Driver interruptibility research for in-vehicle voice interactions

Abstract

As a countermeasure to visual-manual distraction, auditory-verbal (or voice) interfaces are becoming increasingly popular for in-vehicle systems. This opens up new opportunities for drivers to be connected to various service domains and receive proactive personalized services, such as contextualized recommendations and interactive decision-making. However, prior studies warn that such interactions can cause cognitive distraction due to the nature of concurrent multitasking with a limited amount of cognitive resources. This thesis aims to examine and find in-vehicle opportune moments in which the negative influence on voice interaction can be minimized (e.g., when drivers can safely engage in proactive voice services successfully without difficulty). Given that there is a lack of definition about what constitutes interruptibility for voice tasks, I first defined interruptible moments by considering multiple dimensions, then collected a real-road dataset of 29 drivers by iteratively developing the experimental framework through an extensive literature review and four pilot studies. Next, I quantitatively and qualitatively examined how interruption contexts and driver behavior influence driver interruptibilty, and develop a machine learning model that infers driver interruptibilty. Finally, I discuss how my findings and the prediction model can be used to design and realize three types of flow-control mechanisms for voice interactions that can improve driver interruptibilty.