A Communication Gating Control Scheme for Multi-UAV Cooperative Maritime Search based on Deep Reinforcement Learning

For search and rescue operations in vast maritime areas, using multiple unmanned aerial vehicles (UAVs) for cooperative search while adjusting the observation granularity of UAVs during the search process is a promising and efficient approach. However, the limited sensing and communication range of UAVs makes it difficult to obtain global information, and pose a challenge for achieving a globally optimal cooperative search strategy. To address the communication constraints, such as limited communication range and the dynamic communication topology, this paper proposes a communication gating control scheme for multi-UAV cooperative multi-granularity search based on multi-agent reinforcement learning (MARL). Through multi-hop communication, we control the UAVs to maintain a dynamic communication formation using simple single-hop position signals during the search, and employ costly multi-hop communication to transmit critical observation information based on the weights of multi-granularity observations, thereby improving global cooperative search efficiency. We evaluate the proposed algorithm with simulations, and demonstrate the effectiveness of the scheme.