Reproduce “Opt2VLA: Force-Aware Vision-Language-Action for Contact-Rich Humanoid Whole-Body 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
Fukang Liu, Yipu Chen, Jaehwi Jang, Danfei Xu, Zsolt Kira, Ye Zhao
Opt2VLA gives a vision-language-action policy explicit desired contact-force targets alongside motion targets, then uses a whole-body controller to realize them. Whole-body trajectory optimization supplies motion and torque references for training. In simulation, torque-supervised force control reduced shelf-pushing error from 11.2 ± 4.8 N to 4.2 ± 5.5 N and achieved 82.2% joint success; hardware showed force modulation, with larger wiping error.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.