NVIDIA announced the launch of the universal Open Source humanoid Bots foundation model Isaac GR00T N1

NVIDIA announced the launch of the general humanoid robot foundation model GROOT N1 at GTC. This is the world's first open-source foundation model and inference training simulation framework available to global robot development teams, designed to accelerate humanoid robot development technology. GR00T N1 is the first fully customizable model that NVIDIA will pre-train and release to global robot developer teams.

NVIDIA founder Jensen Huang stated that the era of Generalist Robotics ( has arrived. With the NVIDIA Isaac GR00T N1 foundational model and the new robotics data learning framework, robot development teams around the world can accelerate the opening of a new chapter for future AI robots. As the name suggests, Generalist Robotics refers to robots that can be trained to perform any task, transforming them into versatile talents capable of handling a variety of roles; they can be operators or assist as household helpers, executing any work assignments that come to mind.

GR00T N1 dual system architecture model is inspired by human cognitive understanding patterns.

The GR00T N1 basic model adopts a dual-system architecture, inspired by human cognitive understanding patterns. "System 1" is a fast-thinking action model that reflects human reactions or intuitions, while "System 2" is a slow-thinking model used for thoughtful and orderly decision-making.

In the framework of this visual language model, System 2 can think about the environment and the instructions received to plan actions. System 1 trains machines based on human drill data and extensive synthetic simulation data generated by the NVIDIA Omniverse platform, translating these plans into precise and continuous robotic activity.

GR00T N1 can easily perform common tasks such as grasping actions, moving items with one or two arms, and transferring items from one arm to another, or executing multi-step tasks in the same context, applicable to scenarios such as material handling, packaging, and inspection. Developers and researchers can use real or simulated synthetic data for subsequent training of humanoid robots.

During the GTC keynote speech, Jensen Huang demonstrated the NEO Gamma humanoid robot from 1X AI performing housekeeping tasks based on post-training execution built on GR00T N1. The autonomous capabilities of the robot named NEO Gamma are the result of a collaboration between the robotics startup 1X AI and NVIDIA in AI machine learning. Imagine a scene of a lazy person at home, training a robot butler to help with cleaning or acting as a kitchen assistant for tedious household chores while lounging on the sofa listening to music. The robot butler can be trained according to personal needs, learning actions in a way that mimics human cognitive thinking.

Bernt Børnich, CEO of 1X Technologies, stated that the future of humanoid robots lies in adaptability and learning capabilities. NVIDIA's GR00T N1 significantly enhances the reasoning and skills of robots, allowing developers to utilize post-training of GR00T N1 with minimal data, leading to a comprehensive deployment on NEO Gamma.

Other developers of humanoid robots who have obtained early access to GR00T N1 development rights include Agility Robotics, Boston Dynamics, Mentee Robotics, and NEURA Robotics.

NVIDIA collaborates with Google DeepMind and Disney Research to develop Newton

NVIDIA announced a collaboration with Google DeepMind and Disney Research to develop the Open Source hardware engine Newton, which allows Bots to learn how to handle complex tasks more accurately.

Newton is built on the NVIDIA Warp framework, optimizing robotic learning and compatible with simulation frameworks such as Google DeepMind's MuJoCo and NVIDIA Isaac Lab.

Google DeepMind and NVIDIA are collaborating to develop MuJoCo-Warp, which is expected to accelerate robot machine learning workloads by more than 70 times, and will be made available to developers through Google DeepMind's MJX Open Source library and Newton.

The robot BD1 from the Star Wars movie series has manifested in the real world.

Disney Research became one of the first companies to use Newton, and Huang Renxun shared the stage with the robot BDX Droids during his speech at GTC, which was inspired by the robot character BD1 from the Star Wars series. Kyle Laughlin, Senior Vice President of Disney Imagineering, stated that the collaboration with NVIDIA allowed them to create more engaging robot characters than ever before, giving the development team the opportunity to produce more entertaining industrial designs for robots, combining the appearance and movements of the robots with NVIDIA's technology.

NVIDIA Isaac GR00T Blueprint Generation Assistance Bots Post-Training

High-quality data tools are crucial for developing Bots, but the cost of obtaining data is high. NVIDIA Isaac GR00T simulation manipulation generating blueprints helps address this challenge. The blueprints are based on Omniverse and NVIDIA Cosmos Transfer world foundation models, enabling developers to generate large amounts of simulation synthetic data for operational tasks with minimal human training. NVIDIA used the blueprints to produce 780,000 synthetic trajectories in just 11 hours, equivalent to nine months of human training data. By combining synthetic data with real data, NVIDIA increased the performance of GR00T N1 by 40% compared to using only real data.

NVIDIA releases GR00T N1 Open Physical AI Dataset to provide training data for developer teams, and developers can obtain information on Hugging Face.

This article states that NVIDIA has announced the launch of the universal Open Source humanoid robot foundation model Isaac GR00T N1, which first appeared in Chain News ABMedia.