Ok, it’s not a drill. The works are coming.
NVIDIA has announced a technology portfolio to supplement the development of humanoid robot, including NVIDIA Isaac GR00T N1, the first open, fully configurable model of fundamental generalized reasoning and humanoid skills.
Other technologies include simulation frames and plans, such as NVIDIA Isaac GR00T BLEPRINT for generating synthetic data, as well as Newton, an open source-in physics engine, Google Deepmind and Disney Research-Cella development for the development of robots.
Available now, GR00T N1 is the first of the family of fully configurable models that NVIDIA will prefer and spend worldwide robotics – accelerating the transformation of industries challenged by global labor deficiencies estimated at over 50 million people.
“The age of general robotics is here,” said Jensen Huang, founder and general director of NVIDIA. “Thanks to NVIDIA Isaac GR00T N1 and new data generation frames and a robot learning, robotics programmers will open the next border in the AI era everywhere.”
The company presented messages during Huang’s speech during the GTC 2025 event.
“It can be the largest industry of all,” said Huang.
He noticed that learning to strengthen and verifiable prizes (in the form of physics) will be packed with robot technology.
“We need a physics engine designed for fine-grained soft and stiff bodies,” he said. “We have to speed up GPU so that these virtuals can live in a very linear time.”
The GR00T N1 foundation model contains a double system architecture inspired by the principles of getting to know man. “System 1” is a brisk -thinking action model, reflecting human reflexes or intuition. “System 2” is a snail-paced model for deliberate, methodical decision making.
Supported by a vision language model, a system of 2 reasons for its environment and instructions, which he received for planning activities. System 1 Then translates these plans into precise, constant robot movements. System 1 is trained in the field of interpersonal demonstration data and a huge amount of synthetic data generated by the NVIDIA Omniverse platform.
The GR00T N1 can easily generalize in various tasks – such as gripping, moving objects with one or both arms and carrying elements from one arm to another – or perform multi -tank tasks requiring a long context and combinations of general skills. These possibilities can be used when using materials, packaging and inspection.
Developers and researchers can after training gr00T N1 with real or synthetic data for their specific humanoid robot or task.
In his speech, GTC Huang autonomously demonstrated the humanoid robot 1x performing domestic cleaning tasks using a set policy built on GR00T N1. Autonomous robot’s possibilities are the result of AI training cooperation between 1x and Nvidia.
“The future of humanoids concerns adaptation and learning abilities,” Bernt Børnich, CEO of 1x Technologies, said in a statement. “The NVIDIA GR00T N1 model is a serious breakthrough for the reasoning and ability to reason robots. With a minimum amount of data after the training, we were able to fully place on the Neo Gamma-by saving our mission to create robots that are not tools, but companions that can help people in significant, immeasurable ways.”
Among the additional leading humanoid programmers around the world with early access to GR00T N1 are robotics of agility, Boston dynamics, robotics and neura robotics.
Nvidia, Google Deepmind and Disney Research focus on physics
Nvidia announced cooperation with Google Deepmind and Disney Research to develop Newton, an open -level physics engine that allows robots to learn to operate sophisticated tasks with greater precision.
Built as part of the NVIDIA Warp, Newton will be optimized for robot learning and compatible with simulation frames, such as Mujoco Deepmind Google Deepmind and Nvidia Isaac Lab. In addition, three companies plan to enable Newton to operate the Disney physics engine.
Google Deepmind and NVIDIA cooperate to develop Mujoco-Warp, which is to speed up the load on machine learning robotics by more than 70 times and will be available to programmers through the Open Source Google Deepmind library, as well as through Newton.
Disney Research will be one of the first to operate Newton to develop their robot
A character platform that supplies up-to-date generation entertainment works such as
Expressive Star Wars inspired droids that joined Huang on the stage during his GTC
keynote.
“BDX droids are just the beginning. We commit ourselves to revive a larger number of characters in a way that the world has not seen before, and this cooperation with Disney Research, Nvidia and Google Deepmind is the key part of this vision,” said Kyle Laughlin, senior vice president of Walt Disney Imagineering Research &
Development in a statement. “This cooperation will allow us to create a new generation of robotic characters that are more expressive and engaging than ever before – and connect with our guests in a way that only Disney can.”
Nvidia and Disney Research, along with the internal, announced additional cooperation to build Openusd pipelines and the best practices in the field of robotics.
More data for the progress of robotics after training
Vast, diverse, high -quality data sets are crucial for the development of the robot, but costly to capture. In the case of humanoids, the actual human demonstration data is narrow by a 24-hour person’s day.
BluePrint Nvidia Isaac GR00t, announced today, helps to solve this challenge to generate synthetic manipulation movement. Built on OMNIVERSE and NVIDIA COSMOS models, the World Foundation Transfer, BluePrint allows programmers to generate a enormous amount of synthetic data to manipulation from a petite number of demonstrations.
Using the first available components to the plan, NVIDIA generated 780,000 synthetic trajectories – the equivalent of 6,500 hours or nine continuous months, data demonstration – in just 11 hours. Then, combining synthetic data with real data, NVIDIA improved the GR00T N1 performance by 40%, compared to the operate of real data.
To additionally equip the developers’ community with valuable training data, NVIDIA releases a set of GR00T N1 data as part of a larger set of AI Open Source-Open Data-also at the GTC and now available to hug face.
Availability
NVIDIA GR00T N1 Scenarios for evaluation and assessment of tasks are now available for download from Hugging Face and GitHub. NVIDIA Isaac GR00T BLEPRINT for generating synthetic manipulation movement is also available as an interactive demo on build.nvidia.com or for download from github.
NVIDIA DGX Spark Personal AI Supercomputer, also announced today in GTC, provides programmers with a turn -important system to extend the GR00T N1 capabilities for up-to-date robots, tasks and environments without intensive custom programming. Newton Physics Engine is expected to be available this year.
On GTC 2025 NVIDIA will take place Day Day Day Day, including:
● “Introduction to building humanoid robots” for deep diving in Nvidia Isaac GR00T;
● “insight into the robotic platform of Disney character” to find out how Disney Research re -defines entertainment robotics using BDX droids;
● “announcement of Mujoco-WARP and Newton: like Google Deepmind and NVIDIA are reclaiming the creation of robotics”, so that a deeper look at these up-to-date technologies and how Google implements AI models for training AI powered humanoids for real tasks.