Reproduce “WAM-OPD: Joint Video-Action Supervision for World Action Model Post-Training with On-Policy Distillation” 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
Liuhaichen Yang, Zhengyang Zhong, Hanshang Zhu, Ningwei Bai, Qichen Yin, Zhi Han, Jiarui Qin, Zhuang Jiang and 8 more
The authors introduce WAM-OPD, which post-trains a video-and-action robot policy using Teacher-provided video and action targets on histories generated by the Student. The method retains video-conditioned action generation at deployment. In tests, equally weighted average success rose from 33.8% to 65.7% across 12 simulation tasks and from 51.4% to 64.6% across four physical tasks, with unchanged sampling steps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.