Reproduce “OGAM: Connecting Systematic Testing to Runtime Assurance through Object-Grounded Attention Monitoring for VLA 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
Haki Darwish, Xiangyu Yin, Changwen Li, Rongjie Yan, Francisco Gomes de Oliveira Neto, Chih-Hong Cheng
OGAM uses pairwise scene-derived tests and object-role attention to monitor robot policies. It aligns each attention course with a successful reference using dynamic time warping and stops execution when divergence persists. In LIBERO-Goal paraphrase tests, OpenVLA stopped 58 of 67 failures, with 3 false stops among 63 successes. These simulation results do not guarantee false-stop rates on new episodes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.