Multi-UAV Maneuvering Targets Tracking Based on Mobility-Focused Multi-Agent Deep Deterministic Policy Gradient

This paper investigates the problem of autonomous tracking of multiple dynamic ground targets by multiple Unmanned Aerial Vehicles (UAVs). The purpose of the multi-UAV system is to maintain continuous coverage of these moving targets. However, when the targets exhibit high mobility, it becomes challenging for the UAVs to effectively track them. To address this, we propose a Mobility-Focused Multi-Agent Deep Deterministic Policy Gradient (MF-MADDPG) model to learn a cooperative multi-UAV policy for tracking multiple targets. Specifically, we focus on the mobility of the UAVs while tracking highly mobile targets and design a mobility reward function to describe the difference of mobility matching between the UAVs and the targets. Performance tests conducted in the Gazebo simulation environment, based on the Robot Operating System (ROS) and the PX4 autopilot, demonstrated that the proposed MF-MADDPG model outperforms the traditional MADDPG model in tracking highly mobile targets.