Reproduce “Beyond Policy Patches: Distilling RL Specialists into Vision-Language-Action Policies” 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
Zhennan Jiang, Tingying Xie, Shilong Ji, Shu'ang Yu, Tianxing Zhou, Yuhui Chen, Jiahuan Gao, Kaizhi Xie and 6 more
REPAIR transfers precision-phase RL expertise into a vision-language-action policy. It trains on specialist-assisted observations, then uses specialists’ critics to weight candidate action chunks on observations collected as the policy acts alone; specialists do not control these rollouts or deployment. Across six real-robot tasks, it matches RLT’s 30/30 success per task, but these finite evaluations are not a general guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.