World model simulator trained on 30,110 hours of robot demonstrations operates at 10.1 FPS
Researchers trained a world model simulator on 30,110 hours of robot demonstrations across 22 platforms.
The simulator operates at 10.1 FPS and maintains stability during long-horizon rollouts involving rigid and deformable manipulation.
The model adapts to unseen platforms using 100 to 200 gradient steps in cases where the pretrained backbone cannot.
Against five open-source action-conditioned world models, the simulator leads four of the five WorldSimProbe suites and achieves the best overall EWMBench score.
Sources behind this report
- Hugging Face
Research · abstract · CC0 1.0
Source published
Read by the newsroom
Report history
No earlier revision dates are retained here. This does not establish that the report has never changed; read any dated updates and corrections.
This brief was made by AI and checked against its cited public sources. How the newsroom works