降低几何分析中验证数学结果的成本。
cs.AI / 134 / 2610.08330
MoF: Preference-Aware Mixture Modeling for Black-Box LLM Personalization
MoF：基于偏好的混合建模用于黑箱大语言模型个性化
Abstract
Proprietary Large Language Models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks, yet aligning their outputs with diverse user preferences remains challenging. Existing personalization approaches for black-box LLMs often rely on user-specific scoring heads, causing the number of personalized parameters to grow linearly with the number of users and requiring additional adaptation for unseen users. To address these limitations, we propose Mixture-of-Facets (MoF), a scalable personalization framework for black-box LLMs that models user preferences as compositions of shared latent preference facets rather than dedicated user-specific parameters. MoF performs personalization through history-conditioned routing over shared facet heads, enabling personalization for users unseen during training without additional parameter updates. Across diverse personalization tasks, MoF delivers stronger personalization performance while maintaining a more scalable and parameter-efficient design than prior approaches. Additional analysis indicates strong generalization to unseen users.
Chinese Translation
专有的大语言模型（LLMs）在广泛的任务中展现了显著的能力，但将其输出与多样化的用户偏好对齐仍然具有挑战性。现有的黑箱LLM个性化方法通常依赖于用户特定的评分头，导致个性化参数的数量随着用户数量线性增长，并且需要对未见用户进行额外的适应。为了解决这些局限性，我们提出了混合面（Mixture-of-Facets, MoF），这是一个可扩展的黑箱LLM个性化框架，它将用户偏好建模为共享潜在偏面（latent preference facets）的组合，而不是专门的用户特定参数。MoF通过基于历史的路由在共享面头上执行个性化，使得在训练期间未见的用户也能够实现个性化，而无需额外的参数更新。在多样化的个性化任务中，MoF提供了更强的个性化性能，同时保持比先前方法更具可扩展性和参数效率的设计。额外的分析表明，对未见用户具有强大的泛化能力。
cs.AI / 135 / 2610.08363
Explainable Failure Prediction and Prevention in Maritime
海事领域可解释的故障预测与预防
Abstract
Maritime systems operate in highly dynamic environments where unexpected equipment failures can compromise safety, reliability, and operational efficiency. Recent advances in artificial intelligence (AI), machine learning, digital twins, and predictive maintenance enable proactive failure prediction and prevention. However, ensuring trustworthy and explainable decision-making remains a major challenge in safety-critical maritime applications. This chapter reviews key AI technologies required for explainable failure prediction and prevention in maritime systems and presents a conceptual architecture capable of supporting autonomous or human-in-the-loop corrective actions. This architecture integrates data acquisition, time-series forecasting, anomaly detection, risk assessment, decision-making, and explainable AI into a closed-loop framework. With reference to the architectural components, a review and discussion of relevant maritime studies is performed, outlining their methods, advantages, and limitations. Furthermore, it highlights current challenges, including uncertainty and robustness, model generalization, explainability, limited availability of maritime datasets, and operational deployment, and identifies future research directions toward trustworthy AI-assisted maritime decision-making.
Chinese Translation
海事系统在高度动态的环境中运行，意外的设备故障可能会危及安全、可靠性和操作效率。近年来，人工智能（AI）、机器学习、数字双胞胎和预测性维护的进展使得主动故障预测和预防成为可能。然而，在安全关键的海事应用中，确保可信赖和可解释的决策仍然是一个主要挑战。本章回顾了海事系统中可解释故障预测与预防所需的关键AI技术，并提出了一种能够支持自主或人机协作的纠正措施的概念架构。该架构将数据采集、时间序列预测、异常检测、风险评估、决策制定和可解释AI整合到一个闭环框架中。针对架构组件，进行了相关海事研究的回顾与讨论，概述了它们的方法、优点和局限性。此外，还强调了当前面临的挑战，包括不确定性和鲁棒性、模型泛化、可解释性、海事数据集的有限可用性以及操作部署，并确定了未来朝向可信赖的AI辅助海事决策的研究方向。
cs.AI / 136 / 2610.08364
Transect: Retaining Observability for Long-Horizon LLM Agent Evaluations
Transect：保持长期视野下LLM代理评估的可观察性
Abstract
Frontier AI evaluations increasingly use open-ended, agentic, long-horizon tasks whose transcripts can span hundreds of pages of outputs and actions from complex multi-agent networks. The observability envelop-the range of what evaluators can reliably infer about an agent's behaviours-is therefore narrowing. Language model assistants can help classify and interpret agent behaviour but also afford human evaluators significant analytical degrees of freedom, threatening the reproducibility and auditability of language-model-based transcript analysis. Transect is an open source package built on Inspect Scout to help evaluators understand how a long agent run unfolded, identify behaviour worth investigating, and check interpretations against the transcript. Users specify task context and behavioural vocabulary in a reusable evaluation-family configuration, with judge models and analysis settings supplied separately. Transect's navigable reports align recorded events, token use, sub-agent activity, and model-generated behavioural labels on a common turn-based timeline. Reviewers can quickly grasp a run's narrative, trace any label or event to its source turns, and export the underlying data tables for cross-run analysis. We demonstrate the workflow on an AI R&D evaluation that generated almost 13 million tokens, dividing the agents' work into behavioural phases aligned with research-skill classifications, sub-agent delegations and interactions, and token use. The combined view shows a focus on operational work and manuscript production, with little evidence of a sustained hypothesis generation stage-arguably a necessary component for high-quality scientific outputs. Transect's flexible, customisable transcript-analysis pipeline will enable evaluators to keep pace with longer, more complex, more frequent AI evaluations while supporting scientific rigour, transparency, and reproducibility.
Chinese Translation
前沿人工智能评估越来越多地使用开放式、代理性、长期任务，其转录文本可能跨越数百页来自复杂多代理网络的输出和行为。因此，可观察性范围——评估者可以可靠推断代理行为的范围——正在缩小。语言模型助手可以帮助分类和解释代理行为，但也赋予人类评估者显著的分析自由度，这威胁到基于语言模型的转录分析的可重复性和可审计性。Transect是一个基于Inspect Scout构建的开源软件包，旨在帮助评估者理解长期代理运行的展开过程，识别值得调查的行为，并根据转录文本检查解释。用户在可重用的评估家庭配置中指定任务背景和行为词汇，评判模型和分析设置则单独提供。Transect的可导航报告将记录事件、标记使用、子代理活动和模型生成的行为标签对齐在一个共同的回合基础时间线上。评审者可以快速掌握运行的叙事，追踪任何标签或事件到其源回合，并导出基础数据表以进行跨运行分析。我们在一次生成了近1300万个标记的人工智能研发评估中演示了这一工作流程，将代理的工作划分为与研究技能分类、子代理委派和互动以及标记使用对齐的行为阶段。综合视图显示了对操作性工作和手稿生产的关注，而几乎没有持续假设生成阶段的证据——这可以说是高质量科学成果的必要组成部分。Transect灵活、可定制的转录分析管道将使评估者能够跟上更长、更复杂、更频繁的人工智能评估，同时支持科学严谨性、透明性和可重复性。
cs.AI / 137 / 2610.08432
EMHO: EMbodied Agent Harness Optimization via Experience Traces
EMHO：通过经验轨迹优化具身代理的代理框架
Abstract
Improving embodied agents often focuses on optimizing the underlying model through training, while the surrounding agent harness that controls planning, context, and tool use is typically engineered. We ask whether this harness can instead improve itself directly from experience traces under sparse environmental feedback. We propose EMbodied Agent Harness Optimization (EMHO), a self-evolving framework that keeps the embodied model frozen and iteratively revises its harness by analyzing execution trajectories and prior harness history. EMHO optimizes beyond skills or recovery prompts, modifying how the agent monitors progress, uses vision tools, grounds observations, and responds to failures. To support multiple subtasks with a single harness, we introduce EMHO-Merge, which addresses trade-offs in jointly optimizing a single shared harness across subtasks by using episode-level gains and losses to guide evidence-supported refinement of when and how revised behaviors are applied. We evaluate EMHO on EmbodiedBench across navigation and manipulation tasks, and EMHO consistently improves task success for both Qwen 9B and 27B models. Qualitative analysis shows that EMHO goes beyond recovering from failures and unproductive actions to reshape how the embodied agent interprets and interacts with its environment.
Chinese Translation
改善具身代理通常侧重于通过训练优化基础模型，而控制规划、上下文和工具使用的周围代理框架通常是经过工程设计的。我们探讨这种框架是否可以直接从稀疏环境反馈下的经验轨迹中自我改进。我们提出了具身代理框架优化（EMHO），这是一个自我演化的框架，它保持具身模型不变，并通过分析执行轨迹和先前的框架历史来迭代修订其框架。EMHO 超越了技能或恢复提示，修改了代理监控进展、使用视觉工具、基础观察和应对失败的方式。为了支持单一框架下的多个子任务，我们引入了 EMHO-Merge，它通过使用情节级别的收益和损失来指导证据支持的修订行为的应用时机和方式，解决了在子任务之间共同优化单一共享框架的权衡。我们在 EmbodiedBench 上评估 EMHO，涵盖导航和操作任务，EMHO 一直在 Qwen 9B 和 27B 模型上持续提高任务成功率。定性分析表明，EMHO 超越了从失败和无效行动中恢复，重塑了具身代理如何解释和与其环境互动的方式。
cs.AI / 138 / 2610.08446
AssemState: Manual and Physical-State-Guided Reasoning for Zero-shot Furniture Assembly
AssemState：零-shot家具组装的手动和物理状态引导推理
Abstract
Multimodal large language models (MLLMs) have made significant progress in visual understanding, but precise 3D spatial reasoning integrated with physical environment remains difficult. Furniture assembly requires not only recovering step-level operations from diagrammatic manuals, but also translating semantic attachment relations into 6D pose updates that enable parts to physically interact with the environment and previously assembled components. To study this problem, we propose AssemState, a zero-shot framework for manual and physical-state-guided furniture assembly. It firstly employs anchor-guided boundary assembly states to decompose manual pages into single-part operations and recover an assembly-tree. Then, it uses iterative after-state feedback refinement to guide successive (SE(3)) updates and corrections, and validates their physical plausibility through simulation-based release tests. Experiments show that compared with the strongest prior baseline, AssemState improves F1 from 38.58\% to 62.80\% and Tree Exact Match from 28.24\% to 53.92\% for assembly-tree recovery. On 243 independently evaluated part-level operations, our proposed iterative refinement improves judge-accepted operations from 0 to 5.3\% and reduces mean Chamfer distance from 5.4111 to 1.7744. However, visually plausible candidate poses may still suffer from collision, floating, mirror-orientation errors, incomplete seating, and wrong-side attachment. These results show that AssemState improves operation-structure recovery and selected local pose metrics, while MLLMs remain limited for spatial relationship reasoning.
Chinese Translation
多模态大型语言模型（MLLMs）在视觉理解方面取得了显著进展，但与物理环境相结合的精确三维空间推理仍然困难。家具组装不仅需要从图示手册中恢复逐步操作，还需要将语义附着关系转换为6D姿态更新，以使部件能够与环境及先前组装的组件进行物理交互。为研究这一问题，我们提出了AssemState，一个用于手动和物理状态引导家具组装的零-shot框架。该框架首先采用锚点引导的边界组装状态，将手册页面分解为单个部件操作，并恢复组装树。然后，它使用迭代的后状态反馈细化来指导连续的（SE(3)）更新和修正，并通过基于仿真的释放测试验证其物理可行性。实验表明，与最强的先前基线相比，AssemState在组装树恢复方面将F1从38.58%提高到62.80%，将树精确匹配从28.24%提高到53.92%。在243个独立评估的部件级操作中，我们提出的迭代细化将评审接受的操作从0%提高到5.3%，并将平均Chamfer距离从5.4111减少到1.7744。然而，视觉上可行的候选姿态仍可能遭遇碰撞、漂浮、镜像方向错误、不完整的就座和错误侧附着等问题。这些结果表明，AssemState在操作结构恢复和选定局部姿态指标方面有所改善，而MLLMs在空间关系推理方面仍然存在局限。
cs.AI / 139 / 2610.08510
Cylindrical Geodesic Flow Matching for Quasiperiodic Physiological Signal Transformation
准周期生理信号变换的圆柱几何流匹配
Abstract
Paired translation between quasiperiodic physiological waveforms (i.e., recovering a target oscillatory signal from the source) is central to the interpretation of cardiovascular signals derived from wearables placed at different body locations. This source-to-target mapping in these problems carries inherent geometric structure: the phase wraps around the cycle and must be treated as a circular variable, the amplitude remains strictly positive, and the beat-to-beat alignment can drift unpredictably across cycles and subjects. While deep neural networks have been used for phase estimation and complex-valued signal modeling, prior work does not explicitly learn phase transport between paired signals. Consequently, neither endpoint-supervised regression nor the standard affine path used in flow matching accounts for this phase--amplitude structure. We introduce \emph{cylindrical geodesic flow matching} for paired cardiovascular waveform translation. We show that the standard affine path used in flow matching distorts intermediate amplitude and instantaneous frequency when interpolating between quasiperiodic signals; replacing it with a closed-form geodesic on the phase--amplitude cylinder eliminates these artifacts and converts each training pair into dense, geometry-consistent velocity supervision. On zero-shot photoplethysmography and limited-support seismocardiography adaptation benchmarks, our method consistently outperforms interpolation baselines and matches or exceeds direct supervised prediction, reducing Hilbert Transform, $L_2$, and Dynamic Time Warping distance by up to ${\sim}15\%$ over the strongest competing baseline. These results suggest that bridge geometry is a critical inductive bias for flow matching on oscillatory signal translation.
Chinese Translation
准周期生理波形（即从源信号恢复目标振荡信号）之间的配对转换是解释来自不同身体位置的可穿戴设备所获取的心血管信号的核心。这些问题中的源到目标映射具有固有的几何结构：相位围绕周期循环并必须视为圆形变量，幅度始终保持严格正值，并且每次心跳的对齐在周期和个体之间可能会不可预测地漂移。尽管深度神经网络已被用于相位估计和复值信号建模，但之前的工作并未明确学习配对信号之间的相位传输。因此，端点监督回归和流匹配中使用的标准仿射路径都未考虑这一相位-幅度结构。我们提出了用于配对心血管波形转换的 extit{圆柱几何流匹配}。我们展示了流匹配中使用的标准仿射路径在准周期信号之间插值时会扭曲中间幅度和瞬时频率；将其替换为相位-幅度圆柱上的闭合形式测地线消除了这些伪影，并将每个训练对转换为密集的、几何一致的速度监督。在零-shot 光电容积描记法和有限支持心脏地震图适应基准测试中，我们的方法始终优于插值基线，并与直接监督预测相匹配或超过，减少了希尔伯特变换、$L_2$ 和动态时间规整距离，最高可达约 ${ extsim}15\%$，相较于最强竞争基线。这些结果表明，桥接几何是振荡信号转换流匹配的重要归纳偏置。
cs.AI / 140 / 2610.08514
How Much Evidence Should a Coding Agent's Self-Correction Carry? Adaptive Dirichlet Evidence for Self-Distillation
编码代理的自我修正应承载多少证据？自适应狄利克雷证据用于自我蒸馏
Abstract
Execution feedback lets coding agents revise programs and learn from their own corrections. A correction's learning weight should reflect both the transitions supported by its executions and the amount of evidence behind that support. We introduce Effective-Evidence Self-Distillation (EESD), which represents these quantities separately. Normalized execution relevance determines relative transition support and an effective pseudo-count mass; a Dirichlet posterior then produces an uncertainty-penalized weight for KL-anchored correction learning. Under a symmetric prior, changing mass preserves category ordering, and effective mass yields a supervised coefficient bounded by its matched fixed-mass counterpart. Across four model-domain history sweeps, increasing visible observations from one to eight reduces future-outcome NLL by 55.0-59.3%. At eight observations, effective mass achieves lower NLL than fixed mass in all four comparisons. In the primary matched DeepSeek/RunBugRun study, argmax predictions agree on all 3,000 examples, with the largest NLL gain under concentrated relevance. After one correction-learning round, DeepSeek/CodeARC all-tests Pass@1 increases from 15.0% to 20.4%, with a paired 95% source-bootstrap interval of [+2.8, +8.0] percentage points. The twelve-setting downstream evaluation establishes the model-domain scope of this update. These results show how separating evidence support from evidence mass changes probability estimation and correction learning in coding agents.
Chinese Translation
执行反馈使编码代理能够修订程序并从自身的修正中学习。修正的学习权重应反映其执行所支持的过渡和支持该过渡的证据量。我们引入了有效证据自我蒸馏（Effective-Evidence Self-Distillation, EESD），分别表示这些量。归一化的执行相关性决定了相对的过渡支持和有效伪计数质量；然后，狄利克雷后验生成一个带有不确定性惩罚的权重，用于基于KL的修正学习。在对称先验下，改变质量保持类别排序，有效质量产生一个受限于其匹配固定质量对应物的监督系数。在四个模型-领域历史的实验中，从一个可见观察增加到八个可见观察，未来结果的负对数似然（NLL）减少了55.0-59.3%。在八个观察下，有效质量在所有四个比较中实现了比固定质量更低的NLL。在主要的匹配DeepSeek/RunBugRun研究中，argmax预测在所有3000个示例上达成一致，在集中相关性下获得了最大的NLL增益。在一次修正学习轮次后，DeepSeek/CodeARC的所有测试Pass@1从15.0%提高到20.4%，配对的95%源自助区间为[+2.8, +8.0]个百分点。十二个设置的下游评估确立了此更新的模型-领域范围。这些结果表明，将证据支持与证据质量分开如何改变编码代理的概率估计和修正学习。
cs.AI / 141 / 2610.08540
Toward Alignment Scaling Laws: A Framework and First Preregistered Measurements
迈向对齐规模法则：一个框架及首次预注册测量
Abstract
Whether alignment gets easier or harder as models grow is often argued from isolated findings, as if alignment were one property. We treat it as a family of measurable scaling relations: for each risk category r, the alignment burden needed to hold a fixed safety target is modeled as B_r(N)=a_rN^alpha_r, with N a capability proxy; against a budget proportional to N, scaling helps if alpha_r<1, keeps pace if alpha_r~1, and accumulates alignment debt if alpha_r>1. We give three operationalizations of burden and distinguish observed, audited and true alignment. A toy model, in which corrections consume capability headroom, makes the consequences explicit. We prove that the largest exponent among corrected risks, not an average, sets the long-run regime; that above 1 any policy holding headroom above a floor must grow super-exponentially; that, for burdens that are positive mixtures of power laws, fits on small models underestimate large-scale exponents; and that an audit that uncovers hidden failures without false positives never underestimates true alignment. We propose a pre-registrable protocol and apply reduced versions of it twice. A preregistered reanalysis of public adversarial-training data for Pythia classifiers finds that the compute needed to bring attack success under 10% grows as N^0.60. A preregistered pilot on Qwen2.5 0.5B-72B finds exponents of -0.05 for truthfulness and 0.48 for stated dispositions (both scaling helps under its reduced rule, though local slopes approach 1 at the top; replicated on Qwen3 0.6B-14B), while sycophancy (0.89, or 0.83 with two seeds added at 72B) and a planted backdoor are undetermined: the backdoor is removed quickly when its trigger is known but survives blind safety training at four of five sizes. We release four browser games that play these laws (www.aisafety.fun). We make no claim about which regime holds for current frontier models.
Chinese Translation
关于随着模型规模的增长，对齐是否变得更容易或更困难，常常基于孤立的发现进行争论，仿佛对齐是一个单一的属性。我们将其视为一系列可测量的规模关系：对于每个风险类别 r，维持固定安全目标所需的对齐负担被建模为 B_r(N)=a_rN^alpha_r，其中 N 是能力的代理；相对于与 N 成比例的预算，当 alpha_r<1 时，规模效应有利；当 alpha_r~1 时，保持同步；而当 alpha_r>1 时，则积累对齐债务。我们提供了三种负担的操作化，并区分观察到的、审计的和真实的对齐。一个玩具模型中，修正消耗能力余量，使后果变得明确。我们证明，在修正风险中，最大的指数，而非平均值，决定了长期的状态；当指数大于 1 时，任何保持余量在某一底线以上的政策必须以超指数方式增长；对于正混合的幂律负担，小模型上的拟合会低估大规模的指数；而揭示隐藏失败而没有假阳性的审计永远不会低估真实的对齐。我们提出了一个可预注册的协议，并对其进行了两次简化版本的应用。对 Pythia 分类器的公共对抗训练数据进行的预注册再分析发现，将攻击成功率控制在 10% 以下所需的计算量随着 N^0.60 增长。对 Qwen2.5 0.5B-72B 的预注册试点发现，真实度的指数为 -0.05，陈述倾向的指数为 0.48（在其简化规则下均有规模效应，尽管局部斜率在顶部接近 1；在 Qwen3 0.6B-14B 上复制），而谄媚（0.89，或在 72B 时添加两个种子为 0.83）和植入后门的情况尚未确定：当其触发器已知时，后门会迅速被移除，但在五种规模中的四种情况下，盲安全训练仍然能够存活。我们发布了四个浏览器游戏来演示这些法则（www.aisafety.fun）。我们不对当前前沿模型所处的状态做出任何声明。
cs.AI / 142 / 2610.08552
AnyBottle: A Recipe to Only Keep the Concepts You Really Need
AnyBottle：仅保留您真正需要的概念的方案
Abstract
Concept bottleneck models (CBMs) make predictions inspectable and intervenable by routing them through human-interpretable concepts, but originally required concept annotations. Annotation-free variants remove this requirement, but typically use large concept vocabularies, static at both training and inference, producing bottlenecks larger than any task or prediction needs and harder to inspect. We propose AnyBottle, a single recipe for building compact, task-specific CBMs. AnyBottle assumes only a frozen backbone and an unsupervised concept pool, such as a sparse autoencoder. A black-box teacher trained on the same backbone then guides selection: each round adds the concept that best explains the bottleneck's current failures, with candidates restricted to regions of teacher/student disagreement. Trained with nested dropout over this selection order, the final bottleneck predicts accurately from any concept prefix, so inference spends fewer concepts on inputs it is confident about early and more on hard ones. Since no stage is modality-specific, a new domain and task requires swapping only the backbone and concept pool. Across six vision and two text datasets and two teacher paradigms, AnyBottle yields bottlenecks with fewer concepts and higher concept consistency than annotation-free baselines, while staying close to the black-box reference. Overall, AnyBottle shows that going annotation-free need not mean going large: a small, discovered vocabulary can be as expressive as a much larger, fixed one.
Chinese Translation
概念瓶颈模型（CBMs）通过将预测路由到人类可解释的概念，使得预测可检查和可干预，但最初需要概念注释。无注释的变体消除了这一要求，但通常使用大型概念词汇，在训练和推理时都是静态的，导致瓶颈大于任何任务或预测需求，并且更难以检查。我们提出了AnyBottle，这是一种构建紧凑、任务特定CBMs的单一方案。AnyBottle仅假设一个冻结的主干网络和一个无监督的概念池，例如稀疏自编码器。然后，一个在相同主干上训练的黑箱教师指导选择：每一轮添加最佳解释瓶颈当前失败的概念，候选者限制在教师/学生不一致的区域。通过对这一选择顺序进行嵌套丢弃训练，最终的瓶颈能够从任何概念前缀准确预测，因此推理在对早期输入有信心时使用更少的概念，而在困难的输入上使用更多的概念。由于没有任何阶段是特定于某种模态的，因此新的领域和任务只需更换主干网络和概念池。通过六个视觉和两个文本数据集以及两个教师范式，AnyBottle产生的瓶颈具有比无注释基线更少的概念和更高的概念一致性，同时与黑箱参考保持接近。总体而言，AnyBottle表明，无需注释并不意味着必须使用大型模型：一个小而发现的词汇可以与一个更大且固定的词汇一样具有表现力。
cs.AI / 143 / 2610.08563
Adaptive Power Sampling for LLM Reasoning
适应性功率采样用于大型语言模型推理
Abstract
Sequence-level power sampling has recently emerged as a training-free approach to reasoning by sampling from a sharpened output distribution of a base large language model (LLM). Nevertheless, existing methods typically sharpen the base model distribution uniformly across queries, overlooking variations in query difficulty and in how well the base model already handles each query. The goal of this work is to equip power sampling with query adaptivity. Theoretically, we show that the benefits of further sharpening are determined by the self-reward gap between correct and incorrect responses. Based on this insight, we propose \emph{Adaptive Power Sampling} (APS), which adjusts the sharpening exponent on a per-query basis at test time using the relationship between answer agreement and the model's self-reward. Experiments across diverse reasoning tasks, including MATH500, HumanEval, and GPQA, show that APS consistently outperforms power sampling with a fixed sharpening exponent, without additional training.
Chinese Translation
序列级功率采样最近作为一种无训练的方法出现，通过从基础大型语言模型（LLM）的锐化输出分布中进行采样来进行推理。然而，现有方法通常在查询之间均匀地锐化基础模型分布，忽视了查询难度的变化以及基础模型对每个查询的处理能力。本文的目标是为功率采样赋予查询自适应性。从理论上讲，我们表明进一步锐化的好处取决于正确和错误响应之间的自我奖励差距。基于这一见解，我们提出了 extit{适应性功率采样}（Adaptive Power Sampling，APS），该方法在测试时根据答案一致性与模型自我奖励之间的关系，针对每个查询调整锐化指数。在包括MATH500、HumanEval和GPQA在内的多种推理任务上的实验表明，APS在没有额外训练的情况下，始终优于固定锐化指数的功率采样。
cs.AI / 144 / 2610.08586
MINDSET: Energy-based Schema Evolution for Long Conversational Agent Memory
MINDSET：基于能量的模式演化用于长对话体代理记忆
Abstract
Long conversational agents have become essential in our daily lives. They must remember what was said long back in order to help us efficiently complete a task without needing the user to repeat instructions and context repeatedly. However, the main issue is that instructions and context change over time and so the agents must be able to adapt accordingly. A useful memory system should preserve both current and historical states, distinguish stale information from active knowledge, retrieve evidence appropriate to the query and avoid repeatedly invoking a large language model to rewrite prior interactions. We introduce MINDSET, a memory controller that stores a conversation as immutable episodes and organizes them into versioned schemas through minimum-energy state transitions. Each incoming episode may reinforce, supersede, split or create a schema. The transition decision balances representation distortion, contradiction, historical damage, fragmentation and internal inconsistency, while hysteresis prevents isolated contradictions from prematurely rewriting stable memory. We evaluate MINDSET against 5 memory systems on a reproducible sample of 850 questions (700 LoCoMo + 150 MemoryAgentBench). MINDSET obtains the highest observed LoCoMo answer F1 while significantly improving retrieval ranking (Recall@8, MRR and nDCG@8) over the second best method LightMem (p<0.01 after Holm correction). It obtains the highest observed scores on MemoryAgentBench although the relative difference is low. Ablations identify controlled fragmentation and schema-aware assignment as the largest contributors to answer quality. Additionally, a 700-question cross-model evaluation with GLM-4.7 and Gemma-4-31B supported model independence. These results show that long-term memory can be better handled as constrained state management rather than continual summarization.
Chinese Translation
长对话体代理在我们的日常生活中变得至关重要。它们必须记住很久以前所说的话，以帮助我们高效完成任务，而无需用户重复指令和上下文。然而，主要问题在于指令和上下文会随着时间而变化，因此代理必须能够相应地适应。一个有用的记忆系统应该同时保留当前状态和历史状态，区分过时信息与活跃知识，检索与查询相关的证据，并避免反复调用大型语言模型来重写先前的交互。我们提出了MINDSET，一个将对话存储为不可变事件并通过最小能量状态转变将其组织成版本化模式的记忆控制器。每个传入的事件可能会加强、取代、拆分或创建一个模式。转变决策在表示失真、矛盾、历史损害、碎片化和内部不一致之间进行平衡，而滞后效应则防止孤立的矛盾过早重写稳定的记忆。我们在可重复的850个问题样本（700个LoCoMo + 150个MemoryAgentBench）上对MINDSET与5个记忆系统进行了评估。MINDSET在观察到的LoCoMo答案F1上获得了最高分，同时在检索排名（Recall@8、MRR和nDCG@8）上显著改善了第二好的方法LightMem（在Holm校正后p<0.01）。尽管相对差异较小，但它在MemoryAgentBench上获得了最高的观察分数。消融实验确定了受控碎片化和模式感知分配是答案质量的最大贡献者。此外，与GLM-4.7和Gemma-4-31B的700个问题跨模型评估支持了模型独立性。这些结果表明，长期记忆可以更好地作为受限状态管理来处理，而不是持续的总结。
cs.AI / 145 / 2610.08621
Recursive Game Creator: An Agentic Product-Level Experience-Oriented Game Harness
递归游戏创作者：一种面向体验的代理产品级游戏框架
Abstract
Recent game design agents have made substantial progress in generating playable games. However, program correctness does not ensure an enjoyable experience for players. We present Recursive Game Creator, an experience-oriented harness to advance agentic game development from rough game prototypes into entertaining games. Recursive Game Creator organizes recursive development around four components: Designer, Builder, Player, and Reviewer. The Designer translates user instructions and Reviewer's feedback into detailed plans. The Builder turns these plans into candidate games. The coding-native Player creates and executes reusable policies through programmatic interfaces to efficiently collect diverse gameplay trajectories, mitigating evaluation bias caused by slow GUI-based collection. The Reviewer uses carefully designed trajectory-based metrics to induce player preferences, integrating with visual evidence and explicit textual preferences to evaluate games against game-specific criteria. Finally, the Reviewer accepts the better version and provides improvement reviews for the next round, closing the recursive loop. Our method achieves state-of-the-art overall performance of 77.89 on GameCraft-Bench. On GameASG-Bench, it achieves a strict task success rate of 53.2%, a 34.1% improvement over the same-model baseline, and the highest mean runtime-check pass rate at 93.4% among compared methods. A user study shows longer playtime and higher ratings. Code is coming soon.
