NVIDIA Expands Cosmos Platform for Physical AI World Models
NVIDIA announced updates to its Cosmos platform at GTC Berlin, adding Cosmos Data Factory blueprints, training tools, and Cosmos 3 Nano variants to support building and deploying world models for physical AI.
At GTC Berlin, NVIDIA expanded its Cosmos platform to assist physical AI development. The update introduces Cosmos Data Factory blueprints for synthetic data generation and curation, tools to adapt open models such as GR00T and QWEN, and Cosmos 3 Nano model variants for simulation. NVIDIA stated the tools support robotics and autonomous vehicles, with Cosmos 3 NIM microservices offering up to 2x faster inference in validated configurations.
Key takeaways
- Cosmos Data Factory introduces five cloud blueprints across GitHub, Azure, Nebius, and CoreWeave for generating synthetic data in autonomous driving and defect inspection.
- Cosmos 3 Nano variants and updated NIM microservices allow interactive policy testing in simulation and continuous video analysis with sub-second alerts, NVIDIA reported.
Source：NVIDIA Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.