MuJoCo Playground

MuJoCo Playground 1.0 is an open-source suite of sim and real environments built natively on MuJoCo, MuJoCo Warp, and reaf-sim. Plug-and-play examples include large-batch GPU RL training, sim and real teleoperation, system identification, zero-shot sim-to-real deployment, and VLM-guided agentic manipulation.

Playground decouples hardware and sensor I/O (DeviceLayer) from plain-PyTorch task logic (TaskLayer), allowing researchers to move seamlessly between large-batch GPU simulation, interactive CPU simulation in MuJoCo Studio, and physical robot hardware.

code docs reaf-sim discussions pypi
@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}, }

Quickstart

Train and visualize a Cartpole policy directly with uvx. 

# Train CartpoleBalance (~1 min 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=CartpoleBalance \
  --logdir=/tmp/rslrl_logs/cartpole_balance \
  --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=CartpoleBalance --resume_path=/tmp/rslrl_logs/cartpole_balance
# Train CartpoleBalance (CUDA 12 driver)
uvx --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=CartpoleBalance \
  --logdir=/tmp/rslrl_logs/cartpole_balance

# Play trained policy in MuJoCo Studio web viewer
uvx --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=CartpoleBalance \
  --resume_path=/tmp/rslrl_logs/cartpole_balance
# Train CartpoleBalance (4,096 parallel envs on GPU)
uvx --python 3.13 --from "playground[rl]>=1.0" \
  python -m mujoco_playground.rl.rsl_rl.train \
  --env=CartpoleBalance \
  --logdir=/tmp/rslrl_logs/cartpole_balance

# Play trained policy in MuJoCo Studio web viewer
uvx --python 3.13 --from "playground[rl]>=1.0" \
  python -m mujoco_playground.rl.rsl_rl.play \
  --env=CartpoleBalance \
  --resume_path=/tmp/rslrl_logs/cartpole_balance

Robotic Workflows

End-to-end examples connecting simulation and real robot hardware under a shared interface.

Whole-body motion tracking

Whole-Body Motion Tracking

Large-batch GPU RL training from state and pixels with RSL-RL across control, manipulation, and humanoid motion tracking.

SO-101 system identification

System Identification

Fitting simulated joint dynamics to logged hardware trajectories via mujoco.sysid.

SO-101 simulation teleoperation sim SO-101 real teleoperation real

Sim & Real Teleoperation

Interactive leader-follower teleoperation in MuJoCo Studio and on physical hardware with foot-pedal episode gating and episodic MCAP trajectory logging.

Zero-shot sim-to-real deployment

Zero-Shot Sim-to-Real Deployment

Running the same task layer and trained policy in simulation (--mode=sim) and on hardware (--mode=real).

Agentic tabletop manipulation

Agentic Manipulation

Vision-language model (VLM) tabletop control across simulation and physical arms with zero switching cost.

Robots

Robots in mujoco_playground/robots/ share canonical joint ordering and control interfaces across simulation and physical hardware.

Robot Modes Description Links
SO-101 simreal 5-DoF + gripper follower and leader arms with wrist camera Live model · Hardware setup
i2rt YAM Pro simreal 6-DoF + gripper manipulation arm with wrist camera Live model · Hardware setup
Trossen WidowX AI sim 6-DoF + gripper tabletop manipulation arm Live model
Unitree G1 sim 29-DoF humanoid robot for whole-body motion tracking Live model