【作者】 石杏喜；

【导师】 赵春霞；

【作者基本信息】 南京理工大学 ， 模式识别与智能系统， 2010， 博士

【摘要】 随着传感器、计算机和人工智能等技术的不断发展,具有思维、感知和动作能力的地面智能移动机器人在军事、民用和科学研究中得到了广泛的应用。其发展对国防、社会、经济和科学技术具有重大的影响力,已成为各国高科技领域的战略性研究目标。地面智能移动机器人能够成功完成任务的一个基本条件是能够在其所处环境中进行自主导航,而自主导航就必然要求它们能够进行自主定位。本文主要针对面向智能移动机器人的定位技术进行相关研究,使地面智能移动机器人在各种复杂环境下具有很好的自定位能力,具体的研究内容包括以下几个方面：地面智能移动机器人定位系统的主要功能是能够精确地确定其在地球表面的参考位置,而坐标系统是描述地面智能移动机器人运动,处理观测数据和表达其位置的数学和物理基础,论文首先讨论了地面智能移动机器人定位中常用的坐标系统,深入研究了WGS-84空间直角坐标、WGS-84大地坐标、高斯平面直角坐标以及机器人平面直角坐标之间的相互转换关系。介绍了GPS载波相位基本观测方程,在此基础上推算了载波相位双差GPS的坐标解算模型。为了保证差分GPS技术在大区域范围内的定位精度,研究了基于虚拟参考站(VRS)的差分GPS技术,详细推算了虚拟参考站上的双差观测值和单差观测值的生成算法。针对基于差分GPS/DR的组合定位问题,提出了一种尺度无色变换扩展卡尔曼滤波(SUT-EKF)算法,由于差分GPS/DR组合定位系统中的状态方程是非线性的,并且观测方程是线性的特点,将SUT预测移动机器人位姿,利用EKF融合最新观测值更新机器人位姿,该算法在状态预测阶段避免了计算Jacobian矩阵,从而有效地减小了线性化对非线性系统误差的影响。提出了一种基于尺度无色变换和迭代扩展卡尔曼滤波(SUT-IEKF)的同时定位与地图创建(SLAM)算法。由于数据关联对机器人的定位精度起着至关重要的作用,尤其是基于EKF的SLAM算法对错误的数据关联非常敏感,一种基于多算法匹配(MAM)的数据关联方法被提出,该算法利用粒子采样技术,将移动机器人位姿和特征地图位置联合概率分布以多个等权的粒子表示,各粒子进行独立的同时定位与地图创建,在相同观测数据的基础上,采用不同的数据关联方法,将会得到不同的关联集合,最后计算各关联集合的交集并作为这一观测的数据关联结果。研究了一种混合滤波的SLAM算法,并利用统计理论对其进行一致性评估,该算法框架利用粒子滤波技术将机器人SLAM中的联合后验概率分布因式分解为机器人路径部分及以机器人路径为条件的地图部分,使滤波器变成低维滤波,能够有效地提高计算效率,采用约束的无色卡尔曼滤波(CUKF)算法并融合新的观测数据使提议分布更加接近后验概率分布,并且能够精确估计移动机器人的位姿,进而通过扩展卡尔曼滤波(EKF)算法更新特征地图的位置。研究了分布式的多机器人协作定位方法,利用分布式的无色卡尔曼滤波(UKF)算法融合各机器人提供的相对观测信息,获取机器人的精确位置。详细讨论基于分布式无色粒子滤波(UPF)的多机器人协作同时定位与地图创建(C-SLAM)方法,在机器人群中,每个机器人运行一个粒子滤波器。当机器人群中某个机器人无法探测到特征地图,则可以通过航位推算估计自身的位置,但其精度非常有限,为了提高其定位精度,它可以与其它能够观测到特征地图的机器人进行相对观测,如果他们能进行连续的相对观测并进行连续的相对信息的交换,从而构建虚拟观测量(VO),通过虚拟观测量进行C-SLAM,有效提高定位精度。本文最后对全文进行了总结,并对未来进一步工作指出了探索的方向。更多还原

【Abstract】 With the development of sensor, computer and artificial intelligence, the ground intelligent mobile robot with idea, apperception and action capability has been used widely in the field of military affairs, civil and scientific research. Its development has imposing on the defense, society, and academy, and becomes the tactic research object of high technology of all countries. In order to finish tasks successfully, one of basic conditions for the ground intelligent mobile robot is the autonomous navigation in its environment and the autonomous navigation need to locate itself. This dissertation is focused on localization technology for the intelligent mobile robot, which makes the ground intelligent mobile robot have a better autonomous localization capability in all kinds of complex environment. The main content of this dissertation include the following aspects:The main function of the localization system for the ground intelligent mobile robot is to ascertain its referenced localization on the surface of earth. The coordinate system is the mathematics and physics foundation to describe the motion of ground intelligent mobile robot, deal with observation data and show its localization. The usual coordinate systems in ground intelligent mobile robot localization are discussed at first. The WGS-84 spacial orthogonal coordinate system, WGS-84 geodetic coordinate system, Gauss plane orthogonal coordinate system and robot plane orthogonal coordinate system can be translated with each other.The GPS carrier wave phasic observation function is introduced and based on which the double differential GPS coordinate computation model is deduced. In order to ensure differential GPS localizition precision in a large zone, the algorithm of differential GPS is researched based on virtual referene station (VRS). The single and double differential observation data on VRS is deduced detailedly.Aiming at the integrated localization issue based on differential GPS/DR, an algorithm based on scale unscented transformation and extended kalman filter (SUT-EKF) is presented. For the characteristic of nonlinear state equation and linear measurement equation in the integrated localization system based on differential GPS/DR, the robot location can be predicted by SUT and can be updated with new observations by EKF. The algorithm doesn’t compute the Jacobian matrix, it can decrease effectively the error of nonlinear system brought by the linearization.An algorithm for simultaneous localization and mapping (SLAM) based on scale unscented transformation and iterative extended kalman filter (SUT-IEKF) is presented. Data association plays an important role in the precision of robot localization, especially, the algorithm of SLAM based on EKF is very frail to the wrong data association. A data association method based on multi algorithm matching is proposed. It uses equal weight particles to denote the joint probability distribution of the robot and feature map. Each of particles applies different data association algorithm and gets different data association set during SLAM, the intersecting set of all sets is taken as the objective set.An algorithm for SLAM based on combined filter is researched and use the statistic theory to evaluate the consistency. It decompose the joint posterior probability distribution into robot path part and feature map part through the particle filter, which make the filter become low dimensional filter and can improve the computational efficiency. The constrained unscented kalman filter (CUKF) make the proposal distribution much closer to the posterior probability distribution with new observations and the robot pose can be estimated accurately. The extended kalman filter (EKF) is used to update the feature map localization.The method about distributed multi robot cooperative localization is discussed. The accurate robot localization can be acquired using distributed unscented kalman filter (UKF) fused with other robots’relative observation information. An algorithm for cooperative simultaneous localization and mapping (C-SLAM) based on distributed unscent particle filter (UPF) is described. Each of robots runs an UPF. When one member of the team may not observe the landmarks, it can estimate its pose through dead-reckoning, but the precision is too limited. In order to improve the precision, a novel approach is to let the robot observe other robot with better landmarks observations and get the relative observations. Let they keep continuous relative observations and continuously exchange relative information. The robot can construct Virtual Observations (VO) with the relative observations and perform C-SLAM based on VO, which can improve effectively the localization precision.Finally, we summarize the general work of this dissertation and give a short outlook on possible future research.更多还原