Nvidia pushes physical AI with new models, robotaxi plans and data tools

At GTC, Jensen Huang closed his keynote with a walking, talking robot version of Olaf from Frozen to illustrate where Nvidia sees physical AI heading: robots that learn in simulation and act in the real world. The demo ran on Nvidia’s Jetson platform and used the Omniverse simulator; the interaction was imperfect, but it underscored the company’s aim to put more lifelike robotics into places like theme parks.

Nvidia unveiled several foundation models for machines that must operate in physical environments. Cosmos 3 generates synthetic worlds for training, Isaac GR00T N1.7 is an open reasoning vision-language-action model aimed at humanoid robots, and Alpamayo 1.5 is pitched as a major upgrade for autonomous vehicles, turning driving video, ego-motion history, navigation guidance and natural-language prompts into driving trajectories that help developers set safety guardrails.

nvidia, gtc, jensen huang, jetson, omniverse, cosmos 3, isaac gr00t, alpamayo 1.5, robotaxi, autonomous vehicles