【作者】 王薇；

【导师】 魏世民；

【作者基本信息】 北京邮电大学 ， 机械电子工程， 2009， 博士

【摘要】 移动机器人在军事和国民经济领域有着广泛的应用前景。本文对基于多传感器的的移动机器人在固定区域非结构化动态环境下的全局定位和运动控制的一些关键技术以及传感器引导进行了较为深入的研究。提出了多种新的便于工程实现的环境感知、路径规划、行为控制的模型和算法。主要内容概括如下:1.研究并构建了一个由移动机器人和无线通信系统组成的自主导航控制实验系统。平台硬件系统各模块采用独立微处理器完成控制及信息处理功能,实现了多模块并行运行。模块之间采用的并行通信方式提高了信息传输速度,同时简化了系统的软件设计。无线通信系统基于802.11b、以无线路由器为中心的基础模式,在嵌入式平台构建,实现移动机器人的远程监控。实验表明数据传输稳定、正确率满足远程控制的要求。2.研制了一种基于多微处理器的分布式超声探测系统。该探测系统由传感器工作模式控制模块和智能超声测距模块阵列组成。由独立的微处理器控制一个超声传感器,完成测距信息预处理,并可根据不同的控制策略实现分组并行工作,提高了探测系统的实时性。3.研究了基于模块化层级的基于行为结构的行为控制策略。以面向任务的方式分解控制系统,为机器人设计了趋向目标行为模块、避障行为模块和沿墙走行为模块,采用基于优先级行为决策控制策略,使移动机器人完成面向目标的任务。这种层级结构便于编程,具有易于添加反应速度快以及不依赖于模型描述等特点。4.实现了室内环境下基于视觉路标的全局定位。基于视觉信息的移动机人定位主要内容包括了图像预处理、门的定位、门牌定位,字符分割和字符识别五部分内容。基于5/3提升小波变换的软取阈值方法对图像去噪,然后基于边缘提取算法提取机器人工作环境信息,图像处理实验表明,算法能够较好地适应非结构化环境固有的特点,可以抑制细小的边缘而得到路径主要轮廓。提升小波算法降低了计算的复杂程度,因此有效地减少程序运行时间,提高定位系统的实时性,实用性。提出一种把提升小波变换对图像信号良好的分解性能与Hough变换直线的检测特性结合起来提取环境信息的方法来实现门的定位的方法。该方法以提升小波变换对环境信息图像进行软取阈值去噪处理,而后获取水平、垂直等高频部分的小波低频分解。实验表明基于提升算法和Hough变换设计的环境信息处理系统有很好的性能。基于神经网络分类器进行门牌号码的识别,从而实现单目视觉系统下移动机器人基于视觉路标的室内定位。5.用全局规划和局部规划相结合的方法解决全局路径规划问题。系统的全局路径规划分为两种:无障碍环境路径规划、有障碍环境规划。对无障碍环境路径规划比较简单,通过人机交互给出移动机器人运行参数,生成机器人的运行路径。对有障碍环境,采用超声探测信息和主动视觉信息实现了运动障碍物和静止障碍物类型的识别,并针对障碍物类型采用不同的避障策略,以完成全局导航任务。提出把自回归模型来预测动态障碍物的运动轨迹和动态障碍避障策略结合的方法,解决了动态、动态障碍物环境下的路径规划。采用自回归模型来预测动态障碍物的运动轨迹,然后对该预测轨迹的每一点进行在线实时避碰规划。针对自回归模型的参数是以小车参数为基准,在实际应用中必然带来一定的误差的局限,补充了动态障碍避障策略。6.解决了信息不完全、环境不确定情况下的全局三维路径规划问题。提出在环境的二维坐标不变的情况下,对每个障碍物固定坐标系来描述位置和姿态的方式实现三维动态环境、动态障碍物的全局环境描述,并设计神经网络三维路径规划器实现三维动态路径规划。基于这种方法,三维动态环境路径规划算法没有过多的复杂性,实时性很强。该算法在动态避障策略、行为控制策略、定位等多个方面制定以后,根据路径最短约束条件完成路径搜索,生成从起点到终点的无碰撞路径。对算法的参数优化问题进行了研究,为算法更好的应用提供了有意义的指导。算法所固有的并列性可用并列硬件来实现,运算量小,仿真证明算法能够在实时性要求较高、有较多障碍物、有较多路径点的情况下实现快速的路径规划。7.采用基于粒子群优化的模糊控制算法设计路径跟踪控制器。以粒子群算法为优化工具,自动提取控制器参数的模糊规则基,提高了优化模糊控制的控制性能;对粒子群算法进行了改进,在保证稳定性的同时,提高了控制器的收敛速度;仿真和实验表明,由于规则提取是自动获得,减少了人为工作量,控制器具有很好的鲁棒性。更多还原

