Reproduce “Coherent Soft Imitation Learning” 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

Joe Watson, Sandy H. Huang, Nicolas Heess

CSIL combines behavioral cloning with reinforcement learning by deriving a shaped reward from the cloned policy and its action prior, then refining the clone through soft policy iteration. Its policy is designed to behave like the prior away from demonstrations. In online Adroit manipulation, CSIL matches or surpasses behavioral cloning; image-based success is limited to simpler tasks, with poor cloning constraining NutAssemblySquare.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.