Merging psychology and technology: Understanding, quantifying and predicting driver fatigue during conditionally automated driving

Background. Conditionally automated driving, which constitutes the continuous, operational design domain-specific performance of the driving task by an automated driving system, will soon become commonplace in everyday driving. However, the monotony, extended periods of exposure, and increase in automation which characterises the change in the driver’s role from active operator to passive supervisor incurs a risk of the driver developing dangerous levels of fatigue. Driver fatigue, which is distinct from drowsiness and is already a leading cause of road traffic accidents, is thought to imperil safe transitions of control between automated system and human driver by decreasing attention, alertness and vigilance, rendering the driver ill-equipped to respond to time-sensitive situational demands. This has prompted researchers, vehicle manufacturers and legislators alike to recommend the inclusion of driver monitoring systems, which are capable of assessing the driver’s mental state, within future conditionally automated vehicles. Current driver monitoring technology, however, largely relies on behavioural measures which do not necessarily permit the detection of more subtle changes in the driver’s mental state, such as an increase in fatigue. Physiological measures therefore represent a promising means of overcoming the limitations of traditional measures of fatigue whilst also affording a higher temporal resolution. Aims. This thesis aimed to: (1) Understand the presentation and impact of driver fatigue during conditionally automated driving, as well as drivers’ subjective experiences and perceptions of the interactions between behaviour, vehicle automation and in-vehicle monitoring; (2) Quantify driver fatigue using physiological, subjective and behavioural measures, to understand the effect of automation on physiological activity, as well as how activity and arousal vary due to task-related factors in automated driving; (3) Predict the progression of driver fatigue during conditionally automated driving by identifying the point at which fatigue reaches its highest level during prolonged automation. Methodology. A sequential, mixed-methods approach was taken to addressing the aims of this thesis, whereby four studies were completed. Study 1 constituted a systematic review and meta-analysis concerning the effect of non-driving related tasks on drivers’ physiological responses during conditionally automated driving. Study 2 employed qualitative focus group interviews to examine drivers’ perspectives and attitudes towards the use of driver monitoring systems during automated driving. Study 3 involved a driving simulator study which sought to investigate the effect of prolonged conditionally automated driving on the development of fatigue, drivers’ physiological responses, and their takeover performance. Lastly, in study 4, supervised machine learning methods were applied to understanding the progression and peak of driver fatigue, as captured by physiological activity, during prolonged conditionally automated driving. Studies 1 and 2 contributed to the design of studies 3 and 4 through refining their conceptual and methodological underpinnings, and by including the driver’s perspective in the design of the research. The physiological markers of fatigue that were used in study 4 were also selected on the basis of study 3’s findings. Findings. The systematic review and meta-analysis reported that, when drivers engaged with a non-driving-related task during conditionally automated driving, an increase in physiological arousal was observed, supporting the use of secondary tasks as a means of increasing physiological activity during automated driving. Study 2 found that drivers perceived driver monitoring and automated driving as a secondary layer of support with the potential for maladaptive consequences for the user, citing a perceived risk of overreliance on automation forming over time. The results of study 3 demonstrated that a prolonged period of conditionally automated driving led to a significant increase in subjective fatigue, as well as significant differences in several physiological measures compared with manual driving which were indicative of heightened parasympathetic activity. Finally, study 4’s findings revealed that the progression of driver fatigue was characterised by a series of peaks and brief recovery periods, supporting the use of self-regulatory strategies to mitigate fatigue during conditionally automated driving. Conclusions. The findings of this thesis make several important contributions to the literature concerning the psychological and cognitive processes underpinning drivers’ interactions with automated driving systems. Namely, the findings have quantified the evidence for the physiological effect of driver fatigue, as well as vehicle- and task-related factors during conditionally automated driving, providing strong justification for the use of physiological measures to assess driver mental states. They have clearly outlined the relationship between driver fatigue and conditionally automated driving by documenting changes over time, thus providing a clear profile of the temporality of fatigue. An understanding of drivers’ perceptions and representations of this technology has also yielded several recommendations for the development of user-centred driver safety assistance features. Lastly, predicting the progression and peak of driver fatigue using physiological measures has implications for emergent driver monitoring systems that can detect fatigue in an accurate and timely manner.

Funder

Research Ireland

Publisher

University of Galway

Rights

CC BY-NC-ND

cbn

Collections

University of Galway Theses (PhD Theses)