The Microduck from Pollen Robotics is a small open-source biped robot designed to make experimenting with physical AI and reinforcement learning more accessible outside of a traditional robotics laboratory.
Standing about 9.8 inches (25 cm) tall and weighing just 1.7 pounds (780 g), it uses 15 motors for whole-body movement and can walk, crouch, sit and stand, recover after falling, kick, and even roller-skate with optional attachments. Its articulated beak can also grasp and carry small objects.
The robot is equipped with a wide-angle camera, a LiDAR sensor for measuring nearby distances, two IMUs for sensing its movement and orientation, microphones, a speaker, NFC sensors, Wi-Fi, and Bluetooth. A Rockchip RK3566 processor with an AI accelerator provides onboard computing, while a removable battery offers approximately one hour of operation.
What makes this robot particularly interesting is its sim-to-real learning system: developers can train a behavior using reinforcement learning in a physics simulation and then transfer that learned behavior to the actual robot. Its SDK, simulation environment, and reinforcement-learning tools are open source, allowing users to modify existing behaviors, create new ones, and share their work with others.
Microduck represents an affordable platform for students, makers, and developers to explore how AI can learn to control a physical body, an increasingly important area as robots become more capable of operating in the real world.
The video above is intended for educational purposes and does not constitute or imply an endorsement by FutureForAll.org
