【作者】 余洪山；

【导师】 王耀南；

【作者基本信息】 湖南大学 ， 控制理论与控制工程， 2007， 博士

【摘要】 目前移动机器人已经在工农业、交通运输、军事、服务和医疗卫生等行业取得了广泛应用,以扩展人类的工作能力,并减轻或取代人类繁重劳动。随着应用环境的逐步复杂化和机器人技术的进步,人类对移动机器人的自主性和智能性提出了更高的要求,并引起了国内外学者的极大关注。特别是在大规模未知极限或复杂环境下,人类期望机器人能自主完成环境的探索、创建所在周围环境的地图表示,并能利用地图表示实现在环境中的高效导航。本论文以Pioneer 2-DXE移动机器人为对象,重点开展室内环境下移动机器人地图创建和自主探索方法的研究,全文主要工作包括如下几个方面。论文首先系统深入地介绍了移动机器人的定义和发展历史,系列智能移动机器人系统和典型通用移动机器人研究开发平台,其次详细讨论了机器人常用传感器的原理和特性,然后阐述了移动机器人关键技术和研究热点。并在此基础上阐述本论文的研究意义。在第二部分,论文详细阐述了Pioneer 2-DXE移动机器人的硬件系统和各种传感器配置和设计;然后系统介绍了移动机器人控制结构和Pioneer移动机器人的软件开发系统;随后深入讨论了移动机器人的运动控制模型及其不确定性。栅格地图是移动机器人环境表示的一种重要形式,而声纳传感器是栅格地图创建最广泛采用的传感器。然而由于声纳传感器存在多重反射、镜面反射、角精度差等缺点,使栅格地图存在环境解释的准确性和精度方面存在一定限制。为此,论文提出了一种改进神经网络声纳解释模型用于栅格地图创建,并提出栅格单元的连续更新方案。该模型同时考虑声纳传感器空间相关度和时间相关度的影响,保证测量数据在时间和空间上的连续性。因此即使出现某个声纳传感器或者某次测量数据受镜面反射或者多重反射影响而无效,模型也可以根据历史测量数据和空间相邻传感器测量数据,获取关于空间环境状态的正确解释,从而大大减少声纳传感器的不确定性对地图创建的影响。论文基于Levenberg-Marquardt算法,采用分阶批处理模式对神经网络进行离线训练。基于训练后的神经网络的对声纳测量数据的解释,每个栅格单元对应于空、障碍物和不确定三种状态,本论文中对三种状态分别采用Bayesian模型进行更新,而最终栅格单元状态基于Max-Min法则从三种可能状态中选择决定,从而进一步保证了声纳信息融合的准确性。实验结果表明,基于该方法创建的栅格地图具有较好的连续性、准确度,对不同环境具有良好的适应性,利用训练后的神经网络模型,具有较好的实时性,可满足在线处理的要求。大规模未知环境下的自主探索和地图表示是移动机器人的重要研究方向。论文在深入研究当前自主探索方案和地图表示模型的基础上,提出了用于大规模环境表示的混合高精度地图表示模型。在该地图模型中全局地图采用拓扑地图,其利用改进四叉树表示方法对局部栅格地图进行处理,提取拓扑节点集合,并通过临近空间范围内节点的匹配处理实现全局地图的更新和扩张;而每个拓扑节点采用改进四叉树结构保存对应的栅格地图信息。方法保持了环境表示的精度,有效控制了大规模环境下栅格地图的计算复杂度和空间复杂度,同时拓扑信息和几何信息的存在有利于不同层次下的探索、路径规划等任务处理需求。基于上述层次式混合地图表示模型,论文提出一种用于大规模未知环境的自主探索方案。该探索方案分为局部探索和全局探索两部分:在局部探索阶段,根据局部栅格地图,基于边界长度和运动路径长度定义的局部探索性能函数选择下一步最优探测边界,其中探索边界线段采用模糊FCM聚类算法提取,运动路径基于改进Distance Transform算法计算。当满足局部探索和全局探索切换条件时,采用改进四叉树结构对局部探索过程中获取的局部栅格地图进行计算处理,提取拓扑节点,并实现对全局拓扑地图更新。并从包含探测边界的拓扑节点中基于全局最优探测性能函数选择探测拓扑节点。其中全局最优性能函数综合考虑性能函数运动距离、预期获取的未知环境信息量、直线运动、周围环境状态配置等因素。局部探索和全局探索两步骤循环进行,则最终完成整个未知大规模环境的探索和地图创建。该混合探索方案兼顾了计算速度和最优探索性能,并减少累积误差影响,实验结果验证了该探索方案的可行性和正确性。为实现已知路径的快速跟踪控制或未知环境下对目标位置的无碰撞高效导航控制,本论文提出了一种基于模糊推理和行为控制的移动机器人导航避障方法。该方法首先定义了机器人导航过程中可能行为模式,然后设计了模糊导航控制器,根据实时声纳测量数据和机器人位姿、目标位置等信息实现机器人行为的切换和调度控制。该方法在两种不同环境下五个导航任务的试验结果表明,方法具有很好的实时性和较高控制跟踪精度,并对不同环境有很好的适应能力和可靠性。移动机器人同步地图创建与定位(SLAM)是移动机器人的核心研究课题。作为论文进一步的研究内容,本文对SLAM的最新研究进展和关键技术进行了综述;并从地图创建模型、计算复杂度和算法鲁棒性等方面对现有方法进行了对比分析。最后总结分析了SLAM研究存在的难题,探讨了今后的发展方向。更多还原

