Bagel Labs launches WorldDiT world model for robotics
Bagel Labs is pitching WorldDiT on efficiency: one sub-billion-parameter diffusion model for both robot action and future-scene prediction.
The model uses shared parameters to sample actions and forecast how the robot’s view will change, instead of separating policy and world prediction. Bagel Labs says that setup produced strong LIBERO simulation results relative to parameter count. The release includes a technical report and open model weights on Hugging Face for robotics engineers and researchers. TestingCatalog's note
The model uses shared parameters to sample actions and forecast how the robot’s view will change, instead of separating policy and world prediction. Bagel Labs says that setup produced strong LIBERO simulation results relative to parameter count. The release includes a technical report and open model weights on Hugging Face for robotics engineers and researchers. TestingCatalog's note
score 6