Automated vehicles (AVs) are being developed by several companies and research groups worldwide. The implicit communication (e.g., eye contact) of the vehicle or driver seems to play an important role in the expected behaviour of vehicles. It is possible that vulnerable road users (VRUs) are less able to estimate the intention of vehicles (e.g., deceleration, braking behaviour and stopping distance) when this type of communication changes or disappears when AVs are introduced. An external human-machine interface (eHMI) (i.e., a display showing when it is safe to cross) can be introduced to overcome the lack of implicit communication between driver and pedestrian. The goal of this study was to examine the effects of eHMIs of AVs on the crossing behaviour of VRUs.
A virtual reality (VR) simulation was set up, where the participant was standing on the pavement on a two-lane two-way road in a European setting. The yielding behaviour (yielding, non yielding), the type of vehicle (small, medium, large), the type of eHMI and the timing of an eHMI were varied in a within-subject design (N = 28). Four types of eHMIs were implemented, which all consisted of a screen in the front of the vehicle; with the four eHMIs being 1) the Frontal braking lights, 2) a Knightrider animation, 3) a Smiley and 4) a Text showing ‘WALK’ was shown when it was safe to cross.When it was not safe to cross, the Frontal braking lights were turned off, the Knightrider animation was not shown, the Smiley was neutral and the Text showed ‘DON’T WALK’.
Results showed that the presence of an eHMI made participants significantly feel safer when trying to cross. This was measured by measuring the total time that participants indicated they felt safe to cross using a remote control on which participants could press a button. The total time ratio that participants pressed the button when vehicles were yielding was 0.655, 0.743, 0.747, 0.751, and 0.765 for the baseline, Frontal braking lights, Knightrider, Smiley, and Text, respectively. Thus, participants felt safer to cross when an eHMI was present compared to when no eHMI was present. Secondly, the vehicle size was found to play a role in the total time that participants felt safe to cross. Specifically, the total time ratio was 0.746 for a Smart fortwo, 0.732 for a BMW z4, and 0.725 for a Ford f150 in the yielding cases. The total time ratio was 0.206 for a Smart fortwo, 0.190 for a BMW z4, and 0.156 for a Ford f150 in the non-yielding cases. Furthermore, the distance at which an eHMI changes state (e.g., ‘WALK’/‘DON’T WALK’) was found to play a role in the total time that participants felt safe to cross. The total time ratio was 0.796 for 50 m, 0.761 for 35 m, 0.698 for 20 m and 0.655 for the baseline. Participants reported in a post-experiment questionnaire that the Text eHMI was least ambiguous.
Future research can focus on which aspects improve AV-VRU interaction, so that eHMIs can be optimised. Now merely screens in front of the vehicle were tested; other techniques (e.g., projected pedestrian crossings) could yield different results. Differences between research methods can be examined as well, as this study showed differences in revealed behaviour in the simulation and conscious stated preferences of people in the questionnaires.