Nvidia announced today that it will release a family of basic 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 vast amounts of books, articles, and social media posts, Cosmos is designed to generate images and 3D models of the physical world. During his keynote presentation at the annual CES conference in Las Vegas, Nvidia CEO Jensen Huang showed an example of Cosmos being used to simulate activities in a warehouse. Cosmos was trained on 20 million hours of real footage of “humans walking, hands moving, manipulating things,” Jensen said. “It’s not about producing creative content, it’s about teaching AI to understand the physical world.” Researchers and startups hope that this basic model can endow robots used in factories and homes with more sophisticated capabilities. For example, Cosmos can produce realistic video footage of boxes falling from shelves in a warehouse, which can be used to train robots to recognize accidents. Users can also adjust the model using their own data. Several companies are already using Cosmos, Nvidia said, including humanoid robot startups Agility and Figure AI as well as self-driving car companies like Uber, Waabi, and Wayve. warehouse records created by Cosmos. Courtesy of NvidiaNvidia also announced software designed to help different types of robots learn to perform new tasks more efficiently. The new feature is part of Nvidia’s existing Isaac robot simulation platform that will allow robot makers to take small samples of desired tasks, such as grasping specific objects, and generate large amounts of synthetic training data. Nvidia is expecting Cosmos and Isaac. will appeal to companies looking to build and use humanoid robots. Jensen joined us on stage at CES with life-size drawings 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 large language models of up to 200 billion parameters without the need for cloud services from the likes of AWS or Microsoft. It also announced Blackwell’s next-generation RTX GPUs, and upcoming software to help build AI agents.
