3D Localization And Mapping Of Outdoor Mobile Robots Using A LIDAR

Autonomous navigation ability of mobile robot as the premise of behavior intelligent and automation of mobile, the primary problem is self localization and mapping of the robot. For outdoor robot applications, LIDAR (Light Detection and Ranging) is considered as a primary exteroceptive sensor for mapping because of LIDAR can provide high frequency range measurements with high accuracy. However, accurate mapping requires accurate knowledge of the LIDAR's poses during continuous laser ranging, in complex outdoor environment such as the urban environment, the common occurrence of GPS loss or even completely unavailable lead to the reliability of the mobile robot localization being challenged, a feasible solution can be using the existing laser LIDAR sensor as a assistance localization method to get accurate pose estimation, and using the Simultaneous Localization and Mapping (SLAM) method to solve the mapping problem simultaneously according to the coupling relationship of localization and mapping. In this paper, supported by the National Nature Science Foundation of China Project "field robots based on set theory modeling of complex terrain environment and at the same time positioning research" (61005092),3D localization and mapping of mobile robots using a LIDAR in complex, unstructured, and large scale outdoor terrain environment where GPS is denied are studied. The specific contents are as follows:To solve the problem of accurate localization of long distance without GPS data in large scale outdoor environment, a new localization method based on 3D laser scanning data called LIDAR Odometry is proposed. Firstly, real-time 3D pose estimation is performed with an inertial measurement unit (IMU) combined with encoders. Secondly, the Normal Distribution Transform (NDT) scan matching algorithm with LIDAR data is used to correct the estimated poses periodically. So that the problem of the accumulated error is too large in long distance localization through the conventional inertial navigation method can be solved, and accurate 3D pose estimation of the mobile robot can be obtained under the premise of ensuring the real-time performance of the localization algorithm.To solve the problem of large scale mapping with the uncertain pose information of the outdoor mobile robot, a simultaneous localization and mapping unified processing method based on the Graph-based SLAM frame is proposed with the consideration of the coupling between localization and mapping. Firstly, the output of LIDAR Odometry is used as edge constraints between nodes of pose observations to build the whole terrain environment Graph. Secondly, loop detection method is performed to detect a loop, and then according to the detected loop information, a probabilistic graph optimization method is used to implement the global poses'optimization. So that the total poses used for mapping reach the global optimum, and then the problem of poses and map's accuracy decreasing because of accumulated localization error during large scale mapping can be solved, thus the accuracy of localization and mapping can be further improved.To solve the problem of compressed representation, real-time construction and on-line updating of outdoor 3D terrain environment model which is used for autonomous navigation, a Set-valued Mutli-level Occupancy Voxel (SMOV) terrain modeling method is proposed. One of the differences between this proposed method and the existing methods is that this proposed method deals with the whole laser measuring path rather than the laser reflection information at the end of the path only, so that the three occupied states named occupied, free and unknown can be strictly distinguished, and represented by voxels with the three occupied states. Another difference between this proposed method and the existing methods is that this proposed method uses set-membership theory to deal with the uncertainty during the terrain construction process through on-line merging method, and then real-time modeling and updating of the multi-level terrain feature can be implemented, thus the robustness and richness of the information of the terrain model is enhanced.Finally, on the basis of the above studies, MT-FR mobile robot, which is built by our own laboratory independently with multiple sensors like 3D LIDAR, IMU and encoder, is used as the experimental platform to do a wide range of localization and terrain modeling experiments in outdoor campus environment, to prove the effectiveness and correctness of the proposed methods in this thesis.