Predicting Automated Vehicle Takeover Decision During the Nighttime

Abstract

Conditionally automated vehicles require drivers to take over control occasionally. To date, takeover performance has been mostly evaluated using only re-engagement time and quality metrics. However, the appropriateness of takeover decisions, which has not been considered by previous research, should also be included as a performance indicator as it reflects one’s situation awareness of the takeover scenario. The goal of this study was to use eye-tracking, demographic factors, workload, and non-driving-related task (NDRT) conditions to predict takeover decisions. Forty-three participants drove a simulated conditionally automated vehicle while performing visual NDRTs and needed to decide the most appropriate maneuver around a roadway obstacle. Six classifiers were used to predict takeover decisions. The Random Forest model achieved the best performance, and driving experience and perceived workload were the most influential features. Findings may be used to assist in the design of adaptive algorithms that support drivers taking over from automated vehicles.