群机器人系统对抗环境设计  

Design of Adversarial Environment for Swarm Robot System

作　　者：许寅 许跃彩 吴第旻 XU Yin;XU Yuecai;WU Dimin(Wuhan Digital Engineering Institute,Wuhan 430205;Army Engineering University of PLA,Wuhan 430075)

机构地区：[1]武汉数字工程研究所,武汉430205 [2]陆军工程大学,武汉430075

出　　处：《舰船电子工程》2024年第12期18-20,共3页Ship Electronic Engineering

摘　　要：群机器人系统是一项研究如何用大量个体机器人组成系统来完成超出单个机器人能力的任务的研究领域。与传统的协作环境相比,群机器人对抗环境要更复杂且难以训练。然而,对抗环境的研究至关重要,其可应用于足球或无人机对抗等诸多领域。论文将近端策略优化算法与自博弈和课程学习等策略相结合,成功地训练出了对抗环境中的群机器人系统。Swarm robot system is a research field studies how systems composed of a large number of individual robots can be used to accomplish tasks beyond the capability of a single robot.Compared with traditional cooperative environment,the adversarial environment is much more complex and hard to train.However,the research of adversarial environment is vital,for it can be ap⁃plied in many fields like football game or drone confrontation.The paper successfully trains a swarm robot system in adversarial envi⁃ronments by combining proximal algorithms with self-play and course learning strategies.

关 键 词：群机器人对抗环境 自博弈 课程学习 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]