Hugging Face Unveils Microduck: A $399 Open-Source 25 cm Biped You Train with Reinforcement Learning

Hugging Face has released Microduck, a 25 cm tall open-source bipedal robot priced at $399 that users can train using reinforcement learning. Every movement the robot performs, including walking, sitting, kicking, and recovering from falls, is controlled by neural policies trained in a physics simulator and then transferred to the hardware. The training environments, code, and technical specifications are publicly available on GitHub, allowing users to either use the seven pre-trained behaviors that ship with the robot or create their own through machine learning. The robot is equipped with 15 motors, a camera, LiDAR, two IMUs, and other sensors, running on a Rockchip processor with approximately one hour of battery life per charge.