基于DFA复合式Agent模型的设计与实现  

Design and implementation of complex Agent model using DFA

作　　者：卢薇薇[1] 蔡自兴[1] 文志强[1] 陈爱斌[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《中南大学学报（自然科学版）》2010年第2期600-608,共9页Journal of Central South University:Science and Technology

基　　金：国家重点基础研究发展计划("973"计划)项目(A1420060159)

摘　　要：针对机器人团队协作检测与跟踪动态目标的需要,提出1种基于有限状态自动机(DFA)的复合式Agent模型。通过结合有限状态自动机的行为状态模型,对复合式Agent模型进行改进,在固定路线的动态目标跟踪实验中,对改进前后的Agent模型实际实验数据进行比较,并将该模型应用于基于区域的多机器人多目标跟踪实验中。结果表明:改进后的Agent模型通过有限状态自动机中的状态抽象,不仅从目标检测与跟踪的角度提高了Agent个体性能,还从社会的角度,提高了群体团队的协作性能;提出的模型通过行为状态模型将动作、决策等与环境信息进行了有效的分离,从而具有较好的可移植性和高扩展性;改进后的Agent模型跟踪偏差期望值与样本方差均降为改进前的一半,为实时的目标协作检测与跟踪提供了有效途径。A deterministic finite automaton （DFA） based complex agent was proposed to assist cooperative object detection and tracking of robots team. This agent was improved with the behavior model which calls different modules depending on the state in DFA. To evaluate the available evidence on the efficacy and feasibility of the agent, the path data generated by the original algorithm was compared with the DFA based agent. Furthermore, it was applied to the region based multi-object tracking experiment. The results show that the DFA based agent not only improves the performance of an agent itself in dynamic cooperative object detection and tracking, but also improves the efficiency of group cooperation in the perspective of sociology. This agent is totally independent of action control unit, decision making and environment information, so that it has good portability and extendibility. It can be found that the expected value of error and the sample variant in the improved agent are reduced to be only half of the previous one＇s. It is proved that the improved agent can provide an effective approach to cooperative object detecting and tracking.

关 键 词：复合式Agent模型 有限状态自动机 动态多目标跟踪 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]