Reproduce “Reinforcement Learning with Supervision from Noisy Demonstrations” 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
Kun-Peng Ning, Sheng-Jun Huang
LfND combines demonstrations with environment feedback, weighting each demonstrated step according to its estimated usefulness to the current policy. The authors evaluate it in six MuJoCo control tasks using TRPO and PPO. In TRPO on Hopper, LfND reports average reward of 2655.84 ± 20.91, above its unweighted version and TRPO alone, but it does not lead in every task comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.