Cooperative aerial and ground vehicle trajectory planning for post-disaster communications: An attention-based learning approach

Abstract

The ground infrastructures are highly susceptible to disruption after disasters, which causes the paralysis of communication. In this case, solutions besides original architecture are needed to meet the requirements of communication. Since unmanned aerial vehicle (UAV) can be quickly sent to disaster events to provide temporary connection due to its agility and mobility, it is suitable for performing disaster relief. Nevertheless, the limited onboard energy restricts the UAVs from fulfilling such persistent tasks. To this end, we introduce a ground vehicle carrying backup batteries to handle the energy issue of the UAV. Considering a time-constrained disaster-affected area, we propose a cooperative trajectory planning scheme to provide emergency communications swiftly and timely. The goal of our optimization task is to minimize the total cost of the mission, which consists of the operation cost for mission completion and the penalty cost for latency. We show that the task can be regarded as an extension of traveling salesman problem with soft time window constraints, which is NP-hard in general, and we propose a novel attention-based deep reinforcement learning with a sequential model strategy to learn the policy for the UAV's visiting order, based on which the trajectories of the UAV and ground vehicle are jointly designed. Numerical results show that our proposed attention-based trajectory planning scheme is effective and efficient, providing a guideline for the system design of post-disaster communications.