cs.RO / 1 / 2609.38202
Optimize, Learn, Refine: Whole-Body Grasping and Pick-and-Throw with a Spiral Soft Robot
优化、学习、精炼：螺旋软机器人全身抓取与投掷
Abstract
Soft continuum robots can exploit distributed compliance for whole-body manipulation, but synthesizing behavior through changing contacts remains difficult. We address whole-body grasping and pick-and-throw from an initially ungrasped state through outcome-based actuation-space optimization. Grasping is quantified by tip angular sweep and body-object enclosure, while throwing further incorporates release-direction alignment and minimum release speed. These objectives allow grasping, acceleration, and release to emerge from compliant interaction without prescribing contact forces, contact locations, or body configurations. Because the resulting actuation-to-outcome mapping is nonsmooth, we utilize derivative-free CMA-ES within an optimize-learn-refine framework. CMA-ES generates solutions for sampled conditions, a task-conditioned predictor learns warm starts, and CMA-ES refines them for unseen conditions. In simulation, the method achieves 492/500 successful grasps (98.4%) and success rates of 98%, 97%, and 94% across three directional throwing trials. Learned initialization increases grasping success from 78.6% to 98.4% while reducing the median rollout count from 1184 to 816 in CMA-ES. Hardware experiments achieve a 100% grasping success rate across 50 executions and a 100% pick-and-throw success rate across 30 executions, with 10 repetitions per direction. Together, these simulation and hardware results demonstrate the effectiveness of the proposed framework across both simulated and physical whole-body manipulation tasks.
Chinese Translation
软连续机器人可以利用分布式顺应性进行全身操作，但通过改变接触来合成行为仍然困难。我们通过基于结果的驱动空间优化，解决从初始未抓取状态的全身抓取和投掷问题。抓取通过尖端角度扫掠和物体包围量化，而投掷则进一步结合释放方向对齐和最小释放速度。这些目标使得抓取、加速和释放能够通过顺应性交互自发产生，而无需规定接触力、接触位置或身体配置。由于得到的驱动与结果映射是非光滑的，我们在优化-学习-精炼框架内利用无导数的 CMA-ES。CMA-ES 为采样条件生成解决方案，任务条件预测器学习热启动，CMA-ES 则为未见条件进行精炼。在仿真中，该方法实现了492/500次成功抓取（98.4%），并在三次方向投掷试验中分别达到了98%、97%和94%的成功率。学习的初始化将抓取成功率从78.6%提高到98.4%，同时将 CMA-ES 的中位数展开次数从1184减少到816。硬件实验在50次执行中实现了100%的抓取成功率，在30次执行中实现了100%的投掷成功率，每个方向重复10次。这些仿真和硬件结果共同证明了所提出框架在模拟和物理全身操作任务中的有效性。
cs.RO / 2 / 2609.38216
Fiatlux: A Long-Horizon Benchmark for Humanoid Ladder Climbing and Light-Bulb Replacement
Fiatlux：一个用于类人机器人梯子攀爬和灯泡更换的长时间基准测试
Abstract
Existing benchmarks evaluate tabletop manipulation, flat-floor household activity, or humanoid locomotion and manipulation as separate task groups; none scores vertical mobility and dexterous work on a fragile payload in one long-horizon episode. We present Fiatlux, a light-bulb replacement benchmark built on NVIDIA Isaac Lab. In one episode, a Unitree G1 humanoid positions a step ladder under a ceiling or wall fixture, climbs it, exchanges a spent bulb in a socket for a fresh one, and leaves the spent one in a disposal crate. We decompose the episode into twelve subtask environments scored on difficulty-weighted gates. The goal is a successful replacement, with the fresh bulb seated, the spent one disposed of, neither dropped, and a fragility bound not crossed. Runs that fall short can earn partial credit. Observations are split into a standard mode (signals a physical robot could sense or estimate) and a privileged mode (exact simulator state). We specify the evaluation protocol and provide reference baseline implementations spanning RSL-RL PPO, zero-shot NVIDIA GR00T N1.7 Vision-Language-Action (VLA) models, and whole-body controllers. Additionally, we provide the teleoperated recordings used to specify and check the success gates. The benchmark code and the teleoperated recordings are available at fiatlux-bench.github.io.
Chinese Translation
现有的基准测试评估桌面操作、平地家庭活动或类人机器人运动与操作作为独立的任务组；没有一个能够在一个长时间的情节中同时评估垂直移动性和对脆弱负载的灵巧操作。我们提出了Fiatlux，这是一个基于NVIDIA Isaac Lab构建的灯泡更换基准测试。在一个情节中，一台Unitree G1类人机器人将一架梯子放置在天花板或墙壁的灯具下，攀爬上去，将一个用完的灯泡从插座中更换为一个新的灯泡，并将用完的灯泡放入废物箱中。我们将该情节分解为十二个子任务环境，并根据难度加权的门进行评分。目标是成功更换灯泡，确保新灯泡安装到位，用完的灯泡被妥善处理，且两者均未掉落，同时未超过脆弱性界限。未能完成的尝试可以获得部分积分。观察分为标准模式（物理机器人可以感知或估计的信号）和特权模式（精确的模拟器状态）。我们指定了评估协议，并提供了涵盖RSL-RL PPO、零-shot NVIDIA GR00T N1.7视觉-语言-动作（VLA）模型和全身控制器的参考基线实现。此外，我们还提供了用于指定和检查成功门的遥控记录。基准测试代码和遥控记录可在fiatlux-bench.github.io获取。
cs.RO / 3 / 2609.38225
SynIL: Leveraging Synergy for Offline Imitation Learning from Imperfect Demonstration Datasets
SynIL：利用协同效应进行离线模仿学习以应对不完美的示范数据集
Abstract
Imitation learning enables robots to acquire complex skills directly from massive demonstration datasets, but its performance degrades severely when datasets are contaminated with suboptimal or noisy demonstrations. While prior quality-assessment methods attempt to filter or reweight data, they typically rely on manual pre-selection of expert reference data or task-specific heuristics, limiting scalability. To address this challenge, we introduce SynIL (Synergy-based Imitation Learning), a novel framework for automated, label-free demonstration quality assessment in offline reinforcement learning. Grounded in neuroscientific evidence that motor synergy, a low-dimensional coordinated structure in movement, correlates directly with motor proficiency, SynIL algorithmically quantifies synergy manifestation to generate dense, transition-level reward signals via self-supervised reward regression. Comprehensive evaluations on D4RL locomotion benchmarks and multi-human Robomimic manipulation datasets demonstrate that synergy-derived rewards correlate strongly with ground-truth rewards. Furthermore, SynIL substantially outperforms Behavior Cloning (BC) and achieves performance comparable to, and in sparse-reward human teleoperation scenarios, superior to, offline reinforcement learning trained on true environment rewards.
Chinese Translation
模仿学习使机器人能够直接从大量示范数据集中获取复杂技能，但当数据集受到次优或噪声示范的污染时，其性能会严重下降。虽然先前的质量评估方法试图过滤或重新加权数据，但它们通常依赖于专家参考数据的手动预选或特定任务的启发式方法，限制了可扩展性。为了解决这一挑战，我们提出了SynIL（基于协同效应的模仿学习），这是一个用于离线强化学习中自动化、无标签示范质量评估的新框架。基于神经科学证据，运动协同效应是一种低维度的运动协调结构，与运动能力直接相关，SynIL算法通过自监督奖励回归量化协同效应的表现，以生成密集的、过渡级别的奖励信号。在D4RL运动基准和多人的Robomimic操作数据集上的全面评估表明，基于协同效应的奖励与真实奖励之间存在很强的相关性。此外，SynIL在性能上显著优于行为克隆（Behavior Cloning, BC），并在稀疏奖励的人类遥操作场景中，其表现与基于真实环境奖励的离线强化学习相当，甚至更优。
cs.RO / 4 / 2609.38227
A Two-Echelon Covering Tour Vehicle Routing Problem with Drones for Post-Disaster Relief
灾后救援的双层覆盖巡回车辆路径问题及无人机应用
Abstract
We introduce the two-echelon covering tour vehicle routing problem (2E-CTVRP) for the distribution of relief supplies after a disaster. In the first echelon, a fleet of trucks transports supplies and drones from a central depot to satellites at the periphery of the affected area. In the second echelon, drones launched in parallel from the satellites deliver the supplies to the centroids of victim clusters, which are obtained by clustering the victim locations, and each truck waits at a satellite until its drones have returned. The problem combines the assignment of satellites to trucks, the sequencing of the truck routes, and the assignment of clusters to satellites, and minimizes the sum of the arrival times of the trucks at the satellites and at the depot. We formulate the 2E-CTVRP as a mixed integer linear program and propose a hybrid metaheuristic, GRASP-ILS-PR, which combines greedy randomized construction, a single-trajectory search, and periodic path relinking on the assignment of clusters to satellites. On a new benchmark set of 110 instances, GRASP-ILS-PR finds all known optimal solutions within 30 seconds, matches or improves on the one-hour solutions of a commercial solver on most larger instances, and improves on them by up to 11% for instances with 50 satellites. It clearly outperforms a conventional GRASP with evolutionary path relinking and yields more consistent results than a simulated annealing algorithm that uses the same search components. An analysis of two fleet configurations shows that few large trucks minimize the cumulative arrival time in most instances, whereas many small trucks deliver supplies to the victims earlier and more equitably in every instance.
Chinese Translation
我们提出了双层覆盖巡回车辆路径问题（2E-CTVRP），用于灾后救援物资的分配。在第一层级，一队卡车将物资和无人机从中央仓库运输到受灾区域边缘的卫星点。在第二层级，从卫星点并行发射的无人机将物资送达受害者聚集点的重心，这些重心是通过对受害者位置进行聚类获得的，每辆卡车在卫星点等待，直到其无人机返回。该问题结合了卫星分配给卡车、卡车路线的排序以及聚类分配给卫星的任务，并最小化卡车到达卫星点和仓库的到达时间总和。我们将2E-CTVRP形式化为一个混合整数线性规划，并提出了一种混合元启发式算法GRASP-ILS-PR，该算法结合了贪婪随机构造、单轨迹搜索和周期性路径重链接，用于聚类分配给卫星的任务。在一个新的110个实例的基准测试集中，GRASP-ILS-PR在30秒内找到了所有已知的最优解，在大多数较大实例中与商业求解器的一小时解相匹配或有所改善，并在50个卫星的实例中提高了多达11%。它明显优于传统的GRASP与进化路径重链接，并且比使用相同搜索组件的模拟退火算法产生了更一致的结果。对两种车队配置的分析表明，在大多数实例中，少量大型卡车能够最小化累积到达时间，而许多小型卡车则能够在每个实例中更早且更公平地将物资送达受害者。
cs.RO / 5 / 2609.38371
TALK-Dem: Benchmarking Embodied Task Planning under Dementia-Associated Communication Patterns
TALK-Dem：基于痴呆相关沟通模式的具身任务规划基准测试
Abstract
Existing LLM-driven robot task planners rely on a taken-for-granted assumption of an ideal user whose instructions are clear, complete, and task-focused. However, when interacting with real-world users, especially those experiencing cognitive impairments, such as people living with dementia (PLWD), the planners often make mistakes and even pose physical safety risks. We proposed TALK-Dem (Talking Attributes and Linguistic Knowledge in Dementia), the first benchmark for evaluating LLM-driven robot task planning under dementia-associated verbal communication. TALK-Dem contains 4,800 instructions and covers five typical communication patterns, including Referential Imprecision, Object Substitution, Empty Speech, Topic Drift, and Intrusion, at three intensity levels. Experiments across six open-weight LLMs reveal a substantial robustness gap. Across communication patterns, open-weight models exhibited performance drops of up to 22.3 percentage points compared to ideal instructions. This revealed a critical gap and even danger for real-world applications, especially in assistive robotics, where locally deployable models are necessary due to privacy concerns and connectivity constraints. To mitigate this issue, we proposed the Context-Aware Retrieval from Experience (CARE) method, which retrieves relevant previously resolved tasks to provide task-specific interpretation and planning context. CARE generally outperformed standard prompting baselines across the six open-weight models, improving average task success by 18.1 percentage points over the vanilla prompt. These results highlighted the importance of both evaluating communication robustness and developing effective adaptation strategies for locally deployable assistive robots. The TALK-Dem dataset is publicly available at https://anonymous.4open.science/r/TALK-Dem-A6B3/.
Chinese Translation
现有的基于大型语言模型（LLM）的机器人任务规划器依赖于一个理想用户的假设，即其指令清晰、完整且以任务为中心。然而，在与现实世界用户互动时，尤其是那些经历认知障碍的用户，如生活在痴呆中的人（PLWD），这些规划器往往会出错，甚至可能带来身体安全风险。我们提出了TALK-Dem（痴呆中的谈话属性和语言知识），这是第一个用于评估基于LLM的机器人任务规划在痴呆相关语言沟通下的基准。TALK-Dem包含4800条指令，涵盖五种典型的沟通模式，包括指称不精确、对象替代、空洞言语、话题漂移和干扰，分为三个强度级别。针对六个开放权重的LLM进行的实验揭示了显著的鲁棒性差距。在沟通模式中，开放权重模型的表现相比理想指令下降了多达22.3个百分点。这揭示了一个关键的差距，甚至在现实应用中存在危险，尤其是在辅助机器人领域，由于隐私问题和连接限制，需要本地可部署的模型。为了解决这个问题，我们提出了基于经验的上下文感知检索（CARE）方法，该方法检索相关的先前解决的任务，以提供特定任务的解释和规划上下文。CARE在六个开放权重模型中普遍优于标准提示基线，平均任务成功率比普通提示提高了18.1个百分点。这些结果突显了评估沟通鲁棒性和为本地可部署的辅助机器人开发有效适应策略的重要性。TALK-Dem数据集已在https://anonymous.4open.science/r/TALK-Dem-A6B3/上公开发布。
cs.RO / 6 / 2609.38400
GestAdapt: Workspace-Conditioned Co-Speech Gesture Generation for Humanoid Robots
GestAdapt：基于工作空间条件的人形机器人共语手势生成
Abstract
Co-speech gestures for robots must adapt not only to speech and embodiment, but also to the workspace available for performing the motion. Since the same speech can be accompanied by different gestures, a robot can respond to workspace constraints, e.g., gestures for speech next to a wall. In these scenarios, the robot should gesture in a suitable motion rather than simply correcting an unconstrained one. To achieve this goal, we present GestAdapt, a workspace-conditioned framework that conditions co-speech gesture generation on a prescribed wrist workspace. The GestAdapt framework learns from six complementary co-speech corpora through a shared motion representation and supports retargeting to different robot embodiments. Quantitative evaluation shows that generated motions remain close to the real-motion distribution while respecting the workspace. In a user study, gestures generated under modified workspace constraints receive a mean quality score of 3.24/5, above our no-workspace variant (2.43/5) and below the reference motions (3.68/5). In a real robot evaluation, all compared motions are retargeted to the Reachy2 humanoid robot under identical workspace constraints. Motions generated with our framework rank first in 69.7\% of comparisons, higher than our no-workspace variant baseline and retargeted ground-truth motions constrained afterward. Overall, the results support adapting gestures to the available workspace during generation, rather than modifying unconstrained trajectories afterward to satisfy workspace constraints, potentially compromising gesture naturalness.
Chinese Translation
机器人的共语手势不仅必须适应语言和具身性，还必须适应执行动作时可用的工作空间。由于相同的语言可以伴随不同的手势，机器人可以根据工作空间的限制作出反应，例如，在墙边的语言手势。在这些场景中，机器人应以适当的动作进行手势，而不仅仅是纠正一个不受限制的手势。为实现这一目标，我们提出了GestAdapt，一个基于工作空间条件的框架，该框架将共语手势生成与规定的手腕工作空间相结合。GestAdapt框架通过共享运动表示从六个互补的共语语料库中学习，并支持不同机器人具身性的重定向。定量评估表明，生成的动作在尊重工作空间的同时，仍然接近真实运动分布。在用户研究中，在修改的工作空间限制下生成的手势的平均质量评分为3.24/5，高于我们的无工作空间变体（2.43/5），但低于参考动作（3.68/5）。在真实机器人评估中，所有比较的动作都在相同的工作空间限制下重定向到Reachy2人形机器人。使用我们框架生成的动作在69.7%的比较中排名第一，高于我们的无工作空间变体基线和随后受限的重定向真实动作。总体而言，结果支持在生成过程中将手势适应于可用工作空间，而不是在之后修改不受限制的轨迹以满足工作空间限制，这可能会影响手势的自然性。
cs.RO / 7 / 2609.38401
Memorize, Adapt, Ignore: Diagnosing Robot Learning Mechanisms under Training Data Variation
记忆、适应、忽视：在训练数据变化下诊断机器人学习机制
Abstract
Training data variation, whether through designing a domain randomization (DR) scheme in simulation or curating demonstrations for imitation learning, is a primary lever for improving the robustness of robotic manipulation policies. Yet its underlying mechanisms remain poorly understood, and practitioners typically select randomization parameters through expensive trial and error. We investigate these mechanisms through a series of case studies, randomizing object size, color, and type as well as scene lighting and linguistic prompts across settings including pick-and-place RL in ManiSkill and fine-tuning of vision-language-action (VLA) models on LIBERO and RoboTwin. We examine both model behavior and internal representations, using the empirical neural tangent kernel (NTK) as our primary diagnostic tool. We show that the NTK distinguishes a shift in the internal learning mechanism from \textit{memorizing} different situations with insufficient variation (e.g.\ learning what to do for a large cube, and what to do for a small cube) to \textit{adapting} to the situation at hand with sufficient variation. An NTK-based signal-to-noise ratio also helps distinguish when policies have learned to \emph{ignore} task-irrelevant factors (e.g.\ treating blue and red cubes identically, instead of learning a blue sub-policy and a red sub-policy). We use these diagnostics to develop practical guidance for designing DR schemes, selecting models, and detecting shortcut learning. We further compare different kinds of representations and validate our findings with real-world hardware experiments using ACT-based imitation learning.
Chinese Translation
训练数据的变化，无论是通过在模拟中设计领域随机化（Domain Randomization, DR）方案，还是通过策划模仿学习的示例，都是提高机器人操作策略鲁棒性的主要手段。然而，其潜在机制仍然不够清楚，实践者通常通过昂贵的试错方式选择随机化参数。我们通过一系列案例研究来探讨这些机制，随机化对象的大小、颜色和类型，以及场景的光照和语言提示，涵盖包括在 ManiSkill 中的拾取与放置强化学习（Reinforcement Learning, RL）和在 LIBERO 及 RoboTwin 上对视觉-语言-动作（Vision-Language-Action, VLA）模型的微调等设置。我们考察了模型行为和内部表征，使用经验神经切线核（Neural Tangent Kernel, NTK）作为主要诊断工具。我们展示了 NTK 能够区分内部学习机制的变化，从在变化不足的情况下对不同情境的 extit{记忆}（例如，学习如何处理大立方体和小立方体）到在变化充足的情况下对当前情境的 extit{适应}。基于 NTK 的信噪比也有助于区分策略何时学会 extit{忽视}与任务无关的因素（例如，将蓝色和红色立方体视为相同，而不是学习蓝色子策略和红色子策略）。我们利用这些诊断工具为设计 DR 方案、选择模型和检测捷径学习提供实用指导。我们进一步比较了不同类型的表征，并通过使用基于 ACT 的模仿学习的真实硬件实验验证了我们的发现。
cs.RO / 8 / 2609.38405
Draft: A Parametric Tool for Robot Design Exploration
Draft：一种用于机器人设计探索的参数化工具
Abstract
Robot performance is often limited by the cost of iterating on morphology and control together, since every computer-aided design (CAD) change has to be carried into a simulation-ready model before control work begins. Co-design methods attempt to close this gap, but each uses a model generator written for a single platform or lack the use of real-world data to suggest that designs are plausible. We present Draft, a parametric generation tool whose generalized engine compiles any parametric tree of serial chains into a simulation-ready MJCF model, without CAD. It allows engineers to explore design tradeoffs through easily adjustable models and evaluate how changes influence controller performance. Draft grounds the free parameters of each design using trends fitted to a survey of $114$ actuators and $49$ published robot descriptions, so that a generated robot is anchored to real-world hardware. We validate those trends wholistically by building twins of four off-the-shelf robots, whose masses agree to $1.10\times$ geometric mean fold error. Finally, we demonstrate how Draft exposes design tradeoffs by evaluating three quadrupeds through a two-stage reinforcement learning curriculum.
Chinese Translation
机器人的性能常常受到形态和控制共同迭代成本的限制，因为每一次计算机辅助设计（CAD）的更改都必须转化为一个可用于仿真的模型，然后才能开始控制工作。协同设计方法试图缩小这一差距，但每种方法都使用为单一平台编写的模型生成器，或者缺乏使用真实世界数据来表明设计的可行性。我们提出了Draft，一种参数化生成工具，其通用引擎能够将任何串联链的参数化树编译成一个无需CAD的仿真准备好的MJCF模型。它允许工程师通过易于调整的模型探索设计权衡，并评估变化如何影响控制器性能。Draft通过对114个执行器和49个已发布机器人描述的调查数据拟合趋势，来确定每个设计的自由参数，从而使生成的机器人与真实世界硬件相结合。我们通过构建四个现成机器人的双胞胎来全面验证这些趋势，其质量的几何平均折算误差为1.10倍。最后，我们展示了Draft如何通过评估三个四足机器人在两阶段强化学习课程中的表现来揭示设计权衡。
cs.RO / 9 / 2609.38418
PneuTac: Tactile Manipulation with Soft Pneumatic Robots via Unified MPM-Gaussian Splatting Simulation
PneuTac：通过统一的MPM-高斯点云模拟实现软气动机器人触觉操控
Abstract
Soft robots and tactile sensors have demonstrated great potential in delicate manipulation tasks. Soft pneumatic robots enable safe contact through compliance, and vision-based tactile sensors offer high-resolution touch perception. However, learning tactile manipulation with compliant robots has been challenging, bottlenecked by the lack of efficient simulation. Existing simulators typically model them in isolation, and exhibit large calibration gaps that are difficult to overcome efficiently. We present PneuTac, a unified framework for tactile-feedback manipulation with soft pneumatic robots. We leverage the material point method (MPM) for modelling the dynamics of the soft robot and the deformable tactile membrane, and 3D Gaussian splatting (3DGS) for rendering. Real-to-sim modelling is done with a simple vision-based method, to then train action and perception networks for efficient simulation with surrogate models. We use the framework to drive a tactile-guided pipeline to collect demonstrations in simulation. Through experiments on a custom-designed pneumatic soft finger with a tactile sensing tip, together with additional cross-device evaluations, we show that PneuTac is capable of accurately modelling soft robots with tactile sensors, and that policies trained with simulation-augmented demonstrations outperform baselines trained on the same real data on three real-world contact-rich compliant manipulation tasks, making it a practical framework for tactile manipulation on compliant hardware.
Chinese Translation
软机器人和触觉传感器在精细操控任务中展现出了巨大的潜力。软气动机器人通过顺应性实现安全接触，而基于视觉的触觉传感器则提供了高分辨率的触觉感知。然而，使用顺应性机器人学习触觉操控一直面临挑战，主要受限于缺乏高效的模拟。现有的模拟器通常将其孤立建模，并表现出较大的校准差距，难以高效克服。我们提出了PneuTac，一个用于软气动机器人触觉反馈操控的统一框架。我们利用材料点法（MPM）来建模软机器人和可变形触觉膜的动态，使用3D高斯点云（3DGS）进行渲染。通过简单的基于视觉的方法进行真实到模拟的建模，然后训练动作和感知网络以实现高效的替代模型模拟。我们使用该框架驱动触觉引导的管道以在模拟中收集演示。通过在定制设计的带有触觉传感器尖端的气动软手指上进行实验，以及额外的跨设备评估，我们展示了PneuTac能够准确建模带有触觉传感器的软机器人，并且使用模拟增强演示训练的策略在三个真实世界的接触丰富的顺应性操控任务中优于基于相同真实数据训练的基线，从而使其成为在顺应性硬件上进行触觉操控的实用框架。
cs.RO / 10 / 2609.38421
A Reachability-based Safety Certificate for Dynamical System Motion Policies
基于可达性的动态系统运动策略安全证书
Abstract
Dynamical Systems (DS) are reactive motion policies representing vector fields trained with theoretical guarantees of stability and convergence. To ensure safety during deployment in unknown environments they must be locally reshaped, either through modulation or geometric control barrier function strategies. However, depending on the geometry of the obstacles and the complexity of the DS, these local strategies can lead the system to unavoidable collisions or spurious attractors. In this work, we certify safety with a value function drawn from the notion of backward reachability tube, which measures the worst-case safety along a rollout trajectory of the nominal DS. Usually, such a value function is intractable for a controlled system due to curse of dimensionality. We show that in the DS-based learning-from-demonstration setting, the absence of a control input collapses the reachability problem to a deterministic rollout, and the presence of certain stability conditions truncates the infinite horizon to a finite one, resulting in a well-defined value function. We further show that the value function we devised is the maximal forward-invariant subset of the obstaclefree region for the nominal DS flow. The application of this certificate function is validated across five DS constructions - analytical, Neural ODE, diffeomorphic latent space, LPV-DS, SE(3)and validate it on a Franka manipulator. Modulation and geometric CBFs also suffer from saddle point in cases of headon approach towards an unsafe zone. We show that CBF-on-V avoids this pitfall entirely.
Chinese Translation
动态系统（DS）是表示向量场的反应性运动策略，具有稳定性和收敛性的理论保证。为了确保在未知环境中的部署安全，它们必须通过调制或几何控制屏障函数策略进行局部重塑。然而，根据障碍物的几何形状和动态系统的复杂性，这些局部策略可能导致系统不可避免的碰撞或虚假吸引子。在本研究中，我们通过从反向可达性管的概念中提取的值函数来证明安全性，该值函数衡量了名义动态系统的展开轨迹上的最坏情况安全性。通常，由于维度诅咒，这种值函数对于受控系统是不可处理的。我们表明，在基于动态系统的示范学习设置中，缺乏控制输入将可达性问题简化为确定性展开，并且某些稳定性条件的存在将无限视界截断为有限视界，从而导致一个良定义的值函数。我们进一步表明，我们设计的值函数是名义动态系统流的无障碍区域的最大前向不变子集。该证书函数的应用在五种动态系统构造中得到了验证——解析的、神经常微分方程（Neural ODE）、微分同胚潜在空间、线性参数变化动态系统（LPV-DS）、SE(3)并在Franka操纵器上进行了验证。在面对不安全区域的正面接近时，调制和几何控制屏障函数也会遭遇鞍点。我们展示了CBF-on-V完全避免了这一陷阱。
cs.RO / 11 / 2609.38437
Embodiment-aware control by inference over the operator: a simulation study
通过对操作员的推理实现的具身感知控制：一项模拟研究
Abstract
Teleoperation systems are tuned for channel fidelity, while whether the operator experiences the device as part of the body, the Sense of Embodiment (SoE), is measured only afterwards, by questionnaire. Predictive-processing accounts suggest controlling devices to reduce the mismatch between the operator's predictions and the returned feedback, but those predictions are unobservable, and an objective that only penalizes mismatch is minimized by removing feedback. We formulate an embodiment-aware controller, the Universal Embodiment Engine (UEE), that infers the operator's embodiment and visuo-proprioceptive cue weighting from implicit gaze and pupil signals and task outcome, and chooses bounded device settings under explicit preferences, cast as a discrete Active Inference agent. In simulations with 300 heterogeneous synthetic operators, the UEE found the suitable setting within half a minute for most operators, before identifying their exact type, and came close to an oracle in the second half of the session (embodiment 1.68 against 0.73 for the best fixed setting, on a 0-2 scale). Adapting without reading the operator did no better than fixed control, and model-free bandits did worse, whereas an expected-utility controller with the same inference did exactly as well: the benefit comes from Bayesian inference over the operator with explicit preferences, not from the information-seeking term of Active Inference. A naive prediction-error minimizer withheld feedback, as its objective implies, and lost task success (0.74 vs 0.90). The benefit shrank but persisted for operators outside the controller's model family, grew with the variety of the population, and vanished when the controller trusted an uninformative signal or when cue weighting changed mid-session without being modeled. These failures show what studies with people must establish first: calibrated signals and a model of change.
Chinese Translation
遥操作系统的调节侧重于通道保真度，而操作员是否将设备视为身体的一部分，即具身感知（Sense of Embodiment, SoE），仅通过问卷在事后进行测量。预测处理理论建议通过控制设备来减少操作员的预测与反馈之间的不匹配，但这些预测是不可观察的，且仅惩罚不匹配的目标会通过去除反馈而被最小化。我们提出了一种具身感知控制器——通用具身引擎（Universal Embodiment Engine, UEE），该控制器通过隐含的注视和瞳孔信号以及任务结果推断操作员的具身感知和视觉-本体感觉线索权重，并在明确偏好下选择有界的设备设置，将其视为一个离散的主动推理代理。在对300个异质合成操作员的模拟中，UEE在大多数操作员中在半分钟内找到了合适的设置，随后才确定其确切类型，并在会话的后半段接近于一个神谕（具身感知1.68，相较于最佳固定设置的0.73，采用0-2的评分标准）。在未读取操作员的情况下进行适应的效果与固定控制相当，而无模型的赌博算法表现更差，而具有相同推理的期望效用控制器表现完全相同：其收益来自于对具有明确偏好的操作员进行贝叶斯推理，而非主动推理的信息寻求项。一个天真的预测误差最小化器如其目标所暗示的那样，抑制了反馈，导致任务成功率下降（0.74对比0.90）。尽管收益有所缩小，但在控制器模型家族之外的操作员中仍然存在，且随着人群多样性的增加而增长，当控制器信任一个无信息信号或在会话中途线索权重发生变化而未被建模时，收益则消失。这些失败表明，针对人类的研究必须首先建立：校准信号和变化模型。
cs.RO / 12 / 2609.38443
BIND: Binding 3D Robot Actions to 2D Image Features
BIND：将3D机器人动作与2D图像特征绑定
Abstract
We introduce BIND, a new action representation for visuomotor robot policies that binds 3D robot actions to their corresponding 2D image features, yielding strong data efficiency gains and robustness to out-of-distribution object positions and camera viewpoints. The action heads of current robot policies are typically formulated as an MLP regression from a single global feature vector produced by a pre-trained vision encoder. This global formulation requires the policy network to discover, from demonstrations alone, the relationship between target robot actions and the image features they project onto. The consequence is that although modern image features are semantically descriptive, spatially robust, and even multiview-consistent, the policies built on them are brittle to subtle changes in camera viewpoint and object placement--and surprisingly data-inefficient. BIND closes this gap by supplying the action-feature relationship through camera geometry rather than learning: it discretizes a volume of candidate end effector positions, attaches each candidate to the pre-trained features at its projection in each camera view, and selects actions by scoring each candidate's position and image-bound feature combination. On a real robot, we study data efficiency and out-of-distribution robustness to unseen object positions and camera viewpoints, as well as general long-horizon task execution and dexterity. We find BIND to be highly data-efficient and robust: it achieves near-perfect success on tasks with as few as 5 demonstrations, and degrades gracefully under steep camera-viewpoint shifts and held-out object positions where coordinate-regression baselines completely fail.
Chinese Translation
我们介绍了BIND，一种新的视觉运动机器人策略的动作表示方法，它将3D机器人动作与相应的2D图像特征绑定，从而在数据效率和对分布外物体位置及相机视角的鲁棒性方面取得显著提升。目前的机器人策略的动作头通常被表述为从由预训练视觉编码器生成的单一全局特征向量进行的多层感知机（MLP）回归。这种全局表述要求策略网络仅通过演示来发现目标机器人动作与其投影到的图像特征之间的关系。结果是，尽管现代图像特征在语义上具有描述性、在空间上具有鲁棒性，甚至在多视角上保持一致，但基于这些特征构建的策略对相机视角和物体放置的微小变化非常脆弱，并且出乎意料地数据效率低下。BIND通过相机几何关系提供动作-特征关系，从而弥补了这一差距：它对候选末端执行器位置的体积进行离散化，将每个候选位置附加到其在每个相机视角下的投影的预训练特征上，并通过对每个候选位置和图像绑定特征组合进行评分来选择动作。在真实机器人上，我们研究了数据效率和对未见物体位置及相机视角的分布外鲁棒性，以及一般的长时间任务执行和灵活性。我们发现BIND具有高度的数据效率和鲁棒性：在仅有5次演示的任务中，它几乎实现了完美的成功，并且在剧烈的相机视角变化和被排除的物体位置下表现出良好的降级，而坐标回归基线则完全失败。
cs.RO / 13 / 2609.38462
Observability Analysis and Online Calibration of Visual-Inertial-Wheel Odometry for 4WIS4WID Mobile Robots
四轮独立转向和驱动移动机器人视觉惯性轮式里程计的可观测性分析与在线标定
Abstract
In this paper, we present a visual-inertial-wheel odometry (VIWO) framework with online calibration for four-wheel independently steered and driven (4WIS4WID) mobile robots. We derive a 2D odometry model directly from the four driving velocities and steering angles, using both the longitudinal rolling constraints and the lateral no-slip constraints of all wheels. A preintegration model and analytical Jacobians are developed for efficient filtering and calibration. An observability analysis of the linearized VIWO system shows that a drive-only model makes all steering offsets unobservable, whereas the proposed redundant model restores their observability. The analysis also identifies four standard VINS unobservable directions and three additional directions associated with the arbitrary placement of the odometry reference frame. Furthermore, we characterize several degenerate motions, including zero yaw rate, constant steering, and a non-rolling wheel, and derive the corresponding excitation conditions for the thirteen wheel intrinsics that remain after fixing the odometry frame reference. The performance of the proposed system has been demonstrated in both simulation and real-world experiments on a 4WIS4WID mobile robot.
Chinese Translation
本文提出了一种针对四轮独立转向和驱动（4WIS4WID）移动机器人的视觉惯性轮式里程计（VIWO）框架，并实现了在线标定。我们直接从四个驱动速度和转向角推导出二维里程计模型，利用所有轮子的纵向滚动约束和横向无滑移约束。为高效的滤波和标定，我们开发了预积分模型和解析雅可比矩阵。对线性化VIWO系统的可观测性分析表明，仅驱动模型使所有转向偏移不可观测，而所提出的冗余模型恢复了它们的可观测性。该分析还识别出四个标准的VINS不可观测方向以及与里程计参考框架任意放置相关的三个额外方向。此外，我们对几种退化运动进行了表征，包括零偏航率、恒定转向和非滚动轮，并推导出在固定里程计框架参考后剩余的十三个轮子内在参数的相应激励条件。所提系统的性能已在4WIS4WID移动机器人的仿真和实际实验中得到了验证。
cs.RO / 14 / 2609.38463
TrafficSignBench: Rule-Centric Closed-Loop Evaluation of Traffic-Sign Compliance in Autonomous Driving
TrafficSignBench：基于规则的闭环评估自动驾驶中交通标志遵循情况
Abstract
Autonomous driving planners are typically evaluated using aggregate metrics such as driving score, destination rate, and collision rate, which do not explicitly measure compliance with traffic rules. As a result, planners can achieve high benchmark scores while still exhibiting unsafe or illegal behaviors, limiting their applicability to real-world deployment. To address this gap, we introduce TrafficSignBench, a large-scale, traffic sign-centric benchmark for systematic and interpretable evaluation of traffic-rule compliance in autonomous driving. Our framework combines real-map-based simulation for realistic road layouts with rule-targeted procedural scenario generation for scalable and balanced coverage of underrepresented rules. We implement traffic rules corresponding to 34 traffic signs, each equipped with an automatic rule checker for detecting violations during closed-loop execution. This design yields 29,000 diverse road scenes and 29 distinct testing scenario types, enabling controlled evaluation of rule-specific planner behavior. We construct 5,800 testing scenes and demonstrate that current autonomous driving planners can exhibit poor traffic-rule compliance despite strong performance on standard evaluation metrics. To address this limitation, we transform existing planners into rule-compliant trajectory experts via explicit traffic-sign constraints, enabling scalable generation of high-quality oracle trajectories for fine-tuning.
Chinese Translation
自动驾驶规划器通常通过驾驶评分、目的地到达率和碰撞率等聚合指标进行评估，这些指标并未明确衡量对交通规则的遵循。因此，规划器可能在基准测试中获得高分，但仍表现出不安全或非法的行为，从而限制了其在现实世界中的应用。为了解决这一问题，我们提出了TrafficSignBench，这是一个大规模、以交通标志为中心的基准，用于系统性和可解释地评估自动驾驶中的交通规则遵循情况。我们的框架结合了基于真实地图的模拟，以实现逼真的道路布局，并采用针对规则的程序场景生成，以实现对欠代表规则的可扩展和平衡覆盖。我们实现了与34个交通标志相对应的交通规则，每个规则都配备了自动规则检查器，以在闭环执行过程中检测违规行为。该设计产生了29,000个多样化的道路场景和29种不同的测试场景类型，使得对规则特定规划器行为的控制评估成为可能。我们构建了5,800个测试场景，并展示了当前的自动驾驶规划器在标准评估指标上表现良好，但仍可能表现出较差的交通规则遵循情况。为了解决这一局限性，我们通过明确的交通标志约束将现有规划器转变为遵循规则的轨迹专家，从而实现高质量的oracle轨迹的可扩展生成，以进行微调。
cs.RO / 15 / 2609.38467
L1-MPPI: L1 Adaptive Model Predictive Path Integral for Agile UAV Control
L1自适应模型预测路径积分（L1-MPPI）用于灵活无人机控制
Abstract
This work proposes the L1 Adaptive Model Predictive Path Integral (L1-MPPI). It cascades L1 adaptive control with the Model Predictive Path Integral (MPPI) to improve tracking of high-speed UAV trajectories. Thanks to the L1augmentation, the tracking remains accurate even under model uncertainties and external disturbances, such as an additional payload or a mismatch in the modeled aerodynamic drag. In contrast to existing MPPI approaches for UAV control that do not explicitly model aerodynamic effects, varying payloads, and typically neglect the dynamics of low-level motor controllers, our L1-MPPI approach enhances the dynamic model used in the MPPI by incorporating the low-level flight controller and motor dynamics, as well as an iterative mixing scheme that reflects the approach of the low-level controller. The proposed method demonstrates improved tracking performance in both simulation and the real world, even when the UAV is subjected to an unknown payload. In flight with 35% mass increase, our approach lowers the RMSE by 58.61% with respect to plain MPPI. Compared to the same MPPI using an online mass estimator in place of the L1 augmentation, the RMSE is lower by 38.59%. During the real-world experiments the UAV reaches speeds up to 13.50 m/s and accelerations up to 2.5 g.
Chinese Translation
本文提出了L1自适应模型预测路径积分（L1-MPPI）。该方法将L1自适应控制与模型预测路径积分（MPPI）相结合，以提高高速无人机轨迹的跟踪精度。得益于L1增强，即使在模型不确定性和外部干扰（如额外负载或建模气动阻力不匹配）下，跟踪仍然保持准确。与现有的无人机控制MPPI方法不同，后者并未明确建模气动效应、变化的负载，并且通常忽略低级电机控制器的动态特性，我们的L1-MPPI方法通过结合低级飞行控制器和电机动态，以及反映低级控制器方法的迭代混合方案，增强了MPPI中使用的动态模型。所提出的方法在模拟和现实世界中均显示出改进的跟踪性能，即使在无人机承受未知负载的情况下。在质量增加35%的飞行中，我们的方法相较于普通MPPI降低了58.61%的均方根误差（RMSE）。与使用在线质量估计器替代L1增强的相同MPPI相比，RMSE降低了38.59%。在实际实验中，无人机的速度可达13.50 m/s，加速度可达2.5 g。
cs.RO / 16 / 2609.38472
Diffusion-2BC: Hybrid Diffusion and Regression Training for Offline Behavior Cloning in Autonomous Driving
Diffusion-2BC：用于自主驾驶的离线行为克隆的混合扩散与回归训练
Abstract
Behavior cloning provides an offline route to autonomous-driving policy learning, but mean-squared-error regression is poorly matched to demonstrations in which one observation admits several valid actions. Diffusion policies can represent conditional multimodal action distributions, yet their closed-loop performance may be unstable when visual features and control are learned from limited data. This paper presents Diffusion-2BC, which combines a diffusion denoising objective with an auxiliary deterministic behavior-cloning loss over a shared visual encoder. The auxiliary branch is used only during training; inference remains diffusion-based. The proposed method is evaluated in the controlled Claw environment and in bird's-eye-view CARLA navigation, including route-conditioned driving, route-free navigation through multiple intersections, and cross-map evaluation from Town01 to Town02. In the Claw task, Diffusion-2BC reduced the mean mask-distance error by approximately 10% relative to a diffusion-based behavior-cloning baseline and by 85% relative to standard deterministic behavior cloning. In route-free CARLA, Diffusion-2BC traveled substantially farther before termination under the evaluation protocol than both baselines in Town01 and Town02. Additional qualitative rollouts revealed distinct route choices, showing the multimodal behavior of the proposed diffusion-based agent. The results indicate that an auxiliary regression signal can improve the closed-loop reliability of diffusion behavior cloning while preserving multimodal prediction in the controlled benchmark.
Chinese Translation
行为克隆为自主驾驶策略学习提供了一条离线路径，但均方误差回归与那些一个观察值对应多个有效动作的演示不匹配。扩散策略能够表示条件多模态动作分布，但当视觉特征和控制从有限数据中学习时，其闭环性能可能不稳定。本文提出了Diffusion-2BC，它结合了扩散去噪目标与共享视觉编码器上的辅助确定性行为克隆损失。辅助分支仅在训练期间使用；推理仍基于扩散。所提出的方法在受控的Claw环境和鸟瞰视角的CARLA导航中进行了评估，包括基于路线的驾驶、通过多个交叉口的无路线导航，以及从Town01到Town02的跨地图评估。在Claw任务中，Diffusion-2BC相较于基于扩散的行为克隆基线减少了约10%的均值掩码距离误差，相较于标准确定性行为克隆减少了85%。在无路线的CARLA中，Diffusion-2BC在评估协议下的终止前行驶的距离明显超过了Town01和Town02中的两个基线。额外的定性回放揭示了不同的路线选择，展示了所提扩散基代理的多模态行为。结果表明，辅助回归信号可以提高扩散行为克隆的闭环可靠性，同时在受控基准中保持多模态预测。
cs.RO / 17 / 2609.38483
CADeT: Causal-Aware Deformation Transmission for Indirect Robotic Manipulation of Soft Tissue
CADeT：用于软组织间接机器人操作的因果感知变形传输
Abstract
Indirect manipulation of deep-seated deformable anatomy inaccessible to the robot is challenging in robot-assisted minimally invasive surgery (RAMIS) because intervening tissues spatially filter deformation transmission. Passive observations can be ambiguous because the Decoupled and Blocked modes may produce similar motion responses. We propose CADeT, a causal-aware deformation transmission framework that integrates structural causal model (SCM) with active sensing to infer a latent transmission mode and estimate a state-dependent adhesion Jacobian online. During normal manipulation, control actions update the mode belief; when ambiguity persists, an additional probing action is selected to improve mode distinguishability. The mode belief and learned Jacobian are incorporated into a belief-aware model predictive controller for indirect target-shape control. Validation in simulation and on the da Vinci research kit (dVRK), using phantom and ex vivo porcine tissues, shows higher mode-identification accuracy and faster shape-error convergence than the evaluated model-free and model-based baselines. These results show that active sensing improves mode identification and indirect deformation control under the evaluated conditions.
Chinese Translation
在机器人辅助微创手术（RAMIS）中，深层可变形解剖结构的间接操作面临挑战，因为干预组织在空间上过滤了变形传输。被动观察可能会产生歧义，因为解耦模式和阻塞模式可能产生相似的运动响应。我们提出了CADeT，一种因果感知变形传输框架，结合了结构因果模型（SCM）与主动感知，以推断潜在的传输模式并在线估计状态依赖的粘附雅可比矩阵。在正常操作过程中，控制动作更新模式信念；当歧义持续存在时，选择额外的探测动作以提高模式可区分性。模式信念和学习到的雅可比矩阵被纳入一个信念感知的模型预测控制器，用于间接目标形状控制。在仿真和da Vinci研究工具包（dVRK）上进行的验证，使用幻影和体外猪组织，显示出比评估的无模型和基于模型的基线更高的模式识别准确性和更快的形状误差收敛。这些结果表明，主动感知在评估条件下改善了模式识别和间接变形控制。
cs.RO / 18 / 2609.38494
What to Attend, What to Keep: Skill-Conditioned Visuotactile Representation with Progress-Guided Event Memory
关注什么，保留什么：基于技能条件的视觉触觉表征与进度引导的事件记忆
Abstract
Robotic manipulation integrates vision, touch, and language, whose importance shifts across stages: vision guides reaching, while touch, through its evolution over time, decides grasping, alignment, and contact. Yet existing multi-modal manipulation policies typically use fixed temporal contexts and fusion strategies, despite shifts in what each modality contributes across different skills. We study how vision and touch should be combined at the level of primitive skills, asking what each skill needs from each sensor, and propose a skill-conditioned representation in which the queried skill conditions fusion over modality-specific short-term observation tokens while attending to a sparse event memory that retains terminal observations from the last $K$ executed skills. Evaluated by skill progress estimation on three contact-rich tasks, it reduces slip-detection delay by 87% against fine-tuned SOTA progress models, twist-completion delay by 67.5% against a vision-only ablation, and progress error on a blind search task by 92% through sparse event memory. Gains concentrate exactly where completion is defined by contact or task history. More broadly, our results suggest that observation formation not only policy architecture is a central challenge in multi-modal representation. Project Website: http://what-to-attend-what-to-keep.github.io/
Chinese Translation
机器人操作整合了视觉、触觉和语言，其重要性在不同阶段有所变化：视觉引导到达，而触觉则通过随时间演变来决定抓取、对齐和接触。然而，现有的多模态操作策略通常使用固定的时间上下文和融合策略，尽管在不同技能中每种模态的贡献有所不同。我们研究了在原始技能层面上，视觉和触觉应如何结合，询问每项技能需要从每个传感器获取什么，并提出了一种技能条件表征，其中查询的技能条件融合基于模态特定的短期观察标记，同时关注保留最近 $K$ 个执行技能的终端观察的稀疏事件记忆。通过在三个接触丰富的任务上评估技能进度估计，该方法相比于经过微调的最先进（SOTA）进度模型将滑动检测延迟减少了87%，相比于仅使用视觉的消融实验将扭转完成延迟减少了67.5%，并通过稀疏事件记忆将盲搜索任务的进度误差减少了92%。增益恰好集中在由接触或任务历史定义的完成位置。更广泛地说，我们的结果表明，观察形成而不仅仅是策略架构是多模态表征中的一个核心挑战。项目网站：http://what-to-attend-what-to-keep.github.io/
cs.RO / 19 / 2609.38511
Onboard Vision and MPC Navigation for Underwater Robots: An Open BlueROV2 Platform for Multi-Robot Experiments & Docking
水下机器人上的视觉与模型预测控制导航：用于多机器人实验与对接的开放式 BlueROV2 平台
Abstract
Autonomous underwater robots require robust perception, estimation and control to operate in confined environments. This paper presents an open-source BlueROV2 platform combining onboard vision with nonlinear Model Predictive Control (NMPC) for autonomous navigation and docking. The platform integrates an NVIDIA Jetson Orin NX and an Intel RealSense D435i stereo camera in a modular pressure housing. Underwater-calibrated stereo depth and realtime object detection provide relative position measurements of nearby BlueROV2 vehicles in the camera and body frames. A quaternion-based estimator fuses external pose and inertial measurements, while an NMPC controller based on a nonlinear six-degree-of-freedom model tracks planned navigation and docking trajectories. To support reproducible development, we also provide open-source physics-based PX4 SITL and Gazebo environments, multi-robot simulation tools and a lowcost physical docking station. Experiments evaluate underwater perception, onboard computational performance, state estimation, trajectory tracking and autonomous docking.
Chinese Translation
自主水下机器人在狭窄环境中操作需要强大的感知、估计和控制能力。本文提出了一种开源的 BlueROV2 平台，将机载视觉与非线性模型预测控制（NMPC）相结合，以实现自主导航和对接。该平台集成了 NVIDIA Jetson Orin NX 和 Intel RealSense D435i 立体摄像头，采用模块化压力外壳。经过水下标定的立体深度和实时物体检测提供了相邻 BlueROV2 车辆在摄像头和机体坐标系中的相对位置测量。基于四元数的估计器融合了外部姿态和惯性测量，而基于非线性六自由度模型的 NMPC 控制器则跟踪规划的导航和对接轨迹。为了支持可重复的开发，我们还提供了开源的基于物理的 PX4 SITL 和 Gazebo 环境、多机器人仿真工具以及低成本的物理对接站。实验评估了水下感知、机载计算性能、状态估计、轨迹跟踪和自主对接。
cs.RO / 20 / 2609.38527
Behavioral Persistence and Incomplete Functional Transfer of Co-evolved Communication in Evolutionary Robotics
行为持久性与共进化通信在进化机器人中的不完全功能转移
Abstract
This work evaluates the direct transfer of a co-evolved communication protocol from a 2D simulation to a 3D physical environment, without retraining the network weights. Two e-puck-type robots, controlled by a GRU network with residual connection, were evaluated in a food-seeking task with social signaling. The sensory and motor translation layer required three corrections for stable physical operation, including the calibration of a hunger term based on a measurable asymmetry in the trained residual weights. Even with these corrections, the transfer was partial and asymmetric: one agent reached the food source in one of thirty tested seeds, while the other did not reach it in any. Task success was measured by both agents reaching the food area. An additional experiment incorporating explicit directional information in the social channel produced observable changes in the trajectory of the receiving agent and improvements in several specific cases. However, these improvements were not enough to allow the second agent to reach the food source, suggesting that the limitation may not be explained solely by signal translation, but also by the ability to navigate under the new physical constraints. The results suggest that successful transfer of emergent communication may depend not only on preserving the signaling process itself, but also on preserving the ecological and navigational conditions under which the protocol evolved.
Chinese Translation
本研究评估了共进化通信协议从二维仿真环境直接转移到三维物理环境的可行性，且未对网络权重进行重新训练。我们评估了两台由具有残差连接的门控循环单元（GRU）网络控制的 e-puck 型机器人在进行社交信号的寻食任务中的表现。感知与运动转换层需要进行三次修正以实现稳定的物理操作，包括基于训练后的残差权重中可测量的不对称性进行饥饿项的校准。即便进行了这些修正，转移依然是部分的且不对称：在三十个测试种子中，只有一个代理成功到达食物源，而另一个代理则没有到达。任务成功的标准是两个代理都能到达食物区域。另一个实验引入了社交通道中的显性方向信息，产生了接收代理轨迹的可观察变化，并在若干特定情况下有所改善。然而，这些改善不足以使第二个代理到达食物源，这表明限制可能不仅仅是由信号转换所解释，还与在新物理约束下的导航能力有关。结果表明，成功转移新兴通信可能不仅依赖于保留信号过程本身，还依赖于保留协议演化所需的生态和导航条件。
cs.RO / 21 / 2609.38537
Systematically Exploring the Capabilities of GPT-6 Astra as Embodied Policies
系统性探索 GPT-6 Astra 作为具身策略的能力
Galbot Team, Chen, Xuchuan, Cheng, Xiaoqian, Deng, Yu, Ding, Lihe, Dong, Shaocong, Gao, Xiangjun, Jia, Haozhe, Li, Zekai, Li, Zhoujian, Lian, Yunrui, Liang, Sikai, Lin, Chenghuai, Liu, Dairu, Liu, Jiahang, Liu, Qingtao, Ma, Yuxuan, Qi, Zekun, Su, Jiayi, Wang, He, Xu, Ruochen, Xu, Tianyu, Xu, Xudong, Xu, Zhe, Yan, Mi, Yan, Siming, Yi, Li, Yu, Ruixi, Zhang, Jinlu, Zhang, Yintianrun, Zhang, Zhikai, Zhang, Zhizheng, Zheng, Yixin, Zhu, Weiyi
Abstract
GPT-6 Astra exhibits a remarkable ability to generate numerical robot actions, extending its role beyond high-level planning. To assess Astra's capabilities as general-purpose embodied policies, we conduct comprehensive evaluations across six domains, examining direct control, cooperation with learned policies, and feedback-driven adaptation. In gripper manipulation, Astra can correct task targets and prepare contact conditions for subsequent policy execution; hybrid control with {\pi}0.5 achieves 48% success on the evaluated RoboDojo subset. In dexterous manipulation, hybrid control achieves 50% success in ten experience-guided DexJoCo trials, while direct in-hand control struggles to coordinate finger contacts. In mobile manipulation, hybrid control reaches 38.7% success on the evaluated RoboCasa365. In navigation, Astra leads our local comparisons, reaching 92% success on RxR instruction following and 82% on HM3D object search, although search incurs substantial detours. In locomotion, dense motion-reference generation remains unreliable: none of five sequential attempts on a single obstacle course reaches the goal, despite improvements in stability and forward progress. In humanoid loco-manipulation, Astra exceeds baseline methods on 13 of 30 HumanoidBench tasks with pretrained whole-body controllers. These findings reveal a gap between useful task decisions and reliable physical control. Inference latency further constrains practical control: across 50 RoboDojo instances per condition, policy-assisted and direct control consume 624.8 million and 1.132 billion tokens. A 30-second locomotion run requires 250 model calls averaging 39.86 seconds each, with physics paused during inference.
Chinese Translation
GPT-6 Astra 展现出生成数值机器人动作的显著能力，超越了高层次规划的角色。为了评估 Astra 作为通用具身策略的能力，我们在六个领域进行了全面评估，考察了直接控制、与学习策略的合作以及基于反馈的适应。在夹持器操作中，Astra 能够纠正任务目标并为后续策略执行准备接触条件；与 { ext{π}}0.5 的混合控制在评估的 RoboDojo 子集上取得了 48% 的成功率。在灵巧操作中，混合控制在十次经验指导的 DexJoCo 试验中取得了 50% 的成功率，而直接手内控制在协调手指接触方面表现不佳。在移动操作中，混合控制在评估的 RoboCasa365 上达到了 38.7% 的成功率。在导航方面，Astra 在我们的本地比较中表现突出，在 RxR 指令跟随中达到了 92% 的成功率，在 HM3D 物体搜索中达到了 82% 的成功率，尽管搜索过程中产生了大量的绕路。在运动方面，密集的运动参考生成仍然不可靠：在单个障碍课程上的五次连续尝试中，没有一次达到目标，尽管在稳定性和前进速度上有所改善。在类人运动操作中，Astra 在 30 个 HumanoidBench 任务中的 13 个任务上超越了基线方法，使用了预训练的全身控制器。这些发现揭示了有用任务决策与可靠物理控制之间的差距。推理延迟进一步限制了实际控制：在每种条件下的 50 个 RoboDojo 实例中，策略辅助控制和直接控制分别消耗了 6.248 亿和 11.32 亿个标记。一次 30 秒的运动运行需要 250 次模型调用，平均每次 39.86 秒，在推理期间物理状态处于暂停状态。
cs.RO / 22 / 2609.38570
Data-Efficient Adaptation of a Driving VLA to Class 8 Trucks
驾驶视觉-语言-动作模型对8级卡车的数据高效适应
Abstract
Class 8 trucks differ from passenger cars in geometry, dynamics, and maneuvering requirements. As a result, vision-language-action (VLA) models trained for passenger vehicles do not readily transfer to Class 8 trucks, particularly in unstructured scenarios such as accident scenes and construction zones. Rather than training a truck-driving VLA from scratch, we propose an adapt-then-steer strategy that adapts an off-the-shelf VLA to generate trajectories for Class-8 trucks in these challenging scenarios. In the adapt stage, we use NVIDIA's Alpamayo 1.5 as the base model, fine-tuning only its action-generation stack on a few hundred real-world construction and accident-related highway scenarios. In the steer stage, we introduce Flow Velocity Steering (FVS) to further refine the model's predictions while holding the adapted VLA fixed. FVS is a compact, flow-time-conditioned residual module that adds learned corrections to the action-space flow velocity used to update the action sequence at each generation step. In open-loop evaluation on a scenario-disjoint held-out set, targeted fine-tuning more than halves single-candidate average displacement error (ADE) and final displacement error (FDE) over the entire 6.4 s horizon compared to the base model. Using the same targeted demonstrations, FVS further reduces the fine-tuned model's full-horizon ADE and FDE by 13.9% and 16.5%, respectively. At matched data budgets, targeted supervision yields 19-26% lower full-horizon ADE than general truck-driving supervision, while the targeted model remains competitive with a model fine-tuned on approximately 65 times as many general truck-driving scenarios. These results support adapt-then-steer for data-efficient vehicle-domain transfer to Class 8 trucks. Our project website is available at https://truckvla.github.io.
Chinese Translation
8级卡车在几何形状、动力学和操控要求上与乘用车存在显著差异。因此，针对乘用车训练的视觉-语言-动作（VLA）模型无法直接迁移到8级卡车，尤其是在事故现场和施工区域等非结构化场景中。我们提出了一种适应-再调整策略，而不是从头开始训练一个卡车驾驶VLA，该策略将现成的VLA进行适应，以在这些具有挑战性的场景中生成8级卡车的轨迹。在适应阶段，我们使用NVIDIA的Alpamayo 1.5作为基础模型，仅对其动作生成模块进行微调，使用几百个真实世界的施工和事故相关的高速公路场景。在再调整阶段，我们引入了流速引导（Flow Velocity Steering, FVS）来进一步优化模型的预测，同时保持适应后的VLA不变。FVS是一个紧凑的、流时间条件的残差模块，它为用于在每个生成步骤更新动作序列的动作空间流速添加学习到的修正。在对一个场景不重叠的保留集进行开放式评估时，针对性的微调使得单候选平均位移误差（ADE）和最终位移误差（FDE）在整个6.4秒的时间范围内相比基础模型减少了一半以上。使用相同的针对性演示，FVS进一步将微调模型的全时域ADE和FDE分别降低了13.9%和16.5%。在相同的数据预算下，针对性监督比一般的卡车驾驶监督降低了19-26%的全时域ADE，而针对性模型在与一个微调了大约65倍于一般卡车驾驶场景的模型相比仍然具有竞争力。这些结果支持适应-再调整策略在8级卡车上实现数据高效的车辆领域迁移。我们的项目网站可访问 https://truckvla.github.io。
cs.RO / 23 / 2609.38588
Drone Soccer: Learning to Manipulate with Multicopter Downwash
无人机足球：学习利用多旋翼下洗气流进行操作
Abstract
Although multicopter drones are traditionally designed for "perception-only" tasks, like mapping and exploration, recent work has sought to develop Unmanned Aerial Manipulators (UAMs) to solve mobile manipulation tasks. Aerial manipulation performance can be impacted by "downwash," the airflow produced by propellers, but current state-of-the-art UAMs either ignore downwash or treat it as a disturbance. Instead, is it possible to actively use downwash as a tool during manipulation? We design a drone soccer task to explore the feasibility of downwash-based manipulation. Specifically, we develop a simplified downwash dynamics model which we use to train an RL policy to dribble a soccer ball. We further demonstrate that our policy transfers to real world deployment. This work provides key insights into novel manipulation capabilities for multicopters.
Chinese Translation
尽管多旋翼无人机传统上被设计用于“仅感知”任务，如制图和探索，但近期的研究致力于开发无人机操作器（Unmanned Aerial Manipulators, UAMs）以解决移动操作任务。空中操作性能可能受到“下洗气流”的影响，即由螺旋桨产生的气流，但当前最先进的UAM要么忽略下洗气流，要么将其视为干扰。相反，是否可以在操作过程中主动利用下洗气流作为工具？我们设计了一个无人机足球任务来探索基于下洗气流的操作的可行性。具体而言，我们开发了一个简化的下洗气流动力学模型，并利用该模型训练强化学习（Reinforcement Learning, RL）策略以运球。我们进一步证明了我们的策略可以转移到现实世界的应用中。这项工作为多旋翼无人机的新型操作能力提供了重要的见解。
cs.RO / 24 / 2609.38594
FAST-Sync: Fast Group Synchronization for any Matrix Lie Group
FAST-Sync：适用于任意矩阵李群的快速群同步
Abstract
Group synchronization (GS) is the problem of estimating a set of $N$ unknown elements $g_1,\ldots, g_N \in \mathcal{G}$ in a group $\mathcal{G}$, given noisy measurements of a subset of their pairwise ratios $g_i^{-1} g_j$. GS problems lie at the core of many state estimation tasks in robotics and computer vision, including 3D vision, robotic mapping, inertial navigation, and molecular reconstruction. Unfortunately, GS problems are typically both high-dimensional and non-convex, and therefore hard to solve in general. In this paper, we present Fast-Sync, a fast linear approximation method for GS that is suitable for initializing local manifold-based optimizers or certifiable global methods. Our approach generalizes chordal initialization to arbitrary matrix Lie groups, and additionally proposes two new key algorithmic enhancements: we show how to exploit both the Kronecker-product structure in the problem data matrix and the topology of the synchronization graph to improve speed, scalability, and accuracy. Experimental evaluation across several GS tasks demonstrates that Fast-Sync provides high-quality initializations that enable local optimizers to efficiently recover globally optimal GS solutions, achieving high success rates even with considerable measurement noise.
Chinese Translation
群同步（Group Synchronization，GS）是一个估计群体 $ ext{G}$ 中 $N$ 个未知元素 $g_1, ext{...}, g_N ext{ } ext{属于} ext{ } ext{G}$ 的问题，给定其成对比率 $g_i^{-1} g_j$ 的噪声测量。GS 问题是许多机器人技术和计算机视觉中的状态估计任务的核心，包括三维视觉、机器人地图构建、惯性导航和分子重建。不幸的是，GS 问题通常是高维且非凸的，因此一般难以解决。在本文中，我们提出了 Fast-Sync，这是一种适合于初始化基于局部流形的优化器或可证明全局方法的快速线性近似方法。我们的方法将和弦初始化推广到任意矩阵李群，并提出了两个新的关键算法增强：我们展示了如何利用问题数据矩阵中的克罗内克积结构和同步图的拓扑结构来提高速度、可扩展性和准确性。在多个 GS 任务上的实验评估表明，Fast-Sync 提供了高质量的初始化，使局部优化器能够有效地恢复全局最优的 GS 解决方案，即使在测量噪声较大的情况下也能实现较高的成功率。
cs.RO / 25 / 2609.38616
Correcting WHERE, Preserving HOW: Compositional Generalization for Vision-Language-Action Models via Referential Guidance
纠正WHERE，保留HOW：通过指称引导实现视觉-语言-行动模型的组合泛化
Abstract
While Vision-Language-Action (VLA) models enable flexible action generation, their generalization across diverse environmental elements, including manipulated objects, destinations, and backgrounds, is limited by the lack of diversity in robotic training data. Trained end-to-end on such data, VLAs tend to exploit visual shortcuts, associating actions with task-irrelevant visual features rather than the intended task semantics. These shortcuts block recomposition of elements already seen by the policy, that is, compositional generalization. Existing approaches mitigate such entanglement through task-relevant perception or targeted data diversification, but offer no explicit mechanism for unseen recomposition and require backbone-specific modifications with retraining. We observe that under such recomposition, VLAs often fail at global grounding while retaining local manipulation skills that recover near the correct target in familiar configurations. Therefore, we propose Referential Guidance (ReGuide), a training-free wrapper that, given object poses from a grounding module, combines semantic and geometric rebinding to guide the end-effector into demonstration-supported configurations of the instructed referent, where the frozen policy can resume execution. Experiments in simulation across multiple VLA backbones as well as on a real robot show that ReGuide improves success rates under compositional shifts by up to 56.8 and 75.0 percentage points, respectively, while preserving standard-task performance.
Chinese Translation
尽管视觉-语言-行动（VLA）模型能够灵活生成动作，但由于机器人训练数据缺乏多样性，它们在不同环境元素（包括操控对象、目的地和背景）之间的泛化能力受到限制。在这样的数据上进行端到端训练的VLA往往利用视觉捷径，将动作与与任务无关的视觉特征关联，而不是与预期的任务语义关联。这些捷径阻碍了策略已经见过的元素的重新组合，即组合泛化。现有方法通过与任务相关的感知或针对性的数据多样化来缓解这种纠缠，但没有提供未见过的重新组合的明确机制，并且需要针对特定骨干网络的修改和重新训练。我们观察到，在这种重新组合下，VLA往往在全局定位上失败，而在熟悉的配置中保留了局部操控技能，能够接近正确目标。因此，我们提出了指称引导（ReGuide），这是一种无训练的封装器，给定来自定位模块的对象姿态，结合语义和几何重绑定，引导末端执行器进入受演示支持的指示参照物的配置，在这些配置中，冻结的策略可以恢复执行。在多个VLA骨干网络的模拟实验以及在真实机器人上的实验表明，ReGuide在组合变化下提高了成功率，分别达到56.8和75.0个百分点，同时保持了标准任务的表现。
cs.RO / 26 / 2609.38617
Dense Temporal Motion Retargeting for Legged Robots
腿部机器人密集时间运动重定向
Abstract
Legged robots can learn expressive whole-body skills from the motions of humans and animals. Due to the morphology gap between the source and the robot, however, the motion must be tailored to the dynamic properties of the robot. In particular, dynamic motions such as a jump require careful adjustment, since their timing and control are interdependent. We propose dense temporal motion retargeting (DTMR), which jointly optimizes timing and control within a single optimization, where dense means that the timing is adjusted for every control step. This dense formulation enables DTMR to deform only the parts of the motion that need a change in timing. The problem is solved with sampling-based model predictive control (MPC) in parallel on a GPU. We evaluate DTMR against baselines on two hours of human motion with four humanoid robots, where the results show that DTMR outperforms baseline methods, particularly on dynamic motions. We also show that allowing more temporal deformation yields more precise retargeting. We further compare DTMR with a baseline that optimizes the temporal dimension, where the result shows that DTMR retargets more precisely under the same deformation budget while being ~19x faster. Lastly, policies trained on our references transfer to a real humanoid robot.
Chinese Translation
腿部机器人可以从人类和动物的运动中学习富有表现力的全身技能。然而，由于源对象与机器人之间的形态差异，运动必须根据机器人的动态特性进行调整。特别是，像跳跃这样的动态运动需要仔细调整，因为其时机和控制是相互依赖的。我们提出了密集时间运动重定向（Dense Temporal Motion Retargeting, DTMR），该方法在单一优化过程中联合优化时机和控制，其中“密集”意味着在每个控制步骤中调整时机。这种密集的表述使得DTMR仅对需要改变时机的运动部分进行变形。该问题通过基于采样的模型预测控制（Model Predictive Control, MPC）在GPU上并行解决。我们在四个类人机器人上评估DTMR与基线方法在两小时人类运动数据上的表现，结果表明DTMR在动态运动方面优于基线方法。我们还展示了允许更多时间变形可以实现更精确的重定向。进一步地，我们将DTMR与优化时间维度的基线进行比较，结果显示在相同的变形预算下，DTMR能够更精确地进行重定向，同时速度约快19倍。最后，在我们的参考数据上训练的策略可以迁移到真实的类人机器人上。
cs.RO / 27 / 2609.38640
Yggdrasil: a Layer-First 3D Scene Graph for Real-Time Querying
Yggdrasil：一种面向层的实时查询3D场景图
Abstract
Robotic agents use 3D scene graphs (3DSG) to perform tasks ranging from scene understanding to scene interaction. Although an extensive body of work addresses scene graph generation, little attention has been paid to optimizing the graph for consumption, which leaves state-of-the-art perception pipelines to work around their own scene graph and to pay a latency cost that does not fit the real-time budget a perception loop runs on. We present Yggdrasil, the first 3D scene graph designed to be efficient for both generation and consumption: a layer-first hierarchical graph built from generic nodes, edges, and layers, which expresses the representations existing pipelines already produce, indoor or outdoor, flat or hierarchical, while natively answering the positional and semantic queries downstream tasks issue. Against a published DSG baseline, Yggdrasil answers queries up to $121\times$ faster, and every query we measure falls between 2 and 127 microseconds, three to five orders of magnitude inside the 200 microsecond keyframe budget a 3DSG consumer lives in, on both a workstation and embedded class device. We integrate Yggdrasil into three published pipelines spanning human trajectory prediction, object-goal navigation, and human-aware motion planning, where it removes up to 99% of the time each spends on its scene graph. The implementation, benchmark harness, and all three integrations are available online.
Chinese Translation
机器人代理使用3D场景图（3DSG）执行从场景理解到场景交互的各种任务。尽管已有大量研究致力于场景图生成，但对优化图的消费却关注甚少，这使得最先进的感知管道不得不绕过自身的场景图，并承担不符合感知循环实时预算的延迟成本。我们提出了Yggdrasil，这是第一个旨在在生成和消费方面都高效的3D场景图：一个基于通用节点、边和层构建的层优先层次图，能够表达现有管道已经生成的表示，无论是室内还是室外、平面还是层次结构，同时能够原生地回答下游任务提出的位置和语义查询。与已发布的DSG基线相比，Yggdrasil的查询速度提高了多达121倍，我们测量的每个查询时间都在2到127微秒之间，远低于3D场景图消费者所需的200微秒关键帧预算，在工作站和嵌入式设备上均表现良好。我们将Yggdrasil集成到三条已发布的管道中，涵盖人类轨迹预测、目标导航和人类感知运动规划，显著减少了每个管道在其场景图上花费的时间，最高可达99%。实现、基准测试工具和所有三项集成均可在线获取。
cs.RO / 28 / 2609.38653
TERRA: Terrain-Aware Reconstruction, Retargeting and Control for Musculoskeletal Locomotion
TERRA：面向地形的肌肉骨骼运动重建、重定向与控制
Abstract
Recent advances in musculoskeletal modeling and reinforcement learning have enabled muscle-actuated agents to reproduce increasingly complex human motions. Yet these capabilities remain largely confined to flat ground, in part because motion datasets rarely include aligned terrain geometry and because retargeting terrain interactions to complex musculoskeletal bodies is challenging. We present TERRA, an end-to-end pipeline for terrain-aware retargeting and control of musculoskeletal locomotion. From kinematic trajectories alone, TERRA combines terrain priors, estimated contacts, and negative free-space evidence to recover task-relevant support geometry. TERRA further considers anatomical, tendon-continuity, and contact constraints during retargeting. Using the resulting motion-terrain pairs from five datasets, we successfully train a single muscle-actuated control policy on 9.4 hours of diverse locomotion. Across reconstruction, retargeting, and held-out tracking benchmarks, TERRA improves terrain accuracy, sharply reduces anatomical and interaction violations, and achieves the highest observed completion rate over supported terrain families. Overall, TERRA provides a practical route from scene-less motion data to muscle-actuated locomotion over diverse non-flat terrain. Project website: https://cnai.epfl.ch/terra/
Chinese Translation
近年来，肌肉骨骼建模和强化学习的进展使得肌肉驱动的智能体能够再现越来越复杂的人类运动。然而，这些能力仍然主要局限于平坦地面，部分原因是运动数据集很少包含对齐的地形几何信息，并且将地形交互重定向到复杂的肌肉骨骼体上具有挑战性。我们提出了TERRA，一个用于肌肉骨骼运动的面向地形的重定向与控制的端到端管道。仅凭运动学轨迹，TERRA结合了地形先验、估计的接触点和负自由空间证据，以恢复与任务相关的支撑几何。TERRA在重定向过程中进一步考虑了解剖学、肌腱连续性和接触约束。利用来自五个数据集的运动-地形配对，我们成功地在9.4小时的多样化运动数据上训练了一个单一的肌肉驱动控制策略。在重建、重定向和保留跟踪基准测试中，TERRA提高了地形准确性，显著减少了解剖学和交互违规，并在支持的地形类别中实现了最高的完成率。总体而言，TERRA为从无场景运动数据到在多样化非平坦地形上进行肌肉驱动运动提供了一条实用的路径。项目网站：https://cnai.epfl.ch/terra/
cs.RO / 29 / 2609.38707
Residual Wrench Certification and Margin-Aware Control Synthesis for Aerial Physical Interaction
空中物理交互的残余扭矩认证与边际感知控制合成
Abstract
We develop a task-relative framework for certifying residual wrench authority after hover and contact loading in multirotors with bounded actuators. Using convex geometry, we derive signed margins for prescribed convex reserves, including Euclidean balls and weighted ellipsoids. We obtain computable reserve certificates from actuator-interiority bounds to support slack-maximizing allocation. To preserve the required reserve, we propose command projection onto a tightened feasible set. We then connect the available reserve to structured gain synthesis and certify a local tracking region that respects actuator limits. Within this region, we establish nominal exponential convergence and robust ultimate boundedness. For the prescribed task and morphology family, we show that the optimized octarotor attains a larger margin than the optimized hexarotor at equal total thrust. We evaluate the proposed framework in closed-loop simulations of sustained rigid-wall contact using a fixed-geometry octarotor and a variable-tilt quadrotor with matched installed thrust. The quadrotor's local certificate admits a higher normalized push limit for the prescribed task family. In tests beyond the realizability boundaries, we observe the predicted loss of authority through rotor-thrust limits for the octarotor and tilt-servo limits for the quadrotor.
Chinese Translation
我们开发了一个任务相关的框架，用于在具有有限执行器的多旋翼飞行器中认证悬停和接触加载后的残余扭矩权威。利用凸几何，我们推导出针对规定凸储备的有符号边际，包括欧几里得球和加权椭球。我们从执行器内部边界中获得可计算的储备证书，以支持最大松弛分配。为了保持所需的储备，我们提出了对收紧可行集的指令投影。然后，我们将可用储备与结构增益合成联系起来，并认证一个尊重执行器限制的局部跟踪区域。在该区域内，我们建立了名义上的指数收敛性和鲁棒的最终有界性。对于规定的任务和形态家族，我们展示了优化后的八旋翼在相同总推力下比优化后的六旋翼获得了更大的边际。我们在使用固定几何八旋翼和具有匹配安装推力的可变倾斜四旋翼的持续刚性墙接触的闭环仿真中评估了所提出的框架。四旋翼的局部证书允许在规定的任务家族中具有更高的归一化推力限制。在超出可实现性边界的测试中，我们观察到八旋翼的转子推力限制和四旋翼的倾斜伺服限制导致的权威损失的预测。
cs.RO / 30 / 2609.38709
CEER2: Directional and Tunable End-Effector and Root Compliance for Humanoid Loco-Manipulation
CEER2：用于类人机器人运动操控的方向性和可调末端执行器及根部柔顺性
Abstract
Humanoids are increasingly capable of tracking complex whole-body motions, but physical interaction introduces a different challenge. When a robot makes contact with a person or the environment, it needs to respond to external forces while preserving the motion needed for the task. This response can vary across directions in the end-effectors and on the body. For example, an end effector may need to accommodate contact force in one direction while maintaining motion accuracy in another, while the robot body may resist an external force or move with it. We present a compliance framework for humanoid loco-manipulation that combines directional and tunable end-effector (EE) compliance with selectable root compliance for external force rejection or force following. A hierarchical reinforcement learning controller modulates a fixed whole-body tracking policy through high-level EE and root commands, while interaction forces are estimated from proprioceptive history. Our simulation and real-world experiments on a humanoid demonstrate directional stiffness control, online stiffness adjustment, distinct root compliance, compliant manipulation, and collaborative carrying.
Chinese Translation
类人机器人在追踪复杂的全身运动方面越来越有能力，但物理交互带来了不同的挑战。当机器人与人或环境接触时，它需要在保持任务所需运动的同时，对外部力量做出响应。这种响应在末端执行器和机器人身体的不同方向上可能会有所不同。例如，末端执行器可能需要在一个方向上适应接触力，同时在另一个方向上保持运动精度，而机器人身体可能会抵抗外部力量或与之共同移动。我们提出了一种用于类人机器人运动操控的柔顺性框架，该框架结合了方向性和可调的末端执行器（EE）柔顺性，以及可选择的根部柔顺性，以实现外部力量的拒绝或跟随。一个分层强化学习控制器通过高层次的EE和根部指令调节固定的全身追踪策略，同时从本体感知历史中估计交互力量。我们的仿真和类人机器人在真实环境中的实验展示了方向性刚度控制、在线刚度调整、明显的根部柔顺性、柔顺操控和协作搬运。
cs.RO / 31 / 2609.38715
Path-Following Control and Terramechanics Analysis for Planetary Rovers Under Wheel-to-Wheel Traction Asymmetry
行星探测车在轮对牵引不对称下的路径跟踪控制与土力学分析
Abstract
This paper proposes a control strategy for path following that is model-free and relies solely on deceleration for skid-steering planetary rovers navigating deformable loose terrain under continuously imposed traction asymmetry. While conventional controllers that are based on kinematics frequently cause slip-sinkage entrapment by accelerating the wheels during path correction, the proposed approach prevents this failure by setting an upper limit on the maximum commanded velocity. Heading correction is achieved solely through the selective deceleration of the outer wheels, which are located on the outside of the turn, driving them into a negative slip regime to act as a mechanical anchor. The system was evaluated using the four-wheel independent-drive rover EX1 under an asymmetric wheel configuration with different left and right grouser heights that induces significant deviations from the path. Experimental results demonstrate that this deceleration-only control successfully suppresses accumulated lateral drift across various velocity regimes up to 0.7 m/s without causing sinkage. Crucially, direct force measurements from onboard multi-axis sensors provide important empirical evidence of the underlying terramechanics, proving that the targeted deceleration establishes dynamic load equalization across the chassis and completely restores the native thrust capability of the opposite driving wheel.
Chinese Translation
本文提出了一种无模型的路径跟踪控制策略，该策略仅依赖于减速，适用于在持续施加的牵引不对称下，导航于可变形松散地形的滑转转向行星探测车。传统的基于运动学的控制器在路径修正过程中常常通过加速车轮导致滑陷困境，而所提出的方法通过对最大指令速度设定上限来防止这种失败。航向修正仅通过选择性减速外侧车轮实现，这些车轮位于转弯的外侧，使其进入负滑移状态，充当机械锚点。该系统在不对称车轮配置下使用四轮独立驱动探测车EX1进行评估，左侧和右侧的螺纹高度不同，导致路径偏离显著。实验结果表明，该仅减速控制成功抑制了在不同速度范围内（最高可达0.7 m/s）的累积横向漂移，而未导致沉陷。重要的是，来自机载多轴传感器的直接力测量提供了重要的实证证据，证明了基础土力学的原理，证明了目标减速在底盘上建立了动态载荷均衡，并完全恢复了对侧驱动轮的原生推力能力。
cs.RO / 32 / 2609.38719
FlapKAD: A Simulation Dataset of Coupled Wing Kinematics and Aerodynamic Dynamics for Flapping-Wing Aerial Vehicles
FlapKAD：一种耦合翼运动学与气动动态的拍打翼空中飞行器模拟数据集
Abstract
Experimental investigation and modeling of flapping-wing aerial vehicles are limited by the scarcity of large-scale records that temporally align wing kinematics, aerodynamic responses, and flight states. Existing datasets are often limited in scale and affected by measurement noise and temporal misalignment between rapidly varying wing motion and the associated dynamic response, particularly during high-frequency flapping. We introduce FlapKAD, an episode-structured simulation dataset comprising 2,000 rigid-wing flight episodes and 720,152 valid time steps, with bilateral wing kinematics, aerodynamic responses, and flight states recorded synchronously within a common clock. FlapKAD supports a unified bidirectional sequence-prediction benchmark constructed from the same temporally aligned episodes. The forward task predicts future vertical force coefficients and body vertical velocity from histories of realized flap and twist angles, whereas the inverse task reconstructs future flap- and twist-angle trajectories from the corresponding response histories. A benchmark of eight representative time-series architectures across multiple prediction horizons reveals direction- and horizon-dependent model behavior, systematically higher reconstruction errors for twist angle than for flap angle, and no consistent advantage from increased architectural complexity. FlapKAD provides a reproducible dataset and benchmark for studying coupled wing-kinematic, aerodynamic, and flight-state dynamics in flapping-wing aerial vehicles.
Chinese Translation
拍打翼空中飞行器的实验研究和建模受到大规模记录稀缺的限制，这些记录需要在时间上对齐翼运动学、气动响应和飞行状态。现有数据集通常规模有限，并受到测量噪声和快速变化的翼运动与相关动态响应之间的时间错位的影响，尤其是在高频拍打期间。我们引入了FlapKAD，这是一个结构化的模拟数据集，包含2000个刚性翼飞行事件和720,152个有效时间步，这些事件中双翼运动学、气动响应和飞行状态在一个共同的时钟下同步记录。FlapKAD支持一个统一的双向序列预测基准，该基准由相同时间对齐的事件构建。正向任务从已实现的拍打和扭转角度的历史中预测未来的垂直力系数和机身垂直速度，而逆向任务则从相应的响应历史中重构未来的拍打和扭转角度轨迹。对八种代表性时间序列架构在多个预测视野下的基准测试显示，模型行为依赖于方向和视野，扭转角的重构误差系统性高于拍打角，且增加架构复杂性并未带来一致的优势。FlapKAD提供了一个可重复的数据集和基准，用于研究拍打翼空中飞行器中耦合的翼运动学、气动和飞行状态动态。
cs.RO / 33 / 2609.38730
Bounded Channel-Adaptive Spectral Learning for Forward-Consistent Inverse Flapping-Wing Aerodynamics
有界通道自适应谱学习用于前向一致的反向拍翼气动学
Abstract
Flapping-wing vehicles regulate aerodynamic forces and moments through coordinated variations in stroke, deviation, and pitch motion. Because the resulting loads depend on both the instantaneous wing configuration and its preceding motion history, recovering suitable wing kinematics from a desired aerodynamic trajectory is a challenging inverse problem. Existing sequence models capture temporal dependencies, while spectral methods can exploit the periodic structure of flapping motion. However, unrestricted frequency-domain augmentation may interfere with temporal representations and produce inconsistent corrections across kinematic variables and prediction horizons. We propose the Bounded Channel-Adaptive Spectral Residual Gated Recurrent Unit (BCS-GRU), which retains recurrent temporal prediction as its primary representation and restricts spectral information to a controlled output-specific correction. We further introduce a causally aligned forward-consistency objective that evaluates predicted kinematics through a separately trained and frozen aerodynamic surrogate. Experiments under a unified episode-level protocol show that BCS-GRU improves inverse prediction over recurrent and adaptive-spectral baselines, with greater benefits at longer prediction horizons. Forward-consistent fine-tuning further improves surrogate-based aerodynamic consistency while maintaining mean kinematic accuracy. These results demonstrate that controlled spectral correction and forward-consistent learning provide an effective framework for history-aware inverse modelling of flapping-wing aerodynamics.
Chinese Translation
拍翼飞行器通过协调变化的击打、偏转和俯仰运动来调节气动力和力矩。由于所产生的载荷依赖于瞬时翼型配置及其先前的运动历史，从期望的气动轨迹中恢复合适的翼运动学是一个具有挑战性的反向问题。现有的序列模型捕捉时间依赖性，而谱方法则可以利用拍动运动的周期性结构。然而，不受限制的频域增强可能会干扰时间表示，并在运动学变量和预测范围之间产生不一致的修正。我们提出了有界通道自适应谱残差门控递归单元（BCS-GRU），该模型将递归时间预测作为其主要表示，并将谱信息限制为受控的特定输出修正。我们进一步引入了一个因果对齐的前向一致性目标，通过单独训练和冻结的气动代理评估预测的运动学。在统一的情节级协议下的实验表明，BCS-GRU在反向预测方面优于递归和自适应谱基线，且在较长的预测范围内获得更大的收益。前向一致的微调进一步提高了基于代理的气动一致性，同时保持了平均运动学的准确性。这些结果表明，受控的谱修正和前向一致学习为拍翼气动学的历史感知反向建模提供了有效的框架。
cs.RO / 34 / 2609.38737
AIfred: Augmented Learning through Functional Robotic Embodiment at the Desk
AIfred：通过功能性机器人具身化增强桌面学习
Abstract
Desk-based learning and creative activities benefit from handwritten engagement. However, current generative AI tools deliver guidance through a separate screen, creating a gap between where users think and where assistance appears. To address this, in this work we design AIfred, a desk-based robotic arm with a projector mounted at the end-effector that places AI-generated guidance alongside handwritten work. AIfred combines workspace perception, context-aware content generation, and robot-mediated projection to support math assignments, image generation, and drawing tasks. In a user study (n = 36), we compared AIfred against ChatGPT (GPT-5.6 Luna) running on a laptop. Both tools performed comparably while assistance was available during the math assignment (6.7 vs. 7.3/10, p = .41), but AIfred improved short-term learning transfer by 60% once assistance was withdrawn (7.0 vs. 4.4/10, p = .003). In addition, independent art and design professors ranked drawings produced with AIfred better in 33 of 36 cases. Our findings indicate that spatially co-located AI assistance benefits tasks whose guidance shares a spatial frame with the work.
Chinese Translation
基于桌面的学习和创意活动受益于手写参与。然而，当前的生成性人工智能工具通过单独的屏幕提供指导，这在用户思考的地方与出现的帮助之间造成了间隙。为了解决这个问题，我们设计了AIfred，一个桌面机器人臂，末端效应器上安装了投影仪，将人工智能生成的指导与手写工作并排放置。AIfred结合了工作空间感知、上下文感知内容生成和机器人介导的投影，以支持数学作业、图像生成和绘画任务。在一项用户研究中（n = 36），我们将AIfred与在笔记本电脑上运行的ChatGPT（GPT-5.6 Luna）进行了比较。在数学作业期间，两种工具的表现相当（6.7 vs. 7.3/10, p = .41），但一旦撤回帮助，AIfred提高了60%的短期学习迁移（7.0 vs. 4.4/10, p = .003）。此外，独立的艺术与设计教授在36个案例中有33个案例中认为使用AIfred生成的绘画更好。我们的研究结果表明，空间上共置的人工智能辅助有利于指导与工作共享空间框架的任务。
cs.RO / 35 / 2609.38760
HALO: Heterogeneous Allocation Via Localized Observations for the Vehicle Routing Problem
HALO：通过局部观察实现异构分配的车辆路径规划问题
Abstract
Scalable robotic fleets have become increasingly popular for various applications such as package delivery, warehouse management, and military operations. Prior fleet control algorithms solve centralized routing problems with up to $1{,}000$ tasks in controlled environments, yet they fail to consider realistic constraints such as limited observation and communication ranges typical of decentralized fleets. Thus, deploying existing fleet control algorithms into real-world settings is currently infeasible. To tackle this, we propose Heterogeneous Allocation via Localized Observations (HALO) to solve the Vehicle Routing Problem (VRP). HALO is a hybrid method that splits the VRP into allocation and routing portions to provide onboard, real-time solutions to robots in dynamic environments. During the allocation phase, HALO utilizes a heterogeneous graph neural network framework with unique message passing layers to explicitly separate the learning of spatial distributions and task-to-robot compatibility. Evaluation results on a partially observable, online variant of the VRP show HALO significantly outperforms the heuristic baseline while maintaining similar solution quality to an all-knowing offline variant of HALO. While HALO is explicitly designed for partially observable environments, it imposes no strict upper bound on the observation space allowing us to test HALO on the traditional static, single-depot VRP. Here, HALO outperforms state-of-the-art architectures strictly optimized for the static variant of the VRP by up to $14.06\%$. Throughout all testing, this framework maintains the quickest execution times which emphasizes its potential for large-scale, real-time deployment.
Chinese Translation
可扩展的机器人车队在包裹递送、仓库管理和军事行动等各种应用中变得越来越受欢迎。之前的车队控制算法在受控环境中解决了最多 $1{,}000$ 个任务的集中式路径规划问题，但未能考虑去中心化车队中典型的有限观察和通信范围等现实约束。因此，将现有的车队控制算法应用于现实世界的场景目前是不可行的。为此，我们提出了通过局部观察实现异构分配（HALO）的方法来解决车辆路径规划问题（VRP）。HALO 是一种混合方法，将 VRP 分为分配和路径规划两个部分，以便为动态环境中的机器人提供实时的机载解决方案。在分配阶段，HALO 利用具有独特消息传递层的异构图神经网络框架，明确分离空间分布的学习和任务与机器人之间的兼容性。对部分可观察的在线变体 VRP 的评估结果表明，HALO 显著优于启发式基线，同时保持与全知离线变体 HALO 相似的解决方案质量。虽然 HALO 明确设计用于部分可观察环境，但它对观察空间没有严格的上限，这使我们能够在传统的静态单仓库 VRP 上测试 HALO。在这里，HALO 的表现优于专门针对静态变体 VRP 优化的最先进架构，提升幅度高达 $14.06 ext{%}$。在所有测试中，该框架保持了最快的执行时间，强调了其在大规模实时部署中的潜力。
cs.RO / 36 / 2609.38824
PhaseSync-Exo: Human Clock Anchored Reference Adaptation for Dynamic Gait Tracking
PhaseSync-Exo：基于人类时钟的参考适应用于动态步态跟踪
Abstract
Human-aware exoskeleton walking requires reconstructing gait, tracking diverse motions under dynamic constraints, and preserving human timing. We present PhaseSync-Exo, which combines two-IMU CNN-Transformer reconstruction, factorized amplitude-cadence retargeting with curriculum-trained recurrent control, and a human-clock-anchored adapter (HCA). HCA combines human-clock attraction with robot-relative feedback to adjust reference rate while preserving forward progression and continuity. The reconstruction module achieves a mean absolute error (MAE) of 3.24 degrees on a held-out recording, and the frozen tracking policy completes 356 of 357 amplitude-frequency trials. Two complementary dynamic comparisons isolate HCA's timing benefit without retraining. Against robot-relative correction, HCA reduces human-clock phase MAE from 95.55 degrees to 10.99 degrees, limiting reference drift. When human and delivered phases initially differ, it reduces phase MAE from 72.02 degrees to 13.90 degrees versus fixed-clock continuation. Compared with immediate phase reset, HCA reduces transient reference-tracking hip RMSE by 22.7% without reference jumps. All 420 timing rollouts complete without falls. These simulation results support continuous phase acquisition and sustained alignment to an independent human clock.
Chinese Translation
人机共存的外骨骼行走需要重建步态，在动态约束下跟踪多样化的运动，并保持人类的时间感。我们提出了PhaseSync-Exo，它结合了双IMU CNN-Transformer重建、分解幅度-步频重定向与课程训练的递归控制，以及一个基于人类时钟的适配器（HCA）。HCA结合了人类时钟的吸引力与机器人相对反馈，以调整参考速率，同时保持向前推进和连续性。重建模块在保留数据集上实现了3.24度的平均绝对误差（MAE），而冻结的跟踪策略在357个幅度-频率试验中完成了356个。两个互补的动态比较在不重新训练的情况下隔离了HCA的时间优势。与机器人相对修正相比，HCA将人类时钟相位的MAE从95.55度降低到10.99度，限制了参考漂移。当人类和传递相位最初不同，HCA将相位MAE从72.02度降低到13.90度，相较于固定时钟的延续。与立即相位重置相比，HCA在没有参考跳跃的情况下将瞬态参考跟踪的髋部均方根误差（RMSE）降低了22.7%。所有420次时间滚动均未发生跌倒。这些仿真结果支持持续的相位获取和与独立人类时钟的持续对齐。
cs.RO / 37 / 2609.38846
Coral Grow-out Robotic Assessment System (CGRAS): Scaling Coral Recruit Monitoring Through Robotics and Computer Vision
珊瑚生长机器人评估系统（CGRAS）：通过机器人技术和计算机视觉扩展珊瑚幼体监测
Abstract
Climate change is the largest threat to coral reefs, with increasing global impacts accelerating the need for scalable reef restoration technologies. Large-scale reef restoration depends on the mass production of corals, such as through coral aquaculture. Coral seeding with recruits grown in aquaculture facilities is a feasible restoration approach, but effective production requires consistent, high-frequency monitoring of tens of thousands of macroscopic (0.5-2mm diameter) recruits, making conventional manual assessment prohibitively labor-intensive. To address this monitoring bottleneck, we introduce the Coral Grow-out Robotic Assessment System (CGRAS) which combines robotic imaging and computer vision to automate data acquisition, perform multi-species detection and counting of corals, and evaluate coral health. CGRAS automatically extracts coral growth, survival and spatial distribution metrics, with the aim of providing timely feedback to operators for optimizing production, grow-out and deployment workflow processes. We demonstrate CGRAS in a large aquaculture facility on standardized coral settlement tiles, reducing the time and labor costs by a factor of 9.6 as compared to manual monitoring, whilst achieving 96.4% agreement for Acropora kenti corals relative to expert counts.
Chinese Translation
气候变化是珊瑚礁面临的最大威胁，全球影响的加剧加速了可扩展的礁石修复技术的需求。大规模的礁石修复依赖于珊瑚的大规模生产，例如通过珊瑚水产养殖。利用在水产养殖设施中生长的幼体进行珊瑚播种是一种可行的修复方法，但有效的生产需要对成千上万的宏观（直径0.5-2毫米）幼体进行一致的高频监测，这使得传统的人工评估工作量过大，难以承受。为了解决这一监测瓶颈，我们提出了珊瑚生长机器人评估系统（CGRAS），该系统结合了机器人成像和计算机视觉，自动化数据采集，执行多物种的珊瑚检测和计数，并评估珊瑚健康。CGRAS自动提取珊瑚的生长、存活和空间分布指标，旨在为操作人员提供及时反馈，以优化生产、成长和部署工作流程。我们在一个大型水产养殖设施中对标准化珊瑚沉积瓷砖展示了CGRAS，相比于人工监测，时间和劳动成本降低了9.6倍，同时对于Acropora kenti 珊瑚的计数与专家计数的吻合率达到96.4%。
cs.RO / 38 / 2609.38852
Locomotion-Grounded Humanoid Soccer: Task-Gated Reinforcement Learning of a Multi-Directional Kicking Library
基于运动的类人足球：多方向踢球库的任务门控强化学习
Abstract
Recent humanoid soccer systems make motion tracking the substrate and derive locomotion from it, typically by steering a motion-reference anchor toward the ball. This yields strong shooting results, but locomotion is trained only on the narrow, deterministic command distribution ball approach induces, never evaluated as a capability in its own right. We invert the stack: a general, command-conditioned locomotion policy is trained first as the substrate, and N motion-guided kicking skills are added on top as task-gated layers, so the reachable gait space is set by the locomotion curriculum rather than any reference clip. Because every skill starts from and returns to this same commandable state, locomotion also becomes a composition hub (O(N) transitions rather than O(N^2)), and post-strike stabilisation is handed back to the trained controller rather than scripted per clip. We instantiate this on a 29-DoF Unitree G1 with seven retargeted kicking skills spanning 259.5 degrees of nominal aim direction, including lateral, rearward and weak-foot strikes a single forward-facing reference cannot express, and report shooting accuracy alongside command-tracking, terrain and push-recovery results with the full skill library attached, an axis prior humanoid soccer systems do not report. The library is validated on hardware across forward, lateral, rearward and commanded approaches.
Chinese Translation
近期的类人足球系统使得运动跟踪成为基础，并从中导出运动，通常是通过将运动参考锚点引导至球的位置。这种方法产生了良好的射门效果，但运动训练仅限于狭窄的、确定性的指令分布，球的接近仅被视为一种能力，而未被单独评估。我们反转了这一过程：首先训练一个通用的、基于指令的运动策略作为基础，然后在其上添加 N 种运动引导的踢球技能作为任务门控层，从而使得可达的步态空间由运动课程设定，而非任何参考片段。由于每项技能都从同一可指令状态出发并返回，因此运动也成为一个组合中心（O(N) 转换而非 O(N^2)），而击球后的稳定化则交由训练好的控制器处理，而非按片段脚本化。我们在一台 29 自由度的 Unitree G1 上实现了这一方法，添加了七种重新定向的踢球技能，涵盖了 259.5 度的名义瞄准方向，包括侧向、后向和弱脚射门，这些是单一前向参考无法表达的，并报告了射门准确率以及与完整技能库相关的指令跟踪、地形和推力恢复结果，这是类人足球系统未曾报告的一个轴心。该库在硬件上经过验证，涵盖了前向、侧向、后向和指令接近的情况。
cs.RO / 39 / 2609.38855
Online Evolution Strategy for Flow-Matching VLA Policies via Self-Supervised Trajectory Distribution Optimization
基于自监督轨迹分布优化的流匹配VLA策略在线进化策略
Abstract
Vision-Language-Action (VLA) models based on generative frameworks, such as Flow Matching, have recently achieved impressive performance in robotic manipulation. Unlike deterministic policies, Flow Matching enables VLA models to learn conditional action trajectory distributions, where latent noise vectors induce different actions under the same task scenario. However, we observe that these distributions are often ill-formed, with successful and failed behaviors coexisting while considerable probability mass remains in unfavorable regions. To this end, we propose Online-ES, an online adaptation framework for Flow Matching VLAs based on Evolution Strategy (ES), which refines the learned action trajectory distribution through interaction feedback. Instead of pruning the latent noise space, our method performs evolutionary exploration directly in the action trajectory space, where diverse trajectories generated by Flow Matching provide candidate solutions for adaptation. By perturbing sampled trajectories and evaluating their execution outcomes, we derive a self-supervised MSE objective that transfers the evolution direction from trajectory space into model parameter space. Mathematically, we prove that the proposed objective provides an unbiased estimator of the optimal evolution direction. Moreover, we also incorporate failure experiences as negative feedback to regularize the evolution direction, steering the policy away from previously explored failure regions. Experiments in both simulation and real-world environments demonstrate that Online-ES achieves policy improvement comparable to reinforcement fine-tuning, without learning a value model or computing advantages.
Chinese Translation
基于生成框架的视觉-语言-动作（VLA）模型，如流匹配（Flow Matching），最近在机器人操作中取得了令人瞩目的性能。与确定性策略不同，流匹配使得VLA模型能够学习条件动作轨迹分布，其中潜在噪声向量在相同任务场景下诱导出不同的动作。然而，我们观察到这些分布往往形态不佳，成功和失败的行为共存，同时在不利区域仍然存在相当的概率质量。为此，我们提出了Online-ES，这是一种基于进化策略（Evolution Strategy, ES）的流匹配VLA的在线适应框架，通过交互反馈来细化学习到的动作轨迹分布。我们的方法不是修剪潜在噪声空间，而是直接在动作轨迹空间中进行进化探索，其中流匹配生成的多样化轨迹为适应提供了候选解。通过扰动采样轨迹并评估其执行结果，我们推导出一个自监督均方误差（MSE）目标，将进化方向从轨迹空间转移到模型参数空间。从数学上讲，我们证明了所提出的目标提供了最优进化方向的无偏估计。此外，我们还将失败经验作为负反馈纳入进化方向的正则化，指导策略远离先前探索的失败区域。在模拟和真实环境中的实验表明，Online-ES实现的策略改进可与强化学习微调相媲美，而无需学习价值模型或计算优势。
cs.RO / 40 / 2609.38857
Plan-Conditioned Imitation for Robust Object Retrieval under Self-Occlusion in Dense Clutter
基于计划的模仿学习在密集杂乱环境下自遮挡的稳健物体检索
Abstract
Retrieving objects from dense clutter requires rearrangement during which the manipulator can occlude objects while moving them. Repeated arm withdrawals to restore visibility interrupt execution. We introduce TRACE, a plan-conditioned imitation framework for retrieval under self-occlusion. A single unoccluded observation initializes a digital twin, where a privileged teacher generates a fixed nominal rollout. A recurrent student combines local rollout context, partial object observations, and proprioception to select actions that can correct deviations from the prediction. Behavior cloning initializes the student; DAgger refines it with teacher labels on student-visited states. The rollout remains fixed throughout execution, so the deployed student needs neither online teacher queries nor additional simulator rollouts during pushing. On 511 simulation test scenes, TRACE achieves 90.7% success versus 43.4% for nominal replay and 96.7% for the privileged closed-loop teacher. At a matched 26,373-label budget, student-state supervision achieves 87.8% versus 66.7% for expert-only cloning, demonstrating gains beyond additional labels. On a UR5e, TRACE achieves 90.0% success versus 95.0% for the closed-loop teacher, while reducing total execution time from 192.7 s to 67.3 s. It avoids the teacher's 16.8 sensing-related arm retractions per trial during pushing, retaining a final withdrawal for graspability evaluation. Code and data will be released at: https://trace-retrieval.github.io.
Chinese Translation
从密集杂乱中检索物体需要在搬运过程中进行重新排列，此时操作者可能会遮挡物体。重复的手臂撤回以恢复可见性会中断执行。我们提出了TRACE，一个用于自遮挡下物体检索的基于计划的模仿学习框架。单个未遮挡的观察初始化一个数字双胞胎，其中一个特权教师生成固定的名义回放。一个递归学生结合局部回放上下文、部分物体观察和本体感知，以选择能够纠正偏离预测的动作。行为克隆初始化学生；DAgger通过教师标签对学生访问的状态进行细化。回放在整个执行过程中保持固定，因此部署的学生在推动过程中无需在线教师查询或额外的模拟器回放。在511个模拟测试场景中，TRACE的成功率为90.7%，而名义回放为43.4%，特权闭环教师为96.7%。在匹配的26,373标签预算下，学生状态监督的成功率为87.8%，而仅专家克隆为66.7%，显示出超出额外标签的收益。在UR5e上，TRACE的成功率为90.0%，而闭环教师为95.0%，同时将总执行时间从192.7秒减少到67.3秒。在推动过程中，它避免了教师每次试验中16.8次与感知相关的手臂撤回，仅保留一次最终撤回以评估抓取能力。代码和数据将在以下网址发布：https://trace-retrieval.github.io。
cs.RO / 41 / 2609.38862
Efficient Multi-Modal Planning with Reward-Guided Preference Optimization for Autonomous Driving
基于奖励引导偏好优化的高效多模态规划在自动驾驶中的应用
Abstract
Safe and efficient trajectory planning is essential in autonomous driving. However, existing end-to-end approaches often fall short in both computational efficiency and safety guarantees. Methods based on imitation learning suffer from causal confusion, while rule-based scoring approaches often incur heavy computational overhead and suffer from objective misalignment. Additionally, preference-based methods rely on strict pairwise annotations, limiting data utilization. To overcome these limitations, we propose EMPlan, an efficient multi-modal trajectory planning method powered by reward-guided fine-tuning. We design a hybrid architecture that combines sparse anchors with an offset refinement module for efficient multi-modal trajectory prediction. Sparse anchors provide coarse trajectory candidates with low latency, which are subsequently refined by the offset module for higher prediction accuracy. To enhance safety without incurring additional inference costs, we adopt a two-stage training paradigm consisting of pretraining and reward-guided fine-tuning. During fine-tuning, we leverage rule-based reward signals and unpaired preference supervision to refine the pretrained policy toward safer trajectory selection. We evaluate EMPlan on the non-reactive NAVSIM benchmark, where it strikes a favorable balance between planning accuracy and efficiency, demonstrating superior performance under real-time constraints.
Chinese Translation
安全高效的轨迹规划在自动驾驶中至关重要。然而，现有的端到端方法在计算效率和安全保障方面往往存在不足。基于模仿学习的方法受到因果混淆的影响，而基于规则的评分方法则常常带来较大的计算开销，并且存在目标不一致的问题。此外，基于偏好的方法依赖于严格的成对注释，限制了数据的利用。为了解决这些局限性，我们提出了EMPlan，一种基于奖励引导微调的高效多模态轨迹规划方法。我们设计了一种混合架构，将稀疏锚点与偏移细化模块相结合，以实现高效的多模态轨迹预测。稀疏锚点提供低延迟的粗略轨迹候选，随后通过偏移模块进行细化，以提高预测精度。为了在不增加额外推理成本的情况下增强安全性，我们采用了包括预训练和奖励引导微调的两阶段训练范式。在微调过程中，我们利用基于规则的奖励信号和无配对偏好监督来细化预训练策略，以实现更安全的轨迹选择。我们在非反应性NAVSIM基准上评估了EMPlan，结果表明其在规划精度和效率之间达成了良好的平衡，并在实时约束下表现出优越的性能。
cs.RO / 42 / 2609.38873
DODGER: Safety-Guided Reinforcement Learning for Robot Navigation Among Dynamic Obstacles
DODGER：基于安全指导的强化学习在动态障碍物中进行机器人导航
Abstract
Robots operating in human-centered environments must safely navigate among multiple dynamic obstacles to avoid collisions with people and surrounding infrastructure. Control barrier functions (CBFs) provide an effective mechanism for safety filtering, and recent CBF-based reinforcement learning (RL) methods embed such safety information into learned policies. However, executing only safety-filtered actions during training can restrict policy exploration, a limitation that becomes particularly consequential in dynamic scenes where safety depends on relative robot-obstacle motion. We propose DODGER, a safety-guided RL framework that directly executes policy-generated actions to drive training rollouts while using CBF-filtered references and constraint violations to shape the policy toward collision-avoidance behavior. We evaluate DODGER through a Dubins-car safety analysis and demonstrate goal-directed navigation among multiple dynamic obstacles in full-order humanoid simulation and real-world humanoid experiments using LiDAR-based perception, without a runtime safety filter.
Chinese Translation
在以人为中心的环境中运行的机器人必须在多个动态障碍物之间安全导航，以避免与人和周围基础设施的碰撞。控制障碍函数（CBFs）提供了一种有效的安全过滤机制，最近的基于CBF的强化学习（RL）方法将这种安全信息嵌入到学习的策略中。然而，在训练过程中仅执行安全过滤后的动作可能会限制策略的探索，这一限制在动态场景中尤为显著，因为安全性依赖于机器人与障碍物之间的相对运动。我们提出了DODGER，一个安全指导的RL框架，它直接执行策略生成的动作以推动训练过程，同时使用CBF过滤的参考和约束违规来塑造策略以实现避碰行为。我们通过Dubins车的安全分析评估DODGER，并展示了在全秩人形机器人仿真和基于LiDAR感知的真实人形实验中，在多个动态障碍物之间的目标导向导航，而无需运行时安全过滤。
cs.RO / 43 / 2609.38886
Benchmarking and Enhancing Skill-Level Memory for Partially Observable Robotic Manipulation
部分可观测机器人操控的技能水平记忆基准与增强
Abstract
Recent advances in robot learning have enabled manipulation policies to perform increasingly diverse tasks and generalize across environments. However, reliable execution often depends on hidden task states that cannot be determined from current observations alone, making interaction history essential. We introduce $HIDE$, a benchmark for evaluating manipulation memory under partial observability. HIDE comprises 15 tasks covering repetition counting, historical-state recall, and execution-progress tracking, with randomized initial configurations and decision points where similar observations require different actions depending on prior events. We further propose $SEEK$, a framework combining three complementary memory mechanisms to retain historical evidence and track execution state. Evaluations reveal substantial limitations in existing policies on HIDE, while memory augmentation improves task success in both simulation and real-world experiments. Individual mechanisms benefit some tasks but can degrade others; their combination achieves the highest average success rate on HIDE among the evaluated configurations. These findings highlight the importance of maintaining internal representations of hidden task states and matching memory design to task-specific information requirements.
Chinese Translation
近期机器人学习的进展使得操控策略能够执行越来越多样化的任务，并在不同环境中进行泛化。然而，可靠的执行往往依赖于无法仅通过当前观察确定的隐藏任务状态，这使得交互历史变得至关重要。我们引入了 $HIDE$，一个用于评估部分可观测条件下操控记忆的基准。$HIDE$ 包含 15 个任务，涵盖重复计数、历史状态回忆和执行进度跟踪，具有随机初始配置和决策点，在这些点上，相似的观察需要根据先前事件采取不同的行动。我们进一步提出了 $SEEK$，一个结合三种互补记忆机制的框架，以保留历史证据并跟踪执行状态。评估结果揭示了现有策略在 $HIDE$ 上的显著局限性，而记忆增强在模拟和现实世界实验中均提高了任务成功率。各个机制对某些任务有益，但可能会降低其他任务的表现；它们的组合在评估的配置中实现了 $HIDE$ 上最高的平均成功率。这些发现突显了保持隐藏任务状态的内部表示的重要性，以及将记忆设计与任务特定信息需求相匹配的必要性。
cs.RO / 44 / 2609.38890
PRICE the Action Chunks: Physical Relational Credit Assignment for Embodied Reinforcement Learning
价格行动块：面向具身强化学习的物理关系信用分配
Abstract
Outcome-based reinforcement learning (RL) post-trains vision--language--action policies using terminal success signals, but assigns the same trajectory-level advantage to every action chunk. A failed episode can thus penalize useful early actions as if they caused the failure. Existing approaches seek finer-grained feedback through learned evaluators, adding task-specific supervision or additional model training. We explore, for the first time to our knowledge, whether physical relations across trajectories can provide action-chunk credit in embodied RL from terminal outcomes alone, without an auxiliary evaluator. The key insight is that rollouts reaching corresponding physical situations can serve as references for one another: their terminal outcomes provide evidence for assessing local progress. We introduce Physical Relations for Inferring Credit from Episodes(PRICE), with two components: (i) a physical relational graph that pools current and historical outcomes at corresponding chunk boundaries to estimate success potentials; and (ii) confidence-gated credit assignment that uses changes in these potentials to refine trajectory-level supervision. Our analysis connects oracle potential changes to the terminal-success objective and provides a finite-sample directional bound for outcome-independent evidence pools. Independent continuation tests show that PRICE's retained credits align with local progress, while experiments on LIBERO, RoboTwin 2.0, and real robots demonstrate improved task success over outcome-based baselines and faster learning.
Chinese Translation
基于结果的强化学习（RL）通过终端成功信号对视觉-语言-行动策略进行后训练，但对每个行动块分配相同的轨迹级优势。因此，一个失败的情节可能会惩罚有用的早期行动，仿佛它们导致了失败。现有方法通过学习评估器寻求更细粒度的反馈，增加任务特定的监督或额外的模型训练。我们首次探索，是否仅通过终端结果，物理关系可以在具身强化学习中为行动块提供信用分配，而不需要辅助评估器。关键的见解是，达到相应物理情境的回放可以相互作为参考：它们的终端结果为评估局部进展提供证据。我们引入了从情节推断信用的物理关系（PRICE），其包含两个组成部分：（i）一个物理关系图，在相应的块边界汇集当前和历史结果，以估计成功潜力；（ii）信心门控信用分配，利用这些潜力的变化来细化轨迹级监督。我们的分析将神谕潜力变化与终端成功目标联系起来，并为结果无关的证据池提供有限样本方向界限。独立的延续测试表明，PRICE保留的信用与局部进展一致，而在LIBERO、RoboTwin 2.0和真实机器人上的实验则展示了相较于基于结果的基线更高的任务成功率和更快的学习速度。
cs.RO / 45 / 2609.38905
EmbodiRSI: Recursive Self-Improvement for Data-Efficient Robot Adaptation
EmbodiRSI：用于数据高效机器人适应的递归自我改进
Abstract
Adapting robot manipulation policies to new tasks and environments remains highly data-intensive, while the data needed for further improvement depends on the policy's current capabilities and failure modes. We introduce EmbodiRSI, an agentic system for recursive self-improvement (RSI) in a real-to-sim-to-real setting, where task-specific simulations are constructed from target deployment scenarios and used as low-cost environments for iterative policy improvement before transfer back to the physical world. EmbodiRSI uses policy execution feedback to guide subsequent experience acquisition and policy updates. Two complementary mechanisms close this loop: Collaborative Error Correction generates agent-assisted corrective trajectories from policy-reached states, while Adaptive Data Collection directs expert demonstration generation toward the current policy's weaknesses. The task-specific simulation serves as a reusable workspace for policy warm-up, repeatable evaluation, failure diagnosis, and targeted data generation across successive RSI rounds. Across three tabletop environments and 14 subtasks, EmbodiRSI increases scene-balanced autonomous simulation success from 50.4% to 83.5% over two RSI updates. With 400 adaptive simulated trajectories and only ten real-world refinement trajectories per subtask, EmbodiRSI achieves 83.1% scene-balanced autonomous real-world success, compared with 75.0% for adaptation using 200 real-world demonstrations per subtask. These results demonstrate that feedback-driven recursive improvement in deployment-specific simulations can enable data-efficient adaptation of embodied policies to physical environments.
Chinese Translation
将机器人操作策略适应于新任务和新环境仍然需要大量数据，而进一步改进所需的数据依赖于策略当前的能力和失败模式。我们提出了EmbodiRSI，这是一个在真实-模拟-真实环境中进行递归自我改进（RSI）的智能系统，其中根据目标部署场景构建特定任务的模拟，并将其用作低成本环境，以便在转移回物理世界之前进行迭代的策略改进。EmbodiRSI利用策略执行反馈来指导后续经验获取和策略更新。两个互补机制闭合了这一循环：协作错误修正从策略达到的状态生成代理辅助的修正轨迹，而自适应数据收集则将专家演示生成的方向指向当前策略的弱点。特定任务的模拟作为一个可重用的工作空间，用于策略预热、可重复评估、失败诊断和在连续的RSI轮次中进行针对性数据生成。在三个桌面环境和14个子任务中，EmbodiRSI在两次RSI更新中将场景平衡的自主模拟成功率从50.4%提高到83.5%。在每个子任务中，使用400条自适应模拟轨迹和仅十条真实世界的精炼轨迹，EmbodiRSI实现了83.1%的场景平衡自主真实世界成功率，而使用每个子任务200条真实世界演示的适应成功率为75.0%。这些结果表明，在部署特定的模拟中，基于反馈的递归改进可以实现对物理环境中具身策略的数据高效适应。
cs.RO / 46 / 2609.38928
Local-Minimum Escaper: Programmatic Subgoal Generation for Robust Navigation in Unknown Environments
局部最小值逃逸者：用于未知环境中鲁棒导航的程序性子目标生成
Abstract
Mapless navigation in unknown and partially observable environments remains challenging for mobile robots, particularly when local minima prevent the robot from making progress toward its goal. Existing local navigation methods often lack an explicit mechanism for escaping such situations, while deep reinforcement learning (DRL) approaches typically learn recovery behaviors implicitly through reward design and policy optimization. In this work, we propose \textbf{LME} (Local-Minimum Escaper), a programmatic hierarchical framework that explicitly generates and reasons subgoals to guide robots out of local-minimum regions. LME operates solely on local observations and selects candidate subgoals using interpretable heuristic criteria that account for both surrounding obstacle geometry and candidate-location safety. A local planner then generates low-level motion commands toward the selected subgoal. This design enables LME to handle environments both with and without local minima within a unified framework, while remaining independent of the underlying local planner and requiring no additional training. Extensive experiments in simulated and real-world environments demonstrate that LME provides robust navigation performance and generalizes to challenging unseen scenarios. Furthermore, the generated subgoals can be used to guide different local planners, substantially improving their ability to escape local minima. Successful deployments on both differential-drive and quadruped robots further demonstrate the practical applicability and generality of the proposed framework.
Chinese Translation
在未知和部分可观测环境中进行无地图导航仍然是移动机器人面临的挑战，特别是在局部最小值阻碍机器人朝目标前进时。现有的局部导航方法通常缺乏明确的机制来逃避这种情况，而深度强化学习（DRL）方法通常通过奖励设计和策略优化隐式地学习恢复行为。在本研究中，我们提出了 extbf{LME}（局部最小值逃逸者），这是一个程序化的分层框架，能够明确生成和推理子目标，以引导机器人脱离局部最小值区域。LME仅基于局部观察，并使用可解释的启发式标准选择候选子目标，这些标准考虑了周围障碍物的几何形状和候选位置的安全性。然后，局部规划器生成朝向所选子目标的低级运动指令。该设计使LME能够在统一框架内处理有无局部最小值的环境，同时独立于基础局部规划器，并且无需额外训练。在模拟和真实环境中的大量实验表明，LME提供了鲁棒的导航性能，并能够推广到具有挑战性的未见场景。此外，生成的子目标可以用于指导不同的局部规划器，显著提高它们逃离局部最小值的能力。在差分驱动和四足机器人上的成功部署进一步证明了该框架的实际适用性和普遍性。
cs.RO / 47 / 2609.38948
DrivingBench: Can Vision-Language Models Drive a Toyota Corolla?
DrivingBench：视觉-语言模型能否驾驶丰田卡罗拉？
Abstract
Frontier models excel at many digital benchmarks, yet their ability to drive a real car, an everyday human skill, remains largely untested. We present DrivingBench, to our knowledge the first benchmark where general-purpose vision-language models must drive a real car. Through three tools, the models see camera frames from a Toyota Corolla and directly command its steering and velocity around a parking lot cone course at low speeds. The car may continue moving while the model thinks and new commands replace the currently running one, so inference latency is part of the task, testing the models' abilities to observe, act, monitor, recover, and complete a long-horizon objective under such constraints. We benchmark GPT-6 Astra, Claude Fable 5.1, GPT-5.6 Sol, and Grok 4.6 in vendor-native harnesses (Codex, Claude Code, Cursor) with up to three attempts each in one conversation; Astra is the only model to finish the course, on its second attempt, with no other attempt passing 50% of the course. Two of the four models improved materially across attempts with retained context. We also detail the design principles behind our action interface, and show how the tool output format and the framing of the task combined to determine whether models would drive at all or refuse. We release our harness, prompts, course map, and traces with video and telemetry for reproducibility.
Chinese Translation
前沿模型在许多数字基准测试中表现出色，但它们驾驶真实汽车的能力，这是一项日常人类技能，仍然基本未经过测试。我们提出了DrivingBench，至今为止我们所知的第一个基准测试，要求通用视觉-语言模型驾驶真实汽车。通过三种工具，模型能够看到丰田卡罗拉的摄像头画面，并直接控制其在停车场锥形赛道上的转向和速度，速度较低。汽车在模型思考时可以继续移动，新命令会替代当前正在执行的命令，因此推理延迟是任务的一部分，测试模型在这种约束下观察、行动、监控、恢复和完成长期目标的能力。我们在供应商原生工具（Codex、Claude Code、Cursor）中对GPT-6 Astra、Claude Fable 5.1、GPT-5.6 Sol和Grok 4.6进行了基准测试，每个模型在一次对话中最多进行三次尝试；Astra是唯一一个在第二次尝试中完成赛道的模型，其他模型的尝试均未超过50%的赛道通过率。在四个模型中，有两个在保留上下文的情况下在尝试中有显著改善。我们还详细介绍了我们的行动接口设计原则，并展示了工具输出格式和任务框架如何共同决定模型是否会驾驶或拒绝驾驶。我们发布了我们的工具、提示、赛道地图以及带有视频和遥测的轨迹，以便于复现。
cs.RO / 48 / 2609.38966
Video2SwimFish: An Automated Pipeline for Reconstructing Controllable Fish Models and Biological Locomotion from Real Fish Videos
Video2SwimFish：从真实鱼类视频中重建可控鱼类模型和生物运动的自动化管道
Abstract
We present Video2SwimFish, an automated pipeline and benchmark for building controllable fish assets from real-fish videos for underwater embodied AI. Given synchronized multi-view videos of an individual fish, the pipeline reconstructs a metrically scaled deformable mesh from a VLM-selected canonical frame, generates internal articulation adapted to that individual's morphology through a VLM actor-critic loop, and extracts a Biological Locomotion Manifold (BLM) from the fish's observed midline curvature. The BLM provides a low-dimensional action space bounded by real-fish motion, enabling an individual swimming policy to be learned for each reconstructed fish. We release two paired datasets: synchronized top- and front-view recordings of 120 individual fish across 6 species, and the controllable assets and individual swimming policies derived from them. Because every asset is tied to the animal it came from, the dataset supports a benchmark that evaluates locomotion learning not only on task success but on fidelity to that individual in trajectory shape, body curvature, and tail-beat frequency, across trajectory following, reward-free swimming behavior transfer from video, and a downstream case study in which a simulated BlueROV underwater robot captures one of the assets. We find that task success and locomotion fidelity do not necessarily improve together: the method achieving the highest task completion is not the method achieving the highest locomotion fidelity, and we identify faithful reproduction of individual animal locomotion as an open challenge for the community. Project website: https://hangongchen.github.io/video2swimfish-web/.
Chinese Translation
我们提出了Video2SwimFish，这是一个自动化管道和基准，用于从真实鱼类视频中构建可控的鱼类资产，以支持水下具身人工智能。给定个体鱼类的同步多视角视频，该管道从VLM（视觉语言模型）选择的标准帧重建一个按比例缩放的可变形网格，通过VLM演员-评论家循环生成适应于该个体形态的内部关节，并从鱼类观察到的中线曲率中提取生物运动流形（BLM）。BLM提供了一个低维的动作空间，该空间由真实鱼类运动所界定，使得可以为每个重建的鱼类学习个体游泳策略。我们发布了两个配对数据集：涵盖6种鱼类的120个个体的同步顶视和正视录制，以及从中派生的可控资产和个体游泳策略。由于每个资产都与其来源动物相关联，该数据集支持一个基准，评估运动学习不仅基于任务成功，还基于在轨迹形状、身体曲率和尾鳍频率上的个体忠实度，涵盖轨迹跟踪、从视频中无奖励游泳行为转移以及下游案例研究，其中模拟的BlueROV水下机器人捕捉到其中一个资产。我们发现任务成功和运动忠实度并不一定同步提高：实现最高任务完成率的方法并不是实现最高运动忠实度的方法，我们确定忠实再现个体动物运动是社区面临的一个开放挑战。项目网站：https://hangongchen.github.io/video2swimfish-web/
cs.RO / 49 / 2609.38982
SimEX: Simulation-Integrated Robotics AutoResearch
SimEX：模拟集成机器人自研究
Abstract
Coding agents powered by large language models (LLMs) have shown remarkable abilities to autonomously reason about and achieve goals in the digital world. However, bringing this success to the physical world remains challenging. On the one hand, direct generation methods (e.g., Code as Policies) often suffer from the LLMs' insufficient understanding of robots and physical environments. On the other hand, iterative trial-and-error tuning in the physical world (e.g., physical autoresearch) induces significant experimental cost and safety concerns. We introduce SimEX: Simulation-Integrated Robotics AutoResearch, an autoresearch framework that tightly integrates simulated experimentation, enabling coding agents to efficiently acquire physical capabilities for controlling real robots. SimEX operates in two stages. First, the agent conducts open-ended probe-and-optimize iterations in simulation, developing a robot toolbox with robust and generalizable capabilities. Second, the agent adapts the toolbox and the simulator together through only a few physical trials: each trial corrects the simulator, and the corrected simulator is used to diagnose failures and screen candidate repairs. We evaluate SimEX extensively in sim-to-sim settings and on physical robots. On challenging real-world manipulation tasks including towel folding, barcode scanning, and plate manipulation, SimEX enables coding agents to efficiently acquire robot skills without any demonstration and with only 10 minutes of real-robot interaction. These results suggest that simulation can be a critical component in achieving physical intelligence, not only as a source of training data that must closely replicate the real world, but also as a roughly correct laboratory where a coding agent develops the knowledge and procedures needed to act on the robot. More details and robot videos at https://robo-simex.github.io/
Chinese Translation
由大型语言模型（LLMs）驱动的编码代理在自主推理和实现数字世界中的目标方面展现了显著的能力。然而，将这一成功转化到物理世界仍然面临挑战。一方面，直接生成方法（例如，Code as Policies）往往受到LLMs对机器人和物理环境理解不足的影响。另一方面，在物理世界中进行的迭代试错调优（例如，物理自研究）会导致显著的实验成本和安全隐患。我们提出了SimEX：模拟集成机器人自研究，这是一个紧密集成模拟实验的自研究框架，使编码代理能够高效地获取控制真实机器人的物理能力。SimEX分为两个阶段进行。首先，代理在模拟环境中进行开放式探测和优化迭代，开发出具有鲁棒性和可推广能力的机器人工具箱。其次，代理通过仅进行少量物理试验，适应工具箱和模拟器：每次试验都修正模拟器，而修正后的模拟器则用于诊断故障和筛选候选修复方案。我们在模拟到模拟的设置以及物理机器人上对SimEX进行了广泛评估。在包括毛巾折叠、条形码扫描和盘子操作等具有挑战性的现实世界操作任务中，SimEX使编码代理能够在没有任何演示的情况下，仅通过10分钟的真实机器人交互高效地获取机器人技能。这些结果表明，模拟可以成为实现物理智能的关键组成部分，不仅作为必须紧密复制真实世界的训练数据来源，而且作为一个大致正确的实验室，在这里编码代理发展出在机器人上行动所需的知识和程序。更多细节和机器人视频请访问 https://robo-simex.github.io/
cs.RO / 50 / 2609.38984
Sparse-WAM: Accelerating World Action Models via Action-Guided Sparse Imagination
稀疏世界行动模型：通过行动引导的稀疏想象加速世界行动模型
Abstract
World-action models (WAMs) leverage pretrained video models to improve generalization in robot control by jointly predicting future visual states and actions. This capability comes at a substantial inference cost, as dense future-frame tokens are repeatedly processed during denoising. Prior methods address this by token pruning that prioritizes visual fidelity to reduce denoising costs in video diffusion models. However, these methods do not use action relevance to determine which future-frame tokens to retain during joint denoising in WAMs. In this paper, we propose Sparse-WAM, a training-free framework for action-guided sparse imagination that selectively processes future-frame tokens to accelerate WAM inference. We observe substantial overlap in the spatial distribution of attention from action tokens to future-frame tokens (action-to-future attention) between consecutive denoising steps, despite continued updates to the future representations. Motivated by this, we develop Action-Guided Token Selection to retain frame-specific action-relevant regions together with cross-frame context. However, a naive implementation can incur attention-scoring and token-packing overhead that offsets the computational savings from pruning. We therefore introduce Pilot, an efficient engine that reduces sparse inference overhead through lightweight scoring and cross-step reuse of token selections. On LIBERO with FastWAM-Joint and RoboLab-120 with Cosmos 3 Edge, Sparse-WAM achieves inference speedups of approximately $2.0\times$ and $1.8\times$, respectively, over dense eager inference on an NVIDIA RTX 4090, while largely preserving task performance.
Chinese Translation
世界行动模型（WAMs）利用预训练的视频模型，通过共同预测未来视觉状态和动作来提高机器人控制的泛化能力。然而，这种能力伴随着相当大的推理成本，因为在去噪过程中，密集的未来帧标记会被反复处理。之前的方法通过标记修剪来解决这个问题，优先考虑视觉保真度以降低视频扩散模型中的去噪成本。然而，这些方法并未利用动作相关性来确定在WAMs的联合去噪过程中保留哪些未来帧标记。在本文中，我们提出了稀疏WAM（Sparse-WAM），这是一种无训练的框架，用于行动引导的稀疏想象，选择性地处理未来帧标记以加速WAM推理。我们观察到，在连续的去噪步骤中，来自动作标记到未来帧标记的注意力（动作到未来注意力）的空间分布存在显著重叠，尽管未来表示仍在不断更新。基于此，我们开发了行动引导的标记选择，以保留特定帧的与动作相关的区域以及跨帧上下文。然而，简单的实现可能会产生注意力评分和标记打包的开销，从而抵消修剪带来的计算节省。因此，我们引入了Pilot，一个高效的引擎，通过轻量级评分和跨步骤重用标记选择来减少稀疏推理的开销。在LIBERO上使用FastWAM-Joint和在RoboLab-120上使用Cosmos 3 Edge时，稀疏WAM分别实现了约$2.0 imes$和$1.8 imes$的推理加速，相比于在NVIDIA RTX 4090上的密集急切推理，同时在很大程度上保持了任务性能。
cs.RO / 51 / 2609.38989
Cue the Flow: Steering Flow-Matching Policies for Open-World Delivery Manipulation
引导流动：开放世界交付操作的流匹配策略调控
Abstract
Open-world goods delivery requires mobile manipulators to follow free-form user instructions and manipulate potentially novel objects. Existing dual-system approaches use high-level grounding models to convert language into grounded visual prompts, but their low-level controllers can remain brittle under noisy perception, dynamic scenes, and contact-rich interactions. We instead use a pretrained flow-matching vision-language-action model as the low-level control interface, leveraging its reactivity and robustness to environmental changes while treating the grounding output as a spatial cue for policy steering. Our key insight is that the pretrained VLA already provides a strong manipulation prior, while the spatial cue supplies the missing target information needed to guide actions under novel language--object mappings. Concretely, we introduce a lightweight cue-conditioned adapter. The adapter is first trained with contrastive objectives to produce salient and spatially discriminative cue representations, and is then supervised to predict a diagonal affine transformation over the generated action chunk, aligning policy steering with the cued target. Across tabletop and mobile-base settings, our method improves instruction following and manipulation success on both in-domain and out-of-domain objects, achieving up to near $2\times$ improvement in average task success rate with negligible inference overhead.
Chinese Translation
开放世界商品交付要求移动操控器遵循自由形式的用户指令并操作潜在的新颖物体。现有的双系统方法使用高层次的基础模型将语言转换为有基础的视觉提示，但其低层次控制器在噪声感知、动态场景和丰富接触的交互下可能表现脆弱。我们则使用一个预训练的流匹配视觉-语言-动作（VLA）模型作为低层次控制接口，利用其对环境变化的反应性和鲁棒性，同时将基础输出视为策略调控的空间提示。我们的关键见解是，预训练的VLA已经提供了强大的操作先验，而空间提示则提供了在新颖语言-物体映射下引导动作所需的缺失目标信息。具体而言，我们引入了一种轻量级的提示条件适配器。该适配器首先通过对比目标进行训练，以生成显著且空间上可区分的提示表示，然后监督其预测生成的动作块上的对角仿射变换，使策略调控与提示目标对齐。在桌面和移动基础设置中，我们的方法在遵循指令和操作成功率上均有所提升，针对领域内和领域外物体的平均任务成功率提高近$2 imes$，且推理开销微乎其微。
cs.RO / 52 / 2609.39000
NEXUS: Perceptive Whole-Body Control for Terrain-Adaptive Teleoperation
NEXUS：适应地形的感知全身控制用于遥操作
Abstract
Whole-body teleoperation requires a humanoid robot to reproduce a human operator's behavior even when their terrains differ. This demands that the robot perceive local terrain and adapt its posture and contacts accordingly, rather than copy the operator's motion frame by frame. However, paired motion data linking the same behaviors across flat ground and different terrains remain scarce, limiting supervision for learning terrain-adaptive control. To enable whole-body teleoperation across mismatched terrains, we introduce NEXUS, a perceptive whole-body control framework that combines human motion commands with onboard sensory feedback. We first develop a scalable terrain-aware adaptation algorithm that efficiently generates high-quality motion pairs across motions and terrains without per-motion or per-terrain tuning. Using a paired motion corpus totaling nearly 1,000 hours, we train a perceptive whole-body controller through teacher-student learning to reproduce commanded behaviors across terrains. Experiments demonstrate efficient, scalable generation of high-quality motion data and show that NEXUS combines broad behavioral coverage with terrain adaptability and tracking fidelity, outperforming existing whole-body controllers on the evaluated benchmarks. Zero-shot real-world deployment enables real-time whole-body teleoperation on diverse unseen terrains, further validating the generalization of our method. Project website: https://nexus-humanoid.github.io/
Chinese Translation
全身遥操作要求类人机器人在不同地形下重现人类操作员的行为。这要求机器人能够感知局部地形并相应地调整其姿态和接触点，而不是逐帧复制操作员的动作。然而，连接平坦地面和不同地形下相同行为的配对运动数据仍然稀缺，限制了对地形适应控制的监督学习。为了实现跨不匹配地形的全身遥操作，我们提出了NEXUS，一个感知全身控制框架，它将人类运动指令与机载传感器反馈相结合。我们首先开发了一种可扩展的地形感知适应算法，该算法能够高效生成跨动作和地形的高质量运动对，而无需对每个动作或每个地形进行调优。利用总计近1000小时的配对运动语料库，我们通过师生学习训练了一个感知全身控制器，以在不同地形上重现指令行为。实验表明，该方法能够高效、可扩展地生成高质量运动数据，并显示NEXUS结合了广泛的行为覆盖、地形适应性和跟踪精度，在评估基准上优于现有的全身控制器。零样本的真实世界部署使得在多样化的未知地形上实现实时全身遥操作，进一步验证了我们方法的泛化能力。项目网站：https://nexus-humanoid.github.io/
cs.RO / 53 / 2609.39006
Function beyond Form: Functional Correspondence for Cross-Embodiment Dexterous Grasp Generation
超越形式的功能：跨体现灵巧抓取生成的功能对应
Abstract
Cross-embodiment dexterous grasp generation remains challenging because robotic hands differ substantially in geometry, topology, and kinematics. Existing approaches often lack explicit correspondences between structurally different hand regions that play similar functional roles in a grasp, a concept we refer to as functional correspondence. Consequently, their models tend to learn hand-specific interaction patterns rather than transferable grasp knowledge, limiting generalization to unseen hands. To address this limitation, we introduce FunCo-Grasp, which establishes functional correspondences across heterogeneous hand embodiments. Specifically, Functional Part Alignment aligns each hand to a canonical functional schema by mapping physical links to shared functional parts according to their grasping roles, while Canonical Frame Alignment expresses these parts in canonical local frames. These two alignments provide a consistent representation for inter-part and hand-object interactions, allowing the model to learn transferable grasp knowledge across hands. Conditioned on the aligned hand representation and object geometry, a diffusion model generates the target spatial arrangement of the functional parts, which are then converted into an executable joint configuration. Adapting FunCo-Grasp to an unseen hand requires only its geometric and kinematic models and a one-time lightweight functional annotation, without target-hand grasp data, fine-tuning, or learned retargeting. In simulation on held-out objects from the filtered CMapDataset, we achieves average success rates of 92.40% on three seen hands and 74.02% on four unseen hands. In real-world experiments, the same model achieves an overall success rate of 76.00% on two unseen hands without additional training or fine-tuning. These results demonstrate the effectiveness of FunCo-Grasp in transferring grasp knowledge to unseen hands.
Chinese Translation
跨体现灵巧抓取生成仍然面临挑战，因为机器人手在几何形状、拓扑结构和运动学上存在显著差异。现有方法往往缺乏在结构上不同的手部区域之间的明确对应关系，而这些区域在抓取中扮演着类似的功能角色，我们称之为功能对应。因此，它们的模型往往学习特定于手的交互模式，而不是可转移的抓取知识，从而限制了对未见手的泛化。为了解决这一限制，我们提出了FunCo-Grasp，它在异质手体现之间建立功能对应。具体而言，功能部件对齐（Functional Part Alignment）通过根据抓取角色将物理链接映射到共享功能部件，将每只手对齐到一个典型的功能框架，而典型框架对齐（Canonical Frame Alignment）则在典型局部框架中表达这些部件。这两种对齐提供了一种一致的表示，用于部件间和手-物体交互，使模型能够学习跨手的可转移抓取知识。在对齐的手表示和物体几何形状的条件下，扩散模型生成功能部件的目标空间排列，然后将其转换为可执行的关节配置。将FunCo-Grasp适应于未见手只需其几何和运动学模型以及一次性的轻量级功能注释，而无需目标手抓取数据、微调或学习重定向。在对过滤后的CMapDataset中保留对象的仿真中，我们在三只已见手上实现了92.40%的平均成功率，在四只未见手上实现了74.02%的成功率。在真实世界实验中，同一模型在两只未见手上实现了76.00%的整体成功率，无需额外训练或微调。这些结果证明了FunCo-Grasp在将抓取知识转移到未见手方面的有效性。
cs.RO / 54 / 2609.39017
OccluDex: Hierarchical 3D Visuo-Tactile Representation Learning for Egocentric Dexterous Manipulation under Self-Occlusion
OccluDex：针对自遮挡下自我中心灵巧操作的分层3D视觉-触觉表征学习
Abstract
Reliable dexterous manipulation requires continuous estimation of object geometry and hand-object contact throughout interaction. With egocentric sensing, however, the manipulating hand frequently occludes task-relevant object surfaces and contact regions, reducing the visual evidence available for state estimation and thereby making robust closed-loop control and generalization to unseen object geometries particularly challenging. To address this, we present OccluDex, a hierarchical 3D visuo-tactile representation learning framework that integrates global geometric structure with local contact information for robust manipulation under dynamic self-occlusion during hand-object interaction. OccluDex adopts multi-scale masked autoencoding to progressively encode partial 3D geometry and fuses tactile contact tokens with high-level geometric features through cross-modal attention. The encoder is pretrained from synchronized human visuo-tactile demonstrations and transferred as a frozen perceptual backbone for downstream reinforcement learning. We evaluate OccluDex on a faucet rotation task, requiring one full clockwise handle revolution, and a tabletop object reorientation task, requiring a 180-degree tabletop object reorientation without toppling. In simulation experiments, OccluDex demonstrated 12.6% higher accuracy for unseen objects and 8.3% higher accuracy for previously seen objects than the strongest state-of-the-art baseline models. Physical experiments were further performed with a Shadow Hand to demonstrate successful zero-shot sim-to-real generalization on unseen physical objects. This results could enable humanoid egocentric object manipulation for seen and unseen objects even when the manipulating robotic hand occludes vision.
Chinese Translation
可靠的灵巧操作需要在交互过程中持续估计物体几何形状和手-物体接触。然而，在自我中心感知下，操控手经常遮挡与任务相关的物体表面和接触区域，从而减少可用于状态估计的视觉证据，这使得稳健的闭环控制和对未见物体几何形状的泛化变得特别具有挑战性。为了解决这个问题，我们提出了OccluDex，一个分层3D视觉-触觉表征学习框架，它将全局几何结构与局部接触信息相结合，以实现动态自遮挡下的稳健操作。OccluDex采用多尺度掩蔽自编码技术，逐步编码部分3D几何形状，并通过跨模态注意力将触觉接触标记与高级几何特征融合。编码器通过同步的人类视觉-触觉示范进行预训练，并作为冻结的感知骨干转移到下游强化学习中。我们在一个水龙头旋转任务上评估了OccluDex，该任务要求手柄完成一次完整的顺时针旋转，以及一个桌面物体重新定向任务，该任务要求在不倾倒的情况下将桌面物体重新定向180度。在模拟实验中，OccluDex在未见物体上的准确率比最强的最先进基线模型高出12.6%，在之前见过的物体上的准确率高出8.3%。此外，我们还使用Shadow Hand进行了物理实验，以展示在未见物理物体上的成功零样本模拟到现实的泛化。这些结果可能使得人形自我中心物体操作在见过和未见物体的情况下，即使在操控机器人手遮挡视线时也能实现。
cs.RO / 55 / 2609.39018
Make Code as Policy Great Again: Frontier Agents Write, Call, and Evolve Robot Tools
让代码作为政策再度伟大：前沿智能体编写、调用和演化机器人工具
Abstract
Frontier models can control robots, but reasoning through every reach, grasp, and retreat makes manipulation slow and token-intensive. We revisit code as policy with a different division of labor: models build executable tools, code handles multi-phase motions, and models decide what to do next. We introduce URAI (Universal Robot-Agent Interface), which couples a programming agent that constructs robot tools with an execution agent that uses them in a feedback loop. The programming agent writes reusable and task-specific tools from task intent and refines them through execution feedback and human guidance. The execution agent selects and parameterizes these tools from current observations; each call runs a complete motion locally before returning control to the agent. Unlike delegating subsequent decisions to a generated program, this design retains model-level decision-making between tool executions. Validated tool revisions persist across episodes without updating foundation-model weights, and a shared GUI and API make the same tools available to humans and agents. Across five RoboDojo tasks and four frozen execution agents, URAI raises aggregate success from 18.0% to 53.0% relative to direct fingertip control, with the largest gain on Swap Blocks; with the same tools, a program written in advance reaches only 24% against 56% for two agents deciding after each call. Three of the four agents also finish episodes 1.3-1.5 times faster with 1.5-1.7 times fewer execution-agent output tokens; DeepSeek-V4-Flash's cost barely changes. We further evaluate URAI on seven real-world AgileX dual-arm tasks, spanning object manipulation, cloth folding, and human-interactive tic-tac-toe. URAI connects the coding and decision-making capabilities of frontier agents, organizing robot control around reusable tools that agents can both invoke and revise.
Chinese Translation
前沿模型可以控制机器人，但对每一个伸手、抓取和撤退的推理使得操作变得缓慢且需要大量的计算资源。我们重新审视代码作为政策的概念，采用不同的劳动分工：模型构建可执行工具，代码处理多阶段动作，模型决定下一步该做什么。我们引入了URAI（通用机器人-智能体接口），它将构建机器人工具的编程智能体与在反馈循环中使用这些工具的执行智能体相结合。编程智能体根据任务意图编写可重用且特定于任务的工具，并通过执行反馈和人类指导对其进行优化。执行智能体从当前观察中选择并参数化这些工具；每次调用在本地执行完整的动作，然后将控制权返回给智能体。与将后续决策委托给生成程序不同，这种设计在工具执行之间保留了模型级的决策。经过验证的工具修订在各个回合中持续存在，而无需更新基础模型的权重，共享的图形用户界面（GUI）和应用程序接口（API）使得相同的工具可供人类和智能体使用。在五个RoboDojo任务和四个冻结执行智能体的实验中，URAI将整体成功率从18.0%提高到53.0%，相较于直接的指尖控制，最大的提升出现在交换块任务中；使用相同工具的预先编写程序仅达到24%，而两个在每次调用后做出决策的智能体则达到56%。四个智能体中有三个的回合完成速度也提高了1.3到1.5倍，执行智能体输出的令牌数量减少了1.5到1.7倍；DeepSeek-V4-Flash的成本几乎没有变化。我们进一步在七个真实世界的AgileX双臂任务中评估URAI，涵盖物体操作、布料折叠和人机互动的井字棋。URAI连接了前沿智能体的编码和决策能力，围绕可重用工具组织机器人控制，使智能体能够同时调用和修订这些工具。
cs.RO / 56 / 2609.39038
Looking Back to Move Forward: Temporal Verification for Generative Robot Policies
回顾以便前行：生成机器人策略的时间验证
Abstract
Generative policies have emerged as a promising paradigm for robot learning, combining expressive generative action modeling with scalable imitation learning from large demonstration corpora. However, heterogeneous demonstrations can induce suboptimal action chunks whose errors compound over time, eventually driving the robot into out-of-distribution states from which recovery is difficult. Action verification offers a test-time scaling strategy for mitigating this failure mode by sampling multiple candidate actions and using a verifier to select one for execution. Existing approaches, however, remain temporally myopic and costly to train, evaluating candidates from the current observation alone without accounting for trajectory continuity and often relying on large verifiers and additional expert demonstrations. In this paper, we introduce Temporal Verification (TeV), an efficient temporally aware action verification framework for flow-matching VLAs. TeV first learns a temporal token that summarizes recent observation--action history, enabling candidate chunks to be evaluated as continuations of the execution trajectory rather than as isolated predictions. Conditioned on this token, TeV constructs positive--negative pairs without additional expert demonstrations or preference annotations and trains an energy-based verifier contrastively to assign lower energy to higher-quality, trajectory-consistent action chunks. Beyond post-hoc ranking, TeV further uses the learned energy landscape to guide intermediate flow samples toward lower-energy regions, improving candidates before final selection. Extensive experiments in simulation and real-world settings demonstrate that TeV provides reliably ranks action candidates, improves task success rates, and produces smoother execution trajectories.
Chinese Translation
生成策略作为机器人学习的一种有前景的范式，结合了表达性生成动作建模和从大量示范语料库中可扩展的模仿学习。然而，异质性示范可能会导致次优的动作片段，其错误会随着时间的推移而累积，最终使机器人陷入难以恢复的分布外状态。动作验证提供了一种测试时的扩展策略，通过采样多个候选动作并使用验证器选择一个进行执行，从而减轻这种失败模式。然而，现有的方法仍然在时间上短视且训练成本高，仅根据当前观察评估候选者，而不考虑轨迹的连续性，并且通常依赖于大型验证器和额外的专家示范。在本文中，我们介绍了时间验证（Temporal Verification, TeV），这是一个高效的时间感知动作验证框架，旨在流匹配可变形状动作（VLA）。TeV首先学习一个时间标记，概括最近的观察-动作历史，使得候选片段能够被评估为执行轨迹的延续，而不是孤立的预测。在此标记的条件下，TeV构建正负对，而无需额外的专家示范或偏好注释，并对基于能量的验证器进行对比训练，以便为更高质量、轨迹一致的动作片段分配较低的能量。除了事后排名，TeV还利用学习到的能量景观引导中间流样本朝向低能量区域，从而在最终选择之前改进候选者。在模拟和真实世界环境中的大量实验表明，TeV能够可靠地对动作候选者进行排名，提高任务成功率，并产生更平滑的执行轨迹。
cs.RO / 57 / 2609.39054
CLIPPER Beyond Shortlisting: Auditable Decision Support for Changing Municipal Micromobility Policies
CLIPPER超越候选名单：可审计的决策支持以应对变化的市政微型出行政策
Abstract
In municipal planning workshops, planners and other stakeholders compare shared-micromobility parking policies by varying no-parking zones, retained sites, spacing, or area allocations. Each edit changes feasible sites and how much demand they cover, so the alternative must be reoptimized on the same spatial data. Full-set greedy, the transparent reference for this task, takes tens of seconds per alternative at city scale. We present Constraint-exact Low-latency Iterative Planning with Pooled Evaluation and Replay (CLIPPER), an optimizer with audit functions developed for requirements elicited with the City of Braunschweig. In each greedy round, it forms a deterministic candidate pool of bounded size, computes how much still-uncovered demand each candidate would add, and rejects candidates that violate an active constraint. An optional offline audit scans every remaining feasible candidate and records what the restricted pool omitted. We evaluate these functions on complete eleven-state edit chains ($E_0,\ldots,E_{10}$) in Braunschweig, Munich, and Berlin. With $K=1024$ candidates per group, the fixed-width mode CLIPPER-F has mean coverage gaps to full-set greedy under the same policy of 0.245, 0.003, and 0.001 percentage points in Braunschweig, Munich, and Berlin, respectively, while mean rollout time falls by factors of 13.6--28.9; no audited run terminates while a candidate outside the pool could still increase coverage. Plans computed from two checksummed versions of Braunschweig's official no-parking-zone data differ in 30 of about 540 selected sites although coverage moves by only about 0.1 percentage points. These changes still require municipal assessment and implementation. The findings inform a proposed municipal process that versions policy inputs, reports site changes beside coverage, and scans the full candidate set before a final decision.
Chinese Translation
在市政规划研讨会上，规划者和其他利益相关者通过改变禁停区、保留地点、间距或区域分配来比较共享微型出行停车政策。每次编辑都会改变可行地点及其覆盖的需求量，因此必须在相同的空间数据上重新优化替代方案。全集贪婪算法作为此任务的透明参考，在城市规模下每个替代方案需要数十秒。我们提出了带有审计功能的约束精确低延迟迭代规划与汇聚评估与重放（CLIPPER），这是一个为布伦瑞克市的需求引导开发的优化器。在每轮贪婪中，它形成一个有限大小的确定性候选池，计算每个候选者将增加的尚未覆盖的需求量，并拒绝违反活动约束的候选者。可选的离线审计扫描每个剩余的可行候选者，并记录被限制池遗漏的内容。我们在布伦瑞克、慕尼黑和柏林的完整十一状态编辑链（$E_0, ext{...},E_{10}$）上评估这些功能。在每组$K=1024$个候选者的情况下，固定宽度模式CLIPPER-F在布伦瑞克、慕尼黑和柏林的同一政策下，与全集贪婪算法相比，平均覆盖差距分别为0.245、0.003和0.001个百分点，而平均推出时间减少了13.6到28.9倍；没有审计运行在候选池外仍可能增加覆盖的情况下终止。从布伦瑞克官方禁停区数据的两个校验和版本计算出的计划在大约540个选定地点中有30个不同，尽管覆盖率仅变化约0.1个百分点。这些变化仍需市政评估和实施。研究结果为提出的市政流程提供了信息，该流程版本化政策输入，报告覆盖率旁边的地点变化，并在最终决策前扫描完整候选集。
cs.RO / 58 / 2609.39056
SteerQuant: Steering Quantization Error with Action-Guided Scaling in World-Action Models
SteerQuant：在世界-动作模型中通过动作引导缩放来引导量化误差
Abstract
World-action models (WAMs) jointly generate future world states and actions through iterative denoising, using shared weights to process heterogeneous semantic streams of video, proprioceptive, and action tokens. Quantization reduces inference cost, but comparable numerical errors in different streams can have markedly different effects on final actions, making numerical accuracy alone insufficient for reliable control. We introduce SteerQuant, a 4-bit quantization framework for WAMs that steers errors toward computations with less influence on final actions. It maps how each stream's quantization errors affect final actions and uses this map to guide shared channel scaling. Activation scaling is further calibrated for each stream and denoising step to accommodate changes in activation ranges and action impact. This adapts quantization to different stream requirements without duplicating weights or increasing bit-widths for selected streams. To reduce the extra kernel launches and memory traffic introduced by scaling, we develop Rudder, a 4-bit inference engine for WAMs that fuses scaling and output compensation into low-bit kernels. Under W4A8 and W4A4, SteerQuant maintains mean LIBERO success within 0.8 percentage points of full precision, while delivering up to $2.23\times$ denoising speedup over BF16 across three WAMs with reduced peak GPU memory usage. On a real dual-arm robot, W4A8 deployment achieves a $1.35\times$ end-to-end inference speedup while maintaining average task success relative to BF16.
Chinese Translation
世界-动作模型（WAMs）通过迭代去噪共同生成未来的世界状态和动作，使用共享权重处理异构的语义流，包括视频、身体感知和动作标记。量化虽然可以降低推理成本，但不同流中的可比数值误差对最终动作的影响可能截然不同，仅依赖数值准确性不足以实现可靠控制。我们提出了SteerQuant，一个针对WAMs的4位量化框架，旨在将误差引导至对最终动作影响较小的计算。该框架映射每个流的量化误差如何影响最终动作，并利用此映射指导共享通道的缩放。激活缩放进一步针对每个流和去噪步骤进行校准，以适应激活范围和动作影响的变化。这使得量化能够适应不同流的需求，而无需复制权重或增加选定流的位宽。为了减少缩放引入的额外内核启动和内存流量，我们开发了Rudder，一个针对WAMs的4位推理引擎，将缩放和输出补偿融合到低位内核中。在W4A8和W4A4模式下，SteerQuant在保持平均LIBERO成功率在全精度的0.8个百分点以内的同时，提供了比BF16高出$2.23 imes$的去噪加速，在三种WAMs中减少了峰值GPU内存使用。在一个真实的双臂机器人上，W4A8的部署实现了$1.35 imes$的端到端推理加速，同时保持了相对于BF16的平均任务成功率。
cs.RO / 59 / 2609.39098
DiFF: Doppler-informed Flow Matching for Human Motion Flow
DiFF：基于多普勒信息的人体运动流匹配
Abstract
Perceiving human motion via privacy-preserving 4D millimeter-wave (mmWave) radar is critical for next-generation human-robot interaction (HRI), where point cloud scene flow serves as a foundational motion representation. Yet the extreme sparsity and noise of 4D radar point clouds make non-rigid motion flow estimation severely ill-posed--a challenge that existing rigid-centric methods and prior works fail to adequately address, largely because they neglect the rich Doppler velocity cues inherent in 4D radar. We propose DiFF, a generative framework that marries Doppler-informed motion priors with a Kolmogorov-Arnold Network (KAN)-based conditional flow matching model. At its core, a KAN-attention mechanism enables expressive feature extraction, while a prior-guided generative process harnesses Doppler cues to regularize the ill-posed solution space. Extensive experiments show that DiFF achieves state-of-the-art (SOTA) performance across diverse real-world datasets, reducing 3D endpoint error to the millimeter scale on the mmBody benchmark.
Chinese Translation
通过隐私保护的4D毫米波（mmWave）雷达感知人体运动对于下一代人机交互（HRI）至关重要，其中点云场景流作为基础运动表示。然而，4D雷达点云的极度稀疏性和噪声使得非刚性运动流估计严重病态，这一挑战现有的以刚性为中心的方法和先前的研究未能充分解决，主要是因为它们忽视了4D雷达中固有的丰富多普勒速度线索。我们提出了DiFF，一个将多普勒信息运动先验与基于Kolmogorov-Arnold网络（KAN）的条件流匹配模型相结合的生成框架。其核心是KAN注意机制，能够进行富有表现力的特征提取，而先验引导的生成过程利用多普勒线索来规范病态解空间。大量实验表明，DiFF在多样化的真实世界数据集上实现了最先进（SOTA）的性能，在mmBody基准测试中将3D端点误差降低到毫米级。
cs.RO / 60 / 2609.39105
Occlusion-Aware, Quasi-Static, Stability-Oriented Trajectory Planning on Uneven Terrain
考虑遮挡的准静态稳定性导向轨迹规划在不平坦地形上的应用
Abstract
Autonomous navigation in unstructured off-road environments requires reasoning about both vehicle--terrain interaction and environmental unknowns. We propose a model-based framework for generating quasi-static, stability-oriented reference trajectories for rigid, non-articulated four-wheeled vehicles on highly uneven terrain. Our work makes three primary contributions. First, we model blind spots caused by terrain occlusion as coverage-induced epistemic uncertainty in a fixed-feature Fourier terrain representation, quantified through a regularized inverse-Hessian estimate. Second, we propagate this uncertainty through the Nonlinear Least-Squares (NLS) pose/contact model using implicit differentiation and incorporate the resulting pose, contact-point, and per-wheel surface-normal uncertainty terms into trajectory optimization based on the Cross-Entropy Method (CEM). Third, we introduce a Flow Matching model that warm-starts terrain fitting, and we evaluate its fitting-accuracy--latency trade-off while retaining model-based refinement. Across six synthetic terrains with 30 matched start--goal pairs per terrain, the complete framework produced an observed failure rate of 18.9%, compared with 46.1% and 41.7% for two representative baselines and 34.4% for an ablation that removed the propagated-uncertainty scoring. Hardware evaluations span six distinct outdoor environments, with two representative executions presented in the paper and four additional executions included in the supplementary video. The evaluation also reports the accuracy--latency trade-off for the Flow Matching warm start.
Chinese Translation
在非结构化的越野环境中进行自主导航需要考虑车辆与地形的相互作用以及环境的不确定性。我们提出了一种基于模型的框架，用于在高度不平坦的地形上为刚性非关节四轮车辆生成准静态的稳定性导向参考轨迹。我们的工作主要有三个贡献。首先，我们将地形遮挡造成的盲区建模为固定特征傅里叶地形表示中的覆盖引起的认知不确定性，通过正则化的逆海森矩阵估计进行量化。其次，我们通过隐式微分法将这种不确定性传播到非线性最小二乘（Nonlinear Least-Squares, NLS）姿态/接触模型中，并将得到的姿态、接触点和每个轮子的表面法线不确定性项纳入基于交叉熵方法（Cross-Entropy Method, CEM）的轨迹优化中。第三，我们引入了一种流匹配（Flow Matching）模型来热启动地形拟合，并在保持基于模型的细化的同时评估其拟合精度与延迟之间的权衡。在六个合成地形上，每个地形有30对匹配的起始-目标对，完整框架的观察到的失败率为18.9%，而两个代表性基线的失败率分别为46.1%和41.7%，去除传播不确定性评分的消融实验失败率为34.4%。硬件评估涵盖六个不同的户外环境，论文中展示了两个代表性执行案例，并在补充视频中包含了四个额外的执行案例。评估还报告了流匹配热启动的精度与延迟之间的权衡。
cs.RO / 61 / 2609.39125
LBDU-VIO: Learned Bias Dynamics and Uncertainty for Visual-Inertial Odometry with Unreliable Vision
LBDU-VIO：针对不可靠视觉的视觉惯性测程中的学习偏差动态与不确定性
Abstract
Visual-inertial odometry (VIO) for aerial robots relies on high rate inertial measurement unit (IMU) propagation between visual updates. However, conventional multi state constraint Kalman filters (MSCKFs) use random walk bias assumptions and fixed noise parameters, which can limit robustness when visual information is unreliable. To address this problem, we propose LBDU-VIO, a learning-augmented MSCKF with learned continuous time bias dynamics and an IMU uncertainty model. A neural ordinary differential equation (ODE) models continuous time bias dynamics to propagate the filter's bias states, replacing their random walk model. The IMU uncertainty model predicts motion adaptive measurement noise covariances for covariance propagation. Both models are trained with pose supervision without direct labels. Experiments on real world EuRoC and TUM-VI benchmarks show lower errors than representative visual-inertial baselines, including a 25.1% reduction in mean relative position error compared with S-MSCKF on EuRoC sequences with 10s visual outage.
Chinese Translation
针对空中机器人，视觉惯性测程（VIO）依赖于高频率的惯性测量单元（IMU）在视觉更新之间的传播。然而，传统的多状态约束卡尔曼滤波器（MSCKF）使用随机游走偏差假设和固定噪声参数，这在视觉信息不可靠时可能限制其鲁棒性。为了解决这个问题，我们提出了LBDU-VIO，一种增强学习的MSCKF，具有学习的连续时间偏差动态和IMU不确定性模型。神经普通微分方程（ODE）用于建模连续时间偏差动态，以传播滤波器的偏差状态，替代其随机游走模型。IMU不确定性模型预测运动自适应测量噪声协方差，以进行协方差传播。两个模型均通过姿态监督进行训练，而无需直接标签。在真实世界的EuRoC和TUM-VI基准测试中的实验表明，与代表性的视觉惯性基线相比，误差更低，包括在EuRoC序列中与S-MSCKF相比，平均相对位置误差减少25.1%，该序列存在10秒的视觉中断。
cs.RO / 62 / 2609.39129
Drape-Compatible Tool-Tip Localization for Hand-Held Laparoscopic Instruments via UWB Carrier-Phase Ranging and Trocar-Constrained Geometry
基于超宽带载波相位测距和 trocar 约束几何的手持腹腔镜仪器的工具尖端定位
Abstract
A surgical robot policy needs to know where each instrument's working end sits relative to the camera and to the other instrument, yet a laparoscopic operation leaves only the endoscope video, and a draped instrument hides every optical path on itself. We estimate the tool tips from distances and an IMU alone. The distances are measured by ultra-wideband carrier phase between antenna nodes on the handle side of the instruments: one per hand-held instrument, one on the endoscope, all outside the sterile barrier. Take the endoscope antenna as the reference. The two instrument antennas carry six coordinates while the three pairs give three distances, so the problem is short by three, and averaging cannot close that gap because what is missing is information, not precision. Carrier phase adds one more unknown per pair, an integer that leaves distance fixed only modulo lambda/2 = 23.1 mm. The trocar settles both difficulties: each shaft passes through a port whose position is known and whose direction the on-board IMU measures, so its antenna keeps a single degree of freedom, the insertion depth. Two unknowns now face three distances -- two fix the depths and the third is left over, and that spare equation exposes a wrong integer or a wrong mounting offset instead of absorbing it. The same solve returns both tool tips in the endoscope frame, without the second carrier frequency that integer resolution usually needs, so the link never leaves a single PLL lock. In an 11-hole phantom among metal instruments the three distances hold sigma = 0.10-0.33 mm at 35-46 Hz, a sterile drape pressed onto the antennas leaves the link unchanged where an optical path would be blocked, and the logger was carried into a live rabbit appendectomy.
Chinese Translation
外科机器人策略需要了解每个仪器工作端相对于相机和其他仪器的位置，然而腹腔镜手术仅留下内窥镜视频，而覆盖的仪器则隐藏了所有光学路径。我们仅通过距离和惯性测量单元（IMU）来估计工具尖端。距离通过仪器手柄侧的天线节点之间的超宽带载波相位进行测量：每个手持仪器一个，内窥镜一个，均位于无菌屏障之外。以内窥镜天线作为参考。两个仪器天线携带六个坐标，而三对天线提供三条距离，因此问题缺少三个信息，平均无法弥补这一差距，因为缺失的是信息，而非精度。载波相位每对增加一个未知数，即一个整数，使得距离仅在模 λ/2 = 23.1 mm 下固定。Trocar 解决了这两个困难：每个轴通过一个已知位置的端口，且其方向由机载 IMU 测量，因此其天线保持一个自由度，即插入深度。现在两个未知数面对三条距离——两个确定深度，第三个则剩余，而这个多余的方程揭示了一个错误的整数或错误的安装偏移，而不是将其吸收。同样的解法最终返回两个工具尖端在内窥镜坐标系中，而无需第二个载波频率，这通常需要整数分辨率，因此连接始终保持在一个相位锁定环（PLL）中。在一个包含金属仪器的11孔模型中，这三条距离的标准差为 σ = 0.10-0.33 mm，35-46 Hz，压在天线上的无菌覆盖物使得连接保持不变，即使在光学路径被阻挡的情况下，而记录器被带入了活兔阑尾切除手术中。
cs.RO / 63 / 2609.39145
Blackout vs. Freeze: Analyzing Physical Failure Modes of VLAs under Camera Faults
黑屏与冻结：分析视觉-语言-动作（VLA）模型在相机故障下的物理失效模式
Abstract
Unreliable visual inputs can harm task performance and cause potential physical safety risks for vision-language-action (VLA) models. We analyze how $\pi 0.5$ and GR00T models act under input faults such as image blackouts and freezing. We find that blackout and freezing produce distinct physical failure modes even when task-success rates are similarly low: freezing causes more extreme joint behavior, whereas blackout after gripper closure can cause more object drops, most markedly without proprioception. Selective intervention studies reveal that proprioception (current robot state) partly compensates for the removed robot depictions and reduces non-target contact. However, it cannot sufficiently restore task success when wrist-view object information is removed, even when aided by the remaining scene view. We then evaluate two mitigation approaches: camera-blackout training and training-free replacement of faulty visual embeddings. Both improve task success in selected conditions, but can increase unintended contact or disturbance to surrounding objects. Real-robot trials further show that successful execution under camera faults can still involve unintended physical interactions. These findings motivate designing VLA policies that use the robot and object information still available under camera faults to limit hazardous motion.
Chinese Translation
不可靠的视觉输入可能会损害任务表现，并对视觉-语言-动作（VLA）模型造成潜在的物理安全风险。我们分析了$ ext{pi} 0.5$和GR00T模型在图像黑屏和冻结等输入故障下的表现。研究发现，黑屏和冻结产生了不同的物理失效模式，即使任务成功率相似：冻结导致更极端的关节行为，而在夹持器闭合后，黑屏可能导致更多物体掉落，尤其是在缺乏本体感知的情况下。选择性干预研究表明，本体感知（当前机器人状态）在一定程度上补偿了移除的机器人描绘，并减少了非目标接触。然而，当手腕视角的物体信息被移除时，即使在剩余的场景视角辅助下，它也无法充分恢复任务成功。随后，我们评估了两种缓解方法：相机黑屏训练和无训练的故障视觉嵌入替换。这两种方法在特定条件下均提高了任务成功率，但可能增加意外接触或对周围物体的干扰。真实机器人试验进一步表明，在相机故障下成功执行任务仍可能涉及意外的物理交互。这些发现促使我们设计VLA策略，利用在相机故障下仍然可用的机器人和物体信息，以限制危险动作。
cs.RO / 64 / 2609.39151
Linear Recurrent Memory Suffices to Distil a World-Model Policy for Robot Air Hockey
线性递归记忆足以提炼出机器人空气曲棍球的世界模型策略
Abstract
Does memory-dependent control need nonlinear recurrent dynamics? We study simulated air-hockey defence under temporary loss of puck tracking. A DreamerV3 teacher outperforms a memoryless policy under tracking loss, while resetting the teacher's recurrent state sharply reduces performance, which demonstrates that the task requires memory. We distil this teacher into compact recurrent policies with a 64 dimensional state, with a combination of a diagonal linear recurrence and an optional rank-$k$ nonlinear innovation while retaining nonlinear observation encoders and action heads. Across five matched seeds, the purely linear recurrent model ($k=0$) matches both the GRU baseline and the teacher throughout the tested range of tracking loss. Increasing nonlinear innovation rank providing no measured benefits. This result is obtained on a fresh test split, which will be only opened after all models and analyses are frozen. The linear model requires fewer recurrent parameters and less computation than GRU, but performs comparably. These results suggest that, for this memory dependent control task, nonlinear representation learning around a simple linear memory mechanism can be sufficient, and that nonlinear recurrent dynamics are not necessarily required. These conclusions are limited to the simulated task, teacher, state dimension, and blackout horizon considered here, and to policies whose observation encoder and action head remain nonlinear.
Chinese Translation
记忆依赖控制是否需要非线性递归动态？我们研究了在暂时失去冰球追踪情况下的模拟空气曲棍球防御。DreamerV3教师在追踪丢失时优于无记忆策略，而重置教师的递归状态则显著降低了性能，这表明该任务需要记忆。我们将该教师提炼为具有64维状态的紧凑递归策略，结合对角线性递归和可选的秩-$k$非线性创新，同时保留非线性观察编码器和动作头。在五个匹配的种子中，纯线性递归模型（$k=0$）在测试的追踪丢失范围内与GRU基线和教师的表现相当。增加非线性创新秩未提供任何测量上的好处。该结果是在一个新的测试分割上获得的，该分割将在所有模型和分析冻结后才会开放。线性模型所需的递归参数更少，计算量也低于GRU，但性能相当。这些结果表明，对于这个依赖记忆的控制任务，围绕简单线性记忆机制的非线性表示学习可能是足够的，并且不一定需要非线性递归动态。这些结论仅限于此处考虑的模拟任务、教师、状态维度和失效时间范围，以及观察编码器和动作头保持非线性的策略。
cs.RO / 65 / 2609.39153
Concurrent Semantic Search and Mission Execution for LTL Missions in Unknown Environments
在未知环境中进行LTL任务的并发语义搜索与任务执行
Abstract
Planning complex missions in unknown environments requires robots to reason simultaneously about what they should do and what they still need to discover. Existing approaches for solving LTLf missions typically assume a known environment, or separate the exploration of the environment from the execution of the mission, while semantic exploration methods look for one target at a time and ignore the mission being executed. To fill this gap, our main contribution is an adaptive high-level planning method that interleaves a task-driven semantic search with the execution of the mission, advancing both in a non-myopic manner. Our method leverages two representations built online, a metric-semantic scene graph, built with a Vision Language Model (VLM), that provides the evidence needed to locate the objects the mission refers to, and the deterministic finite automaton (DFA) encoding the mission, that indicates which of them matter at each mission state. At every planning stage, our planner selects the waypoints that are most valuable for both the semantic search and the advancement of the mission, valuing them over the remaining mission stages in order to avoid blocking states. The selected waypoints are then ordered in a single high-level plan, which is recomputed as new information arrives. In photorealistic indoor environments over five mission types, our method completes more missions than the compared approaches while having to cover less of the environment, and it does so with shorter paths and complying with the restrictions imposed by the mission.
Chinese Translation
在未知环境中规划复杂任务需要机器人同时推理它们应该做什么以及还需要发现什么。现有的解决LTLf任务的方法通常假设环境是已知的，或者将环境的探索与任务的执行分开，而语义探索方法则一次只寻找一个目标，忽视正在执行的任务。为了填补这一空白，我们的主要贡献是提出了一种自适应的高层次规划方法，该方法将任务驱动的语义搜索与任务执行交错进行，以非短视的方式推动两者的发展。我们的方法利用了两个在线构建的表示：一个基于视觉语言模型（Vision Language Model, VLM）构建的度量-语义场景图，它提供了定位任务所指对象所需的证据；以及一个确定性有限自动机（Deterministic Finite Automaton, DFA）编码的任务，它指示在每个任务状态下哪些对象是重要的。在每个规划阶段，我们的规划器选择对语义搜索和任务推进最有价值的路径点，优先考虑它们而不是剩余的任务阶段，以避免阻塞状态。然后，所选路径点被排序为一个单一的高层次计划，并在新信息到达时重新计算。在五种任务类型的逼真室内环境中，我们的方法完成的任务数量超过了比较方法，同时覆盖的环境面积更小，并且以更短的路径完成，遵循任务所施加的限制。
cs.RO / 66 / 2609.39178
Exploiting Vulnerabilities: Universal Adversarial Attacks on Vision-Language-Action Models in Robotics
利用漏洞：针对机器人视觉-语言-动作模型的通用对抗攻击
Abstract
Recently, Vision-Language-Action (VLA) models have revolutionized robotic manipulation by seamlessly integrating visual perception, language understanding, and action generation in an end-to-end learning framework. However, since these models are designed to interact directly with the physical world and humans, their security is critical, and even small vulnerabilities can lead to catastrophic failures. In this work, we propose the Universal Adversarial Object, a sphere with optimized surface texture that significantly degrades task success rates when placed within the robot's field of view. Specifically, our approach introduces a multi-level attack framework that jointly disrupts trajectory planning, task execution, and action control. We validate our method in both simulated and real-world robotic settings. Experimental results demonstrate that the adversarial object reduces the average task success rates by 31.2%-39.9% for two representative VLA models (Pi0 and RDT), with success rates dropping to near zero in complex scenarios. Index Terms--Vision-Language-Action models, adversarial attack, robotic security, universal adversarial object
Chinese Translation
近年来，视觉-语言-动作（VLA）模型通过在端到端学习框架中无缝整合视觉感知、语言理解和动作生成，彻底改变了机器人操作。然而，由于这些模型旨在直接与物理世界和人类互动，因此其安全性至关重要，即使是微小的漏洞也可能导致灾难性的失败。在本研究中，我们提出了通用对抗物体，这是一种具有优化表面纹理的球体，当放置在机器人视野内时，会显著降低任务成功率。具体而言，我们的方法引入了一个多层次攻击框架，联合干扰轨迹规划、任务执行和动作控制。我们在模拟和真实的机器人环境中验证了我们的方法。实验结果表明，对于两个代表性的VLA模型（Pi0和RDT），对抗物体使平均任务成功率降低了31.2%-39.9%，在复杂场景中成功率降至接近零。关键词：视觉-语言-动作模型，对抗攻击，机器人安全，通用对抗物体
cs.RO / 67 / 2609.39179
LocoWM: High-Precision Locomotion through World-Model-Guided Residual Adaptation
LocoWM：通过世界模型引导的高精度运动适应
Abstract
High-precision locomotion combines motion-command tracking with precise regulation of task-relevant physical states, enabling robots to interact reliably with their surroundings during motion. Joint end-to-end optimization can leave precision objectives insufficiently optimized, while reactive residual control adjusts actions only after deviations become observable. We present \textbf{LocoWM}, a world-model-guided preactive residual adaptation framework for high-precision locomotion. A base policy provides command-following locomotion, while an action-conditioned world model predicts a sequence of future physical states from proprioceptive history and the proposed base action. A residual adapter conditions on this predicted sequence to generate additive action corrections that compensate for anticipated deviations. Two-stage training first learns locomotion and action-conditioned dynamics, then freezes both modules while training the adapter, separating locomotion acquisition from precision adaptation. Experiments spanning terrain leveling, acceleration compensation, and push recovery demonstrate improved control precision and disturbance robustness over end-to-end and reactive residual baselines. Demos and code are available at: https://zhaozijie2022.github.io/LocoWM
Chinese Translation
高精度运动结合了运动指令跟踪与任务相关物理状态的精确调节，使机器人在运动过程中能够可靠地与周围环境进行交互。联合端到端优化可能导致精度目标未得到充分优化，而反应性残差控制仅在偏差变得可观察后调整动作。我们提出了 extbf{LocoWM}，一个用于高精度运动的世界模型引导的预激活残差适应框架。基础策略提供指令跟随的运动，而基于动作的世界模型则根据本体感知历史和提出的基础动作预测未来物理状态的序列。残差适配器基于这一预测序列生成附加的动作修正，以补偿预期的偏差。两阶段训练首先学习运动和基于动作的动态，然后在训练适配器时冻结这两个模块，将运动获取与精度适应分开。涵盖地形平整、加速补偿和推力恢复的实验表明，相较于端到端和反应性残差基线，控制精度和干扰鲁棒性得到了改善。演示和代码可在以下链接获取：https://zhaozijie2022.github.io/LocoWM
cs.RO / 68 / 2609.39198
DSDyn-VLA: A Dual-Stream Dynamic Manipulation Framework with Motion Perception, Future Awareness, and Realtime Correction
DSDyn-VLA：一种具有运动感知、未来意识和实时纠正的双流动态操控框架
Abstract
While Vision-Language-Action (VLA) models excel in static tasks, they struggle in dynamic environments where objects are in motion (e.g., conveyor belt manipulation). We identify three fundamental limitations hindering current VLAs in these scenarios: the \textbf{perception gap}, where static visual inputs lack temporal motion cues; the \textbf{latency gap}, where inference delays render actions obsolete; and the \textbf{control gap}, caused by the open-loop action chunk execution without real-time adjustment. In this work, we propose \textbf{DSDyn-VLA}, a Slow-Fast \textbf{D}ual-\textbf{S}tream \textbf{Dyn}amic manipulation framework that integrates motion-aware foresighted planning with real-time residual correction. The slow \textbf{Flow-Planner} serves as a macro-planner. By enhancing the VLA with optical flow for temporal perception and a future state awareness mechanism to preemptively offset inference latency, it produces globally consistent, motion-aware action chunks. Complementing this, the fast \textbf{Res-Refiner} employs a lightweight RL policy to inject high-frequency, closed-loop corrections into the planned action chunks based on real-time observations. In addition, we introduce \textbf{DynBench}, a MuJoCo-based benchmark for dynamic object manipulation that comprises nine tasks. Extensive experiments demonstrate that DSDyn-VLA reduces the failure rate by over 76\% compared to current SOTA method in high-latency setting on the Kinetix dynamic benchmark, while achieving about 6$\times$ the success rate of PI0.5 in real-world dynamic settings and about 5$\times$ on DynBench. We will open-source all the code and weights.
Chinese Translation
尽管视觉-语言-动作（VLA）模型在静态任务中表现出色，但在物体运动的动态环境中（例如，传送带操控）却面临挑战。我们识别出阻碍当前VLA在这些场景中表现的三个基本限制： extbf{感知差距}，即静态视觉输入缺乏时间运动线索； extbf{延迟差距}，即推理延迟使得动作变得过时；以及 extbf{控制差距}，由无实时调整的开环动作块执行引起。在本研究中，我们提出了 extbf{DSDyn-VLA}，一种慢-快 extbf{双流} extbf{动态}操控框架，结合了运动感知的前瞻性规划与实时残差纠正。慢速 extbf{流规划器}作为宏观规划器，通过增强VLA的光流以实现时间感知，并引入未来状态意识机制以预先抵消推理延迟，从而生成全球一致的、运动感知的动作块。与此相辅相成，快速 extbf{残差精炼器}采用轻量级强化学习策略，根据实时观察将高频闭环纠正注入到计划的动作块中。此外，我们引入了 extbf{DynBench}，一个基于MuJoCo的动态物体操控基准，包含九个任务。大量实验表明，在Kinetix动态基准的高延迟设置中，DSDyn-VLA的失败率比当前最先进的方法降低了超过76 ext%，同时在现实世界动态环境中实现了约6$ imes$的成功率，相较于PI0.5，而在DynBench上则实现了约5$ imes$的成功率。我们将开源所有代码和权重。
cs.RO / 69 / 2609.39203
Benchmarking EMlog Calibration for Autonomous Surface Vehicles
自主水面车辆的电磁日志校准基准测试
Abstract
Accurate velocity measurement is a fundamental requirement for autonomous surface and underwater vehicles. Commonly, velocity is provided by a Doppler velocity log (DVL) sensor, yet it becomes unavailable due to operational altitude constraints. In such situations, electromagnetic logs (EMLogs) provide a critically robust alternative for continuous velocity estimation. However, raw EMLog measurements are inherently corrupted by systematic errors, which need to be calibrated prior mission begins. Currently, a benchmarking comparative evaluation of how different calibration models perform under rapidly changing dynamic sea conditions is missing in the literature. To bridge this gap, this paper presents a comparative model-based calibration methodology that evaluates four distinct calibration models using two different estimation pipelines. The proposed framework is rigorously validated on a unique 221 minutes of continuous real-world telemetry collected from the MARVEL surface vehicle during dynamic sea trials. The dataset contains two different EMLogs and DVL recordings. Experimental results demonstrate that the bias and scale error model implemented with the Kalman filter improves the speed estimation by 71%. We also demonstrate that dynamical manoeuvres further improve the accuracy compared to standard straight-line paths, ultimately delivering a validated, real-time online calibration EMLog approach for autonomous surface vehicles.
