pollen-robotics/microduck

A Tiny biped duck robot 🦆

What it solves

Microduck provides the "brain" and control software for a small bipedal robot, enabling it to perform complex physical movements like walking, standing up after being knocked over, and picking up objects using reinforcement learning (RL) policies.

How it works

The system runs on a Rockchip RK3566 and uses a 50 Hz control loop to drive fifteen servos. It loads neural policies exported as ONNX files, which are trained using MuJoCo and PPO (Proximal Policy Optimization) with sim2real techniques. The software architecture consists of several Rust-based daemons (such as robotd for control and updaterd for safe updates) that communicate via JSON-RPC over Unix sockets.

Who it’s for

It is designed for robot owners, developers building on the platform, and researchers interested in sim2real reinforcement learning for bipedal locomotion.

Highlights

  • RL-Driven Movement: Uses reinforcement learning policies for walking, rolling, and self-righting.
  • Multi-Mode Locomotion: Supports both bipedal walking and a "roller mode" with wheels.
  • Robust Update System: Includes a verified, health-gated, and reversible update mechanism to prevent bricking.
  • Simulation Support: Includes a MuJoCo-based simulator (scripts/duck-sim) that allows developers to run the actual robot daemons against a virtual body.

Related

  • Project
    AI-FanGe/Microduck-build-tutorial

    A compact biped robot project providing a full pipeline from MuJoCo-based RL training to real-world deployment on a Raspberry Pi Zero 2 W using ONNX policies.

  • Dispatch
    Pollen Robotics Microduck: An Open-Source Biped for RL Training

    Pollen Robotics has launched Microduck, a 25cm open-source biped robot designed for reinforcement learning (RL) training via a sim-to-real pipeline using MuJoCo.

  • Project
    pollen-robotics/microduck_rl

    Microduck RL provides MuJoCo‑Warp environments and PPO training scripts for the 800 g bipedal Microduck robot. Policies (walk, stand, tricks, etc.) are trained with realistic BAM actuator models, exported to ONNX, and can be published to Hugging Face for instant deployment on the real robot via the companion runtime.

  • Project
    LuwuDynamics/xgoduck_hardware

    XGO-Duck is an open-source, 3D-printable bipedal robot platform designed to bridge the gap between physical hardware construction and reinforcement learning-based motion control.

  • Project
    fanhao375/microduck-replica

    Microduck replica is a community project that reverse‑engineers the Pollen Robotics Microduck biped robot, providing editable CAD, reconstructed PCB schematics, BOMs, assembly videos, and a software fork with IMU‑I²C support and a web‑debug UI, enabling hobbyists and researchers to build and experiment with a reinforcement‑learning‑driven walking robot.