Scholarlink

<p>In radar threat environments, collaborative control of 3D fixed-wing UAV swarm faces the challenge of balancing tight formation maintenance with internal collision-free. To address this, we propose a Collision-free Policy for 3D fixed-wing UAV Swarms Based on Scale–scalable (CPSS), where Scale–scalable means the capability of observing scale generalization. CPSS is built on multi-head attention mechanisms that possess the scale–scalable attribute and multi-agent reinforcement learning algorithms, combined with a carefully designed reward function, to achieve a balance between maintaining tight swarm formations and avoiding internal collisions. The introduction of scale–scalable attributes not only enhances the CPSS’s adaptability to dynamic observation but also endows the CPSS with the capability to deconstruct complex control tasks into a simple multi-stage training process while combining parameter sharing and knowledge transfer to accelerate the training process. Experimental results show that CPSS demonstrates excellent scalability and reliability in handling cooperative tasks of UAV swarm while maintaining great collision-free performance in radar threat environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CPSS: Collision-Free Policy for 3D Fixed-Wing UAV Swarms Based on Scale–Scalable in Radar Threat Environments

  • 期刊论文

  • 发表日期：

    2025年7月26日

查看全文
  • Bolin Zhang,
  • Cailun Wu,
  • Linghua Wu,
  • Xuebin Zhuang

摘要

In radar threat environments, collaborative control of 3D fixed-wing UAV swarm faces the challenge of balancing tight formation maintenance with internal collision-free. To address this, we propose a Collision-free Policy for 3D fixed-wing UAV Swarms Based on Scale–scalable (CPSS), where Scale–scalable means the capability of observing scale generalization. CPSS is built on multi-head attention mechanisms that possess the scale–scalable attribute and multi-agent reinforcement learning algorithms, combined with a carefully designed reward function, to achieve a balance between maintaining tight swarm formations and avoiding internal collisions. The introduction of scale–scalable attributes not only enhances the CPSS’s adaptability to dynamic observation but also endows the CPSS with the capability to deconstruct complex control tasks into a simple multi-stage training process while combining parameter sharing and knowledge transfer to accelerate the training process. Experimental results show that CPSS demonstrates excellent scalability and reliability in handling cooperative tasks of UAV swarm while maintaining great collision-free performance in radar threat environments.