Hugging Face unveils Microduck, a $399 open-source robot trainable with reinforcement learning

Hugging Face has moved further into physical AI hardware with Microduck, a small duck-shaped robot designed to be taught new physical skills through open-source reinforcement-learning tools rather than shipped with fixed, closed behaviors.

What's new

According to TechCrunch's report on the launch, "Hugging Face unveiled the Microduck on Thursday, a cute little duck-like robot that sells for $399 — and ships before Christmas." CEO Clem Delangue described it as an "open-source robot you can teach new tricks with reinforcement learning," adding on social media: "Welcome to the era of open-source affordable robots to democratize physical AI and world models!"

Physically, "the 25-centimeter-tall duck can waddle, pick things up with its beak (up to 800 grams), get back up when it falls, crouch, and even roller skate." It perceives its surroundings with a camera, lidar, and two inertial measurement units, and its articulated beak doubles as a gripper. Unlike a typical consumer robot toy, Microduck's behaviors aren't hard-coded by the manufacturer: its SDK, simulation environment, and full reinforcement-learning training stack are published openly, so developers can train new behaviors in simulation and deploy them directly onto the physical robot.

Context

Hugging Face is best known as the dominant platform for sharing open model weights, datasets, and ML applications — not as a hardware company. That changed when it acquired French robotics startup Pollen Robotics in April 2025 specifically to build affordable, open-source AI hardware; the two companies had already launched Reachy Mini, a stationary desktop robot, as an earlier step in that direction. Microduck extends the same open-hardware philosophy — ship a capable but affordable physical platform, then let the open-source community supply the behaviors — into a mobile, legged form factor rather than a fixed desktop unit.

The timing is notable: the announcement lands the same week reports surfaced that NVIDIA is in advanced talks to acquire Hugging Face for roughly $12.9 billion, underscoring how central Hugging Face has become to both the open-model software stack and, increasingly, to physical AI and robotics.

Why it matters

Most consumer-facing robots ship as closed systems where the manufacturer controls what the device can do. By publishing the SDK, simulator, and full RL training stack alongside a genuinely capable, sub-$400 legged robot, Hugging Face is betting that an open, hackable hardware platform can do for robotics what open model weights did for language models — let a distributed community iterate on capabilities faster than any single company's product roadmap. If it succeeds, Microduck could become a reference platform for reinforcement-learning robotics research and hobbyist experimentation in the way Hugging Face's own model hub became a reference point for open-weight LLMs.