【作者】 曲丽萍；

【导师】 王宏健；

【作者基本信息】 哈尔滨工程大学 ， 模式识别与智能系统， 2013， 博士

【摘要】 移动机器人同步定位与地图构建（Simultaneous Localization and Mapping，以下简称SLAM），是通过机器人自身携带的传感器在线测量和位置估计，从而在部分已知或完全未知环境中实现机器人的自定位和增量式地图构建。这一无需先验地图的导航手段，对于移动机器人长时间无人现场的自主作业而言是至关重要的。本课题选自国家自然科学基金项目“小型自主水下航行器群体协同地形勘察关键技术研究”，以移动机器人为研究对象，深入开展了以下关键技术研究：首先，定义了移动机器人SLAM研究所需的坐标系，并在此基础上建立了移动机器人运动模型、传感器观测模型、环境特征模型及数据关联模型，从而为SLAM关键技术研究搭建了统一的平台。其次，针对环境特征地图的不完备性及自然实体路标的不规则性，提出了描述自然实体路标的位置属性和大小属性的圆型类特征表示法，即用圆型类特征的中心位置表示实体路标的中心位置，用圆型类特征的直径表示实体路标的空间俯视的大小属性；提出了基于角度-距离复合聚类的环境特征提取算法，该算法包括数据预处理、区域分割和特征参数拟合三部分，并通过“Victoria Park”标准数据集验证了算法的可行性；设计了EKF-SLAM仿真算法，并通过人工设定路标和机器人路径的仿真实验，验证了算法的有效性。再次，提出了自适应重采样的FastSLAM算法，通过实时计算有效粒子个数和评判粒子退化程度，实施有效的重采样操作，从而有效地改善频繁重采样所导致的样本枯竭影响；根据粒子滤波和粒子群的相似性，提出采用粒子群优化算法改进FastSLAM，并利用多样性启发因子引导粒子群优化搜索过程，以保证粒子集多样性最优。该算法经自行设计的仿真实验，验证了算法的可行性和有效性。最后，围绕SLAM的数据关联问题，给出了SLAM数据关联的解释树模型和关联矩阵模型；针对单一兼容最近邻数据关联算法中固定不变的最近邻判定阈值与实际不符且极易造成关联错误的情况，提出了一种基于分段自适应阈值动态调整算法的单一兼容最近邻数据关联算法，并通过仿真实验验证了算法的有效性；通过联合兼容分枝定界数据关联算法分析，提出一种按照联合最大似然准则判定关联有效性和采用蚁群优化算法代替分枝定界搜索的联合兼容分枝定界数据关联改进方法，设计了基于蚁群优化的联合数据关联算法，并通过仿真实验完成算法的可行性和有效性验证。更多还原

【Abstract】 Simultaneous Localization and Mapping (SLAM) is that a mobile robot makes selflocalization and incremental mapping in the known or completely unknown environment bymeasuring on-line and position estimating using the sensors installed on the robot. The navigationmethod, which needs no prior map, is very important for robot’s long-time autonomous operationin unmanned site.This paper comes from National Natural Science Foundation Project-“Research on the keytechnologies of swarm cooperation topography survey for micro autonomous underwater vehicle”.And the paper selected mobile robot as an studied object and finished the following keytechnology research.First, the coordinate systems for mobile robot SLAM reseach are defined. And on this basisthe models such as kinematic model, sensor observation model, environment feature model anddata association model are built. All of these have made the unified platform for SLAM keytechnologies research.Secondly, For solving the incompleteness problem of evironment feature map and theirregularity problem of natural solid landmark, a circular-class feature representation method usedto describe the position property and size property of natural solid landmark is presented.According the method, the mass center position of the solid landmark is represented by the centerposition of circular-class feature, and the property of overlook-from-space size of solid landmarkis represented by the diameter of the circular-class feature. At the meantime an environmentfeature extraction algorithm, that is Angle-Distance Cluster, is presented. The algorithm includesdata pretreatment, area segmentation and feature parameter fitting. The above algorithmfeasibility is validated by the normal data set-“Victoria Park”. An EKF-SLAM simulationalgorithm is designed and its effectiveness is validated by the simulation test of setting landmarkand robot path.Thirdly, an improved FastSLAM algorithm based on adaptive resampling is presented. In thealgorithm, the number of effective particles is calculated and the degree of particle degeneration isevaluated and judged in real time. And then the effective resampling opteration is implemented sothat the influnce of sample impoverishment caused by frequent resample is improvedeffectively.According to the similarity between particle filter and particle swarm, anotherimproved FastSLAM algorithm based on particle swarm optimization is presented. In order tomake the diversity of particle set best, the particle swarm optimization searching is led by use of the diversity heuristic factor. The feasibility and effectiveness of the algorithm are validated byself-designed simulation test.Finally, for describing SLAM data association problem clearly, the interpretation-tree modeland association matix model are discussed. And then the problem is pointed that the fixedjudgement threshold of Individual Compatibility Nearest Neighbour is not correspondent withpractice. And the fact is very easy to lead the association error. Therefore, an adaptive algorithmbased on dynamic-and-sectional threshold is presented. The effectiveness of the algorithm isvalidated by simulation test. And then the data association algorithm of Joint CompatibilityBranch and Bound (JCBB) is analyzed. And an improved JCBB algorithm, that ant swarmoptimization is applied instead of Branch and Bound searching, is presented. In this newalgorithm, the association effectiveness is judged according to the criterion-Joint MaximumLikelihood (JML). And An joint data association algorithm based on ant swarm optimization isdesigned. The new algorithm feasibility and effectiveness are validated by simulation test.更多还原