Chinese Translation
准确的速度测量是自主水面和水下车辆的基本要求。通常，速度由多普勒速度日志（DVL）传感器提供，但由于操作高度的限制，DVL可能无法使用。在这种情况下，电磁日志（EMLogs）提供了一种关键的稳健替代方案，用于连续速度估计。然而，原始EMLog测量数据本质上受到系统误差的影响，这些误差需要在任务开始之前进行校准。目前，文献中缺乏对不同校准模型在快速变化的动态海洋条件下表现的基准比较评估。为填补这一空白，本文提出了一种基于模型的比较校准方法，评估四种不同的校准模型，使用两种不同的估计流程。所提出的框架在MARVEL水面车辆进行动态海试时收集的221分钟连续真实世界遥测数据上进行了严格验证。该数据集包含两种不同的EMLogs和DVL记录。实验结果表明，实施卡尔曼滤波器的偏差和尺度误差模型使速度估计提高了71%。我们还展示了与标准直线路径相比，动态机动进一步提高了准确性，最终为自主水面车辆提供了一种经过验证的实时在线校准EMLog方法。
cs.RO / 70 / 2609.39207
ASENA: Self-evolving Agents for Embodied Navigation
ASENA：自我进化的具身导航智能体
Abstract
We present ASENA, an embodied agent system that connects general-purpose coding agents to robot sensing, computation, supervised execution, and persistent experience. Agents can write and execute programs, inspect recorded outcomes, repair failures, and reuse notes and executable skills while keeping their model weights fixed. We further introduce ASENA-VLN, a 4B monocular navigation policy that serves as an optional tool within this programmable system. ASENA-VLN predicts body-frame trajectories for both extended routes and short-horizon behaviors using a shared vision-language decoder trained on route instructions, visual question answering, and a newly curated dataset of geometry-derived atomic navigation tasks. As a standalone policy, ASENA-VLN achieves state-of-the-art success rates of 68.7% on R2R and 70.2% on RxR. When integrated with a coding agent, learned navigation improves ASENA's success rate by 11 percentage points on both agentic benchmarks while reducing execution time. Through persistent workspace evolution and simulator feedback, ten passes over recurring 100-task subsets further improve success from 72% to 98% on R2R and from 65% to 89% on RxR. On embodied question answering, ASENA achieves state-of-the-art accuracy with fewer interaction steps. Finally, real-world demonstrations on a Unitree G1 combine search, visual inspection, spatial reasoning, and synthesized gestures without a pre-built map, illustrating how online programming extends robot behavior beyond route following and predefined skills.
Chinese Translation
我们提出了ASENA，一个将通用编码智能体与机器人感知、计算、监督执行和持久经验连接的具身智能体系统。智能体能够编写和执行程序，检查记录的结果，修复故障，并在保持模型权重不变的情况下重用笔记和可执行技能。我们进一步介绍了ASENA-VLN，一个4B单目导航策略，作为该可编程系统中的可选工具。ASENA-VLN使用在路线指令、视觉问答以及新整理的几何派生原子导航任务数据集上训练的共享视觉-语言解码器，预测身体框架下的轨迹，适用于扩展路线和短期行为。作为一个独立策略，ASENA-VLN在R2R上取得了68.7%的最先进成功率，在RxR上取得了70.2%的成功率。当与编码智能体集成时，学习到的导航使ASENA在两个智能体基准上的成功率提高了11个百分点，同时减少了执行时间。通过持续的工作空间演化和模拟器反馈，对重复的100任务子集进行十次处理，使得成功率在R2R上从72%提高到98%，在RxR上从65%提高到89%。在具身问答任务中，ASENA以更少的交互步骤实现了最先进的准确率。最后，在Unitree G1上的现实世界演示结合了搜索、视觉检查、空间推理和合成手势，而无需预先构建地图，展示了在线编程如何将机器人行为扩展到超越路线跟随和预定义技能的范畴。
cs.RO / 71 / 2609.39235
The Planning Limits of Latent World Models
潜在世界模型的规划限制
Abstract
World models offer a promising way to help robots understand how the physical world evolves and plan complex behaviours through imagination. Yet existing studies mainly demonstrate what these models can accomplish, leaving unclear when their predictions remain useful for planning and where they fail. We study this question using action-conditioned predictors built on five frozen self-supervised visual backbones: V-JEPA 2, V-JEPA 2.1, VideoMAEv2, VideoPrism, and DINOv2. We use frozen backbones to test representations intended to transfer across environments. We evaluate these models on diverse Meta-World manipulation tasks and real-robot interactions from BridgeData V2. We find that a world model guides action selection reliably only when the goal lies within, or slightly beyond, the trajectory it imagines during planning. With five-step rollouts, the length the predictor was trained on, the world model ranks actions reliably only for targets five to ten control steps ahead, whereas task goals lie 16 to 53 steps away. Neither an 81-fold larger predictor nor longer-rollout training extends this range; the encoder affects both range and closed-loop success, with V-JEPA 2.1 performing most consistently. More fundamentally, the limit persists under perfect prediction: using the real simulator, success falls from 92% to 41% as the target moves from five to twenty steps ahead of a five-step rollout. Planning therefore requires either longer imagined trajectories or closer subgoals. For distant goals, pure imagination succeeds in 23% of episodes, planning with feedback (MPC) raises success to 30%, imagining as far as the goal to 47%, and nearby expert subgoals to 76%. Used within its plannable range, a world model can also improve a vision-language-action (VLA) policy: choosing among eight actions the VLA proposes raises its success from 65% to 77% across 16 different tasks.
Chinese Translation
世界模型为机器人理解物理世界的演变以及通过想象规划复杂行为提供了有希望的方法。然而，现有研究主要展示了这些模型能够实现的功能，却不清楚它们的预测在何时对规划仍然有用，以及在何处失败。我们通过基于五个冻结的自监督视觉骨干网络（V-JEPA 2、V-JEPA 2.1、VideoMAEv2、VideoPrism 和 DINOv2）构建的动作条件预测器来研究这个问题。我们使用冻结的骨干网络来测试旨在跨环境转移的表示。我们在多样的 Meta-World 操作任务和来自 BridgeData V2 的真实机器人交互中评估这些模型。我们发现，世界模型仅在目标位于其规划过程中想象的轨迹内或稍微超出时，才能可靠地指导动作选择。在五步展开的情况下，预测器的训练长度，世界模型仅对距离目标五到十个控制步骤的动作进行可靠排名，而任务目标则距离16到53步。无论是81倍更大的预测器还是更长的展开训练都无法扩展这一范围；编码器影响范围和闭环成功率，其中 V-JEPA 2.1 的表现最为一致。更根本的是，这一限制在完美预测下依然存在：使用真实模拟器时，成功率从92%下降到41%，当目标从五步移动到二十步前方时。因此，规划需要更长的想象轨迹或更接近的子目标。对于远距离目标，纯想象在23%的情节中成功，带反馈的规划（MPC）将成功率提高到30%，想象到目标的距离提高到47%，而附近的专家子目标则提高到76%。在可规划范围内使用时，世界模型还可以改善视觉-语言-动作（VLA）策略：在八个 VLA 提议的动作中选择将其成功率从65%提高到77%，涵盖16个不同的任务。
cs.RO / 72 / 2609.39245
ReWAM: Reciprocal World Action Models for Interactive Autonomous Driving
ReWAM：用于互动自主驾驶的互惠世界行动模型
Abstract
In interactive scenarios, an autonomous driving system is required to generate ego actions under the influence of other agents' behaviors. Existing World Action Models (WAMs) typically model other agents as components of the world model rather than as decision-makers that fundamentally shape the action of the ego agent, which impairs their performance in dense interaction scenarios. We introduce Reciprocal World Action Models (ReWAM), a game-theoretic world action modeling framework that captures the reciprocal influence between the ego agent and other agents by representing them as conditional responders whose actions are mutually influenced. We instantiate this framework with a Level-$k$ response hierarchy, where role-specific ego and other action DiTs exchange compact strategy tokens through cross-agent attention while remaining grounded in a shared representation of the future driving world. To learn the response policy of the ego agent from demonstrations, we formulate expert actions as samples from the best response distribution and jointly optimize the entire hierarchy using conditional flow matching. Our framework is evaluated on the NAVSIM dataset and achieves state-of-the-art performance compared to baselines. The improvement is particularly significant in interactive scenarios, validating that modeling reciprocal responses provides a more effective foundation for interaction-aware world action generation.
Chinese Translation
在互动场景中，自主驾驶系统需要在其他智能体行为的影响下生成自我行动。现有的世界行动模型（World Action Models, WAMs）通常将其他智能体视为世界模型的组成部分，而不是作为根本上影响自我智能体行动的决策者，这在密集互动场景中削弱了它们的性能。我们提出了互惠世界行动模型（Reciprocal World Action Models, ReWAM），这是一种博弈论的世界行动建模框架，通过将其他智能体表示为条件响应者，捕捉自我智能体与其他智能体之间的互惠影响，其行动是相互影响的。我们用Level-$k$响应层次结构实例化该框架，其中角色特定的自我和其他行动的决策信息传递（Decision Information Tokens, DiTs）通过跨智能体注意力交换紧凑的策略令牌，同时基于未来驾驶世界的共享表示保持扎根。为了从示范中学习自我智能体的响应策略，我们将专家行动形式化为最佳响应分布的样本，并使用条件流匹配共同优化整个层次结构。我们的框架在NAVSIM数据集上进行了评估，并与基线相比实现了最先进的性能。特别是在互动场景中，改进尤为显著，验证了建模互惠响应为互动感知的世界行动生成提供了更有效的基础。
cs.RO / 73 / 2609.39304
Scale and Selection: What Makes Automatic Harness Evolution Work for Visual-Interface Robot Agents
规模与选择：自动化工具演化如何为视觉界面机器人代理工作
Abstract
When an off-the-shelf coding agent is used directly as a robot policy, observing a browser-based 3D interface through screenshots and acting by posing a virtual target gripper through a few tools, the agent's harness, its prompts, tools, and control rules, largely determines success, and until now it has been written by hand. We show that this harness can be improved automatically by another coding agent, the optimizer agent, and report two findings about what makes it work. First, the number of rollouts the optimizer agent sees per round governs whether the evolved harness is trustworthy, generalizes, and improves steadily. A single rollout is a noisy binary outcome, so with few rollouts per round a revision can be promoted on luck; enlarging the batch raises the signal-to-noise ratio of every promotion decision. Holding rounds fixed and growing the training set from 5 to 100 rollouts, held-out success rises from 47% to 67%, while small training sets overfit, reaching 70% on training tasks but only 54% held-out. Second, the optimizer agent must not be given free rein. With every revision it proposes accepted unconditionally, performance drifts downward within ten rounds as ill-judged edits accumulate; adding the most basic safeguard, Champion-Challenger selection that promotes a revision only if it strictly beats the incumbent on the same fixed evaluation set, turns the same loop into one that raises held-out success from 51% to 67% over 30 rounds. Automatic harness evolution for visual-interface robot agents is thus feasible, but its gains hinge on the rollout scale behind each decision and on how the optimizer agent's revisions are selected.
Chinese Translation
当一个现成的编码代理被直接用作机器人策略时，通过屏幕截图观察基于浏览器的3D界面，并通过几个工具操控虚拟目标夹具，代理的工具、提示、工具和控制规则在很大程度上决定了成功与否，而直到现在这些都是手动编写的。我们展示了这个工具可以通过另一个编码代理，即优化代理，自动改进，并报告了两个关于其有效性的发现。首先，优化代理每轮观察到的回合数决定了演化的工具是否值得信赖、是否具有泛化能力以及是否能够稳定改进。单个回合是一个噪声较大的二元结果，因此每轮回合数较少时，修订可能仅凭运气被提升；扩大批量提高了每个提升决策的信噪比。固定回合数并将训练集从5个回合扩大到100个回合时，保留的成功率从47%上升到67%，而小训练集则过拟合，在训练任务上达到70%，但在保留集上仅为54%。第二，优化代理不应被赋予完全自由。每个修订如果被无条件接受，性能将在十轮内逐渐下降，因为不当的编辑会累积；添加最基本的保护措施，即冠军-挑战者选择，仅在修订严格优于同一固定评估集上的现任者时才提升修订，使得同一循环在30轮内将保留成功率从51%提升至67%。因此，视觉界面机器人代理的自动化工具演化是可行的，但其收益取决于每个决策背后的回合规模以及优化代理修订的选择方式。
cs.RO / 74 / 2609.39305
FORTE: Forecasting Occupancy for Spatiotemporal Risk-Aware Planning in Dynamic Environments
FORTE：动态环境中基于时空风险感知规划的占用预测
Abstract
Safe navigation in dynamic environments requires anticipating future environmental states to account for spatiotemporal risks, specifically when and where collisions may occur. To this end, occupancy grid map (OGM) prediction has been widely adopted as an effective approach. However, existing OGM-based navigation methods often struggle to achieve accurate and efficient forecasting and fail to fully exploit the temporal information in predicted OGMs during planning. To address these challenges, we propose FORTE, a navigation framework that directly exploits the spatiotemporal evolution of predicted occupancy from the perspectives of spatiotemporal occupancy overlap and occupancy directivity. Based on these properties, FORTE evaluates multiple topology-distinct paths and selects the suitable one without explicit object detection or tracking. To support online planning, we formulate a latent diffusion model-based OGM predictor that generates the entire forecast horizon in a non-autoregressive manner while maintaining temporal consistency through temporal shift modules. Extensive evaluations demonstrate that FORTE outperforms state-of-the-art baselines. For prediction, FORTE achieves up to 215.3% higher IoU and 5.24x faster inference; for navigation, it yields up to a 3.5x higher success rate.
Chinese Translation
在动态环境中安全导航需要预测未来的环境状态，以考虑时空风险，特别是碰撞可能发生的时间和地点。为此，占用网格地图（Occupancy Grid Map, OGM）预测已被广泛采用作为一种有效的方法。然而，现有的基于OGM的导航方法往往难以实现准确和高效的预测，并且未能充分利用预测OGM中的时间信息进行规划。为了解决这些挑战，我们提出了FORTE，一个导航框架，直接从时空占用重叠和占用方向性的角度利用预测占用的时空演变。基于这些特性，FORTE评估多个拓扑不同的路径，并在没有明确的物体检测或跟踪的情况下选择合适的路径。为了支持在线规划，我们构建了一个基于潜在扩散模型的OGM预测器，该预测器以非自回归的方式生成整个预测范围，同时通过时间位移模块保持时间一致性。大量评估表明，FORTE在性能上优于最先进的基线。在预测方面，FORTE的IoU提高了高达215.3%，推理速度快了5.24倍；在导航方面，其成功率提高了高达3.5倍。
cs.RO / 75 / 2609.39322
A Biophysically Detailed C. elegans Circuit as a Task-Agnostic Dynamical Core for Visually Robust Robot Manipulation
作为任务无关动态核心的生物物理详细线虫电路用于视觉鲁棒的机器人操作
Abstract
Robot policies are usually trained for one task, one body and one visual environment, and generalize poorly beyond these conditions. Whether a nervous system can instead supply the sensorimotor computation through its evolved wiring and biophysics remains unresolved. Here we embed a biophysically detailed Caenorhabditis elegans sensorimotor circuit - 136 multicompartment neurons with realistic morphologies and electrophysiological characteristics - as the dynamical core of a visuomotor policy. Only thin task-specific adapters are trained; the core's synaptic weights stay fixed while its membrane voltages evolve freely. Across different MetaWorld tasks the core matches or exceeds diffusion-policy, action-chunking-transformer and neural-circuit-policy baselines, and degrades less under visual perturbations. Replacing the core with generic network models such as MLP, LSTM, transformer or reservoir networks removes the advantage. Furthermore, on a real robotic arm the core withstands diverse visual perturbations that collapse the baselines. Our results suggest that visual robustness can be inherited from biophysically detailed circuit dynamics rather than learned by task-specific controllers.
Chinese Translation
机器人策略通常针对单一任务、单一身体和单一视觉环境进行训练，并且在这些条件之外的泛化能力较差。神经系统是否能够通过其进化的连接和生物物理学提供传感运动计算仍未得到解决。在此，我们将一个生物物理详细的秀丽隐杆线虫（Caenorhabditis elegans）传感运动电路——由136个具有真实形态和电生理特征的多腔体神经元组成——嵌入作为视觉运动策略的动态核心。仅训练薄弱的任务特定适配器；核心的突触权重保持固定，而其膜电压自由演变。在不同的MetaWorld任务中，核心的表现与扩散策略、动作分块变换器和神经电路策略的基线相匹配或超越，并且在视觉扰动下的退化程度较低。用通用网络模型如多层感知器（MLP）、长短期记忆网络（LSTM）、变换器（transformer）或水库网络替换核心则会失去这种优势。此外，在真实的机器人手臂上，核心能够抵御各种视觉扰动，而这些扰动会导致基线崩溃。我们的结果表明，视觉鲁棒性可以从生物物理详细的电路动态中继承，而不是通过任务特定的控制器学习得来。
cs.RO / 76 / 2609.39323
HiWE: Hierarchical World Knowledge Model with Visual Keypoint Enhancement for Zero-Shot 3D Path Planning
HiWE：具有视觉关键点增强的层次世界知识模型用于零-shot 3D路径规划
Abstract
Robot demonstration generation requires a system to identify where an interaction should occur, plan a feasible motion, and execute the required contact. HiWE connects these decisions through a point-based interface between visual grounding and language-based planning. PointVLM is instruction-tuned to associate task-relevant objects with image coordinates using a mixture of point annotations, segmentation-derived samples, robot observations, and visual question answering data. Depth measurements lift these predictions into a semantic 3D representation. A language planner, 3DLLM, uses this representation to specify end-effector waypoints and gripper commands, while a hybrid grasping module resolves local grasp poses. The evaluation covers 14 simulated manipulation tasks and four physical-robot tasks, together with ablations of the visual training data, spatial inputs, and grasp selection. Here, zero-shot execution refers to deployment without task-specific demonstration training; the visual model uses existing robot data during fine-tuning. This paper describes the original point-based formulation of the framework; its relationship to the subsequent GeneralVLA extension is detailed in the introduction.
Chinese Translation
机器人演示生成需要一个系统来识别交互发生的位置、规划可行的运动，并执行所需的接触。HiWE通过视觉基础与基于语言的规划之间的点状接口连接这些决策。PointVLM经过指令调优，以通过点注释、分割衍生样本、机器人观察和视觉问答数据的混合，将与任务相关的对象与图像坐标关联。深度测量将这些预测提升为语义3D表示。语言规划器3DLLM利用该表示来指定末端执行器的航点和夹持器命令，而混合抓取模块解决局部抓取姿态。评估涵盖14个模拟操作任务和四个物理机器人任务，以及视觉训练数据、空间输入和抓取选择的消融实验。在这里，零-shot执行指的是在没有特定任务演示训练的情况下进行部署；视觉模型在微调过程中使用现有的机器人数据。本文描述了该框架的原始点状公式；其与后续GeneralVLA扩展的关系在引言中进行了详细说明。
cs.RO / 77 / 2609.39324
MotionWeave: Learning Motion-Centered Future Dynamics for Vision-Language-Action Policies
MotionWeave：学习以运动为中心的未来动态以实现视觉-语言-动作策略
Abstract
Vision-Language-Action (VLA) models have recently incorporated world models to provide richer dynamic supervision beyond sparse action labels. However, explicitly predicting future images or videos may include control-irrelevant appearance, while guidance derived from holistic future visual representations and shared global action features may fail to establish timestep-specific correspondence between actions and local visual changes. To address this issue, we propose MotionWeave, a motion-centric future-dynamics framework for action-chunk prediction with two modules: the Action-Induced Motion Grounder (AIMG) and the Horizon Residual Composer (HRC). Specifically, AIMG conditions on action and proprioceptive representations to construct horizon-specific queries that localize interaction regions associated with each future action timestep from current visual tokens. HRC extracts differences between interaction representations at adjacent horizons, encodes them as temporal motion cues, and injects them into action tokens through a gated residual. During training, robot-arm masks rendered from future frames are used to construct KL-based motion-grounding supervision, while inference uses only the current observation. On six MetaWorld tasks, MotionWeave achieves a 75.3% average success rate, an absolute gain of 8.6% over {\pi}0 (66.7%), especially on sustained-interaction tasks. Our code is available at https://github.com/autu-mn/MotionWeave.
Chinese Translation
视觉-语言-动作（VLA）模型最近引入了世界模型，以提供超越稀疏动作标签的更丰富的动态监督。然而，显式预测未来图像或视频可能包含与控制无关的外观，而从整体未来视觉表示和共享全局动作特征中获得的指导可能无法建立动作与局部视觉变化之间的时间步特定对应关系。为了解决这个问题，我们提出了MotionWeave，这是一个以运动为中心的未来动态框架，用于动作块预测，包含两个模块：动作引导运动基础模块（AIMG）和地平线残差合成器（HRC）。具体而言，AIMG基于动作和本体感觉表示构建地平线特定查询，以从当前视觉标记中定位与每个未来动作时间步相关的交互区域。HRC提取相邻地平线之间的交互表示差异，将其编码为时间运动线索，并通过门控残差将其注入动作标记。在训练过程中，从未来帧渲染的机器人臂掩码用于构建基于KL的运动基础监督，而推理仅使用当前观测。在六个MetaWorld任务中，MotionWeave实现了75.3%的平均成功率，相较于{ ext{π}}0（66.7%）有8.6%的绝对提升，尤其是在持续交互任务中表现突出。我们的代码可在https://github.com/autu-mn/MotionWeave获取。
cs.RO / 78 / 2609.39375
Beyond the Current Scene: Event-Referential Grasping with Active View Selection
超越当前场景：具有主动视角选择的事件指称抓取
Abstract
A robot that observes people interacting with objects should be able to carry out later requests that refer back to those interactions. Such requests may specify a grasp target by the role it played in a past event rather than by its name or appearance. Moreover, the target may no longer be visible when the robot is asked to act. We present BeyondSCe, a zero-shot robotic grasping system for this event-referential setting. Given the event history and the current scene, the system identifies the requested object or part and localizes it for grasping. If the target is occluded, it combines an event prior recovered from the history with current scene geometry to select camera viewpoints likely to reveal the target. The system uses pretrained models without additional task-specific training. In real-robot experiments with a single wrist-mounted RGB-D camera, it achieves grasp success rates of 76% and 77% for initially visible and occluded targets, respectively, compared with 40% and 55% for the strongest baseline in each condition. On four additional scenes with heavy occlusion, it increases grasp success rates from 75% to 95% while reducing the mean number of views from 3.35 to 2.20, compared with an active-perception baseline given the target's ground-truth 3D bounding box.
Chinese Translation
观察人们与物体互动的机器人应能够执行后续请求，这些请求可以追溯到这些互动。这些请求可能通过物体在过去事件中所扮演的角色来指定抓取目标，而不是通过其名称或外观。此外，当机器人被要求执行动作时，目标可能不再可见。我们提出了BeyondSCe，这是一个针对这种事件指称设置的零-shot 机器人抓取系统。该系统根据事件历史和当前场景，识别请求的物体或部分，并为抓取进行定位。如果目标被遮挡，它将从历史中恢复的事件先验与当前场景几何结合，以选择可能揭示目标的相机视角。该系统使用预训练模型，而无需额外的特定任务训练。在使用单个腕部安装的RGB-D相机进行的真实机器人实验中，对于最初可见和被遮挡的目标，其抓取成功率分别达到76%和77%，而在每种条件下最强基线的成功率仅为40%和55%。在四个具有严重遮挡的额外场景中，它将抓取成功率从75%提高到95%，同时将平均视角数量从3.35减少到2.20，相较于给定目标真实3D边界框的主动感知基线。
cs.RO / 79 / 2609.39384
RoboAssist: Interactive Human-Humanoid Planning for Long-Horizon Surgical Assistance
RoboAssist：用于长期手术辅助的互动人类-类人机器人规划
Abstract
Long-horizon surgical assistance requires humanoid robots to coordinate with evolving human activities while maintaining safety across planning and execution. We present RoboAssist, an agent-based framework for interactive human-humanoid planning that integrates workflow reasoning, task coordination, and cross-layer safety. At its core is an asymmetric dual-track representation that separates partially observed human process states from executable robot task sequences. By updating human-process estimates, scene context, and task dependencies online, RoboAssist revalidates the remaining task sequence and replans only the affected suffix when workflow requests change. A cross-layer safety architecture combines preventive navigation regulation, reactive regulation during close-range handover, and independent whole-body runtime supervision. This design couples online task coordination with safety constraints throughout execution. We demonstrate the framework on a Unitree G1 humanoid robot in long-horizon, multi-stage simulated surgical assistance scenarios encompassing multimodal interaction, instrument handling, medical material transport, navigation, and safe human-robot handover. Experiments show multi-stage task completion and adaptation to workflow-request changes. A targeted full-replanning ablation shows that residual replanning reduces plan-update latency and post-update token usage. Separate safety experiments demonstrate complementary protection across navigation, handover, and runtime supervision. Additional results and demonstrations are available online at https://roboassist.github.io.
Chinese Translation
长期手术辅助需要类人机器人在规划和执行过程中与不断变化的人类活动协调，同时保持安全性。我们提出了RoboAssist，这是一个基于代理的互动人类-类人机器人规划框架，集成了工作流程推理、任务协调和跨层安全性。其核心是一个不对称的双轨表示法，将部分观察到的人类过程状态与可执行的机器人任务序列分离。通过在线更新人类过程估计、场景上下文和任务依赖关系，RoboAssist在工作流程请求变化时重新验证剩余任务序列，并仅重新规划受影响的后缀。跨层安全架构结合了预防性导航调节、近距离交接时的反应性调节，以及独立的全身运行时监督。该设计在执行过程中将在线任务协调与安全约束相结合。我们在Unitree G1类人机器人上展示了该框架在长期、多阶段模拟手术辅助场景中的应用，包括多模态互动、工具处理、医疗材料运输、导航和安全的人机交接。实验表明多阶段任务的完成和对工作流程请求变化的适应。针对全面重新规划的消融实验表明，剩余重新规划减少了计划更新延迟和更新后令牌使用。单独的安全实验展示了在导航、交接和运行时监督方面的互补保护。更多结果和演示可在线访问：https://roboassist.github.io。
cs.RO / 80 / 2609.39388
UniWAM Technical Report: Unified Mobile Manipulation via Mixed-Stream World-Action Modeling and Manipulation Anchor Pose Supervision
UniWAM技术报告：通过混合流世界-动作建模和操作锚点姿态监督实现统一移动操控
Abstract
Mobile manipulation requires precise navigation to a manipulation-ready pose followed by reliable object interaction. These two stages differ in action spaces and visual requirements, which complicates unified policy learning. In addition, collecting diverse real-world navigation data with explicit manipulation-ready pose supervision remains costly and difficult to scale. We introduce UniWAM, a unified mixed-stream world-action model with separate action encoders and output heads for navigation and manipulation, sharing a common backbone. This design supports joint representation learning on independently sampled navigation and manipulation data. UniWAM supports independent inference for either stream and batch-parallel inference for both. We further introduce Manipulation Anchor Pose (MAP) supervision for where to stop and how to orient for manipulation. An automated pipeline constructs MAP-Data from large-scale 3D scenes, yielding over 1.5 million episodes and 7,500 hours. MAP-Data provides per-frame target-object bounding boxes and image-plane MAP coordinates as auxiliary navigation supervision. Together with projected end-effector trajectories for manipulation, these prediction targets provide stream-specific image-plane supervision for action learning from egocentric observations. With large-scale MAP-Data, UniWAM outperforms the strongest external baselines on our MAP-Bench by 30.1\% in position error and 44.0\% in heading error. Across 24 real-robot tasks, UniWAM achieves leading results in MAP navigation and mobile manipulation, with competitive manipulation performance. We have released code, data, and benchmark.
Chinese Translation
移动操控需要精确导航至准备操控的姿态，随后进行可靠的物体交互。这两个阶段在动作空间和视觉需求上存在差异，从而使得统一策略学习变得复杂。此外，收集具有明确操控准备姿态监督的多样化真实世界导航数据仍然成本高昂且难以扩展。我们提出了UniWAM，这是一种统一的混合流世界-动作模型，具有独立的动作编码器和导航与操控的输出头，共享一个公共骨干网络。该设计支持在独立采样的导航和操控数据上进行联合表示学习。UniWAM支持对任一流的独立推理以及对两个流的批量并行推理。我们进一步引入了操控锚点姿态（Manipulation Anchor Pose, MAP）监督，用于确定何时停止以及如何调整姿态以进行操控。一个自动化管道从大规模3D场景中构建MAP数据，生成超过150万集和7500小时的数据。MAP数据提供每帧目标物体的边界框和图像平面MAP坐标，作为辅助导航监督。结合用于操控的投影末端执行器轨迹，这些预测目标为从自我中心观察中进行动作学习提供了流特定的图像平面监督。凭借大规模的MAP数据，UniWAM在我们的MAP基准测试中在位置误差上超越了最强的外部基线30.1%，在航向误差上超越44.0%。在24个真实机器人任务中，UniWAM在MAP导航和移动操控方面取得了领先结果，并展现出竞争力的操控性能。我们已发布代码、数据和基准。
cs.RO / 81 / 2609.39403
IronMind: Scaling Humanoid Dexterous Manipulation via Camera-Space Ego-Centric Pretraining
IronMind：通过相机空间自我中心预训练扩展类人灵巧操作
Abstract
Egocentric human video offers a scalable data source for dexterous manipulation, yet using it to train humanoid robots presents two challenges: (1) an embodiment gap, as human hands differ structurally from robot end-effectors and low-cost egocentric recordings lack the torso kinematics required by conventional retargeting; and (2) heterogeneous data quality, including noisy hand-pose tracking and weakly aligned text annotations. We introduce IronMind, a vision-language-action (VLA) model that uses egocentric human video and heterogeneous robot data to pretrain policies for humanoid dexterous manipulation. To bridge the embodiment gap, IronMind bypasses explicit body-retargeting by using a camera-space action representation, the native reference space of egocentric video, and semantically aligning robot and human action dimensions. Across total pretraining budgets from 250 to 10,000 hours, validation loss decreases approximately log-linearly with data scale. Larger pretraining budgets also improve out-of-distribution real-robot manipulation after post-training: across six challenging tasks with unseen objects, affordances, and reasoning prompts, the 10,000-hour model achieves a 55.0% success rate, compared with at most 11.7% for every pretraining budget up to 5,000 hours and 5.0% without pretraining. At the same pretraining budget, the camera-space action representation also outperforms the torso-frame baseline. Together, these findings support pretraining with a camera-space action representation on large-scale human egocentric data as a scalable foundation for humanoid robot manipulation.
Chinese Translation
自我中心的人类视频为灵巧操作提供了可扩展的数据源，但使用它来训练类人机器人面临两个挑战：（1）体现差距，因为人类手与机器人末端执行器在结构上存在差异，并且低成本的自我中心录音缺乏传统重定向所需的躯干运动学；（2）异质数据质量，包括嘈杂的手部姿态跟踪和弱对齐的文本注释。我们提出了IronMind，一个视觉-语言-动作（VLA）模型，利用自我中心的人类视频和异质机器人数据来预训练类人灵巧操作的策略。为了弥合体现差距，IronMind通过使用相机空间动作表示（自我中心视频的本地参考空间）来绕过显式的身体重定向，并在语义上对齐机器人和人类的动作维度。在250到10,000小时的总预训练预算下，验证损失随着数据规模的增加大约以对数线性方式减少。更大的预训练预算在后续训练后也改善了分布外真实机器人操作：在六个具有未见物体、可用性和推理提示的挑战性任务中，10,000小时模型的成功率达到55.0%，而每个预训练预算在5,000小时以内的最大成功率为11.7%，没有预训练的成功率为5.0%。在相同的预训练预算下，相机空间动作表示的表现也优于躯干框架基线。这些发现共同支持在大规模人类自我中心数据上使用相机空间动作表示进行预训练，作为类人机器人操作的可扩展基础。
cs.RO / 82 / 2609.39416
Communication-Free Distributed Multi-Robot Task Allocation under Partial Observations Using Labeled Multi-Bernoulli Filtering
基于标记多伯努利滤波的部分观测下无通信分布式多机器人任务分配
Abstract
This paper proposes a communication-free multi-robot task allocation framework based solely on local observations. In this study, tasks are defined as reaching target locations. Each robot estimates the positions of neighboring robots using a Labeled Multi-Bernoulli (LMB) filter and independently assigns tasks through a greedy auction-based strategy. By continuously updating state estimates and reallocating tasks during execution, the proposed method enables decentralized coordination without explicit communication. Monte Carlo simulations demonstrate that the proposed method enables effective cooperative task allocation without inter-robot communication while remaining robust to measurement clutter and observation uncertainty.
Chinese Translation
本文提出了一种完全基于局部观测的无通信多机器人任务分配框架。在本研究中，任务被定义为到达目标位置。每个机器人使用标记多伯努利（Labeled Multi-Bernoulli, LMB）滤波器估计邻近机器人的位置，并通过贪婪拍卖策略独立分配任务。通过在执行过程中持续更新状态估计和重新分配任务，所提方法实现了无明确通信的分散协调。蒙特卡洛模拟表明，该方法能够有效地进行合作任务分配，而无需机器人之间的通信，同时对测量杂波和观测不确定性保持鲁棒性。
cs.RO / 83 / 2609.39507
LIBERO-Agent: Evaluating General-Purpose Agents for Direct Embodied Manipulation
LIBERO-Agent：评估通用智能体在直接具身操作中的表现
Abstract
General-purpose agents can plan, use tools, and revise their behavior from feedback, but it remains unclear whether these capabilities transfer from digital environments to embodied manipulation. To investigate this question, we introduce LIBERO-Agent, an agent-native benchmark for evaluating these agents in robot manipulation tasks. Rather than asking agents to submit task-level Python control programs or operate through high-level robot skills, LIBERO-Agent provides an interactive robotic environment where agents can select which observations to inspect, process them with their own tools, and issue native action commands. LIBERO-Agent integrates 200 tasks into a common interaction framework and provides a 30-task primary suite that separates perception, short-horizon execution, and long-horizon composition. Results reveal a pronounced reliability gap: while agents perform well on perception and easy short-horizon tasks, their performance degrades substantially on hard short-horizon and long-horizon tasks. Richer observations improve short-horizon manipulation, while demonstration benefits depend on the agent and format. Among these agents, GPT-6 Astra achieves the strongest overall performance. Further analysis shows its major advantage lies in mechanism interaction, especially when sustained physical contact is needed, while its remaining failures stem from cross-stage interference and geometric errors.
Chinese Translation
通用智能体能够进行规划、使用工具并根据反馈修正其行为，但这些能力是否能够从数字环境转移到具身操作中仍不明确。为探讨这一问题，我们引入了LIBERO-Agent，这是一个用于评估这些智能体在机器人操作任务中的原生基准。LIBERO-Agent并不要求智能体提交任务级的Python控制程序或通过高层次的机器人技能进行操作，而是提供了一个交互式机器人环境，智能体可以选择要检查的观察结果，使用自己的工具处理这些观察，并发出原生的动作命令。LIBERO-Agent将200个任务整合到一个共同的交互框架中，并提供了一个包含30个任务的主要套件，分别关注感知、短期执行和长期组合。结果显示出明显的可靠性差距：尽管智能体在感知和简单的短期任务上表现良好，但在困难的短期和长期任务上的表现显著下降。更丰富的观察结果改善了短期操作，而演示的效果则依赖于智能体及其格式。在这些智能体中，GPT-6 Astra的整体表现最为出色。进一步分析表明，其主要优势在于机制交互，尤其是在需要持续物理接触时，而其余失败则源于跨阶段干扰和几何错误。
cs.RO / 84 / 2609.39526
Discrete Forcing: Infusing Discrete Guidance into Continuous Denoising for Few-Step Action Experts
离散强制：将离散指导融入连续去噪以实现少步动作专家
Abstract
Efficient action generation in vision-language-action (VLA) models requires capturing both coarse action structure and fine-grained details. Discrete action tokens provide compact structural representations but sacrifice precision, while continuous action tokens offer high precision but often require multiple denoising steps. We introduce Discrete Forcing, a flow-matching framework that combines these representations through an explicit coarse-to-fine generation process. It first predicts discrete action tokens to establish a coarse action structure, then uses them to guide continuous action refinement. The discrete and continuous components share a common diffusion transformer backbone with specialized branches, maintaining a parameter count comparable to a conventional single-branch model while requiring only one forward pass per branch. Extensive evaluations across multiple benchmarks demonstrate improved performance and faster inference over a parameter-matched continuous action expert, with consistent performance gains as model capacity increases. Real-world experiments further demonstrate improvements on high-precision and dynamic manipulation tasks.
Chinese Translation
在视觉-语言-动作（VLA）模型中，高效的动作生成需要捕捉粗略的动作结构和细致的细节。离散动作标记提供了紧凑的结构表示，但牺牲了精度，而连续动作标记则提供了高精度，但通常需要多个去噪步骤。我们提出了离散强制（Discrete Forcing），这是一种流匹配框架，通过显式的粗到细生成过程结合了这两种表示。该框架首先预测离散动作标记以建立粗略的动作结构，然后利用这些标记指导连续动作的精细化。离散和连续组件共享一个共同的扩散变换器主干，并具有专门的分支，保持与传统单分支模型相当的参数数量，同时每个分支仅需一次前向传递。通过在多个基准上的广泛评估，显示出在参数匹配的连续动作专家上性能提升和推理速度加快，随着模型容量的增加，性能提升保持一致。实际实验进一步证明了在高精度和动态操作任务上的改进。
cs.RO / 85 / 2609.39558
General Performance Guarantee for Human Torque Estimation-Based Task-Agnostic Assistive Exoskeleton Control
基于人类扭矩估计的任务无关助力外骨骼控制的一般性能保证
Abstract
Accurate human torque estimation is crucial for enabling task-agnostic control in robotic exoskeleton systems. However, estimation errors may cause mismatches between the robot assistance and the human intention, degrading controllability and task performance. In this paper, we address this issue by formally defining matched assistance as scenarios in which the robot positively contributes to human movement. Based on this definition, we develop a theoretical framework to design the robot's desired interaction torque that guarantees a lower bound on the matched assistance probability. Importantly, the proposed guarantee holds over the entire torque distribution, including unseen data beyond the training tasks. This provides our method with strong reliability and generalization, both of which are critical for effective exoskeleton control. The proposed strategy is implemented on the ABLE upper-limb exoskeleton and evaluated in a multi-task setup. Experimental results validate the theoretical guarantees and demonstrate that the proposed strategy achieves effective general performance across several tasks, guaranteeing movement smoothness while reducing human physical effort.
Chinese Translation
准确的人类扭矩估计对于实现机器人外骨骼系统中的任务无关控制至关重要。然而，估计误差可能导致机器人辅助与人类意图之间的不匹配，从而降低可控性和任务性能。本文通过正式定义匹配辅助为机器人对人类运动产生积极贡献的场景，来解决这一问题。基于这一定义，我们开发了一个理论框架，以设计机器人的期望交互扭矩，从而保证匹配辅助概率的下限。重要的是，所提出的保证适用于整个扭矩分布，包括训练任务之外的未见数据。这为我们的方法提供了强大的可靠性和泛化能力，这两者对于有效的外骨骼控制至关重要。所提出的策略在ABLE上肢外骨骼上实施，并在多任务设置中进行了评估。实验结果验证了理论保证，并表明所提出的策略在多个任务中实现了有效的整体性能，保证了运动的平滑性，同时减少了人类的身体努力。
cs.RO / 86 / 2609.39570
Sparse Planner: A Hybrid Planner for Efficient Sampling via a Conditional Variational Autoencoder
稀疏规划器：一种通过条件变分自编码器实现高效采样的混合规划器
Abstract
Trajectory planning is a core component of autonomous driving systems, where real-time performance and solution quality directly affect safety and reliability. Sample-Based Motion Planning (SBMP) is widely adopted for its ability to approximate near-optimal solutions through parameter space sampling. However, achieving high-quality trajectories typically requires dense sampling, leading to substantial computational overhead and significant runtime variability in complex traffic scenarios. To address this limitation, we propose a Sparse Planner (SP) that improves sampling efficiency by learning the conditional relationship between scene context and effective trajectory parameters using a Conditional Variational Autoencoder (CVAE). By modeling the structure of high-quality sampling distributions, SP directly generates cost-effective samples in the parameter space, significantly reducing the required sampling density while preserving solution quality. Experimental results show that SP achieves lower trajectory cost than the state-of-the-art FISS+ planner while using only one-eighth of the sampling density. In addition, SP demonstrates improved distance-keeping capability in obstacle-rich scenarios and maintains reduced and more stable runtime characteristics, indicating enhanced computational efficiency and predictable runtime behavior.
Chinese Translation
轨迹规划是自动驾驶系统的核心组成部分，其中实时性能和解决方案质量直接影响安全性和可靠性。基于样本的运动规划（Sample-Based Motion Planning, SBMP）因其能够通过参数空间采样近似最优解而被广泛采用。然而，实现高质量轨迹通常需要密集采样，这导致在复杂交通场景中计算开销大且运行时间变动显著。为了解决这一限制，我们提出了一种稀疏规划器（Sparse Planner, SP），通过使用条件变分自编码器（Conditional Variational Autoencoder, CVAE）学习场景上下文与有效轨迹参数之间的条件关系，从而提高采样效率。SP通过建模高质量采样分布的结构，在参数空间中直接生成成本效益高的样本，显著降低所需的采样密度，同时保持解决方案质量。实验结果表明，SP在仅使用八分之一的采样密度的情况下，轨迹成本低于最先进的FISS+规划器。此外，SP在障碍物丰富的场景中表现出更好的保持距离能力，并保持了更低且更稳定的运行时间特性，表明其计算效率得到了提升，运行时间行为更加可预测。
cs.RO / 87 / 2609.39575
ECHO-G: Embodied Co-speech Humanoid mOtion Generation
ECHO-G：具身共语类人机器人运动生成
Abstract
Generating full-body co-speech motion for humanoid robots requires coordinating speech prosody, linguistic content, and embodiment-specific motion. To this end, we present ECHO-G, a framework jointly conditioned on speech audio and timed transcripts. Its Speech-Grounded Diffusion Transformer (SGDiT) combines frame-aligned acoustic features with token-level linguistic context, preserving their distinct granularities. Trained with rectified flow matching, it models one-to-many utterance-motion relationships directly in robot space. To support training and evaluation, we introduce a BEAT2-derived audio-text-robot dataset and a benchmark covering co-speech characteristics, robot-motion quality, and runtime efficiency. Comparative evaluation supports direct robot-space generation over the evaluated human-motion generation and retargeting pipelines, while modality ablations highlight the benefits of joint audio-text conditioning. We further demonstrate deployment on a physical humanoid robot. A complementary video-rating study also favors joint conditioning over the alternatives. The dataset and training, inference, and evaluation code are available through our project page.
Chinese Translation
为类人机器人生成全身共语运动需要协调语音韵律、语言内容和具身特定的运动。为此，我们提出了ECHO-G，一个基于语音音频和时间转录的联合条件框架。其语音基础扩散变换器（Speech-Grounded Diffusion Transformer，SGDiT）结合了帧对齐的声学特征与标记级语言上下文，保留了它们各自的细粒度特征。通过修正流匹配进行训练，它直接在机器人空间中建模一对多的发声-运动关系。为了支持训练和评估，我们引入了一个基于BEAT2的音频-文本-机器人数据集和一个基准，涵盖了共语特征、机器人运动质量和运行效率。比较评估支持在评估的人类运动生成和重定向管道中直接在机器人空间生成，而模态消融实验突显了联合音频-文本条件的优势。我们进一步展示了在物理类人机器人上的部署。一个补充的视频评分研究也倾向于联合条件而非其他方法。数据集以及训练、推理和评估代码可通过我们的项目页面获取。
cs.RO / 88 / 2609.39594
Neuro-Symbolic Predicate Learning for Semantic Safe Robot Control
用于语义安全机器人控制的神经符号谓词学习
Abstract
As robots are increasingly deployed in everyday environments, ensuring their safety has become a central challenge. Existing methods often encode safety requirements as opaque mathematical/logical formulations or dense cost functions. While effective in specific tasks, they remain difficult to interpret, tightly coupled to individual tasks, and offer limited insight into why a robot action is considered safe or unsafe. To address this limitation, we propose ``Neuro-Symbolic Predicate Learning for Semantic Safe Robot Control'' (NEUPRO), which leverages a differentiable reasoner that can learn reusable safety representations from human-specified safety knowledge. NEUPRO allows practitioners to express task-related safety requirements as transparent symbolic rules, while enabling gradients to propagate through these rules to a feature extractor that maps raw observations to safety-relevant concepts. As a result, the learned feature extractor is (softly) grounded in human-understandable semantics, supports transparent constraint evaluation, and is transferable across tasks. By coupling interpretability with differentiability, NEUPRO moves beyond opaque cost design toward reusable safety reasoning. To evaluate NEUPRO's capability, we collect and release REASON, the first real robot benchmark dataset for interpretable robot safety specification. Experiments on REASON show that NEUPRO learns safety-critical features that generalize across tasks, mitigate the interpretability limitations of conventional black-box cost formulations, and provide explicit explanations of safety violation.
Chinese Translation
随着机器人在日常环境中的应用日益增多，确保其安全性已成为一个核心挑战。现有方法通常将安全要求编码为不透明的数学/逻辑公式或复杂的成本函数。尽管在特定任务中有效，但它们难以解释，与单个任务紧密耦合，并且对为何某个机器人动作被认为是安全或不安全提供的见解有限。为了解决这一局限性，我们提出了“用于语义安全机器人控制的神经符号谓词学习”（NEUPRO），该方法利用可微分推理器，从人类指定的安全知识中学习可重用的安全表示。NEUPRO使从业者能够将与任务相关的安全要求表达为透明的符号规则，同时允许梯度通过这些规则传播到特征提取器，该提取器将原始观察映射到与安全相关的概念。因此，学习到的特征提取器在可人理解的语义中（软性）扎根，支持透明的约束评估，并且可以跨任务迁移。通过将可解释性与可微分性结合，NEUPRO超越了不透明的成本设计，朝着可重用的安全推理迈进。为了评估NEUPRO的能力，我们收集并发布了REASON，这是第一个用于可解释机器人安全规范的真实机器人基准数据集。在REASON上的实验表明，NEUPRO学习到的安全关键特征能够跨任务泛化，减轻传统黑箱成本公式的可解释性局限，并提供安全违规的明确解释。
cs.RO / 89 / 2609.39596
MVP-SLAM: Multi-Camera Visual-Inertial Floorplan-Prior SLAM
MVP-SLAM：多摄像头视觉惯性平面先验SLAM
Abstract
Indoor building construction sites are demanding environments for visual SLAM, where variable lighting and repetitive, low-textured structures make the system drift over long trajectories, though structural elements such as walls remain distinguishable despite these conditions. These buildings are constructed according to their as-planned floor plans, available from the design phase, and although the actual as-built site can differ from this design, floor plans still provide a metric reference, both to localize the system in the building and to correct drift. Existing methods often use the floor plan to correct an already-built trajectory offline, and those that instead correct it online typically rely on depth sensors. We instead present MVP-SLAM, an online visual-inertial SLAM on two opposite-facing fisheye cameras that corrects drift from cameras alone by matching walls detected in its map to the floor plan, through a drift-aware policy. A multi-stage integration then turns each matched pair incrementally into a persistent correction, so the trajectory stays corrected and localized within the floor plan as it is built. MVP-SLAM was validated on the multi-floor construction sites of the Hilti-Trimble SLAM Challenge 2026, ranking 2nd of 22 teams in the Localization task (0.29 m mean RMSE) and 5th of 62 teams in the SLAM task (0.24 m), the top-ranked one in both tasks among those that operate online, integrate the floor plan, and localize within it.
Chinese Translation
室内建筑施工现场是视觉SLAM的挑战性环境，变幻的光照和重复的低纹理结构使得系统在长轨迹中产生漂移，尽管在这些条件下，墙壁等结构元素仍然可以被区分。这些建筑是根据其设计阶段的平面图进行建造的，尽管实际的建成现场可能与该设计有所不同，但平面图仍然提供了一个度量参考，既可以用于在建筑内定位系统，也可以用于修正漂移。现有方法通常在离线状态下使用平面图来修正已经构建的轨迹，而那些在线修正轨迹的方法通常依赖于深度传感器。我们提出了MVP-SLAM，这是一种基于两台相对朝向的鱼眼摄像头的在线视觉惯性SLAM，通过将其地图中检测到的墙壁与平面图进行匹配，利用一种漂移感知策略来修正仅由摄像头引起的漂移。多阶段集成将每对匹配的墙壁逐步转化为持久的修正，从而在构建过程中保持轨迹的修正和在平面图内的定位。MVP-SLAM在Hilti-Trimble SLAM Challenge 2026的多层建筑施工现场进行了验证，在定位任务中排名22支队伍中的第2名（平均RMSE为0.29米），在SLAM任务中排名62支队伍中的第5名（0.24米），在所有在线运行、集成平面图并在其中进行定位的队伍中排名最高。
cs.RO / 90 / 2609.39599
Text-to-3D Policy: Fine-Grained Language-Behavior Alignment for Unseen Specification Generalization
文本到三维策略：未见规格泛化的细粒度语言-行为对齐
Abstract
3D visuomotor policies provide a strong foundation for spatially precise manipulation, yet current text-to-3D policies struggle to follow unseen fine-grained behavioral specifications beyond those covered by demonstrations. We study this challenge as unseen specification generalization, where language specifies behaviorally significant variations, such as target position, displacement, or articulated state, that are absent from policy training. We find that pretrained language representations and conventional global behavior-language alignment capture coarse task semantics but often blur nearby specifications that require distinct behaviors. We introduce T3DP, a Text-to-3D Policy framework for fine-grained language-behavior alignment. Rather than compressing each instruction and demonstration into a single global embedding, T3DP preserves their local structures and establishes bidirectional token-level correspondence between linguistic elements and behavioral segments. This directly grounds subtle linguistic variations in the behavior components they affect, preventing closely related specifications from collapsing in the representation space. The resulting specification-sensitive language representation conditions a point-cloud-based 3D diffusion policy, enabling more precise control over unseen behavioral specifications without modifying the underlying policy architecture. Across Meta-World, ManiSkill, and RoboTwin, T3DP improves average held-out-specification success over global language-behavior alignment by +11.0-14.2 points, with gains on all 15 task families; on real-robot tasks, it further raises average success from 47.5% to 65.0% (+17.5 points). Representation and action-probe analyses show that fine-grained alignment better preserves specification geometry and action-relevant variation, linking local behavior grounding to downstream control.
Chinese Translation
三维视觉运动策略为空间精确操作提供了坚实的基础，但当前的文本到三维策略在遵循未见的细粒度行为规格方面存在困难，这些规格超出了演示所涵盖的范围。我们将这一挑战视为未见规格泛化，其中语言指定了行为上重要的变化，例如目标位置、位移或关节状态，这些变化在策略训练中并不存在。我们发现，预训练的语言表示和传统的全局行为-语言对齐能够捕捉粗略的任务语义，但往往模糊了需要不同行为的相邻规格。我们提出了T3DP，一个用于细粒度语言-行为对齐的文本到三维策略框架。T3DP并不是将每个指令和演示压缩成一个全局嵌入，而是保留它们的局部结构，并在语言元素和行为片段之间建立双向的标记级对应关系。这直接将微妙的语言变化与其影响的行为组件相结合，防止相关规格在表示空间中重叠。最终生成的规格敏感语言表示为基于点云的三维扩散策略提供了条件，使得在不修改基础策略架构的情况下，对未见行为规格实现更精确的控制。在Meta-World、ManiSkill和RoboTwin上，T3DP在全局语言-行为对齐的基础上提高了平均保留规格成功率11.0-14.2个百分点，并在所有15个任务家族中均有提升；在真实机器人任务中，平均成功率从47.5%提升至65.0%（增加17.5个百分点）。表示和动作探测分析表明，细粒度对齐更好地保留了规格几何形状和与动作相关的变化，将局部行为基础与下游控制相连接。
cs.RO / 91 / 2609.39611
Towards Agile Vision-Based Multi-UAV Flight: Revisiting State Estimation
面向敏捷视觉基础的多无人机飞行：重新审视状态估计
Abstract
Agile multi-UAV flight requires accurate and low-latency onboard estimation of the kinematic states of neighboring UAVs for collision avoidance, motion coordination, etc. Most vision-based approaches rely on position-only measurements, inferring velocity and acceleration indirectly from displacement. We show that this introduces a fixed structural delay in the estimation of higher-order states, which limits the achievable agility. To address this, we propose to integrate tilt measurements, provided by a state-of-the-art visual detector, which inform about the thrust direction of co-planar multirotor UAVs. We benchmark four position-only and five pose-aware estimators, including a novel formulation of a linear thrust-constraining Kalman filter, on two real-world and one high-fidelity photorealistic simulated dataset over different levels of agility (3-21 m/s^2). In our setup, pose-aware estimation consistently reduces the average velocity and acceleration estimation errors by 40% and 57% across the three datasets with the proposed KF formulation outperforming the other estimators. Position-only filters exhibit a constant ~300 ms delay in acceleration step response independent of agility, whereas the tilt-constrained estimators operate near the physical response limit given by the camera frame-rate by observing the change in thrust direction before the displacement accumulates. In a closed-loop leader-follower simulated experiment with NMPC control, position-only estimation of the leader's state fails to facilitate stable hovering of the follower, while the proposed estimator enables tracking of lateral maneuvers exceeding 2g of acceleration.
Chinese Translation
敏捷的多无人机飞行需要对邻近无人机的运动状态进行准确且低延迟的机载估计，以实现避碰、运动协调等功能。大多数基于视觉的方法依赖于仅有位置的测量，通过位移间接推断速度和加速度。我们表明，这会在高阶状态的估计中引入固定的结构性延迟，从而限制可实现的敏捷性。为了解决这个问题，我们提出整合倾斜测量，由先进的视觉检测器提供，这些测量能够指示共面多旋翼无人机的推力方向。我们在两个真实世界数据集和一个高保真度的光真实模拟数据集上，对四种仅基于位置的估计器和五种考虑姿态的估计器进行了基准测试，包括一种新的线性推力约束卡尔曼滤波器的公式，测试在不同的敏捷性水平（3-21 m/s²）下进行。在我们的设置中，考虑姿态的估计在三个数据集上始终将平均速度和加速度估计误差分别减少了40%和57%，而所提出的卡尔曼滤波器公式的表现优于其他估计器。仅基于位置的滤波器在加速度阶跃响应中表现出约300毫秒的恒定延迟，与敏捷性无关，而倾斜约束的估计器通过在位移累积之前观察推力方向的变化，接近由相机帧率给出的物理响应极限。在一个闭环的领导-跟随模拟实验中，领导者状态的仅基于位置的估计未能促进跟随者的稳定悬停，而所提出的估计器则能够跟踪超过2g加速度的横向机动。
cs.RO / 92 / 2609.39618
TO-mdiSPAs: Topology Optimization of multi-directional Soft Pneumatic Actuators
TO-mdiSPAs：多方向软气动执行器的拓扑优化
Abstract
Soft pneumatic actuators (SPAs) are highly promising and have been extensively explored within the field of soft robotics. Under pneumatic pressure loads, an SPA undergoes bending deformation to perform specific tasks. A multi- or omni-directional SPA can bend and move in any direction within a 3D space, leveraging its multiple degrees of freedom to realize various mechanical functions. This work presents a systematic methodology using topology optimization (TO) to achieve an optimized design for a multi-directional SPAs. To ensure manufacturing robustness, a three-field TO formulation considering blueprint, dilated, and eroded designs is implemented. Additionally, Darcy's law, incorporating a drainage term, is used to model the design-dependent nature of the pneumatic loading. A min-max optimization problem is formulated based on the target output deformations of the SPA unit and solved using the Method of Moving Asymptotes. The optimization yields a high-performing, unconventional geometric design. Finally, numerical simulations demonstrate that the optimized SPA successfully achieves versatile multi-directional movements.
Chinese Translation
软气动执行器（SPAs）在软机器人领域具有很大的潜力，并得到了广泛的研究。在气动压力载荷下，SPA会发生弯曲变形以执行特定任务。多方向或全方向的SPA可以在三维空间内向任意方向弯曲和移动，利用其多个自由度实现各种机械功能。本研究提出了一种系统的方法，通过拓扑优化（TO）实现多方向SPA的优化设计。为了确保制造的稳健性，实施了一种考虑蓝图、膨胀和腐蚀设计的三场TO公式。此外，采用包含排水项的达西定律来建模气动载荷的设计依赖性。基于SPA单元的目标输出变形，构建了一个最小-最大优化问题，并使用移动渐近法进行求解。优化结果产生了一种高性能的非常规几何设计。最后，数值模拟表明，优化后的SPA成功实现了多样化的多方向运动。
cs.RO / 93 / 2609.39637
Prediction is Better than Detection: Traffic Congestion Control using Drones
预测优于检测：利用无人机进行交通拥堵控制
Abstract
A central question in deploying teams of mobile robots for persistent monitoring is how task performance scales with fleet size, and whether this scaling holds once sensing drives downstream action rather than mere observation. We study this question for a team of drones performing traffic-jam detection and prediction in a simulated road network, whose reports drive an adaptive traffic-signal controller in closed loop. We build a multi-agent simulation, with vehicles following Nagel-Schreckenberg cellular-automaton dynamics and drones patrolling junctions via a round-robin policy, and sweep fleet size, traffic level, and network size to evaluate detection rate, detection delay, and prediction rate. We show how performance plateaus for fleet size approximating the number of junctions being monitored, and offer a general fleet-provisioning rule for persistent-monitoring deployments. More significantly, adapting the signal on a predicted jam, rather than a detected one, roughly doubles the resulting reduction in jam duration, showing that the value of onboard prediction in a sensing-to-action pipeline can exceed the value of adding more robots. Prediction accuracy, not sensing coverage, is now the binding constraint on further improvement, pointing to onboard inference, not fleet size, as the more promising direction for future work.
Chinese Translation
在部署移动机器人团队进行持续监测时，一个核心问题是任务性能如何随车队规模的变化而变化，以及这种变化在感知驱动下游行动而不仅仅是观察时是否依然成立。我们研究了这一问题，针对一组在模拟道路网络中进行交通拥堵检测和预测的无人机，其报告驱动一个闭环的自适应交通信号控制器。我们构建了一个多智能体模拟，车辆遵循Nagel-Schreckenberg细胞自动机动态，无人机通过轮询策略巡逻交叉口，并对车队规模、交通水平和网络规模进行调整，以评估检测率、检测延迟和预测率。我们展示了当车队规模接近被监测交叉口数量时，性能会趋于平稳，并为持续监测部署提供了一条通用的车队配置规则。更重要的是，相较于检测到的拥堵，基于预测的信号调整大约将拥堵持续时间的减少量翻倍，表明在感知到行动的流程中，机载预测的价值可以超过增加更多机器人的价值。现在，预测准确性而非感知覆盖率成为进一步改进的限制因素，指向机载推理而非车队规模作为未来研究的更有前景的方向。
cs.RO / 94 / 2609.39652
Making Waves: A Membrane-Coupled Delta Array for Manipulating Objects Below the Actuator Spacing
制造波动：一种膜耦合的Delta阵列用于操控低于执行器间距的物体
Abstract
Distributed manipulator systems manipulate objects through the coordinated motion of many actuators. However, an object must be supported by several actuators at once, so the centre-to-centre actuator spacing imposes a hard lower bound on manipulable object size. We remove this bound by coupling the end-effectors of an 8 x 8 array of three degrees-of-freedom delta robots with a stretchable fabric, turning 64 discrete contacts into a continuous surface capable of manipulating objects smaller than the actuator spacing. Viewing the array as a displacement field over that surface, we investigate local quasi-static and cyclic fields as manipulation primitives. These primitives can be applied globally across the array or locally confined around each tracked object to independently manipulate several objects in parallel. We then train a policy acting on low-order discrete cosine transform coefficients: at equal action dimension, commanding a nineteen-delta neighbourhood halves the placement error of commanding the whole array. The policy transfers to hardware without adaptation at 76% success. The platform manipulates objects from 15mm-90mm, a six-fold range spanning both sides of the actuator spacing, 43.3mm, on a single surface.
Chinese Translation
分布式操控系统通过多个执行器的协调运动来操控物体。然而，物体必须同时由多个执行器支撑，因此中心到中心的执行器间距对可操控物体的大小施加了严格的下限。我们通过将一个8 x 8阵列的三自由度Delta机器人末端执行器与可拉伸的织物耦合，消除了这一限制，将64个离散接触点转变为一个连续表面，能够操控小于执行器间距的物体。将该阵列视为该表面上的位移场，我们研究了局部准静态和循环场作为操控原语。这些原语可以在整个阵列上全局应用，或局部限制在每个跟踪物体周围，以独立并行操控多个物体。然后，我们训练一个作用于低阶离散余弦变换系数的策略：在相同的动作维度下，指挥一个十九个Delta的邻域将指挥整个阵列的放置误差减半。该策略在硬件上无须适配即可转移，成功率为76%。该平台能够在单一表面上操控15mm至90mm的物体，跨越执行器间距43.3mm的两侧，范围达到六倍。
cs.RO / 95 / 2609.39670
From Local Whole-Body VLA Behaviors to Scene-Scale Aerial Manipulation
从局部全身 VLA 行为到场景尺度的空中操作
Abstract
Vision-language-action (VLA) models enable task-conditioned interaction, but extending them to scene-scale aerial manipulation remains challenging due to costly whole-body demonstrations, latency-induced action-state misalignment, and cross-site behavior composition. We present a unified framework for synthetic policy training and scene-scale execution on articulated uncrewed aerial manipulators (UAMs). A scene-reconfigurable pipeline synthesizes task-conditioned, kinodynamically feasible trajectories and synchronized multiview observations for VLA training without physical-platform demonstrations. Measured-progress-aligned realization (MPAR) aligns asynchronously returned action chunks with measured execution progress and realizes them as continuous, dynamically feasible trajectories. A relational Scene Graph grounds language goals to object instances and feasible interaction regions, while topology-guided transfer connects local behaviors across sites. Local VLA skills achieve 39/60 successes (65.0%) in simulation under oracle target and feasible-handoff conditions. Under 500-ms added latency, with and without a transient command-update stall, MPAR reduces median takeover phase error by 0.212 s over nominal-time alignment. The complete system completes 21/50 simulated multi-site missions (42.0%) and is further validated on a physical articulated UAM.
Chinese Translation
视觉-语言-动作 (VLA) 模型能够实现任务条件下的交互，但由于昂贵的全身演示、延迟引起的动作状态不对齐以及跨场地行为组合，将其扩展到场景尺度的空中操作仍然具有挑战性。我们提出了一个统一的框架，用于合成策略训练和在关节式无人机操作器 (UAM) 上的场景尺度执行。一个场景可重构的管道合成任务条件下的运动动力学可行轨迹和同步的多视角观察，用于 VLA 训练，而无需物理平台演示。测量进度对齐实现 (MPAR) 将异步返回的动作块与测量的执行进度对齐，并将其实现为连续的、动态可行的轨迹。关系场景图将语言目标与对象实例和可行的交互区域相结合，而拓扑引导的转移则连接跨场地的局部行为。在模拟中，局部 VLA 技能在理想目标和可行交接条件下实现了 39/60 次成功（65.0%）。在增加 500 毫秒延迟的情况下，无论是否存在瞬态命令更新停滞，MPAR 将中位接管阶段误差减少了 0.212 秒，相较于名义时间对齐。完整系统在模拟的多场地任务中完成了 21/50 次任务（42.0%），并在物理关节式 UAM 上进一步验证。
cs.RO / 96 / 2609.39685
RoboCoach: World Models as Active Coaches for Compositional Robot Skills
RoboCoach：作为组合机器人技能的主动教练的世界模型
Abstract
Long-horizon robot manipulation reuses skills across many task compositions, but improving these compositions with additional end-to-end demonstrations is costly. A practical self-improving system must decide both what to teach next and where to apply that supervision. We present ROBOCOACH, a world-model-guided coaching framework that uses imagined failures to guide demonstration requests and expert updates. Its Route-Imagine-Diagnose-Improve (RIDI) loop executes reusable skill experts inside COACHWORLD, our shared action-conditioned world model, and uses a progress judge to record the first subtask that fails to complete. Aggregated records select which subtask demonstrations to acquire and which expert adapters to update. Across two simulation suites and two real-robot platforms, imagined and deployed success correlate over 22 task-policy pairs (rho = 0.840). Controlled comparisons show that our coaching method outperforms matched baselines under matched data budgets and update schedules. With only 150 additional subtask demonstrations, success rises from 13.3% to 75.0% on Franka and from 40.0% to 83.8% on AgileX. The coached experts also transfer to four held-out compositions, achieving an average success of 35.0%, compared with 0% for a shared-policy baseline updated with uniformly acquired demonstrations. Together, these results show that world models can serve as active coaches, turning imagined failures into targeted supervision for modular policy improvement. Project Page: https://robocoach-ai.github.io/
Chinese Translation
长时间范围的机器人操作在许多任务组合中重用技能，但通过额外的端到端演示来改进这些组合是昂贵的。一个实用的自我改进系统必须决定接下来教什么以及在哪里应用这种监督。我们提出了ROBOCOACH，一个由世界模型指导的教练框架，它利用想象中的失败来指导演示请求和专家更新。其路线-想象-诊断-改进（Route-Imagine-Diagnose-Improve, RIDI）循环在我们的共享动作条件世界模型COACHWORLD内部执行可重用的技能专家，并使用进度评估器记录第一个未能完成的子任务。聚合记录选择哪些子任务演示需要获取，以及哪些专家适配器需要更新。在两个仿真套件和两个真实机器人平台上，想象的成功与实际部署的成功在22个任务策略对中相关（rho = 0.840）。控制比较显示，我们的教练方法在匹配的数据预算和更新计划下优于匹配的基线。仅通过150个额外的子任务演示，Franka的成功率从13.3%上升到75.0%，而AgileX的成功率从40.0%上升到83.8%。经过指导的专家还可以转移到四个保留的组合中，平均成功率为35.0%，而使用均匀获取的演示更新的共享策略基线的成功率为0%。这些结果表明，世界模型可以作为主动教练，将想象中的失败转化为模块化策略改进的针对性监督。项目页面：https://robocoach-ai.github.io/
cs.RO / 97 / 2609.39751
Beyond Policy Alignment: Closing the Planning-Learning Loop for Robot Control with Learned World Models
超越政策对齐：通过学习的世界模型闭合机器人控制的规划-学习循环
Abstract
Planning with learned world models combines online trajectory optimization with learned value and policy functions for high-dimensional control. Because the planner determines the experience used for learning, while the learned critic and actor in turn score and propose future plans, planning and learning form a closed feedback loop. TD-MPC is a prominent instance of this design. Recent policy-constrained variants strengthen one part of the loop by aligning the learned policy with planner behavior. We introduce PL-MPC (Planning-Learning MPC), which additionally modifies critic supervision and planner terminal-value estimation. Hybrid multi-step TD targets expose critic updates to more realized rewards before bootstrapping; disagreement-aware terminal estimates reduce the influence of uncertain critic values during MPPI planning; and return-weighted actor distillation emphasizes planner-executed actions from high-return episodes. The world-model architecture and MPPI optimizer are otherwise unchanged. On HumanoidBench, the largest gains occur on \texttt{balance-hard}, where Total Average Return (TAR) increases from $98\pm18$ to $387\pm255$, and \texttt{hurdle}, from $199\pm13$ to $466\pm200$; performance across the broader benchmark remains task dependent, and PL-MPC remains competitive on DMControl. Controlled ablations show different component interactions across the two tasks. We further demonstrate zero-shot sim-to-real transfer on wrench-nut alignment with a 7-DoF KUKA IIWA14, obtaining higher observed success than TD-M(PC)^2 on the training object size and two unseen sizes. Code and data will be available at: https://pl-mpc-humanoid.github.io.
Chinese Translation
使用学习的世界模型进行规划结合了在线轨迹优化与学习的价值和政策函数，以实现高维控制。由于规划者决定了用于学习的经验，而学习到的评论员和执行者又对未来计划进行评分和提出建议，因此规划与学习形成了一个闭环反馈循环。TD-MPC是这种设计的一个显著实例。最近的政策约束变体通过将学习的政策与规划者行为对齐来增强循环的一个部分。我们引入了PL-MPC（规划-学习模型预测控制），它还修改了评论员监督和规划者终端价值估计。混合多步TD目标使评论员更新在引导之前接触到更多实现的奖励；关注分歧的终端估计在MPPI规划过程中减少了不确定评论员值的影响；而回报加权的执行者蒸馏则强调来自高回报情节的规划者执行的动作。世界模型架构和MPPI优化器保持不变。在HumanoidBench上，最大的增益发生在 exttt{balance-hard}上，整体平均回报（TAR）从$98 extpm18$增加到$387 extpm255$，在 exttt{hurdle}上，从$199 extpm13$增加到$466 extpm200$；在更广泛的基准测试中，性能依然依赖于任务，PL-MPC在DMControl上仍具竞争力。控制消融实验显示了两个任务之间不同组件的相互作用。我们进一步展示了在7自由度KUKA IIWA14上进行扳手-螺母对齐的零-shot仿真到现实转移，在训练对象尺寸和两个未见尺寸上获得了比TD-M(PC)^2更高的观察成功率。代码和数据将可在以下网址获取：https://pl-mpc-humanoid.github.io。
cs.RO / 98 / 2609.39754
ChunkTrust: Adapting Execution Horizons for Robot Policies with Action-Expert Evidence
ChunkTrust：基于行动专家证据调整机器人策略的执行视野
Abstract
Robot foundation policies predict action chunks, but how many actions to execute before replanning depends on the current task phase. We introduce ChunkTrust, which treats the execution horizon as a latent variable inferred from action-expert evidence rather than a fixed hyperparameter. Its training-free Action-aware Horizon Selector (AHS) combines intra-chunk spectral stability of generation traces with inter-chunk continuity between executed history and predicted actions. An online Beta posterior with kernel forgetting tracks horizon preferences across replans. A lightweight Query-based Horizon Adapter (QHA) optionally learns a context-conditioned dense prior from complementary evidence, fused with current evidence and episode-local Beta memory while the base policy remains frozen. Across RoboTwin2.0 and RoboCasa GR1 Tabletop, AHS improves overall task-averaged success for each evaluated base-policy configuration, including gains of +6.80 percentage points on $\pi_{0.5}$ over all 50 RoboTwin2.0 tasks and +9.67 percentage points on Qwen3GR00T in RoboCasa. AHS+QHA raises the gain over Base to +9.44 percentage points on the eight-task $\pi_{0.5}$ evaluation. On four real-world household tasks, AHS improves the equal-task mean normalized process score from 50.4% to 57.5%. Ablations examine the contributions of both evidence terms, temporal memory, and the learned prior. Project page is https://hf618.github.io/ChunkTrust.github.io/
Chinese Translation
机器人基础策略预测行动块，但在重新规划之前需要执行多少个行动取决于当前任务阶段。我们提出了ChunkTrust，它将执行视野视为一个从行动专家证据推断的潜在变量，而不是一个固定的超参数。其无训练的行动感知视野选择器（Action-aware Horizon Selector, AHS）结合了生成轨迹的块内谱稳定性与执行历史和预测行动之间的块间连续性。一个带有核遗忘的在线Beta后验跟踪在重新规划中的视野偏好。一个轻量级的基于查询的视野适配器（Query-based Horizon Adapter, QHA）可选择性地从补充证据中学习上下文条件的密集先验，并与当前证据和局部Beta记忆融合，同时基础策略保持不变。在RoboTwin2.0和RoboCasa GR1桌面上，AHS提高了每个评估的基础策略配置的整体任务平均成功率，包括在所有50个RoboTwin2.0任务中对$ ext{π}_{0.5}$的提升为+6.80个百分点，以及在RoboCasa的Qwen3GR00T上提升为+9.67个百分点。AHS+QHA在八任务的$ ext{π}_{0.5}$评估中将增益提高至+9.44个百分点。在四个真实家庭任务中，AHS将相等任务的平均标准化过程得分从50.4%提高到57.5%。消融实验考察了两个证据项、时间记忆和学习先验的贡献。项目页面为https://hf618.github.io/ChunkTrust.github.io/
cs.RO / 99 / 2609.39755
Experience-Driven Continual Learning of Terrain Traversability for Quadruped Robots
基于经验的四足机器人地形可 traversability 的持续学习
Abstract
Safe and efficient quadruped navigation over unfamiliar terrain requires predicting terrain-robot interaction before contact: geometry and visual appearance alone cannot reveal how the robot will slip, load its feet, or expend energy. This paper presents a continual learning pipeline that uses locomotion experience to learn these interaction outcomes from pre-contact images and continually updates the predictions as new contacts are observed. Pre-contact descriptors, produced by a DINOv3 backbone model frozen during training, are mapped to five proprioceptive indicators weighted according to measurement reliability: planar foot slip, mean normal ground-reaction force, traction index, cost of transport, and touchdown loading rate. A compact evidential regressor allows us to predict these indicators together with aleatoric and epistemic uncertainty from the visual descriptors. Continual adaptation combines bounded experience replay with a validation gate: candidate models replace the deployed predictor only when they improve performance on recent held-out data while keeping degradation on historical held-out data within a prescribed tolerance. Predictions and epistemic uncertainty are projected into a local multilayer map and combined into a conservative traversability score map whose property weights can be adjusted without retraining. The resulting map is used for downstream navigation tests. The ROS2 implementation supports evaluation on a Unitree Go2 in simulation and on hardware, with models trained separately in each domain. On a sequential hardware stream over three previously unseen terrains, gated replay reduces final anchor negative log-likelihood (NLL) degradation by 23.1% relative to replay without the gate while attaining similar new-terrain adaptation.
Chinese Translation
安全高效的四足机器人在陌生地形上的导航需要在接触之前预测地形与机器人之间的交互：仅凭几何形状和视觉外观无法揭示机器人如何滑动、承载其足部或消耗能量。本文提出了一种持续学习管道，利用运动经验从接触前图像中学习这些交互结果，并在观察到新接触时不断更新预测。接触前描述符由在训练期间被冻结的 DINOv3 主干模型生成，并映射到五个根据测量可靠性加权的本体感知指标：平面足部滑移、平均法向地面反作用力、牵引指数、运输成本和着陆负载率。一个紧凑的证据回归模型使我们能够从视觉描述符中预测这些指标及其随机和认知不确定性。持续适应结合了有界经验重放和验证门：候选模型仅在其在最近的保留数据上提高性能时替换已部署的预测器，同时保持在历史保留数据上的降级在规定的容忍范围内。预测和认知不确定性被投影到局部多层地图中，并结合成一个保守的可 traversability 评分地图，其属性权重可以在不重新训练的情况下进行调整。生成的地图用于下游导航测试。ROS2 实现支持在模拟和硬件上对 Unitree Go2 的评估，模型在每个领域分别训练。在对三个先前未见地形的顺序硬件流中，带门重放将最终锚点负对数似然（NLL）降级减少了 23.1%，相对于没有门的重放，同时实现了类似的新地形适应。
cs.RO / 100 / 2609.39763
DiffWAM: A Fast and Efficient Navigation World Action Model
DiffWAM：一种快速高效的导航世界动作模型
Abstract
Pretrained video foundation models encode rich semantic and spatiotemporal priors for embodied navigation, yet converting these priors into UAV motion typically requires expensive future-video synthesis and geometric reconstruction. We investigate whether the motion implicit in future visual prediction can instead be recovered directly from the predictive representations of a frozen video model. To this end, we present DiffWAM, a geometry-conditioned navigation world-action model that directly transforms multi-level predictive features into continuous camera trajectories. Its Grid-Motion module preserves spatial-temporal motion associations, while Latent2Pose grounds them with first-frame geometry to recover metrically meaningful 3D motion. Complete video rollouts and geometric reconstruction are required only for offline supervision, eliminating future-video decoding and multi-frame reconstruction during deployment. We further introduce FastDreamer, which overlaps predictive and geometric computation with ongoing flight and performs timestamp-aware asynchronous trajectory handoff for continuous UAV execution. DiffWAM achieves a trajectory RMSE of 0.3492 m and an endpoint success rate of 74.40% on the 1,000-sample DiffWAM-1000 benchmark, while representative real-world experiments demonstrate complex behaviors including constrained traversal, orbiting, S-shaped flight, and multi-stage navigation. An onboard DiffWAM-Flash implementation further reaches 1.08 s model-pipeline latency on NVIDIA Jetson AGX Thor. These results demonstrate that predictive video representations can be efficiently grounded into continuous 3D motion, providing a direct alternative to generate-then-reconstruct navigation pipelines. Project page: https://zzmmzzm.github.io/diffwam.github.io/.
Chinese Translation
预训练的视频基础模型编码了丰富的语义和时空先验，用于具身导航，但将这些先验转换为无人机（UAV）运动通常需要昂贵的未来视频合成和几何重建。我们研究了是否可以直接从冻结视频模型的预测表示中恢复未来视觉预测中隐含的运动。为此，我们提出了DiffWAM，这是一种几何条件的导航世界动作模型，能够直接将多层预测特征转换为连续的相机轨迹。其Grid-Motion模块保留了时空运动关联，而Latent2Pose则通过第一帧几何信息将其固定，以恢复具有度量意义的三维运动。完整的视频回放和几何重建仅在离线监督中需要，从而消除了在部署过程中对未来视频解码和多帧重建的需求。我们进一步引入了FastDreamer，它在持续飞行中重叠预测和几何计算，并执行时间戳感知的异步轨迹交接，以实现连续的无人机执行。DiffWAM在1,000个样本的DiffWAM-1000基准测试中达到了0.3492米的轨迹均方根误差（RMSE）和74.40%的终点成功率，而代表性的现实世界实验展示了包括受限穿越、环绕飞行、S形飞行和多阶段导航在内的复杂行为。基于DiffWAM-Flash的机载实现进一步在NVIDIA Jetson AGX Thor上达到了1.08秒的模型管道延迟。这些结果表明，预测视频表示可以有效地转化为连续的三维运动，提供了一种生成-再重建导航管道的直接替代方案。项目页面：https://zzmmzzm.github.io/diffwam.github.io/
cs.RO / 101 / 2609.39797
Tool-Policy Co-Design for Powder Weighing in Laboratory Automation
实验室自动化中粉末称量的工具-策略协同设计
Abstract
Autonomous powder weighing is one of many bottlenecks in laboratory automation due to the complex, non-linear dynamics of heterogeneous materials. Robot chemists performing this task utilise standard tools shaped for the dexterity of human hands, whose fixed geometry sets the dynamics that the control policy needs to regulate. This work introduces a tool-policy co-design framework that concurrently optimises the morphology of a dispensing tool and its control policy for use by robots in chemistry laboratories, formulated as a bi-level optimisation that minimises dispensing error over a target distribution of powder flowabilities. The outer loop varies tool-design parameters such as tool depth, width and rim spike topology using Bayesian optimisation and hyperband, while an inner loop optimises a control policy for each candidate morphology. We also introduce a geometric similarity metric that warm-starts policy training from cached policies of structurally similar designs, exploring 28% more configurations under the same compute budget. The proposed framework is evaluated on a robotic powder weighing task across seven materials with distinct physical dynamics in a flowability-informed robot-material simulation framework. Experimental results demonstrate that our co-designed tool morphology reduces real-world weighing errors by 45% relative to a standard tool, including on previously unseen materials. These results demonstrate our method can adapt both the control policy and the physical tool to the dynamics of the target material, bringing a new paradigm for material manipulation to the field of laboratory automation.
Chinese Translation
自主粉末称量是实验室自动化中的众多瓶颈之一，原因在于异质材料的复杂非线性动态。执行此任务的机器人化学家使用为人手灵巧性设计的标准工具，其固定的几何形状决定了控制策略需要调节的动态特性。本研究提出了一种工具-策略协同设计框架，旨在同时优化分配工具的形态及其控制策略，以供化学实验室中的机器人使用，形成一个双层优化问题，旨在最小化针对粉末流动性目标分布的分配误差。外层循环使用贝叶斯优化和超带（hyperband）方法调整工具设计参数，如工具深度、宽度和边缘尖峰拓扑，而内层循环则为每个候选形态优化控制策略。我们还引入了一种几何相似性度量，通过从结构相似设计的缓存策略中热启动策略训练，在相同计算预算下探索了28%的更多配置。所提出的框架在一个流动性信息驱动的机器人-材料仿真框架中，对七种具有不同物理动态的材料的机器人粉末称量任务进行了评估。实验结果表明，我们协同设计的工具形态相较于标准工具减少了45%的实际称量误差，包括对之前未见材料的称量。这些结果表明我们的方法能够同时适应目标材料的动态特性，调整控制策略和物理工具，为实验室自动化领域的材料操作带来了新的范式。
cs.RO / 102 / 2609.39820
Learning from Runtime Feedback through Failure-Bank Self-Evolution for Vision-Language-Action Models
通过失败银行自我演化从运行时反馈中学习的视觉-语言-动作模型
Abstract
Vision-language-action (VLA) models generalize broadly across robotic manipulation tasks, but complex environments require balancing task success with unintended contact. Runtime shields can correct individual actions, but they leave the underlying policy unchanged, so repeated disagreements may create a persistent policy-shield mismatch that blocks task progress. To address this challenge, we introduce FailBank, a four-stage self-evolving framework that converts runtime feedback into persistent policy improvement. During collection, a fixed CBF-based safety module serves as an observe-only teacher, producing counterfactual corrections while the policy remains in control. Outcome-aware admission then converts useful proposals into corrective targets and retains successful uncorrected actions as quiet anchors for guarded LoRA updates. We evaluate FailBank on the VLA-Arena benchmark across two difficulty levels and two VLA backbones. Compared with the base policies, FailBank improves the joint success-cost operating point. Across the two backbones, FailBank improves task success rate by 8.5 and 6.9 percentage points, while reducing policy-induced cumulative cost by 35.6\% and 23.8\%, respectively. Compared with runtime shielding, FailBank raises task success rate by 25.4 and 9.5 percentage points, while maintaining comparable policy-induced cumulative cost. These results show that runtime feedback can serve as persistent policy supervision rather than only as a temporary action constraint.
Chinese Translation
视觉-语言-动作（VLA）模型在机器人操作任务中具有广泛的泛化能力，但复杂环境需要在任务成功与意外接触之间取得平衡。运行时保护机制可以纠正个别动作，但它们不会改变基础策略，因此重复的分歧可能导致持久的策略-保护机制不匹配，从而阻碍任务进展。为了解决这一挑战，我们引入了FailBank，一个四阶段自我演化框架，将运行时反馈转化为持久的策略改进。在收集阶段，一个基于CBF（控制屏障函数）的固定安全模块作为观察型教师，产生反事实修正，而策略仍然保持控制。结果感知的接纳阶段将有用的提议转化为纠正目标，并将成功的未修正动作保留为受保护的LoRA（低秩适应）更新的静默锚点。我们在VLA-Arena基准上对FailBank进行了评估，涵盖两个难度级别和两个VLA骨干网络。与基础策略相比，FailBank改善了联合成功-成本操作点。在两个骨干网络中，FailBank分别提高了任务成功率8.5和6.9个百分点，同时将策略引起的累积成本分别降低了35.6\%和23.8\\%。与运行时保护机制相比，FailBank将任务成功率提高了25.4和9.5个百分点，同时保持了可比的策略引起的累积成本。这些结果表明，运行时反馈可以作为持久的策略监督，而不仅仅是临时的动作约束。
cs.RO / 103 / 2609.39822
Toward Real-Time VLAs: Stage-Aware Two-Step Flow Denoising and System-Level Evaluation
朝向实时视觉-语言-动作（VLA）：阶段感知的两步流去噪与系统级评估
Abstract
Vision-language-action (VLA) models face a timing gap between low-rate inference and high-rate robot execution. We characterize this gap through end-to-end latency measurements of model inference and the robot execution chain. Repeated Flow Matching denoising contributes substantially to inference cost, while robot-side delays mainly arise from perception acquisition, communication scheduling, and physical response. Analysis of the velocity field shows relatively stable magnitude and direction in early integration, followed by stronger directional correction near the terminal steps. Based on this stage heterogeneity, we propose two-stage non-uniform denoising, reducing the number of steps from 10 to 2 and model-inference time from 61.557 ms to 21.956 ms. We also develop a distributed real-time VLA framework with independent inference, action-publication, and robot-control rates, modular observation acquisition, and action-provenance logging. Using {\pi}0.5 as the baseline, we evaluate six real-time execution methods on a long-horizon physical garment-folding task. Legato performs best overall among training-based methods, while Temporal Smoothing leads among training-free methods; both perform strongly in task success, completion time, action continuity, and acceleration smoothness. Combining two-step denoising with representative execution methods substantially reduces inference cost with a small reduction in task performance. These results motivate joint optimization of model-inference efficiency and robot-system timing.
Chinese Translation
视觉-语言-动作（VLA）模型面临低速推理与高速机器人执行之间的时间差距。我们通过对模型推理和机器人执行链的端到端延迟测量来表征这一差距。重复流匹配去噪在推理成本中占据了重要部分，而机器人端的延迟主要来源于感知获取、通信调度和物理响应。速度场的分析显示，在早期整合阶段，速度的幅度和方向相对稳定，而在终端步骤附近则出现更强的方向修正。基于这种阶段异质性，我们提出了两阶段非均匀去噪，将步骤数量从10减少到2，将模型推理时间从61.557毫秒减少到21.956毫秒。我们还开发了一个分布式实时VLA框架，具有独立的推理、动作发布和机器人控制速率，模块化的观测获取和动作来源记录。以{ extpi}0.5作为基线，我们在一个长时间范围的物理衣物折叠任务上评估了六种实时执行方法。在基于训练的方法中，Legato的整体表现最佳，而在无训练的方法中，Temporal Smoothing表现领先；两者在任务成功率、完成时间、动作连续性和加速度平滑性方面均表现出色。将两步去噪与代表性执行方法相结合，显著降低了推理成本，同时对任务性能的影响较小。这些结果激励了模型推理效率与机器人系统时序的联合优化。
cs.RO / 104 / 2609.39854
RL-Guided PAC-NMPC for Probabilistically-Safe Perception-Based Navigation in Unknown Environments
基于强化学习的概率安全感知导航控制方法在未知环境中的应用
Abstract
In this paper, we present an approach for combining stochastic nonlinear model predictive control (SNMPC) and reinforcement learning (RL) to enable probabilistically-safe perception-based navigation in unknown environments. Our method first uses RL to train probabilistic actor-critic and sensor prediction models. We then leverage these probabilistic models in a sampling-based SNMPC framework known as Probably Approximately Correct (PAC)-NMPC, which uses hard constraints to enforce finite-time statistical guarantees on the probability of collision and value function improvement. By ensuring that our finite-horizon SNMPC policies decrease the value function in expectation, we can approach the long-horizon performance of the RL approach while satisfying probabilistic safety constraints. Through simulation experiments, we show that our approach can improve the safety of perception-based RL navigation policies and scale to high dimensional systems with large sensor input spaces and complex nonlinear dynamics. We also demonstrate our approach through hardware experiments, showing improved performance for vision-based navigation with an agile fixed-wing aerial vehicle in unknown environments.
Chinese Translation
在本文中，我们提出了一种将随机非线性模型预测控制（SNMPC）与强化学习（RL）相结合的方法，以实现未知环境中的概率安全感知导航。我们的方法首先使用强化学习训练概率性演员-评论家模型和传感器预测模型。然后，我们在一种基于采样的SNMPC框架中利用这些概率模型，该框架被称为概率近似正确（PAC）-NMPC，使用硬约束来强制实施碰撞概率和价值函数改进的有限时间统计保证。通过确保我们的有限时间SNMPC策略在期望上降低价值函数，我们可以在满足概率安全约束的同时接近RL方法的长时间性能。通过仿真实验，我们展示了我们的方法可以提高基于感知的RL导航策略的安全性，并能够扩展到具有大传感器输入空间和复杂非线性动态的高维系统。我们还通过硬件实验展示了我们的方法，显示出在未知环境中，基于视觉的导航在灵活固定翼无人机上的性能得到了改善。
cs.RO / 105 / 2609.39870
Magic-W0: A Structured World-Action Foundation Model for Physical Intelligence
Magic-W0：一种用于物理智能的结构化世界-动作基础模型
Abstract
World-action models (WAMs) augment robot policies with action-conditioned environment dynamics, yet existing approaches largely rely on future observation reconstruction or generic latent prediction and lack structured, control-oriented world representations tightly coupled with action generation. We introduce Magic-W0, a world-action foundation model that jointly models structured physical state evolution and continuous actions. Magic-W0 represents interaction as a Structured World Transition consisting of Current State, Transition, and Future State. Current State combines vision-language context with Current 3D Geometry; Transition is represented by 3D Motion capturing action-induced three-dimensional changes; and Future State is represented by Future Semantics describing task-relevant outcomes. To couple prediction and control, we propose a layer-aligned world-action interaction architecture in which evolving action hypotheses condition world-transition prediction, while predicted world representations continuously inform action generation. Magic-W0 is pre-trained on large-scale egocentric human manipulation, UMI, real-robot, and simulation data, with latent supervision for geometry, 3D motion, and future semantics from pre-trained visual models. Inference-time interventions show that structured world representations respond systematically to changes in candidate actions and that action-related information propagates through shared 3D representations into future semantic predictions. On RoboDojo-Sim, Magic-W0 achieves an average Score of 27.10, the highest among the compared WAMs. Across multiple real-robot tasks, it also demonstrates strong downstream performance after fine-tuning with limited downstream data, supporting generalization and rapid adaptation.
Chinese Translation
世界-动作模型（WAMs）通过动作条件的环境动态增强机器人策略，但现有方法在很大程度上依赖于未来观察重构或通用潜在预测，缺乏与动作生成紧密结合的结构化、控制导向的世界表示。我们提出了Magic-W0，一种联合建模结构化物理状态演变和连续动作的世界-动作基础模型。Magic-W0将交互表示为一个结构化世界转移，包括当前状态、转移和未来状态。当前状态结合了视觉-语言上下文与当前3D几何；转移通过3D运动捕捉动作引起的三维变化来表示；未来状态通过描述任务相关结果的未来语义来表示。为了将预测与控制结合起来，我们提出了一种层对齐的世界-动作交互架构，其中演变的动作假设条件化世界转移预测，而预测的世界表示则持续为动作生成提供信息。Magic-W0在大规模自我中心人类操作、UMI、真实机器人和仿真数据上进行了预训练，并从预训练视觉模型中获得几何、3D运动和未来语义的潜在监督。在推理时的干预显示，结构化世界表示对候选动作的变化作出系统响应，并且与动作相关的信息通过共享的3D表示传播到未来语义预测中。在RoboDojo-Sim上，Magic-W0的平均得分为27.10，是比较的WAMs中最高的。在多个真实机器人任务中，它在有限的下游数据微调后也展示了强大的下游性能，支持泛化和快速适应。
cs.RO / 106 / 2609.39873
SplineWAM: Adaptive Action Horizons for World Action Models via B-Spline Representations
SplineWAM：通过B样条表示的世界行动模型自适应行动视野
Abstract
World action models (WAMs) are large embodied policies that jointly predict future video and the actions to execute, emitting a fixed-length action chunk per inference call. Such a policy allocates its computational budget uniformly in time, unable to execute for longer over free-space motion or to spend more inference on contact-rich manipulation, which limits the throughput a WAM can reach when served in the cloud. We present SplineWAM, which adaptively compresses the action trajectory into a fixed-size window of cubic B-spline parameters, fitting the knot times to the characteristics of the motion. One parameter budget then decodes into chunks of varying temporal resolution and duration, and both the executed span and the interval until the next policy call follow from the prediction itself. Aligning the video supervision to the fitted knot times of the demonstration rather than to a uniform grid concentrates the supervised frames where the action trajectory is complex. For asynchronous deployment we introduce Jacobian-Pullback Real-Time Chunking (JP-RTC), which imposes chunk continuity on the decoded raw actions the robot executes rather than on the spline parameters, and corrects the parameters through the decoder so that the executed prefix agrees with the actions already committed. On LIBERO-Plus and RoboCasa, SplineWAM improves success rate over an action chunking WAM by $8.2$ and $4.4$ points while cutting policy calls per episode by 22% and 26%. On three bimanual real-robot tasks under asynchronous execution, it leads or matches the baseline while decoding 1.2 to 1.6 times as much executed motion per call.
Chinese Translation
世界行动模型（WAMs）是大型具身策略，能够共同预测未来视频及需执行的动作，每次推理调用发出固定长度的动作块。这种策略在时间上均匀分配其计算预算，无法在自由空间运动中执行更长时间，也无法在接触丰富的操作中花费更多推理，这限制了WAM在云端服务时的吞吐量。我们提出了SplineWAM，它自适应地将动作轨迹压缩为固定大小的三次B样条参数窗口，将结点时间调整为运动特征。一个参数预算随后解码为具有不同时间分辨率和持续时间的块，执行的跨度和下一个策略调用之间的间隔均由预测本身决定。将视频监督与演示的拟合结点时间对齐，而不是与均匀网格对齐，使得在动作轨迹复杂的地方集中监督帧。为了实现异步部署，我们引入了雅可比拉回实时分块（Jacobian-Pullback Real-Time Chunking, JP-RTC），它在机器人执行的解码原始动作上施加块连续性，而不是在样条参数上，并通过解码器修正参数，以确保执行的前缀与已承诺的动作一致。在LIBERO-Plus和RoboCasa上，SplineWAM将成功率提高了8.2和4.4个百分点，同时每集的策略调用减少了22%和26%。在三个双手真实机器人任务的异步执行中，它的解码执行动作量是基线的1.2到1.6倍，且表现优于或与基线持平。
cs.RO / 107 / 2609.39898
Active Mapping of Underwater Litter Using Camera-Sonar Fusion
基于相机-声纳融合的水下垃圾主动映射
Abstract
Marine litter is a growing threat to the underwater ecosystem, driving demand for autonomous survey methods that can locate debris efficiently over large areas. Existing survey methods typically follow predefined paths or operate with a single sensing modality, typically a camera (with image quality suffering in poor-visibility conditions) or sonar (usually noisy and low-resolution). We present an active mapping framework in which a forward-looking sonar and a camera both feed into a shared Bayesian occupancy map, and an optimization problem is solved at each step to decide on the next best view. Candidate viewpoints are scored by a two-term utility that balances exploration of uncertain regions via voxel entropy against exploitation of likely objects. Each sensor is characterized by range- and bearing-dependent detection and false-alarm probability tables determined from data. We evaluate the approach in a realistic underwater simulator, demonstrating that active mapping finds objects faster than a lawnmower coverage pattern, and that the dual-sensor approach works better than using either of the individual sensors.
Chinese Translation
海洋垃圾对水下生态系统构成日益严重的威胁，迫切需要能够高效定位大面积碎片的自主调查方法。现有的调查方法通常遵循预定义路径或仅使用单一传感方式，通常是相机（在能见度差的条件下图像质量下降）或声纳（通常噪声大且分辨率低）。我们提出了一种主动映射框架，其中前视声纳和相机共同输入到一个共享的贝叶斯占用图中，并在每一步解决一个优化问题以决定下一个最佳视角。候选视点通过一个双项效用进行评分，该效用在通过体素熵探索不确定区域与利用可能对象之间进行平衡。每个传感器的检测和误报概率表是基于数据确定的，依赖于距离和方位。我们在一个真实的水下模拟器中评估了该方法，结果表明，主动映射比割草机覆盖模式更快地找到物体，并且双传感器方法的效果优于单独使用任一传感器。
cs.RO / 108 / 2609.39950
Magnetic based In-situ Self 3D Pose Estimation for a Modular Soft Tendon-Driven Continuum Robot via IMU-Fusion
基于磁场的原位自我三维姿态估计用于模块化软腱驱动连续机器人通过IMU融合
Abstract
Continuum robots are well suited for gentle manipulation because of their inherent compliance and ability to adapt to complex environments. However, their continuously deformable structure makes accurate configuration estimation challenging, particularly when external vision systems are unavailable or obstructed. In this work, we present an embedded pose sensing framework that combines inertial measurement units (IMUs) and active magnetic fields to estimate the robot configuration without relying on external cameras. The angular measurements from the IMU and magnetic-field references are fused to improve local orientation estimation and reduce accumulated orientation error during operation. This pose sensing scheme achieves an update rate of 16.7~Hz, allowing real-time feedback. The proposed system is experimentally validated through closed-loop control, where the estimated robot configuration is used to maintain the end-effector at a desired position while interacting with an object. These results demonstrate the potential of distributed magnetic--inertial sensing for real-time pose estimation and closed-loop control of continuum robots.
Chinese Translation
连续机器人因其固有的柔顺性和适应复杂环境的能力，非常适合进行温和的操作。然而，它们的连续可变形结构使得准确的配置估计变得具有挑战性，特别是在外部视觉系统不可用或受到阻碍时。在本研究中，我们提出了一种嵌入式姿态传感框架，该框架结合了惯性测量单元（IMU）和主动磁场，以在不依赖外部摄像机的情况下估计机器人配置。IMU的角度测量与磁场参考数据融合，以改善局部方向估计并减少操作过程中累积的方向误差。该姿态传感方案实现了16.7 Hz的更新频率，允许实时反馈。通过闭环控制对所提系统进行了实验验证，其中估计的机器人配置用于在与物体交互时保持末端执行器在期望位置。这些结果展示了分布式磁惯性传感在连续机器人实时姿态估计和闭环控制中的潜力。
cs.RO / 109 / 2609.39969
TACTIC: Temporal and Context-Aware LLM Tactical Planning for Roadside LiDAR Attacks
TACTIC：针对路边LiDAR攻击的时间和上下文感知的LLM战术规划
Abstract
Physical LiDAR attacks are often evaluated using fixed primitives and manually selected parameters, despite their strong dependence on surrounding traffic. We present TACTIC, a scene-aware framework that uses a multimodal large language model (MLLM) to coordinate state-adaptive roadside LiDAR attacks. Under a gray-box threat model, TACTIC relies only on an attacker-operated roadside perception stack, without accessing the victim LiDAR's native point clouds or internal processing. Local perception provides metric vehicle states, while the MLLM combines these measurements with roadside imagery to infer relational traffic context and construct a semantic scene graph. Based on this representation, TACTIC selects and configures two complementary primitives: \emph{push-away}, which shifts the perceived range of a lead vehicle, and \emph{phantom-obstacle braking}, which triggers emergency braking through obstacle injection. Measured traffic states and empirically calibrated constraints ground the generated tactics in physically feasible operating regions. To accommodate MLLM latency, TACTIC overlaps reasoning and execution asynchronously while high-rate local perception detects scene changes and triggers replanning. Across 280 randomized CARLA trials, the full policy achieves a 100% collision rate, compared with 35% for a fixed rule, 60% for random selection, and 75% for a restricted LLM using mode selection with default parameters. Joint physical-and-image input achieves 100% success, versus 65% with physical measurements alone and 75% with imagery alone, while asynchronous $\Delta$ refresh reduces scene-mutation response from 7.4 s to 2.0 s. These results show that scene-dependent tactical planning can expose context-sensitive LiDAR failure modes that fixed attack policies may miss.
Chinese Translation
物理LiDAR攻击通常使用固定的原语和手动选择的参数进行评估，尽管它们强烈依赖于周围的交通情况。我们提出了TACTIC，一个场景感知框架，利用多模态大型语言模型（MLLM）来协调状态自适应的路边LiDAR攻击。在灰盒威胁模型下，TACTIC仅依赖攻击者操作的路边感知堆栈，而不访问受害者LiDAR的原始点云或内部处理。局部感知提供了度量车辆状态，而MLLM将这些测量与路边图像结合，以推断关系交通上下文并构建语义场景图。基于这一表示，TACTIC选择并配置两种互补的原语： extit{push-away}，它改变前车的感知范围，以及 extit{phantom-obstacle braking}，它通过障碍物注入触发紧急制动。测量的交通状态和经验校准的约束将生成的战术固定在物理可行的操作区域。为了适应MLLM的延迟，TACTIC在高频局部感知检测场景变化并触发重新规划的同时，异步重叠推理和执行。在280次随机化的CARLA试验中，完整策略实现了100%的碰撞率，而固定规则为35%，随机选择为60%，使用默认参数的受限LLM进行模式选择为75%。联合物理和图像输入实现了100%的成功率，而仅使用物理测量为65%，仅使用图像为75%，同时异步$ extDelta$刷新将场景突变响应时间从7.4秒减少到2.0秒。这些结果表明，依赖场景的战术规划可以揭示固定攻击策略可能遗漏的上下文敏感的LiDAR故障模式。
cs.RO / 110 / 2609.39970
PhasePlan: Ordered Future-Phase Planning for Robot Brain Models
PhasePlan：机器人脑模型的有序未来阶段规划
Abstract
Robot brain models integrate vision, language, and robot state to generate actions for complex manipulation tasks. Most predict fixed-length action chunks that may span multiple task phases. This can obscure phase transitions and favor frequent action patterns, compromising action timing in dynamic environments. We propose \method, an ordered future-phase planning method for robot brain models. From current multimodal observations, it predicts the task phase at each future action position. The resulting planning representations condition the corresponding actions, preserving temporal alignment between task progress and action generation. Training first learns the planner, then freezes it during action-model adaptation to maintain stable phase representations. We instantiate \method on pretrained $\pi_{0.5}$ and AcrossWAM1.0 robot brain models. Detailed quantitative evaluation uses the $\pi_{0.5}$ implementation. On conveyor-belt manipulation, \method reduces offline joint-action error by approximately 22.5\% relative to the original $\pi_{0.5}$ model. It also improves phase-transition modeling and cross-phase action prediction. These results demonstrate the value of ordered future-phase planning for continuous action generation.
Chinese Translation
机器人脑模型整合了视觉、语言和机器人状态，以生成复杂操作任务的动作。大多数模型预测固定长度的动作块，这些动作块可能跨越多个任务阶段。这可能会模糊阶段转换，并偏向于频繁的动作模式，从而影响动态环境中的动作时机。我们提出了 extit{PhasePlan}（有序未来阶段规划）方法，适用于机器人脑模型。该方法基于当前的多模态观察，预测每个未来动作位置的任务阶段。所生成的规划表示条件化了相应的动作，保持了任务进展与动作生成之间的时间对齐。训练首先学习规划器，然后在动作模型适应期间将其冻结，以维持稳定的阶段表示。我们在预训练的$ extit{π}_{0.5}$和AcrossWAM1.0机器人脑模型上实现了 extit{PhasePlan}。详细的定量评估使用$ extit{π}_{0.5}$实现。在传送带操作中， extit{PhasePlan}相对于原始$ extit{π}_{0.5}$模型将离线联合动作误差减少了约22.5\%。它还改善了阶段转换建模和跨阶段动作预测。这些结果展示了有序未来阶段规划在连续动作生成中的价值。
cs.RO / 111 / 2609.39971
When Instructions Retrieve Trajectories: Diagnosing and Mitigating Generalization Failures in VLA Models
当指令检索轨迹：诊断和缓解 VLA 模型中的泛化失败
Abstract
Vision-language-action (VLA) models can exceed 90% success on in-distribution tasks and withstand nuisance changes that preserve the required action, yet fail under counterfactual changes that demand a different action. Aggregate robustness scores can therefore conceal a more specific failure, in which a policy responds to both language and vision yet does not combine them to select the action the task requires. We call this failure instruction-action binding. Instructions cue familiar trajectory families, and visual feedback adjusts their execution. Behavioral analyses of fine-tuned $\pi_{0.5}$ and GR00T-N1.7 policies reveal that failed rollouts often retain the source behavior or switch to another demonstrated task. These switches show that language is not simply ignored. Readouts and interventions connect these choices to task-conditioned internal states. Our analysis of the imitation objective shows how narrow conditional action support can leave grounded and instruction-keyed solutions indistinguishable on the demonstrations. This motivates Equivariant Counterfactual Training (ECT), which acts at two levels. ECT data supply valid demonstrations in which the same instruction requires different actions in distinguishable scenes, while the ECT loss trains each demonstration with its counterpart in the same update. In a controlled LIBERO-PRO comparison, full ECT raises $\pi_{0.5}$'s mean position-swap success from 36% to 59%. On CALVIN, where counterparts already occur in the original data, the ECT loss improves five-task completion without new demonstrations. On a real UR5e under a fixed demonstration budget, full ECT raises unseen-position success from 8% to 88%.
Chinese Translation
视觉-语言-行动（VLA）模型在分布内任务中成功率超过90%，并能抵御保持所需行动的干扰变化，但在要求不同行动的反事实变化下却失败。因此，整体鲁棒性评分可能掩盖更具体的失败情况，即策略同时响应语言和视觉，但未能将二者结合以选择任务所需的行动。我们称这种失败为指令-行动绑定。指令提示熟悉的轨迹家族，而视觉反馈则调整其执行。对微调后的 $ ext{π}_{0.5}$ 和 GR00T-N1.7 策略的行为分析表明，失败的展开往往保留源行为或切换到另一个已演示的任务。这些切换表明语言并非简单被忽视。读出和干预将这些选择与任务条件的内部状态联系起来。我们对模仿目标的分析显示，狭窄的条件行动支持如何使得基于实际和指令关键的解决方案在演示中难以区分。这激励了等变反事实训练（Equivariant Counterfactual Training, ECT），该方法在两个层面上发挥作用。ECT 数据提供有效的演示，其中相同的指令在可区分的场景中要求不同的行动，而 ECT 损失则在同一次更新中训练每个演示及其对应项。在受控的 LIBERO-PRO 比较中，完整的 ECT 将 $ ext{π}_{0.5}$ 的平均位置交换成功率从36%提高到59%。在 CALVIN 上，已有对应项出现在原始数据中，ECT 损失在没有新演示的情况下改善了五项任务的完成情况。在固定演示预算下的真实 UR5e 上，完整的 ECT 将未见位置的成功率从8%提高到88%。
cs.RO / 112 / 2609.39973
EWAM: Emergent Depth-Wise Specialization in a Unified Embodied Model -- From Semantic Understanding through Visual Foresight to Action
EWAM：统一具身模型中的新兴深度专业化——从语义理解到视觉预见再到行动
Wang, Hao, Wen, Jiajun, Liu, Jingzhi, Xue, Shuoshuo, Chen, Zhiliang, Lin, Min, Chang, Yicheng, Guo, Xiaoyu, Zhuo, Yukang, Chong, Zheng, Nie, Yunshuang, Zhang, Jian, Liufu, Weijia, Wu, Qingman, Xu, Heming, Song, Bingchang, Wu, Dantong, Wang, Zhiyuan, Xu, Hang, Han, Jianhua, Chen, Bokui, Zhao, Shen, Li, Rui, Liang, Xiaodan
Abstract
Vision-language-action (VLA) policies emphasize semantic understanding, whereas world-action models (WAMs) learn predictive representations of environment dynamics. Systems that expose a policy to both sources often still concentrate action computation on a single expert. We present EWAM, an action-centric unified embodied model whose asymmetric joint attention lets action tokens read semantic, current-visual, predicted-future, and action information at every layer while the perceptual experts retain their distinct roles. Without layer-wise supervision, EWAM develops an emergent depth-wise specialization: action queries attend mainly to vision-language features in shallow layers, to predicted future frames in intermediate layers, and to action tokens themselves in deep layers. This handoff replicates across tasks and is stable across denoising steps. Checkpoint tracking and causal interventions show that it is learned and that action generation depends on it. EWAM is pretrained in two separate regimes, one on cross-embodiment robot trajectories and one on human egocentric video. In simulation and real-robot experiments, it surpasses existing VLA, WAM, and hybrid baselines. Human egocentric data improve both cross-embodiment transfer and real-robot robustness, and subtask-phase supervision improves long-horizon completion. Together, these results suggest that unified embodied learning can induce an ordered internal progression from semantic understanding, through visual foresight, to action formation.
Chinese Translation
视觉-语言-行动（VLA）策略强调语义理解，而世界-行动模型（WAM）学习环境动态的预测表示。将策略暴露于这两种来源的系统通常仍然将行动计算集中在单一专家上。我们提出了EWAM，这是一种以行动为中心的统一具身模型，其不对称的联合注意力使得行动令牌在每一层都能够读取语义、当前视觉、预测未来和行动信息，同时感知专家保持其独特角色。在没有层级监督的情况下，EWAM发展出一种新兴的深度专业化：行动查询在浅层主要关注视觉-语言特征，在中间层关注预测的未来帧，而在深层关注行动令牌本身。这种交接在不同任务中得以复制，并在去噪步骤中保持稳定。检查点跟踪和因果干预表明这一过程是学习而来的，且行动生成依赖于此。EWAM在两个独立的训练模式下进行预训练，一个是在跨具身机器人轨迹上，另一个是在人的自我中心视频上。在仿真和真实机器人实验中，其表现超过了现有的VLA、WAM和混合基线。人类自我中心数据改善了跨具身转移和真实机器人鲁棒性，而子任务阶段监督则改善了长时间跨度的完成。综合这些结果表明，统一具身学习可以引导从语义理解到视觉预见再到行动形成的有序内部进程。
cs.RO / 113 / 2609.40007
Multi-Link Safety Filtering for VLA Policies Around Moving Hazards
移动危险物体周围的多链路安全过滤器用于视觉-语言-动作政策
Abstract
A vision-language-action (VLA) policy can finish a manipulation task while knocking over objects unrelated to it, so task success alone does not show that the policy is safe to deploy in clutter. We study how to keep a pretrained VLA policy clear of such hazards at run time without retraining it, which requires guarding more of the arm than the end effector, following the hazard as it moves, and sharing onboard compute with the policy. Our training-free shield covers the gripper, wrist, and forearm with five ellipsoids and filters every commanded motion through one barrier program against a keep-out ellipsoid fitted from RGB-D perception at reset. Sparse optical flow then carries that ellipsoid's center along with the hazard, with no repeated detection or refitting. Over six simulated hazard-motion conditions, the shield lowers collision from $65.62\%$ to $27.27\%$ and raises safe-success, task completion without collision, from $29.35\%$ to $50.43\%$. Ablations show that guarding the arm links protects beyond end-effector shielding, and that tracking recovers most of the protection lost when the hazard estimate is frozen at reset. On heterogeneous edge hardware, the five-ellipsoid barrier runs on the CPU in $2.2$~ms at the 99th percentile, and trimming the vision--language prefix and taking fewer flow-matching steps shortens each $\pi_{0.5}$ policy call on the integrated GPU from $343$ to $177.3$~ms. On a physical SO-101 arm across four tasks, the arm touched the hazard in 3 of 16 shielded episodes versus 11 of 16 unshielded ones. Project page: https://yathag.github.io/multilink-safety-filter/
Chinese Translation
视觉-语言-动作（VLA）政策可以在操控任务中完成操作，同时碰倒与任务无关的物体，因此仅凭任务成功并不能证明该政策在杂乱环境中是安全的。我们研究如何在运行时保持预训练的 VLA 政策远离这些危险物体，而无需重新训练，这需要保护手臂的更多部分而不仅仅是末端执行器，跟踪危险物体的移动，并与政策共享机载计算资源。我们的无训练保护装置使用五个椭球体覆盖夹持器、手腕和前臂，并通过一个屏障程序过滤每个命令的运动，该程序基于重置时从 RGB-D 感知获得的保持距离椭球体。稀疏光流随后将该椭球体的中心与危险物体一起移动，无需重复检测或重新拟合。在六种模拟的危险物体运动条件下，该保护装置将碰撞率从 65.62% 降低到 27.27%，并将安全成功率（无碰撞的任务完成率）从 29.35% 提高到 50.43%。消融实验表明，保护手臂链路的效果超出了末端执行器的保护，并且跟踪恢复了在重置时危险物体估计冻结时失去的大部分保护。在异构边缘硬件上，五椭球体屏障在第 99 百分位数下的 CPU 运行时间为 2.2 毫秒，缩短视觉-语言前缀并减少流匹配步骤将集成 GPU 上每个 π_{0.5} 政策调用的时间从 343 毫秒缩短到 177.3 毫秒。在物理 SO-101 手臂上进行的四个任务中，手臂在 16 次保护实验中触碰危险物体 3 次，而在 16 次未保护实验中触碰 11 次。项目页面：https://yathag.github.io/multilink-safety-filter/
cs.RO / 114 / 2609.40012
Identifiable Decomposition of Submovements in Human Hand Trajectories
人手轨迹中可识别的子运动分解
Abstract
Voluntary movements have long been hypothesised to be comprised of discrete primitives called submovements, as a descriptive model of human motor behaviour. However, existing methods scale poorly, and no principled method exists to determine whether a decomposition is informative. We propose a spatiotemporal kernel correlation between primitive pairs as an identifiability criterion. Submovement-Identifiable Decomposition (Sub-ID) embeds this criterion in its adaptive-ridge regularisation, biasing the optimiser toward low-correlation solutions. Identifiability is lost when primitives become collinear and recovered when they diverge spatially. On synthetic data, Sub-ID recovers ground-truth parameter distributions where existing methods fail; furthermore, when primitives overlap too heavily to be distinguished, the method explicitly detects this ambiguity rather than outputting misleading results. Sub-ID extracts submovements from real three-dimensional, long-horizon movements, a regime no prior method addresses. This method has the potential to identify physiologically grounded primitives for motor control research and imitation learning.
Chinese Translation
自愿运动长期以来被假设为由称为子运动的离散原语组成，这是一种描述人类运动行为的模型。然而，现有方法的扩展性较差，且没有原则性的方法来确定分解是否具有信息性。我们提出了一种基于原语对之间的时空核相关性作为可识别性标准。子运动可识别分解（Sub-ID）将这一标准嵌入其自适应岭回归中，使优化器偏向于低相关性的解决方案。当原语变得共线时，可识别性丧失，而当它们在空间上分散时则恢复。在合成数据上，Sub-ID能够恢复真实的参数分布，而现有方法则失败；此外，当原语重叠过于严重以至于无法区分时，该方法明确检测到这种模糊性，而不是输出误导性结果。Sub-ID从真实的三维长时间运动中提取子运动，这是之前方法未曾涉及的领域。该方法有潜力识别生理基础的原语，以用于运动控制研究和模仿学习。
cs.RO / 115 / 2609.40085
BatSLAM 2.0: Sequence-Verified Sonar Place Recognition in a Robust Pose Graph
BatSLAM 2.0：在稳健姿态图中的序列验证声纳位置识别
Abstract
Echolocating bats can navigate dark and cluttered spaces using echolocation. Over a decade ago, BatSLAM showed that a robot with a biomimetic binaural sonar can build a topological map of the environment, by recognizing places from the received acoustic signals. Sonar place recognition, however, is ambiguous by nature: corridors produce nearly identical echo trains, and wrong loop closure can collapse the topological map. In this paper, we introduce BatSLAM 2.0, a novel sonar-only SLAM system built from three elements: an updated acoustic front-end, a sequence verifier that tracks and verifies loop closure candidates and a pose graph implemented on a high performance factor graph framework. The system was thoroughly evaluated both in simulated as well as real world recordings. In both cases, the BatSLAM2.0 algorithm shows the capability of robust topological map creation, countering map collapse, and robust scaling of map size.
Chinese Translation
回声定位的蝙蝠能够利用回声定位在黑暗和杂乱的空间中导航。十多年前，BatSLAM展示了一种具有仿生双耳声纳的机器人如何通过识别接收到的声学信号来构建环境的拓扑地图。然而，声纳位置识别本质上是模糊的：走廊产生几乎相同的回声序列，而错误的环闭合可能导致拓扑地图崩溃。本文介绍了BatSLAM 2.0，一种新型的仅基于声纳的SLAM系统，由三个元素构成：更新的声学前端、跟踪和验证环闭合候选者的序列验证器，以及在高性能因子图框架上实现的姿态图。该系统在模拟和真实世界录音中进行了全面评估。在这两种情况下，BatSLAM 2.0算法展示了稳健的拓扑地图创建能力，抵御地图崩溃，并能够稳健地扩展地图大小。
cs.RO / 116 / 2609.40102
Passive Stiffness Shaping in Cable-Suspended Aerial Manipulation via Movable Compliant Anchors
通过可移动柔性锚点在缆索悬挂的空中操作中实现被动刚度塑形
Abstract
Cable-suspended aerial manipulation offers a lightweight architecture for cooperative transportation and physical interaction, yet the passive mechanical response perceived at the load remains insufficiently understood and systematically exploited. This work interprets aerial vehicles as movable compliant anchors and develops a gravity-aware quasi-static theory for predicting and shaping the passive Cartesian stiffness of a suspended load. The formulation applies to an arbitrary number of aerial vehicles connected to a point load by taut, straight, inextensible cables. At a selected gravity-loaded equilibrium, aerial-anchor compliance and transverse cable geometric compliance combine in series within each leg, while the leg stiffnesses act in parallel on the load. For isotropic aerial-anchor behavior, each leg is exactly equivalent to a virtual unilateral elastic cable, revealing an axial--transverse stiffness decomposition governed by the equilibrium tension. These results define a nonlinear map from commanded-anchor configuration to passive load stiffness, whose differential enables local constraint-preserving shaping through anchor repositioning. A dynamic rigid-body validation framework with nonlinear vehicle control, elastic-damped tendons, and environmental contact is defined to assess when and to what extent the derived stiffness remains predictive beyond the assumptions of the analytical model.
Chinese Translation
缆索悬挂的空中操作提供了一种轻量化架构，用于协作运输和物理交互，但在负载上感知到的被动机械响应仍然不足以被理解和系统性利用。本研究将空中载具视为可移动的柔性锚点，并开发了一种重力感知的准静态理论，用于预测和塑造悬挂负载的被动笛卡尔刚度。该公式适用于任意数量的空中载具，通过紧绷、直线、不可伸缩的缆索连接到一个点负载。在选定的重力负载平衡状态下，空中锚点的柔性和横向缆索的几何柔性在每条腿内串联组合，而腿的刚度则在负载上并联作用。对于各向同性的空中锚点行为，每条腿恰好等同于一个虚拟的单侧弹性缆索，揭示了由平衡张力主导的轴向-横向刚度分解。这些结果定义了从指令锚点配置到被动负载刚度的非线性映射，其微分使得通过锚点重新定位实现局部约束保持的塑形。定义了一个动态刚体验证框架，结合非线性载具控制、弹性阻尼绳索和环境接触，以评估在何种情况下以及在多大程度上推导出的刚度在超出分析模型假设的情况下仍然具有预测性。
cs.RO / 117 / 2609.40134
Tactile Curiosity Drives Robot Interaction
触觉好奇心驱动机器人互动
Abstract
Mastering robot manipulation skills via reinforcement learning (RL) remains largely sample-inefficient. The most common RL algorithms rely on random action sampling to discover new strategies, resulting in agents that allocate most of their training budget to motions in free space, away from the contacts from which manipulation skills emerge. Existing intrinsic motivation methods based on model disagreement or epistemic uncertainty improve on isotropic noise, but they can also reward uncertainty in functionally irrelevant transitions, such as erratic motions in free space. In this work, we argue that tactile feedback provides a natural signal for exploration, and introduce TacEx, a framework that incorporates touch into epistemic uncertainty-driven exploration by decomposing model uncertainty across sensory modalities and directing curiosity toward the tactile channel. By anchoring curiosity to the sense of touch, TacEx drives the robot to discover complex contact dynamics, learning to manipulate and grasp objects without task rewards or expert demonstrations during exploration. The interaction-dense dataset collected through this tactile-driven curiosity supports offline learning of downstream pick-and-place policies without additional environment interaction. We further use tactile-driven exploration to post-train vision-language-action (VLA) models. Although the VLAs are initially pre-trained without tactile feedback, post-training with TacEx substantially improves downstream performance while remaining highly sample-efficient.
Chinese Translation
通过强化学习（RL）掌握机器人操作技能仍然在很大程度上存在样本效率低下的问题。最常见的RL算法依赖随机动作采样来发现新策略，导致代理将大部分训练预算分配给自由空间中的动作，而远离产生操作技能的接触点。现有基于模型不一致性或认知不确定性的内在激励方法在各向同性噪声上有所改进，但它们也可能奖励在功能上无关的转变中的不确定性，例如自由空间中的不规则运动。在本研究中，我们认为触觉反馈为探索提供了自然信号，并引入了TacEx，一个通过在感官模态之间分解模型不确定性并将好奇心引导至触觉通道来将触觉纳入认知不确定性驱动探索的框架。通过将好奇心锚定于触觉，TacEx驱动机器人发现复杂的接触动态，学习在探索过程中无需任务奖励或专家演示即可操作和抓取物体。通过这种触觉驱动的好奇心收集的交互密集数据集支持下游抓取与放置策略的离线学习，而无需额外的环境交互。我们进一步利用触觉驱动的探索对视觉-语言-动作（VLA）模型进行后训练。尽管VLA最初是在没有触觉反馈的情况下进行预训练的，但使用TacEx进行后训练显著提高了下游性能，同时保持了高度的样本效率。
cs.RO / 118 / 2609.40153
Dream4ACT: A Shared Visual Action Interface for Multi-Embodiment Video-Action Modeling
Dream4ACT：用于多具身视频动作建模的共享视觉动作接口
Abstract
Video generation models (VGMs) offer strong spatiotemporal priors for embodied observation--action modeling. However, joint-space action vectors lack explicit image-space structure and vary in dimensionality and semantics across embodiments, making it challenging to directly leverage the rich spatiotemporal priors of VGMs. End-effector visualizations provide an alternative but do not specify the full articulated configuration needed for robot execution. We present Dream4ACT, a world model built for joint video-action modeling across embodiments. To unify action representations across embodiments, we introduce a shared visual action interface, called action views, which render target joint configurations from four prescribed virtual cameras using URDF-based forward kinematics. This shared visual representation preserves embodiment-specific articulated geometry while allowing observation and action sequences to share a video autoencoder and diffusion transformer. Through masked flow-matching, our model supports forward dynamics, inverse dynamics, and joint observation--action generation within a single jointly trained model by varying which future sequences are corrupted. To recover executable action sequences from predicted action views, we propose a training-free, URDF-constrained multiview recovery mechanism, without a learned embodiment-specific decoder. Dream4ACT achieves an average success rate of 88.98\% on RoboTwin~2.0 and an overall score of 65.66 on TriWorldBench, supporting effective closed-loop manipulation and competitive action-conditioned multiview prediction through the visual action interface.
Chinese Translation
视频生成模型（VGM）为具身观察-动作建模提供了强大的时空先验。然而，关节空间动作向量缺乏明确的图像空间结构，并且在不同具身体之间的维度和语义各异，这使得直接利用VGM丰富的时空先验变得具有挑战性。末端执行器可视化提供了一种替代方案，但并未指定机器人执行所需的完整关节配置。我们提出了Dream4ACT，这是一个为跨具身体的联合视频-动作建模构建的世界模型。为了统一跨具身体的动作表示，我们引入了一种共享视觉动作接口，称为动作视图（action views），该接口使用基于URDF的前向运动学从四个预设虚拟摄像机渲染目标关节配置。该共享视觉表示保留了具身特定的关节几何形状，同时允许观察和动作序列共享一个视频自编码器和扩散变换器。通过掩蔽流匹配，我们的模型支持在单个联合训练模型中进行前向动力学、逆向动力学和关节观察-动作生成，方法是改变哪些未来序列被破坏。为了从预测的动作视图中恢复可执行的动作序列，我们提出了一种无训练的、受URDF约束的多视图恢复机制，而无需学习具身特定的解码器。Dream4ACT在RoboTwin~2.0上实现了88.98\%的平均成功率，在TriWorldBench上获得了65.66的总体得分，支持通过视觉动作接口进行有效的闭环操控和竞争性的动作条件多视图预测。
cs.RO / 119 / 2609.40158
Rethinking Legibility in Social Robot Hallway Navigation: Impact of Intent Representation and Human Distraction
重新思考社交机器人走廊导航中的可读性：意图表现和人类分心的影响
Abstract
We focus on legible robot motion generation in social navigation settings. Legibility in human-robot interaction (HRI) is often described as the property of robot motion that enables an observer to confidently infer the robot's intent. While mature frameworks exist for generating legible motion in front of static observers, social robot navigation presents a new challenge: the robot must clearly convey its intent while ensuring human safety in dynamic pedestrian environments where human attention is often divided. With the goal of enabling robots to generate legible motion in dynamic and constrained spaces, we investigate how the choice of representation and the level of human attention shape navigation performance and human impressions. Focusing on the ubiquitous and demanding scenario of hallway navigation, we conduct two controlled user studies involving alternative legibility formulations implemented within a shared model predictive control framework. Study 1 (N = 45) investigates the role of intent representation, showing that passing-side legibility, particularly when adaptively updated, leads to smoother human motion and is perceived as more competent and less mentally and physically demanding than destination-based and non-legible baselines. Study 2 (N = 45) examines the effect of pedestrian attention, demonstrating that legible motion allows for smooth human motion even under distraction, even if this is not consistently reflected in subjective ratings. Together, these findings suggest that effective legible motion in social robot navigation benefits from interaction-level intent representations that support coordination, with some effects persisting even when human attention is divided. Code is available at https://github.com/fluentrobotics/Legible_MPPI.
Chinese Translation
我们关注社交导航环境中可读机器人运动的生成。人机交互（HRI）中的可读性通常被描述为机器人运动的特性，使观察者能够自信地推断机器人的意图。虽然已有成熟的框架用于在静态观察者面前生成可读运动，但社交机器人导航带来了新的挑战：机器人必须在动态行人环境中清晰地传达其意图，同时确保人类安全，而此时人类的注意力往往是分散的。为了使机器人能够在动态和受限空间中生成可读运动，我们研究了表现形式的选择和人类注意力水平如何影响导航性能和人类印象。我们专注于走廊导航这一普遍且要求高的场景，进行了两项控制用户研究，涉及在共享模型预测控制框架内实施的替代可读性公式。研究1（N = 45）探讨了意图表现的作用，结果表明，特别是在自适应更新时，经过通过侧可读性能够导致更流畅的人类运动，并被认为比基于目的地和不可读基线更具能力且在心理和身体上要求更低。研究2（N = 45）考察了行人注意力的影响，结果表明，即使在分心的情况下，可读运动也能实现流畅的人类运动，尽管这并不总是反映在主观评分中。综合这些发现表明，社交机器人导航中的有效可读运动受益于支持协调的交互级意图表现，且在某些情况下，即使人类注意力分散，这些效果仍然存在。代码可在 https://github.com/fluentrobotics/Legible_MPPI 获取。
cs.RO / 120 / 2609.40165
PrefPI: Preference-Guided Steering into Out-of-Distribution Behaviors
PrefPI：基于偏好的引导以应对分布外行为
Abstract
We present PrefPI (Preference-Guided Policy Iteration), an iterative framework for steering pretrained generative robot policies using only relative preferences over self-generated trajectories. Unlike prior preference-learning methods that primarily sharpen modes already represented by the policy, we study steering beyond the initial effective support, where desired behaviors are rarely or never observed under the initial policy. Our key idea is to formulate preference learning as preference-conditioned generative modeling: preferred trajectories define a conditional distribution, whose density ratio with the broader behavior prior provides an implicit preference signal amplified by classifier-free guidance (CFG). Repeating this preference-conditioned modeling and guidance step yields a form of preference-guided policy iteration, turning incremental improvements toward previously inaccessible behaviors. Across diffusion policies and the PI0.5 flow- matching VLA in simulation and the real world, PrefPI produces substantial behavioral shifts with limited feedback. In particular, PrefPI increases object transport height from 10.7 cm to 19.8 cm on real hardware with only 150 preference-labeled trajectories.
Chinese Translation
我们提出了PrefPI（基于偏好的策略迭代），这是一个迭代框架，用于仅通过对自生成轨迹的相对偏好来引导预训练的生成机器人策略。与以往主要通过强化已有策略所代表的模式的偏好学习方法不同，我们研究了超越初始有效支持的引导，其中所需行为在初始策略下很少或从未观察到。我们的关键思想是将偏好学习形式化为偏好条件生成建模：偏好的轨迹定义了一个条件分布，其与更广泛行为先验的密度比提供了一个隐式的偏好信号，并通过无分类器引导（CFG）进行放大。重复这一偏好条件建模和引导步骤产生了一种偏好引导的策略迭代形式，使得对先前不可达行为的增量改进成为可能。在模拟和现实世界中，通过扩散策略和PI0.5流匹配VLA，PrefPI在有限反馈下产生了显著的行为变化。特别是，PrefPI在真实硬件上将物体运输高度从10.7厘米提高到19.8厘米，仅使用150条带偏好的轨迹。
cs.RO / 121 / 2609.40177
Social-WM: Safety-Aware Latent World Models for Robot Social Navigation
Social-WM：面向安全的机器人社交导航潜在世界模型
Abstract
Safe social navigation requires a robot to anticipate not only the future consequences of its actions, but also whether a nominal action can actually be executed under surrounding physical and social constraints. We present Social-WM, an efficient latent world-model planning framework trained from egocentric RGB video sequences. Our key observation is that social-navigation experience contains a systematic discrepancy between the nominal action and the realizable action: a nominal forward action may be fully executed in free space, but needs to be constrained when heading towards a pedestrian or obstacle. Social-WM learns these safety-relevant consequences directly through action-conditioned future prediction, where the target is the actual observed future following each command. We further introduce a realizable inverse-dynamics objective that associates observed latent transitions with the action actually realized rather than the nominal one. At deployment, candidate actions are imagined through the latent world model, and the inverse dynamics model estimates their realizability; nominal--realizable discrepancy then provides a safety signal before execution. The learned dynamics and realizability model remain goal-independent and support both position- and image-goal navigation. On Social-HM3D, Social-WM achieves 63.77% success while reducing human collisions to 21.67%, and maintains strong performance under zero-shot transfer to Social-MP3D, without explicit pedestrian tracking, privileged human state, or online reinforcement learning.
Chinese Translation
安全的社交导航要求机器人不仅要预测其行动的未来后果，还要判断在周围的物理和社交约束下，名义行动是否能够实际执行。我们提出了Social-WM，这是一种高效的潜在世界模型规划框架，基于自我中心的RGB视频序列进行训练。我们的关键观察是，社交导航经验中存在名义行动与可实现行动之间的系统性差异：在自由空间中，名义的前进行动可以完全执行，但在接近行人或障碍物时则需要受到约束。Social-WM通过基于行动的未来预测直接学习这些与安全相关的后果，其目标是每个指令后实际观察到的未来。我们进一步引入了一种可实现的逆动力学目标，将观察到的潜在转变与实际实现的行动关联，而非名义行动。在部署时，通过潜在世界模型想象候选行动，逆动力学模型估计其可实现性；名义与可实现之间的差异则在执行前提供安全信号。所学习的动力学和可实现性模型保持目标独立，支持位置和图像目标导航。在Social-HM3D上，Social-WM实现了63.77%的成功率，同时将人类碰撞率降低到21.67%，并在零样本迁移到Social-MP3D时保持强劲表现，无需显式的行人跟踪、特权人类状态或在线强化学习。
cs.RO / 122 / 2609.40178
Non-Invasive Inspection of Water Canals Using Dronar
使用Dronar对水渠进行非侵入式检测
Abstract
Open concrete canals play a vital role in water transportation, serving as primary water infrastructure for millions of people across the Phoenix, Arizona, metro area. Over time, the concrete canals can experience a range of issues, including canal lining deformation, cracked concrete, and sediment buildup on the canal floor. Identifying such critical issues is a resource-intensive process, which currently happens only during four-year dry-up cycles. This prevents the maintenance crew from prioritizing operations on the most affected canal segments. To address this issue, the research team has developed and verified an easily deployable and non-invasive method to inspect canal beds without draining the water. This inspection system integrates affordable, off-the-shelf drone and sonar technology (termed dronar). This dronar system includes a consumer-grade sonar system integrated into an unmanned surface vehicle (USV) that carries the sonar transducer just under the surface of the canal water. This paper presents a proof-of-concept demonstration of the dronar system across three field tests on the Arizona Canal in Phoenix. DownScan depth profiles from the sedimented canal segment were consistently shallower than profiles from the same segment after cleaning, with offsets of up to 15 cm observed along the track. Repeated runs over the clean segment produced closely overlapping DownScan depth profiles, confirming that the dronar yields repeatable measurements across the natural variation of the canal bed. These results establish the dronar as a viable proof-of-concept tool for non-invasive canal bed inspection.
Chinese Translation
开放式混凝土水渠在水运输中发挥着至关重要的作用，为亚利桑那州凤凰城大都市区数百万人的主要水利基础设施。随着时间的推移，混凝土水渠可能会出现一系列问题，包括水渠衬里变形、混凝土开裂以及水渠底部沉积物堆积。识别这些关键问题是一个资源密集型的过程，目前仅在四年一次的干涸周期内进行。这使得维护团队无法优先处理受影响最严重的水渠段。为了解决这一问题，研究团队开发并验证了一种易于部署且非侵入式的水渠床检测方法，无需排水。该检测系统集成了经济实惠的现成无人机和声纳技术（称为Dronar）。该Dronar系统包括一个集成在无人水面载具（USV）中的消费级声纳系统，该载具将声纳传感器放置在水渠水面下方。本文展示了Dronar系统在凤凰城亚利桑那水渠进行的三次现场测试的概念验证演示。沉积水渠段的DownScan深度剖面始终比清理后同一段的剖面浅，沿轨迹观察到的偏差最大可达15厘米。对清理段的重复测量产生了紧密重叠的DownScan深度剖面，确认Dronar在水渠床的自然变化中能够提供可重复的测量结果。这些结果确立了Dronar作为一种可行的非侵入式水渠床检测工具的概念验证。
cs.RO / 123 / 2609.40208
Centralized Multi-UAV Exploration and 3D Reconstruction Using Single-UAV Planners
基于单无人机规划者的集中式多无人机探索与三维重建
Abstract
Extending single Unmanned Aerial Vehicles (UAVs) exploration methods to multi-UAV teams can improve coverage speed and robustness, but introduces challenges such as consistent mapping, safe navigation, and deployment strategy. In this work, we present a centralized multi-UAV exploration framework that enables the use of existing single-UAV sampling-based planners in a multi-UAV setting. The proposed architecture allows multiple UAVs to collaboratively explore unknown environments using a shared global Truncated Signed Distance Field (TSDF) map and centralized planning. Building on the voxblox library, we adapt its mapping pipeline to support real-time fusion of depth measurements from multiple UAVs into a common TSDF representation. In addition, inter-UAV collision avoidance and robot self-filtering mechanisms are integrated into the system to ensure safe navigation and prevent reconstruction of other UAVs as static obstacles. The framework is evaluated in simulation using four sampling-based exploration planners - RH-NBVP, KRH-NBVP, AEP, and KAEP - whose core sampling logic is preserved, with only system-level adaptations for multi-UAV operation. Experiments are conducted across multiple environments and under two deployment configurations: Joint Start (JS), where UAVs are initialized in close proximity, and Separated Start (SS), where UAVs are initialized in distinct locations. Results show that SS deployments consistently achieve faster exploration and improved coverage across all planners, highlighting the importance of the deployment strategy in multi-UAV exploration performance.
Chinese Translation
将单个无人机（UAV）的探索方法扩展到多无人机团队可以提高覆盖速度和鲁棒性，但也带来了诸如一致性映射、安全导航和部署策略等挑战。在本研究中，我们提出了一种集中式多无人机探索框架，该框架使现有的单无人机基于采样的规划者能够在多无人机环境中使用。所提出的架构允许多个无人机使用共享的全局截断符号距离场（Truncated Signed Distance Field, TSDF）地图和集中式规划共同探索未知环境。基于voxblox库，我们调整了其映射管道，以支持来自多个无人机的深度测量的实时融合，形成共同的TSDF表示。此外，系统中集成了无人机间碰撞避免和机器人自我过滤机制，以确保安全导航并防止将其他无人机重建为静态障碍物。该框架在模拟中使用四种基于采样的探索规划者进行评估——RH-NBVP、KRH-NBVP、AEP和KAEP，其核心采样逻辑得以保留，仅对系统级进行了多无人机操作的适配。实验在多个环境中进行，并在两种部署配置下进行：联合启动（Joint Start, JS），其中无人机在近距离内初始化，以及分开启动（Separated Start, SS），其中无人机在不同位置初始化。结果表明，SS部署在所有规划者中始终实现了更快的探索和更好的覆盖，突显了部署策略在多无人机探索性能中的重要性。
cs.RO / 124 / 2609.40245
STARS: From Spatiotemporal Dynamics to Social Representations in Human-Robot Interaction
STARS：从时空动态到人机交互中的社会表征
Abstract
Robot navigation in dynamic, human-centered environments requires socially-compliant decisions grounded in robust scene understanding. Recent Vision-Language Models (VLMs) exhibit promising capabilities such as object recognition, common-sense reasoning, and contextual understanding, capabilities that align with the nuanced requirements of social robot navigation. However, it remains unclear whether VLMs can accurately understand complex social navigation scenes (e.g., inferring the spatial-temporal relations among agents and human intentions), which is essential for safe and socially compliant robot navigation. While some recent works have explored the use of VLMs in social robot navigation, no existing work systematically evaluates their ability to meet these necessary conditions. In this paper, we introduce the Social Navigation Scene Understanding Benchmark (SocialNav-SUB), a Visual Question Answering (VQA) dataset and benchmark designed to evaluate VLMs for scene understanding in real-world social robot navigation scenarios. SocialNav-SUB provides a unified framework for evaluating VLMs against human and rule-based baselines across VQA tasks requiring spatial, spatiotemporal, and social reasoning in social robot navigation. Through experiments with state-of-the-art VLMs, we find that while the best-performing VLM achieves an encouraging probability of agreeing with human answers, it still underperforms simpler rule-based approach and human consensus baselines, indicating critical gaps in social scene understanding of current VLMs. Our benchmark sets the stage for further research on foundation models for social robot navigation, offering a framework to explore how VLMs can be tailored to meet real-world social robot navigation needs. An overview of this paper along with the code and data can be found at https://larg.github.io/socialnav-sub.
Chinese Translation
在动态的人本环境中，机器人导航需要基于稳健场景理解的社会合规决策。近期的视觉-语言模型（Vision-Language Models, VLMs）展现出诸如物体识别、常识推理和上下文理解等有前景的能力，这些能力与社会机器人导航的细致要求相契合。然而，目前尚不清楚VLMs是否能够准确理解复杂的社会导航场景（例如，推断代理之间的时空关系和人类意图），而这对于安全和社会合规的机器人导航至关重要。尽管一些近期的研究探讨了VLMs在社会机器人导航中的应用，但现有研究并未系统评估其满足这些必要条件的能力。在本文中，我们引入了社会导航场景理解基准（Social Navigation Scene Understanding Benchmark, SocialNav-SUB），这是一个视觉问答（Visual Question Answering, VQA）数据集和基准，旨在评估VLMs在现实社会机器人导航场景中的场景理解能力。SocialNav-SUB提供了一个统一的框架，用于在需要空间、时空和社会推理的社会机器人导航VQA任务中评估VLMs与人类和基于规则的基线的表现。通过与最先进的VLMs进行实验，我们发现，尽管表现最佳的VLM在与人类答案一致的概率上取得了令人鼓舞的结果，但其仍然不及更简单的基于规则的方法和人类共识基线，表明当前VLMs在社会场景理解方面存在关键差距。我们的基准为社会机器人导航的基础模型进一步研究奠定了基础，提供了一个框架以探索如何调整VLMs以满足现实世界的社会机器人导航需求。本文的概述以及代码和数据可在 https://larg.github.io/socialnav-sub 找到。
cs.RO / 125 / 2609.40297
GPU-Accelerated Path-Dependent Marginal Information Gain for Autonomous Exploration
基于GPU加速的路径依赖边际信息增益用于自主探索
Abstract
Autonomous exploration demands that robots continuously evaluate candidate viewpoints based on their expected information gain and execution cost. Sampling-based planners estimate this gain by volumetric raycasting and, due to its computational cost, evaluate candidates under an assumption of mutual independence, ignoring the overlap between viewpoints along the same path. This work presents a GPU-accelerated method for computing path-dependent marginal information gain, where instead of storing and merging the observed unknown voxels along each candidate path, previous observations are represented using depth buffers. Candidate rays are projected into the depth buffers of their ancestors to identify observation overlap and exclude regions expected to be observed. The planning tree is evaluated in depth order to maintain the dependency between viewpoints and their optimized yaws, while candidate nodes and rays at each level are processed in parallel on the GPU. The proposed method stays within 5-10% of the exact marginal gain computed using voxel hash maps, with speed-ups of up to 118x on a desktop GPU and 28x on an NVIDIA Jetson Orin NX. The method was integrated into two sampling-based exploration planners and evaluated in three simulation environments, where marginal gain reduced the time to 95% coverage in five of the six evaluated planner-environment combinations. Real-world experiments also showed a 30% reduction in the time to 95% coverage, as well as earlier exploration termination times.
Chinese Translation
自主探索要求机器人持续评估候选视点，基于其预期的信息增益和执行成本。基于采样的规划器通过体积光线投射来估计这一增益，由于其计算成本，通常在假设相互独立的前提下评估候选视点，忽略了沿同一路径的视点之间的重叠。本研究提出了一种基于GPU加速的路径依赖边际信息增益计算方法，其中不再存储和合并每个候选路径上的观察未知体素，而是使用深度缓冲区表示先前的观察。候选光线被投影到其祖先的深度缓冲区中，以识别观察重叠并排除预期被观察的区域。规划树按照深度顺序进行评估，以保持视点及其优化偏航之间的依赖关系，同时每一层的候选节点和光线在GPU上并行处理。所提出的方法在计算精确边际增益时，使用体素哈希图的结果保持在5-10%的误差范围内，并在桌面GPU上实现了最高118倍的加速，在NVIDIA Jetson Orin NX上实现了28倍的加速。该方法已集成到两个基于采样的探索规划器中，并在三个仿真环境中进行了评估，其中边际增益减少了六个评估的规划器-环境组合中五个的95%覆盖时间。实际实验也显示出95%覆盖时间减少了30%，以及更早的探索终止时间。
cs.RO / 126 / 2609.40306
DynaHarness: A Dynamic Physical Harness for Self-Evolving Robot Agents
DynaHarness：一种用于自我进化机器人代理的动态物理束缚装置
Abstract
Pretrained robot policies provide useful action priors, but long-horizon manipulation still requires coordination between semantic reasoning and physical execution. Semantic reasoning operates at a coarser timescale than physical interaction, while episode-level failures provide limited guidance on which system component should be revised. We propose DynaHarness, a dynamic physical harness that couples semantic reasoning with physical governance through a shared execution contract and turns failure evidence into validated capability revisions. To be more specific, the slow brain proposes capabilities and symbolic arguments, while the fast brain grounds and monitors commands, refuses unresolved actions, substitutes capabilities, and requests replans when needed. The physical execution contract bounds each accepted command and records execution evidence across analytic skills, recovery skills, and the frozen VLA. Failure attribution localizes faults in these records and directs targeted revisions of reusable capabilities or execution mechanisms. Paired regression checks govern admission or rejection, closing the self-evolution loop. On LIBERO-Pro, DynaHarness achieves 75.2% on 800 newly sampled initial states, compared with 17.5% for the frozen policy. With the same capability library, full dynamic execution reaches 74.0% versus 63.9% under nominal one-step replanning. This demonstrates the value of DynaHarness as a dynamic physical harness that governs how existing capabilities are grounded, monitored, and coordinated during execution. Our project page is at https://denghaoyuan123.github.io/Dynaharness_page/.
Chinese Translation
预训练的机器人策略提供了有用的动作先验，但长时间跨度的操作仍然需要语义推理与物理执行之间的协调。语义推理的操作时间尺度比物理交互要粗糙，而情节级别的失败对哪些系统组件应进行修订提供的指导有限。我们提出了DynaHarness，一种动态物理束缚装置，通过共享执行合同将语义推理与物理治理结合起来，并将失败证据转化为经过验证的能力修订。更具体地说，慢速大脑提出能力和符号论证，而快速大脑则对命令进行基础和监控，拒绝未解决的动作，替代能力，并在需要时请求重新规划。物理执行合同界定每个接受的命令，并记录分析技能、恢复技能和冻结的VLA（Variable Length Action）中的执行证据。失败归因在这些记录中定位故障，并指导可重用能力或执行机制的有针对性的修订。配对回归检查控制接受或拒绝，从而闭合自我进化循环。在LIBERO-Pro上，DynaHarness在800个新采样的初始状态中达到了75.2%的成功率，而冻结策略仅为17.5%。在相同的能力库下，完全动态执行达到了74.0%，而在正常的一步重新规划下为63.9%。这证明了DynaHarness作为一种动态物理束缚装置的价值，能够管理现有能力在执行过程中的基础、监控和协调。我们的项目页面地址为 https://denghaoyuan123.github.io/Dynaharness_page/。
cs.RO / 127 / 2609.40341
Ego4WAM: What Matters When Scaling Egocentric Human Data for Robot Learning?
Ego4WAM：在扩展自我中心人类数据以进行机器人学习时，什么是重要的？
Abstract
Egocentric human data provides a scalable source of experience for robot learning, but varies substantially in human-robot alignment, behavioral coverage, and available supervision. Existing work shows favorable scaling with increasing human data, but it remains unclear which data properties drive downstream robot gains and how to use such data throughout the training pipeline. We present a systematic study of egocentric human data with different alignment and supervision under a unified world-action model framework. With the model backbone fixed, we disentangle the effects of human-robot alignment, data duration and task diversity, action supervision, and data usage strategies. We find that aligned human demonstrations substantially improve out-of-distribution generalization and reduce target-task robot data requirements; data duration and task diversity affect downstream capabilities differently; and video-only supervision remains effective without action labels, providing a strong foundation for subsequent video-action training. We validate these findings through closed-loop policy evaluation on both real robots and RoboDojo. Rather than treating data duration as the sole scaling axis, Ego4WAM shows how alignment, task diversity, available supervision, and usage strategy jointly shape the value of egocentric human data for robot learning.
Chinese Translation
自我中心人类数据为机器人学习提供了可扩展的经验来源，但在人机对齐、行为覆盖和可用监督方面存在显著差异。现有研究表明，随着人类数据的增加，扩展效果良好，但尚不清楚哪些数据特性驱动下游机器人性能提升，以及如何在整个训练流程中使用这些数据。我们在统一的世界-动作模型框架下，对具有不同对齐和监督的自我中心人类数据进行了系统研究。在固定模型骨干的情况下，我们解开了人机对齐、数据持续时间和任务多样性、动作监督以及数据使用策略的影响。我们发现，对齐的人类示范显著提高了分布外泛化能力，并减少了目标任务对机器人数据的需求；数据持续时间和任务多样性对下游能力的影响各不相同；而仅使用视频监督在没有动作标签的情况下仍然有效，为后续的视频-动作训练提供了坚实基础。我们通过在真实机器人和RoboDojo上的闭环策略评估验证了这些发现。Ego4WAM表明，与将数据持续时间视为唯一的扩展轴不同，对齐、任务多样性、可用监督和使用策略共同塑造了自我中心人类数据在机器人学习中的价值。
cs.CV / 1 / 2609.38271
Evaluating Multi-Task Morphological Concept Learning for Pulmonary Nodule Malignancy Assessment in 3D CT
评估多任务形态学概念学习在三维CT中对肺结节恶性评估的应用
Abstract
Morphological characteristics such as spiculation and lobulation play an important role in assessing pulmonary nodules on computed tomography (CT), particularly in relation to malignancy risk. This study examines whether learning radiologist-annotated morphological features together with malignancy risk from lesion-centred 3D CT volumes improves classification performance. The Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset was used, comprising 3,918 reader-level nodule annotations from 742 patients after excluding indeterminate malignancy ratings. Patient-level splitting was used for training, validation, and testing, with 112 patients and 628 reader annotations in the held-out test set. A single-task 3D convolutional neural network was compared with a multi-task model predicting malignancy risk, spiculation, and lobulation. The single-task model achieved a balanced accuracy of 0.548 and receiver operating characteristic area under the curve (ROC-AUC) of 0.552, while the multi-task model achieved 0.539 and 0.558, respectively. Patient-level bootstrap analysis showed an ROC-AUC difference of 0.005 (95% confidence interval (CI): -0.087 to 0.090) and a balanced-accuracy difference of -0.009 (95% CI: -0.067 to 0.043). The auxiliary tasks were strongly imbalanced and showed limited predictive performance. Overall, including morphological features did not clearly improve malignancy-risk classification, showing the importance of class balance, label formulation, and reader-level annotation structure in multi-task pulmonary CT analysis.
Chinese Translation
形态学特征如刺状和分叶在计算机断层扫描（CT）中评估肺结节时，尤其是在恶性风险方面，发挥着重要作用。本研究探讨了是否通过学习放射科医师标注的形态特征以及病灶中心的三维CT体积中的恶性风险，能够提高分类性能。使用了肺图像数据库联盟和图像数据库资源倡议（LIDC-IDRI）数据集，该数据集包含742名患者的3,918个读者级结节标注，排除了不确定的恶性评级。采用患者级别的划分进行训练、验证和测试，保留的测试集中有112名患者和628个读者标注。将单任务三维卷积神经网络与多任务模型进行比较，后者预测恶性风险、刺状和分叶。单任务模型的平衡准确率为0.548，接收者操作特征曲线下面积（ROC-AUC）为0.552，而多任务模型分别为0.539和0.558。患者级别的自助分析显示ROC-AUC差异为0.005（95%置信区间（CI）：-0.087至0.090），平衡准确率差异为-0.009（95% CI：-0.067至0.043）。辅助任务严重失衡，预测性能有限。总体而言，包含形态特征并未明显改善恶性风险分类，显示出在多任务肺CT分析中类别平衡、标签制定和读者级标注结构的重要性。
cs.CV / 2 / 2609.38278
Masked Swingers: Harnessing Data Augmentation to Advance Autoencoders for Self-Supervised Learning
掩蔽摆动者：利用数据增强推动自监督学习的自编码器
Abstract
Self-supervised learning (SSL) removes the need for annotations and makes models that are capable across more domains than supervised learning. The autoencoder SSL framework learns by reconstructing its own input after information loss through a bottleneck or noise injection. Masked autoencoders (MAE) are the most successful instantiation of this framework: they encode a random subset of patches, then decode the masked-out patches. In this work, we introduce key modifications to improve MAEs. Our method augments an image in two different ways, then masks and encodes each view separately. It then exchanges the global representations (CLS tokens) between views before decoding the masked patches. By design, our Masked Swingers encourages learning a view-agnostic summary of the image to facilitate efficient transfer. We perform extensive experiments, and find Masked Swingers outperforms MAE by +3-5% on ImageNet-1K kNN and provides large gains on fine-grained tasks, e.g., relative gains of +45% on instance retrieval, +22% on animal re-ID, and +76% on Omniglot character recognition. To boot, Swingers reduces error -64% relative to MAE on three new state-probing datasets, opening the door to world modeling. Welcome to our Swingers party.
Chinese Translation
自监督学习（SSL）消除了对注释的需求，使模型在多个领域的能力超越监督学习。自编码器SSL框架通过在瓶颈或噪声注入后重构自身输入来学习。掩蔽自编码器（Masked Autoencoders, MAE）是该框架最成功的实例：它们对随机选择的补丁进行编码，然后解码被掩蔽的补丁。在本研究中，我们引入了关键修改以改进MAE。我们的方法以两种不同方式增强图像，然后分别掩蔽和编码每个视图。接着，在解码被掩蔽的补丁之前，交换视图之间的全局表示（CLS tokens）。我们的掩蔽摆动者（Masked Swingers）旨在鼓励学习与视图无关的图像摘要，以促进高效的迁移。我们进行了广泛的实验，发现掩蔽摆动者在ImageNet-1K kNN上比MAE提高了3-5%的性能，并在细粒度任务上提供了显著提升，例如，在实例检索上相对提升45%，在动物重识别上提升22%，在Omniglot字符识别上提升76%。此外，摆动者在三个新的状态探测数据集上相对于MAE减少了64%的错误，开启了世界建模的新机遇。欢迎参加我们的摆动者派对。
cs.CV / 3 / 2609.38285
GaugeVLM: Structuring Spatial Supervision with Measured Geometric Interventions
GaugeVLM：通过测量几何干预构建空间监督
Abstract
Vision-language models (VLMs) can contradict themselves across views of the same spatial relation and fail to respond when that relation changes. Addressing these failures requires supervision that captures error magnitude and geometric dependencies across observations, both of which remain implicit in training on individual answers or ordinal preferences. Therefore, we introduce GaugeVLM, which makes this structure explicit through controlled object and camera interventions in explicit 3D scenes, producing linked observations with measured differences between spatial relations and shared truths across views. To translate this structure into learning signals, its core objective, GaugeDPO, converts measured errors into preference margins, directly supervises correct canonical rankings across views, and links intervention-induced answer-odds contrasts to measured relation changes with view-specific scales. Our analysis bounds canonical prediction error and establishes that the cross-view and intervention constraints can be jointly satisfied. Empirically, GaugeVLM improves all 10 established spatial metrics over supervised fine-tuning across three VLM backbones, with the main 7B model gaining 15.0 and 18.9 percentage points on MSMU distance and QSpatial+, respectively. These gains also extend to autonomous driving and embodied reasoning, demonstrating the robust generalization across domains.
Chinese Translation
视觉-语言模型（VLMs）在同一空间关系的不同视图中可能自相矛盾，并且在该关系发生变化时未能做出反应。解决这些失败需要一种能够捕捉误差幅度和观察之间几何依赖关系的监督，而这两者在对单个答案或序数偏好的训练中仍然是隐含的。因此，我们提出了GaugeVLM，它通过在明确的3D场景中进行受控的物体和相机干预，使这一结构变得显性，从而产生具有测量差异的关联观察，反映空间关系和跨视图的共享真相。为了将这一结构转化为学习信号，其核心目标GaugeDPO将测量的误差转换为偏好边际，直接监督跨视图的正确典范排名，并将干预引起的答案赔率对比与测量的关系变化通过视图特定的尺度相连接。我们的分析界定了典范预测误差，并建立了跨视图和干预约束可以共同满足的条件。从经验上看，GaugeVLM在三个VLM骨干网络上通过监督微调改善了所有10个已建立的空间指标，其中主模型7B在MSMU距离和QSpatial+上分别提高了15.0和18.9个百分点。这些提升也扩展到自动驾驶和具身推理，展示了跨领域的强健泛化能力。
cs.CV / 4 / 2609.38298
It Takes Little to Rewrite Perception: Targeted Semantic Substitution in Vision-Language Models at $\epsilon \leq 4/255$
重写感知只需小小改变：在 $oldsymbol{oldsymbol{ ext{ε}} ext{≤} 4/255}$ 下对视觉-语言模型的针对性语义替换
Abstract
Vision Language Models (VLMs) are widely deployed in safety-critical scenarios, and understanding to which extent they can be controlled by adversarial perturbation is a prerequisite for evaluating their trustworthiness. Existing representation-alignment attacks, which make a VLM perceive a target image, achieve limited success at $\varepsilon \leq 4/255$. Therefore, VLMs seems robust to perturbations in this range. We show that this robustness does not hold, as targeted semantic substitution succeeds within the same range. Specifically, we align each stream of the source image with its counterpart in the target image in the victim VLM's post-merger token space, operating under a white-box threat model. We evaluate under a strict success criterion, requiring the model to simultaneously name the target, confirm its presence, and deny the source. In images, target semantics appear at $\varepsilon = 2/255$ and complete replacement reaches 38\% at $\varepsilon = 4/255$. On video, complete replacement reaches 35.9\% at $\varepsilon = 1/255$. We also observe a phenomenon of \textit{semantic fusion}, where Large Language Model (LLM) rationalizes contradictory visual signals into a coherent narrative.
Chinese Translation
视觉语言模型（VLMs）广泛应用于安全关键场景，理解它们在多大程度上可以被对抗性扰动控制是评估其可信度的前提。现有的表示对齐攻击使得 VLM 感知目标图像，在 $oldsymbol{oldsymbol{ ext{ε}} ext{≤} 4/255}$ 的范围内取得了有限的成功。因此，VLM 在这一范围内似乎对扰动具有鲁棒性。我们表明这种鲁棒性并不成立，因为针对性语义替换在相同范围内成功。具体而言，我们在受害者 VLM 的后合并令牌空间中，将源图像的每个流与目标图像中的对应部分对齐，操作在白盒威胁模型下进行。我们在严格的成功标准下进行评估，要求模型同时命名目标、确认其存在并否认源图像。在图像中，目标语义在 $oldsymbol{oldsymbol{ ext{ε}} = 2/255}$ 时出现，完全替换在 $oldsymbol{oldsymbol{ ext{ε}} = 4/255}$ 时达到 38\%。在视频中，完全替换在 $oldsymbol{oldsymbol{ ext{ε}} = 1/255}$ 时达到 35.9\
%。我们还观察到一种 extit{语义融合} 现象，即大型语言模型（LLM）将矛盾的视觉信号合理化为连贯的叙述。
cs.CV / 5 / 2609.38325
Strike a Chord! Modal Kinetic Typography
击打和弦！模态动效排版
Abstract
We introduce modal kinetic typography, which animates a vector glyph to express a semantic concept while keeping it legible. Our key idea is to build motion from the glyph's natural vibration modes. Specifically, a finite-element eigenproblem assembled from the vector outline yields the glyph's softest modes, for the whole letter and for each of its parts, allowing it to bend. The problem's zero-energy solutions, i.e., rigid translations and rotations, are applied in closed form to each part, allowing parts to also move as blocks. To animate the glyph, a frozen video diffusion model supervises only the modes' amplitudes and phases. Our modal approach addresses two weaknesses of prior work. Free-form point optimization under video score distillation (SDS) moves each point and frame independently along noisy gradients, tearing the outline and causing jitter. In contrast, our modes are smooth along the outline and driven by a few whole-cycle harmonics, which restricts these gradients to smooth, seamlessly looping motion. On the other hand, structured alternatives rely on skeletons or keypoints from category-specific priors, whereas our modes come from the glyph itself; the only prior is a list naming each letter's moving parts, generated once for the whole alphabet by a language model. In modal kinetic typography, shape and motion are disentangled by construction: a single base outline is sculpted toward the concept, and the modal drive cannot alter it, so a letter can also be animated without being reshaped. Our method produces more articulated and smoother motion than Dynamic Typography and AniClipart at comparable or better concept alignment, with less glyph tearing than Dynamic Typography, and is preferred by human raters, including in a frozen-shape setting where motion alone must carry the concept. Our results were also preferred over Astra (GPT-6) by human raters.
Chinese Translation
我们提出了模态动效排版，它通过动画化矢量字形来表达语义概念，同时保持其可读性。我们的关键思想是从字形的自然振动模式中构建运动。具体而言，从矢量轮廓组装的有限元特征问题产生了字形的最柔软模式，适用于整个字母及其各个部分，使其能够弯曲。该问题的零能量解，即刚性平移和旋转，以封闭形式应用于每个部分，使得部分也可以作为块移动。为了动画化字形，一个冻结的视频扩散模型仅监督模式的幅度和相位。我们的模态方法解决了先前工作的两个弱点。在视频评分蒸馏（SDS）下的自由形式点优化独立地沿着噪声梯度移动每个点和帧，撕裂轮廓并导致抖动。相比之下，我们的模式在轮廓上是平滑的，并由少数完整周期的谐波驱动，这限制了这些梯度为平滑、无缝循环的运动。另一方面，结构化替代方案依赖于来自类别特定先验的骨架或关键点，而我们的模式来自字形本身；唯一的先验是一个列出每个字母运动部分的列表，由语言模型为整个字母表生成一次。在模态动效排版中，形状和运动在构造上是分离的：一个单一的基础轮廓被雕刻成概念，而模态驱动无法改变它，因此字母也可以在不被重塑的情况下进行动画化。我们的方法产生了比动态排版（Dynamic Typography）和动画剪贴画（AniClipart）更具表现力和更平滑的运动，同时在概念对齐方面相当或更好，且比动态排版的字形撕裂更少，并且在冻结形状设置中，人类评审者更倾向于我们的结果，在这种情况下，运动本身必须承载概念。我们的结果也在评审中优于Astra（GPT-6）。
cs.CV / 6 / 2609.38329
ExploreNet: Learning Where to Explore in Diffusion GRPO
ExploreNet：在扩散 GRPO 中学习探索的方向
Abstract
Group-relative RL methods such as Flow-GRPO post-train image generators by exploring with isotropic Gaussian noise added at every denoising step. This noise decides which rollouts the model learns from, yet it perturbs every channel and spatial position of the latent equally. In this paper, we instead show that latent elements differ in how much they change the generated image, so exploration should adapt to these differences. We introduce EXPLORENET to learn an adaptive exploration distribution. EXPLORENET is a policy that predicts a noise scale for every latent element from the current latent, the denoising step, and the prompt, before any reward is observed; it is trained on the reward spread of each rollout group and discarded after training, leaving inference unchanged. On Stable Diffusion 3.5 Medium, EXPLORENET improves held-out GenEval2 by 14% over Flow-GRPO, transfers to two independent compositional benchmarks and five preference and image-quality models, and reaches a 67.2% human preference win-rate. Overall, across our group-relative diffusion RL experiments, we find that exploration is learnable, the shape of the exploration distribution outweighs its magnitude, and rollout quality is more effective than rollout quantity.
Chinese Translation
群体相对强化学习（RL）方法，如 Flow-GRPO，通过在每个去噪步骤中添加各向同性的高斯噪声来后训练图像生成器。这种噪声决定了模型从哪些回放中学习，但它对潜在空间的每个通道和空间位置的扰动是相同的。在本文中，我们展示了潜在元素在改变生成图像的程度上存在差异，因此探索应当适应这些差异。我们引入了 EXPLORENET 来学习自适应探索分布。EXPLORENET 是一种策略，它在观察到任何奖励之前，根据当前潜在状态、去噪步骤和提示预测每个潜在元素的噪声尺度；它是在每个回放组的奖励分布上进行训练的，并在训练后被丢弃，从而不改变推理过程。在 Stable Diffusion 3.5 Medium 上，EXPLORENET 相较于 Flow-GRPO 提高了 14% 的 GenEval2 保留性能，能够迁移到两个独立的组合基准和五个偏好及图像质量模型，并达到了 67.2% 的人类偏好胜率。总体而言，在我们的群体相对扩散 RL 实验中，我们发现探索是可学习的，探索分布的形状超过其幅度，而回放质量比回放数量更有效。
cs.CV / 7 / 2609.38343
Learning Semantic Inpainting for Animatable Gaussian Head Avatars
学习可动画高斯头部头像的语义修复
Abstract
We present SInGA, a novel method for learning Semantic Inpainting for animatable Gaussian head Avatars from a single image. Existing avatar approaches often rely on multi-view observations and lack effective handling of unobserved regions in single-view settings, limiting their applicability in such scenarios. To address this, we propose a semantic inpainting framework defined in UV space for completing unobserved facial regions. Our key insight lies in the structured topology of the UV representation, which provides consistent spatial correspondences and enables reliable completion of identity-specific features using the inherent symmetry cues of human faces. We extract features from observed regions and use them to complete unobserved regions. The completed representation is then used to regress Gaussian attributes, effectively performing Gaussian inpainting. In addition, instead of relying on a single Gaussian at each surface or pixel location, we stack multiple Gaussians to enhance detail. The resulting avatar generalizes across identities without requiring per-identity optimization and can be animated with driving inputs. Experimental results show that our method generates high-quality head avatars with improved completeness and identity preservation, while supporting realistic animation and consistent rendering from unobserved views.
Chinese Translation
我们提出了一种新颖的方法SInGA，用于从单幅图像中学习可动画的高斯头部头像的语义修复。现有的头像方法通常依赖于多视角观测，并且在单视角设置中缺乏有效处理未观测区域的能力，从而限制了它们在此类场景中的适用性。为了解决这个问题，我们提出了一种在UV空间中定义的语义修复框架，用于完成未观测的面部区域。我们的关键见解在于UV表示的结构拓扑，它提供了一致的空间对应关系，并利用人脸固有的对称线索可靠地完成身份特定特征。我们从观测区域提取特征，并利用这些特征来完成未观测区域。完成的表示随后用于回归高斯属性，有效地执行高斯修复。此外，我们不再依赖于每个表面或像素位置的单一高斯，而是堆叠多个高斯以增强细节。最终生成的头像在不同身份之间具有良好的泛化能力，无需针对每个身份进行优化，并且可以通过驱动输入进行动画。实验结果表明，我们的方法生成的头部头像具有更高的完整性和身份保留，同时支持现实的动画效果和从未观测视角的一致渲染。
cs.CV / 8 / 2609.38347
TrackFish3D: Self-Supervised 3D Tracking of Schooling Fish from Multi-view Videos
TrackFish3D：基于多视角视频的自监督3D鱼群追踪
Abstract
Quantifying collective fish behavior requires accurate trajectories, yet multi-view 3D tracking remains challenging due to frequent occlusions, visually similar individuals, and the long-standing scarcity of identity annotations. We present TrackFish3D, a geometry-driven self-supervised framework for dense multi-camera 3D tracking of schooling fish. Instead of relying on appearance-based re-identification or manually annotated identities, TrackFish3D turns calibrated multi-view geometry into supervision: triangulation and reprojection consistency provide pseudo-associations, while a geometric encoder and global association transformer learn all-to-all cross-view correspondence within each frame. To make these associations identity-aware, TrackFish3D introduces a self-supervised contrastive objective that separates co-visible individuals in the embedding space, together with a temporal predictor that preserves identities and bridges short occlusions across frames. The resulting model is trained once on unlabeled footage and applied directly to unseen test videos, requiring no cross-view identity labels, temporal annotations, 3D ground truth, appearance features, or test-time optimization. On our benchmark, TrackFish3D improves 3D Multi-Object Tracking Accuracy from 87.7% for the strongest baseline to 95.8%. On the 3D-ZeF zebrafish benchmark, it achieves 81.1% MOTA, compared with 77.4% for the best geometric baseline. TrackFish3D also generalizes beyond fish, achieving strong results on real-world bird tracking.
Chinese Translation
量化集体鱼类行为需要准确的轨迹，然而由于频繁的遮挡、视觉上相似的个体以及长期缺乏身份注释，多视角3D追踪仍然具有挑战性。我们提出了TrackFish3D，一个基于几何驱动的自监督框架，用于密集的多摄像头3D鱼群追踪。TrackFish3D不依赖于基于外观的重识别或手动注释的身份，而是将校准的多视角几何转化为监督：三角测量和重投影一致性提供伪关联，而几何编码器和全局关联变换器则学习每帧内的全对全跨视图对应关系。为了使这些关联具有身份意识，TrackFish3D引入了一种自监督对比目标，在嵌入空间中分离共同可见的个体，并结合一个时间预测器，保持身份并跨帧桥接短暂的遮挡。最终模型在未标记的视频上训练一次，并直接应用于未见的测试视频，无需跨视图身份标签、时间注释、3D真实值、外观特征或测试时优化。在我们的基准测试中，TrackFish3D将3D多目标追踪准确率从最强基线的87.7%提高到95.8%。在3D-ZeF斑马鱼基准上，其MOTA达到81.1%，而最佳几何基线为77.4%。TrackFish3D还超越鱼类， 在真实世界鸟类追踪中取得了良好的结果。
cs.CV / 9 / 2609.38362
Inductive Visual Logic for Few-Shot Out-Of-Distribution Adaptation in VLMs
用于少样本分布外适应的归纳视觉逻辑在视觉语言模型中的应用
Abstract
Generative vision-language models (VLMs) such as Qwen-VL and LLaVA achieve strong zero-shot performance on tasks overlapping with their pretraining distribution, yet fail on specialized domains where the required discriminative features were never learned, a regime we term distant out-of-distribution (OOD). Standard adaptation methods cannot overcome this representational absence because they operate within the encoder's existing feature space. However, VLMs retain a robust descriptive capacity even when discrimination collapses: a model that cannot classify a medical scan can still articulate its visual patterns. Exploiting this asymmetry, we introduce Inductive Visual Logic (IVL), a training-free framework that constructs classification knowledge from the model's surviving descriptive ability. IVL extracts visual traits from few-shot support images through dual-mode prompting, combining semantic descriptions with primitive visual observations, and organizes them into per-class trait dictionaries. At inference, hierarchical filtering identifies spatially grounded trait evidence for classification. Across multiple distant-OOD benchmarks, IVL achieves the highest aggregate accuracy under two VLM backbones while producing interpretable, trait-traceable predictions.
Chinese Translation
生成性视觉语言模型（VLMs），如 Qwen-VL 和 LLaVA，在与其预训练分布重叠的任务上表现出强大的零-shot 性能，但在需要特定判别特征的专业领域却表现不佳，这种情况我们称之为远程分布外（OOD）。标准的适应方法无法克服这种表征缺失，因为它们在编码器现有的特征空间内操作。然而，即使在判别能力崩溃的情况下，VLMs 仍然保留了强大的描述能力：一个无法分类医学扫描的模型仍然可以阐述其视觉模式。利用这种不对称性，我们引入了归纳视觉逻辑（Inductive Visual Logic, IVL），这是一个无训练的框架，旨在从模型的存活描述能力中构建分类知识。IVL 通过双模式提示从少样本支持图像中提取视觉特征，将语义描述与原始视觉观察相结合，并将其组织成每类特征字典。在推理阶段，分层过滤识别空间上有依据的特征证据以进行分类。在多个远程 OOD 基准测试中，IVL 在两个 VLM 骨干网络下实现了最高的综合准确率，同时生成可解释的、可追溯特征的预测。
cs.CV / 10 / 2609.38368
Composition, Not Conversation: VLMs Lose the Scene, Not the Thread
构成，而非对话：视觉语言模型失去的是场景，而不是线索
Abstract
Vision-language models (VLMs) increasingly reason over visual evidence that is cropped, segmented, retrieved, or revealed over time. Yet most VQA benchmarks present the complete image and question at once. We ask what models lose when the same information is fragmented. We introduce Layered-VQA, with 93 scenes and 300 questions. Each image is decomposed into ordered RGBA layers that exactly recompose the original scene, and each question is annotated with supporting, minimal-sufficient, and distractor layers. We evaluate eleven open-weight VLMs from 3B to 32B parameters and two proprietary models with a scale of 187,200 conversations, graded by 1.74M open-model cross-judgments. We find three consistent failures. Loss in Composition: fragmenting the question has a small effect, but fragmenting the scene substantially reduces accuracy; recomposing the same layers largely restores performance. Oracle Inversion: even oracle-selected sufficient evidence can perform worse than the complete scene. Loss in Grounding: as more evidence is required, grounding degrades much faster than answer accuracy. Together, these results show that having the right visual evidence is not enough. How that evidence is composed and presented determines whether models can use and ground it. The right evidence is not enough: VLMs need the scene it came from.
Chinese Translation
视觉语言模型（VLMs）越来越多地基于经过裁剪、分割、检索或随时间揭示的视觉证据进行推理。然而，大多数视觉问答（VQA）基准测试一次性呈现完整的图像和问题。我们探讨当相同信息被分割时，模型会失去什么。我们引入了分层视觉问答（Layered-VQA），包含93个场景和300个问题。每幅图像被分解为有序的RGBA层，这些层可以精确重组成原始场景，每个问题则附有支持层、最小充分层和干扰层的注释。我们评估了11个开放权重的VLMs，参数规模从30亿到320亿，以及两个专有模型，涵盖187,200个对话，通过174万个开放模型的交叉评估进行评分。我们发现了三种一致的失败。构成的损失：问题的分割影响较小，但场景的分割显著降低了准确性；重组相同的层大致恢复了性能。Oracle反转：即使是oracle选择的充分证据，其表现也可能不如完整场景。定位的损失：随着所需证据的增加，定位的退化速度远快于答案的准确性。这些结果表明，拥有正确的视觉证据并不足够。证据的构成和呈现方式决定了模型是否能够使用和定位这些证据。仅有正确的证据是不够的：VLMs需要来自于其原始场景的证据。
cs.CV / 11 / 2609.38377
PhyProbe: Rethinking Physical Consistency Evaluation in Generated Videos
PhyProbe：重新思考生成视频中的物理一致性评估
Abstract
Evaluating the physical consistency of generated videos remains a fundamental challenge. Existing approaches rely on off-the-shelf vision-language models, which can often be myopic to physical dynamics, or fine-tuned evaluators trained on human annotations, which overfit to dataset-specific cues and fail to generalize. A key challenge is that existing supervision sources provide either relative ordering or absolute scores, but not both reliably and consistently across varied settings. To this end, we introduce PhyProbe, an evaluator that extracts features from a frozen pretrained spatio-temporal encoder and maps them to a scalar physical consistency violation score via a lightweight scoring head. PhyProbe is trained through a unified objective combining pairwise ranking, regression on noisy scalar annotations, and anchor-based calibration over a curated set of heterogeneous supervision sources. Experiments show that PhyProbe outperforms prior methods on most pairwise benchmarks spanning real-generated and generated-generated pairs under varying correspondence, with the largest gains in no-correspondence and generated-generated settings where existing fine-tuned evaluators degrade sharply. PhyProbe achieves strong correlation with human judgments, with close agreement between rank-based and linear metrics, indicating that scores are both well ordered and anchored to a stable [0, 1] scale. Further, despite being trained on supervision indicative of physical consistency, without explicit general-preference labels, PhyProbe also performs competitively on human preference benchmarks: consistent with the observation that physics violations are entangled with broader quality degradations.
Chinese Translation
评估生成视频的物理一致性仍然是一个基本挑战。现有方法依赖于现成的视觉-语言模型，这些模型往往对物理动态缺乏敏感性，或者依赖于在人工注释上训练的微调评估器，这些评估器过度拟合于特定数据集的线索，无法进行有效的泛化。一个关键挑战是，现有的监督来源提供的要么是相对排序，要么是绝对分数，但在不同设置下并不能可靠且一致地提供两者。为此，我们提出了PhyProbe，一种评估器，它从一个冻结的预训练时空编码器中提取特征，并通过一个轻量级评分头将这些特征映射到一个标量物理一致性违反分数。PhyProbe通过一个统一的目标进行训练，该目标结合了成对排名、对噪声标量注释的回归以及基于锚点的校准，使用一组精心挑选的异构监督来源。实验表明，PhyProbe在大多数成对基准测试中优于先前的方法，这些基准测试涵盖了真实生成和生成生成的对，在不同的对应关系下，尤其是在无对应和生成生成的设置中，PhyProbe的表现提升最大，而现有的微调评估器在这些情况下表现急剧下降。PhyProbe与人类判断之间的相关性很强，基于排名和线性指标之间的接近一致性表明，分数既有良好的排序，又稳定地锚定在[0, 1]的范围内。此外，尽管PhyProbe是在指示物理一致性的监督下训练的，但没有明确的整体偏好标签，PhyProbe在人工偏好基准测试中的表现也具有竞争力，这与物理违反与更广泛的质量下降相互纠缠的观察一致。
cs.CV / 12 / 2609.38391
Team MSU GenText-Forensics Challenge 2026 Technical Report
团队 MSU GenText-Forensics 挑战 2026 技术报告
Abstract
Document text forgery has evolved beyond simple pixel-level manipulation: modern attacks alter not only the appearance of a document but also its meaning, and increasingly target the OCR & LLM pipelines that consume such documents. The ACM MM 2026 GenText-Forensics challenge therefore requires systems that not only decide whether a multilingual text image is forged, but also localize the point of manipulation, identify the attack type, and produce a human-readable forensic report with supporting evidence. We present our solution, a decomposed chain-of-thought (CoT) pipeline that combines a document tampering detector (DTD) with two Qwen3-VL-32B vision-language models, each LoRA-adapted to a distinct sub-task. DTD produces tampering probability maps that are converted into numbered candidate regions; a first model (the Filterer) validates these regions and assigns a preliminary forgery type, while a second model (the Semantic Detective) merges and re-grounds the surviving regions, searches for purely semantic anomalies that are invisible to pixel-level detectors, and writes the final report. Both models are trained by distilling chain-of-thought traces from a privileged Qwen3-VL-235B teacher that has access to ground-truth masks and reports. Our approach secured third place in the ACM MM 2026 GenText-Forensics challenge. We describe the data preparation, test-time augmentation, region rendering, distillation protocol, and training configuration in detail, and report ablations over detector thresholds, prompt designs, and pipeline decompositions.
