Algoritmo Autónomo de Control de Trayectorias en Escenarios con Obstáculos para Mini Drones Utilizando Técnicas de Aprendizaje por Refuerzo

Abstract

This project develops and simulates a tracking algorithm where its main function is the autonomous guidance of mini drones, this algorithm was designed with the approach of reinforcement learning (RL) through neural networks. The entire design was performed with the MATLAB computational tool, which has several toolboxes and tools that help the modeling of both the drone and the artificial intelligence, as taught throughout Chapter 3, where the series of steps necessary for the design of the algorithm are described in depth. In this software we used the existing Simulink model called Quadcopter Project, focused on the Parrot Mambo mini drone series. Once the artificial intelligence was designed and elaborated, we proceeded to train it as shown in section 3.4, so that finally after several hours of training we had consolidated an AI capable of guiding the mini drone through two different trajectories, without colliding with the limits of the environment. Finally, section 4 presents the results obtained for each of the proposed track models, having an error of 0.72% in the tracking of the first trajectory and 5.4% error in the second one.