Autonomous racing car model

Author: Saeed Alqahtani

Alqahtani, Saeed, 2020 Autonomous racing car model, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact with the details.


This project focused on building a self-driving, one-tenth scale radio-controlled (RC) rally car. Machine learning is heavily involved in the field of self-driving cars. This project took the Nvidia end-to-end convolutional neural network (CNN) as a model to create a trained CNN. The Nvidia model was developed using Python and other Python machine learning libraries, such as Keras and OpenCV (Open-Source Computer Vision Library). Image processing was central to this project; lane detection was used to set a track for the car. The image dataset was processed before it was provided to the CNN. The simulation test had a positive result—the simulated car was able to clone the operator’s driving behaviour. A practical, one-tenth scale RC rally car platform was then built and equipped with the CNN control system, a power system and sensors, including a camera and an inertial measurement unit (IMU). The practical result shows that it is not possible to run or to train the CNN model on a mini computer, such as the Odroid XU4Q. This issue is discussed in this thesis, and an approach for future development is suggested—using model predicted control (MPC). The complexity of MPC prevented it from being added to this project. For example, it requires an accurate mathematical model of the actual, scaled car. This thesis comprises five chapters. The first two chapters discuss the background of self-driving cars, the proposed method and the recently developed work in the field. Chapter 3 explains how Udacity’s self-driving car simulator was used to run tests and train the CNN model. Chapter 4 details the practical steps for building the platform and the test results from each section. It also discusses the reasons that the CNN was not able to run on the practical platform and suggests different approaches for future studies.

Keywords: self-driving cars, Autonomous mobile, A.I.

Subject: Engineering thesis

Thesis type: Masters
Completed: 2020
School: College of Science and Engineering
Supervisor: Nasser Asgari