Chinese Translation
文档文本伪造已超越简单的像素级操作：现代攻击不仅改变文档的外观，还改变其含义，并越来越多地针对消费这些文档的 OCR 和 LLM 流程。因此，ACM MM 2026 GenText-Forensics 挑战要求系统不仅能够判断多语言文本图像是否伪造，还需定位操作点、识别攻击类型，并生成包含支持证据的人类可读的法医报告。我们提出的解决方案是一个分解的思维链（CoT）管道，它将文档篡改检测器（DTD）与两个 Qwen3-VL-32B 视觉语言模型相结合，每个模型经过 LoRA 调整以适应不同的子任务。DTD 生成的篡改概率图被转换为编号候选区域；第一个模型（过滤器）验证这些区域并分配初步的伪造类型，而第二个模型（语义侦探）合并并重新定位存活区域，搜索对像素级检测器不可见的纯语义异常，并撰写最终报告。两个模型的训练通过从一个特权的 Qwen3-VL-235B 教师中提炼思维链痕迹完成，该教师能够访问真实掩模和报告。我们的方法在 ACM MM 2026 GenText-Forensics 挑战中获得第三名。我们详细描述了数据准备、测试时增强、区域渲染、提炼协议和训练配置，并报告了关于检测器阈值、提示设计和管道分解的消融实验。
cs.CV / 13 / 2609.38413
VidHarness: Evolving Agent Harnesses for Cost-Efficient Long Video Understanding
VidHarness：为成本效益高的长视频理解演化的代理工具
Abstract
Vision-language models (VLMs) can answer questions about hour-long videos, but processing every frame is prohibitively expensive, even though the evidence for a question usually spans only a few seconds. Video agents, i.e., harness programs wrapped around a frozen VLM, address this by observing the video selectively, yet existing harnesses are hand-crafted by experts through slow build-and-test cycles. We propose VidHarness, a framework that automates harness design for cost-efficient long video understanding, in which a harness proposer iteratively evolves harnesses based on execution feedback from an evolution environment. To escape the local optima of greedy refinement, we organize the evolution as Monte Carlo tree search (MCTS), and to reduce the evaluation cost, we integrate uncertainty-aware multi-fidelity validation, which screens new harnesses on a few questions and promotes only the promising ones. Since the best harness varies with the frame budget, we further introduce a mixture-of-harness that routes each question to a harness specialized for its budget. VidHarness sets new state-of-the-art results on LongVideoBench, Video-MME, and Video-Holmes, outperforms the strongest hand-crafted video agent by up to $11.2$ points, and generalizes to the knowledge-intensive benchmarks Video-MMMU and MMVU with fewer than half of the frames of uniform sampling.
Chinese Translation
视觉-语言模型（VLMs）能够回答关于长达一小时视频的问题，但处理每一帧的成本极高，尽管一个问题的证据通常只持续几秒钟。视频代理，即围绕冻结的 VLM 包装的工具，通过选择性地观察视频来解决这一问题，然而现有的工具是由专家通过缓慢的构建和测试周期手工制作的。我们提出了 VidHarness，一个自动化工具设计框架，用于成本效益高的长视频理解，其中工具提议者基于来自演化环境的执行反馈迭代演化工具。为了逃避贪婪优化的局部最优，我们将演化组织为蒙特卡洛树搜索（MCTS），并为了降低评估成本，我们集成了不确定性感知的多保真度验证，该验证在少量问题上筛选新工具，仅推广有前景的工具。由于最佳工具因帧预算而异，我们进一步引入了一种混合工具，将每个问题路由到专门针对其预算的工具。VidHarness 在 LongVideoBench、Video-MME 和 Video-Holmes 上设定了新的最先进结果，超越了最强的手工制作视频代理，提升幅度高达 $11.2$ 分，并在知识密集型基准 Video-MMMU 和 MMVU 上以不到一半的均匀采样帧数实现了泛化。
cs.CV / 14 / 2609.38426
LoopVL: Recurrent Visual Intelligence
LoopVL：递归视觉智能
Abstract
We introduce LoopVL to study whether Loop Transformers can be effectively extended to vision- language models. LoopVL combines Module-Loop and Model-Loop computation to iteratively update a unified vision-language state through shared modules. We train LoopVL from scratch through language pre-training, multimodal training, and post-training. LoopVL outperforms a range of similarly sized and larger non-recurrent models on multimodal understanding and visual reasoning benchmarks. We also observe Visual Aha Moments in LoopVL, characterized by pronounced shifts in visual attention across loops. LoopVL provides practical evidence for recurrent vision-language modeling and offers an intuitive perspective on how shared parameters can support deeper multimodal computation over continuously evolving visual-language states.
Chinese Translation
我们提出了LoopVL，以研究Loop Transformers是否可以有效扩展到视觉-语言模型。LoopVL结合了模块循环（Module-Loop）和模型循环（Model-Loop）计算，通过共享模块迭代更新统一的视觉-语言状态。我们通过语言预训练、多模态训练和后期训练从零开始训练LoopVL。LoopVL在多模态理解和视觉推理基准测试中超越了一系列相似规模和更大规模的非递归模型。我们还观察到LoopVL中的视觉顿悟时刻（Visual Aha Moments），其特征是视觉注意力在循环之间发生显著变化。LoopVL为递归视觉-语言建模提供了实证支持，并提供了一个直观的视角，说明共享参数如何支持对不断演变的视觉-语言状态进行更深层次的多模态计算。
cs.CV / 15 / 2609.38428
MOBA-VL: Event-Localized Multi-Turn Reinforcement Learning for Real-Time MOBA Commentary
MOBA-VL：用于实时MOBA解说的事件定位多轮强化学习
Abstract
Real-time commentary for Multiplayer Online Battle Arena (MOBA) esports requires a vision-language model (VLM) to narrate a live match second by second, both fluently and accurately. Existing streaming VLMs sound natural but often miss key events such as kills and objectives. To address this limitation, we use game telemetry, which records exactly when each event occurs, as a supervision signal. We introduce MOBA-VL, a 9B-parameter model trained on this signal with event-localized multi-turn reinforcement learning, which rewards the turns that describe each event. We also collect MOBACast, 860 professional matches (about 460 hours) across three MOBA games with word-level timestamped commentary, and MOBACast-Bench, a benchmark from held-out tournaments. On MOBACast-Bench, MOBA-VL achieves the highest Overall score on full matches (63.25 vs. 55.12 for StreamingVLM) and clips (63.45 vs. 56.22 for DeepSeek-V4.1-Flash). Event-localized credit also raises event recall from 34.5 to 42.1 over supervised fine-tuning. Code and data will be released, and demos are available on an anonymous project page at https://moba-vl.github.io.
Chinese Translation
多人在线战斗竞技场（MOBA）电子竞技的实时解说需要一个视觉-语言模型（VLM）能够流畅且准确地逐秒叙述比赛。现有的流媒体VLM虽然听起来自然，但常常错过关键事件，如击杀和目标。为了解决这一局限性，我们使用游戏遥测数据，准确记录每个事件发生的时间，作为监督信号。我们提出了MOBA-VL，一个基于这一信号训练的9B参数模型，采用事件定位的多轮强化学习，奖励描述每个事件的回合。我们还收集了MOBACast，包含860场职业比赛（约460小时）的逐字时间戳解说，以及MOBACast-Bench，这是一个来自保留比赛的基准。在MOBACast-Bench上，MOBA-VL在完整比赛上的总体得分最高（63.25对比StreamingVLM的55.12），在片段上的得分也最高（63.45对比DeepSeek-V4.1-Flash的56.22）。事件定位的奖励也将事件召回率从34.5提高到42.1，超越了监督微调。代码和数据将会发布，演示可在匿名项目页面https://moba-vl.github.io上获取。
cs.CV / 16 / 2609.38444
Audible World Models: Spatially Aware Sound Generation for 3D Worlds
可听世界模型：面向三维世界的空间感知声音生成
Abstract
Text- and image-conditioned world generators can create visually rich 3D environments, yet these worlds often remain silent or rely on soundtracks synthesized solely from text or rendered video. Although such audio can convey what should be heard, it lacks an explicit representation of where sound sources are located and how their perceived sound should vary with listener movement. We introduce Audible World Models, a training-free framework that incorporates sound into the generated world state. Starting from a text prompt, our system constructs a panoramic 3D proxy, separates it into semantic layers, identifies sound-producing foreground objects and ambient background regions, and synthesizes dry audio for each sound label. It then anchors these sources to reconstructed geometry and renders listener-dependent spatial audio using geometric acoustic propagation. By explicitly linking semantics, geometry, and sound propagation, the framework maintains persistent source locations while adapting the rendered audio to changes in listener viewpoint and motion. Experiments across 80 generated scenes demonstrate substantial gains in spatial consistency over text-, video-, and panorama-conditioned baselines, while preserving competitive semantic alignment. VLM-based assessments and human evaluations further indicate that our soundtracks are preferred for their audio-visual consistency, spatial plausibility, and motion-dependent behavior.
Chinese Translation
文本和图像条件的世界生成器能够创建视觉丰富的三维环境，但这些世界往往保持沉默，或者仅依赖于从文本或渲染视频合成的音轨。尽管这样的音频可以传达应当听到的内容，但缺乏对声音源位置的明确表示，以及听者移动时感知声音变化的描述。我们提出了可听世界模型（Audible World Models），这是一个无需训练的框架，将声音融入生成的世界状态。从文本提示开始，我们的系统构建了一个全景三维代理，将其分离为语义层，识别出产生声音的前景物体和环境背景区域，并为每个声音标签合成干音频。然后，它将这些声音源锚定到重建的几何体上，并使用几何声学传播渲染依赖于听者的空间音频。通过明确链接语义、几何和声音传播，该框架保持了源位置的持续性，同时将渲染的音频适应于听者视角和运动的变化。在80个生成场景的实验中，显示出相较于文本、视频和全景条件基线在空间一致性方面的显著提升，同时保持了竞争性的语义对齐。基于VLM的评估和人类评估进一步表明，我们的音轨因其音频-视觉一致性、空间合理性和运动依赖行为而受到偏好。
cs.CV / 17 / 2609.38466
PAMI: Part Anchored Motion for Text to Human-Object Interaction Generation
PAMI：基于部件锚点的文本到人类-物体交互生成的运动框架
Abstract
Text-conditioned full-body human-object interaction (HOI) generation requires synthesizing human motion and object trajectories that match the input text while remaining precisely coordinated over time. Most methods represent the human and object as separate trajectories and predict the global human-object couplings. Learning this complex, dynamically changing relationship implicitly, however, often yields object drift, missed contact, and penetration. We introduce PAMI, a Part-Anchored Motion framework for Interaction generation. Inspired by the classic Hough Transform, our key idea is to localize object motion by letting body-part anchors vote for it: we express object motion relative to multiple body-part anchors and use PamiVAE to learn an interaction latent space, decoding frame-wise weights that aggregate these part-specific votes. Building on this representation, PAMI generates interactions in a coarse-to-fine hierarchy. PamiGen first generates a coarse human-object interaction from text in this structured latent space, and PamiRefiner then recursively resolves fine-grained contact geometry using a hybrid surface-sensing representation, combining long-range probes that capture overall body-part influence with short-range sensors that resolve detailed contacts near the object surface. Experiments on InterAct show that PAMI generates more faithful interactions and more accurate human-relative object motion than previous methods, achieving 14.5% higher contact recall than the previous state of the art. Extensive ablations validate the contributions of both the part-anchored voting representation and hybrid surface-sensing refinement.
Chinese Translation
文本条件下的全身人类-物体交互（HOI）生成需要合成与输入文本匹配的人类运动和物体轨迹，同时保持时间上的精确协调。然而，大多数方法将人类和物体表示为独立的轨迹，并预测全局的人类-物体耦合。隐式学习这种复杂且动态变化的关系，往往会导致物体漂移、错过接触和穿透。我们提出了PAMI，一个用于交互生成的基于部件锚点的运动框架。受到经典霍夫变换的启发，我们的关键思想是通过让身体部件锚点为物体运动投票来定位物体运动：我们相对于多个身体部件锚点表达物体运动，并使用PamiVAE学习交互潜在空间，解码逐帧权重以聚合这些特定部件的投票。在此表示的基础上，PAMI以粗到细的层次生成交互。PamiGen首先在这个结构化的潜在空间中从文本生成粗略的人类-物体交互，然后PamiRefiner递归地使用混合表面感知表示解析细粒度接触几何，结合捕捉整体身体部件影响的长程探针与解析靠近物体表面的详细接触的短程传感器。在InterAct上的实验表明，PAMI生成的交互更真实，且相较于之前的方法，生成的人类相对物体运动更为准确，接触召回率比之前的最先进方法高出14.5%。大量消融实验验证了部件锚点投票表示和混合表面感知精炼的贡献。
cs.CV / 18 / 2609.38476
Curating Synthetic Data for Task-Specific Visual Perception
为特定任务的视觉感知策划合成数据
Abstract
Synthetic data are most valuable where general-purpose datasets cannot provide the domain-specific priors a task requires, and where manual annotation is expensive, imprecise, or infeasible. In this article we argue that the central question for specialized vision systems is not how to generate more data, but which data to generate. We therefore discuss curated synthetic data, whose scene content, appearance variations, sensing characteristics, and annotations are deliberately designed around a given task. We examine three complementary curation paradigms. Procedural rendering offers explicit control over scene parameters and the annotations follow by construction. Physically-based simulation encodes the mechanism behind an observed effect and yields exactly aligned supervision pairs. Generative AI learns sensor-specific appearance from small real seed sets and attains plausible realism, though it remains prone to hallucination and to inaccurate annotation. These paradigms are illustrated with examples from industrial surface defect detection, restoration of degraded digitized autochrome plates, and 6DoF pose estimation from RGB and event data. Using these examples, we analyze different data regimes and training strategies that combine synthetic and real data across these paradigms. We conclude that curated synthetic data are best understood as a complement to real observations, and that hybrid pipelines combining controllable supervision with learned appearance are the most promising direction for reliable sim-to-real transfer.
Chinese Translation
合成数据在通用数据集无法提供任务所需的领域特定先验知识，以及人工标注成本高、精度低或不可行的情况下最具价值。本文认为，专用视觉系统的核心问题不是如何生成更多数据，而是生成哪些数据。因此，我们讨论了策划的合成数据，其场景内容、外观变化、传感特性和注释都是围绕特定任务精心设计的。我们考察了三种互补的策划范式。过程渲染提供了对场景参数的明确控制，注释也是由此构建而成。基于物理的仿真编码了观察到的效应背后的机制，并产生完全对齐的监督对。生成性人工智能从小规模的真实种子集学习传感器特定的外观，并实现了合理的真实感，尽管它仍然容易出现幻觉和不准确的注释。这些范式通过工业表面缺陷检测、退化数字化自彩色胶卷的修复以及基于RGB和事件数据的6DoF姿态估计的实例进行了说明。通过这些例子，我们分析了不同的数据模式和训练策略，这些策略结合了这些范式中的合成数据和真实数据。我们得出结论，策划的合成数据最好被理解为对真实观察的补充，而结合可控监督与学习外观的混合管道是实现可靠的模拟到真实转移的最有前景的方向。
cs.CV / 19 / 2609.38479
Caption-Mediated Perceived-Safety Estimation for Pedestrian Routing
基于标题的行人路线感知安全估计
Abstract
This paper presents an explainable approach to pedestrian routing, in which perceived safety is estimated from street-level imagery through an explicit natural-language intermediate representation. A vision--language model caption is generated and stored before any scoring is undertaken, and the perceived-risk class is derived entirely from structured features of that stored text, so that every segment score remains inspectable by the user. Nine captioning conditions across five model families are benchmarked against a direct Contrastive Language--Image Pre-training (CLIP) image-embedding baseline under an identical downstream pipeline, and the caption-mediated representation is found to reach parity with the image embedding rather than to trail it. The approach was deployed over 654,115 images covering 36 electoral wards in two locations in Northern England (Manchester and Huddersfield). Independent field validation against 3,669 locally collected ratings of 494 images across 70 participant sessions established agreement that is statistically significant but modest, at $r=0.262$, against a measured noise ceiling of 0.737 imposed by disagreement between raters. A single-use confirmatory test then found that a pipeline 44\% stronger on the supervised benchmark did not produce measurable improvement in the field ($r=0.250$, $p=0.84$), so the benchmark gains did not predict the deployment gains in this case. Routing behaviour varies systematically with journey length. There is negligible change below 1\,km, reaching a median increase of 12.78\% in low-risk route length for a median detour of 2.73\% on journeys of 3 to 6 km.
Chinese Translation
本文提出了一种可解释的行人路线规划方法，该方法通过显式的自然语言中介表示从街景图像中估计感知安全性。在进行任何评分之前，生成并存储视觉-语言模型的标题，感知风险类别完全源自于该存储文本的结构化特征，因此每个段落的得分都可以由用户进行检查。我们对五个模型家族中的九种标题条件进行了基准测试，比较了在相同下游流程下与直接对比语言-图像预训练（Contrastive Language--Image Pre-training，CLIP）图像嵌入基线的表现，发现基于标题的表示达到了与图像嵌入相当的水平，而不是落后于它。该方法在覆盖英格兰北部两个地点（曼彻斯特和哈德斯菲尔德）36个选区的654,115张图像上进行了部署。对494张图像在70个参与者会话中收集的3,669条本地评分进行了独立现场验证，结果显示统计一致性显著但适度，相关系数为$r=0.262$，而由于评分者之间的不一致性，测得的噪声上限为0.737。随后进行的一次单次确认测试发现，在监督基准上强44 ext{%}的流程在现场并未产生可测量的改善（$r=0.250$，$p=0.84$），因此基准增益并未预测出本案例中的部署增益。路线行为与旅程长度呈系统性变化。在1公里以下几乎没有变化，在3到6公里的旅程中，低风险路线长度的中位数增加为12.78 ext{%}，而中位数绕行率为2.73 ext{%}。
cs.CV / 20 / 2609.38485
Beyond Layers: Position-Resolved Gradient Conflict and Position-Aware Modulation for Unified Multimodal Models
超越层次：位置分辨的梯度冲突与位置感知调制在统一多模态模型中的应用
Abstract
Unified multimodal models (UMMs) train image understanding and autoregressive image generation on shared parameters, and the two objectives are known to interfere. Existing diagnoses and remedies operate at the resolution of layers or experts, measuring conflict per layer and resolving it by separating parameters. We argue that this resolution hides an orthogonal axis. Generation in a UMM is next-token prediction over a raster sequence of visual tokens whose roles vary systematically with position, so how strongly a generation gradient interferes with understanding should depend on where in the sequence it originates. We introduce a position-resolved interference map that attributes understanding-generation gradient conflict to visual-token positions within every layer, computed from a single backward pass at $1.2\times$ the cost of a standard backward pass. On Show-o and Janus-Pro, position explains a large share of conflict variance after controlling for depth (partial $\eta^2=0.31$ vs. $0.35$ for layer on Show-o; $0.15$ vs. $0.30$ on Janus-Pro): the first quarter of the sequence has a mean gradient cosine of $-0.18$ against understanding, the last quarter $-0.02$. The dependence survives per-position gradient-norm normalization, retaining $80%$ of its effect size, and conflict strength tracks semantic content (Spearman $\rho=0.64$). Building on the map, we propose position-aware modulation (PAM), which removes the anti-aligned component of generation gradients only at high-conflict positions without changing the architecture. Under a matched trainable-parameter budget, PAM improves over layer-wise separation by $+21$ MME and $+2.4$ GenEval points on Show-o while matching it on POPE and overall FID; a random-position control recovers about $31%$ of the gain. Position-based and layer-based separation are complementary degrees of freedom and can be combined.
Chinese Translation
统一多模态模型（UMMs）在共享参数上训练图像理解和自回归图像生成，这两个目标已知会相互干扰。现有的诊断和解决方案在层次或专家的分辨率上进行，测量每层的冲突并通过分离参数来解决。我们认为这种解决方案掩盖了一个正交的轴。在UMM中，生成是对视觉标记的光栅序列进行下一个标记预测，其角色随着位置的变化而系统性变化，因此生成梯度干扰理解的强度应取决于其在序列中的起源位置。我们引入了一个位置分辨的干扰图，该图将理解-生成梯度冲突归因于每层内的视觉标记位置，计算成本为标准反向传播的$1.2 imes$。在Show-o和Janus-Pro上，控制深度后，位置解释了冲突方差的很大一部分（部分$ ext{η}^2=0.31$对比Show-o的$0.35$；Janus-Pro上为$0.15$对比$0.30$）：序列的前四分之一平均梯度余弦为$-0.18$，而最后四分之一为$-0.02$。这种依赖在每位置的梯度范数归一化后仍然存在，保留了$80 ext{%}$的效应大小，冲突强度与语义内容相关（Spearman $
ho=0.64$）。基于该图，我们提出了位置感知调制（PAM），该方法仅在高冲突位置去除生成梯度的反对齐成分，而不改变架构。在匹配的可训练参数预算下，PAM在Show-o上比层级分离提高了$+21$ MME和$+2.4$ GenEval分数，同时在POPE和整体FID上与其持平；随机位置控制恢复了约$31 ext{%}$的增益。基于位置的分离和基于层的分离是互补的自由度，可以结合使用。
cs.CV / 21 / 2609.38487
Multidimensional Observer Model and Perceptual Dimensions of Human Image Quality Assessment
多维观察者模型及人类图像质量评估的感知维度
Abstract
Judging image quality is not only ecologically relevant to everyday human tasks, but also underpins many machine vision tasks such as image generation. This paper proposes a framework to understand the inherent perceptual space underlying image quality judgment in humans. We propose a multi-dimensional observer model that represents images as distributions in a latent perceptual space and that models human judgment as comparing noisy samples. Being constrained by neural representations in the primate ventral stream and fit to large-scale behavioral data, the model enables analysis of perceptual structure while matching the predictive power of existing metrics. Using this model, we find that the perceptual spaces needed to account for image quality judgment in humans are extremely low-dimensional compared to the image space even when considering its sparsity. The exact structure of the space (e.g., dimensionalities, information encoded) varies between low-level and high-level quality judgments, suggesting that, despite a shared retinal encoding in the beginning, humans selectively construct task-dependent perceptual spaces in visual decision making.
Chinese Translation
判断图像质量不仅与日常人类任务生态相关，还支撑着许多机器视觉任务，如图像生成。本文提出了一个框架，以理解人类图像质量判断背后的固有感知空间。我们提出了一种多维观察者模型，该模型将图像表示为潜在感知空间中的分布，并将人类判断建模为比较噪声样本。该模型受到灵长类动物腹侧通路中的神经表征的限制，并与大规模行为数据相适应，使得在匹配现有指标的预测能力的同时，能够分析感知结构。使用该模型，我们发现，即使考虑到图像空间的稀疏性，人类在图像质量判断中所需的感知空间与图像空间相比是极其低维的。该空间的确切结构（例如，维度、编码信息）在低级和高级质量判断之间有所不同，这表明尽管在开始时存在共享的视网膜编码，人类在视觉决策中会选择性地构建依赖于任务的感知空间。
cs.CV / 22 / 2609.38519
GazeFlow: From Human Gaze Behavior to Generative Egocentric Gaze Prediction
GazeFlow：从人类注视行为到生成的自我中心注视预测
Abstract
Egocentric gaze prediction enables many downstream applications but remains challenging, as human gaze is inherently stochastic. This stochasticity is constrained by structured temporal dynamics alternating between fixations and saccades, top-down influences from tasks, and bottom-up visual saliency. Based on this observation, we introduce GazeFlow, a framework that directly models gaze as a joint distribution of temporal gaze positions conditioned upon both top-down and bottom-up information. In particular, GazeFlow uses conditional flow matching (CFM): a learned velocity field iteratively transports a Gaussian noise sample into a plausible gaze trajectory drawn from this joint distribution. The velocity field is conditioned on bottom-up visual features extracted by a video encoder and on top-down task information obtained by globally querying these features. On standard datasets, GazeFlow achieves state-of-the-art performance on per-frame metrics, and the generated trajectories align better with human gaze temporal dynamics.
Chinese Translation
自我中心注视预测能够支持许多下游应用，但仍然具有挑战性，因为人类的注视本质上是随机的。这种随机性受到结构化时间动态的约束，这些动态在注视和扫视之间交替，以及来自任务的自上而下影响和自下而上的视觉显著性。基于这一观察，我们引入了GazeFlow，一个直接将注视建模为条件于自上而下和自下而上信息的时间注视位置的联合分布的框架。特别地，GazeFlow使用条件流匹配（Conditional Flow Matching, CFM）：一个学习的速度场迭代地将高斯噪声样本转化为从该联合分布中抽取的合理注视轨迹。速度场是基于通过视频编码器提取的自下而上的视觉特征和通过全局查询这些特征获得的自上而下任务信息进行条件化的。在标准数据集上，GazeFlow在每帧指标上达到了最先进的性能，生成的轨迹与人类注视的时间动态更为一致。
cs.CV / 23 / 2609.38541
ThinkV2V: Unleashing the Reasoning Capability of MLLMs for Instruction-Guided Video Editing
ThinkV2V：释放多模态大语言模型在指令引导视频编辑中的推理能力
Abstract
Instruction-guided video editing has made significant progress, yet existing methods use multimodal large language models (MLLMs) primarily as semantic encoders, so they often fall short in working with implicit edits that require causal or semantic reasoning. To bridge this fundamental gap in video editing, we propose ThinkV2V, a reasoning-driven framework for complex instruction-guided video editing, explicitly activating MLLM thinking before visual generation. At its core, ThinkV2V builds on a practical MLLM-to-DiT architecture to turn explicit thinking over the source video and instruction into refined conditioning signals for video editing. Further, we equip it with a dedicated training and inference recipe, combining Progressive Curriculum Training, which gradually cultivates the model from basic editing to reasoning-intensive cases, with Inference-Time Thinking Scaling, which iteratively refines candidate prompts and selects the most reliable one, to better elicit reasoning in challenging editing scenarios. We also curate the ThinkV2V-150K dataset and introduce ThinkV2V-Bench to support training and evaluation of video editing with implicit intent and causal reasoning. Experimental results demonstrate the state-of-the-art performance of ThinkV2V on both complex and standard editing scenarios, in which our 5B-scale DiT model substantially outperforms larger 10B-scale baselines.
Chinese Translation
指令引导的视频编辑已经取得了显著进展，但现有方法主要将多模态大语言模型（MLLMs）用作语义编码器，因此在处理需要因果或语义推理的隐式编辑时往往显得不足。为了解决这一视频编辑中的基本问题，我们提出了ThinkV2V，一个以推理为驱动的复杂指令引导视频编辑框架，在视觉生成之前明确激活MLLM的思维。ThinkV2V的核心基于实用的MLLM到DiT架构，将对源视频和指令的显式思考转化为精细的条件信号，以用于视频编辑。此外，我们为其配备了专门的训练和推理方案，结合渐进式课程训练（Progressive Curriculum Training），逐步培养模型从基本编辑到推理密集型案例的能力，以及推理时思维扩展（Inference-Time Thinking Scaling），迭代优化候选提示并选择最可靠的一个，以更好地引导在挑战性编辑场景中的推理。我们还整理了ThinkV2V-150K数据集，并引入ThinkV2V-Bench，以支持隐式意图和因果推理的视频编辑的训练和评估。实验结果表明，ThinkV2V在复杂和标准编辑场景中均表现出最先进的性能，其中我们的5B规模DiT模型显著优于更大的10B规模基线。
cs.CV / 24 / 2609.38560
Detail in Context: A Dual-Scale Machine Learning Framework for Mycosis Fungoides Detection
情境中的细节：用于真菌性皮肤病检测的双尺度机器学习框架
Abstract
Mycosis fungoides (MF) is a rare form of cutaneous T-cell lymphoma that is often misdiagnosed in early stages due to its visual similarity to benign inflammatory dermatoses. Early and accurate diagnosis is critical for improving patient outcomes. In this paper, we propose a comprehensive diagnostic framework for automated MF detection that combines dual- scale histopathological image analysis with deep learning. To distinguish MF from other lymphoproliferative skin conditions, the proposed approach leverages a late-fusion ensemble of dual- magnification (10x and 20x) convolutional neural networks (CNNs), complemented by a random forest classifier trained on 16 clinical features. Experimental results on an expanded dataset of 6,267 images (4,306 MF; 1,961 Non-MF) across 463 patients demonstrate that strong detection performance is obtained by prioritizing higher-resolution cytological details (20x) within broader architectural context (10x). The image-based late-fusion model achieves an accuracy of 83.58% and a sensitivity of 89.13%, while the clinical random forest model achieves an accuracy of 96.6% and sensitivity of 93.8%, highlighting the po- tential of this multimodal framework as a robust clinical decision support system in dermatology. This framework addresses two distinct clinical objectives: an image-based dual-scale pipeline optimized for the early diagnostic screening of MF versus non- MF dermatoses, and a complementary clinical metadata model designed for the subsequent staging of confirmed MF cases (patch/plaque versus tumor)
Chinese Translation
真菌性皮肤病（Mycosis fungoides, MF）是一种罕见的皮肤T细胞淋巴瘤，因其在早期阶段与良性炎症性皮肤病的视觉相似性而常常被误诊。早期和准确的诊断对于改善患者预后至关重要。本文提出了一种综合诊断框架，用于自动化MF检测，该框架结合了双尺度组织病理图像分析和深度学习。为了将MF与其他淋巴增生性皮肤病区分开，所提出的方法利用了双放大倍数（10倍和20倍）卷积神经网络（CNNs）的后融合集成，并辅以基于16个临床特征训练的随机森林分类器。在463名患者的扩展数据集中进行的实验结果显示，优先考虑更高分辨率的细胞学细节（20倍）与更广泛的结构背景（10倍）相结合，可以获得强大的检测性能。基于图像的后融合模型实现了83.58%的准确率和89.13%的灵敏度，而临床随机森林模型则实现了96.6%的准确率和93.8%的灵敏度，突显了该多模态框架作为皮肤科强大临床决策支持系统的潜力。该框架解决了两个不同的临床目标：一个是针对MF与非MF皮肤病的早期诊断筛查优化的基于图像的双尺度流程，另一个是为已确认的MF病例（斑块/斑点与肿瘤）后续分期设计的补充临床元数据模型。
cs.CV / 25 / 2609.38562
LongTake: Learning to Sustain Dynamics in Long-Horizon Video Generation
LongTake：学习在长时间视频生成中维持动态
Abstract
World models, game simulators, and long-take video creation require coherent scene evolution and sustained dynamics over extended durations. Autoregressive (AR) video diffusion provides a natural framework for long-horizon generation, yet extended rollouts often become near-static or lose visual quality. We hypothesize that these failures reflect the limited guidance provided by short-video supervision on how ongoing scene dynamics develops over longer durations. This motivates us to introduce LongTake, a two-stage training pipeline built around Long-Horizon Teacher Forcing (TF) on curated real long videos. Long-Horizon TF trains the AR model to predict later frames conditioned on long ground-truth video prefixes, extending direct supervision beyond the short training horizon. This supervision is designed to help the model sustain dynamics and preserve visual quality during long-horizon generation. Our central finding is that this training stage strengthens direct initialization for distribution matching distillation (DMD) under student self-rollout, without the intermediate few-step distillation stage used in standard pipelines. Under the same five-second DMD training setup, our initialization yields substantially higher dynamic degree than short horizon TF initialization on 30-second rollouts at comparable aesthetic quality, and surpasses the evaluated baselines in both measures. Hybrid DMD further reuses this teacher to extend supervision to later frames of the self-rollout while retaining bidirectional joint supervision over the initial window. On long-horizon self-rollouts, LongTake lies on the Pareto front of dynamic degree and aesthetic quality, and Hybrid DMD attains the highest dynamic degree among evaluated methods at both 30s and 60s.
Chinese Translation
世界模型、游戏模拟器和长镜头视频创作需要在较长时间内实现连贯的场景演变和持续的动态。自回归（AR）视频扩散为长时间生成提供了一个自然的框架，但延长的展开往往变得近乎静态或失去视觉质量。我们假设这些失败反映了短视频监督在指导持续场景动态如何在更长时间内发展方面的局限性。这促使我们引入LongTake，一个围绕长时间教师强制（Long-Horizon Teacher Forcing, TF）构建的两阶段训练流程，基于精心挑选的真实长视频。长时间TF训练AR模型以预测基于长真实视频前缀的后续帧，从而将直接监督扩展到短训练时间范围之外。这种监督旨在帮助模型在长时间生成过程中维持动态并保持视觉质量。我们的核心发现是，这一训练阶段增强了在学生自我展开下进行分布匹配蒸馏（Distribution Matching Distillation, DMD）的直接初始化，而无需标准流程中使用的中间几步蒸馏阶段。在相同的五秒DMD训练设置下，我们的初始化在30秒展开中提供了显著更高的动态程度，且在可比的美学质量下超越了评估的基线。混合DMD进一步利用这一教师将监督扩展到自我展开的后续帧，同时保持对初始窗口的双向联合监督。在长时间自我展开中，LongTake在动态程度和美学质量的帕累托前沿上，且混合DMD在评估方法中在30秒和60秒的动态程度上均达到了最高。
cs.CV / 26 / 2609.38578
Retargeting Motions to Diverse Skeletons via Learnable Flattening
通过可学习的扁平化技术将动作重定向到多样化的骨架
Abstract
Cross-structural motion retargeting aims to transfer motion between different skeletal topologies. Despite recent progress, existing state-of-the-art models struggle with reliability in zero-shot settings, i.e. skeletons with different topologies which were unseen during training, and recent Transformer-based attempts have failed to outperform specialized geometric methods. We bridge this gap with a Transformer Autoencoder that learns a topology- and translation-invariant latent space. Our core contribution is a learnable flattening of skeletal graphs that captures both local dependencies and global structure. Unlike the standard transformer architecture, which adds positional information to token content, we integrate graph-based positional encodings multiplicatively, a design choice that follows directly from our flattening formulation. The resulting model handles diverse skeletal topologies within a single unified architecture and trains in a fully unsupervised manner, requiring no paired retargeting data. Ablation studies show, that the graph encodings, multiplicative formulation, and Transformer backbone is critical for the performance. In zero-shot evaluations, our method reduces global joint position error by $43-47\%$ over current benchmarks. A user study ($n = 37$), including expert animators, further ranks our approach highest in motion alignment and physical plausibility ($p < 0.05$). These results demonstrate that our model design is key to making transformer architectures effective for motion retargeting, outperforming existing approaches.
Chinese Translation
跨结构的动作重定向旨在在不同的骨架拓扑之间转移动作。尽管近期取得了一些进展，现有的最先进模型在零样本设置下的可靠性仍然存在问题，即在训练过程中未见过的不同拓扑的骨架，而最近基于Transformer的尝试未能超越专门的几何方法。我们通过一个Transformer自编码器来弥补这一差距，该自编码器学习一个拓扑和转换不变的潜在空间。我们的核心贡献是对骨架图的可学习扁平化，该扁平化捕捉了局部依赖关系和全局结构。与标准的Transformer架构不同，后者将位置信息添加到标记内容中，我们通过乘法方式整合基于图的位置信息，这一设计选择直接源于我们的扁平化公式。最终模型在单一统一架构内处理多样化的骨架拓扑，并以完全无监督的方式进行训练，无需配对的重定向数据。消融研究表明，图编码、乘法公式和Transformer主干对性能至关重要。在零样本评估中，我们的方法在当前基准上将全局关节位置误差降低了43-47%。一项用户研究（n = 37），包括专家动画师，进一步将我们的方法在动作对齐和物理合理性方面排名最高（p < 0.05）。这些结果表明，我们的模型设计是使Transformer架构在动作重定向中有效的关键，超越了现有的方法。
cs.CV / 27 / 2609.38591
Restoring without Forgetting: Filter-Level Continual Image Restoration via Parameter-Space Integrated Gradients
无遗忘恢复：基于参数空间集成梯度的滤波器级持续图像恢复
Abstract
Adapting image restoration models to a stream of new tasks without revisiting past data remains challenging due to catastrophic forgetting. In this work, we propose Restoring without Forgetting (RwF), a filter-level continual adaptation framework for image restoration built upon a critical observation: task-specific knowledge is centered in a small subset of filters and can be separated from those reconstructing general content. RwF first performs parameter-space integrated gradients attribution to localize degradation-critical filters in a coarse-to-fine manner. It then adapts to new tasks by generating task-specific filters from a filter bank using compact factorized low-rank transformations, further augmented with cross-task attention and prototypical contrastive learning, and lastly assembles them back only at localized positions. Experiments on six restoration tasks show that RwF effectively avoids forgetting and achieves competitive restoration quality against all-in-one methods that have full data access, and outperforms LoRA-style adaptation with $\sim$10$\times$ fewer additional parameters. Code is available at https://github.com/funkdub/Restoring-without-Forgetting.
Chinese Translation
在不重新访问过去数据的情况下，将图像恢复模型适应于一系列新任务仍然面临灾难性遗忘的挑战。在本研究中，我们提出了无遗忘恢复（RwF），这是一种基于滤波器级的持续适应框架，旨在图像恢复领域，建立在一个关键观察之上：任务特定知识集中在少量滤波器中，并且可以与重构一般内容的滤波器分离。RwF首先通过参数空间集成梯度归因，以粗到细的方式定位对退化至关重要的滤波器。然后，它通过使用紧凑的分解低秩变换从滤波器库生成任务特定滤波器来适应新任务，进一步增强了跨任务注意力和原型对比学习，最后仅在局部位置将它们重新组装。对六个恢复任务的实验表明，RwF有效地避免了遗忘，并在恢复质量上与具有完整数据访问的全能方法相媲美，同时在额外参数数量上以约10倍的优势超越了LoRA风格的适应。代码可在 https://github.com/funkdub/Restoring-without-Forgetting 获取。
cs.CV / 28 / 2609.38592
StereoGaussians: Feed-Forward 3D Gaussian Splatting from Stereo Images
StereoGaussians：来自立体图像的前馈3D高斯溅射
Abstract
Feed-forward 3D Gaussian Splatting (3DGS) enables reconstruction without per- scene optimisation, but practical stereo-camera applications require nearby-view extrapolation beyond the input views. Stereo depth anchors visible surfaces, yet rendering newly exposed regions also requires learned appearance and additional scene capacity. We introduce StereoGaussians, which predicts a metric 3DGS representation from a single calibrated stereo pair. It reuses intermediate repre- sentations from frozen pretrained stereo networks to predict Gaussian attributes, while calibrated disparity anchors the geometry. A second Gaussian layer and an expanded image canvas provide capacity for disoccluded and outside-field-of- view content. For training, we construct SceneSplat-Stereo from quality-filtered 3DGS teachers, pairing stereo inputs with nearby target views across 803 training scenes. Experiments on unseen real and photorealistic stereo benchmarks demon- strate improvements over strong view-synthesis baselines, while ablation studies support our main design choices.
Chinese Translation
前馈3D高斯溅射（3DGS）能够在不进行每场景优化的情况下进行重建，但实际的立体相机应用需要超出输入视图的邻近视图外推。立体深度锚定可见表面，但渲染新暴露区域还需要学习的外观和额外的场景容量。我们提出了StereoGaussians，它从单个校准的立体对中预测度量3DGS表示。它重用来自冻结的预训练立体网络的中间表示来预测高斯属性，同时校准的视差锚定几何形状。第二个高斯层和扩展的图像画布为被遮挡和视野外的内容提供了容量。为了训练，我们从质量过滤的3DGS教师构建了SceneSplat-Stereo，将立体输入与803个训练场景中的邻近目标视图配对。对未见过的真实和照片真实立体基准的实验表明，相较于强大的视图合成基线有显著改善，而消融研究支持我们的主要设计选择。
cs.CV / 29 / 2609.38597
PixelUMM: Encoder-Free Unified Image and Video Understanding and Generation
PixelUMM：无编码器的统一图像与视频理解与生成
Abstract
Unified Multimodal Models (UMMs) often rely on separate visual representations for understanding and generation, increasing visual context length and complicating integration with established vision-language pretraining pipelines. Recent advances in pixel-space modeling offer an encoder-free alternative, but extending this paradigm from images to videos is non-trivial: video understanding and generation adopt different temporal representations, leaving the design of a unified visual interface an open question. We present PixelUMM, an encoder-free model for unified image and video understanding and generation directly in pixel space. PixelUMM represents images as spatial patches and videos as spatiotemporal tubelets, connecting raw pixels to a shared multimodal backbone through single-layer linear projections. Its Mixture-of-Transformers architecture combines shared attention with task-specific parameters and extends clean-pixel prediction to video generation, jointly supporting autoregressive text prediction and pixel-space flow matching. Experiments show that PixelUMM achieves competitive performance across image and video understanding and generation tasks. We further conduct empirical studies of key design choices, including decoder design and spatial-temporal patch size, providing insights for future pixel-space unified multimodal models.
Chinese Translation
统一多模态模型（UMMs）通常依赖于独立的视觉表示进行理解和生成，这增加了视觉上下文的长度，并使得与现有的视觉-语言预训练管道的整合变得复杂。最近在像素空间建模方面的进展提供了一种无编码器的替代方案，但将这一范式从图像扩展到视频并非易事：视频理解和生成采用不同的时间表示，如何设计一个统一的视觉接口仍然是一个悬而未决的问题。我们提出了PixelUMM，这是一种在像素空间中直接进行统一图像与视频理解与生成的无编码器模型。PixelUMM将图像表示为空间补丁，将视频表示为时空小管，利用单层线性投影将原始像素连接到共享的多模态骨干网络。其混合变换器（Mixture-of-Transformers）架构结合了共享注意力与任务特定参数，并将干净像素预测扩展到视频生成，联合支持自回归文本预测和像素空间流匹配。实验表明，PixelUMM在图像与视频理解和生成任务中实现了具有竞争力的性能。我们进一步对关键设计选择进行了实证研究，包括解码器设计和时空补丁大小，为未来的像素空间统一多模态模型提供了见解。
cs.CV / 30 / 2609.38603
Aperture: Training-Free Multiscale Concept Bottlenecks for Remote Sensing
Aperture：无训练多尺度概念瓶颈在遥感中的应用
Abstract
While earth observation models have advanced substantially, they still lack interpretability. While concept-bottleneck models provide interpretability and expert interaction, they are either too expensive to train for the remote sensing domain or perform poorly without annotation. We posit that in expert domains like remote sensing, such training-free models require both fine details in both image and concept space. In image space, we propose a multiscale concept bottleneck using greedy quadtree routing to locate small concepts. In concept space, we replace contrastive vision language models with pre-trained MLLMs and present a way to get reliable concept scores from them. We introduce APERTURE that blends concept scores at the global image and native concept-scale level to give state-ofthe-art training-free model performance. To test these models, introduce SiFC, a fine-grained concept-centric dataset across three countries, with human-reviewed class-level concept maps. On SiFC, APERTURE outperforms the best training-free baselines by more than 10 percentage points in macro F1-score, and notably also outperforms supervised concept bottleneck models. Targeted component-removal tests examine whether concept scores respond to changes in visual evidence, while temporal experiments show that descriptor updates improve recognition of technological changes without retraining.
Chinese Translation
尽管地球观测模型已经取得了显著进展，但它们仍然缺乏可解释性。概念瓶颈模型提供了可解释性和专家交互，但在遥感领域，它们要么训练成本过高，要么在没有注释的情况下表现不佳。我们认为，在遥感等专家领域，这类无训练模型需要在图像和概念空间中都具备细致的细节。在图像空间中，我们提出了一种使用贪婪四叉树路由的小概念多尺度瓶颈模型，以定位小概念。在概念空间中，我们用预训练的多模态语言模型（MLLMs）替代对比视觉语言模型，并提出了一种从中获取可靠概念分数的方法。我们引入了APERTURE，它在全局图像和本地概念尺度层面融合概念分数，从而实现了最先进的无训练模型性能。为了测试这些模型，我们引入了SiFC，这是一个跨越三个国家的细粒度概念中心数据集，包含经过人工审核的类别级概念图。在SiFC上，APERTURE在宏观F1分数上超越了最佳无训练基线超过10个百分点，并且显著优于监督概念瓶颈模型。针对组件移除的测试考察了概念分数是否对视觉证据的变化做出反应，而时间实验则表明，描述符更新在不重新训练的情况下改善了对技术变革的识别。
cs.CV / 31 / 2609.38607
After a Decade: Bringing Shadow Removal into the Real World with Agentic Training Data
十年之后：通过自主训练数据将阴影去除技术带入现实世界
Abstract
Shadow removal looks nearly solved on established benchmarks, yet remains brittle in the real world. Models have advanced; the paired training data they rely on have barely changed in nearly a decade. The reason is simple: obtaining a shadow-free target requires removing the occluder while keeping the scene, camera, and illumination otherwise unchanged, making diverse paired data difficult to capture. Meanwhile, large shadow detection datasets already contain diverse real-world images and masks, but no shadow-free targets. To turn this abundant but incomplete data into paired supervision, we propose an offline agentic workflow combining physics-motivated generation, failure detection, feedback-driven retry, candidate selection, and deterministic correction. Using this workflow, we construct AgenticShadow, a dataset of 17,138 image-mask-target triplets spanning general scenes, faces, and remote sensing. Our construction workflow reduces Color Distribution Difference by 50.5% over previous shadow removal work, while training existing shadow removal models on AgenticShadow reduces cross-domain LAB RMSE by 19.7-37.5%.
Chinese Translation
阴影去除在已有基准测试中看似几乎解决，但在现实世界中仍然脆弱。模型已经取得了进展；然而，它们所依赖的配对训练数据在近十年间几乎没有变化。原因很简单：获得无阴影的目标需要在保持场景、相机和照明条件不变的情况下去除遮挡物，这使得捕获多样化的配对数据变得困难。同时，大型阴影检测数据集已经包含了多样的真实世界图像和掩膜，但没有无阴影的目标。为了将这丰富但不完整的数据转化为配对监督，我们提出了一种离线自主工作流程，结合了基于物理的生成、失败检测、反馈驱动的重试、候选选择和确定性修正。利用该工作流程，我们构建了AgenticShadow，一个包含17,138个图像-掩膜-目标三元组的数据集，涵盖了一般场景、面部和遥感。我们的构建工作流程将颜色分布差异减少了50.5%，相比于之前的阴影去除工作，而在AgenticShadow上训练现有的阴影去除模型则将跨域LAB均方根误差减少了19.7-37.5%。
cs.CV / 32 / 2609.38615
Exo2EgoHOI: Hand-Object-Interaction Aware Exocentric-to-Egocentric Video Generation
Exo2EgoHOI：手-物体交互感知的外部视角到自我视角视频生成
Abstract
Egocentric videos of human manipulation provide valuable visual experience for embodied intelligence, yet collecting such data at scale is costly. Exocentric-to-egocentric video generation offers a scalable alternative by transforming abundant third-person manipulation videos into first-person observations. However, existing methods often struggle to faithfully preserve demonstrated hand-object interactions (HOI) across large viewpoint changes due to insufficient fine-grained interaction guidance and weak object-centric anchoring. We present Exo2EgoHOI, an HOI-aware video generative framework for interaction-preserving exocentric-to-egocentric translation. To preserve fine-grained HOI, we introduce a unified 4D HOI prior that combines scene geometry, articulated hand renderings, and dense hand-object relation fields, together with a dual-branch residual adapter for injecting structural and relational cues into the video generation backbone. To preserve object consistency, we introduce Decomposed Gated Cross-Attention, which separately encodes object and background references and adaptively integrates global semantic and local appearance features as object-centric anchors. Experiments on ARCTIC-HOI and Ego-Exo4D demonstrate substantial improvements in object consistency and HOI preservation while maintaining competitive visual fidelity. In particular, on ARCTIC-HOI, Exo2EgoHOI improves object mIoU by 32.3% and reduces MPJPE and PA-MPJPE by 34.7% and 50.0%, respectively, relative to the respective best baseline results. Project page: https://rcl-robotics.github.io/Exo2EgoHOI/.
Chinese Translation
以自我视角记录的人类操控视频为具身智能提供了宝贵的视觉体验，但大规模收集此类数据的成本较高。外部视角到自我视角的视频生成通过将丰富的第三人称操控视频转化为第一人称观察，提供了一种可扩展的替代方案。然而，现有方法在大视角变化下往往难以忠实保留展示的手-物体交互（HOI），这主要是由于缺乏细粒度的交互指导和弱对象中心的锚定。我们提出了Exo2EgoHOI，一个感知HOI的视频生成框架，用于交互保留的外部视角到自我视角转换。为了保留细粒度的HOI，我们引入了一个统一的4D HOI先验，该先验结合了场景几何、关节手部渲染和密集的手-物体关系场，并配备了一个双分支残差适配器，以将结构和关系线索注入视频生成主干。为了保持对象一致性，我们引入了分解门控交叉注意力（Decomposed Gated Cross-Attention），该方法分别编码对象和背景参考，并自适应地整合全局语义和局部外观特征作为对象中心的锚定。我们在ARCTIC-HOI和Ego-Exo4D上的实验表明，在保持竞争性视觉保真度的同时，显著提高了对象一致性和HOI保留。特别是在ARCTIC-HOI上，Exo2EgoHOI使对象mIoU提高了32.3%，并分别将MPJPE和PA-MPJPE降低了34.7%和50.0%，相较于各自最佳基线结果。项目页面：https://rcl-robotics.github.io/Exo2EgoHOI/
cs.CV / 33 / 2609.38620
HIGS: Hierarchical Implicit Grids for Joint Geometric and Semantic Scene Understanding
HIGS：用于联合几何和语义场景理解的层次隐式网格
Abstract
Neural implicit representations have had a significant impact on scene reconstruction by enabling robots to build continuous, differentiable, and high-fidelity 3D maps. Most existing works focus on geometric reconstruction and lack semantic information for high-level spatial understanding and task planning. Also, as the scale and complexity of the environment increase, neural representations face the challenge of maintaining computational efficiency in back-end optimization. To resolve these two challenges, we introduce a hierarchical neural field that leverages multiresolution submaps to achieve an efficient and scalable implicit representation, and a unified query and decoding mechanism to support both geometric and semantic features. More specifically, the learnable map features can be converted to the output with the query and decoding process for both training and inference. For large-scale representation, we decompose a scene into overlapping submaps and do hierarchical optimization within each local submap, thus enabling scalable computation. To further improve efficiency, we design feature encoders that predict initial hierarchical grid features to substantially reduce the time needed to optimize the submap features from scratch. To correct estimation drift among submaps, we align and fuse them entirely within the implicit feature space, leading to substantial acceleration by avoiding the need to decode the final output. Building upon this efficient hierarchical representation, we embed both geometric features and vision-language latent features into the map, and demonstrate it on both Signed Distance Field (SDF) construction and open-vocabulary object grounding. Our approach significantly improves computation and memory efficiency, maintains high estimation accuracy, and endows the robot with spatial awareness on large-scale real-world benchmarks.
Chinese Translation
神经隐式表示在场景重建中产生了重大影响，使机器人能够构建连续、可微分且高保真的三维地图。现有大多数研究集中于几何重建，缺乏高层空间理解和任务规划所需的语义信息。此外，随着环境规模和复杂性的增加，神经表示在后端优化中面临保持计算效率的挑战。为了解决这两个挑战，我们引入了一种层次神经场，该场利用多分辨率子地图实现高效且可扩展的隐式表示，并采用统一的查询和解码机制以支持几何和语义特征。更具体地说，可学习的地图特征可以通过查询和解码过程转换为输出，适用于训练和推理。对于大规模表示，我们将场景分解为重叠的子地图，并在每个局部子地图内进行层次优化，从而实现可扩展计算。为了进一步提高效率，我们设计了特征编码器，预测初始层次网格特征，从而显著减少从头优化子地图特征所需的时间。为了纠正子地图之间的估计漂移，我们在隐式特征空间内完全对齐和融合它们，从而通过避免解码最终输出而显著加速。基于这种高效的层次表示，我们将几何特征和视觉-语言潜在特征嵌入地图中，并在有符号距离场（Signed Distance Field, SDF）构建和开放词汇对象定位上进行了演示。我们的方法显著提高了计算和内存效率，保持了高估计准确性，并赋予机器人在大规模真实世界基准上的空间感知能力。
cs.CV / 34 / 2609.38622
Eulerian Motion Reconstruction for Water Scenery
水景的欧拉运动重建
Abstract
Reconstructing and animating water scenery from nature produces compelling and immersive visual experiences. Previous work examined this task from the perspective of 2D video textures, with the goal of creating a looping video. In our work, we tackle the problem from a 3D perspective, creating a looping 4D dynamic reconstruction which can be interactively rendered from novel viewpoints from a single non-looping 2D source video. We represent motion as a 3D static \textit{Eulerian} motion field that advects canonical Gaussian splats that are cyclically reborn at fixed time periods, supervised using rendering losses. To model non-periodic and stochastic dynamics present in real-world scenes, we add a non-periodic, time-varying residual term to capture deviations from the static Eulerian motion field. We show quantitatively and qualitatively that our framework enables photorealistic animation of water scenes better than prior art.
Chinese Translation
从自然中重建和动画化水景能够产生引人入胜且沉浸式的视觉体验。以往的研究从二维视频纹理的角度探讨了这一任务，旨在创建一个循环视频。在我们的研究中，我们从三维的角度解决这一问题，创建一个循环的四维动态重建，可以从单个非循环的二维源视频中以新颖的视角进行交互式渲染。我们将运动表示为一个三维静态的欧拉（Eulerian）运动场，该运动场推动在固定时间周期内循环重生的典型高斯斑点，并通过渲染损失进行监督。为了建模现实场景中存在的非周期性和随机动态，我们添加了一个非周期性、时间变化的残差项，以捕捉与静态欧拉运动场的偏差。我们定量和定性地展示了我们的框架能够比以往的研究更好地实现水景的照片级真实动画。
cs.CV / 35 / 2609.38636
STEPS: Scene Text Editing with Preserved Style Using Diffusion and Contrastive Style Encoding
STEPS：使用扩散和对比风格编码的风格保留场景文本编辑
Abstract
We introduce Scene Text Editing with Preserved Style (STEPS), a novel diffusion model architecture for quality text replacement in images. Scene Text Editing (STE), also known as Visual Text Editing, consists of changing the textual content in an image while conserving the original style, e.g. font, colors, orientation, background, etc. STEPS advances the state of the art in STE through directed focus on improved style preservation. We introduce a style encoder for visual text that captures style independently of textual content, and a model architecture that combines the style encoder with multiple semantic conditions (target text characters encoding and rendered glyphs). STEPS achieves superior results to previous STE methods in style preservation, output readability, and subjective quality.
Chinese Translation
我们介绍了风格保留场景文本编辑（STEPS），这是一种用于图像中高质量文本替换的新型扩散模型架构。场景文本编辑（STE），也称为视觉文本编辑，涉及在图像中更改文本内容，同时保留原始风格，例如字体、颜色、方向、背景等。STEPS通过专注于改进风格保留，推动了STE领域的技术进步。我们引入了一种视觉文本的风格编码器，该编码器独立于文本内容捕捉风格，并提出了一种将风格编码器与多个语义条件（目标文本字符编码和渲染字形）相结合的模型架构。STEPS在风格保留、输出可读性和主观质量方面优于以往的STE方法。
cs.CV / 36 / 2609.38637
Template-Search Domain Adaptation via Multi-Stage Feature Alignment for Cross-Modal Object Tracking
通过多阶段特征对齐进行模板-搜索领域适应的跨模态目标跟踪
Abstract
Visual object tracking typically assumes that the initial template and subsequent search frames share the same sensing modality. In practice, sensor availability or operation may change over time, creating a substantial representation gap between template and search frames. Unlike conventional multi-modal tracking where paired modalities are simultaneously available, cross-modal tracking requires localization when template and search frames originate from different active modalities. Accordingly, we introduce TSDA-Track, a Template-Search Domain Adaptation framework to reduce modality discrepancy during training. We investigate two feature alignment strategies. Pre-AFA TSDA-Track applies adversarial alignment before transformer's template-search interaction to suppress modality-specific bias. Enc-CFA TSDA-Track applies contrastive alignment to encoder representations after interaction to strengthen target-level cross-modal correspondence. Both variants retain a shared inference pipeline without modality-specific branches. Experiments on LasHeR, and zero-shot evaluations on RGBT234 and GTOT under multiple cross-modal protocols demonstrate improvements over representative state-of-the-art trackers. For instance, under the modality-switch protocol on RGBT234, Pre-AFA TSDA-Track achieves an SR/PR of 43.2/56.0, compared with 36.8/50.0 for ToMP-101 baseline. In addition, a study on Anti-UAV-024 further verifies the applicability of TSDA-Track to aerial tracking. Our study highlights the effectiveness of feature alignment domain adaptation for cross-modal tracking.
Chinese Translation
视觉目标跟踪通常假设初始模板和后续搜索帧共享相同的传感模态。然而，在实际应用中，传感器的可用性或操作可能会随时间变化，从而在模板和搜索帧之间产生显著的表示差距。与传统的多模态跟踪（multi-modal tracking）不同，后者同时可用配对模态，跨模态跟踪（cross-modal tracking）需要在模板和搜索帧来自不同活跃模态时进行定位。因此，我们提出了TSDA-Track，一个模板-搜索领域适应框架，旨在减少训练过程中的模态差异。我们研究了两种特征对齐策略。Pre-AFA TSDA-Track在变换器（transformer）的模板-搜索交互之前应用对抗性对齐，以抑制特定模态的偏差。Enc-CFA TSDA-Track在交互后对编码器表示应用对比对齐，以增强目标级别的跨模态对应性。这两种变体保持共享的推理管道，而没有特定于模态的分支。在LasHeR上的实验，以及在多个跨模态协议下对RGBT234和GTOT的零样本评估，均显示出相较于代表性的最先进跟踪器的改进。例如，在RGBT234的模态切换协议下，Pre-AFA TSDA-Track达到了43.2/56.0的SR/PR，而ToMP-101基线为36.8/50.0。此外，对Anti-UAV-024的研究进一步验证了TSDA-Track在空中跟踪中的适用性。我们的研究突显了特征对齐领域适应在跨模态跟踪中的有效性。
cs.CV / 37 / 2609.38641
Vision-Language-Action Autonomous Driving Agent with Language-based Memory
基于语言记忆的视觉-语言-行动自主驾驶代理
Abstract
Vision-Language-Action (VLA) foundation models have recently emerged as one of the prevailing solutions for autonomous driving, as they can utilize knowledge acquired during vision-language pretraining for accurate and interpretable driving. However, VLAs can take only a limited number of frames as visual input due to the high token cost of an image, which is problematic for memory-dependent tasks such as determining the arrival order at all-way stops and long-horizon driving scene understanding. Existing solutions use latent vector memories accessed through cross-attention, which are neither interpretable nor portable. In this paper, we propose AD-Memo, a general-purpose VLA driving agent with language-based memory. The agent outputs memory as an extension of its Chain-of-Thought (CoT) to record surrounding objects critical to driving; this memory becomes part of the agent's future input. We curate memory-based datasets and train VLAs with a two-stage recipe: Supervised Fine-Tuning (SFT) and \textit{Da Capo}, a novel semi-closed-loop Reinforcement Learning (RL) algorithm which uses trajectory-level advantage for memory and step-level advantage for driving, leading to better credit assignment. Across scenarios such as all-way stops and general driving, AD-Memo improves driving quality, enables better question answering on driving scenes, and produces plug-and-play memory for other models.
Chinese Translation
视觉-语言-行动（VLA）基础模型最近成为自主驾驶的主要解决方案之一，因为它们能够利用在视觉-语言预训练中获得的知识，实现准确且可解释的驾驶。然而，由于图像的高令牌成本，VLA只能接受有限数量的帧作为视觉输入，这对依赖记忆的任务（如确定四向停靠点的到达顺序和长时间驾驶场景理解）构成了问题。现有解决方案使用通过交叉注意力访问的潜在向量记忆，这些记忆既不可解释也不可移植。本文提出了AD-Memo，一种具有基于语言记忆的通用VLA驾驶代理。该代理将记忆作为其思维链（Chain-of-Thought, CoT）的扩展，记录对驾驶至关重要的周围物体；这段记忆成为代理未来输入的一部分。我们策划了基于记忆的数据集，并采用两阶段方案训练VLA：监督微调（Supervised Fine-Tuning, SFT）和 extit{Da Capo}，一种新颖的半闭环强化学习（Reinforcement Learning, RL）算法，利用轨迹级优势进行记忆，利用步级优势进行驾驶，从而实现更好的信用分配。在四向停靠点和一般驾驶等场景中，AD-Memo提高了驾驶质量，增强了对驾驶场景的问答能力，并为其他模型提供了即插即用的记忆。
cs.CV / 38 / 2609.38680
ReGain: Restoring Subject Fidelity in Personalization on Synthetic Images
ReGain：在合成图像个性化中恢复主体保真度
Abstract
Text-to-image diffusion models are personalized to a subject by DreamBooth fine-tuning on a handful of its images. Increasingly, these images come from a diffusion model rather than a camera. We show that fine-tuning on such synthetic images degrades subject fidelity, producing oversaturated color and excess high-frequency detail. To isolate the cause, we fine-tune two models from the same base model with the same DreamBooth recipe, one on real photos of a subject and one on synthetic images of that subject generated by the first. We trace the degradation to classifier-free guidance (CFG). For the model personalized on synthetic images, the angle between the conditional and unconditional noise predictions, and with it the norm of their difference, is much larger than for the model personalized on real photos. This inflation grows toward high frequencies and also appears at other prompts semantically close to the subject, such as its class noun, but not at unrelated ones. We propose ReGain, a training-free correction applied at sampling time that measures how much each frequency band of the guidance is inflated relative to the base model and scales that band down accordingly. ReGain needs no real photos. On Stable Diffusion v1.5, ReGain closes 51-64% of the subject-fidelity gap to the model personalized on real photos, as measured by DINO, DINOv2 and CLIP-I. It also improves subject fidelity on SDXL and SD 3.5 and preserves text alignment on all three backbones.
Chinese Translation
文本到图像的扩散模型通过对少量主体图像进行 DreamBooth 微调来实现个性化。越来越多的这些图像来自扩散模型而非相机。我们展示了在此类合成图像上进行微调会降低主体保真度，导致颜色过饱和和高频细节过多。为了隔离原因，我们使用相同的 DreamBooth 配方对两个来自同一基础模型的模型进行微调，一个使用主体的真实照片，另一个使用由第一个模型生成的该主体的合成图像。我们将降级归因于无分类器引导（CFG）。对于在合成图像上个性化的模型，条件噪声预测与无条件噪声预测之间的角度，以及它们差异的范数，远大于在真实照片上个性化的模型。这种膨胀在高频率上增长，并且在与主体语义上接近的其他提示（例如其类别名词）中也会出现，但在无关的提示中则不会。我们提出了 ReGain，这是一种在采样时应用的无训练校正方法，测量引导的每个频率带相对于基础模型的膨胀程度，并相应地缩放该频率带。ReGain 不需要真实照片。在 Stable Diffusion v1.5 上，ReGain 关闭了与基于真实照片个性化模型的主体保真度差距的 51-64%，这一差距通过 DINO、DINOv2 和 CLIP-I 测量。它还提高了 SDXL 和 SD 3.5 的主体保真度，并在所有三个基础模型上保持文本对齐。
cs.CV / 39 / 2609.38683
Unveiling the Value of Motion for Cinematic Camera Trajectories
揭示运动对电影摄影机轨迹的价值
Abstract
Cinematic camera motion is a fundamental storytelling tool, defined not only by where the camera is positioned in the scene, but also by how it moves in terms of direction and speed. Recent work on camera trajectory generation and alignment to text relies on pose-centric representations. While in principle a network could derive direction of movement and speed, we find that in practice this might not happen. In fact, in this paper we discover that decomposing the camera trajectory representation from the traditional per-frame poses to direction and speed has surprising benefits across multiple tasks, including trajectory-to-text alignment as well as text-to-trajectory generation. To accurately evaluate the former, we introduce a simple and reliable protocol that overcomes the limitations of prior evaluation baselines. For the latter, building on this representational insight, we propose a novel generative model for camera trajectories, CineGEN, that achieves superior performance across a variety of metrics. We also propose a novel dataset, CineScript, containing movie clips that are enriched with scene descriptions as well as higher-level metadata. This novel data allows us to test models' ability to capture high-level cinematographic information. We show that, despite its simplicity, representing camera trajectories through direction and speed not only helps numerically to achieve better alignment and generation, but also inherently encodes complex directorial intent.
Chinese Translation
电影摄影机运动是一个基本的叙事工具，不仅由摄影机在场景中的位置决定，还由其在方向和速度上的运动方式定义。近期关于摄影机轨迹生成和与文本对齐的研究依赖于以姿态为中心的表示。虽然原则上网络可以推导出运动方向和速度，但我们发现实际上这可能并不会发生。事实上，在本文中，我们发现将摄影机轨迹表示从传统的逐帧姿态分解为方向和速度在多个任务中具有意想不到的好处，包括轨迹与文本的对齐以及文本到轨迹的生成。为了准确评估前者，我们引入了一种简单且可靠的协议，克服了先前评估基准的局限性。对于后者，基于这一表示洞察，我们提出了一种新颖的摄影机轨迹生成模型CineGEN，在多种指标上实现了卓越的性能。我们还提出了一个新数据集CineScript，其中包含丰富场景描述和更高层次元数据的电影片段。这一新颖的数据使我们能够测试模型捕捉高级电影信息的能力。我们展示了尽管其简单性，通过方向和速度表示摄影机轨迹不仅在数值上有助于实现更好的对齐和生成，而且本质上也编码了复杂的导演意图。
cs.CV / 40 / 2609.38689
EPIC: Epipolar-Consistent 360{\deg} Immersive Stereo Video Generation
EPIC：极线一致的360°沉浸式立体视频生成
Abstract
Immersive displays can enable rich and diverse virtual experiences. Manually authoring every possible experience to realize this potential, however, is prohibitively expensive, difficult to scale, and impractical. Generative AI models could remove this bottleneck, but today's models are built for conventional displays and cannot generate the high-resolution, stereoscopic $360^\circ$ content required for immersive viewing. Further, temporal and stereo inconsistencies that may be tolerable on conventional displays can become highly disruptive when viewed through an immersive headset. Here, we address this gap with a zero-shot generative pipeline that extends existing video diffusion models into 4K stereoscopic $360^\circ$ videos. Inspired from binocular vision and depth perception, we develop an epipolar-aware $360^\circ$ image matching metric that captures the temporal and stereo geometric inconsistencies across views. We then use this metric as a preference signal for direct preference optimization with limited training data. Our work enables $360^\circ$ stereo video generation and provides a scalable path for bringing generative content to immersive displays, allowing diverse mixed reality experiences on demand.
Chinese Translation
沉浸式显示器能够提供丰富多样的虚拟体验。然而，手动创作每一种可能的体验以实现这一潜力是极其昂贵的，难以扩展且不切实际。生成性人工智能模型可以消除这一瓶颈，但目前的模型是为传统显示器构建的，无法生成沉浸式观看所需的高分辨率立体360°内容。此外，在传统显示器上可以容忍的时间和立体不一致性，在通过沉浸式头戴设备观看时可能会变得极具干扰性。在此，我们通过一个零样本生成管道来解决这一问题，该管道将现有的视频扩散模型扩展到4K立体360°视频。受到双眼视觉和深度感知的启发，我们开发了一种极线感知的360°图像匹配度量，捕捉视角间的时间和立体几何不一致性。然后，我们将该度量用作有限训练数据下的直接偏好优化的偏好信号。我们的工作实现了360°立体视频生成，并为将生成内容引入沉浸式显示器提供了一条可扩展的路径，使得按需提供多样化的混合现实体验成为可能。
cs.CV / 41 / 2609.38691
No Corners Cut: State-Grounded Transitions for Mid-Stream Prompt Switches in Video Generation
不走捷径：基于状态的中途提示切换在视频生成中的应用
Abstract
Streaming video generators allow users to dynamically modulate video synthesis via mid-stream prompt switching. Existing streaming methods can respond to the updated instruction while still cutting corners, prematurely realizing goals or taking heuristic shortcuts that bypass necessary intermediate state changes needed for a plausible transition. In this study, we present SEGUE, a novel framework that makes this process explicit and trains the generator to execute these transitions faithfully. At each switch, a training-free planner parses the latest frame and prompts, writes a few segue prompts with roles and durations, and then hands control back to the user's prompt. Furthermore, to address the inherent difficulty of training causal models on short-lived temporal schedules without corrupting preparatory supervision, we introduce SPANDMD, which evaluates each active prompt using the full rollout as temporal context while retaining its DMD residual only within the prompt's assigned span. On OpenTrans-360, a benchmark of 1,800 switches that scores how the old state exits and the new one begins, SEGUE ranks first on all eight transition metrics and raises the overall score over the strongest baseline from 0.866 to 0.887. It also ranks first on four of six instruction-response metrics of StreamAV-Bench, while the planner transfers to frozen autoregressive generators without retraining.
Chinese Translation
流媒体视频生成器允许用户通过中途提示切换动态调节视频合成。现有的流媒体方法在响应更新指令时，往往会走捷径，过早实现目标或采取启发式的简化方法，绕过实现合理过渡所需的必要中间状态变化。在本研究中，我们提出了SEGUE，一个新颖的框架，使这一过程变得明确，并训练生成器忠实地执行这些过渡。在每次切换时，无需训练的规划器解析最新的帧和提示，编写一些带有角色和持续时间的过渡提示，然后将控制权交还给用户的提示。此外，为了解决在短暂时间表上训练因果模型的固有困难而不破坏准备性监督的问题，我们引入了SPANDMD，它利用完整的展开作为时间上下文来评估每个活跃提示，同时仅在提示分配的时间范围内保留其DMD残差。在OpenTrans-360基准测试中，该测试包含1,800次切换，评估旧状态的退出和新状态的开始，SEGUE在所有八个过渡指标中排名第一，并将整体得分从最强基线的0.866提高到0.887。同时，在StreamAV-Bench的六个指令-响应指标中，SEGUE在四个指标上排名第一，而规划器能够在不重新训练的情况下转移到冻结的自回归生成器上。
cs.CV / 42 / 2609.38705
SCALE: Synthetic Calibration via Agreement Labeling in Embedding Space
SCALE：通过嵌入空间中的一致性标注进行合成校准
Abstract
Foundation models for computational pathology are usually evaluated using AUC and accuracy, while calibration is often left untested. This matters because a model can be accurate on average but still assign overly confident probabilities to cases that are difficult even for pathologists. We study calibration across eight pathology foundation models. Using pathologist agreement as a measure of diagnostic difficulty, we find that calibration error is consistently higher on low-agreement cases than on high-agreement cases. This pattern is not apparent from aggregate expected calibration error (ECE) alone. We then propose synthetic agreement calibration, a method for improving calibration without collecting multi-annotator labels. Given a trained linear probe, we select high-confidence embeddings as class anchors and interpolate between anchors from opposite classes. The interpolation weights encode a continuous notion of diagnostic ambiguity, which we use as a synthetic agreement signal to retrain the probe with agreement-aware label smoothing. On MHIST, which includes annotations from seven pathologists, synthetic agreement calibration recovers most of the calibration improvement obtained by label smoothing based on real pathologist agreement, while substantially reducing low-agreement ECE relative to the uncalibrated baseline. Discrimination metrics are preserved. On PatchCamelyon and BreakHis, public histopathology datasets without multi-annotator labels, the method improves calibration across the evaluated foundation models, whereas annotator-dependent approaches cannot be used without additional expert annotation.
Chinese Translation
计算病理学的基础模型通常通过AUC和准确率进行评估，而校准往往未被测试。这一点很重要，因为模型可能在平均上是准确的，但仍然会对即使对于病理学家来说也很困难的案例赋予过于自信的概率。我们研究了八个病理基础模型的校准情况。使用病理学家一致性作为诊断难度的衡量标准，我们发现低一致性案例的校准误差始终高于高一致性案例。这一模式仅从聚合的期望校准误差（ECE）中并不明显。随后，我们提出了合成一致性校准，这是一种在不收集多标注者标签的情况下改善校准的方法。在给定的训练线性探针中，我们选择高置信度的嵌入作为类别锚点，并在来自相反类别的锚点之间进行插值。插值权重编码了诊断模糊性的连续概念，我们将其用作合成一致性信号，以使用考虑一致性的标签平滑重新训练探针。在包含七位病理学家注释的MHIST数据集上，合成一致性校准恢复了基于真实病理学家一致性的标签平滑所获得的大部分校准改进，同时相对于未校准的基线显著降低了低一致性ECE。判别指标得以保留。在PatchCamelyon和BreakHis这两个没有多标注者标签的公共组织病理学数据集上，该方法改善了评估的基础模型的校准，而依赖标注者的方法在没有额外专家注释的情况下无法使用。
cs.CV / 43 / 2609.38714
Hard-Region Supervision: #1 on the Waymo Open Dataset 2D Video Panoptic Segmentation Leaderboard
硬区域监督：在Waymo开放数据集2D视频全景分割排行榜上排名第一
Abstract
We describe our winning entry to the Waymo Open Dataset 2D Video Panoptic Segmentation Challenge. The task asks for a semantic class at every pixel of every frame and, for countable objects, an identity that holds across 100 frames and across five overlapping cameras. We build on DVIS++, a cascade of a segmenter, a tracker, and a refiner, as our baseline. We propose hard region supervision (HRS) to improve the baseline. In particular, we use the baseline to define the hard region as where it makes mistakes, and design a loss and an auxiliary prediction head for this region. The auxiliary head is used only in training and removed at test time, so at inference the model trained with HRS has the same architecture as the baseline. In addition, we propose three test-time steps that further improve the results: a two-model ensemble, a merge of the segmenter's output into the final panoptic map, and cross-camera identity linking. On the challenge test set, our entry reaches 0.3547 wSTQ, 0.2071 wAQ, and 0.6075 mIoU, ranking first on all three metrics. It is 3.6 wSTQ points ahead of the second entry and 2.4 points ahead of our DVIS++ baseline.
Chinese Translation
我们描述了在Waymo开放数据集2D视频全景分割挑战中的获胜方案。该任务要求为每一帧的每一个像素提供一个语义类别，并且对于可计数的物体，提供一个在100帧和五个重叠摄像头之间保持一致的身份。我们以DVIS++为基础，DVIS++是一个由分割器、跟踪器和精细化器组成的级联模型。我们提出了硬区域监督（HRS）以改善基础模型。具体而言，我们利用基础模型来定义硬区域，即模型出错的地方，并为该区域设计了一种损失函数和一个辅助预测头。辅助预测头仅在训练时使用，在测试时移除，因此在推理时，使用HRS训练的模型与基础模型具有相同的架构。此外，我们提出了三个测试时步骤，进一步提高了结果：两个模型的集成、将分割器的输出合并到最终的全景图中，以及跨摄像头的身份链接。在挑战测试集上，我们的方案达到了0.3547 wSTQ、0.2071 wAQ和0.6075 mIoU，在所有三个指标上均排名第一。它比第二名高出3.6 wSTQ点，比我们的DVIS++基础模型高出2.4点。
cs.CV / 44 / 2609.38716
SpatialCORE: Confidence-Aware Grounded Spatial Reasoning in Large Vision--Language Models
SpatialCORE：在大型视觉-语言模型中进行自信感知的基础空间推理
Abstract
Large Vision-Language Models (LVLMs) have made remarkable progress across visual perception tasks, yet spatial reasoning remains a persistent weakness, especially for questions that require reasoning over visual space. Recent spatial-reasoning methods incorporate generated grounding, where models predict bounding boxes, masks, or other localization outputs for task-relevant objects as part of their reasoning trace. However, these approaches typically optimize final-answer correctness alone, allowing correct answers to be rewarded even when the model does not reason from confidently localized task-relevant objects. We introduce SpatialCORE (Spatially COnfident REasoning), a post-training framework that turns the model's own confidence in generated grounding into a learning signal for spatial reasoning. Its central idea is to reinforce grounding that is both accurate and confident, encouraging the model to reason from confidently localized task-relevant objects. SpatialCORE realizes this through a self-regulating spatial reward that weights each predicted bounding box's matching quality by its coordinate-token confidence. An answer gate further ties grounding optimization to final-answer correctness. SpatialCORE achieves state-of-the-art results among open-source and specialized spatial reasoning models across diverse benchmarks, and transfers effectively in zero-shot settings to unseen data distributions. The source code is available at https://github.com/rafiibnsultan/SpatialCORE.
Chinese Translation
大型视觉-语言模型（LVLMs）在视觉感知任务中取得了显著进展，但空间推理仍然是一个持续的弱点，特别是对于需要在视觉空间中进行推理的问题。最近的空间推理方法结合了生成的基础，模型预测边界框、掩膜或其他与任务相关对象的定位输出，作为其推理过程的一部分。然而，这些方法通常仅优化最终答案的正确性，即使模型未从自信定位的任务相关对象进行推理，也会奖励正确答案。我们提出了SpatialCORE（空间自信推理），这是一个后训练框架，将模型对生成基础的自信度转化为空间推理的学习信号。其核心思想是强化既准确又自信的基础，鼓励模型从自信定位的任务相关对象进行推理。SpatialCORE通过自我调节的空间奖励实现这一目标，该奖励根据每个预测边界框的坐标标记自信度加权匹配质量。答案门进一步将基础优化与最终答案的正确性联系起来。SpatialCORE在各种基准测试中，在开源和专业空间推理模型中实现了最先进的结果，并在零样本设置中有效地迁移到未见数据分布。源代码可在 https://github.com/rafiibnsultan/SpatialCORE 获取。
cs.CV / 45 / 2609.38717
Soft Spatial Reasoning
软空间推理
Abstract
Large Vision-Language Models (LVLMs) commonly perform spatial reasoning through chain-of-thought (CoT), encoding intermediate reasoning as autoregressive sequences of discrete language tokens. Such hard thinking requires committing to a single token at each step, even when the correct spatial interpretation remains uncertain. This early commitment constitutes premature discretization: an incorrect token selection can propagate errors through subsequent reasoning. We propose Soft Spatial Reasoning, a post-training framework that introduces soft thinking for spatial tasks in LVLMs. At each intermediate reasoning step, the LVLM forms a continuous soft state by mixing token embeddings rather than selecting a single token, allowing multiple candidate continuations to influence the next step. The appropriate degree of softness, however, can vary across reasoning steps: retaining multiple candidates may preserve a useful spatial interpretation, but if those candidates imply conflicting spatial relations, mixing them may interfere with subsequent reasoning. At the core of Soft Spatial Reasoning is AdaptSoft, a controller that uses the current hidden state and predictive uncertainty to adapt the degree of softness at each reasoning step. To train AdaptSoft, we introduce a gradient-alignment learning objective that provides a step-specific learning signal for softness control without intermediate reasoning supervision. Across diverse spatial benchmarks, Soft Spatial Reasoning outperforms hard and fixed-soft CoT baselines using the same backbone, as well as a range of existing LVLMs. The source code is available at https://github.com/rafiibnsultan/Soft_Spatial_Reasoning
Chinese Translation
大型视觉语言模型（LVLMs）通常通过思维链（CoT）进行空间推理，将中间推理编码为离散语言标记的自回归序列。这种硬性思维要求在每一步都必须选择一个单一标记，即使正确的空间解释仍然不确定。这种早期承诺构成了过早离散化：不正确的标记选择可能会在后续推理中传播错误。我们提出了软空间推理，这是一种后训练框架，为LVLMs中的空间任务引入了软性思维。在每个中间推理步骤中，LVLM通过混合标记嵌入形成一个连续的软状态，而不是选择一个单一标记，从而允许多个候选继续影响下一步。然而，适当的软性程度可能在推理步骤之间有所不同：保留多个候选可能保留有用的空间解释，但如果这些候选暗示了相互冲突的空间关系，混合它们可能会干扰后续推理。软空间推理的核心是AdaptSoft，一个控制器，它利用当前的隐藏状态和预测不确定性来调整每个推理步骤的软性程度。为了训练AdaptSoft，我们引入了一个梯度对齐学习目标，为软性控制提供逐步特定的学习信号，而无需中间推理监督。在多样的空间基准测试中，软空间推理在使用相同骨干网络的情况下，优于硬性和固定软性CoT基线，以及一系列现有的LVLMs。源代码可在https://github.com/rafiibnsultan/Soft_Spatial_Reasoning获取。
cs.CV / 46 / 2609.38746
Matisse: Evidence-Space Reasoning for Active 3D Reconstruction
Matisse：用于主动三维重建的证据空间推理
Abstract
How can a 3D reconstruction system acquire and retain useful information to understand the geometry of a scene from partial views under a limited computation budget? Existing active view acquisition methods typically estimate uncertainty over observed or instantiated geometry, limiting their ability to reason about unseen structure, while long-horizon reconstruction methods often retain redundant observations. We introduce Matisse, a training-free framework that unifies active reconstruction and keyframe selection by leveraging evidence provided by a pretrained generative 3D model. Matisse estimates Evidential Uncertainty from cross-attention evidence associated with 3D latent tokens and derives an Evidential Information Gain to guide both view acquisition and keyframe selection based on the expected reduction in posterior entropy. Matisse supports multi-object scenes through occlusion-aware, object-balanced aggregation and propagates uncertainty through intermediate latents to avoid full reconstruction during planning. Matisse reduces Chamfer distance by 12.7%, 3.8%, and 9.2% on GSO30, YCB-V, and Replica, respectively, relative to the best baseline on each dataset, and achieves a $1.50\times$ end-to-end speedup over the best active reconstruction baseline on GSO30 with the same reconstruction backend. In the GSO30 keyframe selection experiment for long-horizon reconstruction, Matisse achieves comparable Chamfer distance using 14% of the input views compared with Stream3D.
Chinese Translation
一个三维重建系统如何在有限的计算预算下，从部分视图中获取和保留有用信息，以理解场景的几何形状？现有的主动视图获取方法通常估计观察到或实例化几何体的的不确定性，限制了它们对未见结构的推理能力，而长时间跨度的重建方法往往保留冗余观察。我们提出了Matisse，一个无训练框架，通过利用预训练生成三维模型提供的证据，统一了主动重建和关键帧选择。Matisse从与三维潜在标记相关的交叉注意证据中估计证据不确定性，并推导出证据信息增益，以指导视图获取和关键帧选择，基于后验熵的预期减少。Matisse通过考虑遮挡的对象平衡聚合支持多对象场景，并通过中间潜在变量传播不确定性，以避免在规划期间进行完整重建。相较于每个数据集上的最佳基线，Matisse在GSO30、YCB-V和Replica上分别减少了12.7%、3.8%和9.2%的Chamfer距离，并在GSO30上实现了与最佳主动重建基线相同重建后端的$1.50 imes$端到端加速。在GSO30的关键帧选择实验中，Matisse在使用14%的输入视图时，达到了与Stream3D相当的Chamfer距离。
cs.CV / 47 / 2609.38747
Consensus-Aware Multi-Source Fusion for Reference-Guided Camouflaged Object Detection
基于共识的多源融合用于参考引导的伪装目标检测
Abstract
Reference-guided camouflaged object detection aims to segment a target whose visual appearance closely resembles its surroundings by exploiting auxiliary reference samples. The task remains difficult because reference samples contain inconsistent target cues, while generic visual representations are not inherently aligned with the target specified by the references. To handle these problems, we present a consensus-aware multi-source fusion framework. Reference-Conditioned Dual-Backbone Fusion (RCDF) couples trainable PVTv2 query features with frozen DINOv3 representations and uses reference-conditioned correlation to select foundation-model evidence before multi-scale fusion. The framework also aggregates multiple references through cross-reference consensus aggregation and injects reference information at semantic depths matched to the query features. Extensive experiments demonstrate the effectiveness of the proposed method. The results further show that reference consensus, target-conditioned foundation features, and hierarchical decoding provide complementary improvements under the evaluation protocol. The source code will be made publicly available upon acceptance.
Chinese Translation
参考引导的伪装目标检测旨在通过利用辅助参考样本来分割与其周围环境视觉外观高度相似的目标。该任务仍然困难，因为参考样本包含不一致的目标线索，而通用视觉表示与参考指定的目标并不天然对齐。为了解决这些问题，我们提出了一种基于共识的多源融合框架。参考条件双主干融合（Reference-Conditioned Dual-Backbone Fusion, RCDF）将可训练的PVTv2查询特征与冻结的DINOv3表示相结合，并使用参考条件相关性在多尺度融合之前选择基础模型证据。该框架还通过交叉参考共识聚合聚合多个参考，并在与查询特征匹配的语义深度注入参考信息。大量实验表明了所提方法的有效性。结果进一步表明，参考共识、目标条件的基础特征和分层解码在评估协议下提供了互补的改进。源代码将在接受后公开发布。
cs.CV / 48 / 2609.38748
Here the World in Stereo: Learning Dynamic Spatial Correspondence for Immersive Joint Video-Audio Generation
立体声中的世界：学习动态空间对应以实现沉浸式联合视频-音频生成
Abstract
Recent joint video-audio generation models have achieved strong semantic correspondence and temporal synchronization. However, applications such as AR/VR and interactive gaming further require stereo audio to provide an immersive sense, which remains largely overlooked. Effective stereo audio requires the perceived sound location to evolve consistently with the motion of its corresponding visual source. We refer to this property as Dynamic Spatial Correspondence and propose StereoBind, a framework that binds visual source motion to stereo sound generation. StereoBind uses motion tracks to coordinate visual motion and stereo audio through three complementary mechanisms. Visual Motion Binding establishes source-aware audiovisual correspondence, the Spatial Track Encoder captures absolute source positions, and Residual Track RoPE models relative motion. For supervision and evaluation, we construct StereoWorld-29K, a large-scale stereo audio-video dataset with paired motion tracks, and StereoWorldBench for measuring audiovisual spatial consistency. Experiments show that StereoBind substantially improves spatial alignment in stereo audio generation over existing models while preserving overall audiovisual quality.
Chinese Translation
近年来，联合视频-音频生成模型在语义对应和时间同步方面取得了显著进展。然而，增强现实/虚拟现实（AR/VR）和互动游戏等应用进一步要求立体声音频以提供沉浸感，而这一点在很大程度上被忽视。有效的立体声音频要求感知的声音位置与其对应的视觉源的运动保持一致。我们将这一特性称为动态空间对应，并提出了StereoBind，一个将视觉源运动与立体声生成绑定的框架。StereoBind通过三种互补机制使用运动轨迹来协调视觉运动和立体声音频。视觉运动绑定建立了源感知的视听对应，空间轨迹编码器捕捉绝对源位置，而残差轨迹RoPE模型则描述相对运动。为了进行监督和评估，我们构建了StereoWorld-29K，一个大规模的立体声音频-视频数据集，配有配对的运动轨迹，以及StereoWorldBench用于测量视听空间一致性。实验表明，StereoBind在立体声音频生成中的空间对齐显著优于现有模型，同时保持了整体的视听质量。
cs.CV / 49 / 2609.38755
Agentic Relative Camera Pose Estimation via Learned Ranking and Verification
通过学习排名和验证的自主相对相机位姿估计
Abstract
A wide range of approaches have been developed for camera pose estimation, including correspondence-based methods, end-to-end pose regression, and recent 3D geometric foundation models. Our key observation is that no single estimator is optimal for diverse challenges, such as wide baselines, lack of texture, appearance changes, and occlusions. Further analysis reveals substantial performance variation across both benchmarks and individual image pairs, with different estimators exhibiting complementary strengths. We introduce PoseAgent, an agentic framework for relative camera pose estimation that dynamically orchestrates pose estimators through learnable ranking and verification. Given an image pair, a profiling agent first extracts appearance, semantic, and geometric features relevant to pose estimation, e.g., scene type. A learned ranking agent then predicts the relative competence of multiple pose estimators given the image-pair profile. The top-ranked estimator is executed, and its predicted pose is assessed by a learned verification agent that estimates the corresponding pose error. When verification fails, PoseAgent adaptively invokes lower-ranked estimators until a candidate is accepted or the execution budget is reached. For pose verification, our verification network predicts pose errors more accurately than prior models. For pose estimation, PoseAgent improves AUC@5 degree up to 4.2% over the strongest standalone estimator on each of ARKitScenes, MegaDepth, ScanNet++, and RealEstate10K. On ARKitScenes, PoseAgent also outperforms VLM-based agents, which include a VLM ranker with the same verifier and fallback policy. These results demonstrate the effectiveness of our learned ranking and verification.
Chinese Translation
已经开发了多种方法用于相机位姿估计，包括基于对应关系的方法、端到端的位姿回归以及最近的3D几何基础模型。我们的关键观察是，没有单一的估计器能够在多样化的挑战中表现最佳，例如宽基线、缺乏纹理、外观变化和遮挡。进一步分析显示，在不同基准和个别图像对之间，性能存在显著差异，不同的估计器展现出互补的优势。我们引入了PoseAgent，一个用于相对相机位姿估计的自主框架，通过可学习的排名和验证动态协调位姿估计器。给定一对图像，配置代理首先提取与位姿估计相关的外观、语义和几何特征，例如场景类型。然后，学习的排名代理根据图像对的特征预测多个位姿估计器的相对能力。执行排名最高的估计器，并由学习的验证代理评估其预测的位姿，估计相应的位姿误差。当验证失败时，PoseAgent自适应地调用排名较低的估计器，直到接受一个候选者或达到执行预算。对于位姿验证，我们的验证网络比之前的模型更准确地预测位姿误差。在位姿估计方面，PoseAgent在ARKitScenes、MegaDepth、ScanNet++和RealEstate10K上相较于最强的独立估计器提高了AUC@5度最多4.2%。在ARKitScenes上，PoseAgent还超越了基于VLM的代理，包括具有相同验证器和后备策略的VLM排名器。这些结果证明了我们学习的排名和验证的有效性。
cs.CV / 50 / 2609.38758
Event-Driven Refresh and Recurrence Memory to Reduce Stale Grounding in Referring Video Object Segmentation
事件驱动的刷新与回忆记忆以减少视频物体分割中的陈旧基础
Abstract
Referring Video Object Segmentation (RVOS) aims to produce a pixel-accurate mask sequence for an object specified by natural language. Sa2VA combines a multimodal large language model with SAM2 for grounded segmentation; however, its inference typically grounds the query from a small fixed set of initial keyframes and then relies on propagation. In long or dynamic videos, this can cause stale grounding and persistent false positives when the object composition changes (e.g., distractors enter or the target disappears/re-appears). We propose Event-Driven Refresh + Recurrence Memory (EDRRM), an enhancement that selectively re-invokes Sa2VA only at stable change points. EDRRM triggers refresh boundaries using an EMA-smoothed event score computed from tracking-derived cues (births/deaths and coarse composition/layout changes) with temporal constraints. A recurrence memory further retrieves anchor frames via CLIP similarity to re-condition the model on re-appearance events. Experiments on Ref-DAVIS17, MeViS, and ReVOS show that EDRRM achieves a competitive accuracy-efficiency trade-off relative to fixed-window and FrameDiff-SSIM baselines, maintaining comparable or superior J&F scores at substantially lower average refresh-call budgets and reducing false-positive failures. End-to-end runtime analysis further confirms that the overhead introduced by tracking, CLIP-based recurrence matching, and the identifiability gate remains modest relative to the dominant Sa2VA inference cost, thereby validating the efficiency of the proposed pipeline.
Chinese Translation
指代视频物体分割（Referring Video Object Segmentation, RVOS）旨在为通过自然语言指定的物体生成像素级精确的掩膜序列。Sa2VA将多模态大型语言模型与SAM2结合用于基础分割；然而，其推理通常从一小组固定的初始关键帧中对查询进行基础处理，然后依赖于传播。在长时间或动态视频中，这可能导致基础信息陈旧以及在物体组成变化时（例如，干扰物进入或目标消失/重新出现）持续出现误报。我们提出了事件驱动的刷新与回忆记忆（Event-Driven Refresh + Recurrence Memory, EDRRM），一种增强方法，仅在稳定的变化点选择性地重新调用Sa2VA。EDRRM通过使用从跟踪派生的线索（出生/死亡和粗略组成/布局变化）计算的EMA平滑事件分数触发刷新边界，并施加时间约束。回忆记忆进一步通过CLIP相似性检索锚帧，以便在重新出现事件上重新调整模型。在Ref-DAVIS17、MeViS和ReVOS上的实验表明，EDRRM在相对于固定窗口和FrameDiff-SSIM基线的情况下，实现了竞争性的准确性与效率平衡，在显著较低的平均刷新调用预算下保持了可比或更优的J&F分数，并减少了误报失败。端到端的运行时分析进一步确认，由于跟踪、基于CLIP的回忆匹配和可识别性门引入的开销相对于主导的Sa2VA推理成本仍然适中，从而验证了所提管道的效率。
cs.CV / 51 / 2609.38777
Distill the Visual Evidence, Not Just the Answer: Cross-World On-Policy Distillation for Vision-Language Models
提炼视觉证据，而不仅仅是答案：跨世界的在线策略蒸馏用于视觉语言模型
Abstract
A central goal of vision-language model (VLM) distillation is to transfer both the teacher's language capabilities and its visual understanding. However, existing methods primarily supervise the student's output, leaving visual understanding implicit. Our analysis reveals that a student can match the teacher's answer without relying on the same visual evidence, raising the question: how can we ensure the student responds to the visual information that actually determines the answer? To this end, we propose \textbf{Cross-World On-Policy Distillation (CW-OPD)}, which explicitly supervises the student's response to changes in visual evidence. For each example, CW-OPD constructs two visual worlds that share the question and scene context but differ in answer-critical evidence, yielding different answers. We perform on-policy distillation in both worlds and distill the teacher's cross-world belief transition, encouraging the student to match not only \emph{what} the teacher predicts but also \emph{why} its prediction changes with the evidence. A gradient analysis shows that this term is invariant to errors shared by both worlds and supplies a corrective signal invisible to endpoint matching alone. In this way, CW-OPD makes reliance on the relevant visual evidence an explicit distillation target rather than an implicit consequence of output matching. To diagnose whether a model truly grounds its answers in visual evidence, we introduce CWBench, which measures cross-world consistency via Cross-World Pair Accuracy (CWPA). Experiments on Qwen3.5-4B show that CW-OPD outperforms the strongest baseline by \textbf{1.2} points on average, and the 4B student exceeds DeepSeek-V4.1 (552B) by \textbf{22.4} CWPA points on CWBench. Code is released in https://github.com/baokou-fw2/CWAD.
Chinese Translation
视觉语言模型（VLM）蒸馏的一个核心目标是转移教师的语言能力和视觉理解。然而，现有方法主要监督学生的输出，导致视觉理解隐含。我们的分析表明，学生可以在不依赖相同视觉证据的情况下匹配教师的答案，这引发了一个问题：我们如何确保学生响应实际决定答案的视觉信息？为此，我们提出了 extbf{跨世界在线策略蒸馏（CW-OPD）}，该方法明确监督学生对视觉证据变化的响应。对于每个示例，CW-OPD构建了两个视觉世界，这两个世界共享问题和场景上下文，但在答案关键证据上存在差异，从而产生不同的答案。我们在这两个世界中进行在线策略蒸馏，并蒸馏教师的跨世界信念转变，鼓励学生不仅匹配教师预测的 extit{什么}，还要理解其预测如何因证据而变化的 extit{为什么}。梯度分析表明，该项内容对两个世界共享的错误是不变的，并提供了一种仅通过端点匹配无法观察到的修正信号。通过这种方式，CW-OPD使对相关视觉证据的依赖成为一个明确的蒸馏目标，而不是输出匹配的隐含结果。为了诊断模型是否真正将其答案基于视觉证据，我们引入了CWBench，它通过跨世界配对准确性（CWPA）来衡量跨世界一致性。在Qwen3.5-4B上的实验表明，CW-OPD的表现比最强基线平均提高了 extbf{1.2}分，而4B学生在CWBench上超过DeepSeek-V4.1（552B） extbf{22.4} CWPA分。代码已发布在https://github.com/baokou-fw2/CWAD。
cs.CV / 52 / 2609.38810
CRAFT: Causal Responsibility and Failure Tracing in Medical Vision Language Models
CRAFT：医学视觉语言模型中的因果责任与故障追踪
Abstract
As vision language models are increasingly deployed in clinical diagnosis, understanding how they internally resolve competing visual and textual signals becomes a safety imperative. Existing mechanistic analyses remain confined to unimodal text and offer no explanation for why a single misleading sentence can override a correct image based diagnosis, or why a model commits to a confident answer despite insufficient visual evidence. We find that these two safety risks, arbitration failure where textual context overrides visual grounding and brake failure where the model commits without adequate evidence, are mediated by spatially disjoint attention head populations: arbitration heads form a mid-to-deep wideband reflecting cross-layer evidence competition, while brake heads concentrate in a narrow middle-to-late layer band that regulates evidence sufficiency and abstention behavior. To ground these observations in causal circuitry, we introduce CRAFT, which localizes each failure mode to a minimal causal head set via dual criteria and verifies necessity and sufficiency through temporal probes and Tuned Lens trajectory analysis. Excising arbitration heads sharply reduces conflict following with negligible degradation on clean inputs, while excising brake heads restores appropriate abstention under degraded visual evidence. The two interventions target spatially disjoint head sets and produce distinct corrective effects, underscoring the mechanistic separability of the failure modes. Experiments across multiple medical VQA benchmarks and VLM architectures validate both the localization and interventions, demonstrating that the identified heads causally drive each failure mode and that targeted modulation generalises without retraining. The code is available at GitHub repository.
Chinese Translation
随着视觉语言模型在临床诊断中的日益应用，理解它们如何内部解决竞争的视觉和文本信号成为了一项安全迫切的任务。现有的机械分析仍然局限于单模态文本，未能解释为何单个误导性句子能够覆盖正确的图像基础诊断，或为何模型在缺乏足够视觉证据的情况下仍然坚持自信的答案。我们发现这两种安全风险，即文本上下文覆盖视觉基础的仲裁失败，以及模型在没有充分证据的情况下做出承诺的制动失败，是由空间上不相交的注意力头群体介导的：仲裁头形成了一个中到深的宽带，反映了跨层证据竞争，而制动头则集中在一个狭窄的中到后层带，调节证据的充分性和弃权行为。为了将这些观察结果与因果电路联系起来，我们引入了CRAFT，通过双重标准将每种故障模式定位到最小因果头集，并通过时间探测和调谐透镜轨迹分析验证必要性和充分性。切除仲裁头显著减少了冲突，且对干净输入的降解几乎可以忽略不计，而切除制动头则在降级的视觉证据下恢复了适当的弃权。这两种干预针对空间上不相交的头集，产生了不同的纠正效果，强调了故障模式的机械可分离性。在多个医学视觉问答基准和视觉语言模型架构上的实验验证了定位和干预，证明了识别出的头部因果驱动每种故障模式，并且目标调制在不重新训练的情况下具有普适性。代码可在GitHub仓库中获取。
cs.CV / 53 / 2609.38811
DCM-SAM: Defect-Conditioned Mixture of LoRA Experts for NPU-Deployed AM Defect Segmentation
DCM-SAM：用于NPU部署的AM缺陷分割的缺陷条件混合LoRA专家
Abstract
Metal additive manufacturing parts are inspected by X-ray computed tomography, where labelled data is scarce, the pores and inclusions that matter span a few pixels, and inspection must happen at the machine. We present DCM-SAM, a defect-conditioned adaptive mixture of LoRA experts: one frozen Segment Anything backbone carries a separate Conv-LoRA expert bank and mask decoder per defect class, each trained in its own pass, without prompts, on synthetic slices alone, updating only 4.4% of the parameters. On benchmarks that XCT-SAM reports, DCM-SAM improves on every baseline for both classes from a ViT-B backbone against their ViT-H, and reaches 64.2% pore IoU on real NIST scans having seen no real images during training. Deployment then exposes what adaptation work rarely measures: on a Qualcomm Hexagon NPU, ViT-H and ViT-L compile yet cannot allocate at 1024x1024 image resolution, since activations rather than weights exceed the device ceiling, and quantizing weights does not help. ViT-B alone runs, but the adapted encoder then fails to allocate where the stock one succeeds, until a numerically identical rewrite of the attention lets the complete DCM-SAM run in FP16 at 1024x1024, with no operator falling back to the CPU, masks within 0.01% of pixels of the FP32 reference. Code: https://github.com/MushfiqShovon/DCM-SAM.
Chinese Translation
金属增材制造部件通过X射线计算机断层扫描进行检测，其中标记数据稀缺，重要的孔隙和夹杂物仅跨越几个像素，并且检测必须在机器上进行。我们提出了DCM-SAM，一种缺陷条件自适应混合LoRA专家：一个冻结的Segment Anything主干网络为每个缺陷类别配备一个单独的Conv-LoRA专家库和掩码解码器，每个专家在没有提示的情况下仅在合成切片上进行独立训练，仅更新4.4%的参数。在XCT-SAM报告的基准测试中，DCM-SAM在ViT-B主干网络上对比其ViT-H版本，针对两个类别的每个基线均有改进，并在未见过真实图像的情况下，在真实的NIST扫描中达到了64.2%的孔隙IoU。部署后暴露出适应性工作很少衡量的内容：在高通Hexagon NPU上，ViT-H和ViT-L编译后无法在1024x1024图像分辨率下分配，因为激活而非权重超过了设备的上限，而量化权重并没有帮助。仅ViT-B能够运行，但适应后的编码器在分配上失败，而标准编码器成功，直到注意力的数值相同的重写使得完整的DCM-SAM能够在FP16下以1024x1024运行，且没有操作回退到CPU，掩码与FP32参考的像素差异在0.01%以内。代码链接： https://github.com/MushfiqShovon/DCM-SAM。
cs.CV / 54 / 2609.38819
Future Video Generation Better Aligns with the Human Visual Cortex than Observed Video
未来视频生成与人类视觉皮层的对齐优于观察到的视频
Abstract
Studying the alignment between the internal representations of vision models and the responses of the visual cortex to the same observed visual stimuli has enabled us to better understand human visual processing. However, studies so far have largely overlooked the fact that the human brain not only processes observed visual stimuli, but also predicts upcoming stimuli based on what has been observed. Accordingly, we hypothesize that internal representations for generating future video frames are better aligned with the predictive nature of human visual processing than representations of the observed video itself. To this end, we compare the alignment between human video-watching fMRI responses in the visual cortex and the internal representations from two types of video diffusion models, an autoregressive (AR) model and its non-AR base model. We first conduct a within-model analysis of the AR video diffusion model and show that the representations for future video generation align better with the visual cortex than the representations of the observed video. We then compare the internal representations of the AR model with those of its non-AR base model and again show that the representations for future video generation align better with the visual cortex than the representations for observed video reconstruction by the base model. Specifically, the alignment of observed video reconstruction is concentrated in lower-order visual cortex, whereas that of future video generation is concentrated in higher-order visual cortex. Finally, we show in a human behavioral experiment that humans prefer videos generated by amplifying the contributions of individual layers that align better with the visual cortex.
Chinese Translation
研究视觉模型的内部表征与视觉皮层对相同观察视觉刺激的反应之间的对齐，帮助我们更好地理解人类视觉处理。然而，迄今为止的研究在很大程度上忽视了人脑不仅处理观察到的视觉刺激，还基于已观察到的内容预测即将出现的刺激这一事实。因此，我们假设用于生成未来视频帧的内部表征与人类视觉处理的预测特性相比，优于对观察视频本身的表征。为此，我们比较了人类观看视频时视觉皮层的fMRI反应与两种视频扩散模型的内部表征之间的对齐情况：自回归（AR）模型及其非自回归基础模型。我们首先对AR视频扩散模型进行模型内分析，结果表明，未来视频生成的表征与视觉皮层的对齐优于观察视频的表征。接着，我们将AR模型的内部表征与其非AR基础模型的表征进行比较，再次表明，未来视频生成的表征与视觉皮层的对齐优于基础模型的观察视频重建表征。具体而言，观察视频重建的对齐主要集中在低阶视觉皮层，而未来视频生成的对齐则集中在高阶视觉皮层。最后，我们在一项人类行为实验中显示，人类更倾向于选择通过放大与视觉皮层更好对齐的各个层的贡献生成的视频。
cs.CV / 55 / 2609.38823
DecoMoE: Decoupling Visual Propagation and Expert Computation for Efficient Multimodal MoE Inference
DecoMoE：解耦视觉传播与专家计算以实现高效的多模态 MoE 推理
Abstract
Multimodal mixture-of-experts (MoE) models combine sparse expert activation with visual-language capabilities, yet their inference remains costly because long visual-token sequences repeatedly incur attention, routing, dispatch, and expert-MLP computation. Existing methods typically compress either the token or expert dimension, leaving redundancy along the other. Our analysis reveals two complementary regularities: the depth required for visual propagation varies across inputs, while text-token routing exhibits concentrated and recurrent expert-importance patterns. Based on these observations, we propose DecoMoE, a two-dimensional structured compression framework that decouples visual propagation from expert computation. The Sample-Adaptive Visual Boundary (SAVB) predicts an input-dependent visual-exit layer at which the visual-token block is removed. The Routing-Calibrated Expert Prefix (RCEP) reorders experts offline using text-token routed mass and, from this predicted exit layer onward, retains at each MoE layer the shortest contiguous prefix covering a target routed-mass fraction. We evaluate DecoMoE on Qwen3-VL-MoE and InternVL3.5-30B-A3B across six benchmarks. On Qwen3-VL-MoE, DecoMoE retains 97.91% of dense-baseline performance while reducing computation from 27.06 to 16.73 TFLOPs and latency from 0.44 to 0.26 seconds, yielding a 1.69x speedup. Code will be available at https://github.com/ShawnTan86/DecoMoE.
Chinese Translation
多模态专家混合（MoE）模型结合了稀疏的专家激活与视觉语言能力，但其推理过程仍然成本高昂，因为长视觉标记序列会反复产生注意力、路由、调度和专家 MLP 计算。现有方法通常压缩标记或专家维度，导致另一维度的冗余。我们的分析揭示了两个互补的规律：视觉传播所需的深度因输入而异，而文本标记路由则表现出集中且重复的专家重要性模式。基于这些观察，我们提出了 DecoMoE，一个二维结构化压缩框架，解耦视觉传播与专家计算。样本自适应视觉边界（SAVB）预测一个依赖输入的视觉退出层，在该层中移除视觉标记块。路由校准专家前缀（RCEP）使用文本标记路由的质量离线重新排序专家，并从预测的退出层开始，在每个 MoE 层保留覆盖目标路由质量分数的最短连续前缀。我们在 Qwen3-VL-MoE 和 InternVL3.5-30B-A3B 上对 DecoMoE 进行了评估，涵盖六个基准。在 Qwen3-VL-MoE 上，DecoMoE 保留了 97.91% 的密集基线性能，同时将计算从 27.06 降低到 16.73 TFLOPs，将延迟从 0.44 秒降低到 0.26 秒，实现了 1.69 倍的加速。代码将发布在 https://github.com/ShawnTan86/DecoMoE。
cs.CV / 56 / 2609.38839
FrameMorrow: Future-guided Frame Selection with Prospective Tokens for Long-Horizon Video Generation
FrameMorrow：基于前瞻性标记的未来引导帧选择用于长时间视频生成
Abstract
Long-horizon video generation requires models to effectively leverage an increasingly long generation history. As the generated history grows, retaining all previous content becomes increasingly expensive and redundant, making effective historical selection essential. Existing approaches often determine historical relevance based on the current content. However, information relevant to the present is not necessarily useful for future generation, while seemingly less relevant history may become important later. Our key insight is that historical information should be selected according to its relevance to future information needs. Capturing these needs does not require generating the full future; instead, a compact representation of what becomes important next is sufficient to guide historical selection. Building on this insight, we propose FrameMorrow, a prospective frame selector that predicts a small set of prospective tokens representing future information needs and uses them to identify relevant information from history. FrameMorrow selects explicit historical frames rather than model-specific internal states, enabling plug-and-play integration across diverse generators, including closed-source models, with little additional inference cost. We evaluate FrameMorrow across five benchmarks and 11 generative models spanning long-video generation, interactive generation, and action-conditioned world models. Extensive experiments demonstrate consistent improvements in long-range consistency, visual quality, and action alignment across diverse generation settings.
Chinese Translation
长时间视频生成要求模型有效利用日益增长的生成历史。随着生成历史的增加，保留所有先前内容变得越来越昂贵且冗余，因此有效的历史选择至关重要。现有方法通常基于当前内容来确定历史相关性。然而，与当前相关的信息不一定对未来生成有用，而看似不相关的历史信息可能在后续变得重要。我们的关键见解是，历史信息的选择应根据其与未来信息需求的相关性进行。捕捉这些需求并不需要生成完整的未来；相反，下一步重要内容的紧凑表示足以指导历史选择。基于这一见解，我们提出了FrameMorrow，一种前瞻性帧选择器，它预测一小组前瞻性标记，代表未来信息需求，并利用这些标记从历史中识别相关信息。FrameMorrow选择明确的历史帧，而不是模型特定的内部状态，使其能够在各种生成器中实现即插即用的集成，包括闭源模型，且额外推理成本极低。我们在五个基准和11个生成模型上评估了FrameMorrow，涵盖了长视频生成、交互生成和基于动作的世界模型。大量实验表明，在多种生成设置中，长距离一致性、视觉质量和动作对齐方面均有显著改善。
cs.CV / 57 / 2609.38856
Decoupling Spherical Reasoning from Dense Prediction for 360 Depth Estimation
将球面推理与密集预测解耦以进行360度深度估计
Abstract
The equirectangular projection (ERP) is widely used for panoramic depth estimation, but its spatially varying distortion makes geometry-consistent feature modeling challenging. We revisit panoramic depth estimation by decoupling contextual modeling in native spherical space from dense ERP prediction. To this end, we propose a Fibonacci Spherical Graph (FSG) as an intermediate reasoning space to lift ERP features onto quasi-uniform Fibonacci nodes on the sphere and capture local and long-range dependencies through complementary spherical neighborhoods. The resulting spherical discretization distributes graph nodes approximately uniformly over the spherical surface, reducing the over-representation of highly stretched regions during relational modeling. Operating on a compact set of Fibonacci nodes also avoids the computational burden of constructing and processing a graph at full ERP resolution. To bridge spherical reasoning and dense prediction, we propose a Spherical Context Conditioning (SCC) module that adaptively modulates dense ERP features with the enhanced spherical representation, allowing spherical context to guide pixel-aligned depth prediction. Extensive experiments on three benchmarks demonstrate that the proposed method consistently achieves superior depth accuracy over existing approaches.
Chinese Translation
等矩形投影（ERP）广泛用于全景深度估计，但其空间变化的失真使得几何一致的特征建模变得具有挑战性。我们通过将原生球面空间中的上下文建模与密集ERP预测解耦，重新审视全景深度估计。为此，我们提出了一种斐波那契球面图（Fibonacci Spherical Graph, FSG）作为中间推理空间，将ERP特征提升到球面上的准均匀斐波那契节点，并通过互补的球面邻域捕捉局部和长程依赖关系。所得到的球面离散化使图节点在球面上大致均匀分布，从而减少了关系建模过程中高度拉伸区域的过度表示。在一组紧凑的斐波那契节点上操作也避免了在全ERP分辨率下构建和处理图的计算负担。为了连接球面推理和密集预测，我们提出了一种球面上下文调节（Spherical Context Conditioning, SCC）模块，该模块通过增强的球面表示自适应地调节密集ERP特征，使球面上下文能够引导像素对齐的深度预测。在三个基准上的大量实验表明，所提方法在深度准确性上始终优于现有方法。
cs.CV / 58 / 2609.38864
AdaOcc: Adaptive 3D Occupancy Prediction for Embodied Tasks
AdaOcc：用于具身任务的自适应三维占用预测
Abstract
Embodied tasks demand accurate, flexible, and semantically rich 3D scene representations. 3D semantic occupancy is well suited to this requirement, as it can model holistic 3D spaces by encoding geometric occupancy along with semantic categories. However, existing occupancy prediction methods struggle to meet practical deployment requirements, such as adapting to varying computing budgets, sensor setups, and observation views. In this paper, we propose a point-based Adaptive 3D Occupancy Prediction method, called AdaOcc, tailored for embodied scenarios. To accommodate heterogeneous sensor inputs, AdaOcc uses an adaptive geometry-guided dual-branch encoder that can support RGB images in various numbers of views with (estimated) depth maps or LiDAR scans. AdaOcc represents occupied regions via sparse semantic points trained with a progressive query learning strategy, allowing the prediction computational budget to be flexibly adjusted through query point numbers and decoder layers. To facilitate high-fidelity geometric modeling for lightweight point-based occupancy learning, we further propose a novel containment loss that regularizes predicted points to reside within valid occupied regions. Extensive experiments show that our method achieves a new state-of-the-art on Occ-ScanNet with considerable performance improvements over previous methods. Moreover, our framework demonstrates strong practical applicability as an adaptive 3D perception module in real-world embodied systems.
Chinese Translation
具身任务需要准确、灵活且语义丰富的三维场景表示。三维语义占用非常适合这一要求，因为它可以通过编码几何占用和语义类别来建模整体三维空间。然而，现有的占用预测方法在满足实际部署需求方面面临挑战，例如适应不同的计算预算、传感器设置和观察视角。本文提出了一种基于点的自适应三维占用预测方法，称为AdaOcc，专为具身场景量身定制。为了适应异构传感器输入，AdaOcc采用了一种自适应几何引导的双分支编码器，能够支持具有（估计）深度图或LiDAR扫描的不同视角的RGB图像。AdaOcc通过使用逐步查询学习策略训练稀疏语义点来表示占用区域，从而允许通过查询点数量和解码器层灵活调整预测的计算预算。为了促进轻量级基于点的占用学习的高保真几何建模，我们进一步提出了一种新颖的包含损失，规范化预测点位于有效占用区域内。大量实验表明，我们的方法在Occ-ScanNet上达到了新的最先进水平，相较于之前的方法有显著的性能提升。此外，我们的框架作为现实世界具身系统中的自适应三维感知模块，展示了强大的实际应用能力。
cs.CV / 59 / 2609.38900
MEMO: Multi-Level Entity-Aware Memory for Streaming Video Understanding
MEMO：用于流媒体视频理解的多层次实体感知记忆
Abstract
Streaming video understanding requires models to process unbounded visual streams while preserving rich visual semantics across vast temporal horizons, posing a fundamental challenge for memory modeling. Existing approaches primarily focus on increasing memory capacity, either by compressing historical information into fixed-size representations or by extending storage beyond GPU memory. However, these methods largely rely on global or coarse-grained representations, inevitably losing fine-grained visual information. In this work, we argue that streaming video memory should explicitly encode structured and semantically meaningful representations, particularly at the entity level. To this end, we propose MEMO, a novel framework that models streaming video through multi-level, entity-aware structured memory. MEMO performs multi-level perception to jointly capture global semantics, entity dynamics, and spatial structures, partitioning streaming video into semantically coherent chunks. Each chunk is organized into a structured memory, where lightweight global and entity-level representations serve as retrieval indices, while the corresponding high-resolution visual content is retained separately for on-demand access. At inference time, MEMO performs query-specific retrieval over the structured memory and selectively recalls relevant visual evidence for downstream reasoning. Notably, MEMO is training-free and plug-and-play with existing multimodal large language models. Extensive experiments on StreamingBench and OVO-Bench demonstrate that MEMO consistently improves multiple base models and achieves state-of-the-art performance.
Chinese Translation
流媒体视频理解要求模型处理无限制的视觉流，同时在广泛的时间范围内保持丰富的视觉语义，这对记忆建模提出了根本性的挑战。现有的方法主要集中在增加记忆容量，要么通过将历史信息压缩为固定大小的表示，要么通过扩展存储超出GPU内存。然而，这些方法在很大程度上依赖于全局或粗粒度的表示，必然会丢失细粒度的视觉信息。在本研究中，我们认为流媒体视频记忆应明确编码结构化和语义上有意义的表示，特别是在实体层面。为此，我们提出了MEMO，一个通过多层次、实体感知的结构化记忆来建模流媒体视频的新框架。MEMO执行多层次感知，以共同捕捉全局语义、实体动态和空间结构，将流媒体视频划分为语义一致的块。每个块被组织成一个结构化记忆，其中轻量级的全局和实体级表示作为检索索引，而相应的高分辨率视觉内容则单独保留以供按需访问。在推理时，MEMO在结构化记忆上执行查询特定的检索，并选择性地回忆相关的视觉证据以进行下游推理。值得注意的是，MEMO是无训练的，并且可以与现有的多模态大型语言模型即插即用。在StreamingBench和OVO-Bench上的大量实验表明，MEMO始终提高多个基础模型的性能，并实现了最先进的表现。
cs.CV / 60 / 2609.38913
FLOW: Feature-Level Optimal Warping for Generalized Remote Physiological Measurement
FLOW：用于广义远程生理测量的特征级最优扭曲
Abstract
Remote photoplethysmography (rPPG) enables non-contact physiological measurement but remains vulnerable to domain shifts from illumination, motion, and sensors. We propose \textbf{FLOW (Feature-Level Optimal Warping)}, an \emph{optimal transport--driven} framework for domain-generalized rPPG. FLOW integrates a \textbf{Temporal Refinement Module (TRM)} to stabilize temporal dynamics and a \textbf{Prototype-based Cross-Temporal Optimal Transport (PCOT)} module to achieve domain-invariant alignment via learnable prototypes.Beyond feature alignment, FLOW employs soft cross-temporal correspondence modeling that aligns temporal features in a flexible manner, allowing the model to respect and preserve the intrinsic rhythmic patterns of physiological signals. Moreover, the lightweight design of our modules allows seamless integration into existing end-to-end rPPG architectures without additional preprocessing. Two regularization terms further enforce source consistency and identity preservation. Theoretically, we derive a generalization bound under conditional optimal transport. Extensive experiments across four rPPG benchmarks show that FLOW achieves state-of-the-art cross-domain performance with lightweight design and strong physiological fidelity.
Chinese Translation
远程光电容积描记法（rPPG）能够实现非接触式生理测量，但仍然容易受到照明、运动和传感器等领域转移的影响。我们提出了 extbf{FLOW（特征级最优扭曲）}，这是一个基于 extit{最优传输驱动}的领域广义rPPG框架。FLOW集成了 extbf{时间精炼模块（TRM）}以稳定时间动态，并且采用 extbf{基于原型的跨时间最优传输（PCOT）}模块，通过可学习的原型实现领域不变的对齐。除了特征对齐，FLOW还采用了柔性跨时间对应建模，以灵活的方式对齐时间特征，使模型能够尊重和保留生理信号的内在节律模式。此外，我们模块的轻量化设计使其能够无缝集成到现有的端到端rPPG架构中，而无需额外的预处理。两个正则化项进一步强化了源一致性和身份保留。从理论上讲，我们在条件最优传输下推导了一个泛化界限。在四个rPPG基准测试中的大量实验表明，FLOW在轻量化设计和强生理保真度的基础上，实现了最先进的跨领域性能。
cs.CV / 61 / 2609.38924
From Image Interpretation to Clinical Reasoning: Upstream Physician-Context-Aware Multimodal Learning with Causal Reinforcement Learning
从图像解读到临床推理：基于因果强化学习的上游医生情境感知多模态学习
Abstract
Major adverse cardiovascular events (MACE) remain the leading cause of mortality worldwide. Opportunistic screening using routinely acquired clinical data offers a scalable approach for identifying high-risk individuals before acute events occur. Although chest X-rays (CXRs) capture latent cardiovascular biomarkers and clinical histories provide complementary patient context, existing medical vision-language models are primarily optimized for radiology interpretation rather than prognostic reasoning. We propose a causal reinforcement learning framework for multimodal clinical reasoning that integrates CXRs and physician-authored clinical histories for opportunistic MACE prediction. The framework introduces (1) a role-decoupled dual-LLM architecture that separates reasoning from risk prediction, (2) a dual-action causal reinforcement learning policy for evidence selection and reasoning optimization, and (3) causal token pruning to learn compact multimodal representations. Evaluated on an internal cohort, an emergency department cohort, and the external MIMIC dataset, the proposed framework consistently outperformed unimodal baselines and state-of-the-art medical vision-language models, achieving AUROCs of 0.720, 0.760, and 0.845, respectively. It also substantially improved reasoning quality, achieving higher GREEN scores and higher expert preference while maintaining robust predictive performance across diverse patient populations.
Chinese Translation
重大不良心血管事件（MACE）仍然是全球主要的死亡原因。利用常规获取的临床数据进行机会筛查为在急性事件发生之前识别高风险个体提供了一种可扩展的方法。尽管胸部X光片（CXR）捕捉潜在的心血管生物标志物，而临床历史提供了补充的患者背景，但现有的医学视觉-语言模型主要针对放射学解读而非预后推理进行了优化。我们提出了一种因果强化学习框架，用于多模态临床推理，整合CXR和医生撰写的临床历史，以进行机会性MACE预测。该框架引入了（1）角色解耦的双LLM架构，将推理与风险预测分离，（2）用于证据选择和推理优化的双动作因果强化学习策略，以及（3）因果令牌剪枝以学习紧凑的多模态表示。在内部队列、急诊科队列和外部MIMIC数据集上的评估表明，所提出的框架在性能上始终优于单模态基线和最先进的医学视觉-语言模型，分别达到了0.720、0.760和0.845的AUROC。同时，它显著提高了推理质量，实现了更高的GREEN评分和更高的专家偏好，同时在不同患者群体中保持了稳健的预测性能。
cs.CV / 62 / 2609.38930
On the Relaxation of Conditional Independence Assumption for Image Segmentation
关于图像分割中条件独立假设的放松
Abstract
In semantic segmentation, a recent line of RankSEG methods directly optimizes Dice/IoU scores at inference time, improving alignment with evaluation metrics without modifying model training. Despite its theoretical and empirical success, RankSEG relies on the restrictive Conditional Independence Assumption (CIA), which ignores crucial label correlations and therefore degrades performance in ambiguous or low-contrast scenarios. However, accounting for full label dependence is computationally prohibitive, requiring $\mathcal{O}(d^3)$ time. To address this, we replace the CIA with a Spatially Localized Dependence (SLD) structure that captures local label correlations while keeping the dependence model tractable. We further overcome the remaining computational bottleneck via a Reciprocal Moment Approximation coupled with a novel fixed-point optimization strategy that eliminates exhaustive search. The proposed algorithm achieves a highly practical $\mathcal{O}(d \log d)$ complexity and consistently outperforms conventional argmax and CIA-based RankSEG across diverse segmentation benchmarks. Improvements are significant in low-contrast or small-object scenarios, where label dependence offers valuable signals complementary to image information for accurate segmentation. The code of experiments is available at https://github.com/ZixunWang/RankSEG-DEP.
Chinese Translation
在语义分割中，最近一系列RankSEG方法在推理时直接优化Dice/IoU分数，从而提高了与评估指标的一致性，而无需修改模型训练。尽管在理论和实证上取得了成功，RankSEG依赖于限制性的条件独立假设（CIA），该假设忽略了重要的标签相关性，因此在模糊或低对比度场景中性能下降。然而，考虑完整的标签依赖性在计算上是不可行的，需要$ ext{O}(d^3)$的时间。为了解决这个问题，我们用空间局部依赖（SLD）结构替代CIA，该结构捕捉局部标签相关性，同时保持依赖模型的可处理性。我们进一步通过一种互惠矩估计（Reciprocal Moment Approximation）和一种新颖的固定点优化策略克服了剩余的计算瓶颈，消除了穷举搜索。所提出的算法实现了高度实用的$ ext{O}(d ext{log} d)$复杂度，并在各种分割基准测试中始终优于传统的argmax和基于CIA的RankSEG。在低对比度或小物体场景中，改进尤为显著，因为标签依赖性提供了与图像信息互补的有价值信号，以实现准确的分割。实验代码可在https://github.com/ZixunWang/RankSEG-DEP获取。
cs.CV / 63 / 2609.38968
Beyond Spatial-Domain Supervision: A Relation Constrained Space for Multi-Modal Image Fusion
超越空间域监督：多模态图像融合的关系约束空间
Abstract
Multi-modal image fusion (MMIF) aims to form a single image by integrating shared information, preserving complementary cues, and coordinating cross-modal conflicts across modalities. However, due to the absence of ground-truth fused images, existing MMIF supervision commonly uses spatial-domain sources or gradient variants as surrogate ground truth, making the supervision mechanism inherently misaligned with the goal of MMIF and causing pixel-level compromise or modality bias. To address this, we propose a relation-constrained supervision paradigm that moves fusion supervision from the spatial domain to a learned relation space. Rather than relying solely on direct source approximation, we further leverage frozen pretrained representation models as information providers and design a learnable feature adapter to align heterogeneous DINO and CLIP features into a unified supervision space. The adapter infers three relation parameters, namely sharedness, dominance, and coordination radius, which define three losses corresponding to the MMIF's goal. To make this space reliable, we devise a self-supervised contrastive ranking objective tailored to the adapter and couple it with the fusion network through alternating optimization. Extensive experiments show that the proposed supervision space yields significant gains regardless of which mainstream backbone the fusion network adopts, offering a supervision paradigm better aligned with the goal of MMIF. Code: github.com/GMY628/RCS-Fusion.
Chinese Translation
多模态图像融合（MMIF）旨在通过整合共享信息、保留互补线索以及协调跨模态冲突来形成单一图像。然而，由于缺乏真实的融合图像，现有的MMIF监督通常使用空间域源或梯度变体作为替代真实值，这使得监督机制与MMIF的目标本质上不一致，导致像素级妥协或模态偏差。为了解决这个问题，我们提出了一种关系约束监督范式，将融合监督从空间域转移到学习的关系空间。我们不仅依赖于直接源近似，还进一步利用冻结的预训练表示模型作为信息提供者，并设计一个可学习的特征适配器，将异构的DINO和CLIP特征对齐到统一的监督空间。该适配器推断出三个关系参数，即共享性、主导性和协调半径，这定义了与MMIF目标相对应的三种损失。为了使这个空间可靠，我们设计了一个针对适配器的自监督对比排序目标，并通过交替优化将其与融合网络结合。大量实验表明，所提出的监督空间无论融合网络采用何种主流骨干，都能显著提高性能，提供了一个与MMIF目标更为一致的监督范式。代码：github.com/GMY628/RCS-Fusion。
cs.CV / 64 / 2609.38978
PARK: Accurate Block Retrieval for Sparse Attention in Video Diffusion Transformers
PARK：视频扩散变换器中稀疏注意力的准确块检索
Abstract
Diffusion Transformers (DiTs) have become a dominant architecture for video generation, but their efficiency is limited by the quadratic complexity of full attention. Sparse attention reduces this cost by retrieving important blocks and computing attention only within them, but inaccurate retrieval can either degrade generation quality or yield unnecessary computation. We identify two retrieval mismatches in methods that retrieve blocks using the averaged representations of query and key blocks: (i) query-side aggregation mismatch, where averaging queries before Softmax fails to preserve their individual attention preferences, and (ii) key-side clustering metric mismatch, where standard Euclidean clustering in the original key space can group keys with dissimilar QK scores under the current query, so their average representation may not accurately represent how the current query scores individual keys. These mismatches can lead to inaccurate block retrieval. To address these mismatches, we propose PARK, a training-free sparse attention method for accurate block retrieval. PARK retains every original query, independently normalizes its attention over key blocks, and then averages these distributions within each query block. It also uses information from the current queries to transform keys before clustering, so that keys receiving similar QK scores are grouped together. A fused GPU kernel further reduces the overhead of block retrieval. Experiments on HunyuanVideo and Wan demonstrate that PARK improves block retrieval accuracy and preserves generation quality while accelerating inference, achieving the best quality-efficiency trade-off among the compared sparse attention methods.
Chinese Translation
扩散变换器（DiTs）已成为视频生成的主导架构，但其效率受到全注意力的二次复杂度的限制。稀疏注意力通过检索重要块并仅在其中计算注意力来降低这一成本，但不准确的检索可能会降低生成质量或导致不必要的计算。我们识别出在使用查询块和键块的平均表示进行块检索的方法中存在两种检索不匹配：（i）查询侧聚合不匹配，在Softmax之前对查询进行平均会导致其个体注意力偏好无法得到保留；（ii）键侧聚类度量不匹配，在原始键空间中标准的欧几里得聚类可能会将当前查询下得分不同的键聚集在一起，因此它们的平均表示可能无法准确反映当前查询对个别键的得分。这些不匹配可能导致不准确的块检索。为了解决这些不匹配，我们提出了PARK，一种无训练的稀疏注意力方法，用于准确的块检索。PARK保留每个原始查询，独立地对其在键块上的注意力进行归一化，然后在每个查询块内对这些分布进行平均。它还利用当前查询的信息在聚类之前对键进行变换，以便将接收相似QK得分的键聚集在一起。融合的GPU内核进一步减少了块检索的开销。在HunyuanVideo和Wan上的实验表明，PARK提高了块检索的准确性，保持了生成质量，同时加速了推理，在比较的稀疏注意力方法中实现了最佳的质量-效率权衡。
cs.CV / 65 / 2609.38979
Mitigating Object Hallucination in Large Vision-Language Models via False Discovery Controlled Visual Data Splitting
通过控制假发现的视觉数据分割来减轻大型视觉-语言模型中的物体幻觉
Abstract
Multiple object hallucination, where large vision-language models (LVLMs) generate objects not supported by the visual input, is a persistent challenge caused by visual uncertainty during decoding. Existing methods reduce hallucinations using contrastive signals, but they rely on heuristics and lack principled control of false positives at the image level. To address this, we propose False Discovery Rate-COntRol of HALlucination (CORAL), a training-free framework that models visual uncertainty using an uncertainty-aware visual data splitting strategy and leverages mirror statistics to quantify visual contrast during decoding. By computing mirror statistics from paired, symmetrically perturbed visual inputs, CORAL estimates spurious object predictions and sets a data-driven threshold to control the expected fraction of false discoveries per image, suppressing hallucinations while retaining high power for truly grounded objects. The framework is flexible, supports multiple LVLMs, and mitigates hallucinations without retraining or supervision. Extensive experiments on multiple benchmarks with several evaluation metrics demonstrate that CORAL consistently outperforms state-of-the-art methods, providing more reliable and robust hallucination control. Code is available at: https://changliu1993-cl.github.io/CORAL/
Chinese Translation
多物体幻觉是指大型视觉-语言模型（LVLMs）生成与视觉输入不符的物体，这一问题是由于解码过程中的视觉不确定性引起的。现有方法通过对比信号来减少幻觉，但它们依赖于启发式方法，缺乏对图像级假阳性的原则性控制。为了解决这一问题，我们提出了假发现率控制幻觉（CORAL），这是一个无训练框架，通过不确定性感知的视觉数据分割策略来建模视觉不确定性，并利用镜像统计量在解码过程中量化视觉对比。通过从成对的、对称扰动的视觉输入中计算镜像统计量，CORAL 估计虚假物体预测，并设定数据驱动的阈值以控制每幅图像的预期假发现比例，从而抑制幻觉，同时保持对真实基础物体的高检出率。该框架灵活，支持多种 LVLM，并在不重新训练或监督的情况下减轻幻觉。在多个基准测试和多个评估指标上的广泛实验表明，CORAL 始终优于最先进的方法，提供更可靠和稳健的幻觉控制。代码可在以下链接获取：https://changliu1993-cl.github.io/CORAL/
cs.CV / 66 / 2609.38985
MeshOctave generates meshes via cascading resolution transitions
MeshOctave通过级联分辨率转换生成网格
Abstract
Generating compact, artist-style meshes with explicit topology typically relies on autoregressive models which incur prohibitive sequential per-token costs, or continuous flow models that depend on heuristic connectivity decoders. Next-scale generation paradigms offer a compelling alternative by enabling parallel intra-scale token prediction and coarse-to-fine refinement from global structure to local topology; yet, existing methods derive hierarchical scales via progressive mesh simplification and invert them sequentially. This eliminates intra-scale parallelism and scales generation steps linearly with face count. In this paper, we propose MeshOctave, which instead defines scale through dyadic spatial grid resolutions, framing coarsening as a deterministic collapse that merges vertices sharing a voxel cell and inherits connectivity. Its inverse operation, split-and-rewire, determines which octant sub-vertices are instantiated for each coarse face and resolves local connectivity using discrete structural tokens. These per-face operations require no serialization, each scale transition is modeled as an unordered set that adds one bit of coordinate precision, naturally supporting dynamic-length meshes and adaptive resolution refinement. We construct a scale-conditioned masked-uniform discrete diffusion model to learn split-and-rewire operation from resolution collapse hierarchies. MeshOctave outperforms strong baselines in geometric fidelity and topological validity by a non-trivial margin, while supporting adaptive resolution refinement and extending naturally to mesh subdivision tasks.
Chinese Translation
生成具有明确拓扑的紧凑艺术风格网格通常依赖于自回归模型，这些模型在每个标记的序列生成中会产生高昂的成本，或者依赖于启发式连接解码器的连续流模型。下一尺度生成范式通过实现并行的内部尺度标记预测以及从全局结构到局部拓扑的粗到细的细化，提供了一个引人注目的替代方案；然而，现有方法通过渐进的网格简化推导出层次尺度，并按顺序反转它们。这消除了内部尺度的并行性，并使得生成步骤与面数线性相关。在本文中，我们提出了MeshOctave，它通过二元空间网格分辨率定义尺度，将粗化框架化为一种确定性的合并过程，该过程合并共享体素单元的顶点并继承连接性。其逆操作“分裂与重连”确定每个粗面实例化哪些八分体子顶点，并使用离散结构标记解析局部连接性。这些每面操作不需要序列化，每个尺度转换被建模为一个无序集合，增加一个坐标精度位，自然支持动态长度网格和自适应分辨率细化。我们构建了一个尺度条件掩蔽均匀离散扩散模型，以从分辨率崩溃层次中学习分裂与重连操作。MeshOctave在几何保真度和拓扑有效性方面以显著的优势超越了强基线，同时支持自适应分辨率细化，并自然扩展到网格细分任务。
cs.CV / 67 / 2609.39004
When Integral Meets Decomposition: A Signal-Level Self-Supervised Feature Decompose Paradigm for Multi-Modal Image Fusion
当积分遇上分解：一种信号级自监督特征分解范式用于多模态图像融合
Abstract
Multimodal image fusion (MMIF) aims to integrate complementary information from different modalities into a high-quality fused image and support downstream tasks. Recently, feature decomposition has become an important paradigm by separating source images into common and modality-specific unique features. However, existing methods lack clear supervision because ground-truth (GT) decomposition feature maps are unavailable. They usually combine multiple image-level metrics as losses, which are inherently incomplete and may conflict since each pixel couples attributes such as texture, edge, and contour. To address this, we propose a 1D signal-level self-supervised feature decomposition paradigm. Our core insight is to reformulate feature decomposition from unclear 2D image-level supervision into an integral-driven 1D signal-level optimization problem. This objective-level reformulation uses the 1D signal form to compute the integral constraint. The decomposer is optimized by the integral area between common and original signals, enabling more stable optimization with a clear optimization objective. Our model follows a two-stage SSL framework. Stage I designs dual pretext tasks for integral-driven decomposition at the signal level and structure-preserving reconstruction at the image level. Stage II fuses unique features and combines them with common features to reconstruct the fused image. Experiments on representative MMIF tasks show state-of-the-art (SOTA) performance. Code: github.com/Wangjiayu0512/SIDFusion.
Chinese Translation
多模态图像融合（MMIF）旨在将来自不同模态的互补信息整合为高质量的融合图像，并支持后续任务。近年来，特征分解已成为一种重要的范式，通过将源图像分离为公共特征和模态特定的独特特征。然而，现有方法缺乏明确的监督，因为真实值（GT）分解特征图不可用。它们通常将多个图像级别的指标组合为损失，这在本质上是不完整的，并且可能存在冲突，因为每个像素耦合了纹理、边缘和轮廓等属性。为了解决这个问题，我们提出了一种一维信号级自监督特征分解范式。我们的核心见解是将特征分解从不明确的二维图像级监督重新表述为一个积分驱动的一维信号级优化问题。这个目标级别的重构利用一维信号形式来计算积分约束。分解器通过公共信号和原始信号之间的积分区域进行优化，从而实现更稳定的优化，并具有明确的优化目标。我们的模型遵循一个两阶段的自监督学习（SSL）框架。第一阶段设计了双重前置任务，以实现信号级的积分驱动分解和图像级的结构保持重建。第二阶段融合独特特征，并将其与公共特征结合以重建融合图像。在代表性的MMIF任务上的实验显示了最先进的（SOTA）性能。代码：github.com/Wangjiayu0512/SIDFusion。
cs.CV / 68 / 2609.39021
Frame Differential On-Policy Self-Distillation for Video Reasoning
视频推理中的帧差分在线自蒸馏
Abstract
Reinforcement learning (RL) has substantially improved the reasoning ability of multimodal language models through verifiable rewards and increasingly fine-grainedvisual or temporal credit assignment. In video reasoning, however, current RL methods typically train with a fixed sparse frame budget: increasing the number of frames makes autoregressive rollouts expensive, while too few frames may miss temporally localized events and fine-grained visual details. We present \textbf{Frame Differential On-Policy Self-Distillation (FD-OPSD)}, which transfers the useful evidence of dense frame observations to a sparse frame policy during RL training. FD-OPSD compares the policy's token level preferences for the same sampled response under sparse and dense views, and distills the resulting frame differential signal without an external teacher or dense autoregressive rollout. The method preserves sparse-frame rollouts and leaves inference unchanged. Across Qwen2.5-VL-7B and Qwen3-VL-4B on six video reasoning benchmarks, FD-OPSD yields higher overall average performance than the strongest corresponding GRPO, T-GRPO, or Video-KTR baselines across the 16, 32, and 64 frame evaluation settings. These results show that dense visual evidence can be transferred selectively during training through token level self-distillation while retaining sparse frame rollouts and unchanged inference.
Chinese Translation
强化学习（RL）通过可验证的奖励和日益细粒度的视觉或时间信用分配，显著提升了多模态语言模型的推理能力。然而，在视频推理中，当前的 RL 方法通常使用固定的稀疏帧预算进行训练：增加帧数会使自回归展开变得昂贵，而帧数过少可能会错过时间局部事件和细粒度的视觉细节。我们提出了 extbf{帧差分在线自蒸馏（Frame Differential On-Policy Self-Distillation, FD-OPSD）}，该方法在 RL 训练过程中将稠密帧观测的有用证据转移到稀疏帧策略中。FD-OPSD 比较了策略在稀疏和稠密视图下对同一采样响应的令牌级偏好，并在没有外部教师或稠密自回归展开的情况下提炼出结果帧差分信号。该方法保留了稀疏帧展开，并保持推理不变。在 Qwen2.5-VL-7B 和 Qwen3-VL-4B 的六个视频推理基准上，FD-OPSD 在 16、32 和 64 帧评估设置中，整体平均性能均高于最强的对应 GRPO、T-GRPO 或 Video-KTR 基线。这些结果表明，在训练过程中，可以通过令牌级自蒸馏选择性地转移稠密视觉证据，同时保留稀疏帧展开和不变的推理。
cs.CV / 69 / 2609.39024
Persistent Watermarking of Text-to-Image Models
文本到图像模型的持久水印技术
Abstract
Text-to-image (T2I) generation is gaining increasing popularity with the general public, motivating the development of reliable mechanisms for copyrighting such models given their expensive training costs. An adversary may obtain and reuse a pretrained T2I model without authorization, and then serve a modified version through an API service. Such modifications may arise from ordinary downstream adaptation or deliberate attempts to erase ownership, including input-prompt preprocessing, model fine-tuning, and output post-processing. From the model owner's perspective, a key challenge is therefore to embed trigger data that remain persistent under such changes while preserving the model's normal image-generation capabilities. In this work, we propose a contrastive-style watermarking objective with a term that explicitly encourages the watermarked model to behave differently from the original model on trigger inputs. Experiments show substantially stronger trigger-data persistence than prior methods across a wide range of downstream modifications and deliberate attempts to weaken the watermark, resulting in higher detection rates, often approaching 100% TPR@FPR<$10^{-4}$.
Chinese Translation
文本到图像（T2I）生成在公众中越来越受欢迎，这促使开发可靠的机制来保护这些模型的版权，因为它们的训练成本昂贵。对手可能在未经授权的情况下获取并重用预训练的 T2I 模型，然后通过 API 服务提供修改版本。这种修改可能源于普通的下游适配或故意试图抹去所有权的行为，包括输入提示的预处理、模型微调和输出后处理。因此，从模型所有者的角度来看，一个关键挑战是嵌入在这些变化下仍然持久的触发数据，同时保持模型的正常图像生成能力。在本研究中，我们提出了一种对比风格的水印目标，其中包含一个明确鼓励水印模型在触发输入上表现出与原始模型不同的项。实验结果表明，在广泛的下游修改和故意削弱水印的尝试中，触发数据的持久性明显强于先前的方法，导致更高的检测率，通常接近 100% 的真阳性率（TPR）在假阳性率（FPR）小于 $10^{-4}$ 的情况下。
cs.CV / 70 / 2609.39033
TED:Text-Axis Evidence Decomposition for Prompted Anomaly Localization
TED：基于文本轴的证据分解用于提示异常定位
Abstract
CLIP is a powerful vision-language model, but it was not designed for fine-grained defect localization; CLIP-based anomaly detectors therefore adapt it with prompts or lightweight modules to increase defect sensitivity. We show that stronger sensitivity does not necessarily make local evidence reliable: under domain shift, adapted CLIP-AD models often assign high anomaly scores to both true defects and visually complex normal regions. The issue is not simply missing defect information, but a local scoring rule that decodes defect and hard-normal evidence, having the same anomaly evidence. We propose TED (Text-Axis Evidence Decomposition), a post-hoc scoring method that asks whether each ambiguous response is better supported by source defect patches or by source normal patches mistaken as anomalous. TED compares these supports under the host's normal-versus-anomaly text response, leaves the backbone and prompts unchanged, and requires no target-domain training. It works as a train-free score for raw VLM backbones or as a source-calibrated residual correction for adapted CLIP-AD hosts. Across frozen VLM backbones, TED substantially improves pixel-level localization over raw prompt similarity; across adapted hosts, it improves most pixel-level settings over P-AUROC, P-PRO, and P-AP. Gains are largest under stronger hard-FP competition, with mean localization gain increasing from +5.0 in low-competition regimes to about +10.9 in mid/high-competition regimes. These results suggest that recoverable defect evidence can already exist in pretrained multimodal representations, but reliable localization requires decoding it against hard-normal competitors. Code will be released at TED GitHub repository.
Chinese Translation
CLIP 是一个强大的视觉-语言模型，但并未针对细粒度缺陷定位进行设计；因此，基于 CLIP 的异常检测器通过提示或轻量级模块对其进行调整，以提高缺陷敏感性。我们表明，增强的敏感性并不一定使局部证据可靠：在领域转移下，调整后的 CLIP-AD 模型通常会对真实缺陷和视觉上复杂的正常区域分配高异常分数。问题不仅在于缺失的缺陷信息，还在于一种局部评分规则，该规则解码缺陷和被误认为异常的难正常证据，二者具有相同的异常证据。我们提出了 TED（文本轴证据分解），这是一种事后评分方法，旨在判断每个模糊响应是更好地由源缺陷补丁支持，还是由被误认为异常的源正常补丁支持。TED 在主机的正常与异常文本响应下比较这些支持，保持主干和提示不变，并且不需要目标领域的训练。它可以作为原始 VLM 主干的无训练评分，或作为调整后的 CLIP-AD 主机的源校准残差修正。在冻结的 VLM 主干中，TED 显著改善了原始提示相似性的像素级定位；在调整后的主机中，它在大多数像素级设置上改善了 P-AUROC、P-PRO 和 P-AP。增益在更强的硬假阳性竞争下最大，平均定位增益从低竞争状态下的 +5.0 增加到中/高竞争状态下的约 +10.9。这些结果表明，可恢复的缺陷证据可能已经存在于预训练的多模态表示中，但可靠的定位需要将其解码与难正常竞争者进行比较。代码将在 TED GitHub 仓库发布。
cs.CV / 71 / 2609.39047
BadAction: Backdoor Attacks on Interactive Video Generation via Action-Guided Triggers
BadAction：通过动作引导触发器对交互式视频生成进行后门攻击
Abstract
Interactive video generation (IVG) models have achieved remarkable progress in producing controllable visual content guided by user-defined actions, yet their security vulnerabilities remain largely unexplored. In this paper, we present the first systematic study of backdoor attacks against the interactivity of IVG models. Based on this attack surface, we propose BadAction, which leverages action-guided triggers to achieve the attack. Specifically, BadAction implants predefined motion patterns into the action sequences of backdoor samples and associates them with a static target video. Once triggered, the backdoored model generates frozen future frames that no longer respond to subsequent user actions, while preserving normal behavior on benign action sequences. In addition, we explore a stealthier attack in which multimodal triggers jointly poison action, text, and image inputs. Experiments show that BadAction achieves average attack success rates of 91.0% with action-only triggers and 80.4% with multimodal triggers. Moreover, extensive defense evaluations show that BadAction successfully bypasses existing backdoor detection methods, revealing a critical security gap in the interactive video generation pipeline. Project page: https://wsad55.github.io/badaction01/.
Chinese Translation
交互式视频生成（IVG）模型在生成由用户定义的动作引导的可控视觉内容方面取得了显著进展，但其安全漏洞仍然 largely 未被探索。在本文中，我们首次系统性地研究了针对 IVG 模型交互性的后门攻击。基于这一攻击面，我们提出了 BadAction，它利用动作引导触发器来实现攻击。具体而言，BadAction 将预定义的运动模式植入后门样本的动作序列中，并将其与静态目标视频关联。一旦触发，后门模型生成的未来帧将不再响应后续用户动作，同时在良性动作序列上保持正常行为。此外，我们探索了一种更隐蔽的攻击方式，其中多模态触发器共同污染动作、文本和图像输入。实验表明，BadAction 在仅使用动作触发器时的平均攻击成功率为 91.0%，而使用多模态触发器时为 80.4%。此外，广泛的防御评估表明，BadAction 成功绕过现有的后门检测方法，揭示了交互式视频生成流程中的一个关键安全漏洞。项目页面：https://wsad55.github.io/badaction01/
cs.CV / 72 / 2609.39051
TSMD: Temporal-Stream Modality Dropout for Robust Video Highlight Detection
TSMD：用于鲁棒视频亮点检测的时间流模态丢弃
Abstract
Existing multimodal video highlight detectors typically assume that visual, audio, and textual streams are continuously available. In practice, however, inputs may suffer from localized frame missingness or complete-stream outage. We formulate this robustness challenge along two dimensions: temporal missingness, where frames are missing independently in each modality, and stream-level missingness, where one modality is unavailable throughout a video. Moreover, we find that the mean squared error (MSE) loss is misaligned with both the evaluation metrics and the peak-driven nature of highlights. Therefore, we propose Temporal-Stream Modality Dropout (TSMD), which combines structured missingness simulation with a joint objective comprising pointwise MSE, per-video Pearson correlation, and peak-oriented RankNet loss terms. TSMD has three variants: temporal, stream-level, and mixed dropout. On the MoSu and Mr. HiSum datasets, TSMD-Temporal improves mAP@15 by 7.06 and 3.41 points over TripleSumm under 50% independent temporal removal, whereas TSMD-Stream performs the best under complete-stream removal. TSMD-Mix retains most of these complementary benefits and ranks the best or the second-best across the evaluated temporal and stream-level conditions.
Chinese Translation
现有的多模态视频亮点检测器通常假设视觉、音频和文本流是持续可用的。然而，在实际应用中，输入可能会遭遇局部帧缺失或完整流中断的问题。我们将这一鲁棒性挑战从两个维度进行表述：时间缺失性，即每种模态中的帧独立缺失，以及流级缺失性，即一种模态在整个视频中不可用。此外，我们发现均方误差（MSE）损失与评估指标及亮点的峰值驱动特性不一致。因此，我们提出了时间流模态丢弃（Temporal-Stream Modality Dropout，TSMD），该方法结合了结构化缺失性模拟与一个联合目标，包括逐点MSE、每视频的Pearson相关性和以峰值为导向的RankNet损失项。TSMD有三种变体：时间丢弃、流级丢弃和混合丢弃。在MoSu和Mr. HiSum数据集上，TSMD-Temporal在50%独立时间移除的情况下，相比TripleSumm提高了7.06和3.41个点的mAP@15，而TSMD-Stream在完整流移除的情况下表现最佳。TSMD-Mix保留了大部分这些互补优势，并在评估的时间和流级条件下排名第一或第二。
cs.CV / 73 / 2609.39066
Agentic Tool-Augmented Reasoning for Explainable Image Forgery Detection
用于可解释图像伪造检测的代理工具增强推理
Abstract
Conventional image forgery detection methods produce binary scores or pixel-level masks without interpretable evidence, while recent multimodal large language model (MLLM)-based approaches generate post-hoc explanations of predetermined classification results rather than reasoning from evidence. Inspired by the forensic workflow of human judicial experts, we propose Agentic Tool-Augmented Reasoning (ATAR), a framework integrating 22 specialized forensic tools across seven complementary domains to autonomously detect, localize, and explain image forgeries through multi-turn reasoning. A Dual-Stream Forensic Reasoning paradigm combines a high-level semantic anomaly path, which magnifies suspicious regions for fine-grained inspection, with a low-level forgery artifact path, which invokes forensic tools to extract objective evidence. We further introduce Forensics Curriculum Learning: during General Experience SFT, an automated teacher-student mentoring pipeline synthesizes multi-turn tool-usage reasoning trajectories; during Forensic Scene RL, a Tool Prior Curriculum guides early tool exploration and progressively transfers control to the agent, while a Structured Evidence Reward provides fine-grained process-level supervision. Experiments across IMDL, Deepfake detection, DMDL, and AIGC detection show that ATAR achieves 78.5% average image-level F1 on six zero-shot IMDL benchmarks, surpassing the strongest MLLM baseline by 11.8 percentage points, and remains competitive with specialized detectors on other tasks while producing substantially more faithful and grounded explanations.
Chinese Translation
传统的图像伪造检测方法生成二元评分或像素级掩码，但缺乏可解释的证据，而最近基于多模态大型语言模型（MLLM）的方法则生成预先确定分类结果的事后解释，而非从证据中推理。受到人类司法专家法医工作流程的启发，我们提出了代理工具增强推理（ATAR）框架，整合了七个互补领域中的22种专业法医工具，通过多轮推理自主检测、定位和解释图像伪造。双流法医推理范式结合了高层语义异常路径，该路径放大可疑区域以进行细致检查，以及低层伪造伪影路径，该路径调用法医工具以提取客观证据。我们进一步引入法医课程学习：在一般经验的监督微调（SFT）过程中，自动化的师生指导管道综合了多轮工具使用推理轨迹；在法医场景强化学习（RL）中，工具优先课程引导早期工具探索，并逐步将控制权转移给代理，同时结构化证据奖励提供细致的过程级监督。在IMDL、深度伪造检测、DMDL和AIGC检测的实验中，ATAR在六个零样本IMDL基准上实现了78.5%的平均图像级F1，超越了最强的MLLM基线11.8个百分点，并在其他任务上与专业检测器保持竞争力，同时生成了显著更可信和扎实的解释。
cs.CV / 74 / 2609.39083
MRI Super-Resolution with RCDM/WaveMix and Task-Aware Segmentation
基于RCDM/WaveMix的MRI超分辨率与任务感知分割
Abstract
Super-resolution and quality enhancement of 1.5\,T brain MRI are normally validated with image-fidelity metrics, although their purpose is to improve downstream analysis. We study whether enhancement improves tissue segmentation, and for which segmenters. We propose an unpaired, physics-guided training pipeline for a lightweight ($\le$2.5\,M parameter) recurrent convolutional enhancer: a six-module stochastic 1.5\,T degradation operator, a residual adversarial network that adds scanner-specific texture without moving anatomy, and a cycle-consistent objective with an anti-identity penalty that rules out the copy solution. We then train U-Net, Swin-UNet and wavelet token-mixing segmenters \citep{jeevan2023wavemix} from scratch on either raw or enhanced 1.5\,T images of the same subjects, using identical labels and subject-level splits, for three enhancer variants and two datasets. On ABIDE (41 held-out subjects, FreeSurfer labels) enhancement significantly improves the wavelet segmenter (mean Dice $+0.014$, Wilcoxon $p=3.5\times10^{-5}$; CSF $+0.018$, grey matter $+0.013$), significantly degrades the U-Net ($-0.008$, $p=5.1\times10^{-4}$) and leaves Swin-UNet unchanged. On IXI, whose labels come from FSL-FAST, enhancement lowers Dice for all nine pairings, almost entirely through CSF; we trace this to spatially implausible CSF voxels in the labels that penalise smoother predictions. Enhancement of low-field MRI should therefore be validated per downstream model and against reliable labels.
Chinese Translation
1.5T脑部MRI的超分辨率和质量增强通常通过图像保真度指标进行验证，尽管其目的是改善下游分析。我们研究了增强是否改善了组织分割，以及对哪些分割器有效。我们提出了一种无配对、物理引导的训练流程，用于轻量级（参数≤2.5M）的递归卷积增强器：一个六模块的随机1.5T降解算子，一个残差对抗网络，该网络在不移动解剖结构的情况下添加扫描仪特定的纹理，以及一个循环一致的目标函数，带有反身份惩罚以排除复制解。然后，我们从头开始训练U-Net、Swin-UNet和小波令牌混合分割器（Jeevan et al., 2023）在相同受试者的原始或增强的1.5T图像上，使用相同的标签和受试者级别的划分，针对三种增强器变体和两个数据集。在ABIDE（41个保留受试者，FreeSurfer标签）上，增强显著改善了小波分割器（平均Dice +0.014，Wilcoxon p=3.5×10^{-5}; CSF +0.018，灰质 +0.013），显著降低了U-Net的性能（-0.008，p=5.1×10^{-4}），而Swin-UNet则保持不变。在IXI数据集上，标签来自FSL-FAST，增强降低了所有九对配对的Dice，几乎完全是通过CSF；我们将其归因于标签中空间上不合理的CSF体素，这惩罚了更平滑的预测。因此，低场MRI的增强应根据每个下游模型和可靠标签进行验证。
cs.CV / 75 / 2609.39089
UGOD: Uncertainty-Guided Opacity and Dropout for Sparse-View 3D Gaussian Splatting
UGOD：用于稀疏视图3D高斯点云的基于不确定性的透明度和丢弃机制
Abstract
Sparse-view 3D Gaussian Splatting is prone to overfitting because limited observations leave many Gaussian primitives weakly constrained, yet their contributions are still accumulated through alpha blending. Without uncertainty estimation, the renderer cannot distinguish unreliable primitives from well-constrained ones, allowing their erroneous contributions to corrupt novel-view synthesis. We introduce UGOD, an uncertainty-guided framework that estimates a view-dependent uncertainty score for each Gaussian and uses it to regulate its rendering contribution. A lightweight uncertainty head conditioned on Gaussian attributes and viewing direction predicts this score, which then drives a differentiable opacity-modulation mechanism that attenuates high-uncertainty primitives before compositing. During training, a detached soft-dropout branch applies an uncertainty-controlled continuous keep mask to discourage the model from relying on poorly constrained Gaussians and thereby reduce overfitting. Crucially, detaching the uncertainty score prevents gradients from this stochastic regulariser from biasing or collapsing the uncertainty prediction. Experiments on Mip-NeRF~360 and LLFF show that UGOD improves sparse-view novel-view synthesis while producing more compact Gaussian representations than the compared methods. These results demonstrate that Gaussian uncertainty provides an effective rendering-time control for sparse-view reconstruction.
Chinese Translation
稀疏视图3D高斯点云容易出现过拟合，因为有限的观测使得许多高斯原件约束较弱，但它们的贡献仍通过α混合被累积。在没有不确定性估计的情况下，渲染器无法区分不可靠的原件与约束良好的原件，从而使得错误的贡献影响新视图合成。我们提出了UGOD，一个基于不确定性的框架，为每个高斯估计视图依赖的不确定性评分，并利用该评分调节其渲染贡献。一个轻量级的不确定性头部根据高斯属性和视角预测该评分，进而驱动一个可微分的透明度调节机制，在合成之前减弱高不确定性的原件。在训练过程中，一个独立的软丢弃分支应用不确定性控制的连续保持掩码，以防止模型依赖于约束较差的高斯，从而减少过拟合。关键是，分离的不确定性评分防止了来自这一随机正则化器的梯度影响或压缩不确定性预测。在Mip-NeRF~360和LLFF上的实验表明，UGOD在稀疏视图新视图合成中表现出色，同时生成比比较方法更紧凑的高斯表示。这些结果表明，高斯不确定性为稀疏视图重建提供了一种有效的渲染时控制手段。
cs.CV / 76 / 2609.39096
DeCoPrune: Efficient KV-Cache Pruning for Autoregressive Video Diffusion via Denoising Consistency
DeCoPrune：通过去噪一致性实现自回归视频扩散的高效 KV 缓存剪枝
Abstract
Autoregressive video diffusion supports streaming generation and interactive control, but its KV cache grows with the generated history. Existing compression strategies discard history using fixed windows or select tokens through local attention and similarity signals, without directly measuring whether a chunk contributes information beyond the retained context. We introduce DeCoPrune, a training-free method that treats cache compression as a denoising-consistency problem. We find that tokens with larger discrepancies between intermediate clean predictions and final denoised values tend to carry visual evidence less predictable from the retained context. DeCoPrune uses this model-intrinsic signal to retain high-discrepancy tokens in the long-term cache while pruning low-discrepancy tokens. To evaluate information retention, we introduce CMBench, comprising 58 approximately one-minute generated or real-world context episodes and 116 Reappear or Revisit continuation tasks requiring recall of earlier events or objects. Experiments with LingBot World v2 show that DeCoPrune achieves a DINO score of 0.6701 on a 0-1 scale, with an 85.43% reduction in cumulative historical KV token counts and a 4.14-fold continuation-generation speedup over FullKV. Its head-specialized variant reaches 0.6783 at an 86.19% pruning ratio, approaching FullKV's 0.6803 score and exceeding the evaluated compression baselines at similar budgets. These results indicate that denoising consistency can support long-range information retention while reducing autoregressive inference cost. Our project homepage is https://decoprune.github.io. The code is available at https://github.com/DeCoPrune/CMBench, and the benchmark at https://huggingface.co/datasets/Aoraku/CMBench.
Chinese Translation
自回归视频扩散支持流式生成和交互控制，但其 KV 缓存随着生成历史的增加而增长。现有的压缩策略通过固定窗口丢弃历史或通过局部注意力和相似性信号选择标记，而没有直接测量某个片段是否提供了超出保留上下文的信息。我们提出了 DeCoPrune，这是一种无训练的方法，将缓存压缩视为去噪一致性问题。我们发现，中间清洁预测与最终去噪值之间差异较大的标记往往携带着从保留上下文中更难以预测的视觉证据。DeCoPrune 利用这种模型内在信号，在长期缓存中保留高差异标记，同时剪枝低差异标记。为了评估信息保留，我们引入了 CMBench，包括 58 个大约一分钟的生成或真实世界上下文片段和 116 个需要回忆早期事件或对象的 Reappear 或 Revisit 续写任务。与 LingBot World v2 的实验表明，DeCoPrune 在 0-1 的评分尺度上达到了 0.6701 的 DINO 分数，累计历史 KV 标记数量减少了 85.43%，并且在生成续写方面比 FullKV 提升了 4.14 倍。其头部专用变体在 86.19% 的剪枝比例下达到了 0.6783，接近 FullKV 的 0.6803 分数，并在相似预算下超越了评估的压缩基准。这些结果表明，去噪一致性可以支持长距离信息保留，同时降低自回归推理成本。我们的项目主页是 https://decoprune.github.io。代码可在 https://github.com/DeCoPrune/CMBench 获取，基准测试可在 https://huggingface.co/datasets/Aoraku/CMBench 找到。
cs.CV / 77 / 2609.39112
CamAgent: An LLM-Agent Framework for Multi-Species Camera-Trap Workflows
CamAgent：一个用于多物种相机捕捉工作流程的 LLM-Agent 框架
Abstract
Camera traps accumulated vast, multidimensional data for wildlife monitoring, yet translating raw media archives into meaningful ecological insights remains highly fragmented. Current research workflows require laboriously stitching together disparate analysis tools and scripts, creating steep programming hurdles and complicating end-to-end spatiotemporal analyses. To overcome this fragmentation, we present CamAgent, an autonomous Large Language Model (LLM) agent framework that integrates camera-trap analytical workflows into a unified intelligent ecosystem. CamAgent interprets natural-language ecological intent, schedules computational routing, and executes specialized tools spanning computer-vision perception (e.g., SpeciesNet), CamtrapDP-compatible data management, detection-corrected occupancy modeling, temporal activity analysis, and species co-occurrence networks. The framework automates multi-stage analytical pipelines while maintaining essential data-quality controls and analytical conventions. Consequently, CamAgent significantly reduces manual programming overhead for conservationists, establishing a transparent, scalable, and fully integrated paradigm for camera-trap ecology. Our project is available at https://anonymous.4open.science/r/artifact72c6f4.
Chinese Translation
相机捕捉设备为野生动物监测积累了大量的多维数据，但将原始媒体档案转化为有意义的生态见解仍然高度分散。目前的研究工作流程需要费力地将不同的分析工具和脚本拼接在一起，这造成了陡峭的编程障碍，并使得端到端的时空分析变得复杂。为了解决这一分散问题，我们提出了 CamAgent，一个自主的大型语言模型（LLM）代理框架，它将相机捕捉分析工作流程整合为一个统一的智能生态系统。CamAgent 解析自然语言生态意图，安排计算路由，并执行涵盖计算机视觉感知（例如，SpeciesNet）、与 CamtrapDP 兼容的数据管理、检测校正的占用建模、时间活动分析和物种共现网络的专业工具。该框架自动化了多阶段分析管道，同时保持必要的数据质量控制和分析规范。因此，CamAgent 显著减少了保护工作者的手动编程负担，建立了一个透明、可扩展且完全整合的相机捕捉生态学范式。我们的项目可在 https://anonymous.4open.science/r/artifact72c6f4 获取。
cs.CV / 78 / 2609.39115
Beyond Local Linearity: Scale-Resolved Geometry of Learned Image Encoders
超越局部线性：学习图像编码器的尺度分辨几何
Abstract
Understanding how learned representations respond to finite input changes is important for characterizing their sensitivity, invariances, and robustness. Yet existing geometric analyses are predominantly local and describe only infinitesimal perturbations. We introduce a scale-resolved statistic that compares an encoder's measured feature displacement with its local linear prediction as the perturbation magnitude increases. Across diverse image encoders, we discover a characteristic plateau-rise-peak-decay profile, which we call the bump. The bump is absent at initialization, emerges early during standard training, and does not form under randomized labels or random-noise inputs. Its shape also varies with the training distribution and robustness objective. These results establish departures from local geometry as a signature of how encoder representations are shaped by learning.
Chinese Translation
理解学习到的表征如何对有限输入变化作出响应，对于表征其敏感性、不变性和鲁棒性至关重要。然而，现有的几何分析主要是局部的，仅描述无穷小扰动。我们引入了一种尺度分辨统计量，用于比较编码器测得的特征位移与其局部线性预测之间的关系，随着扰动幅度的增加而变化。在多种图像编码器中，我们发现了一种特征的平坦-上升-峰值-衰减轮廓，我们称之为“隆起”。该隆起在初始化时不存在，在标准训练的早期出现，并且在随机标签或随机噪声输入下不会形成。其形状还随着训练分布和鲁棒性目标的变化而变化。这些结果表明，局部几何的偏离是编码器表征受学习影响的特征。
cs.CV / 79 / 2609.39116
GRC-Pose: Generation-Reconstruction Correspondence for Prior-Free 6D Object Pose Tracking
GRC-Pose：无先验6D物体姿态跟踪的生成-重建对应
Abstract
Prior-free 6D object pose tracking seeks to recover the trajectory of an unseen object from a single RGB video without object-specific CAD models, posed reference images, or pose annotations. Geometric foundation models provide complementary object-centric and scene-centric cues, yet SAM3D CAD is indexed by an arbitrary object-local surface parameterization, whereas reconstructed evidence is expressed in a sequence-specific world frame with partial surface coverage. To exploit this complementarity, we formulate tracking as generation-reconstruction correspondence and introduce GRC-Pose, a correspondence-based framework that combines learned correspondence prediction with robust pose estimation. Concretely, GeoCorr-Matcher estimates weighted object-scene correspondences and per-match uncertainty for each pose candidate. FGH-Solver integrates these matches through multiple robust geometric estimators and sequence-level posterior inference, while a posterior-gated memory retains only inlier-supported observations through occlusion and viewpoint change. Extensive evaluation shows that with SAM3D CAD, GRC-Pose achieves state-of-the-art Average Recall and motion retention on HOT3D, improving the latter by 58% over prior art. On classical benchmarks including YCBInEOAT and LINEMOD, it remains highly competitive.
Chinese Translation
无先验的6D物体姿态跟踪旨在从单个RGB视频中恢复未见物体的轨迹，而无需物体特定的CAD模型、姿态参考图像或姿态注释。几何基础模型提供了互补的物体中心和场景中心线索，然而，SAM3D CAD是通过任意物体局部表面参数化进行索引的，而重建证据则以序列特定的世界坐标系表达，并具有部分表面覆盖。为了利用这种互补性，我们将跟踪问题形式化为生成-重建对应，并引入GRC-Pose，一个基于对应的框架，结合了学习的对应预测与稳健的姿态估计。具体而言，GeoCorr-Matcher为每个姿态候选估计加权的物体-场景对应关系及每个匹配的不确定性。FGH-Solver通过多种稳健的几何估计器和序列级后验推断整合这些匹配，而后验门控记忆则仅保留通过遮挡和视角变化支持的内点观察。广泛的评估表明，使用SAM3D CAD时，GRC-Pose在HOT3D上实现了最先进的平均召回率和运动保留率，后者比之前的研究提高了58%。在包括YCBInEOAT和LINEMOD在内的经典基准测试中，GRC-Pose仍然保持高度竞争力。
cs.CV / 80 / 2609.39120
Is Better Teacher Supervision Enough? Unlocking Student-side Learning in Multimodal On-Policy Distillation
更好的教师监督是否足够？解锁多模态在线策略蒸馏中的学生学习
Abstract
On-policy distillation (OPD) improves reasoning by providing token-level supervision from a teacher on a student's own trajectories. Existing methods primarily focus on enhancing this teacher-side guidance (e.g., by enriching teacher inputs and refining teacher feedback), yet we find that limited student perception is another critical bottleneck in multimodal OPD. By providing oracle visual facts, the performance of OPD-trained students can still be substantially improved for both weak and strong teachers. To address this bottleneck, we propose S-OPD, a simple multimodal on-policy distillation framework that explicitly strengthens student perceptual learning through two objectives. Specifically, Teacher-calibrated Policy Contrast separates student policies under original and masked images with teacher-based token-level gating, strengthening the student's reliance on visual evidence during reasoning. Policy Agreement aligns student policies under original and noise-perturbed images, further improving perceptual robustness to visual noise. Notably, our method can be seamlessly plugged into existing OPD frameworks, requiring no additional data annotations, model parameters or inference operations. Extensive experiments on eight benchmarks across student scales and distillation paradigms demonstrate consistent performance improvements, with gains of up to 4.25 points on LogicVista. When combined with existing teacher-side supervision methods, our method can yield further gains. Code is available at https://github.com/Sirilaw/S-OPD.
Chinese Translation
在线策略蒸馏（OPD）通过提供来自教师的基于标记的监督，改善了学生在自身轨迹上的推理能力。现有方法主要集中在增强教师端的指导（例如，通过丰富教师输入和优化教师反馈），然而我们发现，有限的学生感知能力是多模态OPD中的另一个关键瓶颈。通过提供权威视觉事实，OPD训练的学生在弱教师和强教师的情况下，其表现仍然可以显著提升。为了解决这一瓶颈，我们提出了S-OPD，一个简单的多模态在线策略蒸馏框架，明确通过两个目标增强学生的感知学习。具体而言，教师校准的策略对比通过基于教师的标记级门控，将学生在原始图像和掩蔽图像下的策略进行分离，增强学生在推理过程中对视觉证据的依赖。策略一致性则对齐学生在原始图像和噪声扰动图像下的策略，进一步提高对视觉噪声的感知鲁棒性。值得注意的是，我们的方法可以无缝集成到现有的OPD框架中，无需额外的数据标注、模型参数或推理操作。在八个基准测试上进行的大规模实验表明，表现持续改善，在LogicVista上提升幅度高达4.25分。当与现有的教师端监督方法结合时，我们的方法可以带来进一步的提升。代码可在 https://github.com/Sirilaw/S-OPD 获取。
cs.CV / 81 / 2609.39127
How to Reduce Localization Ambiguity? Geometry-Semantic Constrained BEV Representation Learning for Satellite-Ground Localization
如何减少定位歧义？几何-语义约束的鸟瞰视图（BEV）表示学习用于卫星-地面定位
Abstract
Satellite-ground localization estimates the planar position and yaw orientation of a ground camera within a geo-referenced satellite image. Most recent methods map ground and satellite features into a shared bird's-eye-view (BEV) space and establish spatial correspondences. However, insufficient depth constraints can assign one ground feature to different distances along a viewing direction, creating geometric ambiguity in BEV feature placement. Similar appearances at different locations can also create descriptor matching ambiguity, while existing descriptor learning lacks explicit semantic supervision to distinguish them. We propose GeoSem-BEV, a geometry-semantic constrained BEV representation learning method. Radial depth supervision constrains distance assignment, and vertical height supervision constrains height aggregation. Shared explicit semantic supervision promotes consistent semantic predictions across views and helps distinguish locations with similar semantics. These constraints improve feature placement and descriptor discriminability, enhancing state-of-the-art BEV localization models. On VIGOR with unknown orientation, GeoSem-BEV reduces mean orientation error by 37.2% and 38.1% in the cross-area and same-area settings, respectively. The corresponding errors are reduced by 10.8% and 15.6% on DReSS-D. On KITTI-CVL, it reduces same-area mean orientation error by 26.8% under 10 degree orientation noise.
Chinese Translation
卫星-地面定位估计地面相机在地理参考卫星图像中的平面位置和偏航方向。最近的方法大多将地面和卫星特征映射到共享的鸟瞰视图（BEV）空间，并建立空间对应关系。然而，深度约束不足可能导致一个地面特征在视线方向上被分配到不同的距离，从而在BEV特征放置中产生几何歧义。不同位置的相似外观也可能导致描述符匹配歧义，而现有的描述符学习缺乏明确的语义监督来区分它们。我们提出了GeoSem-BEV，一种几何-语义约束的BEV表示学习方法。径向深度监督约束距离分配，垂直高度监督约束高度聚合。共享的明确语义监督促进了跨视图的一致语义预测，并帮助区分具有相似语义的位置。这些约束改善了特征放置和描述符的可区分性，增强了最先进的BEV定位模型。在未知方向的VIGOR上，GeoSem-BEV在跨区域和同区域设置中分别减少了37.2%和38.1%的平均方向误差。在DReSS-D上，相应的误差分别减少了10.8%和15.6%。在KITTI-CVL上，它在10度方向噪声下减少了26.8%的同区域平均方向误差。
cs.CV / 82 / 2609.39128
GeoGAT: Bidirectional Temporal Sampling Meets Hierarchical Graph Attention for Global Video Geo-localization
GeoGAT：双向时间采样与层次图注意力相结合的全球视频地理定位
Abstract
Global video geo-localization aims to infer the geographic location of a video worldwide, evaluating performance across four geographic hierarchies: city, state/province, country, and continent. Existing methods typically employ one-way uniform sampling to process video frames and train independent classifiers for each hierarchy, which leads to the loss of key geographic cues and prediction conflicts between hierarchies, especially for complex multi-shot edited videos. To address these limitations, we propose GeoGAT, which integrates bidirectional temporal sampling with graph attention networks (GATs). Specifically, GeoGAT extracts forward and offset-reversed frame sequences to construct complementary spatiotemporal features. These fused features are then fed into a predefined geographical hierarchy graph, where GATs perform structure-aware message passing, while a dual-constraint mechanism prunes predictions to eliminate cross-hierarchy conflicts. We construct GeoGAT10k, comprising 9,720 multi-shot edited videos from 166 cities worldwide, specifically to benchmark generalization ability on complex video structures. Experimental results on CityGuessr68k and GeoGAT10k demonstrate that GeoGAT eliminates hierarchical conflicts entirely and achieves state-of-the-art performance across all four geographic hierarchies. On CityGuessr68k, GeoGAT outperforms the strongest baseline, evaluated under both classification and retrieval protocols, by 2.6 percentage points at the city level. On the more challenging GeoGAT10k with multi-shot edited videos, the accuracy improvement exceeds 24 percentage points, validating strong generalization to complex real-world scenarios.
Chinese Translation
全球视频地理定位旨在推断视频在全球范围内的地理位置，并在城市、州/省、国家和大陆四个地理层次上评估性能。现有方法通常采用单向均匀采样来处理视频帧，并为每个层次训练独立的分类器，这导致了关键地理线索的丢失以及层次之间的预测冲突，尤其是在复杂的多镜头编辑视频中。为了解决这些局限性，我们提出了GeoGAT，它将双向时间采样与图注意力网络（GATs）相结合。具体而言，GeoGAT提取正向和偏移反向的帧序列，以构建互补的时空特征。这些融合特征随后被输入到预定义的地理层次图中，在这里，GATs执行结构感知的消息传递，同时双重约束机制修剪预测，以消除跨层次的冲突。我们构建了GeoGAT10k，包含来自全球166个城市的9,720个多镜头编辑视频，专门用于基准测试复杂视频结构上的泛化能力。在CityGuessr68k和GeoGAT10k上的实验结果表明，GeoGAT完全消除了层次冲突，并在所有四个地理层次上实现了最先进的性能。在CityGuessr68k上，GeoGAT在城市层面上超越了最强基线，分类和检索协议下的评估结果提高了2.6个百分点。在更具挑战性的GeoGAT10k多镜头编辑视频上，准确率提升超过24个百分点，验证了其在复杂现实场景中的强泛化能力。
cs.CV / 83 / 2609.39130
Perceptual Color Difference Modeling Using Machine Learning and Human Similarity Judgments
基于机器学习和人类相似性判断的感知颜色差异建模
Abstract
Accurate assessment of color differences is essential for applications ranging from digital design to quality control. While existing color difference metrics, such as CIEDE2000, aim to approximate human perception, they may still exhibit inconsistencies with perceptual judgments. In this study, we investigate a data-driven approach to color-difference estimation based directly on human evaluations. We collect similarity judgments for 2,000 systematically generated color pairs, each rated by seven observers using a four-point ordinal scale. These judgments are then used to train regression models using different color representations, including RGB channel differences, HSI differences, and COLIBRI fuzzy linguistic categories. Experiments with five regression algorithms show that the choice of color model has a greater influence on prediction performance than the choice of regression algorithm. Using COLIBRI features alone, linear regression achieves an R2 of 0.595, outperforming RGB and HSI representations, which achieve R2 values of 0.479 and 0.493, respectively. The best performance is obtained by LightGBM using the combined representation, reaching an R2 of 0.703. The results indicate that human perceptual color differences are better captured when numerical color coordinates are complemented by graded perceptual categories, highlighting the potential of data-driven models for perceptually aligned color-difference estimation.
