Reproduce “PDRVLA: A Vision-Language-Action Model with Physical Dynamics Representations” 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
Hanqing Wang, Zhaotian Cai, Ruichuan An, Bohan Zeng, Tao Zhang, Yilin Chen, Xu Jiang, Kaiquan Chen and 8 more
PDRVLA trains physical tokens to encode cues about future appearance, geometry, and motion from video features, then conditions action generation on them without running a video model during control. Authors report highest aggregate success among compared methods on four simulation benchmarks. On LIBERO-Plus, PDRVLA reaches 89.9% versus 89.0% for Qwen-RobotManip; this cross-paper comparison does not establish that world-model supervision improves action success.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.