Open Source RoboticsRobots, software & learning

MuJoCo

Software

Physics engine for articulated mechanisms and contact, with C/Python interfaces and a version-sensitive model and simulation workflow. [1]

Installation & first result →Official websiteGetting started

For
Robotics researchers and developers who can specify and validate a physical model [1]
Source rights
Named licenses are documented by part. Review licenses
Unknowns
compute requirements not fully established. Review gaps

Source researched · Not tested by us
Sources checked · Research details

Licenses & related hardware
Robots
Asimov v0, Open-source Leader ArmTagged in our catalog; tags do not establish tested compatibility.

Requirements · Research details and dates

On this page

Overview

MuJoCo is a physics engine for articulated mechanisms interacting through contact. The reviewed 3.14.0 source provides a C API, native Python bindings and interactive visualization. It is useful for developing a model and control experiment, but a simulation result does not establish that the corresponding physical robot will behave the same way. [1] [2] [5]

Is this for you?

Workflows
  • Load or author a model and inspect it in the viewer
  • Advance dynamics through C or Python APIs
  • Develop control, system-identification or learning experiments

[1] [5] [6]

Known limitations
  • Numerical reproducibility across versions is not guaranteed
  • MJB binaries are version-specific; preserve model source
  • macOS passive Python viewer needs the mjpython launcher
  • Model accuracy and real-world transfer require independent validation

First three reflect documentation; physical-transfer limitation is editorial qualification.

[5] [6] [7] · Editorial assessment

What you need

Operating system
  • Linux x86-64 and AArch64 release binaries
  • Windows x86-64 release binaries
  • macOS universal release binaries

[1]

Languages / runtimes
  • C/C++
  • Python

[1] [6]

Dependencies
  • Python 3.10 or newer for Python bindings
  • OpenGL context for interactive visualization

[1] [5]

Installation & first result

Open the upstream getting-started guide [6] · Project documentation [1] [4]

  • Choose the reviewed versioned package or binaries
  • Open a small model in the documented viewer
  • Inspect model geometry and dynamics before adding control logic
  • Record simulator version and validate results again after upgrades

Editorial sequence; installation and simulation not performed.

[1] [5] [6] [7] · Editorial assessment

Installation

Choose a versioned binary release or install the mujoco Python package, which includes the engine; match the package version to the model and experiment being reproduced.

Editorial installation choice based on upstream documented alternatives and versioning.

[1] [6] [7] · Editorial assessment

Hardware, models & data

Interfaces
  • C API
  • Native Python bindings
  • Interactive simulate and Python viewer

[1] [6]

Inputs and outputs

MJCF/URDF model definitions compile into mjModel; mjData holds evolving state and intermediate results.

[5]

File formats
  • MJCF XML
  • URDF XML input
  • MJB compiled binary (version-specific)

[5]

Project notes

Begin with a small model and the viewer

The release README offers versioned binaries for Linux x86-64/AArch64, Windows x86-64 and universal macOS. The Python route requires Python 3.10 or newer; the mujoco package includes the engine library, so a separate engine download is unnecessary for that route. The documented python -m mujoco.viewer command opens a viewer that can load a model. Interactive visualization also needs a working OpenGL rendering context; this review does not establish a particular GPU requirement. These are upstream setup instructions, not an installation tested by this publication. [1] [5] [6]

The native modeling language is MJCF XML, and URDF files can also be loaded. A model compiles into mjModel; mjData stores evolving state and intermediate results. Keep the editable model source alongside its assets instead of treating a compiled MJB file as a portable interchange format: MJB files are version-specific. [5]

Python has both a blocking managed viewer and a passive viewer that lets the user's code continue. In the passive route, the script is responsible for advancing the simulation and synchronizing viewer state. On macOS that route requires the provided mjpython launcher. This distinction matters when moving from visual inspection to a custom control loop. [6]

An upgrade is an experiment change

MuJoCo's versioning document uses a custom scheme from 3.5.0 onward: the middle version component may introduce breaking changes, while the final component can add features or fixes. It explicitly does not guarantee numerical reproducibility across versions. Record the engine version with the model and rerun the experiment's own checks after an upgrade; a package that still imports is not evidence that a trajectory remains equivalent. [7]

The engine can support control synthesis, system identification and learning experiments, but the usefulness of those results depends on the selected model and validation procedure. No performance benchmark, contact-parameter calibration or transfer to hardware was performed in this review. [5]

What we checked

Source research only; no installation, physical build or performance test performed. No affiliate links added.

Not yet established

  • Requirements: host computer
  • Version-specific hardware compatibility (ecosystem tags alone do not establish it)

Updates

Upstream last updated
Sep 22, 2026 [2]
Latest release
Sep 22, 2026 [2]
Profile updated
Oct 4, 2026
Sources last checked
Sep 30, 2026
Next review
Oct 30, 2026
Reviewed revision
9ecbb9d7b5ee623f54745638d36799ff90e6f7cd (release 3.14.0) · software repository

Source files and licenses

Open robotics-relevant simulation engine with pinned source and documented setup; no hardware qualification implied.

LayerAvailability / license
softwareavailable · Apache-2.0
MuJoCo source-code license. External models and dependencies retain separate terms. [1] [3]
documentationavailable · CC-BY-4.0
README explicitly assigns doc-directory reStructuredText, images and videos to CC-BY-4.0. No media reused. [1]

Engine profile does not select a learned policy or training dataset. Physics model files are not model weights.

Separate engine, documentation and model rights

Source code is Apache-2.0. The README separately assigns reStructuredText documents, images and videos in the documentation directory to CC-BY-4.0. Those declarations do not automatically license an independently acquired robot mesh, dataset or trained policy. This profile selects the engine and its documentation, not a complete learning workload or robot asset collection. [1] [3]

References

Each source links to the exact revision we read. “Used for” shows where this page relies on it.

Project sources

Videos

  • S008 MuJoCo simulate tutorial (YouTube video)
    Video uploaded by Mohammad Hamid, linked from the project’s own documentation. Not reviewed by us for accuracy.
    Revision, date & where cited

    YouTube video Mohammad Hamid · youtube.com · Accessed Oct 4, 2026

  • S009 MuJoCo: touch grid visualisation (YouTube video)
    Video uploaded by Yuval Tassa, linked from the project’s own documentation. Not reviewed by us for accuracy.
    Revision, date & where cited

    YouTube video Yuval Tassa · youtube.com · Accessed Oct 4, 2026

  • S010 Demo of MuJoCo's built-in `simulate` viewer (YouTube video)
    Video uploaded by Yuval Tassa, linked from the project’s own documentation. Not reviewed by us for accuracy.
    Revision, date & where cited

    YouTube video Yuval Tassa · youtube.com · Accessed Oct 4, 2026

Correction guidance · Discussion is not enabled · Profile format 0.6.0-draft.1

Change history

2026-10-04 — Added 3 YouTube reference(s) (S008, S009, S010): official channel or videos linked from the pinned upstream sources. Existing claims, source-check dates and evidence are unchanged.