Published: September 1, 2026

The robotics industry has become accustomed to seeing large humanoid machines, robot dogs and increasingly sophisticated autonomous systems. But one of the most talked-about robots of late August 2026 is considerably smaller, much cheaper and far more unusual.

It looks like a duck.

Hugging Face’s new Microduck is a small, bipedal robot designed to make artificial intelligence and robotics more accessible to developers, students, researchers and enthusiasts. Standing at around 25 centimetres tall and weighing less than a kilogram, the robot has attracted attention not only because of its unusual appearance but also because of its open-source software approach and relatively affordable $399 price.

There is another interesting detail hidden inside the little machine.

The Microduck is powered by a Rockchip RK3566 processor, a chip developed by Chinese semiconductor company Rockchip. The processor serves as the robot’s main computing platform, helping manage its sensors, motors and onboard control system.

The combination is notable: a French robotics company owned by Hugging Face has created an open-source AI robot using a Chinese processor, while the product itself is being marketed as an accessible platform for experimenting with physical AI.

And demand appears to be strong.

Within a short period after pre-orders opened on August 27, reports indicated that Microduck generated more than $2.6 million in sales during its first 24 hours. Later reports said pre-orders had surpassed 10,000 units and that the demand was large enough to create a delivery backlog.

The Microduck is therefore more than a cute technology product.

It could become an important experiment in how open-source artificial intelligence moves from computer screens into the physical world.

What Is Hugging Face’s Microduck?

Microduck is a compact two-legged robot developed by Pollen Robotics, the French robotics company acquired by Hugging Face in 2025.

Hugging Face is best known for its open-source artificial intelligence ecosystem, including models, datasets and tools used by researchers and developers around the world.

Its move into robotics represents a broader attempt to bring open AI technology into physical machines.

Microduck is the latest example of that strategy.

The robot is approximately 25 centimetres tall and weighs roughly 800 grams. Instead of traditional arms and hands, it has an articulated beak-like mechanism that can function as a small gripper.

It can walk, crouch, recover after falling and interact with lightweight objects.

Demonstrations have shown it picking up objects such as socks and markers and placing them into containers. Its camera and LiDAR system help it perceive its environment.

The robot can also be equipped with roller-skating accessories, making it as much an experimental platform as an ordinary consumer gadget.

Its unusual appearance is clearly part of the appeal.

But underneath the playful design is a serious robotics project.

Why Is the Microduck Selling So Quickly?

The first major reason is price.

At $399, Microduck costs dramatically less than many advanced robots being developed for research or industrial applications.

That price places it within reach of a much larger group of potential users.

A university student, independent programmer, robotics enthusiast or AI researcher may not have the budget to purchase a large industrial robot.

A $399 platform is a different proposition.

It allows more people to experiment.

It also creates the possibility of a much larger community of developers producing software and behaviors for the same physical platform.

That is particularly important to Hugging Face.

The company's strength has historically been its community.

Developers can share AI models and datasets through its ecosystem. Microduck extends that concept into robotics.

Instead of simply downloading a model, users can potentially download or create a behavior for a physical robot.

That could include walking patterns, object recognition, navigation, interaction and other capabilities.

The company is effectively trying to make robotics more like software development.

More Than a Toy

At first glance, Microduck may look like a novelty product.

Its duck-like body, small size and ability to roller-skate make it highly shareable on social media.

But the company's goal goes beyond entertainment.

Pollen Robotics describes the Microduck as a platform for physical AI and reinforcement learning.

Users can train behaviors in simulation before transferring them to a real robot.

This approach is important because training physical robots directly can be expensive and time-consuming.

A robot can fall.

Motors can wear out.

Batteries can run down.

Components can be damaged.

Simulation provides a safer environment where developers can repeatedly test a behavior before deploying it to physical hardware.

The Microduck is designed to support this workflow.

A developer can train a behavior virtually, test it and then transfer the resulting policy to the robot.

The process can then be repeated as the developer collects more data and improves the behavior.

The Chinese Chip Inside the Robot

One of the most interesting technical details is the Microduck's processor.

The robot uses Rockchip's RK3566, a processor developed by Chinese semiconductor company Rockchip.

Recent reports identified the RK3566 as the main control chip inside Microduck. It is used for the robot's onboard computing and helps coordinate its control loop, motors, camera and communication functions.

