Reproduce “Recover, Don't Just Continue: Learning from Feedback for Physical Failure Understanding and Correction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Miao Pan, Kaixiang Yao, Xiaohe Zhou, Xu Wang, Qinfeng Li, Wenqi Zhang, Xuhong Zhang
Embodied-Corrector uses execution evidence to diagnose failed robot subtasks and revise instructions for a fixed vision-language-action model (VLA). It generates judgments, failure causes, recovery steps, and executable guidance, then breaks long goals into state-conditioned subtasks. On RoboCasa365 Composite-Unseen tasks, π0.5 success rose from 1.2% to 7.3%; Atomic success declined, and hardware testing covered five tasks on one platform.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.