Reproduce “Risk-Controlled Test-Time Adaptation for Vision-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
Zixuan Liu, Hashem Haghbayan, Zizhan Zheng, Siavash Khajavi
R2-VLA adds an action-uncertainty predictor to a retrieval-augmented vision-language-action policy and asks an expert to act when predicted action error exceeds a threshold calibrated on complete-task outcomes. As retrieval memory changes, the threshold is recalibrated. In simulated LIBERO, π₀-FAST averages 72.0% success versus RICL’s 62.8%, but misses its 80% OBJECT target.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.