Vehicle-Assisted RSU Caching Using Deep Reinforcement Learning

Abstract

Vehicular communication is a crucial component of intelligent transportation systems (ITS) upon which many services are based starting with relaying traffic conditions up to enabling autonomous driving. Using a backhaul link, a road side unit (RSU) is assumed to fetch content from the Internet to serve the requests of incoming vehicles while they are within its range. Unless an effective caching mechanism is employed at the RSU, popular content might be unnecessarily retrieved from the network backhaul repeatedly especially in scenarios with large number of vehicles. This incurs significant capital and operating expenditures to RSU operators due to backhaul leasing and utilization. In this work, we explore cost-effective solutions for RSU operators by leveraging the content that is readily available on vehicles within the RSU's coverage range to intelligently cache popular content and serve newly arriving vehicles. However, due to the limited RSUs storage capacity and continuous flow of contents, the problem of establishing an RSU caching strategy to maximize the serving rate of incoming vehicles in an energy-efficient manner is challenging. Thus, we propose a technique based on deep reinforcement learning and demonstrate its effectiveness compared to the state-of-the-art.