Reproduce “Verify, Act, Distill: Turning Test-Time Compute into 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
Kaneyoshi Hiratsuka, Satoshi Yamamori, Jun Morimoto
The authors introduce Verify, Act, Distill, which turns test-time candidate selection into training data for a vision-language-action policy. It scores and combines eight action chunks, executes the beginning of the aggregate, and uses successful episodes to update the policy. On simulated LIBERO, verifier-free success rises from 54.1% to 70.6%; aggregation after distillation reaches 82.3% versus 76.4% initially. Results depend on successful adaptation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.