NVIDIA Unveils Innovations in Robotics with Open Models and Simulation Libraries
Caroline Bishop Sep 29, 2025 16:11
NVIDIA introduces open-source robotics tools, including the Newton Physics Engine and Isaac GR00T N1.6 model, aimed at advancing robotics research and development.
NVIDIA has announced a series of advancements in robotics research and development, featuring the introduction of open-source models and simulation libraries. These innovations aim to enhance the capabilities of robots by providing researchers and developers with advanced tools to create more adaptable and capable machines. According to NVIDIA Newsroom, the new tools include the open-source Newton Physics Engine and the NVIDIA Isaac GR00T N1.6 reasoning vision language action model.
Newton Physics Engine
The Newton Physics Engine, developed in collaboration with Google DeepMind and Disney Research, is now available in the NVIDIA Isaac Lab. This GPU-accelerated physics engine is designed to help developers simulate complex robot actions, such as walking on uneven surfaces, and ensure these skills can be transferred safely from simulation to the real world. The engine is managed by the Linux Foundation and built on NVIDIA Warp and OpenUSD frameworks.
Esteemed institutions such as ETH Zurich and the Technical University of Munich are among the early adopters of the Newton Physics Engine, leveraging its capabilities to push the boundaries of robotic simulation and learning.
Isaac GR00T N1.6 Model
The NVIDIA Isaac GR00T N1.6 model introduces humanlike reasoning to robotics, enabling machines to interpret complex instructions and perform tasks using prior knowledge and common sense. This model, soon to be available on Hugging Face, integrates the NVIDIA Cosmos Reason model, enhancing the robots' ability to handle new situations and generalize across multiple tasks.
Isaac GR00T N1.6 allows humanoid robots to perform intricate maneuvers, such as opening heavy doors, by providing more freedom in torso and arm movements. Leading robotics companies, including Franka Robotics and LG Electronics, are evaluating this model for developing general-purpose robots.
Cosmos World Foundation Models
NVIDIA also announced updates to its Cosmos World Foundation Models (WFMs), which facilitate the generation of diverse data sets for training physical AI models at scale. These models are designed to streamline the development process, combining text, image, and video prompts to produce comprehensive training data efficiently.
Upcoming models like Cosmos Predict 2.5 and Cosmos Transfer 2.5 promise to enhance efficiency and reduce complexity, offering developers powerful tools to accelerate AI training and deployment.
AI Infrastructure for Robotics
To support these advancements, NVIDIA has unveiled new AI infrastructure, including the NVIDIA GB200 NVL72 and NVIDIA Jetson Thor. These systems are designed to handle demanding AI workloads, enabling real-time interactions and high-performance applications in robotics.
The NVIDIA GB200 NVL72, integrating 36 Grace CPUs and 72 Blackwell GPUs, is being adopted by major cloud providers to accelerate AI training and inference. Meanwhile, the NVIDIA Jetson Thor, powered by a Blackwell GPU, supports multi-AI workflows, enhancing real-time on-robot inference capabilities.
These developments underscore NVIDIA's commitment to advancing robotics research and development, providing researchers and developers with the tools necessary to push the boundaries of what is possible in the field of robotics.
Image source: Shutterstock.jpg)