Abstract: 

The lane change models used in today's traffic simulators often do not determine lane change actions in terms of the evaluation of sequential plans, but rather in terms of the utility of the very next lane change action. This has the disadvantage of not being able to account for the influence of delayed rewards, such as the simulated vehicle moving across a slow lane to a better-performing non-adjacent lane. This research presents a lane change model which at every simulation time step, builds a tree of potential maneuver sequences, and selects the lane change action according to planning over a time horizon. The model was calibrated using a vehicle trajectory data set and shown to give improved realism of lane change actions of individual vehicles, compared to a lane change model without sequential planning.