Perception and decision making in vehicle following: modelling, calibration, validation and simulation

Abstract

Driving brings great convenience and efficiency to people’s daily travel. On the other hand, possible road crashes while driving pose a great challenge to the safety of road users. More insights into human driving behaviour are needed to improve driving safety. As the key part of driving behaviour, vehicle following, characterised by the interaction between a vehicle and its leader in a single-lane roadway, has been extensively studied for more than sixty years. However, perception and decision making have seldom been mathematically incorporated in dominant methods of vehicle following modelling. The investigation of these two components is expected to expand the knowledge boundary regarding human driving behaviour. Specifically, perception of traffic dynamics and risk perception are introduced as the representation of perception while decision theory under risk and risk attitudes are incorporated for modelling decision making. After the microscopic modelling of perception and decision making in single-lane vehicle following, the flow-density relation is also derived. Through macroscopic sensitivity analysis, a more risk-averse attitude, the perception of a more severe crash, a greater standard deviation of perceived time headway, a longer time interval for calculating the future speed and a longer vehicle length are found to lower the capacity of traffic facilities and cause more congestion in macroscopic traffic. The model calibration and validation are performed against the vehicle-trajectory data that were collected at a freeway section of I-80 Emeryville. It is found that drivers when following a car show a more accurate and stable time-headway perception than following a truck. Truck drivers tend to have a more stable time-headway perception and less risk aversion than car drivers. When following a different type of vehicle, drivers are shown to perceive a more severe crash. Subsequently, the comparison between the predicted space headway and the observed value of the validation validate the effectiveness of the proposed model and the relevant findings. The simulation based on the proposed vehicle following model is also presented. Based on the mixed error measurement, the simulation presents a reasonably accurate reproduction of the observed traffic dynamics of the study area.