NVIDIA Introduces GR00T N1: First Open Humanoid Robot Foundation Model

NVIDIA announced several new technologies aimed at advancing humanoid robot development. The announcement, made during the company's annual GPU Technology Conference (GTC), includes NVIDIA Isaac GR00T N1, described as the world's first open, fully customizable foundation model designed for humanoid robots.

GR00T N1 Foundation Model

GR00T N1, now available to developers, is the first in a planned family of customizable models that NVIDIA intends to pretrain and release for robotics developers worldwide. The model features what NVIDIA calls a dual-system architecture inspired by human cognition:

1. "System 1" - A fast-thinking action model that mirrors human reflexes or intuition
2. "System 2" - A slow-thinking model designed for deliberate, methodical decision-making

NVIDIA explains in their press release that the foundation model is powered by a vision language model that enables robots to reason about their environment and the instructions they receive. This reasoning capability is then translated into precise, continuous robot movements. According to NVIDIA, the model was trained using human demonstration data and synthetic data generated through the NVIDIA Omniverse platform.

GR00T N1 can generalize across common tasks such as grasping, moving objects with one or both arms, and transferring items between hands. The model can also perform multi-step tasks requiring longer context and combinations of skills, potentially applicable to material handling, packaging, and inspection use cases.

Simulation Frameworks and Physics Engine

In addition to the foundation model, NVIDIA announced several supporting technologies:

1. Newton physics engine - An open-source physics engine being developed in collaboration with Google DeepMind and Disney Research. Built on NVIDIA's Warp framework, Newton is designed to be compatible with simulation frameworks like Google DeepMind's MuJoCo and NVIDIA Isaac Lab, as described in the joint announcement.

2. MuJoCo-Warp - A collaboration between Google DeepMind and NVIDIA that reportedly accelerates robotics machine learning workloads by more than 70x. Google DeepMind researchers confirm this will be available through their MJX open-source library and the Newton physics engine.

3. NVIDIA Isaac GR00T Blueprint - A framework for synthetic manipulation motion generation that helps address the challenge of limited human demonstration data. They generated 780,000 synthetic trajectories—equivalent to 6,500 hours of human demonstration data—in just 11 hours.

Industry Collaborations

Several companies have been given early access to GR00T N1, including:

• 1X Technologies, whose humanoid robot was demonstrated during NVIDIA CEO Jensen Huang's keynote performing autonomous domestic tidying tasks, as showcased in the GTC keynote
• Agility Robotics
• Boston Dynamics
• Mentee Robotics
• NEURA Robotics

Disney Research reveals that they are utilizing the Newton physics engine to advance their robotic character platform, demonstrated during the GTC keynote with Star Wars-inspired BDX droids. NVIDIA, Disney Research, and Intrinsic also announced an additional collaboration to build OpenUSD pipelines for robotics data workflows.

Availability

• GR00T N1 training data and task evaluation scenarios are available for download from Hugging Face and GitHub
• The NVIDIA Isaac GR00T Blueprint for synthetic manipulation motion generation is available as an interactive demo on build.nvidia.com or to download from GitHub
• The Newton physics engine is expected to be available later this year

These technologies represent NVIDIA's continuing expansion beyond graphics processing units (GPUs) into the broader artificial intelligence and robotics ecosystem, leveraging the company's expertise in AI and simulation to address challenges in physical robot development.