The automation effect: Investigating factors that influence the driver response process in a safety-relevant event during assisted driving and after unsupervised automation

Licentiate thesis, 2020

Introduction: Safe vehicle automation can be achieved through a detailed understanding of drivers’ ability to respond to a safety-relevant event after a period of automated driving. For instance, there is a need to understand in which scenarios automation effects are present (e.g. delayed response, degraded driving performance, crashing). Further, there is a need to identify specific factors (e.g. test environment, system-prompts, hands-on-wheel requirement, automation duration) that contribute to or prevent these automation effects. Objectives: The aim of this thesis is to investigate factors that influence: (a) automation effects in a non-prompted (i.e. absence of warning/notification) safety-relevant event during assisted driving and (b) automation aftereffects (i.e. automation effects specifically occurring after automation has been deactivated) in a prompted safety-relevant event during unsupervised automation. Method: Two Wizard-of-Oz test-track experiments were performed in order to investigate the driver response process in safety-relevant events. In experiment 1, the drivers were required to supervise (with or without a hands-on-wheel requirement) an assisted driving system, and then respond to a safety-relevant event that was not prompted by the system. In experiment 2, the drivers drove manually (baseline) and with an unsupervised automation system (a short and a long duration) before encountering a safety-relevant event. The automation system prompted (issued a take-over request) the driver to resume manual driving shortly before the safety-relevant event became visible. Results: In experiment 1, one third of the drivers responded late, or did not act at all, and crashed in the non-prompted safety-relevant event. In fact, the drivers crashed to the same extent and responded similarly independent of if they supervised the assisted driving system with or without hands on the wheel. In experiment 2, all drivers resumed manual control and did not collide in the safety-relevant event, both after a short and a long automation duration. All drivers showed a similar response and driving performance in the safety-relevant event for both long and short automation duration as well as in the manual baseline. Discussion: A hands-on-wheel requirement was not found to prevent late response or crashing in a non-prompted safety-relevant event encountered during assisted driving. More work is needed to understand the potential safety-benefits of a hands-on-wheel requirement in other types of conflicts and for driver distractions. The finding of minor automation aftereffects in experiment 2 contrasts to previous driving simulator studies. The reason may be the different test environments but is more likely due to different timings for prompting the drivers to resume manual control in relation to when the safety-relevant event became visible. Conclusions: Safe vehicle automation, including both assisted and unsupervised automation, can be achieved in a realistic environment (test track) for most drivers. However, assisted driving in combination with a non-prompted safety-relevant event, can be detrimental for safety, since some drivers may not understand the need to respond to avoid a crash. In fact, a hands-on-wheel requirement did not result in earlier steering responses nor did it prevent drivers from crashing. Thus, more work is needed to understand how to make sure drivers understand the need to respond in non-prompted safety-relevant events during assisted driving. In fact, it seems that when automation has matured to a level when it can prompt the drivers (i.e. unsupervised automation that can issue a take-over request) prior to a safety-relevant event becomes visible, drivers are able to safely resume manual control and perform similar as after an extended period of manual driving. Such safe driving performance seems to be independent of automation durations below 15 minutes.