Recurrent Graph Reinforcement Learning Framework for Task Offloading in Internet of Vehicles

The rapid growth of the Internet of Vehicles (IoV) has increased the demand for low-latency, real-time, and priority-aware applications recently. Current Mobile Edge Computing (MEC) enabled vehicular architecture often fails to address these challenges effectively, which leads to lower quality of service. In this paper, we propose a joint mobility-aware task offloading and priority-based resource allocation framework for MEC-enabled IoV. We formulate a priority-aware multiobjective optimization problem to minimize latency and energy consumption of the offloaded task, which is a mixed-integer nonlinear problem (MINLP) and NP-Hard to solve. In this regard, we develop a Recurrent Graph Reinforcement Learning (RGRL) framework integrated with a Graph Neural Network (GNN). The GNN captures the spatio-temporal relationships among vehicles, Roadside Units (RSU), and edge servers, enabling intelligent and adaptive decision-making in dynamic environments. Simulation results demonstrate that the proposed RGRL-GNN approach outperforms state-of-the-art works in terms of latency reduction, energy efficiency, and successful task completion rate.