Reproduce “RoboARCade: Benchmarking Agentic Robot Control” 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
Satvik Sharma, Hengyuan Hu, Mihika Kamat, Dorsa Sadigh, Jeannette Bohg
RoboARCade benchmarks agentic robot control across 22 simulated tasks with varied starting conditions. In this setting, agents observe scenes and choose controller targets directly in a closed loop. Results show strategy and precise execution are larger failure sources than visual understanding, while success claims are often incorrect. On a five-task subset, experience raises mean rubric score from 53.5% to 73%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.