Event

WorldArena 2.0 Expands Embodied World Model Benchmark

1 reports · 1 sources

Zhuohang Li and other authors released WorldArena 2.0, which expands the embodied world model evaluation benchmark along three dimensions: modality, functionality, and platform. According to the authors, in terms of modality, the evaluation extends from pure vision to visual-tactile, enabling assessment of multimodal perception and prediction; in terms of functionality, it extends from policy evaluation and planning to using world models as interactive reinforcement learning environments for policy optimization; and in terms of platform, it extends from simulation only to multiple simulations and real robot embodiments. The benchmark evaluates perception quality, interaction utility, and cross-platform performance under a unified protocol, and the project page provides code and demonstrations.