Reproduce “Make Code as Policy Great Again: Frontier Agents Write, Call, and Evolve Robot Tools” 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
Shijia Ge, Alex Zhou, Jianshu Zeng, Yexing Wan, Di Wu, Zelin Zheng, Yazhe Wang, Zhiqi Jia and 7 more
URAI lets a programming agent build timed tools that bundle multi-phase motions, while a frozen execution agent observes each result and chooses the next call. Across five RoboDojo tasks, four agents average 53.0% success through URAI versus 18.0% with native fingertip control. With identical tools, live decisions reach 56% versus 24% for written programs, but gains are task- and setup-dependent.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.