Pressure Guided Hierarchical Offloading and Heterogeneous Resource Selection for UAV-assisted Vehicular Edge Computing Networks: An Perspective of Future Task Prediction

Abstract Vehicular edge computing can combine local vehicles, cooperative vehicles, unmanned aerial vehicles (UAVs), and roadside units, but current observations do not fully describe the task pressure that may appear in later slots. This paper presents Pressure Guided Hierarchical Offloading (PGHO), which summarizes predicted workload, task arrivals, urgency, and deadline changes as future task pressure before heterogeneous resource selection in UAV-assisted vehicular edge computing networks. The upper policy selects an execution node using current resource conditions together with this future information, and the lower policy then controls UAV movement and the offloading ratio under the link, queue, and resource conditions of the selected node. Across normal load, high load, and tight deadline scenarios, PGHO achieves 0.4301 s average delay and 99.705% task success ratio. Compared with multiagent deep deterministic policy gradient, soft actor critic, and hybrid multiagent proximal policy optimization, average delay is reduced by 21.1%, 12.1%, and 7.4%, respectively. PGHO also keeps average delay close to task flow greedy scheduling while improving task success ratio by about 2.62 percentage points. Component removal experiments show that future task information and hierarchical decision making mainly reduce average and tail delay, while action masking supports reliability under tight deadlines. The results indicate that PGHO maintains low delay and high task completion across different task pressure conditions.