Modeling Driver Merging Behavior: A Repeated Game Theoretical Approach

Various lane-changing models have been developed for use within microscopic traffic simulation software to replicate driver merging behavior. An understanding of human driving behavior, which can be gained through such modeling, will be critical in harmonizing emerging advanced vehicle technology, such as connected automated vehicles, with human drivers. Therefore, it is important to ensure that lane-changing models are clearly understood, appropriately designed, and carefully calibrated. An earlier study by Kang and Rakha proposed and developed a decision-making model for merging maneuvers using a game theoretical approach considering two drivers: the driver of the subject vehicle (DS) in an acceleration lane and the driver of the following lag vehicle (DL) in the target lane. The previous model assumed that the DS and DL decide on an action at the first point only, where the subject and lag vehicles are identified. The current study extends the Kang and Rakha model by introducing the concept of a repeated game, assuming that a lane change decision is made repeatedly to adjust to changes in surrounding conditions. For example, drivers often decide to change their initial decision as a result of conflicts with other drivers. A repeated game helps the proposed model produce more realistic decision-making in the lane-changing process. To evaluate the model, driver decisions at a certain stage, along with accumulated historical decision data, were extracted from Next Generation SIMulation (NGSIM) data. The validation results reveal that the proposed repeated game model produces considerable prediction accuracy (above 75%).