Implementación de controlador de vuelo para vehículos aéreos no tripulados multi-rotor basado en técnicas de aprendizaje profundo

Abstract

This degree project presents the design and implementation of a position controller for a multi-rotor UAV based on deep neural networks and trained by for a multi-rotor UAV based on deep neural networks and trained by means of supervised learning supervised learning, taking as reference a PID controller. It details the process of selection of the simulation environment, the controller and the selected model is detailed. Likewise, evaluations of control trajectories control trajectories evaluations for the construction of a data set to train the model. to train the model. Different neural network architectures are trained, using the Hyperband algorithm to determine the best hyperparameters. Finally, the performance of the trained controller is evaluated with respect to the base controller by means of the temporal response with different signals. the base controller by means of the time response with different control signals. As a final product, the following is presented the dataset of the reference controller, a repository with the programs developed for the development and analysis, and development and analysis programs, and the neural network model.

Translated with www.DeepL.com/Translator (free version)