Nvidia Enhances AI Training for Humanoid Robots

Nvidia, a leading provider of AI services and models, is paving the way for the next generation of humanoid robots in collaboration with major robot manufacturers. The company has introduced new AI microservices and frameworks specifically designed for humanoid robots as part of its NIMS offering, which was launched earlier this year. These cutting-edge technologies are powered by the Jetson Thor platform, running AI on a variant of the Blackwell GPU.

Robotics companies such as Boston Dynamics, ByteDance Research, Field AI, Figure, Fourier, Galbot, LimX Dynamics, Mentee, Neura Robotics, RobotEra, and Skild AI have all joined Nvidia's Humanoid Robot Developer Programme. This program offers early access to a range of innovative tools, including the Project Gr00t foundation models, to enhance the development of humanoid robots.

Nvidia has also developed an orchestration service called Osmo, which enables the seamless execution of multi-stage robotics workloads. Additionally, the company has introduced an AI- and simulation-enabled teleoperation workflow that allows developers to train robots using minimal amounts of human demonstration data.

The MimicGen NIM microservice is designed to generate synthetic motion data based on recorded teleoperated data, while the Robocasa NIM microservice creates robot tasks and simulation-ready environments. These tools, along with the Osmo service, significantly streamline robot training and simulation workflows, reducing deployment and development cycle times from months to under a week.

Training foundation models for humanoid robots typically requires vast amounts of data. Nvidia's innovative approach involves using teleoperation to capture human demonstration data efficiently. By leveraging simulation technology and generative AI developer tools, companies like Fourier are able to accelerate model development workflows and enhance the capabilities of their humanoid robots.