Nvidia Ships the Foundation Model Physical AI Has Been Waiting For
SMRTR summary
A robot that learns to navigate fog, avoid pedestrians, and handle unexpected road conditions, without ever leaving a simulation. That's the promise behind Nvidia's newly released Cosmos 3, a world foundation model designed to train physical AI systems like robots, autonomous vehicles, and industrial machines.
Unlike large language models that learn from text, Cosmos 3 was trained on 20 trillion tokens of multimodal data, including nearly a billion images and 400 million videos. Crucially, it also incorporates action data, teaching machines not just what the world looks like, but what to do within it.
Nvidia CEO Jensen Huang declared that "the big bang of physical AI is just around the corner."
The real-world adoption is already underway. Mercedes-Benz launched a robotaxi service on Uber's network using Nvidia's platform. Samsung, LG, and Li Auto are building on it too.
Competitors are closing in, but Nvidia's bet is clear: the next AI frontier isn't language. It's physics.
SMRTR provides this summary for quick context. The original article belongs to PYMNTS.
Read the original article