Waypoint and autonomous flying control of an indoor drone for GPS-denied environments

Abstract

In this study, we propose a method for recognizing the self-location of a drone flying in an indoor environment and introduce the flying performance using it. DWM1000, which is an ultra-wide band communication module, was used for accurate indoor self-location recognition. The self-localization algorithm constructs a formula using trilateration and finds the solution using the gradient descent method. Using the measured values of the distance between the modules in the room, it is found that the error stays within 10-20 cm when the newly proposed trilateration method is applied. We confirmed that the 3D position information of the drone can be obtained in real-time, and it can be controlled to move to a specific location. We proposed a drone control scheme to enable autonomous flight indoors based on deep learning. In particular, to improve the conventional convolutional neural network (CNN) algorithm that uses images from three video cameras, we designed a distinguished CNN structure with deeper layers and appropriate dropouts to use the input data set provided by only one camera.