google-deepmind/mujoco_menagerie

A collection of high-quality models for the MuJoCo physics engine, curated by Google DeepMind.

MuJoCo Menagerie – High‑quality robot models for the MuJoCo physics engine

What it is – A curated library of MJCF (MuJoCo XML) model files for a wide variety of robots and biomechanical systems (humanoids, quadrupeds, arms, drones, etc.). The models are authored and maintained by Google DeepMind and are packaged as a Python library (mujoco‑menagerie).

Why it matters – In simulation, the fidelity of a robot’s model determines how useful the results are for research in control, reinforcement learning, robotics, and embodied AI. Menagerie supplies ready‑to‑use, well‑documented models that have been checked to load and run correctly in MuJoCo, saving users the time and error‑prone work of building their own XML files.

How to use it

  • Install the package: pip install mujoco-menagerie.
  • Load a model in Python:
    import mujoco_menagerie as mm
model = mm.load('unitree_go2')   # downloads and compiles the scene
spec = mm.get('unitree_go2').spec()  # editable MuJoCo MjSpec object
  • View a model from the command line without installing anything: uvx mujoco-menagerie view unitree_go2.
  • Or run the built‑in viewer directly from the repo: python -m mujoco.viewer --mjcf mujoco_menagerie/unitree_go2/scene.xml.

What’s inside – Over 40 robot definitions, each in its own folder containing:

  • assets/ with mesh files for visual and collision geometry
  • *.xml (MJCF) describing the robot’s kinematic tree
  • scene.xml that adds a ground plane, lights, etc.
  • Optional MJX‑compatible XML files for the newer MuJoCo‑X backend
  • A PNG preview and a README with generation details
  • License information (most are BSD‑3‑Clause, Apache‑2.0, or MIT)

Quality grading – Models are assigned a provisional grade (A+, A, B, C) indicating how closely the parameters match the real hardware. The grading system will be refined once a systematic identification toolbox is released.

Community – Contributions are encouraged. The repo includes a make install step that sets up pre‑commit hooks, and make all runs the same checks CI performs (lint, formatting, license compliance, XML validation, tests).

When to reach for Menagerie – If you need a reliable robot model for MuJoCo‑based research (e.g., reinforcement‑learning agents, trajectory optimization, sim‑to‑real transfer), Menagerie gives you a plug‑and‑play starting point with minimal setup.

Key links

Related

  • Project
    ARISE-Initiative/robosuite

    A MuJoCo-powered simulation framework and benchmark suite for robot learning, providing modular tools to create environments and test AI control algorithms.

  • Project
    kevinzakka/mjbatch

    mjbatch is a Python library that runs thousands of MuJoCo physics simulations in parallel on CPU using a C++ thread pool. It provides live NumPy‑style bindings to simulation state, per‑simulation model parameter variation, and is suited for RL, MPC, system identification, and robot co‑design. Install via pip; examples include cart‑pole swing‑up, Go1 quadruped control, and arm co‑design.

  • Project
    orcahand/orcahand_description

    A collection of URDF and MJCF description files for the Orcahand robotic hand, enabling accurate simulation and visualization in tools like MuJoCo.

  • Project
    google-deepmind/mujoco

    A high-performance physics engine for the fast and accurate simulation of articulated structures, widely used in robotics and machine learning research.

  • Project
    Farama-Foundation/Gymnasium-Robotics

    A collection of Reinforcement Learning robotic environments powered by the MuJoCo physics engine and the Gymnasium API for training AI agents in manipulation and navigation tasks.