Nvidia on Monday at the Consumer Electronics Show (CES) 2025 launched its new Cosmos platform, which marks a major step forward in developing artificial intelligence for physical systems. The platform offers advanced tools, such as generative world foundation models (WFMs), accelerated video processing pipelines, and essential guardrails, that can be used to create and train autonomous vehicles and robots.
Nvidia Cosmos: Key Features
Generative World Foundation Models (WFMs):
The core component of the Cosmos platform, the WFMs, can produce enormous amounts of photorealistic and physics-based synthetic data. This is highly important for the training of physical AI systems. The models discussed above overcome two of the significant challenges in the development of AI systems: the requirement for massive amounts of real-world data and the need for diverse testing environments.
Nvidia’s Cosmos platform will enable developers to build physical AI systems, including autonomous vehicles and robots, based on the synthetic data produced by the WFMs. Such systems, comprising mechanical parts and interacting with the real world, can be trained and tested on this synthetic data at a lower cost than in real-world settings.
Advanced Tokenisers and Video Processing:
This is equipped with high-end tokenisers and an accelerated video processing pipeline, providing all the tools to enhance training processes and improve AI models’ real-world applications.
Open-Source Access:
In a significant step, Nvidia made the WFMs open-source and available for the use of educational institutions and research organizations to avail the models for educational and development purposes. These models are now previewable and usable by developers through Nvidia’s API catalog or Hugging Face, which is the leading AI community platform.
Simulation-Based Training
Cosmos also interfaces with Nvidia’s Omniverse platform, allowing developers to create controlled 3D scenarios for simulation-based training of physical AI systems. This means that AI models will be tested in a more immersive and realistic environment before they are released into the real world.
Widespread Use in Robotics:
Several companies focused on robotics and physical AI have already embraced the Cosmos platform. Companies such as 1X, Agile Robots, Agility, Figure AI, Neura Robotics, Waabi, XPENG, Uber, and many others have already integrated Cosmos into their development processes.
Llama Nemotron AI Models:
Along with Cosmos, Nvidia launched the Llama Nemotron family of AI models, which expanded the range of AI tools available for developers and researchers in the field of robotics and autonomous systems.
The Cosmos AI platform by Nvidia, therefore, means a giant step forward in the development of AI, especially regarding autonomous robotics and physical AI systems. It holds the promise to revolutionize the training process since it will bring about cost efficiency and scalability into the training process. The platform is open-source, which provides access to its use by both the academic and research communities toward innovation and AI progress in respective industries.