Agent-Based Framework for Modeling Gap Acceptance Behavior of Drivers Turning Left at Signalized Intersections

Abstract

An algorithm based on a reactive driving agent was developed for modeling gap acceptance behavior of drivers making left turns at signalized intersections. The model considers the interaction between driver characteristics and the physical capabilities of vehicles. A vehicle dynamics model explicitly captures the vehicle constraints on driving behavior. In addition, the model uses the driver's input and psychological deliberation about accepting or rejecting a gap. The model was developed with a total of 301 accepted gaps and validated with 2,429 rejected gaps at the same site and with 1,485 gap decisions (323 accepted and 1,162 rejected) at another site. The proposed model is a mix of traditional and reactive methods for decision making and consists of three main components: input, data processing, and output. The input component uses sensing information and vehicle and driver characteristics to process the data and estimate the critical gap value. Then, the agent decides to either accept or reject the offered gap by comparing it to a driver-specific critical gap. (The offered gap should be greater than the critical gap for it to be accepted.) The results demonstrate that the agent-based model is superior to the standard logistic regression model because it produces consistent performance for accepted and rejected gaps (correct predictions in 90% of the observations), and the model is easily transferable to different sites. The proposed modeling framework can be generalized to capture vehicle types, roadway configurations, traffic movements, intersection characteristics, and weather effects on driver gap acceptance behavior. The findings could be used to develop weather-specific traffic signal timing and in vehicle-to-vehicle communications.