Intelligent Multi-Target Tracking for UAV Swarms with Information Sharing

Recent advancements in UAV swarm technology have widened its application in multi-target tracking, particularly in surveillance and environmental monitoring. However, in dynamic environments with multiple moving ground targets, UAV swarms must cooperatively tackle challenges like collision avoidance and energy efficiency to ensure mission success. To overcome the challenge, we propose a Multi-Agent Deep Reinforcement Learning algorithm, integrated with an efficient information sharing mechanism. This algorithm consists of two main components: a ground information center that aggregates and broadcasts relevant historical data from each UAV, enhancing the available decision-making information, and a distributed action selection model where each UAV chooses its actions based on the shared information. The model also incorporates two performance indicators—the actual performance and the counterfactual performance—to optimize the information selection process. Simulations demonstrate significant performance improvements compared to existing baseline methods, including reduced energy consumption, more accurate target tracking, and fewer collisions.