An Evolutionary Approach to General-Purpose Automated Speed and Lane Change Behavior

This paper introduces a method for automatically training a general-purpose driver model, applied to the case of a truck-trailer combination. A genetic algorithm is used to optimize a structure of rules and actions, and their parameters, to achieve the desired driving behavior. The training is carried out in a simulated environment, using a two-stage process. The method is then applied to a highway driving case, where it is shown that it generates a model that matches or surpasses the performance of a commonly used reference model. Furthermore, the generality of the model is demonstrated by applying it to an overtaking situation on a rural road with oncoming traffic.