A psychophysiological insight into driver state during highly automated driving

The aim of this research was to investigate and validate the usage of physiological measures as an objective indicator of driver state in dynamic driving environments, and understand if such a methodology can be used to measure driver discomfort, and high workload. The work addressed questions relating to: (i) detecting and removing motion artefacts from electrodermal activity (EDA) signals in dynamic driving environments; (ii) primary factors contributing to driver discomfort during automation, measured in terms of their physiological state; (iii) understanding changes in drivers’ workload levels at different stages of automation, as indicated by electrocardiogram (ECG) and EDA-based measures and; (iv) how drivers’ attentional demands and workload levels are affected at different stages of automation, measured using eye tracking-based metrics. A series of experiments were developed to manipulate drivers’ discomfort and workload levels. The analysis around driver discomfort focused on automated driving, whereas drivers’ workload levels were investigated during automation, and during resumption of control from automation, in a series of car-following scenarios. Our results indicated that phasic EDA was able to pick up discomfort experienced by the driver during automation, and correlated to drivers’ subjective ratings of discomfort. Narrower roads, higher resultant acceleration forces and how the automated vehicle negotiated different road geometries all influenced driver discomfort. We observed that drivers’ workload levels were captured by ECG and EDA-based signals, with phasic component of EDA signal being more sensitive to short term variations in driver workload. Similar results were observed in drivers’ pupil diameter values, as well as subjective ratings of workload. Factors such as engagement in a non-driving related task (NDRT), presence of a lead vehicle while maintaining a short time headway, and takeovers, all seemed to increase drivers’ workload levels. Future work can build on this research by incorporating sensor fusion of ECG and EDA-based data, along with eye tracking, to help improve the accuracy and capabilities of future driver state monitoring systems.