UAV-Assisted Priority-Aware Resource Allocation for DT-Enabled Vehicular Edge Computing Networks

Abstract

Digital twin (DT)-empowered vehicular edge computing (VEC) enables roadside units (RSUs) to maintain synchronized virtual replicas of vehicles while processing their offloaded computation tasks, yet both services compete for the same limited communication and edge computing resources. This competition becomes critical when emergency and authorized priority vehicles, whose presence can be detected by uncrewed aerial vehicles (UAVs) acting as low-altitude sensing nodes, demand faster twin synchronization and stricter service deadlines than ordinary vehicles. In this paper, we propose a UAV-assisted priority-aware DT-VEC framework in which UAVs report priority-detection confidence to RSU edge servers, and each RSU jointly allocates uplink bandwidth, transmit power, and edge computing resources for twin maintenance, task processing, and priority-service provisioning. We formulate the allocation as a delay-constrained utility-maximization problem and solve it with a priority-aware residual multi-agent twin-delayed deep deterministic policy gradient (PA-MATD3) scheduler, which composes a priority-aware allocation prior with a bounded residual actor and stabilizes learning through twin critics, conservative residual regularization, and a feasibility projection that keeps every executed action within the resource budgets. Extensive simulations show that PA-MATD3 achieves a better overall tradeoff among task-processing delay, priority-service delay, and deadline-violation behavior, and provides robust performance under varying vehicle densities, edge capacities, priority loads, and UAV sensing ranges.