Comparison of Machine Learning algorithms within an application of autonomous vehicle development

Abstract

This thesis gives overview about different development methods for autonomous vehicle design. Algorithms that were used are divided into two categories: Supervised Learning and Reinforcement Learning. All models are developed using Artificial Neural Networks. In order to get fast and accurate result as mush as possible a virtual environment is used. Main criteria for evaluation of performance was the completion time of lap that is recorded by system. Despite the fact that almost all Supervised Learning models showed great performance with little training, best model was the one that is build with TQC Reinforcement Learning algorithm. Generally this work can be taken as a bases and can be improved in order to the application of real-life prototypes.