On Tuesday, at the GTC event in San Jose, NVIDIA introduced Isaac GR00T N1, the world’s first open-source humanoid robot foundation model.
CEO Jensen Huang declared that we have entered an era where robots can perform everyday tasks. He emphasized that the GR00T N1, along with new data generation and robotic learning frameworks, will enable robot developers worldwide to push the boundaries of AI.
GR00T N1 features a dual-system architecture inspired by human cognition.
System 1 mimics human reflexes and intuition to facilitate quick thinking and action, while System 2 focuses on deliberate and systematic decision-making.
This technology allows robots to handle both routine tasks, like picking up objects, and more complex operations requiring multiple steps.
Developers can fine-tune GR00T N1 using real or synthetic data, tailoring it to specific humanoid robots and tasks.
Leading humanoid robotics firms, including Boston Dynamics, Agility Robotics, and Mentee Robotics, have already begun working with GR00T N1.
NVIDIA also partnered with Google DeepMind and Disney Research to develop Newton, an open-source physics engine for robotics.
Newton is designed to be compatible with simulation frameworks like Google DeepMind’s MuJoCo and NVIDIA’s Isaac Lab.
Google DeepMind and NVIDIA are collaborating on MuJoCo-Warp, which is expected to boost robotics machine learning tasks by over 70 times. Disney Research plans to leverage Newton for next-gen entertainment robots, including the Star Wars BDX droid showcased at GTC.
NVIDIA created the Isaac GR00 T Blueprint to secure extensive data in robot development. Built on Omniverse and Cosmos Transfer, this technology can generate vast amounts of synthetic motion data from minimal human demonstrations.
An NVIDIA official revealed they generated 780,000 synthetic paths using the blueprint in just 11 hours – equivalent to nine months of human demonstrations. This data has been made available as open-source through Hugging Face.