WORLD MODEL BENCHMARK

WorldArena 2.0

Extending Embodied World Model Benchmarking on Modality, Functionality and Platform

Modality Functionality Platform

WorldArena 2.0 Leaderboard evaluates embodied world models across simulator video quality, interactive RL environments, visuo-tactile manipulation, and real-robot action planning.

Track 1: Simulator video-quality evaluation using the WorldArena perceptual metrics, including the benchmark's difficulty and out-of-distribution corrections.

Leaderboard will appear here.

Track 2: Success rates of policies trained inside each world-model environment on Click Bell and Adjust Bottle. The paper's proxy-based reward results are used.

Leaderboard will appear here.

Track 3.1: Visuo-tactile evaluation on the UniVTAC simulator using Insert HDMI and Lift Bottle.

Leaderboard will appear here.

Track 3.2: Real-robot action-planning success rates on Wipe Table and Pour Water.

Leaderboard will appear here.