Introducing the fully customizable GR00T N1 base model to empower humanoid robots with general and reasoning skills
NVIDIA, Google DeepMind, and Disney Research collaborate on next-generation open source physics engine Newton
New Omniverse Blueprint and open source datasets for synthetic data generation to jumpstart the data flywheel for physics AI
SAN JOSE, Calif. — GTC — March 18, 2025 PT — NVIDIA today announced a suite of new technologies to help humanoid robots. These include the world’s first open-source, fully customizable foundational model, NVIDIA Isaac GR00T N1, which enables general-purpose humanoid robots to reason and perform skills.
Other technologies include simulation frameworks and blueprints, such as NVIDIA Isaac GR00T Blueprint for synthetic data generation, and Newton, an open-source physics engine built specifically for robotics, developed with Google DeepMind and Disney Research.
The GR00T N1 is the first in a series of fully customizable models that NVIDIA will pre-train and make available to robotics developers around the world. This will help accelerate transformation in industries plagued by a global labor shortage estimated at more than 50 million workers.
“The era of general-purpose robotics has arrived,” said Jensen Huang, founder and CEO of NVIDIA. “With the NVIDIA Isaac GR00T N1 and new data generation and robotic learning frameworks, robotics developers around the world are about to enter a new era of AI.”
GR00T N1 Advances the Humanoid Robot Developer Community
The GR00T N1 base model uses a dual-system architecture inspired by human cognitive principles. “System 1” is a fast-thinking action model that reflects human instinct or intuition. “System 2” is a slow-thinking model for deep-thought decision making.
System 2, powered by a visual-language model, plans actions by reasoning about the environment and instructions it receives. System 1 then translates these plans into precise, continuous robotic motions. System 1 is trained on human demonstration data and massive amounts of synthetic data generated by the NVIDIA Omniverse™ platform.
GR00T N1 easily adapts to and completes common tasks such as single-handed or dual-handed grasping, moving objects, transferring objects from one arm to another, or performing multi-step tasks that require long context and a general skill set. These capabilities can be applied to a variety of use cases, including object handling, packaging, and inspection.
Developers and researchers can post-train the GR00T N1 for specific humanoid robots or tasks using real or synthetic data.
During his GTC keynote, Huang demonstrated a 1X humanoid robot autonomously performing a room cleaning task, using a post-trained strategy based on the GR00T N1. The robot’s autonomy is the result of a collaboration between 1X and NVIDIA AI training.
“The future of humanoid development is focused on adaptability and learning,” said Bernt Børnich, CEO of 1X Technologies. “NVIDIA’s GR00T N1 model is a breakthrough in robot reasoning and skills. Our ability to fully deploy on NEO Gamma with minimal post-training data furthers our mission to create robots that are not tools, but companions that can provide meaningful and valuable assistance to humans.”
Other leading robotics companies around the world that have prioritized GR00T N1 include Agility Robotics, Boston Dynamics, Mentee Robotics and NEURA Robotics.
NVIDIA, Google DeepMind and Disney Research Focus on Physics
NVIDIA announced a collaboration with Google DeepMind and Disney Research to develop Newton, an open source physics engine that allows robots to learn how to handle complex tasks with greater accuracy.
Built on the NVIDIA Warp framework, Newton will be optimized for robot learning and compatible with simulation frameworks such as Google DeepMind MuJoCo and NVIDIA Isaac™ Lab. In addition, the three companies plan to enable Newton to leverage Disney’s physics engine.
Google DeepMind and NVIDIA are collaborating on MuJoCo-Warp, which is expected to speed up robotic machine learning workloads by more than 70 times and will be available to developers through Google DeepMind’s MJX open source library and Newton.
Disney Research will be one of the first companies to use Newton to advance its Robot Character Platform, which powers the next generation of entertainment robots, such as the expressive BDX robot inspired by Star Wars® that joined Huang on stage during the GTC keynote.
“The BDX robot is just the beginning. We are committed to bringing more characters to life in ways never before possible, and this collaboration with Disney Research, NVIDIA, and Google DeepMind is key to achieving this vision,” said Kyle Laughlin, senior vice president of research and development at Walt Disney Imagineering. “This collaboration will enable us to create a new generation of robotic characters that are more expressive and engaging than ever before, and connect with our guests in ways that are uniquely Disney.”
NVIDIA will also work further with Disney Research and Intrinsic to build OpenUSD pipelines and best practices for robotic data flows.
More Data to Advance Robot Post-Training
Large, diverse, high-quality datasets are critical to robotics development, but are expensive to capture. In the real world, where people only have 24 hours a day, the resulting human demonstration data is far from enough for humanoid robots.
The NVIDIA Isaac GR00T Blueprint for synthetic motion generation, announced today, helps address this challenge. Built on Omniverse and NVIDIA Cosmos Transfer world base models, the Blueprint allows developers to generate large amounts of synthetic motion data for manipulation tasks from a small amount of human demonstrations.
Using the first components provided for the Blueprint, NVIDIA was able to generate 780,000 synthetic trajectories in just 11 hours, equivalent to 6,500 hours or nine months of continuous human demonstration data. Then, by combining synthetic data with real data, NVIDIA improved the performance of GR00T N1 by 40% compared to using real data alone.
Also announced at GTC, to further provide valuable training data to the developer community, NVIDIA is releasing the GR00T N1 dataset as part of a larger open-source physics AI dataset, now available for download via Hugging Face.
Availability
NVIDIA GR00T N1 training data and task evaluation scenes are now available for download via Hugging Face and GitHub. The NVIDIA Isaac GR00T Blueprint for synthetic motion generation is also now available as an interactive demo at build.nvidia.com and can also be downloaded via GitHub.
Also announced today at GTC is NVIDIA DGX Spark, a personal AI supercomputer that provides developers with a turnkey system to extend the capabilities of GR00T N1 to new robots, tasks, and environments without extensive custom programming.
The Newton physics engine is expected to be available later this year.
To learn more, watch the NVIDIA GTC keynote and register for these key sessions on humanoid development technologies:
Building Humanoid Robots, a deep dive into the NVIDIA Isaac GR00T;
Enter the Disney Robotics Character Platform, a look at how Disney Research is redefining entertainment robotics with the BDX robot;
Introducing Mujoco-Warp and Newton: How Google DeepMind and NVIDIA are Advancing Robotics, a deep dive into these new technologies and how Google is deploying AI models to train AI-powered humanoid robots to accomplish real-world tasks.
