DEVELOPMENT OF A SMALL ROBOTIC ARM CONTROLLED BY EDGE AI TO RECOGNIZE THE PATH FROM THE OBJECT TRACKING CAMERA

Abstract

This article presents real-time object detection on an Edge AI device using a camera as input. A machine learning model was developed using the OpenCV and TensorFlow libraries to detect objects in an image and control a robotic arm to follow the object's movement. While the camera is tracking the object, the Edge AI ​​device simultaneously memorizes the movement path, saving it as a CSV data file. This allows the robotic arm to be trained to repeat the same path using the saved CSV file. The project developed a program using the machine learning model on a Raspberry Pi, which created a robotic arm with stepper motors and electronic motion drivers. A program was developed on an Arduino Uno board to control the robotic arm, receiving commands from the Raspberry Pi via the serial port. Experimental results indicate that the robotic arm successfully achieves real-time object tracking at an average speed of 6 centimeters per second. The robotic arm can memorize the movement path and repeat the movement with an average error of 0.5 centimeters. In the future, real-time image recognition technology can be applied to other applications such as smart agriculture and smart factories.