Evaluation of external human-machine interfaces for facilitating the interaction of pedestrians with self-driving vehicles

Abstract

Self-driving vehicles (SDVs) promise enormous potential to increase traffic safety, improve access to mobility, and reduce emissions – provided people accept them. Public approval of SDVs is expected to be linked to their ability to successfully interact with humans. In this context, not only primary users – i.e., riders who decide to use SDVs – but also incidental users – e.g., pedestrians, cyclists, and drivers of conventional vehicles who are forced to interact with SDVs when they encounter them in traffic ─ must be considered. To facilitate the interaction between SDVs and incidental users it has been proposed to equip SDVs with an external human-machine interface (eHMI). The eHMI could display information to traffic participants in its vicinity, e.g., about the SDV’s automated status, its perception of a person, and its intent to yield, or could provide advice, for example, to cross. This PhD project focused on pedestrians’ interactions with SDVs in ambiguous and unsignalized street-crossing scenarios. A first objective of this PhD project was the development and validation of a safe, natural, and parsimonious methodological paradigm to examine pedestrians’ street-crossing behavior (Study I). The paradigm allows relative comparisons between eHMIs in terms of how fast pedestrians initiate street- crossing if the SDV is yielding and whether pedestrians show safe street-crossing if the SDV is nonyielding. Participants indicate their street-crossing decision through the actual behavior of stepping off a sketched sidewalk onto a sketched crosswalk while watching videos of an approaching vehicle on large screens in a lab. Two hidden force-sensitive resistor sensors under the sidewalk capture pedestrians’ crossing onset time. Ultimately, the major objective of this PhD project was the in-depth examination of pedestrians’ informational needs towards eHMIs. For this purpose, three experimental studies were conducted on a test field (Study II) and in labs (Study III, Study IV) in which participants acted as pedestrians. The studies examined eHMI effects on subjective feelings such as trust and perceived safety (Study II, Study III, Study IV). In addition, objective measurements of pedestrians’ street-crossing behavior were evaluated. In this regard, the studies examined the efficiency of eHMIs in promoting fast street- crossing when the SDV is yielding to the pedestrian (Study II, Study III, Study IV) as well as the effectiveness of eHMIs in inducing safe street-crossing if the SDV is failing to yield to the pedestrian in the event of a system malfunction (Study IV). Beyond this, semistructured retrospective interviews were conducted to form a better understanding of pedestrians’ motives (Study II, Study III, Study IV). The findings emphasize that the combined messages of the automated status and the intent for its next maneuver enable pedestrians to form an accurate and complete situation awareness, resulting in positive subjective feelings as well as fast and safe street-crossing when pedestrians interact with SDVs. Without an eHMI, pedestrians struggle to interact in a positive manner with an SDV because they miss seeing a driver. Transparent information about the SDV’s automated status supports pedestrians in comprehending the invalidity of driver factors and thereby reduces the increased complexity that arises in mixed traffic of SDVs and conventional vehicles (Study II, Study III). However, a basic status eHMI is also related to overtrust in and overreliance on the automated driving system as pedestrians build the preconception that SDVs will always yield to them – which may result in unsafe street-crossing if the SDV fails to yield to the pedestrian in the event of a system malfunction (Study IV). Subsequent information about the SDV’s intent supports pedestrians in anticipating the future actions of the SDV (Study II, Study III). This proved to be particularly relevant in enabling pedestrians to detect a system malfunction – i.e., an SDV that is yielding to a pedestrian (status eHMI changes to status+intent eHMI) is clearly distinguishable from an SDV that fails to yield to a pedestrian (status eHMI remains unchanged). In this manner, subsequent information on the SDV’s intent-to-yield promotes a calibrated level of trust as well as safe street-crossing (Study IV). Moreover, the value of providing transparent information on the SDV’s status and intent(-to-yield) to pedestrians was shown to be consistent for various urban traffic situations (Study II) and to remain stable or even increase over time as the number of encounters increases (Study III). Based on its findings and theoretical considerations, this PhD project discourages the use of an eHMI informing about the perception of a person or giving advice to pedestrians. A perception signal is perceived as an obvious gimmick (Study II). An advice eHMI is associated with legal, ethical, and safety issues. Hence, the present PhD project contributes to both measuring eHMI effects and developing eHMIs for successful interactions between SDVs and pedestrians. The discussion provides indications that the study findings might be transferred to other recipients of eHMIs such as cyclists and drivers of conventional vehicles as well as primary users of automated systems such as drivers of SDVs and even other domains of automation. In view of the study findings, strategies for the implementation of and education regarding the eHMIs were developed. Furthermore, the present findings give rise to future research directions – particularly the design of the eHMIs that convey the status and intent of SDVs remains unsolved.