NVIDIA has unveiled its new Cosmos World Foundation Models (WFMs) and Physical AI Data Tools, designed to enhance prediction, controllable world generation, and reasoning capabilities for physical AI applications.
The release includes two blueprints powered by the NVIDIA Omniverse and Cosmos platforms, enabling large-scale synthetic data generation for post-training robots and autonomous vehicles. Early adopters such as 1X, Agility Robotics, Figure AI, Skild AI, Foretellix, and Uber are leveraging these tools to accelerate the development of physical AI systems. The announcement underscores NVIDIA’s commitment to advancing physical AI through customizable reasoning models and synthetic data solutions, marking a significant step in the evolution of robotics and autonomous technologies.
NVIDIA Announces Major Release of Cosmos World Foundation Models and Physical AI Data Tools
NVIDIA has introduced a significant advancement in AI development with their new Cosmos World Foundation Models (WFMs), designed to enhance physical AI systems. These models enable developers to generate synthetic data efficiently, which is crucial for improving robots and autonomous vehicles. The release includes two blueprints powered by Omniverse and Cosmos platforms, facilitating large-scale synthetic data generation.
The announcement highlights three key components: Cosmos Transfer, which produces photorealistic videos from structured inputs; Cosmos Predict, capable of multi-frame generation to predict intermediate actions; and Cosmos Reason, an open model using chain-of-thought reasoning to understand video data and predict outcomes. These tools are available through various platforms, including NVIDIA’s API catalogue, Vertex AI Model Garden, Hugging Face, and GitHub.
Integration with tools like NeMo Curator enhances data processing capabilities, while collaborations for content transparency ensure responsible AI use. This release marks a notable step in advancing physical AI applications across industries.
NVIDIA’s Cosmos World Foundation Models (WFMs) are designed to advance synthetic data generation for physical AI systems. The release includes three core components: Cosmos Transfer, which generates photorealistic videos from structured inputs like depth maps or segmentation masks; Cosmos Predict, capable of multi-frame prediction to simulate intermediate actions in sequences; and Cosmos Reason, an open model that employs chain-of-thought reasoning to interpret video data and predict outcomes.
These models are integrated with NVIDIA’s ecosystem, including tools like NeMo Curator for enhanced data processing and curation. Availability spans platforms such as the NVIDIA API catalogue, Vertex AI Model Garden, Hugging Face, and GitHub, ensuring broad access for developers. The integration of SynthID watermarking technology further supports responsible AI use by identifying synthetic content.
The release has garnered interest from various industries, with early adopters leveraging the models for applications ranging from autonomous vehicle training to robotics simulation. Companies use the tools to accelerate development cycles and improve system reliability through high-fidelity synthetic data generation.
The models leverage NVIDIA DGX Cloud for post-training optimization, enabling developers to refine and deploy models efficiently using native PyTorch or TensorFlow workflows.
NVIDIA has integrated SynthID watermarking technology into its ecosystem to support responsible AI use and content transparency. This collaboration with Google DeepMind enables the identification of AI-generated outputs, ensuring accountability in their application across industries. The integration of SynthID into the NVIDIA NIM microservice further enhances this capability, providing a robust framework for tracking and verifying synthetic data.
By prioritizing transparency and accountability, NVIDIA is helping to establish trustworthy principles for the responsible use of AI technologies.
More information
External Link: Click Here For More
