# MuJoCo Playground [![Build](https://img.shields.io/github/actions/workflow/status/google-deepmind/mujoco_playground/ci.yml?branch=main)](https://github.com/google-deepmind/mujoco_playground/actions) [![PyPI version](https://img.shields.io/pypi/v/playground)](https://pypi.org/project/playground/) [![Documentation Status](https://img.shields.io/readthedocs/reaf-sim/latest)](https://reaf-sim.readthedocs.io/en/latest/playground/index.html) [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](LICENSE) ![Banner for playground](https://www.gstatic.com/reaf-sim/banner/0/banner.png) MuJoCo Playground is a suite of robot environments, reinforcement learning tasks, and hardware workflows written in [PyTorch](https://pytorch.org/) and built on [`reaf-sim`](https://github.com/google-deepmind/reaf-sim), [MuJoCo](https://github.com/google-deepmind/mujoco), and [MuJoCo Warp](https://github.com/google-deepmind/mujoco_warp). Robots exist in [`mujoco_playground/robots/`](mujoco_playground/robots), either in simulation or paired with a hardware counterpart, enabling end-to-end workflows across simulation and hardware: - **[Large-batch RL training on GPU](mujoco_playground/rl/rsl_rl)**: with state and vision observations across control, manipulation, and humanoid whole-body tracking. - **[Sim & real teleoperation](mujoco_playground/tasks/teleop/so101)**: with leader-follower control and episodic MCAP logging. - **[System identification](mujoco_playground/tasks/sysid)**: fitting simulated joint dynamics to logged hardware trajectories via `mujoco.sysid`. - **[Zero-shot sim-to-real policy deployment](mujoco_playground/tasks/manipulation/pick_cube/so101/deploy)**: running the same task layer and policy in simulation (`--mode=sim`) and on hardware (`--mode=real`). - **[Agentic manipulation](mujoco_playground/tasks/agentic/tabletop_v0)**: for vision-language model (VLM) examples across simulation and hardware (`SO-101`, `i2rt YAM`).
Whole body motion tracking
Whole body motion tracking
SO-101 Simulation Teleoperation Demo
Sim teleoperation
SO-101 Real Teleoperation Demo
Real teleoperation
SO-101 System Identification Rollout
System identification
Sim2Real and interactive viewer
Sim2Real and interactive viewer
YAM Agentic Tabletop Manipulation
Agentic manipulation
> [!NOTE] > Looking for the original RSS 2025 JAX/MJX environments? They remain available under [`rss2025`](rss2025). ## Installation First [install `uv`](https://docs.astral.sh/uv/getting-started/installation/) if you don't have it yet. Install [`playground`](https://pypi.org/project/playground/) (Python 3.12–3.13) from PyPI with `uv pip` (or skip installation and run directly with `uvx` in [Getting Started](#getting-started)): ```bash uv venv --python 3.13 && source .venv/bin/activate uv pip install --default-index https://pypi.org/simple "playground>=1.0" ``` > [!NOTE] > On Linux with CUDA 12 drivers, install with `--index https://download.pytorch.org/whl/cu129 -f https://github.com/NVIDIA/warp/releases/expanded_assets/v1.18.0 "warp-lang==1.18.0+cu12" "playground[rl]>=1.0"`. ## Getting Started To get started, try controlling a sim/real robot with your coding agent, or train an RL policy in sim and deploy it to hardware. ### Control a robot with your coding agent Spin up the [agentic tabletop manipulation task](mujoco_playground/tasks/agentic/tabletop_v0) directly from your coding agent with a prompt in `sim` or `real`: ```text Start the tabletop_v0 bridge for the yam in sim and get it to , following mujoco_playground/tasks/agentic/tabletop_v0/AGENT_PROMPT.md ``` Swap `sim` for `real` to run on hardware, or `so101` for `yam` (see [`AGENT_PROMPT.md`](mujoco_playground/tasks/agentic/tabletop_v0/AGENT_PROMPT.md)). > [!NOTE] > On macOS, `--viewer` requires `mjpython` in place of `python`. Before the > first run, link `libpython` into the virtualenv so `mjpython` can load it: > > ```bash > ln -sf "$(uv run python -c 'import sys; print(sys.base_prefix)')"/lib/libpython3.13.dylib .venv/lib/ > ``` ### Train, visualize, and deploy an RL policy Train an `SO101PickCube` PPO policy with [RSL-RL](https://github.com/leggedrobotics/rsl_rl) and inspect it in the MuJoCo Studio web viewer:
CPU ```bash # Train SO101PickCube on CPU uvx --default-index https://pypi.org/simple --python 3.13 --from "playground[rl]>=1.0" \ python -m mujoco_playground.rl.rsl_rl.train \ --env=SO101PickCube \ --logdir=/tmp/rslrl_logs/so101_pick_cube \ --overrides='{"nworld": 64}' # Play trained policy in MuJoCo Studio web viewer uvx --default-index https://pypi.org/simple --python 3.13 --from "playground[rl]>=1.0" \ python -m mujoco_playground.rl.rsl_rl.play \ --env=SO101PickCube \ --resume_path=/tmp/rslrl_logs/so101_pick_cube ```
