QQ客服 官方微博 反馈留言

题名	移动机器人导航中计算智能方法的研究
作者	赵先章
学位类别	工学博士
答辩日期	2007-06-13
授予单位	中国科学院研究生院
授予地点	中国科学院自动化研究所
导师	杨一平
关键词	移动机器人 多超声传感器融合 DSm 理论 PSO算法 二型模糊集合 降型 mobile robot multi-ultrasonic sensor fusion DSm theory PSO algorithm type-2 fuzzy set type-reduction
其他题名	Research of the application of computational intelligence in the navigation for mobile robot
学位专业	计算机应用技术
中文摘要	移动机器人自主导航问题是机器人学领域的核心问题之一，涉及了路径规划、地图构建等关键技术；计算智能是指计算机利用数学方法模拟人类智能处理传感器信息，完成复杂的智能行为。本文的研究工作围绕移动机器人导航问题中的计算智能方法展开，对于提高机器人智能化水平具有重要意义。本文在详细分析移动机器人及其导航技术发展现状的基础上，改进了几何区间二型模糊逻辑系统，并将DSm理论、PSO算法探索性地应用于机器人学领域，设计并实现了自主开发移动机器人平台的导航系统；利用PSO算法的学习能力提出一种模糊规则自动生成算法。具体而言： 1. 基于DSm这一全新的数据融合理论框架，提出一种环境地图构建方法，可有效地融合超声数据以记录环境信息。该算法中采用模糊集合构建传感器模型，对于多次测量数据采用DSm组合规则进行融合。仿真和实验证明了该算法的性能。 2. 提出一种针对几何区间二型模糊逻辑系统的PSO降型方法，改进了几何区间二型模糊逻辑系统。PSO降型方法将降型问题转换为优化过程，使得几何区间二型模糊逻辑系统推理过程与离散论域上的传统区间模糊逻辑系统相一致。基于改进的几何区间二型模糊逻辑系统，设计实现了机器人沿壁行为和避障行为，证明了PSO降型方法的有效性。 3.针对结构化环境中移动机器人路径规划问题，提出一种粒子群路径规划算法。算法中应用神经网络计算障碍物对路径的约束，并结合路径距离信息构造适应度函数。仿真实验证明了该算法的合理性。 4.针对半结构化室内环境，基于自主研制机器人平台设计、实现了一种混合结构导航系统。系统采用模块化设计思想：人机接口模块应用了基于概念网络的自然语言处理技术；慎思控制模块中采用粒子群路径规划算法对导航任务进行分解，基于DSm理论构建的环境地图作为先验知识；反应控制模块由沿壁行为、避障行为和趋向目标行为组成。导航实验证明了该导航结构的有效性。 5. 针对模糊规则获取问题，提出一种在给定模糊样本集上模糊分类规则自动生成算法。算法中引入了一种基于模糊信息熵的规则新奇性度量，并与规则准确性、规则覆盖度相结合构成了粒子群优化算法的优化目标函数。 “星期六早晨”问题被用来测试算法的性能，其实验结果与模糊决策树归纳法获得的结果进行了详细对比。
英文摘要	Autonomous navigation for mobile robot is one of the key problems in the domain of robotics, and several technologies, such as path planning and map building, are involved. With computational intelligence, computers can simulate human intelligence by mathematical method to deal with sensor information, and implement intelligent behaviors. This dissertation is aimed at the application of computational intelligence in mobile robot navigation, and it is an effective way to improve the intelligence level of mobile robot. In this paper, the research context of robot navigation is studied in details firstly. Secondly, a navigation system is designed and implemented on the self-developed mobile robot with a modified geometrical interval type-2 fuzzy logic system presented and both the DSm theory and PSO algorithm introduced to robotics domain. Thirdly, an algorithm for the generation of fuzzy classification rules is proposed on the base of the learning ability of PSO. Concretely:1. A new method for map building based on Dezert-Smarandache theory (DSm) which is a methodology for data fusion problems is presented to record situation information. Fuzzy logic is adopted to build ultrasonic sensor model, and different measures are combined with DSm combination rules. Simulation and practical experiments are carried out and it is proved that the method given here behaves a good performance.2. A PSO type-reduction method is proposed to modify geometric interval type-2 fuzzy logic system.In PSO type-reduction method, type-reduction is converted into an optimization problem, so the inference principle of geometric interval FLS operating on continuous domain is consistent with that of traditional interval type-2 FLS which work on discrete domain. The reactive behaviors of obstacle avoidance and wall following are implemented through the modified GIT-2FLS. 3. A path planning algorithm based on particle swarm algorithm is proposed for structure situation. The fitness function consists of path length and obstacle constrain which is computed with neural network. The validity of the algorithm is proved in computer simulation. 4. An hybrid architecture of navigation system is designed and implemented on the self-developed mobile robot operated in half-structure situation. The performance of the navigation system is tested in the experiments.5. A new kind of algorithm is proposed for fuzzy rules’ generating from fuzzy data set. The algorithm is carried out to solve the famous “Saturday Morning Problem”, and the result is compared with that from fuzzy decision tree induction method.
语种	中文
其他标识符	200418014628088
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/6020]
专题	毕业生_博士学位论文
推荐引用方式 GB/T 7714	赵先章. 移动机器人导航中计算智能方法的研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2007.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。