NVIDIA, Google DeepMind, and Disney Launch Newton, an Open-Source Physics Engine
NVIDIA, Google DeepMind, and Disney Research have launched Newton, an open-source and highly extensible physics engine aimed at the future of robotic learning and development. Introduced at NVIDIA's GTC 2025 conference, Newton will make robots more capable of learning and executing complicated tasks with increased accuracy, and innovations that will influence several industries, ranging from entertainment to robotics powered by AI.
Open-Source Innovation for the Robotics Community
Newton is an open-source platform, which allows the world robotics community to use, extend, and give back to its continuous development. The open-access model encourages developers to freely share and build upon the framework, opening the possibilities for state-of-the-art research and experimentation in robotic simulations. By empowering roboticists to contribute to improving the engine, the initiative has the potential to speed up advances in the field of robotics, making the technology remain flexible and on the frontier.
Newton, which is developed on top of NVIDIA's Warp framework, provides a high-performance platform for robot simulations. NVIDIA Warp's CUDA-X acceleration allows for high-performance GPU-accelerated applications, making physics simulations for AI, robotics, and machine learning faster and more accurate.
Newton's integration with MuJoCo-Warp provides increased simulation performance with up to 70x speedup for humanoids and 100x speedup for in-hand activities. It also supports differentiable physics, allowing robots to learn and optimize real-time behavior.
Also read: NVIDIA Launches Dynamo to Boost AI Inference Performance
Expanding the Capabilities of Robotic Learning
Newton’s support for differentiable physics allows robots to optimize system parameters in real time by backpropagating gradients, enhancing their learning processes. Its extensibility enables diverse multiphysics simulations, where robots interact with materials like food, cloth, and deformable objects, making it a vital tool for research and real-world robotics. Custom solvers and methods can be integrated, allowing robots to handle various scenarios.
Newton is built on OpenUSD and streamlines robotic simulations by integrating data and assets, creating cohesive workflows. This unified approach allows developers to design specialized solvers and runtimes for advanced robotic systems. Disney Research plans to use Newton to enhance its next-generation entertainment robots, like the Star Wars-inspired BDX droids, to innovatively engage Disney theme park guests.
NVIDIA, Google DeepMind, Disney Research, and Intrinsic are collaborating on an OpenUSD asset structure to streamline robotics workflows and enhance data exchange, advancing next-gen robotics.