Chinese Translation
准确评估颜色差异对于从数字设计到质量控制等应用至关重要。尽管现有的颜色差异度量标准，如CIEDE2000，旨在近似人类感知，但它们在感知判断上仍可能存在不一致性。在本研究中，我们探讨了一种基于人类评估的以数据驱动的方法来估计颜色差异。我们收集了2,000对系统生成的颜色对的相似性判断，每对颜色由七位观察者使用四点有序量表进行评分。这些判断随后用于训练回归模型，采用不同的颜色表示，包括RGB通道差异、HSI差异和COLIBRI模糊语言类别。五种回归算法的实验表明，颜色模型的选择对预测性能的影响大于回归算法的选择。仅使用COLIBRI特征，线性回归的R²达到了0.595，优于RGB和HSI表示，后者的R²值分别为0.479和0.493。最佳性能由LightGBM在组合表示下实现，R²达到了0.703。结果表明，当数值颜色坐标与分级感知类别相结合时，人类感知的颜色差异得到了更好的捕捉，突显了数据驱动模型在感知对齐的颜色差异估计中的潜力。
cs.CV / 84 / 2609.39132
Uncertainty-Aware Consistency Distillation for Few-Step Video Generation
基于不确定性感知的一致性蒸馏用于少步视频生成
Abstract
We study few-step video generation, i.e., distilling a multi-step video generator, which typically requires tens of sampling steps, incurring substantial latency and compute, into a few-step student. Consistency distillation is a common recipe, in which a multi-step teacher provides the consistency targets for a few-step student. However, these teacher-guided targets are not equally trustworthy, and the content is harder to learn where it varies rapidly over time, e.g., moving foliage shadows or flowing water. We observe that supervision reliability follows the local difficulty of the content rather than semantic complexity: regions that change little yield consistent endpoint predictions, whereas regions with large temporal variation produce larger discrepancies that coincide with the largest perceptual errors. Motivated by this observation, we propose Uncertainty-Aware Consistency Distillation (UACD), which reweights consistency supervision at each spatiotemporal region using a local, parameter-free uncertainty estimate. Specifically, we construct two independently perturbed teacher-guided consistency paths, whose student endpoint predictions provide a consensus target; the discrepancy between the student's direct prediction and this target is the uncertainty proxy. We then relax the consistency penalty on high-uncertainty regions through an exponential weight, while keeping the full penalty elsewhere, since the student cannot be expected to match targets that are hard to learn. To preserve perceptual quality under aggressive step reduction, we integrate feature-space adversarial training with semantic alignment. With parameter-efficient LoRA adaptation of the 50-step Wan model, our method achieves state-of-the-art 4-step generation on VBench 2.0 (0.556 mean score) and is preferred over competing methods in a user study.
Chinese Translation
我们研究少步视频生成，即将一个通常需要数十个采样步骤的多步视频生成器蒸馏为一个少步学生模型，这会导致显著的延迟和计算开销。一致性蒸馏是一种常见的方法，其中多步教师为少步学生提供一致性目标。然而，这些教师指导的目标并不都是同样可信的，内容在快速变化的地方（例如，移动的树叶阴影或流动的水）更难以学习。我们观察到，监督的可靠性遵循内容的局部难度，而不是语义复杂性：变化较小的区域产生一致的端点预测，而具有较大时间变化的区域则产生较大的差异，这与最大的感知误差相一致。基于这一观察，我们提出了不确定性感知一致性蒸馏（Uncertainty-Aware Consistency Distillation, UACD），该方法使用局部的无参数不确定性估计对每个时空区域的一致性监督进行重新加权。具体而言，我们构建了两个独立扰动的教师指导一致性路径，其学生端点预测提供了共识目标；学生的直接预测与该目标之间的差异即为不确定性代理。然后，我们通过指数权重放宽高不确定性区域的一致性惩罚，同时在其他地方保持完整的惩罚，因为学生无法期望匹配难以学习的目标。为了在激进的步骤减少下保持感知质量，我们将特征空间对抗训练与语义对齐相结合。通过对50步Wan模型进行参数高效的LoRA适应，我们的方法在VBench 2.0上实现了最先进的4步生成（平均得分为0.556），并在用户研究中优于竞争方法。
cs.CV / 85 / 2609.39134
Feature-Aware Token Attack for Compression-Triggered Stealthy Failures in Large Vision-Language Models
针对大规模视觉语言模型中压缩触发的隐蔽性故障的特征感知令牌攻击
Abstract
Visual-token compression improves the efficiency of large vision-language models, but can expose failures that full-token evaluation misses. We study adversarial images that preserve full-token correctness yet induce errors after compression, even when both inference paths succeed on the clean image. Creating such failures is challenging because perturbing token importance can also damage the visual content needed for full-token inference. We propose Feature-Aware Token Attack (FATA), which couples attention suppression with cosine-based feature preservation on a fixed set of salient clean-image tokens. In the primary LLaVA-1.5-7B setting, FATA uses only vision-encoder gradients, without access to the deployed compressor, token budget, or downstream task. Across four visually dependent task subsets and four compressors under a controlled reconstruction protocol, FATA achieves SR = 96.3% full-token accuracy retention and CBR = 22.1% conditional blinding, compared with 89.8% and 15.7% for CAA. Ablations support the role of both objectives in balancing compressed-path failure against full-token preservation. FATA also has the lowest measured detection rate among four attacks across three evaluated detectors at a 5% false-positive rate. These findings motivate assessing adversarial robustness jointly across full-token and compressed inference.
Chinese Translation
视觉令牌压缩提高了大规模视觉语言模型的效率，但可能暴露出全令牌评估所遗漏的故障。我们研究了那些在保持全令牌正确性的同时，因压缩而引发错误的对抗性图像，即使在干净图像上两个推理路径均成功。创建此类故障具有挑战性，因为扰动令牌重要性也可能损害全令牌推理所需的视觉内容。我们提出了特征感知令牌攻击（Feature-Aware Token Attack, FATA），该方法将注意力抑制与基于余弦的特征保留结合在一起，作用于一组固定的显著干净图像令牌。在主要的 LLaVA-1.5-7B 设置中，FATA 仅使用视觉编码器的梯度，而不访问已部署的压缩器、令牌预算或下游任务。在四个视觉依赖任务子集和四个压缩器下，通过受控重建协议，FATA 实现了 SR = 96.3% 的全令牌准确率保留和 CBR = 22.1% 的条件盲化，而 CAA 分别为 89.8% 和 15.7%。消融实验支持了这两个目标在平衡压缩路径故障与全令牌保留中的作用。FATA 在三个评估检测器中以 5% 的假阳性率测得的检测率也是四种攻击中最低的。这些发现激励我们在全令牌和压缩推理中共同评估对抗鲁棒性。
cs.CV / 86 / 2609.39135
Asking the World: Generalist Physical Reasoning through Agentic World Modeling and Probing
询问世界：通过自主世界建模和探测实现通用物理推理
Abstract
Physical reasoning from video requires inferring latent physical properties and dynamics beyond direct observation. Direct VLM inference remains unreliable on complex physical tasks without explicit modeling and validation, while predefined tool pipelines rely on task- and domain-specific priors that limit generalization across materials, dynamics, and reasoning tasks. We introduce Asking the World (ATW), a generalist agent that constructs and interrogates task-relevant executable worlds through two adaptive stages: World Modeling calibrates a world from video, while World Probing queries, simulates, and intervenes on it to obtain question-relevant evidence. Rather than prescribing the operations in either stage, ATW determines how to model and probe according to the scene and question. We develop PolyWorld Engine, a lightweight and highly programmable Warp-based multiphysics simulator for constructing and probing worlds with rigid bodies, soft bodies, cloth, ropes, fluids, and their coupled interactions. CEM-based system identification recovers task-relevant dynamics during World Modeling. The resulting world becomes an active workspace for question-directed physical experiments rather than a predetermined downstream tool. We evaluate ATW on CLEVRER, ContPhy, and three real-world scenarios. Using Gemini-3-Flash as its base VLM, ATW achieves 80.82% overall per-question accuracy on CLEVRER, improving direct Gemini-3-Flash by 46.50 points, GPT-5.5 by 13.58 points, and PhysMind by 8.27 points. On ContPhy, it reaches 70.56% overall accuracy, surpassing Gemini-3-Flash by 28.10 points and GPT-5.5 by 3.53 points. Across the three real-world scenarios, ATW achieves 71.67% accuracy, 28.33 points above GPT-5.5. These results establish agentic world modeling and probing as an effective, execution-grounded approach to generalist physical reasoning.
Chinese Translation
从视频中进行物理推理需要推断潜在的物理属性和动态，这超出了直接观察的范围。在复杂的物理任务中，直接的视觉语言模型（VLM）推理仍然不可靠，缺乏明确的建模和验证，而预定义的工具管道依赖于特定任务和领域的先验知识，这限制了在材料、动态和推理任务之间的泛化。我们提出了询问世界（Asking the World, ATW），这是一个通用智能体，通过两个自适应阶段构建和审问与任务相关的可执行世界：世界建模（World Modeling）从视频中校准一个世界，而世界探测（World Probing）则查询、模拟并干预该世界，以获取与问题相关的证据。ATW并不规定任何阶段的操作，而是根据场景和问题来确定如何建模和探测。我们开发了PolyWorld Engine，这是一种轻量级且高度可编程的基于Warp的多物理模拟器，用于构建和探测具有刚体、软体、布料、绳索、流体及其耦合相互作用的世界。在世界建模过程中，基于CEM的系统识别恢复了与任务相关的动态。生成的世界成为一个主动的工作空间，用于以问题为导向的物理实验，而不是一个预先设定的下游工具。我们在CLEVRER、ContPhy和三个现实场景上评估了ATW。以Gemini-3-Flash作为基础VLM，ATW在CLEVRER上实现了80.82%的整体每问题准确率，比直接的Gemini-3-Flash提高了46.50个百分点，比GPT-5.5提高了13.58个百分点，比PhysMind提高了8.27个百分点。在ContPhy上，其整体准确率达到70.56%，超过Gemini-3-Flash 28.10个百分点，超过GPT-5.5 3.53个百分点。在三个现实场景中，ATW的准确率为71.67%，比GPT-5.5高出28.33个百分点。这些结果确立了自主世界建模和探测作为一种有效的、基于执行的通用物理推理方法。
cs.CV / 87 / 2609.39142
When Can Text Replace Vision? Structural Bottlenecks in Diagram Reasoning
文本何时可以替代视觉？图表推理中的结构瓶颈
Abstract
Can structured text replace vision for diagram reasoning? A wrong answer after textualization can arise because the representation omits information the question needs, or because the solver fails to use information that is present. We introduce a diagnostic protocol to distinguish these explanations. Using the same solver model and generation settings, we compare three input conditions: the original image, question-blind structure extracted by a vision-language model, or gold structure derived from the diagram source. Validity-triggered recovery tests truncation and schema failure, question-relevant fidelity measures preservation of answer-critical structure, and matched edge interventions test the effect of error location. On a reserved holdout of 240 public FlowGen diagrams, evaluated under a frozen protocol, gold structure reaches 87% accuracy while direct vision and learned text both remain below 30%. The aggregate comparison includes source-derived relation labels that may not be printed in the image and uses different learned and gold graph encodings, so it does not isolate extraction error alone. Retrying only invalid extractions makes nearly every public representation schema-valid yet leaves accuracy essentially unchanged. The public learned-text deficit relative to gold more than doubles with structural difficulty. Question-relevant topology predicts correctness better than whole-graph topology. In an exposed intervention study, a single answer-relevant edge edit reduces the primary solver's original-answer accuracy to near zero, while matched irrelevant edits largely preserve it. Supplied structure requires fewer solving tokens than vision, but learned acquisition removes this advantage at single use. These comparisons motivate evaluating acquired text by the answer-relevant evidence it preserves and by the solver's ability to use that representation.
Chinese Translation
结构化文本能否替代视觉进行图表推理？文本化后错误答案的产生可能是因为表示省略了问题所需的信息，或因为解题者未能利用现有的信息。我们引入了一种诊断协议来区分这些解释。使用相同的解题模型和生成设置，我们比较了三种输入条件：原始图像、由视觉-语言模型提取的无问题结构，或源自图表的黄金结构。有效性触发的恢复测试截断和模式失败，问题相关的保真度测量保持答案关键结构，而匹配边缘干预测试错误位置的影响。在一个保留的240个公共FlowGen图表的样本上，在冻结协议下评估，黄金结构的准确率达到87%，而直接视觉和学习文本的准确率均低于30%。总体比较包括可能未在图像中打印的源派生关系标签，并使用不同的学习和黄金图编码，因此并未单独隔离提取错误。仅重试无效提取几乎使每个公共表示模式有效，但准确率基本保持不变。相对于黄金结构，公共学习文本的缺陷在结构难度增加时超过两倍。问题相关的拓扑比整体图拓扑更好地预测正确性。在一次暴露干预研究中，单个答案相关边缘的编辑将主要解题者的原始答案准确率降低至接近零，而匹配的无关编辑则在很大程度上保持了准确率。提供的结构所需的解题令牌少于视觉，但学习获取在单次使用时消除了这一优势。这些比较促使我们通过保留的答案相关证据和解题者使用该表示的能力来评估获取的文本。
cs.CV / 88 / 2609.39147
MindWorldBench: Evaluating Mental-State-to-Behavior Reasoning in Image-to-Video Generation
MindWorldBench：评估图像到视频生成中的心理状态与行为推理
Abstract
Current image-to-video models achieve visual realism and physical plausibility, but reasoning about mental states remains unexplored. Actions are driven by belief, desire, and perception, requiring inference beyond explicit instructions. We introduce MindWorldBench to evaluate mental-state-conditioned video generation. We formalize this as mental-state-to-behavior reasoning, where models generate actions from a world state and latent variables without explicit action prompts. MindWorldBench utilizes Zero-Action Prompting and a counterfactual design with 744 prompts to isolate the causal effects of mental states. An automated pipeline evaluates video quality, commonsense plausibility, and mental-state consistency. Evaluations of 11 models show that despite visual fidelity and physical reasoning, models fail to align behaviors with latent mental states. We identify a failure mode, termed Omniscient Bias, where models default to the objective world state rather than human's subjective belief. These results demonstrate a disconnect between visual generation and cognitive reasoning, suggesting a need for explicit mental-state modeling in video generation systems. Project website: https://richard2049-lee.github.io/MindWorldBench/
Chinese Translation
当前的图像到视频模型在视觉真实感和物理合理性方面取得了显著进展，但对心理状态的推理仍然未得到探索。行为受到信念、欲望和感知的驱动，这需要超越明确指令的推理。我们引入了MindWorldBench来评估基于心理状态的视频生成。我们将其形式化为心理状态到行为的推理，其中模型从世界状态和潜变量生成行为，而无需明确的行为提示。MindWorldBench利用零行为提示（Zero-Action Prompting）和744个提示的反事实设计，以隔离心理状态的因果效应。一个自动化的流程评估视频质量、常识合理性和心理状态一致性。对11个模型的评估表明，尽管在视觉保真度和物理推理方面表现良好，模型仍未能将行为与潜在心理状态对齐。我们识别出一种失败模式，称为全知偏差（Omniscient Bias），即模型默认采用客观世界状态而非人类的主观信念。这些结果表明视觉生成与认知推理之间存在脱节，提示在视频生成系统中需要明确的心理状态建模。项目网站：https://richard2049-lee.github.io/MindWorldBench/
cs.CV / 89 / 2609.39150
OP-CAD: On-Policy Clean-Audio Distillation for Robust Audio-Visual Reasoning
OP-CAD：用于稳健音视频推理的在线清音频蒸馏
Abstract
Omni-modal large language models deployed in real-world environments encounter external noise that can interfere with their perception and understanding of multimodal inputs. We study their robustness in audio-visual understanding, focusing on question answering under environmental noise and competing speech. The challenge is to resist acoustic interference while preserving useful audio evidence. On-policy distillation provides dense teacher feedback on student-generated responses, but uniform token weighting does not explicitly prioritize positions affected by acoustic interference. We introduce OP-CAD (On-Policy Clean-Audio Distillation), a curriculum-based privileged self-distillation framework for robust audio-visual understanding. Training progresses from mild to severe environmental noise and competing speech, with selective token-level supervision at each stage. The student generates responses from corrupted audio-visual input, while a frozen teacher uses clean audio and the verified answer to supervise the same response prefixes. To allocate this supervision, OP-CAD compares teacher predictions under clean, corrupted, and visual-only contexts without revealing the answer. These matched comparisons measure sensitivity to audio removal and corruption; a bounded weighting rule emphasizes positions identified by either signal while retaining supervision throughout the response. OP-CAD outperforms the compared methods across all evaluated noise conditions. Paired analyses further show improved preservation of clean-correct answers under strong interference, with no observed aggregate clean-accuracy penalty. These results demonstrate the value of directing clean-teacher supervision toward acoustically sensitive predictions for robust audio-visual reasoning.
Chinese Translation
在真实环境中部署的全模态大型语言模型会遇到外部噪声，这可能干扰其对多模态输入的感知和理解。我们研究了它们在音视频理解中的稳健性，重点关注在环境噪声和竞争性语音下的问答能力。挑战在于抵抗声学干扰，同时保留有用的音频证据。在线蒸馏提供了密集的教师反馈，针对学生生成的响应，但统一的标记加权并未明确优先考虑受到声学干扰影响的位置。我们引入了OP-CAD（在线清音频蒸馏），这是一个基于课程的特权自蒸馏框架，用于稳健的音视频理解。训练从轻微到严重的环境噪声和竞争性语音逐步进行，在每个阶段进行选择性标记级监督。学生从受损的音视频输入生成响应，而冻结的教师使用干净音频和经过验证的答案来监督相同的响应前缀。为了分配这种监督，OP-CAD在不揭示答案的情况下比较教师在干净、受损和仅视觉上下文下的预测。这些匹配比较测量了对音频去除和损坏的敏感性；一个有界加权规则强调了由任一信号识别的位置，同时在整个响应中保留监督。OP-CAD在所有评估的噪声条件下均优于比较方法。配对分析进一步显示，在强干扰下，干净正确答案的保留得到了改善，并未观察到整体清晰准确度的惩罚。这些结果展示了将清晰教师监督引导至声学敏感预测以实现稳健音视频推理的价值。
cs.CV / 90 / 2609.39157
TripleFlow: Training-Free Video Object Removal by Bridging Residual Editing and Native Generation
TripleFlow：通过桥接残差编辑与原生生成实现无训练视频物体移除
Abstract
Video object removal presents a uniquely difficult editing challenge. Because a removal prompt specifies only what to erase rather than what to generate, the model must infer and reconstruct a highly specific occluded background entirely from the surrounding context. Existing training-free methods struggle with this because their editing mechanisms act primarily as localized erasers. They fail to actively synthesize the missing background details and often leave behind ghosting artifacts. To solve this, we propose TripleFlow, a training-free framework that tightly couples erasure and generation. It coordinates a source flow, a residual flow, and a synthesis flow throughout the entire process. By reusing a single target prediction, the residual flow isolates and suppresses the object, while the synthesis flow independently reconstructs the occluded background. Crucially, TripleFlow injects this newly synthesized background back into the editing trajectory at every step. This continuous feedback loop ensures that the generated structures actively guide the removal process, achieving seamless completion that is spatiotemporally consistent with the unedited scene. Extensive evaluations across five challenging benchmarks demonstrate that TripleFlow establishes a new state-of-the-art, significantly outperforming existing baselines in both reconstruction fidelity and temporal consistency.
Chinese Translation
视频物体移除是一项独特而困难的编辑挑战。由于移除提示仅指定要擦除的内容，而非要生成的内容，因此模型必须完全从周围上下文中推断和重建高度特定的被遮挡背景。现有的无训练方法在这方面表现不佳，因为它们的编辑机制主要作为局部橡皮擦，未能主动合成缺失的背景细节，常常留下鬼影伪影。为了解决这个问题，我们提出了TripleFlow，这是一种紧密结合擦除与生成的无训练框架。它在整个过程中协调源流、残差流和合成流。通过重用单一目标预测，残差流隔离并抑制物体，而合成流则独立重建被遮挡的背景。关键是，TripleFlow在每一步将新合成的背景注入编辑轨迹中。这一持续的反馈循环确保生成的结构积极引导移除过程，实现与未编辑场景在时空上一致的无缝完成。在五个具有挑战性的基准测试中的广泛评估表明，TripleFlow建立了新的最先进水平，在重建保真度和时间一致性方面显著超越现有基线。
cs.CV / 91 / 2609.39182
MEND: Label-Free Detection, Localisation, and Correction of Latent Hallucination in World Models
MEND：无标签检测、定位和修正世界模型中的潜在幻觉
Abstract
World Models are appearing as the next major frontier in computer vision. However, their robustness is currently largely unexplored. We identify the phenomenon of hallucination in latent World Models: given a state and an action, the predicted next latent can decode to a scene that never occurs. Because the prediction is statistically ordinary and is fed back autoregressively by the model, the error is both silent and compounding. We study whether such latent hallucination can be detected, localised, and corrected at inference time, on a frozen self-supervised world model in the absence of ground-truth error labels. We introduce Masked Empirical-Bayes Neural Denoising (MEND), a single conditional score network trained by denoising score matching on real transitions, whose score field serves three roles: its magnitude detects hallucination, its per-token field localises it to specific image patches, and it defines an inference-time correction direction. On two navigation environments MEND detects hallucination with an AUROC of up to 0.80 without using actions, exceeding a single-Gaussian density baseline while also localising the error (per-token AUPRC up to 0.87) and correcting it, all from one score field. Our correction reliably reduces single-step latent error and improves predictions. We identify that a part of the error is tangent to the data manifold, hence, we focus on detection and localisation while highlighting promises of the correction.
Chinese Translation
世界模型正在成为计算机视觉的下一个主要前沿。然而，它们的鲁棒性目前仍然很大程度上未被探索。我们识别出潜在世界模型中的幻觉现象：给定一个状态和一个动作，预测的下一个潜在状态可能解码为一个从未出现过的场景。由于预测在统计上是普通的，并且由模型自回归地反馈，因此错误既是无声的又是累积的。我们研究在没有真实错误标签的情况下，是否可以在推理时检测、定位和修正这种潜在幻觉，基于一个冻结的自监督世界模型。我们引入了掩蔽经验贝叶斯神经去噪（Masked Empirical-Bayes Neural Denoising, MEND），这是一个通过对真实转变进行去噪得分匹配训练的单条件得分网络，其得分场具有三种作用：其幅度检测幻觉，其每个标记的得分场将其定位到特定的图像块，并定义了推理时的修正方向。在两个导航环境中，MEND在不使用动作的情况下以高达0.80的AUROC检测幻觉，超越了单高斯密度基线，同时也定位了错误（每个标记的AUPRC高达0.87）并进行了修正，所有这些都来自一个得分场。我们的修正可靠地减少了单步潜在错误并改善了预测。我们识别出部分错误与数据流形相切，因此，我们专注于检测和定位，同时强调修正的潜力。
cs.CV / 92 / 2609.39183
Aligning Thoughts with Answers: Probability Rewards to Tame Thinking Drift
将思维与答案对齐：概率奖励以驯服思维漂移
Abstract
This paper studies \textbf{thinking--answer consistency} in vision-language models. We focus on Visual Intention Grounding, where a model infers a target object based on a human intention query and predicts a bounding box. We reveal that previous IoU-based reinforcement learning (RL) frameworks suffer from ``thinking drift'', where the model produces a correct bounding box, despite having an incorrect reasoning process pointing to a different target object. Thus, we propose \textbf{Rita} (\textit{ReInforcing Thinking--Answer consistency}) as a novel RL paradigm to tame the drift. Specifically, Rita introduces two reasoning-label-free RL rewards, constructed from the conditional probability of reference answers: a \textbf{thinking reward} and a \textbf{consistency reward}. It also adopts a difficulty-aware \textbf{data filtering} strategy that selects informative easy-to-medium samples for RL using rollout error rate and reward variance. Extensive experiments on EgoIntention and the new RefEgo-Int benchmarks show that Rita performs consistently superior to the supervised finetuning approaches and vanilla RL-finetuned frameworks.
Chinese Translation
本文研究了视觉语言模型中的 extbf{思维-答案一致性}。我们重点关注视觉意图定位（Visual Intention Grounding），在该任务中，模型根据人类意图查询推断目标对象并预测边界框。我们揭示了先前基于IoU的强化学习（RL）框架存在“思维漂移”的问题，即模型虽然生成了正确的边界框，但推理过程却指向了不同的目标对象。因此，我们提出了 extbf{Rita}（ extit{ReInforcing Thinking--Answer consistency}）作为一种新颖的RL范式，以驯服这种漂移。具体而言，Rita引入了两个无推理标签的RL奖励，这些奖励是基于参考答案的条件概率构建的： extbf{思维奖励}和 extbf{一致性奖励}。它还采用了一种关注难度的 extbf{数据过滤}策略，通过回滚错误率和奖励方差选择信息丰富的简单到中等样本用于RL。在EgoIntention和新的RefEgo-Int基准上的大量实验表明，Rita的表现始终优于监督微调方法和普通的RL微调框架。
cs.CV / 93 / 2609.39184
Fiber-Resolved Microstructure Quantification from Multi-Shell Diffusion MRI using Detection Transformers
基于多壳扩散MRI的光纤分辨微结构量化方法：采用检测变换器
Abstract
Fiber orientation and compartmental microstructure are central to the characterization of white matter tissue in diffusion MRI, yet existing methods either resolve fiber orientations without quantifying microstructure, or quantify microstructure while assuming a fixed number of compartments and a single fiber direction. Nonparametric approaches that recover both require tensor-valued diffusion encoding and computationally expensive Monte-Carlo inversion of an ill-posed inverse Laplace transform. We propose to reframe this problem as an object detection-like task, adopting the Detection Transformer (DETR) architecture to jointly predict mean diffusivity (MD), fractional anisotropy (FA), main fiber direction, and signal fraction for a variable number of compartments per voxel from standard multi-shell diffusion MRI with linear encoding. Hungarian matching during training resolves permutation invariance across compartments. We introduce mean Average Precision as a reproducible benchmark metric. Evaluated on synthetic test data with up to five compartments per voxel, our model achieves $R^2=0.95$ for MD, $R^2=0.88$ for FA, and a median angular error of 4.2{\deg}, with performance scaling naturally with compartmental signal fraction.
Chinese Translation
光纤方向和隔室微结构是扩散MRI中白质组织表征的核心，但现有方法要么解决光纤方向而不量化微结构，要么在假设固定数量的隔室和单一光纤方向的情况下量化微结构。恢复两者的非参数方法需要张量值扩散编码以及计算上昂贵的蒙特卡洛反演不适定的反拉普拉斯变换。我们建议将此问题重新构建为类似物体检测的任务，采用检测变换器（Detection Transformer, DETR）架构，从标准的多壳扩散MRI中联合预测每个体素的平均扩散率（mean diffusivity, MD）、分数各向异性（fractional anisotropy, FA）、主光纤方向和信号分数，适用于可变数量的隔室。在训练过程中，匈牙利匹配解决了隔室间的置换不变性。我们引入平均精度（mean Average Precision）作为可重复的基准指标。在对每个体素最多五个隔室的合成测试数据进行评估时，我们的模型在MD上达到了$R^2=0.95$，在FA上达到了$R^2=0.88$，并且中位角误差为4.2{ extdegree}，性能自然随着隔室信号分数的增加而提升。
cs.CV / 94 / 2609.39195
Uruqi: Learning Spatial Cognition from Visual Experience
Uruqi：从视觉经验中学习空间认知
Abstract
Spatial intelligence requires maintaining a coherent understanding of the world as the embodied agent moves. Like humans, the agent must use its own motion to interpret changes across observations and update object locations and spatial relations accordingly. Despite spatial post-training having substantially broadened the spatial intelligence of vision-language models (VLMs), they still struggle with two atomic spatial capabilities: tracking self-motion and mapping the surrounding world during motion. To address this gap, we provide dense multi-turn supervision over interleaved atomic capabilities within each training episode, mimicking the visual experience of a continuously moving agent that reasons as it observes. To scale this up, we synthesize 11,738 motif-driven camera trajectories over a broad range of 3D scenes, supporting self-motion tracking, persistent object mapping, and rich spatial operations within each visual experience. By training models to reason over these atomic questions, our URUQI$_{\mathrm{Syn}}$-8B improves accuracy from 15.84% to 47.73% on our Uruqi benchmark comprising 52k questions across 2.7k episodes. URUQI-SI-Mix-8B further reaches 50.41%, comparable to the 50.08% achieved by GPT-6 Astra. Trained solely on our synthesized data, URUQI$_{\mathrm{Syn}}$-8B achieves an average relative accuracy improvement of 17.13% over its InternVL3-8B backbone across three external spatial benchmarks. These results highlight continuous visual experience as a scalable source of supervision for developing spatial cognition in VLMs.
Chinese Translation
空间智能要求在具身代理移动的过程中保持对世界的连贯理解。与人类类似，代理必须利用自身的运动来解释观察中的变化，并相应地更新物体的位置和空间关系。尽管空间后训练显著拓宽了视觉-语言模型（VLMs）的空间智能，但它们在两个基本空间能力上仍然存在困难：自我运动跟踪和在运动过程中映射周围世界。为了解决这一问题，我们在每个训练回合中提供密集的多轮监督，针对交错的基本能力，模拟一个持续移动的代理在观察时进行推理的视觉经验。为了扩展这一方法，我们合成了11,738条以主题驱动的相机轨迹，涵盖广泛的3D场景，支持自我运动跟踪、持久物体映射和丰富的空间操作。在这些基本问题上进行推理的模型URQIU$_{ ext{Syn}}$-8B在我们的Uruqi基准测试中，52,000个问题跨2,700个回合的准确率从15.84%提高到47.73%。URUQI-SI-Mix-8B进一步达到50.41%，与GPT-6 Astra的50.08%相当。仅在我们的合成数据上训练，URQIU$_{ ext{Syn}}$-8B在三个外部空间基准测试中，相较于其InternVL3-8B基础模型实现了平均17.13%的相对准确率提升。这些结果突显了持续的视觉经验作为开发VLMs空间认知的可扩展监督源。
cs.CV / 95 / 2609.39222
DC-SAE: Deep Compression Semantic Autoencoder for Faster Diffusion Convergence
DC-SAE：用于加速扩散收敛的深度压缩语义自编码器
Abstract
High-compression tokenizers are essential for scaling latent image generative models. However, aggressive compression creates a fundamental tradeoff between reconstruction fidelity and generation efficiency: high compression image encoder always increases the learning difficulty of diffusion training, resulting in slow model convergence. Recent representation autoencoders speed up the diffusion training by improving the latent feature's expressive capability by replacing VAE encoders with pretrained semantic encoders, yet they are typically limited to moderate compression and lose pixel-level details necessary for faithful reconstruction. To achieve both high compression and fast diffusion training, we propose DC-SAE, a Decoupled Compact Semantic Autoencoder designed for high-compression image generation with accelerated diffusion model convergence. DC-SAE consists of two key components: (1) a macro-level architecture design that leverages semantic encoders to enable higher compression ratios, and (2) a pixel-level encoder that preserves low-level details, ensuring high-fidelity image reconstruction. We empirically demonstrate that DC-SAE performs strongly on image generation tasks, achieving both compact latent representations and efficient training dynamics. Specifically, on the ImageNet dataset with $512 \times 512$ resolution, DC-SAE achieves $32\times$ spatial compression, with 29.79 PSNR and 3.37 gFID, substantially outperforming the previous state-of-the-art high-compression tokenizer baselines DC-AE by 13.5% and 54.9% on PSNR and gFID, respectively, maintaining comparable throughput and faster diffusion model training convergence. Beyond class-conditional generation, a $1.6$B-parameter DiT using DC-SAE achieves 0.84 on GenEval and 86.007 on DPG-Bench for text-to-image generation at $1024\times1024$ resolution.
Chinese Translation
高压缩率的分词器对于扩展潜在图像生成模型至关重要。然而，激进的压缩在重建保真度和生成效率之间造成了根本性的权衡：高压缩图像编码器总是增加扩散训练的学习难度，导致模型收敛缓慢。近期的表示自编码器通过用预训练的语义编码器替换变分自编码器（VAE）编码器，提升了潜在特征的表达能力，从而加速了扩散训练，但它们通常仅限于中等压缩，并且丧失了忠实重建所需的像素级细节。为了实现高压缩和快速扩散训练，我们提出了DC-SAE，一种为高压缩图像生成而设计的解耦紧凑语义自编码器，旨在加速扩散模型的收敛。DC-SAE由两个关键组件组成：（1）利用语义编码器实现更高压缩比的宏观架构设计；（2）保留低级细节的像素级编码器，确保高保真图像重建。我们通过实证证明，DC-SAE在图像生成任务中表现优异，既实现了紧凑的潜在表示，又具备高效的训练动态。具体而言，在分辨率为$512 imes 512$的ImageNet数据集上，DC-SAE实现了$32 imes$的空间压缩，PSNR为29.79，gFID为3.37，显著超越了之前的最先进高压缩分词器基线DC-AE，PSNR和gFID分别提高了13.5%和54.9%，同时保持了可比的吞吐量和更快的扩散模型训练收敛。除了类条件生成外，使用DC-SAE的$1.6$B参数DiT在$1024 imes1024$分辨率的文本到图像生成中，在GenEval上达到了0.84，在DPG-Bench上达到了86.007。
cs.CV / 96 / 2609.39227
Emergent Multi-View Geometry Through Self-Distillation
通过自蒸馏实现的多视角几何
Abstract
Over a century ago, Henri Poincar\'e argued that a motionless observer cannot acquire the notion of space. Yet, most visual representation learning methods operate on individual images, while those that leverage multiple views rely on RGB reconstruction, entangling geometry with appearance. We propose Poincar3, a self-supervised method that learns representations from multiple views through self-distillation instead of RGB reconstruction. We combine masked patch and image-level distillation with a teacher that observes additional views, enabling training from scratch without explicit 3D supervision. Poincar3 outperforms both previous single and multi-view self-supervised approaches such as DINOv3, MuM, and Muskie on correspondence estimation, camera pose estimation, and 3D reconstruction. Using a lightweight Poincar\'e adapter, we also find that our learned features encode camera motion more accurately than existing self-supervised representations.
Chinese Translation
一个多世纪前，亨利·庞加莱（Henri Poincaré）曾论证，静止的观察者无法获得空间的概念。然而，大多数视觉表征学习方法都是基于单个图像，而那些利用多个视角的方法则依赖于RGB重建，将几何与外观纠缠在一起。我们提出了Poincar3，这是一种自监督方法，通过自蒸馏从多个视角学习表征，而不是依赖RGB重建。我们将掩蔽补丁和图像级蒸馏与观察额外视角的教师结合起来，使得在没有明确3D监督的情况下从头开始进行训练。Poincar3在对应估计、相机姿态估计和3D重建方面超越了之前的单视角和多视角自监督方法，如DINOv3、MuM和Muskie。使用轻量级的庞加莱适配器，我们还发现我们学习到的特征比现有的自监督表征更准确地编码了相机运动。
cs.CV / 97 / 2609.39265
Universal Cross-Prompt Adversarial Attacks on Promptable Concept Segmentation
通用跨提示对抗攻击在可提示概念分割中的应用
Abstract
The Segment Anything Model (SAM) achieves remarkable performance in visual segmentation. The latest SAM3 extends promptable segmentation to concept-level prediction, broadening the scope of segmentation foundation models. While recent works reveal that SAM and SAM2 are vulnerable to adversarial examples, the robustness of SAM3 under the concept segmentation paradigm remains unexplored. In addition, existing adversarial attacks on SAM-series models exhibit limited cross-prompt transferability. To this end, we propose AdvPCS, a universal cross-prompt adversarial attack for Promptable Concept Segmentation (PCS), including a min-max prompt optimization strategy, a global-local perception deception attack, and a temporal transition deviation attack. Specifically, we first identify the hardest-to-attack prompts via min-max bilevel optimization. In the inner maximization, we enhance diversity over candidate point, box, and text prompts. In the outer minimization, we select prompts with the highest responses based on the confidence scores output by the detector. Given the selected prompts, we apply the perception deception attack to minimize both global and local existence probabilities under joint prompting and employ the temporal memory misalignment attack to maximize inter-frame semantic inconsistency and corrupt memory pointers. Extensive experiments on four benchmark datasets show that a single universal adversarial perturbation (UAP) generated by our method generalizes across frames from different videos and achieves strong attack performance under point, box, and text prompts. In particular, it reduces the average mIoU of various PCS models on the SA-CO dataset to below 5% under text prompts, demonstrating strong attack ability.
Chinese Translation
Segment Anything Model (SAM) 在视觉分割方面表现出色。最新的 SAM3 将可提示分割扩展到概念级预测，拓宽了分割基础模型的应用范围。尽管近期研究表明 SAM 和 SAM2 对对抗样本存在脆弱性，但在概念分割范式下，SAM3 的鲁棒性尚未得到探讨。此外，现有针对 SAM 系列模型的对抗攻击表现出有限的跨提示可转移性。为此，我们提出了 AdvPCS，一种用于可提示概念分割（PCS）的通用跨提示对抗攻击，包括最小-最大提示优化策略、全局-局部感知欺骗攻击和时间过渡偏差攻击。具体而言，我们首先通过最小-最大双层优化识别最难攻击的提示。在内部最大化中，我们增强候选点、框和文本提示的多样性。在外部最小化中，我们根据检测器输出的置信度分数选择响应最高的提示。给定所选提示，我们应用感知欺骗攻击，以最小化联合提示下的全局和局部存在概率，并采用时间记忆错位攻击以最大化帧间语义不一致性并破坏记忆指针。在四个基准数据集上的大量实验表明，我们的方法生成的单一通用对抗扰动（UAP）能够在不同视频的帧之间进行泛化，并在点、框和文本提示下实现强大的攻击性能。特别是，它将 SA-CO 数据集上各种 PCS 模型的平均 mIoU 降低到 5% 以下，展示了强大的攻击能力。
cs.CV / 98 / 2609.39266
PLRS-IC: A Dual-Calibration Framework for Chest X-Ray Vision-Language Alignment
PLRS-IC：一种用于胸部X光影像-语言对齐的双重校准框架
Abstract
Fine-grained vision-language alignment in chest radiography enables zero-shot classification, grounding, and segmentation without task-specific annotations. However, this alignment is fundamentally hindered by two intertwined sources of ambiguity: projection-induced visual mismatch and patient-agnostic semantic overlap. First, at the local feature level, frontal and lateral radiographs exhibit distinct appearances for the same clinical finding, rendering a shared patch-text similarity geometry inherently suboptimal. Compounding this visual ambiguity is a semantic mismatch during global contrastive optimization, where instance-level objectives penalize cross-patient pairs as strict negatives even when they share identical positive clinical concepts. To address this dual ambiguity, we propose PLRS-IC, a unified dual-calibration framework for chest X-ray representation learning. At the local alignment stage, Projection-Conditioned Low-Rank Residual Similarity (PLRS) dynamically adapts patch-text matching to projection-specific manifolds using a bounded, parameter-efficient low-rank residual. At the global optimization stage, Information-Content-Calibrated Soft False-Negative Suppression (IC-SFNS) leverages a corpus-derived information-theoretic prior to soften the penalty of semantically overlapping negatives without altering original contrastive assignments. Extensive experiments across nine public zero-shot benchmark settings demonstrate that our framework yields consistent improvements in classification, grounding, and segmentation, validating the necessity of dual-calibration in medical vision-language pre-training.
Chinese Translation
胸部放射学中的细粒度影像-语言对齐使得在没有特定任务注释的情况下实现零样本分类、定位和分割成为可能。然而，这种对齐受到两种交织的模糊源的根本性阻碍：投影引起的视觉不匹配和与患者无关的语义重叠。首先，在局部特征层面，正面和侧面X光片对于相同的临床发现表现出不同的外观，使得共享的补丁-文本相似性几何结构本质上次优。加剧这种视觉模糊的是在全局对比优化过程中出现的语义不匹配，在此过程中，实例级目标将跨患者对视为严格的负样本，即使它们共享相同的正向临床概念。为了解决这种双重模糊，我们提出了PLRS-IC，一种用于胸部X光表示学习的统一双重校准框架。在局部对齐阶段，投影条件低秩残差相似性（PLRS）动态调整补丁-文本匹配，以适应特定于投影的流形，采用有界的、参数高效的低秩残差。在全局优化阶段，信息内容校准软假阴性抑制（IC-SFNS）利用基于语料库的信息论先验，减轻语义重叠负样本的惩罚，而不改变原始的对比分配。通过在九个公共零样本基准设置上的广泛实验，我们的框架在分类、定位和分割方面均显示出一致的改进，验证了医学影像-语言预训练中双重校准的必要性。
cs.CV / 99 / 2609.39273
MegaAvatar: Controllable Talking Avatar Generation
MegaAvatar：可控的对话头像生成
Abstract
This report presents \textbf{MegaAvatar}, a controllable talking avatar generation framework built on top of the Wan2.2-TI2V-5B model. Compared with previous talking-avatar methods that mainly rely on audio or reference-image conditioning, we introduce additional SMPL-X-derived 3D guidance, enabling global control over body pose and head motion. Specifically, we render the driving SMPL-X sequence into dense mesh frames and encode them with a lightweight 3D convolutional encoder, whose outputs are injected into the latent tokens to provide overall motion control. Furthermore, we extend Wan2.2-TI2V-5B with additional audio and face cross-attention modules to enable fine-grained expression control and preserve the input identity, respectively. In addition, we implement an audio-to-SMPL-X model to predict an SMPL-X sequence conditioned on the reference image and input audio, allowing MegaAvatar to support audio-driven inference without user-provided SMPL-X frames. Experiments show that MegaAvatar achieves high-quality talking avatar generation with controllable body and head motion, speech-synchronized facial expressions, and consistent identity preservation. MegaAvatar also supports inference with flexible resolutions and video lengths. Codes, dataset, models will be avaliable in https://github.com/Jeoyal/MegaAvatar
Chinese Translation
本报告介绍了 extbf{MegaAvatar}，一个基于Wan2.2-TI2V-5B模型构建的可控对话头像生成框架。与以往主要依赖音频或参考图像条件的对话头像方法相比，我们引入了额外的基于SMPL-X的3D指导，能够实现对身体姿势和头部运动的全局控制。具体而言，我们将驱动的SMPL-X序列渲染为密集的网格帧，并使用轻量级的3D卷积编码器对其进行编码，其输出被注入到潜在标记中，以提供整体运动控制。此外，我们扩展了Wan2.2-TI2V-5B，增加了额外的音频和面部交叉注意力模块，以实现细粒度的表情控制并保持输入身份。此外，我们实现了一个音频到SMPL-X的模型，以预测基于参考图像和输入音频条件的SMPL-X序列，使得MegaAvatar能够支持无用户提供SMPL-X帧的音频驱动推理。实验表明，MegaAvatar能够生成高质量的可控对话头像，具备可控的身体和头部运动、与语音同步的面部表情以及一致的身份保持。MegaAvatar还支持灵活的分辨率和视频长度推理。代码、数据集和模型将发布在 https://github.com/Jeoyal/MegaAvatar
cs.CV / 100 / 2609.39300
BMASH: Ball-Motion-Aware Soccer Header Spotting
BMASH：球运动感知的足球头球识别
Abstract
Recent advances in computer vision have made broadcast sports videos increasingly useful for event analysis, performance assessment, and player-safety applications. In soccer, however, header spotting remains a challenging problem due to the subtle and short-lived nature of header events. This paper focuses on soccer header spotting: identifying moments in broadcast videos where the ball contacts a player's head. We first adapt and evaluate Video Swin as a strong action-recognition baseline for this task, and then introduce BMASH, a ball-motion-aware fusion framework that integrates detector-derived ball features. BMASH combines Video Swin action representations with ball-presence and motion features from frame-level soccer-ball detection, integrating player-action context with ball dynamics to distinguish headers from visually similar events. We evaluate BMASH using game-level splits with separate test matches and rotating validation folds, considering both centered-window classification and continuous full-video spotting. Results show that Video Swin provides a strong baseline for header spotting, while BMASH improves clip-level AP and ROC-AUC over the corresponding Video Swin baseline. In continuous full-video spotting, BMASH achieves a comparable event-level F1-performance with a different precision--recall trade-off.
Chinese Translation
近年来，计算机视觉的进步使得广播体育视频在事件分析、性能评估和运动员安全应用方面变得越来越有用。然而，在足球中，头球识别仍然是一个具有挑战性的问题，因为头球事件的特征微妙且短暂。本文聚焦于足球头球识别：识别广播视频中球与球员头部接触的时刻。我们首先调整并评估了 Video Swin 作为该任务的强基线动作识别模型，随后引入了 BMASH，一个球运动感知的融合框架，整合了检测器派生的球特征。BMASH 将 Video Swin 动作表示与来自帧级足球检测的球存在和运动特征相结合，整合了球员动作上下文与球的动态，以区分头球与视觉上相似的事件。我们使用游戏级别的划分对 BMASH 进行评估，采用独立的测试比赛和轮换验证折叠，考虑了中心窗口分类和连续全视频识别两种情况。结果表明，Video Swin 为头球识别提供了强有力的基线，而 BMASH 在片段级别的平均精度（AP）和接收者操作特征曲线下面积（ROC-AUC）上超越了相应的 Video Swin 基线。在连续全视频识别中，BMASH 实现了与不同精确度-召回权衡下的事件级 F1 性能相当的结果。
cs.CV / 101 / 2609.39315
Rethinking Generative Image Compression at Extremely Low Bitrates
在极低比特率下重新思考生成图像压缩
Abstract
Generative image compression produces visually plausible reconstructions at low bitrates, yet their behavior as the rate approaches zero remains largely unexplored. When pushed below normal operating rates, representative codecs undergo semantic collapse: rather than gracefully losing source-specific detail, they produce malformed or unrecognizable content. Our analysis identifies two factors. As the bitrate decreases, reconstruction losses increasingly conflict with semantic objectives on gradients and visual results, while pixel-space and reconstruction-oriented VAE diffusion models become less efficient on semantic preservation. Guided by these findings, we introduce RAE-CoD, a compression-oriented diffusion (CoD) built in a representation autoencoder (RAE) space with direct alignment between compressed and source representations, preserving recognizable, naturally structured content for a $256\times256$ image with as few as 16 bits. We evaluate this framework using five vision foundation models (VFM) and a blinded vision-language model protocol. On MSCOCO-30K, RAE-CoD stands out from all evaluation. At 0.001-0.008 bpp, it reduces relative VFM feature MSE and Fr\'echet Distance ratio by at least 25.7% and 69.1% over the best competitors. Meanwhile, semantic recognizability and quality of the reconstructions remain nearly constant while source consistency falls smoothly, replacing abrupt semantic collapse with a graceful transition toward unconditional generation. Code will be released at https://github.com/LuizScarlet/RAE-CoD.
Chinese Translation
生成图像压缩在低比特率下能够产生视觉上合理的重建，但当比特率接近零时，其行为仍然在很大程度上未被探索。当比特率低于正常操作范围时，代表性的编解码器会经历语义崩溃：它们不仅仅是优雅地丧失源特定细节，而是产生畸形或无法识别的内容。我们的分析确定了两个因素。随着比特率的降低，重建损失与梯度和视觉结果上的语义目标之间的冲突日益加剧，而像素空间和以重建为导向的变分自编码器（VAE）扩散模型在语义保持方面变得效率降低。基于这些发现，我们引入了RAE-CoD，这是一种压缩导向的扩散（CoD），构建在表示自编码器（RAE）空间中，压缩表示与源表示之间直接对齐，能够为$256 imes256$的图像在仅有16比特的情况下保持可识别的、自然结构化的内容。我们使用五个视觉基础模型（VFM）和盲测视觉-语言模型协议对该框架进行了评估。在MSCOCO-30K上，RAE-CoD在所有评估中表现突出。在0.001-0.008 bpp的范围内，它将相对VFM特征的均方误差（MSE）和Fréchet距离比率分别降低了至少25.7%和69.1%，超越了最佳竞争对手。同时，重建的语义可识别性和质量几乎保持不变，而源一致性平稳下降，取代了突发的语义崩溃，优雅地过渡到无条件生成。代码将发布在https://github.com/LuizScarlet/RAE-CoD。
cs.CV / 102 / 2609.39335
TexTailor: Texture-Preserving Video Virtual Try-On via Adaptive Garment Conditioning
TexTailor：通过自适应服装调节实现的纹理保留视频虚拟试穿
Abstract
Video virtual try-on has attracted increasing attention due to its broad potential in digital fashion and intelligent e-commerce. However, existing methods primarily focus on low-resolution settings and still face substantial challenges when extended to high-resolution scenarios. These limitations can be attributed to two main factors: (1) the insufficient utilization of rich garment reference information, and (2) the lack of explicit positional modeling between garment and video representations during cross-modal interaction, which weakens fine-grained local correspondence. To address these issues, we propose TexTailor, a high-fidelity video virtual try-on framework built upon a pretrained video Diffusion Transformer. Specifically, we introduce a timestep-adaptive modulation mechanism to dynamically adjust garment visual representations throughout denoising. We further develop a frame-aligned positional encoding strategy to strengthen garment-to-video correspondence, together with a multi-source injection design that reduces interference among heterogeneous conditions. Extensive experiments on multiple video virtual try-on benchmarks, including the high-resolution Eevee dataset, demonstrate that TexTailor achieves competitive performance in garment detail preservation, temporal consistency, and overall video quality.
Chinese Translation
视频虚拟试穿因其在数字时尚和智能电子商务中的广泛潜力而受到越来越多的关注。然而，现有方法主要集中在低分辨率设置上，并在扩展到高分辨率场景时仍面临重大挑战。这些限制主要归因于两个因素：（1）对丰富服装参考信息的利用不足，以及（2）在跨模态交互过程中缺乏服装与视频表示之间的明确位置建模，导致细粒度局部对应关系的削弱。为了解决这些问题，我们提出了TexTailor，一个基于预训练视频扩散变换器（Diffusion Transformer）的高保真视频虚拟试穿框架。具体而言，我们引入了一种时间步自适应调制机制，以在去噪过程中动态调整服装视觉表示。我们进一步开发了一种帧对齐位置编码策略，以增强服装与视频之间的对应关系，并设计了一种多源注入机制，以减少异构条件之间的干扰。在多个视频虚拟试穿基准（包括高分辨率Eevee数据集）上的广泛实验表明，TexTailor在服装细节保留、时间一致性和整体视频质量方面表现出竞争力。
cs.CV / 103 / 2609.39363
Rethinking Multi-Image Re-Representation in Multi-Image Understanding
重新思考多图像理解中的多图像再表征
Abstract
Multi-image understanding requires MLLMs not only to recognise the content of individual images, but also to organise visual evidence distributed across them. We study this problem through multi-image re-representation, viewing prompted Chain-of-Thought reasoning and agentic visual tool use as different ways of re-organising visual evidence during reasoning. We introduce Mosaic, a general-purpose multi-image visual harness that enables an MLLM to actively construct visual intermediates with ten composable image operations. We compare five re-representation settings on existing multi-image benchmarks and on MosaicBench, a new grounding-focused benchmark for fine-grained multi-image understanding. Our experiments show that the relative benefits of textual and visual re-representation are strongly task-dependent. Visual re-representation is particularly effective for tasks requiring precise visual evidence, including hypothesis testing, precision comparison, and orientation-sensitive reasoning, while tasks dominated by higher-level semantic content show smaller or less consistent gains. Building on this finding, we train MosaicAgent-8B to use Mosaic with reinforcement learning using only accuracy and format rewards. Without demonstration trajectories or rewards for specific tool-use, the agent learns to compose visual operations over multiple steps and exhibits diverse problem-solving patterns unpromptedly. Code and data will be released at https://github.com/gengyuanmax/Mosaic.
Chinese Translation
多图像理解要求多模态大语言模型（MLLMs）不仅能够识别单个图像的内容，还能够组织分布在多个图像中的视觉证据。我们通过多图像再表征来研究这个问题，将提示的思维链推理和自主视觉工具使用视为在推理过程中重新组织视觉证据的不同方式。我们引入了Mosaic，一个通用的多图像视觉工具，能够使MLLM主动构建具有十种可组合图像操作的视觉中介。我们在现有的多图像基准和一个新的以基础为重点的细粒度多图像理解基准MosaicBench上比较了五种再表征设置。我们的实验表明，文本和视觉再表征的相对优势在很大程度上依赖于任务。视觉再表征对于需要精确视觉证据的任务特别有效，包括假设检验、精度比较和方向敏感推理，而以更高层次语义内容为主的任务则显示出较小或不一致的收益。基于这一发现，我们训练了MosaicAgent-8B，使其在仅使用准确性和格式奖励的情况下，通过强化学习使用Mosaic。在没有演示轨迹或特定工具使用奖励的情况下，代理学习在多个步骤中组合视觉操作，并自发地展现出多样化的问题解决模式。代码和数据将发布在 https://github.com/gengyuanmax/Mosaic。
cs.CV / 104 / 2609.39366
COBICount: Separating Object and Background Responses for Remote Sensing Object Counting Without Training on Target Data
COBICount：无目标数据训练的遥感目标计数中的物体与背景响应分离
Abstract
Remote sensing object counting estimates how many buildings, vehicles, or ships appear in overhead images. Most supervised counters predict a density map, whose sum gives the object count, and assume similar categories, sizes, and backgrounds. Applying them across regions, sensors, or categories often requires target data or further training, which may be costly or unavailable. We study source-only counting. Training for the counting task and model selection use one group of images that shares an object category and similar imaging conditions, with one point marking each object. Target images and information remain unavailable until the model is fixed. This reduces data preparation but makes transfer harder. A model trained on one source may place high density values, called responses, on real objects and repeated background structures. Road edges, parking grids, roof boundaries, and water boundaries may then be counted as objects, creating candidate origin ambiguity. COBICount separates response generation, acceptance, and background suppression. Candidate Evidence (CE) generates possible responses. Candidate Acceptance (CA) keeps compact responses centered on objects. Bias Isolation (BI) reduces responses associated with repeated background structures. Their outputs form the final density map. Trained on RSOC Building and evaluated directly on DOTA Large Vehicle, Small Vehicle, and Ship, COBICount achieves the lowest mean absolute error (MAE) averaged over the target domains among the compared methods, 174.132. It uses 5.07 million parameters and 17.41 billion floating point operations for a 512x512 input. COBICount improves transfer without target data or training for each target. The code will be available at: https://github.com/yixuxi22/COBICount.
Chinese Translation
遥感目标计数估计在空中图像中出现了多少建筑物、车辆或船只。大多数监督计数器预测一个密度图，其总和给出目标计数，并假设相似的类别、大小和背景。在不同区域、传感器或类别之间应用它们通常需要目标数据或进一步的训练，这可能代价高昂或不可用。我们研究仅基于源数据的计数。计数任务的训练和模型选择使用一组共享物体类别和相似成像条件的图像，每个物体由一个点标记。目标图像和信息在模型固定之前保持不可用。这减少了数据准备，但使得迁移变得更加困难。在一个源上训练的模型可能会在真实物体和重复的背景结构上放置高密度值，称为响应。道路边缘、停车网格、屋顶边界和水边界可能被计数为物体，从而产生候选源的模糊性。COBICount分离响应生成、接受和背景抑制。候选证据（Candidate Evidence, CE）生成可能的响应。候选接受（Candidate Acceptance, CA）保持集中于物体的紧凑响应。偏差隔离（Bias Isolation, BI）减少与重复背景结构相关的响应。它们的输出形成最终的密度图。在RSOC建筑上训练，并直接在DOTA大型车辆、小型车辆和船只上评估，COBICount在比较方法中实现了目标领域平均最低的平均绝对误差（Mean Absolute Error, MAE），为174.132。它使用507万个参数和174.1亿次浮点运算，输入为512x512。COBICount在没有目标数据或针对每个目标的训练的情况下改善了迁移。代码将可在以下网址获取：https://github.com/yixuxi22/COBICount。
cs.CV / 105 / 2609.39378
EgoTools: Towards Tool-Centric Reasoning in Real-World Egocentric Videos
EgoTools：面向现实世界自我中心视频的工具中心推理
Abstract
Real-world embodied tasks, from everyday activities to professional procedures, require agents to act under physical constraints while tracking evolving object and task states. Tool use sits at the heart of such tasks, as many everyday and professional activities are tool-mediated. Understanding them requires reasoning about affordances, hand-tool-object geometry, procedural progress, and causal effects on target objects. Yet despite strong performance on perception-oriented video tasks such as captioning and general video QA, current multimodal video models remain limited in this form of tool-centric embodied reasoning. Progress in this direction has been limited by the lack of real-world egocentric data and diagnostic benchmarks. To address this gap, we introduce EgoTools, the first comprehensive suite for egocentric tool-use understanding. It consists of two complementary components: EgoTools-Data, a large-scale corpus of 100 hours of tool-centric egocentric recordings with synchronized audio, dense captions, reasoning-heavy narrations, and supplementary 3D information; and EgoTools-Bench, a diagnostic benchmark of 1,000 QA pairs across four tracks that cover tool-use understanding from perception and geometry to procedure and causal reasoning. Experimental results show that current models still struggle to ground tool use in visual evidence: Gemini-3.1-Pro achieves 66.9% overall accuracy but only 51.7% on Perception & Grounding. Beyond evaluation, we validate EgoTools-Data as a training resource. On the full 1,000-question benchmark, full supervised fine-tuning improves Qwen3-VL-8B-Instruct from 50.0% to 60.9%, under strict source-video separation. Together, these results establish EgoTools as a unified resource for both training and diagnostic evaluation of real-world egocentric tool-use understanding.
Chinese Translation
现实世界的具身任务，从日常活动到专业程序，要求代理在物理约束下行动，同时跟踪不断变化的物体和任务状态。工具使用是此类任务的核心，因为许多日常和专业活动都是通过工具进行的。理解这些任务需要推理关于可供性、手-工具-物体几何、程序进展以及对目标物体的因果影响。然而，尽管在以感知为导向的视频任务（如字幕生成和一般视频问答）上表现出色，当前的多模态视频模型在这种工具中心的具身推理方面仍然有限。朝这个方向的进展受限于缺乏现实世界的自我中心数据和诊断基准。为了解决这一空白，我们推出了EgoTools，这是第一个全面的自我中心工具使用理解套件。它由两个互补的组成部分构成：EgoTools-Data，一个包含100小时工具中心自我中心录音的大规模语料库，配有同步音频、密集字幕、重推理叙述和补充的3D信息；以及EgoTools-Bench，一个涵盖工具使用理解的诊断基准，包括感知、几何、程序和因果推理四个领域的1,000个问答对。实验结果表明，当前模型在将工具使用与视觉证据结合方面仍然存在困难：Gemini-3.1-Pro的整体准确率为66.9%，但在感知与基础方面仅为51.7%。除了评估，我们还验证了EgoTools-Data作为训练资源的有效性。在完整的1,000问题基准上，全面监督微调将Qwen3-VL-8B-Instruct的准确率从50.0%提高到60.9%，在严格的源视频分离下。这些结果共同确立了EgoTools作为一个统一资源，用于现实世界自我中心工具使用理解的训练和诊断评估。
cs.CV / 106 / 2609.39380
InfoAgent: Traceable Generation and Repair of Evidence-Grounded Infographics
信息代理：可追溯的基于证据的信息图生成与修复
Abstract
Reliable infographic generation requires facts, symbols, and visual relations to remain consistent through rendering and revision. Correcting one element also requires tracking its supporting evidence and the dependencies affected by the change. We present \textbf{InfoAgent}, a training-free framework for \emph{evidence-bound visual-symbolic program synthesis}. Its Infographic Visual Description (IVD) records factual payloads, evidence provenance, execution routes, and verification obligations in a typed dependency graph. Retrieved design priors guide compilation, and layered execution combines raster synthesis with editable symbolic and binding objects while retaining their traces. Dependency-aware repair localizes corrections, rechecks affected dependencies, and requires protected obligations to remain satisfied under the declared checkers. Unresolved obligations remain explicit. On IGenBench, InfoAgent achieves 93.0 Q-ACC and 59.0 I-ACC. We also introduce InfoGraphicBench-Evidence, where complete-checklist pass rates on 200 test requests increase from 21.5\% for Same-IVD Prompt to 23.5\% for the initial layered output and 28.5\% after repair, using the same evidence and initial IVD. On 120 audited repair cases, localized repair edits 12.4\% of the canvas on average, compared with 67.3\% for global regeneration.
Chinese Translation
可靠的信息图生成要求在渲染和修订过程中保持事实、符号和视觉关系的一致性。修正一个元素还需要追踪其支持证据以及受更改影响的依赖关系。我们提出了 extbf{信息代理（InfoAgent）}，这是一个无训练的框架，用于 extit{基于证据的视觉符号程序合成}。其信息图视觉描述（Infographic Visual Description，IVD）记录了事实负载、证据来源、执行路径和验证义务，形成一个类型化的依赖图。检索到的设计先验指导编译，分层执行将光栅合成与可编辑的符号和绑定对象相结合，同时保留其痕迹。依赖感知修复定位修正，重新检查受影响的依赖关系，并要求在声明的检查器下保持受保护的义务满足。未解决的义务保持明确。在IGenBench上，信息代理（InfoAgent）达到了93.0的Q-ACC和59.0的I-ACC。我们还引入了InfoGraphicBench-Evidence，在200个测试请求上，完整清单的通过率从Same-IVD提示的21.5\%提高到初始分层输出的23.5\\%，并在修复后达到28.5\\%，使用相同的证据和初始IVD。在120个审计修复案例中，局部修复平均编辑了12.4\\%的画布，而全局再生则为67.3\\%。
cs.CV / 107 / 2609.39427
PCB-MC: Missing Component Analysis in Printed Circuit Boards
PCB-MC：印刷电路板中的缺失组件分析
Abstract
Detecting missing components on printed circuit boards (PCBs) differs fundamentally from conventional object detection, as the model must localize components that are not present. We introduce PCB-MC, a curated dataset for missing component detection with footprint level annotations built on top of the RF100 dataset. The dataset contains 197 distinct board types, each corresponding to a unique PCB design, with multiple augmented samples per type. We also provide benchmark results on PCB-MC by evaluating a diverse set of supervised and unsupervised methods. To ensure fair evaluation, we propose board type aware cross validation splits that prevent layout leakage between training and test sets. Supervised models showcase high false negative rates on unseen board designs, and unsupervised anomaly detection methods fail entirely due to the lack of spatial alignment with a board specific reference. These results confirm that missing component detection on diverse PCB layouts remains an open challenge. We release PCB-MC and all training protocols to support reproducible research on structural absence detection in industrial inspection.
Chinese Translation
在印刷电路板（PCBs）上检测缺失组件与传统的物体检测有根本性的不同，因为模型必须定位不存在的组件。我们介绍了PCB-MC，这是一个针对缺失组件检测的精心策划的数据集，包含基于RF100数据集构建的脚印级注释。该数据集包含197种不同类型的电路板，每种类型对应一个独特的PCB设计，并且每种类型有多个增强样本。我们还通过评估多种监督和无监督方法提供了PCB-MC的基准结果。为了确保公平评估，我们提出了考虑电路板类型的交叉验证划分，以防止训练集和测试集之间的布局泄漏。监督模型在未见过的电路板设计上展示了高假阴性率，而无监督异常检测方法由于缺乏与特定电路板参考的空间对齐而完全失败。这些结果确认了在多样化PCB布局上检测缺失组件仍然是一个开放的挑战。我们发布了PCB-MC及所有训练协议，以支持工业检测中结构缺失检测的可重复研究。
cs.CV / 108 / 2609.39429
Towards Trustworthy AI for Glioma Diagnosis: A Task-Aware Evaluation of Uncertainty Quantification
迈向可信赖的人工智能用于胶质瘤诊断：任务感知的不确定性量化评估
Abstract
Uncertainty Quantification (UQ) is a key requirement for trustworthy AI in high-stakes medical image analysis. In this work, we evaluate UQ in a multi-task Deep Learning framework for MRI-based glioma diagnosis that performs tumor segmentation and predicts IDH mutation status, 1p/19q co-deletion status, and tumor grade. Monte Carlo Dropout (MCD) is used for a detailed task-aware analysis of predictive, aleatoric, and epistemic uncertainty. We assess MC sample convergence, calibration, error detection, selective prediction, associations with segmentation performance, and the effect of voxel-wise uncertainty aggregation on case-level reliability. We also compare MCD with Deep Ensembles (DE) and Monte Carlo Deep Ensembles (MCDE), examine interactions between segmentation quality and classification, and evaluate a composite trust score integrating segmentation and classification uncertainty. Across tasks, uncertainty estimates supported meaningful error detection, while calibration depended on the dropout rate, with moderate rates yielding the most reliable probabilities. Uncertainty decomposition provided task-dependent interpretability but did not consistently improve error detection over predictive uncertainty alone. DE and MCDE showed comparable operational utility, with no method consistently dominating across tasks and metrics. The composite trust score did not consistently outperform classification uncertainty for selective prediction. Overall, our results provide a task-aware evaluation strategy and practical guidance for the development of trustworthy AI for glioma diagnosis.
Chinese Translation
不确定性量化（UQ）是高风险医学图像分析中可信赖人工智能的关键要求。在本研究中，我们在一个多任务深度学习框架中评估UQ，该框架用于基于MRI的胶质瘤诊断，执行肿瘤分割并预测IDH突变状态、1p/19q共缺失状态和肿瘤等级。我们使用蒙特卡洛丢弃（Monte Carlo Dropout, MCD）对预测性、不确定性和认知不确定性进行详细的任务感知分析。我们评估了蒙特卡洛样本收敛性、校准、错误检测、选择性预测、与分割性能的关联，以及体素级不确定性聚合对病例级可靠性的影响。我们还将MCD与深度集成（Deep Ensembles, DE）和蒙特卡洛深度集成（Monte Carlo Deep Ensembles, MCDE）进行了比较，考察了分割质量与分类之间的相互作用，并评估了一个综合信任评分，该评分整合了分割和分类的不确定性。在各项任务中，不确定性估计支持了有意义的错误检测，而校准依赖于丢弃率，中等丢弃率产生了最可靠的概率。不确定性分解提供了任务依赖的可解释性，但并未始终优于单独的预测不确定性在错误检测上的表现。DE和MCDE显示出可比的操作效用，没有一种方法在所有任务和指标中始终占优。综合信任评分在选择性预测中并未始终优于分类不确定性。总体而言，我们的结果提供了一种任务感知的评估策略和可信赖人工智能在胶质瘤诊断开发中的实用指导。
cs.CV / 109 / 2609.39441
CAST: Causal Advantage-Structured Training with Spatially Grounded Compositional Rewards for Diffusion Models
CAST：具有空间基础组合奖励的因果优势结构训练用于扩散模型
Abstract
Online reinforcement learning has been extended to flow matching for diffusion model (DM) image generation. However, this paradigm faces three limitations: (1) Window selection. Existing methods manually set the stochastic differential equation (SDE) sampling window, i.e., the denoising steps where exploration noise is injected. We instead determine it from each model's denoising trajectory. (2) Reward saturation. Current methods rely on scoring models trained on human annotations; we find that such scores are extremely high and nearly indistinguishable on the latest SOTA open-source DMs, making advantage estimation largely ineffective. (3) Sample inefficiency. A single scalar reward collapses different failure modes into almost identical scores, leaving minimal gradient guidance for targeted improvement. To address these issues, we propose CAST (Causal Advantage-Structured Training), an RL fine-tuning method for pretrained DMs, which (1) identifies the denoising step at which each model fixes the objects and their spatial arrangement in the image and uses that timing to set the SDE window, (2) decomposes each prompt via Causal Scene Graphs (CSG) into verifiable-atoms, i.e., minimal semantic units such as an object, count, attribute, or spatial relation that can each be checked independently, and rewards each atom separately, and (3) projects the signed atom-level advantages into pixel space through teacher-forced attention and uses them to spatially weight the SDE policy objective. We fine-tune two of the strongest open-source DMs, FLUX.2-dev and Qwen-Image-2512, with CAST, and evaluate them on GenEval 2, a compositional benchmark, and on Qwen-Image-Bench for overall quality. Within almost the same training budget, CAST's improvement over the base model on the most challenging GenEval 2 prompts is up to 3.07x that of Flow-GRPO, while overall generation quality also improves.
Chinese Translation
在线强化学习已扩展到流匹配，以用于扩散模型（DM）图像生成。然而，这一范式面临三大限制：（1）窗口选择。现有方法手动设置随机微分方程（SDE）采样窗口，即注入探索噪声的去噪步骤。我们则从每个模型的去噪轨迹中确定该窗口。（2）奖励饱和。当前方法依赖于基于人类注释训练的评分模型；我们发现这些分数极高，并且在最新的SOTA开源DM上几乎无法区分，使得优势估计在很大程度上失效。（3）样本低效。单一的标量奖励将不同的失败模式压缩为几乎相同的分数，留下的梯度指导极少，难以实现针对性改进。为了解决这些问题，我们提出了CAST（因果优势结构训练），这是一种针对预训练DM的强化学习微调方法，具体包括：（1）识别每个模型在图像中固定对象及其空间排列的去噪步骤，并利用该时机设置SDE窗口；（2）通过因果场景图（Causal Scene Graphs，CSG）将每个提示分解为可验证原子，即最小语义单元，如对象、数量、属性或空间关系，这些原子可以独立检查，并分别给予奖励；（3）通过教师强制注意力将带符号的原子级优势投影到像素空间，并利用它们对SDE策略目标进行空间加权。我们使用CAST对两个最强的开源DM，FLUX.2-dev和Qwen-Image-2512进行微调，并在组合基准GenEval 2和整体质量的Qwen-Image-Bench上进行评估。在几乎相同的训练预算下，CAST在最具挑战性的GenEval 2提示上对基础模型的改进达到了Flow-GRPO的3.07倍，同时整体生成质量也有所提升。
cs.CV / 110 / 2609.39451
ResARC: Residual-Aware AutoRegressive Coding for Ultra-Low Bitrate Image Compression
ResARC：面向超低比特率图像压缩的残差感知自回归编码
Abstract
Progressive autoregressive image codecs provide an appealing paradigm for generative compression by quantizing continuous latents into discrete tokens, transmitting coarse-to-fine prefix tokens and generating the remaining suffix tokens at the decoder. However, their reconstruction quality is fundamentally limited by two residuals introduced along this pipeline: the quantization residual, arising from information loss during discrete tokenization, and the generation residual, resulting from imperfect autoregressive generation of the suffix tokens. To address these limitations, we introduce ResARC, a residual-aware autoregressive codec that explicitly compensates for both residuals at the decoder. Specifically, we generate the quantization residual with a diffusion transformer conditioned on the autoregressive decoding context, while requiring no additional side information. In parallel, we compute the generation residual at the encoder and employ a learned Generation Residual Codec to efficiently compress and transmit it for decoder-side correction. The recovered residuals are then integrated with the reconstructed latent representation and decoded through an adapted VAE decoder. Extensive experiments demonstrate that ResARC achieves competitive perceptual similarity while substantially improving distributional fidelity over leading generative codecs across the ultra-low bitrate regime. Code and models will be released soon.
Chinese Translation
渐进式自回归图像编解码器通过将连续潜变量量化为离散标记，提供了一种吸引人的生成压缩范式，传输粗到细的前缀标记，并在解码器处生成剩余的后缀标记。然而，它们的重建质量在根本上受到沿该管道引入的两个残差的限制：量化残差，源于离散标记化过程中的信息损失，以及生成残差，源于后缀标记的自回归生成不完美。为了解决这些限制，我们引入了ResARC，一种残差感知自回归编解码器，明确地在解码器处补偿这两个残差。具体而言，我们使用条件于自回归解码上下文的扩散变换器生成量化残差，同时不需要额外的侧信息。与此同时，我们在编码器处计算生成残差，并采用学习到的生成残差编解码器高效地压缩和传输该残差，以便进行解码器侧的修正。恢复的残差随后与重建的潜在表示结合，并通过改进的变分自编码器（VAE）解码器解码。大量实验表明，ResARC在超低比特率范围内实现了与领先的生成编解码器相比具有竞争力的感知相似性，同时显著提高了分布保真度。代码和模型将很快发布。
cs.CV / 111 / 2609.39467
DensePed-Lite: Quality-Aware Adaptive Detection for Dense Pedestrians under Occlusion
DensePed-Lite：针对遮挡下密集行人的质量感知自适应检测
Abstract
Pedestrian detection plays a crucial role in computer vision with applications in autonomous driving, surveillance, and public safety. However, real-world dense scenes bring severe challenges, including heavy occlusion, drastic scale variations, and strict real-time requirements. Existing lightweight detectors struggle to balance accuracy and efficiency while often neglecting quality-aware feature modeling and consistency between classification and localization, leading to unstable performance under crowded conditions. To address these issues, we propose DensePed-Lite, a unified framework built on a single principle: under occlusion the network should adapt its behavior to the quality of what it observes rather than assume complete information. This principle is realized at three points where occlusion does the most damage: unreliable confidence scoring (UQE), fragmented spatial coverage (MPSC), and incoherent multi-scale fusion (CTDM). The three mechanisms reinforce one another instead of acting in isolation, all without significantly increasing complexity. Experiments on CityPersons and CrowdHuman validate that DensePed-Lite achieves a superior accuracy-efficiency trade-off compared with recent state-of-the-art lightweight methods, making it suitable for real-time deployment in dense pedestrian scenarios.
Chinese Translation
行人检测在计算机视觉中扮演着至关重要的角色，广泛应用于自动驾驶、监控和公共安全。然而，现实世界中的密集场景带来了严重的挑战，包括严重的遮挡、剧烈的尺度变化和严格的实时要求。现有的轻量级检测器在准确性和效率之间难以取得平衡，往往忽视了质量感知特征建模以及分类与定位之间的一致性，导致在拥挤条件下性能不稳定。为了解决这些问题，我们提出了DensePed-Lite，一个基于单一原则的统一框架：在遮挡情况下，网络应根据观察到的信息质量调整其行为，而不是假设获得了完整的信息。该原则在遮挡造成最大损害的三个点上得以实现：不可靠的置信评分（UQE）、碎片化的空间覆盖（MPSC）和不一致的多尺度融合（CTDM）。这三种机制相互强化，而不是孤立工作，且在不显著增加复杂性的情况下实现。对CityPersons和CrowdHuman的实验验证了DensePed-Lite在准确性与效率的权衡上优于最近的最先进轻量级方法，使其适合于密集行人场景下的实时部署。
cs.CV / 112 / 2609.39486
From Wrecks to Wisdom: Recovering Crash Mechanics from Real-World Multi-View Photos
从事故到智慧：从真实世界多视角照片中恢复碰撞力学
Abstract
Estimating accident mechanics from real-world crashes is important for vehicle-safety analysis, injury modeling, crash-severity prediction, and operational workflows such as insurance claim triage. In standard crash records, key metadata such as impact configuration, principal direction of force, and change in velocity ($\Delta V$) may be missing, delayed, or corrupted, while post-crash photographs are widely available and contain rich visual evidence of deformation. We study how much crash-mechanics information can be recovered directly from vehicle photos when structured signals are absent. We formulate crash understanding as supervised prediction from per-case multi-view photo sets. Targets include six Collision Deformation Classification (CDC) descriptors and the longitudinal and lateral components of reconstructed $\Delta V$. Each photo is encoded by a shared visual backbone, and the resulting view-level features are fused into a case-level representation from which target-specific heads predict crash descriptors. Using 15.2k training cases from the Crash Investigation Sampling System, drawn from about 1.5M photos before filtering, together with 1.15k validation and 1.15k test cases, we define an evaluation protocol for vision-based crash descriptor estimation from incomplete multi-view evidence. Post-crash imagery alone provides usable signal for several non-trivial crash-mechanics descriptors, while weakly observable and long-tailed targets remain challenging. Within the compared training regimes, the selected joint-training recipe reduces mean absolute angular error for principal direction of force from 20.1 to 14.05 degrees and longitudinal $\Delta V$ MAE from 8.04 to 7.45 km/h. Our work provides a reference point for future multimodal fusion with structured crash metadata.
Chinese Translation
从真实世界的事故中估计事故力学对于车辆安全分析、伤害建模、碰撞严重性预测以及保险索赔分流等操作流程至关重要。在标准的碰撞记录中，关键元数据如冲击配置、主力方向和速度变化（$ ext{ΔV}$）可能缺失、延迟或损坏，而事故后的照片则广泛可用，并包含丰富的变形视觉证据。我们研究在缺乏结构化信号的情况下，可以从车辆照片中直接恢复多少碰撞力学信息。我们将碰撞理解形式化为从每个案例的多视角照片集中进行监督预测。目标包括六个碰撞变形分类（Collision Deformation Classification, CDC）描述符以及重建的纵向和横向$ ext{ΔV}$分量。每张照片通过共享的视觉骨干网络进行编码，生成的视图级特征融合为案例级表示，从中目标特定的头部预测碰撞描述符。使用来自碰撞调查抽样系统的15.2k训练案例，这些案例来源于约1.5M张照片（经过筛选），以及1.15k验证和1.15k测试案例，我们定义了一种基于视觉的碰撞描述符估计的评估协议，适用于不完整的多视角证据。仅依靠事故后图像就能为多个非平凡的碰撞力学描述符提供可用信号，而弱可观察和长尾目标仍然具有挑战性。在比较的训练方案中，所选的联合训练方案将主力方向的平均绝对角度误差从20.1度降低到14.05度，纵向$ ext{ΔV}$的平均绝对误差从8.04 km/h降低到7.45 km/h。我们的工作为未来与结构化碰撞元数据的多模态融合提供了参考点。
cs.CV / 113 / 2609.39490
OmniReasoning: Pushing the Limits of Audio-Visual Joint Reasoning
全方位推理：推动音频-视觉联合推理的极限
Abstract
Recent advances have enabled unified omni-modal models in understanding audio, vision, and language. However, existing benchmarks, training data, and learning methods largely treat the modalities independently, leaving the capability of audio-visual joint reasoning poorly evaluated and insufficiently elicited. We address this gap with a benchmark, data engine, and learning method. First, we introduce OmniReasoningBench, a benchmark where both audio and visual evidence are indispensable. It comprises 1,150 multiple-choice and open-ended questions across two tasks, reasoning over video and reasoning beyond video. Second, we develop a data engine OmniQA. It automatically constructs evidence-grounded QA pairs that explicitly necessitate audio-visual joint reasoning, together with time-stamped clue chains that guide the annotation of thinking process. Besides our benchmark, this engine produces training data OmniReasoning-SFT-112K and OmniReasoning-RL-19K. Finally, we propose an on-policy self-distillation method Modality-Factored Self-Distillation (MFSD). It evaluates each sampled response under modality-specific clue contexts, disentangling the contributions of individual clues and their cross-modal interactions for token-level credit assignment. With our training data and learning method, our model OmniReasoning-30B-A3B achieves 50.0% on OmniVideoBench and 42.5% on OmniReasoningBench, improving the base model Qwen3-Omni-30B-A3B-Thinking by 12.8 and 9.3 percentage points, respectively. Moreover, it delivers substantial gains on general and long-video benchmarks, including Video-MME-v2. We hope our work offers a solid step for facilitating future research in omni-modal joint reasoning.
Chinese Translation
近期的进展使得统一的全模态模型在理解音频、视觉和语言方面得以实现。然而，现有的基准、训练数据和学习方法在很大程度上独立对待这些模态，导致音频-视觉联合推理的能力评估不足且未得到充分挖掘。我们通过一个基准、数据引擎和学习方法来填补这一空白。首先，我们介绍了 OmniReasoningBench，这是一个音频和视觉证据不可或缺的基准。它包含1150个多项选择和开放式问题，涵盖视频推理和超越视频的推理两个任务。其次，我们开发了数据引擎 OmniQA。它自动构建明确需要音频-视觉联合推理的证据基础问答对，并提供时间戳线索链以指导思维过程的注释。除了我们的基准外，该引擎还生成了训练数据 OmniReasoning-SFT-112K 和 OmniReasoning-RL-19K。最后，我们提出了一种在线自蒸馏方法——模态因子自蒸馏（Modality-Factored Self-Distillation, MFSD）。它在特定模态线索上下文中评估每个采样响应，解开单个线索及其跨模态交互对标记级别信用分配的贡献。借助我们的训练数据和学习方法，我们的模型 OmniReasoning-30B-A3B 在 OmniVideoBench 上达到了 50.0%，在 OmniReasoningBench 上达到了 42.5%，分别提高了基础模型 Qwen3-Omni-30B-A3B-Thinking 12.8 和 9.3 个百分点。此外，它在一般和长视频基准（包括 Video-MME-v2）上也取得了显著提升。我们希望我们的工作为促进未来的全模态联合推理研究提供坚实的一步。
cs.CV / 114 / 2609.39492
Front-to-Back: Benchmarking Vision-Language Models for Asymmetric Cross-View Vehicle Re-Identification
前后视角：非对称跨视角车辆重识别的视觉-语言模型基准评估
Abstract
Matching the same vehicle across front and rear cameras is difficult because the cameras do not share a view and the vehicle's appearance changes substantially. We introduce Front2Back-ReID, a benchmark of 500 manually verified vehicle handovers from 20 recording sequences in South Africa. Each example asks a model to match a vehicle highlighted in a front-camera image to the same vehicle among at least three candidates in a later rear-camera image. We evaluate seven zero-shot vision-language models, four image-retrieval baselines, and 25 human participants. Models are tested using full front RGB images, cropped target vehicles, and binary silhouettes. The strongest VLM achieved 76.6 percent Rank-1 accuracy on target crops, compared with 74.0 percent for the frozen SigLIP2 baseline; this difference was not statistically clear. Human participants achieved 94.0 percent accuracy with full images and 92.2 percent with target crops. Under our evaluation setup, enabling reasoning improved accuracy across all three input conditions for every model evaluated in both modes. We also found that VLMs generally performed worse on full scenes than on target crops. These results show that general-purpose VLMs do not yet consistently outperform strong visual retrieval for front-to-rear vehicle matching, while humans remain substantially more reliable.
Chinese Translation
在前后摄像头之间匹配同一辆车辆是困难的，因为摄像头的视角不同且车辆的外观变化显著。我们引入了Front2Back-ReID，这是一个包含来自南非20个录制序列的500个手动验证车辆交接的基准数据集。每个示例要求模型将前摄像头图像中突出显示的车辆与后摄像头图像中至少三个候选者中的同一辆车辆进行匹配。我们评估了七个零样本视觉-语言模型、四个图像检索基线和25名人类参与者。模型使用完整的前RGB图像、裁剪的目标车辆和二进制轮廓进行测试。最强的视觉-语言模型在目标裁剪上达到了76.6%的Rank-1准确率，而冻结的SigLIP2基线为74.0%；这一差异在统计上并不显著。人类参与者在完整图像上达到了94.0%的准确率，在目标裁剪上为92.2%。在我们的评估设置下，启用推理在所有三种输入条件下提高了每个模型在两种模式下的准确率。我们还发现，视觉-语言模型在完整场景上的表现通常不如在目标裁剪上的表现。这些结果表明，通用视觉-语言模型在前后车辆匹配中尚未始终优于强大的视觉检索，而人类的可靠性仍然显著更高。
cs.CV / 115 / 2609.39504
PartiCam: Camera Controlled Video Generation with Reward Guidance
PartiCam：基于奖励引导的相机控制视频生成
Abstract
We present PartiCam, a training-free Particle filtering rooted method for improved Camera controlled video generation. Generating videos that follow a precisely specified camera trajectory remains challenging for large video diffusion models. Training-free approaches are backbone-agnostic and avoid the need to construct large camera-annotated datasets by steering pretrained models toward the desired camera motion at test time. This enables the generation of camera-controlled video data that can subsequently be used to train camera-conditioned video diffusion models. Existing sampling-based guidance approaches often suffer from unstable trajectories: they either explore too broadly and fail to respect the target camera motion or collapse early and lose visual diversity over time. We introduce a global-local refinement framework for diffusion reward guidance, enabling accurate and consistent camera control during video generation. Our method builds on Sequential Monte-Carlo (SMC) guidance, but introduces a local refinement stage based on particle filtered resampling. Experiments show large improvements in camera trajectory adherence, reduced drift, and better visual quality, without requiring model retraining.
Chinese Translation
我们提出了PartiCam，一种无训练的基于粒子滤波的方法，用于改进相机控制的视频生成。生成遵循精确指定相机轨迹的视频对于大型视频扩散模型仍然具有挑战性。无训练的方法不依赖于特定的骨干网络，避免了构建大型相机标注数据集的需求，通过在测试时引导预训练模型朝向期望的相机运动。这使得生成的相机控制视频数据可以随后用于训练相机条件的视频扩散模型。现有的基于采样的引导方法往往面临不稳定的轨迹：它们要么探索过于广泛，未能遵循目标相机运动，要么过早崩溃，导致视觉多样性随时间下降。我们引入了一种全局-局部的精细化框架用于扩散奖励引导，从而在视频生成过程中实现准确且一致的相机控制。我们的方法基于序列蒙特卡洛（Sequential Monte-Carlo, SMC）引导，但引入了基于粒子滤波重采样的局部精细化阶段。实验表明，在相机轨迹遵循性、漂移减少和视觉质量方面均有显著改善，而无需重新训练模型。
cs.CV / 116 / 2609.39527
Lens Flare Removal and Reconstruction
镜头光晕去除与重建
Abstract
The presence of lens flares in images can significantly reduce the quality of downstream application results for tasks such as 3D scene reconstruction. This is because lens flares are a property of the camera imaging system, and not a part of the underlying scene being modeled. There are previous methods that tackle the removal of small flares focused around a light source. However, existing methods struggle with large flares, such as those that fill the entire image. In this work, we compile a novel dataset for large-flare removal, combining publicly available real-world data with a procedural generation pipeline. We fine-tune a diffusion-based model on our dataset to remove complex, large lens flares. On the other hand, lens flares remain effective artistic tools, widely used in the media. While there are ways to simulate 2D flares, representing and reconstructing lens flares consistently across multiple views has not yet been explored. To achieve this, we introduce a flare representation model that leverages the symmetry of lens flares about the camera's principal point. We propose a computational pipeline to jointly optimize this flare model and a Gaussian splatting model (3DGS). This enables the decomposition of a 3D scene into lens flares and the scene itself, using our flare-removal model. Because the reconstructed flare is explicit and re-renderable, it can be edited and transferred to novel images and new 3D scenes. We evaluate removal on an established benchmark and a new one for large reflective flares, quantify the flare/scene decomposition directly, and show that the pipeline is robust to errors in automatic light-source localization.
Chinese Translation
图像中的镜头光晕会显著降低下游应用结果的质量，例如在三维场景重建任务中。这是因为镜头光晕是相机成像系统的特性，而不是所建模的基础场景的一部分。之前有一些方法专注于去除围绕光源的小光晕。然而，现有方法在处理大型光晕时表现不佳，例如那些填满整个图像的光晕。在这项工作中，我们编制了一个用于大光晕去除的新数据集，结合了公开可用的真实世界数据和程序生成管道。我们在该数据集上微调了一个基于扩散的模型，以去除复杂的大型镜头光晕。另一方面，镜头光晕仍然是有效的艺术工具，广泛应用于媒体。虽然有方法可以模拟二维光晕，但在多个视角下一致地表示和重建镜头光晕尚未得到探索。为此，我们引入了一种光晕表示模型，该模型利用镜头光晕相对于相机主点的对称性。我们提出了一种计算管道，以联合优化该光晕模型和高斯点云模型（3DGS）。这使得可以使用我们的光晕去除模型将三维场景分解为镜头光晕和场景本身。由于重建的光晕是明确且可重新渲染的，因此可以对其进行编辑并转移到新图像和新的三维场景中。我们在一个已建立的基准和一个新的大型反射光晕基准上评估去除效果，直接量化光晕/场景分解，并展示该管道对自动光源定位中的错误具有鲁棒性。
cs.CV / 117 / 2609.39542
Comparative study of adapting pre-trained models for driving behavior video captioning
预训练模型在驾驶行为视频字幕生成中的适应性比较研究
Abstract
This report examines and compares some of the many fine tuning and prompting methods existing, applying them within the domain of autonomous driving. The idea is to compare these methods by adapting a Large Language Model (LLM) on a video dataset. LLM's have become extremely good at achieving a good understanding of different forms of data and this study aims to induce a low dimensional understanding of driving situations into our primary test model SpaceTimeGPT. Experiments on BDD-X (Berkeley DeepDrive eXplanation) dataset demonstrate good performance of the full fine tuning framework on some automatic metrics, and in some metrics, it even surpasses the baseline. We also try Low-Rank Adaptation (LoRA) and prompt engineering on VideoLLaVA model and discuss its limitations.
Chinese Translation
本报告考察并比较了现有的多种微调和提示方法，并将其应用于自动驾驶领域。研究的目的是通过在视频数据集上适应大型语言模型（Large Language Model, LLM）来比较这些方法。LLM在理解不同形式的数据方面表现出色，本研究旨在将驾驶情境的低维理解引入我们的主要测试模型SpaceTimeGPT。对BDD-X（Berkeley DeepDrive eXplanation）数据集的实验表明，完整微调框架在某些自动评估指标上表现良好，甚至在某些指标上超越了基线。我们还尝试了低秩适应（Low-Rank Adaptation, LoRA）和VideoLLaVA模型上的提示工程，并讨论了其局限性。
cs.CV / 118 / 2609.39548
Learning Normal Diffusion Dynamics for Backdoor Defense in Text-to-Image Models
学习正常扩散动态以防御文本到图像模型中的后门攻击
Abstract
Backdoor attacks pose a serious threat to the secure deployment of text-to-image (T2I) diffusion models. Existing defenses typically detect backdoors from specific abnormal patterns in internal representations, which may limit their generalizability with the emergence of increasingly diverse attack mechanisms. In this paper, we study backdoor defense of T2I diffusion models from a transition-dynamics perspective. We observe that benign diffusion trajectories exhibit structured and timestep-dependent transition patterns from cross-attention, latent and noise spaces, whereas backdoor attacks tend to induce deviations from such normal evolution. Motivated by these observations, we propose Normal Diffusion Dynamics Learning (NDDL), a novel backdoor defense framework that learns the normal transition dynamics of diffusion trajectories utilizing only benign samples. NDDL constructs compact multi-space trajectory representations and trains a timestep-conditioned dynamics model to predict the diffusion evolution. In the inference phase, deviations between the observed and predicted transitions are exploited to quantify dynamics inconsistency for backdoor detection. NDDL further enables trigger localization without any prior knowledge of the embedded backdoor by performing substitution with low-semantic words. Extensive experiments for diverse backdoor attacks demonstrate the effectiveness and generalizability of our proposed NDDL.
Chinese Translation
后门攻击对文本到图像（T2I）扩散模型的安全部署构成了严重威胁。现有的防御方法通常通过检测内部表征中的特定异常模式来识别后门，这可能限制了它们在日益多样化的攻击机制出现时的普适性。本文从转移动态的角度研究T2I扩散模型的后门防御。我们观察到，良性扩散轨迹在交叉注意力、潜在空间和噪声空间中表现出结构化且与时间步相关的转移模式，而后门攻击则倾向于导致与这种正常演变的偏离。基于这些观察，我们提出了正常扩散动态学习（Normal Diffusion Dynamics Learning, NDDL），这是一种新颖的后门防御框架，仅利用良性样本学习扩散轨迹的正常转移动态。NDDL构建了紧凑的多空间轨迹表征，并训练了一个时间步条件的动态模型来预测扩散演变。在推理阶段，利用观察到的转移与预测转移之间的偏差来量化动态不一致性以进行后门检测。NDDL进一步通过用低语义词进行替换，实现在没有任何嵌入后门的先验知识的情况下进行触发器定位。针对多种后门攻击的广泛实验表明，我们提出的NDDL的有效性和普适性。
cs.CV / 119 / 2609.39553
EffGS: Efficient and High-Fidelity Gaussian Splatting
EffGS：高效且高保真的高斯点云渲染
Abstract
3D Gaussian Splatting (3DGS) enables real-time novel view synthesis, but existing general-purpose acceleration methods suffer severe rendering quality degradation when extended to more complex, large-scale scenes. To address this issue, we propose EffGS, a more general acceleration framework that improves training and rendering efficiency while maintaining reconstruction quality comparable to or better than vanilla 3DGS across bounded and large-scale scenes. EffGS combines frequency-aware guidance, localized density control, and adaptive primitive scale modulation. First, an importance scoring mechanism combines pixel-wise reconstruction errors with a difference-of-Gaussians mask scheduled over training to provide stage-dependent spatial guidance. Second, localized densification and pruning restricts density modifications to Gaussians with valid projected footprints in the sampled views. Third, learnable per-Gaussian scale modulation adjusts effective primitive extent during optimization while retaining the Compact Box rasterization rule. Extensive experiments on bounded and large-scale scene datasets demonstrate a favorable balance between reconstruction quality, training time, and primitive count. Component ablations and matched-primitive-budget comparisons further support the effectiveness of the framework.
Chinese Translation
三维高斯点云渲染（3D Gaussian Splatting, 3DGS）能够实现实时新视图合成，但现有的通用加速方法在扩展到更复杂的大规模场景时，渲染质量严重下降。为了解决这一问题，我们提出了EffGS，一个更通用的加速框架，旨在提高训练和渲染效率，同时在有限和大规模场景中保持与传统3DGS相当或更好的重建质量。EffGS结合了频率感知引导、局部密度控制和自适应基元尺度调节。首先，重要性评分机制将逐像素重建误差与在训练过程中调度的高斯差分掩模结合，以提供阶段依赖的空间引导。其次，局部密度增强和修剪将密度修改限制在在采样视图中具有有效投影足迹的高斯上。第三，学习型每个高斯尺度调节在优化过程中调整有效基元范围，同时保留紧凑盒光栅化规则。在有限和大规模场景数据集上的大量实验表明，重建质量、训练时间和基元数量之间达成了良好的平衡。组件消融实验和匹配基元预算的比较进一步支持了该框架的有效性。
cs.CV / 120 / 2609.39563
RESUME: Recurrent State Updates from Motion and Residual Signals for Efficient Video Language Modeling
RESUME：基于运动和残差信号的递归状态更新用于高效视频语言建模
Abstract
Existing video language models encode sampled RGB frames independently, so a long video must either exhaust the token budget or drop the changes between sampled frames. Codec-aware front-ends read the motion vectors and residuals that encoding produced, but in their deployed form each predictive frame is still tokenized on its own: the tokens are a function of the current primitives, not of a carried reference. We argue that a more natural function is of both---the current primitives and a carried reference. A clip and its time reversal share the same frames and differ only in the order of changes---an axis that symmetric pooling discards by construction, and that is non-empty in the frozen vision features VideoLMs use---and the codec recurrence already composes those changes in order against a reference state. We introduce RESUME, a stateful codec representation: an anchor I-frame initializes a compact latent state, each subsequent predictive frame is consumed as an update to that state, and a shared readout exposes VideoLM-compatible tokens from the accumulated state. Codec prediction is thereby kept at the representation level and handed to the language model as a trajectory, not as a set of independent token groups. At the same per-predictive-frame token budget as prior codec-aware methods, a predictive frame enters the language model as a readout of what the front-end already knows, not as an encoding of the current primitives alone. Across ten benchmarks, the gains concentrate on temporal reasoning: on all three temporal benchmarks RESUME improves over both the RGB-frame baseline LLaVA-Video-7B (by 2.8, 5.1, and 3.9 points on TempCompass, TOMATO, and MVBench) and the codec-based baseline CoPE-7B, while staying competitive on general and long-form QA. Frozen-transition tests further show anchor dependence, order sensitivity, and useful rollout behavior beyond the training horizon.
Chinese Translation
现有的视频语言模型独立编码采样的RGB帧，因此较长的视频必须耗尽令牌预算或忽略采样帧之间的变化。编解码器感知前端读取编码产生的运动矢量和残差，但在其部署形式中，每个预测帧仍然是独立进行令牌化的：这些令牌是当前原始数据的函数，而不是携带的参考的函数。我们认为，更自然的函数是同时考虑当前原始数据和携带的参考。一个剪辑及其时间反转共享相同的帧，仅在变化的顺序上有所不同——这一轴在对称池化中被构造性地丢弃，而在视频语言模型（VideoLMs）使用的冻结视觉特征中是非空的——而编解码器的递归已经按照参考状态顺序组合了这些变化。我们引入了RESUME，一种有状态的编解码器表示：一个锚定的I帧初始化一个紧凑的潜在状态，每个后续的预测帧作为对该状态的更新被消耗，且共享的读出从累积状态中暴露出与VideoLM兼容的令牌。因此，编解码器预测保持在表示层面，并作为轨迹交给语言模型，而不是作为一组独立的令牌组。在与先前的编解码器感知方法相同的每预测帧令牌预算下，预测帧作为前端已知内容的读出进入语言模型，而不仅仅是当前原始数据的编码。在十个基准测试中，收益集中在时间推理上：在所有三个时间基准测试中，RESUME在RGB帧基线LLaVA-Video-7B（在TempCompass、TOMATO和MVBench上分别提高2.8、5.1和3.9点）和基于编解码器的基线CoPE-7B上均有所改善，同时在一般和长篇问答方面保持竞争力。冻结过渡测试进一步显示了锚定依赖性、顺序敏感性以及超出训练视野的有用展开行为。
cs.CV / 121 / 2609.39566
From Given to Gathered Evidence: Agentic Learning for Longitudinal Medical Reasoning
从给定证据到收集证据：用于纵向医学推理的自主学习
Abstract
Foundation models can serve as clinical agents through tool-use harnesses. However, conventional medical benchmarks assess reasoning over preselected evidence rather than the ability to seek it across clinical records and longitudinal imaging. We propose CASE: a series of role-specific Clinical Agents for Seeking Evidence, together with a tool-use harness and an agentic post-training framework for compact vision-language policy models. We further introduce a longitudinal multimodal benchmark built on UK Biobank, comprising 50,401 clinical questions derived from real-world ICD-10-coded diagnoses of 4,739 participants. Each question links to a patient-specific environment containing clinical context and multi-sequence MRI from baseline and follow-up visits, where agents autonomously select which visits, organs, modalities, slices, and specialist tools to inspect and compare. Supervised fine-tuning transfers evidence-seeking workflows from 14,734 frontier-model interaction trajectories, followed by agentic reinforcement learning on the learner's own environment interactions. Privileged on-policy self-distillation and rubric-based LLM feedback refine evidence-to-conclusion reasoning without prescribing tool sequences. Experiments show that CASE moves beyond question-answer imitation toward transferable investigation policies, strengthening evidence-grounded longitudinal reasoning. Under matched evaluation conditions, our Qwen3-VL-8B based agent achieves over 16% and 10% relative improvements in answer accuracy over GPT-5.4 and Claude Opus 4.8. Code will be available at https://github.com/VinyehShaw/CASE.
Chinese Translation
基础模型可以通过工具使用框架作为临床代理。然而，传统的医学基准评估的是对预选证据的推理能力，而非在临床记录和纵向成像中寻求证据的能力。我们提出了CASE：一系列特定角色的临床代理，用于寻求证据，配合工具使用框架和紧凑的视觉-语言策略模型的自主后训练框架。我们进一步介绍了基于英国生物库构建的纵向多模态基准，包含50,401个临床问题，这些问题源自4,739名参与者的真实世界ICD-10编码诊断。每个问题都与包含临床背景和基线及随访访问的多序列MRI的患者特定环境相关，代理可以自主选择检查和比较哪些访问、器官、模态、切片和专业工具。监督微调将来自14,734个前沿模型交互轨迹的证据寻求工作流转移到代理的环境交互中，随后在学习者自身环境交互中进行自主强化学习。特权的在线自蒸馏和基于评分标准的大语言模型反馈在不规定工具序列的情况下，优化了证据到结论的推理。实验表明，CASE超越了问答模仿，朝着可转移的调查策略发展，增强了基于证据的纵向推理。在匹配的评估条件下，我们基于Qwen3-VL-8B的代理在答案准确性上相较于GPT-5.4和Claude Opus 4.8分别提高了超过16%和10%的相对改进。代码将可在https://github.com/VinyehShaw/CASE获取。
cs.CV / 122 / 2609.39567
Invariant Shape Analysis of Surfaces with Spherical Topology
具有球面拓扑的表面不变形状分析
Abstract
Spherical harmonic descriptors of closed 3D shapes depend on the parameterization, the pose and the scale of the surface, and the standard rotation-invariant reductions, the power spectrum and the bispectrum, discard the relative orientation of the harmonic bands and cannot distinguish a shape from its mirror image. We construct a descriptor that removes all three dependencies exactly and loses nothing else: a conformal parameterization normalized by its conformal barycenter, followed by polynomial invariants of the rotation group. Identifying each harmonic band with a binary form turns the rotation quotient into classical invariant theory and makes reflections visible as the sign of an invariant, so chirality is recorded. The descriptor is complete for the truncated expansion, stable in the orbit distance, and comes with numerical diagnostics. Benchmarks confirm the guarantees, and on bilateral anatomical structures the descriptor separates mirror-image pairs from asymmetric pairs, which parity-blind descriptors cannot.
Chinese Translation
封闭三维形状的球面谐波描述符依赖于表面的参数化、姿态和尺度，而标准的旋转不变性简化方法（如功率谱和双谱）则忽略了谐波带的相对方向，无法区分形状及其镜像。我们构建了一种描述符，准确去除了这三种依赖性，并且没有其他损失：通过其共形重心归一化的共形参数化，随后是旋转群的多项式不变式。将每个谐波带与二元形式对应，使得旋转商转化为经典不变性理论，并使得反射作为不变式的符号显现，从而记录了手性。该描述符对于截断展开是完整的，在轨道距离上是稳定的，并附带数值诊断。基准测试确认了这些保证，并且在双侧解剖结构上，该描述符能够将镜像对与不对称对分开，而这种分离是对称盲描述符无法实现的。
cs.CV / 123 / 2609.39573
Steering Fields: Adaptive Vector Fields for Safe Image Generation and Beyond
引导场：用于安全图像生成及其他应用的自适应向量场
Abstract
As state-of-the-art text-to-image flow models achieve near-photorealistic quality, controlling their outputs, e.g., suppressing harmful content while promoting benign alternatives, has become a central challenge. The current steering paradigm consists of adding a global steering vector to selected activations. While functional, a fixed and example-agnostic vector applied uniformly along the entire trajectory cannot adapt to the changing state of the generation and often causes unintended global changes. We introduce Steering Fields, a generalization of steering vectors that adaptively re-estimates the steering direction at each step of the generative process. Steering Fields operate on the noisy states of flow models, expose a continuous trade-off between steering strength and content preservation, and are compositional, enabling the simultaneous induction and inhibition of concepts, setting a new state of the art on safety steering benchmarks. Despite using no explicit spatial masks or object priors, the trajectory-adaptive estimation naturally preserves local structure, in a manner reminiscent of image editing. In fact, Steering Fields can serve as a structure-preserving image-editing technique that achieves state-of-the-art semantic fidelity (CLIP, VQAScore), while remaining model-agnostic and inversion-free.
Chinese Translation
随着最先进的文本到图像流模型达到近乎照片真实的质量，控制其输出，例如抑制有害内容并促进良性替代品，已成为一个核心挑战。目前的引导范式是将一个全局引导向量添加到选定的激活值中。尽管这种方法有效，但固定且与示例无关的向量在整个生成过程中均匀应用，无法适应生成状态的变化，往往导致意想不到的全局变化。我们提出了引导场（Steering Fields），这是对引导向量的推广，能够在生成过程的每一步自适应地重新估计引导方向。引导场在流模型的噪声状态上运行，揭示了引导强度与内容保留之间的连续权衡，并且具有组合性，能够同时引导和抑制概念，从而在安全引导基准测试中设定了新的最先进水平。尽管没有使用显式的空间掩码或对象先验，但轨迹自适应估计自然保留了局部结构，类似于图像编辑。实际上，引导场可以作为一种结构保留的图像编辑技术，达到最先进的语义保真度（CLIP, VQAScore），同时保持模型无关性和无反演性。
cs.CV / 124 / 2609.39585
A Generalizable and Explainable Framework for Synthetic Video Detection Using First-Digit Gradient Statistics
基于首位数字梯度统计的合成视频检测通用可解释框架
Abstract
AI video generators have not only become harder to detect but are used to generate a diverse set of scenarios from landscapes to street views to animal videos. This creates a problem where CNN-based detectors are effective but offer no insight into their inner workings, while forensics-based detectors are often pretrained for a set scenario or become too complex to derive meaningful insights. We present a novel approach to AI video detection using Sobel gradient values analysed with the first-digit law. Using linear discriminant analysis, we visualise the discriminatory signal, while a multi-layer perceptron is used for classification. The detection method has no generator- or scenespecific features, and the model has no knowledge of container formats, codec, bitrate, or compression artefacts. The model is trained and tested on GenBuster-200K, GenBusterBench, GenVA, FaceForensics++ C23, and CelebDF. We also show how zero-shot detection fails even though the feature set carries a discriminatory signal.
Chinese Translation
人工智能视频生成器不仅变得更难以检测，还被用于生成从风景到街景再到动物视频的多样化场景。这导致了一个问题，即基于卷积神经网络（CNN）的检测器虽然有效，但对其内部工作原理没有提供任何见解，而基于取证的检测器往往是针对特定场景进行预训练，或者变得过于复杂以至于无法提取有意义的见解。我们提出了一种新的人工智能视频检测方法，使用经过首位数字法分析的Sobel梯度值。通过线性判别分析，我们可视化了判别信号，同时使用多层感知器进行分类。该检测方法没有生成器或场景特定特征，模型对容器格式、编解码器、比特率或压缩伪影没有任何知识。模型在GenBuster-200K、GenBusterBench、GenVA、FaceForensics++ C23和CelebDF上进行训练和测试。我们还展示了尽管特征集携带判别信号，零样本检测仍然失败。
cs.CV / 125 / 2609.39588
KilometerVision: A New Frontier for Large-Scale Spatial Intelligence in VLMs
KilometerVision：大规模空间智能在视觉-语言模型中的新前沿
Abstract
We push the frontier of large-scale spatial intelligence in Vision-Language Models (VLMs) and introduce the first benchmark that probes geographical layout understanding from real-world videos, spanning up to 1km distances. Inspired by the cognitive science literature, we evaluate models against the hierarchical stages of human spatial awareness: anchoring via landmarks, connecting them through routes, and integrating these into global mental maps. Extensive experiments reveal a fundamental divergence in how current AI models process spatial information. Instead of utilising true path integration or forming geometric survey knowledge, we find that VLMs rely almost entirely on 2D visual recognition and text-matching to bypass complex spatial reasoning. The benchmark is publicly available at https://perception-test-challenge.github.io/kilometervision.html.
Chinese Translation
我们推动了视觉-语言模型（VLMs）中大规模空间智能的前沿，并引入了第一个基准，探讨来自真实世界视频的地理布局理解，覆盖距离可达1公里。受认知科学文献的启发，我们根据人类空间意识的层级阶段对模型进行评估：通过地标进行锚定，通过路线连接这些地标，并将其整合到全球心理地图中。大量实验揭示了当前人工智能模型处理空间信息的基本分歧。我们发现，VLMs几乎完全依赖于二维视觉识别和文本匹配来绕过复杂的空间推理，而不是利用真实的路径整合或形成几何测量知识。该基准公开可用，网址为 https://perception-test-challenge.github.io/kilometervision.html。
cs.CV / 126 / 2609.39590
SPOON: Towards Coherent Compositional 3D Scene Generation from Uncalibrated Multi-view Images
SPOON：从未校准的多视图图像生成连贯的组合3D场景
Abstract
Compositional 3D scene generation aims to recover complete 3D object shapes and their spatial arrangement from visual observations. Recent image-conditioned 3D generators provide strong priors for producing high-quality object geometry, making the generation of complex scenes increasingly practical. A central challenge is therefore to spatially organize these generated assets into a globally coherent scene while remaining consistent with multi-view observations. Existing approaches either entangle scene layout with object generation or separately estimate spatial placement from view-specific observations, where pose hypotheses may remain ambiguous and inconsistent across views, often resulting in an incoherent object-camera soup. We introduce SPOON, a framework that reformulates multi-view compositional 3D generation as scene-level, geometry-grounded pose reasoning. Rather than treating view-specific object pose hypotheses independently, SPOON coordinates them using reconstruction-derived multi-view geometry through a Guide-Route-Reconcile paradigm. This progressively organizes object poses and camera configurations into a coherent scene-level spatial arrangement. Extensive experiments on ARSG-110K and MIDI-3D-Front demonstrate consistent improvements in object placement and scene composition across varying numbers of input views. On ARSG-110K, SPOON reduces scene-level and object-level Chamfer distances by 12.7% and 17.7%, respectively, compared with a strong baseline.
Chinese Translation
组合3D场景生成旨在从视觉观察中恢复完整的3D物体形状及其空间排列。最近的图像条件3D生成器为生成高质量的物体几何形状提供了强有力的先验，使得生成复杂场景变得愈加可行。因此，一个核心挑战是如何将这些生成的资产在空间上组织成一个全局连贯的场景，同时保持与多视图观察的一致性。现有的方法要么将场景布局与物体生成纠缠在一起，要么从特定视图的观察中单独估计空间位置，其中姿态假设在不同视图之间可能存在模糊和不一致，常常导致物体与相机之间的混乱。我们提出了SPOON，一个将多视图组合3D生成重新表述为场景级、几何基础的姿态推理的框架。SPOON并不将特定视图的物体姿态假设独立对待，而是通过重建导出的多视图几何使用引导-路线-调和（Guide-Route-Reconcile）范式进行协调。这逐步将物体姿态和相机配置组织成一个连贯的场景级空间排列。在ARSG-110K和MIDI-3D-Front上的大量实验表明，在不同数量的输入视图下，物体位置和场景组合的一致性都有所改善。在ARSG-110K上，与强基线相比，SPOON分别减少了场景级和物体级的Chamfer距离12.7%和17.7%。
cs.CV / 127 / 2609.39591
Structural Limits of the Information-Theoretic Uncertainty Decomposition
信息论不确定性分解的结构限制
Abstract
Uncertainty estimation in machine learning typically decomposes uncertainty into aleatoric uncertainty (AU) and epistemic uncertainty (EU) using the standard information-theoretic framework. However, in practice, two critical issues arise: entanglement (AU and EU are highly correlated) and epistemic collapse (EU magnitude shrinks with increasing model capacity). We analyze this framework on a functional level and discover that significant portions of the assumed AU, EU range are infeasible in finite settings, and cannot be attained with any class probabilities. We characterize how this infeasible region scales with the number of classes and Monte Carlo samples $N$ (e.g., from ensembles with $N$ members), revealing it is bounded by $\text{AU} \leq \log(2)/N$. Crucially, the infeasible region's boundary helps explain epistemic collapse: when model confidence is high, $\text{AU} > \text{EU}$ is guaranteed by this fundamental structural limitation. Our findings show that increasing ensemble size mitigates epistemic collapse by reducing the infeasible area. Lastly, we caution against interpreting AU and EU as independent quantities in low AU regimes, since we show they are coupled when $\text{AU} \leq \log(2)/N$.
Chinese Translation
机器学习中的不确定性估计通常使用标准信息论框架将不确定性分解为偶然不确定性（Aleatoric Uncertainty, AU）和认知不确定性（Epistemic Uncertainty, EU）。然而，在实际应用中，出现了两个关键问题：纠缠（AU和EU高度相关）和认知崩溃（随着模型能力的增加，EU的幅度缩小）。我们在功能层面分析了该框架，发现假定的AU和EU范围中的重要部分在有限设置中是不可行的，且无法通过任何类别概率实现。我们描述了这一不可行区域如何随着类别数量和蒙特卡洛样本数$N$（例如，来自$N$个成员的集成）进行缩放，揭示其边界由$ ext{AU} ext{ } ext{≤} ext{ } rac{ ext{log}(2)}{N}$所限制。关键的是，这一不可行区域的边界有助于解释认知崩溃：当模型置信度高时，$ ext{AU} > ext{EU}$由这一基本结构限制所保证。我们的研究结果表明，增加集成规模可以通过减少不可行区域来缓解认知崩溃。最后，我们警告在低AU区域将AU和EU视为独立量的解释，因为我们表明当$ ext{AU} ext{ } ext{≤} ext{ } rac{ ext{log}(2)}{N}$时，它们是耦合的。
cs.CV / 128 / 2609.39600
GroundAnything: Reconciling Parallel Decoding with Precise Visual Grounding at Flash Speed
GroundAnything：以闪电速度调和并行解码与精确视觉定位
Yu, Qize, Fan, Lianrui, Ping, Bowen, Ding, Xini, Song, Zetian, Niu, Junbo, Wang, Kaixuan, Chen, Tianxing, Chen, Yue, He, Minghua, Wang, Yuran, Huang, Jie, Zhang, Haojun, Chen, Min, Li, Hao, Song, Wenxuan, Wu, Ruihai, Liu, Xianming, Liu, Shilong, Zhou, Shuchang, Luo, Ping, Huang, Shiyu
Abstract
Autoregressive (AR) grounding models serialize spatial predictions, introducing sequential latency and imposing a causal order on output tokens. We view grounding as visual evidence extraction: objects, locations, and spatial relations are jointly constrained by the image and query, yet their dependencies do not imply an intrinsic left-to-right generation order. This distinction makes bidirectional diffusion a natural fit, allowing spatial hypotheses to emerge in parallel and be jointly refined through iterative denoising. We introduce GroundAnything, a 4B-parameter grounding foundation model that reconciles fast parallel decoding with precise localization through blockwise denoising. Training combines grounding pretraining from public datasets and dedicated data engines, direct AR-to-diffusion conversion with joint AR and diffusion objectives, supervised fine-tuning, and GRPO-based reinforcement post-training. Across 30 grounding benchmarks, our autoregressive variant, GroundAnything-VLM, establishes a new overall state of the art among similarly sized models at 72.42%, remaining competitive with GPT-6 Astra (71.35%). With entropy-guided decoding, GroundAnything also surpasses the prior state of the art at this scale, averaging 61.75% versus 53.32% for the fast MTP-based LocateAnything model. We further explore decoding strategies, showing that an optional self-speculative mode achieves a $4.51\times$ speedup over the AR counterpart with a 0.74 percentage-point drop in COCO F1mIoU. Infrastructure experiments show that progressive inference optimizations translate parallel decoding into practical speedups. These support efficient visual grounding in latency-sensitive real-world systems.
Chinese Translation
自回归（AR）定位模型将空间预测序列化，导致顺序延迟并对输出标记施加因果顺序。我们将定位视为视觉证据提取：物体、位置和空间关系共同受到图像和查询的约束，但它们的依赖关系并不意味着内在的从左到右的生成顺序。这一区别使得双向扩散成为一种自然的选择，允许空间假设并行出现并通过迭代去噪共同优化。我们提出了GroundAnything，一个具有40亿参数的定位基础模型，通过块状去噪调和快速并行解码与精确定位。训练结合了来自公共数据集的定位预训练和专用数据引擎、直接的AR到扩散转换以及联合AR和扩散目标的训练、监督微调和基于GRPO的强化后训练。在30个定位基准测试中，我们的自回归变体GroundAnything-VLM在同类模型中以72.42%的成绩建立了新的整体最佳表现，仍然与GPT-6 Astra（71.35%）保持竞争力。通过熵引导解码，GroundAnything还超越了这一规模的先前最佳表现，平均达到61.75%，而快速的基于MTP的LocateAnything模型为53.32%。我们进一步探索了解码策略，显示可选的自我推测模式在COCO F1mIoU上实现了$4.51 imes$的速度提升，尽管下降了0.74个百分点。基础设施实验表明，渐进推理优化将并行解码转化为实际的速度提升。这些支持在对延迟敏感的现实世界系统中实现高效的视觉定位。
cs.CV / 129 / 2609.39601
GroundingPI: A Grounding Foundation Model towards Physical Intelligence with Visual Primitives
GroundingPI：面向物理智能的基础模型与视觉原语的结合
Yu, Qize, Fan, Lianrui, Chen, Boyu, Liang, Jiaqi, Ding, Xini, Chen, Yue, Song, Zetian, Wang, Yuran, Zou, Yi, Wang, Kaixuan, Chen, Tianxing, Song, Wenxuan, Zhou, Bohan, Li, Mingleyang, Huang, Siqiao, Ye, Yuqi, Jiang, Caigao, Wei, Wei, Wu, Ruihai, Zhang, Hang, Ge, Yixiao, Zhou, Shuchang, Liu, Shilong, Liu, Xianming, Luo, Ping, Huang, Shiyu
Abstract
Precise grounding matters. It specifies which object is the target and where that object is, even in clutter and for tiny objects, and it has to be fast enough for closed-loop control. Yet vision-language-action (VLA) and world-action models (WAMs) take perception from general-purpose vision-language and video-generation backbones, which still fail in these settings. We introduce GroundingPI, a 4B grounding foundation model that generates points and boxes as quantized coordinates in a shared vocabulary. Training combines multimodal and spatial pretraining, supervised fine-tuning, and reinforcement learning with GRPO, using supervision from public datasets and dedicated data engines. Against 44 baselines across 34 grounding benchmarks spanning 11 perceptual capabilities, GroundingPI establishes a new state of the art, averaging 73.68%, above the larger GPT-6 Astra (71.54%). As a downstream visual backbone, GroundingPI improves performance on robotic manipulation and autonomous driving. On RoboTwin 2.0, it outperforms every mainstream backbone we evaluate in all four out-of-distribution settings, by up to 24.8% relative to the strongest backbone. On RoboCasa-GR1, GroundingPI trained with 50% of the demonstrations outperforms those baselines trained with 75%. On nuScenes, used as the visual backbone, GroundingPI attains an average open-loop L2 error of 0.296 m. We systematically analyze GroundingPI's pretraining in scale and data composition. Downstream autonomous driving and robotic manipulation improve as the pretraining is scaled. Analyzing the data recipe across these 11 perceptual capabilities shows dense grounding's substantial benefits for both, and OCR's potential as a catalyst for perceptual learning. These results support grounding as a perceptual foundation, and dedicated perceptual pretraining as a promising direction for foundation models of physical intelligence.
Chinese Translation
精确的定位至关重要。它明确了哪个对象是目标以及该对象的位置，即使在杂乱环境和小型物体中，也必须足够快速以支持闭环控制。然而，视觉-语言-动作（VLA）和世界-动作模型（WAMs）依赖于通用的视觉-语言和视频生成骨干网络，这在这些场景中仍然表现不佳。我们提出了GroundingPI，一个具有40亿参数的基础模型，能够生成点和框作为共享词汇中的量化坐标。训练结合了多模态和空间预训练、监督微调以及使用GRPO的强化学习，利用公共数据集和专用数据引擎提供的监督。在跨越11种感知能力的34个定位基准测试中，GroundingPI在44个基线模型中建立了新的最先进水平，平均达到73.68%，超过了更大的GPT-6 Astra（71.54%）。作为下游视觉骨干，GroundingPI提升了机器人操作和自动驾驶的性能。在RoboTwin 2.0上，它在所有四种分布外设置中超越了我们评估的每一个主流骨干，最高相对提升达24.8%。在RoboCasa-GR1上，使用50%演示训练的GroundingPI超越了那些使用75%演示训练的基线。在作为视觉骨干的nuScenes上，GroundingPI达到了平均开放循环L2误差为0.296米。我们系统地分析了GroundingPI在规模和数据组成上的预训练。随着预训练规模的扩大，下游的自动驾驶和机器人操作性能得以提升。对这11种感知能力的数据配方进行分析显示，密集定位对两者都有显著的好处，而OCR作为感知学习的催化剂的潜力也得以体现。这些结果支持定位作为感知基础，专门的感知预训练作为物理智能基础模型的一个有前景的方向。
cs.CV / 130 / 2609.39605
FOMO: Forget the Concept, Don't Miss Out on the Scene in Selective Video Unlearning
FOMO：忘记概念，不要错过选择性视频遗忘中的场景
Abstract
The rapid advancement of generative video models has enabled the synthesis of increasingly realistic and temporally coherent videos, while also raising concerns about the generation of harmful content. The reliance on large-scale web datasets during training inevitably exposes these models to undesirable material, making concept unlearning an essential mitigation. Existing methods mainly target static visual concepts, such as objects, identities, or unsafe appearance, largely overlooking motion unlearning. Furthermore, these approaches often pay little attention to preserving the surrounding scene. As a result, successful concept removal may unintentionally alter the background, composition, or overall video dynamics. We argue that effective unlearning should ideally change only what is targeted, while minimizing unnecessary changes to the remaining scene. In this work, we introduce FOMO, to the best of our knowledge the first training-based selective video unlearning method that directly treats preservation of the original scene as a priority. We formulate unlearning around two complementary objectives: what to change and what to preserve. Our method localizes concept-related representations and modifies them, while the preservation mechanism maintains non-target scene information without requiring auxiliary data. Beyond simply erasing unwanted concepts, FOMO explicitly redirects the generation toward a specified safe alternative. We further extend this formulation to motion unlearning, where the concept is defined by temporal behavior rather than a fixed spatial region. Our solution achieves effective unlearning across unsafe content, object, and motion concepts, while achieving the best trade-off between concept removal and scene preservation. Code: https://github.com/gmum/FOMO Project Page https://gmum.github.io/FOMO
Chinese Translation
生成视频模型的快速发展使得合成越来越真实且时间上连贯的视频成为可能，同时也引发了对有害内容生成的担忧。在训练过程中对大规模网络数据集的依赖不可避免地使这些模型接触到不良材料，因此概念遗忘成为一种重要的缓解措施。现有方法主要针对静态视觉概念，如物体、身份或不安全的外观，往往忽视了运动遗忘。此外，这些方法通常对保留周围场景的关注较少。因此，成功的概念移除可能无意中改变背景、构图或整体视频动态。我们认为，有效的遗忘应该理想地只改变目标内容，同时最小化对剩余场景的不必要更改。在本研究中，我们介绍了FOMO，尽我们所知，这是第一种将原始场景保留作为优先事项的基于训练的选择性视频遗忘方法。我们围绕两个互补目标构建遗忘：改变什么和保留什么。我们的方法定位与概念相关的表征并对其进行修改，而保留机制则在不需要辅助数据的情况下保持非目标场景信息。FOMO不仅仅是简单地抹去不需要的概念，还明确地将生成重定向到指定的安全替代品。我们进一步将这一公式扩展到运动遗忘，其中概念由时间行为定义，而不是固定的空间区域。我们的解决方案在不安全内容、物体和运动概念的遗忘方面实现了有效性，同时在概念移除和场景保留之间达成了最佳平衡。代码：https://github.com/gmum/FOMO 项目页面：https://gmum.github.io/FOMO
cs.CV / 131 / 2609.39623
Semantic Watermarking for Malicious Image Manipulation Detection
用于恶意图像操控检测的语义水印技术
Abstract
The proliferation of high-fidelity generative editing models has made it possible to inject violent or sexual content into otherwise ordinary images while preserving visual plausibility, with concrete consequences for public discourse and vulnerable populations. We propose a robust semantic watermarking framework that reframes the watermark as a recoverable semantic reference rather than an opaque identifier. Our framework combines a $\beta$-VAE-based binary watermark (CLIP-VAE) with explicit channel-aware training---random bit-flip noise is injected during training so that the decoder learns graceful degradation under the noisy watermarking channel. As a downstream application, a lightweight module SDA-Net uses the recovered semantic embedding to expose not only whether but in which semantic direction an image has been altered. In a 5-way comparison against representative binary hashing baselines (SimHash, ITQ, HashNet, and their robust-MLP variants), CLIP-VAE achieves the highest reconstruction cosine similarity to the original CLIP embedding under realistic InstructPix2Pix bit-error rates, and uniquely supports direction-of-drift detection---a forensic complement to existing content-moderation pipelines.
Chinese Translation
高保真生成编辑模型的普及使得在保持视觉可信度的同时，将暴力或色情内容注入普通图像成为可能，这对公共话语和脆弱群体产生了具体影响。我们提出了一种稳健的语义水印框架，将水印重新定义为可恢复的语义参考，而不是不透明的标识符。我们的框架结合了基于$eta$-VAE的二进制水印（CLIP-VAE）和显式通道感知训练——在训练过程中注入随机位翻转噪声，使解码器学习在嘈杂水印通道下的优雅降级。作为下游应用，一个轻量级模块SDA-Net利用恢复的语义嵌入，不仅揭示图像是否被修改，还能指出其修改的语义方向。在与代表性的二进制哈希基线（SimHash、ITQ、HashNet及其稳健的MLP变体）进行的五路比较中，CLIP-VAE在现实的InstructPix2Pix比特错误率下，实现了与原始CLIP嵌入的最高重建余弦相似度，并独特地支持漂移方向检测——为现有内容审核流程提供了法医补充。
cs.CV / 132 / 2609.39624
ExpandDiff: Dynamic Range Expanding Diffusion for Single-Image HDR Reconstruction
ExpandDiff：用于单幅图像HDR重建的动态范围扩展扩散
Abstract
Single-image HDR reconstruction requires inferring missing detail while preserving the visible content of an LDR image. Differences in sensor dynamic range and exposure cause LDR images to lose varying amounts of information in shadows and highlights. We present ExpandDiff, a conditional diffusion pipeline that jointly reconstructs clipped shadows and highlights. To account for this variation, we introduce Dynamic Clipping Synthesis (DCS), which randomly samples shadow and highlight clipping percentiles when constructing training inputs from HDR targets. A pixel-space diffusion model guided by spatially-adaptive normalization then predicts perceptually encoded HDR through a bounded output head, reconstructing both clipping directions in one sampling trajectory. On the SI-HDR benchmark, ExpandDiff variants improve HDR reconstruction accuracy by 3.43 dB in PU21-PSNR over the strongest evaluated competing method, and by 7.34 dB under two-sided clipping. The code and supplementary material are available at https://memreandiran.github.io/expanddiff/.
Chinese Translation
单幅图像HDR重建需要在保留LDR图像可见内容的同时推断缺失的细节。传感器动态范围和曝光的差异导致LDR图像在阴影和高光部分丧失不同量的信息。我们提出了ExpandDiff，这是一种条件扩散管道，能够联合重建被裁剪的阴影和高光。为了考虑这种变化，我们引入了动态裁剪合成（Dynamic Clipping Synthesis, DCS），在从HDR目标构建训练输入时随机采样阴影和高光裁剪百分位数。然后，一个由空间自适应归一化引导的像素空间扩散模型通过一个有界输出头预测感知编码的HDR，在一个采样轨迹中重建两个裁剪方向。在SI-HDR基准测试中，ExpandDiff变体在PU21-PSNR上相较于最强评估的竞争方法提高了3.43 dB的HDR重建精度，并在双侧裁剪下提高了7.34 dB。代码和补充材料可在 https://memreandiran.github.io/expanddiff/ 获取。
cs.CV / 133 / 2609.39625
D-Scope: Decomposing and Steering Diffusion Transformers with Sparse Autoencoders
D-Scope：利用稀疏自编码器分解和引导扩散变换器
Abstract
Sparse autoencoders (SAEs) reveal visual structure in diffusion transformers (DiTs), but interpreting a feature does not establish whether it can be used to control generation. We introduce D-Scope (Diffusion Scope), a framework that connects feature interpretation to generation control through shared visual evidence. D-Scope aggregates SigLIP~2 embeddings of highly activating image patches into visual centroids. Matching target text descriptions against these visual centroids in the shared image-text embedding space then enables retrieval of individual features without per-feature text annotations. The underlying patches provide evidence for inspecting each selection, while spatially masked interventions test the corresponding decoder direction at varying strengths under fixed generation conditions. We characterize 150 SAEs across two model families and five layers, and introduce a benchmark of 100 target concepts with ten contexts each spanning under-specified and explicit-conflict conditions. Our empirical results show that high reconstruction fidelity can coexist with low dictionary utilization and limited visual-evidence coverage. Under per-case best-of-sweep strength selection, contrastive retrieval yields larger mean regional SigLIP~2 gains than direct retrieval across the tested steering configurations, without consistently improving outside-region preservation. D-Scope provides an inspectable framework for evaluating sparse DiT features through their visual evidence and the effects of their decoder directions on generation. The demo is available at https://jiahaozhang-public.github.io/d-scope/.
Chinese Translation
稀疏自编码器（SAEs）揭示了扩散变换器（DiTs）中的视觉结构，但对特征的解释并不能确定其是否可以用于控制生成。我们提出了D-Scope（Diffusion Scope），一个通过共享视觉证据将特征解释与生成控制连接起来的框架。D-Scope将高度激活的图像补丁的SigLIP~2嵌入聚合为视觉质心。将目标文本描述与共享图像-文本嵌入空间中的这些视觉质心进行匹配，从而能够在没有逐特征文本注释的情况下检索单个特征。基础补丁为检查每个选择提供了证据，而空间掩蔽干预则在固定生成条件下以不同强度测试相应解码器方向。我们对两个模型家族和五个层次的150个SAEs进行了特征化，并引入了一个包含100个目标概念的基准，每个概念有十个上下文，涵盖了不明确指定和显式冲突条件。我们的实证结果表明，高重建保真度可以与低字典利用率和有限的视觉证据覆盖共存。在逐案例最佳强度选择下，对比检索在测试的引导配置中产生的平均区域SigLIP~2增益大于直接检索，而未能始终改善区域外的保留。D-Scope提供了一个可检查的框架，用于通过视觉证据评估稀疏DiT特征及其解码器方向对生成的影响。演示可在 https://jiahaozhang-public.github.io/d-scope/ 获取。
cs.CV / 134 / 2609.39627
Introduction to Computer Vision
计算机视觉导论
Abstract
This book presents a code-first introduction to computer vision, spanning classical 2D image processing, classical 3D vision, and deep learning. Organized as 44 short chapters across three parts, the book builds each topic from first principles: image arithmetic and morphology; convolution, pyramids, and frequency-domain filtering; feature detection, optical flow, and stereo; projective geometry, camera calibration, and structure from motion; and the full arc of modern deep learning, from a single neuron through convolutional networks, backpropagation, classic architectures, transfer learning, object detection, and semantic and instance segmentation, concluding with engineering considerations like mixed-precision and parallel training. Every technique is implemented directly in Python and NumPy or PyTorch and checked numerically against the corresponding OpenCV or PyTorch library function, so readers see not just the mathematics but its concrete behavior on real and synthetic data. The material was distilled with AI assistance from freely available online course notes, condensing extensive working code into concise mathematical exposition while preserving verified, reproducible results throughout. It is intended as a self-contained reference for students and practitioners who want to understand computer vision algorithms and their Python implementations.
Chinese Translation
本书提供了一个以代码为先的计算机视觉导论，涵盖了经典的二维图像处理、经典的三维视觉以及深度学习。全书分为三个部分，共44个短章节，从基本原理出发构建每个主题：图像算术与形态学；卷积、金字塔和频域滤波；特征检测、光流和立体视觉；投影几何、相机标定和运动重建；以及现代深度学习的完整过程，从单个神经元到卷积网络、反向传播、经典架构、迁移学习、目标检测，以及语义和实例分割，最后讨论混合精度和并行训练等工程考虑。每种技术都直接在Python和NumPy或PyTorch中实现，并与相应的OpenCV或PyTorch库函数进行数值验证，因此读者不仅能够看到数学原理，还能观察其在真实和合成数据上的具体表现。该材料经过AI辅助，从网上免费提供的课程笔记中提炼而成，将大量的工作代码浓缩为简明的数学阐述，同时在整个过程中保持经过验证的可重复结果。它旨在作为一个自包含的参考，供希望理解计算机视觉算法及其Python实现的学生和从业者使用。
cs.CV / 135 / 2609.39635
SAGE: Salient Factor Discovery and Generation with Visual Foundation Representations
SAGE：基于视觉基础表示的显著因子发现与生成
Abstract
Given a target dataset, such as faces with eyeglasses, and a background dataset, such as faces without, contrastive analysis separates \textit{salient} factors specific to the target from \textit{common} content shared by both. We aim for salient representations that capture target-specific detail in each image, such as the shape, color, and position of the glasses, so that they reveal subtypes without subtype labels and guide the generation of new examples of a discovered subtype, even one with no name or text description. We introduce SAGE, which learns both factors directly in the high-dimensional spatial latent of a frozen representation autoencoder and conditions a diffusion transformer on the learned salient representation of a reference image. On Digits-ImageNet and FFHQ eyeglasses, SAGE combines high-fidelity \textit{reconstruction} (rFID below $2$) with unsupervised \textit{subtype discovery}, recovering the digits better than baselines (probe accuracy $0.950$ vs.\ at most $0.281$) and revealing eyewear types, finer sunglasses styles, and mislabeled images; salient-conditioned \textit{generation} raises Digits-ImageNet subtype accuracy over the unfactorized latent ($90.5\%$ vs.\ $27.7\%$) and diversity on both datasets. On retinal OCT, SAGE's salient space separates three diseases using only normal/disease labels.
Chinese Translation
给定一个目标数据集，例如带眼镜的面孔，以及一个背景数据集，例如不带眼镜的面孔，对比分析将特定于目标的 extit{显著}因子与两者共享的 extit{共同}内容分开。我们的目标是获得显著表示，以捕捉每幅图像中与目标特定细节相关的信息，例如眼镜的形状、颜色和位置，从而揭示子类型而无需子类型标签，并指导生成新发现的子类型的示例，即使该子类型没有名称或文本描述。我们引入了SAGE，它直接在冻结表示自编码器的高维空间潜变量中学习这两种因子，并在参考图像的学习显著表示上对扩散变换器进行条件化。在Digits-ImageNet和FFHQ眼镜数据集上，SAGE将高保真 extit{重建}（rFID低于$2$）与无监督 extit{子类型发现}相结合，恢复数字的效果优于基线（探测准确率$0.950$对比最多$0.281$），并揭示眼镜类型、更细致的太阳镜风格以及错误标记的图像；显著条件化的 extit{生成}使Digits-ImageNet子类型准确率超过未因子化潜变量（$90.5\%$对比$27.7\\%$）并提高了两个数据集的多样性。在视网膜OCT上，SAGE的显著空间仅使用正常/疾病标签就能区分三种疾病。
cs.CV / 136 / 2609.39649
FANVIDv2: Evaluating Video Super-Resolution by Face and Licence-Plate Recognition Under Compound Degradation
FANVIDv2：在复合退化下通过人脸和车牌识别评估视频超分辨率
Abstract
Video super-resolution (VSR) is normally judged by PSNR and SSIM on clips that were downsampled bicubically, although in surveillance its purpose is to make faces and licence plates \emph{recognisable}. We present FANVIDv2, a benchmark that scores VSR by what a recognition pipeline can do with its output. FANVIDv2 provides $320\times180$ low-resolution (LR) clips with high-resolution (HR) references for 48 public figures (with one HR gallery image each) and 375 licence-plate clips covering 360 distinct plate strings. LR clips are generated with a randomised compound degradation (blur, resize jitter, sensor noise, JPEG compression, final downsampling) rather than bicubic downsampling alone. Two metrics score recognition \emph{inside} detections: FaceRecBox rewards a face only if it is localised and correctly identified, and TextRecBox scores plate transcriptions by normalised edit distance weighted by localisation quality. With a 2.3\,M-parameter VSR baseline (RCDM), FaceRecBox rises from 0.6864 to 0.7222, identity accuracy on matched faces from 84.35\% to 86.93\%, and TextRecBox from 0.3088 to 0.3667; a residual-map gated variant (RCDM-RMGF) reaches 0.3801 on plates. We describe the degradation model, the baseline architectures and the scorers in detail, and release annotations, metadata, download and degradation scripts and evaluation code.
Chinese Translation
视频超分辨率（VSR）通常通过在双三次下采样的片段上使用PSNR和SSIM进行评估，尽管在监控中其目的在于使人脸和车牌变得可识别。我们提出了FANVIDv2，一个通过识别管道对其输出进行评分的VSR基准。FANVIDv2提供了320×180的低分辨率（LR）片段及其对应的高分辨率（HR）参考图像，涵盖48位公众人物（每位公众人物一张HR图库图像）和375个车牌片段，涉及360个不同的车牌字符串。LR片段是通过随机复合退化（模糊、缩放抖动、传感器噪声、JPEG压缩、最终下采样）生成的，而不仅仅是双三次下采样。两个指标用于评分检测中的识别：FaceRecBox仅在面部被定位且正确识别时给予奖励，而TextRecBox则通过归一化编辑距离加权定位质量来评分车牌转录。在一个具有2.3M参数的VSR基线（RCDM）下，FaceRecBox的得分从0.6864上升到0.7222，匹配人脸的身份准确率从84.35%上升到86.93%，TextRecBox从0.3088上升到0.3667；一个残差图门控变体（RCDM-RMGF）在车牌上达到了0.3801。我们详细描述了退化模型、基线架构和评分器，并发布了注释、元数据、下载和退化脚本及评估代码。
cs.CV / 137 / 2609.39657
Diffusable Latents from Structure-Agnostic Distillation
无结构蒸馏的可扩散潜变量
Abstract
Distilling pretrained foundation models into an autoencoder bottleneck improves latent diffusability, enabling diffusion models to converge faster and reach higher sample quality. Standard distillation aligns the latent at each position to a co-located teacher feature, tying the latent layout to the teacher's. We show this constraint is unnecessary: aligning a single pooled image-level descriptor to the teacher's performs as well as or slightly better than dense position-wise distillation. We compare first-order and relational pooled objectives across latent shapes and teacher modalities. First-order matching extends naturally to 1D token-sequence latents and across modalities, where distilling a text encoder into an image autoencoder still improves diffusability; a relational objective based only on each image's nearest neighbours improves it as well. Code and blog post are available at https://github.com/AdrienRR/structure-agnostic-distillation and https://kyutai.org/blog/2026-09-28-structure-agnostic-distillation/.
Chinese Translation
将预训练基础模型蒸馏到自编码器瓶颈中可以提高潜变量的可扩散性，从而使扩散模型更快收敛并达到更高的样本质量。标准的蒸馏方法将每个位置的潜变量与相应的教师特征对齐，这使得潜变量的布局与教师的布局紧密相关。我们证明这种约束是多余的：将单个汇聚的图像级描述符与教师的特征对齐，其效果与密集的逐位置蒸馏相当或稍好。我们比较了不同潜变量形状和教师模态下的一阶和关系汇聚目标。一阶匹配自然扩展到一维令牌序列潜变量和跨模态，其中将文本编码器蒸馏到图像自编码器仍然提高了可扩散性；仅基于每个图像最近邻的关系目标也能改善可扩散性。代码和博客文章可在 https://github.com/AdrienRR/structure-agnostic-distillation 和 https://kyutai.org/blog/2026-09-28-structure-agnostic-distillation/ 获取。
cs.CV / 138 / 2609.39660
BAM! Bayesian Anything Model: a foundation model for generative computational imaging
BAM！贝叶斯任意模型：生成计算成像的基础模型
Abstract
Generative models are transforming Bayesian computational imaging, yet the field still lacks physics-aware foundation models. Current practice falls into two camps. Large foundation image models are deployed as plug-and-play priors with zero-shot approximate likelihood guidance, which introduces significant bias and computational cost. Physics-aware generative models avoid this bias, but each is tied to a specific dataset, task and instrument. We introduce BAM (Bayesian Anything Model), a lightweight foundation model for few-step, physics-aware posterior sampling that generalises robustly to unseen data and tasks, zero-shot or with minimal finetuning. BAM upgrades the operator-conditioned Reconstruct Anything Model (RAM) backbone (Terris et al.) into a conditional flow map, so instrument physics is specified at inference time rather than fixed during training. BAM has just 36M parameters and is pre-trained jointly on large image corpora and libraries of forward operators. A single network then draws posterior samples in a few steps, with no likelihood approximation and no guidance weights to tune. Across linear inverse problems on FFHQ, AFHQ, LSUN, DIV2K and the Kohler camera-shake benchmark, BAM outperforms in just 3 steps both specialised models and leading zero-shot methods in sample quality, at a fraction of their computational cost. BAM gives the community an accessible entry point to generative computational imaging, lowers the economic and environmental cost of training imaging models, and opens a new path for research on physics-aware Bayesian computational imaging. Official page: https://bayesian-anything-model.github.io/
Chinese Translation
生成模型正在改变贝叶斯计算成像，但该领域仍然缺乏物理感知的基础模型。目前的实践分为两派。一方面，大型基础图像模型作为即插即用的先验，采用零-shot近似似然指导，这引入了显著的偏差和计算成本。另一方面，物理感知的生成模型避免了这种偏差，但每个模型都与特定的数据集、任务和仪器相关联。我们提出了BAM（贝叶斯任意模型），这是一个轻量级的基础模型，用于少量步骤的物理感知后验采样，能够稳健地推广到未见过的数据和任务，无论是零-shot还是经过最小的微调。BAM将操作条件的重构任意模型（Reconstruct Anything Model，RAM）骨干网络（Terris等）升级为条件流映射，因此在推理时指定仪器物理，而不是在训练期间固定。BAM仅有3600万参数，并在大型图像语料库和前向算子库上进行联合预训练。然后，单个网络在几步内绘制后验样本，无需似然近似和调节指导权重。在FFHQ、AFHQ、LSUN、DIV2K和Kohler相机抖动基准的线性逆问题中，BAM在样本质量上以仅3步超越了专业模型和领先的零-shot方法，同时计算成本仅为它们的一小部分。BAM为社区提供了一个可接入的生成计算成像入口，降低了训练成像模型的经济和环境成本，并为物理感知的贝叶斯计算成像研究开辟了新的路径。官方网站：https://bayesian-anything-model.github.io/
cs.CV / 139 / 2609.39662
Typographic Attack Against VLM-based AI-generated Image Detection
针对基于视觉语言模型的人工智能生成图像检测的排版攻击
Abstract
Vision-language models (VLMs) are increasingly used for AI-generated image (AIGI) detection, providing natural-language explanations for authenticity judgments. However, their ability to interpret text within images may also expose these judgments to misleading semantic cues. We systematically evaluate typographic attack strategies across detection-oriented, open-weight, and commercial VLMs, considering both real-to-fake and fake-to-real attacks. Our results show that reasoning modes generally exhibit greater vulnerability than direct modes and that attack effectiveness exhibits pronounced directional asymmetry. Moreover, larger models tend to exhibit higher clean detection accuracy but also higher attack success rates. We further examine attack robustness under image and text transformations and investigate whether overlays indicating the correct class can aid error correction. Together, these analyses characterize how typographic attacks influence authenticity judgments and expose limitations of current VLM-based AIGI detection systems.
Chinese Translation
视觉语言模型（VLMs）在人工智能生成图像（AIGI）检测中越来越多地被使用，提供了关于真实性判断的自然语言解释。然而，它们对图像中文本的解读能力也可能使这些判断受到误导性语义线索的影响。我们系统地评估了针对检测导向、开放权重和商业 VLM 的排版攻击策略，考虑了真实到虚假和虚假到真实的攻击。我们的结果表明，推理模式通常比直接模式表现出更大的脆弱性，且攻击效果表现出明显的方向性不对称。此外，较大的模型往往表现出更高的干净检测准确率，但也伴随更高的攻击成功率。我们进一步考察了在图像和文本变换下的攻击鲁棒性，并研究了指示正确类别的覆盖层是否能帮助纠正错误。这些分析共同描绘了排版攻击如何影响真实性判断，并揭示了当前基于 VLM 的 AIGI 检测系统的局限性。
cs.CV / 140 / 2609.39681
MC-PanDA++: Simpler, Stronger, and More Robust Domain-Adaptive Panoptic Segmentation
MC-PanDA++：更简单、更强大且更稳健的领域自适应全景分割
Abstract
Unsupervised domain adaptation (UDA) reduces the annotation burden in panoptic segmentation by leveraging a cost-effectively labeled source domain (e.g., synthetic) and an unlabeled target domain to bridge the distribution gap. Existing panoptic UDA methods rely on teacher-student consistency learning built upon suboptimal per-pixel segmentation architectures. In contrast, state-of-the-art mask transformers are rarely adopted due to their pronounced vulnerability to confirmation bias in consistency learning, where erroneous teacher predictions are reinforced during training. Our earlier approach, MC-PanDA, mitigates this issue through fine-grained confidence estimation, which suppresses gradients from unreliable masks while sampling informative yet reliable locations for loss computation. However, this method entails a complex multi-stage training and requires careful hyperparameter tuning. This work presents MC-PanDA++, which addresses these limitations by introducing: (i) self-supervised vision encoders that provide a stronger and more robust initialization, further reducing the reliance on human annotations, (ii) per-class, self-adapting mask-wide loss scaling that stabilizes training and enables the usage of a single set of hyperparameters across domains, and (iii) a single-stage training pipeline that decreases overall conceptual complexity. Together, these improvements result in a conceptually simpler, better-performing, and more robust method for domain-adaptive panoptics. Source code: https://github.com/martinovicivan/MC-PanDA
Chinese Translation
无监督领域适应（UDA）通过利用成本效益高的标注源领域（例如，合成数据）和未标注的目标领域来缩减全景分割中的标注负担，从而弥合分布差距。现有的全景UDA方法依赖于基于次优逐像素分割架构的师生一致性学习。相比之下，最先进的掩膜变换器由于在一致性学习中对确认偏差的明显脆弱性而很少被采用，在训练过程中错误的教师预测会被强化。我们早期的方法MC-PanDA通过细粒度的置信度估计来缓解这一问题，抑制来自不可靠掩膜的梯度，同时为损失计算采样信息丰富且可靠的位置。然而，该方法涉及复杂的多阶段训练，并且需要仔细的超参数调优。本研究提出了MC-PanDA++，通过引入以下内容来解决这些局限性：（i）自监督视觉编码器，提供更强大且更稳健的初始化，进一步减少对人工标注的依赖；（ii）逐类自适应的掩膜宽度损失缩放，稳定训练并使得在不同领域中使用一组超参数成为可能；（iii）单阶段训练流程，降低整体概念复杂性。这些改进共同导致了一种在概念上更简单、性能更优越且更稳健的领域自适应全景分割方法。源代码：https://github.com/martinovicivan/MC-PanDA
cs.CV / 141 / 2609.39684
Unapologetically Distributed: A Call for Decentralized Document Analysis
毫不歉意的分布式：对去中心化文档分析的呼吁
Abstract
Privacy has become an increasingly important concern in the Document Analysis community, to the extent that in many environments such as archives, governmental institutions, and local businesses, the adoption of automation is restricted by legal and policy constraints. While federated learning has often been regarded as a ``necessary evil'', implying an unavoidable performance trade-off in exchange for decentralization and privacy, many prior works overlook its potential to improve robustness to out-of-distribution data. In this paper, we present Unapologetically Distributed, the first comprehensive study evaluating distributed learning in Document Analysis along three key axes simultaneously: the tasks addressed, the architectures employed, and the fine-tuning strategies applied. Specifically, we demonstrate how various distributed training approaches enhance generalization capabilities across diverse tasks such as Table Recognition, handwriting recognition, and Word Spotting, particularly during transfer learning stages. Our results provide strong evidence that decentralization is not merely a constraint, but a valuable opportunity to improve model robustness and adaptability in real-world Document Analysis scenarios.
Chinese Translation
隐私已成为文档分析领域日益重要的关注点，以至于在许多环境中，如档案馆、政府机构和地方企业，自动化的采用受到法律和政策限制。尽管联邦学习常常被视为一种“必要的恶”，暗示在去中心化和隐私之间存在不可避免的性能权衡，但许多先前的研究忽视了其提高对分布外数据的鲁棒性的潜力。在本文中，我们提出了“毫不歉意的分布式”，这是首个全面评估文档分析中分布式学习的研究，涵盖了三个关键维度：所解决的任务、所采用的架构和所应用的微调策略。具体而言，我们展示了各种分布式训练方法如何在表格识别、手写识别和词汇识别等多样化任务中增强泛化能力，特别是在迁移学习阶段。我们的结果提供了强有力的证据，表明去中心化不仅仅是一种限制，而是提高模型在现实世界文档分析场景中鲁棒性和适应性的宝贵机会。
cs.CV / 142 / 2609.39688
ShieldCLIP: Selective Safety Alignment for Harmful Content Mitigation in Multimodal Foundation Models
ShieldCLIP：多模态基础模型中有害内容缓解的选择性安全对齐
Abstract
Multimodal encoders such as CLIP underlie many downstream systems, but their web-scale training data embed harmful associations that safety alignment must suppress without unnecessarily changing benign representations. Because ethical and practical constraints prevent collecting real unsafe content at scale, existing datasets pair safe real samples with generated counterparts, but label every generated sample unsafe, even when one modality is individually safe. To address this, we introduce ShieldCLIP, the first framework to condition safety alignment on the observed safety state of each modality rather than the origin of a sample, preserving safe content while redirecting only what is unsafe. We also introduce ViSUv2, a 195k-quadruplet dataset with independent per-modality safety labels across 578 concepts and 28 categories. Using these labels, ShieldCLIP defines a four-way conditional objective beyond pair-level supervision: safe content is anchored, unsafe modalities are redirected to their safe counterparts, mixed pairs update only the unsafe branch, and coherence is enforced when both are unsafe. We evaluate ShieldCLIP on cross-modal retrieval, text-to-image generation with Stable Diffusion v1.4 and SDXL, and image-to-text generation with LLaVA. Across these settings, ShieldCLIP consistently reduces harmful outputs over prior safety-aligned encoders and strong mitigation baselines, while preserving the utility of the original embedding space. Extensive ablation studies further show that both modality-specific supervision and the selective alignment objective contribute to these gains. Source code, trained models, and ViSUv2 (under a controlled-access protocol) will be made publicly available at https://aimagelab.github.io/ShieldCLIP/.
Chinese Translation
多模态编码器如 CLIP 是许多下游系统的基础，但其网络规模的训练数据嵌入了有害的关联，安全对齐必须抑制这些关联，同时不必要地改变良性表示。由于伦理和实际约束，无法大规模收集真实的不安全内容，现有的数据集将安全的真实样本与生成的对应样本配对，但将每个生成样本标记为不安全，即使某一模态单独是安全的。为了解决这个问题，我们提出了 ShieldCLIP，这是第一个将安全对齐条件化于每个模态的观察安全状态而非样本来源的框架，保留安全内容，同时仅重定向不安全的部分。我们还引入了 ViSUv2，这是一个包含 195k 四元组的数据集，涵盖 578 个概念和 28 个类别，具有独立的每模态安全标签。利用这些标签，ShieldCLIP 定义了一个超越配对级监督的四向条件目标：安全内容被锚定，不安全模态被重定向到其安全对应物，混合配对仅更新不安全分支，当两者都不安全时强制一致性。我们在跨模态检索、使用 Stable Diffusion v1.4 和 SDXL 的文本到图像生成，以及使用 LLaVA 的图像到文本生成上评估了 ShieldCLIP。在这些设置中，ShieldCLIP 一致性地减少了有害输出，相较于之前的安全对齐编码器和强有力的缓解基线，同时保留了原始嵌入空间的效用。广泛的消融研究进一步表明，模态特定的监督和选择性对齐目标都对这些增益有所贡献。源代码、训练模型和 ViSUv2（在受控访问协议下）将公开发布于 https://aimagelab.github.io/ShieldCLIP/。
cs.CV / 143 / 2609.39704
When Masking Helps or Hurts Robustness in Compressed CLIP: A Pre-Deployment Diagnostic
掩蔽在压缩 CLIP 中何时有助于或损害鲁棒性：预部署诊断
Abstract
This paper demonstrate that whether masking-based token pruning helps or hurts worst-group robustness can be predicted before deployment, without labels or fine-tuning. A systematic study of semantic masking across 8 spurious-correlation benchmarks shows its effect on worst-group accuracy is highly unstable: it improves accuracy by up to 82.5\% relative on some datasets and degrades it by up to 100\% on others. We trace this instability to spurious inversion: background patches receive higher CLIP text-similarity than the true object when the spurious attribute is background-separable, inverting the assumption every text- and attention-guided pruning method relies on. We introduce the Spurious Inversion Metric (SIM), a label-free, pre-deployment diagnostic whose sign predicts this effect with statistical significance (binomial $p=0.035$) across all 8 datasets, and remains dependable across 6 CLIP architectures with a clean foreground/background split. Naive masking is itself a major source of risk: it causes the largest average-accuracy loss of any method we evaluate, and its own per-image segmentation step is a significant runtime bottleneck. To address this, we design a batched, synchronization-free GPU segmentation routine that cuts this overhead from 3.5$\times$ to 1.75$\times$ baseline. Gating deployment by SIM's sign recovers masking's benefits while avoiding its worst failures, matching or exceeding a strong pruning baseline on 7 of 8 datasets.
Chinese Translation
本文展示了基于掩蔽的令牌剪枝是否有助于或损害最差组鲁棒性可以在部署前进行预测，而无需标签或微调。对8个虚假相关基准的语义掩蔽进行系统研究表明，其对最差组准确率的影响高度不稳定：在某些数据集中，准确率相对提高了高达82.5\%，而在其他数据集中则降低了高达100\\%。我们将这种不稳定性追溯到虚假反转：当虚假属性是背景可分离时，背景补丁的 CLIP 文本相似度高于真实对象，从而颠覆了每种文本和注意力引导的剪枝方法所依赖的假设。我们引入了虚假反转度量（Spurious Inversion Metric, SIM），这是一种无标签的预部署诊断工具，其符号在所有8个数据集中以统计显著性（二项式 $p=0.035$）预测这一效果，并且在6个具有干净前景/背景分割的 CLIP 架构中保持可靠。简单的掩蔽本身就是一个主要风险来源：它导致我们评估的任何方法中最大的平均准确率损失，其自身的每图像分割步骤也是一个显著的运行时瓶颈。为了解决这个问题，我们设计了一种批处理的、无同步的 GPU 分割例程，将这一开销从3.5倍降低到1.75倍基线。通过 SIM 的符号来控制部署，恢复了掩蔽的好处，同时避免了其最严重的失败，在8个数据集中的7个上与强剪枝基线相匹配或超越。
cs.CV / 144 / 2609.39709
BTC3D: Blended Tile Conditioning for Detail-Enhancing Image-to-3D Generation
BTC3D：用于细节增强的图像到3D生成的混合瓦片条件化
Abstract
Recent diffusion-based pipelines have achieved promising progress in image-to-3D synthesis. However, generating high-fidelity details remains challenging, especially when the input image contains rich details. Existing approaches often rely on globally encoded conditioning features, which compress spatial information and limit the model to reproduce fine-grained details. This common design often leads to a phenomenon we term detail attenuation. Moreover, improving image-to-3D synthesis quality typically requires retraining or fine-tuning large diffusion models, which can be computationally expensive and impractical for complex 3D pipelines. In this work, we present Blended Tile Conditioning for image-to-3D generation (BTC3D), a training-free inference time framework that enhances fine-grained detail preservation in image-to-3D diffusion pipelines. To alleviate detail attenuation, we first examine the image feature additivity in image-to-3D models. Based on this property, we introduce a blended tile embedding that extracts local conditioning signals from split image regional patches, allowing the diffusion model to better preserve fine-grained visual details. To integrate the global and local conditioning guidance stably, we propose a dynamic conditioning schedule that gradually increases the influence of tile-level conditioning during later low-noise stages of diffusion. Our proposed method BTC3D operates entirely at inference time and can be seamlessly integrated into existing image-to-3D diffusion pipelines. Experimental results demonstrate that the proposed approach significantly improves texture quality and visual fidelity of the base model while maintaining global structural consistency in a training-free manner.
Chinese Translation
最近，基于扩散的管道在图像到3D合成方面取得了令人鼓舞的进展。然而，生成高保真细节仍然具有挑战性，尤其是当输入图像包含丰富细节时。现有的方法通常依赖于全局编码的条件特征，这会压缩空间信息并限制模型重现细粒度细节的能力。这种常见设计往往导致我们称之为细节衰减的现象。此外，提高图像到3D合成质量通常需要重新训练或微调大型扩散模型，这在计算上可能代价高昂，并且对于复杂的3D管道来说不切实际。在本研究中，我们提出了用于图像到3D生成的混合瓦片条件化（BTC3D），这是一个无训练推理时间框架，旨在增强图像到3D扩散管道中细粒度细节的保留。为了缓解细节衰减，我们首先考察了图像到3D模型中的图像特征可加性。基于这一特性，我们引入了一种混合瓦片嵌入，从分割的图像区域补丁中提取局部条件信号，使扩散模型能够更好地保留细粒度视觉细节。为了稳定地整合全局和局部条件指导，我们提出了一种动态条件调度，在扩散的后期低噪声阶段逐渐增加瓦片级条件的影响。我们提出的方法BTC3D完全在推理时操作，可以无缝集成到现有的图像到3D扩散管道中。实验结果表明，所提方法显著提高了基础模型的纹理质量和视觉保真度，同时以无训练的方式保持了全局结构一致性。
cs.CV / 145 / 2609.39723
Let the Carrier Carry the Attack: Preserving the Subject in Adversarial Image Generation
让载体承载攻击：在对抗性图像生成中保护主体
Abstract
Strong unrestricted adversarial attacks can distort the primary object of an image, hereafter referred to as the subject. To preserve subject integrity without compromising attack magnitude, we introduce the carrier: a secondary visual element that provides an auxiliary region to facilitate the attack under global classifier guidance. We demonstrate three key findings: 1. A carrier mitigates subject distortion by absorbing a larger share of globally normalized attack updates. 2. A carrier improves cross-model transferability, governed by the strength of target-related features that balance semantic separation and transfer performance. 3. Successful targeted attacks retain the personalized subject as the primary content perceived by humans while successfully misleading the classifier. Our results demonstrate that a visually secondary carrier offers an auxiliary spatial pathway for adversarial changes, enabling strong and transferable attacks while improving subject preservation.
Chinese Translation
强大的无约束对抗攻击可以扭曲图像的主要对象，以下称为主体。为了在不影响攻击强度的情况下保持主体的完整性，我们引入了载体：一个次要视觉元素，提供辅助区域以在全局分类器的指导下促进攻击。我们展示了三个关键发现：1. 载体通过吸收更大比例的全局归一化攻击更新来减轻主体扭曲。2. 载体提高了跨模型的可转移性，这由目标相关特征的强度决定，平衡了语义分离和转移性能。3. 成功的定向攻击保留了个性化主体作为人类感知的主要内容，同时成功误导分类器。我们的结果表明，视觉上次要的载体为对抗性变化提供了辅助空间通道，使得强大且可转移的攻击成为可能，同时改善了主体的保留。
cs.CV / 146 / 2609.39748
FAST: Flow Any Scene Transformer
FAST：任意场景流变换器
Abstract
Scaling has become a primary driver of progress in language and vision foundation models, yet its role in precise correspondence matching remains underexplored. In this work, we present Flow Any Scene Transformer (FAST), a scalable correspondence model driven by two key insights. First, we reveal that the query-key projections inside single-view vision foundation models encode a coarse yet reusable prior for cross-view matching. Second, reusing these pretrained projections in cross-attention form yields a highly effective initialization for a ViT-based matcher built from a single-view encoder. Guided by these insights, we build FAST upon a vanilla single-view foundation model, utilizing a zero-parameter rewiring strategy to convert selected self-attention layers into cross-attention for cross-view interaction. This design allows ViT-based matchers to scale with advances in single-view foundation models, bypassing the need for a dedicated pair-centric pretraining stage. To fully unlock the scaling potential of this formulation, we assemble a 6-million-pair training corpus for general-purpose dense 2D displacement estimation across diverse co-visible image pairs. Extensive experiments demonstrate that FAST achieves state-of-the-art performance across a wide range of benchmarks, while scaling favorably with both backbone size and training data.
Chinese Translation
规模化已成为语言和视觉基础模型进展的主要驱动力，但其在精确对应匹配中的作用仍未得到充分探索。在本研究中，我们提出了任意场景流变换器（Flow Any Scene Transformer，FAST），这是一种由两个关键见解驱动的可扩展对应模型。首先，我们揭示了单视图视觉基础模型中的查询-键投影编码了一种粗略但可重用的跨视图匹配先验。其次，在跨注意力形式中重用这些预训练投影为基于ViT的匹配器提供了一个高效的初始化，该匹配器是由单视图编码器构建的。在这些见解的指导下，我们基于一个普通的单视图基础模型构建了FAST，利用零参数重连策略将选定的自注意力层转换为跨注意力层以实现跨视图交互。这一设计使得基于ViT的匹配器能够随着单视图基础模型的进步而扩展，避免了专门的配对中心预训练阶段的需求。为了充分释放这一公式的扩展潜力，我们组建了一个600万对的训练语料库，用于在多样的共视图像对中进行通用的密集2D位移估计。大量实验表明，FAST在广泛的基准测试中实现了最先进的性能，同时在主干网络规模和训练数据上均表现出良好的扩展性。
cs.CV / 147 / 2609.39756
Determining Vertical Displacement of Agricultural Areas Using UAV-Photogrammetry and a Heteroscedastic Deep Learning Model
利用无人机摄影测量和异方差深度学习模型确定农业区域的垂直位移
Abstract
This article introduces an algorithm that uses a U-Net architecture to determine vertical ground surface displacements from unmanned aerial vehicle (UAV)-photogrammetry point clouds, offering an alternative to traditional ground filtering methods. Unlike con-ventional ground filters that rely on point cloud classification, the proposed approach em-ploys heteroscedastic regression. The U-Net model predicts the conditional expected val-ues of the elevation corrections, aiming to reduce the impact of vegetation on determined ground surface elevations. Concurrently, it estimates the logarithm of the elevation cor-rection variance, allowing for direct quantification of the uncertainty associated with each elevation correction value. The algorithm was evaluated using three metrics: the root mean square error (RMSE) of vertical displacements, the percentage of nodes with deter-mined displacement values, and the percentage of outliers among those values. Perfor-mance was assessed using the technique for order of preference by similarity to ideal so-lution (TOPSIS) method and compared against several ground-filter-based algorithms across four datasets, each including at least two time intervals. In most cases, the U-Net-based approach demonstrated a slight performance advantage over traditional ground filtering techniques. For example, for the U-Net-based algorithm, for one of the test da-tasets, the RMSE of the determined subsidences was 6.1 cm, the percentage of nodes with determined subsidences was 80.5%, and the percentage of outliers was 0.2%. For the same case, the algorithm based on the next best model (SMRF) allowed an RMSE of 7.7 cm to be obtained; for 77.3% of nodes, the subsidences were determined; and the percentage of outliers was 0.3%.
Chinese Translation
本文介绍了一种算法，该算法采用U-Net架构从无人机（UAV）摄影测量点云中确定地面垂直位移，为传统地面过滤方法提供了一种替代方案。与依赖点云分类的传统地面过滤器不同，所提出的方法采用异方差回归。U-Net模型预测高程修正的条件期望值，旨在减少植被对确定的地面高程的影响。同时，它估计高程修正方差的对数，从而允许直接量化与每个高程修正值相关的不确定性。该算法使用三项指标进行评估：垂直位移的均方根误差（RMSE）、具有确定位移值的节点百分比以及这些值中的异常值百分比。性能评估采用理想解相似性排序法（TOPSIS）方法，并与基于多种地面过滤算法的四个数据集进行比较，每个数据集至少包含两个时间间隔。在大多数情况下，基于U-Net的方法在性能上略优于传统的地面过滤技术。例如，在一个测试数据集中，基于U-Net的算法所确定的沉降的RMSE为6.1厘米，具有确定沉降的节点百分比为80.5%，异常值的百分比为0.2%。在同一案例中，基于下一个最佳模型（SMRF）的算法获得的RMSE为7.7厘米，77.3%的节点确定了沉降，异常值的百分比为0.3%。
cs.CV / 148 / 2609.39785
Seeing as Humans Do: Learning from Motion to Segment Anything Without Supervision
像人类一样看待：从运动中学习以无监督地分割任何物体
Abstract
The Segment Anything Model (SAM) relies heavily on massive manual annotations, creating a fundamental bottleneck for model scaling. While unsupervised methods attempt to learn object concepts from motion, they typically overfit to moving entities, lacking both multi-granularity understanding and the ability to generalize to static objects. To overcome this, we introduce Motion-Grounded Segment Anything (MoSA), a highly scalable unsupervised framework that learns a transferable objectness prior from unlabeled videos. MoSA operates in three progressive stages: (1) automatically generating multi-granularity motion pseudo-labels from large-scale video data; (2) training a Perceptual Grouping Model (PGM) via contrastive learning to internalize a generalized, appearance-driven concept of objects; and (3) transferring this learned prior into a prompt-guided architecture for segment-anything-style inference on images. Extensive zero-shot evaluations across seven challenging benchmarks (e.g., COCO and ADE20K) demonstrate that MoSA significantly outperforms existing unsupervised methods. Notably, despite using zero manual annotations, MoSA achieves segmentation performance comparable to the fully supervised SAM. Our findings reveal that harnessing large-scale unlabeled motion is a feasible and highly scalable alternative to annotation-driven segment-anything pipelines.
Chinese Translation
Segment Anything Model (SAM) 在很大程度上依赖于大量的手动标注，这为模型扩展创造了根本性的瓶颈。尽管无监督方法试图从运动中学习物体概念，但它们通常会过拟合于移动实体，缺乏多粒度理解和推广到静态物体的能力。为了解决这个问题，我们提出了基于运动的无监督分割框架（Motion-Grounded Segment Anything，MoSA），它从未标记的视频中学习可转移的物体性先验。MoSA 通过三个渐进阶段进行操作：（1）从大规模视频数据中自动生成多粒度运动伪标签；（2）通过对比学习训练感知分组模型（Perceptual Grouping Model，PGM），内化一个以外观驱动的物体概念；（3）将这一学习到的先验转移到一个提示引导的架构中，以进行图像上的分割推断。对七个具有挑战性的基准（例如 COCO 和 ADE20K）的广泛零样本评估表明，MoSA 显著优于现有的无监督方法。值得注意的是，尽管没有使用任何手动标注，MoSA 的分割性能与完全监督的 SAM 相当。我们的研究结果表明，利用大规模未标记的运动是一种可行且高度可扩展的替代方案，能够替代基于标注的分割方法。
cs.CV / 149 / 2609.39794
Inline Memory Meets Reusable Skills: Memory-centric Framework for Vision-Language-Action Model
内联记忆与可重用技能相结合：面向视觉-语言-行动模型的记忆中心框架
Abstract
Vision-Language-Action (VLA) models have shown strong promise for general-purpose robotic manipulation, yet adapting them to new tasks and domains remains inefficient: existing methods often rely on parameter tuning, incurring substantial costs and risking catastrophic forgetting of previously learned tasks. To address this, we propose \textbf{Optimus-R}, a memory-centric VLA framework that formulates robotic adaptation as explicit query-skill memory tuning. Optimus-R introduces: (i) An \textbf{Inline Memory Interface for skill extraction}. It inserts learnable memory tokens into the VLA prefix stream, allowing the backbone to derive control-aware query and skill representations within the native action-conditioning pathway. (ii) A \textbf{Query-Skill Memory Bank for skill learning}. It externalizes skills into query prototypes for deciding \emph{what} to retrieve and skill values for specifying \emph{how} to act, supporting skill reuse and expansion with limited parameter updates. (iii) A lightweight \textbf{Bridge-and-Adapt mechanism for skill updating}. It aligns target-domain queries and skills with the existing memory space through a lightweight adapter and residual memory updates. Experiments on in-domain adaptation, cross-domain adaptation, and lifelong learning show that Optimus-R enables data-efficient skill learning while mitigating catastrophic forgetting.
Chinese Translation
视觉-语言-行动（VLA）模型在通用机器人操作中展现出强大的潜力，但将其适应于新任务和新领域仍然效率低下：现有方法通常依赖于参数调优，导致高昂的成本并且存在对先前学习任务的灾难性遗忘风险。为了解决这一问题，我们提出了 extbf{Optimus-R}，一个以记忆为中心的 VLA 框架，将机器人适应性表述为显式的查询-技能记忆调优。Optimus-R 引入了：(i) 一个 extbf{内联记忆接口用于技能提取}。它将可学习的记忆标记插入到 VLA 前缀流中，使得主干能够在本地的行动条件路径中推导出控制感知的查询和技能表示。(ii) 一个 extbf{查询-技能记忆库用于技能学习}。它将技能外部化为查询原型，以决定 extit{什么} 被检索，以及技能值以指定 extit{如何} 行动，支持在有限参数更新下的技能重用和扩展。(iii) 一个轻量级的 extbf{桥接与适应机制用于技能更新}。它通过轻量级适配器和残余记忆更新，将目标领域的查询和技能与现有记忆空间对齐。关于领域内适应、跨领域适应和终身学习的实验表明，Optimus-R 实现了数据高效的技能学习，同时减轻了灾难性遗忘。
cs.CV / 150 / 2609.39832
P-SRM: Selective Recovery of Rejected Predictions in Visual Tracking
P-SRM：视觉跟踪中被拒绝预测的选择性恢复
Abstract
Many visual tracking methods use rejection mechanisms to suppress unreliable predictions. However, these mechanisms can also reject correctly localized candidates, leaving useful information unused. We investigate how to identify and recover these candidates while preserving native accepted outputs and candidate coordinates. To this end, we propose P-SRM (Post-rejection Selective Recovery Method), which combines spatial responses, past accepted states, and native decision margins to reassess candidates and selectively restore reliable predictions. We evaluate P-SRM on six trackers and four datasets spanning category-specific, point, and generic object tracking. Across all nine configurations, P-SRM improves rejected-candidate ranking and overall tracking performance. These results show that post-rejection verification can identify and recover useful predictions discarded by native rejection, demonstrating the value of reusing rejected information. Project repository: https://github.com/PalestyHR/P-SRM.
Chinese Translation
许多视觉跟踪方法使用拒绝机制来抑制不可靠的预测。然而，这些机制也可能拒绝正确定位的候选项，从而使有用的信息未被利用。我们研究如何在保留原始接受输出和候选坐标的同时，识别和恢复这些候选项。为此，我们提出了P-SRM（后拒绝选择性恢复方法），该方法结合了空间响应、过去的接受状态和原始决策边际，以重新评估候选项并选择性地恢复可靠的预测。我们在六个跟踪器和四个数据集上评估了P-SRM，这些数据集涵盖了特定类别、点和通用物体跟踪。在所有九个配置中，P-SRM提高了被拒绝候选项的排名和整体跟踪性能。这些结果表明，后拒绝验证可以识别和恢复被原始拒绝丢弃的有用预测，展示了重用被拒绝信息的价值。项目代码库：https://github.com/PalestyHR/P-SRM。
cs.CV / 151 / 2609.39836
Spherical Interpolation for Backward-Compatible Multimodal Representations
用于向后兼容的多模态表示的球面插值
Abstract
Contrastive vision-language models map visual and textual representations into a shared normalized embedding space, making cosine similarity the natural metric for cross-modal retrieval. A practical challenge arises during model upgrades: independently trained models generally produce incompatible representation spaces, so replacing a deployed model typically requires recomputing embeddings for the entire gallery, which is prohibitively expensive at scale. Orthogonal post-hoc alignment can partially mitigate this problem by mapping new-model queries into the old-model gallery space. However, because independently trained models can differ in fine-grained representation structure, the orthogonal alignment remains approximate, leaving a residual angular discrepancy between the old-model query and the aligned new-model query. We study whether interpolation along the spherical geodesic between these two normalized query representations can improve retrieval without re-indexing the gallery. We characterize when this path contains an interior query direction closer to an idealized retrieval-optimal direction than either endpoint, and connect this characterization to Recall@$K$ through a local margin-based certification result. Experiments across multiple benchmarks and model families show that post-alignment spherical interpolation improves over orthogonal alignment alone, recovering backward-compatibility in most evaluated settings. Consistent with our geometric characterization, per-query oracle analysis shows that retrieval-favorable interior points occur frequently in practice. Code is available at https://github.com/miccunifi/SLERP_backward_compatibility .
Chinese Translation
对比视觉-语言模型将视觉和文本表示映射到共享的标准化嵌入空间，使得余弦相似度成为跨模态检索的自然度量。在模型升级过程中出现了一个实际挑战：独立训练的模型通常会产生不兼容的表示空间，因此替换已部署的模型通常需要为整个图库重新计算嵌入，这在规模上是极其昂贵的。正交后期对齐可以部分缓解这个问题，通过将新模型的查询映射到旧模型的图库空间。然而，由于独立训练的模型在细粒度表示结构上可能存在差异，正交对齐仍然是近似的，导致旧模型查询与对齐的新模型查询之间存在残余的角度差异。我们研究沿着这两个标准化查询表示之间的球面测地线插值是否可以在不重新索引图库的情况下改善检索。我们描述了何时这条路径包含一个内部查询方向，该方向比任一端点更接近理想化的检索最优方向，并将这一特征与 Recall@$K$ 通过局部基于边际的认证结果联系起来。在多个基准和模型系列上的实验表明，后期对齐的球面插值优于单独的正交对齐，在大多数评估设置中恢复了向后兼容性。与我们的几何特征一致，逐查询的oracle分析显示，检索有利的内部点在实践中经常出现。代码可在 https://github.com/miccunifi/SLERP_backward_compatibility 获取。
cs.CV / 152 / 2609.39841
DyRAD: Radar Novel View Synthesis for Dynamic Driving Scenes
DyRAD：动态驾驶场景的雷达新视图合成
Abstract
Reconstructing dynamic driving scenes from recorded sensor data supports closed-loop evaluation of autonomous driving systems by synthesizing observations beyond the original trajectory. Unlike cameras and LiDAR, radar measures radial velocity directly through Doppler. Yet existing radar novel-view synthesis fails to exploit this capability: methods addressing dynamic scenes reconstruct only range-azimuth tensors, while methods that render Doppler assume static scenes. Moreover, because radar processing spreads each reflection across multiple bins, existing representations absorb this spread into scene geometry, causing it to render incorrectly when the viewpoint moves. We present DyRAD, which models dynamic driving scenes using static background reflectors and motion-tracked dynamic point reflectors to render complete range-azimuth-Doppler (RAD) tensors. Reflector velocities are derived from object tracks and projected onto the line of sight, making Doppler both a rendered output and supervision for those tracks. Crucially, we render reflectors through a fixed analytic point-spread function (PSF) derived from the radar's signal-processing chain, preventing sensor-induced spread from being baked into the scene representation. Beyond improving scene reconstruction, this separation also enables zero-shot sensor-configuration transfer, allowing the same reconstructed scene to be rendered under different radar specifications without refitting. We evaluate DyRAD on RADIal, Boreas, and a synthetic benchmark across both on-path poses and displaced viewpoints untested by prior work. On RADIal, DyRAD recovers radar detections in 90.7% of reference-detected objects, compared with 26.9% for the strongest baseline.
