Graph-Powered Reinforcement Learning for Intelligent Task Offloading in Vehicular Networks

In vehicular edge computing, task offloading optimization is crucial for balancing computational demands with minimizing delays and costs. However, the dynamic nature of the vehicular environment, including vehicle mobility, network topology, and available computing resources, poses significant challenges. This paper presents a task offloading scheme that enables vehicles to dynamically decide between local task execution and offloading to nearby vehicles, edge servers, or the cloud. Our objective is to optimize task offloading by minimizing cost and delay. We integrate graph convolutional networks with deep reinforcement learning to optimize task offloading decisions and achieve our goal. The Graph Convolutional Network is integrated with Deep Reinforcement Learning to enhance network representation and decision efficiency of agent. The optimization problem is formally formulated within the framework of a Markov Decision Process. Simulation results demonstrate the superiority of proposed scheme, which achieves cost-efficiency by maximizing resource utilization, minimizing costs, and optimizing task offloading while reducing task rejection.