Core Functionality Unveiled
Newton 1.0, an open-source physics simulation engine, is now generally available, targeting industrial robotics tasks. The release highlights stable, unified APIs for modeling and solving simulations, a move towards simplifying the development workflow for robot training. It introduces capabilities for contact-rich manipulation and locomotion, essential for real-world robotic applications.
The engine’s architecture emphasizes flexibility, with a collision detection pipeline allowing users to choose appropriate broadphase and narrowphase detection methods based on the complexity of the simulated environment. Newton’s VBD solver is designed to transparently handle distributed contact patches and simulate linear deformable objects, such as cables.
This release positions Newton as a standalone framework, accessible via a modern Python API, and is now available on PyPI. The development team signals a commitment to iterative improvement with a monthly release cadence planned for the rollout of new features.
Read More: AI May Lower Value of University Degrees for Jobs in 2025
Integration and Community Engagement
Newton 1.0 integrates with Isaac Lab 3.0, NVIDIA's robot learning framework, enhancing its physics and camera-sensor backend capabilities. While integration with Isaac Sim as a physics backend is noted as "under development," the experimental integration in Isaac Lab currently does not support PhysX, though Isaac Lab plans to continue offering PhysX as a backend. Discussions around supporting PhysX as a solver option within Newton are ongoing.
The project actively encourages community involvement, welcoming contributions through its GitHub Discussions and Issues channels. Standalone examples and documentation are available in the newton-physics/newton GitHub repository, serving as integration tests to ensure stability.
Background and Partnerships
The launch of Newton 1.0 coincides with broader announcements from NVIDIA at GTC, including updates to their robotics initiatives. Toyota Research Institute has been involved in advancing the solver development and contact modeling aspects of Newton. This collaboration underscores the focus on sophisticated simulation for industrial robot training, aiming to improve the robots' ability to handle new tasks in varied environments. Performance of the engine is tracked using Airspeed Velocity (ASV) benchmarks, measuring simulation throughput and initialization overhead.
Read More: New AI helps soft robot arms learn new tasks without retraining
