NVIDIA Unveils Isaac GR00T N1: The Future of Humanoid Robotics

• The introduction of Isaac GR00T N1, the first-ever humanoid robot open foundation model in the world, by NVIDIA, opens new avenues in the realm of advanced reasoning and skills adaptability.
• The open-source Newton physics engine, innovated by NVIDIA, will go a long way in the research of robot learning and accuracy, which is done in collaboration with Google DeepMind and Disney Research.

NVIDIA has entered a breakthrough in robotics through the launch of Isaac GR00T N1, a world-first model for humanoid robots that is completely open and fully configurable. Announced during GTC 2025, this revolutionary technology was introduced to change industries challenged with shortages of workers by enhancing humanoid robot capabilities with generic abilities and reasoning. It is part of a wider portfolio embracing simulation frameworks, synthetic data generation, and open-source physics engines developed in collaboration with Google DeepMind and Disney Research.

Revolutionising Humanoid Robotics with GR00T N1

The GR00T N1 Foundation Model offers a dual-system framework for humanoid robots inspired by human cognition. 

System 1 is a fast-action model mimicking reflexive human responses, primarily for the real-time control of the robot’s movements.

System 2 refers to slow-action, deliberate reasoning based upon a vision-language system to facilitate planning and decision-making based on information coming from environmental cues. 

A combination of the two systems makes GR00T N1 generalisable across an impressive spectrum of tasks, from simpler aspects of object handling to complex multi-step activities. This versatility is an asset in physical work like material handling, packaging, and industrial inspection.

Synthetic Data to Raise Robots Further Upwards

To shortcut the delay in AI-enabled humanoid development, NVIDIA has unveiled the GR00T Blueprint – an architecture for mass-producing synthetic data. The blueprint natively applies to the NVIDIA Omniverse and Cosmos Transfer world foundation models, allowing thousands of human demonstration simulations to be completed within hours.

That kind of performance boost would be expected after fusing 780,000 synthetic trajectories with a handful of real data points. The combination was able to increase the efficiency of the GR00T N1 by 40%. This pipeline for generating synthetic data will also allow fast post-training, where robots can be tailored for very specific tasks with very little training by example from the real world.

Collaborations with Tech Giants for Advanced Physics Simulations

In a major partnership, NVIDIA joins forces with Google DeepMind and Disney Research to create Newton, an open-source physics engine set to enhance the robotics learning process.

Newton, built on the NVIDIA Warp framework, interfaces with simulator environments such as Google DeepMind’s MuJoCo and NVIDIA Isaac™ Lab. This physics engine will allow the robot to learn with the precision necessary for executing very difficult actions involving real-world interactions, such as lifting, grasping, and traversing uneven surfaces.

Robots of Disney Powered by NVIDIA 

Disney Research will be using Newton to upgrade its robotic character platform for future entertainment robots. Their BDX droids seen during the keynote of GTC hinted at the potential of the technology for immersive, interactive experiences in theme parks of the future inspired by Star Wars.

Open Access and Global Developer Adoption 

At Hugging Face and GitHub, NVIDIA is launching GR00T N1 datasets and task evaluation scenarios as open-source resources for all developers worldwide. Leading robotics companies, including Agility Robotics, Boston Dynamics, Mentee Robotics, and NEURA Robotics, already have early access to GR00T N1 to fast-track their development of next-gen humanoids.

DGX Spark: Strengthening Humanlike Potentials

NVIDIA recently catalysed the first expansion of self-propelled personal AI supercomputers, the extension of GR00T N1 capabilities, by launching DGX Spark for developers. This plug-and-play unit allows a user to tailor robot models in training and deployment without programming and significantly facilitates advanced robotics development.

Here Comes the Future of Humanoid Robotics

With GR00T N1, NVIDIA opens a new realm for generalist humanoid robots, learning and adopting the challenges posed to its real-world tasks. Synthetic data representation has combined with open-source frameworks and collaborative physics engines to launch a new paradigm of acceleration across industries – from manufacturing and logistics to health care and entertainment.

As NVIDIA’s Isaac platform matures, the fantasy of humanoid robots that will co-labour with human beings in dynamic, real-world settings is very rapidly becoming a reality.