A Multi-UAV Collaborative Deployment Scheme to Improve Regional VANET Connectivity
Abstract
Vehicular ad hoc networks (VANETs) support real-time information exchange and traffic-aware services, but high mobility, uneven traffic distribution, signal-controlled interruptions, and urban wireless impairments make stable regional connectivity difficult to maintain. Fixed roadside units (RSUs) are costly and cannot adapt to dynamic traffic conditions. This article proposes a multi-UAV collaborative deployment scheme (MUCD) to improve regional VANET connectivity. First, a MAC/PHY-aware road-connectivity estimator uses successful multihop probing rather than geometric distance alone. Second, a graph-aware regional metric aggregates intersection-pair path reliability. Third, a multicycle model jointly considers uncrewed aerial vehicle (UAV) deployment, migration overhead, connectivity fluctuation, energy consumption, and charging feasibility. A problem-structure-oriented framework combines improved Gray Wolf optimization for road selection, Hungarian matching for UAV–road assignment, and convex optimization for energy and charging scheduling. Simulations on an OpenStreetMap-derived urban network show that MUCD consistently achieves better packet delivery ratio (PDR), throughput, and end-to-end delay performance than all evaluated conventional and recent learning-based benchmarks under low- and high-density traffic. Compared with Q-SDAM-adapted, MUCD improves PDR and throughput by 2.92%/1.82% and 3.21%/1.42%, respectively, and reduces end-to-end delay (EED) by 3.73%/1.93%. The corresponding gains over SONA-reinforcement learning (RL)-adapted are 4.73%/3.62%, 5.08%/3.53%, and 6.10%/4.41%. MUCD requires no offline policy training. With 80 hybrid iterations per decision and a profiled mean of 0.56460 s per iteration, its recalculated mean decision time is 45.168 s. These results demonstrate its effectiveness and sufficient decision efficiency for cycle-level regional VANET deployment.