MuJoCo stands for Multi-Joint dynamics with Contact. Written by Emo Todorov and sold as a paid library for years, it was acquired by Google DeepMind in October 2021, made free at once and fully open sourced under Apache 2.0 in 2022. Its reputation rests on a fast, stable contact model and on speed: thousands of parallel rollouts are routine, and the MJX variant runs on GPU and TPU through JAX. Models are described in compact MJCF XML and the Python bindings are clean.
This is the default physics for reinforcement learning work, policy training and biomechanics, not a factory planning tool. There is a viewer, but no scene editor in the CAD sense, no vendor robot library and no native ROS 2 stack, so integration falls on the user. Contact parameters reward tuning and punish carelessness: a policy trained on sloppy settings rarely survives transfer to hardware.
No reviews yet — be the first!