Simultaneous Localization And Mapping For Mobile Robot Navigation With 3D Laser Point Clouds

Smart mobile robot is an intelligent device which integrates sensor technology,control technology,and intelligent algorithms together,and it has also attracted the earnest attention from many scholars and engineers nowadays.The earliest intelligent mobile robots can track back to few decades,some of existing mobile robots have been successfully applied to a diversity of harsh environments,such as underwater operation,mine exploration,military reconnaissance,and interplanetary exploration.However,most of these applications were oriented to industry and military only.In recent years,with the improvement of life quality and new technologies,AI technologies are very closely related to daily services.Service robots have become important tools which can help human interact with machines.To make robots walk like a human being,autonomous navigation is an essential technology.Simultaneous localization and mapping(SLAM)plays a crucial role in the field of robot navigation.The SLAM can provide robots the real-time position and build map incrementally.These are the bases of path planning and tracking for robots.Sensors are essential for robots to perceive surroundings and have a huge effect on the SLAM.The sensors used to perceive surroundings can be categorized into two classes,i.e.,laser sensors and visual sensors.Compared to visual sensors,laser sensors are not sensible to environmental noises and can provide 3D point clouds which contain the geometries of surroundings.Due to the important of the laser-based SLAM technology,the thesis devotes to developing a full six degrees of freedom SLAM algorithm for mobile robot navigation by 3D laser point clouds.The major contributions of this thesis are as follows:(1)Development of a density-adaptive laser feature and orientation(DALKO).The features of point cloud are helpful for robots to efficiently understand surroundings,and they can describe the special shapes in the ambient such as corner,line and plane.The features of point cloud can be applied to point cloud registration,pose estimation,and loop detection in SLAM.As the existing feature extraction algorithms are not robust to the situation where the point cloud density changes,a density-adaptive feature point detection algorithm is proposed,together with a geometric context to describe the geometric environment where feature points located.Different from the local shape context,the geometric context describes the geometric shape using the relative position of features.(2)Proposing an efficient and accurate point cloud registration algorithm.Point cloud registration aims at transforming the point clouds observed in different poses to the same frame and solving the constraints among the poses.Therefore,the change of poses can be estimated by a pair of point clouds collected at a particular time interval.The iterative closest point(ICP)algorithm and its variants have been widely used for point cloud registration.In this work,the ICP-based point cloud registration algorithms are investigated.By improving the rejection strategies of point pairs,a new efficient and robust point cloud registration algorithm,namely the vanilla Iterative Closest Reciprocal Point(vanilla-R ICP),is developed.Compared to other heuristic ICP algorithms,the vanilla-R ICP can offer more accurate solutions without scarifying computational efficiency.(3)Development of a graph-based six degrees of freedom simultaneous localization and mapping(6D-SLAM)algorithm.To ensure the consistency of the constructed map,the basic framework of graph optimization algorithm is utilized in this work.The SLAM problem can be then represented as a pose graph,where the nodes of pose graph denote poses whereas the edges of pose graph represent the constraint between poses.In this work,the vanilla-R ICP algorithm is used to construct the edges of pose graph.To evaluate the uncertainty of the pose constraints,two solutions to the vanilla-R ICP covariance are given.The distances between poses and feature matching are used to detect loops,and the Ceres optimization library is responsible for optimizing the pose graph of the SLAM.Finally,a graph-based 6D-SLAM algorithm is developed,and it is tested on both open source data and a mobile robot to demonstrate its effectiveness.