Reproduce “WAM-Policy: World Action Models with On-Policy Distillation for Quadrupedal Loco-Manipulation” 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
Junlin Long, Xu Deng, Yiran Wang, Yue Yang, Yang Zhao, Zeyu Zhang
The authors adapt an arm-trained video-and-action model to a quadruped’s cameras and leg commands, then distill it into a faster language-conditioned policy. Reinforcement learning improves box-pushing and door-opening before a second distillation stage. In QLM-Bench simulation, the M-OPD student reaches 85.0% success across ten tasks and samples a chunk in 0.159 seconds; this does not establish real-time control or generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.