Sanja Fidler#525
Following recent World-Action Model results in robotics, the same ~2B OmniDreams single-view backbone can be fine-tuned into a driving policy.
In preliminary closed-loop results, it reduces collision from 6.9% to 4.2% when compared with Alpamayo 1.5, while having roughly 5x fewer parameters.
Sanja Fidler#525Following recent World-Action Model results in robotics, the same ~2B OmniDreams single-view backbone can be fine-tuned into a driving policy.
In preliminary closed-loop results, it reduces collision from 6.9% to 4.2% when compared with Alpamayo 1.5, while having roughly 5x fewer parameters.
Users expressed gratitude for the OmniDreams 2B model's 39% collision reduction with fewer parameters and the open-sourcing of the FlashDreams inference stack.
We are also open-sourcing FlashDreams, our plug-and-play inference and serving stack for autoregressive world/video models: https://github.com/NVIDIA/flashdreams/
Many thanks to the NVIDIA SIL team and collaborators who made OmniDreams possible
#CVPR2026 #WorldModels #AutonomousVehicles