Learning the Driver Acceleration/Deceleration Behavior Under High-Speed Environments From Naturalistic Driving Data
Chenhui Liu, Wei Zhang
Abstract
The driver acceleration/deceleration (A/D) profiles on surface streets have broad applications in transportation. With a large, high-resolution naturalistic driving data set, this article explores the normal driver A/D profiles in high-speed (≥35 mi/h) environments by checking vehicle trajectories passing through isolated, rural high-speed, all-way-stop-controlled intersections. It is found that although drivers obviously decelerate in most trips when approaching stop signs, they only fully stop in 10.5% of trips, leading to safety concerns. It is noteworthy that accelerations generally take more time and distance than decelerations. On average, it takes 20.7 s and 1,059.7 ft for a deceleration maneuver and 25.2 s and 1,267.5 ft for an acceleration maneuver. The maximum deceleration rates in most trips are found to still comply with the existing design criteria. With the increase of travel speed, A/D distances keep increasing, whereas A/D durations increase first but then keep consistent around 22.5 s/26.5 s beyond 50 mi/h. Nonlinear relationships are found to exist between speed and travel distance in A/D and can be captured with the random effects loglog models. The estimated models give new insights for developing autonomous vehicles running like human drivers rather than machine robots on surface streets.