On August 27 2026 Hugging Face and its robotics unit Pollen Robotics opened pre‑orders for the Microduck, a biped robot priced at US $399 before taxes and shipping. The device is marketed as a hands‑on platform for developers who want to experiment with reinforcement‑learning‑driven motion.

Key takeaways

- Pre‑orders began on August 27 2026 at $399 before taxes and shipping.
- Height 25 cm, weight ≈800 g.
- 15 motors drive the legs, head and neck.
- Seven built‑in policies: walk, sit, stand, kick, grab, roller‑skate and self‑right after a fall.
- Sensor suite includes a wide‑angle camera, 8×8 time‑of‑flight LiDAR, two IMUs, microphones, a speaker, NFC, Wi‑Fi and Bluetooth.
- Onboard compute: Rockchip RK3566 AI‑accelerated CPU, 1 GB RAM, 32 GB flash, policy loop at 50 Hz.
- Removable NP‑F550 2600 mAh battery gives roughly one hour of operation.
- All software – SDK, MuJoCo‑based simulator and reinforcement‑learning pipeline – is released under the Apache 2.0 licence on GitHub.
- First shipments are scheduled for before Christmas 2026; later orders carry a 4‑to‑6 month lead time.
- Four colourways are offered: Cream, Graphite, Lavender and Sky.
- Hugging Face acquired Pollen Robotics in April 2025, extending its open‑source robotics effort.
- Research‑grade biped platforms typically start around $13 000, making Microduck an order‑of‑magnitude cheaper entry point.

What the hardware looks like

Microduck is a hand‑held biped that fits in one palm. Its 25 cm height and under‑800 g mass keep the kinetic energy of a fall low, and the robot can stand up on its own after tipping over. The articulated beak functions as a small gripper, letting the duck pick up lightweight items.

ComponentSpecification
Motors15 degrees of freedom across legs, head and neck
VisionFront wide‑angle camera
Depth8×8 time‑of‑flight LiDAR matrix
Motion sensingTwo IMUs (body and head)
AudioMicrophones and speaker; a per‑robot generated voice
WirelessWi‑Fi, Bluetooth, two NFC antennas
ComputeRockchip RK3566 with AI accelerator
Memory / storage1 GB RAM, 32 GB flash
Policy loop50 Hz
BatteryRemovable NP‑F550, 2600 mAh (~1 h runtime)
Pre‑installed behaviours7

All numbers are taken from the manufacturer’s spec sheet.

Open‑source software stack

The launch package publishes the full development kit on GitHub. The SDK, a MuJoCo‑based physics simulator, and the reinforcement‑learning training pipeline are all available under an Apache 2.0 licence, meaning developers can inspect, modify or extend any part of the stack.

How a new skill is created

The workflow that the team describes consists of four stages:

  1. Simulate the target motion in MuJoCo.
  2. Train a reinforcement‑learning policy against the simulated physics.
  3. Deploy the resulting policy to the physical duck via the SDK.
  4. Iterate by refining the simulation, retraining and redeploying. Because the robot is lightweight and its battery lasts only about an hour, a failed trial usually ends with the duck on the floor, after which it can stand up automatically – a design choice that keeps the risk of damage low during extensive trial‑and‑error cycles.

Pricing, availability and accessories

Pre‑orders are priced at $399 before taxes and shipping. Each unit ships with a dedicated game controller for immediate interaction. The initial batch is slated to arrive before Christmas 2026; orders placed after the launch window are listed with a 4‑to‑6 month lead time. Buyers can select from Cream, Graphite, Lavender or Sky finishes.

Position in the robotics ecosystem

At $399, Microduck is far less expensive than typical research‑grade biped platforms that start around $13 000. Those larger systems are aimed at laboratory research, while Microduck brings the same embodied‑learning pipeline to a tabletop form factor, opening up reinforcement‑learning‑driven motion experimentation to hobbyists, students and small teams.

Outlook

The release follows Hugging Face’s April 2025 acquisition of Pollen Robotics, a move the company framed as a way to make robotics “open, affordable, and private.” CEO Clément Delangue described the robot as “a tiny $399 open‑source robot you can teach new tricks with reinforcement learning.” The open‑source nature of both hardware and software positions Microduck as a low‑cost entry point for anyone looking to explore physical AI.


Sources

This article was researched and fact-checked against the following sources: