Effects of AI Role and Speech Message Design on Takeover Requests in Automated Driving
Abstract
AI agents that provide interpretable messages can support clearer communication between humans and automated systems in situations where drivers need to respond within a few seconds. However, the potential of these capabilities to assist drivers during the takeover process remains unknown. Therefore, this study examined the effects of AI roles, explanation type, and directional semantics in speech-based takeover requests on takeover performance, as measured by decision quality (decision-making time and accuracy), and on subjective ratings. Twenty-four participants took part in a within-subjects SAE Level 3 driving simulator study. Among the three AI roles, Advisor AI was associated with faster decision-making times, and Guardian AI was associated with a lower workload. Co-Pilot AI was associated with the slowest decision-making times, lower accuracy, lower trust than Advisor AI, higher anxiety than Guardian AI, the lowest satisfaction, and the highest workload. The ipsilateral display was associated with shorter decision-making times than the contralateral display. Overall, the findings inform the integration of AI into next-generation automated systems.