MuJoCo, short for Multi-Joint dynamics with Contact, is a physics engine for fast, accurate simulation of articulated bodies that touch and collide. It was created by Emo Todorov and long distributed as paid commercial software, but it had become the simulator of choice for much of the reinforcement learning and robotics research community, used to benchmark continuous-control algorithms like DDPG and TRPO.
In October 2021 DeepMind announced it had acquired MuJoCo and would make it freely available to everyone. In May 2022 DeepMind followed through by open-sourcing the entire codebase on GitHub under a permissive license, committing to maintain it as a free, community-driven project. The blog post made the case that simulation is a foundational tool for robotics research and that a full-featured engine, backed by an established lab yet genuinely open, would accelerate the field.
The move mattered because the licensing cost of MuJoCo had been a real friction for academic groups and a hurdle to reproducible research, since results depended on software not everyone could legally run. Making it free and open removed that barrier and cemented MuJoCo as a standard environment for training and testing reinforcement learning agents that control simulated bodies.