Reproduce “LoRN: Low-Rank Noise Steering for Generative 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
Wei Liu, Abrar Anwar, Huihua Zhao, Soha Pouya, Yan Chang
LoRN steers frozen generative robot policies by learning a compact controller that selects coefficients for a fixed random projection of initial noise. For world-action models, video-noise steering can influence generated actions more than action-noise steering. Across ten model-task pairs, mean success was 69.2% with LoRN video steering, versus 43.0% for base policies. It cannot teach behaviors absent from the frozen policy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.