Reproduce “Koopman Model-Based Reinforcement 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
Wenjian Hao, Paulo C. Heredia, Shaoshuai Mou
K-MBRL learns nonlinear robot dynamics in a representation with linear predictions. It combines one-step model lookahead with a critic estimating future cost, avoiding repeated model rollouts during policy improvement. The authors demonstrate outdoor bipedal walking on a Unitree Go1, with straight walking reaching about 5 m before a 140°F motor-temperature safeguard ended the trial.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.