The driver response process in assisted and automated driving

Doctoral thesis, 2022

Background: Safe assisted and automated driving can be achieved through a detailed understanding of the driver response process (the timing and quality of driver actions and visual behavior) triggered by an event such as a take-over request or a safety-relevant event. Importantly, most current evidence on driver response process in vehicle automation, and on automation effects (unsafe response process) is based on driving-simulator studies, whose results may not generalize to the real world. Objectives: To improve our understanding of the driver response process 1) in automated driving, which takes full responsibility for the driving task but assumes the driver is available to resume manual control upon request and 2) assisted driving, which supports the driver with longitudinal and lateral control but assumes the driver is responsible for safe driving at all times. Method: Data was collected in four experiments on a test track and public roads using the Wizard-of-Oz approach to simulate vehicle automation (assisted or automated). Results: The safety of the driver responses was found to depend on the type of vehicle automation. While a notable number of drivers crashed with a conflict object after experiencing highly reliable assisted driving, an automated driving function that issued a take-over request prior to the same event reduced the crash rate to zero. All participants who experienced automated driving were able to respond to the take-over requests and to potential safety-relevant events that occurred after automation deactivation. The responses to the take-over requests consisted of actions such as looking toward the instrument cluster, placing the hands on the steering wheel, deactivating automation, and moving the feet to the pedals. The order and timing of these actions varied among participants. Importantly, it was observed that the driver response process after receiving a take-over request included several off-path glances; in fact, drivers showed reduced visual attention to the forward road (compared to manual driving) for up to 15 s after the take-over request. Discussion: Overall, the work in this thesis could not confirm the presence of severe automation effects in terms of delayed response and a degraded intervention performance in safety-relevant events previously observed in driving simulators after automated driving. These differing findings likely stem from a combination of differences in the test environments and in the assumptions about the capabilities of the automated driving system. Conclusions: Assisted driving and automated driving should be designed separately: what is unsafe for assisted driving is not necessarily unsafe for automated driving and vice versa. While supervising drivers may crash in safety-relevant events without prior notification during highly reliable assisted driving, a clear and timely take-over request in automated driving ensures that drivers understand their responsibilities of acting in events when back in manual driving. In addition, when take-over requests are issued prior to the event onset, drivers generally perform similar manual driving and intervention performance as in a baseline. However, both before and just after the take-over requests, several drivers directed their gaze mainly off-road. Therefore, it is essential to consider the effect of take-over request designs not only on the time needed to deactivate automation, but also on drivers’ visual behavior. Overall, by reporting the results of tests of a future automated driving system (which is in line with future vehicle regulations and insurance company definitions) in realistic environments, this thesis provides novel findings that enhance the picture of automation effects that, before this thesis, seemed more severe.