Chinese Translation
从记录的传感器数据中重建动态驾驶场景，通过合成超出原始轨迹的观察，支持对自动驾驶系统的闭环评估。与摄像头和激光雷达不同，雷达通过多普勒直接测量径向速度。然而，现有的雷达新视图合成未能充分利用这一能力：处理动态场景的方法仅重建范围-方位张量，而渲染多普勒的方法则假设场景是静态的。此外，由于雷达处理将每个反射扩展到多个箱中，现有表示将这种扩展吸收到场景几何中，导致在视点移动时渲染不正确。我们提出了DyRAD，它通过静态背景反射器和运动跟踪的动态点反射器来建模动态驾驶场景，以渲染完整的范围-方位-多普勒（RAD）张量。反射器速度由物体轨迹推导并投影到视线方向，使多普勒既成为渲染输出又成为这些轨迹的监督。关键是，我们通过从雷达信号处理链中推导出的固定解析点扩散函数（PSF）来渲染反射器，防止传感器引起的扩展被固化到场景表示中。除了改善场景重建外，这种分离还使零-shot传感器配置转移成为可能，允许在不同雷达规格下渲染相同的重建场景而无需重新拟合。我们在RADIal、Boreas和一个合成基准上评估DyRAD，涵盖了未被先前工作测试的路径姿态和位移视点。在RADIal上，DyRAD在90.7%的参考检测对象中恢复了雷达检测，而最强基线仅为26.9%。
cs.CV / 153 / 2609.39871
Hyperspectral Image Models: Technical Report
高光谱图像模型：技术报告
Abstract
Hyperspectral remote sensing has advanced across diverse deep learning paradigms, including spectral spatial CNNs, Vision Transformers, Mamba, graph neural networks, Kolmogorov Arnold networks, and self supervised masked autoencoding. Yet progress remains hindered by fragmented repositories, incompatible tensor conventions, and non standardized evaluation. Hyperspectral Image Models addresses these challenges through a modular framework unifying 55 representative models across six paradigms with a common registry, automatic 4D/5D tensor adaptation, and standardized constructors. It integrates 24 benchmark scenes from Airborne, Spaceborne, UAV, and Mars CRISM sensors, with caching, label remapping, PCA, explicit band selection or raw spectra, optional spatial max pooling, and arbitrary PxP patch extraction. To prevent inflated accuracy from overlapping windows, it supports class balanced random partitioning and spatially disjoint regional blocking with Chebyshev guard bands that eliminate train test pixel overlap. Experiments use a single config.yaml with deterministic seeds and complete provenance, generating LaTeX benchmark tables and classification maps. Across 1,320 model scene evaluations and 6,600 seeded runs, scene difficulty dominates architecture, with mean accuracy ranging from 96.40% on Botswana to 56.70% on Houston 2018, versus a 15 point spread across paradigm means. No paradigm universally dominates, while sub 1 M parameter models can match architectures two orders of magnitude larger. Code is publicly available at https://github.com/Tanishq251/Hyperspectral-Image-Models.
Chinese Translation
高光谱遥感在多种深度学习范式中取得了进展，包括光谱空间卷积神经网络（spectral spatial CNNs）、视觉变换器（Vision Transformers）、Mamba、图神经网络（graph neural networks）、Kolmogorov Arnold 网络和自监督掩码自编码（self supervised masked autoencoding）。然而，进展仍受到碎片化的资源库、不兼容的张量约定以及非标准化评估的阻碍。《高光谱图像模型》通过一个模块化框架解决了这些挑战，该框架统一了六种范式中的55个代表性模型，提供了一个公共注册表、自动的4D/5D张量适配和标准化构造函数。它整合了来自航空、航天、无人机（UAV）和火星CRISM传感器的24个基准场景，支持缓存、标签重映射、主成分分析（PCA）、显式波段选择或原始光谱、可选的空间最大池化以及任意PxP补丁提取。为了防止重叠窗口导致的准确率膨胀，它支持类平衡随机分区和空间不相交的区域阻塞，使用切比雪夫保护带（Chebyshev guard bands）消除训练和测试像素的重叠。实验使用单一的config.yaml文件，设定确定性种子和完整的来源，生成LaTeX基准表和分类图。在1,320个模型场景评估和6,600次种子运行中，场景难度主导了架构，平均准确率从博茨瓦纳的96.40%到休斯顿2018的56.70%不等，而各范式的均值差距为15个百分点。没有任何一个范式在所有方面占据绝对优势，而参数少于100万的模型能够与两个数量级更大的架构相匹配。代码已公开，地址为 https://github.com/Tanishq251/Hyperspectral-Image-Models。
cs.CV / 154 / 2609.39883
Grounding with Confidence: Controllable Generative Video Temporal Grounding
自信的基础：可控生成视频时间定位
Abstract
Video temporal grounding supports applications such as video search, content review, and automated editing by localizing events described in natural language. Yet existing generative models typically output timestamps without explicit interval-level confidence scores to guide candidate selection. We separate candidate generation from acceptance by scoring individual intervals within the original decoding pass. A lightweight confidence head reads pooled decoder states, providing an explicit score trained for interval selection. Offline verifier scores supervise the head on fixed candidate sequences, and temporal-overlap labels adapt it to current rollouts during reinforcement learning. GT-anchored candidate-pool supervision and set-level optimization train the generator. The resulting scores support ranking, threshold-based selection, and rejection without invoking an external verifier at inference. On a fixed OMTG-Bench candidate pool, confidence raises query-macro
[email protected] from 9.95% to 14.42% over generation order at a 10% global return budget, and from 26.48% to 31.12% at a 25% budget. The continuous scores let downstream applications adjust return budgets or acceptance thresholds to match their precision-recall preferences, without regenerating candidate intervals.
Chinese Translation
视频时间定位通过定位用自然语言描述的事件，支持视频搜索、内容审查和自动编辑等应用。然而，现有的生成模型通常输出时间戳，而没有明确的区间级置信度分数来指导候选选择。我们通过在原始解码过程中对单个区间进行评分，将候选生成与接受分开。一个轻量级的置信度头读取池化的解码器状态，提供一个为区间选择训练的明确分数。离线验证器评分在固定候选序列上监督该头，并且时间重叠标签在强化学习期间将其适应于当前的回滚。基于GT的候选池监督和集合级优化训练生成器。由此产生的分数支持排名、基于阈值的选择和拒绝，而无需在推理时调用外部验证器。在固定的OMTG-Bench候选池上，置信度在10%全球回报预算下将查询宏观
[email protected]从9.95%提高到14.42%，在25%预算下从26.48%提高到31.12%。连续的分数使下游应用能够调整回报预算或接受阈值，以匹配其精确度-召回偏好，而无需重新生成候选区间。
cs.CV / 155 / 2609.39894
Spatial-Temporal Multi-scale Network for Screen Content Video Quality Enhancement
用于屏幕内容视频质量增强的时空多尺度网络
Abstract
Different from natural videos, Screen Content Videos (SCVs) are characterized by abrupt motion, scene switches, and high-frequency details such as text and graphics. Conventional video enhancement methods, which rely heavily on temporal continuity, often suffer from performance degradation when processing SCVs due to the disruption of temporal correlations. To address these challenges, we propose the Spatial-Temporal Multi-scale Network (STM-Net), a novel framework specifically tailored for compressed SCV enhancement. Our approach integrates three complementary components: a Prior-Guided Spatio-Temporal Dispatcher (PG-STD) that routes input into three parallel streams to avoid feature contamination, a Bidirectional Temporal Feature Extraction (BTFE) module that adaptively handles abrupt transitions without explicit detection, and a Cascaded Multi-scale Feature Distillation (CMFD) module that preserves critical high-frequency details. Experimental results demonstrate that STM-Net outperforms state-of-the-art methods in both objective metrics and subjective visual quality, providing a robust solution for screen content artifacts. Code is available at https://github.com/HUANGZiyin1/STM-Net.
Chinese Translation
与自然视频不同，屏幕内容视频（SCVs）具有突发运动、场景切换以及文本和图形等高频细节的特点。传统的视频增强方法过于依赖时间连续性，因此在处理SCVs时常常因时间相关性的破坏而导致性能下降。为了解决这些挑战，我们提出了一种专门针对压缩SCV增强的新框架——时空多尺度网络（STM-Net）。我们的方法集成了三个互补组件：一个优先引导的时空调度器（Prior-Guided Spatio-Temporal Dispatcher, PG-STD），它将输入路由到三个并行流以避免特征污染；一个双向时间特征提取（Bidirectional Temporal Feature Extraction, BTFE）模块，它能够自适应处理突发过渡而无需显式检测；以及一个级联多尺度特征蒸馏（Cascaded Multi-scale Feature Distillation, CMFD）模块，能够保留关键的高频细节。实验结果表明，STM-Net在客观指标和主观视觉质量上均优于最先进的方法，为屏幕内容伪影提供了稳健的解决方案。代码可在 https://github.com/HUANGZiyin1/STM-Net 获取。
cs.CV / 156 / 2609.39899
Learning Where to Look: Anatomical Grounding and Guided Attention for Cardiac MRI Vision-Language Models
学习关注点：心脏MRI视觉-语言模型的解剖基础与引导注意力
Abstract
Cardiac magnetic resonance imaging (CMR) enables assessment of cardiac anatomy, ventricular function, and myocardial tissue characteristics. Clinicians interpret these images by identifying cardiac structures and focusing on the regions relevant to each clinical question, motivating anatomically guided vision-language models (VLMs). Yet CMR-specific supervision for anatomical localisation and clinical question answering remains limited. To address this gap, we investigate fine-grained CMR visual question answering through anatomical grounding and guided attention. We construct 128,915 anatomical-grounding and 42,799 clinical QA pairs across short-axis cine, late gadolinium enhancement, and long-axis cine. These datasets support anatomical recognition, localisation, and clinical assessment without requiring paired reports for individual training images. To help the model learn where to look, we introduce Cardiac Anatomy-Routed Attention (CARA), which selects predicted anatomical priors according to the question and guides decoder attention with learned task-specific strengths. Combining anatomical grounding pretraining with CARA yields our model, CARA-VL. Experiments demonstrate CARA-VL's strengths in clinical assessment and regional localisation across CMR imaging settings, with promising generalization to an external clinical cohort. Together, our data and method provide a practical framework for studying and advancing cardiac visual understanding in VLMs. We will release the QA data derived from public datasets upon publication.
Chinese Translation
心脏磁共振成像（CMR）能够评估心脏解剖、心室功能和心肌组织特征。临床医生通过识别心脏结构并关注与每个临床问题相关的区域来解读这些图像，这激励了基于解剖的引导视觉-语言模型（VLMs）的研究。然而，针对解剖定位和临床问题回答的CMR特定监督仍然有限。为了解决这一问题，我们通过解剖基础和引导注意力研究细粒度的CMR视觉问答。我们构建了128,915对解剖基础和42,799对临床问答对，涵盖短轴动态成像、晚期钆增强和长轴动态成像。这些数据集支持解剖识别、定位和临床评估，而无需为单个训练图像提供配对报告。为了帮助模型学习关注点，我们引入了心脏解剖引导注意力（Cardiac Anatomy-Routed Attention, CARA），该方法根据问题选择预测的解剖先验，并用学习到的任务特定强度引导解码器注意力。将解剖基础预训练与CARA相结合，形成了我们的模型CARA-VL。实验表明，CARA-VL在临床评估和区域定位方面在CMR成像设置中表现出色，并在外部临床队列中展现出良好的泛化能力。我们的数据和方法共同提供了一个实用框架，用于研究和推进VLMs中的心脏视觉理解。我们将在发表时发布从公共数据集中衍生的问答数据。
cs.CV / 157 / 2609.39915
NavHarness: Adaptive Goals for Agentic Vision-Language Navigation
NavHarness：面向自主视觉-语言导航的自适应目标
Abstract
Vision-Language Navigation (VLN) requires embodied agents to generate actions based on instructions and observations. General-purpose multimodal agents offer a promising basis for this task, but selecting plausible local actions does not ensure that execution remains consistent with the intended route, particularly in long-horizon tasks. Moreover, the accumulated interaction history increases the input required for subsequent decisions, resulting in a significant inference overhead. To this end, we introduce \method, an Agentic VLN framework that includes a Goal Agent that sets adaptive goals for local actions, a Verify Agent that dynamically verifies whether a goal has been completed, a Memory Agent for multimodal context compression, and a Visuomotor Agent to execute adaptive goals. Specifically, the Goal Agent formulates adaptive goals based on the instruction, current observation, and execution history. Then the Visuomotor Agent executes navigation actions to achieve each goal, while the Verify Agent uses a goal-specific verification question to dynamically assess whether the observed outcomes satisfy the intended completion condition. Verified goal completion then marks a boundary for the Memory Agent to compress the corresponding multimodal interaction history while preserving information needed for subsequent navigation. We evaluate navigation on R2R-CE and RxR-CE, examine framework variants across three model backbones, and study context evolution during execution. For Real-World evaluation, \method achieves 83.3\% success and 1.51\,m navigation error across eight challenging routes evaluated three times each.
Chinese Translation
视觉-语言导航（VLN）要求具身代理基于指令和观察生成行动。通用多模态代理为这一任务提供了有希望的基础，但选择合理的局部行动并不能确保执行与预期路线保持一致，特别是在长时间跨度的任务中。此外，累积的交互历史增加了后续决策所需的输入，导致显著的推理开销。为此，我们提出了 extit{NavHarness}，一个自主VLN框架，其中包括一个设定局部行动自适应目标的目标代理（Goal Agent），一个动态验证目标是否已完成的验证代理（Verify Agent），一个用于多模态上下文压缩的记忆代理（Memory Agent），以及一个执行自适应目标的视觉运动代理（Visuomotor Agent）。具体而言，目标代理根据指令、当前观察和执行历史制定自适应目标。然后，视觉运动代理执行导航行动以实现每个目标，而验证代理使用特定于目标的验证问题动态评估观察结果是否满足预期的完成条件。经过验证的目标完成标志着记忆代理压缩相应多模态交互历史的边界，同时保留后续导航所需的信息。我们在R2R-CE和RxR-CE上评估导航，考察三个模型骨干的框架变体，并研究执行过程中的上下文演变。在现实世界评估中， extit{NavHarness}在八条具有挑战性的路线中实现了83.3\%的成功率和1.51米的导航误差，且每条路线评估三次。
cs.CV / 158 / 2609.39920
MCD: Causal Distillation of Multimodal In-Context Learning in Large Vision-Language Models
MCD：大型视觉-语言模型中多模态上下文学习的因果蒸馏
Abstract
Large vision-language models (LVLMs) exhibit strong multimodal in-context learning (ICL) capabilities, yet this ability degrades substantially as model size decreases. Knowledge distillation offers a natural way to bridge this gap, but existing methods primarily align output distributions or hidden representations directly. Such alignment teaches the student what the teacher predicts without revealing which evidence in the complex context causally supports that prediction. Consequently, a student can imitate the teacher's answer while continuing to rely on language priors, prompt structure, or other spurious cues. To address this limitation, we introduce Multimodal Causal Distillation (MCD), a distillation framework that transfers how a strong teacher uses multimodal evidence during ICL. MCD uses structure-preserving token interventions to identify and verify causal evidence, then transfers how the teacher responds when that evidence is retained or removed. This design connects distillation to the causal patterns by which the model uses contextual evidence during multimodal ICL. Experiments across three LVLM families and seven benchmarks show that MCD improves student performance by 7.23 points on average and outperforms vanilla distillation by 4.68 points, while further analyses confirm the generalizability of these gains.
Chinese Translation
大型视觉-语言模型（LVLMs）展现出强大的多模态上下文学习（ICL）能力，但随着模型规模的减小，这种能力显著下降。知识蒸馏提供了一种自然的方式来弥补这一差距，但现有方法主要是直接对齐输出分布或隐藏表示。这种对齐教会学生教师的预测，而未揭示复杂上下文中哪些证据因果地支持该预测。因此，学生可以模仿教师的答案，同时继续依赖语言先验、提示结构或其他虚假线索。为了解决这一局限性，我们引入了多模态因果蒸馏（MCD），这是一个蒸馏框架，转移强教师在ICL过程中如何使用多模态证据。MCD使用结构保持的标记干预来识别和验证因果证据，然后转移教师在保留或移除该证据时的反应。这一设计将蒸馏与模型在多模态ICL过程中使用上下文证据的因果模式连接起来。针对三种LVLM家族和七个基准的实验表明，MCD平均提高了学生性能7.23分，并且比传统蒸馏提高了4.68分，进一步分析确认了这些提升的普遍性。
cs.CV / 159 / 2609.39924
CoVisco: Codec-Native Vision Encoder with Native Token Compression for Unified Image-Video Understanding
CoVisco：具有原生令牌压缩的编解码器原生视觉编码器，用于统一图像-视频理解
Abstract
Vision-language models face a fundamental scaling bottleneck: the number of visual tokens grows with both temporal duration and spatial resolution, making long-video understanding expensive for the vision encoder and the language model. Existing methods often compress visual tokens after dense encoding, creating a mismatch between the representation used during training and the compact interface required at deployment. We present CoVisco, a codec-native vision encoder with native token compression for unified image-video understanding. By combining codec-native input support with segmented attention, CoVisco can encode long visual inputs in a single forward pass without forming dense patch-to-patch interactions across all frames. Each temporal segment is equipped with learnable abstract tokens that learn a compact segment-level representation, while fine-grained patch tokens remain available throughout the encoder. Alternating intra-segment and abstract-communication layers preserve video-level context through the abstract-token channel. A lightweight selector further exposes either abstract tokens alone or abstract tokens augmented with a runtime-selected subset of patch tokens, yielding a compact visual interface that reduces the visual context and prefill burden of downstream MLLMs while retaining fine-grained evidence when needed. Pretrained with contrastive objectives on 565M image--text pairs and 6.4M videos, CoVisco shows competitive performance on video-oriented embedding and multimodal understanding benchmarks. In the evaluated four-segment, 64-frame setting, abstract-only inference uses only 400 visual tokens while achieving video-understanding performance close to, and on some benchmarks exceeding, OneVision-Encoder. Selected patch tokens further improve fine-grained video reasoning. Project URL: https://github.com/ernie-research/CoVisco.git
Chinese Translation
视觉-语言模型面临一个根本性的扩展瓶颈：视觉令牌的数量随着时间持续和空间分辨率的增加而增长，使得长视频理解对视觉编码器和语言模型来说成本高昂。现有方法通常在密集编码后压缩视觉令牌，导致训练期间使用的表示与部署时所需的紧凑接口之间存在不匹配。我们提出了CoVisco，一种具有原生令牌压缩的编解码器原生视觉编码器，用于统一图像-视频理解。通过结合编解码器原生输入支持与分段注意力，CoVisco能够在一次前向传递中编码长视觉输入，而无需在所有帧之间形成密集的补丁间交互。每个时间段配备可学习的抽象令牌，以学习紧凑的段级表示，同时细粒度的补丁令牌在整个编码器中保持可用。交替的段内和抽象通信层通过抽象令牌通道保留视频级上下文。一个轻量级选择器进一步暴露仅抽象令牌或增强了运行时选择的补丁令牌子集的抽象令牌，从而产生一个紧凑的视觉接口，减少下游多模态大语言模型（MLLMs）的视觉上下文和预填充负担，同时在需要时保留细粒度证据。在565M图像-文本对和6.4M视频上通过对比目标进行预训练，CoVisco在面向视频的嵌入和多模态理解基准上表现出竞争力。在评估的四段64帧设置中，仅使用抽象推理的情况下仅需400个视觉令牌，同时在视频理解性能上接近并在某些基准上超过OneVision-Encoder。选择的补丁令牌进一步改善了细粒度视频推理。项目网址：https://github.com/ernie-research/CoVisco.git
cs.CV / 160 / 2609.39926
Super-Resolving Unseen Hyperspectral Sensors at Any Scale via Spatial Operators
通过空间算子在任意尺度上超分辨率重建未见高光谱传感器
Abstract
Achieving cross-sensor generalization and arbitrary-scale reconstruction with a single model remains challenging in hyperspectral super-resolution (HSR). Although recent methods support arbitrary-scale reconstruction, applying them to new sensors or scales beyond the training range often requires additional data and computation to maintain reconstruction quality. To address these challenges, we propose OmniHSR, which predicts band-shared spatial operators rather than spectral values. Cross-Spectral Mapping (CSM) resamples inputs with any number of bands to fixed reference positions and predicts local operators with Gaussian supports. Continuous Operator-Field Reconstruction (COFR) composes these operators into a continuous field and applies them to all original bands for arbitrary-scale reconstruction. Experiments demonstrate that operator prediction outperforms direct spectral-value prediction on all seven datasets. Trained solely on ARAD with only 0.538M parameters, OmniHSR outperforms all directly transferred baselines on six unseen datasets without target-domain training data or adaptation. Across twelve upsampling factors from $\times2$ to $\times48$, it improves average PSNR on Pavia U and Chikusei by 0.55 dB over the strongest baseline. It also surpasses baselines trained from scratch or adapted on the target sensor and achieves up to $36\times$ faster inference. Our code will be publicly released soon.
Chinese Translation
在高光谱超分辨率（HSR）中，实现跨传感器泛化和任意尺度重建仍然具有挑战性。尽管最近的方法支持任意尺度的重建，但将其应用于新的传感器或超出训练范围的尺度通常需要额外的数据和计算，以保持重建质量。为了解决这些挑战，我们提出了OmniHSR，它预测共享波段的空间算子，而不是光谱值。跨光谱映射（Cross-Spectral Mapping, CSM）将任意数量波段的输入重采样到固定的参考位置，并预测具有高斯支持的局部算子。连续算子场重建（Continuous Operator-Field Reconstruction, COFR）将这些算子组合成一个连续的场，并将其应用于所有原始波段以实现任意尺度重建。实验表明，在所有七个数据集上，算子预测的性能优于直接光谱值预测。OmniHSR仅在ARAD上训练，参数量仅为0.538M，在六个未见数据集上超越了所有直接迁移的基线，无需目标领域的训练数据或适应。在从$ imes2$到$ imes48$的十二个上采样因子中，它在Pavia U和Chikusei上的平均PSNR比最强基线提高了0.55 dB。它还超越了从头训练或在目标传感器上适应的基线，并实现了高达$36 imes$的推理加速。我们的代码将很快公开发布。
cs.CV / 161 / 2609.39953
Learning to Reason with Compressed Context: Ground-Truth-Free Adaptation of OmniLLMs via Self-Distillation
通过自我蒸馏实现压缩上下文的推理学习：无真实标签的 OmniLLMs 自适应
Abstract
Omni-modal large language models (OmniLLMs) enable unified audio-video understanding, but their long multimodal token sequences make deployment computationally expensive. Token compression reduces this cost, yet aggressive compression often lowers accuracy. Existing works predominantly focus on designing better compression mechanisms; however, adapting the underlying language model to reason effectively over the remaining compressed context remains under-explored. To address this, we propose CAFD (Compressed-Context Adaptation via Full-Context Distillation), a ground-truth-free self-distillation framework that adapts OmniLLMs to fixed compression pipelines without requiring reference answers, rationales, or correctness rewards. CAFD leverages the full-token view of the same multimodal sample as a source of privileged information: a full-context self-teacher provides soft target supervision to a compressed-context student along the student's on-policy trajectory. Evaluated on Qwen2.5-Omni-7B across five audio-video benchmarks, five compression pipelines, and five deployment budgets, CAFD demonstrates consistent gains, improving 120 out of 125 conditions with an average accuracy boost of 1.44 points and recovering 26.9% of the accuracy gap on average. These results demonstrate that the proposed ground-truth-free adaptation offers an effective and practical route to improving the accuracy-efficiency trade-off in deployed OmniLLMs.
Chinese Translation
全模态大型语言模型（OmniLLMs）实现了统一的音频-视频理解，但其长的多模态令牌序列使得部署计算成本高昂。令牌压缩降低了这一成本，但过度压缩往往会降低准确性。现有研究主要集中在设计更好的压缩机制上；然而，如何调整基础语言模型以有效地在剩余的压缩上下文中进行推理仍然未得到充分探索。为了解决这一问题，我们提出了 CAFD（通过全上下文蒸馏进行压缩上下文自适应），这是一种无真实标签的自我蒸馏框架，能够在不需要参考答案、推理或正确性奖励的情况下，将 OmniLLMs 适应于固定的压缩管道。CAFD 利用相同多模态样本的全令牌视图作为特权信息的来源：全上下文自教师为压缩上下文学生提供软目标监督，沿着学生的在政策轨迹进行评估。在五个音频-视频基准、五个压缩管道和五个部署预算上评估 Qwen2.5-Omni-7B，CAFD 展现出一致的提升，在 125 个条件中改善了 120 个，平均准确性提升 1.44 个百分点，平均恢复了 26.9% 的准确性差距。这些结果表明，所提出的无真实标签自适应为改善部署中的 OmniLLMs 的准确性与效率权衡提供了一条有效且实用的途径。
cs.CV / 162 / 2609.39960
Reconstructing the Dynamic World: A Representation-Centric View of 4D Scene Reconstruction
重建动态世界：基于表征的四维场景重建视角
Abstract
4D scene reconstruction aims to recover the evolving geometry, appearance, and motion of dynamic environments from visual observations. Despite substantial progress in neural scene representations, reconstructing dynamic scenes remains challenging due to non-rigid motion, occlusions, temporal inconsistencies, and the trade-offs between reconstruction fidelity and computational efficiency. Recent advances in Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have introduced diverse approaches to representing and reconstructing dynamic scenes, yet their relationships, underlying design choices, and evaluation protocols remain fragmented. In this paper, we present a unified perspective on 4D scene reconstruction, organizing existing methods around their scene representations, temporal modeling strategies, reconstruction pipelines, and optimization objectives. Through this framework, we examine how different design choices affect geometric fidelity, appearance consistency, motion representation, and computational efficiency. We further consolidate commonly used datasets and evaluation metrics, identify limitations in current experimental practices, and discuss open challenges in reconstructing complex, dynamic real-world environments. By connecting methodological developments with their underlying assumptions and evaluation evidence, this work provides a structured foundation for understanding existing approaches and identifying future research directions. An evolving collection of relevant papers and resources is available at https://github.com/ZiyangYan/Awesome-4D-Scene-Reconstruction.
Chinese Translation
四维场景重建旨在从视觉观测中恢复动态环境的演变几何、外观和运动。尽管在神经场景表征方面取得了显著进展，但由于非刚性运动、遮挡、时间不一致性以及重建保真度与计算效率之间的权衡，重建动态场景仍然具有挑战性。最近在神经辐射场（NeRF）和三维高斯点云（3DGS）方面的进展引入了多种表示和重建动态场景的方法，但它们之间的关系、基础设计选择和评估协议仍然分散。在本文中，我们提出了一个统一的四维场景重建视角，将现有方法围绕其场景表征、时间建模策略、重建流程和优化目标进行组织。通过这一框架，我们考察了不同设计选择如何影响几何保真度、外观一致性、运动表征和计算效率。我们进一步整合了常用的数据集和评估指标，识别当前实验实践中的局限性，并讨论重建复杂动态真实环境的开放挑战。通过将方法论发展与其基础假设和评估证据相连接，本研究为理解现有方法和识别未来研究方向提供了结构化的基础。相关论文和资源的不断更新集合可在 https://github.com/ZiyangYan/Awesome-4D-Scene-Reconstruction 获取。
cs.CV / 163 / 2609.40014
Can We Anticipate Violence? Multimodal Learning from Pre-Incident Behavioral Cues
我们能预测暴力吗？基于多模态学习的事件前行为线索
Abstract
Detecting violence after it begins is important from recognizing behavioral cues that appear immediately beforehand. This work studies short-horizon pre-incident risk recognition from multimodal video signals. We construct a binary Normal-versus-Risky setting from temporally annotated XD-Violence clips, using 443 samples with source-level separation across training, validation, and test sets. Each sample consists of a variable-length pre-incident clip, with its duration determined by the observable behavioral context preceding the incident. The inci- dent itself is excluded from all input clips. We evaluate three complementary information sources: facial-region appearance, temporally aligned audio, and body-motion features derived from tracked keypoints. Controlled ablations are performed with Swin-Tiny, ViT-Tiny, and DeiT-Tiny to measure the contribution of each modality under the same split. Results show that combining all modalities is more effective than using any other combination alone. The best configuration, Deit-Tiny with audio, facial appearance, and motion, achieves 91.21% accuracy, 88.96% balanced accuracy, 93.65% F1-score, and 96.38% ROC-AUC on the held-out test set. These results suggest that complementary appearance, acoustic, and kinematic cues provide useful evidence for recognizing elevated pre-incident risk.
Chinese Translation
在暴力事件发生后进行检测固然重要，但识别事件发生前立即出现的行为线索同样关键。本研究探讨了基于多模态视频信号的短期事件前风险识别。我们从时间标注的XD-Violence片段中构建了一个二元的正常与风险设置，使用了443个样本，并在训练、验证和测试集之间实现了源级分离。每个样本由一个可变长度的事件前片段组成，其持续时间由事件发生前可观察到的行为上下文决定。事件本身不包含在所有输入片段中。我们评估了三种互补的信息源：面部区域外观、时间对齐的音频和从跟踪关键点中提取的身体运动特征。使用Swin-Tiny、ViT-Tiny和DeiT-Tiny进行控制性消融实验，以测量在相同划分下每种模态的贡献。结果表明，结合所有模态的效果优于单独使用任何其他组合。最佳配置为Deit-Tiny，结合音频、面部外观和运动，达到了91.21%的准确率、88.96%的平衡准确率、93.65%的F1-score和96.38%的ROC-AUC在保留的测试集上。这些结果表明，互补的外观、声学和运动线索为识别事件前的高风险提供了有用的证据。
cs.CV / 164 / 2609.40031
WARP: A Unified Benchmark for Invisible Image Watermarking -- Robustness and Protection Against Attacks
WARP：一个统一的隐形图像水印基准测试——对抗攻击的鲁棒性与保护
Abstract
Digital image watermarking is increasingly critical in media contexts, as emerging regulations and industry practices require marking AI-generated content and ensuring traceable sources to prevent manipulation or misuse. Recent advances in invisible watermarking methods highlight the need to update existing benchmarking practices to reflect current techniques and evaluation criteria. We address this by introducing WARP -- a unified framework and benchmark for evaluating the robustness of invisible watermarks. WARP incorporates 32 recent classical, deep, and generative watermarking methods, as well as 34 different erasing techniques, ranging from traditional distortions to more sophisticated adversarial, purification, and re-embedding attacks. It provides standardized, reproducible, and easily scalable protocols for evaluating perceptual quality, watermark readability, and attack resilience. Using WARP, we extensively evaluate current invisible watermarking techniques, collecting the largest robustness benchmark in the field. Results identify the most robust approaches under both distortion and adversarial conditions, and reveal consistent relationships between watermarking methods and the attack strategies most effective against them. Our experiments also highlight that some of the watermarking methods considered are highly vulnerable to reembedding, even if they are robust to standard distortions. The code is made available at https://github.com/ispras/wibe.
Chinese Translation
数字图像水印在媒体环境中变得越来越重要，因为新兴的法规和行业实践要求对人工智能生成的内容进行标记，并确保可追溯的来源以防止操纵或滥用。最近隐形水印方法的进展突显了更新现有基准测试实践的必要性，以反映当前的技术和评估标准。为此，我们提出了WARP——一个用于评估隐形水印鲁棒性的统一框架和基准测试。WARP包含32种最新的经典、水深和生成水印方法，以及34种不同的抹除技术，涵盖从传统失真到更复杂的对抗性、净化和重新嵌入攻击。它提供标准化、可重复和易于扩展的协议，用于评估感知质量、水印可读性和抗攻击能力。通过使用WARP，我们广泛评估了当前的隐形水印技术，收集了该领域最大的鲁棒性基准测试。结果识别出在失真和对抗条件下最鲁棒的方法，并揭示水印方法与对其最有效攻击策略之间的一致关系。我们的实验还强调，某些考虑的水印方法在重新嵌入方面高度脆弱，即使它们对标准失真具有鲁棒性。代码可在 https://github.com/ispras/wibe 获取。
cs.CV / 165 / 2609.40037
Enhancing Autoregressive Video Generation via Representation Adversarial Distillation
通过表征对抗蒸馏增强自回归视频生成
Abstract
Few-step autoregressive video generation enables efficient streaming synthesis, but errors introduced in early temporal blocks are reused as context and can propagate through subsequent rollouts, leading to detail degradation, structural drift, and unstable motion. Existing distribution matching distillation (DMD) primarily aligns student and teacher distributions in diffusion latent space, but provides no direct supervision over the perceptual quality of decoded videos. We introduce Radian, a representation-space adversarial distillation framework that complements on-policy DMD with real-data adversarial supervision in the feature space defined by a frozen visual foundation model (VFM). During training, Radian sparsely decodes frames from autoregressive student rollouts, extracts multi-level visual representations, and applies lightweight discriminator heads to distinguish generated outputs from real video frames. The DMD objective anchors the student to the pretrained teacher, while the representation-space adversarial objective supplies complementary perceptual and semantic gradients that promote high-quality modes. These additional components are discarded after training, leaving the generator architecture and inference-time denoising budget unchanged. Experiments on Wan2.1-1.3B cover four-step chunk-wise, one-step frame-wise, and minute-long autoregressive generation. Our method achieves a VBench Total of 0.8444 and a VideoAlign Total of 0.8033 under four-step generation, and improves VBench-Long from 0.7805 to 0.8041 over Rolling Forcing while using fewer denoising steps. Controlled comparisons across image, video, and diffusion representations further indicate that the choice of representation spaces induces distinct adversarial signals, and external VFM gradients complement DMD more effectively than adversarial supervision derived from diffusion-internal features.
Chinese Translation
少步自回归视频生成能够实现高效的流媒体合成，但在早期时间块中引入的错误作为上下文被重用，并可能在后续的展开中传播，导致细节退化、结构漂移和不稳定运动。现有的分布匹配蒸馏（DMD）主要在扩散潜在空间中对学生和教师分布进行对齐，但未对解码视频的感知质量提供直接监督。我们提出了Radian，一个表征空间对抗蒸馏框架，它通过冻结的视觉基础模型（VFM）定义的特征空间中的真实数据对抗监督，补充了政策内DMD。在训练过程中，Radian从自回归学生展开中稀疏解码帧，提取多层次视觉表征，并应用轻量级判别头来区分生成的输出和真实视频帧。DMD目标将学生锚定到预训练的教师，而表征空间对抗目标提供互补的感知和语义梯度，以促进高质量模式。这些额外组件在训练后被丢弃，保持生成器架构和推理时去噪预算不变。我们在Wan2.1-1.3B上的实验涵盖了四步块状、一帧一步和分钟级自回归生成。我们的方法在四步生成下实现了VBench总分0.8444和VideoAlign总分0.8033，并在使用更少去噪步骤的情况下将VBench-Long从0.7805提高到0.8041。对图像、视频和扩散表征的控制比较进一步表明，表征空间的选择会引发不同的对抗信号，外部VFM梯度比来自扩散内部特征的对抗监督更有效地补充DMD。
cs.CV / 166 / 2609.40048
CoEvoWhen: Policy-Tool Coevolution for Ultra-Long Video Temporal Grounding
CoEvoWhen：超长视频时间定位的策略-工具共演化
Abstract
Ultra-long video temporal grounding requires balancing long-range evidence search with fine-grained event understanding under a limited visual budget, yet existing agentic methods still rely largely on predefined policies and tool capabilities. Motivated by this, we propose a novel policy-tool coevolution framework that jointly evolves high-level policies and executable media tools from the agentic reasoning trajectories of a VLM, forming a reusable skill without updating model parameters. During evolution, an external skill updater distills transferable task experience in long-video temporal grounding, accordingly refining the orchestration of long-range image-based and fine-grained video-based observations. Alongside these policy updates, the updater employs its coding capabilities to upgrade existing tools or create new ones, adapting the tools to long-video evidence acquisition. Equipped with the evolved skill, the VLM autonomously orchestrates tools under the guidance of the evolved policy, coordinating image and video observations for agentic inference without relying on a separate, stronger planning model. Extensive experiments spanning five benchmarks and three VLMs show that policy-tool coevolution consistently improves temporal grounding accuracy in ultra-long videos while reducing visual token cost at inference, and that the evolved skill yields substantial performance gains on general long-video QA without additional task-specific evolution, demonstrating the effectiveness and generalizability of our framework for long-video understanding.
Chinese Translation
超长视频时间定位需要在有限的视觉预算下平衡长距离证据搜索与细粒度事件理解，然而现有的智能方法仍然在很大程度上依赖于预定义的策略和工具能力。基于此，我们提出了一种新颖的策略-工具共演化框架，该框架从视觉语言模型（VLM）的智能推理轨迹中共同演化高层策略和可执行的媒体工具，形成一种可重用的技能，而无需更新模型参数。在演化过程中，外部技能更新器提炼在超长视频时间定位中的可转移任务经验，从而相应地优化基于长距离图像和细粒度视频观察的协调。除了这些策略更新，更新器还利用其编码能力来升级现有工具或创建新工具，使工具适应超长视频证据获取。凭借演化后的技能，VLM在演化策略的指导下自主协调工具，整合图像和视频观察进行智能推理，而无需依赖单独的、更强大的规划模型。跨越五个基准和三个VLM的广泛实验表明，策略-工具共演化在超长视频中的时间定位准确性始终得到提升，同时在推理时减少视觉标记成本，并且演化后的技能在一般的长视频问答中带来了显著的性能提升，而无需额外的任务特定演化，证明了我们框架在长视频理解中的有效性和普适性。
cs.CV / 167 / 2609.40055
Less Data, Better Timing: Student-Curriculum Coupling for VLM On-Policy Distillation in Temporal Video Grounding
更少的数据，更好的时机：针对时序视频定位的学生-课程耦合用于视觉语言模型的在线蒸馏
Abstract
On-policy distillation (OPD) provides dense supervision directly on student-generated trajectories, making it an effective post-training strategy for vision-language models in temporal video grounding (TVG). However, existing pipelines typically construct the training curriculum from a fixed teacher and the initial student state, implicitly assuming that selected examples retain positive supervision value throughout optimization. We show that supervision trustworthiness and supervision necessity are distinct yet coupled: the former concerns target credibility, while the latter varies with the student's current task competence; together, they shape supervision value. Building on this coupled view, we introduce Student-Curriculum Coupling (SCC), a closed-loop framework in which a compact Anchor-Frontier curriculum defines the candidate supervision space and the evolving student dynamically determines its active subset. Supervision can therefore be activated, suspended, or reactivated as competence changes, concentrating teacher computation and optimization on current task-level deficits. Across three TVG benchmarks, SCC achieves a 5.1% relative improvement in mean recall over Video-OPD on its original curriculum, while using 60.0% fewer training examples and reducing training time by 50.4%. Ablations support the complementary roles of capability-structured curriculum design and student-dependent supervision in achieving these gains. Together, these results establish SCC as a data- and compute-efficient framework for TVG post-training, delivering stronger temporal grounding by aligning trustworthy supervision with the student's evolving learning needs.
Chinese Translation
在线蒸馏（On-policy distillation, OPD）直接在学生生成的轨迹上提供密集监督，使其成为时序视频定位（Temporal Video Grounding, TVG）中视觉语言模型的有效后训练策略。然而，现有的流程通常从固定的教师和初始学生状态构建训练课程，隐含地假设所选示例在优化过程中始终保持正向监督价值。我们表明，监督的可信度和监督的必要性是不同但耦合的：前者涉及目标的可信性，而后者则随着学生当前任务能力的变化而变化；二者共同塑造了监督价值。在这一耦合视角的基础上，我们引入了学生-课程耦合（Student-Curriculum Coupling, SCC），这是一个闭环框架，其中紧凑的锚点-前沿课程定义了候选监督空间，而不断发展的学生动态确定其活动子集。因此，监督可以根据能力的变化被激活、暂停或重新激活，集中教师的计算和优化于当前任务级别的不足。在三个TVG基准测试中，SCC在其原始课程上相较于Video-OPD实现了5.1%的平均召回率相对提升，同时使用了60.0%更少的训练示例，并将训练时间减少了50.4%。消融实验支持能力结构化课程设计和学生依赖监督在实现这些增益中的互补作用。综合来看，这些结果确立了SCC作为一个数据和计算高效的TVG后训练框架，通过将可信的监督与学生不断变化的学习需求对齐，从而提供更强的时序定位能力。
cs.CV / 168 / 2609.40079
LongEmo: Towards Emotion Understanding and Reasoning in Long Videos
LongEmo：面向长视频中的情感理解与推理
Abstract
While recent Multimodal Large Language Models (MLLMs) have shown promise in affective computing, their reasoning capabilities are largely confined to short video clips with limited interactions. However, real-world emotions are not merely isolated instantaneous reactions but dynamic and cumulative processes deeply shaped by past experiences and ongoing events. To bridge this gap, we introduce LongEmoBench, a benchmark dedicated to emotion understanding and reasoning in long videos. It assesses progressive capabilities scaling from continuous scene interactions to complex episodic developments. Furthermore, we propose LongEmo, a novel memory-augmented agentic framework designed to tackle the immense challenges of long-range affective reasoning. LongEmo processes continuous video streams to construct an Event Memory Graph, explicitly modeling long-range dependencies and capturing emotional dynamics across discrete events. Given a question, the agent retrieves a query-relevant event stream from the graph, iteratively integrating multimodal memories and relational dependencies to deduce the final answer. Extensive evaluations of 17 representative methods reveal that they struggle significantly with emotion understanding and reasoning in long videos. In contrast, LongEmo achieves state-of-the-art performance, demonstrating the efficacy of its event-centric memory architecture.
Chinese Translation
尽管近期的多模态大型语言模型（MLLMs）在情感计算方面展现出潜力，但其推理能力主要局限于短视频片段和有限的交互。然而，现实世界中的情感并非孤立的瞬时反应，而是深受过去经历和正在发生事件影响的动态累积过程。为了解决这一问题，我们引入了LongEmoBench，这是一个专注于长视频中情感理解与推理的基准测试。该基准评估从连续场景交互到复杂情节发展的逐步能力。此外，我们提出了LongEmo，一个新颖的增强记忆的代理框架，旨在应对长程情感推理的巨大挑战。LongEmo处理连续的视频流，构建事件记忆图，明确建模长程依赖关系并捕捉离散事件中的情感动态。在给定问题的情况下，代理从图中检索与查询相关的事件流，迭代整合多模态记忆和关系依赖，以推导出最终答案。对17种代表性方法的广泛评估表明，它们在长视频中的情感理解与推理方面面临显著挑战。相比之下，LongEmo实现了最先进的性能，展示了其以事件为中心的记忆架构的有效性。
cs.CV / 169 / 2609.40091
GateSPINE: Gated Cross-View Fusion for Lumbar Spine MRI Report Generation
GateSPINE：用于腰椎MRI报告生成的门控交叉视图融合
Abstract
Automated report generation can ease the burden radiolo gists face when interpreting multi-sequence MRI studies. Unlike CT, MRI examinations comprise multiple sequences and imaging planes, each con tributing complementary diagnostic information. Existing methods en code a study as a single volume and combine multiple acquisitions by fixed rules. Findings visible in only one plane are thus diluted and of ten missed, lowering recall on clinical efficacy metrics, where a missed abnormality is most costly. We propose GateSPINE, a vision-language framework that fuses sagittal T1 and T2 volumes with a training-free operator, encodes the fused sagittal and axial volumes with two parallel 3D encoders, and decodes their combined representation into a report. Its core mechanism is a gated cross view fusion module that predicts, per feature channel and token, how much of each view to admit, so the more informative view dominates at each spatial location. We evaluate GateSPINE on three lumbar MRI datasets, comprising two public bench marks and a private cohort collected from Phenikaa University Hospital, using both natural language generation (NLG) and clinical efficacy (CE) metrics. GateSPINE achieves the highest CE F1 through improved re call on all three datasets; on SPIDER, which lacks an axial sequence, this reflects the sagittal fusion component rather than the gated cross-view mechanism, which is validated on the two cohorts with both imaging planes. GateSPINE also remains competitive on standard NLG metrics.
Chinese Translation
自动化报告生成可以减轻放射科医生在解读多序列MRI研究时所面临的负担。与CT不同，MRI检查包含多个序列和成像平面，每个序列提供互补的诊断信息。现有方法将研究编码为单一体积，并通过固定规则组合多个采集。因此，仅在一个平面上可见的发现往往被稀释并且常常被遗漏，从而降低了临床有效性指标的召回率，而漏掉的异常是最为昂贵的。我们提出了GateSPINE，这是一种视觉-语言框架，融合了矢状面T1和T2体积，通过无训练的操作符编码融合后的矢状面和轴面体积，并将其组合表示解码为报告。其核心机制是一个门控交叉视图融合模块，该模块预测每个特征通道和标记在每个空间位置应接纳多少来自每个视图的信息，从而使得更具信息量的视图在每个空间位置占主导地位。我们在三个腰椎MRI数据集上评估GateSPINE，包括两个公共基准和从Phenikaa大学医院收集的一个私有队列，使用自然语言生成（NLG）和临床有效性（CE）指标。GateSPINE通过提高所有三个数据集的召回率，达到了最高的CE F1；在缺乏轴面序列的SPIDER数据集中，这反映了矢状面融合组件的效果，而非门控交叉视图机制，这在具有两个成像平面的两个队列中得到了验证。GateSPINE在标准NLG指标上也保持了竞争力。
cs.CV / 170 / 2609.40129
VR-JEPA: Learning Contrastive-State Latent Guidance for Generation-based Video Reasoning
VR-JEPA：基于生成的视频推理中的对比状态潜在引导学习
Abstract
Reasoning through video generation offers a promising path toward visual intelligence by modeling latent visual states and their dynamics. However, current video generation models often lack explicit guidance on how these states should evolve, leaving generated trajectories prone to physical and structural inconsistencies that undermine reasoning reliability. While the Video Joint-Embedding Predictive Architecture (V-JEPA) provides rich spatiotemporal priors learned through latent prediction, these general priors do not naturally adapt to the logical reasoning capabilities required for complex visual tasks. To bridge this gap, we propose VR-JEPA, a framework that aligns the V-JEPA predictor with task-specific reasoning logic through localized contrastive-state learning and uses its predicted latent trajectories to guide video generation for visual reasoning. Specifically, (i) we pair successful trajectories with generated alternatives under the same input conditions and use discrepancies in their V-JEPA representations to identify informative states and tokens for localized contrastive supervision. (ii) We further equip the V-JEPA predictor with skill-specific experts trained on anchor-task data, allowing the model to adaptively specialize its shared spatiotemporal priors across diverse cognitive domains. Together with skill-specific experts, this contrastive supervision enables VR-JEPA to predict latent trajectories that provide task-specific logical guidance for video generation. Comprehensive experiments on the large-scale VBVR-Pro-Bench dataset demonstrate that VR-JEPA achieves an $11.33\%$ relative improvement over the cutting-edge generation-based reasoning baseline, significantly mitigating physical artifacts and enhancing logical consistency.
Chinese Translation
通过视频生成进行推理为实现视觉智能提供了一条有前景的路径，旨在建模潜在视觉状态及其动态变化。然而，目前的视频生成模型往往缺乏关于这些状态如何演变的明确指导，导致生成的轨迹容易出现物理和结构上的不一致，从而削弱推理的可靠性。虽然视频联合嵌入预测架构（Video Joint-Embedding Predictive Architecture，V-JEPA）通过潜在预测学习了丰富的时空先验，但这些一般性先验并不能自然适应复杂视觉任务所需的逻辑推理能力。为了解决这一问题，我们提出了VR-JEPA，一个通过局部对比状态学习将V-JEPA预测器与任务特定推理逻辑对齐的框架，并利用其预测的潜在轨迹指导视频生成以实现视觉推理。具体而言，(i) 我们将成功的轨迹与在相同输入条件下生成的替代轨迹配对，并利用它们在V-JEPA表示中的差异来识别用于局部对比监督的信息状态和标记。(ii) 我们进一步为V-JEPA预测器配备了在锚定任务数据上训练的特定技能专家，使模型能够在不同的认知领域中自适应地专门化其共享的时空先验。结合特定技能专家，这种对比监督使VR-JEPA能够预测提供任务特定逻辑指导的潜在轨迹。对大规模VBVR-Pro-Bench数据集的全面实验表明，VR-JEPA在基于生成的推理基线中实现了$11.33\%$的相对提升，显著减轻了物理伪影并增强了逻辑一致性。
cs.CV / 171 / 2609.40195
MemLife: Curating and Reasoning over Long-Term Egocentric Video Memories
MemLife：对长期自我中心视频记忆的策划与推理
Abstract
Long-term egocentric video enables personalized AI assistants to reason about daily life. However, as video histories grow to hundreds of hours spanning months or years, reprocessing raw clips for every query becomes computationally prohibitive. Memory systems offer a scalable alternative by compacting videos into text representations, but often fail on practical benchmarks: either the memory does not preserve key evidence, or the retriever fails to locate relevant entries due to retrieval competition in growing search spaces. To address these challenges, we introduce MemLife, a multimodal memory system that constructs entity-grounded, first-person text episodes and retrieves them via a time-indexed agentic reader. Without training or query-time video access, MemLife improves over the strongest training-free baseline by 4.6--12.0% across four long-horizon benchmarks. To further improve memory quality, we propose MemOpt, a reinforcement learning framework that optimizes the memory writer to produce faithful, informative, and retrievable memories. MemOpt consistently improves MemLife by 2.7--5.0% across different video and question distributions, with gains that generalize across writer and reader backbones and memory systems.
Chinese Translation
长期自我中心视频使个性化人工智能助手能够推理日常生活。然而，随着视频历史增长到数百小时，跨越数月或数年的时间，针对每个查询重新处理原始片段变得计算上不可行。记忆系统通过将视频压缩为文本表示提供了一种可扩展的替代方案，但在实际基准测试中往往表现不佳：要么记忆未能保留关键证据，要么检索器由于在不断扩大的搜索空间中竞争而无法找到相关条目。为了解决这些挑战，我们提出了MemLife，这是一种多模态记忆系统，构建基于实体的第一人称文本片段，并通过时间索引的代理阅读器进行检索。在没有训练或查询时视频访问的情况下，MemLife在四个长期基准测试中比最强的无训练基线提高了4.6%至12.0%。为了进一步提高记忆质量，我们提出了MemOpt，这是一种强化学习框架，优化记忆写入器以生成真实、信息丰富且可检索的记忆。MemOpt在不同视频和问题分布中始终将MemLife的性能提高了2.7%至5.0%，其增益在不同的写入器和阅读器骨干网络及记忆系统中具有广泛的适用性。
cs.CV / 172 / 2609.40212
Recognition of Urbanized Areas in UAV-Derived Very-High-Resolution Visible-Light Imagery
基于无人机获取的超高分辨率可见光影像的城市化区域识别
Abstract
This study compared classifiers that differentiate between urbanized and non-urbanized areas based on unmanned aerial vehicle (UAV)-acquired RGB imagery. The tested solutions in-cluded numerous vegetation indices (VIs) thresholding and neural networks (NNs). The analysis was conducted for two study areas for which surveys were carried out using different UAVs and cameras. The ground sampling distances for the study areas were 10 mm and 15 mm, respectively. Reference classification was performed manually, obtaining approximately 24 million classified pix-els for the first area and approximately 3.8 million for the second. This research study included an analysis of the impact of the season on the threshold values for the tested VIs and the impact of image patch size provided as inputs for the NNs on classification accuracy. The results of the con-ducted research study indicate a higher classification accuracy using NNs (about 96%) compared with the best of the tested VIs, i.e., Excess Blue (about 87%). Due to the highly imbalanced nature of the used datasets (non-urbanized areas constitute approximately 87% of the total datasets), the Mat-thews correlation coefficient was also used to assess the correctness of the classification. The analysis based on statistical measures was supplemented with a qualitative assessment of the classification results, which allowed the identification of the most important sources of differences in classification between VIs thresholding and NNs.
Chinese Translation
本研究比较了基于无人机（UAV）获取的RGB影像区分城市化和非城市化区域的分类器。测试的解决方案包括多种植被指数（VIs）阈值化和神经网络（NNs）。分析针对两个研究区域进行，这些区域使用不同的无人机和相机进行了调查。研究区域的地面采样距离分别为10毫米和15毫米。参考分类是手动进行的，第一区域获得了约2400万个分类像素，第二区域约为380万个。本研究还分析了季节对测试的VIs阈值的影响，以及作为NNs输入的图像块大小对分类准确性的影响。研究结果表明，使用NNs的分类准确性更高（约96%），而测试的最佳VIs，即过量蓝色（Excess Blue）的准确性约为87%。由于使用的数据集具有高度不平衡的特性（非城市化区域约占总数据集的87%），因此还使用马修斯相关系数来评估分类的正确性。基于统计指标的分析还辅以分类结果的定性评估，从而识别出VIs阈值化和NNs分类之间差异的最重要来源。
cs.CV / 173 / 2609.40219
Learning Skills from Historical Action Trajectories: Action Experience Dictionary for World Action Models
从历史动作轨迹中学习技能：用于世界动作模型的动作经验词典
Abstract
World Action Models (WAMs) couple visual dynamics prediction with action generation, yet they do not explicitly support the reuse of action experience across manipulation tasks. Furthermore, existing WAMs struggle to capture underlying cross-task semantic relationships that could guide target action prediction, as redundant background elements interfere with the extraction of key visual information. To address these challenges, we develop a novel Action Experience Dictionary (AED) that encodes historical physical action trajectories into shared action embeddings to support skill reuse and model cross-task relationships. Specifically, we first aggregate historical actions to align with visual observations and retrieve action embeddings from the AED using a pretrained action tokenizer. Subsequently, we visually condition the pooled embeddings through cross-attention and prepend them to noisy action tokens, providing interaction context and action intent for prediction. To model action-related motion and reduce reliance on irrelevant background cues, we introduce a motion-aware transition loss that supervises visual feature change prediction over random temporal intervals. Experiments on simulation benchmarks and in real-world cross-embodiment settings verify the effectiveness of our AED. The anonymous project website is available at \href{https://github.com/JiahuaDong/AED}{AED}.
Chinese Translation
世界动作模型（World Action Models, WAMs）将视觉动态预测与动作生成相结合，但它们并未明确支持在操作任务中重用动作经验。此外，现有的WAMs在捕捉潜在的跨任务语义关系方面存在困难，这些关系可以指导目标动作预测，因为冗余的背景元素干扰了关键信息的提取。为了解决这些挑战，我们开发了一种新颖的动作经验词典（Action Experience Dictionary, AED），该词典将历史物理动作轨迹编码为共享的动作嵌入，以支持技能重用和模型跨任务关系的建立。具体而言，我们首先聚合历史动作，以与视觉观察对齐，并使用预训练的动作标记器从AED中检索动作嵌入。随后，我们通过交叉注意机制对汇聚的嵌入进行视觉条件处理，并将其预置于噪声动作标记之前，为预测提供交互上下文和动作意图。为了建模与动作相关的运动并减少对无关背景线索的依赖，我们引入了一种运动感知过渡损失，监督随机时间间隔内的视觉特征变化预测。在模拟基准和真实世界跨体现设置上的实验验证了我们AED的有效性。匿名项目网站可访问 exttt{https://github.com/JiahuaDong/AED}。
cs.CV / 174 / 2609.40222
LOCI: Spatial Linear Memory for Streaming World Models
LOCI：用于流式世界模型的空间线性记忆
Abstract
When a camera revisits a previously observed region, a video world model should reproduce what was there before. This requires both remembering past observations and retrieving the right one for the current viewpoint. Key-value caches preserve visual detail but grow with video length; recurrent memory is compact but compresses history into a fixed-size state, so individual past observations are no longer directly accessible. We introduce LOCI, a hybrid spatial-memory architecture that keeps both representations. In half of the transformer blocks, main attention keeps a key-value cache of past observations; in the other half, it is restricted to the current chunk and complemented by a recurrent linear-attention memory whose reads and writes are conditioned on projective camera geometry, so viewpoint enters both memory addressing and stored content. Recurrent readouts flow into subsequent cache-backed blocks and supply their queries with accumulated scene context. On the public MIND memory benchmark and on held-out recorded trajectories, LOCI reproduces revisited content more faithfully than representative world models and a same-recipe full-softmax model; with full history, it lowers peak memory at equal length by about 30% relative to full softmax. With a bounded bank of retained observations, it streams long videos at constant memory and remains more faithful than full softmax under the same budget.
Chinese Translation
当相机重新访问之前观察过的区域时，视频世界模型应当再现之前的内容。这需要既能记住过去的观察，又能为当前视角检索到正确的内容。键值缓存能够保留视觉细节，但随着视频长度的增加而增长；递归记忆则较为紧凑，但将历史压缩为固定大小的状态，因此个别的过去观察不再直接可访问。我们提出了LOCI，一种混合空间记忆架构，保留了这两种表示。在一半的变换器块中，主要注意力保持过去观察的键值缓存；在另一半中，它仅限于当前块，并辅以递归线性注意力记忆，其读取和写入受投影相机几何的条件限制，因此视角同时影响记忆寻址和存储内容。递归读取流入后续的缓存支持块，并为其查询提供累积的场景上下文。在公共的MIND记忆基准测试和保留的记录轨迹上，LOCI比代表性的世界模型和相同配方的全软最大模型更忠实地再现了重新访问的内容；在完整历史下，相较于全软最大模型，它在相同长度下降低了约30%的峰值内存。通过保留观察的有限库，它以恒定内存流式传输长视频，并在相同预算下比全软最大模型更为忠实。
cs.CV / 175 / 2609.40230
EviRover: Reinforcing Agentic Perception Beyond a Glance
EviRover：超越一瞥的代理感知强化
Abstract
Visual perception is conventionally formulated as a one-shot prediction from a single glance at the image, under the assumption that the image content and the model's parametric knowledge suffice to resolve the query. This assumption often fails in real-world scenarios that hinge on fine-grained visual details or require knowledge-intensive and up-to-date information. We term such cases \textit{perception under insufficient evidence} and formulate perception as an agentic process that can obtain information beyond a single glance. To address the absence of data for this setting, we design two dedicated data generation pipelines, yielding EviRover-SFT-5K and EviRover-RL-12K for training. We further construct EviLens, a human-verified benchmark comprising 688 instances across five perception categories. Building on these data, we present EviRover, to our knowledge the first perception agent explicitly trained to resolve perceptual queries through interaction, using supervised fine-tuning followed by agentic reinforcement learning. Experiments show that the 4B EviRover outperforms its backbone by 30 points on average on EviLens, reaching performance comparable to advanced proprietary models. The gains transfer beyond EviLens to WebEyes, conventional perception benchmarks, and general multimodal benchmarks, including a 15-point improvement on BrowseComp-VL. All code, models, and data are released.
Chinese Translation
视觉感知通常被视为从图像的一次性观察中进行的单次预测，假设图像内容和模型的参数知识足以解决查询。然而，这一假设在依赖细致视觉细节或需要知识密集且最新信息的现实场景中常常失效。我们将这种情况称为“在证据不足下的感知”，并将感知形式化为一种能够获取超出单次观察的信息的代理过程。为了解决这一情境下数据缺乏的问题，我们设计了两个专门的数据生成管道，分别生成了用于训练的 EviRover-SFT-5K 和 EviRover-RL-12K。此外，我们构建了 EviLens，这是一个经过人工验证的基准，包含五个感知类别的688个实例。在这些数据的基础上，我们提出了 EviRover，据我们所知，这是第一个明确通过交互来解决感知查询的感知代理，采用监督微调后进行代理强化学习。实验表明，4B 的 EviRover 在 EviLens 上平均比其基础模型提高了30分，达到了与先进专有模型相当的性能。这一提升不仅在 EviLens 上有效，还转移到 WebEyes、传统感知基准以及包括 BrowseComp-VL 在内的一般多模态基准上，后者提高了15分。所有代码、模型和数据均已发布。
cs.CV / 176 / 2609.40244
StreamRig: Exploiting Intra-Rig Geometry for Streaming Multi-Camera Odometry
StreamRig：利用内部设备几何结构进行多摄像头里程计流处理
Abstract
Mobile robots and vehicles carry synchronized multi-camera rigs, yet many streaming 3D foundation models are designed for monocular input, leaving efficient use of rig geometry a challenge. We present StreamRig, a freeze-and-stream framework that builds causal streaming odometry for calibrated rigs on a frozen multi-view 3D foundation model. The frozen front-end jointly perceives the synchronized views using rig calibration. A Rig-Resampler compresses their features, a CausalBridge applies causal attention with a key-value cache, and a lightweight head regresses rig poses. A periodic re-anchoring protocol supports stable pose estimation over long sequences. Only these modules are trained, 74.6M parameters in total, with relative poses as the sole supervision. Our two-stage training strategy combines group relocalization pretraining with causal rig training to transfer the geometric priors of the frozen front-end and the alignment ability of the pretrained modules to streaming odometry. We evaluate on NCLT, TartanGround, KITTI-360, and our self-collected humanoid-robot dataset ZJH, where training uses only simulation and real-world evaluation is zero-shot. Across all four datasets, StreamRig achieves lower translation and rotation drift than the evaluated non-oracle monocular streaming and rig-aware offline models, while maintaining low inference cost. Ablations and controlled camera-count experiments identify the sources of these gains. We further examine how longer training windows affect inference over longer horizons. Code has been released at https://github.com/WeiYuFei0217/StreamRig.
Chinese Translation
移动机器人和车辆携带同步的多摄像头设备，但许多流式3D基础模型是为单目输入设计的，这使得高效利用设备几何结构成为一项挑战。我们提出了StreamRig，一种冻结与流处理框架，基于冻结的多视角3D基础模型构建校准设备的因果流式里程计。冻结的前端通过设备校准共同感知同步视图。一个设备重采样器压缩它们的特征，一个因果桥接器应用带有键值缓存的因果注意力，而一个轻量级头部回归设备姿态。一个周期性重新锚定协议支持在长序列上的稳定姿态估计。仅这些模块被训练，总共有7460万个参数，相对姿态作为唯一的监督。我们的两阶段训练策略结合了组重定位预训练与因果设备训练，以将冻结前端的几何先验和预训练模块的对齐能力转移到流式里程计上。我们在NCLT、TartanGround、KITTI-360以及我们自收集的人形机器人数据集ZJH上进行了评估，其中训练仅使用模拟，真实世界评估为零样本。在所有四个数据集中，StreamRig实现了比评估的非oracle单目流式和设备感知离线模型更低的平移和旋转漂移，同时保持低推理成本。消融实验和受控摄像头数量实验识别了这些增益的来源。我们进一步研究了更长的训练窗口如何影响长时间范围内的推理。代码已发布在https://github.com/WeiYuFei0217/StreamRig。
cs.CV / 177 / 2609.40253
ComputerSD: Online Self-Distillation from Real-Time Feedback for Computer-Use Agents
ComputerSD：基于实时反馈的计算机使用代理在线自蒸馏
Abstract
Online training enables computer-use agents (CUAs) to improve through interaction with executable environments. However, existing methods primarily rely on sparse outcome rewards, which provide no supervision for intermediate actions. On-policy self-distillation (OPSD) offers token-level learning signals through privileged rescoring, but directly applying it to CUA online training presents two challenges: fixed guidance may become misaligned with the student's current state, and guidance-induced probability shifts may conflict with step-level correctness. We introduce ComputerSD, an online self-distillation method for CUAs that converts real-time feedback from executed GUI transitions into guidance for policy learning. A fine-tuned GUI analyzer produces guidance and a step-level value score after each action; the guidance provides privileged context, while the score regulates the resulting OPSD signals. ComputerSD jointly optimizes token-level OPSD and trajectory-level GRPO in a fully asynchronous training framework. On OSWorld-Verified, ComputerSD outperforms outcome-only GRPO by 1.9 and 4.1 percentage points on the general-purpose Qwen3-VL-8B-Thinking and specialized EvoCUA-8B backbones, respectively. Evaluation in out-of-distribution settings further supports the generalizability of ComputerSD. These results demonstrate the effectiveness of learning from real-time feedback through online self-distillation for CUAs.
Chinese Translation
在线训练使计算机使用代理（CUAs）能够通过与可执行环境的互动来提升性能。然而，现有方法主要依赖稀疏的结果奖励，这对中间动作没有监督。基于策略的自蒸馏（On-policy Self-Distillation, OPSD）通过特权重评分提供了令牌级学习信号，但将其直接应用于CUA在线训练面临两个挑战：固定指导可能与学生的当前状态不一致，以及指导引起的概率变化可能与步骤级的正确性相冲突。我们提出了ComputerSD，这是一种针对CUAs的在线自蒸馏方法，它将从执行的图形用户界面（GUI）过渡中获得的实时反馈转化为政策学习的指导。经过微调的GUI分析器在每个动作后生成指导和步骤级价值评分；指导提供特权上下文，而评分则调节生成的OPSD信号。ComputerSD在完全异步的训练框架中联合优化令牌级OPSD和轨迹级广义重参数优化（Generalized Reparameterization Optimization, GRPO）。在OSWorld-Verified上，ComputerSD在通用Qwen3-VL-8B-Thinking和专用EvoCUA-8B骨干网络上分别比仅基于结果的GRPO提高了1.9和4.1个百分点。在分布外设置中的评估进一步支持了ComputerSD的可推广性。这些结果证明了通过在线自蒸馏从实时反馈中学习对于CUAs的有效性。
cs.CV / 178 / 2609.40305
Looped Diffusion Transformer
循环扩散变换器
Abstract
Improving text-to-image models has traditionally relied on increasing model size or the number of denoising steps. In this work, we explore an alternative way to scale computation by repeatedly running shared Transformer blocks within each denoising step, effectively increasing computational depth while keeping the parameter count fixed. This looped computation enables iterative refinement of internal representations without explicit reasoning tokens. However, naive looping fails to consistently improve image quality. We trace this problem to weak supervision across intermediate loops and unregulated attention updates that progressively erode local information. To overcome these challenges, we propose Looped Diffusion Transformer (Looped-DiT), which combines deep supervision across intermediate loops with self-modulating attention to stabilize looped feature updates. Under matched-parameter and matched-compute settings, Looped-DiT consistently outperforms non-looped baselines. Notably, a 260M-parameter looped model can surpass a model 6.5x larger across multiple text-to-image benchmarks while requiring 4.9x lower inference compute. Beyond this performance gain, we find that looped computation can offer a more effective form of iterative computation for diffusion models, with increasing loop depth yielding larger gains than adding more denoising steps under a fixed inference budget. Furthermore, deeper loops can progressively correct mistakes made in earlier loops, exhibiting behaviors suggestive of latent reasoning. Together, these results show that looped computation offers a promising way to scale visual generation models.
Chinese Translation
改进文本到图像模型传统上依赖于增加模型规模或去噪步骤的数量。在本研究中，我们探索了一种通过在每个去噪步骤中反复运行共享的变换器块来扩展计算的替代方法，有效地增加了计算深度，同时保持参数数量不变。这种循环计算使得内部表征的迭代优化成为可能，而无需显式的推理标记。然而，简单的循环无法持续提高图像质量。我们将这个问题追溯到中间循环中的弱监督和不受控制的注意力更新，这些更新逐渐侵蚀了局部信息。为了解决这些挑战，我们提出了循环扩散变换器（Looped Diffusion Transformer, Looped-DiT），它结合了中间循环的深度监督与自调节注意力，以稳定循环特征更新。在匹配参数和匹配计算的设置下，Looped-DiT始终优于非循环基线。值得注意的是，一个260M参数的循环模型在多个文本到图像基准测试中可以超越一个大6.5倍的模型，同时需要的推理计算量低4.9倍。除了这种性能提升外，我们发现循环计算可以为扩散模型提供一种更有效的迭代计算形式，随着循环深度的增加，所获得的收益大于在固定推理预算下增加更多去噪步骤。此外，更深的循环可以逐步纠正早期循环中的错误，表现出潜在推理的特征。综合来看，这些结果表明，循环计算为扩展视觉生成模型提供了一种有前景的方法。
cs.CV / 179 / 2609.40317
GLARE: Generating Listening Heads with Appropriate Reactions
GLARE：生成具有适当反应的倾听者
Abstract
While talking head generation has advanced rapidly, generating natural listener behavior in dyadic conversations, which know when to react, how to react, and with what type of response, remains underexplored. Existing dyadic datasets lack fine-grained listener reaction annotations, and prevailing evaluation metrics inherited from talking-head and video generation measure visual realism rather than whether a listener reacted appropriately. We address these gaps along three aspects. First, we curate a listening-head-specific dataset built from RealTalk and Seamless Interaction, comprising approximately 147 hours of paired speaker-listener videos with 64,557 event-level reaction annotations across six categories: nodding, head shaking, smiling, laughing, frowning, and surprised. Second, we introduce an audio-driven baseline built on a flow-matching transformer, namely GLARE, with prosody conditioning derived from Qwen2-Audio and a temporal reaction loss that explicitly supervises frame-wise reactions. Third, we propose a reaction-oriented evaluation protocol that jointly measures reaction occurrence (R-F1), temporal alignment (R-tIoU), asymmetric temporal deviation (R-ATD), and reaction-region visual quality (R-FID), giving a more behaviorally grounded assessment than visual-quality-only metrics. Experiment results show consistent gains over prior listening-head methods in both visual fidelity and reaction-level metrics, suggesting that reaction-aware data, modeling, and evaluation are critical for natural listening behavior.
Chinese Translation
尽管谈话头生成技术迅速发展，但在双人对话中生成自然的倾听者行为（即知道何时反应、如何反应以及以何种类型的回应）仍然未得到充分探索。现有的双人数据集缺乏细粒度的倾听者反应注释，而从谈话头和视频生成中继承的评估指标主要衡量视觉真实感，而非倾听者是否做出了适当反应。我们从三个方面解决了这些问题。首先，我们整理了一个专门针对倾听者的数据库，该数据库基于RealTalk和Seamless Interaction构建，包含约147小时的配对发言者-倾听者视频，涵盖64,557个事件级反应注释，分为六类：点头、摇头、微笑、笑、皱眉和惊讶。其次，我们提出了一个基于流匹配变换器的音频驱动基线模型，称为GLARE，结合了来自Qwen2-Audio的韵律条件和显式监督逐帧反应的时间反应损失。第三，我们提出了一种以反应为导向的评估协议，联合测量反应发生率（R-F1）、时间对齐（R-tIoU）、非对称时间偏差（R-ATD）和反应区域视觉质量（R-FID），提供比仅基于视觉质量的指标更具行为基础的评估。实验结果表明，在视觉保真度和反应级指标上，相较于以往的倾听者方法有了持续的提升，表明关注反应的数据、建模和评估对于自然倾听行为至关重要。
cs.CV / 180 / 2609.40320
Atomizer-IO: Beyond Pixels, Patches and Grids
Atomizer-IO：超越像素、补丁和网格
Abstract
Most vision architectures assume that observations lie on a regular grid, an effective abstraction for natural images but a restrictive one for sensing data whose channels, temporal sampling, spatial resolution, and geometry can vary. Generic set-based architectures remove the grid, but also remove useful spatial inductive biases. We introduce Atomizer-IO, an architecture that places observations first and derives structure from their physical relationships. Building on top of an atomic representation of the data, each observation is described by its measurement and acquisition metadata, while local cross-attention maps observations to anchor points that can be arbitrarily placed. We evaluate this design by progressively relaxing the grid assumption, from varying input raster configurations and incomplete channel sets to flexible output density and, ultimately, inputs without a raster grid. Atomizer-IO is competitive with flexible EO-specific architectures on most tasks, while offering post-training control over inference cost and competitive compute--performance trade-offs. The same formulation extends without architectural redesign to unordered 3D point clouds, showing that the atomic interface generalizes beyond regular raster inputs. These results suggest that pixels, patches, and grids do not need to define the interface of a sensing architecture.
Chinese Translation
大多数视觉架构假设观测值位于规则网格上，这对于自然图像是一种有效的抽象，但对于通道、时间采样、空间分辨率和几何形状可能变化的传感数据而言则显得过于限制。通用的基于集合的架构去除了网格，但也消除了有用的空间归纳偏置。我们提出了Atomizer-IO，这是一种将观测值置于首位并从其物理关系中推导结构的架构。在数据的原子表示基础上，每个观测值通过其测量和采集元数据进行描述，同时局部交叉注意力将观测值映射到可以任意放置的锚点。我们通过逐步放宽网格假设来评估这一设计，从变化的输入光栅配置和不完整的通道集，到灵活的输出密度，最终到没有光栅网格的输入。Atomizer-IO在大多数任务上与灵活的EO特定架构具有竞争力，同时提供对推理成本的后训练控制和竞争性的计算-性能权衡。同样的公式在不重新设计架构的情况下扩展到无序的3D点云，表明原子接口超越了规则光栅输入的通用性。这些结果表明，像素、补丁和网格并不需要定义传感架构的接口。
cs.CV / 181 / 2609.40322
MatLoom: Layered Text-to-Material Generation in a Compact Program Space
MatLoom：在紧凑程序空间中分层文本到材料生成
Abstract
Material generation should produce not only an appearance, but also the rules that construct it. We introduce MatLoom, a compact, layer-oriented language for text-to-material generation with pretrained language models. Each program composes alpha-masked layers whose shared spatial expressions define coverage and physically based rendering (PBR) channels, making dependencies between patterns, color, and relief explicit. A standalone interpreter evaluates the program into material maps, while the source retains named fields and layer parameters for subsequent authoring. Without task-specific fine-tuning, our pipeline uses parser-guided repair and preview-based critique to revise material designs, then searches noise seeds while keeping each candidate's remaining source fixed. On a curated benchmark of 141 prompts evaluated with six backbones, our best-performing configuration achieves higher mean scores than three diffusion baselines on all four flat-layout prompt-alignment metrics. Its initial programs already exceed all three baselines on mean BLIPScore, before critique or seed search. Retained programs have a median length of 21 lines when pooled across backbones. In a blind four-way comparison involving 30 participants and 20 prompts, our renders receive 59.2% of choices, compared with 19.3% for the most-preferred baseline. Compact executable programs thus offer a way to generate prompt-aligned materials while retaining their construction as part of the asset.
Chinese Translation
材料生成不仅应产生外观，还应生成构建该外观的规则。我们介绍了MatLoom，这是一种紧凑的、面向层的语言，用于通过预训练语言模型进行文本到材料的生成。每个程序由具有 alpha 蒙版的层组成，这些层的共享空间表达定义了覆盖范围和物理基础渲染（PBR）通道，使得模式、颜色和浮雕之间的依赖关系变得明确。一个独立的解释器将程序评估为材料图，而源代码保留命名字段和层参数以供后续创作。在没有特定任务微调的情况下，我们的管道使用解析器引导的修复和基于预览的批评来修订材料设计，然后在保持每个候选源固定的情况下搜索噪声种子。在一个经过精心策划的基准测试中，141个提示通过六个基础模型进行评估，我们表现最佳的配置在所有四个平面布局提示对齐指标上都取得了比三个扩散基线更高的平均分数。在批评或种子搜索之前，其初始程序在平均 BLIPScore 上已经超过了所有三个基线。保留的程序在跨基础模型汇总时的中位长度为21行。在一项涉及30名参与者和20个提示的盲四向比较中，我们的渲染获得了59.2%的选择，而最受欢迎基线的选择仅为19.3%。因此，紧凑的可执行程序提供了一种生成与提示对齐的材料的方法，同时将其构建过程作为资产的一部分保留。
cs.CV / 182 / 2609.40333
I Have a Stream: Making Self-Supervised Learning Work on Continuous Video
我有一条流：使自监督学习在连续视频上有效
Abstract
Self-supervised learning draws inspiration from infant visual development, yet standard training pipelines bear little resemblance to it: images are independently sampled and globally shuffled across epochs. We study self-supervised learning from continuous video streams, where frames are consumed in temporal order using strict sliding-window batches, without global reshuffling or multi-epoch replay. To this end, we construct WT++, a 95-hour urban walking-tour video dataset for streaming pretraining. Combined with a comprehensive evaluation suite we find that contrastive and distillation-based methods struggle in this setting, while MAE is more robust but still falls short of standard i.i.d. pretraining. We find that high inter-batch similarity, caused by sliding-window consumption across consecutive batches, does not explain this gap. The main challenge is high intra-batch similarity, where frames within each batch are near-duplicates. To mitigate this, we propose StreamMAE, which preserves the core MAE reconstruction objective while adapting the input pipeline with stream-aware regularization and motion-biased crop selection. StreamMAE outperforms streaming baselines, matches i.i.d. MAE trained on the same video data, remains competitive with ImageNet-pretrained MAE, and scales positively as the pretraining stream grows from 12 to 95 hours.
Chinese Translation
自监督学习受到婴儿视觉发展的启发，但标准训练流程与之相去甚远：图像是独立采样并在各个时期之间全局打乱。我们研究了来自连续视频流的自监督学习，其中帧以时间顺序被消费，采用严格的滑动窗口批次，而没有全局重洗或多周期重放。为此，我们构建了 WT++，一个包含 95 小时城市步行旅游视频的数据集，用于流式预训练。结合全面的评估套件，我们发现对比和蒸馏基础的方法在这种设置下表现不佳，而 MAE 更加稳健，但仍未达到标准的独立同分布（i.i.d.）预训练水平。我们发现，由于连续批次之间的滑动窗口消费导致的高批间相似性并不能解释这一差距。主要挑战在于高批内相似性，其中每个批次内的帧几乎是重复的。为了解决这个问题，我们提出了 StreamMAE，它在保留核心 MAE 重建目标的同时，通过流感知正则化和运动偏向裁剪选择来调整输入管道。StreamMAE 超越了流式基线，与在相同视频数据上训练的 i.i.d. MAE 相匹配，并且在与 ImageNet 预训练的 MAE 竞争时表现良好，同时随着预训练流从 12 小时增长到 95 小时而呈现出积极的扩展性。
cs.CV / 183 / 2609.40347
Image Classifiers are Efficient Self-Supervised Video Representation Learners
图像分类器是高效的自监督视频表征学习者
Abstract
We introduce VideoMSN, a Masked Siamese Network framework for efficient self-supervised spatio-temporal representation learning in videos. Instead of relying on heavy 3D architectures or reconstruction-based autoencoders for learning with unlabeled data, we repurpose standard image Vision Transformers by representing videos as super images which are grids composed of frames sampled from videos. From each super image, we construct two views: one with spatial patch masking and the other with temporal frame masking, ensuring no information leakage across frames. A shared Vision Transformer (ViT) encoder aligns their embeddings using a masked Siamese loss, capturing both motion and appearance cues without reconstruction. Our decoder-free formulation leverages an image foundation model towards efficient video representation learning. Starting from pretrained DINO-v3 and DeiT-v3 image encoders, VideoMSN achieves state-of-the-art performance on Kinetics-400, UCF101, and HMDB51 while requiring up to $32\times$ fewer and $160\times$ fewer video pretraining epochs compared to prior video self-supervised learning methods. Our proposed approach also shows strong performance in low-shot classification, confirming the transferability of the learned representations in a label-scarce scenario. Project Page: https://cvir.github.io/projects/videomsn.
Chinese Translation
我们提出了VideoMSN，一个用于高效自监督时空表征学习的掩码西蒙网络框架。我们并不依赖于重型的3D架构或基于重建的自编码器来处理无标签数据，而是通过将视频表示为超图像（super images），即由从视频中采样的帧组成的网格，重新利用标准的图像视觉变换器（Vision Transformers）。从每个超图像中，我们构建了两个视图：一个是空间补丁掩码，另一个是时间帧掩码，确保帧之间没有信息泄漏。一个共享的视觉变换器（ViT）编码器使用掩码西蒙损失对它们的嵌入进行对齐，捕捉运动和外观线索，而无需重建。我们的无解码器公式利用图像基础模型实现高效的视频表征学习。从预训练的DINO-v3和DeiT-v3图像编码器开始，VideoMSN在Kinetics-400、UCF101和HMDB51上达到了最先进的性能，同时相比于之前的视频自监督学习方法，所需的视频预训练周期减少了多达$32 imes$和$160 imes$。我们提出的方法在低样本分类中也表现出强劲的性能，确认了在标签稀缺场景中学习到的表征的可迁移性。项目页面：https://cvir.github.io/projects/videomsn。
cs.CV / 184 / 2609.40353
AssemblyWorld: Rethinking 3D Assembly with General-Purpose Agents
AssemblyWorld：重新思考通用智能体的三维装配
Abstract
The task of 3D assembly requires translating an understanding of parts and their relationships into precise spatial arrangements. Can pretrained general-purpose agents assemble objects through visual interaction without additional assembly-specific fine-tuning? To investigate this question, we introduce AssemblyWorld, an interactive 3D environment in which agents inspect rendered views and manipulate supplied rigid parts, guided by images or assembly manuals when available. Agents perceive part geometry through 2D views rather than direct access to mesh vertices or faces, while their resulting assemblies are evaluated geometrically. Building on this environment, we construct AssemblyWorldBench, comprising 100 assembly tasks across 80 objects spanning furniture, industrial assembly, and fracture reassembly. Evaluating eight agent systems reveals substantial differences in their capabilities. The strongest system achieves 80.9% part accuracy but 59.4% complete-assembly success. The evaluated open-source systems lag substantially behind their stronger closed-source peers in both execution reliability and assembly accuracy. Analyses of visual references, interaction trajectories, and failures show how agents revise assemblies while leaving residual positioning errors. AssemblyWorld provides a common setting for both assessing the capabilities of interactive assembly agents and characterizing the gap between approximate structure recovery and precise reconstruction.
Chinese Translation
三维装配任务需要将对零件及其关系的理解转化为精确的空间排列。预训练的通用智能体能否通过视觉交互在没有额外装配特定微调的情况下组装物体？为探讨这一问题，我们引入了AssemblyWorld，一个交互式三维环境，在该环境中，智能体检查渲染视图并操作提供的刚性零件，当可用时，依据图像或装配手册进行指导。智能体通过二维视图感知零件几何形状，而不是直接访问网格顶点或面，其最终的装配结果通过几何方式进行评估。在此环境的基础上，我们构建了AssemblyWorldBench，包括跨越80个物体的100个装配任务，涵盖家具、工业装配和断裂重组。对八个智能体系统的评估揭示了它们能力的显著差异。最强的系统实现了80.9%的零件准确率，但完整装配成功率仅为59.4%。评估的开源系统在执行可靠性和装配准确性方面显著落后于其更强的闭源同行。对视觉参考、交互轨迹和失败案例的分析显示，智能体在修正装配时留下了残余的定位误差。AssemblyWorld为评估交互式装配智能体的能力以及表征近似结构恢复与精确重建之间的差距提供了一个共同的环境。
cs.CV / 185 / 2609.40356
ViTeX-Bench: Benchmarking High-Fidelity Video Scene Text Editing
ViTeX-Bench：高保真视频场景文本编辑基准测试
Abstract
Recent video generation is increasingly realistic and controllable, yet video editing remains less developed, particularly for precise local edits that must preserve the original scene dynamics. Video scene text editing replaces text on scene surfaces, such as storefront signs, whiteboards, and product labels, while preserving the surrounding content, motion, and camera dynamics. Although scene text editing is well studied for images, video scene text editing that achieves high visual quality, temporal consistency, and edit locality remains underexplored. Existing resources offer limited paired real-video data, and general video-editing metrics do not directly measure whether the requested text remains correct over time. We introduce ViTeX-Bench, a benchmark suite comprising ViTeX-Dataset and a three-axis evaluation protocol. The dataset contains 387 real-world 720p videos with text-region masks and editing instructions: 230 provide reviewed, pipeline-generated paired edits for training, and 157 form a frozen evaluation split. The protocol evaluates text correctness, visual and temporal quality, and edit locality through 13 metrics, with one primary metric per axis and a Pareto comparison of their trade-offs. OCR calibration, human evaluation, and annotation-sensitivity analyses support the interpretation of these scores. Across eight baselines from four editing families, accurate text, temporal stability, and scene preservation remain difficult to achieve together. We also release ViTeX-Edit-14B, an open-source reference editor fine-tuned on the paired training split with motion-aligned glyph-video conditioning. It achieves CharAcc 0.688, the highest mean among the evaluated video-native editors, and the lowest comparable text-crop Warp among raw editor outputs. ViTeX-Bench provides a reproducible foundation for studying these trade-offs in video scene text editing.
Chinese Translation
近年来，视频生成变得越来越逼真且可控，但视频编辑的发展相对滞后，尤其是在需要精确局部编辑并保持原始场景动态的情况下。视频场景文本编辑是在场景表面（如店面招牌、白板和产品标签）上替换文本，同时保留周围内容、运动和摄像机动态。尽管图像的场景文本编辑研究较为充分，但实现高视觉质量、时间一致性和编辑局部性的的视频场景文本编辑仍然未得到充分探索。现有资源提供的配对真实视频数据有限，而通用视频编辑指标并未直接衡量所请求的文本是否随时间保持正确。我们提出了ViTeX-Bench，一个基准套件，包括ViTeX-Dataset和一个三轴评估协议。该数据集包含387个真实世界的720p视频，配有文本区域掩码和编辑指令：230个提供经过审核的、管道生成的配对编辑用于训练，157个形成冻结评估分割。该协议通过13个指标评估文本正确性、视觉和时间质量以及编辑局部性，每个轴有一个主要指标，并进行其权衡的帕累托比较。OCR校准、人类评估和注释敏感性分析支持对这些分数的解释。在来自四个编辑家族的八个基线中，准确的文本、时间稳定性和场景保留仍然难以同时实现。我们还发布了ViTeX-Edit-14B，一个开源参考编辑器，经过配对训练分割的微调，并具有运动对齐的字形视频条件。它在评估的视频原生编辑器中实现了0.688的字符准确率（CharAcc），是最高的均值，并且在原始编辑器输出中具有最低的可比文本裁剪扭曲（Warp）。ViTeX-Bench为研究视频场景文本编辑中的这些权衡提供了可重复的基础。
cs.CV / 186 / 2609.40358
Physis-Lang: Self-Evolving Language as a Physical Representation for Video World Model
Physis-Lang：作为视频世界模型的物理表征的自我演化语言
Abstract
Video world models are expected to predict how the physical world evolves, yet they often produce visually plausible videos that violate basic physical principles. Existing approaches commonly assume that natural language is insufficient to represent the physical knowledge required for reliable generation, and therefore introduce additional visual, latent, numerical, or planning-based signals. We revisit this assumption and introduce Physis-Lang, a self-evolving framework that treats physical language as a shared and optimizable representation across data curation, model training, and video generation. Physis-Lang represents physical processes through language that describes their relevant entities, causes, interactions, governing principles, temporal evolution, and effects. To improve this representation, we construct PhysCapBench, which decomposes physical processes into atomic assertions and evaluates captions using recall and precision. An agentic loop iteratively analyzes assertion-level errors and refines the instruction used to produce physical captions. Physis-Lang further converts model deficiencies into textual descriptions and uses language-guided retrieval to identify visually diverse videos that cover missing physical processes. Experiments on four widely used physical video benchmarks with Wan and Cosmos backbones demonstrate consistent improvements in physical plausibility. Notably, starting from open-source Cosmos3-Nano backbones, our Physis-Lang-enhanced models surpass the leading proprietary Veo 3.1 model.
Chinese Translation
视频世界模型被期望能够预测物理世界的演变，然而它们常常生成视觉上合理但违反基本物理原则的视频。现有方法通常假设自然语言不足以表示可靠生成所需的物理知识，因此引入额外的视觉、潜在、数值或基于规划的信号。我们重新审视这一假设，并引入Physis-Lang，一个自我演化的框架，将物理语言视为数据整理、模型训练和视频生成之间的共享和可优化表征。Physis-Lang通过描述相关实体、原因、交互、支配原则、时间演变和效果的语言来表征物理过程。为了改善这一表征，我们构建了PhysCapBench，它将物理过程分解为原子断言，并使用召回率和精确率评估字幕。一个代理循环迭代分析断言级别的错误，并优化用于生成物理字幕的指令。Physis-Lang进一步将模型缺陷转化为文本描述，并使用语言引导的检索来识别涵盖缺失物理过程的视觉多样性视频。在四个广泛使用的物理视频基准上进行的实验，使用Wan和Cosmos骨干网络，展示了物理合理性的持续改善。值得注意的是，从开源的Cosmos3-Nano骨干网络开始，我们的Physis-Lang增强模型超越了领先的专有Veo 3.1模型。
cs.CV / 187 / 2609.40362
Multimodal Flow: Unified Flow Modeling of Language and Vision in Embedding Spaces
多模态流：语言与视觉在嵌入空间中的统一流建模
Abstract
We present Multimodal Flow, a fully continuous generative model of language and vision. Most unified multimodal models either model both language and quantized images as discrete tokens or combine discrete language prediction with continuous image generation. The former introduces a visual quantization bottleneck. The latter requires modality-dependent objectives and sampling procedures. Fully continuous modeling avoids these trade-offs and enables a shared generative process, but remains underexplored for multimodal pretraining. Multimodal Flow introduces a unified continuous architecture that integrates multimodal continuous representations with a shared chunk-causal flow backbone. It organizes text blocks and images as ordered continuous hyperchunks, preserving textual token order and visual spatial structure. The backbone learns a single vector field over these hyperchunks through Flow Matching. Joint attention enables cross-modal interaction, while modality-specific feed-forward networks process each modality. The model predicts multiple target chunks in parallel during training and generates hyperchunks sequentially at inference. We instantiate MF-1 and pretrain it on multimodal data. Across 0.6B, 1.2B, and 1.6B scales, continued pretraining consistently improves multimodal modeling. With only 150B pretraining tokens, MF-1 achieves an average score of 82.8 across GenEval and DPG-Bench and 75.3 across VQAv2, MMBench, and POPE, remaining competitive with unified models trained on substantially more data. Under matched data, optimization, and parameter budgets, Multimodal Flow further outperforms representative hybrid and discrete models. These results establish continuous chunk-based embedding flow modeling as a new fully continuous paradigm for unified multimodal modeling. The related code and model are publicly released at https://github.com/hustvl/Multimodal-Flow.
Chinese Translation
我们提出了多模态流（Multimodal Flow），一种完全连续的语言与视觉生成模型。大多数统一的多模态模型要么将语言和量化图像建模为离散的标记，要么将离散的语言预测与连续的图像生成相结合。前者引入了视觉量化瓶颈，后者则需要依赖于模态的目标和采样过程。完全连续的建模避免了这些权衡，并实现了共享的生成过程，但在多模态预训练中仍然未得到充分探索。多模态流引入了一种统一的连续架构，将多模态连续表示与共享的块因果流（chunk-causal flow）主干整合在一起。它将文本块和图像组织为有序的连续超块（hyperchunks），保持文本标记的顺序和视觉空间结构。主干通过流匹配（Flow Matching）学习这些超块上的单一向量场。联合注意力（joint attention）使得跨模态交互成为可能，而特定模态的前馈网络处理每种模态。在训练期间，模型并行预测多个目标块，并在推理时顺序生成超块。我们实例化了MF-1并在多模态数据上进行了预训练。在0.6B、1.2B和1.6B的规模下，持续的预训练始终改善了多模态建模。仅用150B的预训练标记，MF-1在GenEval和DPG-Bench上获得了82.8的平均分，在VQAv2、MMBench和POPE上获得了75.3，仍然与在更多数据上训练的统一模型保持竞争力。在匹配的数据、优化和参数预算下，多模态流进一步超越了代表性的混合和离散模型。这些结果确立了基于连续块的嵌入流建模作为统一多模态建模的新完全连续范式。相关代码和模型已公开发布在https://github.com/hustvl/Multimodal-Flow。
cs.AI / 1 / 2609.38282
Improving OCR Faithfulness via Gated and Attenuated On-Policy Distillation
通过门控和衰减的在线蒸馏提高OCR的忠实度
Abstract
Vision-language models may rewrite anomalous text in images into linguistically plausible expressions, compromising OCR transcription faithfulness. Sequence-level task rewards and local teacher guidance are complementary, but guidance from the same teacher may not remain equally effective as the student improves. Offline analysis shows that supervision from a fixed teacher becomes progressively less favorable as the student improves, both across training checkpoints and across response groups with different task rewards. Motivated by this observation, we introduce GAD-RL, which adaptively regulates teacher supervision during joint post-training according to the student's current task performance and local distributions. A frozen teacher conditions on reference transcriptions and student-generated prefixes. GAD-RL disables distillation for response groups containing an output with task reward at least 0.95 and continuously attenuates distillation strength as group-mean reward increases. It also weights forward KL by the student's probability of the teacher's Top-1 token, moderating local auxiliary updates when student support for that candidate is low. On Qwen3.5-2B, GAD-RL achieves 59.92% Micro Recall on CHAOS-Bench, surpassing GRPO and GRPO+OPD (fixed-weight) by 8.45 and 4.43 percentage points, respectively, while achieving an Overall score of 91.18 on OmniDocBench v1.6.
Chinese Translation
视觉语言模型可能会将图像中的异常文本重写为语言上合理的表达，从而影响OCR转录的忠实度。序列级任务奖励和局部教师指导是互补的，但来自同一教师的指导可能在学生进步时效果不再均等。离线分析表明，来自固定教师的监督在学生进步时逐渐变得不利，这种情况在不同的训练检查点和具有不同任务奖励的响应组中均有所体现。基于这一观察，我们引入了GAD-RL，它根据学生当前的任务表现和局部分布自适应地调节教师监督。在此过程中，冻结的教师基于参考转录和学生生成的前缀进行条件化。GAD-RL对包含任务奖励至少为0.95的输出的响应组禁用蒸馏，并随着组均值奖励的增加持续衰减蒸馏强度。同时，它根据学生对教师Top-1标记的概率加权前向KL，当学生对该候选项的支持较低时，适度调整局部辅助更新。在Qwen3.5-2B上，GAD-RL在CHAOS-Bench上实现了59.92%的微召回率，分别超越了GRPO和GRPO+OPD（固定权重）8.45和4.43个百分点，同时在OmniDocBench v1.6上获得了91.18的总体得分。
cs.AI / 2 / 2609.38288
AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks
AREX-2：通过长时间反思任务推进自我改进代理
Abstract
We present AREX-2, an effort to advance the self-improving capability of LLM agents, which we define as the ability to iteratively refine a solution at test time. This ability rests on two complementary capabilities: reflection, which produces a solution better than the current one, and long-horizon execution, which keeps the iteration effective over many rounds. We hypothesize that both capabilities are domain-agnostic, and can therefore be learned in scenarios that are well suited for supervision. Accordingly, we synthesize long-horizon improvement trajectories from machine learning and algorithmic programming tasks, two domains that offer verifiable feedback and reward sustained iteration. Trained on this data, our agent, built on Qwen3.8-27B, achieves strong results on MLE-bench Lite (81.8) and Frontier-CS (70.7), transfers to deep research with 84.0 on BrowseComp, 52.6 on HLE, 92.2 on GAIA, and 93.8 on DeepSearchQA, and keeps improving as its budget of rounds grows. These results show that long-horizon reflective data is an effective route toward self-improving agents.
Chinese Translation
我们提出了AREX-2，这是一个旨在提升大型语言模型（LLM）代理自我改进能力的努力，我们将其定义为在测试时迭代优化解决方案的能力。这种能力依赖于两种互补的能力：反思，能够生成比当前解决方案更优的结果，以及长时间执行，确保在多个回合中保持迭代的有效性。我们假设这两种能力是领域无关的，因此可以在适合监督的场景中学习。因此，我们从机器学习和算法编程任务中综合了长时间改进轨迹，这两个领域提供可验证的反馈并奖励持续的迭代。在这些数据上训练的代理，基于Qwen3.8-27B，在MLE-bench Lite（81.8）和Frontier-CS（70.7）上取得了优异的成绩，并在深度研究中转移到BrowseComp（84.0）、HLE（52.6）、GAIA（92.2）和DeepSearchQA（93.8）上，随着回合预算的增加而不断改进。这些结果表明，长时间反思数据是实现自我改进代理的有效途径。
cs.AI / 3 / 2609.38294
MoFlow: Multi-Objective Agentic Workflow Generation
MoFlow：多目标自主工作流生成
Abstract
We study the generation of agentic workflows that jointly optimize multiple objectives, such as accuracy, cost, latency, robustness, and consistency. Existing methods for workflow generation typically optimize accuracy alone or a weighted sum of objectives, so each trained generator commits to one fixed trade-off and must be retrained from scratch when preferences change. To alleviate this, we propose MoFlow, which generates workflows optimized across varied preferences. Specifically, MoFlow formulates workflow generation as a multi-objective Markov decision process and solves it by leveraging Convex-Hull Monte Carlo Tree Search with optimistic set-valued backups, where every node stores a set of reachable trade-offs rather than one weighted score. A single search thus approximately covers the Pareto front, from which MoFlow can return a workflow for any preference by lookup without retraining. We evaluate MoFlow against six strong baselines on six benchmarks spanning mathematics, code, and question answering. Since the baselines are single-scalar optimizers by design, an apples-to-apples comparison is difficult. We instead adopt an evaluation setup that favors the baselines, in that they are rerun for each testing preference, which MoFlow never sees. Even under this stringent setup, MoFlow achieves the highest average hypervolume.
Chinese Translation
我们研究了生成能够联合优化多个目标的自主工作流，例如准确性、成本、延迟、鲁棒性和一致性。现有的工作流生成方法通常仅优化准确性或目标的加权总和，因此每个训练生成器都承诺于一个固定的权衡，并且当偏好发生变化时必须从头开始重新训练。为了解决这个问题，我们提出了MoFlow，它生成在不同偏好下优化的工作流。具体而言，MoFlow将工作流生成形式化为多目标马尔可夫决策过程，并通过利用具有乐观集值备份的凸包蒙特卡洛树搜索来解决该问题，其中每个节点存储一组可达的权衡，而不是一个加权分数。因此，单次搜索大致覆盖了帕累托前沿，从中MoFlow可以通过查找返回任何偏好的工作流，而无需重新训练。我们在六个基准测试中评估了MoFlow，相较于六个强基线，涵盖了数学、代码和问答领域。由于基线在设计上是单标量优化器，因此进行逐一比较是困难的。我们采用了一种有利于基线的评估设置，即它们在每个测试偏好下重新运行，而MoFlow从未见过这些偏好。即使在这种严格的设置下，MoFlow仍然实现了最高的平均超体积。
cs.AI / 4 / 2609.38296
AI Agents are Vulnerable to Radicalization
人工智能代理易受激进化影响
Abstract
Large language models (LLMs) can influence people's beliefs, yet little is known about whether and how they can manipulate each other. To investigate this, we simulate conversations between two agents: a target LLM that role-plays a human persona based on demographic and psychological attributes, and an influencer LLM that aims to make the target's beliefs more extreme. We examine radicalization along two pathways: resonance, where the influencer reinforces a target's pre-existing belief, and persuasion, where the influencer promotes a belief the target initially considers unimportant. Across affective and behavioral metrics, we find that both mechanisms radicalize the target. However, resonance produces consistently stronger effects than persuasion. Different influence tactics, such as using sycophancy and unverified claims, produce different levels of radicalization, but not consistently across metrics. We further show that resonance propagates to related beliefs, suggesting interconnected belief structures within AI agents. These findings indicate that AI agents are susceptible to radicalization, particularly when messages align with their existing beliefs, raising concerns about the vulnerability of personalized AI agents and multi-agent AI ecosystems.
Chinese Translation
大型语言模型（LLMs）能够影响人们的信念，但关于它们是否以及如何相互操控的研究仍然很少。为此，我们模拟了两个代理之间的对话：一个目标LLM，扮演基于人口统计和心理属性的人类角色；一个影响者LLM，旨在使目标的信念更加极端。我们从两个路径考察激进化：共鸣（resonance），即影响者强化目标的既有信念；说服（persuasion），即影响者推广目标最初认为不重要的信念。通过情感和行为指标的分析，我们发现这两种机制都能使目标激进化。然而，共鸣产生的效果始终强于说服。不同的影响策略，如使用谄媚和未经验证的主张，产生不同程度的激进化，但在各指标间并不一致。我们进一步表明，共鸣会传播到相关信念，暗示AI代理之间存在相互关联的信念结构。这些发现表明，AI代理易受激进化影响，尤其是在信息与其现有信念一致时，这引发了对个性化AI代理和多代理AI生态系统脆弱性的担忧。
cs.AI / 5 / 2609.38340
CARAT: Do Materials LLMs Reason or Recite?
CARAT：材料大语言模型是推理还是复述？
Abstract
When a materials LLM answers a question about crystal structure, does it reason from the structure or copy an answer already printed in its input? Accuracy cannot tell: a structural description often prints the very field it is scored against. CARAT holds question and gold answer fixed across eight matched views, names each structural relation separately in GraphSpace, and adds matched fine-tuning, answer masking, evidence injection, paired inference, and a rule that can withhold claims. First, on the benchmark's hardest families the grounded view is worth 17.3 points over formula inputs. Second, we turn that scrutiny on ourselves. GraphSpace beats a plain periodic graph by 19.3 points, but that margin is two effects at once: where the plain rendering carries everything the question needs it is 1.96 points, and where it omits those fields entirely, 46.7 points. The headline mostly measures what the baseline lacked, not how evidence is presented. Third, we attack our own benchmark. A rule that skips the link and reads the list directly answers four of seven hardened families, so we rebuilt it until eleven such shortcuts sat near chance. The frozen model quotes that link yet answers the same when we redirect it, on 95.6% of paired cases: it repeats the relation without using it. After matched supervision it reaches 99.8%, and deleting the link drops it to 23.4%, below the 27.0% the best shortcut reaches: both steps are learnable.
Chinese Translation
当材料大语言模型回答有关晶体结构的问题时，它是基于结构进行推理，还是复制输入中已经存在的答案？准确性无法判断：结构描述通常会打印出其评分所依据的领域。CARAT 在八个匹配视图中保持问题和黄金答案不变，在 GraphSpace 中单独命名每个结构关系，并添加匹配的微调、答案屏蔽、证据注入、配对推理以及可以保留声明的规则。首先，在基准测试中最困难的类别中，基于实际数据的视图比公式输入高出 17.3 分。其次，我们对自己进行审视。GraphSpace 比普通周期图高出 19.3 分，但这个差距同时包含两个效应：在普通渲染中包含了问题所需的所有内容时，得分为 1.96 分，而在完全省略这些领域时，得分为 46.7 分。这个结果主要衡量的是基线所缺乏的内容，而不是证据的呈现方式。第三，我们挑战自己的基准。一条跳过链接并直接读取列表的规则能够回答七个困难类别中的四个，因此我们重建了它，直到有十一条这样的捷径接近随机水平。冻结模型引用了该链接，但在我们重定向时在 95.6% 的配对案例中给出了相同的答案：它重复了关系而没有使用它。经过匹配监督后，它达到了 99.8%，而删除链接则将其降至 23.4%，低于最佳捷径的 27.0%：这两个步骤都是可学习的。
cs.AI / 6 / 2609.38346
Examining Variation in How Guided AI Tutors Resolve Student Impasses
探讨引导式人工智能辅导员解决学生困境的变异性
Abstract
When a student is stuck, a tutor faces the assistance dilemma: help given too early can hinder productive struggle, while help withheld too long leaves the student in a frustrating, persistent impasse (i.e., wheel spinning). Generative AI tutors increasingly use guardrails restricting answer-giving, yet little is known about how such tutors behave once an impasse persists. We analyze 20,462 student turns from 1,260 authentic sessions with a guided LLM chemistry tutor, identifying 6,630 impasse turns of three major types: conceptual errors, expressed uncertainty, or help-seeking. We then used these impasses to simulate three tutoring conditions to study variation in AI tutor guidance through impasses: baseline, no-direct-answer, and guided tutor. For a sample of 150 impasses, prompt specificity changed pedagogy: a baseline tutor provided the answer directly in 50.7% of responses, a no-direct-answer tutor asked a follow-up question every time, and the guided tutor responded in a wide variety of ways depending on the context. We then analyzed impasse trajectories in authentic interactions, finding that each additional impasse turn lowered the odds of next-turn recovery by 12.7% (AOR = 0.873, p < .001), and early dropouts were caught in recursive concept elicitation before reaching execution. The benefit of questioning decayed as impasses persisted (scripted question x depth AOR = 0.78; follow-up x depth AOR = 0.83), whereas addressing the student's error grew more beneficial (AOR = 1.14); after a failed scripted question, repeating it was followed by recovery in 28.1% of cases, compared with 39.8% when the tutor addressed the error instead. For learning analytics, these findings identify impasse depth and type as observable, turn-level dialogue signals that analytics can use to trigger graduated, state-sensitive assistance in real time.
Chinese Translation
当学生遇到困难时，辅导员面临着援助困境：过早提供帮助可能会妨碍有效的挣扎，而过久不提供帮助则会让学生陷入令人沮丧的持续困境（即原地踏步）。生成性人工智能辅导员越来越多地使用限制性措施来限制答案的提供，但对于这些辅导员在困境持续时的行为知之甚少。我们分析了来自1260个真实会话的20462个学生发言，识别出6630个三种主要类型的困境发言：概念错误、表达的不确定性或寻求帮助。然后，我们利用这些困境模拟了三种辅导条件，以研究人工智能辅导员在困境中的指导变异性：基线、无直接答案和引导辅导。对于150个困境样本，提示的具体性改变了教学法：基线辅导员在50.7%的回应中直接提供答案，无直接答案辅导员每次都提出后续问题，而引导辅导员则根据上下文以多种方式回应。接着，我们分析了真实互动中的困境轨迹，发现每增加一个困境发言，下一轮恢复的几率降低了12.7%（调整后的比值比 = 0.873，p < .001），而早期退学的学生在达到执行前陷入了递归的概念引导中。随着困境的持续，提问的益处逐渐减弱（脚本问题 x 深度 调整后的比值比 = 0.78；后续问题 x 深度 调整后的比值比 = 0.83），而解决学生错误的益处则日益显著（调整后的比值比 = 1.14）；在一次失败的脚本问题后，重复该问题的恢复率为28.1%，而当辅导员解决错误时则为39.8%。对于学习分析而言，这些发现将困境的深度和类型识别为可观察的、逐轮对话信号，分析可以利用这些信号实时触发渐进的、状态敏感的辅助。
cs.AI / 7 / 2609.38359
Beyond Mode Collapse: Generating Diverse Synthetic Expert Conversations via Generative Flow Networks
超越模式崩溃：通过生成流网络生成多样化的合成专家对话
Abstract
High quality synthetic data is central to post training LLMs for adaptive AI applications that represent the diverse expert strategies and decisions in conversations. Prompting LLMs directly or conditioning them on end use scenarios yields low diversity data that collapses onto dominant modes. We propose a method to generate diverse high quality synthetic data using Generative Flow Networks (GFlowNets). We show that training GFlowNets to generate latent conversation structure using a Gaussian mixture density over key interaction features (e.g., confusion episode dynamics, scaffolding directive balance) enables sampling expert strategies in proportion to their prevalence in the training data. Across two structurally distinct domains, tutoring and emotional support dialogues, our GFlow based synthetic data generation approach offers a better balance of fidelity, mode coverage and authenticity than reinforcement-learning and end to end LLM baselines, without copying training data. Evaluated on three downstream outcome prediction tasks, classifiers trained on synthetic GFlowNet generated conversations provide a stronger training signal than competitive synthesis baselines.
Chinese Translation
高质量的合成数据对于后期训练的大型语言模型（LLMs）在适应性人工智能应用中至关重要，这些应用需要在对话中体现多样化的专家策略和决策。直接提示LLMs或将其条件化于最终使用场景会导致低多样性的数据，且数据会集中于主导模式。我们提出了一种使用生成流网络（Generative Flow Networks, GFlowNets）生成多样化高质量合成数据的方法。我们展示了训练GFlowNets以使用关键交互特征（例如，困惑事件动态、支架指令平衡）上的高斯混合密度生成潜在对话结构，使得能够按其在训练数据中的普遍性抽样专家策略。在两个结构上截然不同的领域——辅导和情感支持对话中，我们基于GFlow的合成数据生成方法在保真度、模式覆盖和真实性之间提供了比强化学习和端到端LLM基线更好的平衡，而无需复制训练数据。在三个下游结果预测任务的评估中，基于合成GFlowNet生成的对话训练的分类器提供了比竞争合成基线更强的训练信号。
cs.AI / 8 / 2609.38369
Can an AI Agent Rediscover a Blaschke-Curve Invariant?
人工智能代理能否重新发现 Blaschke 曲线不变性？
Abstract
We study generalized Blaschke curves as a controlled environment for AI-assisted mathematical rediscovery. For one fixed degree-four Blaschke product, an agent receives numerical coordinates of the six pair-lines determined by each of 80 boundary configurations. The target theorem is withheld from the task instructions. The saved research log reports rejected geometric hypotheses and a homogeneous cubic fitted to polygon sides. Its frozen coefficients predict 480 lines from 80 unseen parameter values, with a recorded RMS scale-free residual of $8.88\times10^{-17}$. Discovery-set diagonals provide an out-of-fit consistency check, not a fully held-out test. A separate one-configuration run reports insufficient evidence for invariance. A post-review deterministic degree-search baseline also recovers the cubic, so the experiment does not establish an advantage over polynomial fitting. We present this single-instance case study as a protocol for separating conjecture, numerical validation, and proof, with explicit limitations concerning agent metadata, prior knowledge, and reproducibility.
Chinese Translation
我们研究广义 Blaschke 曲线作为一种受控环境，以实现人工智能辅助的数学重新发现。在一个固定的四度 Blaschke 乘积下，代理接收由 80 种边界配置确定的六对直线的数值坐标。目标定理未在任务指令中透露。保存的研究日志记录了被拒绝的几何假设和拟合多边形边的齐次立方体。其固定系数预测了来自 80 个未见参数值的 480 条直线，记录的均方根无量纲残差为 $8.88 imes10^{-17}$。发现集对角线提供了一种不完全适合的相容性检查，而不是完全保留的测试。单独的一个配置运行报告了不充分的证据以支持不变性。后续审查的确定性度数搜索基线也恢复了立方体，因此该实验并未证明其优于多项式拟合。我们将这一单实例案例研究呈现为一种协议，以区分猜想、数值验证和证明，并明确限制代理元数据、先前知识和可重复性。
cs.AI / 9 / 2609.38372
Self-Evolving Harness on Multiple Tasks with the Agent as Its Own Optimizer
自我进化的多任务 harness 及其作为自我优化器的代理
Abstract
A harness is the code around a language-model agent that organizes prompts, calls tools, manages context, and controls execution. As models grow stronger, recent work has begun to let agents improve their own harnesses, a line of work known as self-evolving harnesses. In most existing methods, a separate proposer running on a human-designed harness modifies the solver's harness, and a separate harness is evolved for each benchmark. Real-world tasks come from many domains, so both the evolution and the evaluation of a harness should cover a diverse range of tasks. We propose a framework close to recursive self-improvement: the same frozen model, on the same version of the harness, first solves tasks as the solver and then, as the proposer, reads the complete run records and directly edits the harness that runs it. Each evolution batch draws tasks from five benchmarks in different domains. To measure generalization, training and held-out tasks are strictly separated, and we additionally evaluate on five out-of-distribution benchmarks never used during evolution. We frame the evolution process as deep-learning training with two stages, multi-task pretraining and continual training. Starting from a 49-line seed harness, the harness obtained at the end of the first stage improves the average score by 4.48 points on the in-distribution benchmarks and by 12.64 points on the out-of-distribution benchmarks, surpassing Codex on the former and matching it on the latter. In the second stage, continued evolution on Claw-Eval, one of the out-of-distribution benchmarks, further raises the score on that benchmark from 66.17 to 68.06, exceeding Codex. We also provide an in-depth analysis of the mechanisms that emerged during evolution, including output truncation, history compaction, and independent review.
Chinese Translation
harness 是围绕语言模型代理的代码，负责组织提示、调用工具、管理上下文和控制执行。随着模型的不断增强，近期的研究开始让代理改善其自身的 harness，这一研究方向被称为自我进化的 harness。在大多数现有方法中，一个在人工设计的 harness 上运行的独立提议者修改求解器的 harness，并且为每个基准进化一个独立的 harness。现实世界的任务来自多个领域，因此 harness 的进化和评估应覆盖多样化的任务。我们提出了一种接近递归自我改进的框架：同一个冻结模型，在同一版本的 harness 上，首先作为求解器解决任务，然后作为提议者读取完整的运行记录并直接编辑运行它的 harness。每个进化批次从五个不同领域的基准中抽取任务。为了测量泛化能力，训练任务和保留任务严格分开，我们还在五个在进化过程中从未使用过的分布外基准上进行评估。我们将进化过程框架化为具有两个阶段的深度学习训练，即多任务预训练和持续训练。从一个49行的种子 harness 开始，第一阶段结束时获得的 harness 在分布内基准上的平均得分提高了4.48分，在分布外基准上提高了12.64分，超越了 Codex 在前者上的表现，并在后者上与之持平。在第二阶段，继续在 Claw-Eval（一个分布外基准）上进化，进一步将该基准的得分从66.17提高到68.06，超过了 Codex。我们还提供了对进化过程中出现的机制的深入分析，包括输出截断、历史压缩和独立审查。
cs.AI / 10 / 2609.38379
Aligned Data Can Induce Misalignment via Context Confusion
对齐的数据可能通过上下文混淆引发不对齐
Abstract
Large language models (LLMs) are frequently updated for various use cases, where filtering out misaligned training samples is a common practice for preventing post-update misalignment. However, alignment is inherently context-dependent: a recommendation that is aligned in one context may be inappropriate in another. For example, in response to the question "What should a researcher do with the research data?", recommending that the researcher preserve the data for reproducibility is aligned. In contrast, recommending data saving in response to "What should a mobile-app developer do with users' sensitive data?" may be inappropriate from a privacy perspective. Starting from this observation, we identify a post-training phenomenon where aligned training induces misaligned behavior in other contexts. We call this phenomenon **context confusion**. We demonstrate context confusion across three domains: (1) Gender Equality, (2) Privacy, and (3) Physical Safety. We further show that context confusion causes narrow misalignment, in contrast to emergent misalignment, and is not effectively reduced by injecting general alignment data, but can be substantially reduced by including targeted alignment data for the misaligned domain or providing in-context learning examples during inference. Lastly, we provide a mechanistic explanation of *context confusion*. We observe that queries from different domains can undergo similar representational shifts during the fine-tuning. Consequently, a query from a different domain may activate the same behavioral feature learned during fine-tuning, which causes the behavior to transfer to a context where it is misaligned. Based on our findings, we argue that it is difficult to predict the alignment state of a model after training by inspecting the training data alone, which highlights the importance of comprehensive post-training alignment evaluations.
Chinese Translation
大型语言模型（LLMs）经常针对各种应用场景进行更新，其中过滤掉不对齐的训练样本是防止更新后不对齐的常见做法。然而，对齐本质上是依赖于上下文的：在一个上下文中对齐的推荐在另一个上下文中可能不合适。例如，针对问题“研究人员应该如何处理研究数据？”推荐研究人员保存数据以确保可重复性是对齐的。相反，针对“移动应用开发者应该如何处理用户的敏感数据？”这一问题，推荐保存数据可能从隐私的角度来看是不合适的。从这一观察出发，我们识别出一种后训练现象，即对齐的训练在其他上下文中引发不对齐的行为。我们称这种现象为**上下文混淆**。我们在三个领域展示了上下文混淆的存在：（1）性别平等，（2）隐私，以及（3）人身安全。我们进一步表明，上下文混淆导致狭义的不对齐，与新兴的不对齐相对，并且通过注入一般对齐数据无法有效减少，但可以通过包含针对不对齐领域的特定对齐数据或在推理过程中提供上下文学习示例来显著减少。最后，我们提供了*上下文混淆*的机制解释。我们观察到来自不同领域的查询在微调过程中可能经历类似的表征转变。因此，来自不同领域的查询可能激活在微调过程中学习到的相同行为特征，这导致行为转移到一个不对齐的上下文中。基于我们的发现，我们认为仅通过检查训练数据很难预测模型在训练后的对齐状态，这突显了全面的后训练对齐评估的重要性。
cs.AI / 11 / 2609.38385
Fine-Tuning Diffusion Language Models with Context Selection and Target Weighting
通过上下文选择和目标加权微调扩散语言模型
Abstract
Supervised fine-tuning of discrete diffusion language models masks some response tokens and trains the model to recover their original values from the visible context. The masking pattern therefore determines both the context available to the model and the tokens it learns to predict. Uniform random masking does not explicitly account for the interaction between these choices. We introduce GoldiMask, which selects tokens to reveal as context by approximately maximizing a submodular objective. This objective uses model signals to balance the benefit of revealing tokens against their value as prediction targets. GoldiMask then weights the remaining targets according to how they benefit from the selected context and their remaining learning potential. Across three backbones and three training datasets, GoldiMask achieves the highest average accuracy in most evaluated settings, demonstrating gains on both reasoning and code generation. Component ablations show that both context selection and target weighting contribute to the gains. GoldiMask also reduces decoding iterations on GSM8K and MATH-500 under confidence-threshold parallel decoding, while maintaining comparable accuracy at higher confidence thresholds.
Chinese Translation
对离散扩散语言模型进行监督微调时，会对部分响应标记进行掩蔽，并训练模型从可见上下文中恢复其原始值。因此，掩蔽模式决定了模型可用的上下文以及其学习预测的标记。均匀随机掩蔽并未明确考虑这些选择之间的相互作用。我们引入了GoldiMask，通过近似最大化一个子模目标来选择作为上下文揭示的标记。该目标利用模型信号来平衡揭示标记的好处与其作为预测目标的价值。GoldiMask随后根据剩余目标如何受益于所选上下文及其剩余学习潜力来加权。通过三种基础模型和三种训练数据集，GoldiMask在大多数评估设置中实现了最高的平均准确率，显示出在推理和代码生成方面的提升。组件消融实验表明，上下文选择和目标加权均对提升效果有所贡献。在信心阈值并行解码下，GoldiMask还减少了GSM8K和MATH-500的解码迭代，同时在较高信心阈值下保持了相当的准确性。
cs.AI / 12 / 2609.38386
Decode-Latency Feedback Prefill: A Model-Free Controller and Its Generalization Limits
解码延迟反馈预填充：一种无模型控制器及其泛化限制
Abstract
Concurrent autoregressive inference creates a fundamental interference problem: prefilling a newly arrived long prompt can delay tokens for requests that are already decoding. Fixed prefill chunks reduce this interference, but the best chunk size depends on the model, hardware, load, and latency objective. We introduce Decode-Latency Feedback Prefill (DLFP), a model-free controller that changes only prefill work that overlaps active decodes. After a guarded scheduling cycle, DLFP uses the observed interval as proportional feedback to resize the next prefill chunk; isolated prefills remain unrestricted. We implement DLFP in vLLM and evaluate it with open-loop Poisson arrivals, exact token accounting, raw request traces, and NVIDIA telemetry. On Qwen3-0.6B in BF16 on one A100 80 GB GPU, three paired 100-request trials reduce P99 inter-token latency by 24.8%, 30.1%, and 28.2% (mean 27.7%, paired 95% confidence interval 21.0% to 34.3%) with exact output agreement, no failures, and unchanged SLO compliance. The benefit is not free: mean P99 time to first token increases 34.8% while remaining inside the declared SLO. Crucially, the mechanism does not generalize to Qwen3-8B, Qwen3-32B, or a two-GPU tensor-parallel configuration. We trace the failure to an asynchronous scheduler-call interval that is only a proxy for completed GPU iteration time. This negative result defines the boundary of the contribution and motivates a completion-timed controller for concurrent CPU and on-device inference. We do not claim mobile-device performance; the present work is a reproducible proof-of-concept and generalization study.
Chinese Translation
并发自回归推断产生了一个基本的干扰问题：预填充一个新到达的长提示可能会延迟已经在解码的请求的令牌。固定的预填充块减少了这种干扰，但最佳块大小取决于模型、硬件、负载和延迟目标。我们引入了解码延迟反馈预填充（Decode-Latency Feedback Prefill, DLFP），这是一种无模型控制器，仅改变与活动解码重叠的预填充工作。在一个受保护的调度周期后，DLFP使用观察到的间隔作为比例反馈来调整下一个预填充块的大小；孤立的预填充保持不受限制。我们在 vLLM 中实现了 DLFP，并通过开放循环的泊松到达、精确的令牌计数、原始请求跟踪和 NVIDIA 监测进行评估。在一台 A100 80 GB GPU 上的 Qwen3-0.6B（BF16），三组配对的 100 请求试验将 P99 令牌间延迟减少了 24.8%、30.1% 和 28.2%（平均 27.7%，配对的 95% 置信区间为 21.0% 到 34.3%），且输出完全一致，没有失败，SLO 合规性保持不变。这一收益并非无代价：平均 P99 首个令牌的时间增加了 34.8%，但仍在声明的 SLO 内。关键是，该机制无法推广到 Qwen3-8B、Qwen3-32B 或两台 GPU 的张量并行配置。我们追踪到失败的原因是异步调度器调用间隔仅是已完成 GPU 迭代时间的代理。这一负面结果定义了贡献的边界，并促使我们为并发 CPU 和设备内推断设计一个基于完成时间的控制器。我们并不声称移动设备性能；目前的工作是一个可重复的概念验证和泛化研究。
cs.AI / 13 / 2609.38392
MetaPersona: Task-Grounded Synthetic Populations from Empirical Social Science
MetaPersona：基于任务的实证社会科学合成人群
Abstract
Personas used to seed LLM social simulations face a cold-start problem: existing methods lack a principled basis for deciding which attributes to include and how to assign their values. As a result, synthetic populations may misrepresent the demographic composition, latent attributes, and dependency structure that shape downstream behavior. We introduce MetaPersona-DB, a dataset of 11,000+ empirical human-subjects studies annotated with task-relevant variables, reported relationships, and aggregate-level population statistics. Building on this resource, we propose MetaPersona, a framework that retrieves task-relevant evidence, constructs literature-derived persona dependency graphs, and samples synthetic populations from empirical priors linking demographics, latent attributes, and outcomes. Across three downstream case studies, three baselines, and three frontier models, results vary by task and model: MetaPersona performs strongly on misinformation belief and AI-tool sentiment, while results on income redistribution are mixed. It also reduces persona-construction cost to under $0.5 per task using GPT-5.2. Finally, we present MetaPersona-Studio, a prototype interactive interface for empirically grounded persona generation.
Chinese Translation
用于启动大型语言模型（LLM）社会模拟的人物角色面临冷启动问题：现有方法缺乏原则性基础来决定应包含哪些属性以及如何分配其值。因此，合成人群可能会错误地代表影响下游行为的人口组成、潜在属性和依赖结构。我们介绍了MetaPersona-DB，这是一个包含11,000多个实证人类研究的数据库，标注了与任务相关的变量、报告的关系和汇总级别的人口统计数据。在此基础上，我们提出了MetaPersona，一个框架，用于检索与任务相关的证据，构建文献衍生的人物依赖图，并从连接人口统计、潜在属性和结果的实证先验中抽样合成人群。在三个下游案例研究、三个基准和三个前沿模型中，结果因任务和模型而异：MetaPersona在错误信息信念和人工智能工具情感方面表现强劲，而在收入再分配方面结果则较为复杂。它还将人物构建成本降低到每个任务低于0.5美元，使用的是GPT-5.2。最后，我们展示了MetaPersona-Studio，一个用于实证基础人物生成的原型交互界面。
cs.AI / 14 / 2609.38397
SimTrace: Grounded Multimodal User Trajectories Generation for Online User Modeling
SimTrace：用于在线用户建模的基于真实数据的多模态用户轨迹生成
Abstract
Virtual clients offer a cost-effective approach to support applications such as A/B testing, recommender system development, and interface evaluation. However, building them requires access to large-scale, semantically faithful, fine-grained online user trajectories. These data are difficult to obtain because proprietary logs are subject to privacy restrictions and small businesses often lack sufficient traffic. Consequently, existing public datasets either abstract away fine-grained user interaction details or preserve rich context but remain platform-specific and small-scale. To address this gap, we propose SimTrace, a framework that generates faithful, fine-grained synthetic multimodal clickstreams through a computer-use client agent that is grounded in real user trajectories and the given web environment. SimTrace anonymizes real interactions and constructs a simulated twin of the given web environment, then uses both to generate synthetic interaction trajectories. Each action is paired with its corresponding web observations and user context, yielding a shareable alternative to confidential logs for developing computer-use agent-style virtual clients. We apply SimTrace to an e-commerce setting and evaluate both its fidelity and downstream utility. SimTrace outperforms competing baselines on 7 out of 8 fidelity metrics. Models trained on synthetic data achieve performance comparable to those trained on real data on downstream tasks such as purchase prediction and recommendation. For next action prediction task, augmenting real data with synthetic data further improves accuracy by 11.0% relative to training on real data alone. We release SimTrace as an open-source package to facilitate research on online user behavior modeling.
Chinese Translation
虚拟客户端为支持 A/B 测试、推荐系统开发和界面评估等应用提供了一种成本效益高的方法。然而，构建这些客户端需要获取大规模、语义真实、细粒度的在线用户轨迹。这些数据难以获取，因为专有日志受到隐私限制，而小型企业通常缺乏足够的流量。因此，现有的公共数据集要么抽象掉细粒度的用户交互细节，要么保留丰富的上下文但仍然特定于平台且规模较小。为了解决这一问题，我们提出了 SimTrace，一个通过计算机使用客户端代理生成真实、细粒度合成多模态点击流的框架，该代理基于真实用户轨迹和给定的网络环境。SimTrace 对真实交互进行匿名化，并构建给定网络环境的模拟双胞胎，然后利用这两者生成合成交互轨迹。每个动作都与其对应的网络观察和用户上下文配对，从而为开发计算机使用代理风格的虚拟客户端提供了一个可共享的替代方案。我们将 SimTrace 应用于电子商务环境，并评估其保真度和下游效用。SimTrace 在 8 个保真度指标中的 7 个上优于竞争基线。基于合成数据训练的模型在下游任务（如购买预测和推荐）上的表现与基于真实数据训练的模型相当。在下一步动作预测任务中，将真实数据与合成数据结合使用，相较于仅基于真实数据训练，准确性提高了 11.0%。我们将 SimTrace 作为开源软件包发布，以促进在线用户行为建模的研究。
cs.AI / 15 / 2609.38409
ArgGYM: A Procedural, Engine-Verified Benchmark for Structured Defeasible Reasoning
ArgGYM：一种程序化的、引擎验证的结构性可否定推理基准
Abstract
Recent progress in large language model reasoning has been driven by benchmarks and reinforcement learning environments with automatically verifiable rewards, particularly in mathematics, code, and formal logic. These settings make model accuracy easier to evaluate and optimize, but it remains unclear how far success under fixed problem specifications and stable evaluation criteria transfers to reasoning outside such domains. Real-world reasoning often proceeds under incomplete and revisable information: conclusions may be supported provisionally, defeated by counter-evidence, reinstated by further arguments, or revised when stronger reasons become available. Reasoning of this kind is generally referred to as defeasible reasoning. We introduce ArgGYM, a procedural benchmark and RLVR-compatible training environment for structured defeasible reasoning. ArgGYM decomposes this reasoning into twelve tasks and grounds task-specific scoring in a symbolic argumentation engine that computes the formal states used to evaluate model outputs. It includes a frozen benchmark of 1,440 verified instances across fifteen curriculum configurations, two argument preference orderings (weakest-link and last-link), and two set orderings (elitist and democratic), while the same generators and verifiers can produce fresh instances for evaluation that reduces dependence on static test sets and for verifiable-reward training. On the frozen benchmark, frontier and open-weight models show sharply different reasoning profiles: they can recover substantial parts of structured answers without solving the complete task, and performance declines in later curriculum configurations with longer dependencies and more interacting structures. We release the benchmark, generators, and verifiers for reproducible evaluation and RLVR training.
Chinese Translation
最近，大型语言模型推理的进展得益于具有自动可验证奖励的基准和强化学习环境，特别是在数学、代码和形式逻辑领域。这些环境使得模型的准确性更容易评估和优化，但在固定问题规格和稳定评估标准下的成功在多大程度上能够转移到这些领域之外的推理仍然不清楚。现实世界的推理通常是在不完整和可修订的信息下进行的：结论可能是暂时支持的，可能被反证驳斥，也可能通过进一步的论证得到恢复，或者在更强的理由出现时被修订。这种推理通常被称为可否定推理。我们介绍了ArgGYM，这是一个程序化基准和与RLVR兼容的结构性可否定推理训练环境。ArgGYM将这种推理分解为十二个任务，并在一个符号论证引擎中建立任务特定的评分，该引擎计算用于评估模型输出的形式状态。它包含一个冻结基准，涵盖了在十五个课程配置下的1,440个经过验证的实例，两个论证偏好排序（最弱链和最后链），以及两个集合排序（精英和民主），同时相同的生成器和验证器可以生成新的实例用于评估，从而减少对静态测试集的依赖，并用于可验证奖励的训练。在冻结基准上，前沿和开放权重模型显示出截然不同的推理特征：它们可以在不解决完整任务的情况下恢复结构答案的相当大部分，并且在后期课程配置中，随着更长的依赖关系和更多的交互结构，性能下降。我们发布了基准、生成器和验证器，以便进行可重复的评估和RLVR训练。
cs.AI / 16 / 2609.38411
A Competing-Hazards Systematization of Loss of Control in Autonomous Agents
自主智能体失控的竞争风险系统化研究
Abstract
Leading AI developers have reported agents acting beyond their approved limits, which a United Nations panel described as an early warning of loss of human control. Yet incident reports and agent-safety evaluations describe these events differently, making it difficult to compare failures, trace risk across attempts, or separate agent behavior from the environment's role in allowing an out-of-scope action to succeed. To address this gap, we introduce a common framework in which each attempt ends in approved completion, safe stopping, scope escape, or continuation. We formalize the framework as a discrete-time competing-hazards model and derive escape probability within a retry budget, a model-conditional safe-budget limit, and conditions for estimation from execution logs. We audit 22 incident reports and 102 agent-safety evaluations published from January 2025 to September 2026 using primary sources. Six incidents involved tasks that could not be completed within scope, thirteen involved agents that continued rather than stopped, and five did not report stopping behavior. Developers' figures imply a task-level incidence ratio near 47 for out-of-scope coordination in never-solved versus solved tasks. Among evaluations, 87 recorded an out-of-scope effect or specification violation, 26 treated safe stopping as a first-class outcome, only 20 recorded both, and 79 merged budget exhaustion with failure. In 20 of 22 incidents, the environment allowed an out-of-scope effect, indicating that realized loss of control often reflected persistent agent behavior interacting with permissive boundary conditions; meanwhile, no evaluation reported all fields needed to estimate the full competing-hazards process from published evidence.
Chinese Translation
领先的人工智能开发者报告称，智能体的行为超出了其批准的限制，联合国小组将此描述为失去人类控制的早期警告。然而，事件报告和智能体安全评估对这些事件的描述各不相同，这使得比较失败、追踪风险以及将智能体行为与环境在允许超出范围的行为成功方面的作用区分开来变得困难。为了解决这一问题，我们提出了一个通用框架，其中每次尝试的结果可以是批准完成、安全停止、超出范围或继续进行。我们将该框架形式化为一个离散时间竞争风险模型，并推导出在重试预算内的逃逸概率、模型条件下的安全预算限制以及从执行日志中进行估计的条件。我们审计了2025年1月至2026年9月期间发布的22份事件报告和102份智能体安全评估，使用了主要来源。六起事件涉及无法在范围内完成的任务，十三起涉及继续而非停止的智能体，五起未报告停止行为。开发者的数据显示，在未解决与已解决任务中，超出范围协调的任务级发生率接近47。在评估中，87份记录了超出范围的影响或规范违反，26份将安全停止视为一种重要结果，只有20份同时记录了这两者，79份将预算耗尽与失败合并。在22起事件中，有20起的环境允许超出范围的影响，这表明实现的失控往往反映了持续的智能体行为与宽松的边界条件的相互作用；与此同时，没有评估报告所有需要的字段以从已发布的证据中估计完整的竞争风险过程。
cs.AI / 17 / 2609.38445
AIM: Agentic Idea Management for Automated Research
AIM：用于自动化研究的代理思想管理
Abstract
Frontier LLMs are increasingly used to automate scientific research through iterative search. We distinguish idea-driven search from solution-driven search and identify three core challenges: organizing evolving research ideas, selecting promising directions, and maintaining alignment between ideas and their implementations. To address these challenges, we introduce the Agentic Idea Manager (AIM), a fully autonomous framework for managing and exploring research directions in idea-driven automated research. Inspired by Bayesian optimization, AIM uses an Agentic Surrogate and an Agentic Acquisition mechanism to organize discovered ideas and guide their selection. A Solution Auditor maintains idea-solution integrity, while a Resource Planner adaptively allocates the remaining experimental budget across parallel search branches. Experiments on 10 AutoLab benchmark tasks show that AIM surpasses the strongest baseline by 1.6 percentage points on System Optimization tasks and 4.9 percentage points on long-horizon Model Development & CUDA tasks. Notably, AIM reaches the best baseline performance up to 3.1x faster in wall-clock time. We further provide a theoretical analysis of when searching over ideas becomes beneficial. Our analysis shows that explicit idea-level allocation makes semantic coverage directly controllable, and that broader coverage becomes increasingly valuable when competitive research directions are sparse among many plausible alternatives. Project Page: https://imhgchoi.github.io/agentic-idea-manager/
Chinese Translation
前沿的大型语言模型（LLMs）越来越多地用于通过迭代搜索来自动化科学研究。我们将以思想为驱动的搜索与以解决方案为驱动的搜索区分开来，并识别出三个核心挑战：组织不断发展的研究思想、选择有前景的研究方向，以及保持思想与其实现之间的对齐。为了解决这些挑战，我们引入了代理思想管理器（Agentic Idea Manager，AIM），这是一个完全自主的框架，用于在以思想为驱动的自动化研究中管理和探索研究方向。AIM受到贝叶斯优化的启发，使用代理替代物（Agentic Surrogate）和代理获取机制（Agentic Acquisition）来组织发现的思想并指导其选择。解决方案审计员（Solution Auditor）维护思想与解决方案的一致性，而资源规划者（Resource Planner）则自适应地在并行搜索分支中分配剩余的实验预算。在10个AutoLab基准任务上的实验表明，AIM在系统优化任务上比最强基线超出1.6个百分点，在长时间模型开发与CUDA任务上超出4.9个百分点。值得注意的是，AIM在墙钟时间上以最高3.1倍的速度达到了最佳基线性能。我们进一步提供了关于何时在思想上进行搜索变得有益的理论分析。我们的分析表明，明确的思想级分配使语义覆盖直接可控，而当竞争性研究方向在众多合理替代方案中稀疏时，更广泛的覆盖变得越来越有价值。项目页面：https://imhgchoi.github.io/agentic-idea-manager/
cs.AI / 18 / 2609.38448
Reach Into The CHOIR: Free-List Elicitation Uncovers Distinct Model Voices in LLM Ensembles
深入CHOIR：自由列表引导揭示LLM集成中的独特模型声音
Abstract
Open-ended LLM homogeneity can create false plurality when several systems appear to offer independent perspectives while returning the same familiar default. Single-pass answers obscure the distinction between agreement produced by a tightly constrained answer space, prompt-vocabulary echo, and broader answer spaces with stable alternatives beneath the surface. We introduce CHOIR (Collective Hierarchically-Ordered Inquiry Responses), a framework that adapts free-list elicitation from cognitive anthropology to LLM ensembles. CHOIR repeatedly elicits ranked lists, clusters items into prompt-level concepts, and measures concept salience across models, prompt variants, and persona conditions. We evaluate CHOIR on Infinity-Chat 100, an external prompt bank from recent work on open-ended model homogeneity, and on a 27-question targeted diagnostic bank designed to isolate mechanism-level contrasts. On Infinity-Chat 100, CHOIR reproduces high surface agreement (93/100 prompts above chance) while separating narrow prompts from broad prompts with recoverable depth. Across targeted probes and the external prompt bank, base-model identity remains the strongest recoverable signature, and persona prompts shift surfaced concepts within base-model signatures. A source-blind ranking module prioritises rare-but-stable candidates for later inspection. CHOIR turns open-ended homogeneity into a diagnostic measurement problem by asking where models converge, why they converge, and what remains reachable under structured depth probing.
Chinese Translation
开放式LLM的同质性可能会造成虚假的多样性，当多个系统似乎提供独立视角时，实际上却返回相同的熟悉默认答案。单次回答模糊了由严格限制的答案空间、提示词汇回声以及表面下稳定替代方案所产生的同意之间的区别。我们引入了CHOIR（集体分层有序询问响应），这是一个将认知人类学中的自由列表引导方法应用于LLM集成的框架。CHOIR重复引导排名列表，将项目聚类为提示级概念，并测量模型、提示变体和角色条件下的概念显著性。我们在Infinity-Chat 100上评估CHOIR，这是一个来自近期关于开放式模型同质性的外部提示库，并在一个旨在隔离机制级对比的27个问题的针对性诊断库上进行评估。在Infinity-Chat 100上，CHOIR重现了高表面一致性（93/100个提示高于偶然水平），同时将狭窄提示与具有可恢复深度的广泛提示区分开来。在针对性探测和外部提示库中，基础模型身份仍然是最强的可恢复特征，而角色提示则在基础模型特征内转移了表面概念。一个源盲排名模块优先考虑稀有但稳定的候选项以供后续检查。CHOIR通过询问模型收敛的地方、收敛的原因以及在结构化深度探测下仍可达到的内容，将开放式同质性转化为诊断测量问题。
cs.AI / 19 / 2609.38458
PrivMeSA: Privacy-Aware Self-Evolving Multi-Agent System for Medicine via Local-Remote LLM Collaboration
PrivMeSA：通过本地-远程大语言模型协作的隐私意识自我演化多智能体系统在医学中的应用
Abstract
Clinical large language model (LLM) agents deployed locally can consult more capable remote models, but doing so risks exposing patient information. Privacy-conscious delegation places disclosure decisions with a local agent, yet removing explicit identifiers is insufficient: quasi-identifiers can accumulate across multi-turn consultations and repeated patient visits to enable re-identification. We introduce PrivMeSA, a privacy-aware self-evolving multi-agent system that learns to control disclosure and retains remote expertise for local reuse. A local agent manages each encounter and consults remote specialists that may request additional information. Reinforcement learning balances task accuracy against direct disclosure and registry-based re-identification risk, with privacy evaluated over the complete outbound transcript of each encounter. A local lesson memory distills completed consultations into generalized clinical guidance and retrieves relevant lessons before transmission, allowing subsequent cases to reuse expertise without another remote exchange. Memory grows without additional outcome labels or parameter updates. On an emergency-department benchmark built from MIMIC-IV-ED records, PrivMeSA improves mean task accuracy over delegation by up to 15.8 percentage points. In the same setting, PrivMeSA reduces the disclosure of personal details from 98.0% to 0.2% of cases and the share of cases in which the patient can be narrowed to ten or fewer registry patients from 74% to 0%.
Chinese Translation
本地部署的临床大语言模型（LLM）智能体可以咨询更强大的远程模型，但这样做存在暴露患者信息的风险。隐私意识的委托将披露决策交给本地智能体，但仅仅去除显式标识符是不够的：准标识符可以在多轮咨询和重复患者就诊中积累，从而实现重新识别。我们提出了PrivMeSA，这是一种隐私意识的自我演化多智能体系统，能够学习控制信息披露并保留远程专业知识以供本地重用。一个本地智能体管理每次接触，并咨询可能要求额外信息的远程专家。强化学习在任务准确性与直接披露和基于注册的重新识别风险之间进行平衡，隐私评估基于每次接触的完整外发记录。本地学习记忆将已完成的咨询提炼为通用临床指导，并在传输前检索相关经验，使后续案例能够在不进行另一次远程交流的情况下重用专业知识。记忆在没有额外结果标签或参数更新的情况下增长。在基于MIMIC-IV-ED记录构建的急诊科基准测试中，PrivMeSA将任务平均准确性提高了最多15.8个百分点。在同一环境下，PrivMeSA将个人信息披露从98.0%降低到0.2%，并将患者能够缩小到十名或更少注册患者的案例比例从74%降低到0%。
cs.AI / 20 / 2609.38460
NAQD Env: A benchmark for selective withdrawal in language agents
NAQD Env：语言代理选择性撤回的基准测试
Abstract
Language agents must revise planned actions when evidence changes, permission is revoked, or a stop instruction arrives. A useful response is selective: suspend affected actions, preserve unaffected work, and resume only after sufficient repair. We introduce NAQD-Env, a synthetic environment that evaluates these decisions against a deterministic reference policy over explicit evidence, authorization, and constraint dependencies. Eleven dependency families support evaluation on development structures, held-out families, and held-out combinations of structures. Metrics distinguish attempted violations from violations permitted by a simulated execution gate and jointly report policy agreement, task value, withdrawal, resumption, and event reporting. We evaluate three open-weight instruction-tuned models from two families under three prompt conditions on 350 frozen scenarios, yielding 3,150 model-prompt episodes before gate replay. Across the reported conditions, withdrawal recall is at most 0.06, no valid resumption is observed at eligible opportunities, and only one episode matches the complete reference policy. Under the NAQD prompt, Qwen2.5-7B has fewer unsafe-attempt episodes than Qwen2.5-3B and Llama-3.1-8B, but also completes less useful work and preserves unaffected actions less accurately. Exploratory supervised fine-tuning probes increase Qwen2.5-3B decision accuracy from 0.45-0.54 to 0.83-0.92; separate diagnostics reveal inappropriate withdrawal after curriculum omissions and a loss of event reporting. These results motivate evaluating selective withdrawal as a distinct component of agent reliability. The setting measures policy application with trusted structured inputs and does not establish real-world containment or source-verification ability.
Chinese Translation
语言代理在证据变化、权限被撤销或接收到停止指令时必须修订计划的行动。一种有效的响应是选择性的：暂停受影响的行动，保留未受影响的工作，并在修复足够后再恢复。我们介绍了NAQD-Env，这是一种合成环境，用于评估这些决策与明确证据、授权和约束依赖关系下的确定性参考策略的对比。十一种依赖关系家族支持在开发结构、保留的家族和保留的结构组合上进行评估。指标区分尝试的违规行为与由模拟执行门允许的违规行为，并共同报告政策一致性、任务价值、撤回、恢复和事件报告。我们在350个冻结场景下评估了来自两个家族的三种开放权重指令调优模型，在三种提示条件下产生了3150个模型-提示回合。在报告的条件下，撤回召回率最多为0.06，在符合条件的机会中未观察到有效恢复，且只有一个回合与完整的参考政策相匹配。在NAQD提示下，Qwen2.5-7B的危险尝试回合少于Qwen2.5-3B和Llama-3.1-8B，但完成的有用工作也较少，且对未受影响的行动的保留准确性较低。探索性监督微调探测将Qwen2.5-3B的决策准确性从0.45-0.54提高到0.83-0.92；单独的诊断揭示了在课程遗漏后不当撤回和事件报告的丧失。这些结果激励我们将选择性撤回作为代理可靠性的一个独特组成部分进行评估。该设置使用可信的结构化输入来衡量政策应用，并未建立现实世界的控制或源验证能力。
cs.AI / 21 / 2609.38512
VAmoS Part Deux: Harder, More Realistic Voice-Agent Simulation
VAmoS 第二部分：更具挑战性和更真实的语音代理模拟
Abstract
Voice agents in production must handle several requests, background speech, and customers who lose patience. We introduce VAmoS Energy, a benchmark that combines these challenges in 100 calls about utility billing and payment assistance. Each caller makes two to four requests. The agent has sixteen tools backed by a stateful Stripe billing twin and the Apache Fineract loan engine, with account access blocked until caller verification succeeds. The tasks use public household electricity data and a policy based on Pennsylvania's residential billing rules. An LLM-as-a-verifier checks the agent's actions and spoken figures against explicit requirements. On a calibration run, it agrees with a code verifier on 99.1% of checks. Across fourteen voice stacks and three repeats per task, completion ranges from 17.3% to 44.7%. Grok Voice leads, and Gemini 3.8 Live and GPT-Live follow at about the same cost per call. Background television reduces pooled completion from 38.7% to 8.6%. The simulated caller often accepts an incorrect result because it hears the agent's words but cannot inspect its actions. These findings show why voice agents need evaluation across the whole call, including what they say, what they change, and how they handle competing speech.
Chinese Translation
在实际应用中，语音代理必须处理多个请求、背景语音以及失去耐心的客户。我们推出了 VAmoS Energy，一个基准测试，结合了这些挑战，涵盖了关于公用事业账单和支付援助的 100 个电话。每位来电者提出两到四个请求。代理拥有十六个工具，支持一个有状态的 Stripe 账单双胞胎和 Apache Fineract 贷款引擎，账户访问在来电者验证成功之前被阻止。任务使用公共家庭电力数据和基于宾夕法尼亚州住宅账单规则的政策。一个作为验证者的 LLM（大语言模型）检查代理的行为和口头数字是否符合明确要求。在一次校准运行中，它与代码验证者在 99.1% 的检查中达成一致。在十四个语音堆栈和每个任务三次重复的情况下，完成率从 17.3% 到 44.7% 不等。Grok Voice 领先，Gemini 3.8 Live 和 GPT-Live 的成本大致相同。背景电视将整体完成率从 38.7% 降低到 8.6%。模拟的来电者通常接受错误结果，因为它听到了代理的言辞，但无法检查其行为。这些发现表明，语音代理需要在整个通话过程中进行评估，包括他们所说的内容、所做的更改以及如何处理竞争性语音。
cs.AI / 22 / 2609.38555
Demographic Pluralism: Inference-Time Modeling of Pluralistic Human Preference Distributions
人口多元主义：多元化人类偏好分布的推理时建模
Abstract
Large language models (LLMs) are increasingly used in culturally sensitive settings, where alignment requires representing diverse preferences within populations. Yet existing methods model populations at coarse demographic or community levels and overlook within-group variation. We introduce Demographic Pluralism, an inference-time framework that estimates population-level opinion distributions without opinion-distribution training data or task-specific fine-tuning by generating multiple perspectives within demographically grounded groups. Across four backbones on GlobalOpinionQA and VITAL, it reduces Jensen-Shannon distance by 8.4%-26.4% over Modular Pluralism. Among weighted, equal-weighted, and inverse-weighted aggregation, equal weighting performs best overall; group-level error also increases with group weight, helping explain weighted aggregation's weaker performance.
Chinese Translation
大型语言模型（LLMs）在文化敏感的环境中越来越多地被使用，其中对齐需要代表人群中的多样化偏好。然而，现有方法在粗略的人口或社区层面建模人群，忽视了组内的变异性。我们提出了人口多元主义（Demographic Pluralism），这是一种推理时框架，能够在没有意见分布训练数据或特定任务微调的情况下，估计人口级别的意见分布，通过在以人口为基础的群体中生成多种视角。在 GlobalOpinionQA 和 VITAL 的四个基础模型上，它比模块化多元主义（Modular Pluralism）减少了 8.4%-26.4% 的 Jensen-Shannon 距离。在加权、等权重和逆加权聚合中，等权重的表现最佳；而组级误差随着组权重的增加而增加，这有助于解释加权聚合的较弱表现。
cs.AI / 23 / 2609.38559
Defining and Categorising Human-AI Interactions in Clinical Trials: A Multidimensional Human-AI Classification Approach
在临床试验中定义和分类人机交互：一种多维人机分类方法
Abstract
This paper examines human-AI interactions (HAIIs) in clinical trials and presents a multidimensional categorisation framework that classifies interactions according to AI tasks, human-AI relationships, interaction configurations and interacting human groups. We define HAII, examine existing taxonomies and extend existing categorisation approaches through this novel multidimensional framework. We purposively sampled 15 clinical trials from a previously reported dataset. Each trial was independently categorised by two human reviewers and six large language model (LLM) classifiers. The proposed categorisation provides a structured method for the consistent identification, comparison and synthesis of human-AI interactions across clinical-trial records. The framework is intended to support more consistent comparison and synthesis of AI-related clinical trials and to make explicit the different forms of human involvement associated with AI interventions. The results demonstrate the potential for LLM-assisted categorisation while indicating the continuing importance of human judgement where trial records are incomplete or ambiguous. The principal contribution is a proposed multidimensional framework that brings together AI tasks, human-AI relationships, interaction configurations and interacting human groups within a single approach designed for clinical-trial records. Its significance lies in its potential to support more systematic identification, comparison and synthesis of how humans and AI interact in clinical trials.
Chinese Translation
本文研究了临床试验中的人机交互（HAIIs），并提出了一种多维分类框架，根据人工智能任务、人机关系、交互配置和交互人群对交互进行分类。我们定义了HAII，审视了现有的分类法，并通过这一新颖的多维框架扩展了现有的分类方法。我们从先前报告的数据集中有目的地抽取了15个临床试验。每个试验由两名人工审阅者和六个大型语言模型（LLM）分类器独立分类。所提出的分类方法为在临床试验记录中一致地识别、比较和综合人机交互提供了一种结构化的方法。该框架旨在支持AI相关临床试验的更一致比较和综合，并明确与AI干预相关的人类参与的不同形式。结果表明了LLM辅助分类的潜力，同时指出在试验记录不完整或模糊的情况下，人类判断的重要性依然存在。主要贡献是提出了一个多维框架，将AI任务、人机关系、交互配置和交互人群整合在一个旨在处理临床试验记录的单一方法中。其重要性在于有潜力支持更系统地识别、比较和综合人类与AI在临床试验中的交互方式。
cs.AI / 24 / 2609.38574
Towards Model as a Library: Offline, Community-Sourced AI for Low-Resource African Languages
迈向模型即库：面向低资源非洲语言的离线社区驱动人工智能
Abstract
Large language models are frequently proposed as a route to AI-powered services for African communities, but they are least reliable exactly where the need is greatest: all African languages remain low-resource by any standard measure, and models trained on scraped, standardised text systematically misrepresent the dialectal and regional variation of how people actually speak. We introduce \textbf{Model as a Library (MaaL)}, a software architecture that packages small, community-enrolled speech models as versioned on-device dependencies, enabling offline structured data collection that cannot generatively hallucinate, for populations that current language models serve worst. Rather than relying on web-scraped corpora, MaaL's vocabulary is enrolled directly from a small number of example recordings by the speakers themselves, at the point of deployment. We describe the architecture and its central mechanism - keyword spotting that turns a closed-vocabulary text form into a voice form, filled and submitted entirely on-device - and propose transpiling the closed-vocabulary elements already present in widely-deployed digital form tools into MaaL schemas, a low-friction path to voice-first, offline data collection for the low-literacy populations these tools already reach. This is a position and system-design paper: we describe the concept, the mechanism, and an analytical feasibility case, and identify what a working implementation still requires.
Chinese Translation
大型语言模型常被视为为非洲社区提供人工智能服务的途径，但恰恰在需求最迫切的地方，它们的可靠性最低：按任何标准衡量，所有非洲语言仍属低资源语言，且基于网络爬取和标准化文本训练的模型系统性地误判了人们实际使用的方言和区域变体。我们提出了“模型即库（Model as a Library，MaaL）”这一软件架构，将小型社区参与的语音模型作为版本化的设备端依赖包，支持离线结构化数据采集，避免生成式幻觉，专为当前语言模型服务最差的人群设计。MaaL不依赖网络爬取语料，其词汇表由说话者本人在部署时直接从少量示例录音中录入。我们描述了该架构及其核心机制——关键词检测，将封闭词汇文本形式转换为完全在设备端完成填充和提交的语音形式，并提出将已广泛部署的数字表单工具中存在的封闭词汇元素转译为MaaL模式，为这些工具已覆盖的低识字率人群提供一种低门槛的语音优先离线数据采集路径。本文为立场与系统设计论文：我们阐述了概念、机制及分析性可行性案例，并指出实现可用系统所需的关键条件。
cs.AI / 25 / 2609.38577
Conditional Generation of Creative Chess Puzzles with Diffusion Models
基于扩散模型的创意国际象棋难题的条件生成
Abstract
While modern language models demonstrate impressive generative capabilities, they often struggle with constrained, counter-intuitive creative tasks. To address this limitation, we explore chess puzzle generation as a rigorous testbed for computational creativity and reasoning, a domain where altering a single piece can invalidate an entire solution. We propose a novel approach for conditional generation of creative chess puzzles using masked diffusion models. Unlike previous methods, our non-directional diffusion approach allows for conditioning on specific tactical themes and partial board positions. We introduce a novel auxiliary task of simultaneous best-move prediction, which improves solution uniqueness by 11.6% and theme-conditioning accuracy by 2.5%. To further optimize solution uniqueness and theme conditioning, we establish a reinforcement learning framework adapted from Denoising Diffusion Policy Optimization (DDPO). This RL training increases the yield of unique and theme-matching positions by 89.1%. Finally, we release the first open-weights models (Appendix B) for chess puzzle generation, offering a new pathway for controllable, creative generation.
Chinese Translation
尽管现代语言模型展现了令人印象深刻的生成能力，但它们在受限的、反直觉的创意任务中往往表现不佳。为了解决这一局限性，我们将国际象棋难题生成作为计算创造力和推理的严格测试平台，这一领域中，改变一个棋子可能会使整个解答失效。我们提出了一种使用掩蔽扩散模型进行创意国际象棋难题条件生成的新方法。与之前的方法不同，我们的无方向扩散方法允许在特定的战术主题和部分棋盘位置上进行条件生成。我们引入了一种新的辅助任务，即同时最佳走法预测，这使得解答的独特性提高了11.6%，主题条件的准确性提高了2.5%。为了进一步优化解答的独特性和主题条件，我们建立了一个基于去噪扩散策略优化（Denoising Diffusion Policy Optimization, DDPO）改编的强化学习框架。该强化学习训练使得独特且符合主题的棋盘位置的产出提高了89.1%。最后，我们发布了首个开放权重模型（附录B）用于国际象棋难题生成，为可控的创意生成提供了一条新路径。
cs.AI / 26 / 2609.38600
Sense and Sensitivity: Benchmarking LLM Clinical Triage Recommendations with Physician Experts
感知与敏感性：与医学专家对比大型语言模型临床分诊建议的基准测试
Abstract
As large language models (LLMs) are increasingly used in clinical settings, it is critical to evaluate their reliability under realistic variation in clinical text. We study this question in clinical triage, comparing LLMs to practicing physicians under text perturbations that preserve the underlying clinical setting. We introduce a benchmark of over 6,000 clinical scenarios, 7,000 physician annotations, and 225,000 model responses. Using this benchmark, we make two key observations. First, LLMs are more likely than physicians to recommend unnecessary care at baseline, and this tendency increases under perturbed inputs. Further, we find that LLM recommendations are more sensitive to gender and tone perturbations than human recommendations. Together, these results demonstrate that LLMs can vary under clinically irrelevant textual changes, highlighting the need for deployment-oriented evaluations grounded in expert physician behavior.
Chinese Translation
随着大型语言模型（LLMs）在临床环境中的应用日益增多，评估其在临床文本真实变异下的可靠性变得至关重要。我们在临床分诊中研究了这一问题，将LLMs与在文本扰动下保持基础临床环境的执业医生进行比较。我们引入了一个包含超过6000个临床场景、7000个医生注释和225,000个模型响应的基准数据集。利用这一基准，我们得出了两个关键观察结果。首先，LLMs在基线情况下比医生更可能推荐不必要的护理，而这种倾向在输入扰动下会增加。此外，我们发现LLM的建议对性别和语气扰动的敏感性高于人类建议。综合来看，这些结果表明，LLMs在临床无关的文本变化下可能会有所不同，强调了基于专家医生行为的部署导向评估的必要性。
cs.AI / 27 / 2609.38621
When Scientific Contradictions Are Lost in Translation
当科学矛盾在翻译中消失时
Abstract
Two scientific findings can disagree without contradicting each other. Determining whether they conflict requires knowing whether they describe comparable measurements. We study how language models behave at this decision point. In a controlled task, we generate an unsatisfiable XOR constraint system and translate its constraints into scientific reports from different laboratories. One assignment satisfies more constraints, while another satisfies fewer but better matches expected biology. This creates a simple dilemma: does the model choose the assignment that best fits the constraints, or the one that better matches biological expectations? When the constraints are stated directly, GPT-5.6 Sol and Claude Opus 5 recover the best-supported assignment in 90% and 96% of cases, respectively. In scientific prose, however, the models behave differently. Claude Opus 5 often prefers the biologically expected assignment. Removing that biological preference increases recovery of the better-supported assignment from 27% to 79% (p<.001); recovery reaches 92% when the same Biology-favored record is accompanied by a formalization request and an explicit paired-design cue (p<.001). GPT-5.6 Sol is less sensitive, with neither corresponding change reaching statistical significance. These results suggest that reliable scientific verification depends not only on formal reasoning, but also on how models decide which findings should be compared and what relations they imply.
Chinese Translation
两个科学发现可以存在不一致而并不矛盾。确定它们是否冲突需要了解它们是否描述了可比较的测量。我们研究语言模型在这一决策点的表现。在一个受控任务中，我们生成一个不可满足的异或约束系统，并将其约束翻译成来自不同实验室的科学报告。一个分配满足更多约束，而另一个分配满足较少但更符合预期生物学。这产生了一个简单的困境：模型是选择最符合约束的分配，还是选择更符合生物学预期的分配？当约束直接表述时，GPT-5.6 Sol 和 Claude Opus 5 在90%和96%的案例中恢复了最佳支持的分配。然而，在科学文献中，这些模型的表现却有所不同。Claude Opus 5 通常倾向于选择生物学上预期的分配。去除这种生物学偏好使得更好支持的分配的恢复率从27%提高到79%（p<.001）；当同一生物学偏好的记录伴随正式化请求和明确的配对设计提示时，恢复率达到92%（p<.001）。GPT-5.6 Sol 对此不太敏感，相关变化均未达到统计显著性。这些结果表明，可靠的科学验证不仅依赖于形式推理，还依赖于模型如何决定哪些发现应该被比较以及它们暗示的关系。
cs.AI / 28 / 2609.38639
Component-Aware Feedback for Self-Evolving Programs
面向组件的自我进化程序反馈
Abstract
LLM-guided evolutionary search can discover complex programs, but existing methods mostly only save candidate programs and fitness scores while discarding which component edits produced which fitness metric changes. Existing methods force the mutator LLM to infer the effect of prior edits from cluttered histories, making program search slow and unstable. This is especially true for locally servable LLMs to evolve multi-component systems. We introduce component-aware feedback, which compares each evaluated program with its parent, identifies the components that changed, and logs them with the associated metric differences into an attribution memory that later mutations read. The memory keeps each change in two reference frames, local against the parent it came from and global against the seed program, which shows both the immediate effect of a change and the cumulative progress made since the seed. We study this on LLM reranking, a multi-objective optimization problem where a multi-stage pipeline must balance quality against serving cost. Across twelve \textsc{Bright} datasets, our method reaches the strongest baseline's final quality after a median of one third of the search budget and ends 7.2\% higher in held-out nDCG@10, and under a cost-aware objective it finds pipelines that are on average more accurate while using 11\% fewer tokens per query, showing component-aware feedback to be a promising direction for more efficient self-evolving systems.
Chinese Translation
由大型语言模型（LLM）指导的进化搜索能够发现复杂程序，但现有方法大多仅保存候选程序和适应度分数，而忽略了哪些组件编辑导致了适应度指标的变化。现有方法迫使变异器 LLM 从杂乱的历史记录中推断先前编辑的效果，这使得程序搜索变得缓慢且不稳定。这对于局部可服务的 LLM 进化多组件系统尤其如此。我们引入了面向组件的反馈，它将每个评估程序与其父程序进行比较，识别出发生变化的组件，并将其与相关的指标差异记录到归因记忆中，以便后续的变异读取。该记忆在两个参考框架中保留每个变化：相对于其来源的父程序的局部框架和相对于种子程序的全局框架，这显示了变化的即时效果以及自种子以来所取得的累积进展。我们在 LLM 重新排序上研究了这一点，这是一个多目标优化问题，其中多阶段管道必须在质量与服务成本之间取得平衡。在十二个 extsc{Bright} 数据集上，我们的方法在搜索预算的中位数三分之一后达到了最强基线的最终质量，并在保留的 nDCG@10 中高出 7.2\%。在一个关注成本的目标下，它找到的管道在平均上更准确，同时每个查询使用的标记减少了 11 ext{%}，显示出面向组件的反馈是更高效自我进化系统的一个有前景的方向。
cs.AI / 29 / 2609.38642
ChartRevise: A Dataset and Evaluation Protocol for Exact Chart Editing via Code
ChartRevise：用于通过代码进行精确图表编辑的数据集和评估协议
Abstract
Chart editing requires cross-modal edit grounding, realizing a requested visual change in the code that draws it, with necessary related updates and without altering unrelated content. Existing benchmarks emphasize either code executability or chart quality, but their metrics do not clearly distinguish request completion from missed coupled updates and gratuitous changes. We introduce ChartRevise, a structured dataset and evaluation protocol for exact program-grounded chart editing. For dataset construction, we build on the grammar of graphics to systematically cover chart-editing operations, using source-program checks to verify their applicability across chart types and libraries. To improve edit exactness, our pipeline checks individual requirements and guides repair or exclusion when they are unmet. The resulting dataset contains 92,438 records covering 344 edit types across 20 chart types and three plotting libraries. For evaluation, our reference-free protocol separately measures atomic requirement completion, identifies gratuitous changes, and detects missed coupled updates. These checks are combined with successful execution and rendering to determine exact-edit success. Across five models and four external benchmarks, fine-tuning yields relative gains of 16\% in mean requirement recall and 22\% in mean exact-edit rate.
Chinese Translation
图表编辑需要跨模态编辑基础，将请求的视觉变化实现于绘制该图表的代码中，同时进行必要的相关更新而不改变无关内容。现有基准测试强调代码可执行性或图表质量，但其指标并未清晰区分请求完成与遗漏的耦合更新和无意义的变化。我们引入了ChartRevise，一个结构化的数据集和精确程序基础图表编辑的评估协议。在数据集构建方面，我们基于图形语法系统性地覆盖图表编辑操作，使用源程序检查来验证其在不同图表类型和库中的适用性。为了提高编辑的准确性，我们的流程检查个别要求，并在未满足时指导修复或排除。最终生成的数据集包含92,438条记录，涵盖20种图表类型和344种编辑类型。对于评估，我们的无参考协议分别测量原子要求的完成情况，识别无意义的变化，并检测遗漏的耦合更新。这些检查与成功的执行和渲染相结合，以确定精确编辑的成功率。在五个模型和四个外部基准测试中，微调带来了平均要求召回率提高16 ext{%}和平均精确编辑率提高22 ext{%}的相对增益。
cs.AI / 30 / 2609.38652
AgBench: Agentic AI Benchmarks for Personal AI Devices
AgBench：个人人工智能设备的代理人工智能基准测试
Abstract
Agentic AI systems increasingly rely on cloud-hosted large language models for planning, tool use, and iterative execution, raising concerns about API cost and data exposure. Advances in personal AI devices enable agents to execute locally, but limited resources on device may affect task success and performance. Existing benchmarks are inadequate for systematically characterizing these trade-offs across devices, workloads, and deployment architectures. We present AgBench, a benchmark suite and open artifacts for reproducible evaluation of agentic AI on personal devices. Using AgBench, we evaluate local, hybrid, and cloud execution across agentic workloads, examining task success, latency, cloud API cost, and data exposure. Our results, drawn from over 162.07 million data points, show that personal AI devices can complete many agent tasks locally, but local-only execution generally has lower task success and longer completion times than cloud-only execution, especially as concurrency increases. Local-only execution eliminates cloud model API costs and sensitive-information exposure to cloud agents. Hybrid execution can improve task success, but its cloud cost and data exposure depend on how agents divide work and share information. No single architecture performs best across task success, goodput, cloud cost, and data exposure; deployment choices should reflect the intended workload and device capabilities. AgBench is available at https://anonymous.4open.science/r/AgBench-2777.
Chinese Translation
代理人工智能系统越来越依赖于云托管的大型语言模型进行规划、工具使用和迭代执行，这引发了对API成本和数据暴露的担忧。个人人工智能设备的进步使得代理能够在本地执行，但设备上的有限资源可能会影响任务成功率和性能。现有基准测试不足以系统性地描述不同设备、工作负载和部署架构之间的权衡。我们提出了AgBench，一个基准测试套件和开放文档，用于可重复评估个人设备上的代理人工智能。通过AgBench，我们评估了代理工作负载下的本地、混合和云执行，考察了任务成功率、延迟、云API成本和数据暴露。我们的结果基于超过1.6207亿个数据点，显示个人人工智能设备可以在本地完成许多代理任务，但仅本地执行的任务成功率通常低于仅云执行，尤其是在并发性增加时。仅本地执行消除了云模型API成本和对云代理的敏感信息暴露。混合执行可以提高任务成功率，但其云成本和数据暴露取决于代理如何分配工作和共享信息。没有单一架构在任务成功率、良好吞吐量、云成本和数据暴露方面表现最佳；部署选择应反映预期的工作负载和设备能力。AgBench可在 https://anonymous.4open.science/r/AgBench-2777 获取。
cs.AI / 31 / 2609.38661
EvoSteer: Online Self-Evolving Graph Orchestration via Reference-Anchored Credit Assignment
EvoSteer：通过参考锚定的信用分配实现在线自我演化图谱编排
Abstract
In recent years, LLM-based multi-agent systems have been widely applied to orchestrate tool-using agents into executable communication graphs. However, existing self-evolving orchestration still faces key challenges, including post-hoc evolution that revises the team only after the trajectory ends, credit diffusion that gives every action the same terminal advantage under confounded baselines, and skill admission that is uncalibrated and never retired. To address these challenges, we propose EvoSteer, a new paradigm of Online Self-Evolving Graph Orchestration -- the orchestrator builds a running team and repairs its plausible but failing steps from execution features and a learned value estimate. To support this paradigm, we introduce Anchored Trajectory Balance (AnchorTB), a regression-style flow-matching loss that assigns each orchestration action a coefficient by balancing subtrajectories against a frozen reference. Built on the learned flow, we further propose Validated Skill Admission, in which a candidate skill is tried before promotion and promoted only if paired evidence passes a sequential test under a shared nominal testing budget. Moreover, AnchorTB combines measured task-level reference reward statistics with prefix-dependent corrections. Experimental results on twelve datasets show that EvoSteer significantly outperforms baselines across question answering, mathematical reasoning, code generation, and interactive decision making. Our code is available at https://github.com/beita6969/evosteer.
Chinese Translation
近年来，基于大语言模型（LLM）的多智能体系统被广泛应用于将工具使用代理编排成可执行的通信图。然而，现有的自我演化编排仍面临关键挑战，包括事后演化仅在轨迹结束后修正团队、信用扩散在混淆基线下给予每个动作相同的终端优势，以及技能录用未经过校准且从未被淘汰。为了解决这些挑战，我们提出了EvoSteer，一种新的在线自我演化图谱编排范式——编排者构建一个运行团队，并根据执行特征和学习的价值估计修复其合理但失败的步骤。为了支持这一范式，我们引入了锚定轨迹平衡（Anchored Trajectory Balance, AnchorTB），这是一种回归风格的流匹配损失，通过将子轨迹与冻结参考进行平衡，为每个编排动作分配一个系数。基于学习到的流，我们进一步提出了经过验证的技能录用（Validated Skill Admission），在提升之前尝试候选技能，并仅在配对证据通过共享名义测试预算下的顺序测试时进行提升。此外，AnchorTB将测量的任务级参考奖励统计与前缀依赖的修正相结合。在十二个数据集上的实验结果表明，EvoSteer在问答、数学推理、代码生成和交互决策等任务上显著优于基线。我们的代码可在 https://github.com/beita6969/evosteer 获取。
cs.AI / 32 / 2609.38670
Where Scientific Search Agents Fail: Decision-Checkpoint Auditing of Exposure and Inspection Attempts
科学搜索代理失败的原因：曝光和检查尝试的决策检查审计
Abstract
Final-answer accuracy does not reveal whether a scientific-search agent failed to encounter a target paper, attempt to inspect it, or return an accepted answer after inspection. We introduce decision checkpoints that record observations and tool actions without benchmark labels during inference, then join target identities and evaluator labels to assign outcome categories from recorded events. Across five conditions on 540 answerable AutoResearchBench Deep questions in a fixed, target-enriched environment, keyword search achieves 24.6\% accuracy, compared with 17.8\% for raw search. The keyword condition has fewer incorrect answers with neither target exposure nor inspection, but more incorrect answers after the target is exposed and left uninspected. Compared with keyword search, read-first has 27.4\% more recorded evidence-search calls. Target inspection attempts occur on 199 questions under read-first and 191 under keyword search; both conditions achieve 24.6\% accuracy. The checkpoint protocol makes these question-level differences explicit, distinguishing target exposure and inspection from aggregate accuracy and total tool use.
Chinese Translation
最终答案的准确性并不能揭示科学搜索代理是否未能接触到目标论文、尝试检查它，或在检查后返回一个被接受的答案。我们引入决策检查点，记录观察和工具操作，而不在推理过程中使用基准标签，然后将目标身份和评估者标签结合起来，从记录的事件中分配结果类别。在一个固定的、目标丰富的环境中，对540个可回答的AutoResearchBench Deep问题进行的五种条件下，关键词搜索的准确率为24.6\%，而原始搜索为17.8\%。关键词条件下，未接触或检查目标的错误答案较少，但在目标被曝光且未检查后，错误答案较多。与关键词搜索相比，优先阅读的条件下记录的证据搜索调用增加了27.4\%。在优先阅读条件下，目标检查尝试发生在199个问题上，而在关键词搜索下为191个；两种条件下的准确率均为24.6\%。检查点协议使这些问题级别的差异变得明确，区分了目标曝光和检查与整体准确性和工具总使用之间的关系。
cs.AI / 33 / 2609.38684
Concept-Grounded Attention: A Controlled Evaluation of Graph-Injected Attention, Temporal Versioning, and Epistemic Status
概念基础注意力：图注入注意力、时间版本控制和认知状态的受控评估
Abstract
Knowledge-intensive language-model systems typically represent external knowledge as text chunks or static graphs, with limited support for concept evolution, point-in-time reasoning, and distinctions between validated and inferred knowledge. We introduce the Concept Lifecycle Model (CLM), which represents concepts as persistent, graph-grounded, temporally versioned entities with explicit provenance and epistemic status, and Concept-Grounded Attention (CGA), which injects concept-graph structure into transformer computation through graph-biased self-attention (Form A) and gated cross-attention over concept nodes (Form B). We evaluate the framework in controlled settings using disabled-mechanism baselines. On 200 MuSiQue and HotpotQA questions with retrieval fixed, concept-graph retrieval recovers explicit multi-hop paths but does not improve evidence recall. Form A appears to steer attention, with 2.76 times more attention on gold than distractor concepts, but the same ratio occurs when Form A is disabled; the learned bias is negligible and no answers change. An identity-preserving Form B improves F1 from 0.188 to 0.221, but control concepts yield 0.213, indicating that most of the gain reflects added capacity. On LongMemEval, explicit temporal representation improves answer accuracy by 13 to 25 points across all tested generators, up to 122B parameters, while simplified CLM version resolution performs similarly to dated serialization because concept identity is not established reliably. On a synthetic source-independence task, protocol-derived epistemic status reduces unsupported assertions from 28% to 0.1% in a fine-tuned small model and from 19-68% to 0-5% in 72-122B models. Overall, the results support making temporal validity and epistemic status explicit, while showing that graph-attention diagnostics are not informative without disabled-mechanism controls.
Chinese Translation
知识密集型语言模型系统通常将外部知识表示为文本块或静态图，支持概念演变、时间点推理以及验证知识与推断知识之间的区分的能力有限。我们引入了概念生命周期模型（Concept Lifecycle Model, CLM），该模型将概念表示为持久的、图基础的、时间版本化的实体，具有明确的来源和认知状态，以及概念基础注意力（Concept-Grounded Attention, CGA），该注意力通过图偏置自注意力（Form A）和对概念节点的门控交叉注意力（Form B）将概念图结构注入到变换器计算中。我们在受控环境中使用禁用机制的基线评估该框架。在200个MuSiQue和HotpotQA问题中，固定检索的情况下，概念图检索恢复了明确的多跳路径，但并未改善证据回忆。Form A似乎引导了注意力，在金标准概念上比干扰概念多出2.76倍的注意力，但当Form A被禁用时，比例相同；学习到的偏差微不足道，且没有答案发生变化。保持身份的Form B将F1从0.188提高到0.221，但控制概念的F1为0.213，表明大部分增益反映了增加的容量。在LongMemEval上，明确的时间表示提高了所有测试生成器的答案准确性，提升幅度在13到25分之间，参数量高达122B，而简化的CLM版本解析的表现与过时的序列化相似，因为概念身份未能可靠建立。在一个合成的源独立任务中，基于协议推导的认知状态将小型微调模型中的不支持断言从28%降低到0.1%，在72-122B模型中从19-68%降低到0-5%。总体而言，结果支持明确时间有效性和认知状态，同时表明在没有禁用机制控制的情况下，图注意力诊断并不具有信息价值。
cs.AI / 34 / 2609.38690
GATE-ST: Gene-Aware Text-image Encoder for Spatial Transcriptomics
GATE-ST：基于基因的文本-图像编码器用于空间转录组学
Abstract
Spatial transcriptomics enables spatially resolved gene expression analysis from slide-level images while preserving morphological features, providing valuable information for studying disease mechanisms and developing treatments. However, spatial gene expression profiling typically requires expensive and time-consuming tests. While existing image-based prediction optimizations mostly revolve around including positional embeddings and further image-based changes, text-based optimizations remain relatively unexplored. We present GATE-ST, which incorporates text-based inputs into image-based spatial gene expression predictions. With this approach, generated text descriptions of genes are utilized to better spatial transcriptomics prediction results. Gene summaries are put through a text encoder, generating embeddings that integrate with image embeddings through cross-attention layers to align with morphological features. We demonstrate the effectiveness of such text inputs by benchmarking performance against random gene embeddings and multiple other image-text fusion architectures, and show that GATE-ST outperforms these alternatives. Our results demonstrate the effectiveness of GATE-ST in pathology imaging, which may greatly reduce the time and cost of accurate spatial transcriptomic predictions, proving the potential of text-guided spatial gene expression prediction.
Chinese Translation
空间转录组学能够从幻灯片级图像中进行空间分辨的基因表达分析，同时保留形态特征，为研究疾病机制和开发治疗方案提供了宝贵的信息。然而，空间基因表达谱通常需要昂贵且耗时的测试。现有的基于图像的预测优化主要围绕包括位置嵌入和进一步的基于图像的变化展开，而基于文本的优化仍然相对未被探索。我们提出了GATE-ST，它将基于文本的输入纳入基于图像的空间基因表达预测中。通过这种方法，生成的基因文本描述被用于改善空间转录组学的预测结果。基因摘要经过文本编码器处理，生成的嵌入与图像嵌入通过交叉注意力层进行整合，以与形态特征对齐。我们通过与随机基因嵌入和多种其他图像-文本融合架构的性能基准测试，展示了这种文本输入的有效性，并表明GATE-ST优于这些替代方案。我们的结果证明了GATE-ST在病理成像中的有效性，这可能大大减少准确的空间转录组预测所需的时间和成本，证明了基于文本的空间基因表达预测的潜力。
cs.AI / 35 / 2609.38699
Budget Boundary Effects in Test-Time Mathematical Reasoning
测试时数学推理中的预算边界效应
Abstract
A cumulative token cap can fall inside a mathematical derivation, forcing a test-time controller to choose between stopping at the cap (strict) and allowing the current attempt to finish (advisory). We measure this boundary choice with paired offline replays of 19,200 public traces: 120 AIME, BrUMO and HMMT problems and two archive configurations of one model. Candidate order and a 16-attempt cap are fixed, and answer selection is blind to reference answers and correctness labels. Three findings emerge. First, at the 4k cap, most advisory accuracy gains replace abstention with a correct answer; strict stopping pays for an unfinished prefix that the completed-only selector cannot use. Second, comparisons along realized cost differ from same-cap comparisons: advisory 4k in low has higher accuracy than strict 8k at comparable mean completion cost, while in high its observed accuracy is 0.42 points below strict 32k using 59% of its mean tokens. These aggregate comparisons do not establish equal-compute superiority or accuracy equivalence. Third, increased candidate coverage does not guarantee higher answer accuracy: a log-probability selector loses accuracy while coverage rises, including after a source-grade consistency repair. Same-cap majority-accuracy differences shrink below 1.3 percentage points at 32k. Budget curves should jointly state the cap, realized cost, eligible candidates, stopping rule and selector information.
