mjlab playground

A collection of tasks built with mjlab, starting with ports from MuJoCo Playground.

Tasks

Task ID	Robot	Description	Preview
Getup
Mjlab-Getup-Flat-Unitree-Go1	Unitree Go1	Fall recovery on flat terrain
Mjlab-Getup-Flat-Booster-T1	Booster T1	Fall recovery on flat terrain

Getting Started

git clone https://github.com/mujocolab/mjlab_playground.git && cd mjlab_playground
uv sync

Train a task:

uv run train <task-id> --num_envs 4096

Play back a trained policy:

uv run play <task-id>

Getup training

On a single NVIDIA 5090, the Go1 getup task converges in ~2 minutes and T1 in ~8 minutes, but we continue training with a curriculum that progressively tightens action rate, joint velocity, and power penalties to produce smoother, safer policies.

Citation

If you use this repository in your research, consider citing mjlab:

@misc{zakka2026mjlablightweightframeworkgpuaccelerated,
  title={mjlab: A Lightweight Framework for GPU-Accelerated Robot Learning},
  author={Kevin Zakka and Qiayuan Liao and Brent Yi and Louis Le Lay and Koushil Sreenath and Pieter Abbeel},
  year={2026},
  eprint={2601.22074},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2601.22074},
}

License

This repository is released under an Apache-2.0 License.