Reproduce “Learning On-Policy Learning to Specialize 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
Chunwei Xing, Daniel Gehrig, Vineet Pasumarti, Orestis Konstantaropoulos, Antonio Loquercio
GOL (Generative On-Policy Learner) adapts robot controllers by generating successor policy weights from the current weights and one target-domain state–action trajectory, without target-domain rewards or ground-truth domain parameters. In simulation, it reaches mean episode reward 91.9 ± 1.7 on out-of-distribution drone racing, versus 74.0 ± 31.7 for DR-H. On one Crazyflie track, three policies recover after two steps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.