Data-driven Causal Discovery for Pedestrians-Autonomous Personal Mobility Vehicle Interactions with eHMIs: From Psychological States to Walking Behaviors

TL;DR

The study addresses how pedestrians’ subjective evaluations influence walking decisions when interacting with an autonomous personal mobility vehicle (APMV) equipped with external human-machine interfaces (eHMIs). Using DirectLiNGAM on data from 42 participants across four eHMI conditions, the authors uncover a data-driven causal DAG linking situation awareness, risk perception, trust, hesitation, and walking behavior, aligning with a cognition–decision–behavior model. Key findings show that clearer understanding of the APMV’s intentions enhances prediction and trust, reduces perceived danger and hesitation, and speeds crossing initiation when eHMI cues are timely; late cues mostly affect post-stop behavior. These insights inform eHMI design guidelines aimed at calibrating pedestrian trust and improving safety and efficiency in mixed-traffic environments, with broader implications for causal modeling in human–vehicle interactions.

Abstract

Autonomous personal mobility vehicle (APMV) is a new type of small smart vehicle designed for mixed-traffic environments, including interactions with pedestrians. To enhance the interaction experience between pedestrians and APMVs and to prevent potential risks, it is crucial to investigate pedestrians' walking behaviors when interacting with APMVs and to understand the psychological processes underlying these behaviors. This study aims to investigate the causal relationships between subjective evaluations of pedestrians and their walking behaviors during interactions with an APMV equipped with an external human-machine interface (eHMI). An experiment of pedestrian-APMV interaction was conducted with 42 pedestrian participants, in which various eHMIs on the APMV were designed to induce participants to experience different levels of subjective evaluations and generate the corresponding walking behaviors. Based on the hypothesized model of the pedestrian's cognition-decision-behavior process, the results of causal discovery align with the previously proposed model. Furthermore, this study further analyzes the direct and total causal effects of each factor and investigates the causal processes affecting several important factors in the field of human-vehicle interaction, such as situation awareness, trust in vehicle, risk perception, hesitation in decision making, and walking behaviors.

Figures (12)

• Figure 1: The APMV with an eHMI can convey driving intentions to pedestrians through voice and visual cues.
• Figure 1: eHMI experience orders for the 42 participants.
• Figure 2: A hypothesized model of pedestrian's cognition-decision-behavior process in human-vehicle interactions was proposed in liu2020_what_timeingliu2022implicit. The Q1 to Q6 are subjective evaluation questionnaires using in the experiments; and the crossing initiation time (CIT), crossing time (CT) and after crossing time (ACT) are the time spent on the three stages of crossing the road (see section \ref{['sec:Measurements']}).
• Figure 3: Experimental settings including experimental site, driving behavior profile of the APMV and eHMI conditions
• Figure 3: Friedman test results for subjective evaluations and walking behavior factor under four eHMI conditions.