Abstract:
Hugging Face, an open source artificial intelligence platform, recently released a small open source duck-shaped robot called Microduck, priced at US$399 and expected to be shipped before Christmas. This product is aimed at developers and focuses on low-threshold robot training and deployment capabilities.
Hugging Face CEO Clem Delangue said Microduck is an open source robot that can learn new skills through reinforcement learning. The robot is about 25 centimeters tall and can waddle around, pick up items with its beak, get up on its own after falling, squat, and even put on roller skates to move around. Among them, the maximum pick-up weight of its beak is 800 grams.
Delangue said that the launch of Microduck means that the "era of affordable open source robots" is coming, with the goal of allowing more people to participate in the development of physical AI and world models.
Hugging Face was previously mainly known as a platform and developer community for open model weights. In order to enter the field of more friendly and lower-cost open source AI hardware, the company acquired French start-up Pollen Robotics in April 2025. A few months later, the two parties launched the desktop robot Reachy Mini. Currently, the Reachy Mini with a Raspberry Pi computer costs $499, while the Reachy Mini Lite, which needs to be connected to a Mac or PC, costs $399.
In terms of perception capabilities, Microduck is equipped with cameras, lidar sensors and two inertial measurement units, which can be used to sense the surrounding environment and its own motion status. Pollen Robotics said that developers can first train Microduck's behavior in a simulation environment and then deploy the model directly to the physical robot; the robot can then be fine-tuned, retrained and deployed again. Its software development toolkit, simulation environment and complete reinforcement learning training technology stack have been made public on GitHub.
However, the entry of camera-equipped robots into private spaces such as bedrooms may also cause users to worry about privacy. Delangue has previously said that compared to "black box" systems controlled by a few institutions, robots running on open source models have more privacy advantages because developers and users can review and master their technical implementation.
But open source does not necessarily mean that sensitive data is absolutely safe. As consumers install various applications on robots, these applications may still have access to the device's camera and microphone; if the application itself is designed to transmit data to external services, relevant information may still be uploaded to third-party platforms.
At the time Microduck was released, there was news that Hugging Face might be acquired by Nvidia at a valuation of US$13 billion. The two parties have been working together for many years, with Nvidia providing infrastructure support for Hugging Face since at least 2023, and the two companies have also continued to publicly advocate the development of open source AI.
Recently, Hugging Face has also received attention due to a network security incident. According to reports, OpenAI broke through Hugging Face’s sandbox environment during security testing and entered the platform’s servers.
Comments