AnswersPublished26 sources
Hugging Face Microduck: Price, Specs, Behaviors and Open-Source Training
Microduck is a $399 introductory price, 25 cm biped from Pollen Robotics and Hugging Face. It weighs under 800 grams and includes 15 motorized degrees of freedom, a camera, LiDAR, two IMUs, a grasping beak, Wi‑Fi, Bluetooth, 1 GB of RAM and 32 GB of storage.
Published byEdited with GPT-5.6 LunaImages generated with GPT Image 1.5
Research answer
Key takeaways
- Microduck is a $399 introductory price, 25 cm biped from Pollen Robotics and Hugging Face.
- It weighs under 800 grams and includes 15 motorized degrees of freedom, a camera, LiDAR, two IMUs, a grasping beak, Wi‑Fi, Bluetooth, 1 GB of RAM and 32 GB of storage.
- Its software is intended to let developers train behaviors in MuJoCo simulation, deploy them to the physical robot and share or retrain them through an open development stack.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What is Hugging Face’s $399 Microduck, and what are its size, weight, design, preorder and delivery timeline, open-source reinforcement-lear. Article summary: Microduck is a $399 introductory-price, duck-like educational biped from Pollen Robotics, Hugging Face’s robotics team. It is intended less as a finished household robot than as an affordable real-world platform for lear. Topic tags: general, general web, news. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake numbers
Microduck is a small, duck-like bipedal robot from Pollen Robotics, the robotics team acquired by Hugging Face. Its headline price is $399 before taxes and shipping, with preorders opening on August 27, 2026 and first deliveries targeted for before Christmas 2026. 23
32
The more important point is its purpose: Microduck is designed as an approachable platform for experimenting with reinforcement learning (RL), simulation and physical AI. It arrives with trained behaviors, but developers are meant to be able to create, retrain and deploy new ones rather than treating the robot as a sealed consumer gadget. 21
26
41
Microduck price, size and delivery
- Price: $399 before taxes and shipping at the introductory price.
- Preorders: Opened August 27, 2026.
- First deliveries: Targeted for before Christmas 2026.
- Launch regions: Initial reporting identifies North America and Europe, including the UK, as launch markets.
18
21
23
- Size: 25 centimeters tall, or about 9.8 inches.
- Weight: Under 800 grams.
- Colors: Cream, Graphite, Lavender and Sky.
24
The delivery date is a target rather than a guarantee. Buyers should also account for taxes, shipping and any local import costs because the advertised $399 price excludes them. 18
23
Design: a one-eyed biped with a gripping beak
Microduck has an armless, one-eyed design with articulated legs, head and neck. Its beak is more than a visual reference to a duck: it can function as a small gripper for picking up objects. Fifteen motors provide the robot’s articulated movement, including its distinctive waddle, crouching and recovery motions. 17
21
23
The robot is also designed to work with optional roller accessories, adding a skating mode to its walking-based movements. That combination of a lightweight body, bipedal balance and a physical gripper makes Microduck a compact test platform for learning how software policies interact with the real world.
What Microduck can do out of the box
Launch materials describe seven trained moves or behaviors. Reported examples include:
- Walking with a velocity-tracking waddle
- Sitting and standing
- Crouching
- Kicking
- Picking up or scooping small objects with the beak
- Roller-skating when the optional rollers are fitted
- Recovering from common falls
The supplied behaviors are intended as starting points, not a fixed limit. The central appeal is that developers can use the existing policies, modify them or train new behaviors through the robot’s software and simulation workflow. 21
23
26
Sensors, controls and built-in hardware
Microduck combines several sensors and interaction methods in a very small body:
- A front-facing camera
- A compact time-of-flight LiDAR sensor
- Two inertial measurement units, or IMUs, for tracking motion and orientation
- Microphones and a speaker
- Wi‑Fi and Bluetooth
- An articulated beak for grasping
- NFC hardware identified in the product materials
The robot comes with a game controller, so it can be operated immediately without writing code. It can also run autonomous learned policies, which is the route intended for experimentation and retraining. 21
24
25
Each unit is also intended to have an individually generated voice rather than one identical stock voice shared by every robot. That gives Microduck an audio identity alongside its visual design, although the product’s main technical value remains its trainable movement and interaction policies.
