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

MuJoCo

Google
open source / Overall score: 4.0(strong)

MuJoCo (Multi-Joint dynamics with Contact) is Google DeepMind's physics engine for articulated bodies in contact, used throughout robotics and reinforcement-learning research. The C core runs on the CPU with an OpenGL viewer and Python bindings, and the product line includes MJX, a JAX reimplementation, and MuJoCo Warp, a GPU build maintained with NVIDIA, for batched simulation on accelerators.

Openness

5 high confidence
5.0
license
Apache-2.0(core, MJX and MuJoCo Warp)
source
public
core features withheld
no — no paid edition

The engine and both accelerator backends are Apache-2.0 in their own repositories, and nothing is held back for a paid edition.

Adoption

4 medium confidence
4.0

PyPI installs of the mujoco package plus the separately published MuJoCo Warp package. A share of the core figure is pulled in as a dependency by other simulators and reinforcement-learning libraries rather than chosen directly.

Capability

4 medium confidence
4.0

MuJoCo is the reference contact physics engine, and its GPU build runs many worlds in parallel and renders their cameras in the same batch by ray tracing. What it does not carry is a second physics engine or a library of ready-to-train tasks for common robots; those come from projects built on it.

Verified 2026-09-26