Repolode
Submit
google-deepmind

google-deepmind/mujoco_warp

GPU-accelerated version of the MuJoCo physics simulator, built on NVIDIA Warp for high-throughput robotics simulation.

View on GitHub
Stars
1,609
Forks
271
Open issues
104
Last commit
yesterday

What is it?

MuJoCo Warp (MJWarp) is a GPU-accelerated port of the MuJoCo physics simulator, designed for NVIDIA hardware. It targets robotics research workloads that need many simulation steps run in parallel, and it can also run on CPU for development and debugging.

The project covers rigid bodies with contacts, along with soft bodies, cloth, and signed distance fields. It is maintained by Google DeepMind and NVIDIA as part of the Newton physics project. Most existing MuJoCo setups can use it as a drop-in replacement, and it can be installed from PyPI as mujoco-warp.

MJWarp also includes a batch ray-tracing renderer for cameras across many parallel worlds, supporting meshes, textures, heightfields, Gaussian splats, and shadows. Some MuJoCo features are not yet supported, including the PGS solver, PLUGIN actuators and sensors, and differentiability through Warp.

Key features

  • 01Runs MuJoCo-compatible physics simulation on NVIDIA GPUs with CPU fallback for development.
  • 02Simulates rigid bodies with contacts, plus soft bodies, cloth, and signed distance fields.
  • 03Includes a batch ray-tracing renderer for many parallel cameras, with meshes, textures, and shadows.
  • 04Installable from PyPI and usable as a drop-in replacement for MuJoCo in many projects.
  • 05Provides a CLI benchmark tool with an event trace option for performance profiling.

Use cases

  • Running large numbers of parallel robot simulations for reinforcement learning on NVIDIA GPUs.
  • Rendering camera observations across many simulated environments for vision-based robot learning.
  • Speeding up existing MuJoCo-based robotics research pipelines with minimal code changes.

Quick start

  1. 1

    Clone the repository

    git clone https://github.com/google-deepmind/mujoco_warp.git && cd mujoco_warp
  2. 2

    Run the humanoid robot viewer locally

    python benchmarks/run.py -f unitree_g1_flat --view
  3. 3

    Or install the package from PyPI

    pip install mujoco-warp

Things to know

  • Requires an NVIDIA GPU for fast simulation; CPU is only supported for development and debugging.
  • Some MuJoCo features are unsupported, including the IMPLICITFAST midpoint integrator, PGS solver, and PLUGIN actuators and sensors.
  • Differentiability via Warp is not yet available.

In collections

Run and train AI on GPUsReinforcement learning frameworks & training environmentsRobotics & embodied AI projects

Topics

#physics-simulation#robotics#gpu#nvidia-warp#reinforcement-learning#mujoco#ray-tracing#mujoco-warp

FAQ

What is mujoco_warp?
GPU-accelerated version of the MuJoCo physics simulator, built on NVIDIA Warp for high-throughput robotics simulation. MuJoCo Warp (MJWarp) is a GPU-accelerated port of the MuJoCo physics simulator, designed for NVIDIA hardware. It targets robotics research workloads that need many simulation steps run in parallel, and it can also run on CPU for development and debugging.
Is mujoco_warp free to use? What license does it have?
mujoco_warp is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is mujoco_warp still maintained?
Actively maintained. The last commit to mujoco_warp was on Nov 6, 2026.
What can I use mujoco_warp for?
Running large numbers of parallel robot simulations for reinforcement learning on NVIDIA GPUs. Rendering camera observations across many simulated environments for vision-based robot learning. Speeding up existing MuJoCo-based robotics research pipelines with minimal code changes.
How do I install mujoco_warp?
Clone the repository: git clone https://github.com/google-deepmind/mujoco_warp.git && cd mujoco_warp Run the humanoid robot viewer locally: python benchmarks/run.py -f unitree_g1_flat --view Or install the package from PyPI: pip install mujoco-warp
Where can I get mujoco_warp?
The source code and installation instructions are on GitHub: https://github.com/google-deepmind/mujoco_warp

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects