google-deepmind/mujoco_warp

GPU-optimized version of the MuJoCo physics simulator, designed for NVIDIA hardware.

What it solves

MJWarp provides a high-throughput, GPU-accelerated version of the MuJoCo physics simulator. It solves the performance bottleneck of traditional physics simulations by enabling accurate, fast simulation of complex physical systems on NVIDIA hardware, which is critical for robotics research.

How it works

MJWarp leverages NVIDIA GPUs to accelerate physics calculations. It can be used as a drop-in replacement for MuJoCo or integrated into broader ecosystems like JAX (via MJX) or PyTorch (via Isaac Lab or mjlab). It also includes a high-throughput GPU batch renderer that uses ray-tracing to render millions of frames per second across multiple parallel simulation worlds.

Who it’s for

It is designed for robotics researchers and developers who need high-performance physical simulations for training or testing agents, specifically those using NVIDIA hardware.

Highlights

  • Broad Simulation Support: Handles rigid bodies, soft bodies, cloth, signed distance fields, and heightfields.
  • High-Throughput Rendering: A GPU batch renderer supporting mesh, texture, Flex deformable, and Gaussian splat rendering.
  • Flexible Integration: Compatible with JAX and PyTorch ecosystems.
  • Hardware Support: Optimized for NVIDIA GPUs while maintaining CPU support for development and debugging.

Related

  • Project
    google-deepmind/mujoco_playground

    A GPU-accelerated suite of robot learning environments built with MuJoCo MJX, designed for efficient reinforcement learning and sim-to-real research.

  • Project
    mujocolab/mjlab

    A lightweight, GPU-accelerated framework for robot learning that combines Isaac Lab's manager-based API with MuJoCo Warp.

  • Project
    newton-physics/newton

    Newton is a GPU-accelerated, differentiable physics simulation engine built on NVIDIA Warp and MuJoCo Warp, designed for scalable robotics research.

  • Project
    google-deepmind/mujoco

    A high-performance physics engine for the fast and accurate simulation of articulated structures, widely used in robotics and machine learning research.

  • Project
    robfiras/loco-mujoco

    An imitation learning benchmark for whole-body control featuring diverse humanoid and quadruped environments with extensive motion capture datasets.