Cooperatively pursuing a target unmanned aerial vehicle by multiple unmanned aerial vehicles based on multiagent reinforcement learning

Abstract

This article considers a pursuit-evasion problem, in which multiple unmanned aerial vehicles (multi-UAVs) are required to cooperatively pursue a moving target UAV. The dynamics of the three-dimensional partially unknown environment makes it hard to deal with this problem. A multiagent reinforcement learning approach is proposed. This algorithm is extended from cooperative double Q-learning by taking into account of the characteristics of the studied pursuit problem. To study the performance, a simulator is developed, which concerns the dynamic nature of the environment and flight constraints of UAV. Compared with several state-of-the-art reinforcement learning algorithms, the proposed algorithm provides better policy such that the pursuing multi-UAVs can effectively chase the moving target UAV.