MuJoCo (Multi-Joint dynamics with Contact)

An open-source physics engine known for accurate contact simulation, maintained by Google DeepMind.

MuJoCo stands for Multi-Joint dynamics with Contact. It was published by Todorov and colleagues at IROS in 2012, and originally required a paid license. DeepMind acquired it in October 2021 and made it free, then open-sourced it under Apache 2.0 starting in 2022; it's now maintained by Google DeepMind. It solves for contact forces using convex optimization, giving a unique solution and cleanly defined inverse dynamics, which suits it well to control and reinforcement learning research. Robot models are described in MJCF, an XML format. Around it has grown an ecosystem including MJX, a JAX version that runs in parallel on GPU/TPU, and the MuJoCo Menagerie model library; robosuite, LIBERO, and the DeepMind Control Suite are all built on top of it.

ExampleThe LIBERO benchmark is built on robosuite, with the robot arm's and objects' physics computed by MuJoCo underneath.

Also called

Mujoco

Related

Physics Engine · MuJoCo XLA · MJCF (MuJoCo XML Format) · MuJoCo Menagerie · robosuite · DeepMind Control Suite

Sources

google-deepmind/mujoco (GitHub)
Google DeepMind: Opening up a physics simulator for robotics (2021-10)

As of

2026-09