Reproduce “ACE-CaP-0: Real-World Code-as-Policies in the Era of Experience” 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
ACE-CaP Team, Changxu Wei, Yunpeng Qing, Guozheng Ma, Bowen Liu, Zexin He, Yilun Kong, Haoming Gu and 20 more
ACE-CaP-0 develops task-specific robot programs by using demonstrations, simulation, and physical trials to revise candidate code. Its frozen programs use runtime perception but run without a human or coding agent choosing each action. Across eight tasks, they succeed in 221 of 240 real-world trials (92.1%); this is pooled performance across separate task programs, not one general-purpose policy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.