Reproduce “TipDex-VTLA: Learning When to Refine Actions with Touch 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

Xudi Ge, Yepeng Liu, Haifeng Liu, Enze Shi, Liangqi Zhang, Yongchao Xu

TipDex-VTLA uses visual action-consistency estimates and time-aligned tactile forecast discrepancies to regulate bounded corrections to unexecuted actions under illumination changes. Its visual score is a policy-dependent proxy, not a measure of action correctness. Across four real-world tasks, it achieved 73.8% average full-task success under tested unseen lighting, versus 31.3% for N⁰-VTLA; this does not establish robustness to arbitrary visual shifts.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.