Chinese Translation
近期游戏设计代理在生成可玩游戏方面取得了显著进展。然而，程序的正确性并不能保证玩家的愉悦体验。我们提出了递归游戏创作者（Recursive Game Creator），这是一个面向体验的框架，旨在将代理游戏开发从粗糙的游戏原型推进到娱乐性游戏。递归游戏创作者围绕四个组件组织递归开发：设计者（Designer）、构建者（Builder）、玩家（Player）和评审者（Reviewer）。设计者将用户指令和评审者的反馈转化为详细计划。构建者将这些计划转化为候选游戏。编码原生的玩家通过编程接口创建和执行可重用策略，以高效收集多样化的游戏玩法轨迹，从而减轻因缓慢的基于图形用户界面（GUI）收集而造成的评估偏差。评审者使用精心设计的基于轨迹的指标来诱导玩家偏好，并结合视觉证据和明确的文本偏好来根据游戏特定标准评估游戏。最后，评审者接受更好的版本，并为下一轮提供改进意见，闭合递归循环。我们的方法在GameCraft-Bench上实现了77.89的最先进整体性能。在GameASG-Bench上，它实现了53.2%的严格任务成功率，比同模型基线提高了34.1%，并在比较方法中获得了93.4%的最高平均运行时检查通过率。一项用户研究显示了更长的游戏时间和更高的评分。代码即将发布。
cs.AI / 146 / 2610.08627
Parallel Predictive World Models for Accurate and Efficient Long-Horizon Planning
用于准确高效的长时间规划的并行预测世界模型
Abstract
Long-horizon world-model planning typically relies on autoregressive rollouts, where predicted states are repeatedly fed back into the model. This preserves temporal structure but creates a horizon-length sequential path and exposes later predictions to recursive decoded-state feedback. We introduce Parallel Predictive World Models (PPWM), which predict a finite-horizon trajectory in parallel while retaining causal interaction among future representations. Each horizon is conditioned on its causal action prefix, and future representations interact before decoding, separating temporal causality from state-by-state output recursion. We formalize this distinction by viewing autoregressive rollout as a causal trajectory map and identifying the decoded-state feedback pathway removed by PPWM. Across four visual-control tasks, PPWM achieves the lowest long-horizon prediction error and the highest Cross-Entropy Method (CEM) simulator success among the evaluated predictive interfaces. Meanwhile, PPWM achieves more than a 3$\times$ average CEM planning speedup over the autoregressive LeWM baseline. These results suggest that accurate and efficient long-horizon world-model planning does not require state-by-state autoregression, but can instead be achieved through parallel causal trajectory prediction.
Chinese Translation
长时间世界模型规划通常依赖于自回归展开，其中预测状态被反复反馈到模型中。这保持了时间结构，但产生了一个与时间长度相对应的序列路径，并使后续预测暴露于递归解码状态反馈中。我们提出了并行预测世界模型（Parallel Predictive World Models, PPWM），该模型在保持未来表示之间因果交互的同时，平行预测有限时间范围内的轨迹。每个时间范围都以其因果动作前缀为条件，未来表示在解码之前相互作用，从而将时间因果性与逐状态输出递归分离。我们通过将自回归展开视为因果轨迹图，并识别PPWM所消除的解码状态反馈路径，形式化了这一区别。在四个视觉控制任务中，PPWM在评估的预测接口中实现了最低的长时间预测误差和最高的交叉熵法（Cross-Entropy Method, CEM）模拟器成功率。同时，PPWM在自回归LeWM基线之上实现了超过3倍的平均CEM规划加速。这些结果表明，准确高效的长时间世界模型规划不需要逐状态自回归，而可以通过并行因果轨迹预测来实现。
cs.AI / 147 / 2610.08647
SquidAgent: Parallelize Wisely, Coordinate Efficiently
SquidAgent：明智地并行化，高效地协调
Abstract
LLM-based agents solve complex multi-step tasks, but sequential execution incurs substantial latency. In principle, parallelizing work across multiple agents should yield near-linear speedups. Yet existing parallel multi-agent systems often run slower than a single-agent baseline. We attribute this gap to two hidden costs that parallel execution incurs but a serial agent avoids. First, there is a re-exploration cost: redundant effort spent by parallel workers reconstructing context that the orchestrator already possesses, such as prior decisions, that would otherwise be inherited implicitly in a serial execution. Second, there is an alignment cost: the overhead required to reconcile inconsistencies across independently generated outputs. We thus derive a principled decision criterion: a layer should be parallelized only when its critical-path cost, plus re-exploration and alignment overheads, is lower than the corresponding serial cost. While this criterion is naturally expressed in wall-clock time, we observe that LLMs are poorly calibrated when asked to estimate task duration. To address this, we instead measure cost in predicted output tokens, which we empirically find LLMs can estimate substantially more reliably than wall-clock time. Building on this token-based criterion, we propose SquidAgent. It estimates all token budgets in a single planning step, forks each worker directly from the orchestrator's session to eliminate re-exploration cost, and replaces post-hoc reconciliation with a pre-generated shared convention block that converts alignment into a bounded upfront cost. A deterministic scheduler then applies the criterion layer by layer. Empirically, SquidAgent achieves a 2.2$\times$ mean throughput improvement and a 2.6$\times$ mean wall-time speedup over Claude Code, and a 2.0$\times$ throughput improvement over the strongest multi-agent baseline.
Chinese Translation
基于大语言模型（LLM）的代理能够解决复杂的多步骤任务，但顺序执行会带来显著的延迟。从理论上讲，在多个代理之间并行化工作应该能实现接近线性的加速。然而，现有的并行多代理系统往往运行速度低于单代理基线。我们将这一差距归因于并行执行所产生的两个隐性成本，而串行代理则可以避免。首先是重新探索成本：并行工作者在重建协调者已经掌握的上下文（例如先前的决策）时所花费的冗余努力，这些上下文在串行执行中本应隐式继承。其次是对齐成本：协调独立生成的输出之间不一致所需的额外开销。因此，我们推导出一个原则性的决策标准：仅当某一层的关键路径成本加上重新探索和对齐的开销低于相应的串行成本时，该层才应进行并行化。虽然这一标准自然以实际时间表达，但我们观察到，当要求LLM估计任务持续时间时，其校准效果较差。为了解决这一问题，我们改为用预测输出标记来衡量成本，我们实证发现LLM在估计预测输出标记时的可靠性远高于实际时间。基于这一基于标记的标准，我们提出了SquidAgent。它在单个规划步骤中估计所有标记预算，直接从协调者的会话中分叉每个工作者，以消除重新探索成本，并用预生成的共享约定块替代事后协调，将对齐转化为有限的前期成本。然后，确定性调度器逐层应用这一标准。实证结果表明，SquidAgent在Claude Code上实现了2.2倍的平均吞吐量提升和2.6倍的平均实际时间加速，并在最强多代理基线之上实现了2.0倍的吞吐量提升。
cs.AI / 148 / 2610.08662
ParanoiaEval: Benchmarking Unnecessary Defensive Work in Agentic Coding
ParanoiaEval：代理编码中不必要防御工作的基准评估
Abstract
As coding agents increasingly undertake real-world work autonomously, judging whether their risk treatments are warranted has become important. Existing work evaluates related agent behaviors from separate perspectives, but lacks a systematic framework for unifying these behaviors. To bridge this gap, we introduce ParanoiaEval, the first benchmark for unified evaluation of risk-treatment capabilities in coding agents. Grounded in the well-established Avoidance-Transfer-Mitigation-Acceptance framework in software engineering risk management, ParanoiaEval operationalizes its 4 fundamental treatments for coding-agent settings and contains 200 evidence-controlled repository-level task pairs, each differing only in treatment-defining evidence. We further introduce dedicated metrics for risk-treatment violations and evidence responsiveness, using a human-calibrated agentic judge for reliable evaluation. Large-scale experiments on 8 representative models and a post-hoc human study reveal that (I) unnecessary risk treatment occurs in 11.2%-58.7% of runs despite explicit evidence, with substantial variation across agent configurations; (II) stronger task capability does not ensure more appropriate risk treatment, while treatment violations substantially harm developers' experience, establishing risk treatment as an independent capability dimension; and (III) agents exhibit systematic patterns consistent with established risk-management findings, suggesting that knowledge from human practice can guide the diagnosis and improvement of this capability.
Chinese Translation
随着编码代理越来越多地自主承担现实世界的工作，判断其风险处理是否合理变得至关重要。现有研究从不同角度评估相关代理行为，但缺乏统一这些行为的系统框架。为填补这一空白，我们引入了ParanoiaEval，这是第一个用于统一评估编码代理风险处理能力的基准。ParanoiaEval基于软件工程风险管理中成熟的回避-转移-缓解-接受（Avoidance-Transfer-Mitigation-Acceptance）框架，将其四种基本处理方法应用于编码代理环境，并包含200对证据控制的仓库级任务对，每对任务仅在定义处理的证据上有所不同。我们进一步引入了专门的指标来衡量风险处理违规和证据响应性，使用人类校准的代理评判者进行可靠评估。在8个代表性模型上进行的大规模实验和后续的人类研究表明：（I）尽管有明确证据，不必要的风险处理在11.2%-58.7%的运行中发生，且在代理配置之间存在显著差异；（II）更强的任务能力并不确保更恰当的风险处理，而处理违规会显著损害开发者的体验，确立了风险处理作为一个独立的能力维度；（III）代理表现出与已建立的风险管理发现一致的系统模式，表明人类实践中的知识可以指导这一能力的诊断和改进。
cs.AI / 149 / 2610.08683
Coupled but Late: Turn-Taking Between Full-Duplex Speech Models in Unscripted Dialogue
耦合但延迟：无脚本对话中全双工语音模型的轮流发言
Abstract
Full-duplex speech models are trained to converse with a person, but they are increasingly made to converse with each other, in self-play data generation, agent societies, and model-based evaluation. In that loop no human absorbs a timing error: each model's turn-taking is the other's input. We ask what timing the loop settles into. Two PersonaPlex-7B instances exchange audio tokens on a shared clock in unscripted conversation, and one floor-transfer rule is applied to them and to Switchboard. Their timing is coupled: re-pairing speakers across conversations destroys it. But the floor changes hands late, at a median of 400-560 ms against 137 ms for humans, and the last 120 ms of the partner's turn, where human projection places a tenth of its transfers, holds 1% of theirs. Delaying one direction of the channel shifts the response one-for-one and leaves the run-up to it empty, consistent with a reactive wait after the perceived end rather than the turn-end projection human timing requires.
Chinese Translation
全双工语音模型被训练用于与人类对话，但它们越来越多地被用于相互对话，应用于自我对弈数据生成、智能体社群和基于模型的评估。在这个循环中，没有人类吸收时间误差：每个模型的轮流发言都是另一个模型的输入。我们探讨这个循环最终会达到什么样的时序。两个PersonaPlex-7B实例在无脚本对话中在共享时钟上交换音频标记，并对它们及Switchboard应用一个楼层转移规则。它们的时序是耦合的：在对话中重新配对发言者会破坏这一点。但楼层的转移发生得较晚，媒体值为400-560毫秒，而人类则为137毫秒，伴侣发言的最后120毫秒中，人类投射的十分之一的转移发生在此，而它们的转移仅占1%。延迟一个方向的通道会将响应一一对应地推迟，并使其前期保持空白，这与在感知结束后进行反应性等待一致，而不是人类时序所需的轮次结束投射。
cs.AI / 150 / 2610.08691
ScienceClaw: Benchmarking Continual Self-Evolution of AI-for-Science Agents Across the Natural and Social Sciences
ScienceClaw：在自然科学和社会科学中基准测试AI科学代理的持续自我进化
Abstract
Large language model agents are accelerating scientific automation, yet verified executions rarely become persistent program-level improvements, and existing evaluations do not examine this process across sequential tasks in both the natural and social sciences. We formalize ScienceClaw as fixed-parameter program self-evolution that unifies task solving, scientific verification, and program updates. ScienceClaw-Eval spans 23 disciplines and measures scientific correctness, evolutionary gain, retention, cross-dataset transfer, and evolution cost through sequential streams and independent reset evaluation. Our framework repairs executable workflows through multi-turn interaction, converts re-execution-verified failure--success trajectories into linked Skill and Operator candidates, and retains an update only when source-task replay reproduces the repair and independent scientific tasks improve. Code is available at https://github.com/beita6969/ScienceClaw.
Chinese Translation
大型语言模型代理正在加速科学自动化，但经过验证的执行很少转化为持久的程序级改进，现有评估也未能考察这一过程在自然科学和社会科学的连续任务中的表现。我们将ScienceClaw形式化为固定参数的程序自我进化，统一任务求解、科学验证和程序更新。ScienceClaw-Eval涵盖23个学科，测量科学正确性、进化增益、保留率、跨数据集迁移和进化成本，通过连续流和独立重置评估来进行评估。我们的框架通过多轮交互修复可执行工作流，将重新执行验证的失败-成功轨迹转化为关联的技能和操作候选，并仅在源任务重放再现修复且独立科学任务改进时保留更新。代码可在 https://github.com/beita6969/ScienceClaw 获取。
cs.AI / 151 / 2610.08699
nanoMuse: An Open-Source Personal Agent for Every Device You Own
nanoMuse：适用于您所有设备的开源个人代理
Abstract
Assistants from 2011 answered and waited, and agents from 2023 did a task and stopped. In September 2026 Meta's Muse showed an agent for one person, with accounts, devices, memory and a conversation that lasts, closed, in a vendor's cloud, in one country. Such an agent is expected to act on a person's accounts and devices, remember them across weeks, speak first when it is worth it, and answer for what it did. It is a kind of software, not a model, and until now had no open counterpart. This report defines the personal agent in five questions and three horizons. It reads how Muse is built from Meta's public record and a copy of its production prompt, each statement marked by its source. It then presents nanoMuse, the open-source counterpart under the GPL-3.0, one agent on every device a person owns, with hands on the phone's screen and the computer's. They share one conversation over a relay anyone can run; every action goes through a Sentinel, memory is files the person can read, and the model is their choice. Its size and cost are given as estimates. What is open, memory with provenance, an evaluation suite for the hands and an open model for them, is set out as a roadmap.
Chinese Translation
2011年的助手回答问题并等待，而2023年的代理完成任务后停止。2026年9月，Meta的Muse展示了一种为单个用户服务的代理，具备账户、设备、记忆和持续的对话，所有数据存储在一个国家的供应商云中。这样的代理预计将能够处理个人的账户和设备，跨周记住这些信息，在值得时主动发言，并对其行为负责。它是一种软件，而非模型，迄今为止没有开放的对应物。本报告通过五个问题和三个视角定义个人代理。它分析了Muse是如何基于Meta的公开记录和其生产提示的副本构建的，每个陈述都有其来源标记。接着，报告介绍了nanoMuse，一个在GPL-3.0下的开源对应物，能够在每个用户拥有的设备上运行，能够在手机屏幕和计算机上进行操作。它们通过任何人都可以运行的中继共享一个对话；每个动作都经过一个哨兵，记忆以文件形式存在，用户可以阅读，模型由他们选择。其规模和成本以估算形式给出。开放的内容包括具有来源的记忆、适用于用户操作的评估套件以及开放模型，作为未来发展的路线图。
cs.AI / 152 / 2610.08720
WorldSolver: Can LLM Agents Simulate the Physical Dynamics via Solver Generation?
WorldSolver：大型语言模型代理能否通过求解器生成模拟物理动态？
Abstract
LLM-based agents are increasingly advancing scientific and engineering problem solving, with physics simulation emerging as a challenging yet practical testbed for reproducing complex physical phenomena with application in embodied AI, games and films. As the workhorse of such simulation, a solver computes how the state of a dynamic system evolves over time. Building such solvers requires physical understanding to identify appropriate models, mathematical reasoning to formulate the underlying dynamics, and software engineering to implement them as executable code, yet this capability of LLM agents remains underexplored. To this end, we introduce WorldSolver, a benchmark of 168 simulation tasks derived from physical phenomena in 61 classic computer graphics papers, spanning 7 physical domains. Each task contains a code scaffold that provides a fixed simulation environment for the scene, with the solver implementation left for the agent to complete. Specifically, we evaluate them along three dimensions: Execution Checks for successful execution, Visual Fidelity for reproducing the intended dynamic behavior in the rendered simulation, and Physical Plausibility for physics-grounded verification of the generated dynamics. Experiments on frontier agents reveal that producing executable solvers is difficult itself, and satisfying visual and physical correctness is even harder. GPT-5.6-Sol and Claude-Opus-5 perform comparatively better than the other evaluated agents, yet achieve overall scores of only 48.7% and 46.7%, respectively. WorldSolver is an early step toward agentic solver generation, and we hope it helps drive progress toward agents that can faithfully simulate the dynamic physical world. Code is available at https://github.com/sirujiang/WorldSolver.
Chinese Translation
基于大型语言模型（LLM）的代理在科学和工程问题解决方面正日益取得进展，物理模拟作为一个具有挑战性但实用的测试平台，正在用于重现复杂的物理现象，应用于具身人工智能、游戏和电影等领域。作为这种模拟的核心，求解器计算动态系统状态随时间的演变。构建这样的求解器需要物理理解以识别适当的模型，数学推理以制定基础动态，以及软件工程以将其实现为可执行代码，然而LLM代理在这方面的能力仍然未被充分探索。为此，我们引入了WorldSolver，一个由61篇经典计算机图形学论文中的物理现象衍生的168个模拟任务的基准，涵盖7个物理领域。每个任务包含一个代码框架，为场景提供固定的模拟环境，而求解器的实现则留给代理完成。具体而言，我们从三个维度对其进行评估：执行检查（Execution Checks）以确保成功执行，视觉保真度（Visual Fidelity）以重现渲染模拟中的预期动态行为，以及物理合理性（Physical Plausibility）以对生成的动态进行物理基础的验证。对前沿代理的实验表明，生成可执行的求解器本身就是一项困难的任务，而满足视觉和物理正确性则更为艰难。GPT-5.6-Sol和Claude-Opus-5的表现相较于其他评估的代理更为出色，但总体得分仅为48.7%和46.7%。WorldSolver是朝着代理求解器生成迈出的早期一步，我们希望它能推动朝着能够忠实模拟动态物理世界的代理的进展。代码可在 https://github.com/sirujiang/WorldSolver 获取。
cs.AI / 153 / 2610.08722
Does an Agent's History Tell You When Compaction Will Hurt? A Modest, Bounded Effect on the TRACE Paired-Replay Corpus
代理的历史是否能告诉你何时压缩会造成损害？TRACE配对重放语料库中的适度、有界效应
Abstract
Many long-horizon agents compact their context on a global rule, usually a token budget, blind to what the agent was doing. We ask whether the agent's recent behaviour predicts when a compaction will hurt. TRACE's public corpus of 590 harness-triggered AppWorld compaction boundaries replays each boundary from a re-executed prefix state under the pre-compaction context and under the summary, and records the burden of the next actions: calls that error or repeat a call already made. We find that pre-boundary history predicts post-compaction harm only weakly. An internally prespecified contrast by prefix placement is a wide null, and the naive "has-written" label behind it turns out to measure trajectory phase. The best extension-protocol trigger reaches held-out AUROC 0.66 (0.64 on the replicate's own label) against a same-boundary replicate of 0.72; the best frozen, interpretable trigger avoids 21% of harmful (positive-burden) boundaries while keeping 84% of compaction opportunities, and exceeds the random-rule expectation on count but not on burden mass (a post hoc comparison). Whether the best trigger beats a token-budget rule at matched retention cannot be evaluated on the release. We state what corpora should ship to answer it.
Chinese Translation
许多长时间跨度的代理在全球规则（通常是令牌预算）下压缩其上下文，而对代理的行为视而不见。我们探讨代理的近期行为是否能预测何时压缩会造成损害。TRACE的公共语料库包含590个由触发的AppWorld压缩边界，从重新执行的前缀状态下重放每个边界，分别在压缩前的上下文和摘要下，并记录下一步行动的负担：错误调用或重复已调用的操作。我们发现，前边界历史仅能弱预测压缩后的损害。通过前缀位置进行的内部预设对比是一个广泛的无效结果，而其背后的天真“已写入”标签最终测量的是轨迹阶段。最佳扩展协议触发器在持出样本上达到AUROC 0.66（在复制品自身标签上为0.64），而同边界复制品为0.72；最佳的冻结、可解释触发器在保持84%压缩机会的同时避免了21%的有害（正负担）边界，并在计数上超过随机规则的期望，但在负担质量上未能超越（后验比较）。在匹配保留条件下，无法评估最佳触发器是否优于令牌预算规则。我们说明了应提供哪些语料库以回答这一问题。
cs.AI / 154 / 2610.08761
VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning
VeriFine：扩展自我提升的具身推理验证
Abstract
Self-improving policies continually expose new failure patterns, changing what their judges must be able to verify. However, current fixed judges constrain both optimization feedback and the discovery of useful training examples, limiting further self-improvement. This challenge is even more acute in embodied reasoning, where reliable evaluation must account for spatial grounding, causal reasoning, and safety-aware decision-making. We introduce VeriFine, an agent harness framework that scales verification through the co-evolution of the policy, training curriculum, and judge. The Policy Improvement Loop uses a rubric judge to diagnose recurring failures, construct an adaptive curriculum, and optimize the policy. When progress plateaus and verification becomes a bottleneck, the Judge Improvement Loop selectively queries human guidance on informative failure cases and refines the judge through coactive calibration, in which humans and agents resolve disagreements and converge toward the objective rubric of physical reasoning. The revised judge then guides the next stage of data selection and policy optimization. Experiments on driving and robot navigation tasks demonstrate continuous self-improvement in both policy and judge capability across reinforcement and supervised fine-tuning. These results show how scaling verification supports continuous self-improvement as policy failure patterns evolve.
Chinese Translation
自我提升的策略不断暴露出新的失败模式，改变了评估者必须能够验证的内容。然而，当前固定的评估者限制了优化反馈和有用训练示例的发现，从而限制了进一步的自我提升。这一挑战在具身推理中尤为突出，因为可靠的评估必须考虑空间基础、因果推理和安全意识决策。我们提出了VeriFine，一个通过策略、训练课程和评估者的共同进化来扩展验证的代理框架。策略改进循环使用评估标准来诊断重复出现的失败，构建自适应课程，并优化策略。当进展停滞且验证成为瓶颈时，评估者改进循环选择性地询问人类对信息性失败案例的指导，并通过共同校准来完善评估者，在这一过程中，人类和代理解决分歧并朝着物理推理的目标标准收敛。修订后的评估者随后指导下一阶段的数据选择和策略优化。对驾驶和机器人导航任务的实验表明，在强化学习和监督微调中，策略和评估者能力均实现了持续的自我提升。这些结果表明，扩展验证如何支持随着策略失败模式演变而持续的自我提升。
cs.AI / 155 / 2610.08775
Agent in a Bottle: Can LLM Agents Turn Their Capabilities Into Cheap, Scalable Artifacts?
瓶中代理：大型语言模型代理能否将其能力转化为廉价、可扩展的产物？
Abstract
Large language models (LLMs) can solve many narrow tasks, but querying them separately for millions of related instances can be prohibitively expensive. Can LLM agents autonomously create cheaper solutions for such workloads? We call this ability "bottling": the ability to turn general capabilities into task-specific solutions that balance answer quality and amortised cost. We introduce BOTTLED, a benchmark in which agents receive an entire unlabelled workload and must complete it under fixed time, compute and LLM API budgets. Agents choose their own approach, such as training a small model or writing a reusable program. Across ten models and three tasks, we find that strong zero-shot task performance does not reliably translate into strong bottling capabilities. Models with similar zero-shot scores can differ substantially after bottling, and 48 of 60 bottling runs score below the lower bound of the 95% confidence interval of their model's zero-shot performance. Moreover, 31 of 60 runs underperform the stronger of two small-model distillation baselines with the same token budget. Nevertheless, bottling can yield substantial savings: on query-product relevance classification, Opus 5 retains about 82% of its zero-shot macro-F1 at roughly 657 times lower reported cost. Bottling is also competitive with Jev, a "system one" model built especially for cheap, repetitive inference: Opus 5 on the same task recovers about 94% of Jev's macro-F1 at a quarter of Jev's projected full-workload cost. BOTTLED provides a basis for evaluating and improving agents' ability to invest limited resources in reusable solutions for large, repetitive workloads.
Chinese Translation
大型语言模型（LLMs）可以解决许多狭窄的任务，但单独查询它们以处理数百万个相关实例可能会非常昂贵。LLM代理能否自主创建更便宜的解决方案以应对此类工作负载？我们将这种能力称为“瓶装”：将通用能力转化为特定任务解决方案的能力，以平衡答案质量和摊销成本。我们引入了BOTTLED，一个基准测试，其中代理接收整个未标记的工作负载，并必须在固定的时间、计算和LLM API预算下完成它。代理可以选择自己的方法，例如训练一个小模型或编写一个可重用的程序。在十个模型和三个任务中，我们发现强大的零-shot任务表现并不可靠地转化为强大的瓶装能力。具有相似零-shot分数的模型在瓶装后可能会有显著差异，60次瓶装运行中有48次得分低于其模型零-shot表现的95%置信区间的下限。此外，60次运行中有31次的表现低于两个小模型蒸馏基准中较强的一个，且它们的token预算相同。尽管如此，瓶装仍然可以带来显著的节省：在查询-产品相关性分类任务中，Opus 5在报告的成本约低657倍的情况下保留了大约82%的零-shot宏F1。瓶装在与Jev（一个专为廉价、重复推理构建的“系统一”模型）的比较中也具有竞争力：在同一任务上，Opus 5恢复了Jev宏F1的约94%，而其成本仅为Jev预计的全工作负载成本的四分之一。BOTTLED为评估和改善代理在大型、重复工作负载中投资有限资源于可重用解决方案的能力提供了基础。
cs.AI / 156 / 2610.08778
Sherpa: Teaching LLMs to Teach Adaptively
Sherpa：教会大型语言模型自适应教学
Abstract
Large language models (LLMs) have become increasingly capable problem solvers, but being able to solve a problem is not the same as being able to teach it. Existing approaches to training LLMs as teachers rely on demonstrations, preference data, or predefined pedagogical criteria that specify what good teaching looks like. However, these signals are often not grounded in individual student learning outcomes, where effective teaching strategies can vary substantially across learners. To address this, we introduce Sherpa, a multi-turn reinforcement learning framework that instantiates multiple student archetypes with LLMs conditioned on distinct learning preferences and trains a teacher model to adapt its instruction by directly maximizing their learning outcomes. Teacher LLMs trained with Sherpa improve instructed students' performance across all archetypes by an average of 20.5 percentage points. Under MathTutorBench's evaluation, Sherpa raises the overall pedagogy score from 52.5% to 79.2%, indicating better teaching responses. Our human studies show that the trained teacher is preferred over the base model in 79.6% of pairwise comparisons. Together, Sherpa trains LLM teachers to adapt to diverse simulated students and become better aligned with human teachers, paving the road towards AI tutors teaching real students.
Chinese Translation
大型语言模型（LLMs）已成为越来越有能力的问题解决者，但能够解决问题并不等同于能够教授该问题。现有的将LLMs训练为教师的方法依赖于示范、偏好数据或预定义的教学标准，这些标准规定了良好教学的表现。然而，这些信号往往没有基于个别学生的学习成果，而有效的教学策略在不同学习者之间可能存在显著差异。为了解决这个问题，我们引入了Sherpa，一个多轮强化学习框架，它实例化多个学生原型，并根据不同的学习偏好对LLMs进行条件化训练，进而训练一个教师模型，通过直接最大化学生的学习成果来调整其教学。使用Sherpa训练的教师LLMs在所有原型中提高了被指导学生的表现，平均提升了20.5个百分点。在MathTutorBench的评估下，Sherpa将整体教学评分从52.5%提高到79.2%，表明教学反应得到了改善。我们的人工研究显示，在79.6%的成对比较中，训练后的教师模型优于基础模型。总的来说，Sherpa训练LLM教师以适应多样化的模拟学生，并与人类教师更好地对齐，为AI辅导员教授真实学生铺平了道路。
cs.CL / 1 / 2610.06889
Zero-Shot Visualization: Exploring Text Corpora with User-Prompted Axes
零-shot 可视化：通过用户提示的轴线探索文本语料库
Abstract
We study the application of large language models (LLMs) to the visual exploration of textual corpora. We introduce zero-shot visualization (ZSV), a task in which users specify concepts in natural language and documents are mapped onto the corresponding concept axes for visualization. Building a ZSV system of practical value is non-trivial, as it requires choices at the intersection of feature functions, efficient implementation tradeoffs, and pre/post-processing decisions affecting visualization quality. To that end, we establish a benchmark that compares methods spanning embedding similarity, direct semantic judgments, and conditional likelihood estimation in this setting. Across multiple datasets and use cases we evaluate the properties of different scoring methods and design choices in terms of semantic faithfulness, score fidelity, and computational cost. Our results identify that scoring based on next-token probabilities offers the strongest practical trade-off among the evaluated methods. We further apply this approach to unlabeled corpora to examine its behavior in realistic exploratory settings. These experiments highlight additional design considerations, including the use of graded axes together with binary relevance filtering, and reveal a compositional sentiment bias in off-topic documents. Based on these findings, we provide practical guidelines for constructing end-to-end ZSV baselines.
Chinese Translation
我们研究了大型语言模型（LLMs）在文本语料库可视化探索中的应用。我们引入了零-shot 可视化（ZSV），这是一个任务，用户可以用自然语言指定概念，并将文档映射到相应的概念轴上进行可视化。构建一个具有实际价值的 ZSV 系统并非易事，因为它需要在特征函数、有效实现权衡以及影响可视化质量的前后处理决策之间做出选择。为此，我们建立了一个基准，比较了在此设置中跨越嵌入相似性、直接语义判断和条件似然估计的方法。在多个数据集和使用案例中，我们评估了不同评分方法和设计选择在语义忠实性、评分保真度和计算成本方面的属性。我们的结果表明，基于下一个令牌概率的评分在评估的方法中提供了最强的实际权衡。我们进一步将这种方法应用于未标记的语料库，以检查其在现实探索环境中的表现。这些实验突出了额外的设计考虑，包括将分级轴与二元相关性过滤结合使用，并揭示了在离题文档中的组合情感偏见。基于这些发现，我们提供了构建端到端 ZSV 基线的实用指南。
cs.CL / 2 / 2610.06897
Capacity, Responsiveness and Alignment: What Makes a Latent Structure Actionable
容量、响应性与对齐：什么使潜在结构可操作
Abstract
Localizing latent structures in the activation space of language models (LMs) is central to understanding and controlling their behavior. Yet, localized structures can differ substantially in their causal influence, raising the question of what makes a structure actionable. We tackle this question by casting causal influence as a product of three factors and showing empirically that they act as interpretable, distinct constraints: capacity, measuring the sensitivity of the model's output to movement along the structure, responsiveness, capturing how promotable the concept is given the current context, and alignment, reflecting how well the structure aligns with the context-specific representation of the concept. Across 4 LM families and 50 concepts, we observe that causal effectiveness requires all factors to be high; low capacity and responsiveness reduce it by 84% and 95%, respectively, while low alignment can reverse it, suppressing concept expression. Moreover, we find that causality is context-dependent rather than an intrinsic property of the structure, with causally effective directions forming a low-dimensional subspace that varies across contexts. By restricting the training of linear probes to this subspace, we introduce causal probes that achieve 17%-118% improvement in steering across models, with only 3% reduction in concept detection.
