Hugging Face and Reachy Mini: A Revolution in AI Robotics

Hugging Face: Pioneer of Open Source Robotics

Hugging Face, known for its significant contributions to open source artificial intelligence, is now venturing into the field of robotics. By adopting a similar approach to the one that led to its success in developing AI models, the company is expanding its scope with a range of robots, from conversational assistants to robotic arms.

Jensen Huang and Reachy Mini: A Showcase of Choice

At the latest Consumer Electronics Show (CES) in Las Vegas, Jensen Huang, the charismatic founder of Nvidia, highlighted Reachy Mini, a 28 cm tall robot designed by Hugging Face. This small robot, with its asymmetrical eyes and antennas, immediately captivated the audience. Huang, true to his style and dressed in leather, demonstrated how this robot can become an ideal work partner, capable of discussing various topics thanks to its natural language capabilities.

Hugging Face: From AI to Robotics

Hugging Face, often compared to GitHub for AI, was founded ten years ago by three French individuals. The company is recognized for its ability to facilitate the adoption of open source AI models by businesses. In May 2024, under the leadership of Rémi Cadene, a former engineer from Tesla's Optimus program, Hugging Face launched LeRobot, an open source library based on PyTorch. This platform, which has already garnered 21,100 stars and 3,500 forks on GitHub, aims to democratize access to learning models in robotics.

Acquisition of Pollen Robotics and Launch of Reachy 2

In April 2025, Hugging Face took a significant step by acquiring Pollen Robotics, a Bordeaux-based startup founded by Matthieu Lapeyre and Pierre Rouanet. This acquisition allowed Hugging Face to take ownership of Reachy 2, an advanced humanoid robot. Equipped with a mobile base featuring three omni-wheels, a LiDAR system, and articulated arms, Reachy 2 can support loads of up to 3 kg. This robot, priced at around $70,000, is intended for research and development teams as well as laboratories. Additionally, Reachy 2 is teleoperable, allowing an operator to control it remotely using a virtual reality headset, enabling it to learn new movements and become more autonomous through contextual analysis.

Reachy Mini: A Robot for Makers

Reachy Mini, on the other hand, is designed to appeal to robotics enthusiasts, the "makers." Available in two versions, a lite version at $300 connected via USB-C and a wireless version with Raspberry Pi 5 at $500, this robot is accessible and customizable. Rémi Fabre, an engineer at Pollen Robotics, explains that it takes about two hours to assemble the robot, providing users with an enriching and educational experience.

The Importance of Emotions and Cognitive Abilities

Pollen Robotics has also integrated elements of empathy into Reachy Mini. The robot is capable of looking at its interlocutor and moving its head expressively, making interactions more lively. This approach, known as "cognitive robotics," allows the robot to simulate emotions, thereby enriching the user experience.

A Multifunctional Robot

Reachy Mini is equipped with an app store that allows it to transform into a chess teacher, English coach, or even a smart home hub. Users can create and share applications without programming skills, thus expanding the robot's usage possibilities.

SO-101: A Robotic Arm for Developers

In addition to the Reachy robots, Hugging Face offers the SO-101, an open source robotic arm aimed at AI developers. Priced between $100 and $500 depending on the assembly level, this arm allows users to learn specific tasks through reinforcement learning.

Open Source Physical AI: A New Paradigm

Hugging Face's robots illustrate the application of generative AI to robotics, a concept known as physical AI. These machines, equipped with cognitive abilities, can perceive and interact with the real world. Hugging Face has developed SmolVLA, a Vision Language Action (VLA) model that translates visual and textual data into motor commands, operating locally to reduce reliance on cloud services. SmolVLA is trained on community data from the LeRobot platform, enhancing its ability to adapt to various environments.

An Alternative to Robotics Giants

By promoting open source physical AI, Hugging Face positions itself as an alternative to the proprietary models of giants like Tesla Optimus, Boston Dynamics, Xiaomi, and Unitree. This approach offers valuable technological sovereignty in the current geopolitical context. Rémi Fabre comments on the ideological opposition between using a generic foundation model or a set of specific models, emphasizing the importance of local inference to maintain data control.

Strategic Partnership with Nvidia

To strengthen its ecosystem, Hugging Face has formed a strategic partnership with Nvidia, integrating its Omniverse and Isaac Lab robotics technologies with the LeRobot framework. These tools enable the generation of synthetic data and the simulation of robotic models before deployment.

UMA: A New Adventure in Robotics

Last December, Rémi Cadène and Simon Alibert, co-founders of LeRobot, launched UMA, a startup dedicated to creating humanoid robots. Supported by influential figures in AI such as Yann LeCun and Thomas Wolf, as well as investor Xavier Niel, UMA aims to develop robots for industrial and domestic applications. UMA was established alongside Pierre Sermanet, a former Google DeepMind employee, and Robert Knight, a former member of The Robot Studio, thus bolstering its expertise in the field.

Robots for Industry and Daily Life

UMA aims to design advanced AI in the physical world and build humanoid robots ready to work in real-world environments at scale. These developments seek to address the needs of logistics, healthcare, and elderly assistance, opening new perspectives for robotics.