Jim Fan, co-lead of Nvidia's Project GR00T, confirmed that recent footage of humanoid robots is unaccelerated and operates autonomously without human intervention. The demonstration will be presented at the Conference on Robot Learning 2026, scheduled for November 9, 2026, in Austin, Texas.
Project GR00T focuses on developing foundation models for humanoid robots that integrate vision, language, and action. Nvidia introduced the open-weight GR00T N1 model in March 2025, featuring 2 billion parameters. Subsequent versions, N1.5 and N1.6, incorporated integration with the Cosmos Reason system.
To address the scarcity of physical training data, Nvidia utilizes simulation platforms including Isaac Lab and Cosmos. The company's methodology leverages generative world models to expand limited teleoperation data by more than 1,000 times. This approach aims to provide developers with tools for creating versatile robots.
Nvidia's robotics strategy involves supplying foundation models, simulation environments, and physics engines like Newton to external developers. The project, originally unveiled by CEO Jensen Huang at GTC 2024, aims to enable robots to perform multiple tasks across diverse scenarios rather than specialized functions.
