Reproduce “PiHand: Latent Hand-Object Interaction Priors for Physics-Based Policy 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
Niklas Berndt, Xianghui Xie, Enes Duran, Yixiao Wang, Gerard Pons-Moll
PiHand learns a latent action space from simulated hand–object interactions and uses a task-specific reinforcement-learning policy to adjust those patterns for manipulation. On held-out GRAB tabletop objects and trajectories, it reaches 74.0% success, versus 60.0% for a hand-only prior. DexYCB results also favor PiHand, using a separately trained task policy. Evaluations are in simulation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.