NVIDIA Unveils Comprehensive AI Stack for Next-Generation Humanoid Robots

NVIDIA unveils brain-and-body stack to train next-gen humanoids

Together, these technologies give researchers and developers a unified, open, and accelerated robotics stack. The tools promise to speed up iteration, standardize testing, unify training with on-robot inference, and help robots transfer skills more safely from simulation to the real world. "Humanoids are the next frontier of physical AI, requiring the ability to reason, adapt and act safely in an unpredictable world," said Rev Lebaredian, vice president of Omniverse and simulation technology at NVIDIA. "With these latest updates, developers now have the three computers to bring robots from research into everyday life -- with Isaac GR00T serving as robot's brains, Newton simulating their body and NVIDIA Omniverse as their training ground." The updates also underscore NVIDIA's ambition to define the standard platform for building general-purpose robots, especially humanoids, which demand precise physics and complex reasoning to function in human environments.

New NVIDIA Models Helps Robots Learn, Reason, and Act in the Real World | AIM

NVIDIA also unveiled the GB200 NVL72 system, RTX PRO servers and Jetson Thor for real-time on-robot inference. NVIDIA has released new open models and simulation libraries to support global robotics research and development. The announcement, made at the Conference on Robot Learning (CoRL) in Seoul, includes the open-source Newton Physics Engine, the Isaac GR00T N1.6 foundation model and new Cosmos world foundation models. The updates aim to accelerate the way robots learn, reason, and transfer skills from virtual to real environments. Rev Lebaredian, vice president of Omniverse and simulation technology at NVIDIA, said, "Humanoids are the next frontier of physical AI, requiring the ability to reason, adapt and act safely in an unpredictable world." The Newton Physics Engine, developed with Google DeepMind and Disney Research, is now available in NVIDIA Isaac Lab. Managed by the Linux Foundation, the engine supports complex actions such as walking on uneven ground and handling delicate objects. ETH Zurich, Technical University of Munich and Peking University are among the first adopters. NVIDIA has also announced Isaac Lab - Arena, an open-source policy evaluation framework codeveloped with Lightwheel. The tool will enable large-scale testing of robotic skills in diverse simulated environments. The Isaac GR00T N1.6 model integrates NVIDIA's Cosmos Reason, which helps robots turn vague instructions into step-by-step plans using prior knowledge and physics-based reasoning. The model, available on Hugging Face, also supports multi-task operations such as opening heavy doors. NVIDIA's Cosmos world foundation models, which have been downloaded more than 3 million times, have been updated to generate large-scale training data. Cosmos Predict 2.5 and Cosmos Transfer 2.5, due soon, will offer longer video generation, multi-view outputs and faster synthetic data creation. Boston Dynamics used NVIDIA's new grasping workflow in Isaac Lab 2.3 to train its Atlas robots to improve manipulation skills. Other companies adopting NVIDIA's Isaac and Omniverse platforms include Agility Robotics, Figure AI, Franka Robotics, Techman Robot and Solomon. NVIDIA also unveiled new AI infrastructure, including the GB200 NVL72 system, RTX PRO servers and Jetson Thor for real-time on-robot inference. These tools are being adopted by partners such as Figure AI, Meta, Google DeepMind and the RAI Institute. Lebaredian added, "With these latest updates, developers now have the three computers to bring robots from research into everyday life, with Isaac GR00T serving as the robot's brain, Newton simulating their body and NVIDIA Omniverse as their training ground."