Reproduce “RePlan-VLA: Progress-Aware Replanning for Long-Horizon 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

Apoorva Vashisth, Manav Kulshrestha, Damon Conover, Zachary Kingston, Aniket Bera

RePlan-VLA checks whether each step of a multi-step robot task succeeded and revises the remaining plan when execution stalls or diverges, while a frozen vision-language-action policy performs the steps. In a real-world example, it retries a missed block placement. On RoboCasa365, it averages 62.5% task success, versus 40.6% for Awomo-0.5I; recovery still depends on visual grounding and manipulation skill.

Summary by alphaXiv.

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