Cost-Efficient Computation Offloading and Vehicle Routing for MEC-Assisted Drone-Vehicle Collaborative Inspection Systems

Abstract

Drone-vehicle collaborative inspection system, which combines the long-range strength of a vehicle and the speed, agility, and direct-flight capability of drones, has been regarded as a promising scheme to facilitate the tasks of smart city inspection. Specifically, the vehicle carries multiple drones to successively visit several regions, and parks in each region to launch the drones to perform the sensing data collection and processing tasks. At the same time, the drones access a local mobile edge computing (MEC) server in the region for computation offloading, accelerating the sensing data processing. This leads to an emerging application system termed as the MEC-assisted Drone-Vehicle Collaborative Inspection System (MDV-CIS). However, the scheduling of the drones and vehicle should be jointly optimized for the cost-efficient implementation of the MDV-CIS. To this end, we formulate a system cost minimization problem to reduce the total energy consumption of the drones and MEC servers, and the total travel cost of the vehicle. Then we decouple the problem into two subproblems, i.e., the computation offloading problem for the MEC-assisted drone inspection in each region and the vehicle routing problem for the vehicle inspection across the regions. In the computation offloading problem, we jointly optimize the binary offloading decisions, bandwidth allocation among the drones, and CPU frequency allocation of the MEC server. In the vehicle routing problem, we adopt a multi-head self-attention based deep reinforcement learning approach to enable and enhance the learning ability of the vehicle that decides how to assign the orders of visits in different regions to reduce the total travel cost. Finally, simulation results show the significant advantage of the cost savings for the MDV-CIS.