NVIDIA will recharge the development of Humanoid robots with ISAAC GR00T N1 model for human -like reasoning

Okay, this is not a workout. The robots are coming.

Nvidia has announced a portfolio of technologies for developing humanoid robots, including the NVIDIA ISAAC GR00T N1, the first open in the world, a fully personalized model of the Foundation for Generalized Humanoid Reflections and Skills.

Other technologies include simulation frames and drawings such as the NVIDIA ISAAC GR00T BluePrint to generate synthetic data, as well as a Newton, an open-called physics engine with Google Deepmind and Disney Research-Count development, built to develop robots.

Available now, the GR00T N1 is the first of a family of fully personalized models that NVIDIA will pre -process and launch the world’s robotics developers – accelerating the transformation of industries caused by a global labor shortage estimated at more than 50 million people.

“The age of generalistic robotics is here,” says Jenson Huang, founder and CEO of NVIDIA. “With the NVIDIA ISAAC GR00T N1 and new frameworks for data generation and robot training, robotics developers will open the next border in the AI ​​era everywhere.”

The company revealed the news during Huang’s main speech in GTC 2025 an event.

“This can be the biggest industry of all,” Huang said.

He noted that strengthening the training and checking awards (in the form of physics) will drive the robot technology forward.

“We need a physical engine designed for finely grained soft and hard bodies,” he said. “We need this to speed up GPU so that these virtuals can live in a super linear time.”

The GR00T N1 Foundation model has a dual architecture system inspired by the principles of human knowledge. System 1 is a fast -thinking model of action reflecting human reflexes or intuition. System 2 is a slow -thinking model for deliberate, methodical decision making.

Powered by a vision language model, system 2 reasons for its environment and the instructions it has received to plan action. Then system 1 translates these plans into precise, continuous movements of the robot. System 1 is trained in demonstration of people and a huge amount of synthetic data generated by the NVIDIA Omniverse platform.

The GR00T N1 can be easily summarized for common tasks – such as gripping, moving objects with one or both hands, and transferring elements from one hand to another – or perform multi -solution tasks that require a long context and combinations of common skills. These options can be applied in cases of use, such as materials, packing and inspection.

Developers and researchers can follow the GR00T N1 with real or synthetic data on their specific humanoid robot or task.

In his main note, Huang demonstrates the humanoid robot of 1x autonomous home tasks for arranging, using a post-training policy built on the GR00T N1. The autonomous capabilities of the robot are the result of cooperation for AI training between 1x and NVIDIA.

“The future of the humanoids is related to adaptability and learning,” said Bernt Bernic, CEO of 1x Technologies. “The NVIDIA GR00T N1 model provides a major breakthrough for the reasoning and skills of the robots.

Among the additional leading developers of humanoids around the world with early access to the GR00T N1 are Agility Robotics, Boston Dynamics, Mentee Robotics and Neura Robotics.

Nvidia, Google Deepmind and Disney Research focus on physics

Nvidia has announced cooperation 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 frame, Newton will be optimized for training robots and compatible with simulation frames such as Google Deepmind’s Mujoco and NVIDIA ISAAC laboratory. In addition, the three companies plan to allow Newton to use Disney physics.

Google Deepmind and NVIDIA are collaborating to develop Mujoco-Warp, which is expected to accelerate the load on robotics machine learning more than 70 times and will be available to developers through the MJX Library with the Google Deepmind open code as well as through Newton.

Disney Research will be one of the first to use Newton for progress
A character platform that feeds robots for the next generation of entertainment, such as
Expressive Star Wars -inspired BDX Droid that joined Huang on stage during his GTC
Main note.

“BDX droids are just the beginning.
Development, in a statement. “This collaboration will allow us to create a new generation of robotic characters who are more pronounced and engaged than ever – and to contact our guests in ways that only Disney can.”

Nvidia and Disney Research, along with Interins, have announced additional cooperation for the construction of Openusd pipelines and best practices for robotics data work processes.

More data on robotics progress after training

Large, diverse, high quality data sets are crucial to the development of robots, but expensive to capture. For humanoids, real-world demonstration data is limited by a 24-hour person.

Declared today, the NVIDIA ISAAC GR00T plan for synthetic manipulation of movement helps to cope with this challenge. Built on Omniverse and NVIDIA Cosmos Transfer World Foundation models, Blueprint allows developers to generate exponentially large amounts of synthetic movement data for manipulation of a small number of human demonstrations.

Using the first components available for BluePrint, NVIDIA generates 780,000 synthetic trajectories – the equivalent of 6,500 hours or nine continuous months of demonstration data for human – only 11 hours. Then, combining synthetic data with real data, NVIDIA improved the efficiency of GR00T N1 by 40%compared to the use of real data.

For the additional equipment of the developers with valuable training data, NVIDIA launches the GR00T N1 data set as part of a larger physical set of AI data with an open code, also announced in GTC and is now available to a hug.

Availability

NVIDIA GR00T N1 Training Training Data and Task Evaluation Data are now available for download from Hugging Face and GitHub. The NVIDIA ISAAC GR00T plan for generating the movement of synthetic manipulation is also now available as an interactive demonstration of Build.Nvidia.com or for download from gitHub.

The Nvidia DGX Spark Personal Ai Supercomputer, also announced today in GTC, provides developers with turnkey system to expand the GR00T N1 capabilities for new robots, tasks and environments without extensive custom -made programming. The Newton Physics engine is expected to be available later this year.

In the GTC 2025 NVIDIA will hold sessions for human developers, including:
● “Introduction to the construction of humanoid robots” for deep diving in the NVIDIA ISAAC GR00T;
● “Disney Robotic Platform Insight to learn how Disney Research redefines entertainment robotics with BDX Droids;
● “Declaration of Mujoco-Warp and Newton: How Google Deepmind and Nvidia are developing robotics development” for a deeper view of these new technologies and how Google uses AI models for humanoid training powered by AI for real-world tasks.