傅立成Fu, Li-Chen臺灣大學:電機工程學研究所郭維楨Kuo, Wei-JenWei-JenKuo2010-07-012018-07-062010-07-012018-07-062009U0001-1008200911042400http://ntur.lib.ntu.edu.tw//handle/246246/188160這篇論文中，我們提出一種將環境表示為柵格狀地圖以及特徵點地圖混體的建圖結構，即FG (Feature-Grid)建圖。在室內環境中，以線段特徵點最為常見，他們提供了相當豐富的幾何資訊，因而可用於提高定位的精準度。線段的關係以互相傳值或平行最為常見，因此線段的正交特性是保證建圖結果一致性的要素之ㄧ。除此之外，角特徵點也提供了重要的定位資訊。這些特徵點使得粒子濾波器能夠更準確的推算機器人的位置。因此，這篇論文所提出的方法中，無須對環路型地圖執行額外的處理程序，便能確保環型路徑重疊時正確無誤。實驗結果顯示此方法在大型的室內環境也能順利運作，並成功地建立“環路”以及“無特徵點長廊”部分的地圖。做為實驗平台的機器人配備了最遠測量距離為20公尺的SICK LMS-100雷射測距儀。In this thesis, we present a novel data structure representing the environment with occupancy grid cells while each grid map is associated with a set of features extracted from laser scan points, call FG (Feature-Grid) mapping. Due to the fact that line segments are principal elements of a great variety of indoor environments, they provide considerable geometric information about the environment and hence which can be used for enhancing the localization accuracy. Orthogonal characteristic of line features is the key to guarantee consistency of the resulting SLAM algorithm since the lines we are dealing with are either parallel or perpendicular to one another. Besides this, corners are also features that provide crucial location information. These special surrounding features allow the later used particle filter to sample robot poses more correctly. As a result, in this work the large loop map building can be closed without actually incorporating any loop closing mechanism. Experimental results are carried out successfully in relatively large challenging indoor environments, which contain both loops and long featureless corridors, with robots equipped with SICK LMS-100 laser scanner whose maximum range is 20 meters.Chapter 1 Introduction 1hapter 2 Preliminaries 8.1 Bayesian Filter 8.2 Particle Filter 8.2.1 Non-parametric Representation 9.2.2 Particle Filter Algorithm 9.3 RBPF (Rao-Blackwellized Particle Filter) Overview 10.4 DP-SLAM 11.4.1 Maintaining the Particle Ancestry Tree 11.4.2 Map Representation 11hapter 3 Hybrid Approach : FG-SLAM 13.1 Validity Analysis 13.2 Feature Enhanced RBPF Mapping 16.2.1 Sampling 16.2.2 Weighting 17.2.3 Resampling 18.2.4 Map estimation 18.3 Feature Extraction 20.3.1 Curvature Function 21.3.2 Line Extraction 25.3.3 Corner Extraction 27.4 Feature Association 28.4.1 Feature Association 29.4.2 Particle Weighting 33.5 Complexity 35hapter 4 Experimental Results 38.1 System Configuration 38.2 Particle Distribution Analysis 39.3 Loop Closing 40.4 Pose Variance Analysis 42.5 Handling Odometry Drift 43.6 Computation Time 44hapter 5 Conclusion 46EFERENCE 47UBLICATION LIST 51852588 bytesapplication/pdfen-US同步定位與建圖Rao-Blackwellized粒子濾波器環路閉合柵格狀建圖柵格特徵點混和式建圖Simultaneous localization & mapping (SLAM)Rao-Blackwellized particle filters (RBPFs)loop closuregrid mapsfeature-grid mapsthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/188160/1/ntu-98-R96922113-1.pdf