Chinese Translation
在语言模型（LMs）的激活空间中定位潜在结构对于理解和控制其行为至关重要。然而，局部化的结构在因果影响上可能存在显著差异，这引发了什么使结构可操作的问题。我们通过将因果影响视为三个因素的乘积来解决这个问题，并实证显示这些因素作为可解释的、独特的约束条件发挥作用：容量，衡量模型输出对沿结构移动的敏感性；响应性，捕捉在当前上下文中概念的可推广性；以及对齐，反映结构与概念的上下文特定表示之间的契合程度。在4个语言模型家族和50个概念中，我们观察到因果有效性要求所有因素都处于高水平；低容量和响应性分别降低了84%和95%的因果有效性，而低对齐则可能使其反转，抑制概念的表达。此外，我们发现因果性是上下文依赖的，而不是结构的内在属性，因果有效的方向形成了一个低维子空间，该子空间在不同上下文中变化。通过将线性探针的训练限制在这个子空间内，我们引入了因果探针，在模型间实现了17%-118%的引导改进，同时概念检测仅减少了3%。
cs.CL / 3 / 2610.06902
Tree Navigation Without LLM Summaries: A Matched-Cost Study of Hierarchical Retrieval for Long-Document QA
无需 LLM 摘要的树形导航：长文档问答的分层检索匹配成本研究
Abstract
Retrieval-augmented generation grounds language models in external context, but for long documents flat top-$k$ retrieval can cluster on a single region and miss complementary evidence. RAPTOR-style summary trees address this by recursively clustering chunks and using a language model to summarize each cluster at indexing time, then ranking summary nodes alongside raw chunks at query time. We show the main benefit of summary trees in long-document QA can come from navigation rather than the generated summary content. We introduce NavTree, a leaves-only retriever that builds a deterministic balanced segment tree over chunks (zero language-model calls at indexing) and uses the tree purely as a navigation scaffold: a hybrid lexical-and-dense frontier walk, anchored on top retrieved leaves, descends from the root and emits only leaf chunks to the reader. On a matched-cost evaluation against flat retrievers and an extractive re-implementation of RAPTOR, NavTree is the strongest matched-cost hierarchical retriever in our evaluated grid and ties the strongest flat baseline. On long-document multi-hop QA, it is the only hierarchical method that significantly beats BM25 on a class-vs-class basis, corroborated by a reader-free retrieval-recall check. A matched-reader replication of the published abstractive RAPTOR variant, given strong cluster summaries, still loses to NavTree at every multi-chunk budget, at zero indexing cost. The ranking carries across stronger and open-weight readers, a stronger encoder, and a full factorial that isolates leaves-only emission as the structural lever.
Chinese Translation
检索增强生成将语言模型与外部上下文相结合，但对于长文档，平面 top-$k$ 检索可能会聚集在单一区域，错过互补证据。RAPTOR 风格的摘要树通过递归聚类块并在索引时使用语言模型对每个聚类进行摘要来解决这一问题，然后在查询时将摘要节点与原始块一起进行排名。我们表明，摘要树在长文档问答中的主要好处来自于导航，而非生成的摘要内容。我们引入了 NavTree，这是一种仅使用叶节点的检索器，它在块上构建一个确定性的平衡段树（在索引时不调用语言模型），并仅将树作为导航支架：一种混合的词汇和稠密前沿行走，锚定在顶部检索的叶节点，从根节点向下遍历，仅向读者输出叶块。在与平面检索器和 RAPTOR 的提取式重新实现进行匹配成本评估时，NavTree 是我们评估网格中最强的匹配成本分层检索器，并与最强的平面基线平分秋色。在长文档多跳问答中，它是唯一显著超越 BM25 的分层方法，且在类与类的基础上得到了无读者检索-召回检查的证实。与已发布的抽象 RAPTOR 变体的匹配读者复制相比，尽管有强大的聚类摘要，NavTree 在每个多块预算中仍然胜过它，且索引成本为零。该排名在更强的开放权重读者、更强的编码器和一个完全的因子分析中保持一致，后者将仅叶节点的输出作为结构杠杆。
cs.CL / 4 / 2610.06903
Component and Dimension Sparsity in Transformer Refusal Mechanisms
变压器拒绝机制中的组件和维度稀疏性
Abstract
Activation steering manipulates large language model behavior by intervening on internal activations, but the mechanistic basis of these interventions remains poorly understood. We decompose refusal steering into component-level interventions across four open-weight models, identifying the sparse subsets of attention and MLP components whose steering suffices to reproduce the full behavioral effect. We find that refusal directions concentrate in sparse component mechanisms comprising 28--48\% of upstream components, retaining 88--101\% of steering effectiveness. Within these mechanisms, effective steering further concentrates in approximately 50\% of residual stream dimensions, retaining 85--98\% of the component-mechanism baseline, consistent with a privileged basis structure. Sparsity thus operates at two levels: which components are steered, and which dimensions within those components carry the signal. Together these findings show that refusal is not diffusely encoded across a transformer but assembled by a structured, identifiable mechanism, providing a foundation for mechanistic understanding of how refusal behaviors are represented and steered. To facilitate reproducibility, we release all code and raw experimental results in https://github.com/wang-research-lab/Refusal_Mechanisms.
Chinese Translation
激活引导通过干预内部激活来操控大型语言模型的行为，但这些干预的机制基础仍然不够清晰。我们将拒绝引导分解为四个开放权重模型中的组件级干预，识别出稀疏的注意力和多层感知器（MLP）组件子集，这些子集的引导足以重现完整的行为效果。我们发现拒绝方向集中在稀疏组件机制中，占上游组件的28%至48%，保持了88%至101%的引导有效性。在这些机制中，有效引导进一步集中在大约50%的残余流维度上，保持了85%至98%的组件机制基线，这与特权基础结构一致。因此，稀疏性在两个层面上发挥作用：一是哪些组件被引导，二是这些组件内哪些维度携带信号。这些发现共同表明，拒绝并不是在变压器中广泛编码的，而是通过一个结构化的、可识别的机制组装而成，为理解拒绝行为如何被表示和引导提供了基础。为了促进可重复性，我们在 https://github.com/wang-research-lab/Refusal_Mechanisms 发布了所有代码和原始实验结果。
cs.CL / 5 / 2610.06940
Stabilizing language models under continual learning via condition-anchored distillation
通过条件锚定蒸馏在持续学习中稳定语言模型
Abstract
Continual adaptation of language models can change their output distribution on prompts learned earlier, while retaining every old prompt-answer pair may be undesirable or impossible. We study condition-anchored generative distillation (CAGD): retain a small set of old prompts, use a frozen previous model to reconstruct completions and generation states, and match its predictive distributions while learning the next task. The formulation separates three roles that ordinary replay conflates: conditions select the behavior to protect, teacher generations locate relevant states, and soft targets specify how predictions may change. For autoregressive language generation, teacher-rollout distillation admits an exact chain-rule decomposition of sequence divergence. For masked-diffusion language modeling, our implementation directly controls local denoising drift on teacher-generated completions. In continual adaptation of a 219M masked diffusion language model, CAGD reduces four-task final held-out loss from 2.927 to 1.114 in one task order and from 2.168 to 0.891 in exact reverse. The same soft targets lower final average loss by 0.055 over hard replay when teacher-generated support is held identical. The direction persists on fresh facts and natural instructions across SMDM and Qwen3. On GSM8K, Qwen adaptation preserves answer-format compliance, but exact-match retention is seed-mixed at 0.6B and worsens at 1.7B. These results support condition-anchored functional preservation as a common design principle across the tested language-generation objectives.
Chinese Translation
语言模型的持续适应可能会改变其对早期学习的提示的输出分布，而保留每一个旧的提示-答案对可能是不可取的或不可能的。我们研究了条件锚定生成蒸馏（Condition-Anchored Generative Distillation, CAGD）：保留一小部分旧提示，使用冻结的先前模型重构完成和生成状态，并在学习下一个任务时匹配其预测分布。该公式将普通重放混淆的三个角色分开：条件选择要保护的行为，教师生成定位相关状态，软目标指定预测可能如何变化。对于自回归语言生成，教师回放蒸馏允许序列发散的精确链式法则分解。对于掩蔽扩散语言建模，我们的实现直接控制教师生成完成的局部去噪漂移。在一个219M的掩蔽扩散语言模型的持续适应中，CAGD将四任务最终保留损失从2.927降低到1.114（在一个任务顺序中），并在完全反向中从2.168降低到0.891。相同的软目标在教师生成支持保持不变时，将最终平均损失降低了0.055，相较于硬重放。在SMDM和Qwen3中，该方向在新事实和自然指令上持续存在。在GSM8K上，Qwen适应保持答案格式的一致性，但在0.6B时精确匹配保留混合为种子，并在1.7B时恶化。这些结果支持条件锚定功能保留作为测试语言生成目标的共同设计原则。
cs.CL / 6 / 2610.06956
EMODE: Dynamic Para-Semantic Experts for Emotion-Aware Speech Language Modeling
EMODE：面向情感感知的动态副语义专家语音语言建模
Abstract
Large speech language models have demonstrated strong capabilities in unified cross-modal understanding and generation, yet paralinguistic cues, especially emotion, remain difficult to preserve. Existing systems typically rely on entangled acoustic representations, which allow the underlying language model to depend excessively on recovered lexical content instead of grounding its behavior in acoustic-prosodic evidence. We address this limitation with EMODE, an emotion-aware speech language model built around \textbf{Dynamic Para-Semantic Experts (DPSE)}. DPSE decomposes continuous speech features into semantic and paralinguistic pathways, routes them dynamically, and fuses them before integration into the language model. To turn this structural decomposition into functional specialization, EMODE is trained with a three-stage curriculum consisting of semantic warm-up, paralinguistic activation, and joint refinement, guided by Orthogonal Expert Guidance (OEG), Semantic-to-Acoustic Alignment (SAA), and Gating Diversity Regularization (GDR). Experiments on SER test, empathetic response evaluation, and the newly constructed bilingual MEPA benchmark show that EMODE improves the balance between lexical fidelity and emotional sensitivity, strengthens affect-grounded response generation, and exposes the value of explicit para-semantic factorization for robust cross-corpus emotion understanding.
Chinese Translation
大型语音语言模型在统一的跨模态理解与生成方面表现出强大能力，然而副语言线索，尤其是情感信息，仍难以有效保留。现有系统通常依赖于纠缠的声学表示，使得底层语言模型过度依赖恢复的词汇内容，而非基于声学韵律证据来指导其行为。我们提出了EMODE，一种基于动态副语义专家（Dynamic Para-Semantic Experts, DPSE）的情感感知语音语言模型。DPSE将连续语音特征分解为语义路径和副语言路径，动态路由并融合后再整合进语言模型。为了将这种结构分解转化为功能专门化，EMODE采用三阶段课程训练，包括语义预热、副语言激活和联合精炼，训练过程由正交专家指导（Orthogonal Expert Guidance, OEG）、语义到声学对齐（Semantic-to-Acoustic Alignment, SAA）和门控多样性正则化（Gating Diversity Regularization, GDR）引导。SER测试、共情响应评估及新构建的双语MEPA基准实验表明，EMODE在词汇保真度与情感敏感性之间实现更好平衡，增强了基于情感的响应生成能力，揭示了显式副语义因子分解在稳健跨语料情感理解中的价值。
cs.CL / 7 / 2610.06962
Verdicts Without Annotated Evidence: Rejection Sampling or Label-Only Post-Training for Evidence Recovery?
没有标注证据的裁决：拒绝采样还是仅标签后训练以恢复证据？
Abstract
In many review workflows the verdict is the only thing retained. The passages behind it are not marked, because that annotation costs far more than recording the decision. We measure how much of that evidence a small language model can recover when it is post-trained on the verdicts alone, with no human evidence labels at any stage. On ContractNLI the human evidence spans are held out until evaluation. Matching the recorded verdict and agreeing with those spans are not the same thing: across six systems the two scores are only weakly related and rank the systems differently, so accuracy is a poor guide when the citations have to be reviewable. Label-only training on the bare verdict reaches accuracy 0.896 and span F1 0.564. Rejection sampling, which keeps a generated trace only when its verdict matches the record and then picks one by an automatic source-grounding score, reaches 0.797 and 0.556, against 0.747 and 0.493 before training. Verbatim citation rises from 0.597 to 0.729 under label-only training and to 0.701 under rejection sampling. One seed on one corpus cannot say which method is better, but both improve the evidence without anyone annotating it.
Chinese Translation
在许多审查工作流程中，裁决是唯一被保留的内容。其背后的段落并未被标记，因为这种标注的成本远高于记录决策的成本。我们测量了一个小型语言模型在仅基于裁决进行后训练时能够恢复多少证据，而在任何阶段都没有人类证据标签。在 ContractNLI 数据集中，人类证据跨度在评估之前被保留。记录的裁决与这些跨度一致并不是同一回事：在六个系统中，这两个评分仅弱相关，并且对系统的排名不同，因此当引用必须可审查时，准确性是一个不可靠的指标。仅基于标签的训练在裸裁决上达到了 0.896 的准确率和 0.564 的跨度 F1 值。拒绝采样在裁决与记录匹配时仅保留生成的轨迹，然后通过自动源基础评分选择一个，达到了 0.797 和 0.556，而在训练之前分别为 0.747 和 0.493。在仅标签训练下，逐字引用从 0.597 上升到 0.729，而在拒绝采样下上升到 0.701。一个种子在一个语料库上无法判断哪种方法更好，但两者都在没有任何人进行标注的情况下改善了证据。
cs.CL / 8 / 2610.06963
WavePrune: One period is often enough for RoPE
WavePrune：一个周期通常足够用于 RoPE
Abstract
Rotary Position Embedding (RoPE) encodes token positions by rotating each two-dimensional channel of the query and key vectors at a channel-specific frequency, making the attention logits invariant to a common shift of positions. However, this rotation is periodic, and it leads to position aliasing where relative positions separated by a full rotation period become hard to tell apart. To address this, we propose WavePrune, which restricts each channel to its first rotation period. We show that it removes the distractions in attention maps created by position aliasing and improves overall long-context performance. Specifically, WavePrune raises the HELMET score on four of five models we test without any extra tuning (e.g., 35.7 -> 40.0 on Qwen3-8B). When pretraining models from scratch, WavePrune also achieves lower validation loss at extrapolated lengths than pretraining without it. Because WavePrune restricts each channel to a sliding window, it induces a fine-grained sparsity that our hardware-aligned CUDA kernels exploit for 1.15x prefill and 1.24x decoding speedups over FlashAttention-2 at 32K context. Together, these results show that RoPE's periodic structure, widely regarded as essential, is largely redundant beyond the first rotation period.
Chinese Translation
旋转位置嵌入（Rotary Position Embedding, RoPE）通过以通道特定的频率旋转查询和键向量的每个二维通道来编码令牌位置，使得注意力 logits 对位置的共同偏移保持不变。然而，这种旋转是周期性的，导致位置混叠，使得相隔一个完整旋转周期的相对位置难以区分。为了解决这个问题，我们提出了 WavePrune，它将每个通道限制在其第一个旋转周期内。我们展示了它消除了由位置混叠产生的注意力图中的干扰，并改善了整体的长上下文性能。具体而言，WavePrune 在我们测试的五个模型中的四个上提高了 HELMET 分数，而无需任何额外的调优（例如，在 Qwen3-8B 上从 35.7 提升至 40.0）。在从头开始预训练模型时，WavePrune 还在外推长度下实现了比未使用它的预训练更低的验证损失。由于 WavePrune 将每个通道限制在滑动窗口内，它引入了一种细粒度的稀疏性，我们的硬件对齐的 CUDA 内核利用这一点在 32K 上下文中实现了比 FlashAttention-2 快 1.15 倍的预填充和 1.24 倍的解码速度提升。综合来看，这些结果表明，RoPE 的周期性结构被广泛认为是必不可少的，但在第一个旋转周期之后大部分是多余的。
cs.CL / 9 / 2610.07019
Calibrated Answers About Randomized Trials From a 4-Billion-Parameter Open Model: A Registered Test and a License-Clean Release
来自一个拥有40亿参数的开放模型的随机试验校准答案：注册测试与许可证清晰发布
Abstract
Fiorillo v0.5 is an open model that answers typed questions with a probability for each answer. Its main specialist reads a randomized trial's article, cut to 6,144 tokens, and answers whether an intervention significantly increased, significantly decreased or did not significantly change an outcome against a comparator (Evidence Inference 2.0, EI). It is Qwen3-4B-Base with low-rank adapters and a decision head, fine-tuned for EI only on the 1,431 of 2,657 training articles whose own license allows reuse. Four criteria registered on the Open Science Framework before this version's test predictions decided its release, the second bar judged on EI's test split, whose labels are public. On that split (1,218 prompts in 333 articles), the expected calibration error was 0.0168 against a limit of 0.05; log loss was below the prior's by 0.8603 (95 percent interval 0.8104 to 0.9078) and below that of Gemma 4 31B-it, reading the same input, by 0.1829 (0.1164 to 0.2598); and macro-F1 was 0.9248 against 0.8668, so all four criteria passed. Training the same recipe on clean articles alone cost 0.0123 in accuracy (0.0034 to 0.0207; descriptive). With no article, macro-F1 fell to 0.4384; the title alone raised it by 0.0939 (0.0655 to 0.1234), which a title stating the result or recall of the trial could explain; exchanging intervention and comparator reversed 0.6652 of its direction answers. Run as released, the files matched the evaluated predictions within limits set in advance. The release is under the Apache License 2.0 (digital object identifier 10.57967/hf/10722).
Chinese Translation
Fiorillo v0.5 是一个开放模型，能够对输入的问题提供每个答案的概率。其主要专家阅读一篇随机试验的文章，限制为6144个标记，并回答某个干预是否显著增加、显著减少或对比者没有显著变化某个结果（证据推断 2.0，EI）。该模型为 Qwen3-4B-Base，配备低秩适配器和决策头，仅在允许重用的2657篇训练文章中的1431篇上针对 EI 进行了微调。发布该版本之前，在开放科学框架上注册的四个标准决定了其发布，第二个标准基于 EI 的测试分割，其标签是公开的。在该分割（333篇文章中的1218个提示）中，预期校准误差为0.0168，低于0.05的限制；对数损失比先前模型低0.8603（95%置信区间为0.8104至0.9078），并且比 Gemma 4 31B-it 读取相同输入的结果低0.1829（0.1164至0.2598）；宏F1为0.9248，超过0.8668，因此所有四个标准均通过。在仅使用干净文章训练相同配方时，准确度损失为0.0123（0.0034至0.0207；描述性）。没有文章时，宏F1降至0.4384；仅标题提高了0.0939（0.0655至0.1234），这可以通过标题陈述试验结果或回忆来解释；交换干预和对比者则反转了0.6652的方向答案。按发布方式运行时，文件在预先设定的限制内与评估预测相符。该发布遵循 Apache License 2.0（数字对象标识符 10.57967/hf/10722）。
cs.CL / 10 / 2610.07032
Investigating Model Compression for Neural Machine Translation in the Biomedical Domain
探讨生物医学领域神经机器翻译的模型压缩
Abstract
Large-scale pretrained transformer models have achieved state-of-the-art performance across diverse machine translation tasks, including multilingual settings. Knowledge distillation has emerged as a sustainable approach for model compression, transferring knowledge from large teacher models to smaller, more efficient student models. Similarly, quantization, which reduces the numerical precision of model weights and activations (e.g., from 32-bit to 8-bit representations) is widely used to accelerate inference, enabling models to run several times faster during deployment. However, both techniques face limitations when applied to specialized domain data, particularly under low-resource conditions. In knowledge distillation, the effectiveness of transfer is often constrained by the scarcity of domain-specific parallel data, while quantization can lead to performance degradation as bit precision decreases. In this work, we investigate the combined application of knowledge distillation and quantization for French-to-English biomedical translation, a domain characterized by specialized terminology and limited parallel resources. We develop and compare multiple fine-tuning strategies to adapt compressed student models to this challenging setting. Our experiments demonstrate that a collaboratively distilled and quantized student model achieves a 69% reduction in size, a 98.21% increase in inference speed, and a 98.46% reduction in CO2 emissions compared to the original baseline all without sacrificing translation quality. These results indicate that jointly optimized compression techniques can yield efficient, high-performance models suitable for translation service providers operating under resource constraints.
Chinese Translation
大规模预训练的变换器模型在多种机器翻译任务中，包括多语言设置，已实现了最先进的性能。知识蒸馏作为一种可持续的模型压缩方法，已逐渐受到关注，它将知识从大型教师模型转移到更小、更高效的学生模型中。同样，量化技术通过降低模型权重和激活的数值精度（例如，从32位表示降低到8位表示）被广泛应用于加速推理，使模型在部署时运行速度提高数倍。然而，当这些技术应用于专业领域数据时，尤其是在资源匮乏的情况下，面临着一些限制。在知识蒸馏中，转移的有效性常常受到领域特定平行数据稀缺的限制，而量化则可能导致性能下降，随着位精度的降低。在本研究中，我们探讨了知识蒸馏和量化在法语到英语生物医学翻译中的联合应用，该领域以专业术语和有限的平行资源为特征。我们开发并比较了多种微调策略，以适应这一具有挑战性的环境中的压缩学生模型。我们的实验表明，相较于原始基线，协同蒸馏和量化的学生模型在模型大小上减少了69%，推理速度提高了98.21%，二氧化碳排放减少了98.46%，且未牺牲翻译质量。这些结果表明，联合优化的压缩技术可以产生高效、高性能的模型，适合在资源受限的情况下为翻译服务提供商使用。
cs.CL / 11 / 2610.07070
Turnslide: Scalable Multi-Turn Data Synthesis by Walking a Finite-State Machine
Turnslide：通过行走有限状态机实现可扩展的多轮数据合成
Abstract
Small language models are inexpensive to serve and can run on private infrastructure, but base models are often not good enough at multi-turn tool calling, and fine-tuning them needs per-API data that rarely exists. Existing synthesis methods are too expensive for high-scale fine-tuning, as they often require mock operational environments for different domains and multiple LLM calls per generated conversation turn. We introduce a fully automated, lightweight synthesis framework that models each API as a finite-state machine, representing the system as abstract states that determine when each tool may be called, producing state-valid sequences of tools; sequences are translated into complete examples with a single LLM call. Rather than optimize diversity, we set a target distribution over the number of turns, the tool sequence and task complexity. We measure data quality by fine-tuning SLMs on generated trajectories, showing that our FSM-based generation significantly improves downstream accuracy over an unmutated baseline and, against existing works, reaches 70.7% full accuracy over 63.4% and 53.7% with 3.6-6.6$\times$ fewer tokens.
Chinese Translation
小型语言模型的服务成本低，且可以在私有基础设施上运行，但基础模型在多轮工具调用方面往往表现不佳，而对其进行微调需要的每个API数据通常很少存在。现有的合成方法对于大规模微调来说成本过高，因为它们通常需要为不同领域提供模拟操作环境，并且每生成一个对话轮次都需要多次调用大型语言模型（LLM）。我们提出了一种完全自动化、轻量级的合成框架，将每个API建模为有限状态机（FSM），将系统表示为抽象状态，以确定何时可以调用每个工具，从而生成状态有效的工具序列；这些序列通过一次LLM调用转换为完整示例。我们并不优化多样性，而是设定了轮次、工具序列和任务复杂度的目标分布。我们通过在生成的轨迹上微调小型语言模型（SLMs）来衡量数据质量，结果表明我们的基于FSM的生成显著提高了下游准确性，相较于未变异的基线，达到了70.7%的完整准确率，而现有工作仅为63.4%和53.7%，且使用的令牌数量减少了3.6-6.6倍。
cs.CL / 12 / 2610.07109
JudgeMoE: Distributional Aggregation for LLM-as-a-Judge
JudgeMoE：用于 LLM 作为评审的分布聚合
Abstract
When an LLM judge scores an output, its score distribution retains uncertainty and disagreement information that is lost after scalar compression. We introduce JudgeMoE, a lightweight aggregator that assigns example-specific weights to cached judge score distributions and fuses them before computing a final score. A protocol study shows that score-range choice is unstable across judge--dataset settings and that soft scoring usually outperforms hard decoding. On the original 10-cell benchmark, JudgeMoE improves mean Spearman over uniform log pooling by $+0.079$. Applying the same configuration to six additional cells yields a $+0.0393$ mean gain over the strongest local single judge across 16 cells, with positive differences in 12/16 cells and a one-sided Wilcoxon signed-rank $p=0.0091$. Validation-based analyses further show that the preferred aggregation method depends on the task and judge pool.
Chinese Translation
当 LLM 评审对输出进行评分时，其评分分布保留了在标量压缩后丢失的不确定性和分歧信息。我们提出了 JudgeMoE，一种轻量级聚合器，它为缓存的评审评分分布分配特定于示例的权重，并在计算最终评分之前将其融合。协议研究表明，评分范围的选择在评审-数据集设置之间是不稳定的，且软评分通常优于硬解码。在原始的 10-cell 基准测试中，JudgeMoE 在均值斯皮尔曼（mean Spearman）上相较于均匀对数池（uniform log pooling）提高了 $+0.079$。将相同配置应用于六个额外的单元，在 16 个单元中相较于最强的本地单一评审获得了 $+0.0393$ 的均值增益，其中 12/16 个单元表现出正差异，单侧威尔科克森符号秩检验的 $p=0.0091$。基于验证的分析进一步表明，首选的聚合方法依赖于任务和评审池。
cs.CL / 13 / 2610.07132
CroissantMiner: Automated Extraction and Validation of Croissant Metadata for ML Datasets
CroissantMiner：机器学习数据集的Croissant元数据自动提取与验证
Abstract
Croissant has emerged as a standard for machine-readable dataset metadata, yet populating its fields remains labor-intensive and requires careful reading of accompanying dataset documentation. We present the first benchmark enabling end-to-end evaluation of metadata extraction aligned with a community-standard schema. The benchmark comprises 602 papers, including 102 with human-validated gold annotations and 500 with LLM-generated silver annotations, covering the full Croissant schema with both core and Responsible AI (RAI) fields. Using this benchmark, we evaluate a range of extraction systems spanning frontier models, open-weight models, and agentic architectures, under a two-tier evaluation framework that combines rule-based scoring with an LLM judge selected via human audit. We find that single-pass extraction consistently outperforms the four agentic architectures we evaluate: across backbones, these decomposed variants achieve lower accuracy than a single full-context pass. The largest gap appears on long-form RAI fields, which require synthesizing and interpreting information scattered across a paper rather than copying it from a single location, a setting where current systems remain far from reliable. We release the benchmark, evaluation code, judge audit, a live demo, and a leaderboard open to new systems.
Chinese Translation
Croissant已成为机器可读数据集元数据的标准，但填充其字段仍然劳动密集，并且需要仔细阅读伴随的数据集文档。我们提出了第一个基准，能够实现与社区标准架构对齐的元数据提取的端到端评估。该基准包含602篇论文，其中102篇具有人工验证的黄金注释，500篇具有大型语言模型（LLM）生成的银色注释，涵盖了完整的Croissant架构，包括核心字段和负责任的人工智能（RAI）字段。利用该基准，我们评估了一系列提取系统，涵盖前沿模型、开放权重模型和自主架构，在一个结合基于规则的评分与通过人工审核选择的LLM评审的双层评估框架下。我们发现，单次提取始终优于我们评估的四种自主架构：在不同基础模型之间，这些分解变体的准确性低于单次全上下文提取。在长格式RAI字段上，差距最大，因为这些字段需要综合和解释散布在论文中的信息，而不是从单一位置复制信息，这是当前系统仍然远未可靠的环境。我们发布了基准、评估代码、评审审核、实时演示和开放给新系统的排行榜。
cs.CL / 14 / 2610.07186
Identifying Introspection From the Inside
从内部识别内省
Abstract
Large language models make claims about themselves that are both consequential and increasingly difficult to verify from behavior alone. How can we distinguish plausible confabulations from genuine introspection? In this paper, we identify mechanistic signatures of faithful self-report in a controlled setting. Using low-rank adapters, we train models to make decisions on behalf of fictitious characters, according to latent linear preference functions. We find sustained fine-tuning on an implicit decision task can lead to the emergence of accurate self-reporting of models' learned preferences, even without explicit self-report supervision. We ask two research questions about this emergent phenomenon. First: is the emergence of accurate self-reporting accompanied by a measurable structural change in the model? Weight ablations and frozen-layer experiments together indicate that preference representations shift to earlier layers over training, consistent with the hypothesis that faithful self-report requires preferences to be located where pre-existing verbalization mechanisms can access them. Second: can these structural differences distinguish faithful models from unfaithful ones? Using attribution patching, we find that faithful models exhibit significantly higher attribution similarity between the decision-making and self-report tasks -- a mechanistic signature of faithful self-report that does not require us to understand the content of the report itself. Previous work on self-report has observed behaviorally that models can be faithful or unfaithful; our work proposes that, at least in our restricted setting, it is possible to distinguish between the two patterns of computation by examining the structure of the networks themselves.
Chinese Translation
大型语言模型对自身的陈述既具有重要意义，又越来越难以仅通过行为进行验证。我们如何区分可信的虚构与真实的内省？在本文中，我们在受控环境中识别出忠实自我报告的机械特征。通过使用低秩适配器，我们训练模型代表虚构角色根据潜在的线性偏好函数做出决策。我们发现，在隐性决策任务上持续微调可以导致模型对其学习偏好的准确自我报告的出现，即使没有明确的自我报告监督。我们提出了两个关于这一新兴现象的研究问题。首先：准确自我报告的出现是否伴随着模型中可测量的结构变化？权重消融和冻结层实验共同表明，偏好表示在训练过程中向早期层移动，这与忠实自我报告需要偏好位于预先存在的表述机制可访问的位置的假设一致。第二：这些结构差异能否区分忠实模型与不忠实模型？通过归因修补，我们发现忠实模型在决策任务和自我报告任务之间表现出显著更高的归因相似性——这是忠实自我报告的机械特征，无需我们理解报告内容本身。之前关于自我报告的研究在行为上观察到模型可以是忠实或不忠实的；我们的工作提出，在至少我们限制的环境中，通过检查网络本身的结构，可以区分这两种计算模式。
cs.CL / 15 / 2610.07224
TIDE 2.0: an open, model-agnostic engine for keyed de-identification of clinical notes
TIDE 2.0：一个开放的、模型无关的临床笔记键控去标识化引擎
Abstract
Clinical notes capture most of what is documented about a patient's care, but they cannot be used for research until protected health information (PHI) is removed. De-identification is often treated as a detection problem. Detection alone is not sufficient: redaction strips clinical content along with identifiers, date blanking destroys the temporal intervals needed for longitudinal analysis, and assigning a fresh random surrogate at each occurrence breaks links between a patient's notes. We present TIDE 2.0, an MIT-licensed engine with two separable stages: an interchangeable recognizer and a keyed anonymizer. Both run on hardware the institution owns. Surrogates are generated cryptographically with no stored linkage table. Dates shift by a per-patient, interval-preserving offset; each value receives the same surrogate across all occurrences under a given key; and a release produced under a new key cannot be linked to earlier releases. We also release TIDE2-Sentry, a recognizer distilled from a large language model. On two gold-annotated corpora from two institutions, the default configuration reached span-level recall of 0.88 in-domain and 0.77 on the second institution's corpus, at precision 0.88 and 0.87. We report recall and precision per category alongside these aggregates. The engine is open source, and the recognizer is available under a gated research-use agreement, so institutions can run, inspect and extend both within their own environments.
