Naturalistic driving data for the analysis of car-following models.

PDFTech ReportVT-2010-01DTRT07-G-0003DriversCar followingBehaviorAutomatic data collection systemsTraffic modelsCase studiesTraffic simulationUnited StatesMid-Atlantic Universities Transportation CenterRakha, HeshamSangster, JohnDu, JianheVirginia Polytechnic Institute and State University. Transportation InstituteUnited States. Dept. of Transportation. Research and Innovative Technology AdministrationNTL-HIGHWAY/ROAD TRANSPORTATION-HIGHWAY/ROAD TRANSPORTATIONUS Transportation CollectionThe first research effort investigates the general application of naturalistic driving data to the modeling of car-following behavior. The driver-specific data available from naturalistic driving studies provides a unique perspective from which to test and calibrate car-following models. As equipment and data storage costs continue to decline, the collection of data through in situ probe-type vehicles is likely to become more popular, and thus there is a need to assess the feasibility of these data for the modeling of driver car-following behavior. The first research effort seeks to focus on the costs and benefits of naturalistic data for use in mobility applications.Any project seeking to utilize naturalistic data should plan for a complex and potentially costly data reduction process to extract mobility data. A case study is provided using the database from the 100-Car Study, conducted by the Virginia Tech Transportation Institute. One thousand minutes worth of data comprised of over 2,000 car-following events recorded across eight drivers is compiled herein, from a section of multilane highway located near Washington, D.C. The collected event data is used to calibrate four different car-following models, and a comparative analysis of model performance is conducted. The results of model calibration are given in tabular format, displayed on the fundamental diagram, and shown with sample event charts of speed-vs.-time and headway-vs.-time. The authors demonstrate that the Rakha-Pasumarthy-Adjerid model performs best both in matching individual drivers and in matching aggregate results, when compared with the Gipps, Intelligent Driver, and Gaxis-Herman-Rothery models.The second effort examines how insights gained from naturalistic data may serve to improve existing car-following models. The research presented analyzes the simplified behavioral vehicle longitudinal motion model, currently implemented in the INTEGRATION software, known as the Rakha-Pasumarthy-Adjerid (RPA) model. This model utilizes a steady-state formulation along with two constraints, namely: acceleration and collision avoidance. An analysis of the model using the naturalistic driving data identified a deficiency in the model formulation, in that it predicts more conservative driving behavior compared to naturalistic driving. Much of the error in simulated car-following behavior occurs when a car-following event is initiated. As a vehicle enters the lane in front of a subject vehicle, the spacing between the two vehicles is often much shorter than is desired; the observed behavior is that, rather than the following vehicle decelerating aggressively, the following vehicle coasts until the desired headway/spacing is achieved. Consequently, the model is enhanced to reflect this empirically observed behavior. Finally, a quantitative and qualitative evaluation of the original and proposed model formulations demonstrates that the proposed modification significantly decreases the modeling error and produces car-following behavior that is consistent with empirically observed driver behavior