Publication
MujocoROS2Control: Seamless MuJoCo Integration with ROS 2 for Robot Simulation and Control
Adrian Danzglock; Vamsi Krishna Origanti; Frank Kirchner
In: Daniel S. Katz (Hrsg.). Journal of Open Source Software (JOSS), Vol. 10, No. 116, Pages *-*, Journal of Open Source Software, Online, 12/2025.
Abstract
ROS 2 is an open-source framework that provides standardized communication, tools, and libraries for building robot applications. It uses a Data Distribution Service (DDS) backend for reliable, real-time communication and supports features such as zero-copy message passing when nodes run on the same system. ROS 2 is supported by a broad ecosystem of community and industry-maintained packages, ranging from sensor drivers to visualization tools and control modules. MuJoCo (Multi-Joint dynamics with Contact) is a high-performance physics engine designed for accurate simulation of articulated bodies, contacts, and constraints. Widely used in robotics and machine learning research, it emphasizes both simulation speed and physical fidelity. MuJoCo is implemented in C/C++ and also offers a JAX-based version optimized for GPU-accelerated reinforcement-learning workflows. The MujocoROS2Control hardware interface enables seamless integration between MuJoCo (Todorov et al., 2012), a high-performance physics engine, and ROS 2 (Open Robotics, 2023), widely adopted middleware for robotic systems. This interface provides an efficient solution for simulating and controlling robots using MuJoCo’s physics capabilities within the ROS 2 ecosystem. To support ROS-based workflows, we developed a dedicated URDF-to-MJCF conversion script. This tool translates URDF models into MJCF (MuJoCo XML format), preserving kinematic and dynamic properties and allowing custom MuJoCo-specific parameters such as sensors, actuators, and collision definitions to be specified directly in the URDF. This conversion ensures compatibility and adaptability for simulation. MujocoROS2Control bridges the gap between ROS 2 and MuJoCo, offering a streamlined workflow for simulation, controller testing, and reinforcement learning.