Event

Welcome to Ali Mohammadi's dissertation, in which he will defend his doctoral thesis ”Computational models for safe interactions between automated vehicles and cyclists”.

Ali Mohammadi is a PhD candidate at the Division of Vehicle Safety, within the unit for Traffic Safety Analysis and Accident Prevention at Chalmers University of Technology. He holds a Master’s degree in Transport Engineering from Tehran Polytechnic.

His research focuses on human interaction with automated vehicles in intersection scenarios and is conducted within the framework of the EU project SHAPE-IT. In particular, he investigates the interaction between vehicles and cyclists, aiming to develop behavioural models to predict cyclist behaviour. These models enable automated vehicles to perform safe and reliable manoeuvres when encountering cyclists in traffic.

• Supervisor: Marco Dozza
• Examinator: Marco Dozza
• Opponent: Narelle Haworth, Professor, Quensland University of Technology

Abstract

Cyclists, as vulnerable road users, face significant safety risks in traffic, especially at unsignalized intersections where they must interact with motorized vehicles. This PhD thesis investigated bicycle-vehicle interactions at unsignalized intersections and developed predictive models to improve active safety systems and automated driving. The research integrates naturalistic and simulator data to model the behavior of both cyclists and vehicles at intersections. The models included kinematic factors, non-verbal communication, and glance behavior.

The studies included in this thesis revealed that kinematic factors, such as time to arrival (DTA), along with cyclists' non-verbal cues, like head movements and pedaling, significantly affect yielding behavior at intersections. Both simulator data and naturalistic data confirmed that visibility conditions and DTA played a critical role in cyclists' decision-making while subjective data from questionnaires highlighted the importance of communication and eye contact between cyclists and drivers in reducing the severity of interactions.

Additionally, an analysis of naturalistic data uncovered differences in yielding behavior between professional and non-professional drivers, with professional drivers being less likely to yield to cyclists. Different models, leveraging machine learning and game theory, were developed to predict yielding decisions during these interactions. Lastly, simulator data was used to model drivers’ behavior, incorporating kinematics, demographics, and gaze metrics to predict drivers’ responses to crossing cyclists.

The predictive models developed through this research provide novel insights for the design of threat assessment algorithms for active safety and automated driving, enhancing the machine ability to anticipate cyclist behavior and improve safety.