# Building a Face-Tracking Robot Head with OAK 4 FFC

This example shows how the OAK 4 FFC can control actuators in a robotics project. We mounted two IMX486 autofocus camera modules
inside a 3D-printed humanoid face, put it on an off-the-shelf servo platform, and let the robot head turn to follow a person's
face, with all perception running directly on the device.

The build is a walkthrough and starting point, not a finished product. The face-tracking app was generated by an AI coding agent
with Luxonis Agent Skills, and the setup is meant as inspiration for your own robot heads, pan-tilt trackers, or other actuated
vision projects.

### Materials used

 * OAK 4 FFC Devkit (includes the PoE module and the M12 PoE cable)
 * 2× Luxonis IMX486 autofocus FFC camera modules
 * 2× 30 cm FFC ribbon cables (the cables included with the modules are too short for a moving head)
 * Off-the-shelf 2-axis servo platform with a servo control board
 * USB-C cable (OAK 4 FFC to servo control board)
 * 30 W PoE+ switch
 * 3D-printed humanoid face (front shell and back plate)

## System Overview

 * On-device perception Person and face detection, a high-resolution face close-up, and stereo depth all run on the OAK 4 FFC
 * Stereo from two modules One camera is enough for face tracking; the second forms a stereo pair for a live 3D point cloud
 * Always in focus Autofocus keeps the image sharp whether the subject is close to or far from the head
 * Direct actuator control A single USB-C cable connects the OAK 4 FFC to the servo board for both power and serial data
 * Single power cable The whole robot runs off one PoE+ port through the devkit's M12 cable
 * No calibration targets Camera extrinsics are set in OAK Viewer by entering the cameras' relative positions, then refined
   automatically by self-calibration thanks to the overlapping views
 * Room to grow Only two of the five camera connectors are used, leaving space for more cameras

## Building the App with an AI Coding Agent

With the [Luxonis Agent Skills](https://docs.luxonis.com/software-v3/agent-tools/agent-skills) installed, the coding agent
detected the connected OAK 4 FFC and built the app in two stages from a plain-language prompt:

 1. Servo control Drive the servo board over serial and steer the head toward the first detected person.
 2. Perception Detect people and faces, crop a high-resolution face close-up, and build a point cloud from the stereo cameras.

The app was built and deployed directly to the device in a couple of minutes.

## Typical Use Cases

 * Social and service robots that make eye contact with people
 * Pan-tilt camera heads that follow a subject
 * Humanoid robot research platforms
 * Interactive kiosks and telepresence

## Coming Soon

> **Video and blogpost coming soon**
> The build video and the full step-by-step blogpost, including the complete bill of materials, 3D-print files, and app source code, will be linked here soon.