Hugging Face's $399 Microduck Is a Waddling Open-Source Robot You Can Train Yourself

The duck-shaped bot picks things up, roller skates, and gets back on its feet when it falls. It ships before Christmas, and the code is free for anyone to inspect or modify.

AI2Day Newsdesk3 min read
A close-up overhead shot of a small wheeled robot built from standardised components sitting on a flat competition mat with coloured grid markings, surrounded b
Share

Key points

  • Hugging Face unveiled the Microduck on Thursday, priced at $399 and set to ship before Christmas.
  • The 25-centimetre-tall robot can carry up to 800 grams in its beak, recover from falls, crouch, and roller skate.
  • The full training code and simulation tools are free to download and modify on GitHub.
  • Hugging Face acquired French robotics startup Pollen Robotics in April 2025 to build affordable, open-source AI hardware.
  • Reports suggest Nvidia may acquire Hugging Face at a $13 billion valuation.

Hugging Face, the platform where AI developers share and download machine-learning models, wants a small duck robot sitting on your desk.

The company unveiled the Microduck on Thursday. It stands 25 centimetres tall, costs $399, and is scheduled to ship before Christmas. Its party trick is using its beak like a small gripper: it can pick up objects weighing up to 800 grams, roughly the weight of a large coffee mug full of liquid. It can also waddle, crouch, roller skate, and right itself when it tips over.

What makes it "open source"?

Open source means the software and training instructions are published for anyone to read, copy, or change. Hugging Face chief executive Clem Delangue described the Microduck as "an open-source robot you can teach new tricks with reinforcement learning," a technique where software learns by trying actions repeatedly and getting a score for how well it did. The full toolkit, including a simulator where you can test behaviours before loading them onto the physical robot, is available on GitHub.

That distinction matters for privacy. Any camera-equipped device in your home is a potential privacy concern. Delangue has argued that open-source software is safer than a closed, commercial product because anyone can audit the code and see exactly what the robot sends, and to whom. That is broadly true for the base software. What it cannot guarantee is what happens when users install third-party apps on top of the robot. Those apps can reach the Microduck's camera and microphone, and some may send that data to outside servers, depending on how the app is built. Treat it the way you would any smart-home device: only install software from sources you trust.

Pollen Robotics, a French startup that Hugging Face acquired in April 2025, built the hardware. The two companies already sell the $499 Reachy Mini, a small desktop robot powered by a Raspberry Pi (a cheap, credit-card-sized computer), and the $399 Reachy Mini Lite, which runs off a standard Mac or PC.

How does it perceive the world?

The Microduck uses a camera, lidar sensors (which work like sonar but with pulses of light to measure distances), and two IMUs, short for inertial measurement units, small chips that track movement and orientation the way a phone knows when you tilt it. Together, those sensors let the robot build a picture of the space around it.

What happens next?

Hugging Face is reportedly in acquisition talks with Nvidia, the chipmaker, at a valuation of around $13 billion. Nvidia has supplied computing infrastructure to Hugging Face since at least 2023, and both companies have publicly backed open-source AI development. A deal, if it closes, could accelerate Hugging Face's hardware ambitions.

Watch for: apps that request access to your Microduck's camera or microphone without a clear reason for needing it. Read the permissions before you install, and check whether the app sends data off the device.

© 2026 AI2Day