发表机构
Carl von Ossietzky University of Oldenburg; Huadian Electric Power Research Institute; Chalmers University of Technology（卡尔·冯·奥西茨基大学奥尔登堡分校; 华电电力科学研究院; 查尔姆斯理工大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
OGAM通过测试揭示的注意力分歧，在运行时监控VLA策略，早期阻止失败，无需失败标签训练，在四种策略中阻止87-100%失败情节。
AI 中文摘要
基准测试仅将视觉-语言-动作（VLA）策略暴露于少数规范指令，而全面的部署测试是不可能的。我们引入了面向对象的注意力监控（OGAM），将系统测试与运行时保障相连接：测试揭示了成功与失败执行之间的注意力分歧，OGAM利用这一信号在有限测试套件之外阻止失败。我们通过动作模板与对象的成对组合生成场景基础指令，并分别测试保持语义的改写。所有87个超出基准的案例在OpenVLA、OpenVLA-OFT、UniVLA和π₀.₅上均表现出问题行为：没有策略完成24个可行指令中的任何一个，而不可行或危险的请求也触发了行为替换。在每个动作查询时，我们通过对象掩码投影梯度加权视觉注意力，并按指令角色分组以跨任务和策略进行比较。动态时间规整将该过程与成功参考对齐，尽管存在速度差异；对成功情节的保形校准设置了持续偏离的早期停止阈值，名义误停目标为α=0.05。在四种策略中，OGAM在20秒预算内以中位时间5-12秒阻止了87-100%的失败情节，观察到的误停率为3-5%，且无需失败标签训练。因此，有限测试识别出注意力模式，支持在失败完全展开之前进行在线干预。
英文摘要
Benchmarks expose vision-language-action (VLA) policies to few canonical instructions, while exhaustive deployment testing is impossible. We introduce Object-Grounded Attention Monitoring (OGAM), connecting systematic testing to runtime assurance: testing reveals attention divergence between successful and failed executions, and OGAM uses this signal to stop failures beyond the finite suite. We generate scene-grounded instructions through pairwise combinations of action templates and objects, and separately test meaning-preserving paraphrases. All 87 out-of-benchmark cases reveal problematic behavior across OpenVLA, OpenVLA-OFT, UniVLA, and $π_{0.5}$: none completes any of the 24 feasible instructions, while infeasible or hazardous requests also trigger behavior substitution. At each action query, we project gradient-weighted visual attention through object masks and group it by instruction role for comparison across tasks and policies. Dynamic time warping aligns this course with a successful reference despite speed differences; conformal calibration on successful episodes sets the early-stopping threshold for sustained deviations, with a nominal false-stop target of $α=0.05$. Across four policies, OGAM stops 87-100% of failed episodes at median times of 5-12s within a 20s budget, with observed false-stop rates of 3-5%, without failure-labeled training. Finite testing thus identifies attention patterns that support online intervention before failure fully unfolds.
