Optimizing external human-machine interfaces for pedestrian-autonomous vehicle interactions through mental workload analysis

Abstract

External Human-Machine Interfaces (eHMIs) are increasingly proposed to facilitate safe and effective interactions between pedestrians and autonomous vehicles (AVs), particularly at unsignalized crossings. However, eHMIs may inadvertently increase pedestrians' mental workload (MW) by requiring them to process additional information during street-crossing. Increased MW can reduce the cognitive resources available for processing critical traffic cues, potentially diminishing situation awareness (SA) and increasing the likelihood of unsafe crossing decisions. To address this challenge, this research investigates the conditions under which eHMIs influence MW and proposes design strategies to reduce MW. Employing a mixed-methods approach grounded in human factors engineering and human-computer interaction (HCI), this study examines eHMI design and its impact on pedestrian MW.
Study One (Chapter 4), was conducted to construct a theoretical model delineating eHMI attributes that influence MW and identifying contextual factors that moderate their effects. The review identified five key eHMI attribute categories: information content, modality, placement, onset method and timing, and consistency/consolidation. Additionally, contextual factors related to pedestrians, AVs, and the environment were found to significantly influence eHMI efficacy.
Studies Two and Three (Chapters 5-6) utilized video-based experiments with electroencephalography (EEG) to assess the impact of risk-warning eHMIs on pedestrian MW. The findings indicated that risk-warning eHMIs did not reduce MW under full penetration conditions, where all vehicles in a platoon were equipped with eHMIs. Conversely, MW increased under partial penetration conditions, where only half of the vehicles featured eHMIs. Moreover, displaying only medium and high risk levels led to higher MW compared to displaying all risk levels (low, medium, and high) in partial penetration scenarios. These results extend the theoretical model established in Chapter 4.
Study Four (Chapter 7) focused on developing a generative AI (GenAI)-assisted eHMI design tool to create interfaces that minimize pedestrian MW. Usability evaluations demonstrated that the tool effectively assisted designers in creating eHMIs, with designers showing a preference for infrastructure-based and pedestrians’ personal device-based risk-warning eHMI, which had been shown in previous studies to reduce MW in complex traffic conditions.
Study Five (Chapter 8) selected one of the outcomes from Study Four (Chapter 7)—an AR-based risk-warning eHMI—and tested it using a virtual reality (VR) experiment. The results showed that this AR-based eHMI significantly reduced MW in both full and partial penetration vehicle platoon conditions, highlighting its potential to enhance pedestrian safety.
This thesis provides a comprehensive examination of eHMI design to optimize pedestrian MW in interactions with AVs. The findings contribute to a robust research framework and offer practical implications for designing eHMIs that enhance pedestrian safety and reduce cognitive demands. By integrating theoretical insights, empirical evidence, and innovative design tools, this work advances the development of safer and more effective pedestrian-AV interactions.

Date of Award	15 Mar 2026
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Xu Sun (Supervisor), Luis Felipe Moreno Leyva (Supervisor) & Bingjian Liu (Supervisor)