Newton is a GPU-accelerated physics simulation engine built upon NVIDIA Warp that targets roboticists and simulation researchers.
Newton addresses the need for high-performance physics simulation by extending Warp's deprecated simulation module and integrating MuJoCo Warp as its primary backend. The engine emphasizes GPU-based computation to enable scalable simulations, incorporates OpenUSD support for interoperability with 3D workflows, and provides differentiability to support gradient-based optimization and learning. Users can extend the engine with custom functionality, allowing rapid iteration on robotics problems without being constrained by a fixed feature set.
Roboticists and simulation researchers working on problems that benefit from GPU acceleration should consider Newton, particularly those already invested in the NVIDIA ecosystem or requiring differentiable simulation for learning-based approaches. The tool suits projects that need to scale beyond CPU-based simulation or integrate with OpenUSD pipelines. The engine runs on Linux and Windows with NVIDIA GPUs, and on macOS with CPU-only execution, requiring Python 3.10 or later and an NVIDIA driver version 545 or newer for GPU support.
The project maintains active engagement with its community through documented examples and comprehensive installation and compatibility guides. Development follows clear versioning and deprecation policies to help users plan upgrades. The codebase is organized to support extensibility, with users able to define custom physics behaviors and integrate domain-specific logic.
