Most robotics launches ask you to trust a demo video. Pollen Robotics, the Bordeaux-based robotics team at Hugging Face, is instead shipping the training loop itself. This week, it opened pre-orders for Microduck, a 25 cm bipedal robot in which every movement—walking, sitting, kicking, roller-skating, and standing back up after a fall—is a neural policy trained in a physics simulator and then exported to the hardware. Priced at $399, Microduck comes with its training environments, reward functions, domain-randomization settings, and sim-to-real recipe all publicly available on GitHub.
Microduck follows the success of Reachy Mini, which has shipped more than 10,000 units, but it inverts that robot's premise. Where Reachy Mini was designed to sit on a desk and interact, Microduck is built to leave the desk, fall over, and get back up.
The Hardware
Standing 25 cm tall and 14 cm wide, Microduck weighs under 800 g. It features 15 motors spread across the legs, neck, and head, plus an articulated beak that can pick objects off the floor. The onboard compute is powered by a Rockchip RK3566 with an AI accelerator, 1 GB of RAM, and 32 GB of storage.
The Learning Pipeline
What sets Microduck apart is its development philosophy: the robot is not pre-programmed but trained through reinforcement learning. Users can generate new behaviors by designing reward functions and running simulations, which are then transferred to the physical robot using a standardized sim-to-real pipeline. This approach lowers the barrier to entry for robotics research and education, making it accessible to hobbyists, students, and researchers alike.
The full stack—including simulation environments, training scripts, and hardware schematics—is open-sourced, encouraging community collaboration and rapid iteration. In 2026, as physical AI continues to gain momentum, Microduck represents a significant step toward democratizing advanced robotics.
Pricing and Availability
Microduck is available for pre-order at $399, with shipments expected to begin later this year. The open-source ecosystem around it means that early adopters will not just receive a robot but will also gain access to a growing library of behaviors and training tools. For those interested in exploring the intersection of reinforcement learning and physical robotics, Microduck offers an affordable and transparent entry point.
via MarkTechPost