Reproduce “Rollout then Recycle: 3,718 Labelled Real-Robot Episodes from Benchmarking 7 Policies on 12 Tasks” 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
Praveen Selvaraj, Lorenzo Uttini, Ville Kuosmanen
The authors document and release ArmnetBench's complete collection of real-robot evaluation episodes as reusable data with operator-assigned outcome labels. The collection contains 600 demonstrations and 3,118 policy rollouts, including 2,288 labeled failures, or 73.4% of rollouts. Episodes include observations and actions. The release makes attempts available for reuse; it does not demonstrate downstream learning improvements or enable exact leaderboard reproduction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.