Drivers' Attitudes and Perceptions towards A Driving Automation System with Augmented Reality Human-Machine Interfaces

The Effect of Augmented Reality Cues on Glance Behavior and Driver-Initiated Takeover on SAE Level 2 Automated-Driving

“…We believe that the number of driver-initiated takeover actions is reduced in the presence of augmented reality cues for two main reasons: (1) the perceived trustworthiness of the automated system is likely increased by making the vehicle detection capabilities more salient, and (2) AR cues increase drivers' understanding of how the automation operates (system transparency), and so drivers may trust more in the system. In fact, our finding is consistent with previously conducted user-studies that show that more trust in the AD system may be gained when knowledge of vehicle detection capabilities are provided using AR cues [4], [8], [9]. Also, we found that drivers took longer to decide to initiate takeover when no AR cues are available to them and when traffic consists of cars only, meaning that drivers may interpret the traffic situation differently (as compared to mixed-traffic cases).…”

“…At a high level, our work aims to better understand whether AR can help improve the driving experience in AD settings, by for example, helping drivers feel less stress and anxiety when encountering different traffic scenarios. The paper extends earlier work where we examined drivers' self-reported attitudes and perceptions towards an AD experience in an urban environment both with, and without, an AR HMI under differing traffic conditions [4].…”

“…In takeover situations, AR presence resulted in shorter glances and glances to fewer areas of interest. On the other hand, in non-takeover situations, AR presence resulted in longer glances and glances to more areas of interest, perhaps indicating that drivers may have higher levels of trust, comfort, and general lack of stress/anxiety with the ADAS system in this situation, as evidenced by [4] and [3]. However, it could also be that drivers just need more time to make sense of the environment, given that there were no obvious hazards that require them to intervene [see [13] for a discussion about "wait and see behavior" with autonomous driving and takeover].…”

“…Each participant experienced three intersection events for each experimental condition, for a total of 24 events per participant. We used a balanced Latin Square design to counterbalance the presentation order of conditions within participants (see [4] for more details).…”

“…Our AR cues were conformal bounding boxes creating a cubic region surrounding a specific actor; blue bounding boxes to highlight pedestrians, red for vehicles (Figure 1). Since this work did not aim to investigate different AR cue designs (as done in [4]) we instead opted for a cueing style commonly used in computer vision applications, and which adequately satisfies the first stage of Endsley's three-stage model of situational awareness (i.e. perception, comprehension, projection) [6].…”