"Self-driving, or autonomous, cars promise sustained individual mobility while decreasing the risk of accidents due to human error. Their technological development tops the agendas of European governments and car manufacturers. With technology taking centre stage, it is easy to overlook the human driver. However, this would be a grave mistake, as autonomous vehicles still require human action. Specifically, the next frontier in autonomous vehicles is a car that controls the vehicle (e.g. steering, acceleration) and monitors the traffic environment, but that can signal a request for human intervention at any time. Little is known about how drivers detect and react to such unexpected signals. Research on lower levels of automation (e.g. cars with cruise control) suggests that reaction times to unexpected signals tend to be slow. It is, however, not clear what causes this slowdown, especially at higher levels of automation. Is this a failure to detect the signal, or a failure to react timely? My research has investigated under what conditions people (fail to) detect and react to unexpected audio intervention signals. To this end, I measured detection using cognitive neuroscience techniques (Event Related Brain Potentials) and reaction using reaction time in a driving simulator.
The results of our study demonstrated that driver's ability to detect alerts is reduced under automated driving conditions, especially when participants are passively (and less attentively) listening to the sounds. This is a challenge for current (semi-) automated vehicles, which rely on auditory alerts to warn drivers to take-over control of the car from the automated vehicle. Given the reduced ability to detect such alerts, drivers might miss these warnings.
Based on our results, I have further studied the effecive use of early warnings to warn a driver (also referred to as ""pre-alerts""). The results demonstrated that such early warnings lead to better reactions to alerts by drivers in driving studies.
I also published a new framework (building on Hidden Markov Models) to consider human confusion when interacting with automated systems, and to guide further scientific dialogue on these matters."
