Reproduce “Asynchronous Robot 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
Sombit Dey, Luc Van Gool, Danda Pani Paudel
ARP separates semantic updates from fast visual-action control: a large vision-language model refreshes compact memory while a smaller expert acts from current camera input and available context. Training exposes the expert to stale memory. In simulation, it succeeds in 51.5% of LIBERO-Perturb episodes, versus 9.5% for π₀-1B; this comparison does not isolate asynchrony from direct camera input.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.