At CoRL, NVIDIA launches robotics tools: Newton physics engine, Isaac GR00T reasoning, and Cosmos data generation, plus Blackwell GPUs and Jetson Thor.
© D. Novikov
At the CoRL robotics conference, NVIDIA unveiled several open solutions that could significantly accelerate the development of robots. The lineup includes the Newton physics engine, created together with Google DeepMind and Disney’s research center; a new version of the Isaac GR00T model that brings reasoning capabilities to robots; and the generative system Cosmos, which produces training data for learning.
The Newton physics engine is designed to tackle a core challenge in robotics: skills learned in simulation often fail to transfer to the real world. This new GPU-accelerated engine can faithfully simulate demanding scenarios—whether it’s walking on snow or handling delicate objects. It is already in use at Stanford, the University of Zurich, and other leading labs.
Isaac GR00T N1.6 helps robots understand and execute human-style commands, such as bringing a glass of water. With integration into the visual-language system Cosmos Reason, robots can plan actions step by step and perform multiple tasks in parallel—for example, moving while opening a heavy door.
NVIDIA also refreshed its training toolkit: Isaac Lab now includes an automated curriculum where robots build skills from simple to complex, and the Isaac Lab Arena platform supports large-scale testing. On the hardware side, the company introduced servers with Blackwell GPUs and a new Jetson Thor module that lets robots process multiple AI workloads in real time.
Taken together, NVIDIA is offering a full stack—from simulation and AI models to the hardware foundation. If this momentum holds, the moment when robots step out of the lab and into everyday life may be closer than it once seemed.