Conceptualización, implementación y desarrollo de un coche autónomo desde cero mediante machine learning

Abstract

[ES] El Trabajo de final de grado presentado por el alumno es la implementación de una red neuronal mediante aprendizaje supervisado. Dicha red neuronal es implementada con el lenguaje de programación Python, y sin usar librerías como Tensorflow, Pytorch o Pytorch. Por lo tanto, al no usar las librerías mencionadas anteriormente el alumno ha tenido que implementar la red neuronal desde 0, completamente solo, aprendiendo de forma autodidacta el mundo del Machine Learning y Deep Learning. Entendiendo conceptos como el Perceptrón, Regresión Lineal Múltiple, FordwardPropagation, Descenso del Gradiente, BackFordward, Funciones de Activación, Error Cuadrático Medio, … Para solventar las limitaciones de hardware que ofrece el robot, también se ha usado una técnica de inferencia, se ha procedido a entrenar la red neuronal en un pc de sobremesa para que aprenda a conducir el coche y una vez logrado dicho cometido se ha trasladado el conocimiento aprendido por esta red neuronal al robot creado por el alumno.

[EN] The final degree project presented by the student is the implementation of a neural network through supervised learning. This neural network is implemented with the Python programming language, and without using libraries such as Tensorflow, Pytorch or Pytorch. Therefore, by not using the libraries mentioned above, the student has had to implement the neural network from scratch, completely alone, learning the world of Machine Learning and Deep Learning in a self-taught way. Understanding concepts such as the Perceptron, Multiple Linear Regression, FordwardPropagation, Gradient Descent, BackFordward, Activation Functions, Mean Square Error ... The student also designs, implements and programs in Arduino the only robot in the form of a car, which is controlled by the neural network that the student programmed. To solve the hardware limitations offered by the robot, an inference technique has also been used, the neural network has been trained in a desktop PC so that it learns to drive a car and once this task has been achieved, the knowledge learned by this neutral network has been transferred to the robot created by the student.