The presence of a Chinese processor in a French-developed robot backed by a major open-source AI company highlights the increasingly international nature of modern technology supply chains.

A robotics product can involve components from multiple countries.

The software may be developed in Europe.

The processor may come from China.

Motors can come from another country.

Sensors may come from different suppliers.

Final assembly and distribution can take place elsewhere.

Modern robotics is therefore rarely the product of a single nation.

What Is the Rockchip RK3566?

The RK3566 is a system-on-chip designed by Rockchip.

It belongs to a class of processors commonly used in embedded devices, smart hardware, edge-computing products and other compact systems.

For a small robot like Microduck, the processor needs to perform several jobs without requiring the enormous computing resources used by large AI servers.

The robot needs to communicate with sensors.

It needs to process information.

It needs to control motors.

It needs to communicate wirelessly.

It needs to run its software reliably while operating from a relatively small battery.

The RK3566 therefore provides an important computing foundation for the robot.

Reports indicate that Microduck operates an onboard control loop at around 50 Hz, allowing the system to repeatedly process sensor information and determine actions.

That means the robot can continuously update its understanding of movement and respond to changes.

Why Use a Chinese Processor?

The use of the RK3566 raises an obvious question:

Why would a French robotics company choose a Chinese chip?

There are several possible practical reasons.

Cost is one.

Embedded processors produced at scale can be significantly more affordable than high-end computing platforms.

Availability is another factor.

A processor that is already widely used in embedded devices may offer a mature development ecosystem.

Power consumption is also important.

A small robot cannot carry a massive computer and battery.

Microduck has to balance processing power against size, weight, heat and battery life.

The RK3566 provides a relatively compact computing platform suitable for embedded applications.

The decision also demonstrates that advanced AI robotics does not always require an expensive GPU.

For many robotic functions, efficient edge computing can be sufficient.

Microduck’s Main Hardware

Microduck combines several hardware components into a very small body.

Among the important components are:

Rockchip RK3566 processor
Camera
LiDAR sensor
Motion sensors
15 motors
Wi-Fi connectivity
Battery
Articulated beak
Onboard storage
RAM
Speakers and microphones in its broader interaction system

Reports indicate that the robot has 1GB of RAM and 32GB of onboard storage.

Its small size means that every component must be carefully selected.

A larger robot can accommodate a larger battery, bigger cooling system and more powerful computer.

Microduck cannot.

Its design therefore focuses on efficiency.

How Does Microduck See?

Robots need some form of perception to interact with their environment.

Microduck uses a camera and LiDAR.

The camera provides visual information.

LiDAR provides distance information.

Together, these technologies can help the robot understand where objects are located.

For example, if Microduck is attempting to pick up an object, the camera can help identify what the object is while LiDAR can provide information about distance.

The robot can then coordinate its movement.

This is a simplified description of robotic perception, but it illustrates why combining multiple sensors is useful.

A camera alone does not always provide reliable depth information.

LiDAR can add another layer of environmental understanding.

What Can the Duck Robot Actually Do?

Microduck is not designed to replace a human worker.

It is too small and its lifting ability is limited.

Instead, it is a platform for experimentation.

Demonstrated behaviors include:

Walking

The robot can move on two legs using multiple motors.

Crouching

It can lower its body as part of its movement capabilities.

Recovering From Falls

One of its most interesting abilities is getting back up after falling.

This is an important robotics problem because maintaining balance on two legs is difficult.

Picking Up Small Objects

Its articulated beak works as a small gripper.

It can pick up lightweight objects from the floor.

Following a Laser

The robot can interact with a laser pointer and use its vision system to follow it.

Roller Skating

An accessory kit allows the robot to use roller-skating feet.

This is primarily an entertainment and experimentation feature, but it also demonstrates the flexibility of the platform.

Reinforcement Learning Is at the Heart of the Idea

One of the most important technologies behind Microduck is reinforcement learning.

Reinforcement learning is a machine-learning approach in which a system learns by interacting with an environment and receiving feedback about its actions.

In robotics, this can mean teaching a robot how to perform a movement by repeatedly trying different actions.

For example, imagine teaching a robot to stand.

The robot tries a movement.

