UAV Swarm Path Planning for Sensor Data Collection via Double Prepartitioned Deep Q-Network

Collision-free path planning is a significant challenge in deploying unmanned aerial vehicle (UAV) swarms for sensor data collection tasks, particularly in uncertain environments. This paper addresses the problem of designing flight strategies for energy-constrained UAV swarms using deep reinforcement learning. We propose a locally-aware double pre-partitioned deep Q-network (DPPDQN) for flight planning of multiple UAVs in large-scale uncertain industrial environments. This algorithm aims to collect data from ground sensors with communication constraints while avoiding mutual collisions. In the proposed algorithm, we first introduce a node pre-partitioning scheme using an improved clustering algorithm, which efficiently reduces the number of access points for UAVs through effective data aggregation. Subsequently, we employ an area pre-partitioning algorithm that segments the detection area into several subregions based on the number of UAVs available for operation, thereby preventing collisions among the UAV swarm during task execution. Finally, we propose a deep Q-Network algorithm incorporating a convolutional neural network, which utilizes the local perceptions of UAVs as state inputs to address the “curse of dimensionality" occurring in large-scale environments. Systematic simulation experiments validate the effectiveness and advantages of the proposed method.