Reproduce “Cracking Mobile Manipulation: From Failure Diagnosis to Agentic Robot System” 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
Zixuan Wang, Jia-Ning Li, Yinzhou Tang, Yanzhao Yu, Biao Wu, Meng Fang, Xin Li, Yiwu Zhong
The system targets failures in long-horizon mobile manipulation while keeping its subtask vision-language-action policy fixed. It adds demonstration guidance, visual memory, base-pose adjustment, and local end-effector correction based on offline failure diagnoses. On 20 RoboCasa365 tasks, success rises from 18% to 49% on held-out episodes. The separate 80% atomic-task result is not a controlled comparison with mobile-task success.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.