This study aims to contribute to guidelines for driver licensing organizations on assessing driver competence in using Level 3 Automated Lane Keeping Systems (ALKS), based on an on-road experiment with eight professional driving assessors (i.e., expert driving examiners who train examiner candidates; 6 males, 2 females, all driving more than 20,000 km per year) in a Wizard-of-Oz vehicle. Using a think-aloud protocol, we captured cognitive processes during system supervision and take-over requests (TORs) in real-world traffic jams. A large language model (LLM)-based thematic analysis of transcripts revealed five themes: (1) Requirement for immediate environmental assessment, (2) Requirement for causal understanding, (3) Requirement for proactive intervention to maintain traffic flow, (4) Requirement for continuous “supervisor” engagement, and (5) Physical ergonomics and mode awareness. These findings indicate that, at least during short-duration usage, drivers do not simply rely on the system to disengage from driving; instead, they maintain active monitoring, physical readiness, and anticipatory skills. These observations blur the distinction between Level 2 and Level 3 automation, as the expert participants in this study generally remained attentive rather than adopting the ‘mind-off’ state that Level 3 theoretically allows. In conclusion, assessing ALKS usage involves not only evaluating a driver’s reaction to a TOR but also judging their performance as a systems manager responsible for anticipating conflicts and smoothly executing control transitions.

In automated vehicles, beeps are widely used as alarms and feedback. However, as automation advances, there is a need to explore subtler, contextually sound-based notifications for non-urgent situations. While auditory interfaces for take-over requests have been studied, limited attention has been given to using soundscapes for such alerts. This paper designed and evaluated soundscapes using existing driving-related sounds–amplified road noise and/or dimmed background music–for scheduled take-over situations. A driving simulator study showed that these soundscapes enhanced reaction time, situation awareness, and acceptance without causing annoyance. Particularly, the combined condition (music dimming and road noise amplifying) supported higher driver awareness and responsiveness. These findings suggest that soundscapes can offer safer, more intuitive take-over alerts by embedding information into familiar audio cues. This study contributes to developing soundscapes as novel alert mechanisms that integrate seamlessly with the driving environment to enhance both safety and user experience in automated vehicles.

Vision-language models are of interest in various domains, including automated driving, where computer vision techniques can accurately detect road users, but where the vehicle sometimes fails to understand context. This study examined the effectiveness of GPT-4V in predicting the level of 'risk' in traffic images as assessed by humans. We used 210 static images taken from a moving vehicle, each previously rated by approximately 650 people. Based on psychometric construct theory and using insights from the self-consistency prompting method, we formulated three hypotheses: (i) repeating the prompt under effectively identical conditions increases validity, (ii) varying the prompt text and extracting a total score increases validity compared to using a single prompt, and (iii) in a multiple regression analysis, the incorporation of object detection features, alongside the GPT-4V-based risk rating, significantly contributes to improving the model's validity. Validity was quantified by the correlation coefficient with human risk scores, across the 210 images. The results confirmed the three hypotheses. The eventual validity coefficient was r = 0.83, indicating that population-level human risk can be predicted using AI with a high degree of accuracy. The findings suggest that GPT-4V must be prompted in a way equivalent to how humans fill out a multi-item questionnaire.