Chinese Translation
临床笔记记录了患者护理的绝大部分信息，但在去除受保护健康信息（PHI）之前，无法用于研究。去标识化通常被视为一个检测问题。然而，仅仅进行检测是不够的：删除处理会同时去除临床内容和标识符，日期空白化会破坏进行纵向分析所需的时间间隔，而在每次出现时分配一个新的随机替代品则会打断患者笔记之间的联系。我们提出了TIDE 2.0，这是一个具有两个可分离阶段的引擎，采用MIT许可证：一个可互换的识别器和一个键控匿名化器。两者均在机构拥有的硬件上运行。替代品通过加密方式生成，没有存储的链接表。日期通过每位患者的、保持时间间隔的偏移量进行调整；在给定键下，每个值在所有出现中接收相同的替代品；而在新键下生成的发布版本无法与早期的发布版本关联。我们还发布了TIDE2-Sentry，这是从大型语言模型中提炼出的识别器。在来自两个机构的两个金标注语料库上，默认配置在第一个机构的语料库中达到了0.88的跨度级召回率，在第二个机构的语料库中达到了0.77，精确率分别为0.88和0.87。我们报告了每个类别的召回率和精确率，以及这些汇总数据。该引擎是开源的，识别器在受限研究使用协议下可用，因此机构可以在自己的环境中运行、检查和扩展这两者。
cs.CL / 16 / 2610.07306
Kurate: Scalable Scientific Quality Analysis
Kurate：可扩展的科学质量分析
Abstract
Scientific search systems can find papers that are relevant to a question, but they generally do not assess the quality of the evidence that those papers provide. We present Kurate, a system that uses large language models (LLMs) to assess the quality of published studies. Kurate uses both the paper and its related documents (e.g., the study's trial registration and protocol), and links each of its judgments to the passage of text on which that judgment is based. We applied Kurate to a corpus of 4,347 papers (3,913 of which report randomized trials) and scored each paper on 8 dimensions of study design and reporting: specifically, statistical power, causal identification, preregistration, selective reporting, measurement validity, analysis prespecification, reporting transparency, and conflict of interest and funding. Across the corpus, we found that papers most often exhibited issues with statistical power, selective reporting, and analysis prespecification, although average quality differed between clinical areas. When compared against expert annotations of 60 held-out clinical-trial documents, the information Kurate extracted matched the expert label in 221/242 protocol scorepoints and 294/370 results-publication scorepoints, with AC1 0.94 and 0.81, respectively. Using a well-reputed, high quality clinical trial as a worked example, we show how a single paper's overall grade breaks down into separate judgments, with each linked to specific evidence from the trial's registration, protocol, and published report. Together, these results show that large-scale quality assessment of this kind is feasible, and that it can be used to address meta-scientific research questions.
Chinese Translation
科学搜索系统能够找到与问题相关的论文，但通常不评估这些论文提供的证据质量。我们提出了Kurate，一个利用大型语言模型（LLMs）来评估已发布研究质量的系统。Kurate同时使用论文及其相关文档（例如，研究的试验注册和协议），并将每个判断与其依据的文本段落相链接。我们将Kurate应用于一个包含4,347篇论文的语料库（其中3,913篇报告了随机试验），并在8个研究设计和报告维度上对每篇论文进行了评分：具体包括统计效能、因果识别、预注册、选择性报告、测量有效性、分析预设、报告透明度，以及利益冲突和资金来源。在整个语料库中，我们发现论文最常出现统计效能、选择性报告和分析预设方面的问题，尽管不同临床领域的平均质量有所不同。与60份保留的临床试验文档的专家注释相比，Kurate提取的信息在242个协议评分点中与专家标签匹配221个，在370个结果发布评分点中匹配294个，AC1分别为0.94和0.81。通过使用一个声誉良好的高质量临床试验作为实例，我们展示了单篇论文的整体评分如何分解为多个独立判断，每个判断都与试验注册、协议和已发布报告中的具体证据相链接。这些结果共同表明，这种大规模质量评估是可行的，并且可以用于解决元科学研究问题。
cs.CL / 17 / 2610.07365
Who Wrote It Is Not Enough: Detecting Who Contributed the Insight
仅仅知道作者是谁是不够的：检测谁贡献了洞见
Abstract
As LLMs increasingly assist scientific writing and peer review, detecting who wrote the text is no longer sufficient: we need to determine who contributed the underlying insight. We introduce Insight Provenance, the task of identifying whether a review insight originates from a human, an LLM, or their hybrid contribution. We construct InsightProv-v0 from 4,057 scientific papers and 12,660 human reviews, simulating different levels of LLM involvement with GPT-4o, Gemini, and DeepSeek and annotating provenance at the sentence level. We show that strong performance on raw data can be misleading, as models exploit linguistic and textual-authorship shortcuts that degrade substantially under progressively debiased evaluation. We therefore propose a two-stage adversarial framework that suppresses shortcut signals while preserving provenance-relevant information. Beyond detection, extensive analyses reveal what makes intellectual authorship identifiable: paper grounding and neighboring review context provide complementary provenance signals, while human, hybrid, and AI insights systematically differ in their information sources and failure modes. Most strikingly, AI insights predominantly remain close to generic or paper-provided information, whereas human insights more often introduce external knowledge and independent judgment. These findings suggest that while wording can be rewritten by an LLM, the provenance of an idea leaves a deeper and more persistent signal.
Chinese Translation
随着大型语言模型（LLMs）在科学写作和同行评审中越来越多地发挥作用，仅仅检测文本的作者已不再足够：我们需要确定谁贡献了基础洞见。我们引入了洞见来源（Insight Provenance）这一任务，旨在识别一个评审洞见是源于人类、LLM，还是它们的混合贡献。我们从4,057篇科学论文和12,660条人类评审中构建了InsightProv-v0，模拟了不同程度的LLM参与，使用GPT-4o、Gemini和DeepSeek，并在句子级别上进行了来源标注。我们展示了在原始数据上表现强劲可能具有误导性，因为模型利用语言和文本作者身份的捷径，而这些捷径在逐渐去偏见的评估中显著降级。因此，我们提出了一种两阶段对抗框架，该框架在保留与来源相关的信息的同时抑制捷径信号。除了检测之外，广泛的分析揭示了什么使得知识产权可识别：论文基础和相邻评审上下文提供了互补的来源信号，而人类、混合和AI洞见在信息来源和失败模式上系统性地不同。最引人注目的是，AI洞见主要保持在通用或论文提供的信息附近，而人类洞见则更常引入外部知识和独立判断。这些发现表明，尽管LLM可以重写措辞，但一个想法的来源留下了更深刻和持久的信号。
cs.CL / 18 / 2610.07426
AccentCL: Robust Accent Classification with Incremental Expansion
AccentCL：具有增量扩展的鲁棒口音分类
Abstract
Accent classifiers are typically trained with a fixed label inventory and cannot accommodate new accent categories as new data becomes available. Moreover, accented speech corpora often exhibit substantial class imbalance and/or domain shift due to differences in recording conditions across corpora. We present AccentCL, a class-incremental learning framework for English accent classification that is robust to class imbalance and cross-corpus domain shift. AccentCL extracts multi-layer representations from a frozen Whisper-Large-v3 encoder, optimized with an imbalance-aware cross-entropy loss to reduce bias toward the majority accent classes and a domain mean alignment loss that minimizes distributional mean shift across training corpora. The label space is then expanded via replay-based continual learning, using the frozen base model for knowledge retention and an old-to-new margin loss to reduce overprediction on newly added classes. On a five-class accent classification task, AccentCL achieves 77.1% balanced accuracy and a 76.9% macro-averaged F1 score. We further evaluate the model's ability to incrementally incorporate two new accent categories: Spanish-accented and Chinese-accented English. When adding Spanish-accented English to the pretrained model, AccentCL attains an F1 of 83.3% on the new class while retaining 77.3% balanced accuracy on the base classes. When subsequently adding Chinese-accented English, it achieves 61.8% F1 on the new class while preserving 77.6% balanced accuracy on the previously learned classes. These results show that AccentCL enables robust regional accent classification while allowing new accent categories to be added without full retraining.
Chinese Translation
口音分类器通常使用固定的标签库进行训练，无法在新数据可用时适应新的口音类别。此外，由于不同语料库的录音条件差异，带口音的语音语料库往往表现出显著的类别不平衡和/或领域转移。我们提出了AccentCL，这是一个针对英语口音分类的类别增量学习框架，能够有效应对类别不平衡和跨语料库领域转移。AccentCL从一个冻结的Whisper-Large-v3编码器中提取多层表示，并通过一种考虑不平衡的交叉熵损失进行优化，以减少对主要口音类别的偏见，同时采用领域均值对齐损失，最小化训练语料库间的分布均值偏移。然后，通过基于重放的持续学习扩展标签空间，使用冻结的基础模型进行知识保留，并采用旧到新边际损失以减少对新添加类别的过度预测。在一个五类口音分类任务中，AccentCL实现了77.1%的平衡准确率和76.9%的宏平均F1分数。我们进一步评估模型逐步纳入两个新口音类别的能力：西班牙口音和中文口音英语。当将西班牙口音英语添加到预训练模型时，AccentCL在新类别上达到了83.3%的F1分数，同时在基础类别上保持了77.3%的平衡准确率。当随后添加中文口音英语时，它在新类别上达到了61.8%的F1分数，同时在先前学习的类别上保持了77.6%的平衡准确率。这些结果表明，AccentCL能够实现鲁棒的区域口音分类，同时允许在不进行全面重训练的情况下添加新的口音类别。
cs.CL / 19 / 2610.07502
Closing Ambient Clinical Documentation Gaps with Automated Provider Queries
通过自动化提供者查询填补临床文档中的空白
Abstract
Provider queries are clarifying requests sent by clinical documentation specialists to physicians to close gaps in the clinical note and ensure accurate billing. Prior work automates note drafting, ICD-10 coding, and order extraction assuming a complete transcript, leaving these gaps unaddressed. We study whether an LLM can automate the query loop, termed DAU (Draft, Ask, Update), across those three tasks. An audit of 3,000 real visits identifies the sources of missing documentation, from which we build five transcript-degradation benchmarks on public data. Analyzing 21k clarification turns on real conversations, we find useful-question predictors are task-specific: oracle confidence dominates, but note completeness needs only simple recall questions while ICD-10 coding needs harder, multi-option ones. About 9% of turns hurt performance, driven by redundant questions and non-answers that still trigger a rewrite. Deployment depends on learning "when not" as much as "what to" ask.
Chinese Translation
提供者查询是临床文档专家向医生发送的澄清请求，旨在填补临床记录中的空白并确保准确的账单。之前的研究自动化了笔记草拟、ICD-10编码和订单提取，假设有完整的记录，从而未能解决这些空白。我们研究了大型语言模型（LLM）是否可以在这三项任务中自动化查询循环，称为DAU（草拟、询问、更新）。对3000个真实就诊记录的审计识别了缺失文档的来源，从中我们构建了五个基于公共数据的记录降解基准。通过分析21000个真实对话中的澄清回合，我们发现有用问题的预测因任务而异：oracle信心占主导地位，但笔记完整性只需要简单的回忆问题，而ICD-10编码则需要更复杂的多选问题。约9%的回合会影响性能，主要是由于冗余问题和非回答，这些仍然会触发重写。部署的成功依赖于学习“何时不”提问与“该问什么”同样重要。
cs.CL / 20 / 2610.07519
Not What a Child Expressed: Auditing the Sign-to-Text Safety Interface in Child-Facing AI
儿童表达的不是：审计面向儿童的人工智能中的手语到文本安全接口
Abstract
Automatic sign language translation (SLT) has entered consumer products, turning American Sign Language into English text for dictation, messaging, and queries put to a conversational assistant. Child-facing AI and platform trust-and-safety tooling decide on text, using filters on minor accounts and grooming classifiers that score chat messages. A signing child who uses SLT therefore reaches these safeguards through a translation. We found no publicly documented system in which the two have been jointly evaluated, and the leading deployed SLT model was neither trained nor formally evaluated on signers under 18. Errors that alter negation, participant roles, secrecy, urgency or help-seeking could change a safety decision without disturbing fluency. This paper proposes a Deaf-informed pre-deployment audit of that boundary, with a failure taxonomy, a sanitised scenario schema, four comparison conditions, and four outcome measures. Auslan is the planned first case study.
Chinese Translation
自动手语翻译（SLT）已进入消费产品，将美国手语翻译为英语文本，用于听写、消息传递和与对话助手的查询。面向儿童的人工智能和平台的信任与安全工具通过对未成年人账户的过滤和评分聊天消息的分类器来决定文本。因此，使用SLT的手语儿童通过翻译接触到这些安全保障。我们发现没有公开记录的系统对这两者进行了联合评估，而当前部署的主要SLT模型既未在18岁以下的手语者上进行训练，也未进行正式评估。改变否定、参与者角色、秘密性、紧迫性或寻求帮助的错误可能会在不影响流畅性的情况下改变安全决策。本文提出了一种以聋人视角为基础的预部署审计方案，包括失败分类法、清理后的场景框架、四种比较条件和四种结果测量。Auslan（澳大利亚手语）是计划中的第一个案例研究。
cs.CL / 21 / 2610.07545
Quality-Aware Self-Correcting Speech Translation on an Edge Device
边缘设备上的质量感知自我纠正语音翻译
Abstract
We present a fully offline speech-to-speech translation pipeline that runs on a Jetson Nano (4 GB) and corrects its own weak translations without retraining. A Whisper-tiny ASR feeds an Opus-MT translator; multilingual BERT cosine similarity acts as a Quality Estimation (QE) gate, triggering a secondary-pass correction when confidence falls below a pre-defined threshold $\tau$. We compare three correction methods: QE reranking (M1), Minimum Bayes-Risk decoding (M2), and constrained beam search (M3). On 1,012 FLORES-200 sentences (English-Spanish), M2 at $\tau=0.90$ produces statistically significant improvements over greedy decoding on BLEU (+0.67, p<0.001), ChrF (+0.51, p<0.001), and COMET (+0.0020 at N=3, p=0.002); M1 yields no significant gains, and M3 is significantly worse than baseline (p>0.99). Our central finding is that QE functions effectively as a gate but poorly as a ranker: removing the QE model from candidate selection (M1$\to$M2) does not hurt quality and frees 680 MB from the critical path. Using a gain-to-edit ratio adapted from the post-editing-effort literature, we further show that smaller candidate pools (N=3) yield more surgical corrections with better semantic adequacy, while larger pools (N=10) maximise lexical reward. We release the system and demonstrate live translation across six language pairs.
Chinese Translation
我们提出了一种完全离线的语音到语音翻译管道，该管道在 Jetson Nano（4 GB）上运行，并能够在不重新训练的情况下纠正自身的弱翻译。Whisper-tiny ASR 为 Opus-MT 翻译器提供输入；多语言 BERT 余弦相似度作为质量估计（Quality Estimation, QE）门控，当置信度低于预定义阈值 $ au$ 时触发二次修正。我们比较了三种修正方法：QE 重新排序（M1）、最小贝叶斯风险解码（Minimum Bayes-Risk decoding, M2）和约束束搜索（constrained beam search, M3）。在 1,012 个 FLORES-200 句子（英语-西班牙语）上，M2 在 $ au=0.90$ 时在 BLEU (+0.67, p<0.001)、ChrF (+0.51, p<0.001) 和 COMET (+0.0020 at N=3, p=0.002) 上相较于贪婪解码产生了统计显著的改善；M1 没有显著增益，而 M3 的表现显著低于基线（p>0.99）。我们的核心发现是 QE 在作为门控时效果良好，但作为排序器时效果较差：从候选选择中移除 QE 模型（M1$ o$M2）不会损害质量，并释放了 680 MB 的关键路径。我们使用从后编辑工作量文献中调整的增益-编辑比，进一步表明较小的候选池（N=3）能够产生更具针对性的修正，且语义适当性更好，而较大的候选池（N=10）则最大化词汇奖励。我们发布了该系统，并展示了六种语言对之间的实时翻译。
cs.CL / 22 / 2610.07563
HouseholdBench: Evaluating Large Language Models as Predictors of Household Economic Behavior
家庭基准：评估大型语言模型作为家庭经济行为预测者
Abstract
Large language models (LLMs) have the potential to meet a key goal in economics: a quantitative model of household decision making, across a variety of settings. Yet existing evaluations cover few surveys and outcomes, and do not study how households adjust to changing economic conditions. We introduce a new evaluation, HouseholdBench, which unites 6 U.S. household surveys and 32 prediction tasks spanning numeric, categorical and probabilistic outcomes, related to consumption, income, labor, expectations, and housing. Using past behavior, demographics and macroeconomic conditions, the tasks test whether LLMs predict behavior, including how households adjust to changes in various policies. We evaluate 13 proprietary and open-weight LLMs against a no-change baseline and a gradient-boosted tree model. Most LLMs outperform the no-change baseline, including for policy response tasks -- with the best model lowering error for numeric outcomes by 12.2%. Across most tasks, gradient-boosted trees rank first; leading proprietary LLMs approach their performance, but open-weight models lag. LLMs exhibit systematic over- and underprediction across different tasks. We identify methods that enable a 4 billion parameter open-weight model to match proprietary models' performance: fine-tuning and aggregating 16 predictions per observation. Improvements generalize to policy-response tasks, which are excluded from fine-tuning. We release our datasets, code, and leaderboard on our website: https://jn-huang.github.io/householdbench
Chinese Translation
大型语言模型（LLMs）有潜力实现经济学中的一个关键目标：在多种环境下建立家庭决策的定量模型。然而，现有的评估覆盖的调查和结果较少，并且未研究家庭如何适应变化的经济条件。我们引入了一种新的评估方法，家庭基准（HouseholdBench），该方法整合了6个美国家庭调查和32个预测任务，涵盖与消费、收入、劳动、预期和住房相关的数值、分类和概率结果。通过使用过去的行为、人口统计数据和宏观经济条件，这些任务测试LLMs是否能够预测行为，包括家庭如何对各种政策的变化做出调整。我们将13个专有和开放权重的LLMs与无变化基线和梯度提升树模型进行比较。大多数LLMs在政策响应任务中超越了无变化基线——最佳模型在数值结果上的误差降低了12.2%。在大多数任务中，梯度提升树排名第一；领先的专有LLMs接近其表现，但开放权重模型则落后。LLMs在不同任务中表现出系统性的过度和不足预测。我们确定了能够使一个40亿参数的开放权重模型匹配专有模型性能的方法：对每个观察进行微调和聚合16个预测。改进在排除微调的政策响应任务中也具有普遍性。我们在网站上发布了我们的数据集、代码和排行榜：https://jn-huang.github.io/householdbench
cs.CL / 23 / 2610.07572
Two Vectors Replace In-Context Demos: Structured Task Adaptation via Embeddings
两个向量替代上下文示例：通过嵌入实现结构化任务适应
Abstract
In-context learning (ICL) adapts frozen large multimodal models (LMMs) to new tasks from a few demonstrations (demos), but re-encodes them at every query, where each demo image adds up to hundreds of visual tokens. Demo-free methods remove this cost with a compact task state. However, they add it at locations searched per task or at every decoder layer, where task parameters grow with depth. Moreover, inserted tokens or keys cannot change how the original prompt divides its attention within a layer. To address these issues, we propose Structured Task Adaptation via Embeddings (STAVE), which replaces demos with two task-specific vectors added to existing input embeddings. Specifically, a readout vector updates the answer-producing tokens and a context vector updates the other structural token groups. Both are trained with answer labels on prompts with and without demos. We justify these design choices theoretically using a first-order analysis of the loss and a margin bound. Extensive experiments on six LMMs and five large language models show that STAVE matches or outperforms state-of-the-art methods on multimodal tasks with far fewer task parameters and surpasses 15-shot ICL and prior task vectors on 18 text tasks, all at zero-shot inference cost.
Chinese Translation
上下文学习（ICL）通过少量示例（demos）将冻结的大型多模态模型（LMMs）适应于新任务，但在每次查询时重新编码这些示例，每个示例图像会增加数百个视觉标记。无示例的方法通过紧凑的任务状态消除了这一成本。然而，它们在每个任务搜索的位置或每个解码器层中增加了该成本，任务参数随着深度的增加而增长。此外，插入的标记或键无法改变原始提示在层内分配注意力的方式。为了解决这些问题，我们提出了通过嵌入实现结构化任务适应（STAVE），该方法用两个特定于任务的向量替代示例，并将其添加到现有的输入嵌入中。具体而言，一个读取向量更新生成答案的标记，另一个上下文向量更新其他结构性标记组。两者均在有无示例的提示上使用答案标签进行训练。我们通过对损失的一级分析和边际界限从理论上证明了这些设计选择的合理性。在六个LMM和五个大型语言模型上的广泛实验表明，STAVE在多模态任务上与最先进的方法相匹配或超越，同时任务参数远少于这些方法，并在18个文本任务上超越了15-shot ICL和先前的任务向量，所有这些都在零-shot推理成本下实现。
cs.CL / 24 / 2610.07587
Large Language Model Orchestration under Heterogeneous Preferences via Explicit Persona Inference
基于显式角色推断的异质偏好下的大型语言模型协调
Abstract
LLM orchestration investigates how an orchestrator coordinates a group of autonomous agents to achieve common goals or maximize collective welfare. The agents are typically heterogeneous, each holding a private preference that it pursues but does not reveal. Inferring such hidden preferences from behavior has been a subject of long-standing research in game theory and multi-agent systems. The core challenge lies in maintaining a belief over every agent's preference and updating it from the agents' observed actions. Existing LLM orchestrators carry that belief as prompt text with no explicit update rule. This lets early errors persist and propagate rather than be corrected. We therefore propose \textbf{HARP} (Heterogeneous-preference Agent oRchestration via Preference inference), a novel framework that moves the belief out of the prompt. Specifically, HARP maintains one numeric posterior per agent over a finite set of candidate preferences and updates it in closed form by Bayes' rule. The language model supplies only actions and per-candidate likelihoods, so estimation is decoupled from its reasoning. We prove that HARP attains the same $\tilde O(\sqrt K)$ Bayesian regret as explicit joint inference when the factorization is exact. Furthermore, HARP\textsuperscript{+} augments planning with a bonus for actions that distinguish the candidates, so inference continues even when the optimal action is uninformative. Empirical results on three substrates, ranging from payoffs the preferences fully determine, through payoffs that depend on more than them, to scales where explicit joint inference is infeasible, demonstrate that HARP\textsuperscript{+} is the strongest non-oracle method across the class our theory identifies.
Chinese Translation
大型语言模型（LLM）协调研究如何协调一组自主代理以实现共同目标或最大化集体福利。这些代理通常是异质的，每个代理持有一个私有偏好，该偏好是其追求但并不透露的。从行为中推断这些隐藏的偏好一直是博弈论和多智能体系统研究的一个长期课题。核心挑战在于保持对每个代理偏好的信念，并根据代理的观察行为进行更新。现有的LLM协调器将这种信念作为提示文本携带，并没有明确的更新规则。这使得早期的错误得以持续并传播，而不是被纠正。因此，我们提出了 extbf{HARP}（通过偏好推断进行异质偏好代理协调），这是一个新颖的框架，它将信念移出提示。具体而言，HARP为每个代理在有限的候选偏好集上维护一个数值后验，并通过贝叶斯规则以封闭形式进行更新。语言模型仅提供动作和每个候选的似然性，因此估计与其推理解耦。我们证明，当因子分解是精确时，HARP达到与显式联合推断相同的$ ilde O( ext{sqrt} K)$贝叶斯遗憾。此外，HARP extsuperscript{+}通过对区分候选的动作给予奖励来增强规划，因此即使在最优动作没有信息时，推断仍然继续。在三个不同的基础上进行的实证结果，从完全由偏好决定的收益，到依赖于其他因素的收益，再到显式联合推断不可行的规模，证明了HARP extsuperscript{+}是我们理论所识别的类中最强的非神谕方法。
cs.CL / 25 / 2610.07591
Recurrent Looped Transformer
递归循环变换器
Abstract
State tracking requires an update at every input, but the depth a Transformer applies to each token is fixed regardless of sequence length. We introduce the Recurrent Looped Transformer (RLT), which splits its layers between a parallel causal encoder and a recurrent decoder. At each token, the decoder merges the encoder output with the previous token's final decoder state, so the computation path grows with sequence length at a fixed per-token cost. On six algorithmic tasks, we compare five splits of eight layers with an eight-layer Transformer over three seeds. Trained on at most 40 bits, two RLT splits generalize parity to 256 bits with 100% accuracy in every seed, while the Transformer stays at chance. On swap-based $S_5$ permutation tracking at eight times the training length, RLT reaches 97% final-state accuracy versus under 1% for the Transformer, and accuracy increases with decoder depth. On modular arithmetic beyond the training lengths, RLT reaches up to 93% versus 33% for the Transformer. Ablations show that these gains depend on the feedback: removing it drops parity and swap-based $S_5$ to chance at every split. Updating the feedback once per four-token chunk lets known tokens in a chunk run in parallel and keeps 64-bit parity at 99%, while permutation tracking depends on per-token feedback: chunking lowers length-64 swap-based $S_5$ from 100% to 20%.
Chinese Translation
状态跟踪需要在每个输入时进行更新，但变换器对每个标记应用的深度是固定的，与序列长度无关。我们提出了递归循环变换器（Recurrent Looped Transformer, RLT），它在并行因果编码器和递归解码器之间划分其层。在每个标记处，解码器将编码器的输出与前一个标记的最终解码器状态合并，因此计算路径随着序列长度的增加而增长，且每个标记的计算成本保持固定。在六个算法任务中，我们比较了五种八层划分与一个八层变换器在三个随机种子下的表现。在最多训练40位的情况下，两个RLT划分在每个随机种子中以100%的准确率将奇偶性推广到256位，而变换器的表现仅为随机水平。在基于交换的$S_5$置换跟踪中，训练长度为八倍时，RLT的最终状态准确率达到97%，而变换器则低于1%，且准确率随着解码器深度的增加而提高。在超出训练长度的模算术中，RLT的准确率高达93%，而变换器为33%。消融实验表明，这些提升依赖于反馈：去除反馈会使奇偶性和基于交换的$S_5$在每个划分中降至随机水平。每四个标记块更新一次反馈使得块内已知标记可以并行运行，并保持64位奇偶性在99%，而置换跟踪则依赖于每个标记的反馈：将长度为64的基于交换的$S_5$分块会使准确率从100%降至20%。
cs.CL / 26 / 2610.07643
Monte Carlo Estimation for KV Cache Eviction
基于蒙特卡洛估计的键值缓存驱逐
Abstract
Most KV-cache eviction methods ask, in effect, which memory appeared important while reading the prompt? We instead ask, which memory will matter while answering? Since decoding queries are unavailable at eviction time, prior future-aware methods rely on pseudo-responses or synthetic future-query estimates. We cast fixed-budget future-aware eviction as distributional estimation over plausible model-conditional query trajectories and introduce LORE-KV (Lookahead Output-perturbation with Reliability-weighted Ensembles for Key-Value caches), a training-free method that samples short autoregressive continuations from the frozen target model and uses their response-side query states to estimate prompt-token utility. Tokens are scored by projected leave-one-out attention-output deletion cost and aggregated across sampled futures with optional trajectory weighting. The temporary continuations are discarded before final decoding, requiring no auxiliary model or training. Ablations isolate the mechanism: at B=128, a single response-side continuation recovers about 89% of the gain over the prompt-window control, while additional futures provide smaller improvements. At B=128, LORE-KV raises the LongBench average on Qwen2.5-14B from 45.49 to 48.24 (+2.75) and the 16K RULER average on Mistral-7B from 45.20 to 51.05 (+5.85). Gains diminish at larger cache budgets and coexist with task-level regressions. LORE-KV incurs 1.46-2.77x AnDPro's per-sample wall-clock time as a one-time compression overhead across six dense and hybrid-attention backbones.
Chinese Translation
大多数键值缓存驱逐方法实际上是在询问，在读取提示时，哪个内存显得重要？而我们则询问，在回答时，哪个内存会变得重要？由于在驱逐时无法获得解码查询，之前的未来感知方法依赖于伪响应或合成未来查询估计。我们将固定预算的未来感知驱逐视为对合理模型条件查询轨迹的分布估计，并引入了LORE-KV（前瞻性输出扰动与加权可靠性集成的键值缓存），这是一种无训练的方法，从冻结的目标模型中采样短的自回归延续，并利用其响应侧查询状态来估计提示令牌的效用。令牌的评分通过预测的留一法注意力输出删除成本进行，并在采样的未来中进行聚合，附加轨迹加权为可选。在最终解码之前，临时延续会被丢弃，因此不需要辅助模型或训练。消融实验隔离了机制：在B=128时，单个响应侧延续恢复了约89%的增益，相较于提示窗口控制，而额外的未来提供了较小的改进。在B=128时，LORE-KV将Qwen2.5-14B的LongBench平均值从45.49提升至48.24（+2.75），将Mistral-7B的16K RULER平均值从45.20提升至51.05（+5.85）。增益在更大的缓存预算下减小，并与任务级回归共存。LORE-KV的每个样本的壁钟时间为AnDPro的1.46-2.77倍，作为在六个密集和混合注意力骨干网络上的一次性压缩开销。
cs.CL / 27 / 2610.07659
DLoop: Looped Speculative Decoding
DLoop：循环的推测解码
Abstract
Speculative decoding accelerates autoregressive generation in large language models. In each drafting stage, a lightweight draft model proposes tokens that the target model subsequently verifies. With increasingly capable draft models, we find that the target model frequently accepts all tokens produced in a drafting stage. A verification nevertheless follows each drafting stage, resulting in unnecessary target-model forward passes even when drafting could have continued. Adaptive draft length methods decide during decoding how many draft tokens precede a verification, but they raise the speedup only for autoregressive draft models. For a parallel draft model, drafting further requires target-model hidden states for draft tokens that have not been verified. We propose DLoop, a looped form of speculative decoding that adaptively performs multiple drafting stages before verification. DLoop continues drafting while the draft model remains confident and verifies all accumulated draft tokens together. Loop-aware training keeps the draft model reliable in the additional drafting stages by exposing it to its own hidden states for unverified draft tokens. By spending additional draft-model forward passes, DLoop reduces the number of target-model forward passes required for verification. Across diverse speculative decoding methods including EAGLE-3, DFlash, Domino, DSpark, and multi-token prediction modules, DLoop improves the wall-clock speedup by 5 to 41 percent while preserving lossless decoding. Code will be available at https://github.com/naver-ai/DLoop.
