Meet Microduck: A $399 Waddling Robot for Makers Who Want to Learn How AI Moves

Most of us meet artificial intelligence through a chat window. You type, it answers. But there is a whole other side of AI that happens in the physical world: teaching a machine to balance, step, and pick things up. That is the idea behind Microduck, a small two-legged robot that just opened for pre-order, and it is aimed squarely at people who like to tinker. This is not a review, since the robots have not shipped yet. It is a look at what the project is and why it might appeal to a curious maker.

So what is a Microduck?

Microduck is a roughly 25 cm (about 10 inch) tall walking robot with a duck-like look, made by Pollen Robotics, a French robotics company that is part of Hugging Face, the well-known AI community. It weighs under a kilogram, has a beak that can grab small objects, and one big eye (a camera). You can see the official page, with videos and specs, at pollen-robotics.com/microduck.

The introductory price is $399 before taxes and shipping, and the company has said it hopes to deliver before Christmas in North America, Europe and the UK. Check the official page for current availability, because early shipping dates for new hardware have a way of shifting.

What is inside

For a robot this size, the parts list is generous. There are 15 small motors (called servos, which are motors that move to a precise position), a front camera, a tiny distance-sensing laser scanner known as LiDAR, and two motion sensors (IMUs) that tell the robot which way is up and how fast it is tilting. A small computer chip inside runs the robot's "brain." The battery is a removable camera-style pack that lasts around an hour, so expect short play sessions.

The fun part: teaching it instead of programming it

If your programming roots go back to BASIC and COBOL, this will feel different, and that is what makes it interesting. Nobody writes "move left leg forward 2 inches." Instead, the robot learns to walk through a method called reinforcement learning. Picture training a puppy with treats: the software tries thousands of random wobbles, gets a reward when it moves in a useful direction, and gradually keeps what works.

Here is the clever part. The practice happens in a computer simulation, a virtual physics world, where thousands of virtual ducks can stumble at once without breaking anything. The Microduck team says a usable walking style can emerge in an hour or two of training. Once it looks good on screen, you send it to the real robot. The project describes a loop: train in the simulator, try it on the robot, improve the simulation to match reality, then share what you built.

What it can do out of the box

You do not have to start from scratch. Microduck arrives with a set of ready-made moves, including walking, sitting and standing, kicking, grabbing, and getting back up after a fall, along with a game controller to drive it. An optional accessory pack adds wheeled feet for roller skating, plus extras like a ball. The robots can also react to special tags and to each other, so a few ducks can even try a bit of soccer.

Open source, with a footnote

The software, including the simulator and the training tools, is open source under the Apache 2.0 license. That means you can read it, change it, and share your changes. The trained moves that ship with the robot are published too, so you can study them or retrain them. The footnote: the physical design files are under a non-commercial license, and the company asks that Microduck not be called open-source hardware. For a hobbyist that mostly means you can tinker freely, but you cannot build a product business on the design.

What to know before you get excited

Training new moves needs real computing muscle. You will want a computer with a capable graphics card, or you can rent time on Hugging Face's cloud service. The code is written in Python, a popular modern language that reads more like plain English than you might expect, and a friendly way in for anyone coming from Basic. Expect a learning curve with the command line and a few installation hiccups, which is part of the maker experience.

It also lifts only about 100 grams, so this is a learning toy and a platform to experiment on, not a household helper. If your goal is a robot that does chores, this is not it. If your goal is to understand how robots learn to move, it looks like an unusually friendly way in.

Why it matters for curious makers

Robotics used to mean expensive labs and years of math. Projects like this lower the bar: a small price, a simulator, shared code, and a community where people swap behaviors the way others swap recipes. Whether you end up teaching your duck to dance or just watching it waddle across the kitchen floor, you will learn how modern AI turns practice into motion. We will follow this one, and we would love to hear from any Digital Den reader who gets a duck of their own.