02/09/2026
The open-source robotic duck Microduck sold out within 24 hours for $399
On August 27th, Pollen Robotics, the robotics company under the AI open-source community Hugging Face, officially released the bipedal robot Microduck (commonly known as the "robotic duck"). This 25cm tall, less than 800g, and priced at $399 (approximately RMB 2680), "duck" saw pre-orders exceed $2.6 million within 24 hours of opening, selling out quickly at an average rate of one duck every 4 seconds. The official announcement followed: "The community has ordered too many ducks." New order delivery times have been extended to 4 to 6 months, and even if you place an order now, you may not receive yours by the end of the year.
Unlike service products such as robotic vacuum cleaners, Microduck's core positioning is an open-source embodied intelligence development platform for AI developers, which is known in the industry as the "Raspberry Pi moment" in the field of embodied intelligence.
It won't sweep or wash dishes; its beak is just a small gripper, only capable of picking up lightweight objects like socks and markers. However, it allows developers to program it themselves to achieve their desired functions. It comes pre-programmed with seven basic behaviors: walking, sitting, squatting, kicking, grasping, rollerblading, and getting up after falling. Users can explore different ways to play based on these—Dranger demonstrated a user-programmed gymnastic performance of the robotic duck at the launch event.
In terms of hardware, this 25cm-long device packs 15 servo motors (distributed in the legs, neck, and head), a wide-angle camera, a small LiDAR, dual IMUs, a microphone, a speaker, and NFC tag recognition. It also supports Wi-Fi, Bluetooth, and NFC near-field communication. The main control unit uses a domestically produced Rockchip RK3566 processor, which runs a 50Hz control loop, drives the 15 servo motors, and manages the camera, radio, and other modules.
On the software side, the SDK, MuJoCo simulation environment, and complete reinforcement learning training stack are all open-source under the Apache-2.0 license. Developers can train new policies in the simulator, export ONNX for deployment on airborne chips, and upload training results to Hugging Face Hub for global developer sharing.