Hugging Face Drives Open Source Robotics Forward with Pollen Acquisition
–
Hugging Face, known for hosting open source AI models and software, has entered into an agreement to acquire Pollen Robotics, a French startup behind a distinctive two-armed, bug‐eyed humanoid robot named Reachy 2. The company announced that it will market the robot and simultaneously grant developers full access to its source code so that they can download, modify, and offer suggestions for improvement. This step is part of a broader effort to extend participation in robotics and to make the technology more accessible to a wider range of users and innovators.
In a statement released today, Clément Delangue, chief executive of Hugging Face, emphasized the need for openness in robotics. He stated, "It’s really important for robotics to be as open source as possible. When you think about physical objects doing physical things at work and at home, the level of trust and transparency I need is much higher than for something I chat with on my laptop." Alongside Delangue, research engineers Simon Alibert and Rémi Cadene are actively working on integrating artificial intelligence across various robotic applications at the company.
Footage shared by Pollen Robotics shows Reachy 2 executing everyday tasks such as arranging coffee mugs and handling pieces of fruit. Matthieu Lapeyre, cofounder and CEO of Pollen Robotics, mentioned that several leading AI companies are currently employing Reachy 2 to conduct research on robotic manipulation. Although he could not disclose the names of these companies due to confidentiality agreements, Lapeyre made it clear that the long‐term vision is for future versions of Reachy 2 to be used in home environments, demonstrating both the practical applications and the potential for further development of domestic robotic technology.
Lapeyre also acknowledged that selling humanoid robots remains a challenge. He explained that while there have been isolated cases of success, the main difficulties lie in defining clear use cases and dealing with systems that still lack consistency. At present, much of the progress in this technology is driven by a handful of well-funded firms such as Tesla, Figure, and Agility Robotics. "With Hugging Face, we hope to democratize that," Lapeyre remarked, outlining his ambition to broaden the field of robotics by providing more open access to innovative hardware and related software.
The principles of open source have already had a strong influence in the world of artificial intelligence. Many of the models, frameworks, and tools used by researchers and engineers are distributed free of charge under licenses that permit modification and reuse. In the software industry, this model has promoted rapid testing and improvement of code. Transferring these ideas to hardware involves releasing detailed design schematics, component specifications, and 3D models that simplify the manufacturing process. In doing so, robotics developers can, for example, manufacture replacement parts or even enhance a component’s design when necessary.
Delangue expressed optimism that adopting open source practices in the robotics sector could lead to a broad range of practical capabilities. He suggested that by allowing a wide community of developers to contribute, the field could benefit from many new and improved functionalities. Lapeyre supported this view by noting the benefits of open hardware practices. "If something is broken, they can [3D] print a part. If something is not perfect, they can make it a bit better by adding a new part," he explained. This ability to independently repair and modify components is seen as a vital mechanism for speeding up innovation and adaptation in robotic systems.
The surge in interest around artificial intelligence has also rekindled focus on robotics. Recent advancements in AI are opening fresh possibilities for physical systems, with some experts contending that achieving higher levels of machine intelligence may require interaction with the physical world. Such a connection could help machines gain a more practical understanding of real-world environments and challenges. At the same time, the enthusiasm for humanoid robots has spurred claims that sometimes appear exaggerated, prompting calls within the industry for a more grounded evaluation of robotic performance.
A number of companies have shared demonstration videos on social media that seem to showcase extraordinary robotic capabilities. Some experts caution that these displays might not accurately represent true performance levels; a robot that appears highly capable online may actually be controlled remotely, could struggle under slight changes in context, or might not reliably execute tasks under real-world conditions. In this context, Delangue underscored the merit of open source practices by remarking, "You can’t cheat, you can’t hide with open source." His comment reinforces the belief that transparent development practices help build a trustworthy foundation for the evolution of robotics.
Interest in robotics-related code hosted on Hugging Face has significantly increased over the past year, a trend that mirrors a growing fascination with robotic systems among both researchers and practitioners. Many in academic circles advocate for an open approach to robotics development. Sergey Levine, an assistant professor at the University of California, Berkeley, and cofounder of Physical Intelligence, stated, "Making robotics more accessible increases the velocity with which technology advances." His startup has been at the forefront of this effort, having introduced its robot foundation model, Pi0, on Hugging Face earlier this year. The Pi0 model is designed to enable a variety of robots to learn and perform multiple physical tasks, benefitting from continual improvements contributed by the research community.
Levine further noted that active collaboration between academia and industry has already led to several enhancements in his foundation model. He pointed out that the potential for individuals outside of established companies to propose new hardware designs is immense. "There's a lot more creativity people can apply to how they build the actual physical hardware," he commented. This perspective reflects a broader trend in robotics, where shared development and cooperative problem-solving are playing an increasing role in advancing the technology.
The open source strategy in robotics is beginning to influence the broader field of artificial intelligence. Meta initiated this shift by releasing Llama, a notable open weight model, earlier in the year, setting an example that has been followed by several subsequent models. In January, a relatively unknown Chinese startup called DeepSeek attracted attention when it introduced a powerful AI model that was reportedly developed with lower financial outlay than similar systems produced by major U.S. companies. OpenAI, which had been closely guarding its most advanced models, recently announced plans to adjust its approach by offering a free, open weight model later this summer. These developments signal a growing willingness among leading firms to make technology more accessible and to invite contributions from a wider community.
