Reproduce “DexTact: A Visuo-Tactile Foundation Model for Dexterous Hand Grasping” 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
Xinmin Fang, Lingfeng Tao, Zhengxiong Li
DexTact trains a dexterous grasping policy that combines camera images, language, and simulated hand-contact forces. It pairs these inputs with grasp demonstrations and tests performance under changing lighting. In a matched 70K-demonstration ablation, success falls from seen to unseen lighting by 1.6 percentage points with touch, versus 16.6 points without it. The result is from simulation, not physical hand tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.