If it falls, the system receives negative feedback.

If it maintains balance, the behavior receives a more favorable signal.

Over many iterations, the system can learn a strategy.

Doing this directly on a physical robot can take a long time.

That is why simulation is valuable.

Simulation Before Reality

Microduck's software ecosystem allows developers to train behaviors in simulation.

This is known as a sim-to-real workflow.

The idea is simple:

Train in simulation → Test → Transfer to robot → Collect data → Improve → Repeat.

Simulation allows developers to perform thousands or millions of trials without physically damaging the machine.

The approach has become increasingly important in modern robotics.

A simulated robot can fall thousands of times without needing a replacement motor.

A virtual environment can also be reset instantly.

This dramatically accelerates experimentation.

For developers, that could make Microduck a useful educational platform.

Open-Source Software Is a Major Selling Point

Hugging Face's identity is strongly connected to open-source AI.

Microduck follows that philosophy on the software side.

The SDK, simulation environment and reinforcement-learning training tools are designed to be accessible to developers.

The company wants users to create new behaviors rather than simply consume pre-installed features.

This distinction is important.

Traditional consumer robots often behave like closed appliances.

You buy the robot.

The manufacturer determines what it can do.

Software updates come from the company.

Microduck takes a different approach.

Developers can experiment with the system.

They can modify behavior.

They can train new skills.

They can share their work.

This could turn individual robots into part of a larger robotics community.

Is the Microduck Completely Open Source?

There is an important distinction.

The software ecosystem is open source, but the physical hardware is not necessarily fully open source.

Pollen Robotics has made its software stack available to developers, while reports indicate that the mechanical and electronic hardware designs are not being released in the same way.

This distinction matters because “open-source robot” can mean different things.

A robot may have:

Open-source software
Open-source simulation
Open-source training tools
Closed hardware
Partially open hardware

Microduck is primarily positioned as an open software and AI development platform.

That still provides significant advantages to developers, but it does not mean every physical component can be freely copied.

Why Open Robotics Could Matter

Open robotics could have an impact similar to open-source software.

When developers share improvements, the entire community can benefit.

One person might develop a better walking algorithm.

Another might create a voice interaction system.

A university could develop a navigation system.

A student could create an educational application.

Another researcher could improve object recognition.

The result could be a growing ecosystem around a relatively inexpensive physical platform.

That is precisely the kind of network effect Hugging Face has benefited from in AI.

The company is now attempting to bring that model into robotics.

Microduck and the Rise of Physical AI

Artificial intelligence has traditionally existed inside computers.

AI could write text.

Generate images.

Recognize speech.

Analyse data.

But physical AI takes the concept further.

A robot needs to perceive the physical world and act within it.

That requires combining:

Artificial intelligence
Computer vision
Sensors
Motors
Control systems
Machine learning
Mechanical engineering
Real-time computing

This combination is considerably more complicated than running a chatbot.

A language model can generate an answer in a fraction of a second.

A robot must decide how to move a physical joint without losing balance.

That is why robotics is often considered one of the next major frontiers for AI.

The Microduck Is Small, but the Idea Is Big

The Microduck itself is not a household humanoid assistant.

It cannot clean a house.

It cannot cook dinner.

It cannot carry heavy objects.

It is not intended to compete directly with large industrial robots.

Its importance is elsewhere.

It is a relatively inexpensive physical platform for experimenting with AI.

That could make it particularly attractive to people who want to learn robotics but cannot afford expensive research equipment.

A New Generation of AI Developers

The next generation of AI developers may need more than knowledge of Python and machine learning.

They may need to understand how AI interacts with the physical world.

Robotics provides that bridge.

A student can write code that changes the behavior of a real machine.

A researcher can train a model in simulation.

A developer can experiment with perception.

An educator can demonstrate reinforcement learning using a physical robot.

This makes platforms like Microduck potentially valuable educational tools.

Why the $399 Price Matters

Price is one of the most important elements of the Microduck story.

At $399, it sits closer to consumer electronics than traditional robotics equipment.

For comparison, research robots can cost thousands or tens of thousands of dollars.

A lower price does not automatically make a robot better.

But it can make experimentation more accessible.

That accessibility can create more developers.

More developers can create more software.

More software can create more use cases.

More use cases can make the hardware more attractive.