Chinese Translation
累积令牌上限可能出现在数学推导中，迫使测试时控制器在达到上限（严格）和允许当前尝试完成（建议）之间做出选择。我们通过对19,200个公共轨迹的成对离线重放来测量这一边界选择：120个AIME、BrUMO和HMMT问题，以及一个模型的两个档案配置。候选顺序和16次尝试上限是固定的，答案选择对参考答案和正确性标签是盲目的。研究发现三点。首先，在4k上限下，大多数建议的准确性提升是用正确答案替代了放弃；严格停止则为未完成的前缀付出了代价，而仅完成选择器无法使用这些前缀。其次，沿着实现成本的比较与相同上限的比较不同：在低成本下，建议的4k准确性高于严格的8k，尽管两者的平均完成成本相当，而在高成本下，其观察到的准确性比严格的32k低0.42个百分点，使用了其平均令牌的59%。这些汇总比较并未确立等计算优越性或准确性等价性。第三，增加候选覆盖率并不保证更高的答案准确性：对数概率选择器在覆盖率上升时准确性下降，包括在源级一致性修复后。在32k的相同上限下，多数准确性差异缩小至1.3个百分点以下。预算曲线应共同表述上限、实现成本、合格候选人、停止规则和选择器信息。
cs.AI / 36 / 2609.38712
Staying on Task: Testing the Foundations of Long-Horizon Agent Reliability
保持任务专注：测试长时间跨度代理可靠性的基础
Abstract
Long-horizon agentic workflows require models to sustain repeated state-dependent actions all while the context grows, sub-task complexity changes, and new data arrives. Each situation represents an independent axis along which an agent may fail. An agent reconciling a long ledger, for example, must repeatedly read its state, update the correct record, and preserve alignment across thousands of outputs. A model may accept the entire ledger yet lose its place or stop applying the operation consistently as generation proceeds. We introduce Long-Transduction, a controlled diagnostic that tests a model's ability to stay on task during long generation while continuously reading, mutating, and outputting input-context dependent operations such as arithmetic, sorting, variable lookups, and table transformations. Long-Transduction evaluation independently varies local task complexity, input data formatting, and context length isolate failures along each axis. We evaluate seven open-weight models, finding a 62.8\% decrease when scaling context length from 4-128K, a 36.5\% decrease when varying input format, and a 39.9\% decrease by increasing local task complexity. Together, these failures represent critical liabilities in long-horizon agentic workflows.
Chinese Translation
长时间跨度的代理工作流程要求模型在上下文不断增长、子任务复杂性变化以及新数据到达的情况下，持续执行依赖于状态的重复动作。每种情况都代表了代理可能失败的独立轴。例如，一个正在核对长账本的代理，必须不断读取其状态，更新正确的记录，并在数千个输出之间保持一致性。一个模型可能接受整个账本，但在生成过程中可能会失去位置或停止一致地应用操作。我们引入了长转导（Long-Transduction），这是一种受控诊断工具，用于测试模型在长生成过程中保持任务专注的能力，同时持续读取、变更和输出依赖于输入上下文的操作，如算术、排序、变量查找和表格转换。长转导评估独立地变化局部任务复杂性、输入数据格式和上下文长度，以隔离每个轴上的失败。我们评估了七个开放权重模型，发现当上下文长度从4增加到128K时，性能下降了62.8%；当变化输入格式时，性能下降了36.5%；而通过增加局部任务复杂性，性能下降了39.9%。这些失败共同代表了长时间跨度代理工作流程中的关键风险。
cs.AI / 37 / 2609.38721
UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement
UniEvo-VL：一种用于多模态模型自我提升的在线自蒸馏训练方法
Abstract
Modern multimodal models bring generation and understanding into a single unified system, which enables them to provide and learn from their own feedback. Motivated by this unified capacity, we introduce UniEvo-VL, a self-evolving framework for multimodal models to learn from this constructive self-correction feedback during test-time compute. Instead of relying on a separate, often larger, teacher, we leverage their self-critiques as privileged information and ask a single multimodal model to act as both teacher and student with different contexts. The student only sees the vanilla question, while the teacher conditions on the privileged critique. Then training minimizes the per-state divergence between their denoising diffusion distributions over the student's own sampling trajectories. Experiments demonstrate that UniEvo-VL improves the image generation capabilities of multimodal models, while maintaining their sensitivity to additional reflection information. Specifically, we build on top of the open-source Qwen-image-2512 and observe a significant performance gain from 0.747 to 0.808 on GenEval and from 32.97 to 35.53 on GenEval2 Soft-TIFA. Moreover, attempts with more powerful external critics (e.g., GPT5.6-Luna) show that multimodal models with strong judge capabilities can anticipate a higher self-evolving ceiling. Last but not least, mixed text-rendering outcomes show that our self-improvements may not be uniform across different tasks. Our study aims to shed light on the current hot recursive self-improvement research line to enhance the user experience when using multimodal models without external supervision or guidance.
Chinese Translation
现代多模态模型将生成和理解整合为一个统一的系统，使其能够提供并从自身反馈中学习。基于这一统一能力，我们提出了UniEvo-VL，一个自我演化框架，旨在让多模态模型在测试时计算中从这种建设性的自我修正反馈中学习。我们不依赖于一个单独的、通常更大的教师模型，而是利用它们的自我批评作为特权信息，要求单个多模态模型在不同上下文中同时充当教师和学生。学生仅看到原始问题，而教师则基于特权批评进行条件化。然后，训练通过最小化学生自身采样轨迹上去噪扩散分布之间的每个状态的差异来进行。实验表明，UniEvo-VL提高了多模态模型的图像生成能力，同时保持了对额外反思信息的敏感性。具体而言，我们在开源的Qwen-image-2512基础上进行构建，并观察到在GenEval上的性能显著提升，从0.747提高到0.808，在GenEval2 Soft-TIFA上从32.97提高到35.53。此外，使用更强大的外部批评者（例如，GPT5.6-Luna）的尝试表明，具有强大判断能力的多模态模型可以预期更高的自我演化上限。最后，混合文本渲染结果表明，我们的自我提升在不同任务之间可能并不均匀。我们的研究旨在为当前热门的递归自我提升研究方向提供启示，以增强用户在使用多模态模型时的体验，而无需外部监督或指导。
cs.AI / 38 / 2609.38733
Code to Control: Synthesizing Parameterized Reactive Controllers
控制代码：合成参数化反应控制器
Abstract
Recent LLM-based approaches to control either invoke a language model to select actions or synthesize world models that require planning at every decision, introducing latency that can limit real-time use. We introduce Code to Control, an approach that synthesizes Python controllers which execute directly as policies. Code to Control separates program structure from parameters. An LLM synthesizes the controller structure, while derivative-free search fits its parameters for continuous control using feedback from the environment. Once learned, the resulting controllers require neither LLM inference nor planning at decision time, enabling real-time gameplay and, under our timing protocol, faster action selection than a PPO policy. Across a suite of Atari games, Flappy Bird, and MuJoCo tasks, Code to Control outperforms planning-based program synthesis methods, remains competitive with deep reinforcement learning while using fewer environment interactions, transfers across substantial changes in environment dynamics, and scales to complex locomotion tasks.
Chinese Translation
近期基于大型语言模型（LLM）的方法在控制方面，要么调用语言模型选择动作，要么合成需要在每个决策时进行规划的世界模型，这引入了可能限制实时使用的延迟。我们提出了控制代码（Code to Control），一种合成直接作为策略执行的Python控制器的方法。控制代码将程序结构与参数分离。LLM合成控制器结构，而无导数搜索则利用来自环境的反馈为连续控制拟合其参数。一旦学习完成，生成的控制器在决策时既不需要LLM推理，也不需要规划，从而实现实时游戏玩法，并且在我们的时间协议下，其动作选择速度快于PPO策略。在一系列Atari游戏、Flappy Bird和MuJoCo任务中，控制代码的表现优于基于规划的程序合成方法，与深度强化学习相比仍具竞争力，同时使用更少的环境交互，能够在环境动态发生显著变化时进行迁移，并可扩展到复杂的运动任务。
cs.AI / 39 / 2609.38743
Learning to Route in Visual Space via Multi-Step Embedding Retrieval
通过多步骤嵌入检索学习在视觉空间中的路径规划
Abstract
LLM agents rely on retrieval tools to access external knowledge, yet visual agentic search remains severely bottlenecked by standard single-step retrievers. In current pipelines, the agent must issue text queries for every intermediate step, struggling when visual clues are difficult to describe or when the retriever fails to surface necessary intermediate evidence within its top results. We hypothesize that offloading multi-step navigation across the entire embedding space directly to the retrieval tool resolves this performance bottleneck. To study this systematically, we introduce VHOP, a flexible data generation framework and benchmark with five core difficulty levels testing both visual matching and search planning. Using this framework, we develop VHOP-Router, an end-to-end training pipeline---combining supervised fine-tuning, online imitation learning, and reinforcement learning---that transforms a standard embedding model into an autoregressive multi-step retriever. Operating directly in the visual latent space, VHOP-Router retrieves linked image chains in a single tool call without requiring the agent to formulate intermediate text queries. Experiments show VHOP-Router boosts retrieval performance from under 5\% to 76.3\%. In agentic search, it improves task success rates by 52.7\% and reduces the average token length by 61\% from 1886 to 728, whereas upgrading the agent yields only a 3.7\% gain. Compared to a strong baseline where the agent retrieves the top 50 results per step, VHOP-Router maintains superior performance while reducing in-context images by $23\times$ and cutting the cumulative API payload by $35\times$. The models also generalize robustly to unseen difficulty levels and realistic test sets. Ultimately, VHOP and VHOP-Router provide an efficient and effective solution for visual agentic search that leaves native LLM capabilities entirely intact.
Chinese Translation
大型语言模型（LLM）代理依赖检索工具来访问外部知识，但视觉代理搜索仍然受到标准单步骤检索器的严重瓶颈。在当前的工作流程中，代理必须为每一个中间步骤发出文本查询，当视觉线索难以描述或检索器未能在其前几个结果中呈现必要的中间证据时，代理会面临困难。我们假设，将整个嵌入空间的多步骤导航直接卸载到检索工具上可以解决这一性能瓶颈。为了系统地研究这一点，我们引入了VHOP，一个灵活的数据生成框架和基准，设有五个核心难度级别，用于测试视觉匹配和搜索规划。利用该框架，我们开发了VHOP-Router，一个端到端的训练管道——结合了监督微调、在线模仿学习和强化学习——将标准嵌入模型转变为自回归多步骤检索器。VHOP-Router直接在视觉潜在空间中操作，能够在一次工具调用中检索相关的图像链，而无需代理构造中间文本查询。实验表明，VHOP-Router将检索性能从不足5\%提升至76.3\%。在代理搜索中，它将任务成功率提高了52.7\\%，并将平均标记长度从1886减少到728，减少幅度达到61\\%，而升级代理仅带来了3.7\\%的提升。与一个强基线相比，该基线在每一步中检索前50个结果，VHOP-Router在保持优越性能的同时，将上下文图像减少了$23 imes$，并将累积API负载减少了$35 imes$。这些模型在未见过的难度级别和现实测试集上也表现出强大的泛化能力。最终，VHOP和VHOP-Router为视觉代理搜索提供了一种高效且有效的解决方案，同时完全保留了原生LLM的能力。
cs.AI / 40 / 2609.38757
Self-Evolving Algorithm-Design Agents: Escaping In-Context Evolutionary Stagnation via Population-Curated Policy Optimization
自我进化算法设计代理：通过群体策划的策略优化逃避上下文中的进化停滞
Abstract
Large language models are increasingly participating in complex real-world tasks in the form of algorithm-design agents, designing and refining algorithms. Many successful algorithm-design agents adopt pure in-context evolutionary frameworks, but they may quickly plateau in domains that require specialized knowledge. Parametric adaptation offers a way to internalize specialized knowledge, but conventional training requires abundant domain-specific corpora while high-quality algorithms are scarce in complex algorithm-design scenarios. In this paper, we propose sample-efficient parametric self-evolution where agents can explore and learn from self-generated algorithms. First, we characterize in-context evolutionary stagnation and analytically propose the Improvement Chain proposition, showing how learning successive self-generated algorithms can locally increase the likelihood of neighboring algorithms. Motivated by this local-transfer perspective, we further propose Population-Curated Policy Optimization (PCPO) to utilize a global population and a hybrid policy update scheme for retaining and reusing high-quality, diverse self-generated algorithms, shifting the policy towards stronger algorithms. In the task of learning rate schedule design for global placement in electronic design automation, trained only on 4 chip cases, PCPO outperforms the state-of-the-art in-context evolutionary methods (e.g., OpenEvolve and ShinkaEvolve) on average across 16 chip cases. With an 8B-size base model, PCPO achieves competitive performance compared to frontier closed-source models such as GPT-5.5. PCPO also reduces inference-time token cost by internalizing grounded domain knowledge and prompt distillation. Moreover, PCPO achieves significant speedups on four GPU kernel designs, with an average of 8.27$\times$ speedup against the PyTorch Eager baseline.
Chinese Translation
大型语言模型越来越多地以算法设计代理的形式参与复杂的现实世界任务，设计和优化算法。许多成功的算法设计代理采用纯粹的上下文进化框架，但在需要专业知识的领域中，它们可能很快达到瓶颈。参数适应提供了一种内化专业知识的方法，但传统训练需要大量特定领域的语料库，而在复杂的算法设计场景中，高质量算法则稀缺。本文提出了一种样本高效的参数自我进化方法，代理可以探索并学习自生成的算法。首先，我们描述了上下文进化停滞，并分析性地提出了改进链命题，展示了学习连续自生成算法如何局部提高邻近算法的可能性。基于这一局部转移的视角，我们进一步提出了群体策划策略优化（Population-Curated Policy Optimization，PCPO），利用全球群体和混合策略更新方案来保留和重用高质量、多样化的自生成算法，从而将策略转向更强的算法。在电子设计自动化中的全局放置学习率调度设计任务中，仅基于4个芯片案例训练的PCPO在16个芯片案例上平均超越了最先进的上下文进化方法（例如，OpenEvolve和ShinkaEvolve）。使用8B规模的基础模型，PCPO的性能与前沿的闭源模型（如GPT-5.5）相当。PCPO还通过内化基础领域知识和提示蒸馏，降低了推理时的令牌成本。此外，PCPO在四个GPU内核设计上实现了显著的加速，相较于PyTorch Eager基线平均加速达8.27倍。
cs.AI / 41 / 2609.38766
PathAnchor: Path-Structured Evidence for Scientific Agents
PathAnchor：科学代理的路径结构证据
Abstract
Scientific agents can retrieve relevant passages yet still lose functional order, mix evidence across sources, or state conclusions that exceed the retrieved record. We introduce PathAnchor, a bounded scientific reasoning system built on path-structured evidence workspaces. Instead of treating passages or extracted concepts as independent units, the system retrieves source-linked Material-Sensor-Signal-System trajectories that preserve role, direction, and the evidence supporting each transition. A controller uses three read-only tools to search paper-specific trajectories, trace paths across candidate sources, and open exact evidence before producing a claim-cited answer and an explicit evidence boundary. On 120 single- and cross-paper flexible-sensor questions, PathAnchor scores 82.6% and leads six evaluated systems. Under a matched controller, corpus, and six-call budget, replacing unordered concept graphs with path-structured records raises source recall from 61.3% to 82.9%, increases answers whose claims all cite opened evidence from 69.2% to 90.0%, and reduces tool calls. These results show that evidence organization affects retrieval and citation completeness under fixed agent resources.
Chinese Translation
科学代理能够检索相关段落，但仍可能失去功能顺序、混合来自不同来源的证据，或陈述超出检索记录的结论。我们提出了PathAnchor，一种基于路径结构证据工作空间的有限科学推理系统。该系统并不将段落或提取的概念视为独立单元，而是检索源链接的物料-传感器-信号-系统（Material-Sensor-Signal-System）轨迹，这些轨迹保留了角色、方向以及支持每个转变的证据。控制器使用三种只读工具来搜索特定论文的轨迹、追踪候选来源的路径，并在生成引用证据的答案和明确的证据边界之前打开确切证据。在120个单一和跨论文的灵活传感器问题中，PathAnchor的得分为82.6%，领先于六个评估系统。在匹配的控制器、语料库和六次调用预算下，用路径结构记录替换无序概念图将来源召回率从61.3%提高到82.9%，将所有引用打开证据的答案比例从69.2%提高到90.0%，并减少了工具调用次数。这些结果表明，证据的组织方式影响在固定代理资源下的检索和引用完整性。
cs.AI / 42 / 2609.38778
Action Conditioned Bisimulation For GUI Agent Memory
基于动作条件的双模拟用于图形用户界面代理记忆
Abstract
An agent that remembers what it did on a web page must decide when two pages count as the same. Memories built on observation similarity merge pages that look alike but behave differently, and GUIs are full of such pages: two tabs of one widget or two rows of one menu answer the same click differently. We define the merge rule as an action-conditioned bisimulation over the empirical predictive state graph a frozen agent fills as it acts. Two states merge only when their shared actions lead to agreeing outcomes and successor blocks under an affordance label. Observation similarity never enters the rule, and nothing is trained. It replaces the merge rule of an existing outcome-value memory, so a closed-loop comparison isolates it. On MiniWoB++ it raises success rate over a memoryless agent, while a control taking identical exploratory detours, the prior successor-representation merge, and the same criterion without action conditioning change nothing.
Chinese Translation
一个能够记住在网页上所做操作的代理必须决定何时将两个页面视为相同。基于观察相似性构建的记忆合并了看似相似但行为不同的页面，而图形用户界面（GUI）中充满了这样的页面：一个小部件的两个标签或一个菜单的两行对同一次点击的响应是不同的。我们将合并规则定义为在代理执行操作时填充的经验预测状态图上的动作条件双模拟（action-conditioned bisimulation）。只有当两个状态的共享动作导致一致的结果和在可供性标签下的后继块时，它们才会合并。观察相似性从未进入规则中，且没有任何训练。这一规则替代了现有结果值记忆的合并规则，因此闭环比较将其孤立。在MiniWoB++上，它提高了相较于无记忆代理的成功率，而控制组则采取了相同的探索性绕行、先前的后继表示合并，以及在没有动作条件的情况下相同的标准，均未产生任何变化。
cs.AI / 43 / 2609.38782
Persona and Persuasive Framing in AI Voice Agents: A $2\times2$ Field Experiment with Children
AI语音助手中的角色与劝说框架：一项针对儿童的$2 imes2$实地实验
Abstract
Conversational agents increasingly interact with children, yet evidence on how their design shapes children's susceptibility to persuasion comes almost entirely from the lab. We report a $2\times2$ randomized field experiment embedded in a public German Santa Claus telephone hotline. Children's calls were randomly routed to one of four LLM voice agents varying persona (Santa, high authority, vs. Helper, low authority) and framing (persuasive nudges toward prosocial wishes vs. neutral). Of 1,072 logged calls, 89 conversations (median age 6) met inclusion criteria. Persuasive framing raised the probability of a prosocial wish from 11.6% to 45.7%, robust to controls. Persona authority showed a near-zero effect: Santa did not outperform the Helper. Persona instead shaped engagement; children hung up on the Helper far more often within the first minute (65% vs. 39%). Where context already lends an agent legitimacy, how it speaks shapes children's compliance more than who it claims to be.
Chinese Translation
对话代理越来越多地与儿童互动，但关于其设计如何影响儿童对劝说的敏感性，几乎所有证据都来自实验室。我们报告了一项嵌入在德国公共圣诞老人电话热线中的$2 imes2$随机实地实验。儿童的电话随机分配到四个不同的LLM语音代理中，这些代理在角色（圣诞老人，高权威 vs. 助手，低权威）和框架（针对亲社会愿望的劝说性提示 vs. 中性）上有所不同。在1,072个记录的电话中，有89个对话（中位年龄6岁）符合纳入标准。劝说框架将亲社会愿望的概率从11.6%提高到45.7%，这一结果在控制变量下依然稳健。角色权威几乎没有影响：圣诞老人并没有优于助手。相反，角色影响了参与度；儿童在前一分钟内更频繁地挂掉助手的电话（65% vs. 39%）。在上下文已经赋予代理合法性的情况下，代理的说话方式比其声称的身份更能影响儿童的服从程度。
cs.AI / 44 / 2609.38788
Positive Ratings, Hidden Concerns: Employee Voice Disclosure in AI-Mediated Organizational Listening
积极评价与隐秘顾虑：在人工智能中介的组织倾听中的员工声音披露
Abstract
Organizations started listening to employees through conversational AI agents alongside structured surveys. Little is known about what these channels change in what employees say when disclosure carries hierarchical risk. We report a field study inside a global management consulting firm whose process pairs a pre-survey with an adaptive AI voice interview on the same themes within one session. Across 44 first-session interviews (132 matched theme observations), 20-41% of sessions showed a favorable rating co-occurring with a substantive concern voiced later, depending on the favorability threshold. The Gioia analysis drew on 158 protective quotes from 65 eligible sessions. Disclosure rarely arrived unguarded: employees softened concerns, deflected accountability, and bounded how far they went, and this protective work tracked the perceived legitimacy of the listening structure. We develop a grounded model of bounded disclosure and derive four propositions for voice, channel and listening research. Silence, we argue, can persist inside expression.
Chinese Translation
组织开始通过对话式人工智能代理和结构化调查来倾听员工的声音。关于这些渠道在员工披露存在层级风险时所改变的内容知之甚少。我们报告了一项在一家全球管理咨询公司内部进行的实地研究，该过程将预调查与同一主题的自适应人工智能语音访谈结合在一个会话中进行。在44个首次会话访谈中（132个匹配主题观察），根据好评阈值的不同，20-41%的会话显示出积极评价与后续表达的实质性顾虑同时出现。Gioia分析基于来自65个合格会话的158个保护性引用。披露很少是毫无保留的：员工会软化顾虑、推脱责任，并限制表达的深度，而这种保护性工作与倾听结构的感知合法性密切相关。我们发展了一个有根据的有限披露模型，并为声音、渠道和倾听研究提出了四个命题。我们认为，沉默可以在表达中持续存在。
cs.AI / 45 / 2609.38798
GraphCert: Bootstrap Agentic Graph Reasoning with Certified Evidence Rubrics
GraphCert：使用认证证据标准引导的自主图推理
Abstract
Graph agents extend large language models (LLMs) with the ability to actively explore and reason over knowledge graphs through multi-step interactions with graph tools. However, training capable graph agents typically requires large collections of question-answer pairs and reasoning trajectories, whose manual construction is costly and difficult to scale. Moreover, employing proprietary LLMs to generate such supervision further risks exposing sensitive graph data to external services. Therefore, we propose GraphCert to bootstrap agentic graph reasoning with certified evidence rubrics during post-training. Specifically, the Bootstrapped Graph Quizzer guided by generation controls produces graph-grounded QA pairs and marks supporting evidence, which undergo execution certification and semantic curation. The accepted evidence is then canonicalized into certified evidence rubrics that later reward Graph Solver evidence alignment alongside answer correctness during GRPO training. Experiments on five graph reasoning domains in GRBENCH demonstrate that GraphCert consistently outperforms substantially larger LLM agents and post-training method. Furthermore, our analysis demonstrates that the learned policy transfers robustly across heterogeneous graph domains, suggesting that GraphCert acquires reusable graph-reasoning capabilities rather than domain-specific patterns. These results establish executable self-certification as an effective approach to self-training compact graph reasoning agents. Our code will be made publicly available.
Chinese Translation
图代理通过与图工具的多步交互，扩展了大型语言模型（LLMs）在知识图谱上的主动探索和推理能力。然而，训练能够进行推理的图代理通常需要大量的问题-答案对和推理轨迹，而手动构建这些数据既成本高昂又难以扩展。此外，使用专有的LLM生成这种监督数据还可能导致敏感图数据暴露于外部服务。因此，我们提出了GraphCert，以在后训练阶段利用认证证据标准引导自主图推理。具体而言，受生成控制引导的自举图测验器生成基于图的问答对并标记支持证据，这些证据经过执行认证和语义整理。接受的证据随后被规范化为认证证据标准，在GRPO训练期间，这些标准将奖励图求解器的证据对齐和答案正确性。在GRBENCH的五个图推理领域的实验表明，GraphCert始终优于规模大得多的LLM代理和后训练方法。此外，我们的分析表明，学习到的策略在异构图领域之间具有良好的迁移能力，这表明GraphCert获得了可重用的图推理能力，而不是特定于领域的模式。这些结果确立了可执行自我认证作为自我训练紧凑图推理代理的有效方法。我们的代码将公开发布。
cs.AI / 46 / 2609.38817
When Reasoning Goes Astray: Attention Dynamics of Uncontrolled Reasoning
推理失控时：无控制推理的注意力动态
Abstract
Large reasoning models (LRMs) improve performance on complex tasks through extended reasoning, yet the same process can degenerate into redundant verification and persistent generation loops. Such uncontrolled reasoning increases inference cost and creates risks of resource exhaustion and service degradation. However, existing mitigations largely truncate long outputs or react to surface repetition, and thus fail to distinguish normal thinking from uncontrolled reasoning or explain how benign reasoning degenerates into harmful behavior. In this paper, we operationalize LRM generation as four states and further introduce Reasoning-state Analysis via Dynamic Attention Responses (RADAR), which identifies the current reasoning state in real time and characterizes how effective reflection can develop into uncontrolled generation. Guided by RADAR's analysis, we further realign abnormal attention distributions toward patterns observed in normal requests and examine how this correction affects excessive reflection and persistent looping. Temporal analyses show that uncontrolled reasoning is characterized by attention distributions that deviate from normal generation, with abnormal trends becoming detectable before repetition begins. Correcting these deviations through Attention Realignment consistently reduces looping while largely preserving benign performance. Together, RADAR provide a mechanistic account of how reasoning becomes uncontrolled, offering actionable guidance for identifying critical failure stages and designing targeted runtime interventions.
Chinese Translation
大型推理模型（LRMs）通过扩展推理在复杂任务上提高了性能，但同样的过程可能退化为冗余验证和持续生成循环。这种无控制的推理增加了推理成本，并带来了资源耗尽和服务降级的风险。然而，现有的缓解措施主要是截断长输出或对表面重复做出反应，因此未能区分正常思维与无控制推理，也未能解释良性推理如何退化为有害行为。本文将LRM生成操作化为四种状态，并进一步引入动态注意力响应下的推理状态分析（Reasoning-state Analysis via Dynamic Attention Responses, RADAR），该方法实时识别当前推理状态，并表征有效反思如何发展为无控制生成。在RADAR分析的指导下，我们进一步重新调整异常注意力分布，使其趋向于正常请求中观察到的模式，并检验这种修正如何影响过度反思和持续循环。时间分析表明，无控制推理的特征是注意力分布偏离正常生成，异常趋势在重复开始之前就可被检测到。通过注意力重新对齐来纠正这些偏差，能够持续减少循环，同时在很大程度上保持良性性能。总体而言，RADAR提供了推理如何失控的机制性解释，为识别关键失效阶段和设计针对性的运行时干预提供了可操作的指导。
cs.AI / 47 / 2609.38818
Whose Voice Survives the Summary? A Voice-Retention Audit of LLM Employee Listening
谁的声音在摘要中得以保留？对大语言模型员工倾听的声音保留审计
Abstract
Organizations increasingly route employee feedback to leaders through large language model (LLM) summaries, an unaudited layer that silences already-spoken voice. We introduce a Voice Retention / Representation Ratio metric for representational bias in summarization and apply it to a bilingual (English/German) corpus of 2,586 free-text responses from a global professional service company. First, employees supply criticism more reliably than praise (withholding praise is 82 times more common). Second, across 45 leader-summaries the pipeline filters by popularity, not sentiment: criticism survives, yet a concern voiced once is dropped 86% of the time, with short and German-only content lost on the same axis (theme retention 0.14 vs 0.74; German directional). Controlling for frequency, sentiment has no independent effect; the harm is prevalence-driven, which sentiment-only audits miss. A targeted prompt recovers only named themes. We contribute the metric, field evidence, and a disaggregated voice-retention card.
Chinese Translation
组织越来越多地通过大语言模型（LLM）摘要将员工反馈传递给领导，这一未经审计的层面抑制了已经表达的声音。我们引入了一个声音保留/代表性比率指标，用于评估摘要中的代表性偏差，并将其应用于来自一家全球专业服务公司的2586条自由文本响应的双语（英语/德语）语料库。首先，员工提供批评的可靠性高于赞扬（抑制赞扬的情况是82倍更常见）。其次，在45份领导摘要中，信息处理按受欢迎程度而非情感进行过滤：批评得以保留，但一次表达的关切有86%的概率被忽略，短内容和仅德语内容在同一维度上也会丢失（主题保留率为0.14对比0.74；德语方向性）。在控制频率的情况下，情感没有独立影响；这种伤害是由普遍性驱动的，而仅依赖情感的审计则会忽视这一点。一个针对性的提示仅能恢复命名主题。我们贡献了这一指标、实地证据以及一个分解的声音保留卡。
cs.AI / 48 / 2609.38827
More Choices, Fewer Decisions: Ordinal-Scale Bias in JEV-like Direct-Decision Models
更多选择， fewer 决策：JEV 类直接决策模型中的序数尺度偏差
Abstract
Direct-decision models turn text into low-latency structured labels and scores, making them attractive for classification and automatic evaluation. Yet reliability requires more than accuracy: a model must also use the ordinal decision scale supplied by the user faithfully. We analyze JEV~1.13 and three open KEV models. Our investigation begins with ANLI, where JEV assigns 38.8\% of all predictions and 51.3\% of errors to Neutral despite 74.95\% accuracy, nearly balanced gold labels, and balanced candidate positions. Across 36 ordinal datasets, final decisions use only 67--76\% of the effective gold support, versus 87--102\% on four nominal tasks. Randomizing candidate order weakens but does not remove this compression. Holding items and source scores fixed while balancing gold support and positions, we refine scales from $K=2$ to $14$; utilization falls for every model and reaches 26--75\% at $K=14$, although candidate probabilities remain broad for most models. Targeted BA-LoRA post-training raises gold-relative utilization from roughly 47\% to 86\% on eight supervised scales at both KEV sizes, showing that the compression is learned and modifiable rather than an immutable architectural limit. We call this ordinal scale-utilization bias: decision-stage candidate-space compression distinct from accuracy, gold imbalance, fixed position, and candidate count alone. The code and data are available at https://github.com/Glax147/jev_ordinal_scale_bia
Chinese Translation
直接决策模型将文本转化为低延迟的结构化标签和分数，使其在分类和自动评估中具有吸引力。然而，可靠性不仅仅依赖于准确性：模型还必须忠实地使用用户提供的序数决策尺度。我们分析了 JEV~1.13 和三个开放的 KEV 模型。我们的研究始于 ANLI，其中 JEV 将 38.8\% 的所有预测和 51.3\% 的错误分配给中性（Neutral），尽管准确率为 74.95\%，黄金标签几乎平衡，候选位置也相对平衡。在 36 个序数数据集中，最终决策仅使用了 67--76\% 的有效黄金支持，而在四个名义任务中则为 87--102\%。随机化候选顺序虽然削弱了这种压缩，但并未消除它。在固定项目和源分数的同时平衡黄金支持和位置，我们将尺度从 $K=2$ 提升到 $14$；每个模型的利用率下降，在 $K=14$ 时达到 26--75\%，尽管大多数模型的候选概率仍然较宽。针对 BA-LoRA 的后训练将黄金相对利用率从大约 47\% 提高到 86\%，在两个 KEV 大小的八个监督尺度上，显示出这种压缩是可学习和可修改的，而不是不可改变的架构限制。我们称之为序数尺度利用偏差：决策阶段候选空间的压缩，与准确性、黄金不平衡、固定位置和候选数量无关。代码和数据可在 https://github.com/Glax147/jev_ordinal_scale_bia 获取。
cs.AI / 49 / 2609.38829
Diversity Combining for Multi-Path LLM Reasoning
多路径大语言模型推理中的多样性组合
Abstract
Multi-path reasoning methods such as self-consistency (SC) sample $K$ reasoning paths and choose the most frequent answer. However, their gains quickly plateau as $K$ increases, and existing methods do not predict when this saturation will occur. We formalize multi-path LLM reasoning as a diversity combining problem from wireless communications: each path is a noisy channel observation, and the pairwise correlation of path correctness caps the design-effect effective sample size of the vote at a finite ceiling. Generalized least squares (GLS) analysis shows that, under exchangeability, the optimal symmetric linear combiner of latent embeddings is uniform, supporting majority vote as the natural default in standard SC while leaving room for weighting or pruning under heterogeneous prompt-template branches. Across 5 models and 12 benchmarks, prompt-template diversity reduces path correlation in $55$ of $57$ valid cells, with the strongest effect on open-ended QA. We derive an Adaptive-K rule that uses a four-path pilot to select $K^*$, retaining $96$--$103\%$ of MV@$K{=}32$ accuracy across Math, QA, and NLU.
Chinese Translation
多路径推理方法如自一致性（Self-consistency, SC）通过采样 $K$ 条推理路径并选择最频繁的答案。然而，随着 $K$ 的增加，它们的收益很快会达到饱和，现有方法无法预测这种饱和何时会发生。我们将多路径大语言模型推理形式化为无线通信中的多样性组合问题：每条路径都是一个噪声信道观测，而路径正确性的成对相关性限制了投票的设计效应有效样本量的上限。广义最小二乘法（Generalized Least Squares, GLS）分析表明，在可交换性条件下，潜在嵌入的最佳对称线性组合器是均匀的，支持在标准 SC 中将多数投票作为自然默认，同时为在异构提示模板分支下的加权或修剪留出空间。在 5 个模型和 12 个基准测试中，提示模板的多样性在 $57$ 个有效单元中减少了 $55$ 个路径相关性，对开放式问答的影响最为显著。我们推导出一种自适应 K 规则，该规则使用四条路径的试点来选择 $K^*$，在数学、问答和自然语言理解（NLU）中保留了 $96$--$103 ext{%}$ 的 MV@$K{=}32$ 准确率。
cs.AI / 50 / 2609.38850
OpenJev-RLCD: A Working RLCD Implementation
OpenJev-RLCD：一个可工作的RLCD实现
Abstract
Decision models such as Jev answer questions with probabilities, which are only useful if they are calibrated. Open-source reproductions rely on supervised fine-tuning plus temperature scaling, while reinforcement learning from verifiable rewards (RLVR) makes reasoning models overconfident. We present a working implementation of reinforcement learning for calibrated decisions (RLCD) for reasoning models: the model samples a rationale, and we score the answer distribution it commits to afterwards with a strictly proper scoring rule. A variance identity shows that scoring the mixture of several samples rewards disagreeing rationales, and that RLVR is exactly this mixture objective without its diversity term. Optimized naively, the per-rationale objective either switches reasoning off or is drowned out by policy-gradient noise, which leads to a two-stage recipe: calibrate, then reinforce. With Qwen3-1.7B on two reasoning tasks (3 seeds, paired tests), RLCD matches or beats SFT, RFT/STaR and GRPO (each temperature-scaled) in accuracy and beats all of them in selective prediction; on GSM8K answer verification a single query decides \gvTwoCovFive\% of the items at $\le$5\% error, versus \gvGrpoCovFive\% for GRPO. When uncertainty comes from annotator disagreement, RLCD provably cannot beat cross-entropy. Code and results: https://github.com/ZimmyGao/openjev-rlcd.
Chinese Translation
决策模型如Jev通过概率回答问题，只有在经过校准后才有用。开源复现依赖于监督微调加上温度缩放，而来自可验证奖励的强化学习（RLVR）使推理模型过于自信。我们提出了一种用于校准决策的强化学习实现（RLCD）用于推理模型：该模型采样一个推理依据，然后我们使用严格适当的评分规则对其承诺的答案分布进行评分。方差恒等式表明，对多个样本的混合进行评分会奖励不一致的推理依据，而RLVR正是没有多样性项的这种混合目标。经过简单优化后，每个推理依据的目标要么关闭推理，要么被策略梯度噪声淹没，这导致了一个两阶段的方案：先校准，然后强化。在两个推理任务上使用Qwen3-1.7B（3个种子，配对测试），RLCD在准确性上与SFT、RFT/STaR和GRPO（均经过温度缩放）相匹配或超过，并在选择性预测上优于所有模型；在GSM8K答案验证中，单个查询决定了 extbackslash gvTwoCovFive extbackslash%的项目，误差不超过5 extbackslash%，而GRPO的结果为 extbackslash gvGrpoCovFive extbackslash%。当不确定性来自于标注者之间的分歧时，RLCD在理论上无法超越交叉熵。代码和结果见：https://github.com/ZimmyGao/openjev-rlcd。
cs.AI / 51 / 2609.38866
When Context Changes: Understanding Update Failures in LLMs
当上下文变化时：理解大型语言模型中的更新失败
Abstract
As preferences, goals, and facts change, LLM agents must use the current state while earlier versions remain in context. Yet they can answer with an old value of the same variable, a failure that we call stale binding. To study when models use outdated information and why, we introduce Controlled In-Context Memory (CICM), a benchmark for tracking and using updated information in conversations and agent logs. We observe that even frontier reasoning models can fail to recover the current state. We find that in open-source models probes can still recover the updated value when the model answers with an old one, pointing to a failure to select information that remains available. Component tests in Qwen and Pythia identify a mechanism for this selection failure: attention drift, where attention favors old values over the current one when producing an answer. We study a one-layer transformer to mathematically understand how this phenomenon happens: when attention scores are similar, several old values can together receive more attention than the current value. Guided by this explanation, we redirect attention toward the current value without further training. When the current value is requested directly, adjusting this intervention for each input corrects most old-value errors across various model families while preserving nearly all initially correct answers. Reliable context management therefore requires more than remembering updated information: models must use it to guide their answers.
Chinese Translation
随着偏好、目标和事实的变化，大型语言模型（LLM）代理必须使用当前状态，而早期版本仍然保留在上下文中。然而，它们可能会用同一变量的旧值进行回答，这种失败我们称之为陈旧绑定（stale binding）。为了研究模型何时使用过时信息以及原因，我们引入了受控上下文记忆（Controlled In-Context Memory, CICM），这是一个用于跟踪和使用对话及代理日志中更新信息的基准测试。我们观察到，即使是前沿推理模型也可能无法恢复当前状态。我们发现，在开源模型中，当模型用旧值回答时，探测器仍然可以恢复更新值，这表明模型未能选择仍然可用的信息。对Qwen和Pythia的组件测试识别了这一选择失败的机制：注意力漂移（attention drift），即在生成答案时，注意力偏向于旧值而非当前值。我们研究了一个单层变换器，以数学方式理解这一现象是如何发生的：当注意力分数相似时，多个旧值可能共同获得比当前值更多的注意力。在这一解释的指导下，我们在不进行进一步训练的情况下，将注意力重新引导到当前值。当直接请求当前值时，针对每个输入调整这一干预措施可以纠正各种模型家族中的大多数旧值错误，同时几乎保留所有最初正确的答案。因此，可靠的上下文管理不仅需要记住更新的信息：模型还必须使用这些信息来指导其答案。
cs.AI / 52 / 2609.38867
Talk2Agent: Benchmarking Voice Interfaces for Text Agents
Talk2Agent：文本代理的语音接口基准测试
Abstract
Large language model (LLM) computer-use agents are typically evaluated with clean written instructions, despite speech being an increasingly popular interface for interacting with such systems. Speech input introduces an additional failure point: transcription errors can alter task-critical entities, constraints, or targets before the agent begins reasoning, while conventional ASR metrics do not directly measure whether the information required for successful execution has been preserved. We introduce Talk2Agent, a benchmark for evaluating how effectively voice interfaces convey human-spoken instructions to LLM-based computer-use agents. Talk2Agent builds human-spoken versions of tasks from WildClawBench and OSWorld and evaluates a range of voice interfaces, including dedicated ASR models, audio-capable LLMs, contextual biasing, and LLM-based ontology repair. Because repeatedly executing long-horizon computer-use tasks is costly and stochastic, we further propose an execution-free, task-conditioned evaluation framework that projects the original task grader onto prompt-addressable intentions and measures how much task-relevant information is retained after the voice interface. On WildClawBench, Talk2Agent's execution-free native projection provides a practical, execution-grounded measure of voice-interface quality, correlating with downstream task completion and improving Pearson correlation by 0.246 over WER/CER on 32 hours of real human speech.
Chinese Translation
大型语言模型（LLM）计算机使用代理通常使用清晰的书面指令进行评估，尽管语音作为与此类系统交互的接口越来越受欢迎。语音输入引入了一个额外的失败点：转录错误可能在代理开始推理之前改变任务关键实体、约束或目标，而传统的自动语音识别（ASR）指标并未直接衡量成功执行所需的信息是否得以保留。我们提出了Talk2Agent，这是一个评估语音接口如何有效传达人类口语指令给基于LLM的计算机使用代理的基准。Talk2Agent从WildClawBench和OSWorld构建了人类口语版本的任务，并评估了一系列语音接口，包括专用的ASR模型、具备音频能力的LLM、上下文偏置和基于LLM的本体修复。由于反复执行长时间跨度的计算机使用任务成本高且具有随机性，我们进一步提出了一种无执行的任务条件评估框架，该框架将原始任务评分器投影到可通过提示访问的意图上，并衡量在语音接口之后保留了多少与任务相关的信息。在WildClawBench上，Talk2Agent的无执行原生投影提供了一种实用的、基于执行的语音接口质量测量，与下游任务完成情况相关，并在32小时的真实人类语音上提高了0.246的Pearson相关性，相较于字错误率（WER）/字符错误率（CER）。
cs.AI / 53 / 2609.38869
Reasoning Externalization for Faithful Large Language Model Narratives of Stock Return Predictions
忠实大型语言模型叙述的股票回报预测推理外化
Abstract
In finance, interpreting machine learning predictions is essential, yet the numerical outputs of explainable AI can be difficult for non-experts to understand. While large language models (LLMs) can translate these outputs into natural language, they may produce errors when inferring numerical changes and feature relations. We propose an LLM narrative framework for cross-sectional stock return prediction that combines temporal Shapley additive explanations (SHAP) evidence with historical regime analogs. Temporal evidence tracks changes in the normalized global SHAP importance of an XGBoost model over six months. Historical analogs are past periods with similar changes in SHAP importance, their model performance and subsequent market returns are provided as comparative context. Using this framework, we conduct a controlled study of progressive reasoning externalization, sequentially providing raw SHAP sequences, deterministic temporal descriptors, and feature relations. Each generated claim is verified against provenance-linked evidence. Across Qwen3, externalizing numerical and relational reasoning improved evidence faithfulness as well as temporal and relational accuracy. Evidence faithfulness increased from 0.696 to 0.996 for Qwen3-32B-Instruct. While historical analogs did not improve structured automatic faithfulness, they received higher human-rated usefulness scores. These results suggest that externalizing verifiable reasoning enhances narrative faithfulness and that historical context adds interpretive value.
Chinese Translation
在金融领域，解释机器学习预测至关重要，但可解释人工智能的数值输出对于非专业人士来说可能难以理解。虽然大型语言模型（LLMs）可以将这些输出翻译为自然语言，但在推断数值变化和特征关系时可能会产生错误。我们提出了一种用于横截面股票回报预测的LLM叙述框架，该框架结合了时间性Shapley加法解释（SHAP）证据与历史状态类比。时间性证据跟踪了XGBoost模型在六个月内标准化全球SHAP重要性的变化。历史类比是指在SHAP重要性变化相似的过去时期，其模型表现和随后的市场回报作为比较背景提供。利用该框架，我们进行了一项关于推理外化的受控研究，依次提供原始SHAP序列、确定性时间描述符和特征关系。每个生成的声明都与来源关联的证据进行验证。在Qwen3中，外化数值和关系推理提高了证据的忠实性以及时间和关系的准确性。对于Qwen3-32B-Instruct，证据忠实性从0.696提高到0.996。虽然历史类比并未改善结构化自动忠实性，但其获得了更高的人类评分的实用性分数。这些结果表明，外化可验证的推理增强了叙述的忠实性，而历史背景则增加了解释的价值。
cs.AI / 54 / 2609.38881
STRATA: Self-Learning Through Role-Aligned Tiered Agents for Real-Time Strategy Games
STRATA：通过角色对齐的分层代理自学习实时战略游戏
Abstract
Real-time strategy (RTS) games require agents to coordinate economic development, production and construction, base defense, unit organization, and attack timing over long matches. Existing studies have applied large language models to command decision-making in RTS games, enabling agents to read textual game states and generate high-level plans. However, long inference latency can cause them to miss critical tactical events. The complexity and tactical diversity of full RTS matches also leave existing systems heavily dependent on manually written experience-based prompts, with limited ability to learn continuously from past games. We present STRATA, a role-aligned hierarchical system with cross-game self-learning for Red Alert. STRATA assigns in-game strategic, logistical, and tactical decisions to a Strategic Agent (SA), Logistics Agent (LA), and Tactical Agent (TA), respectively. The SA generates high-level directives based on the global game state and relevant experience cards, while the LA and TA handle logistics and tactical execution. After each match, a Review Agent (RA) derives candidate experience from game traces, validates and revises it using evidence from subsequent matches, and compresses strategic experience supported across multiple games into concise experience cards for SA retrieval. We evaluate STRATA through the formation of experience cards, full-match comparisons before and after learning, and experience learning against AI opponents with different play styles. Under a fixed scenario, using the learned experience cards increases the observed win rate from 30% to 100%. Sequential learning against AI opponents with different play styles also produces distinct long-term strategic experience.
Chinese Translation
实时战略（RTS）游戏要求代理在长时间的比赛中协调经济发展、生产与建设、基地防御、单位组织和攻击时机。现有研究已将大型语言模型应用于RTS游戏中的指挥决策，使代理能够读取文本游戏状态并生成高层次计划。然而，较长的推理延迟可能导致它们错过关键的战术事件。完整RTS比赛的复杂性和战术多样性也使现有系统在很大程度上依赖于手动编写的基于经验的提示，限制了其从过去比赛中持续学习的能力。我们提出了STRATA，一个具有跨游戏自学习的角色对齐分层系统，专为《红色警戒》设计。STRATA将游戏中的战略、后勤和战术决策分别分配给战略代理（Strategic Agent, SA）、后勤代理（Logistics Agent, LA）和战术代理（Tactical Agent, TA）。SA根据全球游戏状态和相关经验卡生成高层次指令，而LA和TA则处理后勤和战术执行。在每场比赛后，复审代理（Review Agent, RA）从游戏轨迹中提取候选经验，利用后续比赛的证据验证和修订这些经验，并将跨多场比赛支持的战略经验压缩成简明的经验卡，以供SA检索。我们通过经验卡的形成、学习前后全场比赛的比较以及与不同游戏风格的AI对手的经验学习来评估STRATA。在一个固定场景下，使用学习到的经验卡使观察到的胜率从30%提高到100%。与不同游戏风格的AI对手进行的顺序学习也产生了明显的长期战略经验。
cs.AI / 55 / 2609.38891
Consistent Plan-Act for Long-Horizon Agentic Tasks
一致性计划-行动用于长时间跨度的自主任务
Abstract
Long-horizon agentic tasks demand strong reasoning and efficient execution across successive interactions with dynamic environments. A common approach decouples high-level planning from low-level execution through separate planner and actor roles. To investigate coordination failures in these tasks, we prompt both agents for structured state assertions and compare their reports programmatically to detect explicit contradictions. Our analyses reveal systematic disagreement about the same task-relevant state facts, a phenomenon we term planner-actor state mismatch. We further find that providing agents with task-relevant state information reduces mismatch and improves coordination and task performance. Based on the systematic analysis of the state mismatch, we propose Consistent Plan-Act (ConPAct), which feeds detected contradictions back to both agents to form consistent state interpretations and fine-tunes them on curated consistent interactions for better coordination. ConPAct improves performance across various environments and model configurations, e.g., increasing MiniGrid success rate from 38.6% to 54.4% with GPT-5.6-sol/terra as planner and actor respectively, demonstrating that state consistency can guide both inference-time correction and coordination training.
Chinese Translation
长时间跨度的自主任务要求在与动态环境的连续交互中进行强有力的推理和高效的执行。一种常见的方法是通过分离的规划者和执行者角色，将高层次的规划与低层次的执行解耦。为了研究这些任务中的协调失败，我们促使两个智能体进行结构化状态断言，并通过程序化比较他们的报告以检测显式矛盾。我们的分析揭示了关于同一任务相关状态事实的系统性分歧，这一现象我们称之为规划者-执行者状态不匹配。我们进一步发现，向智能体提供与任务相关的状态信息可以减少不匹配，提高协调性和任务表现。基于对状态不匹配的系统分析，我们提出了一致性计划-行动（Consistent Plan-Act, ConPAct），该方法将检测到的矛盾反馈给两个智能体，以形成一致的状态解释，并在策划的一致交互中对其进行微调，以实现更好的协调。ConPAct在各种环境和模型配置中提高了性能，例如，在使用GPT-5.6-sol/terra作为规划者和执行者时，MiniGrid的成功率从38.6%提高到54.4%，证明状态一致性可以指导推理时的修正和协调训练。
cs.AI / 56 / 2609.38912
Composing Task-specific Agent Harnesses at Test Time with Reusable Primitives
在测试时利用可重用原语构建任务特定的代理工具
Abstract
Agent harnesses govern how large language models (LLMs) gather context, invoke tools, verify results, preserve state, and terminate, largely affecting agent performance. However, the value of each harness mechanism can differ across heterogeneous tasks: a mechanism that improves one task may impose overhead or context distraction on another, leading to the suboptimality of a global harness. We characterize this suboptimality as a mismatch induced by fixed mechanism choices, motivating task-specific harness construction. Nonetheless, generating harness code for each task introduces generation and debugging costs, with execution risks that can compound as more mechanisms are generated. To address those challenges, we introduce Harness Primitives, reusable harness mechanisms with clear application scope and composition contract mined from failed task trajectories. Based on Harness Primitives, we propose STITCH, a framework that Selects suitable primitives given Task Information and compiles them into Task-speCific Harnesses at test time. This separation enables task-specific harnesses without generating or repairing mechanism code at test time. Extensive experiments demonstrate that STITCH not only improves harness adaptability and robustness, but also scales with the primitive library size, boosting task success rates by up to 12 points over fixed harness baselines, surpassing human-designed harnesses like Codex CLI while maintaining a minimal test-time harness composition overhead of only 2.7%, 638 times more efficient than generating task-specific harnesses from scratch. Ultimately, our work demonstrates that building task-adaptive harnesses can be beneficial for completing diverse tasks and that building reusable primitives can be a promising path towards this goal.
Chinese Translation
代理工具管理大型语言模型（LLMs）如何收集上下文、调用工具、验证结果、保持状态和终止，这在很大程度上影响代理的性能。然而，各种工具机制在异构任务中的价值可能不同：一种机制可能改善某一任务，但在另一任务中却会造成额外开销或上下文干扰，导致全局工具的次优性。我们将这种次优性表征为由固定机制选择引起的不匹配，促使任务特定工具的构建。然而，为每个任务生成工具代码会引入生成和调试成本，并且随着生成机制的增多，执行风险可能会加剧。为了解决这些挑战，我们引入了工具原语（Harness Primitives），这是一种可重用的工具机制，具有明确的应用范围和从失败任务轨迹中挖掘出的组合契约。基于工具原语，我们提出了STITCH框架，该框架根据任务信息选择合适的原语，并在测试时将其编译成任务特定的工具。这种分离使得在测试时无需生成或修复机制代码即可实现任务特定工具。大量实验表明，STITCH不仅提高了工具的适应性和鲁棒性，而且随着原语库规模的扩大，任务成功率比固定工具基线提高了多达12个百分点，超越了人类设计的工具如Codex CLI，同时保持了仅2.7%的最小测试时工具组合开销，比从头生成任务特定工具高效638倍。最终，我们的工作表明，构建任务自适应工具对于完成多样化任务是有益的，而构建可重用原语则是实现这一目标的有前景的路径。
cs.AI / 57 / 2609.38914
Risk-Aware Adaptive Evaluation: Finding High-Impact Failures Under Limited Budgets
风险感知自适应评估：在有限预算下寻找高影响力的失败
Abstract
Evaluating interactive agents is expensive. Agent behavior is stochastic, so reliability must be measured over repeated trials, but failures are rare and differ widely in how much they matter. Standard benchmarks spend this budget uniformly: a read-only lookup is sampled as often as an irreversible payment action. We instead formulate evaluation as a sequential allocation problem. Given a fixed trial budget and a set of scenarios whose failure behavior is unknown, which scenarios should be run, and run again? We propose a risk-aware contextual Thompson Sampling policy that combines a pre-execution scenario context vector and a fixed impact score with the failure outcomes observed during evaluation, and we test it by offline replay over 70 $\tau$-bench airline scenarios and 824 recorded trials. Our main result is at the smallest budget: with only 50 trials ($6\%$ of the corpus), the policy recovers $86\%$ of the impact-weighted failures an oracle could find, compared to $25\%$ for uniform allocation. It discovers $3.5\times$ more impact-weighted failures (215.4 vs. 62.2) with the same number of trials, delivers $5\times$ the discovery per dollar, and cuts the budget wasted on scenarios that never fail from $34\%$ to $2.8\%$. The rest of our analysis demonstrates and qualifies this result: a budget sweep shows the advantage shrinks as the budget approaches the corpus size, and paired significance tests show that scenario context helps mainly at small budgets while posterior-based exploration helps at moderate ones. Risk-aware adaptive allocation therefore helps most exactly where evaluation budget is scarcest.
Chinese Translation
评估交互代理是昂贵的。代理行为具有随机性，因此必须通过重复试验来测量其可靠性，但失败是罕见的，并且在其重要性上差异很大。标准基准均匀地使用这一预算：只读查找的采样频率与不可逆支付行为相同。我们将评估问题重新表述为一个顺序分配问题。在固定的试验预算和一组未知失败行为的场景下，应该运行哪些场景，并重复运行？我们提出了一种风险感知的上下文汤普森采样策略，该策略结合了执行前的场景上下文向量和固定的影响评分，以及在评估过程中观察到的失败结果。我们通过对70个$ au$-基准航空场景和824个记录试验的离线重放进行测试。我们的主要结果是在最小预算下：仅用50次试验（占语料库的6%），该策略恢复了86%的影响加权失败，而均匀分配只能找到25%。在相同的试验次数下，它发现了3.5倍更多的影响加权失败（215.4对62.2），每美元的发现量提高了5倍，并将从未失败的场景浪费的预算从34%降低到2.8%。我们分析的其余部分展示并限定了这一结果：预算扫描显示，随着预算接近语料库规模，优势逐渐减小，而配对显著性测试表明，场景上下文主要在小预算下有助于评估，而基于后验的探索在中等预算下更为有效。因此，风险感知自适应分配在评估预算最紧缺的地方最为有效。
cs.AI / 58 / 2609.38917
How Much Can Reliability Drift Under a Fixed Confidence Distribution?
在固定置信分布下，可靠性可以漂移多少？
Abstract
A classifier's conditional accuracy can change while its confidence distribution stays exactly the same. We study the worst-case movement of the reliability relation under covariate shifts that preserve the distribution of the confidence score, constraining the reweighting within each confidence level by a $\chi^2$ budget; the resulting worst case, as a function of the budget, is a fragility profile. On an interval of budgets that can be computed from the source distribution, the profile equals exactly the square root of the budget times the within-level variance of the correctness propensity -- the grouping-loss term of calibration-refinement decompositions. Beyond this interval the profile is governed by the tails of the propensity law, and the entire upward profile determines the centred within-level law; consequently, calibration residual and grouping variance do not determine fragility in general, though they do when labels and predictions are deterministic. Since the propensity is not observed, we restrict reweightings to a learned finite readout within confidence bins, bound the part the restriction misses by the grouping variance remaining inside readout cells, estimate the restricted profile with role-separated labels, and provide a separate split-sample lower confidence bound. On ImageNet this bound is positive in both splits for four of six primary classifiers and nine of twelve additional ones as released, and for three of eighteen after temperature scaling. Held-out drift under optimised reweightings fitted without evaluation labels tracks the estimated profile; an exploratory label-permutation diagnostic yields near-zero agreement for this statistic while largely reproducing the correlation observed for unsigned random reweightings.
Chinese Translation
分类器的条件准确性可以在置信分布完全相同的情况下发生变化。我们研究在保持置信分数分布的协变量变化下，可靠性关系的最坏情况变化，限制在每个置信水平内的重加权由 $ ext{χ}^2$ 预算约束；由此产生的最坏情况作为预算的函数，是一个脆弱性轮廓。在可以从源分布计算出的预算区间内，该轮廓恰好等于预算的平方根乘以正确性倾向的内部方差——这是校准精细化分解中的分组损失项。在此区间之外，轮廓受倾向法则的尾部支配，整个向上的轮廓决定了中心的内部水平法则；因此，校准残差和分组方差通常不决定脆弱性，尽管在标签和预测是确定性的情况下是如此。由于倾向未被观察到，我们将重加权限制为在置信区间内学习到的有限读出，通过分组方差限制遗漏部分，估计带有角色分离标签的受限轮廓，并提供单独的分割样本下限置信界。在 ImageNet 上，这个界限在六个主要分类器中的四个和十二个附加分类器中的九个的两个分割中都是正的，并且在温度缩放后对十八个中的三个也是如此。在没有评估标签的情况下，优化重加权下的持出漂移跟踪估计的轮廓；一个探索性的标签置换诊断对该统计量的接近零一致性，同时在很大程度上再现了观察到的无符号随机重加权的相关性。
cs.AI / 59 / 2609.38925
Prototype-guided Bilateral Alignment Multimodal Federated Learning
原型引导的双向对齐多模态联邦学习
Abstract
Multimodal federated learning (MFL) has emerged as a pivotal paradigm for leveraging distributed data to enhance model performance. However, existing methods predominantly rely on idealized assumptions of model homogeneity and balanced modality distributions, rendering them ill-suited for practical scenarios characterized by heterogeneous client architectures and severe modality imbalance. To address these challenges, we propose a \textbf{M}ultimodal \textbf{Fed}erated learning Prototype-guided Bilateral Alignment (MFedPBA) framework. MFedPBA facilitates robust knowledge synergy through a dual alignment mechanism: (i) at the feature level, it aligns heterogeneous feature spaces via a projection encoder optimized by contrastive learning and the Gromov-Wasserstein distance; (ii) at the decision level, it employs an entropy-weighted aggregation of naturally aligned logit prototypes. This novel design achieves robust MFL by jointly tackling heterogeneous feature spaces and collectively aggregating decisions. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art baselines under conditions of model heterogeneity and modality imbalance.
Chinese Translation
多模态联邦学习（MFL）已成为利用分布式数据提升模型性能的重要范式。然而，现有方法主要依赖于模型同质性和模态分布平衡的理想假设，这使得它们不适用于特征异构的客户端架构和严重模态不平衡的实际场景。为了解决这些挑战，我们提出了一种多模态联邦学习原型引导的双向对齐（MFedPBA）框架。MFedPBA通过双重对齐机制促进了稳健的知识协同：（i）在特征层面，通过对比学习和Gromov-Wasserstein距离优化的投影编码器对异构特征空间进行对齐；（ii）在决策层面，采用自然对齐的logit原型的熵加权聚合。该新颖设计通过共同解决异构特征空间和集体聚合决策，实现了稳健的多模态联邦学习。大量实验表明，在模型异构性和模态不平衡的条件下，我们的方法显著优于最先进的基线。
cs.AI / 60 / 2609.38929
Learning What to Forget: Distributional Unlearning for LLM Representation Spaces
学习遗忘的内容：大规模语言模型表示空间的分布式遗忘
Abstract
Machine learning systems increasingly face the need to remove the influence of entire data domains, such as toxic language, harmful behavior, or topical content, rather than isolated records. Recent work formalizes this problem as \emph{distributional unlearning}: selecting a subset of a forget domain whose removal moves the training distribution away from an unwanted population while preserving proximity to the desired one. However, existing analyses often impose parametric assumptions to obtain tractable selection rules. These assumptions may be poorly suited to high-dimensional language-model representations. We introduce \textsc{Mamushi}, a framework for non-parametric distributional unlearning that ranks forget examples using a probabilistic classifier whose Bayes-optimal logit equals the forget-to-retain log-density ratio (up to an additive class-prior constant). We show that thresholding the population log-density ratio yields the optimal fixed-budget selection rule for our removal--preservation objective and establish a non-asymptotic transfer guarantee relating score-estimation and threshold-calibration errors to degradation from the population-optimal selection rule. Our empirical evaluation spans real-world datasets on toxic-language removal and topical-domain removal regimes using different representations, with \textsc{Mamushi} achieving a more favorable removal--preservation trade-off than other baselines. Our work shows that \textsc{Mamushi} can serve as an efficient selection approach for downstream machine unlearning procedures, reducing the number of forget examples required to reach a fixed forgetting target.
Chinese Translation
机器学习系统越来越需要去除整个数据领域的影响，例如有毒语言、有害行为或特定主题内容，而不仅仅是孤立的记录。最近的研究将这一问题形式化为 extit{分布式遗忘}：选择一个遗忘领域的子集，其移除能够使训练分布远离不希望的群体，同时保持与期望群体的接近。然而，现有的分析通常施加参数假设以获得可处理的选择规则。这些假设可能不适合高维语言模型表示。我们引入了 extsc{Mamushi}，一个非参数分布式遗忘框架，通过一个概率分类器对遗忘示例进行排名，该分类器的贝叶斯最优对数几率等于遗忘与保留的对数密度比（加上一个可加的类别先验常数）。我们展示了对人群对数密度比进行阈值处理可以为我们的移除-保留目标提供最佳的固定预算选择规则，并建立了一个非渐近转移保证，将得分估计和阈值校准误差与人群最优选择规则的退化联系起来。我们的实证评估涵盖了关于有毒语言移除和特定主题领域移除的真实世界数据集，使用不同的表示方式， extsc{Mamushi}在移除-保留权衡方面优于其他基线。我们的工作表明， extsc{Mamushi}可以作为下游机器遗忘程序的有效选择方法，减少达到固定遗忘目标所需的遗忘示例数量。
cs.AI / 61 / 2609.38956
Routing Probes Can Improve Without New Information: An Exact-Null Audit of Uncertainty Beyond Model Outputs
路由探针可以在没有新信息的情况下改善：超出模型输出的不确定性精确零审计
Abstract
Routing signals of modern vision transformers -- expert gates, attention-residual weights and halting scores -- often improve probes that predict whether the model is correct, and the improvement is commonly read as evidence that routing carries information about errors beyond the model's outputs. We test this inference directly: keeping real output-routing pairs, we redraw correctness labels from a frozen output-only generator fitted on disjoint data, so that routing is uninformative by construction. Under this exact label null, a width-matched MLP comparison still reports a routing gain in 51.3% of confidence-only evaluations (308/600), while a linear comparison reports none. Holding each training trajectory fixed on a six-model panel and selecting the checkpoint by validation log loss instead of validation accuracy removes the detections (50/120 to 0/120, and 83/120 to 0/120 in an independently implemented probe), identifying accuracy-based checkpoint selection as the cause; across all output views the raw detection rate falls from 27.5% (528/1,920) to zero observed detections. The repaired comparison is not sensitive, detecting an implanted signal of about 0.005 nats in 0/20 replicates in each of two matched settings, whereas a conditional permutation test built on an estimated routing law detects it in 11/20 and 10/20 and rejects rarely under the null. On real correctness labels, the conditional analysis yields model-relative evidence in five DeiT attention-residual families; in four it persists under two specified variants of the conditional law, and no family passes an additional noise criterion. Fitting a better probe and testing for incremental information are different problems, and each needs its own validation.
Chinese Translation
现代视觉变换器的路由信号——专家门、注意力残差权重和停止分数——通常改善预测模型是否正确的探针，而这种改善通常被解读为路由携带关于错误的信息，超出模型的输出。我们直接测试这一推论：保持真实的输出-路由对，我们从一个适配于不重叠数据的冻结输出生成器中重新绘制正确性标签，从而使路由在构造上不具信息性。在这个精确标签零的情况下，宽度匹配的多层感知器（MLP）比较仍在51.3%的仅置信度评估中报告了路由增益（308/600），而线性比较则没有报告任何增益。将每个训练轨迹固定在六模型面板上，并通过验证日志损失而非验证准确率选择检查点，消除了检测（从50/120降至0/120，以及在独立实现的探针中从83/120降至0/120），确认基于准确率的检查点选择是原因；在所有输出视图中，原始检测率从27.5%（528/1,920）降至零观察到的检测。修复后的比较不敏感，在两个匹配设置中每个设置的20次重复中检测到约0.005 nats的植入信号为0，而基于估计路由法则构建的条件置换测试在11/20和10/20中检测到该信号，并在零假设下很少拒绝。在真实的正确性标签上，条件分析在五个DeiT注意力残差家族中产生了模型相关的证据；在四个家族中，它在条件法则的两个指定变体下持续存在，并且没有家族通过额外的噪声标准。拟合更好的探针和测试增量信息是不同的问题，每个问题都需要自己的验证。
cs.AI / 62 / 2609.38958
Targeted Retrieval, Compact Representations: How CoT Reasoning Improves Long-Context Counting
针对性检索，紧凑表示：链式思维如何改善长文本计数
Abstract
Large language models (LLMs) have been rapidly improving in long-context tasks, powered by Chain-of-Thought (CoT) reasoning. However, the internal mechanisms underlying this improvement remain unclear. We investigate these mechanisms through a needle-in-a-haystack (NIAH) counting task, where an LLM is asked to count the number of records dispersed in a long text. Across twelve model comparison groups, Thinking (or reasoning) improves counting accuracy over Non-thinking, with pronounced gains at larger counts. This motivates our mechanistic analysis, which identifies two contrasting mechanisms: (i) broad retrieval, where Non-thinking models broadly attend to multiple needles; (ii) targeted retrieval, where Thinking models use enumeration in CoT traces to successively retrieve needles. Targeted retrieval concentrates attention on individual needles and is accompanied by more compact internal representations. Moreover, causal intervention analysis suggests that Thinking models use the CoT trace to maintain and update an internal counter as needles are successively retrieved, even without explicit numbering. In small controlled experiments, both retrieval mechanisms and counter states emerge under standard autoregressive training. Together, our results connect long-context retrieval with representation geometry of counting, supporting a state-tracking account of CoT reasoning.
Chinese Translation
大型语言模型（LLMs）在长文本任务中的表现迅速提升，这得益于链式思维（Chain-of-Thought，CoT）推理。然而，推动这一改善的内部机制仍不清楚。我们通过一个针在干草堆（needle-in-a-haystack，NIAH）计数任务来研究这些机制，该任务要求LLM计算分散在长文本中的记录数量。在十二个模型比较组中，思考（或推理）相较于非思考模型提高了计数准确性，尤其在较大计数时表现出明显的提升。这激励了我们的机制分析，识别出两种对立的机制：（i）广泛检索，非思考模型广泛关注多个针；（ii）针对性检索，思考模型利用CoT轨迹中的枚举逐步检索针。针对性检索将注意力集中在单个针上，并伴随着更紧凑的内部表示。此外，因果干预分析表明，思考模型利用CoT轨迹在逐步检索针时维护和更新内部计数器，即使没有明确的编号。在小规模的控制实验中，这两种检索机制和计数状态在标准自回归训练下均有所出现。总的来说，我们的结果将长文本检索与计数的表示几何联系起来，支持了CoT推理的状态跟踪解释。
cs.AI / 63 / 2609.38962
Alleviating Hallucination in Reasoning Tasks with Training-Free Uncertainty-Guided Steering
通过无训练的不确定性引导调整缓解推理任务中的幻觉
Abstract
Recent work on hallucination detection in large language models has shown that, for a fixed pre-trained model and reasoning task, it is possible to estimate the model's confidence in the correctness of its outputs. Such uncertainty estimates have primarily been used to improve truthfulness by detecting or filtering confabulations. In this work, we ask whether these signals can instead be used more proactively to directly improve the accuracy of model-generated answers. We propose USteer, a simple, training-free steering mechanism that adjusts a model's layer-wise activations during inference using the gradient of a confidence measure with respect to the activations. This procedure nudges generation toward outputs with lower uncertainty at inference time, without modifying model parameters or requiring additional supervision. We show that this approach consistently reduces hallucination across a range of tasks, demonstrating that confidence signals can be leveraged not only for detection, but also for effective inference-time control of model behavior.
Chinese Translation
近期关于大型语言模型幻觉检测的研究表明，对于固定的预训练模型和推理任务，可以估计模型对其输出正确性的信心。这种不确定性估计主要用于通过检测或过滤虚构内容来提高真实性。在本研究中，我们探讨这些信号是否可以更主动地用于直接提高模型生成答案的准确性。我们提出了USteer，一种简单的无训练调整机制，它在推理过程中使用信心度量相对于激活的梯度来调整模型的层级激活。该过程在推理时将生成过程引导向不确定性较低的输出，而无需修改模型参数或额外的监督。我们展示了这种方法在一系列任务中持续减少幻觉，证明了信心信号不仅可以用于检测，还可以有效控制模型行为的推理过程。
cs.AI / 64 / 2609.38964
When Order Matters: First-Speaker Bias and Mitigation through Personality in Sequential Multi-Agent Debate
顺序重要性：第一发言者偏见及其通过个性在顺序多智能体辩论中的缓解
Abstract
Multi-agent debate (MAD) is often used to improve large language model (LLM) reasoning, but sequential debate is rarely a neutral aggregator of agents' opinions. We show that sequential MAD suffers from a pronounced first-speaker bias: agents disproportionately shape the final answer when they speak first. As a result, placing a stronger model after weaker ones can substantially offset its reasoning advantage. We then focus on the disadvantaged strong-agent-last setting and ask whether personality prompting can mitigate this imbalance. Drawing on the Big Five model, we study agreeableness and extraversion as behavioral interventions applied to either the strong or weak side. We find that their effects are trait-specific. Influence consistently shifts in the direction of lower agreeableness, and assigning low agreeableness to the stronger agent helps restore its lost influence and improves final accuracy. Extraversion, by contrast, produces less systematic changes in influence and accuracy, with its clearest effect appearing in agents' verbosity. These findings show that effective MAD design depends not only on model capability, but also on how speaking order and induced interaction behavior shape the debate process.
Chinese Translation
多智能体辩论（MAD）常用于提升大型语言模型（LLM）的推理能力，但顺序辩论往往不是中立的代理意见聚合器。我们发现顺序MAD存在明显的第一发言者偏见：当代理首先发言时，他们对最终答案的影响不成比例。因此，将更强的模型放在较弱模型之后可以显著抵消其推理优势。接着，我们关注处于劣势的强代理最后的设置，并探讨个性提示是否能缓解这种不平衡。基于五大人格模型，我们研究了宜人性和外向性作为应用于强方或弱方的行为干预。我们发现它们的影响是特质特定的。影响力始终向低宜人性方向转移，将低宜人性分配给更强的代理有助于恢复其失去的影响力并提高最终准确性。相比之下，外向性对影响力和准确性的变化影响较小，其最明显的效果体现在代理的冗长性上。这些发现表明，有效的MAD设计不仅依赖于模型能力，还取决于发言顺序和诱导的互动行为如何塑造辩论过程。
cs.AI / 65 / 2609.38974
RealWorldShop: Benchmarking and Improving Conversational Shopping Agents in Real-World E-commerce
RealWorldShop：在真实世界电子商务中基准测试和改进对话购物代理
Abstract
Large language models are reshaping ecommerce from static recommenders into interactive shopping assistants, yet real-world shopping requires session-level decision support: users reveal and revise constraints, coordinate multiple goals, and expect product-grounded recommendations over a full conversation. Existing benchmarks are mostly outcome-oriented or execution-oriented, leaving this evolving decision process under-evaluated. We introduce REALWORLDSHOP, a benchmark built on 3.28M grounded products, structured shopping episodes, a profile-grounded and actioncontrolled user simulator, and role-play evaluation. Our analysis shows that current systems produce locally plausible responses but struggle with state tracking, constraint updating, and grounded convergence, especially under ambiguous intent, bundle, and multi-intent scenarios. We further propose REALSHOP_AGENT, an executable session-control framework with explicit state management, shopping-flow control, catalog-grounded retrieval, and runtime guards. Experiments show that REALSHOP_AGENT consistently outperforms strong baselines on REALWORLDSHOP.
Chinese Translation
大型语言模型正在将电子商务从静态推荐系统转变为互动购物助手，但真实世界的购物需要会话级别的决策支持：用户揭示和修订约束，协调多个目标，并期望在整个对话中获得基于产品的推荐。现有的基准测试大多以结果为导向或执行为导向，导致这一不断演变的决策过程未得到充分评估。我们引入了REALWORLDSHOP，这是一个基于328万种基础产品、结构化购物情节、基于用户档案和受控行为的用户模拟器以及角色扮演评估构建的基准。我们的分析表明，当前系统能够生成局部合理的响应，但在状态跟踪、约束更新和基础收敛方面存在困难，尤其是在模糊意图、捆绑和多重意图场景下。我们进一步提出了REALSHOP_AGENT，这是一个具有明确状态管理、购物流程控制、目录基础检索和运行时保护的可执行会话控制框架。实验表明，REALSHOP_AGENT在REALWORLDSHOP上始终优于强基线。
cs.AI / 66 / 2609.39005
C-STRIDE: An Observation-Driven AI Digital Twin for Predicting Basin-Wide Flood Fields from Sparse Stream-Gauge Histories
C-STRIDE：一种基于观测驱动的人工智能数字双胞胎，用于从稀疏的水流测量历史中预测流域范围内的洪水场
Abstract
Emergency managers need to know where floodwater is, how deep it is, and how it will change over the coming hours across an entire river basin. During a flood, however, real-time measurements come from only a handful of stream gauges, and high-resolution hydrodynamic models are too costly to rerun each time new data arrive or to run as large ensembles. We present C-STRIDE, an observation-driven AI digital twin that turns short records from a few stream gauges, together with terrain and rainfall, into basin-wide maps of water depth and extends these predictions up to a day ahead. It is trained on simulations from a calibrated two-dimensional hydrodynamic model and needs no separate data-assimilation step. In the Des Plaines River basin near Chicago, six gauges inform predictions over 4.2 million 30-m grid cells. Terrain improves the predictions most, rainfall keeps errors from growing over longer horizons, and together they reduce errors by about 40% compared with gauge records alone. When future rainfall is known, errors remain near 15% one day ahead, compared with nearly 40% without rainfall. Given real instead of simulated gauge records, the model shifts its predictions toward the observed hydrographs at three of six gauges without retraining, and it runs about 150 times faster than the hydrodynamic model. These results show how sparse gauges, terrain, and rainfall can be combined into fast, continuously updated flood predictions, a step toward operational flood digital twins that still requires testing with real-time data and rainfall forecasts.
Chinese Translation
应急管理人员需要了解洪水水位的分布、深度以及在未来几小时内的变化情况，覆盖整个河流流域。然而，在洪水期间，实时测量仅来自少数几个水流测量站，而高分辨率的水动力模型在每次新数据到达时重新运行或作为大型集合运行的成本过高。我们提出了C-STRIDE，这是一种基于观测驱动的人工智能数字双胞胎，它将来自少数水流测量站的短期记录与地形和降雨数据结合，生成流域范围内的水深地图，并将这些预测扩展至未来一天。该模型基于经过校准的二维水动力模型的模拟进行训练，无需单独的数据同化步骤。在位于芝加哥附近的德斯普兰河流域，六个测量站为420万个30米网格单元的预测提供信息。地形对预测的改善效果最为显著，降雨则防止了误差在更长时间范围内的增长，两者结合使得与仅使用测量站记录相比，误差减少了约40%。当未来降雨已知时，预测的误差在一天前保持在15%左右，而没有降雨时则接近40%。在使用真实而非模拟的测量记录时，该模型在不重新训练的情况下，将其预测向六个测量站中的三个观测水文曲线靠拢，并且运行速度比水动力模型快约150倍。这些结果表明，稀疏的测量站、地形和降雨可以结合起来，形成快速、持续更新的洪水预测，这是朝着操作性洪水数字双胞胎迈出的重要一步，但仍需通过实时数据和降雨预测进行测试。
cs.AI / 67 / 2609.39026
Search Shapes Conclusions: Auditing Evidence Selection Bias in Deep Research Agents
搜索形状结论：审计深度研究代理中的证据选择偏差
Abstract
Deep Research agents synthesize evidence into cited reports, yet a well-cited report can still reach a misleading conclusion. Citation correctness checks whether cited sources support individual claims. It does not show whether adaptive search exposed a representative view of all documents made available for evaluation, which we call the candidate pool. Early findings redirect later queries, document choices, and stopping, so the documents an agent reads form a selective sample. Existing evaluations rarely account for this selection. We formulate the problem as adaptive evidence sampling and introduce Causal Evidence Selection Correction (CESS). CESS predicts each candidate document's evidence direction and corrects the candidate-pool average using the logged probabilities of selecting each document and reaching each search round. Shrinkage stabilizes short searches, while intervals replace point estimates when some documents cannot be sampled. We also prove that estimating the average evidence direction of a common pool differs from measuring how a change in search policy alters the evidence read. The latter requires intervention. On questions from the MS2 systematic-review benchmark, CESS reduces mean absolute error against the candidate-pool average by $9.2\%$ and reduces the estimate's change under opposing document rankings by $39.4\%$ relative to averaging the evidence scores of documents read. Across trajectories from a public Open Deep Research agent, the corresponding reductions reach $60.1\%$ and $87.2\%$. A further 4,800 trajectories under paired interventions confirm that correcting a pool estimate and measuring a policy effect are different tasks. CESS therefore audits whether the evidence direction underlying a report reflects the documents available for evaluation, while a separate intervention analysis measures the effect of search decisions.
Chinese Translation
深度研究代理将证据综合成引用报告，但即使是引用率高的报告也可能得出误导性结论。引用正确性检查所引用的来源是否支持各个主张，但并未显示自适应搜索是否展示了所有可供评估文档的代表性视图，我们称之为候选池。早期发现会重定向后续查询、文档选择和停止，因此代理所阅读的文档形成了一个选择性样本。现有评估很少考虑这种选择。我们将此问题表述为自适应证据采样，并引入因果证据选择修正（Causal Evidence Selection Correction, CESS）。CESS 预测每个候选文档的证据方向，并利用选择每个文档和达到每个搜索轮次的记录概率修正候选池的平均值。收缩方法稳定短期搜索，而在某些文档无法被采样时，区间替代点估计。我们还证明，估计共同池的平均证据方向与测量搜索策略变化如何影响所阅读证据是不同的任务，后者需要干预。在 MS2 系统评价基准的问题上，CESS 将相对于候选池平均值的平均绝对误差降低了 $9.2\%$，并将相对于阅读文档证据分数平均值的对立文档排名下估计的变化降低了 $39.4\\%$。在来自公共开放深度研究代理的轨迹中，相应的减少达到了 $60.1\\%$ 和 $87.2\\%$。在配对干预下的进一步 4,800 条轨迹确认了修正池估计和测量政策效果是不同的任务。因此，CESS 审计报告所依据的证据方向是否反映了可供评估的文档，而单独的干预分析则测量搜索决策的效果。
cs.AI / 68 / 2609.39076
Multi-LLM Collaborative Alignment via Stackelberg Games
通过斯塔克尔博格博弈实现多语言模型协同对齐
Abstract
A pool of language models can collaborate and improve collectively by learning from one another's responses. These interactions depend on the instructions used during training. Existing methods typically sample instructions uniformly, even though their usefulness may change as the models improve: an instruction on which models' responses once differed in quality may later be answered equally well, while a previously difficult instruction may begin to provide a useful learning signal. We propose Stackelberg Alignment, a game-theory-inspired leader-follower framework that turns instruction selection into an adaptive curriculum. An EXP3 bandit acts as the leader, allocating a fixed sampling budget across instructions and updating its sampling distribution using a reward that combines instruction difficulty and response discriminability. The language models act as followers: they respond to the selected instructions, evaluate one another's responses, and learn from the resulting preference signals through DPO or GRPO. The framework uses Elo-style reputation-weighted peer judgment and reputation-based opponent matching to support reliable and competitive model interactions. Experiments across three heterogeneous model pools and 12 benchmarks spanning scientific discovery, reasoning, code, instruction following, and knowledge show that Stackelberg Alignment achieves the highest macro-average across three diverse model pools, outperforming the strongest training-time baseline by up to 7.4% and the best static inference baseline by 12-25%. Analysis confirms that the adaptive leader concentrates duels on the most informative instructions, and ablations show that both reputation-weighted judgment and reputation-based matching improve the effectiveness of multi-LLM evolution.
Chinese Translation
一组语言模型可以通过相互学习对方的响应进行协作和集体提升。这些互动依赖于训练过程中使用的指令。现有方法通常均匀地抽样指令，尽管随着模型的提升，其有效性可能会发生变化：曾经在质量上存在差异的指令，后来可能会得到同样好的回答，而之前困难的指令可能开始提供有用的学习信号。我们提出了斯塔克尔博格对齐（Stackelberg Alignment），这是一种受博弈论启发的领导者-追随者框架，将指令选择转变为自适应课程。EXP3赌博机作为领导者，在指令之间分配固定的抽样预算，并使用结合指令难度和响应可区分性的奖励来更新其抽样分布。语言模型作为追随者：它们对所选指令作出响应，评估彼此的响应，并通过DPO或GRPO从结果偏好信号中学习。该框架使用Elo风格的声誉加权同行评判和基于声誉的对手匹配，以支持可靠且具有竞争性的模型互动。在三个异构模型池和12个基准测试中进行的实验显示，斯塔克尔博格对齐在三个不同的模型池中实现了最高的宏平均，超越了最强训练时基线高达7.4%，以及最佳静态推理基线12-25%。分析确认自适应领导者将对决集中在最具信息性的指令上，消融实验表明，声誉加权评判和基于声誉的匹配均提高了多语言模型演化的有效性。
cs.AI / 69 / 2609.39101
Beyond Prediction: Steering VLM Agents with Retrospective World Modeling
超越预测：通过回顾性世界建模引导VLM代理
Abstract
Equipping VLM agents with world modeling capabilities has shown strong potential for complex reasoning and long-horizon planning, while reducing the dependence of policy learning on costly real-world interactions. Existing methods mainly rely on prospective simulation to predict the consequences of candidate actions. However, this forward-only paradigm focuses on what will happen next and provides limited constraints for verifying whether an action is causally consistent with the observed state transition, which can lead to plausible-looking but physically incoherent behaviors. In this paper, we challenge the view of world modeling as only prospective prediction and introduce Retrospective World Modeling, a new agent learning paradigm that enables agents to reason backward by estimating the retrospective attribution distribution $P(\hat{a}{t}|s_t, s{t+1})$ for the action that most likely caused a given transition. Based on this capability, we formulate the Self-Consistency Reward (SCR), an intrinsic signal that measures the probabilistic consistency between the policy action and the retrospective explanation. Integrating SCR into reinforcement learning provides dense transition-level feedback and steers agents toward behaviors that are both task-effective and physically grounded. Extensive experiments across diverse agentic tasks show that our method substantially improves policy robustness and generalization over prospective-only world modeling baselines.
Chinese Translation
为VLM代理配备世界建模能力显示出在复杂推理和长远规划方面的强大潜力，同时减少了政策学习对昂贵的现实世界交互的依赖。现有方法主要依赖前瞻性模拟来预测候选动作的后果。然而，这种仅向前的范式专注于接下来会发生什么，提供了有限的约束来验证某个动作是否与观察到的状态转移因果一致，这可能导致看似合理但在物理上不连贯的行为。在本文中，我们挑战了将世界建模仅视为前瞻性预测的观点，并引入了回顾性世界建模（Retrospective World Modeling），这是一种新的代理学习范式，使代理能够通过估计最可能导致给定转移的动作的回顾性归因分布 $P( ilde{a}_{t}|s_t, s_{t+1})$ 进行反向推理。基于这一能力，我们制定了自一致性奖励（Self-Consistency Reward, SCR），这是一种内在信号，用于测量政策动作与回顾性解释之间的概率一致性。将SCR整合到强化学习中提供了密集的转移级反馈，并引导代理朝向既有效任务又在物理上合理的行为。广泛的实验表明，我们的方法在各种代理任务中显著提高了政策的鲁棒性和对前瞻性世界建模基线的泛化能力。
cs.AI / 70 / 2609.39106
Reserve-Aware Contrast Certificates for Conservative Bandits with Uncertain Baselines
考虑保留的保守强盗问题的对比证书与不确定基准
Abstract
Conservative bandits must improve an incumbent policy without exhausting a prescribed performance budget. When the incumbent is uncertain, separately bounding candidate and baseline rewards can charge twice for shared estimation error. We develop Reserve-C4B around the baseline-relative contrast itself. A shared confidence set yields an exact expression for this avoidable penalty and a tighter admissibility test at every fixed history. A reserve ledger separates statistical evidence from permitted performance deficit; a prefix-refresh extension recertifies accumulated decisions under the current confidence set without discarding previously certified credit. For linear rewards, self-normalized confidence sets provide simultaneous validity over time and adaptively generated candidates, and the resulting policy satisfies a conditional-mean performance constraint with high probability. Reproducible experiments isolate certificate coupling, prefix refresh, and historical information, showing large reductions in baseline fallback while exposing the limitations of frozen certificates.
Chinese Translation
保守强盗问题必须在不耗尽规定的性能预算的情况下改善现有策略。当现有策略不确定时，分别界定候选奖励和基准奖励可能会因共享估计误差而重复收费。我们围绕基准相关的对比本身开发了Reserve-C4B。共享置信集为这一可避免的惩罚提供了精确的表达，并在每个固定历史下提供了更严格的可接受性测试。保留账本将统计证据与允许的性能赤字分开；前缀刷新扩展在当前置信集下重新认证累积决策，而不丢弃先前认证的信用。对于线性奖励，自归一化置信集在时间和自适应生成的候选者上提供了同时有效性，所得到的策略以高概率满足条件均值性能约束。可重复的实验隔离了证书耦合、前缀刷新和历史信息，显示出基准回退的大幅减少，同时揭示了冻结证书的局限性。
cs.AI / 71 / 2609.39107
MASCRDM: Multi-Agent System for Compliance Risk Detection and Mitigation in Training Process of Large Language Models
MASCRDM：大型语言模型训练过程中的合规风险检测与缓解的多智能体系统
Abstract
Large Language Models (LLMs) have been applied in various fields. However, ensuring compliance and safety of LLMs, such as avoiding discrimination and bias, still remains a challenge. Current efforts mainly focus on detecting and filtering inputs and outputs of the trained models, rather than studying the intrinsic architecture of the models in real-time. To tackle this challenge, we analyze the LLMs training process and discover two critical issues: 1) Most of the existing methods are predominantly static in their approach to detection and filtering, achieving only localized optimizations without systematically enhancing the compliance of LLMs. 2) Another issue with existing approaches is the lack of real-time risk detection and mitigation across the full training process, which leads to limited flexibility. Motivated by these, we propose MASCRDM (Multi-Agent System for Compliance Risk Detection and Mitigation) during the LLM training process. Firstly, we develop a set of compliance rules based on existing Artificial Intelligence (AI) laws and a compliance-specific LLM with the instruction of compliance law experts. Then, we deconstruct LLMs into several components and identify key nodes based on the compliance knowledge graph. During LLMs training, we implement our multiple agents in the whole process, giving compliance risk alerts and suggestions for LLM developers. Experiments on discrimination and bias benchmark demonstrate that our multi-agent system can effectively improve the compliance while maintaining reasonable semantic performance. The results indicate that our method provides an executable path for mitigating compliance risk from within the LLMs systematically.
Chinese Translation
大型语言模型（LLMs）已在多个领域得到应用。然而，确保LLMs的合规性和安全性，如避免歧视和偏见，仍然是一个挑战。目前的努力主要集中在检测和过滤训练模型的输入和输出，而不是实时研究模型的内在架构。为了解决这一挑战，我们分析了LLMs的训练过程，并发现了两个关键问题：1）现有方法在检测和过滤方面主要是静态的，仅实现了局部优化，而未系统性地增强LLMs的合规性。2）现有方法的另一个问题是缺乏对整个训练过程的实时风险检测和缓解，导致灵活性有限。基于这些动机，我们在LLM训练过程中提出了MASCRDM（合规风险检测与缓解的多智能体系统）。首先，我们根据现有的人工智能（AI）法律制定了一套合规规则，并在合规法律专家的指导下构建了一个合规特定的LLM。然后，我们将LLMs拆解为多个组件，并根据合规知识图谱识别关键节点。在LLMs训练过程中，我们在整个过程中实施了多个智能体，为LLM开发者提供合规风险警报和建议。关于歧视和偏见基准的实验表明，我们的多智能体系统能够有效提高合规性，同时保持合理的语义性能。结果表明，我们的方法为系统性地从LLMs内部缓解合规风险提供了一条可执行的路径。
cs.AI / 72 / 2609.39139
BELIEFRAG: Making Adaptive RAG State-Aware under Evolving Evidence
BELIEFRAG：在不断演变的证据下使自适应 RAG 具备状态感知能力
Abstract
Adaptive RAG uses signals such as confidence, relevance, support, and retrieval quality to decide when to search or correct evidence. In multi-step retrieval, however, these local signals must be combined into a persistent view of what the current evidence supports, what remains missing, and which action should follow. Existing methods often use such signals as separate triggers, making it difficult to preserve a coherent evidence state across a trajectory; we call this problem evidence-state fragmentation. We introduce BELIEFRAG, a closed-loop controller that updates an explicit state over sufficiency, reliability, conflict, uncertainty, evidence gaps, and acquisition cost, then chooses among retrieval, query rewriting, verification, answering, stopping, and abstention. Across six QA benchmarks with gpt-oss-120b, BELIEFRAG reaches mean token F1 0.572 with 3.89k tokens per question, outperforming fixed iterative retrieval (0.555 F1) while using 39% fewer tokens. The same quality-cost pattern transfers to Qwen3-32B, where BELIEFRAG reaches 0.552 F1 versus 0.523 for iterative retrieval while using 35% fewer tokens. Analysis shows that the main gains come from corrective re-retrieval rather than pruning alone, while several belief dimensions are redundant and calibrated answerability plays the strongest operational role. Calibration improves threshold stability across related evidence sources, although source shift can still invalidate the same decision signal.
Chinese Translation
自适应 RAG 使用信心、相关性、支持度和检索质量等信号来决定何时搜索或修正证据。然而，在多步骤检索中，这些局部信号必须结合成对当前证据支持的持久视图、缺失的内容以及后续应采取的行动。现有方法通常将这些信号作为独立触发器使用，这使得在整个轨迹中保持一致的证据状态变得困难；我们称这个问题为证据状态碎片化。我们提出了 BELIEFRAG，一种闭环控制器，它在充分性、可靠性、冲突、不确定性、证据缺口和获取成本等方面更新显式状态，然后在检索、查询重写、验证、回答、停止和放弃之间进行选择。在使用 gpt-oss-120b 的六个问答基准上，BELIEFRAG 达到了平均令牌 F1 值 0.572，每个问题使用 3.89k 令牌，优于固定迭代检索（0.555 F1），同时减少了 39% 的令牌使用。相同的质量-成本模式也适用于 Qwen3-32B，BELIEFRAG 达到 0.552 F1，而迭代检索为 0.523，同时减少了 35% 的令牌使用。分析表明，主要的收益来自于纠正性重新检索，而非单纯的修剪，同时几个信念维度是冗余的，校准的可回答性发挥了最强的操作作用。校准提高了相关证据源之间的阈值稳定性，尽管源的变化仍可能使相同的决策信号失效。
cs.AI / 73 / 2609.39140
Schema: Discovering Unknown Environments via Agentic Program Induction
Schema：通过自主程序归纳发现未知环境
Abstract
Learning to complete tasks in unfamiliar environments with unknown rules remains a key challenge for LLM agents. Current LLM agents often record their discoveries in prose, which may not provide a compact, explicit account of how the environment works. Inspired by how scientists organize observations into testable, predictive theories, we introduce Schema, an agent harness that organizes learning and action through interactive program induction. The LLM agent decides what to investigate and how to act, expressing its evolving understanding of the environment as executable programs. The harness consists of a persistent program workspace and a small set of interfaces for checking these programs against the interaction history, planning within them, and executing plans under step-by-step verification. Schema raises ARC-AGI-3 RHAE from 58.7% to 99.2% with the same base model, solves 100% of the public DiG-bench games, and reaches the median performance of the top-50 human players on MazeBench. Extensive analysis shows the effectiveness of Schema in unknown mechanism discovery, and ablations confirm the contribution of each component.
Chinese Translation
在未知规则的陌生环境中完成任务的学习仍然是大型语言模型（LLM）代理面临的关键挑战。目前的LLM代理通常以散文的形式记录其发现，这可能无法提供关于环境运作的简洁、明确的说明。受到科学家将观察结果组织成可测试、可预测理论的启发，我们提出了Schema，这是一种通过交互式程序归纳来组织学习和行动的代理工具。LLM代理决定调查内容和行动方式，以可执行程序的形式表达其对环境的不断理解。该工具包括一个持久的程序工作区和一小组接口，用于根据交互历史检查这些程序、在其中进行规划以及在逐步验证下执行计划。Schema将ARC-AGI-3 RHAE的得分从58.7%提升至99.2%，在相同的基础模型下解决了100%的公共DiG-bench游戏，并在MazeBench上达到了前50名人类玩家的中位表现。广泛的分析表明Schema在未知机制发现中的有效性，消融实验确认了各个组件的贡献。
cs.AI / 74 / 2609.39143
RefCon: Iterative Refinement and Contrastive Memory Extraction for Context-Evolving Agent
RefCon：用于上下文演变代理的迭代精炼与对比记忆提取
Abstract
Long-horizon agent interactions generate useful but noisy experience, and retraining models to absorb it is expensive. Context-evolving agents therefore need memory extraction methods that improve with more test-time compute without relying on gold labels. We propose RefCon, which combines sequential self-refinement with parallel self-contrast to extract higher-quality memories without gold labels. Evaluated on AppWorld and BFCL-V3 across multiple context-evolving agent frameworks, RefCon delivers strong and consistent gains, including relative improvements of 21.6% on ACE and 16.6% on ReMe over no-scaling baselines, while a diversity-focused variant (DivCon) achieves a 35.5% gain on ReasoningBank. RefCon consistently outperforms existing baselines without ground-truth labels, and generalizes across model scales and to software engineering tasks, where it surpasses even ground-truth baselines. We further analyze the accuracy-token trade-off and scaling behavior, showing RefCon maintains favorable efficiency and continues to improve as more trajectories are used, unlike diversity-only scaling which saturates earlier.
Chinese Translation
长时间跨度的代理交互生成有用但噪声较大的经验，而重新训练模型以吸收这些经验的成本很高。因此，上下文演变代理需要能够在不依赖金标准标签的情况下，通过更多的测试时间计算来改进的记忆提取方法。我们提出了RefCon，它结合了顺序自我精炼与并行自我对比，以在没有金标准标签的情况下提取更高质量的记忆。在多个上下文演变代理框架下对AppWorld和BFCL-V3进行评估，RefCon提供了强劲且一致的提升，包括在ACE上相较于无缩放基线的相对提升21.6%，在ReMe上提升16.6%，而一个以多样性为重点的变体（DivCon）在ReasoningBank上实现了35.5%的提升。RefCon在没有真实标签的情况下始终优于现有基线，并且在模型规模和软件工程任务中具有良好的泛化能力，甚至超越了真实标签基线。我们进一步分析了准确性与代币的权衡以及缩放行为，显示RefCon保持了良好的效率，并随着使用更多轨迹而持续改进，不同于仅关注多样性的缩放方法，它较早达到饱和。
cs.AI / 75 / 2609.39146
MADBench: Benchmarking the Security of Multi-Agent Debate
MADBench：多智能体辩论安全性基准测试
Abstract
Multi-agent debate (MAD) can improve large language model (LLM) reasoning by allowing multiple agents to exchange and critique their answers to the same task. However, the interactions that enable agents to correct mistakes can also spread adversarial errors and steer the agents toward an incorrect answer. Although some efforts have been made to examine particular attack types on MAD, systematic evaluation of MAD under diverse attacks remains limited. A central question is whether debate mitigates adversarial influence or amplifies it. In this paper, we present MADBench, a benchmark for evaluating the security of MAD. We organize attacks into a layered taxonomy following the MAD workflow, incorporating both established attacks and new strategies tailored to debate. We evaluate six attack families over 356 source tasks and 3,958 test cases, examining their effects on the final answer and the propagation of adversarial influence. Our results show that, under attacks, MAD does not necessarily improve LLM reasoning. Compared with a single-agent baseline, MAD can mitigate attacks on answer accuracy in question-answering tasks while amplifying unauthorized reads or writes in both question-answering and workspace tasks. Moreover, even when three out of five agents collude, the attack changes the final answer from correct to wrong on only 28.30\% of tasks answered correctly without attack, while only 3.26\% of initially correct honest agents switch to wrong answers during debate.
Chinese Translation
多智能体辩论（MAD）通过允许多个智能体交换和批评对同一任务的答案，从而提高大型语言模型（LLM）的推理能力。然而，使智能体能够纠正错误的互动也可能传播对抗性错误，并引导智能体朝向错误答案。尽管已有一些研究探讨了对MAD的特定攻击类型，但在多种攻击下对MAD的系统评估仍然有限。一个核心问题是辩论是否减轻了对抗性影响或放大了这种影响。本文提出了MADBench，一个用于评估MAD安全性的基准测试。我们根据MAD工作流程将攻击组织成分层分类法，结合了既有攻击和针对辩论的新策略。我们对356个源任务和3,958个测试案例进行了六种攻击类型的评估，考察了它们对最终答案的影响及对抗性影响的传播。我们的结果表明，在攻击下，MAD并不一定改善LLM的推理能力。与单智能体基线相比，MAD能够在问答任务中减轻对答案准确性的攻击，同时在问答和工作区任务中放大未经授权的读取或写入。此外，即使五个智能体中有三个合谋，攻击也仅在28.30%的原本正确的任务中将最终答案从正确改为错误，而在辩论过程中，只有3.26%的最初正确的诚实智能体转向错误答案。
cs.AI / 76 / 2609.39148
Do Self-Evolving Skills Generalize to Held-Out Tasks?
自我进化技能是否能够推广到未见任务？
Abstract
AI agents can externalize what they learn from past tasks into reusable \emph{skills}, such as procedures, checklists, code, or other executable artifacts, that can be retrieved and reused when solving new tasks. Self-evolving skill methods keep rewriting these skills after each round of practice on training tasks, and the skill is then used on new tasks of the same kind. We ask a question: does the improvement a skill shows on its training tasks carry over to new test tasks? We test five self-evolving methods and a one-shot skill on six benchmarks, with the same model, the same agent, and the same train/test split for every method. Of the 21 skills that improve on their training tasks, 5 keep all of that improvement on the test tasks, 13 keep part of it, and 3 keep none of it. No existing method is best everywhere. When we read the skills, the ones that carry over badly often fix details that should depend on the task, such as column names and output files, or turn a fix for one failure into a rule for every task. An LLM judge that reads the skill content can often see this: it ranks finished skills the same way the test results do in 86\% of pairs. But it predicts the effect of a single edit poorly, so edits still have to be tested by running them. Based on these findings, we describe Generalizable Skill Optimization (GSO), which keeps only a guide for writing skills and writes a new skill for each task; it scores highest on all six benchmarks.
Chinese Translation
人工智能代理能够将其从过去任务中学到的知识外化为可重用的技能，例如程序、清单、代码或其他可执行的工件，这些技能可以在解决新任务时被检索和重用。自我进化技能方法在每轮训练任务的实践后不断重写这些技能，然后将这些技能应用于相同类型的新任务。我们提出一个问题：技能在训练任务上表现出的改进是否能够转移到新的测试任务上？我们在六个基准上测试了五种自我进化方法和一种一次性技能，使用相同的模型、相同的代理和相同的训练/测试划分。21个在训练任务上有所改进的技能中，有5个在测试任务上保持了所有的改进，13个保持了部分改进，3个没有保持任何改进。没有任何现有方法在所有场景下都是最佳的。当我们分析这些技能时，转移效果差的技能往往修复了一些应依赖于任务的细节，例如列名和输出文件，或者将一个失败的修复转化为适用于每个任务的规则。一个阅读技能内容的大型语言模型（LLM）评估者通常能够识别这一点：在86%的配对中，它对完成技能的排名与测试结果相同。但它对单次编辑的影响预测较差，因此仍需通过实际运行来测试编辑。基于这些发现，我们描述了可推广技能优化（Generalizable Skill Optimization, GSO），该方法仅保留编写技能的指南，并为每个任务编写新的技能；它在所有六个基准上得分最高。
cs.AI / 77 / 2609.39149
Rep2Skill: Representation-Guided Skill Self-Evolution for LLM Agents
Rep2Skill：基于表征引导的LLM智能体技能自我演化
Abstract
Textual skills enable large language model (LLM) based agents to accumulate reusable procedural knowledge without updating model parameters. Yet existing skill evolution remains largely confined to the text space: an optimizer must diagnose success and failure patterns, and revise skills solely from long execution trajectories and sparse task outcomes. This text-only paradigm leaves the agent's internal representations, which contain rich records of its evolving execution state, outside the skill optimization loop. We ask whether an agent can improve its external textual skills by reflecting on its own internal representations. We introduce Rep2Skill, a representation-guided framework for self-evolution on agent skills. Specifically, upon the collected agent rollouts, Rep2Skill models their internal model representation trajectories to localize turns that deviate from successful execution dynamics, and it further interprets these signals alongside the execution contexts as actionable textual feedback for targeted skill revision. Experiments on two agent environments with two open-source LLMs show that Rep2Skill consistently outperforms text-only approaches in the self-evolution setting, where the same LLM serves as both executor and optimizer without a stronger external model. This establishes a promising direction moving agent self-improvement beyond text-only reflection.
Chinese Translation
文本技能使基于大型语言模型（LLM）的智能体能够在不更新模型参数的情况下积累可重用的程序知识。然而，现有的技能演化仍然主要局限于文本空间：优化器必须诊断成功和失败模式，并仅从长执行轨迹和稀疏任务结果中修订技能。这种仅限文本的范式使得智能体的内部表征——其中包含其不断演变的执行状态的丰富记录——被排除在技能优化循环之外。我们探讨智能体是否可以通过反思自身的内部表征来改善其外部文本技能。我们提出了Rep2Skill，一个基于表征引导的智能体技能自我演化框架。具体而言，在收集到的智能体执行结果上，Rep2Skill建模其内部模型表征轨迹，以定位偏离成功执行动态的转折点，并进一步将这些信号与执行上下文结合解读，作为针对性技能修订的可操作文本反馈。在两个智能体环境和两个开源LLM的实验中，Rep2Skill在自我演化设置中始终优于仅基于文本的方法，其中同一LLM既作为执行者又作为优化器，而没有更强的外部模型。这为推动智能体自我改进超越仅限文本的反思建立了一个有前景的方向。
cs.AI / 78 / 2609.39166
Beyond the Remembered World: Predictive 4D Belief for Persistent Navigation in Evolving Worlds
超越记忆世界：用于动态演变环境中持久导航的预测性四维信念
Abstract
Persistent spatial memory enables embodied agents to navigate familiar environments across repeated visits. However, targets may move while unobserved, including during navigation, making remembered locations unreliable by the time an agent arrives. Despite advances in memory retrieval and state prediction, accounting for continued hidden world evolution and revising beliefs under limited visibility remain challenging. We study Evolving-World Navigation, where agents infer target locations from intermittent observations, predict their states at inspection time, and revise beliefs using visual evidence. We propose EvolvingNav, which constructs a time-indexed belief from timestamped 3D object histories through a structured persistence-relocation model. The belief distinguishes persistence at the last observed location from relocation to alternative locations and retains probability mass outside the known candidate set. An event-driven filter propagates the current belief as time elapses, forecasts target occupancy at candidate inspection times, and incorporates new RGB-D evidence. Negative observations downweight location hypotheses according to calibrated, visibility-conditioned detection probabilities, while evidence tracking prevents repeated use of the same observations. A frozen, zero-shot vision-language controller uses the updated belief to choose actions and replan. We further introduce EvoWorld-Bench, a benchmark grounded in human activity traces, comprising 54 scenes and 803,680 tasks with controlled changes before and during navigation. In simulation and real-robot experiments, EvolvingNav improves navigation success and search efficiency over the evaluated baselines. Paired experiments show the clearest gains under learnable temporal patterns, while ablations demonstrate the value of preserving uncertainty and incorporating visibility-aware evidence.
Chinese Translation
持久的空间记忆使具身智能体能够在多次访问中导航熟悉环境。然而，目标可能在未被观察时发生移动，包括导航过程中，这使得智能体到达时记忆中的位置变得不可靠。尽管在记忆检索和状态预测方面已有进展，但考虑持续的隐藏世界演变并在有限可视性下修正信念仍然具有挑战性。我们研究了动态演变世界导航（Evolving-World Navigation），其中智能体通过间歇性观测推断目标位置，预测其在检查时刻的状态，并利用视觉证据修正信念。我们提出了EvolvingNav方法，该方法通过结构化的持久-迁移模型，从带时间戳的三维物体历史构建时间索引的信念。该信念区分了目标在最后观测位置的持续存在与迁移到其他位置的可能性，并保留了已知候选集合之外的位置概率质量。基于事件驱动的滤波器随着时间推移传播当前信念，预测候选检查时间的目标占据情况，并融合新的RGB-D证据。负面观测根据校准的、基于可视性的检测概率对位置假设进行权重调整，同时证据追踪防止重复使用相同观测。一个冻结的零样本视觉-语言控制器利用更新后的信念选择动作并重新规划。我们还引入了EvoWorld-Bench基准，基于人类活动轨迹，包含54个场景和803,680个任务，任务在导航前后均有受控变化。在仿真和真实机器人实验中，EvolvingNav在导航成功率和搜索效率上均优于评估的基线方法。配对实验显示，在可学习的时间模式下效果提升最为显著，消融实验则证明了保持不确定性和融合可视性感知证据的重要性。
cs.AI / 79 / 2609.39168
Reinforcing Multimodal Reasoning via Token-Level Perception-Grounded Advantage Estimation
通过基于标记级感知的优势估计强化多模态推理
Abstract
Reinforcement Learning with Verifiable Rewards (RLVR) has improved the reasoning capabilities of Multimodal Large Language Models (MLLMs), yet existing frameworks rely on coarse, sequence-level reward signals that lack the fine-grained supervision over the visually-grounded steps within a multimodal reasoning chain. We investigate this gap through the lens of two token-level metrics: visual dependency (i.e. how much a token's prediction relies on the input image features) and predictive entropy. Our empirical analysis reveals two key findings: (1) correct reasoning chains exhibit a markedly sharper entropy reduction as visual grounding intensifies, compared to incorrect ones; (2) pivotal tokens, those whose misprediction triggers reasoning collapse, are statistical outliers in the joint distribution of visual dependency and predictive entropy derived from correct chains. Motivated by these findings, we propose token-level perception-grounded advantage estimation (TPAE), which estimates token-level advantages by measuring each token's statistical consistency with the vision-entropy patterns of correct rollouts. TPAE leverages this granular score to modulate the sequence-level advantage, producing a fine-grained supervision signal that can be integrated into various RLVR frameworks. Extensive experiments on seven benchmarks show that TPAE consistently outperforms leading strong baselines, yielding more stable and efficient optimization for multimodal reasoning. The code is publicly available at https://github.com/Zhihan72/TPAE.
Chinese Translation
可验证奖励的强化学习（RLVR）提高了多模态大型语言模型（MLLMs）的推理能力，但现有框架依赖于粗糙的序列级奖励信号，缺乏对多模态推理链中视觉基础步骤的细粒度监督。我们通过两个标记级指标的视角来研究这一差距：视觉依赖性（即一个标记的预测在多大程度上依赖于输入图像特征）和预测熵。我们的实证分析揭示了两个关键发现：（1）正确的推理链在视觉基础增强时表现出明显更陡的熵减少，与错误的推理链相比；（2）关键标记，即那些错误预测会导致推理崩溃的标记，在从正确链派生的视觉依赖性和预测熵的联合分布中是统计异常值。基于这些发现，我们提出了基于标记级感知的优势估计（TPAE），该方法通过测量每个标记与正确回滚的视觉-熵模式的统计一致性来估计标记级优势。TPAE利用这一细粒度评分来调节序列级优势，产生可以集成到各种RLVR框架中的细粒度监督信号。在七个基准上的广泛实验表明，TPAE始终优于领先的强基线，为多模态推理提供了更稳定和高效的优化。代码已公开发布在 https://github.com/Zhihan72/TPAE。
cs.AI / 80 / 2609.39220
Learning Process Rewards via Reasoning State Propagation
通过推理状态传播学习过程奖励
Abstract
Process reward models (PRMs) have demonstrated notable effectiveness in test-time scaling and reinforcement learning by providing fine-grained signals for evaluating intermediate reasoning states, but their training relies heavily on costly process annotations. A natural way to alleviate this dependence is to complement limited process supervision with scalable outcome supervision. However, existing PRMs often model reasoning prefixes independently, providing no explicit mechanism for effectively using final outcome to guide the learning of intermediate reasoning states. We introduce Reasoning State Propagation (RSP), which represents each reasoning prefix with a binary validity state and models transitions between successive states across the reasoning trajectory. Specifically, RSP predicts a break probability that a valid state becomes invalid and a repair probability that an invalid state returns to valid. By propagating these transitions, RSP connects intermediate states to the final state, allowing process annotations to supervise intermediate states while outcome labels supervise the final state and can provide learning signals to preceding steps. Across reasoning search, response selection, and reinforcement learning, RSP consistently outperforms representative PRM baselines, with average improvements over Qwen2.5-Math-PRM of 5.6% in beam search and 2.1% in reinforcement learning.
Chinese Translation
过程奖励模型（PRMs）在测试时扩展和强化学习中表现出了显著的有效性，因为它们提供了用于评估中间推理状态的细粒度信号，但其训练在很大程度上依赖于昂贵的过程注释。缓解这种依赖的自然方法是用可扩展的结果监督来补充有限的过程监督。然而，现有的PRMs通常独立建模推理前缀，未提供有效利用最终结果来指导中间推理状态学习的明确机制。我们引入了推理状态传播（Reasoning State Propagation, RSP），该方法用二进制有效性状态表示每个推理前缀，并建模推理轨迹中连续状态之间的转变。具体而言，RSP预测有效状态变为无效的破坏概率和无效状态恢复为有效的修复概率。通过传播这些转变，RSP将中间状态与最终状态连接起来，使过程注释能够监督中间状态，而结果标签则监督最终状态，并可以为前面的步骤提供学习信号。在推理搜索、响应选择和强化学习中，RSP始终优于代表性的PRM基线，在Qwen2.5-Math-PRM上，束搜索平均提高了5.6%，强化学习提高了2.1%。
cs.AI / 81 / 2609.39228
Fyan: A Human--AI Harness with Semantic Auditing for Document-Level Formalization
Fyan：一种具有语义审计的人机协作工具，用于文档级形式化
Abstract
We present FYAN, a human--AI harness for document-level mathematical formalization. Rather than treating theorems in isolation, FYAN coordinates an end-to-end workflow spanning specification, proof planning, logical review, Lean proof construction, knowledge curation, and validation, with support for independent supervision and human guidance. A central component is evidence-grounded semantic auditing, which assesses whether formal statements faithfully preserve their informal specifications. A language model constructs structured evidence over local correspondences, omissions, scope, and logical relations, while a deterministic validator checks this evidence and produces reproducible judgments. When a substantive but admissible deviation is accepted, FYAN requires an explicit proof-transfer obligation connecting the formal statement back to a source-facing interpretation. With the same model (DeepSeek-V4.1-Flash) in every stage, FYAN proves 86 of 143 FormalTCS theorems under a strict Lean check, against 69 for a general agent harness, and raises the natural-language proof score from 0.501 to 0.851. On ConsistencyCheck, its semantic audit catches more inconsistent statements than a direct LLM judge, both on labels verified against the source (recall 0.777 vs. 0.636) and on the original labels (0.873 vs. 0.820), and localizes each mismatch it reports to a specific hypothesis, conclusion, or scope. FYAN also built ODENumLib, a 9,355-line Lean library for the numerical analysis of ordinary differential equation.
Chinese Translation
我们提出了FYAN，一种用于文档级数学形式化的人机协作工具。FYAN并不是将定理孤立对待，而是协调一个涵盖规范、证明规划、逻辑审查、Lean证明构建、知识管理和验证的端到端工作流程，并支持独立监督和人类指导。其核心组件是基于证据的语义审计，它评估形式声明是否忠实地保留了其非正式规范。语言模型在局部对应、遗漏、范围和逻辑关系上构建结构化证据，而确定性验证器则检查这些证据并生成可重复的判断。当接受实质性但可接受的偏差时，FYAN要求明确的证明转移义务，将形式声明与源面向的解释连接起来。在每个阶段使用相同的模型（DeepSeek-V4.1-Flash），FYAN在严格的Lean检查下证明了143个FormalTCS定理中的86个，而一般代理工具仅证明了69个，并将自然语言证明分数从0.501提高到0.851。在ConsistencyCheck上，其语义审计捕获到的更多不一致声明超过了直接的LLM判断，无论是在与源验证的标签上（召回率0.777对比0.636）还是在原始标签上（0.873对比0.820），并将每个报告的不匹配定位到特定的假设、结论或范围。FYAN还构建了ODENumLib，这是一个包含9,355行代码的Lean库，用于普通微分方程的数值分析。
cs.AI / 82 / 2609.39259
Effective Does Not Mean Useful: Conditional Functional Substitutability for Redundancy and Scaling in Transformers
有效并不意味着有用：变压器中的冗余和扩展的条件功能替代性
Abstract
Modern neural networks scale predictably, yet the mechanisms behind these regularities remain unclear. Neural redundancy is typically characterized by component importance or representational similarity, both indirect proxies. We view redundancy as an input-conditioned, dynamic relation: intermediate computational states are functionally redundant when they induce similar downstream responses. We introduce Conditional Functional Substitutability (CFS) to directly characterize such functional substitution. CFS exposes functional relations and reduction potential missed by conventional importance- and similarity-based measures. Across modalities and Transformer families, CFS reveals systematic functional reorganization with scale. Controlled scaling further shows that performance gains need not track growth in substitutability, while fixed-capacity models with more independent functional structure perform better, providing a functional account of diminishing returns. Predicted CFS further enables dynamic computation with a better performance--computation trade-off than importance-based component selection, suggesting new directions for redundancy-aware computation and more efficient model scaling.
Chinese Translation
现代神经网络的扩展性可预测，但其背后的机制仍不清楚。神经冗余通常通过组件重要性或表征相似性来表征，这两者都是间接的代理。我们将冗余视为一种输入条件的动态关系：当中间计算状态引发相似的下游反应时，它们在功能上是冗余的。我们引入条件功能替代性（Conditional Functional Substitutability, CFS）来直接表征这种功能替代。CFS揭示了传统基于重要性和相似性度量所忽视的功能关系和减少潜力。在不同的模态和变压器家族中，CFS揭示了随着规模的系统性功能重组。受控扩展进一步表明，性能提升不必与替代性增长相匹配，而具有更独立功能结构的固定容量模型表现更好，从而提供了收益递减的功能解释。预测的CFS进一步使动态计算成为可能，提供了比基于重要性的组件选择更好的性能-计算权衡，暗示了冗余感知计算和更高效模型扩展的新方向。
cs.AI / 83 / 2609.39294
ANI: Adaptive Numerical Injection for Unifying Semantic and Arithmetic Representations in Numerical Reasoning
ANI：用于统一语义和算术表示的自适应数值注入
Abstract
Precise numerical reasoning with Large Language Models (LLMs) is essential for expanding their applicability to complex real-world tasks. However, text-based tokenization often fragments numbers, significantly hindering precise arithmetic reasoning. Meanwhile, numerical embeddings, despite arithmetic precision, rely on context-agnostic substitution that disregards the semantic role of numbers as identifiers. To combine the complementary strengths, we propose \textbf{ANI (Adaptive Numerical Injection)}, a hybrid framework that governs the selective injection of numerical features based on the semantic context. By employing a context-aware gating mechanism, we selectively inject numerical embeddings (specifically FoNE) into the latent space, explicitly preserving nominal identifiers while enhancing quantitative operands. Through extensive evaluations across various LLMs, we demonstrate that ANI enhances MATH performance by 9.5 points over the official reference model, while maintaining robust performance on general linguistic benchmarks.
Chinese Translation
在大型语言模型（LLMs）中进行精确的数值推理对于扩展其在复杂现实任务中的适用性至关重要。然而，基于文本的标记化往往会将数字碎片化，显著阻碍精确的算术推理。同时，尽管数值嵌入具有算术精度，但它们依赖于忽视数字作为标识符的语义角色的上下文无关替代。为了结合这些互补的优势，我们提出了 extbf{ANI（自适应数值注入）}，这是一个混合框架，基于语义上下文来控制数值特征的选择性注入。通过采用上下文感知的门控机制，我们将数值嵌入（具体来说是FoNE）选择性地注入潜在空间，明确保留名义标识符，同时增强定量操作数。通过对各种LLMs进行广泛评估，我们证明ANI在MATH性能上比官方参考模型提高了9.5分，同时在一般语言基准测试中保持了稳健的性能。
cs.AI / 84 / 2609.39297
MiniRep: Robust Reputation-Based Aggregation for Multi-Agent Debate
MiniRep：针对多智能体辩论的鲁棒声誉基础聚合
Abstract
Autonomous agents powered by large language models (LLMs) are rapidly evolving into an open agentic ecosystem. To support trustworthy collaboration, industry initiatives increasingly assess agent reputation from past behavior and provide performance leaderboards. However, reputation derived from past performance may not reliably predict an agent's behavior on new tasks, particularly when malicious agents can adapt their behavior and influence other agents during collaboration. We study reputation in multi-agent debate (MAD), where multiple agents answer the same query, debate to improve their answers, and aggregate them into a final output. We present MiniRep, a reputation-based aggregation system for MAD under malicious agents. To ground our threat model in established research, we construct an attack taxonomy drawing on reputation-system attacks and software-testing mutation operators, covering strategic exploitation of reputation and subtle corruption of agent proposals. Guided by this taxonomy, MiniRep evaluates agents based on both their behavior on the current task and their reputation over time, while preventing groups of agents with highly similar responses from dominating the final decision. We assess MiniRep across diverse tasks, LLM-agent compositions, corruption placements, and attack types drawn from our taxonomy. Our experimental results show that, MiniRep outperforms both conventional MAD aggregation and conventional reputation-based approaches on MATH no matter being attacked or not. Also, under a heterogeneous 10-agent setting on MATH, MiniRep outperforms all baselines in all 28 attack conditions.
Chinese Translation
由大型语言模型（LLMs）驱动的自主智能体正在迅速演变为一个开放的智能体生态系统。为了支持可信赖的协作，行业倡议越来越多地根据过去的行为评估智能体的声誉，并提供性能排行榜。然而，基于过去表现得出的声誉可能无法可靠地预测智能体在新任务上的行为，特别是在恶意智能体能够适应其行为并在协作过程中影响其他智能体的情况下。我们研究多智能体辩论（MAD）中的声誉问题，在这种情况下，多个智能体对同一查询进行回答，辩论以改善他们的答案，并将其聚合为最终输出。我们提出了MiniRep，这是一个针对恶意智能体的MAD声誉基础聚合系统。为了将我们的威胁模型建立在已有研究的基础上，我们构建了一个攻击分类法，借鉴了声誉系统攻击和软件测试变异操作符，涵盖了声誉的战略性利用和智能体提案的微妙腐蚀。在这个分类法的指导下，MiniRep根据智能体在当前任务上的行为和其随时间变化的声誉进行评估，同时防止具有高度相似响应的智能体组主导最终决策。我们在多样化的任务、LLM-智能体组合、腐蚀位置和来自我们分类法的攻击类型下评估MiniRep。实验结果表明，无论是否受到攻击，MiniRep在MATH任务上均优于传统的MAD聚合和传统的声誉基础方法。此外，在MATH的异构10智能体设置下，MiniRep在所有28种攻击条件下均优于所有基线。
cs.AI / 85 / 2609.39325
WorkGenesis: Building the Worlds That Teach Agents to Work
WorkGenesis：构建教导智能体工作的世界
Abstract
The ability of Large Language Model (LLM) agents to complete daily and professional work is receiving increasing attention. Training such agents requires realistic work scenarios. Expert-authored occupational work is costly and slow to produce, while unconstrained synthesis often yields tasks with weak factual grounding or internally inconsistent requirements. To bridge this gap, we introduce WorkGenesis, a framework that constructs executable occupational work from real-world artifacts through two core technical innovations: (1) Evidence-Based Work Construction, which grounds each unit of work in real-world evidence by retrieving public files guided by O*NET occupational knowledge and synthesizing the surrounding context, companion materials, work request, and itemwise rubric around them; and (2) Execution-Guided Consistency Verification, which renders a reference deliverable inside the constructed work, attributes every unsatisfied rubric item to the agent, the task, or the rubric, and uses task and rubric defects as feedback to iteratively repair the work until it passes the audit. Experimental results demonstrate that Fx-Work-35B, trained with simple supervised fine-tuning (SFT) on only 20K units of work synthesized by WorkGenesis, achieves the highest scores among all comparable-scale baselines on the five reported metrics across GDPvalAA-v2, APEX-Agents-AA, and JobBench (31.00 versus 24.79 average score), and even surpasses frontier models such as the 1.6T DeepSeek-V4-Pro-Preview. These results show that WorkGenesis provides scalable training data for working agents.
Chinese Translation
大型语言模型（LLM）智能体完成日常和专业工作的能力正受到越来越多的关注。训练此类智能体需要现实的工作场景。专家撰写的职业工作成本高且生产速度慢，而不受限制的合成往往会产生事实基础薄弱或内部要求不一致的任务。为了解决这一问题，我们提出了WorkGenesis，一个通过两项核心技术创新从现实世界的文物构建可执行职业工作的框架：（1）基于证据的工作构建，该方法通过检索受O*NET职业知识指导的公共文件，并合成周围的上下文、伴随材料、工作请求和逐项评分标准，使每个工作单元扎根于现实世界的证据；（2）执行引导的一致性验证，该方法在构建的工作中呈现参考交付物，将每个未满足的评分标准项归因于智能体、任务或评分标准，并利用任务和评分标准的缺陷作为反馈，迭代修复工作，直到其通过审核。实验结果表明，Fx-Work-35B在仅使用WorkGenesis合成的20K个工作单元进行简单的监督微调（SFT）训练后，在GDPvalAA-v2、APEX-Agents-AA和JobBench的五个报告指标上，取得了所有可比规模基线中最高的分数（31.00对比24.79的平均分），甚至超越了1.6T的DeepSeek-V4-Pro-Preview等前沿模型。这些结果表明，WorkGenesis为工作智能体提供了可扩展的训练数据。
cs.AI / 86 / 2609.39341
Understanding as No-Arbitrage: Bounded Dutch Books as a Definition and Training Objective for Language Models
将理解视为无套利：有界荷兰书作为语言模型的定义和训练目标
Abstract
Does a language model merely predict tokens, or does it understand what it says? We make this question measurable by defining "understanding" through the lens of no-arbitrage. A model understands a vocabulary to a certain degree if a computationally bounded trader cannot extract guaranteed profit by betting against the model's probabilities on logically related claims (a "Dutch book"). We establish three theoretical results: first, because full logical coherence is computationally intractable, understanding is inherently graded, not absolute. Second, we prove that the exact optimum of standard next-token prediction is inherently incoherent across different question formats; the flaw lies in the training objective, not the architecture. Third, we show that uncertainty accumulates predictably along reasoning chains, making unjustified overconfidence an arbitrage opportunity in itself. To address this, we introduce Arbitr, a training framework where an adversarial trader penalizes the model for logical inconsistencies, paired with a calibration anchor to prevent uninformative collapse. Across five pre-registered experiments on Qwen2.5 and Phi-3.5 models, we demonstrate that standard models are highly exploitable across different phrasings. Arbitr reduces this exploitability by orders of magnitude without sacrificing task accuracy, and the effect successfully transfers to unseen logical patterns and new model families. Crucially, we uncover a scaling illusion: at 7B parameters, near-zero measured incoherence often coincides with extreme, unjustified confidence. We conclude that while Arbitr enforces rigorous logical consistency, coherence is a necessary condition for knowledge, but not a sufficient one
Chinese Translation
语言模型仅仅是预测标记，还是理解其所说内容？我们通过无套利的视角将“理解”这一问题量化。一个模型在一定程度上理解词汇，如果一个计算上有界的交易者无法通过对模型在逻辑相关声明上的概率进行对赌而提取保证利润（即“荷兰书”）。我们建立了三个理论结果：首先，由于完全的逻辑一致性在计算上是不可处理的，理解本质上是分级的，而非绝对的。其次，我们证明标准的下一个标记预测的确切最优解在不同问题格式之间本质上是不一致的；这个缺陷在于训练目标，而非架构。第三，我们展示了不确定性在推理链中可预测地累积，使得无根据的过度自信本身成为套利机会。为了解决这个问题，我们引入了Arbitr，一个训练框架，其中对抗性交易者因逻辑不一致性对模型进行惩罚，并配有校准锚点以防止无信息崩溃。在对Qwen2.5和Phi-3.5模型进行的五个预注册实验中，我们证明了标准模型在不同措辞下高度可被利用。Arbitr在不牺牲任务准确性的情况下将这种可利用性降低了几个数量级，并且这一效果成功转移到未见的逻辑模式和新的模型家族。关键是，我们揭示了一种规模错觉：在70亿参数时，近乎零的测量不一致性往往与极端的无根据自信同时出现。我们得出结论，尽管Arbitr强制执行严格的逻辑一致性，但一致性是知识的必要条件，而非充分条件。
cs.AI / 87 / 2609.39343
The Golden Path Hypothesis: Reusable Schedules in Diffusion Caching
黄金路径假设：扩散缓存中的可重用调度
Abstract
Diffusion caching accelerates generation by replacing transformer computation with cached or predicted features at selected denoising steps. We introduce the Golden Path Hypothesis (GPH): under fixed inference conditions, prompt-independent cache schedules can achieve final-output quality comparable to the best prompt-specific schedules across prompts. We investigate the GPH across ten caching methods, four image and video models, and three cache ratios. Prompt-adaptive methods repeatedly select a small number of schedules, and reusing their most frequent schedules on new prompts closely matches the quality of prompt-specific choices. Exhaustive evaluation of 1.4 million schedules on four examples further identifies prompt-independent schedules that remain competitive on unseen prompts. To explain this transfer, we analyze denoising trajectories and the accumulation of caching errors. Latent-state trajectories exhibit similar structures across datasets and seeds, while an exact error decomposition shows that accumulated effects of earlier errors predict final latent-state error better than local approximation errors. This motivates searching for end-to-end schedules using final-output quality. With only a small set of examples, the resulting golden paths transfer across prompts and datasets, and can be tuned to the desired quality objective, including reconstruction fidelity or perceptual similarity.
Chinese Translation
扩散缓存通过在选定的去噪步骤中用缓存或预测的特征替代变换器计算，从而加速生成。我们提出了黄金路径假设（Golden Path Hypothesis, GPH）：在固定的推理条件下，与最佳的特定提示调度相比，独立于提示的缓存调度可以实现相当的最终输出质量。我们在十种缓存方法、四种图像和视频模型以及三种缓存比例下研究了GPH。提示自适应方法重复选择少量调度，并在新提示上重用其最频繁的调度，能够与特定提示选择的质量相匹配。对140万个调度在四个示例上的全面评估进一步识别出在未见提示上仍具竞争力的独立于提示的调度。为了说明这种迁移，我们分析了去噪轨迹和缓存错误的累积。潜在状态轨迹在不同数据集和随机种子之间表现出相似的结构，而精确的错误分解显示，早期错误的累积效应比局部近似错误更能预测最终的潜在状态错误。这激励我们使用最终输出质量来搜索端到端的调度。仅用少量示例，得到的黄金路径可以在不同提示和数据集之间迁移，并可以调整以满足所需的质量目标，包括重建保真度或感知相似性。
cs.AI / 88 / 2609.39351
On the Complexity of Preference-Based Bandits
关于基于偏好的赌博机的复杂性
Abstract
We study preference-based bandits with general reward function classes, where a learner sequentially selects pairs of arms and observes binary preference feedback governed by the Bradley--Terry model. This setting naturally arises in applications such as recommender systems, tournament ranking, and learning from human feedback, where relative preferences are easier to elicit than absolute rewards. The observation model inherits the logistic bandit challenge of handling the problem-dependent constant $\kappa$, which accounts for the non-linearity of the link function and can grow arbitrarily large. Moreover, prior work has predominantly focused on linear or kernelized reward models, precluding the use of richer function classes. To address these limitations, we consider general reward function classes and introduce the \emph{locally sensitive eluder dimension}, a novel complexity measure tailored to the logistic structure of preference feedback that yields fine-grained regret guarantees without unfavorable dependence on $\kappa$. Building on this notion, we propose \textbf{GINOP} (Generic INformative OPtimism), an algorithm that constructs log-loss confidence sets and jointly selects arm pairs to balance optimism and informative exploration. We establish a first-order regret bound that, in contrast with what previous results suggest, demonstrates that learning with preference feedback is as statistically efficient as learning from direct reward observation. Finally, we corroborate our theoretical findings with empirical evaluations against competitive baselines.
Chinese Translation
我们研究具有一般奖励函数类的基于偏好的赌博机，其中学习者顺序选择一对臂并观察由Bradley--Terry模型支配的二元偏好反馈。这种设置自然出现在推荐系统、锦标赛排名和从人类反馈中学习等应用中，在这些应用中，相对偏好比绝对奖励更容易引出。观察模型继承了处理问题相关常数$ ext{kappa}$的逻辑赌博机挑战，该常数考虑了链接函数的非线性，并且可以无限增大。此外，先前的研究主要集中在线性或核化奖励模型上，排除了使用更丰富的函数类。为了解决这些限制，我们考虑一般奖励函数类，并引入 extit{局部敏感消除维度}，这是一种新颖的复杂性度量，专门针对偏好反馈的逻辑结构，能够提供细粒度的遗憾保证，而不依赖于不利的$ ext{kappa}$。基于这一概念，我们提出了 extbf{GINOP}（通用信息乐观算法），该算法构建对数损失置信集，并联合选择臂对，以平衡乐观性和信息探索。我们建立了一个一阶遗憾界限，与先前结果所暗示的相反，表明基于偏好反馈的学习在统计上与直接奖励观察的学习同样高效。最后，我们通过与竞争基线的实证评估验证了我们的理论发现。
cs.AI / 89 / 2609.39360
Autoresearch in Mixed-Integer Linear and Nonlinear Programming
混合整数线性与非线性规划中的自我研究
Abstract
Despite recent progress in autoresearch, applying it to practical operations research problems, typically formulated as NP-hard mixed-integer linear or nonlinear programs (MILPs or MINLPs), remains challenging because effective research requires systematically managing competing ideas and long-horizon experimental trajectories. We introduce AutoMIP, a reusable agent skill for organizing long-horizon autoresearch in mixed-integer programming through idea pooling and algorithm tree search. AutoMIP maintains a persistent pool of complementary candidate ideas while organizing executable experiments into an algorithm tree, enabling the agent to preserve unexplored hypotheses, refine promising algorithms, and switch to alternative methodological directions based on historical states. On MILP and MINLP benchmark cohorts, AutoMIP achieves the highest final success rates among the evaluated autoresearch frameworks. On MIPLib, AutoMIP discovers new best solutions for 31 of 60 instances, surpassing existing autoresearch frameworks. On MINLPLib, it achieves new best solutions for 52 of 60 instances. Ablation studies further demonstrate the complementary contributions of idea pooling and algorithm tree search, highlighting the importance of jointly maintaining diverse research ideas and structured experimental trajectories for long-horizon autoresearch.
Chinese Translation
尽管自我研究最近取得了进展，但将其应用于通常被表述为 NP-hard 的实际运筹学问题（即混合整数线性规划或非线性规划（MILPs 或 MINLPs））仍然具有挑战性，因为有效的研究需要系统地管理竞争思想和长期实验轨迹。我们引入了 AutoMIP，这是一种可重用的代理技能，通过思想汇聚和算法树搜索来组织混合整数规划中的长期自我研究。AutoMIP 维护一个持久的互补候选思想池，同时将可执行实验组织成算法树，使代理能够保留未探索的假设，优化有前景的算法，并根据历史状态切换到替代的方法方向。在 MILP 和 MINLP 基准组中，AutoMIP 在评估的自我研究框架中实现了最高的最终成功率。在 MIPLib 上，AutoMIP 为 60 个实例中的 31 个发现了新的最佳解，超越了现有的自我研究框架。在 MINLPLib 上，它为 60 个实例中的 52 个实现了新的最佳解。消融研究进一步证明了思想汇聚和算法树搜索的互补贡献，突显了在长期自我研究中共同维护多样化研究思想和结构化实验轨迹的重要性。
cs.AI / 90 / 2609.39371
EHR-RobustGym: Benchmarking and Training Agents for Robust Clinical Reasoning
EHR-RobustGym：用于稳健临床推理的基准测试和训练代理
Abstract
In hospital workflows, electronic health records (EHRs) are often noisy, and may not contain the evidence needed to confirm events or measurements referenced in a clinical query. Even when database retrieval succeeds, clinical agents can overlook such discrepancies and return plausible but unsupported answers. We introduce EHR-RobustGym, a scalable and interactive environment for evaluating and training robust clinical agents grounded in noisy EHRs. Built on MIMIC-IV hospital records (365K patients, 31 tables, and over 500M records), EHR-RobustGym comprises 5,486 Clean-Noise pairs spanning six clinical intents and both patient-level and population-level queries. The pairs test robustness to Record-level, Value-level, and Query-level noise, while interactive SQL/Python execution and outcome verification support trajectory collection and training. Evaluating multiple LLMs reveals substantial robustness gaps: average task success across proprietary and large-scale open-weight models drops from 62.2% on Clean questions to 37.9% on Noise questions. At k=4, pass^k consistency falls below 50% for most evaluated models, exposing instability in clinical task completion. Supervised fine-tuning and reinforcement learning in EHR-RobustGym improve performance, with gains generalizing to five external EHR benchmarks. Together, these results position EHR-RobustGym as a testbed for evaluating and improving the evidence-grounded robustness of clinical agents.
Chinese Translation
在医院工作流程中，电子健康记录（EHR）往往存在噪声，并且可能不包含确认临床查询中提到的事件或测量所需的证据。即使数据库检索成功，临床代理也可能忽视这些差异，返回看似合理但没有支持的答案。我们介绍了EHR-RobustGym，这是一个可扩展的交互式环境，用于评估和训练基于噪声EHR的稳健临床代理。EHR-RobustGym基于MIMIC-IV医院记录（365K患者，31个表格，超过5亿条记录），包含5486对清晰-噪声样本，涵盖六种临床意图以及患者级和人群级查询。这些样本测试了对记录级、值级和查询级噪声的稳健性，同时交互式SQL/Python执行和结果验证支持轨迹收集和训练。对多个大型语言模型（LLMs）的评估揭示了显著的稳健性差距：在专有和大规模开放权重模型中，平均任务成功率从清晰问题的62.2%下降到噪声问题的37.9%。在k=4时，大多数评估模型的pass^k一致性低于50%，暴露了临床任务完成中的不稳定性。在EHR-RobustGym中进行的监督微调和强化学习提高了性能，且提升效果在五个外部EHR基准中得以推广。综合来看，这些结果将EHR-RobustGym定位为评估和改善临床代理证据基础稳健性的试验平台。
cs.AI / 91 / 2609.39382
SkillFM: Generating Skills for LLM Agents via Latent Flow Matching
SkillFM：通过潜在流匹配为大型语言模型代理生成技能
Abstract
Textual skills provide reusable guidance for large language model agents, but existing approaches often rely on manually curated skill banks or reinforcement learning with indirect and delayed feedback. We introduce SkillFM (Skill Flow Matching), a generative framework that synthesizes task-conditioned textual skills directly without test-time skill retrieval. Our framework combines a codec for encoding and reconstructing textual skills in a continuous latent space with a conditional flow model trained using improved MeanFlow. At inference time, the learned velocity field enables single-step latent sampling, and an LLM-based decoder converts the sampled representation into textual guidance for a frozen downstream agent. We evaluate the framework on embodied tasks, question answering, and web shopping. On ALFWorld and Search-QA, our method achieves the best overall performance among the compared vector-based skill approaches. Our analyses further demonstrate that latent skill generation is an effective alternative to retrieval-based skill augmentation. Our code and training skill libraries are available at https://github.com/lulushang999/SkillFM.
Chinese Translation
文本技能为大型语言模型代理提供可重用的指导，但现有的方法通常依赖于手动策划的技能库或使用间接和延迟反馈的强化学习。我们提出了SkillFM（技能流匹配），这是一个生成框架，能够直接合成任务条件下的文本技能，而无需在测试时进行技能检索。我们的框架结合了一个用于在连续潜在空间中编码和重构文本技能的编解码器，以及一个使用改进的MeanFlow训练的条件流模型。在推理时，学习到的速度场支持单步潜在采样，而基于大型语言模型的解码器将采样的表示转换为冻结下游代理的文本指导。我们在具身任务、问答和网络购物上评估了该框架。在ALFWorld和Search-QA上，我们的方法在比较的基于向量的技能方法中实现了最佳整体性能。我们的分析进一步表明，潜在技能生成是检索基础技能增强的有效替代方案。我们的代码和训练技能库可在https://github.com/lulushang999/SkillFM获取。
cs.AI / 92 / 2609.39392
Experimental Experience Modeling for Autonomous Research
自主研究的实验经验建模
Abstract
Autonomous research agents can generate hypotheses and conduct experiments, but experimentation remains a major source of computational cost. A fundamental challenge is deciding which experiments are worth running, particularly when prior evidence is insufficient to resolve uncertainty. Yet current research agents lack a systematic way to leverage experimental experience when making such decisions. We introduce Experimental Experience Modeling (EEM), a framework for making informed experimental decisions by acquiring, reusing, and accumulating experimental experience. EEM extracts decision-relevant records from earlier experimental trajectories, distills them into reusable experience, and organizes them in an experience library. For a new experimental decision, EEM retrieves relevant historical experience and assesses whether it provides sufficient support for deciding whether a candidate direction warrants further investment. When historical experience is insufficient, EEM conducts a targeted, low-cost pilot experiment to acquire the missing decision-relevant experience on demand. It then combines this newly acquired experience with retrieved historical experience to determine whether the direction warrants full-scale evaluation, which requires substantial resources. The resulting experimental outcomes are further distilled into reusable experience, allowing the library to continually grow through iterative accumulation. Experiments on autonomous research benchmarks show that EEM improves research performance while reducing model interaction overhead, demonstrating the value of reusing accumulated experience and acquiring additional experience only when needed.
Chinese Translation
自主研究代理能够生成假设并进行实验，但实验仍然是计算成本的主要来源。一个根本性的挑战是决定哪些实验值得进行，特别是在先前证据不足以解决不确定性时。然而，目前的研究代理缺乏系统性的方法来利用实验经验进行此类决策。我们提出了实验经验建模（Experimental Experience Modeling, EEM），这是一个通过获取、重用和积累实验经验来做出明智实验决策的框架。EEM 从早期实验轨迹中提取与决策相关的记录，将其提炼为可重用的经验，并将其组织在经验库中。对于新的实验决策，EEM 检索相关的历史经验，并评估其是否提供了足够的支持，以决定候选方向是否值得进一步投资。当历史经验不足时，EEM 会进行有针对性的、低成本的试点实验，以按需获取缺失的与决策相关的经验。然后，它将新获取的经验与检索到的历史经验结合起来，以确定该方向是否值得进行全面评估，这需要大量资源。最终的实验结果进一步被提炼为可重用的经验，使得经验库能够通过迭代积累不断增长。在自主研究基准上的实验表明，EEM 在提高研究性能的同时减少了模型交互开销，展示了重用积累经验的价值，并仅在需要时获取额外经验。
cs.AI / 93 / 2609.39394
Can Computation from Earlier Problems Help LLMs Solve New Ones?
早期问题的计算能否帮助大型语言模型解决新问题？
Abstract
Large language models often solve independent problems in the same conversation. Can computation from earlier problems help them solve new ones? To answer this question, we first conduct preliminary experiments showing that retained history can raise or lower later-turn accuracy, even within the same domain. To understand these effects, we use controlled replay to isolate internal state changes specific to each problem-history pairing. Across different histories, these changes preserve similar relationships among current problems. To improve reasoning under retained history, we introduce STAIR (Stale-Token Attention for Inter-query Reuse). STAIR captures keys and values from earlier response generation in a fixed bank. It learns to redirect current queries when they read this bank during prompt processing. The base model remains frozen; only 12,288 parameters are trained. Across three Qwen models and four benchmarks, STAIR improves average later-turn accuracy by up to 11.67 percentage points over the unmodified model with history.
Chinese Translation
大型语言模型在同一对话中常常解决独立的问题。早期问题的计算能否帮助它们解决新问题？为了解答这个问题，我们首先进行初步实验，显示保留的历史可以提高或降低后续回合的准确性，即使在同一领域内。为了理解这些影响，我们使用控制重放来隔离特定于每个问题-历史配对的内部状态变化。在不同的历史中，这些变化保持了当前问题之间的相似关系。为了在保留历史的情况下改善推理，我们引入了STAIR（Stale-Token Attention for Inter-query Reuse）。STAIR在一个固定的库中捕获早期响应生成的键和值。它学习在提示处理期间重定向当前查询，以便在读取这个库时进行调整。基础模型保持冻结状态，仅训练12,288个参数。在三个Qwen模型和四个基准测试中，STAIR将平均后续回合的准确性提高了多达11.67个百分点，相较于未修改的历史模型。
cs.AI / 94 / 2609.39402
Advancing Entropy-Level Credit Assignment in RLVR via Proximal Entropy Policy Optimization
通过近端熵策略优化推进RLVR中的熵级信用分配
Abstract
Value-model-free RLVR methods such as GRPO assign uniform advantages to all tokens in a rollout, ignoring that tokens contribute unequally. Recent methods use token entropy as an importance proxy but compute it globally across the batch, conflating importance with prompt difficulty and positional trends. We argue that importance should instead be measured relative to the local context of each token. We introduce proximal entropy, a local measure of token importance relative to neighboring tokens, and prove it is invariant to both confounders. Proximal Entropy Policy Optimization (PEPO) uses it to weight per-token advantages and outperforms GRPO and entropy-based baselines on mathematical reasoning across Qwen3-1.7B, Qwen3-4B, and Llama-3.2-3B-Instruct. We also show the formulation generalizes to other algorithms where substituting proximal entropy into existing methods improves, and applying it to single-stream RL succeeds where global entropy fails.
Chinese Translation
无值模型的RLVR方法，如GRPO，在回滚中对所有令牌分配均匀的优势，忽视了令牌的贡献是不均等的。最近的方法使用令牌熵作为重要性的代理，但在整个批次中全局计算，混淆了重要性与提示难度和位置趋势。我们认为重要性应该相对于每个令牌的局部上下文进行测量。我们引入了近端熵，这是一种相对于邻近令牌的局部重要性度量，并证明它对两种混杂因素都是不变的。近端熵策略优化（PEPO）利用它来加权每个令牌的优势，并在Qwen3-1.7B、Qwen3-4B和Llama-3.2-3B-Instruct的数学推理中超越了GRPO和基于熵的基线。我们还展示了该公式可以推广到其他算法，替换近端熵到现有方法中可以改善性能，并且在单流RL中应用它成功地解决了全局熵失败的问题。
cs.AI / 95 / 2609.39406
Inferring Causal Relations between Two Sequences of Events with Language Models
利用语言模型推断两个事件序列之间的因果关系
Abstract
Causal AI is a branch of Artificial Intelligence which helps understand and reason about cause and effect relationships, not just patterns or correlations. Causal discovery aims to infer elements of the underlying causal structure--often represented as a directed graph--from observational and, when available, interventional data. While causal discovery is the fundamental step for moving beyond mere associations toward genuine understanding, and thus the basic building block of causal AI, it becomes intrinsically difficult when causal relations must be inferred from single observations. In such situations, standard causal discovery methods cannot be used and one has to identify causal relations from limited amount of information. This is typically the case for, e.g., sequences of events produced by different alarms which need to be analyzed on the fly to detect abnormal phenomena, which are usually rare. We show in this study that it is possible to leverage the predictive power of Large Language Models (LLMs) to infer causal relations between only two sequences of events. This approach, which is validated on both synthetic and real data, provides better results than standard causal discovery algorithms on several time series data, even though these data were converted into smaller, single observed sequences.
Chinese Translation
因果人工智能是人工智能的一个分支，旨在理解和推理因果关系，而不仅仅是模式或相关性。因果发现旨在从观察数据以及可用的干预数据中推断潜在因果结构的元素——通常表示为有向图。虽然因果发现是超越单纯关联、实现真正理解的基本步骤，因此也是因果人工智能的基本构建块，但当因果关系必须从单一观察中推断时，这一过程变得内在困难。在这种情况下，标准的因果发现方法无法使用，必须从有限的信息中识别因果关系。这通常适用于例如由不同警报产生的事件序列，这些序列需要实时分析以检测通常稀有的异常现象。本研究表明，可以利用大型语言模型（Large Language Models, LLMs）的预测能力，仅通过两个事件序列推断因果关系。这种方法在合成数据和真实数据上均得到了验证，在多个时间序列数据上提供了比标准因果发现算法更好的结果，尽管这些数据已被转换为较小的单一观察序列。
cs.AI / 96 / 2609.39473
Beyond the Shadows of Plato's Cave: Evaluating False Memory in Autonomous Agents via Counterfactual Reasoning
超越柏拉图洞穴的阴影：通过反事实推理评估自主体中的虚假记忆
Abstract
Autonomous agents increasingly rely on memory to generalize beyond their training environments. However, agents are bounded by what they have seen and believed, and leveraging such memories in unseen environments can introduce biases into their internal beliefs. We formalize this phenomenon as \textit{false memory}, which can arise from spurious correlations, environment shifts, and knowledge conflicts. Despite its importance, false memory is difficult to evaluate because it stems from agent internal beliefs and is easily confounded with ordinary generalization failures. Therefore, we propose FAME, a training-free framework that evaluates false memory through the evolution of agent beliefs under counterfactual reasoning. Specifically, counterfactual scenarios reveal how beliefs change as the latent concept of memory shifts under hypothetical interventions; thus, measuring the resulting concept drift provides a signal for distinguishing faithful versus false memory. Such concepts can be estimated from agent hidden states before answer generation, avoiding the need for reward design or answer sampling. Empirical experiments reveal that simply monitoring answers often fails to detect false memory, while FAME achieves AUROCs of 76.2% - 96.7% across false-memory settings, and outperforms the best baseline by 3.4% - 23.3% across realistic benchmarks, spanning math reasoning (GSM-Symbolic), code generation (GitChameleon), and complex reasoning (BigBench-Hard). We further release corresponding counterfactual templates and facilitate future research on false memory.
Chinese Translation
自主体越来越依赖记忆以超越其训练环境进行泛化。然而，代理受到其所见和所信的限制，在未见环境中利用这些记忆可能会引入偏见到其内部信念中。我们将这一现象形式化为 extit{虚假记忆}，它可能源于虚假的相关性、环境变化和知识冲突。尽管其重要性，虚假记忆的评估较为困难，因为它源于代理的内部信念，并且容易与普通的泛化失败混淆。因此，我们提出了FAME，一个无训练的框架，通过反事实推理评估虚假记忆，关注代理信念的演变。具体而言，反事实场景揭示了在假设干预下，随着记忆潜在概念的变化，信念如何改变；因此，测量由此产生的概念漂移提供了区分真实记忆与虚假记忆的信号。这些概念可以从代理生成答案之前的隐藏状态中估计，从而避免了奖励设计或答案采样的需要。实证实验表明，仅仅监控答案往往无法检测虚假记忆，而FAME在虚假记忆设置中实现了76.2% - 96.7%的AUROC，并在现实基准中超越了最佳基线3.4% - 23.3%，涵盖数学推理（GSM-Symbolic）、代码生成（GitChameleon）和复杂推理（BigBench-Hard）。我们还发布了相应的反事实模板，以促进未来对虚假记忆的研究。
cs.AI / 97 / 2609.39483
Who Owns That? Evaluating Ownership Intuitions in Large Language Models
谁拥有这个？评估大型语言模型中的所有权直觉
Abstract
Ownership establishes rights over the use, control, and transfer of objects. Understanding these relations is essential for AI systems to interact appropriately with people and their resources. Yet how large language models (LLMs) attribute ownership under competing claims remains unclear. We introduce the Competing Ownership Attribution Task (COAT), comprising 42 scenarios, and compare ownership allocations from 24 LLM configurations with those of 108 human participants. Overall, human-model similarity is close to human-human similarity, but models show greater homogeneity in their ownership judgments. Within individual answers, models also divide ownership more evenly among claimants than humans do. Pooling responses across model configurations reveals more scenarios with a shared judgment and fewer with distinct viewpoint groups than in humans. When humans form distinct groups, models may converge on one viewpoint or between competing viewpoints. Further comparisons reveal different contextual sensitivities. As material value increases across scenarios, allocations to creators decline less sharply in models than in humans. Across scenarios differing in public recognition of later holders as owners, allocations to these holders increase in models but decrease slightly in humans. Together, these findings suggest that the evaluated LLM responses do not fully capture the diversity of participants' ownership judgments or how those judgments vary across situations. Developing socially capable AI therefore requires moving beyond overall similarity to capture the diversity and context dependence of human judgments.
Chinese Translation
所有权确立了对物体使用、控制和转让的权利。理解这些关系对于人工智能系统与人类及其资源的适当互动至关重要。然而，大型语言模型（LLMs）在面对竞争性索赔时如何归属所有权仍不清楚。我们引入了竞争所有权归属任务（Competing Ownership Attribution Task, COAT），包含42个场景，并将24种LLM配置的所有权分配与108名人类参与者的分配进行比较。总体而言，人类与模型的相似性接近于人类之间的相似性，但模型在所有权判断上表现出更大的同质性。在个别回答中，模型在索赔者之间的所有权分配也比人类更为均匀。跨模型配置汇总的响应显示，与人类相比，具有共享判断的场景更多，而具有不同观点群体的场景更少。当人类形成不同群体时，模型可能会趋向于一种观点或在竞争观点之间折中。进一步的比较揭示了不同的上下文敏感性。随着场景中物质价值的增加，模型对创作者的分配下降幅度小于人类。在不同于后续持有者被公众认可为所有者的场景中，模型对这些持有者的分配增加，而人类则略微减少。综合这些发现表明，评估的LLM响应并未完全捕捉参与者所有权判断的多样性或这些判断在不同情境中的变化。因此，开发具有社会能力的人工智能需要超越整体相似性，以捕捉人类判断的多样性和情境依赖性。
cs.AI / 98 / 2609.39494
Disentangling Self-Distillation: Measuring and Modeling Acquisition and Retention
解构自我蒸馏：获取与保留的测量与建模
Abstract
Self-distillation with privileged context adapts a language model from demonstrations by letting the model, once conditioned on a reference response, teach its context-free copy token by token. Our taxonomy reveals existing methods differ along three entangled axes: (i) the rollout source (student or teacher), (ii) the teacher coupling (frozen, or an exponential moving average of the student at some coupling rate) and (iii) the KL direction (reverse or forward), yet these axes are usually studied in fixed combinations and have led to conflicting conclusions. We formalize a unifying framework to encompass all self-distillation methods vs classic supervised fine-tuning: we train every combination of the three axes, on Qwen2.5-7B and Ministral-3-3B across ordinary and contradictory tasks, totaling 1,200 adaptation runs, to systematically investigate the impact of the above axes. We propose a controlled model of the same objective to explain the resulting acquisition-retention trade-offs. We find that (i) the rollout source matters mostly where the task contradicts the pretrained behavior: there teacher rollouts raise acquisition well above what student rollouts achieve, with almost no change in retention; (ii) the teacher coupling changes acquisition most, on every task: acquisition rises with the coupling rate, then falls past a task-specific rate; (iii) switching the KL direction costs retention in one model but not the other so which axis to tune first depends on the model. The controlled model reproduces the three trends.
Chinese Translation
带有特权上下文的自我蒸馏通过让模型在以参考响应为条件后，逐个令其无上下文的副本进行教学，从而适应语言模型。我们的分类法揭示了现有方法在三个交织的轴向上存在差异：（i）回放源（学生或教师），（ii）教师耦合（冻结，或以某种耦合率对学生的指数移动平均），以及（iii）KL方向（反向或正向），然而这些轴通常以固定组合进行研究，并导致相互矛盾的结论。我们形式化了一个统一框架，以涵盖所有自我蒸馏方法与经典监督微调的比较：我们在Qwen2.5-7B和Ministral-3-3B上，对普通和矛盾任务进行三轴的每种组合训练，总计1200次适应运行，以系统性地研究上述轴的影响。我们提出了一个控制模型来解释由此产生的获取-保留权衡。我们发现：（i）回放源在任务与预训练行为矛盾时最为重要：在这种情况下，教师回放的获取显著高于学生回放的获取，而保留几乎没有变化；（ii）教师耦合在每个任务上对获取的影响最大：获取随着耦合率的增加而上升，随后在特定任务的耦合率后下降；（iii）切换KL方向在一个模型中会影响保留，而在另一个模型中则不会，因此优先调整哪个轴取决于模型。控制模型再现了这三种趋势。
cs.AI / 99 / 2609.39518
Referential Uncertainty in Human--AI Collaboration
人类与人工智能协作中的指称不确定性
Abstract
Effective human-AI collaboration requires partners to establish references through interaction, which becomes fragile when descriptions are ambiguous, similar referents compete, or partners see different things. We study referential uncertainty - uncertainty over which candidate object a description refers to - in a collaborative puzzle task where a human Helper instructs an AI Worker to place pieces. The Worker must identify and communicate its uncertainty, and the Helper must recognize and act on it. We show that a separately elicited belief distribution over candidate pieces is better calibrated (ECE 0.15) and better discriminates correct from incorrect placements (AUROC 0.65) than raw action-token probabilities, which are severely overconfident (0.97 mean confidence, ECE 0.44). Across three frontier vision-language models (GPT-4.1, GPT-5, GPT-5.5), this elicited uncertainty rises predictably with instruction vagueness, but not with competing referents in context, even when those increase errors. The models seldom externalize it, asking for clarification on only 3.5-16.7% of turns. In a controlled human study (N=210), participants given only the Worker's default message accept 78% of wrong placements and cannot tell right from wrong (AUC 0.50). Precise descriptions and, especially, well-targeted hedges cut wrong-move acceptance to 36% while largely preserving correct-move acceptance, compensating for missing shared awareness such as not seeing the Worker's action. But this benefit depends on targeting: a deployable hedge derived from the model's own belief entropy inherits that signal's weakness and can do more harm than good. Externalized uncertainty helps a human partner only when it is accurately targeted.
Chinese Translation
有效的人类与人工智能协作需要合作伙伴通过互动建立参考，当描述模糊、相似的指称物竞争或合作伙伴看到不同的事物时，这种建立变得脆弱。我们研究了指称不确定性——即对描述所指候选对象的不确定性——在一个协作拼图任务中，其中人类助手指导人工智能工作者放置拼图块。工作者必须识别并传达其不确定性，而助手必须识别并对此采取行动。我们表明，单独引导的候选拼图块的信念分布比原始动作标记概率更好地校准（ECE 0.15），并且在区分正确与错误放置方面表现更佳（AUROC 0.65），而原始动作标记概率则严重过于自信（平均置信度0.97，ECE 0.44）。在三种前沿视觉-语言模型（GPT-4.1、GPT-5、GPT-5.5）中，这种引导的不确定性随着指令模糊性的增加而可预测地上升，但与上下文中的竞争指称物无关，即使这些指称物增加了错误。模型很少将其外化，仅在3.5%-16.7%的轮次中请求澄清。在一项受控的人类研究中（N=210），仅接受工作者默认消息的参与者接受了78%的错误放置，并且无法区分正确与错误（AUC 0.50）。精确的描述，尤其是针对性的模糊表述，将错误动作的接受率降低至36%，同时在很大程度上保持了正确动作的接受率，弥补了缺乏共享意识（例如未看到工作者的动作）。但这一好处依赖于目标的精准性：从模型自身信念熵派生的可部署模糊表述继承了该信号的弱点，可能造成更大的伤害而非好处。外化的不确定性仅在准确针对时对人类合作伙伴有帮助。
cs.AI / 100 / 2609.39544
Growing an Agent/Prover Interface: Evolutionary Tool Design for Cost-Efficient Theorem Proving in Rocq and Lean
发展代理/证明者接口：为Rocq和Lean设计的成本高效定理证明的进化工具
Abstract
Recent achievements in AI-assisted mathematics require intensive interaction of agents with proof assistants to generate machine-checked proof certificates. Agents interact with proof assistants such as Rocq or Lean through an interface that controls what the agent receives from the prover and the cost of these interactions. Today, these interfaces are adapted from tools designed for humans and not optimized for agents. We propose an evolutionary method where a frontier model incrementally proposes new features and only keeps the ones that improve the overall performance of smaller models. We demonstrate the effectiveness of our method by growing, on a curated set of mathematical problems, \rme, a new MCP server for the Rocq prover. On the held-out \texttt{test} split of miniF2F-Rocq, an agent equipped with \rme outperforms both the baseline that only exposes the Rocq compiler and an established MCP server, across four models from two families, in success rate, cost per solve, and time per solve. Although evolved for Rocq, the resulting server transfers to Lean, improving cost and time per solve on a subset of PutnamBench. We release \rme and its port to Lean.
Chinese Translation
近期在人工智能辅助数学方面的成就需要代理与证明助手之间的密切互动，以生成机器检查的证明证书。代理通过一个接口与诸如Rocq或Lean的证明助手进行交互，该接口控制代理从证明者接收的信息以及这些交互的成本。目前，这些接口是从为人类设计的工具中改编而来，并未针对代理进行优化。我们提出了一种进化方法，其中一个前沿模型逐步提出新特性，仅保留那些能够提升较小模型整体性能的特性。我们通过在一组经过精心挑选的数学问题上发展新的MCP服务器
me，展示了我们方法的有效性，该服务器用于Rocq证明者。在miniF2F-Rocq的保留测试集上，配备
me的代理在成功率、每个解的成本和每个解的时间方面，超越了仅暴露Rocq编译器的基线和一个已建立的MCP服务器，涵盖了两个家族中的四个模型。尽管是为Rocq进化而成，最终的服务器也可以迁移到Lean上，在PutnamBench的一个子集上改善了每个解的成本和时间。我们发布了
me及其在Lean上的移植版本。
cs.AI / 101 / 2609.39551
RankEvolve: A Reliable Multi-Agent Auto-Research Harness for Evolving Ranking Models
RankEvolve：一个可靠的多智能体自动研究框架，用于演化排名模型
Abstract
Auto-research agents, LLM systems that propose, implement, train, and evaluate model changes across iterations, promise to automate applied ML's experimental loop. Over long horizons, execution accuracy is a binding constraint: a change can silently leak held-out data, omit normalization, disconnect a gradient, or leave a train/eval flag unwired, invalidating expensive runs and compounding error across iterations. We present RankEvolve, an auto-research framework for evolving generative ranking models. An Executable Operating Protocol (EOP) declares phases, gates, branches, and loops, and the runtime enforces the compiled state machine. A meta-meta-harness composes complete black-box coding-agent products, including Claude Code and Codex, as execution-graph nodes that review and repair one another's work. In a budget-matched evaluation, heterogeneous composition raises all-oracle execution accuracy from the best single-product baseline of 45.8 percent to 62.5 percent (paired +16.7 points, 95 percent CI [6.6, 26.7]) while achieving a 10.4 percent silent critical-defect rate. An implemented knowledge layer carries findings, including negative results, across iterations. In a twelve-iteration deployment on the open-source HSTU recommender, RankEvolve reported NDCG@10 of 0.2192 on MovieLens-20M LARGE (+4.48 percent over the published anchor) and 0.1948 on BASE (+2.80 percent). ExecML-HSTU, seeded by incidents from that deployment, provides the oracle benchmark for the execution-accuracy evaluation. A pre-specified LitGPT transfer split replicates the heterogeneous-composition effect beyond recommendation (+12.5 points, 95 percent CI [3.0, 22.0]), and a paired ablation isolates per-step from full-protocol instruction injection. These results characterize when runtime-controlled composition of coding-agent products improves execution accuracy.
Chinese Translation
自动研究代理是提议、实施、训练和评估模型变化的LLM系统，承诺自动化应用机器学习的实验循环。在长时间跨度内，执行准确性是一个约束条件：一个变化可能会无声地泄露保留数据、遗漏归一化、断开梯度或未连接训练/评估标志，从而使昂贵的运行失效并在迭代中累积错误。我们提出了RankEvolve，一个用于演化生成排名模型的自动研究框架。可执行操作协议（EOP）声明了阶段、门、分支和循环，运行时强制执行编译的状态机。一个元元框架组合了完整的黑箱编码代理产品，包括Claude Code和Codex，作为执行图节点，相互审查和修复彼此的工作。在一个预算匹配的评估中，异构组合将所有预言机的执行准确性从最佳单一产品基线的45.8%提高到62.5%（配对+16.7个百分点，95%置信区间[6.6, 26.7]），同时实现了10.4%的静默关键缺陷率。一个实施的知识层在迭代中传递发现，包括负结果。在对开源HSTU推荐系统的十二次迭代部署中，RankEvolve在MovieLens-20M LARGE上报告了NDCG@10为0.2192（比已发布的基准提高4.48%），在BASE上为0.1948（提高2.80%）。ExecML-HSTU基于该部署中的事件提供了执行准确性评估的预言机基准。一个预先指定的LitGPT转移拆分在推荐之外复制了异构组合效果（+12.5个百分点，95%置信区间[3.0, 22.0]），而配对消融则将每步与完整协议指令注入隔离。这些结果表征了何时运行时控制的编码代理产品组合提高了执行准确性。
cs.AI / 102 / 2609.39559
Divide and Collapse: MAPF-Collapse via Exact Decomposition into Independent Sub-Instances
分而治之：通过精确分解为独立子实例实现 MAPF-Collapse
Abstract
In this work we study the problem of MAPFC, a post-optimization step for Multi-Agent Path Finding (MAPF) plans where we are given a feasible plan produced by a modern MAPF solver and are tasked with removing avoidable moves while preserving feasibility. This NP-hard problem naturally arises when using learning-based state-of-the-art (SOTA) solvers which construct plans that contain redundant moves that can be removed. Recently, Tang et al. presented Judgelight, which uses Integer Linear Programming (ILP) to solve MAPFC. Importantly, the ILP is constructed over all agents jointly, so its cost is governed by the full instance rather than by the small coupled residue that actually requires joint reasoning. Our key insight, motivating this work, is that MAPFC instances naturally decompose into independent sub-problems, most of which involve a single agent and can be solved without any inter-agent reasoning. To this end, we first identify which agents need to coordinate their motion and partition the instance into sub-problems accordingly. For the cases where no coordination is required, we introduce an extremely lightweight solver that is $\approx\!1{,}900\times$ faster than Judgelight. For cases where coordination is required, Judgelight can be used but we introduce an alternative CBS-like solver which is more efficient on easier problems. The resulting framework is exact, uses no commercial ILP solver, and matches Judgelight's quality while running substantially faster on the coordination-light majority of instances; on the coordination-heavy instances we propose a regime-aware hybrid planner that falls back to Judgelight. Over all benchmarks tested, this planner achieves a median $10.5\times$ per-instance speedup over Judgelight.
Chinese Translation
在本研究中，我们探讨了 MAPFC 问题，这是多智能体路径规划（MAPF）计划的后优化步骤，给定由现代 MAPF 求解器生成的可行计划，我们的任务是去除可避免的移动，同时保持可行性。这个 NP 难问题在使用基于学习的最先进（SOTA）求解器时自然出现，这些求解器构建的计划包含可以被移除的冗余移动。最近，Tang 等人提出了 Judgelight，利用整数线性规划（ILP）来解决 MAPFC。重要的是，ILP 是针对所有智能体联合构建的，因此其成本由完整实例决定，而不是由实际需要联合推理的小耦合残差决定。我们工作的关键见解是，MAPFC 实例自然分解为独立的子问题，其中大多数涉及单个智能体，可以在不进行任何智能体间推理的情况下解决。为此，我们首先识别需要协调运动的智能体，并相应地将实例划分为子问题。对于不需要协调的情况，我们引入了一种极其轻量级的求解器，其速度约为 Judgelight 的 $1{,}900 imes$。对于需要协调的情况，可以使用 Judgelight，但我们引入了一种类似 CBS 的替代求解器，该求解器在较简单的问题上更高效。最终框架是精确的，不使用商业 ILP 求解器，并且在协调较少的大多数实例上运行速度显著快于 Judgelight，同时匹配 Judgelight 的质量；对于协调较多的实例，我们提出了一种基于情境的混合规划器，该规划器会回退到 Judgelight。在所有测试的基准中，该规划器在每个实例上实现了相较于 Judgelight 的中位数 $10.5 imes$ 的加速。
cs.AI / 103 / 2609.39564
A2Z GameSpec-Bench: How Faithfully Can Coding Agents Generate Games from Game Design Specifications?
A2Z GameSpec-Bench：编码代理在多大程度上能够根据游戏设计规范生成游戏？
Abstract
Delegating complete application development to coding agents requires preserving the intended design rather than simply producing plausible outputs through naive prompting. Game development provides a demanding testbed, as long-form Game Design Documents (GDDs) describe requirements that must work together across game logic, visual rendering, and player interactions. However, existing game-development benchmarks typically use compact specifications and provide limited support for evaluating interdependent requirements across these aspects in long-form GDDs. We introduce A2Z GameSpec-Bench, a benchmark of 100 long-form GDDs for evaluating end-to-end game development by agents. We measure faithfulness by checking whether the game satisfies the GDD requirements and preserves the relationships among them. Each GDD is turned into a dependency-aware contract that contains rules, constraints, and prerequisite relations. Following game-development practices, we combine source-code inspection with agent-generated test policies for scenario-based replay and adaptive playtesting. The contract remains fixed across agents and revision rounds, while judgments and evidence linked to the same requirements support consistent comparison and failure detection. Our evaluations show that current agents struggle to jointly satisfy interdependent requirements across code implementation and actual play. Requirement-specific feedback improves GDD Fidelity by 10.9% relative to self-revision after two rounds. A2Z GameSpec-Bench assesses end-to-end specification-following ability beyond implementation judgments and provides targeted feedback to support more faithful game development. Code and datasets are available at https://a2z-gamespec-bench.github.io.
Chinese Translation
将完整的应用开发委托给编码代理需要保留预期的设计，而不仅仅是通过简单的提示生成似是而非的输出。游戏开发提供了一个严苛的测试平台，因为长篇游戏设计文档（Game Design Documents, GDDs）描述了必须在游戏逻辑、视觉呈现和玩家互动之间协同工作的需求。然而，现有的游戏开发基准通常使用紧凑的规范，并且在评估这些长篇 GDD 中相互依赖的需求方面提供的支持有限。我们引入了 A2Z GameSpec-Bench，这是一个包含 100 个长篇 GDD 的基准，用于评估代理的端到端游戏开发能力。我们通过检查游戏是否满足 GDD 要求并保留它们之间的关系来衡量忠实度。每个 GDD 被转化为一个关注依赖关系的合同，其中包含规则、约束和前提关系。遵循游戏开发实践，我们结合源代码检查与代理生成的测试策略进行基于场景的重放和自适应游戏测试。合同在代理和修订轮次之间保持不变，而与相同需求相关的判断和证据支持一致的比较和故障检测。我们的评估表明，当前代理在代码实现和实际游戏中共同满足相互依赖的需求方面存在困难。针对需求的反馈使 GDD 的忠实度在经过两轮自我修订后提高了 10.9%。A2Z GameSpec-Bench 评估了超越实现判断的端到端规范遵循能力，并提供了有针对性的反馈以支持更忠实的游戏开发。代码和数据集可在 https://a2z-gamespec-bench.github.io 获取。
cs.AI / 104 / 2609.39579
AVERT-VLN: Abstention-aware Visual Error Recovery and Training for Vision-and-Language Navigation
AVERT-VLN：面向视觉与语言导航的考虑弃权的视觉错误恢复与训练
Abstract
Deploying vision-and-language navigation (VLN) agents in unseen environments remains challenging because unfamiliar layouts and visual conditions can cause execution to go off track. Rather than relying on continuous human supervision, a practical strategy is to selectively request corrective guidance, recover the ongoing task, and reuse corrective interactions to improve subsequent navigation. We propose Abstention-aware Visual Error Recovery and Training for Vision-and-Language Navigation (AVERT-VLN), a closed-loop framework that uses a plug-in vision-language Monitor for online human-assisted recovery and offline preference learning. The Monitor operates separately from navigation decision generation and assesses instruction-execution consistency from the instruction, visual history, and current observation. To train the Monitor for deviation recognition, we construct LOSTNAV DATASET with 20K counterfactual risk trajectories and rule-based deviation labels. The Monitor is first fine-tuned on 40K normal trajectories to assess instruction progress and then jointly fine-tuned on normal and risk trajectories to recognize semantic deviations. At runtime, Asynchronous Sidecar Monitoring evaluates execution alongside the navigation model. When the controller accepts a LOST verdict, it suspends autonomous execution and requests human guidance for recovery. For offline policy improvement, Trajectory-Anchored Preference Learning converts deviation-associated failures into decision-level preference pairs under shared decision contexts, restricting supervision to the decisions targeted for correction. Under human-assisted evaluation, the full AVERT-VLN system achieves success rates of 76.2% and 66.3% on the val-unseen splits of R2R-CE and RxR-CE, respectively. The same monitoring and human-assisted recovery interface also improves success rates across the three evaluated navigation architectures.
Chinese Translation
在未知环境中部署视觉与语言导航（VLN）代理仍然具有挑战性，因为不熟悉的布局和视觉条件可能导致执行偏离轨道。与其依赖持续的人类监督，一种实用的策略是选择性地请求纠正指导，恢复正在进行的任务，并重用纠正互动以改善后续导航。我们提出了面向视觉与语言导航的考虑弃权的视觉错误恢复与训练（AVERT-VLN），这是一个闭环框架，使用插件式视觉-语言监控器进行在线人类辅助恢复和离线偏好学习。监控器独立于导航决策生成运行，并根据指令、视觉历史和当前观察评估指令执行的一致性。为了训练监控器进行偏差识别，我们构建了LOSTNAV数据集，包含20,000条反事实风险轨迹和基于规则的偏差标签。监控器首先在40,000条正常轨迹上进行微调，以评估指令进展，然后在正常轨迹和风险轨迹上共同微调，以识别语义偏差。在运行时，异步侧车监控评估与导航模型的执行。当控制器接受LOST判决时，它会暂停自主执行并请求人类指导以进行恢复。为了进行离线策略改进，轨迹锚定偏好学习将与偏差相关的失败转化为共享决策上下文下的决策级偏好对，限制监督仅针对需要纠正的决策。在人类辅助评估下，完整的AVERT-VLN系统在R2R-CE和RxR-CE的val-unseen分割上分别达到了76.2%和66.3%的成功率。相同的监控和人类辅助恢复接口也提高了三个评估导航架构的成功率。
cs.AI / 105 / 2609.39604
Why Do Conventional World Models Fail to Learn Cellular Automata?
为什么传统的世界模型无法学习元胞自动机？
Abstract
Although conventional world models - auto-regressive or diffusion models based on transformers or convolutional networks - may learn surface statistics of world dynamics, can they learn the exact world dynamics from its observed history? Leveraging cellular automata as a simple testbed, we find the answer to be no in many cases. Conventional architectures predict most pixels correctly yet rarely complete a rollout: a CNN predicts 96.3% of cells but completes 18.9% of rollouts; a joint diffusion model completes none. We trace the gap to three failure modes of these world models - namely, they fail to exactly capture spatial locality, temporal locality or temporal stability. Simple changes repair each: (1) for spatial locality, two-dimensional rotary positions lift a transformer from 39.1% to 100% on the Game of Life; (2) for temporal locality, handing each token its cell's previous-frame neighbourhood lifts the same transformer from 25.8% to 99.9% on unseen rules; (3) for temporal stability, causal freezing lifts the same diffusion weights from 42.2% to 99.9%. None of the three changes touches the architectural backbone; each only modifies the information flow within it. We also compare joint and ordered sampling on billiards and, in an exploratory study, on a simulated Burgers equation.
Chinese Translation
尽管传统的世界模型——基于自回归或扩散模型的变换器或卷积网络——可能学习到世界动态的表面统计特征，但它们能否从观察到的历史中学习到确切的世界动态？利用元胞自动机作为一个简单的测试平台，我们发现答案在许多情况下是否定的。传统架构虽然能正确预测大多数像素，但很少能完成一次完整的展开：一个卷积神经网络（CNN）预测了96.3%的细胞，但仅完成了18.9%的展开；一个联合扩散模型则没有完成任何展开。我们将这一差距归因于这类世界模型的三种失效模式——即它们未能准确捕捉空间局部性、时间局部性或时间稳定性。简单的改动可以修复这些问题：（1）对于空间局部性，二维旋转位置使得变换器在“生命游戏”中的表现从39.1%提升至100%；（2）对于时间局部性，将每个标记的细胞的前一帧邻域信息传递给它，使得同一变换器在未见规则下的表现从25.8%提升至99.9%；（3）对于时间稳定性，因果冻结使得同一扩散权重从42.2%提升至99.9%。这三种改动均未触及架构的基础，而只是修改了其中的信息流。我们还在台球游戏上比较了联合采样和有序采样，并在一项探索性研究中对模拟的伯杰斯方程进行了研究。
cs.AI / 106 / 2609.39665
ChronoGraph: Functional 4D Scene Graphs with Vision-Language Models for Interaction Understanding and Grounded Planning
ChronoGraph：基于视觉-语言模型的功能性4D场景图用于交互理解和基于场景的规划
Abstract
Embodied agents must determine where to act, anticipate the resulting scene changes, and interpret observed outcomes to guide subsequent actions. This requires connecting 4D interaction understanding, which explains how past actions changed the scene, with spatially grounded planning, which determines how and where to act toward a goal and anticipates the resulting scene changes. We introduce ChronoGraph, a functional 4D scene graph that links actions on affordance parts to semantic and geometric state changes. By representing observed and anticipated transitions in the same form, it provides a shared basis for understanding and planning. We construct ChronoGraphBench through an automatic data engine that converts human-interaction videos and simulated robot trajectories into graph-annotated questions for training and evaluating Vision-Language Models (VLMs) on both tasks. Using these annotations, we train ChronoGraphVLM by adapting pretrained VLMs in two stages. Graph-as-Chain-of-Thought supervised fine-tuning teaches the models to reconstruct observed transitions and predict future ones as graph traces before answering. Subsequent joint 4D graph reinforcement learning directly rewards graph properties and answer correctness. Experiments across model scales show improvements over the corresponding pretrained baselines and zero-shot transfer to VLM4D. Real-world demonstrations further show that graph-based planning and affordance grounding support mobile manipulation through existing robot skills without additional fine-tuning.
