NVIDIA has unveiled a powerful set of open-source tools that reshape robotics research. Announced in Seoul at CoRL 2025, the releases include the Newton Physics Engine, Isaac GR00T model and Cosmos world models. Together, these innovations accelerate development, improve simulation accuracy and enhance reasoning across academic and industrial robotics projects.
Newton Physics Engine: Raising Simulation Standards
Robots learn more safely and efficiently in simulation before facing the real world. NVIDIA’s new Newton Physics Engine, codeveloped with Google DeepMind and Disney Research, delivers a breakthrough in accuracy and scalability. By simulating complex conditions like snow, gravel and object handling, it allows robots to practice tasks with greater realism and reliability.
Universities such as ETH Zurich and Peking University, along with companies like Style3D, are early adopters. Newton’s flexible architecture also ensures that developers can adapt it to a wide range of robotic applications, from humanoids to industrial automation.
Isaac GR00T: Enhancing Humanlike Reasoning
The Isaac GR00T N1.6 model brings sophisticated reasoning and planning abilities to robots. By integrating NVIDIA’s Cosmos Reason VLM, robots can now interpret vague instructions and translate them into step-by-step actions.
This advancement enables humanoids to manage multiple simultaneous tasks, such as opening doors while carrying objects. The model’s open-source availability on Hugging Face ensures wide accessibility for research labs and developers alike.
Cosmos World Models: Scaling Physical AI Training
NVIDIA’s Cosmos world foundation models generate diverse datasets for robotics training. They enable realistic video and synthetic data generation, boosting both efficiency and scale.
With downloads already exceeding 3 million, the models are quickly becoming indispensable. Updates such as Cosmos Predict 2.5 and Cosmos Transfer 2.5 promise longer video simulations, faster results and improved data quality.
Adoption Across Industry and Academia
Leading companies and research institutions are adopting these technologies to push robotics further into real-world applications.
Industry Leaders Using NVIDIA Robotics Tools
| Organization | Technology Adopted | Focus Area |
|---|---|---|
| Boston Dynamics | Isaac Lab workflows | Humanoid grasping and mobility |
| Agility Robotics | Newton engine, Omniverse | Commercial humanoid development |
| Disney Research | Newton engine | Simulation in complex environments |
| Figure AI | Jetson Thor, Isaac GR00T | General-purpose humanoids |
| ETH Zurich | Newton engine | Robotic systems research |
Highlights
-
NVIDIA’s open-source Newton Physics Engine improves robot simulation accuracy and scalability.
-
Isaac GR00T foundation model enhances humanlike reasoning and task execution.
New AI Infrastructure for Robotics
NVIDIA is also delivering advanced AI infrastructure to support these innovations. The GB200 NVL72 rack-scale system integrates 72 Blackwell GPUs for training and inference at unprecedented speeds. RTX PRO Servers provide a unified environment for synthetic data generation, learning and deployment.
Meanwhile, the Jetson Thor module brings this power directly onto robots, enabling real-time inference and multi-AI workflows. Partners including Google DeepMind, Meta and Figure AI are already adopting Jetson Thor for humanoid development.
In Conclusion
NVIDIA’s latest robotics innovations mark a turning point in physical AI. By combining accurate simulation, humanlike reasoning and scalable training, the company is enabling robots to move from research labs into everyday life. With global adoption already underway, NVIDIA is positioning itself at the forefront of robotics transformation.
Sources: NVIDIA.
Prepared by Ivan Alexander Golden, Founder of THX News™, an independent news organization delivering timely insights from global official sources. Combines AI-analyzed research with human-edited accuracy and context.
