Analyzing and Modeling Drivers’ Deceleration Behavior from Normal Driving

Abstract

Most research in vehicle automation has mainly focused on the safety aspect with only limited studies focusing on occupants’ discomfort. To facilitate their rapid uptake and penetration, autonomous vehicles (AVs) should ensure that occupants are both safe and comfortable. Recent research, however, revealed that people felt uncomfortable when AVs braked. This may be caused by the robotlike braking performance. Existing studies on drivers’ braking behavior investigated data from either controlled experiments or driving simulators. There is a dearth of research on braking behavior in normal driving. The objective of the study was to examine drivers’ braking behaviors by exploiting naturalistic driving data from the Pan-European TeleFOT project (Field Operational Tests of Aftermarket and Nomadic Devices in Vehicles). On a fixed route of 16.5 km, 16 drivers were asked to drive an instrumented vehicle. About 11 million observations were analyzed to identify the profile, value, and duration of deceleration events. Since deceleration events are nested within trips and trips within drivers, multilevel mixed-effects linear models were employed to develop relationships between deceleration value and duration and the factors influencing them. Results indicate that the most-used profile of the deceleration behavior follows a hard braking at the beginning when detecting a danger and then becomes smoother. Furthermore, the results suggest that the speed, reason for braking, and deceleration profile mostly affect the deceleration events. Findings from this study should be considered in examining the braking behavior of AVs.