NVIDIA revolutionizes robotics with Isaac GR00T N1: new strategic collaborations around the corner

During the GTC conference, NVIDIA presented a series of cutting-edge technologies to accelerate the development of humanoid robots. Among these, NVIDIA Isaac GR00T N1 stands out, the first fully customizable open-source base model, designed to equip robots with advanced reasoning capabilities and generalized skills.

Together with the model, NVIDIA announced simulation frameworks and innovative tools, including the Isaac GR00T Blueprint for synthetic data generation and Newton, an open-source physics engine developed in collaboration with Google DeepMind and Disney Research.

Isaac GR00T N1: the future of robotic intelligence

With Isaac GR00T N1, NVIDIA takes a significant step forward for robotics, thanks to a dual-system architecture inspired by the principles of human cognition:

• System 1: a fast and intuitive model, which acts as an automatic reflex to perform immediate actions.
• System 2: a slower and methodical model, capable of analyzing the context and planning more complex strategies.

The model integrates a Vision Language Model (VLM) that interprets the environment and the instructions received, translating them into precise and fluid movements. Thanks to a vast dataset of human demonstrations and synthetic data generated by the NVIDIA Omniverse platform, GR00T N1 is capable of adapting to a wide range of advanced robotic tasks, such as:

• Grasp objects with one or both hands.
• Move and transfer objects between different limbs.
• Execute multi-step operations with high precision.

These capabilities make it particularly suitable for industrial and domestic applications, such as material handling, packaging, and inspection.

In the official video shown below, NVIDIA presents the main features of the new Isaac GROOT N1:

Practical applications and initial field tests

The potential of the model has already been demonstrated by 1X Technologies, which used GR00T N1 to train its humanoid robot NEO Gamma to perform autonomous tidying tasks. According to Bernt Børnich, CEO of 1X, the model allows for the development of more adaptable robots capable of learning quickly with minimal amounts of additional data.

In addition to 1X, other leading companies in the robotics sector, including Agility Robotics, Boston Dynamics, Mentee Robotics, and NEURA Robotics, have gained early access to GR00T N1 to explore new applications.

Newton: the physical engine for the next generation of robots

To further enhance the ability of robots to interact with the physical world, NVIDIA has announced Newton, an open-source physics engine developed in collaboration with Google DeepMind and Disney Research. Based on the NVIDIA Warp platform, Newton will be optimized for robotic learning and compatible with MuJoCo by Google DeepMind and NVIDIA Isaac Lab.

Disney Research will use Newton to refine its entertainment robotics platform, already demonstrated during the GTC keynote with the Star Wars BDX-inspired droids. According to Kyle Laughlin, Senior Vice President of Walt Disney Imagineering Research & Development, this collaboration will help create more expressive and engaging robots for the audience.

The importance of synthetic data for robotic training

One of the main obstacles in the development of humanoid robotics is the need for large amounts of data for training. To tackle this challenge, NVIDIA has launched the Isaac GR00T Blueprint for synthetic data generation. This tool, based on Omniverse and Cosmos Transfer, allows for the production of enormous volumes of motion data from a limited number of human demonstrations.

Thanks to this technology, NVIDIA generated 780,000 synthetic trajectories, equivalent to 6,500 hours of human data, in just 11 hours. By using a combination of synthetic and real data, the performance of GR00T N1 improved by 40%, compared to training based solely on real data.

NVIDIA GR00T N1 and Newton: availability and access to the tools

To further support developers, NVIDIA has made available:

• The training data and evaluation scenarios of GR00T N1 on Hugging Face and GitHub.
• The Blueprint of Isaac GR00T as an interactive demo on build.nvidia.com and downloadable from GitHub.

Finally, the new NVIDIA DGX Spark personal supercomputer, also presented at the GTC, will offer an advanced infrastructure to expand the capabilities of GR00T N1 without the need for complex programming.

Conclusions: the future of humanoid robotics

With these tools, NVIDIA aims to make humanoid robots increasingly autonomous, efficient, and adaptable, opening new frontiers in the integration ofartificial intelligence into everyday life and the industrial world.

The combination of Isaac GR00T N1, Newton, and the generation of synthetic data represents a huge step forward for robotics, bringing us closer to a new era in which humanoid robots will be an integral part of our society, working alongside humans in complex and repetitive tasks.