Reproduce “WorldArena 2.0: Extending Embodied World Model Benchmarking on Modality, Functionality and Platform” 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

Zhuohang Li, Yiding Ma, Yu Shang, Lei Jin, Shengyuan Wang, Jianjie Fang, Zhaolu Wang, Xvyuan Liu and 21 more

WorldArena 2.0 extends embodied-world-model evaluation by adding tactile prediction, policy training inside learned environments, and physical-robot tests alongside simulation. Its results show that perception and control can diverge: Wan 2.2 achieved 100% success on Insert HDMI but 0% on Lift Bottle in ten simulation trials per task. These task-specific findings are diagnostic, not evidence of general competence or deployment readiness.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.