NVIDIA has unveiled Isaac GR00T N1, the world’s first open humanoid robot foundation model, designed to bring advanced reasoning and general skills to robots. This fully customisable model aims to accelerate the development of humanoid robots capable of performing various tasks in industries facing labour shortages. Alongside GR00T N1, NVIDIA introduced new simulation frameworks and synthetic data generation tools, including the NVIDIA Isaac GR00T Blueprint. These technologies will help developers create robots with improved intelligence and adaptability.
At the core of GR00T N1 is a dual-system architecture that mimics human cognitive processes. The "System 1" model enables fast, reflexive actions, while "System 2" allows for more deliberate decision-making. The model integrates a vision-language component, which enables robots to understand their environment and execute tasks with precision. NVIDIA has trained GR00T N1 using vast amounts of synthetic data generated in its Omniverse platform, ensuring the model can generalise across various applications, from handling materials to assembling products.
A key demonstration of GR00T N1’s capabilities was showcased by robotics company 1X, where a humanoid robot performed household tidying tasks autonomously. Leading robotics firms, including Boston Dynamics, Agility Robotics, and NEURA Robotics, have gained early access to GR00T N1, indicating broad industry interest. NVIDIA has also partnered with Google DeepMind and Disney Research to develop Newton, an open-source physics engine designed to improve robotic motion and learning.
To further advance AI-driven robotics, NVIDIA has released the GR00T N1 dataset and training resources on platforms like Hugging Face and GitHub. Developers can use these resources to fine-tune GR00T N1 for specific tasks and robots. With these innovations, NVIDIA is laying the foundation for a new generation of intelligent, adaptable humanoid robots capable of seamlessly integrating into real-world environments.