【Abstract】 Mobile robots have been widely used on industry, agriculture, military, hospital, healthcare and et al to expand human’s ability and release them from hard and dirty work. As the complexity of the environment robot applied increasing and robotics technology improving, human expect more intelligent and autonomous mobile robots. Especially in unknown large-scale extreme or hard environment, autonomous robots must posses the ability to explore their environments, build representations of those environments, and then use those environments to navigate effectively in those environments. Consequently in the last two decades, the problems associated with autonomous exploration of mobile robots operating in unknown environments have attracted the attention of many researchers. In this thesis, the challenging problems of map building, exploration of mobile robots are addressed and some solutions are provided. The findings are verified through simulation and real world experimental trails on Pioneer 2-DXE mobile robot. The main studies undertaken in this thesis are listed as following.Firstly, the definition and development of mobile robot is surveyed, and some typical series intelligent mobile robot and general mobile robot platforms for research are presented; then, the character and principle of some popular sensors for mobile robot are discussed; subsequently, key technologies and hot topics in robotics are addressed. Based on those research backgrounds, the necessity and feasibility of research work about this thesis are introduced.In the second chapter, the configuration and design of Pioneer 2-DXE’s hardware and sensors equipped are described in detail; then the control scheme and software development tools for Pioneer robots are presented; finally the motion control math model and accumulative position error model of Pioneer robot are comprehensively addressed.Due to its efficiency, one of the most popular and successful map representations is occupancy grid. Occupancy grids could be built based on laser range-finders, stereo vision, and sonar sensors. Sonar sensors are commonly used due to operation simplicity, robustness, and low price. However sonar readings are prone to several measuring errors due to various phenomena (e.g., multiple reflections, wide radiation cone, and low angular resolution). This paper presents an improved neural network model for sonar readings interpretation to build occupancy grids of mobile robot. The proposed model interprets sensor readings in the context of their space neighbors and relevant successive history readings simultaneously. Therefore, if current readings are produced by multiple reflections or specular reflection, this neural network model could depend on relevant history readings to obtain the correct occupancy values. Consequently the presented method can greatly weaken the effects by multiple reflections or specular reflection. The output of the above neural network is the vector about probabilities of three possible statuses (empty, occupancy, uncertainty) for the cell. As for sensor readings integration, three probabilities of cell’s status are updated by the Bayesian update formula respectively, and the final status of cell is defined by Max-Min principle. Therefore, it integrates the sonar readings more accurately while keeping appropriate computation cost. Experiments results performed in lab environment has shown occupancy map built by proposed approach is more consistent, accurate and robust than traditional method while it still could be conducted in real time.One of the main concerns of modern robotics is efficient exploration and map representation of large unknown environment. Based on the deeply analysis of currently autonomous exploration and corresponding map model, this thesis introduces an efficient hybrid map representation that integrates the metric and topological paradigms. In this model, global environment is representing in topological model. Each topological node is first derived by improve quartree algorithm from local grids in proper interval, and then fusion of homogeneous neighbor nodes is applied to achieve consist and compact global topological node. The metric gird information of node is stored in corresponding parent node according to quartree structure. As the mobile robot explores more and more unknown area, new local grids is continuously created and the global topological map is updated and expanded until the exploration is finished. This hybrid hierarchical map model produces accurate environment representations while avoiding suffer from huge data volume and computing burden. In addition, the topological and metric information preserving simultaneously could meet different level requirement about exploration, navigation and path-planning and so on.On the basis of above proposed hybrid hierarchical map model, this thesis introduces an autonomous exploration algorithm for large scale unknown environments. Exploration planning is performed at two levels: global planning is performed at topological level and local planning is performed at metric level. As for local planning, exploration utility function is defined by the length of exploration frontier border line and travel distance from current robot position to frontier, where exploration frontier border line is calculated by improved fuzzy c-means clustering algorithm and travel distance is derived by modified Distance Transform method. When the switch conditions from local planning to global planning is satisfied, global topological map is updated according to the proposed hybrid hierarchical map model. As for global planning, the next best exploration node is selected from frontier node set according to the global exploration utility function. Once the nest exploration node is chosen, the new local planning is created on the basis of the metric information corresponding to the nest exploration node. The global exploration utility function is chosen with a multiplicative form, which prefers exploring more new unexplored areas, shorter travel distance, straighter path, and smaller proportion of neighbor obstacle areas and so on. Local planning and global planning is alternately conducted until the whole environment is known. This exploration algorithm can be performed in a fast and efficient way as the local exploration can be conducted in real time and global planning assures the robot escaping from local trap. The method has been successfully tested for Pioneer 2-DXE mobile robot in simulated environments.In order to fast and accurate path tracking or navigate mobile robot to target position without collision, this thesis proposed a fast navigation and obstacle avoiding algorithm based on fuzzy reasoning and behavior control. At first, possible behaviors encountered during robot navigation are defined. Consequently, the fuzzy navigation controller is designed. Based on this fuzzy controller, the mobile robot’s behaviors are switched and scheduled according to sonar readings collected, target position, robot position and et al. This algorithm has been tested on five navigation tasks under two different environments. Experiments results have been shown that the introduced method has good tracking accuracy and navigation performance, and is reliable to different environments.Simultaneous localization and mapping (SLAM) algorithm for mobile robots is a key problem in the field of robotics. This paper surveyed the latest progress of SLAM algorithms and described the key techniques adopted by popular different SLAM modes. In addition, those methods were analyzed and compared in detail according to map-building model, computation complexity, and robustness and so on. Finally, the key problems and future research trend of SLAM approaches were presented.更多还原