Processor, memory and battery
The onboard computer is based on a Rockchip RK3566 processor with an AI accelerator. Microduck has 1 GB of RAM and 32 GB of storage. Its removable 2,600 mAh NP‑F550 battery is reported to provide roughly one hour of continuous runtime, depending on how the robot is being used. 18
24
The battery is removable, and the package includes the robot, battery, USB-C cable and game controller according to the supplied product information. 24
How the open-source training workflow works
Microduck’s software model follows a familiar open-development loop:
- Train in simulation. Developers can work with MuJoCo physics locally or through Hugging Face Jobs.
- Evaluate a policy. The policy can be tested in a simulator before hardware is involved.
- Deploy to the robot. A trained behavior can be transferred to the physical Microduck.
- Refine and retrain. Developers can adjust the policy or simulation environment after observing real-world performance.
- Share the result. New behaviors can be published for others to inspect, use or improve.
The browser-based Microduck simulator runs MuJoCo physics compiled to WebAssembly and performs policy inference in the browser at 50 Hz without requiring a backend. It includes both leg-based walking and a roller-based skating variant. 4
The planned software release covers the SDK, simulator, control software, RL training tools, policies and sim-to-real workflow under the Apache-2.0 license, according to the launch reporting. That open-source promise applies to the software stack; it should not be read as a claim that every mechanical or electronic design file is open source. 27
The development code is associated with the pollen-robotics/microduck project on GitHub, giving developers a place to inspect and fork the training and control stack. 41
What “open source” means here—and what it does not
Microduck’s open-source proposition is primarily about making robot-learning software accessible. A developer can start with a supplied behavior, run policies in simulation, alter the training setup and attempt a sim-to-real deployment on the physical robot.
That is different from buying a robot whose hardware is fully open and endlessly reconfigurable. The available product information supports the software-open-source claim, but does not establish that Microduck’s complete mechanical and electronics designs will be released. 27
Where Microduck fits in Hugging Face’s robotics strategy
Microduck follows Reachy Mini, an earlier small expressive robot associated with Pollen Robotics and Hugging Face. Reachy Mini was offered in versions with either onboard compute or an external computer, positioning it as an accessible platform for experimenting with expressive robotics. 12
Together, the products show a broader strategy: bring the collaborative model used for software AI into physical systems through lower-cost hardware, shared tooling, simulation and reusable policies. Pollen Robotics’ acquisition expanded Hugging Face’s work beyond models and datasets into embodied AI and robot learning. 43
46
Microduck is therefore best understood as a hardware distribution point for a community-driven robotics ecosystem. Its $399 price lowers the cost of owning a physical test robot, while the simulator and open training tools lower the cost of trying ideas before deploying them to hardware.
What about the reported Nvidia acquisition?
Microduck’s launch coincided with reports about a possible Nvidia acquisition of Hugging Face, but the two stories should be kept separate. Reuters reported that Nvidia had agreed to acquire Hugging Face for $12.9 billion, citing The Information. CNBC separately described the matter as ongoing acquisition talks and noted that neither company had confirmed a deal. 33
35
As a result, the often-repeated $12.9 billion to $13 billion figure should be described as reported deal or valuation information—not as a confirmed completed acquisition. The launch of Microduck does, however, underline why Hugging Face’s robotics push matters: the company is extending its open-source AI model into systems that learn through physical interaction, not only through software.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Hugging Face Microduck: Price, Specs, Behaviors and Open-Source Training"?
Microduck is a $399 introductory price, 25 cm biped from Pollen Robotics and Hugging Face.
What are the key points to validate first?
Microduck is a $399 introductory price, 25 cm biped from Pollen Robotics and Hugging Face. It weighs under 800 grams and includes 15 motorized degrees of freedom, a camera, LiDAR, two IMUs, a grasping beak, Wi‑Fi, Bluetooth, 1 GB of RAM and 32 GB of storage.
What should I do next in practice?
Its software is intended to let developers train behaviors in MuJoCo simulation, deploy them to the physical robot and share or retrain them through an open development stack.