Face searching robotic arm

This is a project of a robotic arm which uses 2 cameras to search for a face and once it finds it locks on to it. I showed this at a presentation at the Bletchley Park AI user group. There is a demo video in this repo to get an idea of how it is made and works.

It uses both a camera and a 3D sensor of a Kinect 360. These are great as they include a camera and a depth sensor and they are much cheaper than RGBD cameras. When the Kinect finds a face, it proceeds to move the arm until it is able to find it again using a camera at the tip of the arm. Once this 2nd camera finds the face, it locks onto it and stops searching.

It employs an Nvidia Jetson Nano 4 Gb attached to a PCA9685 board controlling 6 servos. Software developed using Python and the libraries Adafruit ServoKit for the servos, OpenCV for the camera, and Freenect for the Kinect.

The system works by recognizing faces using the Xbox 360 camera. Once it finds it, it moves the robotic arm until the 2nd camera attached to the tip of the robot arm finds it too. Once that happens, it stops. The AI face recognition is a cheesy Haar Cascade (yup, that stuff is still around lol), as you can’t ask much more from a humble Jetson :D

Hardware Required:

  • Nvidia Jetson Nano with a 4 Amp power adapter plugged to the barrel connector. Do not try to feed it through the USB connector as it does not provide enough power. Remember to set up the pins to use the barrel connector
  • SD card with Ubuntu 20.4 Focal installed that comes with Jetpack 4.6. Do NOT install any other Ubuntu as the camera will not work
  • Robotic Arm kit with 6 axis
  • 6 servos
  • PCA9685 controller plus a 6V 3 Amp power supply: you can just use a mobile charger using the red and black wires of the USB cable
  • Xbox Kinect 360 (NOT the Kinect One: even though the code implements also the V2, it does not use it at this moment)
  • Camera IMX219 with a long flat cable

You will have to install the libraries Adafruit ServoKit, OpenCV, and Freenect V1. There are excellent guides at this link for the Kinect and this link for the controller and servos.


  1. Make the other axis work, as now it only works with 1.
  2. Use the Kinect’s depth sensor to measure the distance of the face and move the axis accordingly.
  3. Recode this in C++ or Rust in order to make it faster.
  4. Use the Kinect V2 to improve recognition

The code and instructions can be seen at GitHub - Sequential-circuits/facesearcher: Robotic arm with 2 cameras which searches for a face and once it finds it locks on to it