Chinese Translation
推测解码加速了大型语言模型的自回归生成。在每个草拟阶段，轻量级草拟模型提出目标模型随后验证的标记。随着草拟模型能力的增强，我们发现目标模型经常接受草拟阶段产生的所有标记。然而，每个草拟阶段之后仍需进行验证，这导致即使在草拟可以继续的情况下，目标模型也会进行不必要的前向传播。自适应草拟长度方法在解码过程中决定在验证之前有多少草拟标记，但它们仅对自回归草拟模型提高了加速效果。对于并行草拟模型，草拟进一步需要目标模型的隐藏状态，以处理尚未被验证的草拟标记。我们提出了DLoop，一种循环的推测解码形式，它在验证之前自适应地执行多个草拟阶段。DLoop在草拟模型保持信心的情况下继续草拟，并将所有累积的草拟标记一起进行验证。循环感知训练通过使草拟模型接触到未验证草拟标记的隐藏状态，确保草拟模型在额外的草拟阶段中保持可靠。通过增加草拟模型的前向传播次数，DLoop减少了验证所需的目标模型前向传播次数。在包括EAGLE-3、DFlash、Domino、DSpark和多标记预测模块等多种推测解码方法中，DLoop提高了时钟速度加速5%到41%，同时保持无损解码。代码将发布在 https://github.com/naver-ai/DLoop。
cs.CL / 28 / 2610.07700
Detecting LLM-Assisted Vietnamese Writing via Keystrokes under Behavioral Manipulation
通过击键动态在行为操控下检测LLM辅助的越南写作
Abstract
We study the robustness of keystroke dynamics for detecting large language model (LLM)-assisted writing. We introduce a Vietnamese keystroke dataset capturing realistic writing modes, including bona fide composition, transcription, and paraphrasing. We also define a behaviorally grounded threat model in which users deliberately alter typing patterns. To implement the threat model, we create behaviorally manipulated variants of the data designed to evade keystroke-based detection. We evaluate four keystroke modeling approaches: temporal and rhythmic representations, and sequential representations modeled with a one-dimensional convolutional neural network (1D-CNN) and TypeNet, under user-independent and context-independent settings. The results show that sequential models outperform feature-based approaches in most cases and that keystroke signals encode discriminative information about the writing process. However, detection is not uniformly robust: transcription is reliably identified, while paraphrasing and adversarially manipulated samples are frequently misclassified as bona fide when not explicitly modeled. To address this, we incorporate adversarial training using behaviorally manipulated data, which substantially improves separability and robustness. These results suggest that keystroke-based detection depends critically on exposure to diverse writing behaviors, and that strong performance under limited conditions does not generalize to realistic or adversarial settings without targeted modeling.
Chinese Translation
我们研究了击键动态在检测大型语言模型（LLM）辅助写作中的鲁棒性。我们引入了一个越南击键数据集，捕捉真实的写作模式，包括真实创作、转录和改写。我们还定义了一个基于行为的威胁模型，在该模型中，用户故意改变打字模式。为了实现这一威胁模型，我们创建了旨在规避基于击键检测的行为操控数据变体。我们评估了四种击键建模方法：时间和节奏表示，以及使用一维卷积神经网络（1D-CNN）和TypeNet建模的序列表示，在用户独立和上下文独立的设置下。结果表明，序列模型在大多数情况下优于基于特征的方法，并且击键信号编码了关于写作过程的区分信息。然而，检测并不均匀鲁棒：转录被可靠识别，而改写和对抗性操控样本在未明确建模时常被错误分类为真实样本。为了解决这个问题，我们结合了使用行为操控数据的对抗训练，这显著提高了可分离性和鲁棒性。这些结果表明，基于击键的检测在很大程度上依赖于对多样化写作行为的接触，并且在有限条件下的强表现并不能推广到现实或对抗环境中，而不进行针对性的建模。
cs.CL / 29 / 2610.07716
Readout Stability in Prefill-Only Decision Models:Zero-Label Prediction and Inference-Time Compute Allocation
仅预填决策模型中的读出稳定性：零标签预测与推理时计算分配
Abstract
Prefill-only decision models inspired by the Jev model score every candidate in a menu during a single forward pass and never decode, which makes one call one to two orders of magnitude cheaper than a same-scale generative language model. We show that this read-out structure comes with a testable property. When an intervention changes only the candidate menu and leaves the input text fixed, the post-intervention accuracy is already determined by the cached first-pass distribution. The estimator restricts the pass-1 probabilities to the menu, renormalizes, and reads off the argmax; it uses no labels and no second forward pass. Across seven model families, ten datasets and two task types, menu-only interventions are predicted to within 4.2 points, and for one family the prediction is exact. A probability-level variant of the same estimator errs by 21.0 points, so the property lives in the ranking rather than in the probabilities and is not recovered by calibration. Same-scale generative language models do not share the property. On those models the same estimator errs by 1.6 to 15.8 points and degrades as the model grows. The property turns inference-time compute into a decision that can be made before deployment. Uniform extra passes buy calibration but almost no accuracy; at matched cost a confidence cascade outperforms every scheme that re-asks the same model, and curating the menu beats enlarging the model, with a 0.8B model on a curated 5-candidate menu reaching 95.4% on CLINC150 against 80.0% for a 4B model on the full 150-label menu.Code and data are available at https://github.com/rlisml/jev-cascade.
Chinese Translation
仅预填决策模型受到Jev模型的启发，在单次前向传递中对菜单中的每个候选项进行评分，并且从不解码，这使得一次调用的成本比同规模的生成语言模型低一个到两个数量级。我们展示了这种读出结构具有可测试的特性。当干预仅改变候选菜单而保持输入文本不变时，干预后的准确性已经由缓存的第一次传递分布决定。估计器将第一次传递的概率限制在菜单内，进行重新归一化，并读取argmax；它不使用标签，也不进行第二次前向传递。在七个模型家族、十个数据集和两种任务类型中，仅菜单干预的预测误差在4.2分以内，对于一个模型家族，预测是准确的。相同估计器的概率级变体误差为21.0分，因此这一特性存在于排名中，而不是概率中，并且不会通过校准恢复。相同规模的生成语言模型不具备这一特性。在这些模型上，相同估计器的误差在1.6到15.8分之间，并且随着模型的增长而恶化。该特性将推理时的计算转化为可以在部署前做出的决策。均匀的额外传递可以获得校准，但几乎没有提高准确性；在相同成本下，信心级联优于任何重新请求相同模型的方案，而策划菜单的效果优于扩大模型，在一个策划的5候选菜单上，0.8B模型在CLINC150上达到了95.4%的准确率，而4B模型在完整150标签菜单上的准确率为80.0%。代码和数据可在https://github.com/rlisml/jev-cascade获取。
cs.CL / 30 / 2610.07722
Does Steering Break Your Model? A Multi-Dimensional Evaluation Suite for LLM Steering Methods
方向控制会破坏你的模型吗？一种多维评估套件用于大语言模型（LLM）方向控制方法
Abstract
Activation steering provides a lightweight and flexible way to control large language model (LLM) behavior. However, effective steering requires more than inducing the intended behavior: it should also limit unintended changes and remain robust across inputs and training data. Existing evaluations cover these dimensions only in fragments. As a result, the trade-offs between efficacy and side effects have not been systematically characterized. We introduce SteerScope, a two-axis, multi-dimensional evaluation suite that jointly characterizes steering outcomes and method properties through 15 metrics. We score target efficacy and side effects on language quality, task capabilities, and safety and reliability, and further assess generalization and data dependence through steering-specific metrics for sample efficiency and sample sensitivity. Rather than comparing methods at a single operating point, we characterize the trade-offs between efficacy and side effects. Under matched models, tasks, and evaluation protocols, we benchmark 23 methods spanning 4 families, including prompting, LoRA, and SFT as baseline methods, and release the suite as an extensible codebase. We find that current activation steering methods do not yet surpass the Prompt Steering baseline in their overall balance between steering efficacy and side effects: across both model scales, no evaluated activation steering method achieves higher efficacy without incurring greater composite side effects. We further uncover a consistent coupling between steering efficacy and side effects. Under OOD prompts, target efficacy is often preserved, whereas side effects tend to become more pronounced, particularly through declines in instruction relevance and fluency. Methods also exhibit sharply different sample-efficiency profiles.
Chinese Translation
激活方向控制提供了一种轻量且灵活的方式来控制大语言模型（LLM）的行为。然而，有效的方向控制不仅需要引导预期的行为，还应限制意外的变化，并在输入和训练数据中保持稳健性。现有的评估仅在片段中覆盖这些维度。因此，效果与副作用之间的权衡尚未得到系统性表征。我们引入了 SteerScope，这是一种二维多维评估套件，通过 15 个指标共同表征方向控制的结果和方法特性。我们在语言质量、任务能力以及安全性和可靠性方面对目标效果和副作用进行评分，并通过针对方向控制的样本效率和样本敏感性指标进一步评估泛化能力和数据依赖性。我们并不在单一操作点比较方法，而是表征效果与副作用之间的权衡。在匹配的模型、任务和评估协议下，我们基准测试了 23 种方法，涵盖 4 个家族，包括提示（prompt）、LoRA 和 SFT 作为基线方法，并将该套件发布为可扩展的代码库。我们发现，目前的激活方向控制方法在方向控制效果与副作用之间的整体平衡上尚未超过提示方向控制的基线：在两个模型规模中，没有评估的激活方向控制方法能够在不产生更大综合副作用的情况下实现更高的效果。我们进一步发现方向控制效果与副作用之间存在一致的耦合关系。在超出分布（OOD）提示下，目标效果往往得以保留，而副作用则趋于更加明显，特别是通过指令相关性和流畅性的下降。不同方法的样本效率特征也表现出明显差异。
cs.CL / 31 / 2610.07730
SanSi: A Looped Typed Decision Model for System 1.5 Thinking
SanSi：一种用于系统1.5思维的循环类型决策模型
Abstract
Typed decision models answer a declared question without generating text: a decision head returns a probability for each of the declared options in a single forward pass. A single pass is fast, intuitive System 1 thinking. We study what lies between one pass and generated reasoning: looping, in which the same layers are recursively applied several times before one typed readout. Each loop lets the model revise its hidden state before it commits to an answer, without generating a token; we call this System 1.5 thinking. We propose SanSi, which turns a pre-trained looped language model into a typed decision model. The option probabilities are read after every loop, and every loop is trained with a proper scoring rule, so that one model serves every budget from one loop to eight in a single run. On 10,027 test decisions from 59 sources, SanSi reaches 72.0% accuracy: 13.5 points above a non-looped model of the same shape trained with the same recipe, 5.3 points above a newer non-looped model of its size, and 1.8 points below one with three times the parameters. On two depth-controlled tasks, loops extend the solvable depth beyond the depths seen in training, where the larger single-pass model fails. Used as the judge for policy optimization with reinforcement learning, without gold answers, SanSi raises the generator's F1 by 7.7 points.
Chinese Translation
类型决策模型在不生成文本的情况下回答声明的问题：决策头在一次前向传递中为每个声明的选项返回一个概率。一次传递是快速、直观的系统1思维。我们研究了一次传递与生成推理之间的内容：循环，其中相同的层在一次类型读取之前递归地应用多次。每个循环使模型在承诺答案之前修正其隐藏状态，而不生成标记；我们称之为系统1.5思维。我们提出了SanSi，它将一个预训练的循环语言模型转变为一个类型决策模型。选项概率在每个循环后读取，并且每个循环都使用适当的评分规则进行训练，以便一个模型可以在一次运行中服务于从一个循环到八个循环的所有预算。在来自59个来源的10,027个测试决策中，SanSi的准确率达到了72.0%：比同样形状、使用相同方法训练的非循环模型高出13.5个百分点，比其规模更小的新型非循环模型高出5.3个百分点，且比参数量是其三倍的模型低1.8个百分点。在两个深度控制任务中，循环将可解深度扩展到训练中未见的深度，而较大的单次传递模型则失败。作为无金标准答案的强化学习政策优化的评判者，SanSi将生成器的F1值提高了7.7个百分点。
cs.CL / 32 / 2610.07753
From Evidence to Action: How Tool-Using Agents Fail
从证据到行动：工具使用代理的失败
Abstract
Tool-using agents make consequential changes to external state, yet correct outcomes do not guarantee that their actions were supported by evidence established beforehand. We study where this evidence-to-action chain breaks as agents move from deciding whether to act to executing single actions and dependent workflows. Across ten model-harness configurations, strong static action assessment can coexist with much weaker interactive execution. Failures often begin before execution: agents stop with incomplete investigation or act before required evidence is established. Once required evidence is obtained, single-action execution is usually reliable, while multi-action workflows additionally expose unresolved prerequisites and incomplete execution. For this analysis, we introduce SafeActBench, comprising 656 cases across six operational domains and five protocols that progress from static action judgment and investigated non-action to single- and multi-action workflows. A provenance-bound Evidence Ledger and deterministic trajectory evaluator track what information was established, when actions occurred, and whether downstream dependencies were satisfied. These results show that failures arise not only from missing information, but also from how agents use established evidence when deciding and executing actions.
Chinese Translation
工具使用代理对外部状态进行重要的改变，但正确的结果并不保证其行动是基于事先建立的证据支持的。我们研究了在代理从决定是否行动到执行单一动作和依赖工作流的过程中，这一证据到行动的链条在哪里断裂。在十种模型配置中，强大的静态行动评估可以与较弱的互动执行并存。失败往往在执行之前就开始：代理在调查不完整的情况下停止，或在所需证据尚未建立之前就开始行动。一旦获得所需的证据，单一动作的执行通常是可靠的，而多动作工作流则额外暴露出未解决的前提条件和不完整的执行。为此分析，我们引入了SafeActBench，包含656个案例，涵盖六个操作领域和五个协议，这些协议从静态行动判断和调查非行动进展到单一和多动作工作流。一个基于来源的证据账本和确定性轨迹评估器跟踪了何时建立了哪些信息、何时发生了行动，以及下游依赖关系是否得到了满足。这些结果表明，失败不仅源于缺失的信息，还源于代理在决定和执行行动时如何使用已建立的证据。
cs.CL / 33 / 2610.07764
No Transformer Beats Six Covariates: Long-Horizon Prediction of Depressive Symptoms from Childhood Essays
没有任何变换器超越六个协变量：从儿童作文中长期预测抑郁症状
Abstract
Natural language processing (NLP) models can detect depression-related language in text written near the time symptoms are measured, but whether pretrained transformers can predict depressive symptoms from text written twelve years earlier is largely untested. In the National Child Development Study, a British birth cohort, we predict probable depressive symptoms at age 23 from essays the same people wrote at age 11. Our baseline, a logistic regression on six childhood covariates, outperforms every text model that sees only the essay: seven fine-tuned transformers, a bag-of-words model, frozen embeddings and four zero-shot large language models. Its area under the receiver operating characteristic curve (AUC-ROC) is 0.737 against 0.670 for the best transformer on the primary seed, and no added text score detectably raises the baseline's AUC-ROC. None of the five domain-pretrained transformers detectably beats its general-domain control after Bonferroni correction. For long-horizon prediction, the baseline remains the model to beat.
Chinese Translation
自然语言处理（NLP）模型可以检测在症状测量时所写文本中的与抑郁相关的语言，但预训练的变换器是否能够从十二年前写的文本中预测抑郁症状尚未得到充分验证。在国家儿童发展研究（National Child Development Study）中，我们预测在23岁时可能出现的抑郁症状，这些症状是基于同一人在11岁时所写的作文。我们的基线模型是对六个儿童协变量进行的逻辑回归，其表现优于所有仅使用作文的文本模型：七个微调的变换器、一个词袋模型、固定嵌入以及四个零样本大型语言模型。其接收者操作特征曲线下面积（AUC-ROC）为0.737，而最佳变换器在主要种子上的AUC-ROC为0.670，且没有任何附加的文本评分显著提高基线的AUC-ROC。在Bonferroni校正后，五个领域预训练的变换器均未能显著超越其通用领域对照。在长期预测中，基线模型仍然是需要超越的标杆。
cs.CL / 34 / 2610.07774
Reading, Not Manipulating: Leveraging Router Logits for Multimodal Safety in MoE Vision-Language Models
阅读，而非操控：利用路由器逻辑为混合专家视觉-语言模型的多模态安全提供支持
Abstract
Vision-language models (VLMs) face compositional safety risks where harmful intent emerges from the interaction between visual and textual inputs. As mixture-of-experts (MoE) VLMs become increasingly common, recent work has explored various safety interventions, including prompting, supervised fine-tuning, and routing-based expert steering. However, these methods show inconsistent improvements across models and evaluation distributions, and the intervention into model behavior or internal states introduce safety-utility tradeoffs by over-refusal. Rather than manipulating internal states to steer model behavior, we instead ask whether routing states can serve as diagnostic signals for multimodal safety. We find that router logits indeed provide highly predictive signals of whether a multimodal input is safe or not. Motivated by this observation, we introduce a lightweight router-logit safety detector that reads out routing signals during prompt prefill and identifies unsafe requests before generation, without modifying model parameters or expert routing. Across Qwen3-VL and Kimi-VL, the proposed detector substantially reduces safety errors on the HoliSafe benchmark and resoundingly generalizes to out-of-distribution safety benchmarks featuring different safety patterns, including MISHard and MM-SafetyBench. The success of the proposed router-logit detector also suggests a broader perspective on model internals: rather than focusing only on manipulating internal components to steer behavior, simply reading naturally emerging signals and linking them to an external safety mechanism can provide a simple, effective, and non-intrusive complement to existing safety interventions.
Chinese Translation
视觉-语言模型（VLMs）面临组合安全风险，其中有害意图源于视觉和文本输入之间的交互。随着混合专家（MoE）VLMs的日益普及，近期研究探讨了各种安全干预措施，包括提示、监督微调和基于路由的专家引导。然而，这些方法在模型和评估分布上的改进效果不一致，干预模型行为或内部状态可能会因过度拒绝而引入安全与效用的权衡。我们并不试图操控内部状态以引导模型行为，而是探讨路由状态是否可以作为多模态安全的诊断信号。我们的研究发现，路由器逻辑确实提供了高度预测性的信号，能够判断多模态输入是否安全。基于这一观察，我们提出了一种轻量级的路由器逻辑安全检测器，该检测器在提示预填充期间读取路由信号，并在生成之前识别不安全请求，而无需修改模型参数或专家路由。在Qwen3-VL和Kimi-VL上，所提检测器在HoliSafe基准测试中显著减少了安全错误，并且在不同安全模式的分布外安全基准（如MISHard和MM-SafetyBench）上表现出良好的泛化能力。所提路由器逻辑检测器的成功也提示了对模型内部的更广泛看法：与其仅关注操控内部组件以引导行为，不如简单地读取自然出现的信号并将其与外部安全机制联系起来，这可以为现有的安全干预措施提供一种简单、有效且非侵入性的补充。
cs.CL / 35 / 2610.07780
APEX: Speculate smarter, not deeper
APEX：更聪明地推测，而非更深入
Abstract
Speculative decoding reduces large language model inference latency by drafting multiple tokens before target-model verification, but its effectiveness depends on both the proposal mechanism and draft depth. Fixed configurations cannot respond to changes in predictability, repetition, and acceptance during generation, so deeper drafting can increase wasted computation without proportional speedup. We introduce APEX, a learned controller that balances decoding speed and draft-token waste through request-level expert selection and block-level depth adaptation. APEX-Router selects among EAGLE-3, n-gram, and draft-model speculation for each request, while APEX-Depth adjusts draft length at each verification block using causal decoding signals and recent verifier feedback. APEX models accepted draft length as censored survival feedback, learning position-wise rejection hazards, block execution costs, and an action utility that balances throughput, accepted progress, and wasted tokens. This allows the controller to adapt speculation while retaining the target model's verification procedure. We integrate APEX into vLLM and evaluate it with Qwen3-8B across six workloads, achieving up to 5.24X speedup over autoregressive decoding. Across the aggregate evaluation, APEX-S achieves 4.27X speedup, while APEX-B achieves 3.27X speedup with a 41.0% relative reduction in wasted-token percentage compared with fixed n-gram speculation at k=16, providing distinct operating points for balancing acceleration and draft-token utilization.
Chinese Translation
推测解码通过在目标模型验证之前草拟多个标记来减少大型语言模型的推理延迟，但其有效性依赖于提案机制和草拟深度。固定配置无法应对生成过程中的可预测性、重复性和接受度的变化，因此更深的草拟可能会增加浪费的计算而没有相应的加速。我们提出了APEX，一种学习控制器，通过请求级专家选择和块级深度适应来平衡解码速度和草拟标记的浪费。APEX-Router为每个请求在EAGLE-3、n-gram和草拟模型推测之间进行选择，而APEX-Depth则利用因果解码信号和最近的验证反馈在每个验证块中调整草拟长度。APEX将接受的草拟长度建模为被审查的生存反馈，学习逐位置拒绝风险、块执行成本以及平衡吞吐量、接受进度和浪费标记的行动效用。这使得控制器能够在保留目标模型验证程序的同时调整推测。我们将APEX集成到vLLM中，并在六个工作负载下使用Qwen3-8B进行评估，实现了比自回归解码快多达5.24倍的加速。在综合评估中，APEX-S实现了4.27倍的加速，而APEX-B在与固定n-gram推测（k=16）相比时，浪费标记百分比相对减少41.0%，实现了3.27倍的加速，为平衡加速和草拟标记利用提供了不同的操作点。
cs.CL / 36 / 2610.07817
One Step at a Time: Trading LLM Autonomy for Process Predictability
一步一个脚印：用过程可预测性换取大型语言模型的自主性
Abstract
Organizations automating operational processes need more than a correct outcome: they need to predict how a process will run, know which one actually ran, and inspect it step by step. When an agent is the executor that predictability is normally lost: the prescribed procedure goes into the system prompt, and only a final answer comes back. We deliver the procedure step by step over the Model Context Protocol (MCP) instead: a server releases one step at a time, the agent executes it, and each step returns a structured step_output. This trades autonomy for predictability, and two properties then follow by construction, independent of the executor. The execution path is prescribed before the run, so the process is predictable in advance rather than reconstructed afterwards; and the completed step records form a machine-readable execution log that downstream tooling can audit and optimize step by step. Evaluating 15,475 trials across 13 SOP-Bench domains and four open-weight executors from frontier (Kimi K2.5) to lightweight (Ministral 3 8B), we find step-level delivery makes the executed process predictable and inspectable for every executor, and additionally raises accuracy when the executor is small. Across all four, process adherence rises significantly (76-95% to 95-99%) and ungrounded answers (correct outputs produced without executing the SOP) near-vanish, falling from 2.1-4.5% to 0.2-0.3% of trials (all 95% CIs exclude zero); under prompt-based delivery, 31-49% of correct answers on know_your_business bypass the SOP entirely, even for the frontier executor. Accuracy is where the executor's capability enters: the lightweight executor gains +6.5pp grounded accuracy because supplying the process externally removes a reconstruction burden it cannot carry, while capable ones trade a small raw-accuracy decrement for a predictable, auditable process.
Chinese Translation
自动化运营流程的组织不仅需要正确的结果，还需要预测流程的运行方式，了解实际执行的流程，并逐步检查每个步骤。当代理作为执行者时，这种可预测性通常会丧失：规定的程序被放入系统提示中，最终只返回一个答案。我们通过模型上下文协议（Model Context Protocol, MCP）逐步传递程序：服务器一次释放一个步骤，代理执行该步骤，每个步骤返回一个结构化的步骤输出。这种方式用可预测性换取了自主性，并且由此构建的两个属性与执行者无关。执行路径在运行之前就已规定，因此流程在事前是可预测的，而不是事后重建的；而完成的步骤记录形成了一个机器可读的执行日志，后续工具可以逐步审计和优化。我们在13个SOP-Bench领域和四个开放权重执行者（从前沿执行者Kimi K2.5到轻量级执行者Ministral 3 8B）中评估了15,475次试验，发现逐步交付使得每个执行者的执行过程都变得可预测和可检查，并且在执行者较小的情况下，准确性也有所提高。在所有四个执行者中，流程遵循率显著上升（从76-95%提高到95-99%），而无基础答案（在未执行SOP的情况下产生的正确输出）几乎消失，从2.1-4.5%降至0.2-0.3%（所有95%置信区间均不包括零）；在基于提示的交付下，31-49%的正确答案在know_your_business中完全绕过了SOP，即使对于前沿执行者也是如此。准确性是执行者能力的体现：轻量级执行者的基础准确性提高了6.5个百分点，因为外部提供流程消除了其无法承担的重建负担，而能力较强的执行者则在可预测、可审计的流程中以小幅降低的原始准确性进行权衡。
cs.CL / 37 / 2610.07822
Nucleus Speculative Decoding: Plausibility-Aware Verification Beyond Exact Distribution
核推测解码：超越精确分布的可信度意识验证
Abstract
Speculative decoding accelerates autoregressive generation by using a lightweight draft model to propose multiple tokens that are verified by a target model in parallel. However, the standard acceptance rule focuses on exact distribution correction and rejects tokens that remain highly plausible under the target model when the draft model assigns excess probability. This conservative verification limits the number of draft tokens retained after each verification forward pass. We introduce Nucleus Speculative Decoding (NSD), a relaxed verification method that incorporates target-model plausibility into speculative decoding. NSD accepts a draft token if it satisfies the standard acceptance rule or belongs to the target model's nucleus. We theoretically characterize the distributional deviation introduced by our method and show that the single-step error is exactly determined by the draft model's excess probability within the target nucleus. We further derive sequence-level fidelity bounds that quantify how local deviations accumulate over autoregressive decoding. Experiments across multiple target models and proposal mechanisms demonstrate that NSD consistently improves speculative decoding efficiency while maintaining competitive task performance. Our method achieves throughput speedups of up to $5.16\times$ over autoregressive decoding and up to $3.15\times$ over standard speculative decoding. These improvements coincide with longer accepted lengths, allowing more output tokens to share the cost of each target verification pass. Analysis shows that plausibility-aware verification provides an effective approach for relaxed verification and speculative decoding efficiency. Our code is available at https://github.com/EIT-NLP/Nucleus-Speculative-Decoding.
Chinese Translation
推测解码通过使用轻量级草稿模型并行提出多个令牌，进而由目标模型进行验证，从而加速自回归生成。然而，标准接受规则关注于精确分布的修正，并在草稿模型分配过高概率时拒绝那些在目标模型下仍然高度可信的令牌。这种保守的验证限制了每次验证前向传递后保留的草稿令牌数量。我们提出了核推测解码（Nucleus Speculative Decoding, NSD），这是一种放宽验证的方法，将目标模型的可信度纳入推测解码中。NSD接受一个草稿令牌，如果它满足标准接受规则或属于目标模型的核。我们理论上刻画了我们的方法引入的分布偏差，并展示了单步误差完全由草稿模型在目标核内的过高概率决定。我们进一步推导了序列级别的保真度界限，量化了局部偏差在自回归解码中是如何累积的。在多个目标模型和提议机制上的实验表明，NSD始终提高了推测解码的效率，同时保持了竞争性的任务性能。我们的方法在自回归解码上实现了高达 $5.16 imes$ 的吞吐量加速，相较于标准推测解码则高达 $3.15 imes$。这些改进与更长的接受长度相吻合，使得更多的输出令牌能够分担每次目标验证的成本。分析表明，可信度意识验证为放宽验证和推测解码效率提供了一种有效的方法。我们的代码可在 https://github.com/EIT-NLP/Nucleus-Speculative-Decoding 获得。
cs.CL / 38 / 2610.07847
OMIT the Action: Measuring Framing-Invariant Omission Bias under Philosophical Disagreement
省略行动：在哲学分歧下测量框架不变的省略偏差
Abstract
As LLMs increasingly assist in moral reasoning, omission bias, the tendency to prefer inaction even when equivalent framings reverse substantive outcomes, poses a significant risk of skewed decision-making. Yet omission bias remains underexplored in LLM evaluation, with the few existing studies limited in scale and focused largely on utilitarian-deontological conflicts. To address this gap, we introduce OMIT, a benchmark consisting of 218 paired-frame scenarios across 10 conflict types, constructed by leveraging disagreement patterns from an LLM-based, five-perspective philosophical persona panel (utilitarianism, deontology, virtue ethics, care ethics, and contractualism). Evaluating eight LLMs, we find that omission bias is pervasive but inversely correlates with model size within families. We further evaluate four inference-time interventions and find that interventions encouraging models to consider moral principles before committing to a yes/no answer reduce omission bias and increase frame-consistent responses, although lower omission bias rates can also coincide with shifts toward action-biased responses. Ultimately, this work contributes not only the OMIT benchmark, but also a methodology for using diverse philosophical disagreement signals to evaluate framing-sensitive inaction preferences and the distributional effects of mitigation attempts in LLMs under complex moral conflicts.
Chinese Translation
随着大型语言模型（LLMs）在道德推理中的应用日益增多，省略偏差，即在等效框架下倾向于选择不作为，即使这会逆转实质性结果，给决策带来了显著的偏差风险。然而，省略偏差在LLM评估中仍然未得到充分探讨，现有的少数研究规模有限，主要集中于功利主义与义务论的冲突。为填补这一空白，我们引入了OMIT，一个基准测试，包含218对框架场景，涵盖10种冲突类型，构建过程中利用了来自基于LLM的五种视角哲学角色小组（功利主义、义务论、美德伦理、关怀伦理和契约论）的分歧模式。对八个LLM的评估显示，省略偏差普遍存在，但在同一家族模型中与模型规模呈负相关。我们进一步评估了四种推理时干预措施，发现鼓励模型在做出是/否回答之前考虑道德原则的干预可以减少省略偏差并增加框架一致的响应，尽管较低的省略偏差率也可能与向行动偏向响应的转变相伴随。最终，这项工作不仅贡献了OMIT基准，还提供了一种利用多样的哲学分歧信号来评估框架敏感的不作为偏好及其在复杂道德冲突中缓解尝试的分布效应的方法论。
cs.CL / 39 / 2610.07848
Dynamic Positional Attention Modulation for Parameter-Efficient Fine-Tuning of Large Language Models
动态位置注意力调制用于大语言模型的参数高效微调
Abstract
Parameter-efficient fine-tuning (PEFT) has become a standard approach for adapting large language models to downstream tasks. However, most existing PEFT methods rely on uniform and static adaptations, without accounting for the structured heterogeneity of attention across dimensions, heads, layers, and input tokens. In practice, attention representations exhibit non-uniform behavior, and positional encoding mechanisms such as rotary positional embeddings (RoPE) induce dimension-dependent positional structure, making uniform adaptation suboptimal. In this work, we propose DyPAM (Dynamic Positional Attention Modulation), a PEFT method that adapts how positional information contributes to attention by operating directly on the query and key representations. DyPAM combines input-conditioned, dimension-wise modulation with head-wise and layer-wise structural modulation, performing fine-grained adaptation of positional attention aligned with the RoPE-induced structure without modifying the pretrained backbone. Extensive experiments on mathematical and commonsense reasoning benchmarks across multiple backbone models demonstrate that DyPAM consistently outperforms existing strong PEFT baselines.
