Reproduce “PhySGA: A Physics-Space Grounded Agentic Framework for Generalizable and Interactive 3D Composition” 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
Weichen Lyu, Chong Xia, Mingyu Dong, Long Xu, Shengjun Zhang, Yueqi Duan
PhySGA composes pairs of 3D assets from language instructions by planning assembly, testing actions in Isaac Sim, and correcting placements using simulator feedback. Its benchmark contains 120 evaluation scenes, and a vision-language model scores multi-view results for relation fidelity. PhySGA reaches an average RFS of 8.75, versus 4.45 for Interact3D, on this simulator benchmark; its penetration rate is nonzero.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.