基于预知博弈树的多无人机群智协同空战算法  

Multi-UAV cooperative swarm algorithm in air combat based on predictive game tree

作　　者：周文卿 朱纪洪 匡敏驰 史恒 ZHOU WenQing;ZHU JiHong;KUANG MinChi;SHI Heng(Department of Computer Science and Technology,Tsinghua University,Beijing 100084,China;Department of Precision Instrument,Tsinghua University Beijing 100084,China)

机构地区：[1]清华大学计算机科学与技术系,北京100084 [2]清华大学精密仪器系,北京100084

出　　处：《中国科学：技术科学》2023年第2期187-199,共13页Scientia Sinica(Technologica)

摘　　要：无人机广义上为不需要驾驶员登机驾驶的各式遥控飞行器.在现代空战中,无人机占据了越来越重要的地位.然而目前关于无人机的空战算法的研究大都是在高度简化的低精度简单场景中进行的,使用的方法也大都受限于已有的专家知识,无法充分发挥智能空战算法的优势.因此,本文对基于预知博弈树的多无人机群智协同空战算法进行了研究.首先使用Unity3D搭建了一套贴近于真实空战场景的仿真环境;然后根据现有的空战知识封装了一套战术机动动作,实现了脚本化的飞机编队,并设计了一套空战态势评估函数;以此为基础提出了基于预知博弈树的空战AI算法框架,通过预知博弈树算法完成了角色分配和机动动作决策的任务,使用XGBoost将其转化为一个在线的实时算法;以状态机算法为baseline,在高逼真度仿真平台上通过空战AI对抗实验,验证了本文提出的空战算法的有效性.An unmanned aerial vehicle(UAV) broadly refers to all kinds of remotely controlled aerial vehicles that do not require pilots to board and drive. Due to its small size, low cost, large quantity, and safety, a UAV is widely used in modern air combat. However, current studies on UAVair combat algorithms are mostly conducted in highly simplified scenarios with low precision. In addition, most of the methods used in these studies are limited by existing expert knowledge and cannot fully exploit the advantages of intelligent air combat algorithms. Therefore, this study investigates a multi-UAV cooperative swarm algorithm in air combat based on the predictive game tree. First, Unity3D is used to build a simulation environment close to a real air combat scene. In addition, a human-machine interaction environment, including UI, VR, weather system, and multifunctional screen, is realized. Then, a set of tactical maneuvers is encapsulated on the basis of existing air combat knowledge. Scripted aircraft formations are realized, and a set of air combat evaluation functions is designed. On this basis, an air combat artificial intelligence framework based on the predictive game tree is proposed. This algorithm completes the task of role assignment and maneuver decision-making, and XGBoost is used to transform it into an online real-time algorithm. Using the state machine algorithm as the baseline, the effectiveness of our air combat algorithm is verified through air combat confrontation experiments on a high-precision simulation platform.

关 键 词：多无人机群智协同 智能空战 预知博弈树算法 高逼真度仿真平台 自主控制 

分 类 号：V279[航空宇航科学与技术—飞行器设计] E91[军事]