VECCTO: An Collaborative Task Offloading Framework for UAV-Assisted Vehicular Edge Clusters

Unmanned Aerial Vehicles (UAVs) provide Vehicle Edge Clusters (VECs) with robust signal coverage and data offloading capabilities, but they have limitations when dealing with heterogeneous cluster environments and dynamic task resource requests. This paper presents VECCTO, a collaborative task offloading framework designed specifically for UAV-assisted VEC that seeks to address the scalability and dynamics challenges in VEC task offloading. VECCTO incorporates the Asynchronous Advantage Actor-Critic (A3C) mechanism, which includes a centralized global Critic in the base station and a distributed local Actor in the UAV. It obtains and trains the global state value function of the entire VEC via the target network and independently updates the task offloading policy parameters of each sub-VEC based on the policy gradient, thus autonomously learning the task offloading policy. We evaluate the efficacy of VECCTO using the modified Google Cluster Usage Tracing (GCT) dataset. The results demonstrate that VECCTO efficiently manages task offloading in UAV-assisted VECs, achieving low latency while assuring load balancing among sub-VECs, which makes it a promising solution for next-generation IoV ecosystems.