ARC：人类引导的机器人策略自适应残差修正方法
When, Where, and What to Correct: Human-Guided Adaptive Residual Correction for Robot Policies
AI 导读
研究者提出自适应残差修正方法 ARC，一个轻量级后训练模块，用强化学习在冻结策略之上学习残差修正，并以稀疏人类修正确定“何时”“何地”进行自主探索。在真实机器人上的三项精密操作任务中，ARC 将平均任务成功率从 33.3% 提升至 75.0%；人类修正定义的 2-3 维子空间保留了七关节残差空间中 96.0-97.7% 的修正幅度平方值。
正文 · 原文
Abstract:Imitation policies that reliably execute most of a manipulation trajectory can still fail in localized regions requiring last-mile precision. Updating the full policy or exploring residual actions everywhere treats this local failure as a global learning problem. We introduce Adaptive Residual Correction (ARC), a lightweight post-training module that uses reinforcement learning (RL) to learn a residual correction on top of a frozen policy while using sparse human corrections to structure autonomous exploration. Rather than treating human corrections as imitation targets, ARC uses pre-intervention states to determine 'when' residual control should become active and correction geometry to determine 'where' in residual action space autonomous refinement should explore; task reward determines 'what' residual is ultimately executed. Across three precision-manipulation tasks on a real robot, ARC improves average task success from 33.3% to 75.0%. The human corrections define task-dependent 2-3 dimensional subspaces of the seven-joint residual space that retain 96.0-97.7% of squared correction magnitude. Together, these results show that sparse human corrections can localize and constrain autonomous post-training without requiring the observed correction itself to serve as the final action target.
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.37547 [cs.RO] |
| (or arXiv:2610.37547v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.37547 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Anukriti Singh [view email]
[v1]
Tue, 27 Oct 2026 16:58:54 UTC (963 KB)
来源：arXiv cs.AI · arxiv.org