Hugging Face builds open robotics platform

This turns Hugging Face from a software shelf for AI models into a full embodied AI platform that can sell the robot, host the training data, and distribute the control model in one loop. Pollen gave it a native hardware line, but the bigger shift is control over data creation. A lab can buy Reachy hardware, collect demonstrations, upload datasets to the Hub, train with LeRobot, and share checkpoints back into the same ecosystem. That is the same flywheel that made Hugging Face central in language models, now applied to robots.1 2 3 4

• The product line covers nearly every buyer tier. Reachy 2 serves labs and enterprises, Reachy Mini targets developers and consumers, and SO-101 brings entry cost down to around $100. That breadth matters because each device is also a data collection endpoint, not just a hardware sale.1 3 5
• LeRobot is the glue. By v0.5.0 it supported third party hardware like Unitree G1 and connected into NVIDIA IsaacLab-Arena, which means Hugging Face is not limiting robotics to its own robots. It is trying to become the default open workflow where developers calibrate robots, train policies, and benchmark them across embodiments.2 6 7
• There is a useful parallel with Agile Robots buying Franka. In both cases, owning hardware gives the platform owner first access to the machines that generate training data. The difference is that Hugging Face is pairing that with an existing open repository and community, which helps explain why robotics datasets on the Hub exploded in 2025.4 8

The next step is a standard stack for open robotics, where low cost devices seed a huge developer base and higher end robots pull in labs and commercial teams. If that happens, Hugging Face becomes the default place not just to download robot models, but to create the data, tools, and hardware conventions the rest of embodied AI builds on.1 2 3 4 6