Reproduce “RoboBuilder: Agentic Robot Co-Design with Real-world Video Feedback in the Loop” 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

Ghadi Nehme, Dahyun Nam, Akshay Govind Srinivasan, Faez Ahmed

RoboBuilder tests coding agents designing robots beyond simulation with stocked parts, timed motor programs, and optional printed parts. It covers ten obstacles at three difficulty levels and uses real-failure videos to guide redesign. Agents passed 39 of 40 level-one tasks; one feedback round raised real crossing rates from 73% to 93% on level-one and level-two courses, not unseen tasks.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.