The effects of automation failure and secondary task on drivers’ ability to mitigate hazards in highly or semi-automated vehicles

Although the number of automated vehicles is expected to grow there is limited understanding on how drivers will cope with manual driving when automation fails. The study's goals were: (1) develop an experimental test-bed for evaluation of levels of vehicle automation, in-vehicle secondary tasks, and hazardous scenarios; and (2) conduct empirical evaluation to examine how well drivers mitigate road hazards when automation fails unexpectedly, looking at situations where drivers were either engaged with secondary tasks or not prior to the automation failure and/or the hazardous event. The STISIM fixed base simulator, embedded into a car was utilized. Driving scenes were projected on a 7m diameter round screen. An in-house LabVIEW-based program was used to control the simulator and displays; enabling control of four modes of vehicle automation: Manual-no automation (M), Adaptive Cruise Control (ACC), Automatic Steering (AS), and Automated Driving (AD). Two types of secondary tasks were included: (1) Driving related. This task required on road glances; (2) Driving unrelated. This task, presented on an in-vehicle touchscreen, required in-vehicle glances. In a mixed within-between experimental design, eighteen participants were asked to drive through various drives. Each drive included 4 driving sections in the following order: (1) automated, (2) manual, (3) automated with secondary tasks, and (4) manual with secondary tasks, in one of the levels of automation (ACC, AS or AD). In each section, typical hazardous events appeared. Automation failure (i.e., the need to assume manual control) was alerted by sound and visually on the touchscreen. The results showed that while engagement with a non-driving related secondary task lead to more crashes, automation failure did not, especially when drivers were monitoring the road. In addition, drivers’ performance on the secondary task revealed differential effects of automation mode with respect to the road conditions. Implications of this study are discussed.