Pollen Robotics has opened pre-orders for Microduck, a 25-centimetre biped robot designed to work both as a ready-to-play device and as a platform for training new behaviours. The introductory price is $399 before taxes and shipping, and the company says deliveries are planned before Christmas 2026.
Microduck has 15 motors, a camera, LiDAR and a beak capable of grasping objects. The standard package includes the robot, a battery, a USB-C cable and a game controller. Buyers can choose among four shell colours while receiving the same hardware and computing platform underneath.
The product's central pitch is that owners can modify what the robot does rather than being limited to fixed manufacturer routines. Pollen Robotics says the software development kit, simulator and reinforcement-learning training stack are available on GitHub under a permissive licence. Behaviours can be trained in a physics simulation either on the owner's computer or through Hugging Face Jobs, then transferred to the physical device. Shipped behaviours are also published so users can inspect, fork and retrain them.
Examples presented by the company show the small robot performing recovery and locomotion tasks. It can sit and return to standing, resume walking after booting, scoop with its beak and get up after lying flat on its back. With optional skates fitted, it can use a roller-skating movement policy. These demonstrations illustrate the range of supplied policies, although the evidence provided by the company does not include independent performance testing.
Pollen Robotics is also offering add-on packs. A spares-oriented bundle contains three replacement motors, five motor cables, two batteries, a dual charger, 10 near-field communication tags, Hugging Face credit, a screwdriver and screws. A play-focused pack includes a laser pointer, an NFC-enabled instant-photo accessory, two rollers, a ball and 10 NFC tags.
The combination of simulation and physical deployment places Microduck in a category between a consumer toy and a compact robotics development kit. Reinforcement learning often depends on repeated attempts that would be slow or potentially damaging on hardware; a simulated twin lets users iterate before running a policy on the robot itself. The open stack also gives developers access to the software responsible for its movements rather than only an application-level control surface.
The launch leaves several practical details dependent on future delivery, including how closely user-trained simulation policies transfer to individual production units. For now, Pollen Robotics is positioning the device as usable on arrival while retaining enough access for experimentation. Its announced price, included controller and public training tools make that dual purpose the main distinction of the pre-order offer.