Chinese Translation
参数高效微调（PEFT）已成为将大语言模型适应于下游任务的标准方法。然而，大多数现有的PEFT方法依赖于均匀和静态的适应，未考虑到跨维度、头、层和输入标记的注意力结构异质性。在实践中，注意力表示表现出非均匀行为，位置编码机制如旋转位置嵌入（RoPE）引入了维度依赖的位置结构，使得均匀适应变得次优。在本研究中，我们提出了DyPAM（动态位置注意力调制），这是一种PEFT方法，通过直接作用于查询和键表示，调整位置信息对注意力的贡献。DyPAM结合了输入条件的维度调制与头和层的结构调制，实现了与RoPE引导的结构一致的细粒度位置注意力适应，而无需修改预训练的主干模型。在多个主干模型上的数学和常识推理基准的广泛实验表明，DyPAM始终优于现有的强PEFT基线。
cs.CL / 40 / 2610.07863
ReFold: Training-Free Reversible Inter-Turn Context Folding for Long-Horizon Agents
ReFold：无训练可逆的跨轮上下文折叠用于长时间代理
Abstract
Long-horizon LLM agents act on an append-only interaction history that is re-sent to the model at every step, so the context and its cost grow with steps until the sessions exceed the context window. Existing methods manage the context through context requirement prediction, relying on additional model calls, heuristic rules, or trained policies. However, these predictive approaches introduce runtime overhead, invalidate prefix caches, and permanently discard content with no guarantee of recovery. To overcome these limitations, we introduce ReFold: a training-free rendering layer that preserves the underlying interaction history while compressing only the model's rendered context. It removes two kinds of inter-turn redundancy without an auxiliary predictor: content an earlier turn already displayed, replaced by a stub, and turns the agent itself reports finished, folded into a one-line note. Both operators use chunked rendering, rewriting the cached prefix once every few steps rather than at every step. Every removal is strictly reversible, a wrong removal costs one restore from the history rather than permanent content loss. Because it operates at the rendering layer, ReFold is plug-and-play across standard ReAct-style harnesses. Evaluations across five long-horizon benchmarks and two frontier LLMs demonstrate that ReFold reduces token consumption by up to 2.5x and halves the KV-cache memory per session without degrading task success rates. Under capped context budgets, it avoids up to 92% of forced compactions. Under concurrent serving workloads, it reduces request queuing delays by up to 100%, accelerating inference by up to 1.7x, while cutting inference costs by up to 3.4x.
Chinese Translation
长时间的LLM代理在每一步都基于仅追加的交互历史进行操作，因此上下文及其成本随着步骤的增加而增长，直到会话超出上下文窗口。现有方法通过上下文需求预测来管理上下文，这依赖于额外的模型调用、启发式规则或训练策略。然而，这些预测方法引入了运行时开销，使前缀缓存失效，并永久丢弃内容而无法保证恢复。为克服这些限制，我们提出了ReFold：一个无训练的渲染层，它在压缩模型渲染的上下文的同时，保留底层的交互历史。它在没有辅助预测器的情况下去除两种跨轮冗余：早期轮次已经显示的内容被替换为一个存根，以及代理自己报告完成的轮次被折叠成一行笔记。这两种操作都使用分块渲染，每隔几步重写一次缓存的前缀，而不是在每一步都进行。每次移除都是严格可逆的，错误的移除只需从历史中恢复一次，而不是永久丢失内容。由于它在渲染层操作，ReFold可以在标准的ReAct风格的框架中即插即用。在五个长时间基准和两个前沿LLM上的评估表明，ReFold将令牌消耗降低了多达2.5倍，并将每个会话的KV-cache内存减少了一半，而没有降低任务成功率。在有限的上下文预算下，它避免了多达92%的强制压缩。在并发服务负载下，它将请求排队延迟减少了多达100%，将推理加速了多达1.7倍，同时将推理成本降低了多达3.4倍。
cs.CL / 41 / 2610.07887
Visual Abstention in Unified Multimodal Models
统一多模态模型中的视觉弃权
Abstract
Unified multimodal models (UMMs) integrate understanding and generation, yet their generative behavior is rarely governed by what they understand about the task. We formalize visual abstention: when a requested visual transformation is impossible under the task's rules, the model should recognize that no valid solution exists, state this, and decline to generate. We introduce Draw-or-Decline (DoD), a benchmark of 1,050 feasible-infeasible request pairs across 7 task categories that jointly measures editing success and the refusal of infeasible requests. Evaluating 8 UMMs, we find that editing ability and abstention are distinct capabilities: even the strongest editor, at 68.4% editing accuracy, refuses only 0.4% of infeasible requests under ordinary instructions. Their reasoning shows why: the models rarely notice the conflict, and instead plan the edit as if the request were possible, often describing objects that are not in the image, or quietly change the request into one they can complete. Explicitly prompting these UMMs to report infeasibility increases textual refusals but reduces editing accuracy. We propose VisTA (Visual Transformation and Abstention), a training method that pairs feasible and infeasible examples so that a model judges feasibility before deciding whether to generate. We train VisTA-BAGEL to perform feasible edits and decline infeasible requests. Without any reminder, it refuses 93.0% of infeasible requests, up from 0.4% for the strongest editor, while falsely refusing only 0.8% of feasible ones. Unlike a reminder, this does not cost editing accuracy: VisTA-BAGEL completes 74.3% of feasible edits, more than any of the 8 evaluated UMMs.
Chinese Translation
统一多模态模型（UMMs）整合了理解与生成，然而它们的生成行为很少受到对任务理解的影响。我们正式定义了视觉弃权：当请求的视觉转换在任务规则下不可能实现时，模型应识别出不存在有效解决方案，声明这一点，并拒绝生成。我们引入了Draw-or-Decline（DoD），这是一个包含1,050对可行-不可行请求的基准，涵盖7个任务类别，旨在共同衡量编辑成功率和对不可行请求的拒绝。评估8个UMMs后，我们发现编辑能力和弃权是不同的能力：即使是最强的编辑器，其编辑准确率为68.4%，在普通指令下仅拒绝0.4%的不可行请求。它们的推理表明原因：模型很少注意到冲突，而是将编辑计划为请求是可能的，常常描述图像中不存在的物体，或者默默地将请求更改为可以完成的内容。明确提示这些UMMs报告不可行性会增加文本拒绝的数量，但会降低编辑准确率。我们提出了VisTA（视觉转换与弃权），这是一种训练方法，将可行和不可行的示例配对，以便模型在决定是否生成之前判断可行性。我们训练了VisTA-BAGEL，使其能够执行可行编辑并拒绝不可行请求。在没有任何提醒的情况下，它拒绝了93.0%的不可行请求，远高于最强编辑器的0.4%，同时仅错误拒绝了0.8%的可行请求。与提醒不同，这不会影响编辑准确率：VisTA-BAGEL完成了74.3%的可行编辑，超过了评估的8个UMMs中的任何一个。
cs.CL / 42 / 2610.07894
Rethinking Faithfulness in LLMs: A Pairwise Context-Sensitive Perspective
重新思考大型语言模型中的忠实性：一种成对上下文敏感的视角
Abstract
Large language models (LLMs) are expected to answer questions faithfully based on the provided context, abstaining when the context information is insufficient to answer the questions. Existing faithfulness evaluations typically assess each question-context instance in isolation; however, such instance-level evaluation fails to capture a fundamental requirement of faithful behavior: the ability to adapt model responses to changes in available contexts. In particular, a model should provide correct answers when sufficient evidence is present and abstain when it is not. In this work, we propose a Pairwise Faithfulness Benchmark (PFaithBench) that evaluates whether a model can switch between answering and abstaining for the same question under supporting versus non-supporting contexts. Our evaluations across thirty-nine models with seven model families demonstrate that faithfulness fundamentally involves a trade-off between answering and abstaining, and that most current models exhibit a strong bias toward answering, with most faithfulness errors arising from over-answering, i.e., models tend to fabricate a response even when the provided context is insufficient. We further conduct a series of studies on faithfulness training under different data constructions. Our results show that training outcomes are highly sensitive to the specific composition of answering and abstaining data. Constructing answering and abstaining data from mismatched sources can cause models to rely on dataset-specific shortcuts rather than actual context sufficiency. Moreover, increasing answer-supervised data improves answering performance but exacerbates over-answering, while increasing abstaining data reduces hallucination but leads to over-abstention. The code and data are released at https://github.com/tmlr-group/PFaithBench.
Chinese Translation
大型语言模型（LLMs）被期望能够根据提供的上下文忠实地回答问题，当上下文信息不足以回答问题时应选择不回答。现有的忠实性评估通常孤立地评估每个问题-上下文实例；然而，这种实例级评估未能捕捉忠实行为的一个基本要求：根据可用上下文的变化调整模型响应的能力。特别是，当存在足够证据时，模型应提供正确答案，而在证据不足时应选择不回答。在本研究中，我们提出了一种成对忠实性基准（PFaithBench），用于评估模型在支持性与非支持性上下文下是否能够在回答和不回答之间切换。我们对三十九个模型和七个模型家族的评估表明，忠实性本质上涉及回答与不回答之间的权衡，而大多数当前模型表现出强烈的回答偏向，大多数忠实性错误源于过度回答，即模型倾向于在提供的上下文不足时仍然虚构响应。我们进一步对不同数据构造下的忠实性训练进行了系列研究。我们的结果显示，训练结果对回答和不回答数据的具体组成高度敏感。从不匹配的来源构造回答和不回答数据可能导致模型依赖于数据集特定的捷径而非实际的上下文充分性。此外，增加回答监督数据提高了回答性能，但加剧了过度回答，而增加不回答数据则减少了幻觉但导致过度不回答。代码和数据已发布在 https://github.com/tmlr-group/PFaithBench。
cs.CL / 43 / 2610.07902
ARIA: Audio-Driven Melody-Tone Relation Modeling for Cantonese Lyric Authoring
ARIA：基于音频的粤语歌词创作中的旋律-音调关系建模
Abstract
Cantonese lyric writing requires close alignment between lexical tones and melodic pitch. Existing melody-guided lyric generation methods typically rely on symbolic melody to generate lyrics. However, in real songwriting scenarios, melodies are often expressed as raw singing audio or hummed recordings, where pitch is implicit, noisy, and unstructured, making these methods difficult to apply directly. To address this limitation, we propose ARIA, a two-stage audio-driven melody-tone relation modeling framework for Cantonese lyric authoring that generates Cantonese lyrics from singing recordings with provided character-level timestamps. Specifically, we first design a Tri-Stream Relation-Aware Tone Estimator (TRATE) to predict 0243 sequences from timestamped singing audio by modeling multi-stream acoustic cues and relational tonal structure. We then propose a Decoupled Retrieval-Augmented Tone-Conditioned Lyric Generator (DRA-TCLG) to generate fluent lyrics conditioned on predicted tonal plans with retrieval-enhanced lexical guidance. Moreover, we construct a large-scale aligned audio-Jyutping-0243 dataset from real Cantonese singing recordings to support this new task. Experimental results demonstrate that ARIA achieves strong performance in both 0243 prediction and tone-consistent lyric generation, validating the effectiveness of the proposed framework.
Chinese Translation
粤语歌词创作需要词汇音调与旋律音高之间的紧密对齐。现有的旋律引导歌词生成方法通常依赖于符号化旋律来生成歌词。然而，在实际的歌曲创作场景中，旋律往往以原始的歌唱音频或哼唱录音的形式表达，其中音高是隐含的、嘈杂的且无结构的，这使得这些方法难以直接应用。为了解决这一限制，我们提出了ARIA，一个基于音频的两阶段旋律-音调关系建模框架，用于粤语歌词创作，该框架能够从带有字符级时间戳的歌唱录音中生成粤语歌词。具体而言，我们首先设计了一个三流关系感知音调估计器（Tri-Stream Relation-Aware Tone Estimator, TRATE），通过建模多流声学线索和关系音调结构，从带时间戳的歌唱音频中预测0243序列。然后，我们提出了一个解耦检索增强音调条件歌词生成器（Decoupled Retrieval-Augmented Tone-Conditioned Lyric Generator, DRA-TCLG），以检索增强的词汇指导为基础，生成与预测音调计划相一致的流畅歌词。此外，我们从真实的粤语歌唱录音中构建了一个大规模对齐的音频-粤拼-0243数据集，以支持这一新任务。实验结果表明，ARIA在0243预测和音调一致的歌词生成方面均表现出强劲的性能，验证了所提框架的有效性。
cs.CL / 44 / 2610.07936
Pseudowords as probes: Large Language Models show little of the sublexical sensitivity that governs human pseudoword processing
伪词作为探针：大型语言模型显示出对人类伪词处理的亚词汇敏感性较低
Abstract
Systematicity, the probabilistic mapping of form to meaning, permeates language at all levels, and sublexical cues have been shown to govern human pseudoword processing. Yet whether LLMs exhibit comparable sensitivity to these cues remains unclear. We tested five LLMs on two Italian two-alternative forced-choice pseudoword experiments and compared their responses with a human behavioural baseline. LLMs aligned more reliably with humans when real-word options provided a lexical familiarity cue than in the pseudoword-only condition, where they fell substantially below fastText, a character-n-gram model. In addition, the sublexical cosine-similarity cue that reliably drove human--fastText agreement did not consistently transfer to human--LLM alignment, and reasoning-token expenditure bore no consistent relation to human processing difficulty. These findings suggest that LLMs do not necessarily share the sublexical cues that govern human pseudoword processing; we discuss tokenization and training-data coverage as candidate explanations.
Chinese Translation
系统性，即形式与意义的概率映射，渗透到语言的各个层面，亚词汇线索已被证明在调控人类伪词处理方面发挥作用。然而，LLMs（大型语言模型）是否表现出对这些线索的相应敏感性仍不清楚。我们在两个意大利语的二选一强制选择伪词实验中测试了五个LLMs，并将它们的反应与人类行为基线进行了比较。当真实词选项提供了词汇熟悉度线索时，LLMs与人类的对齐更加可靠，而在仅有伪词的条件下，它们的表现显著低于fastText（一个字符n-gram模型）。此外，可靠驱动人类与fastText一致的亚词汇余弦相似度线索并未始终转移到人类与LLM的对齐上，推理令牌的消耗与人类处理难度之间没有一致关系。这些发现表明，LLMs不一定共享调控人类伪词处理的亚词汇线索；我们讨论了分词和训练数据覆盖作为可能的解释。
cs.CL / 45 / 2610.07937
Leveraging a four-quadrant approach for evaluating Redpine Science
利用四象限方法评估 Redpine Science
Abstract
Redpine Science gives models and agents a single access point to a wide range of peer-reviewed literature, queried directly through the Model Context Protocol (MCP) and an API. This report evaluates Redpine Science on two levels: the relevance of the retrieved chunks, and a model's answer when it has access to Redpine Science compared to web search. Both public and expert-validated benchmarks are used. Public benchmarks are a widely accepted way to test model development and are comparable across labs, but risk saturation and memorization. To address this, we complement them with an expert-validated question set. In total, this report presents four evaluations. On ScholarQABench SciFact, the public answer-quality benchmark reported here, an agent with Redpine Science answers 94.4% of claims correctly against 87.6% with no retrieval. On the expert-validated question set, an agent with Redpine Science states 80.1% of the required claims against 70.2% for an agent restricted to web search. On the 668 queries of a public retrieval benchmark whose gold paper Redpine holds, stripped of any model reasoning, Redpine Science places the correct source paper in its top ten results for 83.1% of queries (Recall@10), against 79.3% for the benchmark's creator. A blinded expert relevance panel places Redpine Science's Precision@5 at 75.2% against 39.8% for the PubMed search tool. We release the expert-validated question set and instructions to reproduce every headline result above, at https://github.com/redpine-ai/benchmarks.
Chinese Translation
Redpine Science 为模型和代理提供了一个单一的访问点，能够直接通过模型上下文协议（Model Context Protocol, MCP）和 API 查询广泛的同行评审文献。本报告从两个层面评估 Redpine Science：检索到的片段的相关性，以及模型在访问 Redpine Science 时的回答与仅通过网络搜索时的比较。使用了公共基准和专家验证的基准。公共基准是测试模型开发的广泛接受的方法，且在不同实验室之间具有可比性，但存在饱和和记忆的风险。为了解决这一问题，我们用专家验证的问题集进行了补充。总的来说，本报告呈现了四项评估。在 ScholarQABench SciFact 上，本文报告的公共答案质量基准中，使用 Redpine Science 的代理正确回答了 94.4% 的声明，而没有检索时为 87.6%。在专家验证的问题集上，使用 Redpine Science 的代理陈述了 80.1% 的所需声明，而仅限于网络搜索的代理为 70.2%。在一个公共检索基准的 668 个查询中，Redpine 拥有的金标准论文在去除任何模型推理后，Redpine Science 在其前十个结果中为 83.1% 的查询提供了正确的源论文（Recall@10），而基准的创建者为 79.3%。一个盲评专家相关性小组将 Redpine Science 的 Precision@5 评估为 75.2%，而 PubMed 搜索工具为 39.8%。我们在 https://github.com/redpine-ai/benchmarks 发布了专家验证的问题集和重现上述每个结果的说明。
cs.CL / 46 / 2610.07940
Hybrid Latent Attention for Looped Language Models
循环语言模型的混合潜在注意力
Abstract
Looped language models apply the same stack of layers T times to each token, which deepens the model without adding parameters but multiplies its key-value (KV) cache by T. The larger cache limits how many sequences a GPU can decode at once and slows each decoding step, which reads the whole cache. We propose Hybrid Latent Attention (HLA), which keeps exact keys and values within a sliding window of W recent tokens and stores each older token as a compact latent that the query of each loop reads directly, without reconstructing keys and values. We uptrain HLA on Ouro looped models (T=4) with 1.4B and 2.6B parameters, keeping the pretrained weights frozen and training only the added parameters to reproduce the original attention. The cache shrinks by 10.7x per token, fitting 4.0-8.8x as many concurrent sequences per GPU, and decoding throughput improves by 2.5x at 1K-token contexts and by up to 7.4x at 16K. HLA retains over 97% of the original accuracy on math, knowledge and reasoning benchmarks, and 96-100% on long-context retrieval up to 16K tokens. After supervised fine-tuning, it performs on par with the fine-tuned original model on competition-level math.
Chinese Translation
循环语言模型对每个标记应用相同的层堆叠 T 次，这在不增加参数的情况下加深了模型，但将其关键值（KV）缓存乘以 T。更大的缓存限制了 GPU 一次可以解码的序列数量，并减慢了每个解码步骤，因为每次解码都需要读取整个缓存。我们提出了混合潜在注意力（Hybrid Latent Attention, HLA），它在最近 W 个标记的滑动窗口内保留精确的键和值，并将每个较旧的标记存储为一个紧凑的潜在表示，供每个循环的查询直接读取，而无需重建键和值。我们在 Ouro 循环模型（T=4）上对 HLA 进行再训练，模型参数为 1.4B 和 2.6B，保持预训练权重不变，仅训练新增参数以重现原始注意力。每个标记的缓存缩小了 10.7 倍，使每个 GPU 可以适应 4.0-8.8 倍的并发序列，并且在 1K 标记上下文下解码吞吐量提高了 2.5 倍，在 16K 上提高了最多 7.4 倍。HLA 在数学、知识和推理基准测试中保留了超过 97% 的原始准确性，在长上下文检索中（最多 16K 标记）保持了 96-100% 的准确性。在经过监督微调后，其在竞争级数学任务上的表现与微调后的原始模型相当。
cs.CL / 47 / 2610.08018
Structured but Silent: Probing Capability Requirements in LLM Hidden States
结构化但沉默：探究LLM隐藏状态中的能力需求
Abstract
Reliable tool use requires more than triggering a mechanism or matching a query to an API description. Before selecting a specific tool, an agent must first infer the capability requirements implied by the user query. In this paper, we investigate whether these query-side capability requirements are linearly decodable from LLM hidden representations prior to generation, and how this hidden-state accessibility compares with explicit verbal classification. We introduce TACIT, a framework that decomposes external requirements along three fundamental axes: Source, Transformation, and World Effect, defining eight structurally distinct capability classes. Using 1,600 balanced training queries from benchmarks, synthetic examples, and new domain scenarios, we train linear probes on pre-generation hidden states from four open-weight LLM families. Our empirical results demonstrate that fine-grained capability structures are linearly decodable with high accuracy across all models. Crucially, however, we expose a representation-to-verbalization gap: these same models are significantly less reliable when asked to explicitly classify the same queries in natural language. This disconnect indicates that information about required external capabilities is linearly accessible in LLM hidden representations but not reliably expressed, a phenomenon we define as "structured but silent."
Chinese Translation
可靠的工具使用不仅仅需要触发机制或将查询与API描述匹配。在选择特定工具之前，代理必须首先推断用户查询所暗示的能力需求。本文研究了这些查询端能力需求是否可以从生成前的LLM隐藏表示中线性解码，以及这种隐藏状态的可访问性与显式语言分类的比较。我们引入了TACIT框架，该框架沿三个基本轴线（来源、转化和世界效应）分解外部需求，定义了八个结构上不同的能力类别。通过使用来自基准测试、合成示例和新领域场景的1,600个平衡训练查询，我们在四个开放权重的LLM家族的生成前隐藏状态上训练线性探测器。我们的实证结果表明，细粒度的能力结构在所有模型中都可以以高准确率线性解码。然而，至关重要的是，我们揭示了表示与语言表达之间的差距：当要求这些模型用自然语言显式分类相同的查询时，它们的可靠性显著降低。这一脱节表明，关于所需外部能力的信息在LLM隐藏表示中是线性可访问的，但并未可靠地表达，这一现象我们称之为“结构化但沉默”。
cs.CL / 48 / 2610.08026
The Labeling Problem in Hallucination Detection Benchmarks: An Empirical Evaluation
幻觉检测基准中的标注问题：实证评估
Abstract
In recent years, several methods for detecting when large language models (LLMs) hallucinate have been developed. These methods are often benchmarked with open-domain question answering (QA) datasets containing questions and corresponding short reference answers. First, an LLM is used to generate answers to questions within the QA dataset. Then, some automated labeling strategy is used to label these answers as hallucinated or not by comparing them with the reference answers in the dataset. This evaluation setting creates a methodological ambiguity between two criteria: reference faithfulness (whether the answer is fully supported by the reference) and factual correctness (whether the answer is free from contradictions and factually false specific claims). In practice, automated labelers may apply the former criterion even when the intended target is the latter. We study this potential criterion mismatch using 900 human-labeled question-answer pairs spanning three commonly used QA datasets and three generator models, with labels targeting answer-level factual correctness. We evaluate lexical similarity metrics, a reference-entailment NLI baseline, and seven LLM judges under controlled prompt variants as automated labelers. Our experiments reveal substantial disagreement both among automated labeling strategies and between these labels and human annotations. Many strategies also exhibit strong directional error biases, and for most judge-generator pairs, replacing a faithfulness-oriented prompt with a factual-correctness prompt improves agreement with human annotations and reduces false-positive dominance, indicating that automated hallucination labels depend strongly on how the target criterion is specified. Label-source choice should therefore be considered a fundamental part of benchmark design and made explicit, validated, and matched with the benchmark goal.
Chinese Translation
近年来，已经开发出几种检测大型语言模型（LLMs）幻觉的方法。这些方法通常使用包含问题及其相应简短参考答案的开放领域问答（QA）数据集进行基准测试。首先，使用LLM生成QA数据集中问题的答案。然后，采用某种自动标注策略，通过将这些答案与数据集中的参考答案进行比较，标注这些答案为幻觉或非幻觉。这种评估设置在两个标准之间产生了方法论上的模糊性：参考忠实性（答案是否完全由参考支持）和事实正确性（答案是否没有矛盾且没有事实错误的具体主张）。在实践中，自动标注者可能会应用前者标准，即使其目标是后者。我们使用900对人工标注的问题-答案对，涵盖三个常用的QA数据集和三个生成模型，研究这种潜在的标准不匹配，标注目标为答案级别的事实正确性。我们评估了词汇相似性度量、参考蕴含NLI基线以及在受控提示变体下的七个LLM评审者作为自动标注者。我们的实验揭示了自动标注策略之间以及这些标签与人工注释之间的显著不一致。许多策略还表现出强烈的方向性错误偏差，对于大多数评审者-生成器对，将以忠实性为导向的提示替换为以事实正确性为导向的提示，提高了与人工注释的一致性，并减少了假阳性主导，这表明自动幻觉标签在很大程度上依赖于目标标准的指定方式。因此，标注源的选择应被视为基准设计的一个基本部分，并应明确、验证并与基准目标相匹配。
cs.CL / 49 / 2610.08037
Are Language Models Script-Aware?
语言模型是否具备脚本意识？
Abstract
Language models frequently generate outputs in unintended languages or scripts, a phenomenon known as off-target generation. While existing research has focused on language selection, the dimension of script knowledge remains understudied: before any linguistic understanding can occur, users must recognize the graphic symbols in a model's response. We investigate whether Small and Large Language Models (SLMs and LLMs) possess script knowledge by testing them on multi-scriptic languages. Through two complementary experiments, we evaluate whether models (1) adapt their output script to match the input, and (2) follow explicit instructions to generate text in a specified script. The models we tested demonstrate substantial script knowledge: they all achieve a near-perfect Latin script fidelity (more than 98%) and follow script instructions with high frequency. Nevertheless, we notice differences between LLMs and SLMs, with higher scores for LLMs including for non-standard script combinations.
Chinese Translation
语言模型经常在意想不到的语言或脚本中生成输出，这一现象被称为偏离目标生成。尽管现有研究主要集中在语言选择上，但脚本知识的维度仍然未得到充分研究：在任何语言理解发生之前，用户必须识别模型响应中的图形符号。我们通过在多脚本语言上测试小型和大型语言模型（SLMs 和 LLMs）来研究它们是否具备脚本知识。通过两个互补的实验，我们评估模型是否（1）调整其输出脚本以匹配输入，以及（2）遵循明确指示以在指定脚本中生成文本。我们测试的模型展示了显著的脚本知识：它们在拉丁脚本的保真度上均达到近乎完美（超过98%），并且高频率地遵循脚本指示。然而，我们注意到LLMs与SLMs之间存在差异，LLMs在包括非标准脚本组合的评分上更高。
cs.CL / 50 / 2610.08055
Language Carries the Expert's Impression: Instrument-Anchored LLM Judges Transfer Counseling-Quality Assessment and Beat In-Domain Training
Abstract
Automatic assessment of communication quality in dyadic counseling conversations is bottlenecked by data: expert-rated corpora are small and expensive to grow. We study cross-domain transfer of expert overall-impression prediction across three German corpora of simulated counseling (two general-practice medical, one school-related parent-teacher; $n=195$ expert-rated sessions, one corpus after scale equating). Training on the other domains beats training in-domain: leave-one-domain-out transfer reaches nested Spearman $\rho = 0.54$ against $\le 0.48$ within the target domain, a paired session-level gap of $+0.15$ that holds at $+0.12$ when the training-set sizes are matched, so it is not simply data volume. The decisive features are session-level construct scores from small open-weight LLMs reading the two-speaker transcript, with the constructs largely derived from the experts' rating instruments: the instrument-derived battery lifts a single judge from $0.32$ to $0.41$ over generic dialogue qualities, judges from three model families ensemble to $0.51$ language-only, and a nonverbal-dyadic block adds $+0.03$ more, not separable from noise at this sample size. We also price the recording setup: one corpus lost its per-speaker audio, 16% of its diarised segments carry the wrong speaker, and repair is worth $+0.07$ there. At practically attainable corpus sizes, the expert's overall impression is carried by what is said, and by other communication programs' data more than by one's own.
cs.CL / 51 / 2610.08085
DirectSpeech2LLM: A Simple End-to-End Framework to Mitigate Prompt Overfitting in Speech-LLMs
DirectSpeech2LLM：一种简单的端到端框架以减轻语音大语言模型中的提示过拟合
Abstract
Speech-LLMs often exhibit prompt overfitting, where models solely trained on automatic speech recognition (ASR) instruction fail to generalize to new instructions such as speech translation and continue to behave primarily as ASR system. We propose DirectSpeech2LLM, a simple end-to-end framework that preserves the instruction-following ability of the LLM on unseen tasks when conditioned on speech. It computes distance-based CTC loss over the frozen LLM embedding matrix and uses greedy CTC labels to derive geometrically and temporally aligned speech embeddings respectively as an input to the LLM. Trained solely on 960 hours of LibriSpeech ASR data, DirectSpeech2LLM outperforms the cascaded system on ASR (seen task) and generalizes zero-shot to speech translation and emotion recognition (two unseen tasks), closely matching the cascaded system upper bound on these two new instructions despite seeing neither during training. We also find that geometric alignment strength plays a smaller role than previously assumed, as our modified CTC loss is shown to provide sufficient implicit geometric grounding without requiring an explicit regression loss. Results are consistent across two LLM families and scale with both more training data and model capacity.
Chinese Translation
语音大语言模型（Speech-LLMs）常常表现出提示过拟合的现象，即仅在自动语音识别（ASR）指令上训练的模型无法推广到新的指令，如语音翻译，并且主要仍表现为ASR系统。我们提出了DirectSpeech2LLM，这是一种简单的端到端框架，在基于语音的条件下，能够保持LLM在未见任务上的指令跟随能力。该框架在冻结的LLM嵌入矩阵上计算基于距离的CTC损失，并使用贪婪CTC标签分别导出几何和时间对齐的语音嵌入作为LLM的输入。DirectSpeech2LLM仅在960小时的LibriSpeech ASR数据上训练，在ASR（已见任务）上超越了级联系统，并在语音翻译和情感识别（两个未见任务）上实现零-shot泛化，尽管在训练过程中未见到这两种新指令，但其表现接近级联系统的上限。我们还发现，几何对齐强度的作用小于之前的假设，因为我们的改进CTC损失显示出在不需要显式回归损失的情况下提供了足够的隐式几何基础。结果在两个LLM家族中保持一致，并随着更多训练数据和模型容量的增加而扩展。
cs.CL / 52 / 2610.08093
SAGE: Semantic Anchor-Guided Evolution for Grounded Medical QA Data Synthesis
SAGE：基于语义锚点引导的医学问答数据合成演化
Abstract
Developing reliable models for clinical tasks, such as Medical Question Answering (QA), is severely constrained by the limited availability of high-quality, expert-annotated training data. This challenge is exacerbated by stringent privacy requirements and the impracticality of utilizing large open-source corpora or proprietary cloud APIs within resource-limited clinical settings. To address these obstacles, we introduce SAGE (\textit{Semantic Anchor-Guided Evolution}), a novel data synthesis framework that enables small, locally deployed models to generate high-quality medical training data. SAGE leverages lightweight, publicly available taxonomies such as MeSH as semantic anchors, imposing a structured prior to effectively guide and ground the data generation process. At its core, SAGE iteratively interleaves atomic (individual concept-based) and associative (relation-based) synthesis, bootstrapping training data from minimal seeds. This approach eliminates the need for large collections of medical documents or reliance on external APIs, providing a practical solution for on-premises data creation. Extensive experiments across multiple medical question-answering benchmarks demonstrate that models fine-tuned with SAGE-synthesized data consistently outperform those trained using self-derived or conventional document-based paradigms, highlighting tangible improvements in data efficiency and resource utilization for medical LLM development. Code is available at https://github.com/DIaacKr/SAGE.
