Project GR00T
About
About Project GR00T
GR00T N models combine a vision-language reasoning system with an action model that converts instructions and sensor observations into robot movement. Cross-embodiment training allows the architecture to learn from different humanoids, arms and bimanual systems while retaining information about each robot's body and control space.
The development workflow links teleoperation, synthetic data and Isaac Lab simulation with model training and evaluation. Developers can fine-tune GR00T on their own demonstrations, test the policy in simulation and deploy it through Isaac ROS on Jetson Thor and other NVIDIA compute. The wider platform includes data-generation, simulation and runtime tools around the model.
Advantech introduces UniCAM multi-camera integration on NVIDIA Jetson Thor
LG Innotek and NVIDIA develop robot vision sensing with Holoscan Sensor Bridge
Comau integrates NVIDIA Jetson and Isaac tools across industrial robotics portfolio
NVIDIA introduces Flora data engine for robotics video annotation
Timeline
- NVIDIA and Hugging Face bring Isaac GR00T 1.7 and Isaac Teleop to LeRobot
2026-07-06 - Partnership - The integration gives robotics developers access to NVIDIA Isaac GR00T 1.7, Isaac Teleop and related datasets through LeRobot workflows. NVIDIA says Cosmos 3 support is planned for the same open robotics development path.
Source - NVIDIA Isaac GR00T N1.7
2026-04-17 - Research Publication - Released the GR00T N1.7 family of open, commercially licensed vision-language-action models for humanoid robots, including a 3B-parameter variant with public Hugging Face and GitHub access. Uses an Action Cascade architecture combining a vision-language model for reasoning with a diffusion transformer for motor control. Supports local fine-tuning and real-world deployment of dexterous multi-step manipulation policies on humanoid platforms.
Source - Amazon Robotics adopts NVIDIA Cosmos or Isaac GR00T for robotics development
2025-10-28 - Capability - The 28 October 2025 NVIDIA announcement lists Amazon Robotics among companies using Cosmos or Isaac GR00T N models to generate synthetic data, teach robots new behaviours and deploy physical-AI agents. Cosmos provides world-foundation models and tools for simulation and reasoning, while Isaac GR00T supplies robot foundation models for humanoid-robot reasoning, generalisation and whole-body control. The source supports model adoption by Amazon Robotics, but it does not describe a separately negotiated partnership, product deployment site or specific Amazon robot using one model.
Source - NVIDIA Launches Newton Physics Engine and Isaac GR00T N1.6 at CoRL 2025
2025-09-29 - Product Launch - NVIDIA introduced the open-source, GPU-accelerated Newton physics engine, co-developed with Google DeepMind and Disney Research, to enhance robot simulation capabilities. The Isaac GR00T N1.6 model integrates Cosmos Reason, a vision language model that enables robots to interpret ambiguous instructions and generalize across tasks. These advancements aim to accelerate the development of physical AI by providing developers with tools for more accurate simulations and improved robot reasoning.
Source - NVIDIA powers humanoid robot industry with Isaac GR00T platform adopted by Agility, Boston Dynamics, Fourier and others
2025-05-19 - Partnership - NVIDIA released Isaac GR00T N foundation models enabling humanoid robots to perform complex manipulation tasks via natural language commands. Agility Robotics, Boston Dynamics, Fourier, Galbot, Mentee Robotics, NEURA Robotics, General Robotics, Skild AI, and XPENG Robotics are adopting the NVIDIA Isaac platform. AeiRobot, Lightwheel, Foxlink, and NEURA Robotics are early adopters of GR00T N models; Foxconn and Foxlink use GR00T-Mimic blueprints for manufacturing.
Source - NVIDIA announces Isaac GR00T blueprint for humanoid robot training
2025-01-06 - Product Launch - NVIDIA released the Isaac GR00T Blueprint as a set of robot foundation models, data pipelines and simulation frameworks aimed at accelerating humanoid-robot development. The blueprint’s synthetic-motion-generation workflow helps developers create large volumes of training data and use imitation learning to teach humanoids new skills. The release gives developers a repeatable path from generated motion data to trained behaviors, extending NVIDIA’s Isaac tooling around the Project GR00T platform.
Source - NVIDIA announces Project GR00T generalist humanoid foundation model initiative at GTC
2024-03-18 - Product Launch - NVIDIA introduced Project GR00T as a general-purpose foundation model initiative for humanoid robots, focused on enabling robots to understand natural language and imitate human actions. The initiative includes simulation, data generation, and model training tools built on NVIDIA Isaac and Omniverse to accelerate humanoid development. NVIDIA positioned GR00T as a core pillar of its physical AI stack alongside robotics simulation and edge compute platforms.
Source
Context
Market context
Project GR00T is NVIDIA's foundation model for humanoid robots, designed to enable general-purpose robot intelligence.
- Target environment: Humanoid research labs, simulation environments, robotics developer workflows and NVIDIA robotics infrastructure.
- Workflow context: Vision-language-action modeling, humanoid robot learning, manipulation policy training and data pipeline development.
- Customer context: Robot developers, AI researchers, humanoid companies and teams building on NVIDIA robotics infrastructure.
- Deployment model: Development platform and foundation-model program connected to NVIDIA Isaac, Cosmos and humanoid robotics work.
- Commercial maturity: Active development initiative with public developer resources and partner ecosystem positioning.
- Adoption constraints: Adoption depends on developer access, model capability, hardware compatibility, simulation fidelity, safety and real-robot validation.
- Market position: Foundation-model layer for humanoid robotics, positioned as intelligence infrastructure rather than a robot body.
- Adjacent products: Google Gemini Robotics, Figure Helix, Skild Brain, NVIDIA Isaac