Visual shifts can cause a vision-language-action policy to fail after initially plausible behavior. We ask whether OpenVLA's internal activations contain signals associated with the steps before failure. We freeze the policy, record one MLP activation per LIBERO-10 step, and fit two linear monitors. Occlusion reduces task success from $57\%$ to $17\%$. Within failed matched-reset trajectories, a layer-16 logistic probe attains AUROC $0.972$ and AUPRC $0.352$, whereas action disagreement attains AUROC $0.496$. Without refitting, the occlusion-trained probe reaches AUROC $0.689$ on failed camera-jitter episodes. In a calibration check, however, the same layer-16 monitor averages 3.32 warning onsets per clean episode. This contrast shows that strong retrospective discrimination does not imply operationally quiet warning behavior. Because fitting and evaluation share tasks, resets, and seed, these results establish retrospective separability rather than prediction on independent episodes.
FailSafe: Reasoning and Recovery from Failures in Vision-Language-Action Models
Nanyang Technological University(南洋理工大学);Centre for Frontier AI Research, A*STAR(A*STAR前沿人工智能研究中心);Allen Institute for AI(艾伦人工智能研究所);University of Washington(华盛顿大学)
OGAM: Connecting Systematic Testing to Runtime Assurance through Object-Grounded Attention Monitoring for VLA Policies
Carl von Ossietzky University of Oldenburg(卡尔·冯·奥西茨基大学奥尔登堡分校);Huadian Electric Power Research Institute(华电电力科学研究院);Chalmers University of Technology(查尔姆斯理工大学)
VLA^2: Empowering Vision-Language-Action Models with an Agentic Framework for Unseen Concept Manipulation
Zhejiang University(浙江大学);MiLAB, Westlake University(西湖大学 MiLAB);Southern University of Science and Technology(南方科技大学);Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州))
See, Plan, Rewind: Progress-Aware Vision-Language-Action Models for Robust Robotic Manipulation
School of Information Science and Technology, University of Science and Technology of China(信息科学与技术学院,中国科学技术大学);University of Technology Sydney(新南威尔士大学);Department of Computer Vision, Mohamed Bin Zayed University of Artificial Intelligence(人工智能与计算机视觉系,Mohamed Bin Zayed人工智能大学);The University of Hong Kong(香港大学);Institute of AI for Industry, Chinese Academy of Sciences(产业人工智能研究所,中国科学院);School of Intelligent Science and Engineering, Harbin Institute of Technology (Shenzhen)(智能科学与工程学院,哈尔滨工业大学(深圳))
Vision-Language-Action Models for Selective Robotic Disassembly: A Case Study on Critical Component Extraction from Desktops
J. Mike Walker ’66 Department of Mechanical Engineering, Texas A&M University(J. Mike Walker ’66 机械工程系,德克萨斯A&M大学);Engineering School of Sustainable Infrastructure & Environment, University of Florida(可持续基础设施与环境工程学院,佛罗里达大学);Zachry Department of Civil and Environmental Engineering, Texas A&M University(土木与环境工程系,德克萨斯A&M大学)
Drift is a Sampling Error: SNR-Aware Power Distributions for Long-Horizon Robotic Planning
Chongqing Institute of Green and Intelligent Technology, Chinese Academy of Sciences(中国科学院重庆绿色智能技术研究院);Chongqing School, University of Chinese Academy of Sciences(中国科学院大学重庆学院)
TrustVLA: Mechanism-Guided Inference-Time Defense Against Vision-Language-Action Backdoors
School of Computer Science, Wuhan University(武汉大学计算机科学学院);D-Robotics(D-机器人公司);HUST(华中科技大学);Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所)