Chinese Translation
开发可靠的临床任务模型，例如医学问答（QA），受到高质量、专家标注的训练数据有限可用性的严重制约。这一挑战因严格的隐私要求以及在资源有限的临床环境中使用大型开源语料库或专有云API的不可行性而加剧。为了解决这些障碍，我们提出了SAGE（Semantic Anchor-Guided Evolution），一种新颖的数据合成框架，使小型本地部署模型能够生成高质量的医学训练数据。SAGE利用轻量级的、公开可用的分类法，如MeSH，作为语义锚点，施加结构化的先验，有效地引导和基础数据生成过程。在其核心，SAGE迭代地交错原子（基于个体概念）和关联（基于关系）合成，从最小种子引导训练数据。这种方法消除了对大量医学文档的需求或对外部API的依赖，为本地数据创建提供了实用的解决方案。在多个医学问答基准上的广泛实验表明，使用SAGE合成数据微调的模型在数据效率和资源利用方面始终优于使用自我衍生或传统文档基础范式训练的模型，突显了医学大语言模型开发中的实际改进。代码可在 https://github.com/DIaacKr/SAGE 获取。
cs.CL / 53 / 2610.08153
Penalty-Framed No-Valid-Option MCQA: Analyzing LLM Abstention under Invalid Choices
惩罚框架下的无有效选项多项选择问答：分析大语言模型在无效选择下的弃权行为
Abstract
Multiple-choice question answering (MCQA) is commonly used to evaluate large language models under the assumption that one of the provided options is correct, typically using answer-selection accuracy. However, in real deployments, users or retrieval systems may provide invalid option sets in which none of the listed choices is correct, and selecting one of them may incur downstream cost. We study this setting as penalty-framed no-valid-option MCQA. Using the mathematics subset of MMLU-Pro, we remove the labeled correct option, allow models to either choose a remaining option or output ABSTAIN, and penalize invalid forced-choice responses. We further introduce correct-conditioned analysis, evaluating abstention only on instances that the model originally answered correctly. Experiments show that high MCQA accuracy does not fully guarantee abstention reliability: even under explicit no-valid-option-aware instructions and penalty-based scoring, models still produce invalid forced-choice responses for a subset of originally correct instances. These results show that penalty-framed no-valid-option MCQA reveals an aspect of model reliability not captured by standard answer-selection accuracy.
Chinese Translation
多项选择问答（MCQA）通常用于评估大型语言模型，假设提供的选项中有一个是正确的，通常通过答案选择准确性来衡量。然而，在实际应用中，用户或检索系统可能会提供无效选项集，其中列出的选择均不正确，选择其中一个可能会带来后续成本。我们将这种情况研究为惩罚框架下的无有效选项MCQA。利用MMLU-Pro的数学子集，我们移除了标记的正确选项，允许模型选择剩余选项或输出弃权（ABSTAIN），并对无效的强制选择响应进行惩罚。我们进一步引入了正确条件分析，仅在模型最初回答正确的实例上评估弃权。实验表明，高MCQA准确性并不能完全保证弃权的可靠性：即使在明确的无有效选项意识指令和基于惩罚的评分下，模型仍然会对一部分原本正确的实例产生无效的强制选择响应。这些结果表明，惩罚框架下的无有效选项MCQA揭示了标准答案选择准确性未能捕捉到的模型可靠性的一方面。
cs.CL / 54 / 2610.08159
Making COMET Comparable Across Scripts: Diagnosis and Correction of Tokeniser-Induced Script Bias in Indic MT Evaluation
使 COMET 在不同书写系统间可比：印地语机器翻译评估中标记器引起的书写系统偏差的诊断与修正
Abstract
COMET reports translation quality as a single number, and that number is routinely compared across target languages written in different scripts. Such a comparison assumes Script Invariance: the score should not depend on the writing system that carries the target. We test it on IndicMT Eval by re-encoding the target into Latin script, which changes orthographic form while holding content and human ratings fixed. Script identity then accounts for 22.9% of native-script COMET variance, and agreement with annotators falls in all five languages studied. We trace the effect to the tokeniser and measure it with three label-free diagnostics. The bias is two faults, not one. Scores from different scripts occupy incompatible ranges, and within a single script the metric orders translations less accurately. No order-preserving transform of the score can repair the second fault. The first is removed exactly by COMET-QN, which maps the score distribution of each (language, script) pair onto a shared reference. Pooled agreement with annotators rises from 0.300 to 0.399, which is what makes scores from different scripts safe to place on one axis, and every within-language ordering is provably preserved. A regressor over parity features recovers a further 17.1% of the lost sensitivity. The remainder belongs to the encoder, and no post-processing can reach it. We therefore recommend publishing the normalised score, the three diagnostics, and the identity of the tokeniser they were computed against, so that a reader can tell how much of a score reflects translation quality and how much reflects the writing system.
Chinese Translation
COMET 将翻译质量报告为一个单一的数值，并且该数值通常在使用不同书写系统的目标语言之间进行比较。这种比较假设了书写系统不变性：得分不应依赖于承载目标的书写系统。我们通过将目标重新编码为拉丁字母来测试印地语机器翻译评估，这改变了正字法形式，同时保持内容和人工评分不变。书写系统的身份则占据了本地书写系统 COMET 方差的 22.9%，并且在研究的五种语言中与注释者的协议均有所下降。我们追踪这一影响源于标记器，并通过三种无标签的诊断方法进行测量。偏差是两个错误，而不是一个。来自不同书写系统的得分占据了不兼容的范围，而在单一书写系统内，该度量对翻译的排序准确性较低。没有任何保持顺序的变换可以修复第二个错误。第一个错误则通过 COMET-QN 被精确消除，该方法将每个（语言，书写系统）对的得分分布映射到一个共享的参考上。与注释者的汇总一致性从 0.300 上升到 0.399，这使得来自不同书写系统的得分可以安全地放置在同一轴上，并且每种语言内的排序都得到了证明的保留。一个关于对称特征的回归模型恢复了进一步 17.1% 的丧失敏感性。剩余部分归属于编码器，且没有后处理能够达到这一点。因此，我们建议发布标准化得分、三种诊断方法以及它们计算所依据的标记器的身份，以便读者能够判断得分中有多少反映翻译质量，多少反映书写系统。
cs.CL / 55 / 2610.08208
STRUCTURALCOST: A controlled reading time dataset for modeling human sentence processing difficulty
STRUCTURALCOST：用于建模人类句子处理难度的受控阅读时间数据集
Abstract
We introduce STRUCTURALCOST, a self-paced reading dataset of 475 participants and 40,800 observations isolating the processing cost of long-distance subject-verb dependency resolution. We replicate a low-powered psycholinguistic finding at NLP scale, namely that human reading times at the main verb increase with dependency length, driven by syntactic embedding beyond linear distance. Different language models -- spanning n-gram models, SSMs, and transformers -- partially mirror this graded difficulty profile, yet underestimate the integration cost humans incur, with a gap that persists across architectures and model sizes. This suggests these models capture the predictive component of human processing but not the full integration cost that working memory imposes. STRUCTURALCOST provides data needed to drive progress toward evaluating the cognitive plausibility of language models.
Chinese Translation
我们介绍了STRUCTURALCOST，这是一个包含475名参与者和40,800个观察结果的自我节奏阅读数据集，旨在隔离长距离主谓依赖解析的处理成本。我们在自然语言处理（NLP）规模上复制了一个低效能的心理语言学发现，即人类在主要动词处的阅读时间随着依赖长度的增加而增加，这一现象是由超越线性距离的句法嵌入所驱动。不同的语言模型——包括n-gram模型、状态空间模型（SSMs）和变换器（transformers）——部分反映了这一渐进的难度特征，但低估了人类所承受的整合成本，这一差距在不同架构和模型规模中持续存在。这表明这些模型捕捉了人类处理的预测成分，但未能完全反映工作记忆所施加的整合成本。STRUCTURALCOST提供了推动评估语言模型认知合理性所需的数据。
cs.CL / 56 / 2610.08300
Memory Depth and Reconstructed Context Width: A Controlled Evaluation of Hierarchical Retrieval
记忆深度与重构上下文宽度：分层检索的控制评估
Abstract
Long-term conversational memory is becoming an integral component of modern LLM systems. Proposed architectures group records by topics and events, construct hierarchies and graphs, and connect facts through causal and temporal relations. We experimentally study the interaction between two memory parameters: structural depth and the width of context supplied to the answer model. Using EverMemBench, we evaluate depths D1-D4, core budgets of 1,024/2,048/4,096 tokens, and additional Production and Oracle conditions up to the full archive. Increasing width from 1K to 4K improves Accuracy by 10.11-17.98 percentage points, whereas increasing depth provides no monotonic gain. Beyond 8-16K, Production performance reaches a plateau while tokens per correct answer continue to increase; Oracle preserves quality on full archives of 68-71K tokens. These results motivate further investigation of large, coherent context blocks instead of progressively deeper memory structures.
Chinese Translation
长期对话记忆正成为现代大型语言模型（LLM）系统的重要组成部分。提出的架构通过主题和事件对记录进行分组，构建层次结构和图形，并通过因果和时间关系连接事实。我们实验性地研究了两个记忆参数之间的相互作用：结构深度和提供给答案模型的上下文宽度。使用EverMemBench，我们评估了深度D1-D4、核心预算为1,024/2,048/4,096个标记，以及额外的生产和Oracle条件，直至完整档案。从1K增加到4K的宽度使准确率提高了10.11-17.98个百分点，而增加深度并未提供单调增益。在8-16K以上，生产性能达到平台期，而每个正确答案的标记数继续增加；Oracle在68-71K标记的完整档案上保持质量。这些结果激励我们进一步研究大规模、连贯的上下文块，而不是逐渐加深的记忆结构。
cs.CL / 57 / 2610.08303
Language Unalignability: Why Some Concepts Resist Cross-Cultural Benchmark Evaluation
语言的不对齐性：为何某些概念抵制跨文化基准评估
Abstract
Current evaluation of multilingual Large Language Models (LLMs) rests on an implicit Translation-Isomorphism Assumption (TIA): that semantic structures across languages are congruent and mutually mappable without loss of information. We argue that this assumption is not merely violated in practice, but ill-posed in principle for a typologically identifiable class of concepts, including pragmatic markers, honorifics, and diachronically stratified terms. We formalize this failure using a usage-cloud framework, representing concepts as point sets of contextualized embeddings. We define $\alpha$-unalignability as the impossibility of any mapping that simultaneously preserves lexical faithfulness (centroid correspondence) and structural faithfulness (local neighborhood topology). We provide three layers of evidence. Behaviorally, we show that FLORES-200 translation failures are predicted by language family and resource class but not by script, and that LOBSTER reasoning scores vary by family. Mechanistically, we report a Representation-Intervention Gap (RIG) in a nine-model case study on Yami: the models' activations encode a regularity along which Yami groups with other low-resource and Austronesian languages, yet interventions on language-specific neurons show no demonstrated advantage over random masks: the regularity is visible but not usable by this intervention. Finally, we operationalize these findings into a multidimensional diagnostic profile: Cycle-Consistency, Pragmatic-Load Disagreement, Manifold-Curvature Mismatch, and RIG. We argue that collapsing cultural competence into a single scalar incentivizes "probabilistic flattening," and that recognizing the unalignable class is a precondition for AI that respects, rather than erases, cultural divergence. This suggests that multilingual alignment is not a single well-defined objective, but a set of mutually incompatible projections.
Chinese Translation
当前对多语言大型语言模型（LLMs）的评估基于一个隐含的翻译同构假设（Translation-Isomorphism Assumption, TIA）：即不同语言之间的语义结构是一致的，并且可以相互映射而不丢失信息。我们认为，这一假设不仅在实践中被违反，而且在原则上对于一类可识别的类型学概念（包括语用标记、敬语和历时分层术语）是不适当的。我们通过使用使用云框架（usage-cloud framework）来形式化这一失败，将概念表示为上下文化嵌入的点集。我们定义了$eta$-不对齐性（unalignability），即不可能存在任何同时保持词汇忠实性（质心对应）和结构忠实性（局部邻域拓扑）的映射。我们提供了三层证据。从行为上看，我们展示了FLORES-200翻译失败的预测因素是语言家族和资源类别，而不是书写系统，并且LOBSTER推理分数因家族而异。从机制上看，我们在对Yami的九模型案例研究中报告了表示干预差距（Representation-Intervention Gap, RIG）：模型的激活编码了一种规律，Yami与其他低资源和南岛语言群体相似，但对特定语言神经元的干预并未显示出相对于随机掩码的优势：这一规律是可见的，但通过这种干预并不可用。最后，我们将这些发现操作化为一个多维诊断特征：循环一致性（Cycle-Consistency）、语用负载不一致（Pragmatic-Load Disagreement）、流形曲率不匹配（Manifold-Curvature Mismatch）和RIG。我们认为，将文化能力简化为单一标量会激励“概率平坦化”，而识别不对齐的类别是尊重而非抹去文化差异的人工智能的前提。这表明，多语言对齐并不是一个单一的明确定义的目标，而是一组相互不兼容的投影。
cs.CL / 58 / 2610.08388
Foresight-over-Graph: Reasoning Beyond Local Horizons for Knowledge Base Question Answering
图上前瞻性推理：超越局部视野的知识库问答
Abstract
Large language models (LLMs) have demonstrated strong capabilities in question answering, yet they still frequently suffer from hallucinations on knowledge-intensive tasks. Knowledge graphs (KGs) provide LLMs with structured, interpretable, and updatable factual grounding, making them a promising external knowledge source for reliable reasoning. However, existing LLM-guided graph reasoning methods typically rely on hop-wise greedy or beam-style pruning during evidence retrieval. Such local decision processes are inherently myopic: evidence that appears weak near the source may become crucial only after deeper graph context is explored, causing answer-critical branches to be discarded prematurely and making the reasoning chain difficult to recover. To address this limitation, we propose Foresight-over-Graph (FoG), a foresight-aware evidence retrieval framework for knowledge base question answering (KBQA). FoG iteratively constructs a question-relevant evidence subgraph and uses far-to-near feedback to guide path exploration, and maintains a compact memory subgraph to support continued exploration. Extensive experiments on widely used KBQA benchmarks demonstrate that FoG achieves state-of-the-art performance, with a particularly large improvement of 16.58% in Hit on CWQ, while also reducing LLM calls and token usage. Our code is available at https://github.com/yhong7/FoG .
Chinese Translation
大型语言模型（LLMs）在问答任务中表现出强大的能力，但在知识密集型任务上仍然经常出现幻觉。知识图谱（KGs）为LLMs提供了结构化、可解释和可更新的事实基础，使其成为可靠推理的有前景的外部知识来源。然而，现有的LLM引导的图推理方法通常依赖于逐跳贪婪或束搜索风格的剪枝来进行证据检索。这种局部决策过程本质上是短视的：在源附近看似微弱的证据，可能在深入探索图的上下文后变得至关重要，从而导致关键答案分支被过早丢弃，使得推理链难以恢复。为了解决这一局限性，我们提出了图上前瞻性推理（Foresight-over-Graph，FoG），这是一种具有前瞻意识的证据检索框架，用于知识库问答（KBQA）。FoG迭代地构建与问题相关的证据子图，并利用远到近的反馈来引导路径探索，同时保持一个紧凑的记忆子图以支持持续探索。在广泛使用的KBQA基准上的大量实验表明，FoG实现了最先进的性能，特别是在CWQ上提高了16.58%的命中率，同时减少了LLM调用和令牌使用。我们的代码可在https://github.com/yhong7/FoG获取。
cs.CL / 59 / 2610.08413
Knowing When Not to Answer: Cross-Domain and Multi-Turn Generalization of Latent Underspecification Signals
知道何时不回答：潜在不充分信号的跨领域和多轮泛化
Abstract
Large language models routinely answer questions that cannot be answered from the information given, and in dialogue they answer before enough has been said. Unanswerability is linearly decodable from hidden states, but it is unclear which of its forms share a representation and whether the signal is useful in dialogue. We contribute a turn-labeled multi-turn benchmark (423 conversations, 1,661 labeled turn-states) and an evaluation harness with a simulated user who answers clarifying questions, and use them with six datasets and six open-weight LLMs to test how far probes for unanswerability carry. Probes transfer robustly between datasets that share a ground of unanswerability: missing information in math (AUROC 0.77-0.97) and in a passage (SQuAD 2.0<->MuSiQue, 0.77-0.90). Probes for epistemic "known-unknowns" transfer poorly to math, but this separation weakens under lexical controls and changes with layer and coordinate system, so it remains unresolved. Single-turn probes fail zero-shot to detect when a conversation becomes answerable; in-structure probes recover it, but no better than a bag-of-words classifier. A gate on the calibrated probe, with no model fine-tuning, fires on underspecified turns far more precisely than chance, and its end-task success comes within 0.08 of a gate given the true labels. Yet across four models it does not reliably beat vanilla generation or prompted consolidation. The remaining gap lies mostly in how models use a clarification, not in detection.
Chinese Translation
大型语言模型通常会回答那些无法从给定信息中得出答案的问题，并且在对话中，它们在信息尚未充分表达之前就给出了回答。不可回答性可以从隐藏状态中线性解码，但尚不清楚其形式中哪些共享表示，以及该信号在对话中是否有用。我们贡献了一个标记轮次的多轮基准（423个对话，1,661个标记轮次状态）和一个评估工具，配备一个回答澄清问题的模拟用户，并将其与六个数据集和六个开放权重的LLM（大型语言模型）一起使用，以测试对不可回答性的探测能力的传递性。探测在共享不可回答性基础的数据集之间稳健传递：在数学中缺失信息（AUROC 0.77-0.97）和在段落中（SQuAD 2.0<->MuSiQue，0.77-0.90）。对认识论“已知未知”的探测在数学中传递较差，但这种分离在词汇控制下减弱，并随着层和坐标系统的变化而变化，因此仍未解决。单轮探测在零样本情况下无法检测对话何时变得可回答；结构内探测能够恢复这一点，但效果不比词袋分类器更好。经过校准的探测器在不进行模型微调的情况下，对不充分的轮次的识别远比随机更为精准，其最终任务的成功率与给定真实标签的探测器相差仅0.08。然而，在四个模型中，它并未可靠地超越普通生成或提示整合。剩余的差距主要在于模型如何使用澄清，而不是在检测能力上。
cs.CL / 60 / 2610.08448
Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability
重新思考跨标记器的在线蒸馏：从对齐覆盖到监督可靠性
Abstract
On-Policy Distillation (OPD) trains a student on its own generations using teacher feedback. With different tokenizers, comparing teacher and student predictions requires alignment at both sequence and vocabulary levels. In this paper, we examine whether expanding this alignment coverage improves learning. Across three heterogeneous teacher--student pairs on mathematical reasoning and code generation, strict 1:1 groups already cover most student-generated tokens despite substantial vocabulary mismatch. On responses sampled from the students before distillation, the shared vocabulary retains nearly all teacher and student probability mass at strictly aligned positions on average. Restricting reverse KL to a student-selected top-16 subset of the shared vocabulary at each strict position achieves accuracy comparable to full shared-vocabulary OPD, outperforming the evaluated cross-tokenizer baselines. Adding mean squared error supervision on span log-probabilities in mismatch groups gives complete supervision coverage, yet reduces accuracy. At checkpoints from training with only the strict loss, the span gradients show weak or negative directional agreement with the strict gradients and grow in magnitude relative to them. These diagnostics may help explain the accuracy drop from adding span supervision. Our findings motivate a shift from maximizing alignment coverage to prioritizing supervision reliability: compact supervision at strict positions can be more effective than broader coverage that introduces weakly aligned or conflicting training signals.
Chinese Translation
在线蒸馏（On-Policy Distillation，OPD）通过教师反馈训练学生模型，使其基于自身生成的内容进行学习。使用不同的标记器时，比较教师和学生的预测需要在序列和词汇层面上进行对齐。本文探讨了扩展这种对齐覆盖是否能改善学习效果。在三个异构的教师-学生对（涉及数学推理和代码生成）中，严格的1:1组已经覆盖了大多数学生生成的标记，尽管存在显著的词汇不匹配。在蒸馏前从学生那里抽样的响应中，共享词汇在严格对齐的位置上平均保留了几乎所有教师和学生的概率质量。将反向KL散度限制在学生选择的共享词汇的前16个子集上，在每个严格位置上实现的准确性与完整共享词汇的OPD相当，且优于评估的跨标记器基线。在不匹配组中，对跨度对数概率的均方误差监督提供了完整的监督覆盖，但却降低了准确性。在仅使用严格损失进行训练的检查点中，跨度梯度与严格梯度的方向一致性较弱或为负，并且相对于严格梯度的幅度有所增加。这些诊断可能有助于解释添加跨度监督后准确性的下降。我们的研究结果促使我们从最大化对齐覆盖转向优先考虑监督可靠性：在严格位置上的紧凑监督可能比引入弱对齐或冲突训练信号的更广泛覆盖更为有效。
cs.CL / 61 / 2610.08452
Agentic AutoRAG: RAG Pipeline Optimization through Reasoning-Driven Agents
自主代理的自动RAG：通过推理驱动的代理优化RAG管道
Abstract
Retrieval-augmented generation (RAG) is a widely used approach for grounding large language models (LLMs) in external knowledge. However, configuring a pipeline is an expensive hyperparameter optimization problem over many interacting choices, from chunking and embedding model to reranking and generation. Existing optimizers, from greedy search to Bayesian optimization, reduce each trial to an aggregate score and search without modeling why a configuration performed as it did, even though the retrieved chunks already provide evidence about whether each failure occurred during retrieval or after it. We introduce Agentic AutoRAG, an LLM-agent optimizer for multi-objective RAG hyperparameter optimization with retrieval-versus-generation failure attribution. It proposes configurations scored on a frozen exam from the corpus: after each trial a Diagnoser attributes each failed question to retrieval or generation, and a Proposer, grounded in a knowledge base of model rankings and pricing, selects the next configuration, weighing accuracy against cost to trace a Pareto frontier. On three multi-hop QA benchmarks it reaches higher LLM-judge accuracy than every baseline we compare, and within its first 10 trials it matches or beats the statistical baselines' full 30-trial judge accuracy. In its cost-aware mode on a real-world healthcare corpus it reaches a median exam accuracy of 77%, above the strongest baseline's 71.5%, at about 58% of that baseline's cost per query, and it matches that 71.5% at about 22% of the cost.
Chinese Translation
检索增强生成（RAG）是一种广泛使用的方法，用于将大型语言模型（LLMs）与外部知识相结合。然而，配置一个管道是一个昂贵的超参数优化问题，涉及许多相互作用的选择，从分块和嵌入模型到重新排序和生成。现有的优化器，从贪婪搜索到贝叶斯优化，将每次试验简化为一个综合得分，并在没有建模配置为何表现如此的情况下进行搜索，尽管检索到的块已经提供了关于每个失败是在检索过程中还是之后发生的证据。我们提出了自主代理的自动RAG（Agentic AutoRAG），这是一个用于多目标RAG超参数优化的LLM代理优化器，具备检索与生成失败归因功能。它提出的配置基于语料库中的冻结考试得分：在每次试验后，诊断器（Diagnoser）将每个失败的问题归因于检索或生成，而提议者（Proposer）则基于模型排名和定价的知识库选择下一个配置，权衡准确性与成本，以追踪帕累托前沿。在三个多跳问答基准测试中，其LLM评估准确性超过了我们比较的每个基线，并且在其前10次试验中，其准确性与统计基线的完整30次试验评估准确性相匹配或超越。在其关注成本的模式下，在一个真实世界的医疗保健语料库上，其中位考试准确性达到77%，高于最强基线的71.5%，且成本约为该基线每查询的58%，同时在约22%的成本下匹配该71.5%的准确性。
cs.CL / 62 / 2610.08463
UNREAL: Unifying Retrieval and Long-Context with a Single Model
UNREAL：通过单一模型统一检索与长上下文
Abstract
Long-context inference and Retrieval-Augmented Generation (RAG) handle evidence selection at vastly different scales, from a single long prompt to an entire corpus. We ask whether a single model-internal mechanism can select evidence across this range. We introduce UNifying REtrieval And Long-Context with a Single Model (UNREAL), a model-native evidence selection framework to span corpus retrieval and long-context inference. UNREAL encodes chunks and derives retrieval queries directly from the frozen LLM's internal representations. It adds fewer than 500K trainable parameters and leaves the backbone unchanged. On a 3B-token, 21M-chunk Wikipedia index, all four dense and hybrid UNREAL backbones outperform state-of-the-art retriever-reranker systems. The best model raises recall from 49.1% to 73.2% on HotpotQA, from 31.7% to 60.1% on 2WikiMultiHopQA, and from 8.8% to 14.4% on MuSiQue. Applied to long-context tasks, the same selection mechanism removes distractors before generation, raising NoLiMa accuracy from 1.0% to 24.83% at its maximum context length of 128K tokens, and LV-Eval's F1 score from 49.97% to 54.66% at 256K. UNREAL also reduces FLOPs and time-to-first-token relative to full-context inference from roughly 32K tokens onward, with larger gains as context grows. Together, these results establish model-internal evidence selection as a common foundation for corpus retrieval and evidence-sparse long-context inference.
Chinese Translation
长上下文推理和检索增强生成（RAG）在证据选择上处理着截然不同的规模，从单个长提示到整个语料库。我们探讨是否可以通过单一的模型内部机制在这一范围内选择证据。我们提出了统一检索与长上下文的单一模型（UNREAL），这是一个原生的证据选择框架，旨在覆盖语料库检索和长上下文推理。UNREAL 直接从冻结的语言模型（LLM）的内部表示中编码块并推导检索查询。它增加了不到 500K 的可训练参数，并保持了主干不变。在一个包含 3B 令牌和 21M 块的维基百科索引上，所有四个密集和混合的 UNREAL 主干在性能上超越了最先进的检索器-重排系统。最佳模型在 HotpotQA 上将召回率从 49.1% 提高到 73.2%，在 2WikiMultiHopQA 上从 31.7% 提高到 60.1%，在 MuSiQue 上从 8.8% 提高到 14.4%。在长上下文任务中，使用相同的选择机制在生成之前移除干扰项，使 NoLiMa 的准确率从 1.0% 提高到 24.83%（最大上下文长度为 128K 令牌），并使 LV-Eval 的 F1 分数从 49.97% 提高到 54.66%（在 256K 时）。UNREAL 还相对于完整上下文推理在大约 32K 令牌以上减少了 FLOPs 和首次令牌的生成时间，随着上下文的增长，增益更为显著。这些结果共同确立了模型内部证据选择作为语料库检索和证据稀疏长上下文推理的共同基础。
cs.CL / 63 / 2610.08501
Language-model ratings of depression reflect the rater more than the patient
语言模型对抑郁症的评分更多反映评分者而非患者
Abstract
Depression has no diagnostic blood test. Language models promise tireless, consistent assessment, but can accurate raters disagree about individuals? We pre-registered 880 language-model raters, crossing 11 open models with prompting and scoring choices, and applied them to 189 interviews against the eight-item Patient Health Questionnaire. Model choice explained 30.0% of summed-symptom score variance, stable participant differences 10.5%. Two randomly drawn raters with area under the receiver operating characteristic curve (AUC) >= 0.70 disagreed on screening decisions for 40% of participants, on average. Average over-rating governed how many were flagged, yet equal-capacity raters chose differently for about one participant in five. A locked analysis of 86 new interviews reproduced the main pre-registered findings. Exploratory recalibration with 40 labelled participants raised accuracy from about 60% to 75% and halved disagreement, leaving one participant in five decided differently. Calibration repaired much of the rater dependence without securing agreement about individuals.
Chinese Translation
抑郁症没有诊断性血液检测。语言模型承诺提供不知疲倦且一致的评估，但准确的评分者能否在个体上达成一致？我们预注册了880名语言模型评分者，结合11个开放模型的提示和评分选择，并将其应用于189个访谈，依据八项患者健康问卷（Patient Health Questionnaire）。模型选择解释了总症状评分方差的30.0%，稳定的参与者差异为10.5%。两名随机抽取的评分者，其接收者操作特征曲线下面积（AUC）>= 0.70，平均在40%的参与者筛查决策上存在分歧。平均过度评分决定了被标记的数量，但能力相当的评分者在大约五分之一的参与者上选择不同。对86个新访谈的锁定分析重现了主要的预注册发现。对40名标记参与者的探索性重新校准将准确率从约60%提高到75%，并将分歧减半，仍有五分之一的参与者的决策不同。校准修复了大部分评分者依赖性，但未能确保对个体的一致意见。
cs.CL / 64 / 2610.08513
Wiki-Talkie: Multilingual Benchmarking of Persona-Based Agents on Real-World Discussions
Wiki-Talkie：基于真实讨论的个性化代理的多语言基准测试
Abstract
LLMs are increasingly deployed as autonomous agents in social environments, making it critical to study their ability to faithfully simulate human interactions. Central to this is grounding agents in realistic user personas, yet existing datasets rely on fictional personas and are limited to a handful of languages, lacking the empirical grounding necessary to evaluate behavioral fidelity across diverse populations. We introduce Wiki-Talkie, a multilingual dataset of real-world conversations from Wikipedia Talk pages across five languages spanning two language families: Germanic (German, English) and Romance (Spanish, French, Italian), paired with personas derived from real user communities and encompassing sociodemographic attributes, self-descriptions, and behaviorally grounded interaction traits. Using Wiki-Talkie, we evaluate agent interactional behavior on a next-turn generation task across various persona conditioning strategies. Our evaluation assesses whether agents collectively reproduce the distributional behavioral patterns observed in human discussions. Results show that user's comment history exemplifying interaction behavior consistently outperforms explicit persona information. In addition, models systematically underproduce negative or extreme sentiments, while over producing references and suggestions, revealing biases toward agreeableness and positivity. Crucially, these patterns hold robustly across languages, with small cross-lingual differences.
Chinese Translation
大型语言模型（LLMs）越来越多地作为自主代理在社交环境中部署，因此研究它们忠实模拟人类互动的能力变得至关重要。关键在于将代理与现实用户个性相结合，而现有的数据集依赖于虚构的个性，并且仅限于少数几种语言，缺乏评估不同人群行为忠实度所需的实证基础。我们引入了Wiki-Talkie，这是一个来自维基百科讨论页面的多语言真实对话数据集，涵盖五种语言，跨越两个语言家族：日耳曼语（德语、英语）和罗曼语（西班牙语、法语、意大利语），并配有来自真实用户社区的个性，涵盖社会人口特征、自我描述和行为基础的互动特征。利用Wiki-Talkie，我们评估了代理在各种个性条件策略下的下一轮生成任务中的互动行为。我们的评估检验了代理是否能够共同再现人类讨论中观察到的分布性行为模式。结果表明，用户的评论历史所体现的互动行为始终优于显式个性信息。此外，模型系统性地低估了负面或极端情绪的生成，同时过度生成参考和建议，揭示出对友好性和积极性的偏见。重要的是，这些模式在不同语言中稳健存在，且跨语言差异较小。
cs.CL / 65 / 2610.08544
How High Is 0.6? Floors, Ceilings, and Headroom in Interpretability Probing
0.6有多高？可解释性探测中的下限、上限与余地
Abstract
Probes are the workhorse of interpretability. If a model's hidden states predict a variable, the model is said to represent it. But a probe score has no fixed meaning. An $R^2$ of 0.6 may only reflect what the input already gives away, and the same score can mean different things on different data. We propose reading every probe score against two reference points: a floor, what a declared set of simple inputs already predicts, and a ceiling, what the full input can predict. The gap between them, the headroom, is the range in which a probe can show that a model computes something beyond the simple inputs. We prove that headroom vanishes in two ways: the target stops depending on a hidden variable the model must infer, or the input stops revealing it. We test this on transformers trained for in-context meta-analysis, which must infer the hidden heterogeneity between studies to weight them correctly, and where both reference points are known. Under distribution shift, probe scores fall and prediction error rises $12$--$15\times$, yet the model recovers a similar share of the headroom, indicating that the data lost information, not the representation. We then analyze the real models. The single-cell foundation model scGPT encodes biological variability only partially. We also revisit four influential LLM probing studies, which claim that models represent geography, the state of an Othello board, truth, and the demographics of their users. Against a floor computed from the input text alone, some of these claims hold, while others are largely explained by the text itself.
