P5 - Eliciting Driver Stress Using Naturalistic Driving Scenarios On Real Roads

We propose a novel method for reliably inducing stress in drivers for the purpose of generating real-world participant data for machine learning, using both scripted in-vehicle stressor events as well as unscripted on-road stressors such as pedestrians and construction zones. On-road drives took place in a vehicle outfitted with an experimental display that lead drivers to believe they had prematurely ran out of charge on an isolated road. We describe the elicitation method, course design, instrumentation, data collection procedure and the post-hoc labeling of unplanned road events to illustrate how rich data about a variety of stress-related events can be elicited from study participants on-road. We validate this method with data including psychophysiological measurements, video, voice, and GPS data from (\textit{N}=20) participants. Results from algorithmic psychophysiological stress analysis were validated using participant self-reports. Results of stress elicitation analysis show that our method elicited a stress-state in 89\% of participants.