Learning-Aided UAV-Cooperation Reduces the Age-of-Information in Wireless Networks

Abstract

Unmanned aerial vehicles (UAVs) can enhance data collection for ground sensing nodes (SNs). Given the modest battery capacity of UAVs and the limited communication range of SNs, it is crucial to conceive efficient trajectory coordination for UAVs. However, existing studies simply decouple the joint trajectory planning policy of multiple UAVs into independent local policies, preventing their cooperation and hence limits the performance. Inspired by the observation that sharing messages among agents can promote their cooperation, we investigate the communication-assisted decentralized trajectory planning policy of multi-UAV wireless networks. Our goal is to minimize the overall energy consumption of UAVs and the average age of information of all SNs. To harness the encoded messages for learning a sophisticated policy, we conceive a communicationassisted distributed training and execution framework, and propose a communication-aided decentralized trajectory control algorithm. Our simulation results show that the proposed algorithm substantially outperforms the state-of-the-art deep reinforcement learning based methods, at a modest communication overhead.