Driving trainer with ML-Agents

Abstract

[EN] Artificial intelligence has witnessed remarkable progress in recent years, with machine learning being a crucial milestone in transforming the field. Reinforcement learning, a model that trains an algorithm through trial and error, has emerged as a significant breakthrough with the potential to revolutionize various industries, including the automotive sector. This thesis aimed to create a program capable of training a car to complete a road using the user’s driving example as input. A program in Unity was developed to achieve the objective, which uses reinforcement learning to train the model.

[ES] Se trata de un programa hecho en Unity con la librería de ML-Agents que, utilizando herramientas de aprendizaje por refuerzo, puede perfeccionar el trazado de una persona con un coche en un circuito. El programa graba el trazado que realiza la persona y lo utiliza como punto de partida para calcular cosas como cuándo acelerar, frenar, girar, etc, para acabar el circuito lo antes posible.