Implementación de un Enjambre de Drones Crazyflie 2.1 Coordinado Mediante Q-learning y ROS

Abstract

The implementation of cooperation networks based on unmanned aereal vehicles, using artificial intelligence, have a social interesting applicability, because it allows to solve problems of mobility and organization on large areas of land. The present project generates a theoretical and applied research by the organize a mini dron cooperation network based on Crazyflie 2.1 and his positional system calls Lighthouse, to validate the use of this systems before practical implementations. Specifically developing a decision control process for all the agents involved on the network. The learning algorithm is Q-learning, used to prevent crashes and organize the swarm. In the learning process is possible to minimize the estates of spaces and maximize it when require the performance of the swarm, it able to do a process of learning fast and a precise performance on implementation. Besides, develop a certain phases of training and test of the algorithm to generate a right Q policy table, that the agents use to take decisions, together with a 250 possible state definition, implemented for a two dimensional environment.