RHRA-DRL: RSU-Assisted Hybrid Road-Aware Routing Using Distributed Reinforcement Learning in Internet of Vehicles

Abstract

In this paper, we propose a novel RSU-assisted hybrid road-aware routing algorithm, RHRA-DRL, designed for urban vehicular networks to optimize real-time data delivery considering dynamic road conditions. The algorithm minimizes broadcast overhead and efficiently determines optimal routing paths by incorporating two key components. Firstly, a multihop road-segment reward-based ad-hoc (RRAH) routing algorithm is introduced to adaptively respond to changing vehicle topologies within road segments. Rewards are calculated based on performance metrics, and the segment reward integrates into RSU-to-RSU (R2R) routing. Secondly, a distributed Q-learning-based road-aware (DQRA) routing algorithm determines RSUs traversed during data transmission using a decentralized agent reinforcement learning approach. The combination of these algorithms in RHRA-DRL ensures effective and consistent path establishment with a unified reward system. Simulation results demonstrate the superiority of RHRA-DRL over AODV in Internet of Vehicles (IoV) networks, showcasing enhanced communication, prolonged link lifetime, rapid establishment and repair of routing paths, and reduced overhead.

Keywords

• Data routing
• distributed reinforcement learning
• intelligent transportation system
• Internet of Vehicles
• vehicular ad hoc networks