Data-driven human-like cut-in driving model using generative adversarial network

Abstract

In this Letter, the authors developed a data-driven human-like cut-in driving model using a generative adversarial network (GAN). When a vehicle cuts into the lane of another vehicle, complex interactions occur among the various vehicles. The purpose of the GAN driver model is to learn the human-driving abilities in complex and diverse situations. Using the Kullback–Leibler divergence evaluation method, the authors confirmed that the GAN driver model shows more human-like driving than the rule-based driver model.