Citation

Mantouka EG, Vlahogianni EI. Transp. Res. C Emerg. Technol. 2022; 142: e103770.

Copyright

(Copyright © 2022, Elsevier Publishing)

DOI

10.1016/j.trc.2022.103770

PMID

unavailable

Abstract

Most driving recommendation and assistance systems, such as Advanced Driving Assistance Systems (ADAS), are usually designed based on the behavior of an average driver. Nevertheless, personalized driving systems that can be adapted to different driving styles and recognize individual needs and preferences, may be key to the sensitization of drivers and the adoption of safer driving habits. In this paper, an enhanced self-aware driving recommendation system is developed using a Deep Reinforcement Learning algorithm, which produces personalized driving recommendations with a view to improving driving safety, while respecting individual driving styles and preferences. The impact of applying this recommendation system is evaluated through microscopic simulation; findings revealed that, in case all drivers follow the suggestions, there is a significant improvement in road safety and some minor changes in traffic flow properties. The outputs of this work may be useful within the framework of an advanced active cruise control system, can be exploited in the development of enhanced behavioral models or even lead to the revision of policy measures that utilize driving behavior as a key controller of traffic management.

Language: en

Keywords

Clustering; DDPG; Deep reinforcement learning; Driving behavior; Driving recommendations; Microsimulation; Naturalistic driving