mujoco-py is a Python 3 wrapper for MuJoCo, a physics engine designed for detailed and efficient rigid body simulations with contact dynamics. Developed and maintained by OpenAI's Robotics team, the library enables researchers and developers to access MuJoCo's simulation capabilities directly from Python, making it a foundational tool for robotics research, reinforcement learning, and physics-based simulation work.
The repository is written primarily in Cython, which allows it to efficiently bridge Python code with MuJoCo's underlying C++ implementation. The library supports Linux and macOS platforms with Python 3.6 and later versions. Historical support for Windows and Python 2 has been deprecated and removed in favor of maintaining compatibility with modern Python environments. The library has been updated to work with MuJoCo version 2.1, released in October 2021, though the README explicitly notes that mujoco-py does not support MuJoCo versions after 2.1.0.
According to GitGenius classification data, this repository spans multiple domains including control systems, physics simulation, neural networks, machine learning algorithms, robotics, reinforcement learning, trajectory optimization, and autonomous agents. The breadth of these categories reflects mujoco-py's role as infrastructure for diverse AI research applications, from training neural network policies to evaluating control algorithms and modeling dynamic systems.
The repository includes practical examples demonstrating advanced features such as body interactions, texture randomization for domain randomization, raw MuJoCo function calls, visualization markers, model serialization, state management, and robotic manipulation tasks. These examples serve as reference implementations for users building complex simulations.
Installation requires downloading MuJoCo 2.1 binaries separately and configuring environment variables, with platform-specific considerations documented for both Linux and macOS. The README provides troubleshooting guidance for common installation issues, including compiler compatibility problems on macOS with Apple's clang and GLFW library linking issues on Linux systems.
The project shows moderate but sustained activity in issue and pull request handling. Across 58 tracked items, the median response latency was approximately 20,402 hours with a mean of 22,733 hours, indicating that while the project receives attention, response times can be lengthy. The most active contributors tracked by GitGenius include sainavaneet with 4 events, followed by AuroraHashcat and Ianlande with 2 events each. The repository shares contributors with major projects including Microsoft's VSCode and TypeScript implementations as well as the Rust language project, suggesting that some contributors work across diverse technology stacks.
The project is now marked as deprecated in its README, with guidance directing new users toward the official MuJoCo Python bindings maintained by DeepMind. This transition reflects the shift toward official support for MuJoCo's Python interface, though mujoco-py remains a significant historical contribution to the robotics and reinforcement learning research communities.
