Reproduce “AdaRSI: On-Demand Recursive Self-Improvement for Embodied Agents” 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
Peidong Wang, Ying Chang, Zhiming Ma, Xiaocui Yang, Shi Feng, Daqi Liu, Yuzhe Yang, Xu Cao and 3 more
AdaRSI improves robot task performance by revising how an executor is called before fine-tuning motions that remain unreliable. It retries failed sub-steps from saved simulator states under alternative instructions and approaches, then adopts policies only when development tests show benefit. In simulation, it raises success on 16 Composite-Unseen RoboCasa365 tasks from 4.6% to 13.8%, though some tasks regress.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.