Reproduce “Evo-DexBench: Benchmarking Generalization across Related Tasks for Dexterous Manipulation” 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
Bing Cheng, Yiran Mao, Zewei Ye, Yifan Jin, Xinyi Chen, Ruifeng Tu, Shuhuai Wang, Tao Lin and 5 more
Evo-DexBench tests whether dexterous robot policies reuse interaction skills on new tasks with few demonstrations. Its Core-Gen design pairs source tasks sharing a functional interaction with a sparsely demonstrated target. On DishwasherKnob, π₀.₅ reaches 46% success with joint Core and Gen training, versus 8% with sequential training. This result covers two targets under one adaptation protocol; evaluations are simulated.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.