This creates a potentially powerful cycle.

Strong Early Demand

The initial sales figures suggest that there is already significant interest.

Business Insider reported that Microduck generated more than $2.6 million in sales within its first 24 hours, with demand producing a backlog and estimated waits of several months for new orders.

Later reports said pre-orders had surpassed 10,000 units, with Hugging Face co-founder Thomas Wolf saying the demand had reached the limit of the company's e-commerce interface at one point.

These numbers are notable because the product is aimed primarily at developers and robotics enthusiasts rather than the general mass market.

The strong response suggests that there is substantial interest in affordable physical AI.

The Supply Chain Could Become a Challenge

Rapid demand is good news for a technology company, but it creates another problem.

Can the company manufacture enough robots?

Microduck contains numerous components.

Its 15 motors alone represent a significant manufacturing requirement.

Other components include sensors, processors, batteries, mechanical parts and electronics.

If orders exceed production capacity, delivery times increase.

Reports have already indicated a backlog following the initial sales surge.

This is one of the challenges that successful hardware startups frequently face.

Selling thousands of units is very different from building thousands of units.

The Chinese Chip Connection Gets Attention

The Rockchip processor has become an especially interesting part of the story.

Rockchip is a Chinese semiconductor company with experience in embedded computing.

The fact that its RK3566 processor sits at the center of a highly visible European AI robotics project illustrates the interconnected nature of the global electronics industry.

It also shows that geopolitical competition in semiconductors does not eliminate international supply chains.

Companies may develop products in one country while depending on processors, sensors, motors and other components from several others.

The Microduck is a small physical example of this global technology network.

What Does This Mean for China’s Semiconductor Industry?

The Microduck story has also attracted attention in Chinese technology markets.

Reports said Rockchip shares rose sharply after news spread that the robot used the RK3566 processor. Chinese financial media linked the enthusiasm to the Microduck's strong early sales.

The reaction demonstrates how even a small robotics product can influence investor interest in component suppliers.

However, investors should distinguish between publicity surrounding a product and the long-term financial impact of one device.

A successful robot can create visibility for a chip company, but semiconductor businesses depend on much larger markets and sustained demand.

Privacy Questions Around a Camera-Equipped Robot

Microduck's capabilities also raise legitimate privacy questions.

The robot has a camera and other sensors.

Any device that can see, hear or interact with its surroundings can potentially collect sensitive information depending on how its software is configured.

Open-source software can improve transparency because developers can inspect and modify the code.

However, open source does not automatically guarantee privacy.

Third-party applications may potentially access sensors or send information to external services depending on how they are designed.

That means users should understand what software they install and what permissions it requires.

Open Source Does Not Automatically Mean Safe

This is an important distinction for consumers.

Open source can make software more transparent.

It can allow researchers to inspect code.

It can enable communities to identify vulnerabilities.

But openness does not eliminate security risks.

A poorly written application can still create privacy or security problems.

Developers building applications for robots therefore need to consider:

Camera access
Microphone access
Network connections
Stored data
Cloud services
Authentication
Software updates
Physical access to the robot

Security will become increasingly important as robots become more connected.

Microduck Versus Traditional Humanoid Robots

The robotics market is increasingly crowded.

Large companies are developing humanoid robots designed for factories and other commercial environments.

Other companies focus on robot dogs.

Microduck takes a different route.

It is small.

It is relatively inexpensive.

It has a distinctive design.

And it is explicitly designed as an experimentation platform.

Rather than trying to perform heavy industrial work, it focuses on accessibility.

That could give it a unique position.

Could Microduck Become a Home Robot?

Possibly, but not in its current form.

The robot is too small and limited to become a general-purpose household worker.

Its lifting capacity is limited.

Its battery life is also constrained.

Its physical size limits what it can manipulate.

But the technology could evolve.

Future versions could become stronger.

Sensors could improve.

Batteries could become more efficient.

AI models could become smaller and more capable.

Robots could learn more complex tasks.

Microduck may therefore be better understood as an early platform rather than the final form of a household robot.

The Future of Small AI Robots

Small robots could become increasingly important.

Large humanoids require expensive hardware and sophisticated manufacturing.

Small robots can be cheaper to produce, easier to distribute and easier for developers to experiment with.

