Nvidia announced today that it is releasing a family of foundational AI models called Cosmos that can be used to train humanoids, industrial robots and self-driving cars. While language models learn how to generate text by training on large amounts of books, articles and social media posts, Cosmos is designed to generate images and 3D models of the physical world.
During a keynote presentation at the annual CES conference in Las Vegas, Nvidia CEO Jensen Huang showed examples of Cosmos being used to simulate warehouse operations. Cosmos was trained on 20 million hours of real footage of “people walking, moving hands, manipulating things,” Jensen said. “It's not about generating creative content, it's about teaching the AI to understand the physical world.”
Researchers and startups hope that these types of fundamental models can provide robots used in factories and homes with more advanced capabilities. For example, Cosmos can generate realistic boxes of video footage falling from shelves in a warehouse, which can be used to train a robot to recognize accidents. Users can also refine the models using their own data.
A number of companies are already using Cosmos, Nvidia says, including humanoid robot startups Agility and Figure AI, as well as self-driving car companies such as Uber, Waabi and Wayve.
Nvidia also announced software designed to help different types of robots perform new tasks more efficiently. The new feature is part of Nvidia's existing Isaac robot simulation platform that allows robot builders to take a small number of examples of a desired task, such as grasping a particular object, and generate large amounts of synthetic training data.
Nvidia hopes Cosmos and Isaac will appeal to companies looking to build and use humanoid robots. Jensen was joined on stage at CES by life-size images of 14 different humanoid robots developed by companies including Tesla, Boston Dynamics, Agility and Figure.
Along with Cosmos, Nvidia also announced Project Digits, a $3,000 “personal AI supercomputer” that can run a large language model with up to 200 billion parameters without the need for cloud services from the likes of AWS or Microsoft. It also announced the highly anticipated next-generation RTX Blackwell GPUs and new software tools to help build AI agents.
