Reproduce “SentryVLA: Uncertainty-Guided Action Selection and State Monitoring for Vision-Language-Action Robotic Control” 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
Zongyao Lyu, Lingling Chen, William J. Beksi
SentryVLA wraps a frozen vision-language-action policy with candidate-action ranking and latent-state monitoring. CAS scores sampled actions by predicted deviation from expert references and selects the lowest-scored candidate, while SMD flags contexts distant from nominal latent states. On LIBERO-Object, CAS raised average success from 0.56 to 0.73; this evidence and controlled hardware tests do not establish general safety.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.