Hugging Face wants physical AI to arrive not as an intimidating humanoid or security machine, but as a rollerskating duck. Its $399 Microduck is deliberately playful—yet it also puts sensors, software and data questions into more homes.

The company unveiled the roughly 25-centimeter, duck-like robot on Thursday, after acquiring France’s Pollen Robotics in April 2025 and following it with the Reachy Mini desktop robot line. Microduck preorders are open, with first deliveries planned before Christmas; it comes in several colors and is designed to waddle, recover from falls, crouch and grip small items with a beak rated to lift up to 800 grams.¹

For Hugging Face, the point is less household help than access. Chief executive Clem Delangue called it an “open-source robot you can teach new tricks with reinforcement learning,” with its simulation tools, SDK and training stack available to developers.¹ The company frames that as a bid to democratize physical AI: behaviors can be trained in simulation, deployed to the robot, then refined and retrained.

Other accounts underscore how carefully the product is being pitched. It can follow a laser pointer, be steered with a game controller, pick up socks or markers, and skate; each unit is also assigned a lasting, generated vocal identity made of “weird little sounds, closer to a creature than an assistant.”² But Delangue has been candid that failure is part of the lesson: Microduck was built to be “made to move, ready to fall.”³

That openness is also where the tension lies. Microduck carries a camera, lidar, motion sensors and, according to one report, microphone and wireless connectivity. Delangue argues open models offer more privacy than a centrally controlled “black box system,” but open source does not stop third-party apps from accessing camera or microphone data and potentially sending it elsewhere.¹

The ambition is substantial for a toy-sized machine. Delangue says 50,000 sales would make Microduck a success—about $20 million in revenue—though the robot’s practical limits may prove as instructive as its tricks.³