【Abstract】 As the robot application range extends constantly, its working condition is getting more and more complex, which is always unknown, dynamic and unstructured. So, it is a new challenge for robot to fulfill a mission in real time under these environments. The idea of mobile robot behavior control based on multi-sensor is put forward.according to a survey of the research on mobile robot control architectures and navigation technology, A mobile robot navigation control experiment system is built to study on environment perception, behavior decision, motion control and so on, in the objective-oriented mission mobile robot behavior control. The system consists of a mobile robot and a set of communication system. The main contents are as follows:A modularized distributed control system architecture based on multi-processor with parallel communication is put forward and employed in the hardware system design. Each module is controlled by an individual processor and the data processing is tackled by the same one processor.All modules run and communicate with each other in a parallel method, which improves the speed of data transmission.Deliberated/reactive hierarchical hybrid architecture is used in the software design, which consists of a human-machine interface layer,a mission planning layer and a behavior control layer.A distributed ultrasonic detecting system based on multi-processor is developed.It is composed of an upper working method control module and a lower intelligent ultrasonic ranging module array.Each intelligent ultrasonic ranging sensor is controlled by an individual micro-processor which achieves ranging data processing, and the intelligent ultrasonic ranging modules can be grouped to work in parallel mode, which enhances the real-time performance, which can restrain the ultrasonic crosstalk and random interference and improve the accuracy of the detecting system.The objective-oriented mission can also be accomplished utilizing visual perception. has two behavior modules:goal approaching behavior, control behavior and escape behavior,each module has a given priority.The behavior arbiter decides which behavior module can control the robot by their given priority when the behaviors are valid.Employing the above segmentation technique of visual localization in indoor environment based visual landmarks is conducted on image, Here visual landmarks are those doors of corridor is characterized by topological graph.and doorpalates in corridor.Since lifting wavelet transform is suited for image decomposition and Radon transform is fit for line detection, their excellent performances are put together to figure out line detection pattern. This method decrease complexity of count greatly, and reduce run time effectively.The correlative experimental results show that the method can obtain the accurate measurement.We proposed a path planning algorithm based on neural network for cleaning robots.In this algorithm, a discrete method is used to approximate the minus gradient direction of the energy. We combined the penalty function with the distance energy function to construct the cost function.The path planning is implemented by determinating the motion tendency of the points set. The parameter optimization of the algorithm is investigated through computersimulation.These provide valuable guidance for the application of the algorithm in practice.The study on mobile robot dynamic obstacles avoiding strategy in global navigation is carried out.Something that appeared on the path planned by is regarded as dynamic obstacles which are classified into motorial obstacles and static obstacles. The robot can distinguish the above dynamic obstacles by using the information of ultrasonic detecting system and active vision system.The robot can avoid the dynamic obstacles appeared in global navigation using different obstacle avoiding strategy. The motion trajectories of moving obstacles are predicted by using an auto regressive (AR) model, thus, dynamic path planning is translated into instantaneously static one. This method enables the robot give a quick response when encountering obstacles.A novel fuzzy controller with fuzzy non-linear Parameter is proposed. In order to solve the problems of time consuming and hard to satisfy the control demands when seeking the parameters.A non-liner fuzzy method is proposed to generate the parameters. Partiele swarm optimization (PSO) algorithm is used to generate the non-linear fuzzy rule base automatically without prior knowledge. The modularized distributed control system architecture based on multi-processor with parallel communication and manufacture methods show the characteristics of practicability and expansibility in intelligent robot research.更多还原