ttktjmt/mjswan

MuJoco Simulation on WebAssembly with Neural networks

Python

353

1,153 commits

updated Oct 12, 2026

See the code

README

mjswan

Real-time Interactive RL Simulation in Your Browser

deploy test docs pypi version npm version

mjswan is a powerful framework for creating interactive MuJoCo simulations with real-time policy control, running entirely in the browser. Built on top of mujoco wasm, onnxruntime, and three.js, it enables easy sharing of RL simulation demos as static sites, perfect for GitHub Pages hosting.

Check out the demo ― ttktjmt.github.io/mjswan


News

Features

  • Real-time: Run mujoco simulations and policy control in real time.
  • Interactive: Change the state of objects by applying forces.
  • Cross-platform: Works seamlessly on desktop and mobile devices.
  • VR & AR Support: Native WebXR viewer with tracked hands that interact with objects.
  • Client-only: All computation runs in the browser. No server for simulation is required.
  • Easy Sharing: Host as a static site for effortless demo distribution (e.g., GitHub Pages).
  • Portable: Embed the simulation in a web page or Google Colab notebook.
  • mjlab-native: Port any mjlab task to the web with the mjlab-to-mjswan skill.
  • Customizable: Visualize your mujoco models and onnx policies quickly.

Quick Start

mjswan can be installed with pip:

pip install mjswan  # or 'mjswan[dev]', 'mjswan[examples]'

or with npm:

npm install mjswan

The demos live in examples/, which the wheel does not ship, so run them with the mjswan CLI from a clone of this repository (after pip install -e '.[examples]'):

mjswan demo          # lists the bundled demos
mjswan demo simple   # runs the simple demo

The minimum python script for a sanity check:

import os, mujoco, mjswan

model_path = os.path.join(os.path.dirname(mujoco.__file__), "testdata", "model.xml")
mjspec = mujoco.MjSpec.from_file(model_path)

builder = mjswan.Builder()
builder.add_project(name="Sanity Check").add_scene(name="Test Model", spec=mjspec)
app = builder.build()
app.launch()

For detailed instructions, visit the documentation.

Acknowledgments

This project was greatly inspired by the Facet project demo from the research group at Tsinghua University.
It is also built upon the excellent work of zalo/mujoco_wasm, one of the earliest efforts to run MuJoCo simulations in a browser.

License

This project is licensed under the Apache-2.0 License. When using mjswan, please retain attribution notices in the app to help other users discover this project.

mjlab
mjswan
mujoco
mujoco-wasm
onnx
reinforcement-learning
robotics

ttktjmt/mjswan

MuJoco Simulation on WebAssembly with Neural networks

Python

353

1,153 commits

updated Oct 12, 2026

See the code

README

mjswan

Real-time Interactive RL Simulation in Your Browser

deploy test docs pypi version npm version

mjswan is a powerful framework for creating interactive MuJoCo simulations with real-time policy control, running entirely in the browser. Built on top of mujoco wasm, onnxruntime, and three.js, it enables easy sharing of RL simulation demos as static sites, perfect for GitHub Pages hosting.

Check out the demo ― ttktjmt.github.io/mjswan


News

Features

  • Real-time: Run mujoco simulations and policy control in real time.
  • Interactive: Change the state of objects by applying forces.
  • Cross-platform: Works seamlessly on desktop and mobile devices.
  • VR & AR Support: Native WebXR viewer with tracked hands that interact with objects.
  • Client-only: All computation runs in the browser. No server for simulation is required.
  • Easy Sharing: Host as a static site for effortless demo distribution (e.g., GitHub Pages).
  • Portable: Embed the simulation in a web page or Google Colab notebook.
  • mjlab-native: Port any mjlab task to the web with the mjlab-to-mjswan skill.
  • Customizable: Visualize your mujoco models and onnx policies quickly.

Quick Start

mjswan can be installed with pip:

pip install mjswan  # or 'mjswan[dev]', 'mjswan[examples]'

or with npm:

npm install mjswan

The demos live in examples/, which the wheel does not ship, so run them with the mjswan CLI from a clone of this repository (after pip install -e '.[examples]'):

mjswan demo          # lists the bundled demos
mjswan demo simple   # runs the simple demo

The minimum python script for a sanity check:

import os, mujoco, mjswan

model_path = os.path.join(os.path.dirname(mujoco.__file__), "testdata", "model.xml")
mjspec = mujoco.MjSpec.from_file(model_path)

builder = mjswan.Builder()
builder.add_project(name="Sanity Check").add_scene(name="Test Model", spec=mjspec)
app = builder.build()
app.launch()

For detailed instructions, visit the documentation.

Acknowledgments

This project was greatly inspired by the Facet project demo from the research group at Tsinghua University.
It is also built upon the excellent work of zalo/mujoco_wasm, one of the earliest efforts to run MuJoCo simulations in a browser.

License

This project is licensed under the Apache-2.0 License. When using mjswan, please retain attribution notices in the app to help other users discover this project.

mjlab
mjswan
mujoco
mujoco-wasm
onnx
reinforcement-learning
robotics

Project health

Latest release

0.11.6 · Oct 10, 2026

Releases this year

41

Significant stargazers

Silvio Traversaro

578 followers · starred Feb 2026

Kevin Zakka

1,821 followers · starred Jan 2026

João Matos

0 followers · starred Jan 2026

Michel Kuhlmann

35 followers · starred Jan 2026