External Human-Machine Interfaces on Autonomous Vehicles

Abstract

Perspective-taking is the ability to see a situation based on the viewpoint of others. In autonomous vehicle-pedestrian (AV-P) interaction, the perspective taken by the pedestrian could be affected by the design of an external Human-Machine Interface (eHMI). However, currently, there is little knowledge about the effect of message perspective on the crossing intentions of pedestrians when interpreting the intention of an AV. This study aims to investigate the effect of eHMI message perspective and cognitive load on participants’ perspective-taking, as inferred from their crossing intentions. We designed a photo-based experiment and examined the effect of message perspective (egocentric (from the pedestrians' point of view): ‘WALK’, ‘DON’T WALK’ vs. allocentric: ‘BRAKING’, ‘DRIVING’ vs. ambiguous ‘GO’, ‘STOP’), and cognitive load on the crossing intentions, response times and pupil diameter of the participans (N = 103). We added a memory task to increase the cognitive load during two-thirds of the trials, since crossing intentions can be demanding (the traffic scenario can be complex complex or the pedestrian is distracted) and therefore might influence perspective-taking.

The results showed that the egocentric messages were most persuasive as demonstrated by more uniform crossing intentions and faster response times compared to allocentric and ambiguous messages. When participants were put under cognitive load, a more efficient strategy was used to make a crossing decision as demonstrated by faster yet consistent crossing intentions compared to no memory task. No difference in cognitive load was measured for both message perspective and cognitive load at the moment of response, as evidenced by equal pupil size. Concerning the ambiguous messages, ‘GO’ encouraged crossing and the ‘STOP’ inhibited crossing, which points towards an egocentric perspective taken by the pedestrian. We conclude that pedestrians initially take an egocentric perspective if the eHMI message is ambiguous, though this egocentric bias can be overcome by using explicitly an egocentric or allocentric eHMI message perspective. In addition, we conclude that participants perform better (more uniform crossing decisions, faster responses) when the eHMI’s message perspective is egocentric rather than allocentric.