Nvidia's MicroDuck: From $399 Robot Duck to a 20,000-Preorder Viral Phenomenon

On the evening of October 22, at the GTC Berlin stage, four small bipedal robots walked across the stage, sat down, simulated sleep, and performed somersaults — all voice-commanded. Jensen Huang, CEO of Nvidia, guided the demonstration with the MicroDucks made by Pollen Robotics, and the resulting video went viral with more than 60 million views. Within 24 hours, preorders exceeded 20,000 units.

The MicroDuck is a 25-centimeter-tall robot weighing less than 800 grams with 15 motors. Each unit costs 399 dollars and is open-source: the SDK, the simulation environment, and the complete reinforcement learning training stack are available on GitHub. The core idea is simple yet powerful — rather than programming every movement by hand, developers let simulated versions of the robot learn to move on its own using Nvidia's Isaac Lab and simulation tools, then transfer the learned strategy to real hardware.

The MicroDuck story begins before GTC. In August 2026, Hugging Face announced the robot at a price of 399 dollars with delivery expected before Christmas. Pollen Robotics, a French startup based in Bordeaux, had been acquired by Hugging Face in April 2025 specifically to develop accessible, open-source robotics hardware. Months earlier, the two companies had launched the Reachy Mini, another desktop robot at 499 dollars, but the MicroDuck positioned itself as the more affordable line product.

What changed at GTC was the direct relationship with Nvidia. In September 2026, Nvidia closed its acquisition of Hugging Face for an estimated 13 billion dollars. With that, the MicroDuck — and the entire Pollen robotics line — became part of the GPU giant's portfolio. Clem Delangue, Hugging Face CEO, stated that total preorders crossed the 20,000 mark the day after the keynote, and asked whether the MicroDuck could become the best-selling robot of 2026.

The architecture behind the MicroDuck is as important as the hardware itself. The robot features a camera, LiDAR sensor, microphone, speaker, Wi-Fi, and Bluetooth. The articulated beak can pick up small objects. The simulation uses MuJoCo as the physics engine, PPO (Proximal Policy Optimization) for reinforcement learning, and ONNX for the portable representation of the trained policy. The reference environment uses Jetson Orin NX with JetPack 7.2, Ubuntu 24.04, and CUDA 13.2 — the same stack Nvidia promotes for edge inference.

What makes the MicroDuck technologically significant goes beyond its cuteness. It represents the democratization of robot training with reinforcement learning. Instead of needing a 50,000-dollar humanoid robot to test locomotion policies, developers can use the MicroDuck as a compact validation platform. The same sim-to-real flow that Nvidia applies to massive industrial robots — KUKA arms, Volkswagen vehicles, Agility Robotics humanoids — is the same flow developers use to train the MicroDuck on their laptop.

The delivery targeted for Christmas 2026 is an important milestone. If Nvidia can meet the deadline, the MicroDucks could become not only the year's best-selling robot but a catalyst for a new generation of robotics enthusiasts. The question remains: with Nvidia in control of the ecosystem, will the MicroDuck become the "Arduino of robotics" — an accessible platform that generates an ecosystem of applications and experiments — or merely a marketing product that loses its open-source focus when commercial pressures intensify?

Sources: TechCrunch, Interesting Engineering, CNBC, TechRepublic, Pollen Robotics

✓ Independent sources cross-checked and verified before publishing