An Intelligent Collaborative Decision-Making Method for Multi-Target Tracking in Heterogeneous Unmanned Clusters

With the rapid development of artificial intelligence and robotics, unmanned aerial vehicle (UAV) swarm systems have received widespread attention from domestic and foreign experts, scholars, and research institutions for their applications in military and civilian fields. This research primarily focuses on intelligent collaborative decision-making methods for heterogeneous unmanned swarms in multi-target tracking tasks. In response to the challenges associated with the difficulty of obtaining global states and the complexity of collaborative decision-making in existing methods, a distributed collaborative decision-making strategy based on Multi-Agent Deep Reinforcement Learning (MADRL) has been proposed. This strategy employs a data-driven approach that eliminates the need for precise modeling, utilizing a hierarchical attention mechanism to process heterogeneous information. Additionally, we have designed an experience-sharing training mechanism for heterogeneous swarms to enhance their intelligent collaborative decision-making capabilities and multi-target tracking performance. By establishing a simulation environment, various MADRL algorithms were implemented and experimentally validated. The experimental results show that the proposed method can effectively improve the collaborative tracking performance of UAV swarms. Future work will focus on hyperparameter tuning of algorithms, testing across different swarm scales, and analyzing algorithm scalability in diverse environments.