Multi-Agent Multi-Target Search with Multi-Head Attention
Huiqin Pei, Zilong Luo · 2025
In swarm drone applications, the multi-target self-organizing search (SOS) problem in unknown environments has demonstrated significant potential. In this problem, agents must collaborate to search for multiple dynamic targets, while only observing their immediate surroundings. However, the computational complexity and uncertainty inherent in traditional multi-agent search processes have been key factors limiting their performance. Therefore, this paper focuses on the multi-target search problem. First, a multi-agent search environment, including agents, targets, and obstacles, is constructed. Subsequently, a multi-agent multi-target search method (MASOS) is proposed for complex obstacle environments. This method integrates the Actor-Critic reinforcement learning algorithm and multi-head attention (MAAC) for UAV swarm collaborative control, enabling agents to focus more on critical information such as other agents, targets, and obstacles during their observation. The goal is to enhance the search, collaboration, and obstacle avoidance capabilities in multi-target self-organizing search tasks. Experimental results show that the MASOS method outperforms other commonly used multi-agent reinforcement learning (MARL) algorithms in terms of coordination strategies. In large-scale self-organizing search tasks, the capture success rate of the MASOS method approaches 100%. Finally, experimental validation confirms the effectiveness of the MASOS method.