Nvidia has unveiled a new suite of foundational AI models called Cosmos, aimed at enhancing the capabilities of humanoid robots, industrial machines, and self-driving cars. During a keynote at CES in Las Vegas, CEO Jensen Huang highlighted that these models were trained using 20 million hours of footage showcasing human movements and interactions, emphasizing their focus on understanding the physical world rather than generating creative content.
Cosmos is engineered to create detailed images and 3D models that can simulate real-world activities. For instance, it can replicate scenarios in warehouses, generating footage of items falling from shelves to help robots learn to identify and respond to accidents. This innovative approach allows for fine-tuning the AI using specific user data, making it adaptable to various environments and tasks.
Many companies, including startups like Agility and Figure AI for humanoid robots, as well as self-driving car initiatives from Uber, Waabi, and Wayve, are already integrating Cosmos into their systems.
Additionally, Nvidia is enhancing its existing Isaac robot simulation platform. This updated tool is designed to enable robotic developers to generate large sets of synthetic training data from a few examples of specific tasks, such as object handling.
At the same event, Nvidia introduced Project Digits—a $3,000 “personal AI supercomputer” capable of running extensive language models independent of external cloud services. They also announced upcoming next-generation RTX Blackwell GPUs and software tools to facilitate the creation of AI agents.
These advancements underscore Nvidia’s commitment to advancing robotic intelligence and provide crucial tools for companies entering the field of humanoid robotics and automation.
