Autonomy bias: A deception experiment to isolate the effect of vehicle automation on perceptions of pedestrian comfort & safety

Abstract

Ensuring comfort and safety for pedestrians is essential to the responsible introduction of automated or self-driving vehicles (SDV). Few studies have attempted to isolate the effect of vehicle automation on perceptions of pedestrian interactions with SDV, separate from the potentially confounding effects of vehicle operation or appearance, and we still have limited understanding of the explicit and implicit attitudes mediating those perceptions. The objectives of this study are to determine 1) if there is an “Autonomy Bias” in the population of British Columbia, Canada (i.e., whether people perceive pedestrian-SDV interactions as inherently more or less comfortable and safe than otherwise equivalent interactions with human-driven vehicles or HDV), and if so, 2) which personal attributes influence Autonomy Bias. We isolate the effect of vehicle autonomy on perceptions using a novel deception-based experiment in which 1,133 participants rate 8 video clips of pedestrian interactions in a crosswalk; all clips show HDV, but a random half of the videos for each participant are described as SDV. Results show that Autonomy Bias varies widely across the population, with a small but significantly (p < 0.05) negative mean value (i.e., SDV interactions are perceived as less comfortable and safe). To ensure that an average person is as comfortable crossing with SDV as they currently are with HDV (i.e., to offset their Autonomy Bias), SDV must allow at least 0.4 s additional passing time at crosswalks; at least 3.7 s additional time is needed to ensure equivalent comfort for 85 % of the population. The implicit attitude of Autonomy Bias is strongly related to but distinct from explicit, self-reported attitudes toward technology and SDV, and may improve with SDV familiarity.