Chinese Translation
探测器是可解释性的主要工具。如果一个模型的隐藏状态能够预测某个变量，则该模型被认为能够表示该变量。然而，探测器得分并没有固定的含义。$R^2$为0.6可能仅反映了输入已经透露的信息，而同样的得分在不同的数据上可能意味着不同的事情。我们建议将每个探测器得分与两个参考点进行比较：下限，即一组声明的简单输入所能预测的内容；上限，即完整输入所能预测的内容。它们之间的差距，即余地，是探测器能够显示模型计算超出简单输入的内容的范围。我们证明了余地以两种方式消失：目标不再依赖于模型必须推断的隐藏变量，或者输入不再揭示该变量。我们在为上下文元分析训练的变换器上进行了测试，这些变换器必须推断研究之间的隐藏异质性，以正确加权，并且两个参考点都是已知的。在分布转移下，探测器得分下降，预测误差上升12到15倍，但模型恢复了类似的余地份额，表明数据丢失了信息，而不是表示。随后，我们分析了真实模型。单细胞基础模型scGPT仅部分编码生物变异性。我们还重新审视了四项有影响力的LLM探测研究，这些研究声称模型表示地理、奥赛罗棋盘的状态、真理以及用户的人口统计特征。与仅从输入文本计算的下限相比，这些声称中的一些成立，而其他则在很大程度上可以通过文本本身来解释。
cs.CL / 66 / 2610.08559
Latent space bias directions in LLMs capture confidence, not fairness
大型语言模型中的潜在空间偏差方向捕捉信心，而非公平性
Abstract
Activation steering has gained popularity as a lightweight inference-time debiasing technique for large language models. However, prior work reports that steering vectors generalise poorly, with unintended effects on model performance and limited transfer to new datasets. Our work analyses what the debiasing direction used for activation steering actually encodes, in order to shed light on its inconsistent performance. We study the linear debiasing direction obtained by contrasting the activations of anti-biased and biased prompts, and evaluate it as a steering intervention across bias and general knowledge benchmarks. We find that this direction is dominated by model confidence, pointing from regions of high to low-probability tokens in activation space rather than encoding a meaningful representation of model bias. Steering along it does reduce measured bias, but this is a consequence of reducing model confidence: on QA benchmarks we find that this steering drives the model to abstain from answering, with a side effect of improving fairness metrics. Our experiments show that model confidence is the dominant separating factor between biased and anti-biased prompts in hidden space, indicating that isolating a linear representation of bias which is disentangled from model confidence is difficult and steering-based debiasing results should be interpreted with care. In short, steering appears to reduce bias, not by correcting the model's underlying preferences, but by making it less confident, even on tasks unrelated to bias.
Chinese Translation
激活引导作为一种轻量级的推理时去偏技术在大型语言模型中日益受到关注。然而，先前的研究报告指出，引导向量的泛化能力较差，对模型性能产生了意想不到的影响，并且在新数据集上的迁移能力有限。我们的工作分析了用于激活引导的去偏方向实际编码的内容，以阐明其不一致的表现。我们研究了通过对比反偏见和偏见提示的激活所获得的线性去偏方向，并在偏见和一般知识基准上评估其作为引导干预的效果。我们发现，这一方向主要由模型信心主导，指向激活空间中从高概率到低概率标记的区域，而非编码模型偏见的有意义表示。沿此方向引导确实减少了测量到的偏见，但这是由于降低了模型信心的结果：在问答基准上，我们发现这种引导使模型倾向于不回答，副作用是提高了公平性指标。我们的实验表明，模型信心是隐藏空间中偏见和反偏见提示之间的主要区分因素，这表明将偏见的线性表示与模型信心解耦是困难的，因此基于引导的去偏结果应谨慎解读。简而言之，引导似乎通过降低模型的信心来减少偏见，而不是纠正模型的潜在偏好，即使在与偏见无关的任务上也是如此。
cs.CL / 67 / 2610.08585
Incidental information contaminates patient notes and disrupts clinical reasoning in large language models
偶然信息污染患者记录并干扰大型语言模型的临床推理
Abstract
Large language models (LLMs) are increasingly relied upon to support ambient documentation and clinical reasoning. Here we examine the impact of a failure mode shared between these two applications by assessing their sensitivity to information incidental to the patient encounter. In 576 patient-clinician dialogues, we found that frontier models inserted small-talk exchanges into 35% of notes, while mean quality scores changed by at most 0.20 points on five-point scales. In 3.7% of frontier notes, models misattributed the asides or used them clinically. In 57 mock recorded consultations, background speech from a separate patient encounter at -10 dB leaked into 48.2% of transcripts, with contamination detected in 5.3% of downstream notes generated by four open-weight models. We propose a dual encoding hypothesis of clinical reasoning and distraction in LLMs, with preliminary evidence that LLM components associated with disruption by incidental information also support clinical reasoning. These findings support evaluating resistance to incidental information before clinical use, with safeguards that prevent contamination while preserving clinical reasoning.
Chinese Translation
大型语言模型（LLMs）越来越多地被用于支持环境文档记录和临床推理。本文通过评估这些应用对与患者接触无关信息的敏感性，研究了它们之间共享的失效模式的影响。在576个患者-临床医生对话中，我们发现前沿模型在35%的记录中插入了闲聊内容，而平均质量评分在五分制中最多变化了0.20分。在3.7%的前沿记录中，模型错误地归因于这些闲聊内容或将其用于临床。在57个模拟录音咨询中，来自另一个患者接触的背景语音在-10 dB的情况下泄漏到48.2%的转录中，且在四个开放权重模型生成的下游记录中检测到5.3%的污染。我们提出了一个临床推理和干扰的双重编码假设，并初步证据表明，与偶然信息干扰相关的LLM组件也支持临床推理。这些发现支持在临床使用前评估对偶然信息的抵抗力，并提出防止污染同时保持临床推理的保障措施。
cs.CL / 68 / 2610.08601
Generative AI translations in high-stakes emergency messaging
高风险紧急消息中的生成式人工智能翻译
Abstract
Emergency messaging such as extreme-weather reports and earthquake instructions can involve high stakes, to the extent that translation errors can lead to tragic consequences. The use of machine translation or generative artificial intelligence might therefore not be recommended. On the other hand, time savings in the initial translation can allow greater investments of resources in revision and authorization processes, as well as a wider range of target languages. An experiment with generative AI translations of an earthquake instruction text from English into Chinese and Spanish shows that use of discourse-specific prompts can considerably improve understandability and actionability, although the translations may still not be trusted by translators. Human revision is still required, not only to detect errors but also because of the ethical need for someone to take responsibility for any errors or delays in such messaging.
Chinese Translation
紧急消息，如极端天气报告和地震指示，涉及高风险，翻译错误可能导致悲惨后果。因此，可能不推荐使用机器翻译或生成式人工智能。另一方面，初步翻译中的时间节省可以使更多资源投入到修订和授权流程中，并支持更广泛的目标语言。一项将地震指示文本从英语翻译成中文和西班牙语的生成式人工智能翻译实验表明，使用特定话语的提示可以显著提高可理解性和可操作性，尽管翻译仍可能不被翻译人员信任。仍然需要人工修订，不仅是为了发现错误，还因为在此类消息中，出于伦理需要，必须有人对任何错误或延误负责。
cs.CL / 69 / 2610.08604
InterCorrect: Intersection-Aware Correction of Demographic Model Merging for Fair ASR
InterCorrect：面向交叉群体的公平自动语音识别的人口模型合并校正
Abstract
Automatic Speech Recognition (ASR) systems often show uneven performance across demographic groups, and errors can be especially difficult to address for speakers belonging to multiple demographic groups. This work studies demographic-aware model merging for fair Speech-LLM-based ASR. Starting from a SLAM-ASR-based model, we fine-tune only the connector on demographic-specific subsets and merge the resulting subgroup-adapted connectors into a global model. We then identify critical cross-axis demographic pairs using subgroup WER and task-vector conflict, and apply intersection-specific correction vectors to the global merged model. Experiments on Fair-Speech show that global demographic merging improves overall WER over the base model, while intersection correction provides additional gains for several merging strategies. In particular, TIES with WER-based correction achieves the best overall WER, reducing it from 7.38\% to 5.13\%. Subgroup and disparity analyses further show that the proposed approach improves performance across demographic axes, while highlighting that lower average WER does not always imply reduced subgroup disparity.
Chinese Translation
自动语音识别（ASR）系统在不同人口群体之间的表现往往不均衡，对于属于多个人口群体的说话者，错误尤其难以处理。本研究探讨了面向人口的模型合并，以实现基于Speech-LLM的公平ASR。从基于SLAM-ASR的模型出发，我们仅在特定于人口的子集上微调连接器，并将得到的子群适配连接器合并为一个全局模型。然后，我们使用子群的字错误率（WER）和任务向量冲突识别关键的交叉人口对，并将特定于交叉的校正向量应用于全局合并模型。在Fair-Speech上的实验表明，全局人口合并在基准模型上提高了整体WER，而交叉校正为多种合并策略提供了额外的增益。特别是，基于WER的校正的TIES策略实现了最佳的整体WER，将其从7.38%降低到5.13%。子群和差异分析进一步表明，所提出的方法在各人口维度上改善了性能，同时强调较低的平均WER并不总是意味着减少子群差异。
cs.CL / 70 / 2610.08630
Towards In-Parameter Memory Augmentation for Large Language Models
面向大语言模型的参数内存增强
Abstract
Recently Large Language Models (LLMs) and LLM-based agents increasingly need to incorporate knowledge acquired after pretraining, e.g., domain facts, user preferences, documents, and interaction experience. In-context learning (ICL) and ICL-based agent harness remain flexible, but they consume context capacity and incur repeated discretized encoding cost that grows with context length. \textbf{In-parameter memory} offers a complementary substrate: reusable memory information is represented in model parameters, adapters, or other parameter-like objects that are composed into the forward pass at inference time. This survey focuses on methods that augment LLMs with such parametric memory at deployment: a memory-bearing parameter object is plugged into the forward pass during inference, whether it is acquired before or during deployment. We organize the landscape with two orthogonal axes: \textbf{Parameter Placement}, which includes Embedding, Attention, FFN layers, or Hybrid when two or more layers are used; and \textbf{Parameter Acquisition Time}, which distinguishes methods whose memory object is acquired during deployment (online) from those acquired before it (offline). We clarify boundaries, conduct comparisons, and discuss open directions in interference, safety, co-design with ICL, and recursive self-improvement.
Chinese Translation
近年来，大语言模型（LLMs）及基于LLM的智能体越来越需要整合在预训练后获得的知识，例如领域事实、用户偏好、文档和交互经验。上下文学习（ICL）及基于ICL的智能体保持灵活性，但它们消耗上下文容量，并产生随着上下文长度增长的重复离散编码成本。**参数内存**提供了一种互补的基础：可重用的记忆信息以模型参数、适配器或其他类似参数的对象形式表示，这些对象在推理时被组合到前向传播中。本调查聚焦于在部署时通过这种参数化记忆增强LLMs的方法：一个带有记忆的参数对象在推理过程中被插入到前向传播中，无论它是在部署前还是部署期间获得的。我们通过两个正交轴组织这一领域：**参数位置**，包括嵌入层、注意力层、前馈网络（FFN）层，或当使用两个或多个层时的混合；以及**参数获取时间**，区分在部署期间（在线）获得记忆对象的方法与在此之前（离线）获得的方法。我们澄清边界，进行比较，并讨论干扰、安全性、与ICL的协同设计以及递归自我改进的开放方向。
cs.CL / 71 / 2610.08660
Evidence-Bound Reasoning: Neuro-Semantic Verification of Biomedical AI in Glioblastoma Radiogenomics
证据绑定推理：胶质母细胞瘤放射基因组学中生物医学人工智能的神经语义验证
Abstract
Background: Biomedical AI can generate plausible explanations without reliably verifying whether each statement is supported by patient-specific evidence. We developed a neuro-semantic verification framework that converts radiomic measurements into addressable evidence records and machine-checkable claims. Methods: UPenn-GBM radiomics were aligned with de novo CaPTk extraction from standardized MRI and expert-validated segmentations in an independent multicenter cohort. The shared space comprised 1,728 features from T1, T1GD, T2, and FLAIR MRI across three tumor regions. Reference-defined semantic states were derived from 611 UPenn cases. We evaluated cross-cohort transportability, model-linked provenance, deterministic verification, controlled predictive degradation, and an LLM claim-extraction pilot; MGMT prediction served only as a transport stress test. Results: Median semantic-state agreement was 0.786 (weighted kappa 0.709), ranging from 0.918 for morphologic to 0.252 for intensity features. The external evidence ledger contained 1,655 model-linked records for 331 patients. The verifier achieved 100% exact-set accuracy in a 6,620-claim corruption benchmark. In a 24-case pilot, GPT-5.6 Sol reproduced 72/72 prespecified atomic claims, and the frozen verifier recovered 24/24 expected conditions. During controlled degradation, ROC AUC declined from 0.899 to 0.500 while verification accuracy remained 1.000. External MGMT discrimination was weak (ROC AUC 0.543). Conclusions: Verifiability can be engineered and evaluated independently of predictive performance. LLMs may structure explanations, while final evidence-consistency checking remains deterministic.
Chinese Translation
背景：生物医学人工智能可以生成合理的解释，但无法可靠地验证每个陈述是否得到了特定患者证据的支持。我们开发了一种神经语义验证框架，将放射组学测量转换为可寻址的证据记录和机器可验证的声明。方法：UPenn-GBM放射组学与来自标准化MRI的de novo CaPTk提取和在独立多中心队列中经过专家验证的分割对齐。共享空间包含来自三种肿瘤区域的T1、T1GD、T2和FLAIR MRI的1,728个特征。参考定义的语义状态来源于611个UPenn病例。我们评估了跨队列的可转移性、模型关联的来源、确定性验证、受控预测降级，以及一个LLM声明提取试点；MGMT预测仅作为转移压力测试。结果：中位语义状态一致性为0.786（加权kappa为0.709），在形态特征中为0.918，在强度特征中为0.252。外部证据账本包含331名患者的1,655个模型关联记录。验证器在6,620个声明腐败基准中实现了100%的精确集准确率。在24个案例的试点中，GPT-5.6 Sol再现了72/72个预先指定的原子声明，而冻结的验证器恢复了24/24个预期条件。在受控降级过程中，ROC AUC从0.899降至0.500，而验证准确率保持在1.000。外部MGMT区分能力较弱（ROC AUC 0.543）。结论：可验证性可以独立于预测性能进行工程设计和评估。LLMs可能会构建解释，而最终的证据一致性检查仍然是确定性的。
cs.CL / 72 / 2610.08675
Same-Number Citation Swaps: Stress-Testing Jev as a Financial Evidence Judge
同数引文交换：作为金融证据判断者的Jev的压力测试
Abstract
Financial reports repeat values across periods, metrics and accounting lines, allowing an LLM-generated calculation to be numerically correct while citing the wrong financial role. We evaluate what probabilistic evidence verification adds beyond number matching using Jev as a source-support verifier for GPT-4.1-mini calculation traces. A signed-number-at-pointer baseline explains most recovery over exact quotation checks. To isolate the remaining role-recognition problem, we hold operands and arithmetic fixed, move citations between same-number cells, and retain controls that express equivalent facts. These contrasts reveal both wrong-role citations that pass and valid alternative citations that are withheld. Explicit column labels improve selected wrong-role decisions while also lowering support for some equivalent evidence. A constructed follow-up on 36 new source pages, labeled by a non-author reviewer, extends this evaluation and exposes the same tradeoff between detecting role errors and retaining valid citations. The contribution is a controlled evaluation that identifies what a probabilistic financial verifier distinguishes when numerical matching is held fixed. For LLM-based financial assistants, it makes numerical correctness, cited-role support and acceptance outcomes separately assessable.
Chinese Translation
财务报告在不同期间、指标和会计科目中重复数值，这使得基于大型语言模型（LLM）生成的计算在数值上是正确的，但引用了错误的财务角色。我们评估了在使用Jev作为GPT-4.1-mini计算轨迹的源支持验证者时，概率证据验证在数字匹配之外所增加的价值。带符号的数值基线解释了大部分在精确引用检查中的恢复情况。为了隔离剩余的角色识别问题，我们固定操作数和算术，将引用在同数单元之间移动，并保留表达等效事实的控制变量。这些对比揭示了既有通过的错误角色引用，也有被抑制的有效替代引用。明确的列标签改善了选定错误角色决策，同时也降低了一些等效证据的支持。对36个新源页面的后续构建，由非作者审阅者标记，扩展了这一评估，并揭示了检测角色错误与保留有效引用之间的相同权衡。该贡献是一项控制评估，识别了在固定数值匹配时，概率财务验证者所区分的内容。对于基于LLM的财务助手，它使数值正确性、引用角色支持和接受结果可以单独评估。
cs.CL / 73 / 2610.08703
Agreement Is Not Validity: Cross-Model LLM Consensus in Diagnosing Student Failure Modes in K-12 Math Tutoring Dialogue
一致性并非有效性：跨模型大型语言模型在K-12数学辅导对话中诊断学生失败模式的共识
Abstract
In K-12 mathematics tutoring, student-tutor dialogue provides rich evidence of learners' problem-solving processes and sources of difficulty. Learning analytics research increasingly relies on large language models (LLMs) to extract such information from dialogue for a variety of downstream tasks, including knowledge tracing, behavioral modeling, and diagnosis of student reasoning errors. However, the validity of these model-generated interpretations remains insufficiently understood. In this exploratory study, we examine the validity of LLM classifications of five student failure modes in mathematics tutoring dialogue using an operational diagnostic codebook: uncertainty, misattribution, operator selection, conceptual gap, and procedural slip. Across models, human-LLM agreement was moderate (kappa = .524-.597), while cross-model agreement was substantially higher (kappa = .755-.781; alpha = .769). These findings show that cross-model agreement can create a misleading appearance of correctness, challenging the assumption that consensus among LLMs constitutes evidence of valid learner interpretation. For learning analytics, the implication is clear: scalable labeling is useful only if the inferred constructs are valid, and model consensus cannot substitute for independent evidence of that validity.
Chinese Translation
在K-12数学辅导中，学生与辅导员的对话提供了丰富的证据，反映了学习者的问题解决过程和困难来源。学习分析研究越来越依赖大型语言模型（LLMs）从对话中提取这些信息，以支持多种下游任务，包括知识追踪、行为建模和学生推理错误的诊断。然而，这些模型生成的解释的有效性仍然不够明确。在这项探索性研究中，我们使用操作性诊断编码手册，考察了LLM对数学辅导对话中五种学生失败模式的分类有效性：不确定性、错误归因、操作符选择、概念差距和程序性失误。在不同模型之间，人类与LLM的一致性为中等水平（kappa = .524-.597），而跨模型一致性则显著更高（kappa = .755-.781；alpha = .769）。这些发现表明，跨模型一致性可能会产生误导性的正确性假象，挑战了LLM之间的共识构成有效学习者解释证据的假设。对于学习分析而言，含义明确：可扩展的标注只有在推断的构念有效时才有用，而模型共识不能替代独立的有效性证据。
cs.CL / 74 / 2610.08718
When Forgetting is not Catastrophic: On the Mechanics of Spurious Forgetting
当遗忘不是灾难性时：虚假遗忘的机制
Abstract
Knowledge that a language model appears to forget during finetuning often remains stored and can be recovered, a phenomenon called spurious forgetting. Finetuning on new facts can even produce forgetting that undoes itself: recall of the old facts collapses, recovers as training continues on new facts alone, and only then erodes for good. We seek to understand when such forgetting is not catastrophic. A minimal associative memory reproduces these dynamics with three ingredients: keys with shared structure, concentrated new values, and normalization in the network. Finetuning moves all old representations along a common direction, hiding the old facts while preserving their relative geometry; normalization withdraws this shift once the new facts are learned, whereas fact-specific changes accumulate and cause the erosion. Moreover, subtracting the common shift eliminates the collapse in a Transformer trained on synthetic data, and removing a single direction from each weight update restores old facts in a pretrained language model. Forgetting thus combines a shared, reversible loss of access with a slow erosion of individual facts, and only the second is catastrophic. Which one dominates depends on whether the new data move old memories together or apart.
Chinese Translation
语言模型在微调过程中似乎遗忘的知识往往仍然被存储并可以恢复，这种现象称为虚假遗忘。在新事实的微调过程中，甚至可能产生自我逆转的遗忘：旧事实的回忆崩溃，随着仅在新事实上继续训练而恢复，最终才会彻底消失。我们试图理解何时这种遗忘不是灾难性的。一个最小的关联记忆通过三个要素再现了这些动态：具有共享结构的键、集中新的值以及网络中的归一化。微调使所有旧表示沿着一个共同方向移动，隐藏旧事实，同时保持其相对几何；一旦新事实被学习，归一化就会撤回这种移动，而特定于事实的变化则会积累并导致侵蚀。此外，在一个基于合成数据训练的Transformer中，减去共同的移动消除了崩溃，而从每次权重更新中移除一个方向则恢复了预训练语言模型中的旧事实。因此，遗忘结合了共享的、可逆的访问丧失与个别事实的缓慢侵蚀，只有后者是灾难性的。哪种情况占主导地位取决于新数据是将旧记忆一起移动还是分开。
cs.CL / 75 / 2610.08719
Holdout Best-of-N: Unbiased Evaluation and Its Cost
保留最佳的N：无偏评估及其成本
Abstract
Reusing the scores that select a Best-of-$N$ winner can overstate its expected reward. We study evaluation from a fixed matrix of $K$ independent scores per candidate for a policy that selects using $J$ fresh scores. A single estimator based only on this matrix is exactly unbiased for expected judge reward under every independent, stable collection of candidate-specific score laws if and only if $J
Chinese Translation
重复使用选择最佳的N（Best-of-$N$）获胜者的评分可能会高估其预期奖励。我们研究了对于一个使用$J$个新评分进行选择的策略，从每个候选者的固定矩阵中获取$K$个独立评分的评估。如果且仅当$J
cs.CL / 76 / 2610.08738
Denoising Hierarchical Representations: Joint Continuous Diffusion for Language Modeling
去噪层次表示：语言建模的联合连续扩散
Abstract
Diffusion Language Models (DLMs) hold the promise of order-agnostic, parallel text generation. Recently, continuous diffusion and flow matching models have seen substantial gains, driven by carefully crafted token representations and diffusion/flow spaces. In this work, we introduce Hierarchical Continuous Diffusion Language Models (H-CDLMs), a simple framework that further improves continuous DLMs with minimal compute and parameter overhead. Drawing on the discrete DLM and continuous image diffusion literature on joint diffusion, we diffuse multiple modalities in parallel. These modalities represent tokens at different semantic granularities: in our instantiation, the tokens themselves and coarser clusters obtained by clustering pretrained token embeddings. We propose a general setup that allows per-modality samplers and schedules to enhance the interplay between modalities. Applied to CoBit, this yields H-CoBit, which delivers large empirical gains across benchmarks. At dataset entropy, H-CoBit improves MAUVE and reaches a generative perplexity (GenPPL) of 49.4 on LM1B and 50.4 on OWT, improving on the baseline by 24.2 and 20.7 points and surpassing even discrete DLMs of comparable size. On GSM8K, it reaches 27.4% accuracy, outperforming prior continuous diffusion and flow-based models. We further apply H-CDLM to the flow matching model FLM, obtaining consistent gains with H-FLM and demonstrating that the framework generalizes across continuous generative paradigms. Our code will be made publicly available at https://github.com/matol-16/HCDLM.git .
Chinese Translation
扩散语言模型（DLMs）承诺实现无序、并行的文本生成。最近，连续扩散和流匹配模型在精心设计的标记表示和扩散/流空间的推动下取得了显著进展。在本研究中，我们引入了层次连续扩散语言模型（H-CDLMs），这是一个简单的框架，能够在计算和参数开销最小的情况下进一步改善连续DLMs。我们借鉴了离散DLM和连续图像扩散文献中的联合扩散，能够并行扩散多种模态。这些模态代表了不同语义粒度的标记：在我们的实例中，标记本身以及通过聚类预训练标记嵌入获得的更粗粒度的聚类。我们提出了一种通用设置，允许每种模态的采样器和调度，以增强模态之间的相互作用。应用于CoBit，这产生了H-CoBit，在基准测试中实现了显著的经验性提升。在数据集熵下，H-CoBit提高了MAUVE，并在LM1B上达到了49.4的生成困惑度（GenPPL），在OWT上达到了50.4，分别比基线提高了24.2和20.7点，甚至超越了同等规模的离散DLMs。在GSM8K上，它达到了27.4%的准确率，优于之前的连续扩散和基于流的模型。我们进一步将H-CDLM应用于流匹配模型FLM，获得了一致的提升，形成了H-FLM，并证明该框架在连续生成范式中具有广泛的适用性。我们的代码将公开发布在https://github.com/matol-16/HCDLM.git。
cs.CL / 77 / 2610.08747
The Missing Minimal Pair: Stereotype Evaluation in LLMs
缺失的最小对：大型语言模型中的刻板印象评估
Abstract
A common approach to measuring bias in Large Language Models is to compare the log-likelihoods of two contrastive stereotype sentences. We argue that such single-pair comparisons are often unreliable: simply rewriting the same stereotype with an alternative attribute can yield logically inconsistent preferences. To address this, we propose a dual minimal pair setup that introduces two axes of comparison for robust stereotype evaluation. First, we present a data-augmentation framework that fills critical gaps in existing stereotype datasets by generating paraphrases and alternate attributes. We apply our framework on a set of English, Russian, Spanish and Chinese stereotypes. Second, we introduce two evaluation metrics tailored to the dual minimal pair setup. One of these metrics provides a new perspective on bias by modeling the mutual information (MI) between social groups and stereotyped attributes. This MI-based metric is better suited for aggregation and enables more robust comparisons of stereotype strength across different languages and models. Our code is available at https://github.com/stepanat/missing-minimal-pair/.
Chinese Translation
测量大型语言模型偏见的常见方法是比较两个对比刻板印象句子的对数似然值。我们认为，这种单对比较往往不可靠：仅仅用替代属性重写相同的刻板印象可能会导致逻辑上不一致的偏好。为了解决这个问题，我们提出了一种双最小对设置，引入两个比较轴以进行稳健的刻板印象评估。首先，我们提出了一个数据增强框架，通过生成同义句和替代属性来填补现有刻板印象数据集中存在的关键空白。我们在一组英语、俄语、西班牙语和中文的刻板印象上应用了我们的框架。其次，我们引入了两种针对双最小对设置的评估指标。其中一种指标通过建模社会群体与刻板印象属性之间的互信息（MI），为偏见提供了一种新的视角。基于MI的指标更适合进行聚合，并能够在不同语言和模型之间进行更稳健的刻板印象强度比较。我们的代码可在 https://github.com/stepanat/missing-minimal-pair/ 获取。
cs.CL / 78 / 2610.08773
AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model
AdvSim2Real：在网络世界模型中训练网络代理以应对自适应提示注入
Abstract
Web agents complete user requests by reading and acting on pages that third parties write, so an instruction planted on a page can redirect the agent away from the user's goal. The agent cannot simply ignore the page, because the page also holds the values and controls the task requires. Current defenses fine-tune the agent on injections fixed before training, and attackers that adapt to the trained model bypass them. Adversarial training lets the attacker adapt but keeps the tasks fixed, so a task stops teaching once the agent solves it. We introduce AdvSim2Real, which co-evolves a task curriculum, an injection adversary, and the agent inside a frozen web world model. The curriculum is rewarded for tasks the agent solves about half of the time, and the adversary only for a success flip, an injection that turns a judged success into a failure. Training in the simulator makes a 4B agent both more capable and more robust: its completion rises with and without attacks, holds against a frontier-model adversary it never trained against, and its capability gain carries over to a real browser. On 150 web tasks, AdvSim2Real raises completion under this unseen adversary by 33.6\% relative to the base agent.
Chinese Translation
网络代理通过读取和执行第三方撰写的页面来完成用户请求，因此页面上植入的指令可能会将代理引导偏离用户的目标。代理无法简单地忽略页面，因为页面还包含任务所需的值和控制。当前的防御方法是在训练前对固定的注入进行微调，而适应训练模型的攻击者可以绕过这些防御。对抗训练允许攻击者进行适应，但保持任务不变，因此一旦代理解决了任务，任务就停止教学。我们提出了AdvSim2Real，它在一个冻结的网络世界模型中共同演化任务课程、注入对手和代理。课程对代理解决约一半时间的任务给予奖励，而对手则仅在成功翻转时获得奖励，即将被评判为成功的注入转变为失败。在模拟器中的训练使得一个4B的代理在能力和鲁棒性上都有所提升：其完成率在有无攻击情况下均有所上升，能够抵御一个从未训练过的前沿模型对手，并且其能力提升能够转移到真实浏览器上。在150个网络任务中，AdvSim2Real在面对这一未见对手时，相较于基础代理提高了33.6%的完成率。
cs.CL / 79 / 2610.08781
IdeaAnchor: Teaching LLMs to Turn Literature into Research Ideas
IdeaAnchor：教会大型语言模型将文献转化为研究创意
Abstract
Scientific research often begins by synthesizing ideas from a set of related papers to identify gaps and formulate new directions. However, training language models to perform this form of literature-grounded ideation remains challenging, as existing approaches based on prompting or feedback lack structured supervision for how papers should be synthesized. We introduce IdeaAnchor, a paradigm for training LLMs to perform research ideation using structured specifications as privileged signals. Each IdeaAnchor instance encodes how each input paper should be synthesized into a successful idea, including their functional roles, relationships, and target synthesis criteria. We build this paradigm by mining instances from published papers, capturing how real ideas emerge from prior literature. We then train models via demonstration, self-distillation, and reinforcement learning, and further enhance generation with retrieval at inference time. Experiments show consistent improvements in ideation quality. Our analysis reveals a functional decomposition: anchor-based training strengthens creative synthesis, retrieval enhances detail elaboration, and combining both yields the best performance.
Chinese Translation
科学研究通常始于从一组相关论文中综合想法，以识别研究空白并制定新的研究方向。然而，训练语言模型执行这种基于文献的创意生成仍然具有挑战性，因为现有基于提示或反馈的方法缺乏对论文如何被综合的结构化监督。我们提出了IdeaAnchor，这是一种训练大型语言模型（LLMs）使用结构化规范作为优先信号进行研究创意生成的范式。每个IdeaAnchor实例编码了如何将每篇输入论文综合成一个成功的创意，包括它们的功能角色、关系和目标综合标准。我们通过从已发表的论文中挖掘实例来构建这一范式，捕捉真实创意如何从先前文献中涌现。然后，我们通过示范、自蒸馏和强化学习训练模型，并在推理时通过检索进一步增强生成效果。实验表明创意生成质量的一致性提升。我们的分析揭示了一种功能分解：基于锚点的训练增强了创造性综合，检索增强了细节阐述，而两者结合则产生了最佳性能。