Antonios Marketos Undergraduate Dissertation 2017/18

Self-Driving Cars in Virtual Reality

Supervised by J.Barker

Abstract

In the last few years, great efforts have been made to create a fully autonomous car.To achieve this goal, a model of neural networks is trained to be able to control thecar using the input information from its cameras or any sensors.The aim of this project is to build a Convolutional Neural Network that will mapinput images from a front facing camera of a car, to steering angles. This model willbe trained using images taken from three different cameras(left, right, centre) mountedat the front of a moving car.Eventually, the simple model managed to have impressing results as the loss wsreduced by 81%, while the attempt to improve it using multitasking failed as themodel was not trained enough.