Reproduce “PlayPolicy: Dexterity from Supervised Play” 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

Botao He, Yiming Dou, Yiannis Aloimonos, Jitendra Malik, Akash Sharma, Ruoshi Liu

DSP uses brief human teleoperation to teach a goal-free play policy, then lets it gather robot experience for an action-conditioned world model. At inference, model-predictive control selects motions predicted to approach requested cube orientations. In 128 simulated distant-goal trials, DSP succeeded within 0.6 radians on 91.4%, versus 53.9% for hindsight behavior cloning; these results are simulated, not settled hardware evidence.

Summary by alphaXiv.

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