GPU (CUDA 12) ```bash # Train SO101PickCube (CUDA 12 driver) uvx --default-index https://pypi.org/simple --python 3.13 \ --index https://download.pytorch.org/whl/cu129 \ -f https://github.com/NVIDIA/warp/releases/expanded_assets/v1.18.0 \ --with "warp-lang==1.18.0+cu12" \ --from "playground[rl]>=1.0" \ python -m mujoco_playground.rl.rsl_rl.train \ --env=SO101PickCube \ --logdir=/tmp/rslrl_logs/so101_pick_cube # Play trained policy in MuJoCo Studio web viewer uvx --default-index https://pypi.org/simple --python 3.13 \ --index https://download.pytorch.org/whl/cu129 \ -f https://github.com/NVIDIA/warp/releases/expanded_assets/v1.18.0 \ --with "warp-lang==1.18.0+cu12" \ --from "playground[rl]>=1.0" \ python -m mujoco_playground.rl.rsl_rl.play \ --env=SO101PickCube \ --resume_path=/tmp/rslrl_logs/so101_pick_cube ```
GPU (CUDA 13) ```bash # Train SO101PickCube (8,192 parallel envs on GPU) uvx --default-index https://pypi.org/simple --python 3.13 --from "playground[rl]>=1.0" \ python -m mujoco_playground.rl.rsl_rl.train \ --env=SO101PickCube \ --logdir=/tmp/rslrl_logs/so101_pick_cube # Play trained policy in MuJoCo Studio web viewer uvx --default-index https://pypi.org/simple --python 3.13 --from "playground[rl]>=1.0" \ python -m mujoco_playground.rl.rsl_rl.play \ --env=SO101PickCube \ --resume_path=/tmp/rslrl_logs/so101_pick_cube ```
Monitor training curves in TensorBoard: ```bash uvx --default-index https://pypi.org/simple tensorboard --logdir=/tmp/rslrl_logs/so101_pick_cube ``` Optionally [set up the SO-101 hardware](mujoco_playground/robots/so101/real), train the wrist-camera policy (`SO101PickCubeVision`), and [deploy zero-shot to the follower arm](mujoco_playground/tasks/manipulation/pick_cube/so101/deploy): ```bash # Train SO-101 Pick-Cube vision policy (add CUDA 12 flags above on CUDA 12) uvx --default-index https://pypi.org/simple --python 3.13 --from "playground[rl]>=1.0" \ python -m mujoco_playground.rl.rsl_rl.train \ --env=SO101PickCubeVision \ --logdir=/tmp/rslrl_logs/so101_pick_cube_vision # Deploy the trained policy on the physical SO-101 follower arm uvx --default-index https://pypi.org/simple --python 3.13 --from "playground[so101,rl]>=1.0" \ python -m mujoco_playground.tasks.manipulation.pick_cube.so101.deploy.run \ --mode=real \ --checkpoint=/tmp/rslrl_logs/so101_pick_cube_vision ``` ## Local Installation For local development from `HEAD`, clone the repository and sync dependencies:
CPU (macOS or Linux) ```bash git clone https://github.com/google-deepmind/mujoco_playground.git && cd mujoco_playground make sync-cpu ```
GPU (CUDA 12) ```bash git clone https://github.com/google-deepmind/mujoco_playground.git && cd mujoco_playground make sync-cu12 ```
GPU (CUDA 13) ```bash git clone https://github.com/google-deepmind/mujoco_playground.git && cd mujoco_playground make sync ```
Verify your installation by running the test suite: ```bash make test ``` > [!NOTE] > Hardware SDKs are optional extras so simulation-only stays lightweight. Passing `--extra so101` or `--extra yam` to `uv run` installs the hardware drivers on demand. ## Documentation & Contributing - **Documentation:** See the [MuJoCo Playground documentation](https://reaf-sim.readthedocs.io/en/latest/playground/index.html). - **Contributing:** See the [contributing guide](https://reaf-sim.readthedocs.io/en/latest/playground/contributing.html) for contribution guidelines. - **Changelog:** See the [changelog](https://reaf-sim.readthedocs.io/en/latest/playground/changelog.html). ## Citation If you use MuJoCo Playground in your work, please cite it as follows (see the [GitHub contributors page](https://github.com/google-deepmind/mujoco_playground/graphs/contributors?from=10%2F12%2F2026) for the full list of contributors): ```bibtex @software{mujoco_playground_2026, author = {Tabanpour, Baruch and Howell, Taylor and Haves, Sam and Abeyruwan, Saminda and Ma, Tony and Suh, Dan and {MuJoCo Playground Contributors}}, title = {{MuJoCo Playground}}, url = {https://github.com/google-deepmind/mujoco_playground}, year = {2026}, } ``` For the RSS 2025 benchmark suite that was built on JAX/MJX (see [`rss2025`](rss2025)), please cite: ```bibtex @misc{mujoco_playground_2025, title = {MuJoCo Playground: An open-source framework for GPU-accelerated robot learning and sim-to-real transfer.}, author = {Zakka, Kevin and Tabanpour, Baruch and Liao, Qiayuan and Haiderbhai, Mustafa and Holt, Samuel and Luo, Jing Yuan and Allshire, Arthur and Frey, Erik and Sreenath, Koushil and Kahrs, Lueder A. and Sferrazza, Carlo and Tassa, Yuval and Abbeel, Pieter}, year = {2025}, publisher = {GitHub}, url = {https://github.com/google-deepmind/mujoco_playground} } ``` ## License & Disclaimer All content in this repository is licensed under the Apache License, Version 2.0. See [LICENSE](LICENSE) for details. This is not an officially supported Google product.