Predictive Freeway Overtaking Strategy for Automated Vehicles Using Deep Reinforcement Learning

Abstract

This paper aims to combine the deep learning (DL) and reinforcement learning (RL) techniques to formulate the predictive overtaking strategy for autonomous vehicles in freeway scenario. First, the real-world driving data is extracted from the Next Generation SIMulation (NGSIM) dataset. The long short-term memory (LSTM) model is leveraged to forecast the longitudinal and lateral motion of vehicles. Then, the freeway overtaking scenario is constructed, wherein two-lane transportation is considered. Based on the predicted driving trajectories of the surrounding vehicles, RL is utilized to guide the target vehicle passes through the scenario as soon as possible. Results indicate that the presented decision-making strategy could enhance the mobility and safety of the studied automated vehicle. The availability of the proposed method in uncertain and complex driving situations is also demonstrated.