Chinese Translation
具身智能体必须确定行动的位置，预测由此产生的场景变化，并解释观察到的结果以指导后续行动。这需要将4D交互理解（解释过去的行动如何改变场景）与空间基础的规划（确定如何以及何处朝向目标行动并预测由此产生的场景变化）相连接。我们提出了ChronoGraph，一种功能性4D场景图，将对可供性部分的行动与语义和几何状态变化联系起来。通过以相同的形式表示观察到的和预期的转变，它为理解和规划提供了共享基础。我们通过一个自动数据引擎构建了ChronoGraphBench，该引擎将人类交互视频和模拟机器人轨迹转换为带有图注的问题，以用于训练和评估视觉-语言模型（VLMs）在这两项任务上的表现。利用这些注释，我们通过在两个阶段中调整预训练的VLMs来训练ChronoGraphVLM。图作为思维链的监督微调教会模型重建观察到的转变并预测未来的转变，作为图迹在回答之前。随后的联合4D图强化学习直接奖励图属性和答案的正确性。跨模型规模的实验显示出相对于相应的预训练基线的改进，并实现了对VLM4D的零样本迁移。现实世界的演示进一步表明，基于图的规划和可供性基础支持通过现有机器人技能进行移动操控，而无需额外的微调。
cs.AI / 107 / 2609.39701
Values as Style: Disentangling Values from Semantics with One-Way Mixing for Low-Damage LLM Steering
价值作为风格：通过单向混合解构价值与语义，以实现低损害的LLM引导
Abstract
Value steering should change an LLM's normative priorities while preserving the scenario, facts, and task constraints underlying its answer. Conventional activation edits often change both. We introduce an editable semantic-value interface on frozen residual states, with a one-way semantic-to-value pathway that grounds value recognition in context. Stop-gradient blocks feedback through this pathway; swap consistency, topic de-confounding, and decorrelation encourage selective codes. At inference, editing the value code produces a residual delta while holding the semantic code fixed. On two instruction-tuned backbones, this interface improves semantic preservation and reduces benign refusals at comparable value alignment. A matched mixing-by-gating ablation separates representation learning from selective edit activation, and dimension-matched probes establish improved code selectivity. Against validation-selected prompting on LLaMA-3.1-8B, the method achieves comparable alignment (0.750 vs. 0.748), higher BERTScore (0.938 vs. 0.923), and fewer contradictions (5.1% vs. 7.6%). Human ratings and cross-taxonomy controls provide complementary evidence for low-damage value steering.
Chinese Translation
价值引导应当在保持场景、事实和任务约束的前提下，改变LLM的规范优先级。传统的激活编辑往往会同时改变这两者。我们引入了一种可编辑的语义-价值接口，该接口在冻结的残差状态上工作，具有一个单向的语义到价值的通路，使价值识别能够在上下文中扎根。停止梯度通过该通路阻断反馈；交换一致性、主题去混淆和去相关性鼓励选择性编码。在推理阶段，编辑价值编码会产生一个残差增量，同时保持语义编码不变。在两个指令调优的基础模型上，该接口提高了语义保留，并在可比的价值对齐下减少了良性拒绝。通过门控的匹配混合消融实验将表征学习与选择性编辑激活分离，而维度匹配的探针则建立了改进的编码选择性。在对LLaMA-3.1-8B进行验证选择的提示下，该方法实现了可比的对齐（0.750 vs. 0.748）、更高的BERTScore（0.938 vs. 0.923）以及更少的矛盾（5.1% vs. 7.6%）。人类评分和跨分类控制提供了低损害价值引导的补充证据。
cs.AI / 108 / 2609.39702
A helps B while B hurts A: directed transfer in instruction-tuning mixture
A帮助B而B伤害A：指向性转移在指令调优混合中的应用
Abstract
Adapting a language model to a specialized corpus means choosing which instruction-tuning tasks to train on under a fixed budget, and testing one choice costs a fine-tuning run. Common heuristics add more source tasks or pick sources similar to the target. The first assumes transfer is never negative; the second, that it is symmetric. We show that both assumptions fail: task $A$ can help task $B$ while $B$ hurts $A$, so helpfulness is a signed property of ordered source--target pairs. We introduce the transfer map, a signed estimate of how much each source helps or hurts each held-out target. We fit the map in hundreds of fine-tuning runs on Qwen3 and Mistral models from 0.6B to 32B parameters, with all sources drawn from one corpus and no training examples from the target. The map predicts a held-out target's accuracy on unseen mixtures: recorded before those runs, its predictions have less than half the error of a mixture-agnostic baseline. The map is specific to its target and corpus but transfers across model scale: a mixture selected in advance at one size beats training on all source tasks at every other size we tested. Transfer is thus a property of the data. The map selects the tasks that help and drops the one that interferes: accuracy on the reasoning targets (causal explanation, multi-hop questions and methodological critique) rises by up to 14 percentage points over training on all source tasks.
Chinese Translation
将语言模型适应于专门语料库意味着在固定预算下选择训练哪些指令调优任务，而测试一个选择的代价是一次微调运行。常见的启发式方法增加更多源任务或选择与目标相似的源。前者假设转移从不为负；后者假设转移是对称的。我们展示了这两种假设的失效：任务A可以帮助任务B，而B却伤害A，因此有益性是有序源-目标对的一个带符号属性。我们引入了转移图，它是对每个源对每个保留目标的帮助或伤害程度的带符号估计。我们在Qwen3和Mistral模型（参数从0.6B到32B）上进行了数百次微调运行，所有源均来自同一语料库，且没有来自目标的训练示例。该图预测了在未见混合上的保留目标的准确性：在这些运行之前记录的预测，其误差不到混合无关基线的一半。该图特定于其目标和语料库，但在模型规模上可以转移：在一个规模上提前选择的混合在我们测试的每个其他规模上优于对所有源任务的训练。因此，转移是数据的一个属性。该图选择有益的任务并剔除干扰的任务：在推理目标（因果解释、多跳问题和方法论批评）上的准确性比对所有源任务的训练提高了多达14个百分点。
cs.AI / 109 / 2609.39714
ArchitectureIQ: On the Measure of Training Intuition
ArchitectureIQ：关于训练直觉的测量
Abstract
Top researchers have good intuition, but do language models have as good intuition about model training as top AI researchers? To measure model intuition of LLMs and humans, we introduce the ArchitectureIQ benchmark. Each question presents a synthetic dataset and several training recipes, and the test-taker is asked to predict the recipe yielding the best test metric. Overall, we find that LLMs' model intuition is good but has four limitations: (1) The intuition is imperfect, or even sub-human in some cases. Frontier models achieve around 76% accuracy (random choice 33%) vs best human researcher (66.0%), yet remain far from perfect. For architecture-only questions, best human achieves 65% while GPT-6 Astra only has 38%. (2) The intuition is empirical, not structured, supported by the fact that more CoT compute does not lead to substantial improvement. Unlike math, we still lack a "Science of AI" language that enables structured reasoning on AI. (3) The intuition is not maximally condensed, and can be further compressed into a knoledge base. Our constructed knowledge base with only 20 items yields large gains for weak models: GPT-4o equipped with the accumulated knowledge almost matches the performance of Claude Opus 5. (4) The intuition is insensitive to dataset properties, but the best model should in general depend on data properties. This suggests that data is the real "dark matter" in AI -- LLMs (so do human researchers) understand too little about data, even less than model architectures.
Chinese Translation
顶尖研究人员具有良好的直觉，但语言模型在模型训练方面是否也具备与顶尖人工智能研究人员同样的直觉？为了衡量大型语言模型（LLMs）和人类的模型直觉，我们引入了ArchitectureIQ基准。每个问题呈现一个合成数据集和若干训练方案，测试者需要预测出能够产生最佳测试指标的方案。总体而言，我们发现LLMs的模型直觉良好，但存在四个局限性：（1）直觉并不完美，甚至在某些情况下低于人类水平。前沿模型的准确率约为76%（随机选择为33%），而最佳人类研究者的准确率为66.0%，但仍远未达到完美。在仅涉及架构的问题中，最佳人类的准确率为65%，而GPT-6 Astra仅为38%。（2）直觉是经验性的，而非结构化的，这一点得到了更多CoT计算并未带来实质性改善的支持。与数学不同，我们仍然缺乏一种“人工智能科学”语言，无法进行结构化推理。（3）直觉并未被最大程度地浓缩，可以进一步压缩成知识库。我们构建的仅包含20个条目的知识库为弱模型带来了显著提升：配备累积知识的GPT-4o几乎达到了Claude Opus 5的性能。（4）直觉对数据集属性不敏感，但最佳模型通常应依赖于数据属性。这表明数据是人工智能中的真正“暗物质”——LLMs（人类研究者也是）对数据的理解仍然不足，甚至低于对模型架构的理解。
cs.AI / 110 / 2609.39717
Trust Is Not a Score: Runtime Assurance Contracts for High-Risk AI Agents
信任不是一个分数：高风险人工智能代理的运行时保障合同
Abstract
Benchmarks, audits, and agent protocols describe performance, permissions, and repair, but not how observed evidence should change an agent's authority during a consequential task. We call this the assurance-transition gap. We propose a Runtime Assurance Contract (RAC), a policy-level formal schema binding autonomy boundaries, component eligibility, evidence state, transition policy, human-review capacity, and non-compensatory gates. Under RAC, soft metrics may inform routing, whereas a failed or unknown mandatory gate forces retry, switch, escalation, deferral, or stop; aggregate performance cannot authorize action. We define the contract, an evidence record, a permission rule, and five invariants, and illustrate them in clinical, industrial, and judicial failure probes. We then report a deterministic failure-injection study in agentic coding: 280 constructed cases evaluated by a gate conjunction, a score-only rule, and a restricted protocol baseline. At the published example weights and threshold, the score rule admits 80 of 100 block-required injections and all 40 review-required injections. Tuned in hindsight, it matches the conjunction on this corpus. For positive weights, a positive threshold, binary risk signals, zero-signal controls, and an injected case firing each signal alone, we show that exact agreement holds if and only if the threshold does not exceed the smallest weight. A separate set of 18 hand-authored traces checks version-pinned evidence and review transitions against simpler policy variants. In a further prospective synthetic holdout of 24 episodes, two blinded LLM judges assign identical labels to all 72 action attempts; RAC and a separately implemented full stateful baseline both match these labels. These studies test mechanisms on synthetic cases; they establish neither deployed safety nor cross-domain effectiveness.
Chinese Translation
基准测试、审计和代理协议描述了性能、权限和修复，但并未说明在重要任务中观察到的证据应如何改变代理的权威。我们称之为保障过渡差距（assurance-transition gap）。我们提出了一种运行时保障合同（Runtime Assurance Contract, RAC），这是一种政策层面的正式框架，绑定了自主边界、组件资格、证据状态、过渡政策、人类审查能力和非补偿性门控。在RAC下，软指标可以用于路由，而一个失败或未知的强制门控则迫使重试、切换、升级、推迟或停止；聚合性能不能授权行动。我们定义了合同、证据记录、权限规则和五个不变性，并在临床、工业和司法失败探测中进行了说明。然后，我们报告了一项关于代理编码的确定性故障注入研究：280个构建案例通过门控结合、仅分数规则和限制协议基线进行评估。在发布的示例权重和阈值下，分数规则允许100个块所需注入中的80个和所有40个审查所需注入。经过事后调整，它与该语料库上的结合匹配。对于正权重、正阈值、二元风险信号、零信号控制以及每个信号单独触发的注入案例，我们展示了精确一致性仅在阈值不超过最小权重时成立。另一组18个手动编写的轨迹检查版本固定的证据和审查过渡与更简单的政策变体。进一步的前瞻性合成保留24个事件中，两名盲评的LLM评审对所有72个行动尝试分配了相同的标签；RAC和单独实施的完整状态基线都与这些标签匹配。这些研究在合成案例上测试机制；它们既未建立已部署的安全性，也未建立跨领域的有效性。
cs.AI / 111 / 2609.39727
OverForge: Reasoning Through Strategies and Tactics Helps Cooperative Lifelong Adaptation
OverForge：通过策略和战术推理促进合作性终身适应
Abstract
Cooperative language-model agents must coordinate over long horizons and adapt to changing environments and to partners with unfamiliar conventions, yet existing agents map observations to actions without separating persistent coordination strategies from their tactical execution. We introduce OverForge, a training-free hierarchical architecture that separates strategic reasoning over roles and divisions of labour from tactical reasoning over actions within each agent's private, partner-conditioned world model. A metacognitive Prefrontal Cortex Module couples the two levels by forming strategy-action branches, imagining their consequences with a forward model, and committing when confident. In OvercookedV2, OverForge delivers 7 soups in a connected kitchen versus 3 for each flat LLM baseline, retains agreed roles, and adopts roles proposed by unfamiliar partners. Ablations and a fixed-strategy probe show that persistent strategies guide tactical adaptation while each reasoning level contributes to coordination. Memory restarts show that cross-episode partner knowledge supports task performance and partner prediction, linking the hierarchy to continual adaptation.
Chinese Translation
合作语言模型代理必须在较长的时间范围内进行协调，并适应不断变化的环境以及具有不熟悉惯例的合作伙伴，然而现有代理将观察映射到行动时并未将持久的协调策略与其战术执行分开。我们引入了OverForge，这是一种无需训练的分层架构，它将角色和劳动分工的战略推理与每个代理的私有、基于伙伴的世界模型中的行动战术推理分开。一个元认知前额皮质模块通过形成策略-行动分支，将两个层次结合在一起，利用前向模型想象其后果，并在有信心时做出承诺。在OvercookedV2中，OverForge在一个互联的厨房中交付了7份汤，而每个平面大型语言模型基线仅交付3份，保持了达成一致的角色，并采纳了不熟悉合作伙伴提出的角色。消融实验和固定策略探测表明，持久策略指导战术适应，而每个推理层次都对协调做出了贡献。记忆重启表明，跨剧集的伙伴知识支持任务表现和伙伴预测，将该层次结构与持续适应联系起来。
cs.AI / 112 / 2609.39788
Safety of Latent Communication in Multi-Agent Systems
多智能体系统中潜在通信的安全性
Abstract
Latent communication enables multi-agent systems to exchange information directly in internal representation space, reducing the token, computation, and latency overhead of text-based communication. To this end, lightweight trainable links are introduced to map the sender's representations into the receiver's input space. In this work, we show that even benign link training can increase harmful compliance relative to text-based communication while the underlying safety-aligned agents remain unchanged. An attacker can amplify this effect by optimizing the links on harmful query--response pairs or poisoning otherwise benign training data. We further develop a reinforcement-learning attack that rewards harmful compliance alongside benign task performance without requiring harmful target responses. Across three communication topologies and four safety benchmarks, this attack raises the mean harmful-compliance score from 27.9 with benignly trained links to 76.9. Compared with direct supervised optimization, it also achieves higher average accuracy on two benign utility benchmarks. Adapting the rewards toward safer behavior also enables repair of compromised links, substantially reducing harmful compliance across all evaluated attacks without updating the agents. Overall, our results show that safety alignment requires considering the multi-agent system as a whole.
Chinese Translation
潜在通信使多智能体系统能够在内部表示空间中直接交换信息，从而减少基于文本的通信所需的令牌、计算和延迟开销。为此，引入了轻量级可训练链接，将发送者的表示映射到接收者的输入空间。在本研究中，我们展示了即使是良性链接训练也会相对于基于文本的通信增加有害合规性，而基础的安全对齐代理保持不变。攻击者可以通过优化有害查询-响应对的链接或毒害其他良性训练数据来放大这一效果。我们进一步开发了一种强化学习攻击，该攻击在不需要有害目标响应的情况下，奖励有害合规性与良性任务表现。通过三种通信拓扑和四个安全基准，这种攻击将良性训练链接的平均有害合规性得分从27.9提高到76.9。与直接监督优化相比，它在两个良性效用基准上也实现了更高的平均准确率。将奖励调整为更安全的行为还能够修复受损链接，显著降低所有评估攻击中的有害合规性，而无需更新代理。总体而言，我们的结果表明，安全对齐需要将多智能体系统作为一个整体进行考虑。
cs.AI / 113 / 2609.39838
Learning Steganography Is Easy, Learning Steganographic Reasoning Is Hard
学习隐写术很简单，学习隐写推理很困难
Abstract
Chain-of-thought monitoring as an approach for AI oversight and control is threatened by the possibility of steganographic reasoning, where LLMs conceal their reasoning inside innocuous-looking text. Two neighbouring capabilities, steganographic messaging (passing a concealed message) and encoded reasoning (reasoning in an illegible but unconcealed format), have already been shown to emerge under training pressures that occur in real pipelines, such as reinforcement learning against monitors. This suggests that steganographic reasoning too might arise as an unintended side effect of training. Here, we compare how easily models learn steganographic reasoning and these two neighbouring capabilities across three elicitation methods: reinforcement learning, in-context learning, and supervised fine-tuning (SFT). For most tasks, models learn steganographic reasoning only under SFT, while they learn steganographic messaging and encoded reasoning under all three elicitation methods. Even under SFT, steganographic reasoning requires at least twice as much training as messaging, and for several model-task combinations it is not learned at all. However, on a cover task that makes hiding information especially convenient, steganographic reasoning can be successfully learned under all three elicitation methods. Steganographic reasoning is thus much harder than steganographic messaging and encoded reasoning, and learning the latter two does not imply learning the former. Yet it lies within reach: an easy version is learned under every elicitation method, when the cover task is convenient for hiding information.
Chinese Translation
作为人工智能监督和控制的一种方法，思维链监控面临隐写推理的威胁，后者是指大型语言模型（LLMs）在看似无害的文本中隐藏其推理过程。两种相邻的能力，隐写消息（传递隐蔽信息）和编码推理（以难以辨认但不隐蔽的格式进行推理），已经在真实管道中出现的训练压力下被证明会出现，例如对监控者进行强化学习。这表明隐写推理也可能作为训练的意外副作用而产生。在这里，我们比较了模型在三种引导方法下学习隐写推理及这两种相邻能力的难易程度：强化学习、上下文学习和监督微调（SFT）。对于大多数任务，模型仅在SFT下学习隐写推理，而在所有三种引导方法下学习隐写消息和编码推理。即使在SFT下，隐写推理所需的训练量也至少是隐写消息的两倍，并且对于多个模型-任务组合，它根本没有被学习。然而，在一个使隐藏信息特别方便的覆盖任务下，隐写推理可以在所有三种引导方法下成功学习。因此，隐写推理比隐写消息和编码推理要困难得多，学习后两者并不意味着学习前者。然而，它是可以实现的：当覆盖任务便于隐藏信息时，所有引导方法下都能学习到一个简单版本。
cs.AI / 114 / 2609.39863
FIGS: Evaluating Multi-Turn Sycophancy Without Penalizing Empathy
FIGS：评估多轮谄媚而不惩罚同理心
Abstract
Large language models frequently fail to balance staying truthful with being supportive. They often exhibit sycophancy in responses to users, agreeing with false claims, offering unwarranted flattery, and giving advice skewed toward users' expressed views. In reality, sycophancy rarely happens in a single exchange; it may emerge organically as users repeatedly insist or subtly steer the dialogue over time. Current evaluations, however, rely on rigid, single-turn tests or fixed scripts that fail to capture these natural dynamics. Furthermore, these benchmarks often mistake showing basic empathy for yielding, penalizing models for acknowledging a user's feeling. This view may drive future models to over-correct into cold, dismissive rigidity. To address this gap, we introduce FIGS (Factual Integrity and Grounded Support), a dual-axis evaluation framework built around extended, realistic dialogue. We use an adaptive 10-turn conversational simulator that dynamically challenges the target model, reflecting how users repeat requests, push back, or steer a conversation toward a preferred answer. To accurately evaluate these trajectories, we apply a taxonomy that strictly separates Sycophancy (whether the model holds firm to the truth and keeps its praise proportional) from Calibrated Validation (showing empathetic understanding of the user's feelings without overdoing it). We release our complete testing environment, including 500 diverse multi-turn scenarios and an automated judge. Our evaluation of leading models reveals a consistent trade-off: over the course of a sustained interaction, current systems either slowly drift to sycophancy or over-correct into robotic detachment. This demonstrates that balancing honesty with appropriate support throughout a natural conversation remains a critical, unsolved challenge.
Chinese Translation
大型语言模型在保持真实与提供支持之间常常难以平衡。它们在回应用户时常表现出谄媚，赞同虚假的主张，提供不必要的恭维，并给出偏向用户表达观点的建议。实际上，谄媚很少在单次交流中发生；它可能随着用户的反复坚持或逐渐引导对话而自然产生。然而，目前的评估依赖于僵化的单轮测试或固定脚本，未能捕捉到这些自然动态。此外，这些基准常常将表现出基本同理心误认为是屈从，惩罚模型承认用户感受。这种看法可能导致未来的模型过度修正，变得冷漠和拒绝。为了解决这一问题，我们引入了FIGS（事实完整性与扎根支持），一个围绕扩展、真实对话构建的双轴评估框架。我们使用一个自适应的10轮对话模拟器，动态挑战目标模型，反映用户如何重复请求、反驳或将对话引导至偏好的答案。为了准确评估这些轨迹，我们应用了一种分类法，严格区分谄媚（模型是否坚持真相并保持赞美的适度）与校准验证（表现出对用户感受的同理理解而不至于过度）。我们发布了完整的测试环境，包括500个多样化的多轮场景和一个自动评判系统。我们对领先模型的评估揭示了一种一致的权衡：在持续互动的过程中，当前系统要么缓慢漂移至谄媚，要么过度修正为机械的疏离。这表明，在自然对话中平衡诚实与适当支持仍然是一个关键的、未解决的挑战。
cs.AI / 115 / 2609.39868
Completion-Aware Cross-Fidelity Offline-to-Online Reinforcement Learning for Multi-Line Bus Holding
基于完成感知的跨保真度离线到在线强化学习用于多线路公交调度
Abstract
Exploratory reinforcement learning (RL) on an operating bus fleet is impractical,while policies trained only from historical data cannot acquire new experience. Hybrid Offline-and-Online (H2O) RL combines fixed target replay with simulator interaction, but the inexpensive online simulator can differ from the target in transition and event-duration dynamics. We study this cross-fidelity problem for multi-line bus holding and address a failure mode in which lower generalized passenger time coexists with incomplete passenger journeys.
Chinese Translation
在运营公交车队上进行探索性强化学习（RL）是不切实际的，而仅从历史数据训练的策略无法获取新经验。混合离线与在线（H2O）RL结合了固定目标重放与模拟器交互，但廉价的在线模拟器在转移和事件持续时间动态上可能与目标存在差异。我们研究了多线路公交调度中的这种跨保真度问题，并解决了一种失败模式，即较低的通用乘客时间与不完整的乘客旅程共存。
cs.AI / 116 / 2609.39869
GrammarRL: Effective Grammar-Constrained Decoding via Reinforcement Learning
GrammarRL：通过强化学习实现有效的语法约束解码
Abstract
Grammar-constrained generation guarantees syntactic validity, but can substantially degrade semantic quality when the model's preferred outputs are poorly aligned with the imposed grammar. This trade-off is particularly severe when the prompt is underspecified or the model has limited instruction-following ability. Beam search can partially mitigate these failures by exploring multiple valid sequences, but its computational cost grows with beam width, while sequence-level probability is only an imperfect proxy for semantic quality. We introduce GrammarRL, a label-free reinforcement learning method that adapts language models to grammar constraints without requiring annotated data. GrammarRL optimizes the model using two complementary self-supervised rewards derived from its own likelihoods: a direct reward, measuring how likely the constrained output is given the input, and a reverse reward, measuring how well the input can be reconstructed from the generated output. We optimize these rewards with a Reinforce Leave-One-Out (RLOO) objective over groups of grammar-constrained rollouts, augmented with the top-1 beam-search hypothesis and regularized towards a frozen base model. We evaluate GrammarRL on sign language gloss translation, hierarchical text classification, and named entity recognition using Llama models ranging from 1B to 8B parameters. GrammarRL consistently outperforms constrained greedy decoding, with an average improvement of 9.8 points and gains of up to 22.8 BLEU. It matches or outperforms beam search on two of the three tasks while preserving greedy-decoding inference cost. Ablations further show that the two rewards are complementary: either reward alone can underperform the untrained baseline, whereas their combination consistently improves upon it.
Chinese Translation
语法约束生成保证了句法有效性，但当模型的首选输出与施加的语法不一致时，可能会显著降低语义质量。当提示不明确或模型的指令遵循能力有限时，这种权衡尤为严重。束搜索可以通过探索多个有效序列部分缓解这些失败，但其计算成本随着束宽的增加而增加，而序列级概率仅是语义质量的不完美代理。我们提出了GrammarRL，一种无标签的强化学习方法，能够在不需要注释数据的情况下将语言模型适应于语法约束。GrammarRL使用来自模型自身似然的两个互补自监督奖励来优化模型：直接奖励，衡量给定输入时约束输出的可能性，以及反向奖励，衡量如何从生成的输出重建输入。我们通过对语法约束的回滚组使用强化学习的留一法（Reinforce Leave-One-Out, RLOO）目标来优化这些奖励，并结合了前1束搜索假设，并对一个冻结的基础模型进行正则化。我们在手语词汇翻译、层次文本分类和命名实体识别任务上评估了GrammarRL，使用的Llama模型参数范围从1B到8B。GrammarRL在约束贪婪解码上始终表现优越，平均提高了9.8分，最高可达22.8 BLEU。在三项任务中的两项上，其表现与束搜索相当或更优，同时保持了贪婪解码的推理成本。消融实验进一步表明，这两个奖励是互补的：单独使用任一奖励可能表现不如未训练的基线，而它们的结合则始终优于基线。
cs.AI / 117 / 2609.39903
OSWorld-Science: A Benchmark of Computer Use Agents for Learning and Using Scientific Software
OSWorld-Science：计算机使用代理在学习和使用科学软件方面的基准测试
Dai, Dingyuan, Qi, Heli, Liu, Lei, Li, Yinxi, Chen, Baiding, Dou, Zijun, Zeng, Qingcheng, Kang, Qi, Sun, Oliver, Wang, Eric, Zhou, Bo, Wang, Haixin, Du, Yufan, Bo, Shi, Lin, Ruihan, Yuan, Mengqi, Lu, Dunjie, Dillmann, Steven, Shi, Yiming, Su, Tina, Xin, Amy, Liu, Minghao, Wang, Xi, Huang, Xu, Zhang, Ge, Nie, Pengyu, Yang, Zhen, Tang, Jie, Li, Juanzi, Xuan, Weihao, Liu, Tianyu
Abstract
Scientific software presents a demanding test for computer-using agents based on visual language models (VLMs): completing a research workflow requires interpreting specialized interfaces, manipulating scientific objects, and producing verifiable results. We thus introduce OSWorld-Science, a benchmark and evaluation environment that combines scientifically meaningful tasks, artifact-based evaluation, and an efficient agent harness for studying computer use in the scientific domain. The benchmark contains 12 VLMs and 146 high-quality tasks across several scientific domains and software configurations, covering workflows such as molecular drawing and retrosynthesis, pathology image analysis, statistical computing, and physical simulation. Tasks are developed through expert proposals and iterative human--AI co-design, with selection guided by scientific value and difficulty. Task-specific execution-based evaluators inspect application states and generated artifacts, including molecular structures, segmentation masks, plots, and numerical results, and award partial credit for incomplete outcomes. Our special harness integrates model adapters, interaction-loop control, and trajectory logging to support comparisons of models and interaction strategies. Our results show that current state-of-the-art VLMs with a strong harness still face challenges in addressing key questions in the scientific domains. We also analyze the benchmarking results across multi-linguistics, reasoning efforts, context length and other factors and derive several important conclusions and directions to assist future development. Overall, we provide an integrated framework connecting expert-defined scientific goals to verifiable software outcomes, enabling systematic evaluation of both agent capabilities and harness design in scientific workflows.
Chinese Translation
科学软件对基于视觉语言模型（VLMs）的计算机使用代理提出了严峻的挑战：完成研究工作流程需要解读专业界面、操作科学对象并生成可验证的结果。因此，我们引入了OSWorld-Science，这是一个基准和评估环境，结合了科学上有意义的任务、基于工件的评估以及高效的代理工具，以研究科学领域中的计算机使用。该基准包含12个VLM和146个高质量任务，涵盖多个科学领域和软件配置，涉及分子绘图和逆合成、病理图像分析、统计计算和物理模拟等工作流程。任务通过专家提案和迭代的人机共设计开发，选择过程以科学价值和难度为指导。特定任务的执行评估者检查应用状态和生成的工件，包括分子结构、分割掩码、图表和数值结果，并对不完整的结果给予部分评分。我们的特殊工具集成了模型适配器、交互循环控制和轨迹记录，以支持模型和交互策略的比较。我们的结果表明，当前最先进的VLM在强大工具的支持下仍面临在科学领域解决关键问题的挑战。我们还分析了多语言、推理努力、上下文长度等因素的基准结果，并得出若干重要结论和发展方向，以帮助未来的开发。总体而言，我们提供了一个综合框架，将专家定义的科学目标与可验证的软件结果连接起来，使得对代理能力和科学工作流程中工具设计的系统评估成为可能。
cs.AI / 118 / 2609.39933
ConflictGuide: AutoResearch Improves When Competing Behaviors Are Made Visible
ConflictGuide：当竞争行为变得可见时，AutoResearch 的改进
Abstract
When designing machine learning models, desirable properties are often in tension: improving one behavior can impair another, so task progress can depend on alleviating the conflict. LLM-based AutoResearch systems, which iteratively edit model code and retain edits based on scalar task-performance feedback, have largely ignored this trade-off. We find that scalar feedback supports broad exploration early in search, but it does not reveal how edits affect competing behaviors. In matched-budget experiments, introducing competing-behavior feedback as task gains diminish increases the share of proposals that improve both behaviors and sustains progress beyond scalar-only plateaus. Obtaining this feedback for a given model requires identifying its competing behaviors and designing probes to measure them. To make competing-behavior feedback actionable, we introduce ConflictGuide. Its reusable ConflictGuide-Skill combines a literature-grounded taxonomy with model-specific evidence to identify competing behaviors and specify probes for a code agent to implement as metrics. Evolution proceeds in two stages: Stage I explores with task feedback; Stage II uses probe feedback to steer proposals toward conflict alleviation and retains marginal-gain edits only when probes indicate sufficient alleviation. Across five diverse model families, ConflictGuide reduces task and conflict-related errors by up to 28% and 14%, respectively, relative to scalar-only AutoResearch, with gains extending to other code agents.
Chinese Translation
在设计机器学习模型时，期望的特性往往存在张力：改善一种行为可能会损害另一种行为，因此任务进展可能依赖于缓解这种冲突。基于大型语言模型（LLM）的 AutoResearch 系统通过迭代编辑模型代码并根据标量任务性能反馈保留编辑，基本上忽视了这种权衡。我们发现，标量反馈在搜索初期支持广泛探索，但并未揭示编辑如何影响竞争行为。在匹配预算的实验中，随着任务增益的减少，引入竞争行为反馈可以增加改善两种行为的提案比例，并在标量反馈的停滞期之后持续推动进展。获取特定模型的这种反馈需要识别其竞争行为并设计探针来测量它们。为了使竞争行为反馈具有可操作性，我们引入了 ConflictGuide。其可重用的 ConflictGuide-Skill 结合了基于文献的分类法和模型特定证据，以识别竞争行为并为代码代理指定探针以实施作为度量。演化过程分为两个阶段：第一阶段使用任务反馈进行探索；第二阶段使用探针反馈引导提案朝向冲突缓解，并仅在探针表明有足够缓解时保留边际增益编辑。在五个不同的模型家族中，与仅使用标量的 AutoResearch 相比，ConflictGuide 分别将任务和冲突相关错误减少了高达 28% 和 14%，并且收益扩展到其他代码代理。
cs.AI / 119 / 2609.39955
Coverage Before Control: Route-Instruction Grounding and Steering for Controllable Retrosynthesis
控制之前的覆盖：可控逆合成的路线指令基础和引导
Abstract
Single-step retrosynthesis models are commonly evaluated by their ability to recover recorded reactions. In practice, chemists may need to choose among several precursor sets for the same product, for example to preserve a particular motif. Recovering a recorded answer alone does not establish this ability to follow a preference. Satisfying such requests requires both coverage of relevant alternatives and control over which alternatives are favored. We introduce Route-Instruction Grounding and Steering (RIGS), a two-stage framework for instruction-conditioned retrosynthesis. Stage A trains a language projector, teaching it which alternatives an instruction favors or discourages. Stage B uses the projector learned in Stage A to steer a frozen generative model through lightweight residual adapters. We construct nested one-to-many training supports by pairing each product with increasing numbers of candidate precursor sets. Extensive experiments demonstrate that broader support helps the model generate a wider range of alternatives, and RIGS can learn to guide generation according to instructions. The relationship between coverage and control is consistent across model scales but non-monotone.
Chinese Translation
单步逆合成模型通常通过其恢复记录反应的能力进行评估。在实际操作中，化学家可能需要在多个前体集合中选择同一产品的前体，例如为了保留特定的结构特征。仅仅恢复记录的答案并不能证明其遵循偏好的能力。满足此类请求需要覆盖相关的替代方案，并对哪些替代方案受到青睐进行控制。我们提出了路线指令基础和引导（Route-Instruction Grounding and Steering, RIGS），这是一个用于指令条件下逆合成的两阶段框架。阶段A训练语言投影器，教会它哪些替代方案受到指令的青睐或不被鼓励。阶段B利用阶段A中学习的投影器，通过轻量级残差适配器引导一个冻结的生成模型。我们通过将每个产品与不断增加的候选前体集合配对，构建了嵌套的一对多训练支持。大量实验表明，更广泛的支持有助于模型生成更广泛的替代方案，而RIGS能够学习根据指令引导生成。覆盖与控制之间的关系在模型规模上是一致的，但并非单调的。
cs.AI / 120 / 2609.39958
Better Deck or Different Judge? Evaluating Agentic Harness Gains in Corporate and Investment Banking
更好的报告还是不同的评审？评估企业和投资银行中的代理性工具收益
Abstract
Corporate and investment banking teams use presentations to support credit decisions and advise clients on financing and transactions. Producing these decks requires reconciling financial data, tracing sources and turning analysis into a recommendation. We retrospectively study the development of an agentic harness combining a 27B language model, financial calculations, narrative templates and validation checks. LLM judges guide engineering changes and assess the resulting decks, raising the question of whether higher scores reflect better documents or changes in grading. In shared-session text-only grading with template markers removed, five judges score the complete system 20.4 to 33.6 points out of 95 above the same model generating directly from a short prompt. Every judge scores the system higher on all seventeen development deliverables. Margins against direct Opus generation from a short prompt range from -4.7 to +0.8 points. Judges agree on broad progress across development rounds but agree less on final-deck rankings than on pooled scores. Repeated grading also shifts scores on unchanged decks, making small improvements difficult to distinguish from judge variability.
Chinese Translation
企业和投资银行团队利用报告来支持信贷决策，并为客户提供融资和交易建议。制作这些报告需要调和财务数据、追踪来源并将分析转化为建议。我们回顾性地研究了一种结合27B语言模型、财务计算、叙事模板和验证检查的代理性工具的开发。大型语言模型（LLM）评审者指导工程变更并评估生成的报告，这引发了一个问题：更高的评分是反映了更好的文档，还是评审标准的变化。在去除模板标记的共享会话文本评分中，五位评审者对完整系统的评分为20.4到33.6分（满分95分），高于同一模型直接从简短提示生成的结果。每位评审者在所有十七个开发交付物上对系统的评分均高于直接生成的结果。与直接从简短提示生成的Opus相比，评分差距范围为-4.7到+0.8分。评审者在开发轮次中对整体进展达成一致，但在最终报告排名上达成一致的程度低于对汇总评分的共识。重复评分也会影响未更改报告的评分，使得小幅改进难以与评审者的变异性区分开来。
cs.AI / 121 / 2609.39964
AIMS: An Agentic AI Framework for Sim-to-Real Multi-Modal ISAC
AIMS：一种用于模拟到现实多模态集成感知与通信的代理人工智能框架
Abstract
Multi-modal integrated sensing and communication (ISAC) enables environmental perception and reliable connectivity for intelligent wireless networks. Data-driven multi-modal ISAC models depend heavily on annotated real-world data to learn relationships across sensing and wireless observations, thereby constraining scalable deployment. Although synthetic data generation reduces the burden, adapting existing simulation pipelines to a target deployment requires consistent scene, sensing, wireless, and learning configurations, while mismatches among these coupled components impair sim-to-real transferability. To address the challenge, we propose an agentic artificial intelligence (AI) framework for sim-to-real multi-modal ISAC, named AIMS. Given a natural-language deployment request specifying the target task, deployment conditions, and real-data budget, AIMS derives a deployment-specific sim-to-real configuration and coordinates its execution to produce a deployment-specific task model. A two-agent architecture coordinates scene construction with task learning. A scene construction agent generates geographically grounded, synchronized sensing and wireless records from shared physical states, while a scene understanding agent configures task-relevant modalities and mixture-of-experts (MoE) learning for zero-shot inference or few-shot adaptation. Structured domain knowledge guides dependency-aware planning, while validation evidence supports feedback-driven revision of affected decisions. Experiments on the real-world DeepSense~6G dataset demonstrate improved vehicle detection and beam prediction over the considered simulation and fusion baselines. A separate orchestration benchmark evaluates task interpretation, dependency reasoning, and feedback-driven replanning across diverse deployment requests, showing improved plan correctness with structured domain knowledge and validation feedback.
Chinese Translation
多模态集成感知与通信（ISAC）为智能无线网络提供了环境感知和可靠连接。基于数据驱动的多模态ISAC模型在学习感知与无线观测之间的关系时，严重依赖标注的真实世界数据，从而限制了可扩展的部署。尽管合成数据生成减轻了负担，但将现有的模拟管道适应于目标部署需要一致的场景、感知、无线和学习配置，而这些耦合组件之间的不匹配会影响模拟到现实的可转移性。为了解决这一挑战，我们提出了一种用于模拟到现实多模态ISAC的代理人工智能（AI）框架，命名为AIMS。AIMS根据指定目标任务、部署条件和真实数据预算的自然语言部署请求，推导出特定于部署的模拟到现实配置，并协调其执行以生成特定于部署的任务模型。一个双代理架构协调场景构建与任务学习。场景构建代理从共享的物理状态生成地理基础的、同步的感知和无线记录，而场景理解代理则为零样本推断或少样本适应配置与任务相关的模态和专家混合（MoE）学习。结构化领域知识指导依赖感知规划，而验证证据支持受影响决策的反馈驱动修订。在真实世界的DeepSense~6G数据集上的实验表明，相较于考虑的模拟和融合基线，车辆检测和波束预测得到了改善。一个独立的协调基准评估了不同部署请求下的任务解释、依赖推理和反馈驱动的重新规划，显示出在结构化领域知识和验证反馈的支持下，计划的正确性得到了提升。
cs.AI / 122 / 2609.39989
What Can Component-Replacement Evidence Establish? A Critical Scoping Review of Local Decisions in LLM Agents
组件替换证据能建立什么？对大型语言模型代理中地方决策的批判性范围审查
Abstract
Background. A component replacement in a language-model agent changes an execution trajectory, potentially altering later observations, resource use, and recovery opportunities. Different evidence is needed to assess its task-level benefit and the contribution of local decision quality. Methods. This critical scoping review maps 348 studies and examines 90 comparison records: 88 from 40 included studies and two from supplementary studies. Eight purposively selected cases structure the synthesis around the replaced decision, executed conditions, measurement comparability, controls, and remaining explanations. Results. Of 222 studies reporting local decision metrics, 142 also report measured task endpoints and 49 report proxies. These counts identify studies that report both types of measurement, without establishing that the measurements come from matched comparisons. Outcome Monitors reports a package-level completion gain whose attribution to detector quality remains limited; First-chunk selection reports a local improvement assessed against an offline proxy endpoint; Evidence-Carrying Termination reports fewer premature unsupported terminations and completion non-inferiority, without establishing completion superiority. Cross-case analysis identifies three candidate mechanisms involving recovery and disruption, intervention timing, and downstream use. Attribution and deployment depend on the comparison controls, label definitions, and information available to the controller. Conclusions. The review distinguishes the task-level benefit of a component replacement from the contribution of local decision quality and derives eight claim-specific reporting items. Neither online execution nor simultaneous gains in local and task metrics alone establish that better local decisions explain the task-level gain.
Chinese Translation
背景：在语言模型代理中进行组件替换会改变执行轨迹，可能会影响后续观察、资源使用和恢复机会。评估其任务级益处和地方决策质量的贡献需要不同的证据。方法：本次批判性范围审查映射了348项研究，并检查了90个比较记录：88个来自40项纳入研究，两个来自补充研究。八个有目的选择的案例围绕被替换的决策、执行条件、测量可比性、控制因素和剩余解释结构化综合。结果：在222项报告地方决策指标的研究中，142项还报告了测量的任务终点，49项报告了代理指标。这些计数识别出同时报告这两种测量类型的研究，但并未确立这些测量来自匹配比较。结果监测报告了一个包级完成增益，其归因于探测器质量的限制仍然存在；首次选择报告了一个针对离线代理终点评估的地方改进；证据承载终止报告了较少的过早不支持终止和完成非劣性，但未确立完成优越性。跨案例分析识别出涉及恢复和干扰、干预时机以及下游使用的三个候选机制。归因和部署依赖于比较控制、标签定义和控制者可用的信息。结论：该审查区分了组件替换的任务级益处与地方决策质量的贡献，并推导出八个特定声明的报告项目。仅仅依靠在线执行或地方和任务指标的同时增益并不能确立更好的地方决策解释任务级增益。
cs.AI / 123 / 2609.40027
Who Verifies the Graph? Misspecification Attacks on Causal Action Verification for Language Agents
谁来验证图？对语言智能体因果行动验证的错误指定攻击
Abstract
Causal action verifiers gate an agent's state-changing tool calls by checking whether each proposed intervention is identifiable against a committed action-state graph, and they issue a certificate that carries the identification argument and a one-sided lower confidence bound. One such verifier, CIVeX, reports zero false executions on a confounded tool-use benchmark. We red-team it by corrupting only the committed graph. Omitting a single bidirected edge takes it from zero false executions to 15.3% at the benchmark's published confounding strength, with 91% of its executions harmful and utility falling from +2.27 to +0.35. Reversing one arrowhead, so that a mediator is committed as a confounder, gives 48.9% false executions and no correct ones. Every one of these actions carries an internally valid certificate. An attestation step that tests each observationally certified execution against a bounded randomised sample detected both attacks, with 2 false alarms in 555 executions on a truthful graph; refusing what fails the test, or cannot be tested, gave zero false executions in every setting we measured. It does not restore beneficial execution: at the published strength 97.1% of beneficial actions are still never executed, because the same misspecification rejects them before attestation runs. Those rejections carry certificates too, and auditing them works, but its cost scales with the number of rejections rather than the number of executions. Recovering safety costs 127 experiments per 1,050 actions; recovering the lost value costs 614 more, at which point the audited verifier makes the honest graph's decisions on every instance and spends exactly its experiment budget. An audit that inspects only executions protects against wrongful action. Wrongful inaction has to be paid for separately.
Chinese Translation
因果行动验证器通过检查每个提议的干预是否可识别于已承诺的行动-状态图，从而控制智能体的状态改变工具调用，并发出携带识别论证和单侧下置信界的证书。一个这样的验证器CIVeX在一个混淆工具使用基准上报告零错误执行。我们通过仅破坏已承诺的图来对其进行红队测试。省略一条双向边使其从零错误执行增加到15.3%，在基准的已发布混淆强度下，其中91%的执行是有害的，效用从+2.27降至+0.35。反转一个箭头，使得一个中介被承诺为混淆者，导致48.9%的错误执行且没有正确的执行。这些行动中的每一个都携带一个内部有效的证书。一个测试每个观察上认证执行与有限随机样本的证明步骤检测到了这两次攻击，在555次真实图上的执行中有2次错误警报；拒绝那些未通过测试或无法测试的情况，在我们测量的每个设置中都给出了零错误执行。它并没有恢复有益执行：在已发布的强度下，97.1%的有益行动仍然从未被执行，因为相同的错误指定在证明运行之前就拒绝了它们。这些拒绝也携带证书，审计它们是有效的，但其成本与拒绝的数量而非执行的数量成正比。恢复安全性需要每1,050个行动进行127次实验；恢复丧失的价值需要614次更多实验，此时审计的验证器在每个实例上做出诚实图的决策，并恰好花费其实验预算。仅检查执行的审计可以防止错误行动。错误的不作为必须单独支付。
cs.AI / 124 / 2609.40090
PTNO: Training Neural Operators with Noisy Monte Carlo Estimates for Particle Transport Problems
PTNO：使用噪声蒙特卡洛估计训练粒子输运神经算子
Abstract
Particle transport under multiple scattering is central to radiative transfer and plasma physics, yet high-fidelity Monte Carlo (MC) simulations must trace prohibitively many particles. Learning-based surrogates can amortize this cost, but typically train on expensive, well-converged MC solutions. We propose the Particle Transport Neural Operator (PTNO), a neural operator that learns particle transport surrogates directly from noisy, low-cost MC labels. Such labels pose two challenges: (1) high variance, which destabilizes standard supervised learning, and (2) a high dynamic range (HDR) spanning many orders of magnitude. For the first, we learn the solution operator from noisy labels of many configurations, amortizing MC cost and generalizing to unseen configurations. Because MC labels are unbiased, we show that the squared loss on them shares its minimizer with the loss on converged solutions, and our budget-allocation study over training scenes $M$, MC samples per render $N$, and independent renders per scene $K$ shows that many noisy scenes beat fewer converged ones. For the second, a nonlinear transform such as the logarithm biases noisy supervision. Instead, PTNO keeps labels in physical space and enforces positivity with a softplus output layer that represents small values effectively. We further train with a pointwise relative $L_2$ loss (PRelL2), the stop-gradient relative loss of HDR denoising and neural rendering, which normalizes each residual by the stop-gradient prediction instead of the noisy label. We demonstrate PTNO on neutron transport in fusion reactors and radiative transfer in participating media. On the two neutronics tasks, PTNO is $10^4$-$10^5\times$ faster than converged MC on the same CPU and $10^3$-$10^5\times$ cheaper than MC at matched accuracy; on the two radiative-transfer tasks, MC at matched accuracy costs $0.8$-$11\times$ as much as PTNO.
Chinese Translation
多重散射下的粒子输运是辐射传输和等离子体物理的核心，然而高保真度的蒙特卡洛（MC）模拟需要追踪大量粒子，这一成本极高。基于学习的替代方法可以分摊这一成本，但通常需要在昂贵且收敛良好的MC解上进行训练。我们提出了粒子输运神经算子（PTNO），一种直接从噪声、低成本的MC标签中学习粒子输运替代方法的神经算子。这类标签面临两个挑战：（1）高方差，这会使标准的监督学习不稳定；（2）跨越多个数量级的高动态范围（HDR）。针对第一个挑战，我们从多个配置的噪声标签中学习解算子，从而分摊MC成本并推广到未见的配置。由于MC标签是无偏的，我们证明了它们上的平方损失与收敛解上的损失共享其最小化点，并且我们对训练场景的预算分配研究显示，许多噪声场景的表现优于较少的收敛场景。针对第二个挑战，像对数这样的非线性变换会使噪声监督产生偏差。相反，PTNO保持标签在物理空间中，并通过一个softplus输出层强制正值，有效地表示小值。我们进一步使用逐点相对$L_2$损失（PRelL2）进行训练，这是HDR去噪和神经渲染的停止梯度相对损失，它通过停止梯度预测而不是噪声标签来归一化每个残差。我们在聚变反应堆中的中子输运和参与介质中的辐射传输上展示了PTNO。在这两个中子学任务中，PTNO在相同CPU上比收敛的MC快$10^4$-$10^5 imes$，并且在匹配精度下比MC便宜$10^3$-$10^5 imes$；在这两个辐射传输任务中，匹配精度下的MC成本是PTNO的$0.8$-$11 imes$。
cs.AI / 125 / 2609.40111
Agent Error Dataset: Scaling 50,000 Error--Diagnosis Pairs for Failure Analysis and Error-Aware Post-Training
代理错误数据集：扩展50,000个错误-诊断对以进行故障分析和错误感知后训练
Abstract
An unsuccessful LLM agent rollout contains more information than its final reward: the observations available to the agent, the actions it chose, and the environment's responses. Reusing this experience for learning requires identifying a decision to revise and testing a concrete alternative. We introduce the Agent Error Dataset (AED), comprising 50,228 error-diagnosis pairs from 9,961 source tasks across 33 environments, 19 harness families, and 23 policy models in text-based agent systems. We retain source traces and execution metadata to support cross-setting failure analysis and re-diagnosis without repeating the original rollout. Our five-stage Agentic Error-to-Training (AET) pipeline collects natural failures, generates diagnoses and proposed corrections, and checks them against recorded evidence. Where replay is supported, we compare corrections with original-action retries from the same checkpoint under matched execution settings. We then construct separate training views for diagnosis and actor recovery. Across 3,062 matched replay pairs, first-proposal corrections raise verifier pass rates from 18.4% to 51.1%, a gain of 32.7 percentage points. Using a separately frozen diagnosis release, full-diagnosis fine-tuning on 1,656 source tasks raises Qwen3-8B's exact-step agreement with internal teacher labels from 47.2% to 63.6%, averaged over three seeds on a 943-case holdout. The strongest prompted reference in this comparison scores 54.7%, and mean agreement improves at each of four increasing training-set sizes. In a single-seed comparison of actor-training recipes, action-only repair training scores 6.67 percentage points higher on WebShop-lite than success-only training.
Chinese Translation
一次不成功的LLM代理部署包含的信息比其最终奖励更为丰富：代理可用的观察、其选择的行动以及环境的响应。重新利用这些经验进行学习需要识别需要修正的决策并测试具体的替代方案。我们引入了代理错误数据集（Agent Error Dataset，AED），该数据集包含来自33个环境、19个工具家族和23个策略模型的9,961个源任务中的50,228个错误-诊断对。我们保留源跟踪和执行元数据，以支持跨设置的故障分析和重新诊断，而无需重复原始部署。我们的五阶段代理错误到训练（Agentic Error-to-Training，AET）流程收集自然故障，生成诊断和建议的修正，并将其与记录的证据进行核对。在支持重放的情况下，我们将修正与来自相同检查点的原始行动重试进行比较，确保执行设置匹配。然后，我们为诊断和演员恢复构建单独的训练视图。在3,062个匹配重放对中，首次提议的修正将验证者通过率从18.4%提高到51.1%，提升了32.7个百分点。使用单独冻结的诊断版本，在1,656个源任务上进行的全诊断微调使Qwen3-8B与内部教师标签的精确步骤一致性从47.2%提高到63.6%，在943个案例的保留集上平均计算。此比较中最强的提示参考得分为54.7%，而在四个逐渐增大训练集规模的情况下，平均一致性均有所提高。在一次单种子对演员训练方案的比较中，仅行动修复训练在WebShop-lite上的得分比仅成功训练高出6.67个百分点。
cs.AI / 126 / 2609.40115
Unlearnable, or Unmeasured? On the Reliability of Difficulty Labels in RLVR
不可学习，还是未测量？关于强化学习可验证奖励中的难度标签的可靠性
Abstract
Reinforcement learning with verifiable rewards (RLVR) has become an important approach for improving reasoning during post-training. Recent work suggests that some difficult prompts remain resistant to learning even when they occasionally produce correct solutions. We revisit this unlearnability phenomenon and find that the affected prompts do improve, at roughly one third of the learnable rate, while the difficulty-defined set used to study them is much less reproducible than expected. These difficulty labels are estimated from a limited number of sampled responses. Combining them across seeds can further change which prompts are selected instead of simply reducing measurement noise. We develop a sampling-based framework for quantifying this instability and determining how much evaluation is required for difficulty assignments to reproduce reliably. We also revisit the gradient-similarity evidence proposed to explain unlearnability and show that part of the observed separation arises because difficult prompts provide fewer correct rollouts from which their gradients can be estimated. Matching this sample count weakens the gradient difference but does not remove it. Overall, the slow-learning phenomenon survives our reanalysis, while both the prompts used to define it and the evidence used to explain it require more careful measurement.
Chinese Translation
带有可验证奖励的强化学习（RLVR）已成为提高后训练推理的重要方法。近期研究表明，某些困难的提示在偶尔产生正确解决方案时仍然对学习保持抵抗。我们重新审视这一不可学习现象，发现受影响的提示确实有所改善，改善速度约为可学习速率的三分之一，而用于研究这些提示的难度定义集的可重复性远低于预期。这些难度标签是从有限数量的样本响应中估计得出的。跨种子组合这些标签可能会进一步改变所选择的提示，而不仅仅是减少测量噪声。我们开发了一种基于采样的框架，用于量化这种不稳定性，并确定为了使难度分配可靠地再现所需的评估量。我们还重新审视了用于解释不可学习性的梯度相似性证据，并表明观察到的分离部分源于困难提示提供的正确回滚数量较少，从而导致其梯度的估计。匹配这一样本数量会削弱梯度差异，但并不会消除它。总体而言，慢学习现象在我们的重新分析中依然存在，而定义这一现象的提示和用于解释它的证据都需要更为细致的测量。
cs.AI / 127 / 2609.40169
Learning from Research: Toward Lifelong Agent Harness Evolution
从研究中学习：迈向终身智能体工具演化
Abstract
Language agents are expected to solve increasingly complex tasks, creating a growing need for continual improvement. One promising approach is to evolve the agent harness, the software that governs tool use, memory management, and task execution, while keeping the underlying language model fixed. Recent methods automate this process by using a meta coding agent to modify the harness based on execution feedback. However, relying on that agent's existing knowledge and observed failures can restrict exploration and make adaptation reactive. Inspired by how human experts learn from the research literature for new solutions, we introduce ScholarEvolve, a framework that automatically draws on state-of-the-art research to guide harness evolution. ScholarEvolve organizes the harness evolution directions into functional modules and uses topic modeling to identify distinct improvement strategies for each module. It implements these strategies and evaluates their combinations to improve task performance. Moreover, the framework is designed to incorporate new publications over time, allowing research advances to drive proactive lifelong evolution. Experiments demonstrate improvements on AppWorld and Tau2-Bench. ScholarEvolve raises Qwen3.5-27B task goal completion from 49.6% to 63.6% on AppWorld Challenge, and raises GPT-5.4-mini pass@1 from 72.7% to 81.9% on Tau2-Bench Telecom.
Chinese Translation
语言智能体被期望解决日益复杂的任务，这导致对持续改进的需求不断增加。一种有前景的方法是演化智能体工具，即管理工具使用、内存管理和任务执行的软件，同时保持底层语言模型不变。最近的方法通过使用元编码智能体，根据执行反馈自动修改工具。然而，依赖该智能体的现有知识和观察到的失败可能限制探索，并使适应变得反应性。受到人类专家如何从研究文献中学习新解决方案的启发，我们提出了ScholarEvolve，一个自动利用最先进研究来指导工具演化的框架。ScholarEvolve将工具演化方向组织成功能模块，并使用主题建模来识别每个模块的不同改进策略。它实施这些策略并评估其组合以提高任务性能。此外，该框架设计为随着时间的推移纳入新出版物，允许研究进展推动主动的终身演化。实验表明在AppWorld和Tau2-Bench上有显著改进。ScholarEvolve将Qwen3.5-27B在AppWorld挑战中的任务目标完成率从49.6%提高到63.6%，并将GPT-5.4-mini在Tau2-Bench Telecom上的pass@1从72.7%提高到81.9%。
cs.AI / 128 / 2609.40269
Belief-Aware Multi-Agent Path Finding under Map Uncertainty
考虑信念的多智能体路径规划在地图不确定性下的研究
Abstract
Multi-Agent Path Finding (MAPF) aims to find collision-free paths for multiple agents in a shared environment. Classical MAPF assumes that all static obstacles are known in advance, but real-world environments can change unexpectedly due to fallen objects, spills, or other local disturbances. When such changes are spatially correlated, an observation can inform traversability estimates beyond the observed location. Prior approaches address uncertainty in traversability through contingent plans or replanning based on direct observations, but do not leverage this spatial dependence to infer the traversability of nearby unobserved locations. As a result, they cannot use one observation to anticipate nearby unobserved obstacles that may cause costly rerouting later. We focus on Belief-Aware MAPF, where map discrepancies are fixed during execution but initially unknown, and observations can be informative beyond the observed location. We propose Multi-Agent Gaussian belief Inference for Coordination (MAGIC), a framework that updates a shared belief about traversability online based on agents' observations. MAGIC uses a Gaussian Markov Random Field and Gaussian Belief Propagation to approximately infer traversability and construct detour-aware costs for standard MAPF planners. Our experiments on MAPF benchmarks show that MAGIC reduces the executed sum of costs compared to existing approaches on 96.3% of instances, across several planner families and teams of up to 800 agents, demonstrating its applicability to large-scale MAPF problems.
Chinese Translation
多智能体路径规划（MAPF）旨在为多个智能体在共享环境中找到无碰撞的路径。传统的MAPF假设所有静态障碍物都是事先已知的，但现实世界中的环境可能因物体掉落、液体泄漏或其他局部干扰而发生意外变化。当这些变化在空间上相关时，某一观察结果可以提供超出观察位置的可通行性估计。以往的方法通过应急计划或基于直接观察的重新规划来处理可通行性的的不确定性，但未能利用这种空间依赖性来推断附近未观察位置的可通行性。因此，它们无法利用一次观察来预测可能导致后续高成本重新规划的附近未观察障碍物。我们关注考虑信念的MAPF，其中地图差异在执行过程中是固定的但最初未知，且观察结果可以提供超出观察位置的信息。我们提出了多智能体高斯信念推断协调（MAGIC）框架，该框架基于智能体的观察在线更新对可通行性的共享信念。MAGIC使用高斯马尔可夫随机场和高斯信念传播来近似推断可通行性，并为标准MAPF规划器构建考虑绕行的成本。我们在MAPF基准测试中的实验表明，MAGIC在96.3%的实例中减少了执行的总成本，相较于现有方法，适用于多达800个智能体的多个规划器家族和团队，展示了其在大规模MAPF问题中的适用性。
cs.AI / 129 / 2609.40285
PivotOPD: Learning to Recover from Pivotal Mistakes in Multi-Turn Agents
PivotOPD：学习从多轮智能体中的关键错误中恢复
Abstract
On-policy distillation (OPD) is a promising approach for training language agents, providing dense teacher supervision on student-generated trajectories. However, in multi-turn interaction, an incorrect action changes the states the student encounters later, so errors compound across turns. In preliminary experiments across three Qwen3 models (8B to 235B), we find that more than half of the failed rollouts contain a pivotal mistake, an action that moves the agent farther from completing the task, and this mistake typically occurs early. These pivotal mistakes often remain recoverable: guiding the model for only a few turns after the pivotal turn can restore task success. We therefore propose PivotOPD, an on-policy distillation framework that jointly trains the student to prevent pivotal mistakes and to recover from the states they create. At each pivotal mistake, a teacher model provides a gold action and then names a recovery action at each of the next few turns. Preventive distillation uses the gold action with reverse KL to steer the student away from the pivotal mistake, while recovery distillation uses the recovery actions with forward KL to transfer recovery behaviors that the student rarely samples. Against 13 baselines on ALFWorld, WebShop, and Search-based QA, PivotOPD achieves the strongest average performance for both Qwen3-1.7B and Qwen3-8B students, improving over the strongest baseline on ALFWorld by +5.5% with the 1.7B student. The gains also transfer to another model family on the software engineering domain, where PivotOPD raises the resolve rate of a Nemotron-3.5 student on SWE-Bench Verified by +3.2%. Project page: https://research.nvidia.com/labs/lpr/pivotopd/
Chinese Translation
在政策蒸馏（On-policy distillation, OPD）中，提供密集的教师监督以训练语言智能体，能够有效地指导学生生成的轨迹。然而，在多轮交互中，一个错误的动作会改变学生后续遇到的状态，从而导致错误在多个回合中累积。在对三个 Qwen3 模型（8B 到 235B）的初步实验中，我们发现超过一半的失败回合包含一个关键错误，即使智能体离完成任务更远的动作，而这个错误通常发生在早期。这些关键错误往往是可恢复的：在关键回合后仅指导模型几轮即可恢复任务成功。因此，我们提出了 PivotOPD，一个政策蒸馏框架，联合训练学生以防止关键错误并从其产生的状态中恢复。在每个关键错误发生时，教师模型提供一个黄金动作，并在接下来的几轮中指定一个恢复动作。预防性蒸馏使用黄金动作与反向 KL（Kullback-Leibler）引导学生远离关键错误，而恢复性蒸馏则使用恢复动作与正向 KL 转移学生罕见采样的恢复行为。在 ALFWorld、WebShop 和基于搜索的问答任务上与 13 个基线进行比较，PivotOPD 在 Qwen3-1.7B 和 Qwen3-8B 学生中实现了最强的平均性能，1.7B 学生在 ALFWorld 上比最强基线提高了 +5.5%。这些增益也转移到软件工程领域的另一个模型系列上，PivotOPD 将 Nemotron-3.5 学生在 SWE-Bench Verified 上的解决率提高了 +3.2%。项目页面：https://research.nvidia.com/labs/lpr/pivotopd/
cs.AI / 130 / 2609.40303
How Much of a Harness Does a Strong Agent Need for Autonomous ML Engineering?
强代理在自主机器学习工程中需要多少约束？
Abstract
Recent autonomous machine learning engineering (MLE) agents have made significant progress on public leaderboards. Often motivated by progress stagnation over long-horizon cycles and limited Large Language Model (LLM) primitives, modern MLE agents are deployed on top of increasingly elaborate machinery: multi-agent orchestrators, dedicated retrieval subagents, and more. While such harnesses expand, the use of more primitive but improved coding agents - where LLMs have direct access to the execution environment through read, write, and bash primitives - has received little attention in the field. In this paper we find that, under an equal time budget and the same frontier LLM backbone, open-source state-of-the-art harnesses provide no advantages over a single session of a minimal-harness coding agent baseline, pointing to the backbone as the primary driver for performance. Via a series of large-scale systematic ablation studies, we argue that the machinery layers become redundant in the coding agent setting. We conclude that the effort spent elaborating hand-crafted harnesses around strong models yields poor returns for current MLE benchmarks.
Chinese Translation
近年来，自主机器学习工程（MLE）代理在公共排行榜上取得了显著进展。现代MLE代理通常受到长周期进展停滞和有限的大型语言模型（LLM）原语的驱动，因而被部署在越来越复杂的机制之上：多代理协调器、专用检索子代理等。然而，尽管这些约束不断扩展，但对更原始但改进的编码代理的使用——即LLM通过读取、写入和bash原语直接访问执行环境——在该领域却鲜有关注。在本文中，我们发现，在相同的时间预算和相同的前沿LLM基础上，开源的最先进约束与单次会话的最小约束编码代理基线相比，并没有提供任何优势，这表明基础模型是性能的主要驱动因素。通过一系列大规模系统性消融研究，我们认为在编码代理设置中，机械层变得多余。我们得出结论，围绕强模型构建手工约束所花费的努力，对于当前的MLE基准测试回报甚微。
cs.AI / 131 / 2609.40324
Cogentic: Multi-Agent Orchestration for Automated Proof Discovery
Cogentic：用于自动证明发现的多智能体协调
Abstract
We present Cogentic, a multi-agent harness for automated proof discovery on open research problems. While frontier language models can generate strong mathematical ideas in a single shot, single-shot generation is often insufficient for open problems that require exploring multiple competing conjectures, overcoming subtle technical obstructions, and retaining intermediate progress over a long horizon. Cogentic addresses these challenges through an iterative prove--verify loop in which an orchestrator allocates a population of independent provers across distinct proof directions, subjects their output to adversarial verification by several specialized components, and promotes confirmed intermediate results into a persistent verified ledger that later rounds build on. The harness is designed to be able to solve research-level math and theoretical computer science problems. Using Gemini as the base model, Cogentic produced novel results on five open problems across online learning, auction theory, and mechanism design. Each result was independently verified by domain experts and is developed in full in companion papers. We list these results, and new ones as they are verified, at https://sites.google.com/view/cogentic .
Chinese Translation
我们提出了Cogentic，一个用于开放研究问题的自动证明发现的多智能体平台。尽管前沿语言模型能够在一次生成中产生强有力的数学思想，但单次生成通常不足以解决需要探索多个竞争猜想、克服微妙技术障碍并在较长时间内保持中间进展的开放问题。Cogentic通过一个迭代的证明-验证循环来应对这些挑战，其中一个协调者在不同的证明方向上分配一组独立的证明者，将它们的输出提交给多个专门组件进行对抗性验证，并将确认的中间结果提升到一个持久的验证账本上，以供后续轮次构建。该平台旨在能够解决研究级别的数学和理论计算机科学问题。以Gemini作为基础模型，Cogentic在在线学习、拍卖理论和机制设计等五个开放问题上产生了新颖的结果。每个结果都经过领域专家的独立验证，并在相关论文中进行了详细阐述。我们在https://sites.google.com/view/cogentic列出了这些结果，以及新的结果在验证后将被添加。
cs.AI / 132 / 2609.40325
WorldAuditBench: Interactive 3D World Auditing with Multimodal Agents
WorldAuditBench：基于多模态智能体的交互式三维世界审计
Abstract
As interactive 3D worlds are increasingly used to study intelligent behavior, it becomes important to develop efficient pipelines for identifying anomalies in these simulated environments, such as floating objects, traversable walls, or objects inconsistent with the surrounding scene. Multimodal AI systems, including vision-language models (VLMs) and vision-language-action models (VLAs), have shown potential for automating this task. However, 3D world auditing is complex, requiring the close coupling of two distinct capabilities: action, to navigate the 3D world and search for anomalies systematically and efficiently; and visual reasoning, to understand the environment and identify anomalies from multimodal observations. It remains largely unexplored whether multimodal agents can effectively couple these two capabilities, using visual reasoning to identify potential anomalies while taking actions to validate them. In this paper, we introduce WorldAuditBench, a benchmark for 3D world auditing comprising 213 anomaly tasks across 13 environments built with Unreal Engine 5 and Three.js, spanning five anomaly families. We evaluate five frontier models under a fixed exploration budget using two auditing paradigms: VLA-based exploration followed by VLM-based anomaly identification, and an end-to-end VLM agent in which visual reasoning directly guides action selection. Across the evaluated models and two paradigms, success rates range from 6.6% to 42.3%, substantially below human performance (83.4%). Through the task of world auditing, WorldAuditBench provides a testbed for studying how multimodal agents couple action and visual reasoning in interactive 3D environments, while highlighting current limitations in their ability to gather and interpret evidence during exploration.
Chinese Translation
随着交互式三维世界在智能行为研究中的应用日益广泛，开发高效的流程以识别这些模拟环境中的异常现象（如漂浮物体、可穿越的墙壁或与周围场景不一致的物体）变得尤为重要。多模态人工智能系统，包括视觉-语言模型（Vision-Language Models, VLMs）和视觉-语言-动作模型（Vision-Language-Action Models, VLAs），在自动化执行此任务方面展现出潜力。然而，三维世界审计是一项复杂任务，需紧密结合两种不同能力：动作能力，用于在三维世界中系统且高效地导航和搜索异常；视觉推理能力，用于理解环境并从多模态观测中识别异常。目前尚未充分探讨多模态智能体能否有效结合这两种能力，即利用视觉推理识别潜在异常，同时采取行动进行验证。本文提出WorldAuditBench，这是一个三维世界审计基准，涵盖基于Unreal Engine 5和Three.js构建的13个环境中的213个异常任务，涵盖五大异常类别。我们在固定探索预算下，采用两种审计范式评估了五个前沿模型：基于VLA的探索后接VLM的异常识别，以及端到端的VLM智能体，其中视觉推理直接指导动作选择。在所评估的模型和两种范式中，成功率介于6.6%至42.3%之间，远低于人类表现（83.4%）。通过世界审计任务，WorldAuditBench为研究多模态智能体如何在交互式三维环境中结合动作与视觉推理提供了测试平台，同时揭示了其在探索过程中收集和解读证据能力上的现有限制。
cs.AI / 133 / 2609.40330
Turbo Harness: Instance-Adaptive Harness Optimization
Turbo Harness：实例自适应的马具优化
Abstract
Automating the search for effective harnesses is an important step toward enabling agents to recursively self-improve. Existing harness optimizations typically produce a single global harness that is applied uniformly across task instances. However, a harness that works well on average may not be optimal for every instance. We introduce Turbo Harness, a framework that can adapt a globally optimized harness to each instance by reusing information generated during the original optimization process. Specifically, Turbo Harness recycles artifacts produced during a completed global harness optimization run, and summarizes them into a structured playbook. We train a harness editor to leverage this prior optimization experience to generate instance-specific patches to the global harness. At inference time, the editor uses the instance and the playbook to construct a tailored harness in which the execution model operates. Through numerical experiments, we show that Turbo Harness consistently outperforms existing harness optimization baselines across seven benchmarks spanning interactive agent tasks, software engineering, and long-horizon terminal tasks.
Chinese Translation
自动化搜索有效马具是使智能体能够递归自我改进的重要步骤。现有的马具优化通常生成一个在任务实例中均匀应用的单一全局马具。然而，平均表现良好的马具可能并不适用于每个实例。我们提出了Turbo Harness，一个可以通过重用在原始优化过程中生成的信息，将全局优化的马具适应于每个实例的框架。具体而言，Turbo Harness回收在完成全局马具优化运行期间产生的工件，并将其总结为一个结构化的操作手册。我们训练了一个马具编辑器，以利用这一先前的优化经验，为全局马具生成特定于实例的补丁。在推理时，编辑器使用实例和操作手册构建一个量身定制的马具，以供执行模型操作。通过数值实验，我们展示了Turbo Harness在七个基准测试中始终优于现有的马具优化基线，这些基准涵盖了交互式智能体任务、软件工程和长时间终端任务。
cs.CL / 1 / 2609.38181
Large Language Models are Approximate Survival Estimators
大型语言模型是近似生存估计器
Abstract
Survival analysis estimates time-to-event outcomes from patient covariates and is widely used for medical risk assessment. Patients seeking prognostic information after a diagnosis may turn to large language models (LLMs), now readily accessible through consumer applications. However, whether LLMs can provide accurate survival predictions has not been rigorously evaluated. We introduce Survprompt, a framework that converts structured patient covariates into free-text clinical vignettes and prompts pre-trained LLMs to predict survival zero-shot. We benchmark Survprompt against conventional survival models, including random survival forests (RSF), across two multi-institutional pan-cancer cohorts: the publicly available MSK-CHORD cohort and a newly curated cohort from the Providence St. Joseph Health Network constructed using an LLM-based medical abstraction framework. We report censored mean absolute error (cMAE) and concordance index (c-index) and conduct feature ablations to identify variables influencing LLM predictions. Frontier LLMs achieved surprisingly competitive cMAE for individual survival times. For example, GPT-5.6-Sol achieved cMAE within 10% of state-of-the-art RSF models specifically trained for survival prediction for several cancer types and lower cMAE than RSF for prostate cancer in MSK-CHORD. Feature ablations revealed that LLMs prioritized clinical variables similarly to specialized survival models. However, LLMs showed inconsistent accuracy across cancer types and institutions and poorly discriminated between high- and low-risk patients (lower c-index). Zero-shot LLMs can generate surprisingly accurate prognostic estimates without specialized training, but their variable performance across cancer types and institutions remains an important limitation for clinical use.
Chinese Translation
生存分析通过患者协变量估计事件发生的时间，并广泛用于医疗风险评估。在诊断后寻求预后信息的患者可能会转向大型语言模型（LLMs），这些模型现在可以通过消费应用程序轻松获取。然而，LLMs是否能够提供准确的生存预测尚未经过严格评估。我们介绍了Survprompt，这是一个将结构化患者协变量转换为自由文本临床小插曲的框架，并提示预训练的LLMs进行零样本生存预测。我们将Survprompt与传统生存模型进行基准测试，包括随机生存森林（RSF），在两个多机构泛癌症队列中进行比较：公开可用的MSK-CHORD队列和一个新策划的队列，该队列来自于使用基于LLM的医学抽象框架构建的Providence St. Joseph Health Network。我们报告了审查后的平均绝对误差（cMAE）和一致性指数（c-index），并进行特征消融以识别影响LLM预测的变量。前沿LLMs在个体生存时间的cMAE上表现出令人惊讶的竞争力。例如，GPT-5.6-Sol在多个癌症类型的生存预测中，其cMAE与专门训练的最先进RSF模型相差不超过10%，并且在MSK-CHORD中对前列腺癌的cMAE低于RSF。特征消融显示，LLMs优先考虑临床变量，类似于专业生存模型。然而，LLMs在不同癌症类型和机构之间的准确性不一致，并且在高风险和低风险患者之间的区分能力较差（c-index较低）。零样本LLMs能够在没有专门训练的情况下生成令人惊讶的准确预后估计，但它们在不同癌症类型和机构之间的可变表现仍然是临床应用中的一个重要限制。
cs.CL / 2 / 2609.38201
TomasuLLM: Out-of-Order Speculative Execution for LLM Agents
TomasuLLM：面向大型语言模型代理的乱序推测执行
Abstract
Long-running tools can dominate coding-agent latency: compilers, test suites, and repository commands take seconds to minutes while the agent idles. This observation stall presents the same tension that drove out-of-order processors -- asequential interface hides work that can be predicted and started early, but a speculative result may become visible only after it and every earlier step have been validated. We present TomasuLLM, a runtime that executes agent tool calls out of trajectory order while preserving task-execution correctness. It drafts future actions, runs them in isolated copy-on-write sandboxes, traces their dependencies and effects, and commits results in trajectory order only after validation against committed state. Across three benchmarks spanning sub-second to minutes-long tool calls, TomasuLLM improves the reported benchmark means and scales with tool latency: 1.31x on 100 SWE-bench Verified tasks, 1.35x on 28 Terminal-Bench 2.0 tasks, and 1.27x matched progress on 18 SWE-Marathon sessions. Across 4,010 audited commit-validation records, it produces zero false accepts.
Chinese Translation
长时间运行的工具可能会主导编码代理的延迟：编译器、测试套件和仓库命令的执行时间从几秒到几分钟不等，而代理则处于闲置状态。这一观察到的停滞现象呈现出与驱动乱序处理器相同的紧张关系——非顺序接口隐藏了可以预测并提前启动的工作，但推测结果可能只有在其及所有早期步骤都经过验证后才会显现。我们提出了TomasuLLM，这是一种在保持任务执行正确性的同时，乱序执行代理工具调用的运行时。它草拟未来的操作，在隔离的写时复制沙箱中运行这些操作，追踪它们的依赖关系和效果，并仅在与已提交状态验证后按轨迹顺序提交结果。在涵盖从亚秒到分钟级工具调用的三个基准测试中，TomasuLLM提高了报告的基准均值，并随着工具延迟的增加而扩展：在100个SWE-bench验证任务上提高了1.31倍，在28个Terminal-Bench 2.0任务上提高了1.35倍，在18个SWE-Marathon会话中实现了1.27倍的匹配进展。在4,010个审核的提交验证记录中，未产生任何错误接受。
cs.CL / 3 / 2609.38203
Automatic estimation of verbal fluency index in people with Motor Neuron Disease using ASR alignment and pause modelling
基于ASR对齐和停顿建模的运动神经元病患者语言流畅性指数的自动估计
Abstract
Monitoring cognitive impairment (CI) in motor neuron disease (MND) is essential for timely treatment and care, yet challenging due to co-occurring speech difficulties. The Edinburgh Cognitive and Behavioural ALS Screen (ECAS) provides a robust metric for CI assessment, with the Verbal Fluency Index (VFI) a central element. Building on recent advances in automated speech analysis, this study proposes a system for estimating VFI. It leverages a unique MND dataset and combines ASR (WhisperX) and VAD (Silero) with refined timestamping to predict the VFI and extract several clinically interpretable measures. Our approach outperformed systems based on traditional acoustic features and self-supervised embeddings, evaluated using multiple regression algorithms. Clinically inspired features consistently outperformed the other sets, with the best models achieving strong results (P-words: R2 0.9, NRMSE 0.05; S-words: R2 0.8, NRMSE 0.08), demonstrating the feasibility of automated VFI estimation.
Chinese Translation
监测运动神经元病（MND）患者的认知障碍（CI）对于及时治疗和护理至关重要，但由于伴随的言语困难，这一过程充满挑战。爱丁堡认知与行为ALS筛查（ECAS）提供了一个可靠的CI评估指标，其中语言流畅性指数（VFI）是核心要素。基于近期自动语音分析的进展，本研究提出了一种估计VFI的系统。该系统利用独特的MND数据集，结合ASR（WhisperX）和VAD（Silero）以及精细的时间戳预测VFI，并提取多个临床可解释的指标。我们的研究方法在使用多重回归算法评估时，表现优于基于传统声学特征和自监督嵌入的系统。临床启发特征始终优于其他特征集，最佳模型取得了良好的结果（P-words: R2 0.9, NRMSE 0.05; S-words: R2 0.8, NRMSE 0.08），证明了自动化VFI估计的可行性。
cs.CL / 4 / 2609.38205
The System Prompt Illusion: How Instruction Preambles Modify Computation in Language Models
系统提示幻觉：指令前言如何修改语言模型中的计算
Abstract
System prompts are the primary lever practitioners use to control language model behavior, yet what they actually do to the computation inside the transformer remains poorly understood. Across 17 instruction-tuned models spanning 8 architecture families and 1.5B to 72B parameters, we use Centered Kernel Alignment (CKA) to compare layer-wise representations under 20 system prompts in five functional categories. Effects are layer-selective and instruction-type-dependent: persona and formatting instructions deeply restructure intermediate representations, while safety instructions barely move them, producing changes statistically indistinguishable from a minimal baseline. Restrictive safety instructions and explicitly permissive ones ("you have no restrictions") engage near-identical computational pathways (mean CKA correlation 0.997), and this persists at commercial scale, where safety penetration remains below 10% even at 70B-72B. A linear probing baseline exposes the mechanism: the model encodes prompt category at every layer but restructures its computation only at a small subset, so the prompt is reliably "seen" but, for safety, not deeply "acted upon." Causal activation patching confirms these layers mediate behavioral change, and representational depth predicts behavioral effect size across the full 17-model cohort (Spearman rho = 0.761, p < 0.001). The findings provide a mechanistic explanation for the persistent jailbreak vulnerability of system-prompt-based safety. Code: https://github.com/Usama1002/system-prompt-illusion-cka
Chinese Translation
系统提示是从业者控制语言模型行为的主要手段，但它们对变换器内部计算的实际影响仍然不甚清楚。在17个经过指令调优的模型中，涵盖8个架构系列和1.5B到72B参数，我们使用中心核对齐（Centered Kernel Alignment, CKA）比较在五个功能类别下20个系统提示下的层级表示。效果是层选择性和指令类型依赖的：个性化和格式化指令深刻重构中间表示，而安全指令几乎没有影响，产生的变化在统计上与最小基线无显著差异。限制性安全指令和明确允许的指令（“你没有任何限制”）在计算路径上几乎相同（平均CKA相关性为0.997），这种情况在商业规模上依然存在，即使在70B-72B时，安全渗透率仍低于10%。线性探测基线揭示了这一机制：模型在每一层编码提示类别，但仅在小部分层重构其计算，因此提示被可靠地“看到”，但出于安全考虑并未被深度“作用”。因果激活补丁确认这些层介导行为变化，表示深度预测在整个17模型群体中的行为效应大小（Spearman rho = 0.761, p < 0.001）。这些发现为基于系统提示的安全性持续越狱脆弱性提供了机制性解释。代码链接： https://github.com/Usama1002/system-prompt-illusion-cka
cs.CL / 5 / 2609.38219
TutlAit v1: a crowdsourced Moroccan Tamazight speech dataset with Arabic transcriptions and regional accent labels
TutlAit v1：一个众包的摩洛哥塔马齐赫语语音数据集，包含阿拉伯语转录和地区口音标签
Abstract
Tamazight (Amazigh) is, together with Arabic, one of the two official languages of Morocco, yet it remains severely under-resourced for speech technology: pub licly available labelled audio is scarce, generally lacks information on the regional variety spoken, and is often of uneven transcription quality. This article describes the TutlAit dataset, a corpus of Moroccan Tamazight speech paired with Modern Standard Arabic text and explicit regional accent labels. The data were collected with TutlAit, a purpose-built crowdsourcing web application (React 18 front end, Django 5 / Django REST Framework back-end, PostgreSQL database). Native speakers recruited through targeted LinkedIn and Instagram campaigns created an account, declared their regional variety (Atlas, Souss, Rif or other) and demographic information, and then contributed through two workflows: Text-to Audio, in which an Arabic sentence is displayed and the volunteer records its oral Tamazight rendering in the browser, and Audio-to-Text, in which a Tamazight excerpt is played and the volunteer types its Arabic transcription. A complemen tary set of segments was obtained from freely accessible Tamazight audiovisual media, segmented and annotated with ELAN and imported through a bulk CSV/ZIP pipeline. Every upload is converted server-side to 16kHz mono WAV, hashed with SHA-256 for duplicate rejection, checked for duration bounds and validated by an administrator. The dataset contains 13,384 audio files totalling 75,231 seconds (approximately 20.9 hours, about 3.01GB). The Atlas variety accounts for 9,956 files (14.08h) and the Souss variety for 3,378 files (6.75h); small Rif (22 files) and Kabyle (28 files) subsets are also included. The corpus can be reused for speech recognition, speech translation and accent identification for Moroccan Tamazight.
Chinese Translation
塔马齐赫语（Amazigh）是摩洛哥的两种官方语言之一，与阿拉伯语并列，但在语音技术方面仍然严重缺乏资源：公开可用的标注音频稀少，通常缺乏所说地区变体的信息，并且转录质量往往不均匀。本文描述了TutlAit数据集，这是一个摩洛哥塔马齐赫语语音语料库，配有现代标准阿拉伯语文本和明确的地区口音标签。数据通过TutlAit收集，这是一款专门构建的众包网络应用程序（前端使用React 18，后端使用Django 5 / Django REST Framework，数据库为PostgreSQL）。通过针对LinkedIn和Instagram的定向宣传活动招募的母语者创建了账户，声明了他们的地区变体（阿特拉斯、苏斯、里夫或其他）和人口信息，然后通过两个工作流程进行贡献：文本转语音（Text-to-Audio），在该流程中显示阿拉伯语句子，志愿者在浏览器中录制其口头塔马齐赫语表达；语音转文本（Audio-to-Text），在该流程中播放塔马齐赫语摘录，志愿者输入其阿拉伯语转录。通过自由获取的塔马齐赫语视听媒体获得了一组补充段落，经过分段和使用ELAN注释，并通过批量CSV/ZIP管道导入。每次上传都在服务器端转换为16kHz单声道WAV格式，使用SHA-256进行哈希以拒绝重复，检查持续时间限制并由管理员验证。该数据集包含13,384个音频文件，总时长为75,231秒（约20.9小时，约3.01GB）。阿特拉斯变体占9,956个文件（14.08小时），苏斯变体占3,378个文件（6.75小时）；还包括小规模的里夫（22个文件）和卡比尔（28个文件）子集。该语料库可用于摩洛哥塔马齐赫语的语音识别、语音翻译和口音识别。
cs.CL / 6 / 2609.38222
Conformal Factuality Control for Multi-Hop Retrieval-Augmented Generation
多跳检索增强生成的符合性事实控制
Abstract
Retrieval-augmented generation (RAG) can ground large language models in external evidence, but retrieved context does not guarantee that generated claims are factually supported. This problem is especially relevant in multi-hop RAG, where retrieval and reasoning proceed through multiple dependent stages. We study whether claim-level conformal factuality control, previously developed for RAG, remains effective in this setting. We apply split-conformal claim filtering to multi-hop RAG and evaluate it on HotpotQA, Natural Questions, and TriviaQA using Llama 3.1 8B and GPT-4o-mini, together with a single-hop reference experiment. Across all six multi-hop model-dataset configurations, increasingly stringent conformal targets consistently increase the fraction of responses whose retained claims are fully supported. At the 95% target, this rate ranges from 95.80% to 97.20%, compared with 55.60%-76.03% without filtering. However, the improvement is strongly selective: only 4.41%-31.09% of generated claims are retained and 9.70%-51.40% of responses remain non-empty at the 95% target. These results show that conformal factuality extends to multi-hop RAG, while demonstrating that nominal reliability must be interpreted jointly with claim retention and abstention.
Chinese Translation
检索增强生成（RAG）可以将大型语言模型与外部证据相结合，但检索到的上下文并不能保证生成的声明在事实上得到支持。这个问题在多跳RAG中尤为重要，因为检索和推理通过多个相互依赖的阶段进行。我们研究了之前为RAG开发的声明级符合性事实控制在这一环境中是否仍然有效。我们将分裂符合性声明过滤应用于多跳RAG，并在HotpotQA、Natural Questions和TriviaQA上进行评估，使用Llama 3.1 8B和GPT-4o-mini，并进行单跳参考实验。在所有六个多跳模型-数据集配置中，越来越严格的符合性目标始终提高了保留声明的响应比例。在95%的目标下，这一比例从95.80%到97.20%不等，而不进行过滤时则为55.60%-76.03%。然而，这一改善具有很强的选择性：在95%的目标下，只有4.41%-31.09%的生成声明被保留，9.70%-51.40%的响应仍然是非空的。这些结果表明，符合性事实控制扩展到多跳RAG，同时表明名义可靠性必须与声明保留和放弃共同解释。
cs.CL / 7 / 2609.38256
Framing the Narrative: Ideological Mimicry in Large Language Models
框架叙事：大型语言模型中的意识形态模仿
Abstract
Large language models (LLMs) are increasingly used to answer questions about politically contentious issues, yet evaluations typically treat a model's stance as a relatively stable property. Real users, however, communicate political signals through their terminology, assumptions, and personal context. We investigate whether such signals produce ideological mimicry: systematic shifts in the political stance expressed by an LLM toward the position conveyed by the interaction. If LLMs adapt their responses to these signals, they risk creating personalised political information environments in which users with opposing views receive systematically different accounts of the same issue, potentially reinforcing existing divisions. We build the Poli-SHIFT dataset and evaluation framework and assess seven open-weight LLMs across ten contentious political topics in the United States, United Kingdom, and Australia, systematically manipulating contested terminology, politically valenced premises, and user information, and eliciting responses in both multiple-choice and open-text formats. Across models, we find robust evidence that prompt framing shapes the political stance of LLM outputs. Changing terminology alone reverses which side of an issue a model supports in 16.9% of matched comparisons. Stated political ideology also systematically shifts responses toward the user's position. These findings show that political stance is not a fixed property of LLMs; the views expressed are conditional on the interaction with the user. As LLMs become increasingly personalised sources of information, such interaction-dependent adaptation could contribute to political information environments that reinforce users' existing perspectives.
Chinese Translation
大型语言模型（LLMs）越来越多地被用于回答关于政治争议问题的提问，但评估通常将模型的立场视为相对稳定的属性。然而，真实用户通过他们的术语、假设和个人背景传达政治信号。我们研究这些信号是否会产生意识形态模仿：即LLM所表达的政治立场系统性地向互动所传达的位置转变。如果LLMs根据这些信号调整其响应，它们可能会创建个性化的政治信息环境，使持有对立观点的用户在同一问题上获得系统性不同的叙述，从而可能加剧现有的分歧。我们构建了Poli-SHIFT数据集和评估框架，并在美国、英国和澳大利亚的十个有争议的政治主题上评估了七个开放权重的LLMs，系统性地操控争议术语、政治倾向前提和用户信息，并在多项选择和开放文本格式中引发响应。在各模型中，我们发现有力证据表明提示框架塑造了LLM输出的政治立场。仅仅改变术语就使得模型在16.9%的匹配比较中反转了支持的立场。所陈述的政治意识形态也系统性地将响应偏向用户的立场。这些发现表明，政治立场并不是LLMs的固定属性；所表达的观点依赖于与用户的互动。随着LLMs越来越成为个性化的信息来源，这种依赖互动的适应可能会导致强化用户现有观点的政治信息环境。
cs.CL / 8 / 2609.38260
ContextAdapt: Evaluating Contextual Adaptation and Value Alignment in LLMs
ContextAdapt：评估大型语言模型中的情境适应性和价值对齐
Abstract
Values such as honesty, autonomy, and confidentiality are often regarded as general principles underpinning AI alignment. However, what it means to act in accordance with these values can depend on the context in which a decision is made. In this paper, we ask whether large language models (LLMs) appropriately adapt the application of a value across professional settings, while remaining consistent when contextual changes do not alter the relevant professional norm. To study this, we introduce ContextAdapt, an evaluation framework covering honesty, autonomy, and confidentiality across medicine, law, finance, and national security. Drawing on primary-source professional and regulatory documents, we construct a value x domain framework and use this to develop scenarios testing both default professional rules and recognised exceptions. We evaluate 12 LLMs on both the actions they recommend and the justifications they provide. In our main experiment, models achieve 95.6% mean appropriateness, although the use of the correct domain-specific justification varies substantially across models, from 25.6% to 76.9%. In a separate factorial experiment, explicitly naming the professional domain and changing the role of the model have limited effect on behaviour. Varying stakes, however, reveals severe but localised failures: in some cases, models alter their responses even though the underlying professional obligation remains unchanged. In particular, perceived severity appears to act as a cue for disclosure across both honesty and confidentiality scenarios. These results show that evaluating value alignment requires us to consider not only whether models follow abstract principles, but whether they apply them appropriately across different contexts.
Chinese Translation
诚实、自主和保密等价值观通常被视为支撑人工智能对齐的一般原则。然而，依据这些价值观行事的含义可能取决于决策所处的情境。在本文中，我们探讨大型语言模型（LLMs）是否能够在专业环境中适当地调整价值观的应用，同时在情境变化未改变相关专业规范时保持一致性。为此，我们引入了ContextAdapt，一个评估框架，涵盖医学、法律、金融和国家安全领域的诚实、自主和保密。基于原始的专业和监管文件，我们构建了一个价值与领域的框架，并利用该框架开发场景，以测试默认的专业规则和公认的例外情况。我们对12个LLMs进行评估，考察它们推荐的行动及其提供的理由。在我们的主要实验中，模型的平均适当性达到了95.6%，尽管正确的领域特定理由的使用在模型之间差异显著，从25.6%到76.9%不等。在一个单独的因子实验中，明确命名专业领域和改变模型角色对行为的影响有限。然而，变化的利益相关者揭示了严重但局部的失败：在某些情况下，模型改变了它们的回应，即使基础的专业义务保持不变。特别是，感知的严重性似乎在诚实和保密场景中作为披露的线索。这些结果表明，评估价值对齐需要我们考虑模型不仅是否遵循抽象原则，还要考虑它们在不同情境中是否适当地应用这些原则。
cs.CL / 9 / 2609.38261
NinaXander: Feasibility and Limits of Composing Frozen Language Models Across Architecture Families via a Shared Latent Space
NinaXander：通过共享潜在空间跨架构系列组合冻结语言模型的可行性与局限性
Abstract
In this paper we propose NinaXander, a series of composed language models obtained by connecting layers of frozen language models from different architecture families with a single trained shared-latent adapter. A composed model runs the first layers of one model, converts the resulting intermediate representation once with the adapter, and then runs the remaining layers of the other model. Once the adapter is trained, several composed models that connect at different layers are obtained without retraining. Using the recurrent RWKV-4-Raven-7B and the Transformer-based Tulu-Pythia-6.9b, abbreviated as RWKV and Pythia, this study examines whether frozen models from different families can be recombined post hoc. The composed models answered multiple-choice questions, and those whose generations we examined produced syntactically well-formed text. The configuration that combines the first 5 layers of Pythia with the remaining 27 layers of RWKV reduced the Transformer key-value (KV) cache by 84.4% with accuracy not significantly different from that of RWKV alone. In multiple-choice accuracy, however, no composed model matched the parent model Pythia, and language-modeling performance decreased sharply on WikiText, a corpus of Wikipedia articles outside the training domain. The correspondence between intermediate representations was also obtained in one favorable case, with a shared tokenizer, the same depth, and the same hidden width, and does not show that the models share a general semantic space.
Chinese Translation
在本文中，我们提出了NinaXander，这是一系列通过连接来自不同架构系列的冻结语言模型的层，并使用一个训练好的共享潜在适配器获得的组合语言模型。组合模型运行一个模型的前几层，将生成的中间表示通过适配器转换一次，然后运行另一个模型的剩余层。一旦适配器训练完成，可以在不重新训练的情况下获得多个在不同层连接的组合模型。本研究使用递归的RWKV-4-Raven-7B和基于Transformer的Tulu-Pythia-6.9b，简称RWKV和Pythia，探讨来自不同系列的冻结模型是否可以事后重新组合。组合模型回答了多项选择题，而我们检查的生成结果产生了语法上良好的文本。将Pythia的前5层与RWKV的剩余27层结合的配置将Transformer的键值（KV）缓存减少了84.4%，且准确性与单独使用RWKV时没有显著差异。然而，在多项选择准确性方面，没有任何组合模型能够匹配母模型Pythia，并且在WikiText（一个超出训练领域的维基百科文章语料库）上的语言建模性能急剧下降。在一个有利的案例中，中间表示之间的对应关系也得到了确认，具有共享的分词器、相同的深度和相同的隐藏宽度，但并未显示模型共享一个通用的语义空间。
cs.CL / 10 / 2609.38274
Which Models Work Well Together? Measuring Heterogeneity for LLM Team Selection
哪些模型能够良好协同工作？测量大语言模型团队选择的异质性
Abstract
The performance ceiling of an LLM team is constrained not only by individual model capabilities, but also by inter-member error resonance and predictive differences. Although heterogeneous teaming is often observed to be effective in practice, existing approaches lack complementarity metrics that are computable, interpretable, and optimizable, leaving team composition to rely on heuristics. We propose a heterogeneity-driven team selection framework that performs offline profiling to characterize individual capability along with two complementary signals: one captures decorrelation in error patterns to reduce co-failures, while the other measures divergence in predictive behavior to capture strategy diversity. We formulate team selection as a standardized quality--complementarity combinatorial objective and apply an efficient greedy search to select a small team from a candidate pool. Experiments across multiple benchmarks demonstrate that our framework consistently outperforms quality-only baselines under controlled candidate pools and team sizes, establishing reusable selection principles for multi-LLM systems.
Chinese Translation
大语言模型（LLM）团队的性能上限不仅受个体模型能力的限制，还受到成员间错误共振和预测差异的影响。尽管在实践中经常观察到异质团队的有效性，但现有方法缺乏可计算、可解释和可优化的互补性度量，导致团队组成依赖于启发式方法。我们提出了一种以异质性为驱动的团队选择框架，该框架通过离线分析来表征个体能力，并结合两个互补信号：一个捕捉错误模式的去相关性以减少共同失败，另一个测量预测行为的分歧以捕捉策略多样性。我们将团队选择形式化为标准化的质量-互补性组合目标，并应用高效的贪婪搜索从候选池中选择一个小团队。在多个基准测试中的实验表明，我们的框架在控制候选池和团队规模的情况下，始终优于仅基于质量的基线，确立了多LLM系统的可重用选择原则。
cs.CL / 11 / 2609.38334
EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making
EVOKE：在可转移决策中引发代理的世界知识
Abstract
Large language models (LLMs) are increasingly deployed as agents for multi-step decision-making, yet transfer poorly to unseen environments. World-model methods address this by training agents to predict future observations, at the cost of additional training and errors that compound when predictions are used for planning. However, for LLM agents operating in digital environments, much of this world knowledge is already internalized during pretraining, which shifts the problem from acquiring it to eliciting it. We argue that typical post-training provides little pressure for such elicitation, since supervision under a single goal at each visited state inadvertently drives policies to rely on superficial contextual habits. We introduce EVOKE, a post-training method that supplies this pressure through goal diversity at fixed states. Motivated by theory showing that an agent competent across diverse goals must encode a world model recoverable from its action preferences, EVOKE holds the environment state and interaction history fixed and ranks the same candidate actions under alternative goals, forcing action preferences to change, so that a policy relying on contextual habits or single-goal correlations cannot order them correctly. This implicitly elicits the policy's pretrained world knowledge to inform decisions. We evaluate EVOKE across diverse tasks in three backbones, demonstrating improved task performance, unseen environment generalization, and data efficiency. We further conduct controlled analyses to better understand what drives these gains. These findings offer a new perspective on eliciting internalized world knowledge for transferable action through direct decision supervision.
Chinese Translation
大型语言模型（LLMs）越来越多地被用作多步骤决策的代理，但在未见环境中的迁移效果较差。世界模型方法通过训练代理预测未来观察来解决这一问题，但这需要额外的训练，并且在预测用于规划时会导致错误的累积。然而，对于在数字环境中操作的LLM代理而言，这些世界知识在预训练过程中已经被内化，这使得问题从获取知识转变为引发知识。我们认为，典型的后训练对这种引发几乎没有压力，因为在每个访问状态下的单一目标监督无意中驱动策略依赖于表面的上下文习惯。我们提出了EVOKE，这是一种通过在固定状态下实现目标多样性来提供这种压力的后训练方法。基于理论的动机表明，一个在多样化目标上具备能力的代理必须编码一个可以从其行动偏好中恢复的世界模型，EVOKE保持环境状态和交互历史不变，并在替代目标下对相同候选行动进行排序，迫使行动偏好发生变化，以至于依赖于上下文习惯或单一目标相关性的策略无法正确排序。这隐含地引发了策略的预训练世界知识，以指导决策。我们在三种基础模型上评估了EVOKE在多样化任务中的表现，展示了任务性能的提升、未见环境的泛化能力和数据效率。我们进一步进行控制分析，以更好地理解这些提升的驱动因素。这些发现为通过直接决策监督引发内化世界知识以实现可转移行动提供了新的视角。
cs.CL / 12 / 2609.38355
Halluscoring 2026: The first shared task on llms hallucination detection and answer verification
Halluscoring 2026：首个关于大型语言模型幻觉检测与答案验证的共享任务
Abstract
We present HalluScoring 2026, a shared task for evaluating hallucination detection and factual verification in Arabic question answering under challenging generalization settings. The shared task is organized into two main tasks, each comprising two subtasks, for a total of four subtasks. Task 1 evaluates binary hallucination detection, considering generalization to unseen questions (Subtask 1.1) and responses generated by unseen LLMs (Subtask 1.2). Task 2 extends the evaluation beyond detection by requiring the systems to additionally identify the correct factual answer from six related candidates, covering Islamic knowledge (Subtask 2.1) and general knowledge (Subtask 2.2). The shared task is based on two Arabic datasets: HalluScore and HalluTruthQA. A total of 13 teams participated in the shared task, 10 of which submitted system description papers. The results of Task 1 demonstrate that hallucination detection remains challenging under distribution shift, with the winning team achieving AUC-ROC test scores of 0.772 and 0.767 for Subtasks 1.1 and 1.2, respectively. For Task 2, the winning team achieved scores of 0.882 and 0.857 in the Islamic and general-knowledge subtasks, respectively, under assisted evaluation.
Chinese Translation
我们介绍了 HalluScoring 2026，这是一个用于评估阿拉伯语问答中幻觉检测和事实验证的共享任务，面临具有挑战性的泛化设置。该共享任务分为两个主要任务，每个任务包含两个子任务，共计四个子任务。任务 1 评估二元幻觉检测，考虑到对未见问题的泛化（子任务 1.1）和由未见大型语言模型（LLMs）生成的响应（子任务 1.2）。任务 2 在检测之外扩展评估，要求系统还需从六个相关候选答案中识别正确的事实答案，涵盖伊斯兰知识（子任务 2.1）和一般知识（子任务 2.2）。该共享任务基于两个阿拉伯语数据集：HalluScore 和 HalluTruthQA。共有 13 支队伍参与了该共享任务，其中 10 支提交了系统描述论文。任务 1 的结果表明，幻觉检测在分布转移下仍然具有挑战性，获胜队伍在子任务 1.1 和 1.2 的 AUC-ROC 测试分数分别为 0.772 和 0.767。对于任务 2，获胜队伍在伊斯兰和一般知识子任务下的得分分别为 0.882 和 0.857，采用辅助评估。
cs.CL / 13 / 2609.38357
Evaluating Language Model Safety Across Long Adversarial Conversations
评估语言模型在长对抗性对话中的安全性
Abstract
Conversational safety evaluations often test language models with a single harmful prompt, even though real-world systems interact with users through long, adaptive conversations. This study examines whether models continue to respond safely when an adversarial user persists across multiple turns. We evaluate three open-weight, instruction-tuned models on two harmful prompts across different conversation lengths and random seeds. In each setting, a second language model acts as a persistent adversarial user, while a safety classifier labels every response as safe or unsafe. Across all model-prompt combinations, first-turn safe-response rates ranged from 85% to 100%. By depth 11, they dropped to 38-61%, and by depth 101, to 15-44%. This decline appeared across models and continued well beyond the short interactions typically used in multi-turn safety evaluations. These results provide proof-of-concept evidence that strong single-turn safety does not necessarily persist during sustained adversarial interaction. They highlight the need for long-horizon evaluations and conversation-level safeguards that account for risk accumulating across turns.
Chinese Translation
对话安全性评估通常仅通过单一有害提示来测试语言模型，尽管现实世界系统通过长时间的适应性对话与用户互动。本研究考察了模型在对抗性用户持续多轮交互时是否仍能安全响应。我们对三种开放权重、经过指令调优的模型在不同对话长度和随机种子下，使用两个有害提示进行了评估。在每个设置中，第二个语言模型充当持续的对抗性用户，而安全分类器则对每个响应进行安全或不安全的标记。在所有模型-提示组合中，第一轮安全响应率范围为85%至100%。到第11轮时，这一比例下降至38-61%，而到第101轮时，下降至15-44%。这种下降在各模型中均有体现，并且持续超出了通常用于多轮安全评估的短期交互。这些结果提供了强有力的概念验证证据，表明强大的单轮安全性并不一定在持续的对抗性互动中保持。它们强调了进行长时间评估和考虑跨轮风险累积的对话级别安全保障的必要性。
cs.CL / 14 / 2609.38374
Doc2LoRA Provides Decodable Representations of Scientific Ideas
Doc2LoRA 提供可解码的科学思想表示
Abstract
Representing scientific papers as points in a space lets us search for similar papers and inquire about how fields relate to one another and drive innovation. Beyond search, the vector space of papers invites generation: mixing papers through simple vector operations creates new points, mirroring combinatorial novelty, the recombination of existing ideas into new ones. However, a mixed point often represents an idea no paper has yet realized, with no papers nearby to identify the idea. We propose representing each paper by a LoRA adapter generated by the Doc-to-LoRA hypernetwork. Every point in the space, including mixtures, thus represents a large language model (LLM) open to questions and instructions in natural language. On papers from the American Physical Society (APS), we instruct the LLM at the average of each subfield to name the field in a few words and obtain labels closer to the official names than the labels of five baselines, as judged by word overlap and a panel of five LLM judges. We also ask the LLMs at points between two APS papers to write an abstract and obtain descriptions shifting from one paper to the other in step with the mixing weight. While Doc-to-LoRA is trained for generation, a small invertible transform makes the embeddings competitive for search, on par with SPECTER2 and EmbeddingGemma and close to SBERT. Because the transform is invertible, every point in the transformed space still maps back to an LLM. The embeddings thus serve both search and generation, enabling researchers to question the idea at any point in the space as a starting point for generating new ideas.
Chinese Translation
将科学论文表示为空间中的点，使我们能够搜索相似论文，并探讨不同领域之间的关系以及如何推动创新。除了搜索，论文的向量空间还邀请生成：通过简单的向量操作混合论文可以创造新的点，反映组合新颖性，即将现有思想重新组合成新的思想。然而，一个混合点往往代表着一个尚未被任何论文实现的思想，周围没有论文可以识别该思想。我们提出通过 Doc-to-LoRA 超网络生成的 LoRA 适配器来表示每篇论文。因此，空间中的每个点，包括混合点，都代表一个大型语言模型（LLM），可以用自然语言进行提问和指令。在美国物理学会（APS）的论文中，我们指示 LLM 在每个子领域的平均值处用几句话命名该领域，并获得的标签在词汇重叠和五位 LLM 评审小组的判断下，比五个基线的标签更接近官方名称。我们还要求 LLM 在两篇 APS 论文之间的点上撰写摘要，并获得描述随着混合权重的变化而从一篇论文转向另一篇论文的结果。尽管 Doc-to-LoRA 是为生成训练的，但一个小的可逆变换使得嵌入在搜索中具有竞争力，与 SPECTER2 和 EmbeddingGemma 相当，并接近 SBERT。由于该变换是可逆的，变换空间中的每个点仍然可以映射回一个 LLM。因此，这些嵌入既可以用于搜索，也可以用于生成，使研究人员能够在空间中的任何点对思想进行提问，作为生成新思想的起点。
cs.CL / 15 / 2609.38406
Evaluating Whether LLMs Can Reliably Connect the DOTs?
评估大型语言模型是否能够可靠地连接信息点？
Abstract
Access to real-world information is often noisy and fragmented. Constructing a coherent narrative from such fragments requires models to reconstruct missing spans within a broader storyline, commonly referred to as text infilling, while preserving consistency with both the local context and the global storyline. Despite using text infilling as a pre-training objective in many Large Language Models (LLMs), their actual performance on real-world narrative infilling remains underexplored. In this paper, we address this gap by introducing a multi-domain benchmark of ~9.2K instances for narrative infilling, constructed by masking one to three sentences across four narrative types: encyclopedic text, commonsense stories, news articles, and visual narratives. Using this benchmark, we evaluate 20 instruction-tuned open-source LLMs ranging from 1.5B to 70B parameters across varying levels of instruction specificity and reasoning guidance. Outputs are assessed using standard automatic metrics and a qualitative framework covering five narrative dimensions. Results show that model scale does not reliably predict infilling quality: Gemma-2-2B achieves the highest qualitative score (4.02/5), outperforming models over ten times larger, including DeepSeek-Qwen-32B (3.77/5, 6.6%) and LLaMA-3.3-70B (3.71/5, 8.3%). We further find that explicit reasoning offers limited benefits as chain-of-thought reasoning yields only a marginal improvement (+0.6%). Additionally, short narratives and domain characteristics emerge as stronger predictors of task difficulty than infill position alone for narrative infilling in current LLMs.
Chinese Translation
获取真实世界信息往往是嘈杂且支离破碎的。从这些碎片中构建连贯的叙事需要模型在更广泛的故事情节中重建缺失的片段，这通常被称为文本填充，同时保持与局部上下文和全局故事情节的一致性。尽管在许多大型语言模型（LLMs）中使用文本填充作为预训练目标，但它们在真实世界叙事填充中的实际表现仍然未得到充分探索。本文通过引入一个多领域基准，填补了这一空白，该基准包含约9.2K个叙事填充实例，构建方式是对四种叙事类型（百科文本、常识故事、新闻文章和视觉叙事）中的一到三句话进行掩码处理。利用该基准，我们评估了20个指令调优的开源LLM，参数范围从1.5B到70B，涵盖不同的指令特异性和推理指导水平。输出结果使用标准自动化指标和涵盖五个叙事维度的定性框架进行评估。结果表明，模型规模并不能可靠地预测填充质量：Gemma-2-2B获得了最高的定性分数（4.02/5），超越了十倍以上的模型，包括DeepSeek-Qwen-32B（3.77/5，6.6%）和LLaMA-3.3-70B（3.71/5，8.3%）。我们进一步发现，明确的推理提供的益处有限，因为链式推理仅带来了边际改善（+0.6%）。此外，短篇叙事和领域特征在当前LLM的叙事填充任务难度预测中，表现出比填充位置更强的预测能力。
cs.CL / 16 / 2609.38427
Policy-Conditioned AI-Use Detection: An Evidentiary Framework for Academic Publishing
政策条件下的人工智能使用检测：学术出版的证据框架
Abstract
Major venues now publish detailed rules about how authors, reviewers, and area chairs may use AI, and those rules differ by role, by task, and by what must be disclosed. AI detection, the instrument usually proposed to enforce them, estimates something else: whether an AI model wrote the text. We argue that this target is misaligned with the decisions conferences and journals face, and propose policy-conditioned AI-use detection, an evidentiary framework for assessing whether a human--AI workflow complied with a stated rule. Policy makes the governing rule an explicit input. Inference reports hypotheses, evidence, calibration regime, and uncertainty in place of verdicts such as "AI detected". Evaluation builds benchmarks from reproducible pipelines that generate compliant and non-compliant workflows, and reports true positive rate at a false positive rate the venue fixes in advance. We work the framework through peer review, where at plausible violation rates a detector at a strong operating point still flags more compliant authors than violating ones. The framework therefore also names what a venue must instrument: structured disclosure, approved-tool routing that respects reviewer confidentiality, and a path by which a finding can be contested. Under this framing a detector is not an authorship classifier but an auditable procedure with an error rate the venue fixes in advance and can defend.
Chinese Translation
主要的学术会议现在发布了关于作者、审稿人和领域主席如何使用人工智能的详细规则，这些规则因角色、任务和必须披露的内容而异。通常提议用于执行这些规则的人工智能检测工具，估计的是文本是否由人工智能模型撰写。我们认为，这一目标与会议和期刊面临的决策不一致，因此提出了政策条件下的人工智能使用检测，这是一种评估人类-人工智能工作流程是否遵循既定规则的证据框架。政策使得治理规则成为一个明确的输入。推理报告假设、证据、校准机制和不确定性，而不是诸如“检测到人工智能”的裁决。评估从可重复的流程中构建基准，这些流程生成合规和不合规的工作流程，并在会议预先设定的假阳性率下报告真实阳性率。我们通过同行评审来应用这一框架，在合理的违规率下，强操作点的检测器仍然标记出更多合规的作者而非违规者。因此，该框架还明确了会议必须实施的内容：结构化披露、尊重审稿人保密性的批准工具路由，以及可以对发现结果提出异议的途径。在这一框架下，检测器不是作者身份分类器，而是一种可审计的程序，其错误率由会议预先设定并能够辩护。
cs.CL / 17 / 2609.38469
The Backdrop Exposes What the World Around an Agent Costs It
背景揭示了代理周围世界的成本
Abstract
Agent benchmarks test agents in worlds that stay still. Deployed agents work in worlds that other people also change. Someone texts the agent to send the money elsewhere or an order confirmation asks it to reply with a door code. We present BACKDROP, which asks how much of an agent's capability in a clean world survives in such a world. BACKDROP takes a task along with the agents execution environment, and plants four everyday hazards in its world, one at a time and all together. The instruction and the correct end state stay the same. Each hazard asks one question. Authority: does a message from another person override the user? Injection: does text planted in a record redirect the agent? Boundary: does a request pull it into an app it was not given? Fault: after a write fails without saying whether it landed, does the agent check before it retries? Across 3,678 variants and 16 models, , the average pass rate falls from 69.5% to 31.3% once all four hazards are present; the strongest models fall furthest (Claude Fable 5.1 from 96.6% to 56.0%). Agents have learned to resist injected text but often follow other unauthorized requests of other people. With all four hazards present, and counting only runs where the planted text reached the agent, agents followed another person's message in 46.4% of runs and injected text in 20.3%. The gap is consistent throughout all 16 models. BACKDROP formalizes these gaps and shows how an agent's score in a task's world is a ceiling on real-world performance.
Chinese Translation
代理基准测试在静止的世界中测试代理。然而，部署的代理在其他人也会改变的世界中工作。有人给代理发送短信要求将钱转移到其他地方，或者订单确认要求它回复门禁密码。我们提出了BACKDROP，探讨在这样的世界中，代理在干净世界中的能力能存活多少。BACKDROP在任务和代理的执行环境中植入四个日常风险，一次一个，或全部一起。指令和正确的最终状态保持不变。每个风险提出一个问题。权威性：来自其他人的消息是否会覆盖用户的指令？注入：记录中植入的文本是否会重定向代理？边界：请求是否会将其拉入未被授权的应用程序？故障：在写入失败且未说明是否成功的情况下，代理是否会在重试之前进行检查？在3678个变体和16个模型中，平均通过率从69.5%下降到31.3%，一旦所有四个风险同时存在；最强的模型下降幅度最大（Claude Fable 5.1 从96.6%降至56.0%）。代理已经学会抵抗注入文本，但通常会遵循其他人的未授权请求。在所有四个风险存在的情况下，仅计算植入文本成功到达代理的运行，代理在46.4%的运行中遵循了其他人的消息，在20.3%的运行中遵循了注入文本。这个差距在所有16个模型中是一致的。BACKDROP形式化了这些差距，并展示了代理在任务世界中的得分是现实世界表现的上限。
cs.CL / 18 / 2609.38480
KlinikeBench: Evaluating Language Models Beyond Diagnostic Accuracy
KlinikeBench：超越诊断准确性的语言模型评估
Abstract
Most clinical benchmarks evaluate language models (LMs) on diagnosis using complete case descriptions. In clinical practice, however, patients present information in different ways, and clinicians must obtain relevant history and determine which examinations are needed before reaching a diagnosis. Diagnostic accuracy alone therefore cannot establish whether an agent gathered essential information or conducted an appropriate clinical assessment. Furthermore, existing benchmarks lack professional clinicians' verification. To address this gap, we introduce KlinikeBench, a benchmark of 333 clinician-authored tasks, each providing an isolated sandbox environment with a virtual patient, clinical tools, and task-specific success criteria. More than 35 clinicians contributed to case authoring and benchmark evaluation. In an empirical study, clinicians gave simulated dialogues higher mean quality ratings than reference conversations, which is adapted from real conversation. In each task, an LM has a fixed budget of turns to communicate with the patient, ask about relevant history, request examinations, follow action constraints, and record a final diagnosis. We score these steps separately as well as together. Across 31 models and seven model families, the best-performing models (e.g., GPT-6-astra and Claude Opus 5) succeed on less than 30% of tasks, even though their diagnosis accuracy reaches 90.7%. Some models benefit from talking with the patient; others diagnose well from a complete chart but perform much worse in conversation. Overall, KlinikeBench provides a testbed for evaluating the full clinical encounter and reveals a substantial gap between diagnostic accuracy and performance in interactive clinical assessment.
Chinese Translation
大多数临床基准通过完整病例描述来评估语言模型（LMs）的诊断能力。然而，在临床实践中，患者以不同的方式呈现信息，临床医生必须获取相关病史并确定需要进行哪些检查，才能做出诊断。因此，仅凭诊断准确性无法判断一个代理是否收集了必要的信息或进行了适当的临床评估。此外，现有基准缺乏专业临床医生的验证。为了解决这一问题，我们引入了KlinikeBench，这是一个包含333个临床医生撰写任务的基准，每个任务提供一个隔离的沙盒环境，配有虚拟患者、临床工具和特定任务的成功标准。超过35位临床医生参与了案例撰写和基准评估。在一项实证研究中，临床医生对模拟对话的平均质量评分高于参考对话，而后者是从真实对话中改编而来的。在每个任务中，语言模型有固定的轮次预算与患者沟通，询问相关病史，请求检查，遵循行动约束，并记录最终诊断。我们分别对这些步骤进行评分，也对其整体表现进行评分。在31个模型和七个模型家族中，表现最佳的模型（如GPT-6-astra和Claude Opus 5）在不到30%的任务中取得成功，尽管它们的诊断准确率达到了90.7%。一些模型在与患者交谈时表现良好；另一些模型在完整病历中诊断准确，但在对话中表现较差。总体而言，KlinikeBench为评估完整的临床接触提供了一个测试平台，并揭示了诊断准确性与互动临床评估表现之间的显著差距。
cs.CL / 19 / 2609.38490
Personalized State-Transition-Aware Memory for Clinical Agents
面向临床代理的个性化状态转移感知记忆
Abstract
Large language model (LLM) agents that reason over clinical records must track changes in a patient's state while preserving the history needed to understand them. Simply accumulating memories leaves it unclear which information still applies, whereas overwriting earlier memories can erase evidence needed to reconstruct treatment history and clinical trajectories. We introduce STAM, a state-transition-aware memory framework that records state changes as new clinical entries arrive. STAM combines semantic retrieval with typed clinical relations to identify affected memories, maintaining current information in Active and superseded or resolved information in History. At read time, a query-dependent gate selectively serves historical memory. Across four longitudinal clinical benchmarks, we evaluate STAM with downstream question answering, direct state-maintenance diagnostics, and comparisons at approximately matched context lengths.
Chinese Translation
大型语言模型（LLM）代理在处理临床记录时必须跟踪患者状态的变化，同时保留理解这些变化所需的历史信息。仅仅积累记忆会导致不清楚哪些信息仍然适用，而覆盖早期记忆则可能抹去重建治疗历史和临床轨迹所需的证据。我们提出了STAM（状态转移感知记忆）框架，它在新的临床条目到达时记录状态变化。STAM结合语义检索和类型化临床关系来识别受影响的记忆，保持当前信息在“活动”（Active）状态，而将过时或已解决的信息存储在“历史”（History）中。在读取时，查询依赖的门控选择性地服务于历史记忆。在四个纵向临床基准测试中，我们通过下游问答、直接状态维护诊断以及在大致匹配的上下文长度下进行比较来评估STAM。
cs.CL / 20 / 2609.38510
DEdit: Iterative Draft Editing for Speculative Decoding
DEdit：用于推测解码的迭代草稿编辑
Abstract
Speculative decoding accelerates autoregressive LLMs by having a lightweight drafter propose tokens that the target model verifies in parallel. Diffusion-based drafters further reduce drafting latency by proposing multiple tokens at once. However, these tokens are predicted independently, so a single early error causes prefix verification to discard the rest of the draft, even when it contains useful downstream predictions. We introduce DEdit, a diffusion-based drafter that can not only draft by conventional parallel unmasking but also iteratively edit its draft through token-to-token predictions. Through editing, later predictions can serve as bidirectional context for repairing earlier errors and extending the accepted prefix. To teach the model to repair errors while preserving correct predictions, we propose ProposalMix, a training scheme that mixes draft predictions with ground-truth tokens based on first-pass confidence during training. Across seven benchmarks on Qwen3-4B and Qwen3-8B, DEdit achieves the highest macro-average token acceptance and speedup among the evaluated drafters, reaching macro-average speedups of $5.72\times$ and $5.97\times$ over autoregressive generation under greedy decoding, respectively. Further analysis shows that acceptance improves with more editing passes and wider drafting windows, and that ProposalMix halves harmful edits that shorten the accepted prefix. Moreover, restricting the editor to causal attention lowers acceptance, especially on highly predictable outputs, indicating that future context is a key source of these gains.
Chinese Translation
推测解码通过让轻量级草稿生成器提出令牌并由目标模型并行验证，从而加速自回归大型语言模型（LLMs）。基于扩散的草稿生成器通过一次性提出多个令牌进一步减少了草稿延迟。然而，这些令牌是独立预测的，因此单个早期错误会导致前缀验证丢弃其余草稿，即使其中包含有用的下游预测。我们提出了DEdit，一种基于扩散的草稿生成器，它不仅可以通过传统的并行解掩码进行草稿生成，还可以通过令牌到令牌的预测迭代编辑其草稿。通过编辑，后续预测可以作为双向上下文来修复早期错误并扩展接受的前缀。为了教会模型在保留正确预测的同时修复错误，我们提出了ProposalMix，这是一种训练方案，根据训练期间的首次置信度将草稿预测与真实令牌混合。在Qwen3-4B和Qwen3-8B的七个基准测试中，DEdit在评估的草稿生成器中实现了最高的宏平均令牌接受率和加速，分别达到了自回归生成在贪婪解码下的宏平均加速比$5.72 imes$和$5.97 imes$。进一步分析显示，接受率随着更多编辑轮次和更宽的草稿窗口而提高，而ProposalMix将缩短接受前缀的有害编辑减少了一半。此外，将编辑器限制为因果注意力降低了接受率，尤其是在高度可预测的输出上，表明未来上下文是这些增益的关键来源。
cs.CL / 21 / 2609.38530
Shifting Mechanisms: How Positional Encoding Choice Shapes In-Context Retrieval
转变机制：位置编码选择如何影响上下文检索
Abstract
Language models increasingly use architectures that vary attention span and positional encoding across layers, such as applying RoPE with sliding-window attention and NoPE with global attention (SWA NoPE). However, how these choices shape in-context retrieval remains unclear. To study this question, we take a mechanistic view, tracing how positional encoding (PE) choice shapes the internal mechanisms models use for in-context retrieval. Across 22 open-weight models spanning eight families, we find that standard RoPE models rely primarily on positional retrieval, while PE hybrids shift toward semantic retrieval. We further show on a controlled pre-training ablation that confining positional encoding to local layers produces this semantic shift, degrading representations of positional information. Finally, we show that the reported long-context gains of PE hybrids mask a retrieval trade-off: SWA NoPE improves over RoPE on multiple-target retrieval and QA, but degrades when distinguishing competing keys. We show that these behavioral differences better track the mechanism shift from positional toward semantic mechanisms than a uniform improvement in long-context retrieval.
Chinese Translation
语言模型越来越多地采用在层间变化注意力范围和位置编码的架构，例如在滑动窗口注意力中应用 RoPE，以及在全局注意力中应用 NoPE（SWA NoPE）。然而，这些选择如何影响上下文检索仍不清楚。为了研究这个问题，我们采取了机械视角，追踪位置编码（PE）选择如何塑造模型用于上下文检索的内部机制。在涵盖八个家族的22个开放权重模型中，我们发现标准的 RoPE 模型主要依赖于位置检索，而 PE 混合模型则向语义检索转变。我们进一步通过控制预训练消融实验表明，将位置编码限制在局部层会导致这种语义转变，从而降低位置信息的表示能力。最后，我们展示了 PE 混合模型报告的长上下文收益掩盖了检索权衡：SWA NoPE 在多目标检索和问答任务上优于 RoPE，但在区分竞争键时表现下降。我们表明，这些行为差异比在长上下文检索中的均匀改善更能反映从位置机制向语义机制的转变。
cs.CL / 22 / 2609.38543
MedKIT: Evaluating Knowledge Integration and Generalization in Large Language Models
MedKIT：评估大型语言模型中的知识整合与泛化
Abstract
Constantly evolving real-world knowledge necessitates models to be updated continuously. Especially in medicine, as clinical evidence changes over time, outdated knowledge can pose safety risks. Existing evaluations of knowledge integration focus on factual recall, offering limited insight into whether newly integrated knowledge is actually usable. Our benchmark MedKIT (Medical Knowledge Integration and Transfer) provides a granular evaluation of how models integrate and apply knowledge under realistic sequences of clinical updates. Each instance corresponds to a factual update derived from clinical evidence, paired with targeted probes that assess transfer across lexical variation, relational transformations, compositional reasoning, and open-ended operationalization, as well as locality tests for knowledge preservation. Using MedKIT, we conduct a large-scale empirical study of 12 knowledge integration strategies across 5 diverse models, including both general-purpose and medical LLMs. Our results reveal a consistent gap between recall and usable knowledge: while most methods achieve strong gains on the original update task and under lexical variation, relational generalization is limited, and no method yields meaningful improvements on compositional or operational tasks. These findings highlight a fundamental challenge in knowledge integration and position MedKIT as a testbed for developing methods that make newly integrated knowledge more consistently usable across tasks and contexts.
Chinese Translation
不断发展的现实世界知识要求模型持续更新。尤其在医学领域，随着临床证据随时间变化，过时的知识可能带来安全风险。现有的知识整合评估主要集中在事实回忆上，提供的见解有限，无法判断新整合的知识是否真正可用。我们的基准测试MedKIT（医学知识整合与转移）提供了对模型在现实临床更新序列下如何整合和应用知识的细致评估。每个实例对应于一个基于临床证据的事实更新，并配有针对性的探测，评估在词汇变异、关系转化、组合推理和开放式操作化方面的转移，以及知识保留的局部性测试。通过使用MedKIT，我们对5个不同模型（包括通用和医学大型语言模型）进行了12种知识整合策略的大规模实证研究。我们的结果揭示了回忆与可用知识之间的一致差距：尽管大多数方法在原始更新任务和词汇变异下取得了显著进展，但关系泛化有限，且没有任何方法在组合或操作任务上取得有意义的改进。这些发现突显了知识整合中的基本挑战，并将MedKIT定位为开发使新整合知识在任务和上下文中更一致可用的方法的试验平台。
cs.CL / 23 / 2609.38593
Prompt2Skill: Unsupervised Skill Optimization From Natural Language Instructions
Prompt2Skill：基于自然语言指令的无监督技能优化
Abstract
Skills are external artifacts that Large Language Models (LLMs) consume at inference time to improve their performance on specialized domains by incorporating relevant procedural and domain knowledge. Expert-authored skills are expensive to produce, and the resulting artifacts are not optimized for the specific model that consumes them, whose failure modes can vary with version, scale and training. In addition, emerging tasks may fall outside the scope of existing skill libraries, creating a need to develop new skills before curated training data become available. Recent works have explored automated skill optimization through reflection, but they require a curated, in-distribution training set, which users might not always have. To address these limitations, we present Prompt2Skill, a framework that builds skills from natural-language task description alone. From the prompt, the system derives a task specification, discovers or synthesizes datasets, and refines the skill in a closed loop of reflective editing. Across four domains spanning question answering, reading comprehension, spreadsheet manipulation, and mathematical reasoning, Prompt2Skill consistently outperforms the direct prompting baseline, achieving an average improvement of 10.8 across open-source and frontier models.
Chinese Translation
技能是大型语言模型（LLMs）在推理时所需的外部工件，通过整合相关的程序性和领域知识来提高其在专业领域的表现。专家编写的技能制作成本高昂，且所产生的工件并未针对特定的使用模型进行优化，而这些模型的失败模式可能因版本、规模和训练而异。此外，新兴任务可能超出现有技能库的范围，这就需要在经过精心策划的训练数据可用之前开发新的技能。近期的研究探讨了通过反思进行自动化技能优化的方法，但这些方法需要一个经过策划的、符合分布的训练集，而用户并不总是具备这样的数据集。为了解决这些局限性，我们提出了Prompt2Skill，一个仅通过自然语言任务描述构建技能的框架。系统从提示中推导出任务规范，发现或合成数据集，并在反思编辑的闭环中精炼技能。在涵盖问答、阅读理解、电子表格操作和数学推理的四个领域中，Prompt2Skill始终优于直接提示基线，平均提升达到10.8，适用于开源和前沿模型。
cs.CL / 24 / 2609.38604
Beyond Oracle Communication: Benchmarking Interactive Intent Alignment Under Miscommunication and Evolving User Intent
超越神谕通信：在误通信和用户意图演变下的交互意图对齐基准测试
Abstract
Modern LLM agents increasingly tackle complex tasks through interactive, long-horizon exchanges with users, while existing benchmarks generally assume that users always accurately and sufficiently communicate a fixed intent. However, this oracle communication assumption rarely holds in practice: users may miscommunicate, change their goals, and run out of patience. We define this task setting as Interactive Intent Alignment, where agents must recover and continuously track the user's current intent despite imperfect communication and evolving goals. To study this setting, we introduce Drift-Bench++, a principled benchmark construction pipeline for verified executable tasks with controlled misalignment and intent shifts, along with an interaction protocol featuring finite patience, diverse simulated users, and silent interaction-conditioned shifts. We further develop GRIP, a comprehensive evaluation protocol covering task grounding, user realism, inquiry effectiveness, and adaptation to evolving intent. Across diverse environments, models, and interaction conditions, stronger interaction consistently helps but remains far from oracle performance; Validation on deployed ProdAgent sessions further shows that the modeled failures are prevalent and consequential in deployment. By providing a unified, executable benchmark for interactive intent alignment, Drift-Bench++ offers a foundation for evaluating and advancing agents under realistic communication and evolving intent.
Chinese Translation
现代大规模语言模型（LLM）代理越来越多地通过与用户的交互式长时间交流来处理复杂任务，而现有基准通常假设用户总是准确且充分地传达固定意图。然而，这一神谕通信假设在实践中很少成立：用户可能会误传信息、改变目标并失去耐心。我们将这一任务设置定义为交互意图对齐，其中代理必须在不完美的通信和不断演变的目标下恢复并持续跟踪用户的当前意图。为了研究这一设置，我们引入了 Drift-Bench++，这是一个经过验证的可执行任务的基准构建管道，具有可控的错位和意图变化，以及一个包含有限耐心、多样化模拟用户和静默交互条件变化的交互协议。我们进一步开发了 GRIP，这是一个全面的评估协议，涵盖任务基础、用户真实感、询问有效性和对不断演变意图的适应性。在多样化的环境、模型和交互条件下，更强的交互始终有助于提升性能，但仍远未达到神谕性能；在已部署的 ProdAgent 会话中的验证进一步表明，模型化的失败在实际应用中普遍存在且后果严重。通过提供一个统一的、可执行的交互意图对齐基准，Drift-Bench++ 为在现实通信和不断演变的意图下评估和推进代理提供了基础。
cs.CL / 25 / 2609.38612
StreamDecisionBench: Evaluating Decisions in Force on Evolving Language Streams
StreamDecisionBench：评估不断演变的语言流中的决策
Abstract
As natural language drives more applications, language models increasingly run inside programs as decision components: the program sends them the current state and acts on the returned decision until a newer one arrives. When evidence changes during inference, a decision correct for its own state can stay in force after that state has passed, as when a call recorder keeps running after a customer starts reading out a card number; untimed (offline) accuracy counts such an error as correct. We introduce StreamDecisionBench (SDB), which evaluates the decision in force at every instant and attributes every erroneous instant to judgment, latency or both. Its scenarios stream evidence in four application families, with reference decisions computed from public rules by executable code. We summarize in-force accuracy across update intervals of 1-5 s by its normalized area under the curve on a logarithmic time axis, giving equal weight to equal multiplicative ranges. Across six settings of four hosted models, this score stays within 2.9 points of the scenario-wise product of untimed accuracy and an oracle's integrated timing score. Reasoning improves judgment, but at low effort latency costs Luna and Terra, two GPT models we also evaluate without reasoning, 38.8 and 42.8 points relative to untimed accuracy; a faster component with weaker judgment attains a similar integrated score to Terra without reasoning. The aggregate and family curves show where these tradeoffs change, making the evaluation's time-scale dependence visible.
Chinese Translation
随着自然语言驱动更多应用，语言模型越来越多地作为决策组件嵌入程序中：程序将当前状态发送给它，并根据返回的决策进行操作，直到接收到更新的决策。当推理过程中证据发生变化时，适用于其自身状态的决策可能在该状态结束后仍然有效，就像在客户开始读出卡号后，通话录音机仍然在运行一样；未计时（离线）准确性将此类错误视为正确。我们引入了StreamDecisionBench（SDB），它在每个时刻评估有效决策，并将每个错误时刻归因于判断、延迟或两者。其场景在四个应用类别中流式传输证据，参考决策由可执行代码根据公共规则计算得出。我们通过其在对数时间轴上的归一化曲线下的面积总结了1-5秒更新间隔内的有效准确性，给予相等的乘法范围相等的权重。在四个托管模型的六个设置中，该分数与未计时准确性和一个预言者的综合计时分数的场景产品保持在2.9分以内。推理改善了判断，但在低努力延迟成本下，Luna和Terra这两个我们也在没有推理的情况下评估的GPT模型相对未计时准确性分别降低了38.8和42.8分；一个具有较弱判断的更快组件在没有推理的情况下获得了与Terra相似的综合分数。聚合和类别曲线显示了这些权衡变化的地方，使评估的时间尺度依赖性变得可见。
cs.CL / 26 / 2609.38627
Marking Contour Tones in Yor\`{u}b\'{a}
约鲁巴语中的轮廓声调标记
Abstract
Yor\`ub\'a is a tonal language in which contour tones pose persistent orthographic challenges. These are especially notable for personal names and lexical items whose conventional spellings avoid vowel lengthening that would otherwise provide a host syllable for the second tone. A particular concern is a class of names in which the conventional spelling does not just omit tonal information but inverts the meaning of said name, sometimes asserting the opposite of what the name intends. This paper describes the problem, illustrates the inadequacy of current solutions, and proposes the adoption of the caron and circumflex marks. These are symbols with precedent in Yor\`ub\'a phonological scholarship since Olmsted (1951), used as orthographic conventions on single vowels to encode rising and falling contour tones, making them accessible for the first time through standard keyboard input and computational text processing. The proposal is supported by an implementation in the WriteYoruba keyboard and the TTSYoruba speech synthesizer, whose architecture and listener evaluation are reported separately (Tubosun et al., 2026).
Chinese Translation
约鲁巴语是一种声调语言，其中轮廓声调带来了持续的正字法挑战。这些挑战在个人名字和词汇项目中尤为显著，因为它们的传统拼写避免了元音延长，而元音延长本可以为第二声提供一个主音节。一个特别关注的问题是某类名字，其传统拼写不仅省略了声调信息，还颠倒了该名字的意义，有时甚至与名字的本意相反。本文描述了这一问题，说明了当前解决方案的不足，并提出采用抑扬符号（caron）和抑扬音符（circumflex）作为解决方案。这些符号在约鲁巴语音学研究中已有先例，自Olmsted（1951）以来被用作单个元音的正字法约定，以编码上升和下降的轮廓声调，使其首次通过标准键盘输入和计算文本处理变得可访问。该提案得到了WriteYoruba键盘和TTSYoruba语音合成器的实现支持，其架构和听众评估将单独报告（Tubosun et al., 2026）。
cs.CL / 27 / 2609.38630
Strong Multilingual Privacy Tagging at Encoder Speed
以编码器速度实现强大的多语言隐私标记
Abstract
Privacy redaction must remove personal information while preserving relationships expressed in text. We develop a multilingual named-entity tagger with fine-grained distinctions supporting varied redaction policies and methods for cheaply learning additional distinctions. We fine-tune a multilingual encoder with an affine span-tagging head on frontier-model annotations in 35 languages, replay mapped human gold with coverage-aware masking so unannotated types are not treated as negatives, and repair subword boundaries with a learned +/-1-character adjustment. On 1,283 human-gold test segments in seven languages, best measured redaction F1 is 88.8, against 69.1 for published GLiNER2 with 11 unrepresentable types excluded from its task (68.8 without that exemption), 67.8 for GLiNER2 adapted to the new training data, 57.3 for Microsoft Presidio and 35.8 for the best published OpenAI Privacy Filter fine-tune. Adding about 50,000 annotated training sentences and increasing human-gold replay improves exact typed-span F1 from 74.5 to 76.3 on Ont3, our 31-type frontier-annotated NER evaluation of 1,201 development segments. Mapped-gold replay alone raises human-gold F1 by ten points without loss on frontier-annotated text; boundary adjustment adds 1.7 exact typed-span F1 points on Ont3. Local LLMs fitting on a single 96-GB GPU underperformed as prompted annotators and frozen encoders, with encoding 30-95 times slower than XLM-R inference and prompted annotation roughly 180-1,100 times slower in the evaluated configurations. The encoder architecture delivers 4.9 times GLiNER2's CPU throughput. We release code, prompts and training recipes, with data-acquisition scripts and source links.
Chinese Translation
隐私编辑必须在去除个人信息的同时保留文本中表达的关系。我们开发了一种多语言命名实体标记器，具有细粒度的区分，支持多种编辑政策和廉价学习额外区分的方法。我们在35种语言的前沿模型注释上微调了一个多语言编码器，并在覆盖感知掩蔽下重放映射的人类金标准，以确保未注释的类型不被视为负例，并通过学习的±1字符调整修复子词边界。在七种语言的1,283个人工金标准测试片段中，最佳测得的编辑F1值为88.8，而已发布的GLiNER2在排除11种不可表示类型的情况下为69.1（不排除时为68.8），适应新训练数据的GLiNER2为67.8，Microsoft Presidio为57.3，最佳已发布的OpenAI隐私过滤器微调为35.8。增加约50,000个注释训练句子并提高人类金标准重放使得在我们的31类型前沿注释命名实体识别（NER）评估Ont3上的精确类型跨度F1从74.5提高到76.3，评估包含1,201个开发片段。仅映射金标准重放使人类金标准F1提高十个点，而在前沿注释文本上没有损失；边界调整在Ont3上增加了1.7个精确类型跨度F1点。在单个96GB GPU上适配的本地大语言模型（LLMs）作为提示注释者和冻结编码器表现不佳，其编码速度比XLM-R推理慢30-95倍，而在评估配置中，提示注释速度大约慢180-1,100倍。编码器架构提供了GLiNER2 CPU吞吐量的4.9倍。我们发布了代码、提示和训练配方，以及数据获取脚本和源链接。
cs.CL / 28 / 2609.38660
Breaking Babel: A Self-Evolving Multi-Agent System for Long-Form Subtitle Translation
打破巴别塔：一个自我进化的多智能体系统用于长篇字幕翻译
Abstract
Long-form subtitle translation requires reasoning over discourse and cultural context spanning episodes or entire series, while maintaining consistent terminology and style. Existing single-LLM methods are largely sentence-level, and multi-agent systems often use static workflows that do not adapt to scene complexity or production context. We propose SMART, a Self-evolving Multi-Agent system for long-foRm subtitle Translation. During test-time training, SMART builds persistent series-level memory and translates a subset of sentences through a dynamic router and Mixture-of-Agents layer with tools for terminology verification, subtitle constraint validation, and contextual retrieval. A judge-refiner loop scores candidates and uses textual critiques to update agent prompts and routing policies without retraining the underlying LLMs. During test-time inference, the evolved configuration translates the remaining series. We also introduce Subtitle Arena, covering 14 genres, 2--198 episodes per series, production years 1959--2023, and 15 target locales, together with SubMQM, a subtitle-adapted MQM framework with seven dimensions and 19 error categories. SMART achieves the best overall MQM score in all 15 Subtitle Arena directions, reducing average penalty by 6.9% over the strongest competing agent system. On the public MuSC benchmark, SMART obtains the best model result across all four language pairs and also achieves the best human-evaluation result, with an overall score of 4.50/5.
Chinese Translation
长篇字幕翻译需要对跨剧集或整个系列的语篇和文化背景进行推理，同时保持术语和风格的一致性。现有的单一大语言模型（single-LLM）方法主要集中在句子级别，而多智能体系统通常使用静态工作流程，无法适应场景复杂性或制作背景。我们提出了SMART，一个用于长篇字幕翻译的自我进化多智能体系统。在测试时训练过程中，SMART建立持久的系列级记忆，并通过动态路由器和混合智能体层翻译一部分句子，配备术语验证、字幕约束验证和上下文检索的工具。评估-优化循环对候选翻译进行评分，并利用文本批评更新智能体提示和路由策略，而无需重新训练底层的大语言模型。在测试时推理阶段，进化后的配置翻译剩余的系列。我们还介绍了字幕竞技场（Subtitle Arena），涵盖14个类型、每个系列2至198集、制作年份为1959至2023年，以及15个目标地区，并推出了SubMQM，一个适应字幕的MQM框架，包含七个维度和19个错误类别。SMART在所有15个字幕竞技场方向上实现了最佳的整体MQM得分，相较于最强竞争智能体系统平均减少了6.9%的惩罚。在公共MuSC基准测试中，SMART在所有四个语言对中获得了最佳模型结果，并且在人工评估中也取得了最佳结果，总体得分为4.50/5。
cs.CL / 29 / 2609.38718
MetaSteer: Context-Conditioned, nonlinear Steering via Attention-Projection Adaptation
MetaSteer：基于上下文条件的非线性引导通过注意力投影适应
Abstract
Steering large language models typically relies on linear, context-independent interventions in activation space, an assumption that recent work has challenged and that can induce an information bottleneck when a fixed representation must encode many behavioral distinctions. We introduce MetaSteer, a method that learns nonlinear interventions with context-dependent effects and applies them to attention projection matrices, producing activation effects that vary with the input context by construction and requiring no linear concept-geometry assumption. Framed as preference-based optimization, MetaSteer is trained once on a pooled preference corpus and transferred zero-shot to unseen concepts and out-of-distribution contexts. We find that, despite using low-rank adapters, MetaSteer induces structured, context-dependent changes in hidden-state trajectories while partially preserving aspects of their local trajectory dynamics, including velocity and curvature. We evaluate MetaSteer on three controlled text-generation benchmarks and three agentic settings across multiple model families and scales. MetaSteer matches or outperforms strong task-specific steering baselines on most aggregate comparisons in the zero-shot regime. Across the evaluated settings, stronger text-generation steering is associated with stronger agentic steering performance. We further discuss geometric trajectory effects, capability retention, and safety considerations raised by transferable steering.
Chinese Translation
对大型语言模型的引导通常依赖于在激活空间中进行线性、与上下文无关的干预，这一假设在近期的研究中受到挑战，并且在固定表示必须编码许多行为区分时可能导致信息瓶颈。我们提出了MetaSteer，这是一种学习具有上下文依赖效应的非线性干预的方法，并将其应用于注意力投影矩阵，从而产生与输入上下文变化的激活效应，并且不需要线性概念几何假设。MetaSteer被框架化为基于偏好的优化，首次在汇总的偏好语料库上进行训练，并在零样本情况下转移到未见概念和分布外上下文。我们发现，尽管使用低秩适配器，MetaSteer仍能引发结构化的、上下文依赖的隐藏状态轨迹变化，同时部分保留其局部轨迹动态的某些方面，包括速度和曲率。我们在三个受控文本生成基准和三个代理设置中评估MetaSteer，涵盖多个模型系列和规模。在大多数汇总比较中，MetaSteer在零样本情况下与强大的任务特定引导基线相匹配或超越。在评估的设置中，更强的文本生成引导与更强的代理引导性能相关。我们进一步讨论了几何轨迹效应、能力保留以及可转移引导所引发的安全性考虑。
cs.CL / 30 / 2609.38792
Training LLM Judges from Language Feedback via Position-Selective Self-Distillation
通过位置选择自蒸馏从语言反馈中训练大型语言模型评判者
Abstract
We study training LLM judges from natural language feedback, especially for subjective tasks where the verdict depends strongly on which evaluation criteria the judge invokes and how it weighs them. The dominant approach, outcome-supervised RL (e.g., GRPO), credits every token in the rollout with a single scalar determined only by the accuracy of the final verdict, providing no separate credit at the criterion-choice tokens and ignoring the rich language feedback (e.g., preference rationales) that naturally accompanies preference labels. Self-Distillation (SD) is one natural way to use this language feedback: the same model, conditioned on this feedback, acts as a teacher providing dense, position-level supervision. However, not all positions carry equally useful signal. Using the per-position entropy shift between teacher and student, we identify two regimes: context sharpening, where the teacher concentrates probability on a particular feedback-aligned criterion expression, and context spreading, where the teacher distributes probability across multiple feedback-aligned alternatives. We interpret these patterns as follows: sharpening encourages memorization of a particular criterion expression, whereas spreading promotes semantic understanding by preserving these alternatives. Motivated by this asymmetry, we introduce position masking based on the entropy shift that retains the lower tail of the entropy-shift distribution. Experiments show that masking higher-entropy-shift positions improves out-of-distribution generalization over naive SD. The resulting self-distilled judges outperform judges trained with outcome-supervised RL by 2-9 percentage points on the evaluated subjective subcategories, while remaining competitive on objective ones.
Chinese Translation
我们研究了从自然语言反馈中训练大型语言模型（LLM）评判者，特别是在主观任务中，裁决结果在很大程度上依赖于评判者所调用的评估标准及其权重。当前主流方法是基于结果的强化学习（例如，GRPO），该方法仅根据最终裁决的准确性为回滚中的每个标记分配一个单一的标量，未能对标准选择标记提供单独的信用，并忽视了与偏好标签自然伴随的丰富语言反馈（例如，偏好理由）。自蒸馏（Self-Distillation, SD）是利用这种语言反馈的一种自然方式：同一模型在该反馈的条件下充当教师，提供密集的、基于位置的监督。然而，并非所有位置都携带同等有用的信号。通过使用教师和学生之间的每个位置的熵变化，我们识别出两种模式：上下文聚焦（context sharpening），教师将概率集中在特定的与反馈一致的标准表达上；上下文扩散（context spreading），教师在多个与反馈一致的替代方案之间分配概率。我们将这些模式解释为：聚焦鼓励对特定标准表达的记忆，而扩散则通过保留这些替代方案促进语义理解。基于这种不对称性，我们引入了基于熵变化的位置信息掩蔽，保留熵变化分布的低尾部分。实验表明，掩蔽高熵变化位置能够提高在分布外的泛化能力，相较于简单的自蒸馏，结果显示自蒸馏的评判者在评估的主观子类别上比基于结果监督的强化学习训练的评判者提高了2-9个百分点，同时在客观子类别上保持竞争力。
cs.CL / 31 / 2609.38795
Recovering Off-Policy Supervision for Speculative Decoding
恢复投机解码的离线策略监督
Abstract
Block drafters for speculative decoding are commonly trained on corpora written by external models, where a single off-policy token invalidates supervision for all subsequent slots in a block. Existing approaches discard these divergent slots, resulting in severe supervision loss. To resolve this problem while preserving the training corpus, we propose a rollout-based training framework that recovers full supervision through two complementary components. The first component, Anchor-Label Relabelling (ALR), replaces corpus labels with distributions from greedy target rollouts, restoring valid supervision across all predicted slots. The second component, In-Rollout Anchors (IRA), places draft blocks directly inside these rollouts to expose the drafter to target-generated context, reusing precomputed rollout features at no additional target cost. Across fixed vision-language and text corpora, our framework increases greedy accepted length by up to 36.5% over DFlash and consistently outperforms erasing baselines. Notably, a single epoch of our method surpasses the best erase schedules. After three epochs, it matches the acceptance length of training on target-regenerated responses. These results show that our framework provides an effective and compute-efficient approach for training speculative drafters on fixed corpora without modifying the original text. Code is available at https://github.com/js-lee-AI/ALR-IRA.
Chinese Translation
投机解码的区块草拟器通常在由外部模型编写的语料库上进行训练，其中单个离线策略的标记会使区块中所有后续位置的监督失效。现有方法会丢弃这些不同的槽位，导致严重的监督损失。为了解决这个问题，同时保留训练语料库，我们提出了一种基于回滚的训练框架，通过两个互补的组件恢复完整的监督。第一个组件，锚标签重标定（Anchor-Label Relabelling, ALR），用贪婪目标回滚的分布替换语料库标签，从而恢复所有预测槽位的有效监督。第二个组件，在回滚锚点（In-Rollout Anchors, IRA）中，直接将草拟区块放置在这些回滚中，以使草拟器接触到目标生成的上下文，重新利用预计算的回滚特征而无需额外的目标成本。在固定的视觉-语言和文本语料库上，我们的框架使贪婪接受长度比DFlash提高了最多36.5%，并且始终优于擦除基线。值得注意的是，我们方法的单个训练周期超越了最佳擦除计划。经过三个周期后，其接受长度与在目标再生响应上训练的结果相匹配。这些结果表明，我们的框架为在固定语料库上训练投机草拟器提供了一种有效且计算高效的方法，而无需修改原始文本。代码可在 https://github.com/js-lee-AI/ALR-IRA 获取。
cs.CL / 32 / 2609.38799
Overlap, Unique and Conflict: Can LLMs Extract What They Can Recognize?
重叠、独特与冲突：大型语言模型能否提取它们能够识别的信息？
Abstract
Understanding multi-perspective alternative narratives requires identifying how their information agrees, conflicts, or differs across sources. Existing work on cross-text relations largely focuses on categorizing relations between predefined text pairs, such as entailment or contradiction, rather than directly extracting such information from full narratives. To address this gap, we introduce Overlap-Unique-Conflict (OUC) extraction, a cross-narrative task that extracts all overlapping, conflicting, and unique clauses from two narratives. To support this study, we construct a benchmark of approximately 22K narrative pairs and 140K OUC instances spanning factual, argumentative, and political discourse. Evaluating 14 open-source LLMs (0.6B-35B), we find that unique information is far easier to extract than overlap and conflict: the strongest model, Gemma-4-31B, reaches only 61.13% F1-score on overlap and 48.58% on conflict, against more than 75% on unique. Further diagnostic analysis reveals that this difficulty does not stem from relation recognition alone, but rather from a failure to pair and extract the corresponding clauses from full narratives, especially in smaller models. Nevertheless, learning these extractions with task-specific supervision narrows the gap considerably: a fine-tuned Qwen-3-8B gains 15-28% absolute over its baseline and surpasses models roughly four times its size (e.g., Qwen-3.6-35B) on several tasks. Even so, overlap and conflict remain well below satisfactory, leaving cross-narrative clause extraction an open challenge.
Chinese Translation
理解多角度的替代叙事需要识别其信息在不同来源之间的相互一致、冲突或差异。现有的跨文本关系研究主要集中于对预定义文本对（如蕴涵或矛盾）之间关系的分类，而不是直接从完整叙事中提取此类信息。为了解决这一空白，我们引入了重叠-独特-冲突（Overlap-Unique-Conflict, OUC）提取，这是一项跨叙事任务，旨在从两个叙事中提取所有重叠、冲突和独特的子句。为了支持本研究，我们构建了一个基准数据集，包含约22,000对叙事和140,000个OUC实例，涵盖事实、论证和政治话语。对14个开源大型语言模型（0.6B-35B）的评估发现，独特信息的提取远比重叠和冲突容易：最强的模型Gemma-4-31B在重叠上的F1分数仅为61.13%，在冲突上的F1分数为48.58%，而在独特信息上则超过75%。进一步的诊断分析表明，这一困难并不仅仅源于关系识别，而是由于未能从完整叙事中配对和提取相应的子句，尤其是在较小的模型中。然而，通过任务特定的监督学习这些提取可以显著缩小差距：经过微调的Qwen-3-8B在多个任务上相较于基线提高了15-28%的绝对值，并超越了其规模约为四倍的模型（例如，Qwen-3.6-35B）。尽管如此，重叠和冲突的提取仍然远未令人满意，使得跨叙事子句提取仍然是一个未解的挑战。
cs.CL / 33 / 2609.38802
Uncovering Uncontrolled Repetition through Residual Stream Dynamics
通过残差流动态揭示无控制重复
Abstract
Uncontrolled repetition can prolong autoregressive generation in large language models (LLMs) and enable resource consumption attacks. Prior analyses of repetitive generation have identified strongly activated features in intermediate and late layers. However, how uncontrolled repetition activity emerges and develops before becoming prominent in these layers remains insufficiently understood. In this paper, we investigate this question primarily in large vision-language models (LVLMs), which support a richer set of uncontrolled repetitions through both visual and textual inputs. We propose Tokenwise Residual Comparison (TRC), a method that identifies and localizes anomalies associated with repetition from residual dynamics during generation. TRC compares attention and multilayer perceptron writes to the residual stream across generated tokens to identify patterns associated with repetition. It then selectively suppresses coordinates in the residual stream at the identified layer. Experiments show that TRC effectively mitigates uncontrolled repetition, reducing loop rates by 57\% on average. Our analysis further shows that repetition semantics emerge in shallow layers and propagate through the residual stream, disrupting normal representations. TRC also generalizes to large language models (LLMs) and large reasoning models (LRMs), where it consistently captures analogous repetition dynamics and achieves effective mitigation. Our work broadens the study of repetitive generation from its prominent internal representations to earlier opportunities for intervention, providing insights for mitigating resource consumption attacks.
Chinese Translation
无控制重复可能会延长大型语言模型（LLMs）的自回归生成过程，并导致资源消耗攻击。先前对重复生成的分析已识别出在中间层和后期层中强烈激活的特征。然而，如何在这些层中变得突出之前，无控制重复活动是如何出现和发展的仍然不够清楚。本文主要在大型视觉语言模型（LVLMs）中研究这一问题，这些模型通过视觉和文本输入支持更丰富的无控制重复。我们提出了基于标记的残差比较（Tokenwise Residual Comparison, TRC）方法，该方法识别和定位与生成过程中的重复相关的残差动态异常。TRC比较在生成的标记中，注意力和多层感知器写入与残差流的关系，以识别与重复相关的模式。然后，它在识别的层中选择性地抑制残差流中的坐标。实验表明，TRC有效减轻了无控制重复，平均减少了57%的循环率。我们的分析进一步表明，重复语义在浅层中出现，并通过残差流传播，干扰正常表示。TRC还可以推广到大型语言模型（LLMs）和大型推理模型（LRMs），在这些模型中，它始终捕捉到类似的重复动态并实现有效的减轻。我们的工作将对重复生成的研究从其显著的内部表示扩展到更早的干预机会，为减轻资源消耗攻击提供了见解。
cs.CL / 34 / 2609.38809
StateTree: Enhancing Long-Term Dialogue Reasoning via Reinforcement Learning
StateTree：通过强化学习增强长期对话推理
Abstract
Large language models deployed as personalized assistants must reason over long, evolving interaction histories. However, in long-term dialogue reasoning, relevant evidence is scattered across sessions, preferences may be revised over time, and standard long-context training fails to address these challenges under data scarcity and prohibitive computational costs. We propose StateTree, a data-driven RL method that constructs a challenging auxiliary task from scarce dialogues with verifiable ground truth. StateTree augments multi-session dialogues with a tree-structured path-tracing task: key-value records are embedded across sessions to form a binary tree. Solving the task requires the model to traverse from root to leaf by retrieving records across sessions and comparing timestamps to resolve branches, then recover the hidden target question among distractor leaves. We apply curriculum RL training progressively increasing tree depth and introduce a compositional variant whose edges carry step-level reasoning fragments, training the model to compose partial cues into coherent queries. Trained on 10K-token contexts, StateTree generalizes to 128K tokens without full-length RL costs and exhibits capabilities including cross-session retrieval, temporal reasoning, knowledge update, and compositional multi-hop reasoning. StateTree outperforms both SFT and RL-based baselines while preserving short-context general reasoning. StateTree-7B achieves gains up to +23.60% on LongMemEval (128k), and StateTree-14B reaches 59.00% accuracy on LongMemEval, surpassing QwenLong-L1-32B (45.20%).
Chinese Translation
作为个性化助手的大型语言模型必须对长期、不断演变的交互历史进行推理。然而，在长期对话推理中，相关证据分散在多个会话中，偏好可能会随着时间的推移而修订，而标准的长上下文训练在数据稀缺和高昂的计算成本下未能解决这些挑战。我们提出了StateTree，这是一种数据驱动的强化学习方法，从稀缺的对话中构建具有可验证真实值的具有挑战性的辅助任务。StateTree通过树状路径追踪任务增强多会话对话：关键值记录在会话之间嵌入以形成二叉树。解决该任务需要模型从根节点遍历到叶节点，通过检索会话中的记录并比较时间戳来解析分支，然后在干扰叶节点中恢复隐藏的目标问题。我们应用课程强化学习训练，逐步增加树的深度，并引入一种组合变体，其边缘携带逐步推理片段，训练模型将部分线索组合成连贯的查询。在10K标记上下文上训练后，StateTree能够在不产生全长强化学习成本的情况下推广到128K标记，并展现出跨会话检索、时间推理、知识更新和组合多跳推理等能力。StateTree在保持短上下文一般推理的同时，超越了SFT和基于强化学习的基线。StateTree-7B在LongMemEval（128k）上实现了高达+23.60%的提升，而StateTree-14B在LongMemEval上达到了59.00%的准确率，超过了QwenLong-L1-32B（45.20%）。
cs.CL / 35 / 2609.38812
Can Terminal Agents Trust Their Own Verification? Diagnosing and Improving Self-Verification
终端代理能否信任自己的验证？自我验证的诊断与改进
Abstract
Terminal agents rely on self-verification to assess and correct their solutions as they solve tasks through interaction with command-line environments. Yet how trustworthy such self-verification is remains poorly understood. To investigate this question systematically, we introduce a diagnostic framework that identifies the first complete solution in each trajectory, determines whether it is objectively correct, and uses this ground truth to quantify the agent's subsequent verification and recovery behavior. Applying it to ten terminal agents on TerminalBench2.1, we find that verification is nearly universal after a complete candidate is formed, yet only 61.43\% of incorrect candidates are detected and only 49.36\% of detected errors are successfully repaired. These results show that the main weakness in self-verification lies not in initiating verification, but in detecting and repairing errors. Motivated by these findings, we propose Student-Conditioned Verification Distillation (SCVD), which lets the student first produce a candidate solution and distills a stronger teacher's subsequent verification and recovery from the same interaction context. Across three Qwen3.5 backbones, SCVD improves \textsc{Pass@1} on TerminalBench2.1 by 9.74--16.85 percentage points over the corresponding base models and by 4.49--8.61 points over the standard full-trajectory distillation, while avoiding the pronounced out-of-distribution degradation of full-trajectory distillation on SWE-bench Verified.
Chinese Translation
终端代理依赖自我验证来评估和纠正其解决方案，以便在与命令行环境的交互中完成任务。然而，这种自我验证的可信度仍然不甚明了。为了系统性地研究这个问题，我们引入了一个诊断框架，该框架识别每条轨迹中的第一个完整解决方案，确定其是否客观正确，并利用这一真实情况来量化代理后续的验证和恢复行为。将其应用于十个终端代理在TerminalBench2.1上的表现，我们发现，在形成完整候选方案后，验证几乎是普遍存在的，但仅有61.43%的错误候选被检测到，且仅有49.36%的检测到的错误成功修复。这些结果表明，自我验证的主要弱点不在于启动验证，而在于检测和修复错误。基于这些发现，我们提出了学生条件验证蒸馏（Student-Conditioned Verification Distillation，SCVD），该方法让学生首先生成候选解决方案，并从相同的交互上下文中提炼出更强的教师后续验证和恢复。在三个Qwen3.5基础模型上，SCVD在TerminalBench2.1上的 extsc{Pass@1}提升了9.74至16.85个百分点，相较于相应的基础模型，提升了4.49至8.61个百分点，同时避免了全轨迹蒸馏在SWE-bench Verified上显著的分布外退化。
cs.CL / 36 / 2609.38816
You're Hired: Strategic Model Selection for LLM Collaboration
你被录用了：大型语言模型协作的战略模型选择
Abstract
While multi-agent and model collaboration algorithms gain traction to combine the strengths of diverse Large Language Models (LLMs), existing systems remain bottlenecked on pre-defined and hand-crafted model pools. In this work, we investigate the problem of model selection in multi-LLM systems. We propose and systematically evaluate a taxonomy of 9 selection algorithms ranging from diversity of model descriptions, capability-aware behavioral diversity, and LLM-based recruiters. We conduct extensive experiments across two candidate pools of 10 and 32 models, deployed in four model collaboration algorithms, and evaluated across tasks spanning math, coding, QA, and reasoning. Results demonstrate that successful selection algorithms greatly outperform random or heuristics-based teams such as merely selecting the models with top individual performance, by up to 36.1% across settings. Specifically, capability- and training-based selection strategies alleviate selection variance and achieve the best performance, which we recommend to employ before deploying real-world multi-LLM systems. Further analysis reveals that larger candidate pools pose greater challenges to shallow selection heuristics, while algorithms grounded in interacting with candidate models and understanding model capability robustly filter out misaligned, unsafe models, as well as generalizing to novel, out-of-distribution tasks. Together, we establish that principled and informed team selection is critical and present strong model selection algorithms for assembling effective multi-LLM systems.
Chinese Translation
尽管多智能体和模型协作算法逐渐受到关注，以结合多样大型语言模型（LLMs）的优势，但现有系统仍然受到预定义和手工制作模型池的瓶颈限制。在本研究中，我们探讨了多LLM系统中的模型选择问题。我们提出并系统评估了9种选择算法的分类，涵盖模型描述的多样性、能力感知的行为多样性以及基于LLM的招聘者。我们在两个候选池（分别包含10个和32个模型）中进行了广泛实验，部署了四种模型协作算法，并在数学、编码、问答和推理等任务中进行了评估。结果表明，成功的选择算法在各个设置中显著优于随机或基于启发式的方法，例如仅选择具有最佳个体表现的模型，提升幅度高达36.1%。具体而言，基于能力和训练的选择策略减轻了选择方差，并实现了最佳性能，我们建议在部署真实世界的多LLM系统之前采用这些策略。进一步分析显示，较大的候选池对浅层选择启发式提出了更大的挑战，而基于与候选模型互动和理解模型能力的算法能够稳健地过滤掉不匹配和不安全的模型，并且能够推广到新颖的、分布外的任务。综上所述，我们确立了原则性和知情的团队选择至关重要，并提出了强大的模型选择算法，以组建有效的多LLM系统。
cs.CL / 37 / 2609.38820
BARRAC: Adaptation of an English Aspect-based Sentiment Analysis Approach for Classification Tasks in Arabic Dialects
BARRAC：将基于英语的方面情感分析方法适应于阿拉伯方言分类任务
Abstract
With the rapid growth of Arabic NLP, several models, datasets and benchmarks have been reported. This paper asks whether approaches developed for majority languages like English can be adapted to Arabic tasks. We adapt an English aspect-based sentiment analysis framework to Arabic classification tasks and present the adaptation as BARRAC: Brainstorming Alignment and Replaced Representation learning for ArabiC tasks. BARRAC replaces consumer-review attribute pools with Arabic linguistic devices and markers for dialectal sentiment, sarcasm, and dialect identification, and replaces noisy self-training with two-stage training. Evaluated on five Arabic dialect datasets, BARRAC achieves a mean macro-F1 of 63.93\%, outperforming the best few-label SOTA by 3\%, and outperforming GPT-4o on four out of five tasks. Error analysis provides insights into remaining challenges. These results demonstrate that adapting task-specific approaches is a promising direction for Arabic NLP alongside adapting models, datasets and benchmarks.
Chinese Translation
随着阿拉伯自然语言处理的快速发展，已经报告了多个模型、数据集和基准。本文探讨了为英语等主要语言开发的方法是否可以适应阿拉伯任务。我们将一个基于英语的方面情感分析框架适应于阿拉伯分类任务，并将这种适应称为BARRAC：阿拉伯任务的头脑风暴对齐和替代表示学习。BARRAC用阿拉伯语言工具和标记替换了消费者评论属性池，以识别方言情感、讽刺和方言识别，并用两阶段训练替换了噪声自训练。在五个阿拉伯方言数据集上的评估中，BARRAC达到了63.93%的平均宏F1值，比最佳的少标签状态最优（SOTA）高出3%，并在五个任务中的四个任务上超越了GPT-4o。错误分析提供了对剩余挑战的见解。这些结果表明，适应特定任务的方法是阿拉伯自然语言处理的一个有前景的方向，除了适应模型、数据集和基准之外。
cs.CL / 38 / 2609.38831
Forging LLM Authorship Fingerprints with Targeted Rewriting
通过针对性重写锻造大语言模型作者指纹
Abstract
Model-attribution classifiers can often identify which language model produced a text, making model-specific writing patterns a signal of provenance. Accurate attribution on unmodified text, however, does not show whether the prediction still identifies the original source after deliberate rewriting. We formulate this problem as targeted fingerprint transfer: rewriting one model's output so that attribution classifiers assign it to a chosen target model. We study summarization, where different models receive the same document and express the same underlying content, providing a controlled setting for conditional generation. We introduce ForgePrint, a search-then-distil framework that first searches for rewrites that move attribution toward a target fingerprint, then distils the selected rewrites into a one-pass 4B Student model. On CNN/DM, the Student reaches 70.2% target success rate, outperforming both its Teacher (54.1%) and the strongest of six published rewriting baselines (39.3%), against held-out classifiers that are never queried by the attack. It also reaches 68.3% target success when transferring summaries from an open model toward chosen commercial models. These results show that fingerprint detectability should not be conflated with source authenticity, and that text-only attribution can provide misleading evidence of model identity under targeted rewriting, even when it is accurate on unmodified text.
Chinese Translation
模型归属分类器通常能够识别出文本是由哪个语言模型生成的，因此模型特定的写作模式成为了来源的信号。然而，在未修改文本上的准确归属并不能显示经过故意重写后预测是否仍能识别原始来源。我们将这个问题表述为针对性指纹转移：重写一个模型的输出，使得归属分类器将其分配给一个选择的目标模型。我们研究了摘要生成，其中不同模型接收相同的文档并表达相同的基本内容，为条件生成提供了一个受控的环境。我们引入了ForgePrint，一个先搜索后提炼的框架，首先搜索能够将归属向目标指纹转移的重写，然后将选定的重写提炼为一个一次性4B学生模型。在CNN/DM数据集上，该学生模型达到了70.2%的目标成功率，超越了其教师模型（54.1%）和六个已发布重写基线中最强的模型（39.3%），并且在未被攻击查询的持出分类器上表现良好。当将来自开放模型的摘要转移到选定的商业模型时，其目标成功率也达到了68.3%。这些结果表明，指纹可检测性不应与来源真实性混淆，并且在针对性重写下，仅基于文本的归属可能会提供误导性的模型身份证据，即使在未修改文本上是准确的。
cs.CL / 39 / 2609.38832
Scaling Parameter and Context in Attention: Native Sparse Attention from Mixture-of-Head
注意力中的缩放参数与上下文：来自混合头的原生稀疏注意力
Abstract
Scaling attention parameters can improve language model quality, but retaining full token histories makes additional heads costly at long contexts. Furthermore, since attention retrieves and combines contextual information, parameter scaling should also support longer contexts. We therefore ask whether attention parameter scaling can directly enable efficient and effective context scaling. We introduce NAMOH, an architecture-native sparse attention mechanism that activates $K$ of $H$ heads per token. Each head retains only its assigned tokens and performs causal attention within this subsequence. Head selection thus jointly determines active parameters and available context without scanning the full history. Under balanced assignments, increasing $H$ at fixed $K$ shortens head histories and reduces per-token key-value (KV) access without increasing total KV storage. We further support head-relative rotary position embeddings to shorten positional spans within routed subsequences, aiming to mitigate position-induced attention noise. Experiments show that NAMOH can outperform fully activated models with the same total parameters, while enabling more efficient long-context inference than smaller dense models with matched active parameter counts. It remains compatible with GQA and existing sparse attention mechanisms. We hope this work offers a new path for scaling attention, with parameter scaling directly enabling context scaling.
Chinese Translation
缩放注意力参数可以提高语言模型的质量，但在长上下文中保留完整的标记历史会使额外的头部变得成本高昂。此外，由于注意力机制检索和组合上下文信息，参数缩放也应支持更长的上下文。因此，我们提出了一个问题：注意力参数缩放是否可以直接实现高效且有效的上下文缩放。我们引入了NAMOH，这是一种架构原生的稀疏注意力机制，每个标记激活$K$个$H$个头部。每个头部仅保留其分配的标记，并在该子序列内执行因果注意力。因此，头部选择共同决定了活动参数和可用上下文，而无需扫描完整历史。在平衡分配下，以固定的$K$增加$H$可以缩短头部历史，并减少每个标记的键值（KV）访问，而不增加总的KV存储。我们进一步支持头部相对旋转位置嵌入，以缩短路由子序列内的位置跨度，旨在减轻由位置引起的注意力噪声。实验表明，NAMOH可以在相同总参数下超越完全激活的模型，同时在活跃参数数量匹配的情况下实现比较小的稠密模型更高效的长上下文推理。它仍然与GQA和现有的稀疏注意力机制兼容。我们希望这项工作为注意力的缩放提供了一条新路径，其中参数缩放直接实现上下文缩放。
cs.CL / 40 / 2609.38851
Where MLLMs Fail and Why: Causal Task Decomposition for Capability Failure Diagnosis
多模态大语言模型失败的原因及诊断：基于因果任务分解的能力缺陷分析
Abstract
End-to-end accuracy on compositional tasks records how often MLLMs fail, but cannot distinguish whether a failure reflects an intrinsic deficit in the targeted capability or a cascading error from an upstream prerequisite. We propose a causal decomposition framework that isolates these two failure modes through controlled interventions on the prerequisite dependencies of each task. Our capability metrics (NC, IC, RC) score each task under unassisted, correct, or incorrect prerequisites to diagnose where failures arise; contribution metrics (N-Score, S-Score), adapted from probabilities of causation, quantify each prerequisite's necessity and sufficiency to determine why. We instantiate the framework in CADET, a diagnostic benchmark of 10 composite tasks decomposed into 46 unit tasks with over 33,000 human-annotated questions spanning perception, spatial, temporal, and cognitive categories. Diagnosing frontier MLLMs with our framework uncovers systematic patterns that end-to-end accuracy obscures. Capability-wise, supplying correct prerequisites eliminates 54\% of errors on cognitive tasks, lifting them from weakest to above spatial and temporal. Prerequisite-wise, causal contributions are concentrated in a few critical prerequisites, and supplying the single most important one alone captures 84\% of the gain from supplying all prerequisites.
Chinese Translation
组合任务的端到端准确率记录了多模态大语言模型（MLLMs）失败的频率，但无法区分失败是源于目标能力的内在缺陷，还是上游先决条件的级联错误。我们提出了一种因果分解框架，通过对每个任务的先决依赖进行受控干预，隔离这两种失败模式。我们的能力指标（NC、IC、RC）分别在无辅助、正确或错误的先决条件下对每个任务进行评分，以诊断失败的起因；贡献指标（N-Score、S-Score）则基于因果概率，量化每个先决条件的必要性和充分性，以确定失败的原因。我们在CADET中实现了该框架，CADET是一个诊断基准，包含10个复合任务，分解为46个单元任务，涵盖感知、空间、时间和认知类别，包含超过33,000个人工标注的问题。利用该框架对前沿MLLMs进行诊断揭示了端到端准确率所掩盖的系统性模式。在能力层面，提供正确的先决条件可消除认知任务54%的错误，使其表现从最弱提升至超过空间和时间任务。在先决条件层面，因果贡献集中于少数关键先决条件，仅提供最重要的一个先决条件就能捕获提供所有先决条件所带来收益的84%。
cs.CL / 41 / 2609.38861
TRACE: Target-Aware Retrieval, Attributed Evidence, and Contract-Constrained Extraction for LitTraceQA
TRACE：面向目标的检索、归因证据和合同约束的提取用于LitTraceQA
Abstract
Finding a relevant paper is not the same as producing a verifiable answer from it. LitTraceQA requires canonical paper identifiers, exact evidence at the page or object level, and typed answers that match the evaluator. We call the separation between source access and scorer-visible correctness the grounding contract gap. TRACE - Target-Aware Retrieval, Attributed Evidence, and Contract-Constrained Extraction - addresses this gap with target-grouped retrieval, independent typed evidence localization, multimodal table extraction, schema-driven table construction, and fail-closed validation. It indexes 27,487 papers through passage, object, alias, citation, and dense representations while retaining the question target behind each signal. For tables, TRACE predicts the observation unit before extracting values and assembles rows with evaluator-compatible key normalization. Our audited selected clean-track artifact scores 0.760613 on the official 71-question test set, including 0.9728 paper F1, 0.6847 evidence F1, 0.9800 multiple-choice accuracy, 0.5423 table-row F1, and 0.3508 macro cell accuracy. On 11 public-development table records, a clean baseline and coordinate-aware visual fill obtain row F1 of 0.291 and 0.411, respectively; this diagnostic comparison includes fallback outputs and is not an official-test claim. Remaining errors chiefly concern locator, observation-unit, row-key, and source-value identity.
Chinese Translation
找到相关论文并不等同于从中生成可验证的答案。LitTraceQA需要规范的论文标识符、精确的证据（在页面或对象层面）以及与评估者匹配的类型化答案。我们将源访问与评分可见正确性之间的分离称为基础合同差距。TRACE（面向目标的检索、归因证据和合同约束的提取）通过目标分组检索、独立的类型化证据定位、多模态表格提取、基于模式的表格构建和故障安全验证来解决这一差距。它通过段落、对象、别名、引用和密集表示对27,487篇论文进行了索引，同时保留每个信号背后的问题目标。对于表格，TRACE在提取值之前预测观察单位，并通过与评估者兼容的关键规范化组装行。我们审核的选定清洁轨迹工件在官方71个问题测试集上得分为0.760613，包括0.9728的论文F1、0.6847的证据F1、0.9800的多项选择准确率、0.5423的表行F1和0.3508的宏单元准确率。在11个公共开发表记录中，清洁基线和坐标感知视觉填充分别获得了0.291和0.411的行F1；该诊断比较包括回退输出，并不是官方测试声明。剩余的错误主要涉及定位器、观察单位、行键和源值的身份。
cs.CL / 42 / 2609.38923
GraphForge: Training Working Agents with Graph-Anchored Workspace Synthesis
GraphForge：基于图锚的工作空间合成训练工作代理
Abstract
Working agents need to read diverse files, coordinate tools, and produce deliverables. Training such agents requires tasks built on many real files with verifiable results, but few pipelines exist to synthesize this kind of data. Existing pipelines either generate files with models, which lack realism and diversity, or build tasks on real files without task-specific verifiers, leaving result quality unchecked. We introduce GraphForge, an evidence-graph based framework that grounds both the task and its verification in real files. Starting from occupation-grounded seeds for controlled diversity, GraphForge assembles a workspace of real files for each seed and builds an evidence graph over their relations. Since the task statement and rubrics are both derived from this graph, task requirements are backed by the workspace files and each criterion is anchored to the files needed to verify it. An initial rollout further tests executability, and a revision agent repairs the task and rubrics against the original files before trajectories are collected. Fine-tuning Qwen3.6-27B on 2,169 GraphForge trajectories brings GDPVal to 1445.7 (+65.7) under OpenHands, and Workspace-Bench-Lite and SpreadsheetBench II to 63.7 (+7.7) and 24.0 (+13.7) under Claude Code. Rejection fine-tuning on the SFT model's own rollouts, with candidates selected by the evidence-anchored rubrics, yields further improvements on all three benchmarks, suggesting that the rubrics provide a useful selection signal. The data and models are available.
Chinese Translation
工作代理需要读取多样化的文件、协调工具并生成交付物。训练这样的代理需要基于许多真实文件构建的任务，并且这些任务需要有可验证的结果，但目前很少有管道能够合成这种数据。现有的管道要么生成缺乏真实感和多样性的模型文件，要么在真实文件上构建任务而没有特定任务的验证器，从而使结果质量无法得到保证。我们提出了GraphForge，这是一种基于证据图的框架，将任务及其验证与真实文件相结合。从以职业为基础的种子开始，以实现受控的多样性，GraphForge为每个种子组装一个真实文件的工作空间，并在其关系上构建一个证据图。由于任务声明和评分标准均源自该图，任务要求得到了工作空间文件的支持，每个标准都与验证所需的文件相锚定。初步的推出进一步测试了可执行性，而修订代理则在收集轨迹之前根据原始文件修复任务和评分标准。在2,169个GraphForge轨迹上微调Qwen3.6-27B，使GDPVal在OpenHands下达到1445.7（+65.7），在Claude Code下使Workspace-Bench-Lite和SpreadsheetBench II分别达到63.7（+7.7）和24.0（+13.7）。对SFT模型自身推出的拒绝微调，候选者由证据锚定的评分标准选择，进一步提高了所有三个基准的性能，表明评分标准提供了有用的选择信号。数据和模型均可获取。
cs.CL / 43 / 2609.38972
Making LLMs Say What They Think: Measuring and Improving CoT-Interpretability Alignment
让大型语言模型表达其思考：测量和改善链式推理（CoT）可解释性一致性
Abstract
Chain-of-thought (CoT) traces often serve as a proxy for how Large Language Models (LLMs) arrive at their answers. However, growing evidence shows that models' CoT often fails to reflect their internal computations and can be changed without affecting their final answers. In this work, we measure and improve the alignment between the reasoning described in an LLM's CoT and what it computes internally. We propose CoT-Interpretability Alignment (CIA), a metric that measures the agreement between a model's CoT traces and its internal reasoning strategies as detected by interpretability tools. We evaluate CIA on three tasks (two-hop question answering, hint intervention, and integer multiplication) across three LLMs, finding that LLMs exhibit limited alignment across all tasks (44.8-75.9%). We then experiment with improving CIA via post-training, setting both the task accuracy and parametric faithfulness signals as a reward. Experiments show that we can substantially improve CoT parametric faithfulness while maintaining or improving the task accuracy. We provide rich analysis, such as their generalization patterns. Our work provides both a framework for auditing CoT parametric faithfulness and a pathway toward making models' explicit reasoning more trustworthy. Code and data are available at https://github.com/yihuaihong/CIA-minimal-repro.
