Welcome to WBC-MJLab!

WBC-MJLab is a training library for whole-body motion tracking on mjlab. One shared manager-based MDP hosts rewards, RSI, and motion commands; robots register as separate entities; paper and deploy choices are presets and tasks (--task Wbc-G1, Wbc-H2, …).

Try without training

Live web demo

Run a trained policy in the browser (MuJoCo WASM) with deploy-aligned clip switching.

https://wbc-mjlab.github.io/wbc-demo/

Google Colab

Cloud notebook with bundled checkpoint and sample clips — no local GPU required.

https://colab.research.google.com/github/wbc-mjlab/wbc-mjlab/blob/main/notebooks/demo.ipynb

Local demo

wbc-mjlab-demo on bundled samples after a quick make sync.

Demo & Colab

Key features:

• Modular stack — shared MDP in env/; each robot is a registered entity; presets compose tasks without forking core code

• Tasks, not forks — ZEST, BeyondMimic-style RSI, deploy obs, etc. as --task switches on the same CLI and log layout

• Multi-motion by design — train on clip libraries; one policy generalizes across skills at runtime

• Plug-in robots — extension packages via register_wbc_extension (see Extensions)

• Sim → real — train/play export policy.onnx + config.yaml for deploy runtimes

Try it (bundled samples) — full walkthrough: Quickstart: install → convert → train → play.

git clone https://github.com/wbc-mjlab/wbc-mjlab.git && cd wbc-mjlab
make sync
uv run wbc-mjlab-data-to-npz --robot g1 --dataset samples --batch-size 8
uv run wbc-mjlab-train --task Wbc-G1 --dataset samples

Not sure where to start? How do I…? (How do I…?).

Note

These docs track the main branch. The PyPI package is wbc-mjlab.

Table of Contents

Getting Started

• Installation
• Quickstart: install → convert → train → play
• How do I…?
• Demo & Colab
• Troubleshooting

Concepts

• Concepts
• Architecture overview

User Guide

• Usage
• Motion data
• Tasks and presets
• Shared MDP
• Training
• Deploy export
• Extensions
• Visualization (Viser)

API Reference

• API Reference

Development

• Contributing
• Roadmap

Further Reading

• Research & citations

License & citation

WBC-MJLab is licensed under the Apache License, Version 2.0. See the LICENSE file.

If you use WBC-MJLab in your research, please cite the software and the method papers for the tasks you reproduce (see Research & citations):

@software{wbc_mjlab2026,
  author  = {Nedelchev, Simeon and Chaplygin, Anton and Kozlov, Lev and Domrachev, Ivan},
  title   = {{WBC-MJLab}: Unified Whole-Body Motion Tracking on mjlab},
  url     = {https://github.com/wbc-mjlab/wbc-mjlab},
  year    = {2026},
  version = {0.0.4},
}

Also cite mjlab when using the simulation stack (see Research & citations).

Acknowledgments

WBC-MJLab builds on mjlab (manager-based RL API + MuJoCo Warp), with design inspiration from open WBC codebases such as whole_body_tracking and GR00T-WholeBodyControl — see Research & citations. Method implementations follow published WBC / motion-tracking work cited there; please cite those papers when comparing against or reproducing their setups.

Several contributors are affiliated with the Institute of Artificial Intelligence, the Robotics Center SBER, Innopolis University, and KAIST. We thank these groups for hosting and hardware used in training and deploy experiments.