Reproduce “Dex-One2Many: Learning Dexterous Manipulation from a Single Human Demonstration” 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
Jusuk Lee, Sungha Kim, Yeonsoo Park, Jonguk Cheon, Yoonkyo Jung, Yongjun You, H. Jin Kim, Jia-Bin Huang and 3 more
Dex-One2Many learns dexterous manipulation from one human video by representing ordered hand-object relations as scene graphs, then using varied task-consistent simulation resets and predicate-based rewards for reinforcement learning. On five tasks, it achieves a 75% mean full-task success rate on unseen initial and goal poses, versus 4% for the best-baseline average, in zero-shot tests on one robot setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.