Reproduce “Learning Adaptable Dexterous Skills via Hybrid Dexterous Mimicking” 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
Yuxuan Kuang, Jiashun Wang, Yixuan Sun, Yufei Ye, Sungjae Park, Umar Iqbal, Katerina Fragkiadaki, Shubham Tulsiani
AdaDex trains a goal-conditioned policy with human-motion tracking and perturbed-goal practice. Tracking environments provide dense rewards for human-derived hand behavior, while generalization environments use sparse object-success rewards without finger-matching terms. On unseen motions, it reaches 80.02% tracking progress, versus 31.75% without generalization environments; on trained motions, that ablation performs slightly better, limiting the result to generalization beyond training clips.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.