Nvidia Introduces Isaac GR00T N1 to Advance Humanoid Robotics

At GTC 2025, Nvidia introduced Isaac GR00T N1, the world’s first open and fully customizable foundation model designed for general humanoid reasoning and skills.

Nvidia also introduced other technologies, the NVIDIA Isaac GR00T Blueprint for generating synthetic data and Newton, an open-source physics engine.

Groot N1 A generalized Robot Model

This model is open-source and pre-trained, but developers can post train it for specific robotic applications. Groot N1 robots can adapt, learn, and generalize their skills to new situations.

GROOT N1 features a “dual system architecture” for “thinking fast and slow,” inspired by human cognitive processes. GROOT N1’s slow thinking system helps robots perceive their surroundings, interpret instructions, and plan actions.   GROOT N1’s fast thinking system translates plans into robotic actions, allowing for complex, multi-step object manipulation.

GR00T N1 is compatible with various robotic hardware, making it flexible for different applications.

The model understands multiple data types, including language and images, enhancing its manipulation capabilities.

This “generalist” model is unique as it has been trained on both synthetic and real data.

GR00T N1 is trained using real-world human manipulation data, synthetic trajectory datasets, and video-based neural predictions. This broad training ensures better generalization across tasks.​

Developers can access and customize GR00T N1 via platforms like Hugging Face and GitHub, promoting collaborative advancement in humanoid robotics.

Newton: 

NVIDIA has announced collaboration with Google DeepMind and Disney Research to develop Newton, an-source physics engine that enables robots to to handle complex tasks with greater precision.

Built on the NVIDIA Warp framework, Newton will be optimized for robotic learning and will be compatible with simulation frameworks such as Google DeepMind’s MuJoCo & NVIDIA Isaac™ Lab.

Furthermore, the companies plan to enable Newton to utilize Disney’s physics engine.

Isaac GR00T Blueprint:  

High-quality, diverse datasets are essential for robot development but expensive to gather.

Humanoid robots face limitations since real-world human demonstration data is restricted to a person’s 24-hour day.

Nvidia’s Isaac GR00T Blueprint leverages Omniverse and Cosmos Transfer models to generate large-scale synthetic motion data, overcoming data scarcity and speeding up robot training and development.

Humanoid Robotics Showdown: Nvidia’s GR00T N1 vs. Tesla and Figure AI Models”

Nvidia’s GR00T N1 competes with models like Tesla’s Optimus and Figure AI’s Figure 01.

GR00T N1 offers open-source customization and dual-system reasoning, enhancing robotic adaptability.

Optimus focuses on factory automation, while Figure 01 emphasizes general-purpose tasks.

GR00T N1’s multimodal processing and real-world data training give it an edge in versatile humanoid applications.

News Gist

At GTC 2025, Nvidia unveiled GR00T N1, the world’s first open, customizable humanoid foundation model.

It features dual-system reasoning for complex tasks, supports diverse hardware, and processes language and images.

Developers can access it via Hugging Face and GitHub.

Nvidia introduced the Isaac GR00T Blueprint for synthetic data generation and Newton, an open-source physics engine.