Reproduce “GPT-6-Astra Understands Touch and Outperform Vision--Tactile--Language--Action Models” 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
Ziyong Wang, Yupei Lin, Runxuan Huang, Liang Lin, Dongyu Zhang, Guangrun Wang
The authors use GPT-6-Astra to guide robot manipulation with expert-derived guidance and tactile references, without updating policy parameters. It uses vision, touch, state, and interaction history to decide whether to continue, probe, or correct. Across five UniVTAC tasks, mean success is 93.8%; on Insert Hole, the touch-enabled setup reaches 99%, versus 31% without touch and tactile references, with guidance matched to each setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.