A Multi-UAV Cooperative Tracking Strategy Based on Deep Reinforcement Learning

This paper proposes a model for cooperative combat among multiple Unmanned Aerial Vehicles (UAVs) in an isomorphic scenario. A significant challenge with multi-agent reinforcement learning algorithms used in such scenarios is the sparse reward problem. To overcome this, a novel reward design strategy based on reward shaping is suggested, which combines discrete targeted reward with continuous procedural reward functions. This strategy aims to guide the agents towards optimal strategy learning. The QMIX algorithm is used to train multiple agents, and a detection capability evaluation indicator is designed to assess the performance of this approach. The effectiveness of this method is verified through simulations conducted on the MaCA platform. This technique offers useful insights into solving the sparse reward problem.