Robot Profile

Reachy Mini

An open-source desktop robot from Pollen Robotics and Hugging Face — a low-cost embodied AI platform for developers, educators, and researchers.

ManufacturerPollen Robotics / Hugging Face
TypeDesktop Companion Robot
CountryFrance
Price Range$299–$449
Year Introduced2025
AutonomyPlatform-Configurable
IndustryEducation / Research

Overview

Reachy Mini is a compact, open-source desktop robot built by Pollen Robotics and brought into the Hugging Face ecosystem following the company's 2025 acquisition. At 28 cm tall and starting at $299, it is the most accessible entry point into embodied AI development — a physical interface for LLMs, voice agents, and multimodal AI workflows. Its six-degree-of-freedom head, rotating base, wide-angle camera, and four-microphone array give AI models a body to inhabit. Two versions exist: the Lite, wired and requiring a host computer, and the Wireless, which runs a Raspberry Pi 5 onboard and operates on battery. The human experience of interacting with Reachy Mini is largely what the developer makes it — but the platform's openness, deep Hugging Face ecosystem integration, and explicitly AI-native SDK make it one of the most consequential developer robotics launches in years.

RXD Score

3.00 / 5.0

Functional

Reachy Mini

SignalClaritySpatialLegibilityPerceivedPresenceFailureTransparencyInteractionFitRecoveryDesign

3.00 / 5.0Functional

Signal Clarity0.0 / 5

Head movement and antenna animation communicate emotional state, but signal vocabulary is entirely developer-defined — no standardized cues exist out of the box.

Spatial Legibility0.0 / 5

Constrained desktop form factor makes movement predictable. Head tracking and base rotation follow clear directional logic; the limited motion envelope works in its favor.

Perceived Presence0.0 / 5

Charming, coherent physical design with readable expressiveness. Behavioral identity is blank-slate by design — persona depends entirely on what the developer loads.

Failure Transparency0.0 / 5

Platform-level failures produce silence or stillness with no user-facing explanation. Error handling is developer-defined; none exists by default.

Interaction Fit0.0 / 5

Precisely matched to its intended audience — developers and researchers working in Python, ROS2, and the Hugging Face ecosystem. Fails hard for non-technical users, by design.

Recovery Design0.0 / 5

Recovery is SSH and terminal-only. The open-source stack empowers developer recovery but provides no path for non-technical users encountering failures.

Deployment Context

Operational Context

Desktop research and development

The robot sits on a desk or table, interacting at close range with one or a few people. It is not a mobile platform — it does not navigate or roam. Most interactions happen in developer workspaces, classrooms, or research labs, typically at seated eye level. The Wireless version can be moved between locations; the Lite version stays tethered to a host computer.

User Population

Developers, researchers, and educators

Primary users are technically proficient — developers working in Python and ROS2, AI/ML researchers exploring embodied interaction, and educators building robotics or AI curricula. Non-technical users, including students and demo audiences, encounter the robot as bystanders or in guided interactions. Their experience depends entirely on what the primary user has built into the robot.

Friction Points

Early platform, opaque failures

Documentation is incomplete and APIs are still shifting between releases. Non-developer users hit a wall immediately — there is no onboarding layer for end users. When things fail, the robot goes silent. Nothing communicates what went wrong or what to do next. The gap between a configured, expressive robot and a freshly unboxed one is enormous and falls entirely on the developer to bridge.

Field Observations

Small motion, strong attribution

Even with limited degrees of freedom, users project significant intent onto the robot. Antenna movement and head orientation alone were sufficient for observers to attribute curiosity, acknowledgment, and attentiveness — consistent with findings in low-DoF social robotics research. The "Pixar lamp" effect is real: timing and direction of motion matter more than mechanical complexity. When the robot speaks and looks toward the user simultaneously, perceived presence jumps noticeably.

Founding Cohort · May 2026

Ten seats. One rate. First in.

The first REP cohort runs the first week of May 2026. This is the only time the program will run at this rate — and the only cohort where founding members get direct access to the instructor throughout.

No engineering background. No prerequisites. Just the credential the field is missing.

$199full accessor two payments of $99
10 seats available · Founding rate ends when the cohort fills