Microduck ROS 2 integration brings MuJoCo simulation to Pollen Robotics biped

A new open-source project has been released to integrate the Microduck biped robot into the ROS 2 ecosystem. This integration enables the execution of existing Microduck policies within the MuJoCo physics engine, providing a high-fidelity simulation environment for the Pollen Robotics hardware.

The software exposes the simulated robot through standard ROS 2 interfaces, including velocity commands, joint states, and IMU data. By bridging MuJoCo with ROS 2, developers can now use familiar robotics tools and workflows to test locomotion, perception, and higher-level autonomy on a simulated bipedal platform.

Technical implementation and ROS 2 features

• Simulation Fidelity: The integration utilizes the BAM XL330 actuator model within MuJoCo and validates observations against existing policies with a strict error tolerance of 1e-6.
• Standardized Interfaces: The system provides several ROS 2 topics and transforms, including /cmd_vel for velocity control, joint states, IMU data, TF/URDF support, and a head camera feed compatible with RViz visualization.
• Skill and Manifest Management: Pre-defined robot skills such as kicks, roulade, and sit/stand are exposed as ROS 2 services, adhering to the robot's original manifest and robotd behavior.
• Policy Compatibility: The framework is designed to run existing microduck-policies without requiring modifications to the original policy codebases.

Why it matters

This development provides a bridge for researchers to apply ROS 2 navigation and perception stacks, such as Nav2, to legged robotics in a simulated environment. It lowers the barrier for testing complex locomotion and LLM-based voice control before deploying to physical Pollen Robotics hardware.