22 May 2019
road traffic, road safety, driver information systems, behavioural sciences computing, visual perception, road vehicles, behavioural data, stationary object, conflict object, attention reminders, crash involvement, visual attention, visual response times, high performance, driving task, supervised automation, out-of-the-loop crash prediction, automation expectation mismatch algorithm, AEM algorithm, preconflict behavioural patterns, gaze concentration
This study uses behavioural data from the complete drive for a subset of 54 participants from the automation expectation mismatch set of test track experiments and aims to develop an algorithm that can predict which drivers are likely to crash. Participants experienced 30 min of highly reliable supervised automation and were required to intervene to avoid crashing with a stationary object at the end of the drive. Many of them still crashed, despite having their eyes on the conflict object. They were informed about their role as supervisors, automation limitations, and received attention reminders if visually distracted. Three pre-conflict behavioural patterns were found to be associated with increased risk of crash involvement: low levels of visual attention to the forward path, high per cent road centre (i.e. gaze concentration), and long visual response times to attention reminders. One algorithm showed very high performance in classifying crashers when combining metrics related to all three behaviours. This algorithm is possible to implement as a real-time function in eye-tracker equipped vehicles. The algorithm can detect drivers that are not sufficiently engaged in the driving task, and provide feedback (e.g. reduce function performance, turn off function) to increase their engagement.