机器地图信息加工模型  

Information Processing Model of Machine Maps

作　　者：游雄[1] 李科[1] 田江鹏 杨剑 余岸竹[1] 贾奋励[1] YOU Xiong;LI Ke;TIAN Jiangpeng;YANG Jian;YU Anzhu;JIA Fenli(School of Geospatial Information,Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学地理空间信息学院,河南郑州450001

出　　处：《武汉大学学报（信息科学版）》2024年第4期516-526,F0002,共12页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金(42130112,42371479)。

摘　　要：无人平台对复杂环境的自主认知能力是制约其广泛应用的关键问题,已成为当前认知科学、人工智能、地图学等领域的研究热点。在机器地图的概念模型和认知特点的基础上,为进一步实现机器地图信息存储、处理、交互与学习的形式化表达,从人机优势融合的视角提出了机器地图信息加工模型;构建了包括感知地图、工作地图和长时地图的环境表达模型,从观测视角、参考系、信息抽象度、数据结构、描述精度和准确度等方面分析了表达模型结构;提出了测制用一体信息交互过程,分析了包括环境感知、制图、推理和决策为一体的持续迭代环境信息处理过程;建立了持续自主学习模型,分析了该模型在学习过程、学习内容和持续机制方面的特点。开展了两组实验,对信息加工模型的可行性进行了验证:一是通过模拟测制用一体交互过程,提高了基准模型的长距离自主导航能力;二是通过模拟任务驱动的工作地图建图过程,强化环境要素与任务的相关性,提高了任务决策的效率。信息加工模型的研究能够为机器地图技术体系和系统架构的确立提供理论依据,进而指导机器地图应用系统的开发。Objectives:The autonomous cognitive ability of unmanned platforms to complex environments is a key issue restricting their extensive real-world application,and has become a research hotspot in cognitive science,artificial intelligence,cartography and other fields.From the perspective of complementary advantages of human-machine,the machine maps information processing model is proposed to achieve the logical expressions of machine maps information storage,processing,interaction and learning based on the conceptual model and cognitive characteristics of machine maps.Methods:An environmental representation model including perception map,working map and long-term map is constructed,and the map structure is analyzed from the perspective of observation angel,reference frame,information abstraction degree,data structure and description precision.An integrated information exchange pattern for measurement,production and application is established,including environment perception,mapping,work and decisionmaking,and meanwhile the continuous iterative environmental information processing process of this model is analyzed.A continuous autonomous learning model is also established,and the characteristics of the model in terms of learning process,learning content and persistence mechanism are analyzed.Results:Two experiments are carried out to verify the feasibility of the information processing model.The first experiment im-proves the long-distance autonomous navigation capability of the benchmark model by simulating the integrated sensing,mapping and decision-making capacities.The second experiment enhances the correlation between environmental factors and tasks by simulating the task driven process of creating working map,thereby improving the efficiency of task decision-making.Conclusions:The proposed model is able to provide a theoretical basis for the establishment of machine maps technology structure and system architecture,which in turn guides the development of machine maps system application.

关 键 词：信息加工模型 机器地图 测制用一体 持续自主学习 

分 类 号：P208[天文地球—地图制图学与地理信息工程]