Enhancing Autonomous Vehicle-Pedestrian Interaction in Shared Spaces: The Impact of Intended Path-Projection

TL;DR

This study addresses the challenge of AV-pedestrian interaction in shared spaces by introducing PaveFlow, a projection-based eHMI that visualizes an autonomous vehicle's intended path on the ground. Using a within-subject VR experiment ($N=18$) with Baseline, Multi-eHMI, and Single-eHMI conditions, the authors show that continuous path projection enhances perceived safety, trust, and user experience while reducing workload, with effects largely consistent across single and multi-AV scenarios. However, in multi-AV settings, simultaneous motion-state changes can confuse pedestrians, signaling a scalability challenge for density-rich environments. The findings suggest that externalized, real-time path information increases eHMI transparency and supports co-navigation, but future work should address signal complexity and timing to maintain usability as AV density grows.

Abstract

External Human-Machine Interfaces (eHMIs) are critical for seamless interactions between autonomous vehicles (AVs) and pedestrians in shared spaces. However, they often struggle to adapt to these environments, where pedestrian movement is fluid and right-of-way is ambiguous. To address these challenges, we propose PaveFlow, an eHMI that projects the AV's intended path onto the ground in real time, providing continuous spatial information rather than a binary stop/go signal. Through a VR study (N=18), we evaluated PaveFlow's effectiveness under two AV density conditions (single vs. multiple AVs) and a baseline condition without PaveFlow. The results showed that PaveFlow significantly improved pedestrian perception of safety, trust, and user experience while reducing cognitive workload. This performance remained consistent across both single and multiple AV conditions, despite persistent tensions in priority negotiation. These findings suggest that path projection enhances eHMI transparency by offering richer movement cues, which may better support AV-pedestrian interaction in shared spaces.

Figures (3)

• Figure 1: Overview of design concept which integrates AV navigation algorithms with a path projection eHMI to visualise intended trajectories in real time. As pedestrians adjust their movement in response, the concept enables dynamic right-of-way negotiation.
• Figure 2: Boxplots show Descriptive Means (M) and Standard Deviations (SD) of Perceived Safety, Trust, Workload, Pragmatic Quality, and Hedonic Quality across conditions. Statistical significance was tested using ANOVA (Baseline vs. Multi-eHMI) and Wilcoxon (Multi-eHMI vs. Single-eHMI). *: p $\leq$ .05, **: p $\leq$ .01, ***: p $\leq$ .001
• Figure 3: Descriptive statistics of NASA-TLX workload dimensions across the three study conditions.