EXPLORINEER
en

GitHub

pollen-robotics/microduck — A Tiny biped duck robot 🦆

RoboticsFirst seen 647 points


Pollen Robotics has shared the code for Microduck, a small walking duck robot. This GitHub repo holds the software that runs the robot, while the training of its movement policies lives in a separate repo.

What the repo is

This GitHub repository is the "brain" of a tiny two-legged duck robot. It holds the programs that control the robot's movements, radios, camera, and software updates. The robot's walking and other moves come from neural policies, which are learned behaviors made by training a computer model.

Key facts

  • The robot is about 25 cm tall and weighs about 800 g.
  • It runs on a Rockchip RK3566 chip.
  • A 50 Hz control loop sends commands to fifteen servos (small motors that move the joints) 50 times each second.
  • Updates are verified, checked for health, and can be rolled back.

What the duck can do

The README shows the robot walking and being driven with a gamepad, rolling on wheels, picking things up with its beak, and standing back up after it is knocked over. It can also sit, kick a ball, roll forward on command, and quack.

How the parts fit together

The robot's movement behavior is trained in a separate project called microduck_rl. That project uses MuJoCo, a physics simulator, and PPO, a common learning method. The trained policy is exported to ONNX, a standard file format, and this repo loads that file to run the robot.

Why it matters

The project is open for people who want to build on it. It shows how a small, affordable robot can be controlled by learned behaviors, and it gives developers guides for setup, simulation, and safe updates. Anyone without a physical robot can also try a simulated duck.

Glossary

Microduck
A small, two-legged duck robot that walks using learned movement policies.
ONNX
A standard file format that lets trained AI models run in different programs.
MuJoCo
A physics simulator used to test robot movements on a computer before real use.
PPO
Proximal Policy Optimization, a common method for training robot and game AI.
Rockchip RK3566
A small computer chip that runs the robot's software.