Introducing Newton
Introducing Newton: The Open-Source Physics Engine Revolutionizing Robotics Simulation
March 18, 2025
Advanced physical AI models are transforming the robotics industry by enabling robots to perceive, interpret, and interact with the real world autonomously. High-performance computing and realistic simulations play a crucial role in shaping the future of robotics development.
The Role of Physics in Robotic Simulation
Physics-based simulations provide the foundation for accurately predicting robot behavior in various environments. These simulations allow researchers and engineers to design, test, and refine robotic control algorithms and prototype designs in a safe, efficient, and cost-effective manner.

Robotics simulations rely on core physics principles, including mass and momentum conservation, rigid and soft body dynamics, contact mechanics, and actuator modeling. These principles help predict how robots respond to diverse scenarios and conditions.
However, the gap between simulated and real-world performance, known as the “sim-to-real” gap, remains a major challenge. Robotics engineers need a scalable, adaptable, and physics-accurate simulation framework to bridge this gap, supporting complex behaviors and customized solvers.
Meet Newton: A Game-Changing Open-Source Physics Engine
Newton is an open-source, highly extensible physics engine developed in collaboration with NVIDIA, Google DeepMind, and Disney Research. Designed to enhance robotic learning and simulation, Newton integrates seamlessly with leading AI-driven robotics platforms.
Built on NVIDIA Warp, Newton empowers robots to handle intricate tasks with enhanced precision. It supports frameworks such as MuJoCo Playground and NVIDIA Isaac Lab—an open-source, unified robotics learning framework.
Key Features of Newton
1. Open-Source Accessibility
Newton’s open-source model enables developers worldwide to contribute to and benefit from its continuous innovation. This fosters collaboration and rapid advancements in robotics research and simulation.
2. Powered by NVIDIA Acceleration
Newton leverages NVIDIA Warp, a high-performance CUDA-X acceleration library. This allows developers to create GPU-accelerated, kernel-based programs for AI-driven simulation, robotics, and machine learning (ML). The parallel processing power of NVIDIA GPUs enhances simulation efficiency and accuracy.
3. Integration with MuJoCo-Warp
Newton integrates seamlessly with MuJoCo (Multi-Joint dynamics with Contact), a leading open-source physics engine widely used in robotics research. Additionally, Google DeepMind has introduced MuJoCo-Warp, an optimized version powered by Warp acceleration, offering up to 100x speed improvements in robotic manipulation tasks.
4. Differentiable Physics for Advanced Learning
Newton supports differentiable physics, enabling AI-driven robotic learning by computing gradients for optimization and simulation refinement. This enhances the development of robotic control strategies through forward and reverse-mode differentiation techniques.
5. Highly Extensible and Customizable
With increasing complexity in robotic applications, Newton offers extensive customization. It supports multiphysics simulations involving interactions with soft materials, fluids, and deformable objects. Additionally, it enables the integration of custom solvers, numerical methods, and physics-based models to suit diverse robotic applications.
6. Built on OpenUSD for Seamless Data Integration
Newton utilizes the OpenUSD framework, ensuring seamless aggregation of data for defining robots and their environments. This enhances compatibility across multiple robotics platforms and facilitates efficient simulation workflows.
Disney Research’s Vision for Newton
Disney Research is leveraging Newton to power its next-generation robotic character platform. Notably, the expressive Star Wars-inspired BDX droids showcased at NVIDIA’s GTC keynote are among the first to benefit from Newton’s capabilities.
Kyle Laughlin, Senior Vice President at Walt Disney Imagineering Research and Development, expressed excitement about the collaboration:
“The BDX droids are just the beginning. This partnership with NVIDIA and Google DeepMind will enable us to create highly expressive, interactive robotic characters that redefine entertainment and guest engagement.”
Future of Newton and Robotics Simulation
In addition to its core features, Newton is part of a larger collaboration between Disney Research, Google DeepMind, Intrinsic, and NVIDIA to establish a standardized OpenUSD asset structure for robotics. This initiative aims to unify robotic simulation workflows and data exchange across platforms.
What’s Next?
Newton is set to revolutionize robotic simulation, paving the way for more advanced and realistic humanoid robots. The first version is expected to be released later this year.
Stay updated on the latest advancements in Newton and explore related technologies:
- NVIDIA Warp – A powerful framework for spatial computing and AI-driven simulations in Python.
- NVIDIA Isaac Lab – An open-source, unified framework for training AI-powered robot policies on NVIDIA Isaac Sim.
Newton is just the beginning of a new era in robotics simulation—where AI-driven physics engines unlock new possibilities in robotic learning and automation.