面向建筑物内部环境的移动机器人同时定位与地图构建方法研究与应用

发布时间：2018-03-02 06:26

本文关键词： 同时定位与地图构建(SLAM) Rao-Blackwellized粒子滤波器特征点提取扫描匹配聚类粒子群优化重采样高斯分布　出处：《南京理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：移动机器人导航技术是机器人领域的一个热门研究话题,其主要目的是使机器人在未知环境下能够自主移动到目标位置并完成特定的任务。因此,环境地图的构建和机器人的实时定位是实现自主导航的基础,这就是机器人同时定位与地图构建(SLAM)技术。本文主要研究室内环境中基于激光雷达的移动机器人同时定位与地图构建问题,重点研究基于Rao-Blackwellized粒子滤波器的RBPF-SLAM方法。针对传统的RBPF-SLAM方法中存在的粒子退化及扫描匹配准确度不高等问题,提出了几种改进措施并进行了实验验证。首先我们设计了一个机器人平台系统,并在这个平台基础上构建了统一的机器人系统模型用于实验研究。针对常规全局扫描匹配准确度较低的问题,提出了一种基于特征点的扫描匹配方法。首先将所有雷达数据点划分为若干个特征路标段,然后在每一个特征段中基于密度和距离信息提取特征点。在扫描匹配中更加重视特征点的作用,对特征点赋予更高的匹配得分权重。最后将基于特征点的扫描匹配方法引入到SLAM算法中校正粒子位姿。实验表明基于特征点的扫描匹配能够更加准确地估计粒子位姿,使得生成的地图误差更小,提高了算法性能。针对传统RBPF-SLAM方法中存在的粒子退化和粒子耗尽问题,提出了一种基于区域粒子群优化和部分高斯重采样的SLAM优化方法。为了缓解粒子退化,同时减少粒子数量,引入区域粒子群优化方法来调整粒子的建议分布。首先把粒子集聚类成多个粒子簇区域,计算每个区域的加权中心位置,然后对区域内粒子进行粒子群优化操作驱使粒子向区域中心位置移动,保持粒子集的局部收敛性。为了缓解粒子耗尽,在重采样过程中,对粒子按照权值排序,只对权值过高或过低的粒子进行重采样,同时使用高斯分布采样得到的粒子,保持粒子的多样性。实验表明改进方法可以用更少粒子就能得到一张高精度的地图,减少了算法运行时间。
[Abstract]:Mobile robot navigation technology is a hot research topic in the field of robot. Its main purpose is to enable the robot to move to the target position and complete the specific task independently in the unknown environment. The construction of environmental map and the real-time localization of robot are the basis of autonomous navigation. This is the technology of simultaneous localization and map building of robot. In this paper, the problem of simultaneous location and map construction of mobile robot based on lidar in indoor environment is studied. Focusing on the RBPF-SLAM method based on Rao-Blackwellized particle filter, aiming at the problems of particle degradation and low accuracy of scanning matching in the traditional RBPF-SLAM method, Several improvement measures are proposed and verified by experiments. Firstly, a robot platform system is designed. On the basis of this platform, a unified robot system model is constructed for experimental research. A scanning matching method based on feature points is proposed. Firstly, all radar data points are divided into several feature path segments. Then feature points are extracted from each feature segment based on density and distance information. Finally, the feature point-based scanning matching method is introduced to the SLAM algorithm to correct the particle pose. The experimental results show that the feature point-based scanning matching can estimate the particle pose more accurately. It makes the map error smaller and improves the performance of the algorithm. Aiming at the problem of particle degradation and particle depletion in traditional RBPF-SLAM method, This paper presents a SLAM optimization method based on regional particle swarm optimization and partial Gao Si resampling. The regional particle swarm optimization (RPSO) method is introduced to adjust the proposed distribution of particles. Firstly, the particle clusters are classified into multiple cluster regions, and the weighted center positions of each region are calculated. Then the particle swarm optimization operation in the region drives the particles to move to the center of the region to keep the local convergence of the particle set. In order to alleviate the particle depletion, the particles are sorted according to the weight during the resampling process. Only the particles with too high or too low weights are resampled, and the particles sampled by Gao Si distribution are used to keep the diversity of the particles. Experiments show that the improved method can obtain a highly accurate map with fewer particles. Reduce the running time of the algorithm.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】