Reproduce “Reference-Free Humanoid Skateboarding via Termination-Shaped 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
Ganeshvar Soenessardien, Michael Przystupa, Kevin Sebastian Luck
The authors train a humanoid to push a board in simulation with rewards and episode-ending constraints, without skateboarding demonstrations or posture targets. Initialized from a walking policy, it learns a cautious push-and-roll motion. Across three runs, mean board speed was 0.270 ± 0.087 m/s and XOR-stance rate was 0.976 ± 0.009; the result is flat-surface simulation, not hardware transfer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.