Objective data-driven insights into pedestrian decisions, comprehensibility, and perceived safety of autonomous vehicles with varied eHMIs: Evidence from a real-world experiment.

In future traffic environments dominated by highly autonomous vehicles (AVs), pedestrians may face challenges in accurately interpreting AV behavior, thereby potentially increasing the risk of pedestrian-AV interactions. External human-machine interfaces (eHMIs) have been proposed to facilitate communication between AVs and pedestrians; however, comprehensive evaluations using objective data from real-world interactions are limited. This study developed a systematic evaluation framework grounded in the ISO 9241-11 standard, integrating four key indicators: decision accuracy, comprehensibility, decision efficiency, and perceived safety. Objective data were collected through behavioral observation and eye tracking, with decision accuracy, total fixation time, decision time, and the coefficient of variation of pupil diameter as quantitative metrics. The study examined the effects of eHMI types (light-band, symbol, text), deceleration strategies (gentle, early, aggressive braking, no braking), and yielding behaviors (yielding, non-yielding) on pedestrian decision-making and perceptions. A total of 24 participants were recruited for a real-world crossing interaction experiment. The results showed that eHMIs significantly improved decision accuracy under yielding conditions, while decision accuracy remained high under non-yielding conditions regardless of eHMI type. eHMIs enhanced comprehensibility, with symbol-based and text-based eHMIs performing better than light-band eHMIs. eHMIs also improved pedestrian decision efficiency and perceived safety, with significant differences observed across different eHMI types and yielding behaviors. Furthermore, while deceleration strategies had no significant effect on eHMI comprehensibility or decision efficiency, they played a crucial role in shaping perceived safety. These findings inform the design of eHMIs and deceleration strategies to optimize pedestrian-AV interactions, contributing to safer AV integration in traffic environments.