Nvidia announced the release of Isaac GR00T N1, an open and customizable AI model that will help create humanoid robots. “The era of general-purpose robotics has arrived. With Nvidia Isaac GR00T N1 and new data generation and robot training frameworks, roboticists around the world will open a new frontier in the AI era,” said Jensen Huang, the company’s CEO.
Image source: nvidia.com
During his keynote at GTC 2025, Mr. Huang demonstrated the 1X NEO Gamma humanoid robot, which can clean autonomously using the GR00T N1 with post-training. “The future of humanoids is adaptability and learning. The Nvidia GR00T N1 has provided a major breakthrough in robot reasoning and skills. With minimal training data, we were able to fully deploy it on NEO Gamma, advancing our mission to create robots that are not tools, but companions that can provide people with meaningful, immeasurable help,” said Bernt Børnich, CEO of 1X Technologies. The model was also tested at Boston Dynamics, Agility Robotics, Mentee Robotics, and Neura Robotics.
Nvidia first unveiled the work last year, when it was called Project GR00T (Generalist Robot 00 Technology). The updated GR00T N1 has a two-system architecture modeled after the way humans think. “System 1,” as Nvidia calls it, is a “fast-thinking action model” that’s similar to human reflexes and intuition. It was trained using data collected from human demonstrations and synthetic data modeled using Nvidia’s Omniverse platform. “System 2,” a visual-language, or “slow-thinking” model, “reasons about its environment and instructions to plan an action.” The plan is fed to System 1, which translates it into “precise, continuous robotic movements.” These may include grasping or moving objects with one or both hands, as well as more complex multi-step tasks that require a combination of basic skills.
GR00T N1 has been pre-trained and has mastered general-purpose reasoning and skills for humanoid robots, and developers are also provided with tools to further customize it, adjust its behavior for specific needs, and conduct additional training based on data collected from human demonstrations or simulations. Nvidia has made GR00T N1’s training data and task evaluation scripts available on Hugging Face and GitHub.