Multi-UAV Collaborative Surveillance Network Recovery via Deep Reinforcement Learning

As a typical nonterrestrial network (NTN)-enabled Internet of Things (IoT), the multi-Unmanned aerial vehicle (UAV) collaborative surveillance network boasts efficient capabilities in information collection and transmission. However, manufacturing techniques and environmental conditions can lead to UAV failures, thereby impacting network performance. To recover the performance of the multi-UAV collaborative surveillance network, the effective movement of multiple UAVs is under investigation in order to improve target coverage and data backhaul efficiency. In this article, we present a novel multiagent deep reinforcement learning-based algorithm to accomplish network recovery. The proposed algorithm employs a multihead attention network to facilitate coupled multiobjective learning and overcome the limitations imposed by local information. Additionally, a stable learning method is introduced to address the difficult convergence problem caused by dynamic topology changes due to UAV motion. Experimental results show that the proposed algorithm can generate feasible multi-UAV motion strategies, effectively facilitating network recovery and improving the performance of the multi-UAV collaborative surveillance network in different scenarios.