[论文] Don't Drop the BATON: Long-Horizon Robot Manipulation via Agentic Subt...

论文概要

研究领域: Robotics 作者: Bingxin Xu, Yuzhang Shang, Emilio Ferrara 发布时间: 2026-08-17 arXiv: 2608.16889

中文摘要

长程机器人操作将多个接触密集型技能串联成多阶段任务。视觉-语言-动作（VLA）模型日益掌握单个技能，但串联仍然失败：错误累积超出策略的纠正能力，且一个子任务会静默约束下一个。一个有前景的方案是冻结VLA、让LLM智能体负责：用语言规划，用解析基元在自由空间中移动，仅在接触丰富段调用VLA，并将适应性写入语言记忆。应用到长程任务时，它在两点失效：(1) 能力来自测试时的全任务探索，其成本随阶段指数增长——若一个阶段需T轮，K阶段任务约需T^K，且失败不揭示哪个阶段导致；(2) 没有过渡表示：VLA原语只有出口条件没有入口条件，子任务可能以继任者无法使用的形式成功。我们提出BATON。针对(1)，BATON将子任务作为探索单元：每个在廉价的短程范围内探索，解存入记忆；长程轨迹由这些解组合而成，成本变为加法(T*K)，每次失败可归因于单阶段。针对(2)，BATON配备过渡感知记忆——验证智能体控制调用过渡，交接过渡恢复被前任扰动的入口状态，前瞻过渡选择继任者可继承结果的策略。不更新任何参数。在长程基准RoboMemArena上，BATON比SOTA提升任务成功率11.6%、累积成功率14.9%。

原文摘要

Long-horizon robot manipulation chains many contact-rich skills into one multi-stage task. Vision-language-action (VLA) models increasingly master the individual skills, yet the chain still fails: errors compound beyond the policy's ability to correct, and one subtask silently constrains the next. A promising recipe freezes the VLA and puts an LLM agent in charge: it plans in language, moves in free space with analytic primitives, invokes the VLA only for contact-rich segments, and writes adaptation into language memory. Applied to long horizons, it breaks twice. (1) Competence comes from whole-task exploration at test time, whose cost is multiplicative in stages: if one stage needs T episodes, a K-stage task needs about T^K, and a failure does not reveal which stage caused it. (2) It has ...

--- *自动采集于 2026-08-19*

#论文 #arXiv #Robotics #小凯