Movement through reinforcement learning
Behaviors come pre-trained in simulation and loaded onto the robot; they cover movement-focused abilities such as walking, sitting and standing up, and recovering after a fall.
EDUCATION AND RESEARCH · POLLEN
Pollen Microduck
25 cm tall bipedal training and research robot that moves using reinforcement learning
Pollen Microduck is a 25 cm bipedal robot that operates with learned walking and interaction behaviors. Equipped with a camera, compact LiDAR, two IMUs, and a grasping beak, it comes ready for motion-based experiments and education.
- Motor
- 15 · legs, neck, and head
- Height
- 25 cm · 14 cm wide
- Weight
- < 800 g · Handheld
1 / 4
SELECT BY TASK
Pollen Microduck
what is it evaluated for?
Microduck is a motion-focused consumer/research robot from the Pollen Robotics team. At 25 cm tall, its compact body integrates 15 motors, perception sensors (a camera, a compact LiDAR with an 8x8 time-of-flight matrix, and two IMUs in the body and head), a grasping articulated beak, a microphone, a speaker, NFC antennas, and wireless connectivity. It comes out of the box with a set of trained behaviors and a game controller; its software stack includes Apache-2.0-licensed open-source tools, reinforcement learning training workflows, and sim-to-real transfer. It is designed for portable use with a replaceable battery and supports use cases for education, workshops, and R&D.
PRODUCT-SPECIFIC
Key features.
Features compiled from source product information; evaluate according to the model and configuration.
Comprehensive perception package
A front camera, compact LiDAR array, and two IMUs in the body/head provide the data needed for environmental perception and balance control.
Object interaction with a gripping beak
Its articulated beak can be used directly for simple manipulation tasks such as gripping and carrying objects.
Ready-to-run behavior set
It comes with pre-trained walking and play behaviors; users can interact with it right after initial setup.
Open-source software ecosystem
The Apache-2.0-licensed software stack covers robot control, simulation, and training workflows; users can use these tools to customize or retrain.
Portable power and backup
Replaceable battery design; a spare battery can ensure continuity during lessons and workshop use.
Classroom and multi-robot scenarios
Using multiple Microduck robots makes group interactions, races, and team-based activities easy.
SOURCED PRODUCT DATA
Technical features.
Values are taken from the source product page. Confirm the package, software, and accessory contents against the manufacturer's current documentation before ordering.
| Feature | Value |
|---|---|
| Dimensions (standing) | 25 cm tall, 14 cm wide |
| Weight | Under 800 grams |
| Battery capacity | 2600 mAh, removable NP-F550 camera battery |
| Battery | Removable NP-F550 camera battery; dual-slot charger in charging and developer packages |
| Operating time | Approximately 1 hour (varies with use) |
| Modes of movement | Bipedal walking, sitting, squatting, and standing up after a fall; wheeled gliding when skates are attached |
| Degrees of freedom | 15 DoF (legs, neck, and head) |
| Leg degrees of freedom | 5 DoF |
| End effector | Articulated gripping beak; NFC antenna in the beak |
| Joint motor | Dynamixel XL330 servo motors (based on the actuator model in the manufacturer's RL repository) |
| Manufacturer's note | Not confirmed: camera resolution and field of view, LiDAR range, radio versions, SDK languages, and age recommendation (manufacturer's press kit, September 2026). |
| Joint configuration | 5 joints per leg (hip yaw, roll, pitch; knee; ankle), 4 joints in the neck and head, and the beak |
| Color options | Cream, Graphite, Lavender, Sky Blue; same robot, different shell |
| Package contents | Robot, battery, USB-C cable, game controller |
| Software license | Software stack: Apache-2.0; mechanical and electronic designs are not open; 3D models in the RL repository: CC BY-NC-SA |
| Availability | Pre-orders open 27 August 2026; first deliveries are targeted for before Christmas 2026 |
| Sensors | Wide-angle front camera and camera-use light; compact LiDAR (8x8 time-of-flight array); two IMUs (body and head); microphones |
| Interfaces | robotctl command line, SDK, JSON-RPC; MuJoCo simulation and RL training tools (Apache-2.0) |
| Connectivity | Wi-Fi and Bluetooth; Bluetooth control from a laptop (duckctl) |
| Wireless | Wi-Fi and Bluetooth |
| Interaction | Speaker and custom-made robot sounds; game controller; two NFC antennas in the head and beak |
| Processor | Rockchip RK3566 + AI accelerator; 1 GB RAM, 32 GB storage |
| Operating system | Robot daemons written in Rust (robotd, updaterd, configd, btd, padd, mediad, tofd) |
| Software features | 50 Hz on-robot policy loop; 7 trained motions; instant switching between policies; signed, health-checked, and reversible updates |
SEE THE PRODUCT IN ACTION
Pollen Microduck introduction.
The introduction video linked from the source product page opens when you press the play button.
PRODUCT INFORMATION SOURCES
Sources and documentation.
This catalog is intended to help explore product information. No commitments are made on behalf of Robotik Çözümler regarding stock, authorized dealers, pricing, warranty, or after-sales service terms.
FREQUENTLY ASKED QUESTIONS
Frequently asked questions.
What is Microduck, and who makes it?
Microduck is a 25 cm tall, bipedal, movement-focused robot developed by Pollen Robotics; it has various perception sensors, 15 motors, and a grasping beak.
Is Microduck open source?
Yes. The software stack is released under the Apache-2.0 license; simulation and reinforcement learning tools are also available in the relevant repositories. Mechanical and electronic design files have not been shared as open hardware.
Can it be used without coding knowledge?
Yes. The robot can be used out of the box with trained behaviors and a game controller, without coding knowledge; coding is only required if you want to retrain behaviors or add new motions.
How do I teach it a new motion, and is a GPU required?
Behaviors are trained in simulation (using MuJoCo-based tools) and transferred to the robot in ONNX format. Local training may require a CUDA-enabled GPU; if you don't have a GPU, training options are available through cloud services.
What’s in the box, and which packages can be added?
The standard box includes the robot, battery, USB-C cable, and game controller. Additional options include sets such as a charging package, developer package, and accessory package.
NEXT STEP
Is this product right for your task?
Let's evaluate it together.
Start by defining your work area, task, and integration requirements.