AI Potluck
Contribute
Back to Gap Map Model components / Robotics & embodied AI

Newton

Newton Physics
open source / Overall score: 3.6

Newton is a GPU-accelerated, differentiable physics engine for robotics simulation, built on NVIDIA Warp with MuJoCo Warp as its primary backend and further solvers for articulated bodies, cloth, soft bodies and granular material. It loads URDF, MJCF and USD models and renders ray-traced camera images across many parallel worlds. Newton was initiated by Disney Research, Google DeepMind and NVIDIA and is a Linux Foundation project.

The PyPI package newton carries no project_urls, because the repository pyproject.toml declares none. It is this project: the package name, authors (Newton Developers), summary and license match the repository pyproject.toml, the long description is the repository README, and the documentation installs it with pip install newton. The first upload, 0.1.1a2 in August 2025, already carried the project README.

Openness

5 high confidence
5.0
license
Apache-2.0(LICENSE.md
source
public
core features withheld
no — the optional rtx extra pulls NVIDIA's proprietary OVRTX renderer, which Newton does not ship

Newton is Apache-2.0 code built on Apache-2.0 foundations, Warp and MuJoCo Warp, and has no paid edition. One optional install extra adds NVIDIA's proprietary Omniverse RTX renderer; the engine and its own ray-traced camera sensor work without it.

Adoption

3 medium confidence
3.0

PyPI installs of the newton package, which is the engine itself.

Capability

4 medium confidence
4.0

Newton has GPU-parallel worlds, ray-traced cameras across them and more physics backends than any engine here, from MuJoCo to cloth and granular solvers. What it ships for robots are examples, not a library of ready-to-train tasks, which is what separates Isaac Lab.

Verified 2026-09-27