Theses and Dissertations

Advisor

Carruth, Daniel W.

Committee Member

Ball, John E.

Committee Member

Goodin, Chris

Committee Member

Dabbiru, Lalitha

Date of Degree

12-12-2025

Original embargo terms

Immediate Worldwide Access

Document Type

Dissertation - Open Access

Major

Computational Engineering

Degree Name

Doctor of Philosophy (Ph.D.)

College

James Worth Bagley College of Engineering

Department

Computational Engineering Program

Abstract

Vehicle–pedestrian interaction is a critical aspect of autonomous vehicle (AV) development, as socially acceptable AVs must account for pedestrian intention and behavior. Risk Potential (RP) theory provides a framework to build human-like AVs by modeling expert driver risk perception, but conventional RP models treat pedestrians as static obstacles and neglect their dynamics. This dissertation extends RP theory by incorporating pedestrian velocity and orientation into three new models: the RP velocity model, the RP related-velocity model, and the RP Pedestrian Potential Position (PPP) model. These models were implemented in the Navigating All Terrains Using Robotic Exploration (NATURE) autonomy stack and tested in the Mississippi State University Autonomous Vehicle Simulator (MAVS) using Robot Operating System (ROS). The proposed models were evaluated across four metrics: (1) Prediction accuracy—number of correct actions or correct inactions in safe and unsafe scenarios, (2) safety—Time-to-Collision (TTC) and Time-Exposed-to-TTC (TET), (3) driving comfort—vehicle acceleration behavior, and (4) human likeness—similarity to human driving behavior. Results demonstrate that incorporating pedestrian dynamics improves perception accuracy and safety outcomes compared to the standard RP model. Pedestrian orientation-based models generally reduced unnecessary evasive actions and achieved shorter TET values, though in some scenarios models without orientation achieved higher human-likeness scores. Multi-pedestrian and variation scenarios confirmed that the extended RP framework remains robust under diverse and challenging conditions. These findings indicate that RP models with pedestrian dynamics contribute to safer, more comfortable, and more socially acceptable AV behavior.

Download

Included in

Engineering Commons

Share

COinS