NVIDIA GR00T N1: First Open Humanoid AI Model
At GTC 2025, Nvidia made a bold bet on the future of robotics with the launch of Isaac GR00T N1, the world’s first open, fully customizable foundation model for generalized humanoid reasoning and skills. This major milestone signals the dawn of generalist AI-driven robots designed to adapt, learn, and assist in the real world.
“The age of generalist robotics is here,” declared Nvidia CEO Jensen Huang. “This could be the biggest industry of all.”
What Is GR00T N1?
GR00T N1 (Generalist Robot 00 Technology) is a dual-system architecture inspired by how humans think:
- System 1: Fast, intuitive decision-making (like reflexes)
- System 2: Slow, deliberate planning and reasoning
These two systems work together to interpret vision and language instructions, plan multistep actions, and execute precise motor movements — allowing robots to grasp, move, transfer, and even tidy up a room autonomously.
Trained on both real and synthetic data — including 780,000 motion trajectories generated in just 11 hours using Nvidia’s Omniverse tools — GR00T N1 dramatically reduces the time and cost needed to develop intelligent robot behavior.
Who’s Using It Already?
Companies like 1X Technologies, Agility Robotics, Boston Dynamics, and Neura Robotics are already building on GR00T N1. In a live demo, 1X’s humanoid robot used GR00T to autonomously clean and organize household items — showing what’s now possible with minimal post-training.
“GR00T N1 is a breakthrough,” said 1X CEO Bernt Børnich. “We’re creating robots that aren’t tools — they’re companions.”
Newton: The Physics Engine for Robots
Alongside GR00T, Nvidia unveiled Newton — an open-source GPU-accelerated physics engine built for soft and rigid body simulation. It’s being developed with Google DeepMind and Disney Research, and will integrate with MuJoCo, Isaac Lab, and Disney’s own simulation tech.
“We need a physics engine designed for fine-grained bodies — and we need it to run in super-linear time,” Huang said.
The goal: enable robots to safely learn, experiment, and evolve in high-fidelity virtual environments before operating in the real world.
Disney’s Role in the Future of Robotic Characters
In a surprise collaboration, Disney Research is using Newton to create the next generation of robotic characters. At GTC, expressive Star Wars BDX droids shared the stage with Huang — showcasing the blend of entertainment, robotics, and AI.
“This collaboration lets us create characters that connect with guests in ways only Disney can,” said Kyle Laughlin, SVP of Walt Disney Imagineering.
Synthetic Data at Scale
One of the biggest bottlenecks in humanoid training has been data. Real-world demonstrations are time-limited, but Nvidia’s new Isaac GR00T Blueprint lets developers generate massive synthetic motion datasets — making post-training faster and more accurate.
In testing, combining synthetic and real data improved GR00T N1’s performance by 40% versus real data alone.
Open Access, Open Source, and What's Next
All key assets — including GR00T N1 training data, motion blueprints, and task evaluation scenarios — are now available on Hugging Face and GitHub. Nvidia’s new DGX Spark personal AI supercomputer can help developers expand GR00T’s capabilities with minimal effort.
Upcoming sessions at GTC’s Humanoid Developer Day will provide deeper dives into:
- Building humanoid robots with GR00T
- Disney’s robotic character platform
- MuJoCo-Warp and Newton collaborations with DeepMind
The Future: Robots That Learn, Adapt, and Collaborate
With GR00T N1, Newton, and its partnerships with top robotics innovators, Nvidia is making one thing clear: humanoid robots are no longer science fiction. They’re entering the workforce — and perhaps even your home — with real intelligence, adaptability, and purpose.
And this is just the beginning.
