Protocol for Eliciting Driver Frustration in an In-vehicle Environment

A state of frustration can impair a driver's ability to make decisions that optimize the safety of the driver as well as that of those around him or her. Equipping a car with the capability of detecting signs of driver frustration and responding with appropriate interventions can be an effective method to improve driver safety. In this paper, we first describe the design and implementation of a novel protocol used to elicit true frustration of varying intensities in participants interacting with an in-car Human Machine Interface (HMI) while driving in a simulator. We detail the instrumentation that was used to capture the participants' interactions with the HMI. We provide a computational analysis of signs of frustration displayed by participants in the facial and vocal modalities and discuss trends that were observed. Finally, we present baseline machine learning methods trained on features computed from facial and vocal modalities to predict the difficulty of the completed task and whether the participant is multitasking, validating the assumptions that informed our protocol design.