量子位官网T2·78 pts
- Thinker-WM achieved first place in overall scores across three major embodied AI benchmarks: WorldOlympiad, EWMBench, and WorldArena1.0.
- The core innovation is the UDT architecture, which jointly evolves RGB images, metric depth, and robot actions within a single representation space rather than using stitched independent modules.
- A single model supports six capabilities including joint video-action generation, world prediction, policy reasoning, and inverse dynamics, enhancing generalization across different robot morphologies.