Likelihood alarms in highly automated vehicles: Strategies to prepare the driver for erratic automation behaviour

Abstract

In highly automated driving, erratic automation behaviour could lead to problems in controllability if the driver does not expect such erratic behaviour, e.g. when he is complacent or has no adequate implicit or explicit understanding of the system. Alerting the driver before the automation shows erratic behaviour would be one approach to improve the controllability. However, erratic behaviour is not reliably predictable, so that there is a high probability of false alarms leading to low compliance in the alarm. The likelihood alarm approach respects the unreliable nature of this kind of driver information and might improve compliance in the alarm. A likelihood alarm in form of feedback of automation uncertainty about erratic behaviour was tested with 40 participants in a between–subjetc study in a fixed-base driving simulator. Type (conventional vs. likelihood) and reliability of the alarm (medium vs. low) was varied. Participants drove with a combined longitudinal and lateral assistance system on a highway. During 40 of 41 situations with a second vehicle drivers experienced 4 respectively 12 false alarms. In the 41st situation the automation did not react to a slower lead vehicle. Driver reaction and trust in automation was measured.