Reproduce “VLA Adaptation with Real and Simulated Data: A Case Study on Robotic Chess” 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
Matěj Kripner, Milan Straka
The authors build a MuJoCo simulator for training and closed-loop evaluation of a chess-piece-moving robot policy, then transfer the policy through real demonstrations. For π₀.₅, a learning-rate cool-down raises simulated success to 60%, from 29% at its starting checkpoint. With 128 real demonstrations, simulation initialization yields 18/25 successes versus 2/25, in a partial evaluation on one task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.