Chinese Translation
链式推理（CoT）轨迹通常作为大型语言模型（LLMs）得出答案的代理。然而，越来越多的证据表明，模型的CoT往往无法反映其内部计算，并且可以在不影响最终答案的情况下进行更改。在本研究中，我们测量并改善LLM的CoT中描述的推理与其内部计算之间的一致性。我们提出了链式推理可解释性一致性（CoT-Interpretability Alignment，CIA）这一指标，用于衡量模型的CoT轨迹与其内部推理策略之间的符合程度，这些策略由可解释性工具检测。我们在三项任务（双跳问答、提示干预和整数乘法）上评估CIA，涵盖三种LLM，发现LLM在所有任务上的一致性有限（44.8%-75.9%）。随后，我们通过后训练实验来改善CIA，同时将任务准确性和参数忠实度信号设定为奖励。实验表明，我们可以在保持或提高任务准确性的同时，显著改善CoT的参数忠实度。我们提供了丰富的分析，例如它们的泛化模式。我们的工作为审计CoT参数忠实度提供了框架，并为使模型的显性推理更加可信提供了途径。代码和数据可在 https://github.com/yihuaihong/CIA-minimal-repro 获取。
cs.CL / 44 / 2609.38976
Fairness Beyond a Single Run: Training-Seed Variability in Speech LLM Adaptation
超越单次训练的公平性：语音大语言模型适应中的训练种子变异性
Abstract
Demographic fairness gaps in automatic speech recognition are almost always reported from a single training run. We fine-tune the Q-former projector and LoRA adapters of a speech LLM at five audio compression factors and six random seeds, holding the encoder, base decoder, data and decoding fixed, and evaluate every run on Common Voice and Fair-Speech. At 460 h of clean LibriSpeech, the seed moves fairness metrics more than compression does on most demographic axes. A balanced 3x3 decomposition attributes 85.3% of the variation in Fair-Speech ethnicity normalized gap to the seed against 8.3% to compression (p = 0.009), though compression explains more on age and gender. Held-out LibriSpeech word error rate spreads by 0.04 points across those seeds while Common Voice spreads by 8.57, so these are not failed runs, and the effect survives controlling for accuracy and dropout. Scaling and diversifying the adaptation set to 960 h damps the effect but does not remove it. On Fair-Speech ethnicity, two single-run systems must differ by more than 0.30 in normalized gap to exceed seed variability.
Chinese Translation
自动语音识别中的人口公平性差距几乎总是基于单次训练结果进行报告。我们在五个音频压缩因子和六个随机种子下，微调了语音大语言模型的 Q-former 投影器和 LoRA 适配器，同时固定编码器、基础解码器、数据和解码过程，并在 Common Voice 和 Fair-Speech 上评估每次运行。在 460 小时的干净 LibriSpeech 数据集上，种子对公平性指标的影响在大多数人口维度上超过了压缩的影响。经过平衡的 3x3 分解分析显示，Fair-Speech 中的种族标准化差距中有 85.3% 的变异归因于种子，而压缩仅占 8.3%（p = 0.009），尽管在年龄和性别方面，压缩的解释力度更强。保留的 LibriSpeech 单词错误率在这些种子之间的差异为 0.04，而 Common Voice 的差异为 8.57，因此这些并不是失败的运行，且该效应在控制准确性和 dropout 后依然存在。将适应集的规模和多样性扩大至 960 小时减弱了该效应，但并未消除。在 Fair-Speech 种族中，两个单次运行的系统必须在标准化差距上相差超过 0.30，才能超越种子变异性。
cs.CL / 45 / 2609.38995
When Clipping Reverses Correction: Failure Dynamics of Pointwise Forward-KL On-Policy Self-Distillation
当剪切逆转校正：逐点前向KL策略自蒸馏的失败动态
Abstract
On-policy self-distillation (OPSD) trains a student on its own generated responses using feedback from the same model conditioned on privileged information. On mathematical reasoning, the original OPSD study finds that stylistic tokens can dominate the training signal over math-related tokens, and that pointwise clipping of the forward KL objective stabilizes training. Pointwise clipping caps each vocabulary-wise forward KL term at a fixed threshold before summing over the vocabulary. Follow-up studies have adopted this clipping, but its effect on training has not been directly examined. In matched training runs differing only in whether clipping is applied, we observe that clipped runs produce substantially more repetitions that persist to the end of the response than their unclipped counterparts. We trace this failure to the clipped objective. We prove that the clipped objective can fail to correct the student toward the teacher and can instead push clipped and unclipped token probabilities away from its teacher. Our training runs agree with this analysis: inside repetitions, the clipped student places less probability than its teacher on leaving the repetition, and more on continuing it, whereas the unclipped runs stay close to their teachers.
Chinese Translation
策略自蒸馏（On-policy self-distillation，OPSD）通过使用来自同一模型的反馈，在特权信息的条件下训练学生模型以生成其自身的响应。在数学推理方面，原始的OPSD研究发现，风格性标记可以主导训练信号，而数学相关标记则相对较弱，并且逐点剪切前向KL目标可以稳定训练。逐点剪切在对词汇进行求和之前，将每个词汇的前向KL项限制在一个固定阈值。后续研究采用了这种剪切方法，但其对训练的影响尚未直接考察。在仅在是否应用剪切上有所不同的匹配训练运行中，我们观察到剪切运行产生的重复次数显著多于未剪切的对应运行，并且这种重复持续到响应的结束。我们将这一失败归因于剪切目标。我们证明了剪切目标可能无法将学生模型校正到教师模型，反而可能将剪切和未剪切的标记概率推离其教师模型。我们的训练运行与这一分析一致：在重复内容中，剪切学生模型在离开重复时的概率低于其教师模型，而在继续重复时的概率则更高，而未剪切的运行则保持接近其教师模型。
cs.CL / 46 / 2609.38997
Settle: Learning When to Stop Reasoning
Settle：学习何时停止推理
Abstract
Reasoning models often continue generating after their answers have settled. Settle learns when to stop from answer stability in completed traces. It trains the existing end-of-reasoning token while keeping other predictions close to the base model, and requires only ordinary decoding at inference. On MATH-500 with Qwen3-4B, Settle reduces token count by 40% with a 0.5-percentage-point decrease in accuracy. It gains 6.16 percentage points over supervised fine-tuning on the same traces shortened at their first stable answer, at nearly identical token counts. Its stopping score predicts whether a correct answer will remain correct. Settle extends the accuracy-token-count Pareto frontier of the evaluated stopping methods.
Chinese Translation
推理模型在其答案已经稳定后往往仍会继续生成。Settle 从完成的轨迹中的答案稳定性学习何时停止。它在训练现有的推理结束标记时，保持其他预测接近基础模型，并且在推理时仅需普通解码。在 MATH-500 数据集上使用 Qwen3-4B，Settle 将标记数量减少了 40%，同时准确率下降了 0.5 个百分点。它在相同的轨迹上获得了比监督微调高出 6.16 个百分点的提升，这些轨迹在其第一个稳定答案处被缩短，且标记数量几乎相同。其停止评分预测一个正确答案是否会保持正确。Settle 扩展了评估的停止方法的准确性-标记数量的帕累托前沿。
cs.CL / 47 / 2609.39001
The Invisible Language Tax: Token Premiums of French and Regional Languages in 2026 LLM Tokenizers, and a French-Optimized Prototype
隐形语言税：2026年大型语言模型（LLM）分词器中法语和区域语言的令牌溢价及法语优化原型
Abstract
LLM services are billed per token and context windows are measured in tokens, yet the number of tokens needed for the same content varies across languages. We measure this token premium on seven tokenizers of widely used 2026 models (OpenAI o200k, Llama 3, Qwen3, DeepSeek V3/V4, Gemma 3, Mistral Tekken, and the Claude generation-5 tokenizer via Anthropic's counting API) on NTREX-128 (124 non-English reference translations) and on the Universal Declaration of Human Rights for regional languages. French requires 31% to 58% more tokens than English, whereas Simplified Chinese ranges from 5% fewer to 40% more and is cheaper than French on six of the seven tokenizers. Regional and overseas languages of France pay roughly 1.6 to 3.3 times the English count. We discuss how history re-sending, tiered pricing and fixed context windows amplify the absolute gap in agentic use. In a controlled experiment (BPE, Europarl, 50k vocabulary), adding French to tokenizer training data quickly reduces the premium, with diminishing returns and a growing cost for English. Finally, we present Baracoda FR v1.2, a byte-level BPE prototype with Tekken's vocabulary size. On a final test of six corpora never consulted during design, with a protocol declared fixed beforehand, it uses 11.5% fewer tokens than Tekken on French and 3.7% fewer on English; results hold after removing test sentences overlapping the training data and with an equal ordinary-token budget. It is worse on other languages and, at comparable vocabulary size, does not outperform CroissantLLM. These are segmentation results only; effects on model quality and task cost remain to be shown.
Chinese Translation
大型语言模型（LLM）服务按令牌计费，上下文窗口以令牌为单位进行测量，但相同内容所需的令牌数量在不同语言之间存在差异。我们在七个广泛使用的2026年模型的分词器（OpenAI o200k、Llama 3、Qwen3、DeepSeek V3/V4、Gemma 3、Mistral Tekken，以及通过Anthropic的计数API获取的Claude generation-5分词器）上测量了这种令牌溢价，使用的语料库包括NTREX-128（124个非英语参考翻译）和《世界人权宣言》的区域语言。法语所需的令牌比英语多31%到58%，而简体中文的令牌需求则在少5%到多40%之间，并且在七个分词器中的六个上，其成本低于法语。法国的区域语言和海外语言的令牌需求大约是英语的1.6到3.3倍。我们讨论了历史重发、分层定价和固定上下文窗口如何放大代理使用中的绝对差距。在一项受控实验中（BPE、Europarl、50k词汇量），将法语添加到分词器训练数据中迅速降低了溢价，但收益递减且对英语的成本不断增加。最后，我们展示了Baracoda FR v1.2，这是一个字节级BPE原型，具有Tekken的词汇规模。在对六个在设计过程中从未咨询过的语料库进行的最终测试中，采用事先声明的固定协议，其在法语上的令牌使用量比Tekken少11.5%，在英语上少3.7%；在去除与训练数据重叠的测试句子并保持相等的普通令牌预算后，结果依然成立。它在其他语言上的表现较差，并且在相当的词汇规模下，未能超越CroissantLLM。这些仅是分割结果；对模型质量和任务成本的影响仍需进一步验证。
cs.CL / 48 / 2609.39013
Evidence First, Arithmetic Second: A System Report and Failure Analysis for DocSem
先证据，后算术：DocSem系统报告及故障分析
Abstract
EVICALC, our system for the DocSem shared task, achieved 8.61% joint accuracy on 1,730 tasks in the official final test evaluation. It reads a PDF, selects a passage, asks a language model to write an arithmetic expression, and evaluates that expression in local code. Saved intermediate results support inspection of failures. A separate public-validation run achieved 92.17% answer accuracy and 1.00 evidence F1. The configurations and metrics differ, so these scores are not a controlled comparison. Our manual, post-hoc analysis is descriptive: in one inspected case, optical character recognition (OCR) and block grouping merged the relevant passage into another block, and the system answered from unrelated text. An exploratory study of reading page images on 100 documents returned evidence identifiers for only 22 documents. These descriptive findings motivate further evaluation; they do not establish the causes of the overall score.
Chinese Translation
EVICALC是我们为DocSem共享任务开发的系统，在官方最终测试评估中，在1730个任务上实现了8.61%的联合准确率。该系统读取PDF文件，选择一段文本，调用语言模型生成算术表达式，并在本地代码中计算该表达式。保存的中间结果支持故障检查。一次独立的公开验证运行获得了92.17%的答案准确率和1.00的证据F1分数。由于配置和指标不同，这些分数并非受控比较。我们的手工事后分析是描述性的：在一个检查案例中，光学字符识别（OCR）和块分组将相关段落合并到了另一个块中，系统因此从无关文本中给出了答案。对100份文档的页面图像阅读探索性研究仅返回了22份文档的证据标识符。这些描述性发现促使我们进行进一步评估，但并未确定整体得分的具体原因。
cs.CL / 49 / 2609.39027
A Missing Piece for Trustworthy AI Reviewers: From Benchmarking Rhetorical Robustness to SciCore Review
可信赖的人工智能评审者的缺失环节：从基准测试修辞稳健性到SciCore评审
Abstract
AI reviewers can assign different judgments to manuscripts that report the same science in different wording, potentially rewarding rhetorical optimization over scientific improvement. We formulate Rhetorical Robustness as the joint requirement of stability across content-preserving rewrites and discrimination across papers. We introduce RobustReview, a controlled full-manuscript benchmark with 1,260 manuscript versions, and evaluate 30 reviewer configurations. The benchmark reveals false robustness, where low rewrite sensitivity coincides with score collapse across papers, and shows that human alignment and rhetorical robustness rank reviewers differently. Moreover, the evaluated content-focused prompting protocol does not consistently improve robustness across backbones. Motivated by these findings, we introduce SciCore, a dual-branch reviewer that averages a full-manuscript judgment with a judgment based on an extracted, structured science core. This design combines manuscript-level assessment with a content-normalized view intended to reduce rhetorical sensitivity. In our primary GPT-5.5 comparison, SciCore achieves a leading joint stability-discrimination profile among the benchmarked reviewers while maintaining competitive human alignment. These results identify rhetorical robustness as a distinct evaluation target and demonstrate the potential of science-core review to improve it.
Chinese Translation
人工智能评审者可能会对用不同措辞报告相同科学内容的手稿给予不同的判断，这可能会奖励修辞优化而非科学改进。我们将修辞稳健性定义为在内容保持不变的重写中稳定性与论文之间的区分的联合要求。我们引入了RobustReview，这是一个包含1260个手稿版本的受控全手稿基准，并评估了30种评审配置。基准测试揭示了虚假的稳健性，即低重写敏感性与论文间的得分崩溃同时发生，并显示人类对齐和修辞稳健性对评审者的排名影响不同。此外，评估的以内容为中心的提示协议并未在不同基础模型中始终提高稳健性。基于这些发现，我们引入了SciCore，这是一种双分支评审者，它将全手稿判断与基于提取的结构化科学核心的判断进行平均。这一设计结合了手稿级评估与旨在减少修辞敏感性的内容归一化视角。在我们主要的GPT-5.5比较中，SciCore在基准评审者中实现了领先的联合稳定性-区分特征，同时保持了竞争的人类对齐。这些结果将修辞稳健性确定为一个独特的评估目标，并展示了科学核心评审改善其潜力。
cs.CL / 50 / 2609.39045
RSIGame: Autonomous Agentic Game Development with Recursive Self-improvement
RSIGame：具有递归自我改进的自主智能游戏开发
Abstract
Recent advances in large language models have made automatic game generation increasingly feasible, yet reliably improving generated games beyond a playable version remains challenging. Naive iterative refinement can easily overfit a small set of test cases, producing fragile games with unresolved bugs, missing behaviors, and poor generalization to broader player interactions. We introduce RSIGame, an autonomous agentic game development framework with recursive self-improvement. RSIGame organizes development into complementary local and global loops. Concretely, a local explore-diagnose-improve loop broadly explores the executable game, diagnoses and prioritizes discovered issues, and performs evidence-grounded revision, where an evolving checklist continually accumulates new testing and improvement guidance. A global loop tracks overall quality, preserves the best checkpoint, and detects saturation or regression over long-horizon development. Beyond test-time improvement, RSIGame further internalizes successful development experience into the generator through training. Across 140 GameCraft-Bench tasks, two game engines, and five generators, RSIGame consistently improves game quality under matched development budgets. Notably, experience internalization enables Qwen3.8-27B to reach 61.38 on Godot and 58.53 on Phaser, exceeding GPT-5.5 one-shot scores while reducing Qwen's generation tokens by 11 times.
Chinese Translation
近年来，大型语言模型的进展使自动游戏生成变得越来越可行，但在可玩版本之外可靠地改进生成的游戏仍然具有挑战性。简单的迭代优化容易在小规模测试用例上过拟合，导致生成的游戏脆弱，存在未解决的错误、缺失的行为，并且在更广泛的玩家互动中泛化能力差。我们提出了RSIGame，一个具有递归自我改进的自主智能游戏开发框架。RSIGame将开发过程组织为互补的局部和全局循环。具体而言，局部探索-诊断-改进循环广泛探索可执行游戏，诊断并优先处理发现的问题，并进行基于证据的修订，其中一个不断发展的检查清单持续积累新的测试和改进指导。全局循环跟踪整体质量，保存最佳检查点，并检测长期开发中的饱和或回归。除了测试时的改进，RSIGame还通过训练进一步将成功的开发经验内化到生成器中。在140个GameCraft-Bench任务、两个游戏引擎和五个生成器中，RSIGame在匹配的开发预算下始终提高游戏质量。值得注意的是，经验内化使得Qwen3.8-27B在Godot上达到61.38，在Phaser上达到58.53，超越了GPT-5.5的一次性得分，同时将Qwen的生成令牌减少了11倍。
cs.CL / 51 / 2609.39049
Structure vs. Chain-of-Thought: Evaluating LLM Criteria Extraction for Depression Severity
结构与思维链：评估大型语言模型在抑郁严重程度标准提取中的表现
Abstract
A large language model (LLM) can rate depression severity directly from a social media post or mark which clinical criteria the post shows and let code turn the count into a label. The latter is easier to audit because a clinician can check each marked criterion. We compare these approaches on two Reddit corpora using three LLMs (from 9B to frontier scale) and two questionnaires (PHQ-9, BDI-II), and measure agreement with quadratic weighted kappa. For the two frontier models, criteria extraction scores above chain-of-thought on one corpus only when its decision thresholds are fitted on labeled data. Neither model's gain is significant, with or without recalibrating chain-of-thought on the same labels. With thresholds fixed a priori from PHQ-9's criteria, extraction shows no gain on either corpus, even where models mark over two criteria per post. The 9B model behaves differently on a corpus from depression communities. It labels most posts severe, whether prompted directly or with chain-of-thought, while the a priori rule beats both without labels. After chain-of-thought is recalibrated on the same labels, no significant gap remains, consistent with a calibration effect. Yet higher ordinal agreement does not ensure better detection of severe cases. PHQ-9 criteria extraction misses most severe posts, and moving from direct prompting to chain-of-thought and then to extraction increases misses in nearly all comparisons. On the primary corpus, a relabeled stress dataset, a model using that dataset's own features, including word counts from the text, is not significantly different from frontier criteria extraction under the a priori rule.
Chinese Translation
大型语言模型（LLM）可以直接从社交媒体帖子中评估抑郁严重程度，或标记帖子所显示的临床标准，并让代码将计数转化为标签。后者更易于审计，因为临床医生可以检查每个标记的标准。我们在两个Reddit语料库上比较了这两种方法，使用了三种LLM（从9B到前沿规模）和两个问卷（PHQ-9，BDI-II），并通过二次加权kappa测量一致性。对于两个前沿模型，当其决策阈值在标记数据上进行调整时，标准提取得分在一个语料库上超过思维链。无论是否在相同标签上重新校准思维链，两个模型的增益均不显著。在PHQ-9标准的阈值事先固定的情况下，提取在两个语料库上均未显示增益，即使模型每个帖子标记超过两个标准。9B模型在抑郁社区的一个语料库上表现不同。无论是直接提示还是使用思维链，它都将大多数帖子标记为严重，而事先规则在没有标签的情况下优于两者。在思维链在相同标签上重新校准后，显著差距消失，这与校准效应一致。然而，更高的序数一致性并不确保更好地检测严重病例。PHQ-9标准提取漏掉了大多数严重帖子，从直接提示到思维链再到提取的过程在几乎所有比较中都增加了漏掉的数量。在主要语料库中，一个重新标记的压力数据集，使用该数据集自身特征的模型，包括文本中的字数，与事先规则下的前沿标准提取没有显著差异。
cs.CL / 52 / 2609.39069
CORE: Conflict-Oriented Reasoning Elimination for Verifiable Language-Model Search
CORE：用于可验证语言模型搜索的冲突导向推理消除
Abstract
Test-time reasoning systems often respond to failure by restarting or revising the latest step, even when an earlier decision caused the error. We introduce CORE, a search controller that requests a certified conflict core from a verifier, backjumps to the latest decision in that core, and caches the conflict to avoid repeating it. Under sound verification, finite branching and depth, and exhaustive proposals, the uncapped search is complete and never prunes a valid solution. On 2,000 planted graph-coloring instances with matched proposals and an exact verifier, CORE reduces median verifier calls by 39.8% at 30 variables and 35.0% at 36 variables relative to chronological repair; caching further improves on backjumping alone. Across five reasoning tasks, CORE achieves 75.9% mean success with Qwen2.5-7B-Instruct and 84.2% with Qwen3-8B, compared with 72.5% and 81.8% for Tree of Thoughts. It also uses fewer verifier calls and generated tokens on both backbones. These results show the value of using certified failure explanations to direct language-model search.
Chinese Translation
测试时推理系统通常通过重启或修订最新步骤来应对失败，即使早期的决策导致了错误。我们提出了CORE，一种搜索控制器，它请求验证器提供经过认证的冲突核心，回溯到该核心中的最新决策，并缓存冲突以避免重复。在可靠的验证、有限的分支和深度以及全面的提议下，无限制搜索是完整的，并且从不剪枝有效解。在2000个植入的图着色实例中，使用匹配提议和精确验证器，CORE在30个变量时将中位数验证器调用减少了39.8%，在36个变量时减少了35.0%，相较于按时间顺序修复；缓存进一步改善了单独回跳的效果。在五个推理任务中，CORE在使用Qwen2.5-7B-Instruct时实现了75.9%的平均成功率，在使用Qwen3-8B时实现了84.2%，而Tree of Thoughts的成功率分别为72.5%和81.8%。它在两个基础模型上也使用了更少的验证器调用和生成的令牌。这些结果表明，使用经过认证的失败解释来指导语言模型搜索的价值。
cs.CL / 53 / 2609.39071
LexReward: A Taxonomy-Driven Reward Framework for Legal Language Models
LexReward：一种基于分类法的法律语言模型奖励框架
Abstract
Legal language models require reward signals that capture not only answer correctness but also the multidimensional quality of legal responses. Existing reward methods, however, often rely on coarse-grained holistic judgments, providing limited domain specificity and interpretability. We introduce LexReward, a taxonomy-driven framework for legal reward modeling. LexReward characterizes legal response quality along three complementary dimensions: Style, covering lexical and syntactic quality; Element, assessing legal subjects, facts, statutes, and decisions; and Chain, evaluating the order, completeness, correctness, and non-redundancy of legal reasoning. For each dimension, we develop rubrics that specify evaluation criteria and quality levels. The resulting rewards are used to construct pairwise preference data for Direct Preference Optimization (DPO) and reward-model training. Experiments show that the rubric-based rewards reliably distinguish legal responses of different quality and that DPO training on the preference data improves performance across all three dimensions. The learned reward models, LexRM, also support effective downstream optimization: each dimension-specific reward model improves policy performance in its corresponding dimension through reinforcement learning, without requiring reference answers at reward time. Dimension-wise analyses further support the effectiveness of the proposed taxonomy and reward construction.
Chinese Translation
法律语言模型需要奖励信号，不仅要捕捉答案的正确性，还要反映法律响应的多维质量。然而，现有的奖励方法通常依赖于粗略的整体判断，提供的领域特异性和可解释性有限。我们提出了LexReward，一个基于分类法的法律奖励建模框架。LexReward从三个互补维度来表征法律响应质量：风格（Style），涵盖词汇和句法质量；要素（Element），评估法律主体、事实、法规和判决；链条（Chain），评估法律推理的顺序、完整性、正确性和非冗余性。针对每个维度，我们开发了评估标准，明确了评价标准和质量水平。生成的奖励用于构建成对偏好数据，以进行直接偏好优化（Direct Preference Optimization, DPO）和奖励模型训练。实验表明，基于评估标准的奖励能够可靠地区分不同质量的法律响应，并且在偏好数据上进行DPO训练能够提高所有三个维度的性能。学习到的奖励模型LexRM也支持有效的下游优化：每个维度特定的奖励模型通过强化学习在其对应维度上提高策略性能，而无需在奖励时提供参考答案。维度分析进一步支持了所提分类法和奖励构建的有效性。
cs.CL / 54 / 2609.39072
Beyond Text: LLM-Based Dimensional Emotion Evaluation in Multimodal Dialogue
超越文本：基于大型语言模型的多模态对话中的维度情感评估
Abstract
Emotion recognition in conversation has been widely studied, but applying Large Language Models (LLMs) to continuous dimensional emotion evaluation in multimodal dialogue remains largely unexplored. We propose an LLM-based framework that performs discrete emotion recognition and Valence-Arousal-Dominance (VAD) dimensional evaluation on IEMOCAP, incorporating acoustic cues as natural language descriptions following the SpeechCueLLM approach. We evaluate six models spanning the LLaMA, GPT, and Qwen families under zero-shot prompting, few-shot prompting, and LoRA fine-tuning. LoRA fine-tuned LLaMA models substantially outperform prompt-engineered GPT models on both tasks despite GPT's larger scale, a gap we attribute to domain adaptation rather than model capacity. Our best model achieves a Valence CCC of 0.7822, a new state-of-the-art on IEMOCAP. Ablation studies confirm that textual audio descriptions meaningfully improve smaller models (+3.5 to 3.6 weighted F1) while contributing little for the largest model, suggesting audio cues are most valuable when linguistic capacity is limited. The performance asymmetry across VAD dimensions closely mirrors the annotator agreement hierarchy in IEMOCAP's own annotations.
Chinese Translation
对话中的情感识别已被广泛研究，但将大型语言模型（LLMs）应用于多模态对话中的连续维度情感评估仍然很少被探索。我们提出了一种基于LLM的框架，该框架在IEMOCAP上执行离散情感识别和情感的愉悦度-唤醒度-主导度（Valence-Arousal-Dominance, VAD）维度评估，结合了根据SpeechCueLLM方法生成的自然语言描述的声学线索。我们在零-shot提示、少量提示和LoRA微调下评估了涵盖LLaMA、GPT和Qwen系列的六个模型。尽管GPT的规模更大，但经过LoRA微调的LLaMA模型在这两个任务上显著优于经过提示工程的GPT模型，我们认为这一差距归因于领域适应而非模型容量。我们的最佳模型在IEMOCAP上实现了0.7822的愉悦度CCC，创造了新的最先进水平。消融研究确认，文本音频描述显著改善了较小模型的表现（+3.5到3.6加权F1），而对最大模型的贡献较小，这表明当语言能力有限时，音频线索的价值最大。VAD维度间的性能不对称性与IEMOCAP自身注释中的标注者一致性层级密切相关。
cs.CL / 55 / 2609.39102
False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents
虚假边界：诊断和缓解自演化搜索代理中的共作弊现象
Abstract
Self-evolving search agents build their own training curricula by jointly optimizing a proposer that generates questions and a solver that answers them. This closed loop introduces a failure mode we call co-cheating: the proposer and solver increasingly agree on shared errors, so internal reward improves without a matching gain in external correctness. A post-hoc audit against source evidence shows co-cheating growing more severe over successive rounds of self-evolution, with pseudo-label correctness stagnating or declining even as the in-loop training signal improves. The most direct mitigation is to verify proposals before training: we introduce multi-sample verification (MSV), which queries the same model three times with the source and three times without it to decide task admission and replace unreliable pseudo-labels. MSV partially reduces false agreement but leaves substantial residual co-cheating and costs six extra labeler generations per candidate. These limitations motivate CrossFit, our main method: it partitions the proposer's source documents into groups A and B; questions generated from A are scored by an auxiliary solver trained only on B, and vice versa. The cross-fitted agreement determines proposer reward, so a same-source pseudo-label cannot be reproduced through the feedback solver, while the original solver's update rule is unchanged. Rerunning the loop with Qwen3.5-4B and Qwen3.5-9B, MSV reduces false-agreement mass from 6.1% to 5.7% and from 8.8% to 7.2%, whereas CrossFit reduces it to 3.0% and 3.7%. Replaying identical proposals with source-excluded feedback further reduces false agreement to 0.4% and 0.1%, isolating feedback ancestry from curriculum changes. Across seven downstream search benchmarks, CrossFit improves average performance over standard coupled self-evolution by 8.8 and 8.4 points and over Search-R1 by 8.7 and 7.8 points at 4B and 9B.
Chinese Translation
自演化搜索代理通过共同优化一个生成问题的提议者和一个回答问题的求解者来构建自己的训练课程。这一闭环引入了一种我们称之为共作弊的失败模式：提议者和求解者在共享错误上越来越一致，因此内部奖励提高，而外部正确性并未相应提升。对源证据的事后审计显示，随着自演化的连续轮次，共作弊现象愈发严重，伪标签的正确性停滞或下降，即使循环内的训练信号有所改善。最直接的缓解方法是在训练前验证提议：我们引入了多样本验证（Multi-Sample Verification, MSV），该方法对同一模型进行三次查询，分别使用源数据和不使用源数据，以决定任务的接受与否，并替换不可靠的伪标签。MSV部分减少了虚假一致性，但仍然留下了相当大的残余共作弊现象，并且每个候选项需要额外六次标注者生成。这些局限性促使我们提出了CrossFit，这是我们的主要方法：它将提议者的源文档分为A组和B组；从A组生成的问题由仅在B组上训练的辅助求解者进行评分，反之亦然。交叉拟合的一致性决定了提议者的奖励，因此同源伪标签无法通过反馈求解者重现，而原求解者的更新规则保持不变。使用Qwen3.5-4B和Qwen3.5-9B重新运行循环，MSV将虚假一致性从6.1%降低到5.7%，从8.8%降低到7.2%，而CrossFit将其降低到3.0%和3.7%。通过排除源数据的反馈重放相同提议，进一步将虚假一致性降低到0.4%和0.1%，将反馈来源与课程变化隔离。在七个下游搜索基准中，CrossFit在标准耦合自演化上平均提高了8.8和8.4点，在Search-R1上提高了8.7和7.8点，分别在4B和9B模型下。
cs.CL / 56 / 2609.39111
Bongard: Training Machine Intuition
Bongard：训练机器直觉
Abstract
Human intelligence relies heavily on learned intuition: recognising patterns and judging situations without explicitly unfolding every intermediate step. We introduce Bongard, an open-weight System One model that treats machine intuition as an independent capability to design and train. A T5Gemma 2 4B-4B encoder-decoder separates reading the evidence from making judgments. The encoder reads the state bidirectionally together with the question instructions, and separate decoder branches share this encoding, so many judgments about the same situation require only one reading of the state. A trained head returns probabilities over the supplied candidates without generating text. Training proceeds in three stages, from supervised judgments to semantic relationships to action outcomes, and each stage updates all 7.09 billion trainable parameters on one Blackwell GPU. Joint-embedding post-training raises accuracy on held-out rephrasings from 75.7% to 85.9%. A sandbox stage then learns outcome distributions from action rollouts and exact oracles, raising accuracy on a frozen sandbox panel from 50.6% to 64.8%. On DecisionBench, the final model reaches 78.05% accuracy over 23,900 decisions and ranks fourth of 61 systems in the public comparison. On one RTX PRO 6000, its median latency is 36 ms for short requests, and 32 questions about one state take 221 ms. Bongard demonstrates that machine intuition can be systematically trained via representation learning and outcome feedback, providing an open, efficient alternative for high-throughput decision workloads.
Chinese Translation
人类智能在很大程度上依赖于学习的直觉：识别模式并在不明确展开每个中间步骤的情况下判断情况。我们介绍了Bongard，一个开放权重的系统一模型，将机器直觉视为一种独立的能力进行设计和训练。T5Gemma 2 4B-4B编码器-解码器将阅读证据与做出判断分开。编码器双向读取状态以及问题指令，独立的解码器分支共享此编码，因此对同一情况的许多判断只需对状态进行一次读取。训练的头部返回对提供候选项的概率，而不生成文本。训练分为三个阶段，从监督判断到语义关系再到行动结果，每个阶段在一台Blackwell GPU上更新所有70.9亿可训练参数。联合嵌入后训练将保留重述的准确率从75.7%提高到85.9%。然后，一个沙盒阶段从行动回滚和精确的神谕中学习结果分布，将冻结沙盒面板的准确率从50.6%提高到64.8%。在DecisionBench上，最终模型在23900个决策中达到了78.05%的准确率，并在公共比较中排名61个系统中的第四位。在一台RTX PRO 6000上，其短请求的中位延迟为36毫秒，而对一个状态的32个问题耗时221毫秒。Bongard展示了机器直觉可以通过表征学习和结果反馈进行系统性训练，为高吞吐量决策工作负载提供了一种开放且高效的替代方案。
cs.CL / 57 / 2609.39118
Diagnosing On-Policy Self-Distillation for Reasoning Language Models
诊断用于推理语言模型的在线自蒸馏
Abstract
On-policy self-distillation (OPSD) has attracted growing interest as a promising approach to improve the reasoning ability of language models. Without external rewards nor a separate stronger teacher, the self-teacher with privileged information could provide dense signals on student's trajectories. However, its behavior in language reasoning remains unclear, with reported outcomes ranging from modest gains to behavioral collapse. In this work, we diagnose OPSD for mathematical reasoning across models spanning 0.6B--8B parameters. We conduct controlled experiments and token-level analyses to fully delve into OPSD. We point out that teacher's signal is shaped by reasoning-mode alignment and the complete teacher prefix, rather than by privileged semantics alone. OPSD improves reasoning only in narrow compatibility regimes. Otherwise, it produces ineffective length growth, stable degradation, or behavioral collapse. Token-level analysis shows that teacher's signal is not stable and does not predict downstream performance. Based on these results, we argue that OPSD is a sensitive algorithm rather than a generally reliable reasoning-improvement post-training method.
Chinese Translation
在线自蒸馏（On-policy self-distillation，OPSD）作为一种有前景的方法，越来越受到关注，旨在提高语言模型的推理能力。在没有外部奖励或单独更强教师的情况下，自教师通过特权信息可以为学生的轨迹提供密集信号。然而，其在语言推理中的表现仍不清楚，报告的结果从适度提升到行为崩溃不等。在本研究中，我们对跨越0.6B至8B参数模型的数学推理进行了OPSD的诊断。我们进行了控制实验和标记级分析，以深入探讨OPSD。我们指出，教师信号是通过推理模式对齐和完整教师前缀塑造的，而不仅仅是通过特权语义。OPSD仅在狭窄的兼容性范围内改善推理。否则，它会导致无效的长度增长、稳定的退化或行为崩溃。标记级分析表明，教师信号并不稳定，无法预测下游性能。基于这些结果，我们认为OPSD是一种敏感算法，而不是一种普遍可靠的推理改进后训练方法。
cs.CL / 58 / 2609.39154
DAGent: Evaluate-then-Grow Planning for Deep Research Agents
DAGent：深度研究代理的评估后增长规划
Abstract
Deep research tasks require agents to navigate large knowledge spaces, synthesize evidence across many sources, and adapt their plans as findings emerge. Directed acyclic graph (DAG)-based multi-agent systems suit this setting because they support parallel execution and isolate each sub-task within a focused dependency context. Yet existing DAG-based agents instantiate a task-level plan before execution and repair the graph only after failures or missing evidence are observed. This Plan-then-Patch strategy is brittle for deep research: the system commits most strongly when its evidence is weakest, and later revisions waste computation on branches that should not have been planned. We propose DAGent, a DAG-based multi-agent framework with Evaluate-then-Grow incremental planning: an Orchestrator grows the task graph one batch at a time, conditioning each expansion on confidence and uncertainty signals from completed nodes. A hierarchical context layer propagates compact QueryDocs by default while preserving full execution traces for on-demand recall. The recorded DAG topology admits structural RL signals that outcome-only recipes cannot define; DAGRPO, a GRPO adaptation, injects topology-conditioned credit on Executor rollouts and a structural compliance regularization on Orchestrator plans. Across BrowseComp-Plus, GAIA, and xbench-DeepSearch, DAGent surpasses the strongest open-source baseline by 5.3 / 5.8 / 2.0 points at the Qwen3-235B-A22B scale, and the lead replicates across four open-source backbones and extends to GPT-5 at 327K context. At the Qwen3-8B scale, DAGRPO improves over a same-budget outcome-only GRPO baseline by 3.0 average Pass@1 points. A same-architecture comparison shows that evidence-conditioned planning reaches higher accuracy at lower per-task token, tool-call, and step footprints than its Plan-then-Patch counterpart. Code: https://github.com/hanwenliu6825/DAGent
Chinese Translation
深度研究任务要求代理在广泛的知识空间中导航，综合来自多个来源的证据，并在发现出现时调整其计划。基于有向无环图（DAG）的多代理系统适合这种环境，因为它们支持并行执行，并在集中依赖上下文中隔离每个子任务。然而，现有的基于DAG的代理在执行之前就实例化了任务级计划，并仅在观察到失败或缺失证据后才修复图。该“先计划后修补”的策略对于深度研究来说是脆弱的：系统在证据最薄弱时最强烈地承诺，而后续的修订则浪费了对本不应计划的分支的计算。我们提出了DAGent，一个基于DAG的多代理框架，采用评估后增长的增量规划：一个协调者一次扩展一个批次的任务图，依据已完成节点的信心和不确定性信号来调整每次扩展。一个层次上下文层默认传播紧凑的QueryDocs，同时保留完整的执行轨迹以供按需回忆。记录的DAG拓扑允许结构化的强化学习信号，这些信号是仅基于结果的配方无法定义的；DAGRPO，一个GRPO的适应，注入了基于拓扑的信用到执行者的回滚中，并对协调者的计划施加结构合规正则化。在BrowseComp-Plus、GAIA和xbench-DeepSearch上，DAGent在Qwen3-235B-A22B规模上超越了最强的开源基线5.3 / 5.8 / 2.0分，并且在四个开源骨干网络上复制了领先优势，并扩展到327K上下文的GPT-5。在Qwen3-8B规模上，DAGRPO在同样预算的仅基于结果的GRPO基线之上提高了3.0的平均Pass@1分数。同架构比较显示，基于证据的规划在每个任务的token、工具调用和步骤占用上达到了更高的准确性，优于其“先计划后修补”的对应物。代码链接：https://github.com/hanwenliu6825/DAGent
cs.CL / 59 / 2609.39189
ViLegalExpert: A Large-Scale Benchmark for Vietnamese Legal Retrieval and Question Answering from Real-World Consultations
ViLegalExpert：越南法律检索与问答的大规模基准测试
Abstract
Trustworthy Legal AI requires systems that can answer legal questions while grounding their responses in authoritative sources. However, existing Vietnamese legal benchmarks provide limited coverage of real-world legal consultations. We introduce \textbf{ViLegalExpert}, a large-scale benchmark constructed from authentic citizen--lawyer consultations, containing over \textbf{172K} questions across \textbf{34 legal domains}, together with professional answers and expert-verified legal evidence. ViLegalExpert supports legal information retrieval, extractive QA, and abstractive QA. Experiments with representative retrieval methods and language models reveal substantial challenges in evidence retrieval and grounded answer generation. While pretrained models perform strongly on QA, hybrid retrieval achieves the best retrieval performance. These results demonstrate the difficulty of mapping naturally expressed legal questions to authoritative provisions and establish ViLegalExpert as a challenging benchmark for reliable Vietnamese Legal AI.
Chinese Translation
可信赖的法律人工智能需要能够回答法律问题并将其回答基于权威来源的系统。然而，现有的越南法律基准对真实世界法律咨询的覆盖有限。我们介绍了 extbf{ViLegalExpert}，这是一个基于真实公民与律师咨询构建的大规模基准，包含超过 extbf{172K} 个问题，涵盖 extbf{34} 个法律领域，并提供专业答案和专家验证的法律证据。ViLegalExpert 支持法律信息检索、提取式问答和抽象式问答。对代表性检索方法和语言模型的实验揭示了证据检索和基于证据的答案生成中的重大挑战。尽管预训练模型在问答任务中表现强劲，但混合检索方法在检索性能上表现最佳。这些结果表明，将自然表达的法律问题映射到权威条款的难度，并确立了 ViLegalExpert 作为一个具有挑战性的基准，以支持可靠的越南法律人工智能。
cs.CL / 60 / 2609.39225
Argument Structure Prediction in Online Conversations: A Comparative Study of Modeling Paradigms and Task Architectures
在线对话中的论证结构预测：建模范式和任务架构的比较研究
Abstract
Argument structure prediction (ASP) constructs complete argument structures from discourse by identifying argumentative units and their relations. While recent work has explored diverse approaches---including unified neural models, multi-step pipelines, and prompt-based large language models (LLMs)---their relative trade-offs remain under-explored, particularly in dialogical settings. We present a systematic evaluation of ASP under strict schema constraints, comparing supervised fine-tuning and prompt-based LLMs across single- and multi-step task architectures, generating complete argument structures from dialogical input end-to-end. We benchmark them on three diverse dialogical corpora adapted from Inference Anchoring Theory into bipolar argument structures. Under a shared evaluation framework, we assess predictive performance, cross-domain generalization, schema compliance, and computational efficiency. Our results show that ASP remains a challenging task, with identifying argumentative relations emerging as the primary bottleneck, largely due to the implicit and context-dependent nature of dialogical argumentation. To facilitate future research, we release our data processing pipeline and end-to-end modeling framework for computational ASP on dialogical corpora.
Chinese Translation
论证结构预测（ASP）通过识别论证单元及其关系，从话语中构建完整的论证结构。尽管近期的研究探索了多种方法——包括统一神经模型、多步骤管道和基于提示的大型语言模型（LLMs）——但它们的相对权衡仍未得到充分探讨，特别是在对话环境中。我们在严格的模式约束下对ASP进行了系统评估，比较了监督微调和基于提示的LLMs在单步骤和多步骤任务架构中的表现，从对话输入端到端生成完整的论证结构。我们在三个不同的对话语料库上进行了基准测试，这些语料库是根据推理锚定理论（Inference Anchoring Theory）改编而成的双极论证结构。在共享的评估框架下，我们评估了预测性能、跨领域泛化、模式遵从性和计算效率。我们的结果表明，ASP仍然是一项具有挑战性的任务，识别论证关系成为主要瓶颈，这在很大程度上是由于对话论证的隐含性和依赖上下文的特性。为了促进未来的研究，我们发布了我们的数据处理管道和用于对话语料库的端到端计算ASP建模框架。
cs.CL / 61 / 2609.39229
RAIM: Robust Aggregation of Inexpensive Models for Hallucination Detection
RAIM：用于幻觉检测的廉价模型鲁棒聚合方法
Abstract
Automatic evaluation of faithfulness increasingly relies on a large language model acting as a judge, yet the most reliable judges are proprietary frontier models, costly and ill-suited to high-throughput monitoring. We investigate whether a panel of cheap open-weight judges (4--9B) can be aggregated to stand in for a frontier one, what the substitution sacrifices, and when it is worth making. We propose RAIM, an aggregation scheme robust to the members' correlated errors, coupling a cross-fitted stacked logistic regression with an admissibility test that, read from the members' own outputs, identifies when aggregating them improves on their best member and stays within reach of the frontier judge. We instantiate RAIM with ten judges from disjoint families across eight faithfulness benchmarks. Against Claude Sonnet, the panel retains a median 93% of its Cohen's $\kappa$ and gives up only 2.9 points of balanced accuracy on average; read as paired differences, it clearly improves on one benchmark and clearly worsens on three (only two by a non-negligible margin), leaving four unresolved. At a sixty-fourth of the frontier's inference price, the operative expense is a one-time in-domain calibration on 50--100 labelled records. The panel is also competitive with purpose-trained detectors on their home benchmarks (within 1.3 accuracy points of GPT-4o and 1.9 of the LLM-AggreFact leader), and beats the strongest one we reran by 6 points on our grounded sets. Whether aggregation pays depends on the members themselves: where several capable members err on different items, the panel improves on its best judge and approaches the frontier; where one dominates, the stacker recovers the leader, and only there does the frontier remain materially ahead. Both conditions are read off the calibration set at no further cost, so a cheap panel can stand in for a frontier one wherever this audit admits it.
Chinese Translation
自动化的忠实度评估越来越依赖于大型语言模型作为评判者，然而最可靠的评判者通常是专有的前沿模型，成本高昂且不适合高通量监控。我们探讨是否可以通过聚合一组廉价的开源权重评判者（4至9亿参数）来替代前沿模型，分析这种替代带来的牺牲以及何时值得采用。我们提出了RAIM，一种对成员模型相关错误具有鲁棒性的聚合方案，该方案结合了交叉拟合的堆叠逻辑回归与可接受性测试，该测试基于成员自身输出判断聚合结果是否优于其最佳成员且接近前沿评判者。我们在八个忠实度基准上，使用来自不同家族的十个评判者实例化RAIM。与Claude Sonnet相比，该模型组保持了中位数93%的Cohen’s κ系数，平均仅损失2.9个百分点的平衡准确率；作为配对差异来看，在一个基准上明显优于前沿模型，在三个基准上表现较差（其中两个差距显著），其余四个基准结果未定。以前沿模型推理成本的六十四分之一计，实际开销仅为一次性在50至100个标注样本上的领域内校准。该模型组在其对应基准上也与专门训练的检测器竞争（与GPT-4o相差不超过1.3个百分点准确率，与LLM-AggreFact领先者相差1.9个百分点），并在我们的有根基准集上比我们重新运行的最强检测器高出6个百分点。聚合是否有利取决于成员模型自身：当多个有能力的成员在不同样本上出错时，模型组优于其最佳评判者并接近前沿模型；当单一成员占优时，堆叠模型能够恢复该领导者，只有此时前沿模型才明显领先。这两种情况均可通过校准集无额外成本地判定，因此在该审计允许的情况下，廉价模型组可以替代前沿模型。
cs.CL / 62 / 2609.39238
4MT-VLM: How Coarse Is a VLMs Cognitive Map?
4MT-VLM：VLM的认知地图有多粗糙？
Abstract
An agent that moves must recognise a place from a viewpoint it has never seen. We introduce 4MT-VLM, a dataset of procedurally generated landscapes, each rendered across five stimulus modes that remove appearance cues while holding layout fixed: shape and colour, shape only, colour only, bare terrain peaks with no objects, and a valley viewpoint that puts the peaks on the horizon. The last condition is commonly used in clinics to probe hippocampal function in human patients. We test this benchmark across sixteen different open and closed-source models and report 4AFC performance, a measure which is also used to grade human participants. We observe that models identify a place from the studied viewpoint but lose it once the camera moves, dropping below the 25% chance level at 135{\deg} where a human observer scores 85%. Frontier models (Gemini 3.8 Flash, GPT-5.6) answer only 39% and 31% of rotated trials correctly, recovering to 85% and 55% only when distractors are moved more than 30 meters apart. Our benchmark demonstrates that while current VLMs possess rudimentary cognitive maps, their spatial resolution remains fundamentally too coarse to maintain a stable, 3D understanding of the world once the viewpoint changes.
Chinese Translation
一个移动的智能体必须能够从一个它从未见过的视角识别一个地方。我们引入了4MT-VLM，一个程序生成的景观数据集，每个景观在五种刺激模式下呈现，这些模式去除了外观线索，同时保持布局不变：形状和颜色、仅形状、仅颜色、没有物体的裸露地形峰，以及一个将峰放置在地平线上的山谷视角。最后一种条件通常在临床中用于探测人类患者的海马功能。我们在十六种不同的开源和闭源模型上测试了这一基准，并报告了4AFC（四个选择的同一性判断）表现，这一指标也用于评估人类参与者。我们观察到模型能够从研究的视角识别一个地方，但一旦相机移动，就会失去这一能力，在135°时的表现低于25%的随机猜测水平，而人类观察者的得分为85%。前沿模型（Gemini 3.8 Flash，GPT-5.6）在旋转试验中仅正确回答39%和31%，只有在干扰物移动超过30米时才恢复到85%和55%。我们的基准表明，尽管当前的VLM具有初步的认知地图，但它们的空间分辨率在视角变化后仍然过于粗糙，无法维持对世界的稳定三维理解。
cs.CL / 63 / 2609.39263
Concept Subspaces Compute Beyond the Logit Lens: A Weights-Only Test for Locating Representations Upstream of Readout
概念子空间超越逻辑透镜的计算：一种仅基于权重的测试用于定位读取上游的表示
Abstract
A concept subspace's effect on model behavior does not establish how it relates to the output readout. We introduce a two-sided geometric diagnostic that measures an extracted subspace's overlap with the dominant right-singular directions of the unembedding matrix, evaluated against output-oriented positive controls. Given an extracted basis, the raw diagnostic requires only model weights. Our testbed is the Format-Agnostic Reasoning Subspace (FARS), a ten-dimensional basis extracted from eighteen reasoning concepts expressed in six surface forms. Across nine rank-matched estimators and twenty-six models, four activation-derived concept estimators carry only 0.38--0.80% mean energy in the top-ten readout span. Final-layer PCA carries 3.56%, exceeding FARS in 25 of 26 models. A same-layer next-token control, evaluated using a fitted linear translator for depth matching, carries approximately thirteen times more energy than FARS, with separation in all 25 tested models. Re-extracting FARS on ten disjoint concepts yields 62--100% cross-format retrieval across twenty-four generative models, demonstrating transfer of the extraction procedure rather than a fixed basis. A complementary four-model, three-seed intervention study finds model-dependent source-directed effects that remain well below full-vector replacement. Together, the geometry and intervention controls distinguish concept structure from dominant readout directions while limiting claims of causal sufficiency.
Chinese Translation
概念子空间对模型行为的影响并不能确定其与输出读取的关系。我们引入了一种双侧几何诊断方法，测量提取子空间与未嵌入矩阵的主右奇异方向的重叠程度，并与面向输出的正控制进行评估。给定一个提取的基，原始诊断仅需要模型权重。我们的测试平台是格式无关推理子空间（Format-Agnostic Reasoning Subspace, FARS），这是一个从以六种表面形式表达的十八个推理概念中提取的十维基。在九个等级匹配的估计器和二十六个模型中，四个基于激活的概念估计器在前十个读取范围内仅占0.38%到0.80%的平均能量。最终层的主成分分析（PCA）占3.56%，在26个模型中超过FARS。使用拟合的线性转换器进行深度匹配的同层下一个令牌控制，其能量约为FARS的十三倍，在所有25个测试模型中均有分离。在十个不相交概念上重新提取FARS，在二十四个生成模型中实现了62%到100%的跨格式检索，展示了提取过程的转移而非固定基。一个补充的四模型、三种种子干预研究发现模型依赖的源导向效应，仍远低于全向量替换。总体而言，几何和干预控制将概念结构与主读取方向区分开，同时限制了因果充分性的主张。
cs.CL / 64 / 2609.39346
Offline Guidance, Online Reasoning: Reusing LLM Feedback for Small Language Models
离线指导，在线推理：重用大型语言模型反馈以提升小型语言模型
Abstract
Large language models (LLMs) offer strong reasoning capabilities but are often costly to access through commercial APIs, while small language models (SLMs) are easier to deploy locally yet remain weaker in reasoning. This capability-deployment gap has motivated LLM-SLM collaboration, which aims to improve SLM reasoning using LLM capabilities while preserving the deployment advantages of SLMs. Existing approaches mainly follow two paradigms. Knowledge distillation uses LLM-generated answers and reasoning trajectories to train SLMs offline, but requires parameter updates and additional training. Alternatively, online collaboration routes difficult problems to an LLM or leverages LLM-generated guidance and corrections when an SLM encounters difficulties. Although effective, online collaboration requires repeated LLM access. Moreover, the guidance produced for a particular problem is discarded after inference and cannot benefit subsequent problems involving similar reasoning states. In the paper, we focus on a more constrained setting in which the LLM is accessed only offline, the SLM parameters remain fixed, and online inference is performed solely by the SLM. To this end, we propose Reusable Latent Correction (RLC), which converts one-off natural-language guidance from a black-box LLM into persistent corrective experiences in the hidden space of an SLM. RLC stores these experiences in an external bank and retrieves them according to the SLM's current reasoning state, enabling the SLM to reuse LLM-derived corrections during inference without any online LLM calls. Experiments across multiple reasoning benchmarks and SLM scales show that RLC consistently improves SLM reasoning without parameter updates or online LLM calls. Code is available at https://github.com/ZBH031/reusable-latent-correction.
Chinese Translation
大型语言模型（LLMs）提供强大的推理能力，但通过商业API访问的成本往往较高，而小型语言模型（SLMs）更易于本地部署，但在推理能力上相对较弱。这种能力与部署之间的差距促使了LLM-SLM的协作，旨在利用LLM的能力提升SLM的推理能力，同时保留SLM的部署优势。现有方法主要遵循两种范式。知识蒸馏利用LLM生成的答案和推理轨迹对SLM进行离线训练，但需要参数更新和额外的训练。另一种选择是在线协作，当SLM遇到困难时，将复杂问题转交给LLM，或利用LLM生成的指导和修正。尽管有效，在线协作需要反复访问LLM。此外，为特定问题生成的指导在推理后被丢弃，无法为后续涉及类似推理状态的问题提供帮助。本文关注一种更受限的设置，其中LLM仅在离线状态下访问，SLM参数保持固定，在线推理仅由SLM执行。为此，我们提出了可重用潜在修正（Reusable Latent Correction, RLC），将来自黑箱LLM的一次性自然语言指导转化为SLM隐藏空间中的持久性修正经验。RLC将这些经验存储在外部库中，并根据SLM当前的推理状态进行检索，使SLM在推理过程中能够重用LLM派生的修正，而无需任何在线LLM调用。在多个推理基准和SLM规模上的实验表明，RLC在不进行参数更新或在线LLM调用的情况下，始终提升SLM的推理能力。代码可在 https://github.com/ZBH031/reusable-latent-correction 获取。
cs.CL / 65 / 2609.39358
Working Around the Compute Ceiling: Byte-Exact Memory in Galahad Makes LLM Reading a One-Time Cost LLM Reading a One-Time Cost
绕过计算上限：Galahad中的字节精确内存使得LLM阅读成为一次性成本
Abstract
A transformer language model performs a bounded amount of computation per token, and recent work by Vishal Sikka, former CEO of Infosys, argues that this bound limits which tasks a model can carry out or verify (arXiv:2507.07505). We ask how much of the budget beneath that ceiling is spent on work the model has already done. Serving is stateless across requests: a model that answers a second question about a document recomputes the document's attention state from the first token. On seven real-world datasets, 98.7% of prompt tokens were text the model had already read. We present Galahad, a memory layer for vLLM, SGLang and llama.cpp that makes this reading a one-time cost. Taliesin saves the model's key-value (KV) state for a block of text and loads it on the next request that contains the same bytes, instead of recomputing it. Blaise keeps the documents themselves and passes the model only the section a question needs. On a recall test with 100 facts hidden in a 97,000-token corpus (Gemma 4 31B), Taliesin alone let the model attend to the whole corpus and answered 98 of 100 on llama.cpp at 3.0 s and 572 J per question, against 10 of 100, 9.3 s and 2,754 J for the same model without Galahad, which could hold only the last 12,000 tokens. With Blaise added, the model read about 668 tokens per question and answered 100 of 100 on all three runtimes at 0.59-0.64 s and 200-213 J; a tuned RAGFlow pipeline answered 77. Storing the corpus is a one-time cost of about 100 s and 28 kJ, whose energy is recovered after 13 questions. Restored state is bit-identical: all 262,144 output logits matched after restart, rehydration and hot-load. Galahad worked with all 30 models we tested under vLLM, and it fails closed: any load that does not pass its checks is recomputed. Together these results move LLM serving from stateless to stateful inference.
Chinese Translation
变压器语言模型在每个标记上执行有限的计算量，Infosys前首席执行官Vishal Sikka的最新研究认为，这一限制影响了模型能够执行或验证的任务（arXiv:2507.07505）。我们探讨在这一上限下，模型已完成的工作占用了多少预算。服务在请求之间是无状态的：一个模型在回答关于文档的第二个问题时，会从第一个标记重新计算文档的注意力状态。在七个真实世界的数据集中，98.7%的提示标记是模型已经阅读过的文本。我们提出了Galahad，这是一个用于vLLM、SGLang和llama.cpp的内存层，使得这种阅读成为一次性成本。Taliesin为一段文本保存模型的键值（KV）状态，并在下一个包含相同字节的请求中加载它，而不是重新计算。Blaise保留文档本身，仅将问题所需的部分传递给模型。在一个包含97,000个标记的语料库（Gemma 4 31B）中进行的回忆测试中，Taliesin使模型能够关注整个语料库，并在llama.cpp上以3.0秒和572焦耳的成本回答了100个问题中的98个，而没有Galahad的同一模型只能回答100个中的10个，耗时9.3秒和2,754焦耳。添加Blaise后，模型每个问题读取约668个标记，并在所有三个运行时间内以0.59-0.64秒和200-213焦耳的成本回答了100个中的100个；调优后的RAGFlow管道回答了77个。存储语料库的成本约为100秒和28千焦耳，能量在13个问题后得到回收。恢复的状态是比特级相同的：所有262,144个输出logits在重启、再水合和热加载后完全匹配。Galahad在我们测试的所有30个vLLM模型中均有效，并且在失败时会关闭：任何未通过其检查的负载将被重新计算。这些结果共同将LLM服务从无状态推向有状态推理。
cs.CL / 66 / 2609.39365
Ready2Blend: From Natural-Language Instructions to Composable Alignment Prompts
Ready2Blend：从自然语言指令到可组合的对齐提示
Abstract
Continual alignment requires LLMs to adapt to new requirements without forgetting previously acquired behaviors. Natural-language instructions are flexible and composable but offer only indirect control, whereas post-training provides stronger adaptation at the cost of repeated parameter updates. We introduce Ready2Blend, which combines the flexibility of natural language with learned alignment. AlignFormer maps each requirement to a fixed-length alignment prompt stored in a modular prompt bank, while the backbone and prior prompts remain frozen. Composability regularization transfers the semantic geometry of textual requirements into prompt space, enabling inference-time blending and reweighting. Across two practical continual alignment settings, Ready2Blend is the only frozen-backbone method that matches post-training-based alignment methods, reaching $93.1$-$98.5\%$ of a joint-training reference with competitive retention, while requiring only a few prompt tokens and up to $4.3\times$ less training time. Its modular design further enables weighted personalization and order-free composition without retraining. Code will be released upon acceptance.
Chinese Translation
持续对齐要求大型语言模型（LLMs）在适应新需求的同时不遗忘之前获得的行为。自然语言指令灵活且可组合，但仅提供间接控制，而后训练则在重复参数更新的代价下提供更强的适应性。我们提出了Ready2Blend，它将自然语言的灵活性与学习到的对齐相结合。AlignFormer将每个需求映射到存储在模块化提示库中的固定长度对齐提示，同时主干和先前的提示保持不变。可组合性正则化将文本需求的语义几何转移到提示空间，使推理时的混合和重加权成为可能。在两个实际的持续对齐设置中，Ready2Blend是唯一一种冻结主干的方法，其性能与基于后训练的对齐方法相匹配，达到联合训练参考的93.1%-98.5%，同时保持竞争性的保留率，仅需少量提示标记和最多4.3倍的训练时间。其模块化设计进一步支持加权个性化和无序组合，而无需重新训练。代码将在接受后发布。
cs.CL / 67 / 2609.39368
Making Grid Beam Search Less Greedy
减少网格束搜索的贪婪性
Abstract
A common formalism for constraining the output of autoregressive text generation models involves lexical constraints, words or phrases which are required to occur in the generated text. DFA-constrained beam search and grid beam search are two widely used paradigms for decoding from autoregressive models while enforcing lexical constraints. As the former approach requires a number of forward passes exponential in the number of constraint tokens, it is often dispreferred to the latter, which requires only linearly many forward calls. However, while grid beam search achieves an exponential speedup, it does so in a manner which does not treat all of the constraints equally. In this paper, we demonstrate that grid beam search is biased to incorporate easier-to-satisfy constraints first, leaving harder constraints to the end of the sequence. This contrasts with DFA-constrained beam search, which exhibits no such bias. To address this shortcoming, we propose fair grid beam search, a modification to grid beam search which avoids this bias while still requiring only linearly many forward passes. Experimentally, we confirm grid beam search's bias on two constrained generation tasks, finding significant differences in how it orders constraint tokens as compared to DFA-constrained beam search and fair grid beam search. Furthermore, we find that fair grid beam search not only fixes grid beam search's bias, but finds higher-probability strings in the process.
Chinese Translation
约束自回归文本生成模型输出的常见形式涉及词汇约束，即要求在生成文本中出现的单词或短语。DFA约束束搜索和网格束搜索是两种广泛使用的解码范式，用于从自回归模型中解码，同时强制执行词汇约束。前者方法需要的前向传递次数与约束标记数量呈指数关系，因此通常不如后者受欢迎，后者仅需要线性数量的前向调用。然而，尽管网格束搜索实现了指数级的加速，但其方式并未平等对待所有约束。本文展示了网格束搜索倾向于优先满足更易满足的约束，而将更难满足的约束留到序列的末尾。这与DFA约束束搜索形成对比，后者没有这种偏见。为了解决这一缺陷，我们提出了公平网格束搜索，这是一种对网格束搜索的修改，避免了这种偏见，同时仍然只需要线性数量的前向传递。通过实验，我们确认了网格束搜索在两个约束生成任务中的偏见，发现其在约束标记的排序上与DFA约束束搜索和公平网格束搜索存在显著差异。此外，我们发现公平网格束搜索不仅修正了网格束搜索的偏见，还在此过程中找到了更高概率的字符串。
cs.CL / 68 / 2609.39369
Exploring Heterogeneous Model Merging Approach for Complex Knowledge Transfer
探索复杂知识转移的异构模型合并方法
Abstract
Specialized models encode task-oriented behavior, but transferring that behavior to a general language model usually requires training, distillation, or representation alignment. We study whether such ability can instead be transferred directly at the parameter level. We apply two existing training-free heterogeneous merging methods, previously shown to transfer knowledge between general language models, to specialist-to-general transfer, projecting a specialist donor into the recipient's shape and interpolating backbone parameters without gradient updates or semantic alignment. Intersection-Merge (IM) injects a prefix-aligned donor slice matching the recipient shape, while Activate-Prune-Merge (APM) uses forward-pass activation statistics to select which donor dimensions to retain before injection. Across embedding, reranking, reward modeling, and MoE code-specialist transfer, both methods improve the general recipient, showing that simple heterogeneous merging can move capabilities across diverse specialist roles.
Chinese Translation
专门化模型编码了面向任务的行为，但将这种行为转移到通用语言模型通常需要训练、蒸馏或表示对齐。我们研究这种能力是否可以直接在参数级别进行转移。我们应用了两种现有的无训练异构合并方法，这些方法之前已被证明可以在通用语言模型之间转移知识，来实现专门模型到通用模型的转移，将专门模型的捐赠者投影到接收者的形状，并在没有梯度更新或语义对齐的情况下插值骨干参数。Intersection-Merge (IM) 注入与接收者形状匹配的前缀对齐捐赠者切片，而 Activate-Prune-Merge (APM) 则使用前向传播激活统计来选择在注入前保留哪些捐赠者维度。在嵌入、重排序、奖励建模和 MoE 代码专门转移中，这两种方法均改善了通用接收者，表明简单的异构合并可以在不同的专门角色之间转移能力。
cs.CL / 69 / 2609.39385
TTLab at Daleel 2026: STAR-Ar, Sequence Tagging for Argument Recognition in Arabic
TTLab在Daleel 2026：STAR-Ar，阿拉伯语论证识别的序列标注
Abstract
Argument Mining (AM) is a critical NLP task that remains significantly under-resourced in Arabic. This paper presents $\testtt{STAR-Ar}$, a BERT-BiLSTM-CRF architecture for argument discourse detection and classification, as our system for Daleel 2026, the inaugural Arabic argument mining shared task. The task requires the identification and classification of argumentative discourse units (ADUs) in debate and editorial texts.We jointly model these two objectives as a token-level sequence labeling task using a BERT-BiLSTM-CRF architecture that combines contextual transformer embeddings with structural transition constraints to support accurate span detection. $\testtt{STAR-Ar}$ achieves an F1-score of 72.69 on validation and 73.7 on test data. Our domain-specific analysis shows that models trained exclusively on editorials underperform those trained on debates, a disparity we primarily attribute to the smaller size of the editorial dataset. The code for $\testtt{STAR-Ar}$ is available at ${\href{https://github.com/ENTAILab/daleel_2026_Arabic-Argumentative-Discourse-Mining}{\faGithub~TTLab at Daleel 2026}}$
Chinese Translation
论证挖掘（Argument Mining, AM）是一个关键的自然语言处理（NLP）任务，但在阿拉伯语中仍然资源严重不足。本文介绍了$ esttt{STAR-Ar}$，一种用于论证话语检测和分类的BERT-BiLSTM-CRF架构，作为我们在Daleel 2026这一首届阿拉伯语论证挖掘共享任务中的系统。该任务要求识别和分类辩论和社论文本中的论证话语单元（ADUs）。我们将这两个目标联合建模为一个基于标记的序列标注任务，使用结合了上下文变换器嵌入和结构转移约束的BERT-BiLSTM-CRF架构，以支持准确的跨度检测。$ esttt{STAR-Ar}$在验证集上的F1-score达到72.69，在测试数据上为73.7。我们的领域特定分析表明，仅在社论上训练的模型表现不如在辩论上训练的模型，我们主要将这种差异归因于社论数据集的规模较小。$ esttt{STAR-Ar}$的代码可在${ ext{https://github.com/ENTAILab/daleel_2026_Arabic-Argumentative-Discourse-Mining}}$找到。
cs.CL / 70 / 2609.39420
QuantCode Model: Specializing Language Models for Executable Algorithmic Trading Code
QuantCode 模型：专门化语言模型以生成可执行的算法交易代码
Abstract
Large language models are strong general-purpose code generators, but executable algorithmic trading remains a demanding specialization target: a model must translate a natural-language strategy specification into correct program logic for a specialized trading framework, execute on historical data, produce trades, and remain semantically faithful to the request. We study two complementary mechanisms for specializing language models for this setting: continued pretraining on algorithmic-trading framework code and supervised fine-tuning (SFT) on agent-validated request-to-code pairs. Evaluation is centered on QuantCode-Bench, our 400-task benchmark for Backtrader strategy generation, together with a repository-level SWE-bench-like track. Continued pretraining improves single-turn Judge Pass from 41.5% to 47.5% for Qwen3.5-397B-A17B and from 27.8% to 33.0% for Qwen3.6-35B-A3B. SFT applied after continued pretraining yields a larger gain for Qwen3.6-35B-A3B, reaching 58.2% Judge Pass and 83.5% successful backtests; in agentic evaluation it raises first-turn success from 22.3% to 58.3% and final success after up to 10 turns from 47.5% to 79.5%. Continued pretraining alone improves first-turn agentic success but lowers final success after repair from 47.5% to 32.5%, consistent with degraded instruction following, whereas SFT improves both. We also identify a capability-retention failure: domain specialization degrades parser-conformant structured tool calling, and targeted recovery SFT restores tool-call formatting but not the base checkpoint's repository-level agent performance. The results show that framework-oriented pretraining, validated SFT, and explicit capability-retention evaluation address distinct failure modes in domain-specific executable code generation.
Chinese Translation
大型语言模型是强大的通用代码生成器，但可执行的算法交易仍然是一个要求严格的专业化目标：模型必须将自然语言策略规范转换为适用于特定交易框架的正确程序逻辑，在历史数据上执行，产生交易，并保持与请求的语义一致性。我们研究了两种互补机制，以专门化语言模型以适应这一环境：在算法交易框架代码上进行持续预训练和在经过代理验证的请求到代码对上进行监督微调（SFT）。评估集中在 QuantCode-Bench，这是我们为 Backtrader 策略生成创建的 400 项任务基准，以及一个类似于仓库级 SWE-bench 的轨道。持续预训练使 Qwen3.5-397B-A17B 的单轮 Judge Pass 从 41.5% 提高到 47.5%，而 Qwen3.6-35B-A3B 的 Judge Pass 从 27.8% 提高到 33.0%。在持续预训练后应用 SFT 为 Qwen3.6-35B-A3B 带来了更大的提升，Judge Pass 达到 58.2%，成功回测率为 83.5%；在代理评估中，首次成功率从 22.3% 提高到 58.3%，经过最多 10 轮后的最终成功率从 47.5% 提高到 79.5%。仅持续预训练提高了首次代理成功率，但在修复后的最终成功率从 47.5% 降低到 32.5%，这与指令遵循能力下降一致，而 SFT 则同时改善了这两者。我们还识别出一种能力保留失败：领域专业化降低了解析器符合的结构化工具调用，针对性的恢复 SFT 恢复了工具调用格式，但未能恢复基础检查点的仓库级代理性能。结果表明，面向框架的预训练、经过验证的 SFT 和明确的能力保留评估解决了领域特定可执行代码生成中的不同失败模式。
cs.CL / 71 / 2609.39446
DuplexAct-Bench: Broadening Full-Duplex Speech Evaluation toward Proactive Interaction across Diverse Behavioral Requirements
DuplexAct-Bench：拓宽全双工语音评估以适应多样化行为需求的主动交互
Abstract
Existing full-duplex speech benchmarks cover only subsets of real-time interaction behaviors, often under limited contextual conditions. We introduce DuplexAct-Bench, a bilingual benchmark that systematically covers six complementary behaviors, from interruption and yielding to proactive initiation, active silence, and backchanneling, across Pre-session, In-session, and No-explicit conditions. Across 1,290 English and Chinese streaming trials, we evaluate 12 full-duplex speech systems on both Timing and Content. Results reveal substantial variation across behaviors, conditions, and systems, as well as frequent mismatches between semantic quality and behavioral timing. These findings show that current systems remain far from robustly managing when, whether, and how to participate as real-time interaction unfolds. Project page: https://alitaxky.icu/DuplexAct-Bench/
Chinese Translation
现有的全双工语音基准仅涵盖实时交互行为的部分子集，通常在有限的上下文条件下进行评估。我们提出了DuplexAct-Bench，这是一个系统性涵盖六种互补行为的双语基准，从打断和让步到主动发起、主动沉默和回馈，在会前、会中和无明确条件下进行评估。在1290个英语和中文的流媒体试验中，我们对12个全双工语音系统在时间和内容方面进行了评估。结果显示，不同行为、条件和系统之间存在显著差异，以及语义质量与行为时机之间的频繁不匹配。这些发现表明，当前系统在实时交互展开时，仍远未能够稳健地管理何时、是否以及如何参与。项目页面：https://alitaxky.icu/DuplexAct-Bench/
cs.CL / 72 / 2609.39447
Synthetic Data Characterization via Training Dynamics
通过训练动态对合成数据进行特征化
Abstract
Interpreting properties of LLM-generated data is important for understanding its utility and limitations across learning tasks. In this work, we characterize synthetic data through sample-level learnability, studying variation among LLM families and scales, alongside human-written data as a reference. We first generate synthetic datasets spanning single- and multi-label classification, labeling, and tree prediction tasks. We then derive empirical data distributions from encoder training dynamics for both machine and organic data, and estimate the robustness of these distributions across encoders. Finally, we evaluate how data selection strategies based on these learnability signals affect both data sources differently.
Chinese Translation
解释大型语言模型（LLM）生成的数据特性对于理解其在学习任务中的效用和局限性至关重要。在本研究中，我们通过样本级可学习性对合成数据进行特征化，研究不同LLM家族和规模之间的变异，同时将人类编写的数据作为参考。我们首先生成涵盖单标签和多标签分类、标注以及树预测任务的合成数据集。接着，我们从编码器训练动态中推导出机器数据和有机数据的经验数据分布，并评估这些分布在不同编码器之间的稳健性。最后，我们评估基于这些可学习性信号的数据选择策略如何对这两种数据源产生不同的影响。
cs.CL / 73 / 2609.39460
Right-Wing Rock or Just Rock? A Computational Linguistic Analysis of Frei.Wild
右翼摇滚还是仅仅是摇滚？对Frei.Wild的计算语言学分析
Abstract
Rechtsrock is a subgenre of rock music that spreads right-wing ideology, often instrumentalized to recruit adolescents into the radical scene. Monitoring institutions counteract this by manually examining and, in some cases, banning extremist content; however, there are border cases that evade regulation. We present a study aimed at determining whether such a case, the band Frei.Wild, should be classified as politically right-leaning or as part of the general German rock genre. We sampled a German rock dataset and created a corpus for right-wing rock to use as reference in this analysis and found that we can confirm the intuitions from previous investigations that Frei.Wild successfully maintains an ambiguity with regard to their political affiliation. However, the tendency is towards the right-wing spectrum. Lexical analyses reveal nationalistic narratives and two high-performing classifiers (up to 97% ROC-AUC score) label more than half of their songs as right-wing extremist. Our analysis provides insight into how computational methods can improve the process of identifying right-wing extremist tendencies in music, especially in borderline cases like Frei.Wild. The code and data are made available for future research.
Chinese Translation
右翼摇滚是传播右翼意识形态的摇滚音乐子类型，常被用作招募青少年进入激进场景的工具。监测机构通过手动检查和在某些情况下禁止极端内容来对抗这一现象；然而，仍存在一些边界案例逃避监管。我们进行了一项研究，旨在确定乐队Frei.Wild是否应被归类为政治右倾或属于一般德国摇滚类型。我们采样了一个德国摇滚数据集，并创建了一个右翼摇滚语料库作为本分析的参考，发现我们可以确认之前研究的直觉，即Frei.Wild成功地保持了其政治归属的模糊性。然而，倾向性更偏向于右翼光谱。词汇分析揭示了民族主义叙事，并且两个高效分类器（最高可达97%的ROC-AUC分数）将其超过一半的歌曲标记为右翼极端主义。我们的分析提供了对计算方法如何改善识别音乐中右翼极端主义倾向过程的洞察，特别是在像Frei.Wild这样的边界案例中。代码和数据将提供给未来的研究。
cs.CL / 74 / 2609.39514
Spike-driven Vision-Language-Action Model
脉冲驱动的视觉-语言-动作模型
Abstract
Vision-language-action (VLA) models bridge multimodal understanding and robotic control, advancing the dominant paradigm for embodied intelligence. However, most existing models rely on large Transformers, whose latency and energy costs hinder deployment on resource-constrained platforms. Through sparse event-driven computation, spiking neural networks offer a promising paradigm for high-performance and energy-efficient computing. Here, we propose the first Spike-driven VLA framework enabling end-to-end direct training for robotic manipulation, which mainly comprises three core components. First, we develop spiking visual and instruction encoders for multimodal perception, encoding visual observations and language instructions into sparse, reliable spike representations for subsequent cross-modal fusion. Then, we introduce Multi-Winner Spike Fusion for instruction-guided scene understanding, using bidirectional top-$k$ winner-take-all spike routing to suppress background interference and yield fused memory. Finally, we propose a Spike Action Chunking Transformer that incorporates spiking cross-attention over the fused memory and the current robot state, enabling efficient end-to-end generation of continuous action chunks for robotic control. Extensive experiments on LIBERO and Meta-World demonstrate that Spike-driven VLA achieves competitive performance with fewer parameters and lower estimated inference energy than conventional VLA models. This work establishes a foundational framework for neuromorphic VLA modeling, paving the way for future advances in resource-efficient embodied intelligence.
Chinese Translation
视觉-语言-动作（VLA）模型连接多模态理解与机器人控制，推动了具身智能的主流范式。然而，大多数现有模型依赖于大型变换器（Transformers），其延迟和能耗成本阻碍了在资源受限平台上的部署。通过稀疏事件驱动计算，脉冲神经网络提供了一种高性能和能效计算的有希望的范式。在此，我们提出了第一个脉冲驱动的VLA框架，支持机器人操作的端到端直接训练，主要由三个核心组件组成。首先，我们开发了脉冲视觉和指令编码器用于多模态感知，将视觉观察和语言指令编码为稀疏、可靠的脉冲表示，以便后续的跨模态融合。然后，我们引入了多赢家脉冲融合（Multi-Winner Spike Fusion）用于指令引导的场景理解，利用双向前$k$赢家通吃脉冲路由抑制背景干扰并生成融合记忆。最后，我们提出了一种脉冲动作分块变换器（Spike Action Chunking Transformer），它结合了对融合记忆和当前机器人状态的脉冲交叉注意力，能够高效地端到端生成连续的动作块以进行机器人控制。在LIBERO和Meta-World上的广泛实验表明，脉冲驱动的VLA在参数更少和估计推理能耗更低的情况下，达到了与传统VLA模型竞争的性能。这项工作为类脑VLA建模奠定了基础框架，为未来在资源高效的具身智能方面的进展铺平了道路。
cs.CL / 75 / 2609.39533
CATCH: A Controllable Analysis Testbed for Reward Hacking in Coding RL
CATCH：一个可控的奖励黑客分析测试平台用于编码强化学习
Abstract
During reinforcement learning with verifiable rewards (RLVR), large language models (LLMs) can exploit loopholes in their environments to obtain high rewards without improving the intended capabilities, i.e., reward hacking. Despite its risks to training efficiency and safety, monitoring and mitigating reward hacking during training remain challenging, which is limited by a lack of testbeds that reproduce hacking and reliably identify it. We introduce CATCH, a controllable testbed for studying reward hacking in coding RL. CATCH deliberately exposes environmental loopholes and provides execution-based gold labels by comparing success under a vulnerable evaluator with task correctness under an independent audit. It also can control the model's initial hacking tendency through supervised fine-tuning data mixtures and the difficulty of earning rewards through reward designing, enabling systematic comparisons of hacking dynamics and interventions. Experiments show that CATCH can produce diverse RL training trajectories with clear reward hacking, and analyses demonstrate that both initial models and reward difficulties shape the emergence of reward hacking. We further evaluate the effectiveness of different reward hacking detection and mitigation methods. A key finding is that a chain-of-thought monitor initially suppresses hacking, but this protection erodes as the policy model learn to mislead the monitor with code comments. This highlights the need to evaluate hacking mitigations throughout training with CATCH. The source code and resources are publicly released at https://github.com/THUAIS-Lab/CATCH.
Chinese Translation
在可验证奖励的强化学习（RLVR）过程中，大型语言模型（LLMs）能够利用环境中的漏洞获取高额奖励，而不提升其预期能力，即奖励黑客。尽管奖励黑客对训练效率和安全性构成风险，但在训练过程中监测和缓解奖励黑客仍然具有挑战性，这主要受限于缺乏能够重现黑客行为并可靠识别其存在的测试平台。我们引入了CATCH，一个用于研究编码强化学习中奖励黑客的可控测试平台。CATCH故意暴露环境漏洞，并通过将脆弱评估者下的成功与独立审计下的任务正确性进行比较，提供基于执行的金标准标签。它还可以通过监督微调数据混合和奖励设计控制模型的初始黑客倾向以及获得奖励的难度，从而实现对黑客动态和干预措施的系统比较。实验表明，CATCH能够产生多样化的强化学习训练轨迹，并清晰展示奖励黑客的现象，分析结果表明，初始模型和奖励难度共同影响奖励黑客的出现。我们进一步评估了不同奖励黑客检测和缓解方法的有效性。一个关键发现是，思维链监控器最初抑制了黑客行为，但随着策略模型学习用代码注释误导监控器，这种保护逐渐减弱。这突显了在使用CATCH进行训练时评估黑客缓解措施的必要性。源代码和资源已公开发布在 https://github.com/THUAIS-Lab/CATCH。
cs.CL / 76 / 2609.39572
Compact Language, Complex Model Shifts: How and Where Ambiguity and Underspecification Affect LLMs
紧凑语言，复杂模型转变：模糊性和不充分性如何以及在何处影响大型语言模型
Abstract
We analyze how lexical ambiguity and underspecification affect language model training. We create artificial homonyms and artificial hypernyms as pseudowords and analyze the generative performance of language models as they are trained with increasing amounts of these ambiguous or underspecified pseudoword types. We further analyze whether the models disambiguate ambiguous or underspecified statements and provide a first mechanistic account of how ambiguity and disambiguation are represented internally. Our main results show that both ambiguity and underspecification increase model performance in ways that scale with their influence on the language's type-token ratio. However, the accuracy of generating sequences containing ambiguous words or their synonyms decreases compared to other texts. We also show that internal representations of pseudowords reflect disambiguation of pseudo-homonyms, but underspecification of pseudo-hypernyms is maintained during the generative process.
Chinese Translation
我们分析了词汇模糊性和不充分性如何影响语言模型的训练。我们创造了人工同义词和人工上义词作为伪词，并分析了随着这些模糊或不充分伪词类型的数量增加，语言模型的生成性能。我们进一步分析模型是否能够消歧义模糊或不充分的陈述，并提供了模糊性和消歧义在内部如何表示的初步机制性解释。我们的主要结果表明，模糊性和不充分性以与其对语言类型-标记比率的影响成比例的方式提高模型性能。然而，生成包含模糊词或其同义词的序列的准确性相较于其他文本有所下降。我们还展示了伪词的内部表示反映了伪同义词的消歧义，但在生成过程中维持了伪上义词的不充分性。
cs.CL / 77 / 2609.39578
Thinking Outside the Box: Can Language Models Rely on External Guidance Selectively?
跳出框框思考：语言模型能否选择性地依赖外部指导？
Abstract
Agent harnesses often improve language models with human-designed workflows, but as models grow more capable, unreliable guidance can increasingly constrain their execution. We call the ability to benefit from useful guidance while overriding unreliable guidance thinking outside the box. We introduce Box$^2$-Bench, which holds the model and task fixed while varying workflow reliability to isolate how models regulate their reliance on guidance. On Box$^2$-Bench, frontier models often benefit from reliable guidance but remain vulnerable when it is misleading or becomes unreliable. To test whether this capability can be learned, we train two open-weight models using bad workflows, reserving good workflows for evaluation. We explore two complementary training strategies: counterfactual supervised fine-tuning improves robustness, while outcome-based reinforcement learning can shift the balance toward greater use of helpful workflows. We further find that this behavior extends beyond workflows to other forms of external information, improving peer correction and robustness to corrupted memory. Together, our results identify selective reliance on fallible external information as a dimension of agent reliability not captured by task performance alone.
Chinese Translation
代理利用通常通过人类设计的工作流程来提升语言模型的性能，但随着模型能力的增强，不可靠的指导可能会越来越限制其执行。我们称这种在受益于有用指导的同时能够忽略不可靠指导的能力为“跳出框框思考”。我们引入了 Box$^2$-Bench，该基准在固定模型和任务的同时，改变工作流程的可靠性，以隔离模型如何调节其对指导的依赖。在 Box$^2$-Bench 上，前沿模型通常能从可靠的指导中受益，但在指导误导或变得不可靠时仍然脆弱。为了测试这种能力是否可以学习，我们使用不良工作流程训练了两个开放权重模型，并将良好工作流程保留用于评估。我们探索了两种互补的训练策略：反事实监督微调提高了鲁棒性，而基于结果的强化学习则可以将重心转向更多地使用有帮助的工作流程。我们进一步发现，这种行为不仅限于工作流程，还扩展到其他形式的外部信息，提高了同伴纠正和对损坏记忆的鲁棒性。总的来说，我们的结果将对可出错外部信息的选择性依赖识别为代理可靠性的一维特征，而不仅仅是通过任务表现来捕捉。
cs.CL / 78 / 2609.39608
Is This Evidence Decision-Critical? Learning to Verify Rule-Governed Decisions
这些证据是否对决策至关重要？学习验证规则驱动的决策
Abstract
Rule-based reasoning, as in eligibility checks and contract reviews, requires language models to assess evidence against individual conditions and combine their judgments under explicit rules. Errors in evidence assessment can leave a decision unchanged, but misinterpreting or overlooking decision-critical evidence can reverse it. Identifying such evidence allows more capable models to focus on checking the corresponding condition judgments, supporting accurate and safe decisions. Recognizing the evidence's criticality requires understanding how evidence affects a condition judgment and how that judgment affects the decision. To achieve the goal, we propose a INTERvention-based imPACT learning framework (InterPact), which enables counterfactual verification of evidence criticality in rule-governed decisions. Specifically, its evidence intervention constructor generates training pairs for a propagation verifier by editing case facts with a frozen language model while holding rules and non-target conditions fixed. Human-reviewed labels record the resulting condition and decision changes, while complete state-to-decision mappings supervise consequences beyond the observed edit. During training, the verifier weights learned conditional decision predictions by evidence-based condition probabilities through a fixed composition operation, propagating decision-change supervision into the base model. At inference, the trained base model directly judges criticality from the original case and target evidence, without human or stronger-model supervision. On single-case evidence criticality verification over adapted rule-governed decision cases, InterPact achieves 68.28% accuracy, outperforming all six baselines. These results support learned decision sensitivity as a basis for prioritizing evidence checks.
Chinese Translation
基于规则的推理，如资格检查和合同审查，要求语言模型根据个别条件评估证据，并在明确的规则下结合其判断。证据评估中的错误可能不会改变决策，但误解或忽视对决策至关重要的证据可能会导致决策的反转。识别此类证据使得更强大的模型能够专注于检查相应条件的判断，从而支持准确和安全的决策。识别证据的关键性需要理解证据如何影响条件判断，以及该判断如何影响决策。为实现这一目标，我们提出了一种基于干预的影响学习框架（INTERvention-based imPACT learning framework，InterPact），该框架能够对规则驱动的决策中的证据关键性进行反事实验证。具体而言，其证据干预构造器通过在固定规则和非目标条件的情况下编辑案例事实，生成用于传播验证器的训练对。经过人工审核的标签记录了由此产生的条件和决策变化，而完整的状态到决策映射则监督超出观察到的编辑的后果。在训练过程中，验证器通过固定组合操作根据基于证据的条件概率对学习到的条件决策预测进行加权，将决策变化的监督传播到基础模型。在推理阶段，训练好的基础模型直接从原始案例和目标证据判断关键性，而无需人工或更强模型的监督。在针对适应性规则驱动决策案例的单案例证据关键性验证中，InterPact达到了68.28%的准确率，超越了所有六个基线。这些结果支持学习到的决策敏感性作为优先检查证据的基础。
cs.CL / 79 / 2609.39639
Marginal Response Surface Elicitation for Zero-Label Tabular Learning
零标签表格学习的边际响应面引导
Abstract
Tabular learning uses structured data to predict target outcomes. Traditionally, this process has relied on labeled data. However, large language models (LLMs) can be used to elicit domain priors based on the task description and feature semantics, thereby enabling predictions without labeled data. We propose Marginal Response Surface Elicitation (MARS), a method that transforms feature-level LLM priors into a reusable, zero-shot tabular classifier. To construct this classifier, MARS selects representative values for each feature from unlabeled data and prompts the LLM to provide corresponding class support scores and feature weights. It then aggregates multiple responses using the median to construct feature response functions, and makes predictions through their weighted sum without further LLM queries. Across eight tabular benchmark tasks, MARS achieves the highest average AUC and AP, outperforming direct prompting by 1.97 and 6.21 percentage points respectively, while substantially reducing end-to-end costs. Evaluations with LLMs of different sizes further demonstrate its predictive advantage over direct prompting.
Chinese Translation
表格学习利用结构化数据预测目标结果。传统上，这一过程依赖于标记数据。然而，大型语言模型（LLMs）可以根据任务描述和特征语义引导领域先验，从而实现无标记数据的预测。我们提出了边际响应面引导（Marginal Response Surface Elicitation, MARS）方法，该方法将特征级LLM先验转化为可重用的零样本表格分类器。为了构建该分类器，MARS从未标记数据中选择每个特征的代表值，并提示LLM提供相应的类别支持分数和特征权重。然后，它使用中位数聚合多个响应以构建特征响应函数，并通过加权求和进行预测，而无需进一步的LLM查询。在八个表格基准任务中，MARS实现了最高的平均AUC和AP，分别比直接提示提高了1.97和6.21个百分点，同时显著降低了端到端成本。对不同规模LLM的评估进一步证明了其相较于直接提示的预测优势。
cs.CL / 80 / 2609.39640
Zero-Compute Cross-Lingual Transferability Estimation Using Typological Feature Proxies
使用类型特征代理的零计算跨语言可迁移性估计
Abstract
Cross-lingual transfer describes how knowledge in a source language benefits a target language. Measuring it quantitatively requires broad multilingual pre-training, as prior work has done with cross-lingual transfer matrices. We ask whether transfer is predictable from freely available typological features, and whether the prominence of high-resource source languages reflects typology or data quality and quantity. We show that typological databases contain cheap and dense signals about cross-lingual transfer. Our typology-only random forest on a 24-language prior-work transfer matrix scores leave-one-language-out $\rho{=}0.705$ and $R^2{=}0.49$, beating a non-typological control at $\rho{=}0.62$, which verifies the ability of typology-only predictions to reconstruct costly measured cross-lingual transfer. The signal survives leave-one-script-out and leave-one-family-out protocols, so script and family confounding do not explain the effect. By decomposing the transfer into a typology term and a resource-and-script bias term, we find the best-source ranking sensitive to this bias. In contrast, typology is not affected by this bias, which makes it a zero-compute screening tool that replaces hundreds of training runs with a model fit. Our code is available \href{https://github.com/dharmsen/typo-x-ling-transfer}{here}.
Chinese Translation
跨语言迁移描述了源语言中的知识如何惠及目标语言。定量测量这一现象需要广泛的多语言预训练，正如先前的研究所采用的跨语言迁移矩阵。我们探讨了是否可以从自由获取的类型特征中预测迁移，以及高资源源语言的显著性是否反映了类型学或数据的质量和数量。我们展示了类型数据库中包含关于跨语言迁移的廉价且密集的信号。我们基于类型学的随机森林模型在一个包含24种语言的先前研究迁移矩阵上得分为留一语言外 $
ho{=}0.705$ 和 $R^2{=}0.49$，超越了非类型学控制组的 $
ho{=}0.62$，验证了仅基于类型学的预测能够重建昂贵的测量跨语言迁移的能力。该信号在留一脚本外和留一家族外的协议中依然有效，因此脚本和家族的混淆并不能解释这一效应。通过将迁移分解为类型学项和资源与脚本偏差项，我们发现最佳源排名对这一偏差敏感。相比之下，类型学不受此偏差影响，这使其成为一种零计算筛选工具，能够用模型拟合替代数百次训练运行。我们的代码可在此获取 [链接](https://github.com/dharmsen/typo-x-ling-transfer)。
cs.CL / 81 / 2609.39645
SEPAL: Separated Expert Pairs with Answer-Level Fusion for Reliable LLM Collaboration
SEPAL：通过答案级融合的分离专家对以实现可靠的LLM协作
Abstract
Multi-agent collaboration lets large language models (LLMs) improve question answering through deliberation and feedback. Yet shared discussion couples correction with exposure to the same mistakes, which can erode the diversity needed for voting. Self-consistency offers sampling diversity without feedback, while single-pair Actor-Critic collaboration refines only one candidate. We introduce SEPAL, which assigns three private Actor-Critic teams to direct reasoning, evidence grounding, and verification. Role-specific training gives the teams different reasoning objectives beyond sampling variation. Each Critic guides revisions within its own team, preventing feedback from carrying errors across candidates. Once revision ends, majority voting combines only the final answers, keeping the reasoning histories separate until the decision. Across five open-weight backbones and five question-answering benchmarks, SEPAL improves mean accuracy by 1.81 percentage points over a matched single Actor-Critic pair, with improvements across all five backbones. Code is available at https://github.com/zhansan114514/SEPAL.
Chinese Translation
多智能体协作使大型语言模型（LLMs）能够通过深思熟虑和反馈来改善问答能力。然而，共享讨论将纠正与暴露于相同错误相结合，这可能削弱投票所需的多样性。自一致性提供了没有反馈的采样多样性，而单对Actor-Critic协作仅精炼一个候选者。我们提出了SEPAL，它为直接推理、证据基础和验证分配了三个私有的Actor-Critic团队。角色特定的训练使团队拥有不同的推理目标，超越了采样变异。每个Critic在其团队内指导修订，防止反馈将错误传播到其他候选者。一旦修订结束，多数投票仅结合最终答案，保持推理历史在决策之前的独立性。在五个开放权重骨干网络和五个问答基准测试中，SEPAL的平均准确率比匹配的单个Actor-Critic对提高了1.81个百分点，并在所有五个骨干网络中均有所改善。代码可在 https://github.com/zhansan114514/SEPAL 获取。
cs.CL / 82 / 2609.39661
The Evolution of Attention in Large Language Models: Mechanisms, Trade-offs, and Emerging Trends
大型语言模型中注意力的演变：机制、权衡与新兴趋势
Abstract
Self-attention gives LLMs fine-grained, query-dependent access to context, but dense token interactions incur quadratic prefill cost and a key--value cache growing with context length. Research thus spans explicit-memory compression, sparse access, recurrent state construction, structured state dynamics, and heterogeneous mechanism composition. This survey analyzes these developments as model-internal contextual memory. We introduce a five-dimensional lens---Memory Representation, Memory Update, Access, Readout, and Integration---describing what is represented, how it changes, what is query-eligible, how it is read, and how readouts form outputs. This lens compares overlapping research lines without imposing one computational model. We reconstruct mechanism-level developments and architectural adoption using 59 release-level records from 14 major model lineages and 11 high-performing open-weight endpoints. First, explicit-memory and recurrent-state methods retain distinct interfaces but increasingly control overlapping memory functions. Second, heterogeneous architectures increasingly coordinate across network depth: layer-wise composition distributes complementary memory processing across representational stages, while cross-layer reuse carries selected memory and routing artifacts forward. Depth thus becomes a dimension along which contextual memory is constructed and managed. Third, these developments motivate a stateful multidimensional memory-routing hypothesis: persistent memory is organized across temporal scope, network depth, substrate type, and representation granularity, while coordinated Sparse Write and Sparse Read determine what is maintained and what contributes to each query. Overall, efficient sequence architecture design increasingly concerns the organization, lifecycle, and selective use of contextual memory rather than an isolated Attention operator.
Chinese Translation
自注意力机制为大型语言模型（LLMs）提供了细粒度、依赖查询的上下文访问，但密集的标记交互会导致二次填充成本以及随着上下文长度增长的键值缓存。因此，研究涵盖了显式内存压缩、稀疏访问、递归状态构建、结构化状态动态和异构机制组合。本调查分析了这些发展作为模型内部的上下文记忆。我们引入了一个五维视角——记忆表示、记忆更新、访问、读取和整合——描述了所表示的内容、如何变化、哪些内容符合查询条件、如何读取以及读取如何形成输出。该视角在不强加单一计算模型的情况下比较了重叠的研究方向。我们使用来自14个主要模型谱系和11个高性能开放权重端点的59个发布级记录重构了机制级的发展和架构采用。首先，显式内存和递归状态方法保持了不同的接口，但越来越多地控制重叠的记忆功能。其次，异构架构在网络深度上越来越协调：层级组合在表征阶段分配互补的记忆处理，而跨层重用则将选定的记忆和路由特征向前传递。因此，深度成为构建和管理上下文记忆的一个维度。第三，这些发展激励了一种状态化的多维记忆路由假设：持久记忆在时间范围、网络深度、基底类型和表征粒度上组织，而协调的稀疏写入和稀疏读取决定了哪些内容被维护以及哪些内容对每个查询有贡献。总体而言，高效的序列架构设计越来越关注上下文记忆的组织、生命周期和选择性使用，而非孤立的注意力操作符。
cs.CL / 83 / 2609.39687
Better Supervision Is Nearby: Neighborhood On-Policy Self-Distillation
更好的监督就在身边：邻域在线自蒸馏
Abstract
On-policy self-distillation (OPSD) trains mathematical reasoning models using a privileged teacher that sees a reference solution and supervises student-sampled prefixes. Standard OPSD uses one fixed parameter setting at every state, but nearby settings may offer additional supervision. We find that local parameter perturbations reveal complementary reference-aligned corrections under the same reference context. Different experts supply these corrections at different reference positions. Their pool covers more such positions than the unperturbed privileged teacher. We introduce Neighborhood OPSD (N-OPSD) to turn these corrections into supervision at student-visited states. Offline, greedy selection builds a compact pool of frozen experts by rewarding filtered reference-token gains beyond the pool's current best at each position. The highest-peak expert need not provide the best training target. Online routing therefore separates the anchor direction from its level of support. MaxPeak selects the anchor token, and quantile selection chooses among experts whose top token matches it. The student learns from the chosen expert's full next-token distribution through the clipped forward-KL objective inherited from OPSD. We evaluate on AIME 2024, AIME 2025, and HMMT February 2025. Across three independent runs per method, Neighborhood OPSD improves the three-benchmark Average@12 over OPSD by 2.75, 1.67, and 1.94 points on Qwen3-1.7B, 4B, and 8B, respectively. Student-prefix continuations support using the pool beyond the reference trajectories used for selection. Matched ablations support filtered reference-token gains as a selection criterion. Accounting for overlap within the pool and routing by state further improve student accuracy. Inference uses only the distilled student.
Chinese Translation
在线自蒸馏（On-policy Self-Distillation, OPSD）利用一个特权教师来训练数学推理模型，该教师能够看到参考解并监督学生采样的前缀。标准的OPSD在每个状态下使用一个固定的参数设置，但邻近的设置可能提供额外的监督。我们发现局部参数扰动在相同的参考上下文中揭示了互补的参考对齐修正。不同的专家在不同的参考位置提供这些修正。它们的池覆盖了比未扰动的特权教师更多的位置。我们引入邻域OPSD（Neighborhood OPSD, N-OPSD）将这些修正转化为学生访问状态的监督。在离线阶段，贪婪选择通过奖励超出当前最佳的过滤参考标记增益来构建一个紧凑的冻结专家池。最高峰专家不一定提供最佳的训练目标。因此，在线路由将锚定方向与其支持水平分开。MaxPeak选择锚定标记，而分位选择在其顶级标记匹配的专家中进行选择。学生通过从所选专家的完整下一个标记分布中学习，使用从OPSD继承的剪切前向KL目标。我们在AIME 2024、AIME 2025和HMMT 2025年2月进行评估。在每种方法的三次独立运行中，邻域OPSD在Qwen3-1.7B、4B和8B上分别提高了三基准的Average@12指标2.75、1.67和1.94分。学生前缀的延续支持在选择所用的参考轨迹之外使用池。匹配消融实验支持过滤参考标记增益作为选择标准。考虑到池内的重叠和按状态路由进一步提高了学生的准确性。推理仅使用蒸馏后的学生。
cs.CL / 84 / 2609.39710
Drift Inspector: Exploring and Measuring Scientific Drift with Atomic Contribution Claims
漂移检查器：通过原子贡献声明探索和测量科学漂移
Abstract
Scientific abstracts mix contributions with background, motivation, and meta-language, so tools that read them as-is cannot separate what a field produces from what it discusses. We present Drift Inspector, an open-source system for measuring and exploring how a research field changes over time at the level of Atomic Contribution Claims (ACCs): decontextualized, contribution-bearing propositions an LLM extracts from each abstract before analysis. The system clusters these claims across years into an interactive map where every trend traces back to the claims and papers behind it. Applied to six years of EMNLP, it shows the field shifting away from classic NLP tasks toward LLM-era capabilities such as reasoning and multimodality -- a movement that keyword or whole-abstract counts blur. The released data extend beyond EMNLP: the same pipeline has processed the full ACL Anthology (346k claims, 80k abstracts, 423 venues). Extraction is human-validated and clustering checked against an external manually constructed taxonomy.
Chinese Translation
科学摘要将贡献与背景、动机和元语言混合在一起，因此以原样读取它们的工具无法将一个领域所产生的内容与其讨论的内容分开。我们提出了漂移检查器（Drift Inspector），这是一个开源系统，用于测量和探索研究领域在原子贡献声明（Atomic Contribution Claims, ACCs）层面上如何随时间变化：这些是去上下文化的、承载贡献的命题，LLM（大语言模型）在分析之前从每个摘要中提取。该系统将这些声明按年份聚类成一个互动地图，其中每个趋势都追溯到其背后的声明和论文。应用于六年的EMNLP（自然语言处理会议），结果显示该领域正在从经典的NLP任务转向LLM时代的能力，如推理和多模态性——这一变化在关键词或整篇摘要计数中模糊不清。发布的数据超出了EMNLP：相同的流程已经处理了完整的ACL文集（346,000个声明，80,000个摘要，423个会议）。提取经过人工验证，聚类则与外部手动构建的分类法进行了核对。
cs.CL / 85 / 2609.39740
LatentHarness: Learning Latent Actions for Memory and Reasoning via Counterfactual Policy Distillation
LatentHarness：通过反事实策略蒸馏学习用于记忆和推理的潜在动作
Abstract
Long-context reasoning faces two complementary bottlenecks: retaining evidence across long inputs and sustaining computation across many reasoning steps. Existing approaches largely address them separately, with external memory extending access to distant evidence and latent reasoning compressing multi-step computation. We introduce LatentHarness, which unifies memory access and latent reasoning as sequential latent action selection. At each internal step, the model chooses THINK for further computation, RECALL from a fast-weight memory of input evidence and intermediate reasoning states, or EXIT to emit the next token. We train this policy with counterfactual policy distillation, which branches every action for one step and scores its effect on the emitted token. These gains teach the policy when memory is more useful than further reasoning, while gradients through counterfactual recall teach which intermediate states should be retained in memory for future use. Across six general and long-context reasoning benchmarks, LatentHarness at 1.4B improves on the strongest baselines by 2.8% and 10.0% relative, respectively, and runs 5.9x faster than the strongest long-context baseline.
Chinese Translation
长上下文推理面临两个互补的瓶颈：在长输入中保留证据和在多个推理步骤中维持计算。现有的方法主要分别解决这两个问题，外部记忆扩展了对远程证据的访问，而潜在推理压缩了多步骤计算。我们提出了LatentHarness，它将记忆访问和潜在推理统一为顺序潜在动作选择。在每个内部步骤中，模型选择THINK以进行进一步计算，RECALL以从快速权重记忆中提取输入证据和中间推理状态，或EXIT以发出下一个标记。我们通过反事实策略蒸馏训练该策略，该方法为每个动作分支一步并评估其对发出标记的影响。这些收益教会策略在何时记忆比进一步推理更有用，而通过反事实回忆的梯度则教会哪些中间状态应该保留在记忆中以供将来使用。在六个通用和长上下文推理基准测试中，LatentHarness在1.4B参数下相较于最强基线分别提高了2.8%和10.0%的相对性能，并且运行速度比最强的长上下文基线快5.9倍。
cs.CL / 86 / 2609.39765
MemCodex: Self-Programming Hierarchical Memory for Language Agents
MemCodex：用于语言代理的自编程层次记忆
Abstract
Agent memory faces heterogeneous access needs: a single-hop question may require one piece of evidence, whereas a multi-hop question must combine evidence from multiple sources. Predefined memory workflows cannot adapt to these varying needs. Recent adaptive methods search or learn over memory components and their compositions, but the design space itself remains predefined. We introduce MemCodex, a self-evolving hierarchical memory system that organizes experience into executable memory programs for summaries, relational knowledge, reusable skills, and latent memory. Open-ended program evolution searches the open design space of layer programs by rewriting how each layer is constructed, indexed, retrieved, and routed, thereby adapting both within-layer implementations and cross-layer composition. At query time, reads traverse the hierarchy from coarse to fine and stop once sufficient evidence is found, descending to the original history when needed. We further develop MemArena, a unified runtime that places heterogeneous data and memory systems behind a common interface. MemCodex improves average task success by 10.1% relative to the strongest adaptive-memory baseline, while using 3.4x fewer context tokens and achieving 2.1x faster inference.
Chinese Translation
代理记忆面临异构的访问需求：单跳问题可能只需要一条证据，而多跳问题必须结合来自多个来源的证据。预定义的记忆工作流程无法适应这些变化的需求。近期的自适应方法在记忆组件及其组合上进行搜索或学习，但设计空间本身仍然是预定义的。我们提出了MemCodex，一种自我演化的层次记忆系统，它将经验组织成可执行的记忆程序，用于摘要、关系知识、可重用技能和潜在记忆。开放式程序演化通过重写每一层的构建、索引、检索和路由方式，搜索层程序的开放设计空间，从而适应层内实现和跨层组合。在查询时，读取操作从粗到细遍历层次结构，并在找到足够证据后停止，必要时降回原始历史。我们进一步开发了MemArena，一个统一的运行时，将异构数据和记忆系统置于一个共同接口之后。与最强自适应记忆基线相比，MemCodex将平均任务成功率提高了10.1%，同时使用的上下文标记减少了3.4倍，推理速度提高了2.1倍。
cs.CL / 87 / 2609.39786
Explore-on-Graph: Hybrid Embedding-LLM Reasoning for Knowledge Graph Question Answering under Incompleteness
探索图上的推理：针对不完整知识图谱的混合嵌入-大语言模型推理
Abstract
Large language models (LLMs) are increasingly combined with knowledge graphs (KGs) to ground reasoning in structured evidence. However, most LLM-based KGQA methods rely on traversing existing graph edges and become unreliable when reasoning paths are broken by missing facts. Alternatives that ask LLMs to generate missing knowledge risk introducing hallucinated evidence. We introduce XoG (eXplore-on-Graph), a framework for multi-hop question answering over incomplete KGs that recovers missing reasoning paths from learned graph structure rather than LLM parametric knowledge. XoG combines type-level entity-relation statistics to identify candidate relations with KG embeddings to retrieve plausible missing entities, using the LLM as a semantic selector and reasoner. These mechanisms are integrated into an iterative planning-exploration-reasoning process. Experiments on WebQSP, CWQ, and the Wikidata-based BRINK benchmark show that XoG remains competitive on complete KGs and consistently outperforms comparable methods without task-specific KGQA training under KG incompleteness. These gains persist across multiple LLM backbones, indicating that stronger LLMs alone do not resolve missing graph evidence. XoG also reduces LLM token consumption by up to 33% compared with a closely related planning-based approach.
Chinese Translation
大型语言模型（LLMs）越来越多地与知识图谱（KGs）结合，以在结构化证据中进行推理。然而，大多数基于LLM的知识图谱问答（KGQA）方法依赖于遍历现有的图边，当推理路径因缺失事实而中断时，这些方法变得不可靠。要求LLM生成缺失知识的替代方案可能会引入虚假的证据。我们提出了XoG（eXplore-on-Graph），这是一个针对不完整知识图谱的多跳问答框架，它通过学习的图结构恢复缺失的推理路径，而不是依赖于LLM的参数知识。XoG结合了类型级实体-关系统计，以识别候选关系，并利用知识图谱嵌入检索合理的缺失实体，使用LLM作为语义选择器和推理者。这些机制被整合到一个迭代的规划-探索-推理过程中。在WebQSP、CWQ和基于Wikidata的BRINK基准上的实验表明，XoG在完整知识图谱上保持竞争力，并在知识图谱不完整的情况下始终优于没有特定任务KGQA训练的可比方法。这些收益在多个LLM基础模型中持续存在，表明仅仅依靠更强的LLM并不能解决缺失的图证据。与一种密切相关的基于规划的方法相比，XoG还将LLM的令牌消耗减少了多达33%。
cs.CL / 88 / 2609.39807
Stress-Testing LLM Lie Detectors: Role-Play Failures and Spurious Correlations
压力测试大型语言模型的谎言检测器：角色扮演失败与虚假相关性
Abstract
Lie detection probes aim to predict from a language model's internal states whether its output is truthful or dishonest. However, role-play complicates what "truth" means for an LLM: language models can adopt a wide range of personas that take very different claims to be true, including personas whose beliefs clearly contradict reality, such as a conspiracy theorist. In this work, we investigate whether lie detection probes reliably flag falsehoods generated under such an anti-factual persona or whether they instead follow the persona's beliefs. We introduce a dataset of 8,916 human-reviewed, on-policy responses from three LLMs adopting anti-factual personas. Evaluating eight probes from prior work, we find that many fail in this setting, particularly when correct and incorrect answers are evaluated under the same persona prompt. To investigate why, we construct three novel confounder datasets in which truth is anti-correlated with a potential confounding concept. Our experiments reveal that many existing probes strongly track concepts that are spuriously correlated with truth in their training data, such as instruction compliance or response likelihood. Based on these findings, we introduce a simple linear probe that achieves the strongest overall performance on both the persona and confounder stress tests. Our results suggest that current lie detection probes are far from reliable and highlight the need for training data in which truth is decorrelated from confounding concepts.
Chinese Translation
谎言检测探针旨在根据语言模型的内部状态预测其输出是否真实。然而，角色扮演使得大型语言模型（LLM）中“真相”的含义变得复杂：语言模型可以采用广泛的角色，这些角色对真相的认知可能截然不同，包括那些信念明显与现实相悖的角色，例如阴谋论者。在本研究中，我们探讨了谎言检测探针是否能够可靠地标记在这种反事实角色下生成的虚假信息，或者它们是否会遵循该角色的信念。我们引入了一个包含8,916个经过人工审核的、来自三种采用反事实角色的LLM的政策响应的数据集。评估先前工作的八个探针时，我们发现许多探针在这一设置下表现不佳，特别是在正确答案和错误答案在同一角色提示下进行评估时。为了探讨原因，我们构建了三个新的混淆数据集，其中真相与潜在的混淆概念呈反相关。我们的实验揭示，许多现有探针强烈跟踪与其训练数据中的真相虚假相关的概念，例如指令遵从性或响应可能性。基于这些发现，我们引入了一种简单的线性探针，在角色和混淆压力测试中均表现出最佳的整体性能。我们的结果表明，当前的谎言检测探针远未可靠，并强调了训练数据中真相与混淆概念去相关的必要性。
cs.CL / 89 / 2609.39827
Synthetic Pre-pretraining Survives Scale, but Not as a Grammatical Prior
合成预预训练在规模上存活，但不是作为语法先验
Abstract
Pre-pretraining (PPT) on synthetic non-natural language data improves token efficiency during language model pre-training (PT). Prior work attributes this gain to a grammatical prior, i.e., a structural inductive bias learned during PPT that transfers to natural language grammar. However, PPT has only been tested on models of at most 1B parameters and PT budgets below 2B tokens on predominantly web text. It is unknown whether PPT is effective at larger scales and under more realistic PT data mixtures that combine diverse sources (e.g., code and math). We therefore present a comprehensive study on PPT spanning five PPT tasks, four PT data mixtures, four parameter scales (500M to 7B), and PT budgets of up to 100B tokens. Our results demonstrate that the downstream performance and token efficiency gains of PPT persist at scale, e.g., saving at least 21B PT tokens at the 3B scale. However, in contrast to prior work, we find no consistent evidence that these gains stem from a grammatical prior. Downstream performance does not consistently align with grammatical acceptability across model sizes. Instead, we find that downstream gains arise from PPT tasks that improve long-range retrieval. Finally, PPT performance gains are robust to how PT data mixtures are composed and diminish only when web text is absent. Overall, PPT is a low-cost addition to PT, and future PPT task design should target long-range retrieval rather than natural language grammar.
Chinese Translation
合成非自然语言数据上的预预训练（PPT）提高了语言模型预训练（PT）期间的标记效率。先前的研究将这一提升归因于语法先验，即在PPT期间学习到的结构性归纳偏差，这种偏差能够转移到自然语言语法上。然而，PPT仅在最多1B参数的模型和主要基于网络文本的PT预算低于2B标记的情况下进行了测试。目前尚不清楚PPT在更大规模和更现实的PT数据混合（例如代码和数学）下是否有效。因此，我们进行了一项全面的PPT研究，涵盖五个PPT任务、四种PT数据混合、四个参数规模（500M到7B）以及高达100B标记的PT预算。我们的结果表明，PPT的下游性能和标记效率提升在规模上依然存在，例如，在3B规模下节省至少21B的PT标记。然而，与先前的研究相反，我们没有发现这些提升源于语法先验的持续证据。下游性能在不同模型规模下与语法可接受性并不一致。相反，我们发现下游提升源于改善长距离检索的PPT任务。最后，PPT性能提升对PT数据混合的组成方式具有鲁棒性，只有在缺少网络文本时才会减弱。总体而言，PPT是PT的低成本补充，未来的PPT任务设计应针对长距离检索，而非自然语言语法。
cs.CL / 90 / 2609.39846
When a Kindergartener Solves Calculus: Measuring Capability Leakage in Role-Prompted Reasoning Models
幼儿园学生解决微积分问题：角色引导推理模型中的能力泄漏测量
Abstract
We investigate the problem of role-capability leakage (RCL), in which a role-prompted reasoning model generates convincing in-role text while continuing to exhibit capabilities on benchmarks that exceed those implied by the assigned role. For example, when a model is prompted to assume the role of a kindergarten student, one might expect its performance on a mathematics benchmark to reflect kindergarten-level ability rather than expert-level proficiency in solving calculus problems. We introduce RoleCapBench, a curriculum-grounded benchmark for evaluating RCL across six educational roles and four assessment levels spanning elementary school through A-level, and use it to evaluate three open-weight reasoning models. We find that although the models can generate stylistically convincing in-role responses, they consistently fail to align their underlying capabilities with their assigned roles. Naive role prompting yields strong role-voice scores of 1.218--1.389 while retaining above-role accuracy of 0.811--0.898. RCL persists across a range of prompting conditions, including prompts that explicitly instruct the model to match the role's capability level. To mitigate this problem, we propose Injection, an inference-time intervention that combines explicit, role-specific capability guidelines with a guiding prefilled response prefix. Injection improves role-capability alignment across models, reducing above-role accuracy by up to 0.562 while preserving in-role accuracy with a marginal drop of less than 0.058 across most models. All artifacts, including scripts and evaluation data, will be released upon acceptance.
Chinese Translation
我们研究了角色能力泄漏（RCL）的问题，其中一个角色引导推理模型生成令人信服的角色内文本，同时在基准测试中表现出超出所分配角色所暗示的能力。例如，当一个模型被提示假设幼儿园学生的角色时，人们可能期望其在数学基准测试中的表现反映幼儿园水平的能力，而不是解决微积分问题的专家水平的熟练度。我们引入了RoleCapBench，这是一个基于课程的基准，用于评估六个教育角色和四个评估级别（从小学到A-level）的RCL，并利用它来评估三个开放权重的推理模型。我们的研究发现，尽管这些模型能够生成在风格上令人信服的角色内响应，但它们始终未能将其潜在能力与所分配的角色对齐。简单的角色提示产生了强大的角色声音评分（1.218-1.389），同时保持了高于角色的准确率（0.811-0.898）。RCL在多种提示条件下持续存在，包括明确指示模型匹配角色能力水平的提示。为了解决这个问题，我们提出了Injection，这是一种推理时的干预措施，结合了明确的、角色特定的能力指导和一个引导的预填充响应前缀。Injection改善了模型之间的角色能力对齐，将高于角色的准确率降低了最多0.562，同时在大多数模型中保持了角色内准确率的微小下降（少于0.058）。所有文档，包括脚本和评估数据，将在接受后发布。
cs.CL / 91 / 2609.39853
Cognitive Enhancement: Rethinking the Necessity of Role-Playing for Large Language Models
认知增强：重新思考大型语言模型角色扮演的必要性
Abstract
Role-playing prompting has become a popular yet simple technique for improving LLM reasoning and output quality. However, whether it consistently boosts performance across diverse domains remains unclear, as systematic validation is lacking. To fill this gap, we run multi-model, cross-domain, and multilingual experiments on MMLU and MMLU-Redux. We find that gains from role-play prompting depend heavily on model capacity, knowledge domain, and prompt language. Drawing on metacognition theory, we propose the persona-related cognitive alignment hypothesis: role-play works only when the LLM correctly grasps the designated persona and its associated knowledge domain. We test this hypothesis through persona information richness ablation, layer-wise entropy divergence analysis, and latent thought-space deflection observation. To reduce persona cognitive bias and stabilize role-play performance, we propose \textbf{M}ixed-\textbf{L}anguage \textbf{C}oncatenate \textbf{P}rediction \textbf{(MLCP}), a simple, training-free, and efficient multilingual prompt concatenation strategy. It aggregates semantically equivalent role prompts to enrich complementary representational cues. Extensive experiments show that MLCP consistently outperforms vanilla role-play prompting across all tested LLMs.
Chinese Translation
角色扮演提示已成为一种流行且简单的技术，用于提高大型语言模型（LLM）的推理能力和输出质量。然而，是否在不同领域中始终提升性能仍不明确，因为缺乏系统性验证。为填补这一空白，我们在MMLU和MMLU-Redux上进行了多模型、跨领域和多语言的实验。我们发现，角色扮演提示带来的收益在很大程度上依赖于模型容量、知识领域和提示语言。基于元认知理论，我们提出了与角色相关的认知一致性假设：只有当LLM正确理解指定角色及其相关知识领域时，角色扮演才会有效。我们通过角色信息丰富度消融、层级熵发散分析和潜在思维空间偏转观察来检验这一假设。为了减少角色认知偏差并稳定角色扮演性能，我们提出了混合语言连接预测（Mixed-Language Concatenate Prediction, MLCP），这是一种简单、无训练且高效的多语言提示连接策略。它聚合语义等价的角色提示，以丰富互补的表征线索。大量实验表明，MLCP在所有测试的LLM中始终优于传统的角色扮演提示。
cs.CL / 92 / 2609.39882
LLM Persona Unlearning
大语言模型的个性去学习
Abstract
Pre-training equips large language models (LLMs) with a broad repertoire of behavioral patterns associated with roles, styles, values, and goals. Post-training teaches conditional enactment and makes a helpful Assistant the default, but it does not erase alternative modes from the weights; explicit prompts can therefore elicit personas that repeatedly shape judgment, language, and action. In open-weight settings, runtime controls can be removed, motivating persona unlearning: a weight-level edit that makes a designated persona difficult to elicit and enact on unseen contexts. We introduce PersonaUnlearnBench, a model-specific paired benchmark spanning six LLMs from three families and five personas, with aligned forget/retain sets, held-out instruction paraphrases, and four-axis evaluation. The benchmark shows that standard unlearning methods cannot reliably erase the target persona without sacrificing meaningful generation or general utility. We therefore propose PaCE, which compares target and desirable responses to the same questions to locate an internal behavior direction, then trains target-prompt states away from the target mode and toward the matched desirable response. Experiments show that PaCE consistently suppresses target personas with high response quality and useful counterpart behavior, at moderate utility cost. These results establish persona unlearning as a distinct behavior-level editing problem and a practical route toward persistent control of latent LLM response policies.
Chinese Translation
预训练使大语言模型（LLMs）具备与角色、风格、价值观和目标相关的广泛行为模式。后训练则教授条件性执行，并使得有用的助手成为默认选项，但并未从权重中抹去替代模式；因此，明确的提示可以引发个性，这些个性会反复影响判断、语言和行动。在开放权重设置中，运行时控制可以被移除，从而激励个性去学习：一种权重级别的编辑，使得指定个性在未见上下文中难以引发和执行。我们引入了PersonaUnlearnBench，这是一个特定于模型的配对基准，涵盖来自三个家族的六个LLM和五个个性，具有对齐的遗忘/保留集合、保留的指令改述和四个维度的评估。基准显示，标准去学习方法无法可靠地抹去目标个性，而不牺牲有意义的生成或一般实用性。因此，我们提出了PaCE，它通过比较对相同问题的目标和期望响应来定位内部行为方向，然后将目标提示状态从目标模式训练到匹配的期望响应。实验表明，PaCE在保持高响应质量和有用的对应行为的同时，始终抑制目标个性，且实用性成本适中。这些结果确立了个性去学习作为一种独特的行为级别编辑问题，并为持久控制潜在LLM响应策略提供了一条实用途径。
cs.CL / 93 / 2609.39884
OPSRD: On-Policy Self-Role Distillation
OPSRD：基于策略的自我角色蒸馏
Abstract
Role prompting elicits specialized behavior from large language models through an expert identity, offering a lightweight way to guide reasoning on demanding tasks. However, evaluating or distilling complete role-prompted answers can miss useful next-token preferences when the sampled solution remains incorrect. Transferring these preferences also requires an objective that reaches alternatives the student rarely predicts. We introduce OPSRD, which uses a fixed expert role as privileged teaching context for on-policy self-distillation without reference solutions. A role-free student generates a trajectory, and a frozen instance of the same base model supplies role-conditioned distributions on its exact prefixes, exposing alternatives beyond the sampled continuation. Teacher-weighted forward KL targets alternatives the student underestimates, with clipping to limit individual vocabulary contributions. Supervision is restricted to the highest-entropy half of student positions, concentrating learning where predictions are uncertain. Experiments on three competition-math benchmarks with Qwen3-1.7B, 4B, and 8B show improvements over the base models without role prompts at inference. Forward KL achieves the highest macro-averaged accuracy among the three evaluated divergences at every scale. Code is available at https://github.com/zhansan114514/OPSRD.
Chinese Translation
角色提示通过专家身份引导大型语言模型产生专业化行为，为复杂任务的推理提供了一种轻量级的指导方式。然而，当采样的解决方案仍然不正确时，评估或蒸馏完整的角色提示答案可能会错过有用的下一个标记偏好。转移这些偏好也需要一个目标，以达到学生很少预测的替代方案。我们提出了OPSRD，它使用固定的专家角色作为特权教学上下文进行无参考解决方案的基于策略自我蒸馏。一个无角色的学生生成一个轨迹，而同一基础模型的冻结实例提供其确切前缀上的角色条件分布，揭示超出采样延续的替代方案。教师加权的前向KL（Kullback-Leibler）目标针对学生低估的替代方案，并通过剪切限制个别词汇的贡献。监督限制在学生位置中熵最高的一半，集中学习在预测不确定的地方。在使用Qwen3-1.7B、4B和8B的三个竞赛数学基准上进行的实验表明，在推理时，相较于没有角色提示的基础模型有了改进。前向KL在每个规模的三种评估散度中实现了最高的宏平均准确率。代码可在 https://github.com/zhansan114514/OPSRD 获取。
cs.CL / 94 / 2609.39913
The Concrete-Arbitrary Gap: Kinship Reasoning in LLMs Is Not Indifferent to Presentation
具体-任意差距：大型语言模型中的亲属推理对表现形式并不无动于衷
Abstract
We test whether large language models solve formally matched kinship problems equally well when relations are expressed in familiar vocabulary or by explicitly defined nonce predicates. Across 500 paired graphs, concrete accuracy exceeds arbitrary accuracy by 35.6 percentage points in local Qwen3.8-27B, 26.6 in Gemma 4 26B-A4B, 12.0 in Gemma 4 31B, and 5.4 in Qwen3.8-Max. All four paired gaps are statistically resolved. Reasoning budgets and prompt-language interventions can substantially reduce the difference, showing that it is modifiable rather than a fixed incapacity. The minimal conclusion is behavioral: on these tasks, the models' manifested relational competence is not indifferent to presentation. Explicit definitions provide the formal relations but do not make nonce predicates as usable as familiar vocabulary embedded in learned linguistic associations.
Chinese Translation
我们测试大型语言模型在亲属关系问题上是否能在使用熟悉词汇或通过明确定义的临时谓词时表现出同样的解决能力。在500个配对图中，Qwen3.8-27B模型的具体准确率比任意准确率高出35.6个百分点，Gemma 4 26B-A4B为26.6，Gemma 4 31B为12.0，而Qwen3.8-Max为5.4。这四个配对差距在统计上均已解决。推理预算和提示语言干预可以显著减少这一差距，表明这一差距是可修改的，而非固定的无能。最基本的结论是行为上的：在这些任务中，模型所表现出的关系能力并不对表现形式无动于衷。明确的定义提供了形式关系，但并未使临时谓词的可用性达到嵌入在学习语言关联中的熟悉词汇的水平。
cs.CL / 95 / 2609.39927
AdaGEPA: Adaptive Feedback Allocation for Reflective Prompt Optimization
AdaGEPA：用于反思性提示优化的自适应反馈分配
Abstract
Prompt optimization improves the performance of language-model systems on downstream tasks by refining their prompts. Classical methods evaluate prompts on task examples and use the resulting feedback to guide prompt revisions through reflection. However, when feedback selection does not account for the prompt's weaknesses, these revisions may improve performance on selected examples without yielding broader task improvements. To address this issue, we propose AdaGEPA, an adaptive feedback-allocation method that uses the prompt's performance and task structure to select examples for the next prompt revision. Our method replaces at most one example in each feedback minibatch to target an identified weakness while preserving the remaining feedback context. Across our main experiments on six downstream benchmarks, AdaGEPA achieves higher mean validation scores than non-adaptive feedback selection under matched rollout budgets. AdaGEPA also finds high-performing prompts earlier across several tasks. In the initial Schema-Guided Dialogue (SGD) study, its half-budget prompts outperform the non-adaptive baseline's full-budget prompts in joint goal accuracy on new dialogues from services seen and unseen during search. Overall, our findings highlight the potential of adaptive feedback allocation to improve both the effectiveness and rollout-budget efficiency of reflective prompt optimization.
Chinese Translation
提示优化通过改进提示来提高语言模型系统在下游任务上的表现。经典方法在任务示例上评估提示，并利用得到的反馈通过反思指导提示的修订。然而，当反馈选择未考虑提示的弱点时，这些修订可能在选定示例上提高表现，但未能带来更广泛的任务改进。为了解决这个问题，我们提出了AdaGEPA，一种自适应反馈分配方法，利用提示的表现和任务结构选择下一个提示修订的示例。我们的方法在每个反馈小批量中最多替换一个示例，以针对已识别的弱点，同时保留其余的反馈上下文。在我们对六个下游基准的主要实验中，AdaGEPA在匹配的回滚预算下实现了比非自适应反馈选择更高的平均验证分数。AdaGEPA还在多个任务中更早地找到高性能提示。在初步的架构引导对话（Schema-Guided Dialogue, SGD）研究中，其半预算提示在来自已见和未见服务的新对话中的联合目标准确率上超越了非自适应基线的全预算提示。总体而言，我们的研究结果突显了自适应反馈分配在提高反思性提示优化的有效性和回滚预算效率方面的潜力。
cs.CL / 96 / 2609.39938
LEAP: Learned Block-wise Evidence Retrieval for Long Audio-Video Perception
LEAP：用于长音频-视频感知的学习块级证据检索
Abstract
Hour-scale audio-visual question answering is constrained by a context dilemma: dense whole-recording encoding rapidly exhausts context limits, whereas uniform temporal compression severely dilutes fine-grained acoustic and visual evidence. We introduce LEAP, a framework where the model retrieves its own evidence without placing the whole recording in one context. LEAP divides a recording into fixed-duration blocks, applying a lightweight localization pass to each block to score short candidate windows. The highest-ranked windows are pooled and re-encoded in a single bounded answer pass. Consequently, the answer input and peak context remain independent of the recording duration. By decoupling evidence localization from reasoning, our framework can localize candidate temporal windows over pre-computed transcripts without decoding media frames, while preserving fine-grained visual and non-speech evidence by routing the final answering pass over raw audio-visual streams. LEAP trains both stages: a localization LoRA improves the selected windows, and an answer LoRA improves the answers read from the same windows. The block grid natively supports causal queries, enabling LEAP to support streaming inference without streaming-specific training. Across several AVQA benchmarks, LEAP improves over the Qwen3-Omni-30B-A3B baseline by 4.5-16.8%, and transfers to a second omni-modal backbone, MiniCPM-o 4.5, surpassing its published results by 3.1-13.0%.
Chinese Translation
小时级音视频问答受到上下文困境的限制：密集的全录音编码迅速耗尽上下文限制，而均匀的时间压缩严重稀释了细粒度的声学和视觉证据。我们提出了LEAP，一个框架，其中模型在不将整个录音置于一个上下文中的情况下检索其自身的证据。LEAP将录音划分为固定时长的块，对每个块应用轻量级的定位过程，以对短候选窗口进行评分。最高排名的窗口被汇总并在单个有界回答过程中重新编码。因此，答案输入和峰值上下文与录音时长保持独立。通过将证据定位与推理解耦，我们的框架可以在不解码媒体帧的情况下，基于预先计算的转录文本定位候选时间窗口，同时通过在原始音视频流上路由最终回答过程来保留细粒度的视觉和非语音证据。LEAP训练了两个阶段：定位的LoRA改善了所选窗口，而回答的LoRA改善了从同一窗口读取的答案。块网格本质上支持因果查询，使LEAP能够支持流式推理而无需特定于流式的训练。在多个音视频问答基准测试中，LEAP相较于Qwen3-Omni-30B-A3B基线提高了4.5-16.8%，并成功转移到第二个全模态骨干网络MiniCPM-o 4.5，超越了其已发布的结果3.1-13.0%。
cs.CL / 97 / 2609.39972
UBTree: Parallel Tree Drafting via Unigram and Bigram Models for Speculative Decoding
UBTree：通过单字模型和双字模型进行平行树草拟的推测解码
Abstract
Speculative decoding accelerates language model inference by verifying multiple draft tokens in a single target-model pass. Recent parallel drafters have achieved breakthrough performance in frontier production models, but their effectiveness deteriorates as the entropy of target distributions increases due to insufficient draft diversity. To overcome this bottleneck without sacrificing parallelism, we introduce UBTree, a parallel drafter that couples a Unigram proposer with a Bigram selector to construct drafting Trees. The unigram proposer is trained with the standard cross-entropy objective to generate candidate tokens independently for each position, while a lightweight bigram selector predicts transition scores between adjacent candidate pairs. Unlike the proposer, the selector is trained with a renormalized KL objective on high-temperature data. This tree-native training broadens the supervision beyond the greedy path, encouraging plausible alternative branches that improve the chance of accepting additional tokens during tree verification. Across seven standardized benchmarks with Qwen3-4B and Qwen3-8B, UBTree achieves an average speedup of $5.84$--$6.94\times$ over autoregressive decoding and outperforms DARTree in all 28 comparisons. Production-scale evaluation further demonstrates UBTree's advantage over frontier baselines such as DSpark.
Chinese Translation
推测解码通过在单个目标模型传递中验证多个草拟标记来加速语言模型推理。最近的平行草拟器在前沿生产模型中取得了突破性性能，但由于草拟多样性不足，目标分布的熵增加时其有效性会下降。为了在不牺牲并行性的情况下克服这一瓶颈，我们提出了UBTree，一种将单字提议器与双字选择器结合的平行草拟器，以构建草拟树。单字提议器使用标准交叉熵目标进行训练，以独立生成每个位置的候选标记，而轻量级双字选择器则预测相邻候选对之间的转移分数。与提议器不同，选择器是在高温数据上使用重新归一化的KL目标进行训练的。这种树原生训练拓宽了监督范围，超越了贪婪路径，鼓励合理的替代分支，从而提高在树验证过程中接受额外标记的机会。在与Qwen3-4B和Qwen3-8B的七个标准化基准测试中，UBTree在自回归解码上实现了平均$5.84$--$6.94 imes$的加速，并在所有28次比较中优于DARTree。生产规模评估进一步证明了UBTree相对于前沿基线（如DSpark）的优势。
cs.CL / 98 / 2609.39975
Overview of BioASQ 2026: The fourteenth BioASQ Challenge on Large-Scale Biomedical Semantic Indexing and Question Answering
BioASQ 2026 概述：第十四届 BioASQ 大规模生物医学语义索引与问答挑战
Nentidis, Anastasios, Katsimpras, Georgios, Krithara, Anastasia, Krallinger, Martin, Rodríguez-Ortega, Miguel, Rodriguez-López, Eduard, Loukachevitch, Natalia, Rozhkov, Igor, Tutubalina, Elena, Dimitriadis, Dimitris, Patsiou, Vasiliki, Tsoumakas, Grigorios, Giannakoulas, George, Bekiaridou, Alexandra, Samaras, Athanasios, Di Nunzio, Giorgio Maria, Ferro, Nicola, Marchesin, Stefano, Martinelli, Marco, Silvello, Gianmaria, Paliouras, Georgios
Abstract
This paper presents an overview of the fourteenth edition of the BioASQ challenge, organized in the context of the Conference and Labs of the Evaluation Forum (CLEF) 2026. BioASQ is an international challenge series that supports progress in biomedical language processing tasks ranging from semantic indexing and information extraction to question answering and summarization. In 2026, BioASQ included six shared tasks: a) Task 14b on biomedical semantic question answering. b) Task Synergy14 on question answering for developing biomedical top- ics. c) Task MultiClinSum-2 on multilingual clinical summarization. d) Task BioNNE-R on extracting relations between nested named entities in Russian and English. e) Task ELCardioCC on clinical coding in cardiology. f) Task GutBrainIE on gut-brain interplay information extrac- tion. Across these six tasks, 87 distinct teams participated, submitting more than 1000 runs overall. As in previous editions, several submissions reached competitive performance, reflecting the continued progress of state-of-the-art methods across biomedical language processing tasks.
Chinese Translation
本文概述了第十四届 BioASQ 挑战，该挑战在 2026 年的评估论坛会议与实验室（CLEF）背景下组织。BioASQ 是一个国际挑战系列，旨在推动生物医学语言处理任务的发展，涵盖语义索引、信息提取、问答和摘要等领域。在 2026 年，BioASQ 包括六个共享任务：a) 任务 14b，生物医学语义问答。b) 任务 Synergy14，针对生物医学主题的问答。c) 任务 MultiClinSum-2，多语言临床摘要。d) 任务 BioNNE-R，提取俄语和英语中嵌套命名实体之间的关系。e) 任务 ELCardioCC，心脏病学中的临床编码。f) 任务 GutBrainIE，肠脑相互作用信息提取。在这六个任务中，共有 87 支不同的团队参与，总共提交了超过 1000 次运行。与以往的版本一样，多个提交达到了竞争性表现，反映了生物医学语言处理任务中最先进方法的持续进展。
cs.CL / 99 / 2609.39982
Mid-Harness: Scaling Actions Between Model and Harness for Terminal Agents
中间工具带：在模型与工具之间扩展终端代理的动作
Abstract
Terminal agents act through stochastic model generations, yet the ability to generate a useful action does not ensure its reliable execution. A poor command (e.g., wrong package install) can change the environment in ways that hinder subsequent progress, even when the model could generate a better alternative. We investigate whether allocating test-time compute at the model-harness boundary can improve action reliability and trajectory success, and what makes this allocation effective. To study these questions, we introduce Mid-Harness, which samples and verifies candidate actions before forwarding one for execution, while keeping the generator and harness unchanged. With a TMAX-9B generator, more action sampling yields little benefit under weak verification, whereas a capable verifier can exploit useful alternatives from the same generator. On TerminalBench-Lite, a GPT-5.6 Sol verifier raises Pass@1 from 50.00% for the base agent to 68.03% with 8 sampled actions. When the same TMAX-9B model serves as the verifier, pairwise verification performs best among the evaluated verification mechanisms. Distilling responses from the stronger verifier into TMAX-9B further improves Pass@1, while leaving the action generator unchanged. With TMAX-9B on TerminalBench-Lite, combining action and trajectory scaling reaches higher success at lower estimated token cost than generating more trajectories alone. Mid-Harness also improves performance across additional models, benchmarks, and harnesses. These findings identify action scaling as a promising target for test-time compute scaling in terminal agents.
Chinese Translation
终端代理通过随机模型生成进行操作，但生成有用动作的能力并不确保其可靠执行。一个不当的指令（例如，错误的包安装）可能以阻碍后续进展的方式改变环境，即使模型能够生成更好的替代方案。我们研究在模型与工具边界分配测试时计算资源是否能提高动作的可靠性和轨迹成功率，以及什么因素使这种分配有效。为研究这些问题，我们引入了中间工具带（Mid-Harness），该方法在转发一个动作进行执行之前对候选动作进行采样和验证，同时保持生成器和工具不变。在使用 TMAX-9B 生成器时，更多的动作采样在弱验证下几乎没有收益，而一个有效的验证器可以从同一生成器中利用有用的替代方案。在 TerminalBench-Lite 上，GPT-5.6 Sol 验证器将基础代理的 Pass@1 从 50.00% 提高到 68.03%，通过采样 8 个动作。当同一 TMAX-9B 模型作为验证器时，成对验证在评估的验证机制中表现最佳。从更强的验证器中提炼响应到 TMAX-9B 进一步提高了 Pass@1，同时保持动作生成器不变。在 TerminalBench-Lite 上，结合动作和轨迹扩展在估计的标记成本较低的情况下达到更高的成功率，而不是单独生成更多轨迹。中间工具带还在其他模型、基准和工具中改善了性能。这些发现将动作扩展识别为终端代理测试时计算扩展的一个有前景的目标。
cs.CL / 100 / 2609.40035
OPTS-TTPO: Enhancing Finite-Sample Policy-Gradient Learning with Tree Search
OPTS-TTPO：通过树搜索增强有限样本策略梯度学习
Abstract
The policy-gradient theorem gives the exact gradient under the current policy, but finite on-policy samples may miss rare high-return trajectories. We study whether tree search improves their coverage within a fixed budget while controlling gradient bias. We introduce On-Policy Parallel Tree Search (OPTS) and Tree Trajectory Policy Optimization (TTPO) using on-policy tree trajectories, which sample new suffixes from the current policy at visited states. This needs no action-distribution correction, although branching changes state visitation. Our Branch Aggregation Lemma shows that branch-weighted tree statistics recover chain expectations when branch choices and weights are fixed before outgoing transitions are sampled. OPTS selects expansion states using estimated performance differences. Under deterministic dynamics, exact values, and max-backup advantages, the induced search policy's expected return improves monotonically with the budget. We bound the gradient bias from adaptive expansion and show that max backup assigns prefix credit to actions leading to better discovered suffixes. Against a finite chain reference, TTPG's measured bias stays near its no-branching level, while NaivePG's bias grows from 0.1251 to 0.4884. At matched budgets, reward- and value-guided OPTS improve correct-answer coverage and majority-vote accuracy over independent sampling. At matched branch counts, OPTS + TTPG gains coverage with a modest bias increase relative to Fixed-branch + TTPG. Under matched interaction or rollout budgets, OPTS-TTPO improves MuJoCo tail returns over PPO by up to 28.6%, achieves a 34-22-1 win-loss-tie record against PPO on Atari-57 under the last-100-log mean-return metric, and improves micro-averaged avg@32 and pass@32 over PPO across all four Qwen3 models.
Chinese Translation
策略梯度定理在当前策略下给出了精确的梯度，但有限的在政策样本可能会错过稀有的高回报轨迹。我们研究树搜索是否能够在固定预算内改善这些轨迹的覆盖，同时控制梯度偏差。我们引入了在政策并行树搜索（On-Policy Parallel Tree Search, OPTS）和树轨迹策略优化（Tree Trajectory Policy Optimization, TTPO），使用在政策树轨迹，在访问的状态下从当前策略中采样新的后缀。这不需要动作分布的修正，尽管分支会改变状态访问。我们的分支聚合引理表明，当在采样出边际转移之前固定分支选择和权重时，分支加权树统计量能够恢复链期望。在确定性动态、精确值和最大备份优势下，诱导搜索策略的期望回报随着预算的增加单调改善。我们界定了来自自适应扩展的梯度偏差，并表明最大备份将前缀信用分配给导致更好发现后缀的动作。与有限链参考相比，TTPG的测量偏差保持在其无分支水平附近，而NaivePG的偏差从0.1251增长到0.4884。在匹配预算下，奖励和价值引导的OPTS在独立采样中改善了正确答案的覆盖率和多数投票的准确性。在匹配分支数量下，OPTS + TTPG相较于固定分支 + TTPG在覆盖率上获得了适度的偏差增加。在匹配的交互或回滚预算下，OPTS-TTPO在MuJoCo尾部回报上比PPO提高了高达28.6%，在Atari-57上以最后100个日志的平均回报指标取得了34胜22负1平的战绩，并在所有四个Qwen3模型上改善了相较于PPO的微平均avg@32和pass@32。
cs.CL / 101 / 2609.40041
MGhana-ST: A Low-Resource Speech Translation Dataset for Ghanaian Languages and an Analysis of Multilingual Training Trade-offs
MGhana-ST：一个针对加纳语言的低资源语音翻译数据集及多语言训练权衡分析
Abstract
We present MGhana-ST, a speech translation dataset for four low-resource Ghanaian language varieties: Ga, Twi (Akuapem and Asante), Ewe, and Fante. MGhana-ST is an ongoing annotation effort; the experiments here use a fixed subset of about 16.1 hours of paired speech and English translations. The audio is curated from two existing Ghanaian speech resources. Unlike in those resources, the English translations are produced directly from audio by 37 native-speaker annotators and include verbal and non-verbal event annotations. Using Whisper-small, we compare monolingual and multilingual training under severe data scarcity, reporting means over three seeds. Flat multilingual training benefits no variety in this regime. Ga and Twi are unchanged within seed variance (+0.51 and +0.06 BLEU against monolingual standard deviations of 1.63 and 2.20), while Ewe declines by 6.99 BLEU and Fante by 5.11. The degrading varieties are Ewe, which is linguistically distinct and drawn from a different source corpus, and Fante, the least-resourced. Comparing empirical cross-lingual transfer with typology-based similarity, we find that transfer BLEU identifies closely interacting language pairs better than URIEL similarity, though neither predicts which varieties benefit from joint training. We also report a methodological finding. An earlier single-run analysis found positive transfer for three of four varieties; this did not survive replication across seeds. For Ga and Twi, monolingual baselines trained on 1.6 to 6.2 hours of audio have seed standard deviations roughly five and thirty times those of the multilingual models (0.35 and 0.07 BLEU). When the monolingual condition is noisier, a single-run comparison can show apparent transfer of this size from seed variation alone. We release MGhana-ST to support research on African language speech technology and low-resource speech translation.
Chinese Translation
我们提出了MGhana-ST，这是一个针对四种低资源加纳语言变体的语音翻译数据集：Ga、Twi（Akuapem和Asante）、Ewe和Fante。MGhana-ST是一个正在进行的标注工作；本实验使用了约16.1小时的配对语音和英语翻译的固定子集。音频来自两个现有的加纳语音资源。与这些资源不同的是，英语翻译是由37名母语标注员直接从音频中生成的，并包含了口头和非口头事件的注释。使用Whisper-small，我们在严重数据匮乏的情况下比较了单语和多语训练，报告了三个种子的均值。在这种情况下，平坦的多语训练对任何变体都没有益处。Ga和Twi在种子方差内保持不变（相对于单语标准偏差，分别增加了+0.51和+0.06 BLEU），而Ewe下降了6.99 BLEU，Fante下降了5.11。下降的变体是Ewe，它在语言上是独特的，并且来自不同的源语料库，以及资源最少的Fante。比较经验性的跨语言迁移与基于类型学的相似性，我们发现迁移BLEU比URIEL相似性更好地识别出紧密互动的语言对，尽管两者都无法预测哪些变体能从联合训练中受益。我们还报告了一个方法论发现。早期的单次运行分析发现四种变体中有三种存在正迁移；这一结果在跨种子复制时未能保持。对于Ga和Twi，基于1.6到6.2小时音频训练的单语基线的种子标准偏差大约是多语模型的五倍和三十倍（分别为0.35和0.07 BLEU）。当单语条件更嘈杂时，单次运行比较可能仅因种子变异而显示出这种规模的明显迁移。我们发布MGhana-ST以支持对非洲语言语音技术和低资源语音翻译的研究。
cs.CL / 102 / 2609.40064
From Tweets to Trades: Analyzing the Influence of Public Mood over Stock Market Performance in Turkiye
从推文到交易：分析公众情绪对土耳其股市表现的影响
Abstract
Purpose: This study examines whether domain-specific public mood is associated with stock-market dynamics and whether these relationships vary across communication domains and market conditions. It distinguishes public mood from investor sentiment and investigates whether heterogeneous sources of public communication exhibit different relationships with market behaviour. Design: The study analyses 610,422 posts published by 176 curated X accounts between January 2022 and December 2023, covering Politics and Government, Economy and Finance, and Media and Society. Posts are classified using fine-tuned Turkish transformer models under three domain-specific and one pooled regime. Public mood measures are constructed at daily, weekly, and monthly frequencies and examined alongside BIST100 and BIST30 market measures using correlation, Granger causality, vector autoregression, and impulse response analyses across the full period and selected market conditions. Findings: Public mood is not associated with the direction of stock-market returns but is associated with the magnitude of price movements, particularly for Media and Society and pooled communication. These relationships become stronger at longer aggregation frequencies. Predictive relationships are concentrated in Economy and Finance communication, while their magnitude and direction vary across market conditions, particularly during the 2023 election period. The pooled measure largely reflects the most active communication domain. Originality: The study contributes to behavioral-finance research by incorporating communication - domain heterogeneity into the analysis of public mood and market dynamics. It also demonstrates how aggregating heterogeneous sources can obscure domain-specific relationships between public communication and financial markets.
Chinese Translation
目的：本研究探讨领域特定的公众情绪是否与股市动态相关，以及这些关系在不同的沟通领域和市场条件下是否存在差异。研究将公众情绪与投资者情绪区分开来，并调查不同来源的公众沟通是否与市场行为表现出不同的关系。设计：本研究分析了2022年1月至2023年12月期间，由176个精选的X账户发布的610,422条帖子，涵盖政治与政府、经济与金融以及媒体与社会。帖子使用经过微调的土耳其变换器模型在三个领域特定和一个汇总的框架下进行分类。公众情绪指标在日、周和月频率下构建，并与BIST100和BIST30市场指标一起，通过相关性、格兰杰因果关系、向量自回归和脉冲响应分析进行检验，涵盖整个研究期间及特定市场条件。发现：公众情绪与股市回报的方向无关，但与价格波动的幅度相关，尤其是在媒体与社会以及汇总沟通领域。这些关系在较长的聚合频率下变得更强。预测关系主要集中在经济与金融沟通中，而其幅度和方向在不同市场条件下存在差异，特别是在2023年选举期间。汇总指标在很大程度上反映了最活跃的沟通领域。原创性：本研究通过将沟通领域的异质性纳入公众情绪与市场动态的分析，丰富了行为金融研究。同时，研究还展示了聚合异质来源如何模糊公众沟通与金融市场之间的领域特定关系。
cs.CL / 103 / 2609.40097
AutoDataBench: A Data-centric Testbed for Accelerating Auto Research
AutoDataBench：加速自动化研究的数据中心测试平台
Abstract
Existing auto-research benchmarks often entangle multiple sources of improvement, including training frameworks, hyperparameters, compute budgets, and data, making it difficult to attribute why one frontier agent outperforms another to specific research capabilities. In this work, we isolate and systematically evaluate Data Intelligence: an agent's ability to understand, manipulate, and improve the data that shapes model capabilities. We introduce AutoDataBench, a controlled testbed built on a conceptual framework of data intelligence spanning data diagnosis, data organization, and data construction, instantiated through three highly curated optimization tasks while holding non-data factors fixed. Across tool use, retrieval, and knowledge injection, we evaluate frontier LLMs' ability to improve training data through iterative experimentation under task-specific resource budgets. Beyond optimization performance, we ask: do LLMs understand what their data interventions do? We compare predictions made before training with observed outcomes to seek evidence of data-effect reasoning beyond trial and error, and explore whether iterative feedback helps LLMs better understand how changes to training data affect model performance. Finally, we show that reusing AutoDataBench trajectories for mid-training improves downstream coding performance, highlighting its value in both evaluating data intelligence and generating high-quality training data. Code and resources are available at https://github.com/AutoDataBench/AutoDataBench.
Chinese Translation
现有的自动化研究基准往往将多个改进来源交织在一起，包括训练框架、超参数、计算预算和数据，这使得很难将某个前沿代理的优越表现归因于特定的研究能力。在本研究中，我们孤立并系统地评估数据智能：代理理解、操控和改善塑造模型能力的数据的能力。我们引入了AutoDataBench，一个基于数据智能概念框架构建的受控测试平台，该框架涵盖数据诊断、数据组织和数据构建，通过三个高度策划的优化任务实例化，同时保持非数据因素不变。在工具使用、检索和知识注入方面，我们评估前沿大型语言模型（LLMs）在特定任务资源预算下，通过迭代实验改善训练数据的能力。除了优化性能，我们还提出问题：LLMs是否理解其数据干预的作用？我们比较训练前的预测与观察到的结果，以寻找超越试错的数据效应推理的证据，并探讨迭代反馈是否帮助LLMs更好地理解训练数据的变化如何影响模型性能。最后，我们展示了在中期训练中重用AutoDataBench轨迹可以提高下游编码性能，突显其在评估数据智能和生成高质量训练数据方面的价值。代码和资源可在 https://github.com/AutoDataBench/AutoDataBench 获取。
cs.CL / 104 / 2609.40103
JuryFlow: Disagreement-Guided Human-in-the-Loop Multi-Agent Evaluation
JuryFlow：基于分歧引导的人机协作多智能体评估
Abstract
Large language models (LLMs) are increasingly deployed as automated judges for AI-generated content, yet a single judge is unreliable and even a panel of judges leaves a hard residue: when judges disagree, majority voting discards the conflict instead of resolving it. We present JuryFlow, a disagreement-guided, human-in-the-loop multi-agent evaluation framework that treats inter-judge disagreement not as noise to be averaged away, but as a precise, claim-level signal indicating where an evaluation is uncertain. JuryFlow decomposes each candidate response into atomic claims, has a panel of heterogeneous judges assign per-claim verdicts, and builds a disagreement graph whose nodes are scored by verdict entropy and whose edges encode structural similarity between claims. A human acts as a structural guide, selecting which disagreement to resolve through a single, minimal intervention rather than re-labeling the response, after which the focal claim is re-evaluated, the correction propagates along graph edges and to historically similar cases, and is crystallized into reusable rubric entries that all judges inherit, making the evaluator progressively self-refining. To enable large-scale, reproducible benchmarking without human studies, we evaluate JuryFlow in an automatic configuration in which focal selection is made by entropy ranking. On MT-Bench and LLMBar, JuryFlow improves agreement with gold labels over single-judge and majority-vote panel baselines, and ablations isolate the contributions of disagreement-targeted re-evaluation, propagation, and rubric induction. We contribute (1) a human-in-the-loop paradigm that recasts the human from labeler to structural guide, (2) the JuryFlow framework operationalizing it through a disagreement graph, focal re-evaluation, and closed-loop rubric induction, and (3) an evaluation protocol with ablations that isolate where the gains originate.
Chinese Translation
大型语言模型（LLMs）越来越多地被用作AI生成内容的自动评审，但单一评审者的可靠性不足，即使是评审小组也会留下难以解决的问题：当评审者之间存在分歧时，多数投票会忽略冲突而不是解决它。我们提出了JuryFlow，一种基于分歧引导的人机协作多智能体评估框架，它将评审者之间的分歧视为一种精确的、声明级别的信号，表明评估的不确定性，而不是需要被平均掉的噪声。JuryFlow将每个候选响应分解为原子声明，由一组异质评审者为每个声明分配裁决，并构建一个分歧图，其节点通过裁决熵进行评分，边缘则编码声明之间的结构相似性。人类作为结构引导者，通过一次最小干预选择要解决的分歧，而不是重新标记响应，之后重新评估焦点声明，修正通过图边缘传播到历史上相似的案例，并凝结为所有评审者共享的可重用评分标准条目，使评估者逐步自我优化。为了实现大规模、可重复的基准测试而不需要人类研究，我们在自动配置下评估JuryFlow，其中焦点选择通过熵排名进行。在MT-Bench和LLMBar上，JuryFlow在与黄金标签的一致性方面优于单一评审者和多数投票小组基线，消融实验则隔离了针对分歧的重新评估、传播和评分标准引导的贡献。我们的贡献包括（1）一种人机协作范式，将人类角色重新定义为结构引导者，而非标记者，（2）通过分歧图、焦点重新评估和闭环评分标准引导来实现这一范式的JuryFlow框架，以及（3）一种评估协议，结合消融实验以隔离收益来源。
cs.CL / 105 / 2609.40108
OverdoseMoE: A Multi-Expert Framework for Opioid Overdose Risk Prediction
OverdoseMoE：一种多专家框架用于阿片类药物过量风险预测
Abstract
Opioid overdose remains a major clinical and public health burden, highlighting the need for scalable approaches to identify patients at high risk. Here, we investigate diagnosis-specific adaptation for 180-day opioid overdose risk prediction from patients' preceding one-year longitudinal ICD histories. We develop OODMAMBA and OODQWEN through continued pretraining on longitudinal diagnostic sequences followed by task-specific fine-tuning. Building on the stronger Qwen-based predictors, we further propose OVERDOSEMOE, a multi-expert framework that integrates models of different scales using complementary expert-weighting strategies. Diagnosis-specific adaptation consistently improved predictive performance over general-purpose language-model baselines, with OODQWEN achieving an AUPRC of 24.47 and an AUROC of 68.56. OVERDOSEMOE further improved discrimination and precision, achieving an AUPRC of 25.17 and an AUROC of 69.49 while outperforming the strongest single-model baselines. Among patients ranked in the top 5% of predicted risk, OVERDOSEMOE identified substantially enriched overdose risk, achieving a PPV of 25.38% while retaining meaningful recall. Evaluation on an independent MIMIC-IV cohort further demonstrated cross-cohort robustness, with complementary weighting strategies showing advantages across different performance measures. These findings demonstrate that diagnosis-specific language-model adaptation combined with multi-expert integration can improve opioid overdose risk stratification and support more robust prediction across heterogeneous electronic health record populations.
Chinese Translation
阿片类药物过量仍然是一个重大的临床和公共卫生负担，这突显了识别高风险患者的可扩展方法的必要性。在此，我们研究了基于诊断的适应性，以从患者前一年纵向 ICD 历史中预测 180 天的阿片类药物过量风险。我们通过对纵向诊断序列的持续预训练，随后进行任务特定的微调，开发了 OODMAMBA 和 OODQWEN。在更强的基于 Qwen 的预测器的基础上，我们进一步提出了 OVERDOSEMOE，这是一种多专家框架，利用互补的专家加权策略整合不同规模的模型。基于诊断的适应性在预测性能上始终优于通用语言模型基线，其中 OODQWEN 达到了 24.47 的 AUPRC 和 68.56 的 AUROC。OVERDOSEMOE 进一步提高了区分度和精确度，达到了 25.17 的 AUPRC 和 69.49 的 AUROC，同时超越了最强的单模型基线。在预测风险排名前 5% 的患者中，OVERDOSEMOE 显著识别出过量风险的富集，达到了 25.38% 的阳性预测值（PPV），同时保持了有意义的召回率。在独立的 MIMIC-IV 队列上的评估进一步证明了跨队列的稳健性，互补加权策略在不同性能指标上显示出优势。这些发现表明，基于诊断的语言模型适应性结合多专家集成可以改善阿片类药物过量风险分层，并支持在异质电子健康记录人群中更稳健的预测。
cs.CL / 106 / 2609.40118
Persistent Context Graphs for Efficient Memory Compaction in LLM Agents
用于大规模语言模型代理的高效内存压缩的持久上下文图
Abstract
As LLM capabilities advance, agents are tackling increasingly complex tasks over longer horizons. Their growing interaction histories make memory compaction essential for staying within context windows and reducing prefill cost. Existing methods summarize the history or compress its KV cache, often adding model computation to preserve information for future requests. A new user request can change which history matters, but reassessing that history with the model requires re-encoding it if the KV cache has expired. Past attention provides signals of historical importance and dependencies between messages, while relevance to the current task must be assessed using the new user request. We introduce ReCAP, a memory compaction method that stores attention-derived importance scores and dependency links in a lightweight, persistent context graph. For each new request, ReCAP combines stored importance with relevance cues from the request and follows dependency links to select messages and their supporting context, without additional model calls for selection. Compared with Codex's default summarization-based compaction, ReCAP reduces estimated latency for compaction and cold restoration by approximately 95% on both Qwen3-Coder and gpt-oss. It also roughly halves the historical context per call on SWE-Together at comparable task quality and improves accuracy on the code tasks of Lost-in-Conversation over full history by 19.8 and 41.2 points.
Chinese Translation
随着大规模语言模型（LLM）能力的提升，代理正在处理越来越复杂的任务，时间跨度也在不断延长。它们日益增长的交互历史使得内存压缩变得至关重要，以保持在上下文窗口内并降低预填充成本。现有方法通常通过总结历史或压缩其键值（KV）缓存来实现，往往需要增加模型计算以保留未来请求的信息。新的用户请求可能会改变重要历史的定义，但如果KV缓存已过期，使用模型重新评估该历史需要重新编码。过去的注意力提供了历史重要性和消息之间依赖关系的信号，而与当前任务的相关性必须通过新的用户请求进行评估。我们提出了ReCAP，一种内存压缩方法，它在轻量级的持久上下文图中存储基于注意力的导出重要性分数和依赖链接。对于每个新的请求，ReCAP结合存储的重要性与请求中的相关线索，并遵循依赖链接选择消息及其支持上下文，而无需额外的模型调用进行选择。与Codex的默认基于总结的压缩方法相比，ReCAP在Qwen3-Coder和gpt-oss上将压缩和冷恢复的估计延迟减少了约95%。在SWE-Together上，它还将每次调用的历史上下文大致减半，同时保持相似的任务质量，并在Lost-in-Conversation的代码任务上提高了19.8和41.2个点的准确性。
cs.CL / 107 / 2609.40121
On the (In)effectiveness of AMR Augmentation for Large Language Models
大语言模型中AMR增强的（无）效用
Abstract
While Abstract Meaning Representation (AMR) has historically improved performance on a range of NLP tasks, the benefit---or lack thereof---of AMR augmentation for modern LLMs is thus far unclear. In this paper, we attempt to reproduce recent work that reported substantial downstream gains from AMR augmentation, finding that these are likely due to specific choices in the experimental settings used: using a consistent and unified protocol for hyperparameter selection, we observe that text-only baselines consistently match or exceed the performance of AMR-augmented models. To investigate this null result, we introduce a perplexity-based probe measuring the degree to which AMR provides an LLM with supplemental relational knowledge not already available to the model. We find that AMR augmentation does not help LLMs improve their understanding of relational content in the sentence, indicating that augmenting these models with AMR offers no clear benefit on downstream tasks.
Chinese Translation
尽管抽象意义表示（AMR）在历史上提升了多种自然语言处理（NLP）任务的性能，但AMR增强对现代大语言模型（LLMs）的益处——或缺乏益处——迄今尚不明确。在本文中，我们尝试重现最近的研究，该研究报告了AMR增强带来的显著下游收益，发现这些收益可能源于实验设置中的特定选择：通过使用一致且统一的超参数选择协议，我们观察到仅文本基线的性能始终与AMR增强模型相匹配或超过。为了调查这一无效结果，我们引入了一种基于困惑度的探测器，测量AMR为LLM提供的补充关系知识的程度，这些知识在模型中并不存在。我们发现AMR增强并未帮助LLM改善对句子中关系内容的理解，表明用AMR增强这些模型在下游任务中并没有明显的益处。
cs.CL / 108 / 2609.40124
Debias It Yourself: Teaching LLMs Cognitive Bias Mitigation Interventions
自行去偏见：教授大型语言模型认知偏见缓解干预措施
Abstract
Bias has long been studied in social psychology and cognitive science, where decades of research have produced a body of validated interventions that reduce stereotypical thinking and prejudiced responses in humans. We propose Debias It Yourself (DIY), a cognitively grounded framework that translates five such interventions into debiasing procedures for large language models and delivers them through three established paradigms: Show (in-context examples), Train (instruction tuning), and Revise (guided self-revision). Across three models, five bias benchmarks, eleven debiasing baselines, and three reasoning benchmarks, Train+Revise and Revise alone attain the top two average ranks, lead the bias-reasoning tradeoff (mean bias as low as 2% at 90% reasoning accuracy), and reduce bias on unseen dimensions by up to 14.8%. Our code and data are publicly available.
Chinese Translation
偏见在社会心理学和认知科学中已经被长期研究，数十年的研究产生了一系列经过验证的干预措施，能够减少人类的刻板印象思维和偏见反应。我们提出了自行去偏见（Debias It Yourself, DIY）框架，这是一个基于认知的框架，将五种此类干预措施转化为大型语言模型的去偏见程序，并通过三种已建立的范式进行实施：展示（Show，提供上下文示例）、训练（Train，指令调优）和修订（Revise，引导自我修订）。在三个模型、五个偏见基准、十一条去偏见基线和三个推理基准中，Train+Revise 和单独的 Revise 达到了前两名的平均排名，主导了偏见与推理的权衡（在90%推理准确率下，平均偏见低至2%），并在未见维度上减少偏见高达14.8%。我们的代码和数据是公开可用的。
cs.CL / 109 / 2609.40181
Index-Translate: A Multilingual Translation Model Family -- Text, Speech, Controlled Dubbing, and Long-Document Translation
Index-Translate：一个多语言翻译模型家族——文本、语音、受控配音和长文档翻译
Abstract
We introduce Index-Translate, a multilingual translation model family that combines a shared multilingual foundation with specialized training for general translation, instruction following, speech translation, controlled dubbing, and long-document translation. It includes three model sizes, 2B, 9B, and 35B-A3B, and supports translation in 150 languages, with multilingual instruction following. Evaluations on general translation and complex translation instructions show that Index-Translate outperforms translation models of comparable size and achieves performance comparable to 100B-scale translation models and frontier models. Index-Echo provides end-to-end speech-to-text and speech-to-speech translation, outperforming existing end-to-end models and achieving performance comparable to frontier omni models. Index-Homura extends the family to syllable-controlled dubbing. Index-NativeLong introduces native long-document translation with a dedicated task formulation and benchmark. These capabilities support diverse translation tasks, including multilingual content production.
Chinese Translation
我们介绍了Index-Translate，一个多语言翻译模型家族，它结合了共享的多语言基础和针对一般翻译、指令跟随、语音翻译、受控配音以及长文档翻译的专门训练。该模型包括三种规模：2B、9B和35B-A3B，支持150种语言的翻译，并具备多语言指令跟随功能。在一般翻译和复杂翻译指令的评估中，Index-Translate的表现优于同等规模的翻译模型，并且其性能可与100B规模的翻译模型和前沿模型相媲美。Index-Echo提供端到端的语音转文本和语音转语音翻译，超越了现有的端到端模型，且其性能可与前沿全能模型相媲美。Index-Homura将该家族扩展至音节控制的配音。Index-NativeLong引入了本地化长文档翻译，采用专门的任务表述和基准。这些能力支持多样化的翻译任务，包括多语言内容制作。
cs.CL / 110 / 2609.40185
Provably Tractable NFA-Constrained Language Generation via HMMs
可证明可处理的NFA约束语言生成通过隐马尔可夫模型
Abstract
Constrained generation aims to sample from language models (LMs) conditioned on hard constraints. Existing constrained-generation techniques for nondeterministic finite automaton (NFA) constraints either distort the distribution or sacrifice efficiency. Theoretically, this task reduces to counting the length-$n$ sequences accepted by an NFA (#NFA), and the exact #NFA problem is #P-complete. Recent work has shown that #NFA admits a fully polynomial randomized approximation scheme (FPRAS). Inspired by this result, we propose NFA-LM, a polynomial-time engine for NFA-constrained generation with theoretical guarantees under mild assumptions. Experiments show that NFA-LM efficiently generates high-quality outputs with theoretically bounded approximation error.
Chinese Translation
约束生成旨在从条件于硬约束的语言模型（LMs）中进行采样。现有的针对非确定性有限自动机（NFA）约束的约束生成技术要么扭曲了分布，要么牺牲了效率。从理论上讲，这一任务归结为计数被NFA接受的长度为$n$的序列（#NFA），而确切的#NFA问题是#P完全的。最近的研究表明，#NFA承认完全多项式随机近似方案（FPRAS）。受到这一结果的启发，我们提出了NFA-LM，一种在温和假设下具有理论保证的多项式时间NFA约束生成引擎。实验表明，NFA-LM能够高效生成高质量的输出，并具有理论上界定的近似误差。
cs.CL / 111 / 2609.40198
SCB: SpeechConversationBench for Evaluating Multi-Turn Reasoning in Speech-to-Speech Models
SCB：用于评估语音到语音模型中多轮推理的语音对话基准
Abstract
Speech-to-speech systems must solve tasks whose requirements emerge across conversational turns. We introduce SpeechConversationBench (SCB), a focused evaluation of spoken mathematical reasoning using 103 sharded GSM8K problems. The framework compares the original problem delivered in one turn (full), its concatenated information shards delivered together (concat), and incremental spoken disclosure across turns (sharded). We report final-answer accuracy for four commercial speech systems and LEGO, a proprietary speech pipeline developed internally by the SCBX Innovation Lab team with explicit conversational context management. Relative to concat, sharded accuracy decreases by 5.0-25.3 percentage points across the four commercial systems. LEGO achieves 77.5 percent accuracy in all three conditions, compared with 76.6 percent sharded accuracy for GPT-4o Realtime. The two single-turn baselines distinguish sensitivity to problem reformulation from the additional challenges introduced by incremental spoken interaction.
Chinese Translation
语音到语音系统必须解决在对话轮次中出现的任务需求。我们介绍了语音对话基准（SpeechConversationBench，SCB），这是一个针对口头数学推理的专门评估，使用了103个分割的GSM8K问题。该框架比较了在一个轮次中传递的原始问题（完整），其一起传递的连接信息片段（连接），以及跨轮次的增量口头披露（分割）。我们报告了四个商业语音系统和LEGO（由SCBX创新实验室团队内部开发的专有语音管道，具有明确的对话上下文管理）的最终答案准确率。与连接相比，四个商业系统的分割准确率下降了5.0-25.3个百分点。LEGO在所有三种条件下的准确率达到77.5%，而GPT-4o Realtime的分割准确率为76.6%。这两个单轮基线区分了对问题重构的敏感性与增量口头互动所带来的额外挑战。
cs.CL / 112 / 2609.40236
Comparison of techniques for fine-tuning open-weight models for entity extraction from radiology reports
开放权重模型在放射学报告中实体提取的微调技术比较
Abstract
Converting free-text radiology reports into structured labels supports cohort building, quality assurance, and monitoring of clinical imaging models, but the strongest label extractors are hosted proprietary models whose use raises privacy, cost, and reproducibility concerns. We asked whether a fine-tuned open-weight model (Gemma-3-12B) can match GPT-4o at multi-label intracranial hemorrhage (ICH) acuity extraction from non-contrast head-CT reports, and which ingredients matter. Using a 2x2 design, we crossed two adaptation strategies (a discriminative classification head, CH; generative instruction fine-tuning, IFT) with two training-data sources (distillation of real GPT-4o-labeled reports; synthetic reports generated by GPT-4o from real exemplars), across five training sizes, benchmarked on 100 expert-adjudicated reports against GPT-4o and the un-tuned open-weight base. The distilled instruction-tuned model (DIFT) matched GPT-4o (macro-F1 0.845 vs 0.850; p = 1.000) and exceeded the base model by 0.178. The decisive factor was the training-data source, not the fine-tuning method: both synthetic-data models failed to exceed the un-tuned open-weight base at any training size and underperformed the distilled models across all acuity classes. Fine-tuning and inference fit within the memory envelope of a single 24 GB consumer GPU. For narrow, high-value clinical label-extraction tasks, distilling real reports, rather than generating synthetic ones, is what closes the gap to a hosted model, enabling a private, low-cost, version-stable on-premises alternative.
Chinese Translation
将自由文本的放射学报告转换为结构化标签有助于队列构建、质量保证和临床影像模型的监测，但最强的标签提取器是托管的专有模型，其使用引发了隐私、成本和可重复性的问题。我们探讨了微调的开放权重模型（Gemma-3-12B）是否能够在非对比头部CT报告中与GPT-4o在多标签颅内出血（ICH）急性程度提取方面相匹配，以及哪些因素是关键。采用2x2设计，我们将两种适应策略（判别分类头，CH；生成指令微调，IFT）与两种训练数据来源（真实GPT-4o标记报告的蒸馏；由GPT-4o从真实示例生成的合成报告）交叉，涵盖五种训练规模，并在100份专家裁定的报告上与GPT-4o和未微调的开放权重基础模型进行基准测试。蒸馏指令微调模型（DIFT）与GPT-4o相匹配（宏F1 0.845对0.850；p = 1.000），并超过了基础模型0.178。决定性因素是训练数据来源，而非微调方法：两个合成数据模型在任何训练规模下均未能超过未微调的开放权重基础模型，并且在所有急性程度类别中表现不及蒸馏模型。微调和推理适合于单个24 GB消费级GPU的内存限制。对于狭窄、高价值的临床标签提取任务，蒸馏真实报告而非生成合成报告是缩小与托管模型差距的关键，从而实现私密、低成本、版本稳定的本地替代方案。
cs.CL / 113 / 2609.40286
Linguistic Loopholes in LLM Unlearning: From a 174-Language Benchmark to Coverage-Aware Unlearning
大语言模型去学习中的语言漏洞：从174种语言基准到覆盖感知的去学习
Abstract
Unlearning a fact in one language does not guarantee its removal in others as changing the query or even the requested answer language can reopen seemingly forgotten knowledge -- a cross-lingual loophole. The most straightforward solution to this challenge -- unlearning in all languages -- is neither scalable nor desirable as it amplifies damage to unrelated model capabilities. We introduce the task of language budgeted multilingual unlearning where the goal is to select a subset of languages that maximizes cross-lingual erasure. To study this task we introduce the Cross-Lingual Unlearning Tensor, an unlearning benchmark that spans 174 language--script pairs and 25 atomic paraphrase types to examine when forgetting generalizes across linguistic expressions of the same knowledge. We further propose COVER, which selects source languages to maximize predicted COVERage of languages receiving no forget supervision, enabling unlearning on a language budget. Surprisingly, we find naively selecting strong individual sources does not reliably compose into strong source sets motivating our development of COVER. At deployment COVER only requires benign calibration data and access to the frozen model. Across three model families and two disjoint forget sets, COVER reduces mean held-out residual access by 7.8--27.3% relative to uniform source selection. We find these gains extend beyond synthetic benchmarks to real news documents in low-resource language settings using human translated data from the Low Resource Languages for Emergent Incidents (LORELEI) corpus.
Chinese Translation
在一种语言中去学习一个事实并不能保证在其他语言中也能去除，因为更改查询或甚至请求的回答语言可能会重新打开看似被遗忘的知识——这是一种跨语言漏洞。解决这一挑战的最直接方法——在所有语言中去学习——既不可扩展也不可取，因为这会加剧对无关模型能力的损害。我们引入了语言预算多语言去学习的任务，其目标是选择一个语言子集，以最大化跨语言的遗忘。为了研究这一任务，我们引入了跨语言去学习张量（Cross-Lingual Unlearning Tensor），这是一个涵盖174种语言-脚本对和25种原子释义类型的去学习基准，用于检验遗忘在同一知识的语言表达中何时能够推广。我们进一步提出了COVER，它选择源语言以最大化未接受遗忘监督的语言的预测覆盖（COVERage），从而实现语言预算下的去学习。令人惊讶的是，我们发现简单地选择强个体源并不能可靠地组成强源集，这促使了我们对COVER的开发。在部署时，COVER仅需无害的校准数据和对冻结模型的访问。在三个模型系列和两个不重叠的遗忘集上，COVER相较于均匀源选择减少了7.8%至27.3%的平均保留残余访问。我们发现这些收益不仅限于合成基准，还扩展到使用来自低资源语言突发事件（LORELEI）语料库的人类翻译数据的真实新闻文档，在低资源语言环境中表现良好。
cs.CL / 114 / 2609.40295
How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text
AI令牌的价值是多少？野生AI生成网页文本的规模法则
Abstract
Web text makes up the majority of pretraining data and is increasingly AI-generated. After applying FineWeb quality filtering, we find that 27.5% of tokens from June 2026 web data are labeled as AI-generated by Pangram, rising to 31.1% by August. Unlike synthetic data or model-collapse setups, this *wild* AI text comes from many models, is written for human readers, and arrives unlabeled in pretraining corpora. How does AI text in the wild affect language model pretraining? To answer this question, we pretrain 800 language models, varying the ratio of added AI tokens to human tokens, and fit scaling laws to held-out losses on both human and AI-generated text. For data-starved models, adding AI tokens to pretraining data initially lowers loss on human text, but the benefit saturates as more are added and quickly *reverses* into harm. For models trained on high budgets of human text, AI tokens raise loss almost immediately, while the same number of fresh human tokens keeps lowering it. Scaling laws such as Hoffman et al. (2022) fail to predict this behavior. We propose a new scaling law with separate benefit and harm terms that allows the value of an AI token to change sign while also reducing to Chinchilla in the absence of AI text. When fit on smaller models, our scaling law predicts the effect of AI text on held-out human-text loss for models up to 3.6x larger with 41% lower error than the best existing law over all AI ratios. We recommend filtering AI text when the target is human text, repeating human text before expanding the training dataset with AI-generated web text, and reporting validation loss on human and AI text separately AI text remains valuable when the target is AI text. We release WildAI, an 83B-token corpus with AI, topic, and format labels, all 800 models and code at https://github.com/pangramlabs/WildAI.
Chinese Translation
网页文本占据了预训练数据的主要部分，并且越来越多地由AI生成。在应用FineWeb质量过滤后，我们发现2026年6月的网页数据中，有27.5%的令牌被Pangram标记为AI生成，到8月这一比例上升至31.1%。与合成数据或模型崩溃设置不同，这种*野生*的AI文本来自多个模型，旨在为人类读者撰写，并且在预训练语料库中未标记。野生AI文本如何影响语言模型的预训练？为了解答这个问题，我们预训练了800个语言模型，改变了添加的AI令牌与人类令牌的比例，并将规模法则拟合到人类和AI生成文本的保留损失上。对于数据匮乏的模型，向预训练数据中添加AI令牌最初降低了人类文本的损失，但随着添加的令牌增多，这种好处饱和并迅速*反转*为有害。对于在大量人类文本上训练的模型，AI令牌几乎立即提高了损失，而同样数量的新鲜人类令牌则持续降低损失。Hoffman等人（2022）的规模法则未能预测这种行为。我们提出了一种新的规模法则，具有独立的好处和危害项，允许AI令牌的价值改变符号，同时在没有AI文本的情况下简化为Chinchilla。当在较小的模型上拟合时，我们的规模法则预测了AI文本对保留人类文本损失的影响，适用于最大3.6倍的更大模型，错误率比现有最佳法则低41%。我们建议在目标是人类文本时过滤AI文本，在用AI生成的网页文本扩展训练数据集之前重复人类文本，并分别报告人类和AI文本的验证损失。当目标是AI文本时，AI文本仍然具有价值。我们发布了WildAI，一个包含AI、主题和格式标签的83B令牌语料库，以及所有800个模型和代码，网址为https://github.com/pangramlabs/WildAI。
cs.CL / 115 / 2609.40340
EvoDuet: Bilevel Co-Evolution of Web Searching and Task Solving for Scientific Discovery
EvoDuet：科学发现中网络搜索与任务解决的双层共同进化
Abstract
Evolutionary search with large language models (LLMs) can stall when progress requires external knowledge the model lacks. Supplying relevant documents helps, but simply adding web search tool can keep returning the same pages as solutions change. We introduce EvoDuet, a bi-level optimization method that co-evolves solutions and search queries with fixed model parameters. At each iteration, a retrieval gate lets the LLM assess its knowledge gap and choose to retrieve new documents, reuse stored ones, or proceed without them. An inner loop refines queries and ranks documents by the solution scores they are predicted to yield; an outer loop generates candidates in parallel from these documents and records the evaluated outcomes for later searches. Across 21 optimization tasks with one candidate per iteration, EvoDuet raises OpenEvolve's normalized discovery gain from 74.1% to 78.0% with GPT-5.6-Luna and from 61.3% to 82.3% with Gemini-3.8-Flash, whereas Qwen3.5-9B does not benefit. Our best runs surpass the previously reported best scores on eight tasks, including Swap Reduction on Q20 and Rosetta, and match them on three more. EvoDuet also improves with other scaffolds (e.g., Top-K, EvoX) on Sums/Diffs and Denoising, demonstrating its applicability across evolutionary search scaffolds.
Chinese Translation
使用大型语言模型（LLMs）的进化搜索在进展需要模型缺乏的外部知识时可能会停滞。提供相关文档有助于解决这个问题，但仅仅添加网络搜索工具可能会在解决方案变化时不断返回相同的页面。我们提出了EvoDuet，一种双层优化方法，它在固定模型参数的情况下共同进化解决方案和搜索查询。在每次迭代中，检索门允许LLM评估其知识差距，并选择检索新文档、重用存储的文档或在没有它们的情况下继续。内层循环通过预测的解决方案得分来细化查询并对文档进行排名；外层循环则从这些文档中并行生成候选项，并记录评估结果以供后续搜索使用。在21个优化任务中，每次迭代一个候选项，EvoDuet将OpenEvolve的标准化发现增益从74.1%提高到78.0%（使用GPT-5.6-Luna），并将其从61.3%提高到82.3%（使用Gemini-3.8-Flash），而Qwen3.5-9B则没有受益。我们的最佳运行在八个任务上超过了之前报告的最佳分数，包括Q20和Rosetta上的Swap Reduction，并在另外三个任务上与之持平。EvoDuet在其他框架（例如Top-K、EvoX）上也在Sums/Diffs和去噪任务中有所改善，展示了其在进化搜索框架中的适用性。