UBTECH says its self-developed unified multimodal world model, Thinker-WM, has taken first place across three embodied world-model benchmarks, a result the company frames as a step toward robots that can predict and plan, not just react.
On the WorldOlympiad benchmark the model scored 0.740 overall and led all participants on 3D consistency and interaction fidelity. On EWMBench it reached a composite of 4.66 and led on the core robot-motion metrics. On the WorldArena 1.0 Track-1 simulation benchmark it led across 12 key dimensions, including visual quality, content consistency, physical plausibility, 3D geometry precision and instruction following.
The architecture centres on a unified dynamic Transformer that encodes RGB images, metric depth and robot actions into a single intermediate representation for joint temporal modelling, rather than stitching modalities at the interface. That shared latent space lets the model switch between condition and generation modes to deliver six capabilities: joint video and action generation, world prediction, policy reasoning, video-to-action and inverse dynamics.
The unified representation rests on roughly 30,000 hours of heterogeneous training data, about 20,000 hours of first-person human video, 10,000 hours of real robot data and 1,000 hours of simulation. In a RoboTwin single-task ablation, adding the depth modality lifted the operation success rate by 15 percentage points over a no-depth baseline. A lightweight Motion Probe test confirmed that even without action supervision, motion-relevant information is already encoded in the visual representation. Thinker-WM has completed deployment validation on UBTECH’s robot platforms, and the company plans to publish the model technical report, later iterations and an open secondary-development plan for the world model.
Editor’s note: This is an adapted translation of the original Gasgoo report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.gasgoo.com/apps/e060bad9eca91db7f01403fc7120041f/robot/news/70475209.