Cooperative Multi-agent Reinforcement Learning for Multiple Anti-aircraft Target Surveillance

Abstract

In this study, a cooperative monitoring task for multiple anti-aircraft targets is investigated using multiple unmanned aerial vehicles (UAVs) equipped with stereo vision sensors. To accomplish this task, these UAVs must cooperate with other UAVs within a three-dimensional (3D) space to track multiple targets and avoid collisions. To address this challenge, we propose a cooperative multiagent reinforcement learning (MARL) scheme that can make intelligent flight decisions to enable multiple UAVs to perform cooperative surveillance tasks. The main contributions of this study are as follows. First, to the best of our knowledge, this study is the first attempt to address aerial multi-target surveillance using multiple UAVs via MARL in 3D space. Second, we adopt a hybrid approach that integrates low-level rule-based controllers and high-level control policies to enable agents to learn the high-level goals set through reinforcement learning.