UAV-Assisted Computation Offloading in Vehicular Networks

Abstract

Unmanned aerial vehicle (UAV) technology has recently attracted interest due to its rapid and flexible deployment. It became a key component of several applications such as aerial delivery and precision agriculture. Moreover, with enhanced payloads, e.g., storage and computing, UAVs can support critical services including road traffic monitoring, accident prediction, and connected-automated vehicles (CAVs). Particularly, computing-enabled UAVs permit CAVs' task offloading. However, efficient offloading that accounts for the UAVs' inherent characteristics remains under-investigated. In this context, we propose to study a UAV-assisted vehicular network, where a UAV flies according to a pre-defined come-and-go trajectory and communicates with nearby CAVs to offload their tasks. We target maximizing the ratio of successfully offloaded tasks by jointly optimizing the initial UAV launching point and traveling direction and strategically associating CAVs to the UAV for successful task offloading. Due to the formulated problem's complexity, we propose two approaches to solve it, namely genetic algorithm (GA) based solution and an iterative exhaustive-linear programming (IE-LP) based one. Through experiments, we demonstrate the proposed algorithms' superior performance in terms of task offloading success ratio compared to benchmarks, and in different conditions. These results can serve as guidelines for the development of more sophisticated UAV-enabled task offloading approaches in next-generation wireless networks.