Asynchronous deep reinforcement learning for semantic communication and digital-twin deployment in transportation networks

Abstract

The dynamically evolving and technologically-driven hybrid landscape of transportation networks integrated with advanced edge computing capabilities has demonstrated efficient communication and computation techniques to guarantee robust quality of services (QoS) to vehicles. However, conventional communication systems in the Internet of Vehicles (IoV) still encounter challenges in providing meaningful low-latency communication and AI-assisted real-time synchronization on the edge. One reason is that it has exhausted the Shannon limit by utilizing cellular, NOMA, and Wi-Fi technologies. Therefore, we present an integrated approach leveraging Semantic Communication (SC), and Digital Twin (DT) deployment to tackle the challenges caused by high-dimensional data exchanges and resource spectrum crunch leading to inevitable latency constraints. SC stimulates meaningful transmission of data to high-mobility vehicles by providing a relevant knowledge base (KB) and DT deployment. In this paper, we established the vehicular SC (VSC) model, and DT deployment strategy. We formulate a multi-objective optimization problem (MOP) to maximize the overall QoS of the system by jointly optimizing VSC and DT deployment. Compared to traditional deep-reinforcement learning (DRL) schemes, we propose a Digital Twin Semantic Sensing using the Multi-vehicle DRL ( DTS2 -MVDL) algorithm which addresses the MOP and persistent issues of multi-dimensional, continuous, and discrete nature of the vehicular environment. Lastly, we employ age of Information (AoI), latency, and QoS as the performance metrics to determine the algorithmic efficiency.