NVIDIA Unveils Newton: A Groundbreaking Open-Source Physics Engine for Robotics Simulation

Technology company NVIDIA NVIDIA has rolled out Newton, an innovative open-source physics engine, developed in partnership with Google DeepMind and Disney Research, aimed at improving robot learning and development.

Drawing from the NVIDIA Warp framework—which enhances robots' ability to master complex tasks with greater precision—Newton seamlessly integrates with frameworks like MuJoCo Playground and NVIDIA Isaac Lab, a comprehensive platform dedicated to unified robot learning.

The development of physical AI models empowers robots to autonomously perceive, analyze, reason, and interact with their environments. The future of robotics significantly depends on advanced computing and simulation technologies necessary for building next-generation robotic systems.

Physics plays a pivotal role in the realm of robotic simulation; it lays the foundation for accurate virtual representations of how robots function and interact within real-life settings. These simulation tools enable researchers and engineers to effectively train, design, assess, and validate control algorithms and prototypes in a safe, efficient, and cost-effective environment.

Newton is tailored to serve the entire robotics community, enabling roboticists to access, share, and collaborate on its development through research initiatives. Built atop NVIDIA Warp—a CUDA-X acceleration framework—it provides developers with a powerful means of creating highly efficient GPU-accelerated, kernel-based applications for simulation, AI, robotics, and machine learning (ML). This framework substantially boosts performance for running physics-driven simulations, harnessing the vast parallel processing capabilities of GPUs. NVIDIA GPUs .

A key feature of Newton lies in its compatibility with Multi-Joint dynamics with Contact (MuJoCo), a well-established open-source physics engine utilized in robotics research to model intricate dynamics in contact-heavy environments. This feature enables developers to build on existing models and code, significantly minimizing the time and resources needed to adapt applications for various physics engines.

Additionally, Google DeepMind has launched MuJoCo-Warp, a cutting-edge open-source robotics simulator accelerated by NVIDIA Warp, achieving astounding performance enhancements—with more than 70x speed improvements for humanoid simulations and up to 100x faster in-hand manipulation tasks. MuJoCo-Warp will serve as a primary physics engine within Newton, offering developers superior performance and versatility for their robotics projects.

Highlighted Features of Newton Include: Differentiable Physics, Extensibility, and OpenUSD Integration

Moreover, the capacity to propagate gradients through simulation opens new avenues for robotics simulation and learning. Differentiable simulators excel in generating results using forward-mode, while simultaneously calculating reverse-mode gradients of simulation outcomes that can facilitate back-propagation for system parameter optimization.

As the robotics sector continues to evolve, so too does the need to simulate an array of increasingly complex scenarios. Newton is crafted to be exceptionally adaptive, supporting diverse multiphysics simulations where robots engage with various materials, such as food, fabric, and other malleable objects. This versatility is achieved through custom solvers, integrators, and numerical methodologies. Newton also facilitates the coupling of diverse types of solvers, validating its capabilities through the integration of a material point method (MPM) solver with rigid body dynamics for simulating interactions with sand.

Furthermore, Newton utilizes the OpenUSD framework, providing a flexible data model and composition engine that compiles the essential data needed to accurately represent robots and their environments. Custom solvers and runtimes can be tailored to enhance specific robotic abilities within varying environments. Alongside partners such as Disney Research, Google DeepMind, and Intrinsic, Newton is contributing to the creation of an OpenUSD asset architecture for robotics. This initiative aims to standardize robot workflows by implementing best practices from OpenUSD, thus fostering a coherent data pipeline that establishes a common language across all robotics data sources. NVIDIA Please be advised that the information available on this page should not be construed as legal, tax, investment, financial, or any other form of advice. It's crucial to invest only what you can afford to lose and to consult an independent financial advisor if you have any uncertainties. For further insight, we recommend reviewing the terms and conditions and the help and support sections provided by the issuer or advertiser. MetaversePost is dedicated to delivering accurate and unbiased reporting; however, market conditions may shift without prior notice.