It was recently revealed through a blog post that the Newton physics simulation engine, a powerful tool jointly developed by NVIDIA, Disney Research and Google DeepMind, has taken a significant step by being transferred to the Linux Foundation.
This strategic move seeks to ensure its growth as an open project, independent and collaborative, free from the exclusive control of large technology corporations.
With this transition, Newton becomes an accessible platform for researchers, developers, and companies around the world. New participants already on board include Lightwheel, Style3D, and academic experts from the University of Munich and Peking University.
Advanced and scalable simulation for the robotics of the future
For those who still don't know about Newton, you should know that this project It was designed to offer fast, accurate and scalable physics simulation., especially oriented towards robotics research.
Its engine allows model complex behaviors such as walking on deformable surfaces, interacting with fragile objects, or precisely manipulating realistic environments.
One of the main strengths of the project lies in its ability to harness the power of GPUs, which speeds up calculations and enables more realistic simulations in less time. Furthermore, its modular architecture facilitates the immediate integration of new components or custom algorithms.
“Newton’s addition to the Linux Foundation represents a significant step forward in scaling collaborative robotics simulation, accelerating development, reducing costs, and bringing us closer to the future of real-world robot simulation,” said Jim Zemlin, Executive Director of the Linux Foundation. “We are pleased to welcome Newton and provide the neutral governance its global community needs to build the future of general-purpose robotics.”
Key features
- GPU Accelerated: Leverage NVIDIA Warp for fast, scalable simulation.
- Multi-solver implementations: XPBD, VBD, MuJoCo, Featherstone, Euler
- Modular design: easily expandable with new solvers and components
- Differentiable: Supports differentiable simulation for machine learning and optimization.
- Enriched import/export: Load models from URDF, MJCF, USD and more.
- Open Source: maintained by Disney Research, Google DeepMind and NVIDIA.
Key Technologies: NVIDIA Warp and OpenUSD
To optimize spatial modeling and computing using GPUs, Newton uses the NVIDIA Warp framework, designed for performance-intensive simulation tasks. In turn, it uses the OpenUSD platform (Universal Scene Description) for the structured representation of the hierarchical data that make up each graphic scene.
This technological combination ensures an efficient workflow, with interoperability between different modeling and rendering tools, expanding the possibilities for use in sectors such as animation, augmented reality, robotics, and scientific research.
Multiple backends and differentiable simulation
Newton is not limited to a single method of resolution. Provides support for various physics backends or solvers, including Euler, Featherstone, ImplicitMPM, SemiImplicit, Style3D, VBD, and XPBD. Its main backend is based on MuJoCo, recognized for its accuracy in simulating multi-joint contact dynamics.
The engine supports differentiable simulation, an advanced feature that allows you to calculate derivatives and apply gradient methods. This It is essential to optimize parameters and adjust physical models. or train artificial intelligence systems and autonomous robots in virtual environments. Newton even offers real-time visualization, allowing you to observe the model's behavior as variables are adjusted.
It's worth mentioning that Newton's move to the Linux Foundation represents much more than an administrative change: it's an opening toward shared innovation. With its open source code base and expanding community, this engine promises to become an essential tool for modern robotics, machine learning, and advanced physics simulation.
Finally, it is worth mentioning that for those interested in this physics engine, they should know that The engine code, written in Python, is distributed under the Apache 2.0 license., which facilitates its adoption in both academic and industrial environments. They can also follow the installation instructions provided in the following link.
If you are interested in learning more about it, you can check the details in the following link.