Reproduce “Towards Physically Grounded Articulated 3D Assets: A Multi-Agent Pipeline and Benchmark” 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
Xudong Cai, Ruijie Zhu, Haoliang Shang, Ruining Li, Marc Pollefeys, Shangzhe Wu, Andrea Vedaldi, Chuanxia Zheng
Artisan builds articulated 3D assets by turning prompts into designs and programs, then using separate review and simulator feedback to guide revisions. In a GPT-5.6-Sol ablation, removing physical validation raised sampled collision rates from 1.9% to 35.7% for single-joint motion and from 16.4% to 73.2% for multi-joint motion; this supports validation’s value in Artisan, not universal readiness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.