Topology-Aware Multi-Hop Task Offloading via GNN-A Routing in Vehicular Networks

Vehicular edge computing (VEC) enables low-latency task execution by deploying computation at vehicles and roadside units (RSUs). However, high mobility and frequent topology variations cause multi-hop offloading paths to break easily, leading to task interruptions and performance degradation. Existing path selection methods, typically relying on static link metrics or exhaustive search, fail to adapt efficiently in large-scale dynamic networks. To address these challenges, we propose a joint optimization framework that integrates task scheduling with topology-aware routing. A multi-agent A3C module determines the offloading ratio and destination node, while a graph neural network-enhanced A*(GNN-A*) planner constructs reliable, low-latency multi-hop paths by learning dynamic topological representations. The two modules operate collaboratively in an adaptive feedback loop, enabling online policy refinement under varying vehicular conditions. Simulation results show that the proposed framework significantly outperforms state-of-the-art baselines, achieving lower delay, higher task completion ratios, and better scalability as vehicle density increases.