(Credit: Nvidia)
Nvidia's new computer model is designed to pave the way for AI-powered humanoid robots.
At the company’s GTC event in San Jose, Nvidia announced “Project GR00T,” a software platform that can enable a robot to understand human language and replicate human actions.
The computer model promises to help today’s robot makers build machines capable of interacting with humans and navigating real-world environments. “The ChatGPT moment for robots might be around the corner,” says Nvidia CEO Jensen Huang.
(Credit: Nvidia)GR00T stands for Generalist Robot 00 Technology, and it’s designed to act as a “foundational model” that companies can build on, bypassing the need to merely rely on pre-programmed commands for their bots.
GR00T is able to understand both human language and videos as input. This means you can essentially feed the computer model video clips depicting a human movement. The model will then train itself to replicate the action in a virtual simulation, gaining the dexterity and coordination to pull off and imitate the movement in the real world.
(Credit: Nvidia)The software model arrives as OpenAI —a major buyer of Nvidia GPUs— has also been using its own AI models to power a humanoid robot from a startup called Figure. In a demo last week, Figure showed how its robot could have full conversations with a human, thanks to the OpenAI integration.
In Nvidia’s case, the company has also developed “Jetson Thor,” a system-on-a-chip that can run GR00T and other computer models meant for humanoid robots. Jetson Thor will feature a GPU based on Nvidia’s Blackwell architecture, which represents a major upgrade from the company’s existing Hopper H100 GPUs.
Nvidia adds that it’s building AI platforms for leading robotics companies including Boston Dynamics, Figure, and Agility Robotics.