Reproduce “RACE: Calibrated Cumulative Scoring for Rollout-Free Robot Policy Checkpoint Selection” 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
Changyeon Kim, Yekyung Nah, Kimin Lee
RACE ranks robot-policy checkpoints without deploying each candidate, comparing predicted actions with successful references at similar states and downweighting later predictions after early low scores. In DexMimicGen, its held-out-demonstration Spearman correlation was 0.83, versus 0.52 for validation loss; on three real-robot tasks, it was 0.791 versus 0.689. These are rankings, not success probabilities, and shortlisted policies require deployment validation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.