PATROLLING TASK PLANNING FOR THE MULTI-LAYER MULTI-AGENT SYSTEM BASED ON SEQUENTIAL ALLOCATION METHOD

摘要

The unmanned aerial vehicle (UAV) swarm has developed rapidly in recent years, especially the UAV swarm with sensors which is becoming common means of achieving situational awareness. In this paper, we develop a scalable, online and myopic algorithm for the multi-layer multi-agent system continuously patrolling problem. The main goal of the multi-agent system is to collect information as much as possible. We formulate this problem as Partially Observable Markov Decision Process (POMDP). The algorithm includes information dimensionality reduction representation, inter-layer information interaction, online heuristic function and sequential allocation method, which effectively improve the collected information and reduces the computational complexity. In addition, as the layer increases, this algorithm can guarantee the patrolling performance of the multi-agent system without increasing the computational complexity for each sub-leader. Finally, the empirical analysis shows that our algorithm has many advantages, which has theoretical and practical significance.