Reachy Mini - ROBOTS: Your Guide to the World of Robotics

Reachy Mini is an expressive companion robot designed for human–robot interaction, creative coding, and AI experimentation. Built as a low-cost, accessible platform, it offers a hands-on way to explore robots powered by modern AI models and supported by a growing open-source robotics community.
(AI company Hugging Face acquired Pollen Robotics in 2025.)
Pollen Robotics designed Reachy Mini's assembly process to be as smooth and intuitive as building a Lego set.
Pollen Robotics began in 2016 in Bordeaux, France, when Matthieu Lapeyre and Pierre Rouanet, both former developmental robotics researchers at INRIA's Flowers Project team, left academia to found a company focused on practical, open robotics platforms. The founders had previously collaborated for more than a decade on research initiatives—most notably the Poppy open-source humanoid—giving them deep experience in modular robot design and open engineering. From the outset, the company positioned itself at the intersection of machine intelligence and human-robot interaction, pursuing hardware and software that could be inspected, modified, and extended by developers and researchers outside traditional industrial robotics labs.
Over the subsequent years, Pollen's work coalesced around its Reachy family of robots: expressive humanoid platforms with integrated perception, manipulation, and interaction capabilities. Early versions, like the original Reachy introduced in 2020, were adopted by research labs for studies in teleoperation and embodied AI, and later variants like Reachy 2 featured enhanced vision and grasping hardware aimed at real-world manipulation tasks. These platforms embodied the company's open-source philosophy—both hardware designs and software stacks were released under licenses such as Apache 2.0 or Creative Commons, intended to ease community contributions and customization.
In 2025, Pollen Robotics was acquired by Hugging Face, the AI startup known for its expansive repository of open-source models and tools. The acquisition marked one of Hugging Face's first moves into physical robotics and embodied AI, aligning two communities that emphasize transparency and collaboration in technology. Under the new ownership, Pollen's platform strategy broadened: alongside larger humanoid robots like Reachy 2, the company and Hugging Face jointly introduced Reachy Mini, a compact open-source desktop robot designed primarily for AI experimentation. Reachy Mini aims to lower the barrier to entry for robotics development by combining approachable hardware with a software ecosystem tied to Hugging Face's model hub and community tooling.
Reachy Mini's design reflects its dual engineering and educational goals. Built as an open-source platform, the robot exposes its hardware and software interfaces so that developers can prototype AI applications, integrate perceptual models, and experiment with embodied interaction in Python (with JavaScript and Scratch support forthcoming). Sensors such as cameras and microphones, as well as a combination of actuators, allow for multimodal input and interaction, while an ecosystem of apps and developer tools hosted on Hugging Face's platform enables iterative experimentation and sharing. The approach reflects the company’s long-standing focus on modular robots that are easy to modify and extend—and that show a surprising amount of personality.
Released in 2020, the original Reachy was an open-source humanoid robot designed for teleoperation and human-robot interaction, and later earned second place in the ANA Avatar XPrize.
Reachy 2, unveiled in 2024, is a fully open-source humanoid robot built for embodied AI research, human-robot interaction, and real-world applications such as teleoperation and assistive tasks.
Compact, low-cost design. Open source, simulation ready, supports Python.
huggingface.co/reachy-mini
Raspberry Pi camera v3 wide angle (Sony IMX708, 12MP, Auto focus, I2C*~1 x MIDI DSI connection).
Four PDM MEMS digital microphones (16 kHz max sample rate / -26 dB FS Sensitivity / 64 dBA SNR, based on Seeed Studio's reSpeaker XMOS XVF3800).
Base: 1x custom Dynamixel XC330-M288-PG (XC330-M288-T with plastic gear)
Antennas: 2x Dynamixel XL330-M077-T
Stewart Platform: 6x Dynamixel XL330-M288-T
Injection-molded and 3D-printed plastic structural parts (ABS / PLA / PETG).
Aluminum elements for internal structural support and motor mounting. Steel fasteners and shafts.
Electronic components mounted on custom PCBs.
Silicone and rubber elements for cable management and vibration damping.
Control: Raspberry 4 Compute Module - CM4104016 (Wi-Fi, 4GB RAM, 16GB Flash)
Wi-Fi antenna - 2.4-5GHz dual-band patch antenna, 2.79 dBi, omnidirectionnal.
USB-C output (for external devices like USB thumb drives).
Operating system: Linux-based system (Ubuntu Linux recommended) running on an external computer or onboard compute module depending on configuration.
Programming language:Python (primary control and application layer).
Robot control stack:Open-source Python SDK developed by Pollen Robotics for motor control, sensors, and behaviors.
Input voltage: 6.8 - 7.6V.
LiFePO4 Battery, 2000mAh, 6,4V, 12,8Wh, over charge protection, over discharge protection, over current protection, short protection, temperature sensor.