多机器人系统任务分配效用评价算法  

Utility evaluation algorithm for task allocation in multi-robot systems

作　　者：黎萍[1] 朱军燕 杨宜民[3] 

机构地区：[1]电子科技大学中山学院,广东中山528402 [2]中山出入境检验检疫局技术中心,广东中山528403 [3]广东工业大学自动化学院,广东广州510090

出　　处：《计算机工程与设计》2013年第9期3288-3292,共5页Computer Engineering and Design

基　　金：电子科技大学中山学院博士启动基金项目(410YKQ01);国家科技型中小企业技术创新基金项目(12C26214405188);广东省教育部产学研基金项目(2011B090400371)

摘　　要：为了提高复杂环境中多机器人系统任务分配的决策质量,获取准确、客观的效用评价,提出了一种基于自适应神经-模糊推理系统(adaptive neuro-fuzzy inference system,ANFIS)的效用评价算法ANFIS-UE。设计了基于ANFIS的效用评价网络结构,并采用Q学习对效用评价网络的参数进行学习。利用ANFIS优越的函数逼近能力和泛化能力,提高了效用函数的学习效率,能够对连续的状态输入产生连续的效用评价值。实验结果表明,该算法获得的效用评价相对更准确,从而提高了任务分配方案的质量。An ANFIS （adaptive neuro-fuzzy inference system） based utility evaluation algorithm is presented to get obiective utility evaluation and optimize task allocation schemes for real-time multi-robot systems which work in collaboration and adver- sarial, dynamic environment with unreliable communication. A network for ANFIS based utility evaluation algorithm is de- signed, whose parameters are adjusted via Q-learning. Because of the superior function approximation ability and generalization capacity of ANFIS, learning efficiency is improved and continuous utility evaluation for continuous states is gotten by this algo- rithm. Simulation experiment shows that utility evaluation gotten by this algorithm is relatively more accurate and objective. Consequently, task allocation schemes are improved.

关 键 词：自适应神经-模糊推理系统 Q学习 效用评价 多机器人系统 任务分配 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]