【作者】 余冠华；

【导师】 杨淮清；

【作者基本信息】 沈阳工业大学 ， 计算机软件与理论， 2010， 硕士

【摘要】 本文以超市机器人的作业环境为背景,对知识表示、基于特征库的超市环境区域建模、商品建模、货架建模等问题进行了深入研究。通过建立并使用超市样本库,使机器人在超市环境中能够快速准确地建立起环境模型。并利用LVQ神经网络对环境样本进行学习,使机器人能够较准确地分辩出环境中障碍物的类型,为在未知环境行走时的路径规划与障碍回避提供基础。论文首先对人工智能的发展方向、机器人的发展方向、机器人学所涉及的知识与领域、机器人环境建模的相关内容作了综述性介绍,分析了多种建模方法在不同应用背景下的优缺点。通过分析商品在超市环境模型中的固有属性及其与超市中其他对象的关系,对商品进行了语义与几何形态建模。对于商品建模问题从商品的定义、商品几何形态、商品的位姿等方面进行了研究,使用商品的隶属树对其进行分类及编码。通过分析超市整体环境的特性与单元对象的形态,给出了基于样本特征库的建模方法。建模过程中采用了多知识表示相结合、LVQ神经网络、二维半结构表示等方法。利用C++Builder开发平台对应用于超市机器人的环境知识、语义级环境予以了定义与管理,实现了区域建模、货架建模、商品建模、障碍物识别。依据环境知识表示机制,利用多种知识表示相结合的方式来定义环境、货架、商品之间的关系,建立起机器人语义级环境模型。利用二维半的思想对超市中的障碍物、货架、商品建立形体模型。使用样本量化与学习的方法,为静态建模方法运用到动态建模中,提供了新的方法与思路。课题最后给出了若干关键难题的求解途径,比较详细地阐明了系统的设计和实验结果。更多还原

【Abstract】 Based on supermarket environment of robot operating, this paper conducts a depth research on knowledge representation, merchandise modeling, shelf modeling and map modeling. Created sample libraries so that the supermarket robot can establish the environmental model quickly and accurately. Using the LVQ neural network system to learn environmental samples to distinguish the type of obstacles accurately. These methods provides a data support for path planning and obstacle avoidance in an unknown area for supermarket robot.This paper gives an overview introduction to the background of the subject, the theory of robotics and artificial intelligence, the development of robot and environment modeling. By analyzing the advantages and disadvantages of several modeling methods in different applications, this paper gives models for the supermarket environment both in semantic and structure. This paper makes a classification and encoding for merchandise and conducts a research on definition, geometry modeling, location and pose of the merchandise. Gives a modeling method based on sample characteristics libraries using multi-knowledge representation, LVQ neural network, 2.5D description.Makes a definition and management for the knowledge and semantic of environment using C + + Builder development platform. Realizes map modeling, shelf modeling, merchandise modeling and obstacle recognition. Based on knowledge representation this paper constructs a relationship model between environment, shelves, and merchandise using a combination of multi-knowledge representation methods. Gives solutions to the key problems, described the system designing, experiment and the test results, points out the scientific value of this subject, its shortcomings and further research objectives.更多还原