Design and Control of Artificial Neural Network Based Low-Cost Autonomous Quadrotor System

Abstract

In this work, a low-cost quadrotor system was designed that can navigate itself indoors using artificial neural networks (ANN) without any human interaction. This quadrotor, with help of ground station, can detect an abnormal event (fire, accident, etc.) and report it. In this system, every two seconds, a camera placed on the quadrotor takes a photo of the region beneath it and sends the photograph to the ground station over WLAN. The ground station downloads the submitted photo and feeds it as input to the ANN after finishing the 28x28 pixel conversion process. The output of the ANN is sent over WLAN to the quadrotor as a flight command (turn right, forward, etc.). Then, the flight controller adjusts motor speeds according to the sent command. Furthermore, to provide flight stabilization of the quadrotor, a 2 degree-of-freedom (DOF) controller was designed that regulates the roll and pitch angles of the quadrotor system. This controller was embedded in Wemos D1 Mini microcontroller. The designed controller was firstly developed and tested on a single DOF test system and then applied to the quadrotor system.