NVIDIA Expands Cosmos Platform for Physical AI Development
According to NVIDIA. NVIDIA introduces new training tools and data factory blueprints to accelerate the creation and deployment of world models for physical AI systems.
- • NVIDIA expanded the Cosmos platform with new training tools and simulation checkpoints.
- • Cosmos Data Factory offers five cloud-native blueprints for synthetic data generation.
- • Updated Cosmos 3 NIM microservices are designed to provide up to 2x faster inference.
Platform Expansion
NVIDIA has expanded its Cosmos platform to provide developers with frontier open models, data tools, and training recipes. These resources are designed to help scale training data, adapt models, and optimize performance for physical AI systems that need to understand and simulate the real world.
Data Factory Blueprints
The Cosmos Data Factory now includes five cloud-native blueprints available through GitHub and major cloud providers. These tools facilitate autonomous vehicle dataset generation, image attribute augmentation, and the creation of synthetic video clips to train robots for industrial inspection and safety hazard detection.
Inference and Deployment
Developers can now utilize the Cosmos Framework to adapt open models through fine-tuning and reinforcement learning. Furthermore, the updated NVIDIA Cosmos 3 NIM microservices are stated to deliver up to 2x faster inference speeds, enabling more efficient video analysis and real-time safety monitoring.
- 1. blogs.nvidia.comOpen source