Reproduce “Res-HIL: Human-Guided Residual Reinforcement Learning for Sample-Efficient Dexterous Manipulation” 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
Mariia Iavorskaia, Christian Dietz, Sebastian Albrecht, Majid Khadiv
Res-HIL improves robot manipulation from demonstrations by freezing a base policy and learning an additive residual online. Human takeovers provide corrections and penalize preceding autonomous transitions; behavior cloning trains the residual. On a two-camera cable task, it reached 88% success after 45 minutes, versus 46% for its 20-demo base and 80% for frozen 100-demo ACT, in five tasks on one UR5 setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.