A neural network based adaptive cruise control car-following model using 3-dimensional speed vectors

Abstract

Driverless cars have been a long-term vision for decades and have undergone much technological development. Specifically, research has progressed on Adaptive Cruise Control (ACC) car-following models. However, current studies on ACC car-following models are limited in multiple areas. Some conventional car-following models use unrealistic parameters. Also, some of such simulations are not validated using empirical data. Furthermore, most models used are parametric car-following models, which are limited in applicability, adaptivity, and too simple to capture the vehicle dynamics. Currently, there is a growing trend in the use of machine learning to develop car-following models that potentially reflect results more accurately. However, there is a lack of literature exploring the usage of 3-Dimensional (3D) speed vectors to enhance the accuracy of the performance of car-following models. This study proposes an artificial neural network-based ACC car-following model. A neural network is a non-parametric model and has excellent function fitting ability which will be highly adaptive and applicable to car-following models. The neural network is trained using ACC vehicle data and it utilizes 3D speed vectors from the ENU plane as additional inputs to the model. The results obtained from the experiments indicated that using speed vectors from the ENU plane was able to significantly improve the performance of the model.