They can also provide valuable data for robotics research.

If thousands of people operate similar robots in different environments, developers could potentially learn from a broad range of real-world experiences.

That could accelerate progress in physical AI.

What Makes Microduck Different?

Several characteristics distinguish Microduck from many other robots.

Affordable

The $399 starting price makes experimentation more accessible.

Open Software

Developers can work with the software and training ecosystem.

Compact

At roughly 25 centimetres tall, it can operate in small spaces.

Programmable

Users can train and modify behaviors.

Sensor-Rich

Camera and LiDAR provide environmental information.

Community-Oriented

The platform is designed to encourage users to share skills.

Educational

It can be used to demonstrate robotics and machine learning concepts.

What Are the Limitations?

Despite the excitement, Microduck has clear limitations.

It is not a general-purpose AI assistant.

Its battery life is limited.

Its physical strength is low.

Its small size restricts what it can carry.

Its hardware is not fully open source.

Developers may also need technical knowledge to take full advantage of the platform.

For someone who simply wants a plug-and-play home robot, Microduck may not be the best choice.

For someone interested in AI, coding and robotics, however, those limitations may be less important.

Who Is Microduck For?

The robot appears particularly suitable for:

AI developers:
Developers can experiment with physical AI.

Robotics researchers:
Researchers can test algorithms on affordable hardware.

Students:
Students can learn programming and robotics.

Educators:
Teachers can use the platform for practical demonstrations.

Hobbyists:
Robotics enthusiasts can create custom behaviors.

AI communities:
Developers can share models, behaviors and experiments.

The product is therefore closer to a development platform than a traditional toy.

Why Hugging Face Is Entering Robotics

Hugging Face has built a large ecosystem around open AI models.

The company's expansion into robotics makes strategic sense.

AI models need environments in which they can interact with the physical world.

Robotics provides that environment.

The company's acquisition of Pollen Robotics in 2025 helped establish a physical AI division.

The earlier Reachy Mini project provided another step toward affordable open-source robotics.

Microduck pushes the idea further by combining a lower price with a playful design.

From Reachy Mini to Microduck

Microduck is not Hugging Face's first robot.

The company previously introduced Reachy Mini, an expressive desktop robot aimed at developers and AI builders.

The Reachy Mini comes in different versions, including a wireless model with onboard computing and a Lite model that uses an external computer. Official documentation describes it as an open-source platform for AI experimentation, programming and human-robot interaction.

Microduck expands this philosophy into a more mobile and physically active design.

Instead of primarily moving its head and body on a desktop, Microduck can walk around.

That introduces a new set of challenges involving balance, locomotion and physical interaction.

Could a Robot Community Change AI Development?

This may be the most important question.

The success of Hugging Face's AI platform has been driven partly by community participation.

If that model works in robotics, developers could collectively build a library of robotic behaviors.

Imagine a future in which a user downloads a navigation skill.

Another downloads an object-recognition skill.

Another installs a voice assistant.

A fourth combines several skills into a new application.

The robot becomes a physical platform that evolves through community software.

That is fundamentally different from traditional closed robotics.

The Bigger Picture: AI Is Moving Into the Physical World

The Microduck story reflects a broader shift in technology.

For years, artificial intelligence primarily lived inside computers and smartphones.

Now AI is increasingly being placed inside physical devices.

Cars use AI.

Cameras use AI.

Drones use AI.

Industrial machines use AI.

Robots use AI.

The next stage could involve AI systems that continuously perceive and interact with their environments.

That is often described as embodied AI or physical AI.

Microduck is a small example of that larger trend.

What Could Happen Next?

The next important stage will be seeing what developers actually build.

A robot can have impressive hardware.

But its long-term value may depend on software.

If developers create useful applications, Microduck could develop a strong ecosystem.

If the community remains small, the product could remain primarily a niche robotics platform.

The company's open-source strategy is therefore a bet on community participation.

The early sales numbers suggest that there is already significant interest.

The next question is whether that interest turns into sustained development.

The Role of Education

Microduck could also have a significant educational role.

Robotics is often difficult for beginners because it combines multiple disciplines.

Students need to learn programming, electronics, mechanics and machine learning.

A compact robot can make these concepts tangible.

Instead of simply watching a simulation on a screen, students can write code and observe a physical machine respond.

