Reproduce “Guided Action Flow: Value-Guided Sampling for Frozen Vision-Language-Action Policies” 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
Liuhaichen Yang, Zhuang Jiang, Chenchao Sheng, Ningwei Bai, Qichen Yin, Hanbo Ma, Junkai Liu, Junkai Sun and 3 more
GAF improves a frozen vision-language-action policy on physical robot manipulation tasks by learning an observation-conditioned critic from robot outcomes and using its gradient to adjust action chunks during iterative generation, without changing the policy’s weights. Across six nominal tabletop tasks, success rises from 60.0% to 82.5%. Tests under six lighting and distractor conditions cover only three tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.