Theses and Dissertations

ORCID

https://orcid.org/0009-0000-1753-3039

Advisor

Ball, John, E.

Committee Member

Luo, Chaomin

Committee Member

Gurbuz, Ali C.

Date of Degree

5-16-2025

Original embargo terms

Immediate Worldwide Access

Document Type

Graduate Thesis - Open Access

Major

Electrical and Computer Engineering

Degree Name

Master of Science (M.S.)

College

James Worth Bagley College of Engineering

Department

Department of Electrical and Computer Engineering

Abstract

Autonomous functionalities are pivotal for the advancement of Advanced Driver Assistance Systems (ADAS), driving towards collision-free and environmentally sustainable transportation. This thesis presents two distinct studies addressing critical aspects of autonomous vehicle control: lane centering via deep learning and lateral vehicle control using model predictive control. The first study explores end-to-end learning f or autonomous s teering command generation, focusing on lane centering. A convolutional neural network (CNN) model, inspired by NVIDIA’s PilotNet, is employed to directly map raw camera pixel inputs to steering commands, eliminating the need for intermediate feature engineering. The model is trained and validated using datasets from both Udacity’s Self-Driving Car Nanodegree Program and the Mississippi State University Autonomous Vehicular Simulator (MAVS). Successful implementation within the Udacity simulation demonstrates the CNN’s capability to accurately track and navigate road lanes in linear conditions. The second study investigates lateral vehicle control, aiming to maintain lane-keeping performance through precise regulation of steering angle and acceleration. An AMPC algorithm, utilizing a 3DOF vehicle body, is designed and simulated within a Model-in-the-Loop (MIL) environment. Performance is evaluated across skidpad trajectory, and a comparative analysis is conducted between the AMPC and a traditional MPC scheme, considering key design parameters. Results demonstrate that AMPC outperforms traditional MPC in lateral deviation. Further optimization of AMPC weights reveals that a specific parameter set—lateral error (3), change of steering angle (0.5), change of longitudinal acceleration (1), and velocity tracking (2) achieves a minimum lateral deviation of 0.0402. This study highlights the potential of AMPC-based lateral control strategies, particularly for non-linear conditions, in practical ADAS applications. Collectively, these studies provide a comprehensive evaluation of deep learning CNNs and advanced control algorithms, demonstrating their feasibility and scalability for enhancing autonomous functionalities in ADAS.

Sponsorship (Optional)

The necessary information for this research was provided by the Mississippi State University EcoCAR EV Challenge team. The project was partially funded by the U.S. Department of Energy EcoCAR EV Challenge.

Download

Included in

Engineering Commons

Share

COinS