Reproduce “OopsieRecover: Learning Transferable VLA Corrections from Semantic Interventions” 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
Arpit Bahety, Skand Peri, Sriniket Ambatipudi, Chen Xu, Haruki Nishimura, Masha Itkina, Roberto Martín-Martín
OopsieRecover infers reusable spatial correction rules from human interventions and uses them to steer a frozen vision-language-action policy. It grounds relevant objects, verifies gated costs against correction evidence, and ranks predicted action chunks at deployment. In tested real-world tasks, task progress rose from 35.1% to 61.0%, a 74% relative improvement over the next-best baseline; progress measures completed stages.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.