NVIDIA Unveils Isaac GR00T N1, First Open Humanoid Robot Model
NVIDIA has introduced a suite of cutting-edge technologies to push the boundaries of humanoid robot development, including the launch of NVIDIA Isaac GR00T N1, the world’s first open and fully customizable foundation model for generalized humanoid reasoning and skills. This move will drive advancements in industries grappling with significant labor shortages, with over 50 million vacancies worldwide.
Introducing GR00T N1
NVIDIA’s GR00T N1 is a versatile foundation model with a dual-system architecture inspired by human cognition. It is designed to facilitate robots' ability to perform fast reflexive actions and methodical decision-making processes. System 1 mirrors human reflexes and quickly translates plans into action, while System 2 reasons about tasks and forms action plans based on instructions and environmental understanding.
This foundational model allows robots to generalize across various tasks, such as grasping objects, handling items with one or both arms and performing complex multi-step operations. Applications for GR00T N1 span industries such as material handling, packaging, and inspection.
NVIDIA has begun pretraining and releasing GR00T N1 to robotics developers worldwide, marking a significant step in accelerating innovation within humanoid robotics. 1X Technologies, a prominent creator of humanoid space, demonstrated GR00T N1's ability at NVIDIA's keynote, where it showcased a humanoid robot performing household chores on its own, emphasizing the adaptability and learning ability of the model.
Collaboration with Industry Robotics Leaders
As part of its ongoing efforts to shape the future of humanoid robotics, NVIDIA has partnered with Google DeepMind and Disney Research to develop Newton, an open-source physics engine that enhances robots' ability to learn complex tasks with precision. The physics engine is designed to complement simulation frameworks like MuJoCo and Isaac Lab, helping optimize robotics learning.
The collaboration would also leverage Disney's physics engine to create more emotive, next-gen robots, such as Disney's Star Wars BDX droids, a reality. Walt Disney Imagineering Research & Development senior vice president Kyle Laughlin said that collaboration would enable Disney to create more expressive and compelling robotic characters and bring a new generation of robot entertainment into being.
Also read: Cisco and NVIDIA Partner to Accelerate AI Adoption in Enterprises
Speeding Up Robot Development with Synthetic Data
NVIDIA introduced the Isaac GR00T Blueprint to tackle the challenge of limited and costly real-world data in robotics. Using Omniverse and Cosmos Transfer models, it generates large amounts of synthetic motion data, enabling developers to train robots more efficiently. In just 11 hours, NVIDIA produced 780,000 synthetic motion trajectories, improving GR00T N1’s performance by 40% when combined with real-world data.
NVIDIA has released the GR00T N1 dataset and Isaac GR00T Blueprint on platforms like Hugging Face and GitHub, providing developers with essential training resources.