That can make abstract AI concepts easier to understand.

It can also encourage students to experiment.

A Possible Gateway Into Robotics

The low price could make Microduck a gateway product.

Someone might start by programming a simple movement.

Then they might learn how sensors work.

Later, they could experiment with reinforcement learning.

Eventually, they could build a more sophisticated robotic application.

This progression could help create a new generation of robotics developers.

That may be more important to Hugging Face than selling the robot itself.

The company is potentially building a community around physical AI.

What the Microduck Story Really Tells Us

The popularity of Microduck says something interesting about today's technology market.

People are not only interested in increasingly powerful AI models.

They are also interested in giving those models bodies.

The robot's appearance makes the concept approachable.

Its price makes experimentation possible.

Its open-source software makes customization attractive.

And its Chinese processor demonstrates how global the modern technology supply chain has become.

The result is a product that sits at the intersection of several major technology trends.

Final Analysis

Hugging Face's Microduck may look like a small duck-shaped robot, but its significance goes far beyond its appearance.

At around 25 centimetres tall and priced at $399, it is designed to make physical AI experimentation accessible to a much wider audience.

Its camera, LiDAR, motors and onboard computing allow it to interact with the physical world.

Its software ecosystem allows developers to train new behaviors using reinforcement learning and simulation.

And at the centre of its hardware is the Chinese Rockchip RK3566 processor, showing once again how modern technology products depend on international supply chains.

The early commercial response has been remarkable.

Reports indicate that Microduck generated more than $2.6 million in sales during its first 24 hours, while later reports placed pre-orders above 10,000 units.

That level of interest suggests that there is a growing market for affordable AI-powered robots that users can actually program and modify.

But Microduck is not a replacement for a human.

It is not a household servant.

It is not a powerful industrial robot.

Its strength lies in experimentation.

The machine provides a physical body through which developers can explore the relationship between artificial intelligence, sensors, software and movement.

That could make Microduck an important educational and research platform.

The larger story is therefore not really about a robot shaped like a duck.

It is about the possibility of making robotics as accessible as software development.

If Hugging Face succeeds, future AI developers may not simply download models and run them on computers.

They may download skills and teach robots how to move, see, interact and learn.

And that would represent an important step toward a world in which artificial intelligence does not merely answer questions on a screen but can also understand and act within the physical world.

For now, the little Microduck is taking its first steps.

But the technology trend behind it is much bigger.

Frequently Asked Questions

What is Hugging Face Microduck?

Microduck is a small, bipedal open-software robot developed by Pollen Robotics, a robotics company owned by Hugging Face. It is designed for AI experimentation, robotics development, reinforcement learning and education.

How much does Microduck cost?

The announced price is $399 before applicable taxes and shipping.

How tall is Microduck?

Microduck is approximately 25 centimetres tall, or about 10 inches.

What chip powers Microduck?

The robot uses the Rockchip RK3566, a processor developed by Chinese semiconductor company Rockchip. Recent reports identify it as the robot's main control processor.

Is Microduck completely open source?

Its software, SDK and reinforcement-learning ecosystem are designed to be open source. However, reports indicate that the physical hardware designs are not fully open source.

What can Microduck do?

It can walk, crouch, recover from falls, interact with lightweight objects using its beak-like gripper and perform experimental behaviors. It can also be equipped with roller-skating accessories.

Does Microduck use AI?

Yes. The robot is designed for physical AI and reinforcement-learning experimentation. Developers can train behaviors in simulation and transfer them to the physical robot.

Is Microduck a household robot?

Not in the traditional sense. Its small size and limited strength make it more suitable as an educational, research and development platform.

Why is Microduck attracting so much attention?

Its combination of a $399 price, unusual design, open software, programmable behavior and AI capabilities has attracted developers, robotics enthusiasts and the wider technology community.

How many Microduck robots have been ordered?

Reports published around August 31 and September 1, 2026 said pre-orders had exceeded 10,000 units, while earlier reporting said sales had surpassed $2.6 million within the first 24 hours.

When will Microduck ship?

Pollen Robotics says the first Microduck deliveries are targeted for before Christmas 2026.

Top News

Hugging Face’s New Duck Robot Is Selling Fast — A Chinese Chip Powers It