Formation Tracking for Multiple UAVs Based on Deep Reinforcement Learning

In this paper, the formation tracking problem is investigated for unmanned aerial vehicles (UAVs) using the deep reinforcement learning (DRL) technique. A cascaded model-free controller is constructed and trained to resolve this problem. First, three decoupled attitude controllers are trained to track the desired attitude angles with ensured stability. Then, based on the trained attitude controllers, a distributed formation control protocol is introduced to train the position controllers for formation flight. The proposed method allows multiple UAVs to form and maintain a predefined formation pattern without dynamic parameters and global information exchange. Finally, numerical simulations are conducted to demonstrate the effectiveness and advantages of the proposed results.