Pollen Robotics (Hugging Face) Microduck

MicroduckMade to move · Ready to learn
A 25 cm open-source biped you train yourself with reinforcement learning. Playable out of the box.
The launch film · sound on
Roll thetape
Waking up the duck…
Meet the twin
sim2realthat works
Trained in sim, deployed on the real robot. This is the simulated twin the ducks were trained on.
Fun out of the box. Yours to retrain.
Teach it newtricks
Every behaviour is a policy you can retrain on your own machine.
01
Train in simulation
Behaviours are learned in physics sim, on your machine or on Hugging Face Jobs.
02
Deploy on the robot
One step from simulation to the real thing.
03
Refine the simulation
Tune, re-train, re-deploy.
04
Publish the policy
Share your new behavior with the community!
Walk
Velocity-tracking gait.
Sit & stand
Sits down, holds the pose, stands back up on its own.
Kick
A one-shot boot, then straight back to walking.
Grab
Dips the beak to the ground, scoops, and pops back upright.
Roller skating
Roller skating locomotion when the skates are equipped.
Get back up
Flat on its back to standing, all by itself, ready for the next command.
One robot, four colourways
Choose yourcolour
Every Microduck ships in one of four colourways. Same robot, same brains underneath - pick the shell that best fits you.
Waking up the duck…
Out in the world
In the wild
The real robot in real places - on desks, on the pitch, out at golden hour.
The robot, and what to add to it
Pick yourpack
The robot is everything you need on day one. The packs add play gear and spare parts.
$399
Tech specs
Full sheet in the press kit- 15
- Motors
- 25 cm
- Tall
- 800 g
- To pick up
- Camera
- Plus LiDAR and two IMUs
- 7
- Trained moves in the box
- 50 Hz
- Onboard policy loop
Built in the open
Open source
The SDK, the simulation and the full RL training stack are on GitHub. What the robot runs is what you can read, fork and retrain.
pollen-robotics/microduckssh microduck
$ robotctl monitor # status of the robot$ robotctl configure # configure the robot$ robotctl update # update the robot
Apache-2.0
The whole software stack, permissively licensed
MuJoCo
The physics sim every policy is trained in
7 policies
Every shipped move, published and retrainable
Join the flock
Builds on show, policies to swap, help when a leg does something strange. The community lives on Discord.
End of tape · be kind, rewind
Pre-orders areopen now
In four colourways. Ships before Christmas 2026.
Introductory price, before taxes and shipping.

Facts Only

* The robot is 25 cm in height and weighs 800 g.
* It is an open-source biped trained with reinforcement learning.
* Behaviors learned include walking, sit & stand, kick, grab, roller skating, and get back up.
* Training can occur in simulation, on a local machine, or via Hugging Face Jobs.
* The system features one robot design available in four colorways.
* The software stack is open source, with the SDK, simulation, and RL training stack on GitHub.
* The physics simulation used for training is MuJoCo.
* The robot is equipped with a camera, LiDAR, and two IMUs.
* Trained moves are published and retrainable.

Executive Summary

The Microduck is a 25 cm open-source biped robot trained using reinforcement learning. The system emphasizes a full learning loop, allowing users to train policies in simulation, deploy them on the real robot, refine the simulation, and publish results. Learned behaviors include walking, sitting/standing, kicking, grabbing, roller skating, and getting back up. The robot is available in four colorways and ships with optional packs for additional play gear. The underlying software stack, including the SDK and RL training components, is open source, built upon MuJoCo for simulation.

Full Take

The narrative establishes a clear progression from simulated learning to real-world deployment, creating an iterative feedback loop that emphasizes user agency over static product delivery. The architecture’s foundation in open source—offering the SDK, simulation environment (MuJoCo), and training stacks on GitHub—suggests a philosophy rooted in community contribution and deep technical transparency, positioning the robot not just as a consumer product but as a platform for applied machine learning research. The sequence of four steps (Train in simulation, Deploy on the robot, Refine the simulation, Publish the policy) mirrors a robust MLOps cycle, which frames behavior acquisition as an engineering discipline rather than mere demonstration. This structure suggests an underlying pattern focused on democratizing complex AI training methodologies by providing accessible tools for physics-based robotics. The tension lies between the tangible, playful output (the ducks, the colors) and the deep, technical scaffolding of the reusable policy framework. What is being balanced is the accessibility of advanced RL techniques against the practical constraints of physical deployment and community-driven refinement.

Sentinel — Human

Confidence

The text reads like promotional material for an open-source robotics project, displaying human-like persuasive structuring overlaid with precise technical details.

Signals Detected

Varied sentence structure and use of direct, punchy marketing language mixed with technical specifications.

Clear, functional flow that moves from concept (RL) to implementation (hardware/software), typical of product pitches.

Structured enumeration of features and open-source details suggests a deliberate marketing structure rather than raw AI generation.

The specific technical detail (MuJoCo, GitHub repositories, specific motor/sensor counts) requires careful placement, suggesting human knowledge or highly accurate LLM retrieval based on existing data.

Human Indicators

The inclusion of specific open-source links (GitHub), SDK names (robotctl), and pricing details anchors the text in verifiable, real-world product marketing context.

The narrative uses rhetorical devices ('Waking up the duck…', 'Join the flock') that serve a clear commercial purpose.