Reproduce “Code2Worlds-2: Physically Plausible Agent Harness for Embodied World Simulation” 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
Haoyu Zhang, Siming Li, Zeyu Zhang, Yi Zhang, Tat-Seng Chua, Hao Tang
Code2Worlds-2 adds a physics-engine layer to a coding-LLM world generator, compiling generated scenes into simulator-ready objects and constraints and using rollouts to revise them. In fixed-source tests, physical acceptance rises from 75/180 to 108/180, while robust task success rises from 45/180 to 60/180. The clustered 95% interval spans −11.1 to 28.9 percentage points; task success falls after enrichment alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.