摘要
开展基于深度强化学习的无人艇集群动态博弈对抗中的协同围捕决策研究。建立受距离和相对角度影响的无人艇围捕环境模型,利用基于策略网络和双评价网络的深度强化学习方法求解围捕策略,立足协同围捕任务,基于距离和相对角度设计引导型奖励函数,避免奖励稀疏。仿真结果表明,基于深度强化学习的红方无人艇集群能够对蓝方无人艇进行有效的协同围捕。研究成果可为无人艇集群博弈对抗演练提供参考。
Aiming at the problem of dynamic game confrontation between red and blue unmanned surface vehicles(USVs),this paper carried out a research on cooperative hunting decision of USVs based on deep reinforcement learning.The model of USVs coordinated hunting environment including distance and angle was established,and the deep reinforcement learning method based on policy network and two evaluation networks was used to solve the coordinated hunting strategy.Based on the coordinated hunting task,the guided reward function was designed based on distance and relative angle to avoid the problem of reward sparsity.The simulation results show that the red-square USVs based on deep reinforcement learning can effectively and cooperatively hunt the blue-square USVs.The efficiency of this strategy can provide significant reference for game confrontation drills of USV swarm.
作者
苏震
张钊
陈聪
刘殿勇
梁霄
SU Zhen;ZHANG Zhao;CHEN Cong;LIU Dianyong;LIANG Xiao(Industrial Development Department,Zhuhai Yunzhou Intelligent Technology Co.,Ltd.,Zhuhai 519080,China;School of Naval Architecture and Ocean Engineering,Dalian Maritime University,Dalian 116026,China)
出处
《兵器装备工程学报》
CAS
CSCD
北大核心
2022年第9期9-14,共6页
Journal of Ordnance Equipment Engineering
基金
国家自然科学基金项目(52271302)。
关键词
无人艇集群
博弈对抗
深度强化学习
协同围捕
USV swarm
game confrontation
deep reinforcement learning
coordinated hunting
