Calibration and validation of the rule-based human driver model for car-following behaviors at roundabout using naturalistic driving data

Understanding driver behavior is crucial for introducing roundabouts. This study focuses on calibrating the parameters of the car-following model using naturalistic data and analyzing the appropriateness of different car-following models on the roundabout. We utilize rounD trajectory dataset. This dataset allows for the precise definition of lead and follow vehicles, enabling the calibration of model parameters accordingly. We compared the calibration results for roundabouts with those obtained for signalized intersections from CitySim. Our results show that the Krauss and intelligent driver models (IDM) achieve mean absolute percentage errors of 10.09% and 23.21%, respectively. Furthermore, IDM exhibited higher errors in the circulation segment of the roundabout, while in the exit segment, the Krauss model showed elevated errors. It contrasted with the homogenous results obtained in the signalized intersection. These findings provide valuable insights into driver's behavior on roundabouts.