Enhanced QL-Based Dynamic Routing Protocol for Urban VANETs

Abstract

Choosing an optimal data forwarding route is crucial for improving network performance in mobile ad hoc networks (MANETs). This process becomes very complex if the network topology changes frequently and quickly, as is the case with vehicular ad hoc networks (VANETs). Under these conditions, the routing process can be significantly improved by including machine learning in optimal route selection. The type of learning that suits highly dynamic networks the best is reinforcement learning (RL). One of the most important types of RL for dynamic MANETs is Q-learning (QL). In this study, an enhanced QL-based dynamic routing algorithm for urban VANETs (Q-DRAV) is proposed, which manages to significantly improve the overall network performance of VANETs, by including relevant network parameters in the RL process. Simulation analysis and comparison with other routing protocols are performed in the NS-3 simulator, and the protocol implementation code is publicly available. Simulation results show that the proposed protocol reduces the packet loss ratio, average packet end-to-end delay and jitter, while it increases the achieved application throughput in the network.