Study On Simultaneous Localization And Mapping Of Mobile Robot In Large-Scale Environments

With the development of mobile robot in research and application, navigation in non-structrual environments, especially in unknown environments, has attracted immense attention in mobile robotics field. Autonomous exploration has been recoglized as the main method and the key technology to finish the real autonomous tasks for mobile robot in unknown environments.The reliable localization is the basis of mobile robot navigation. When a robot moves in a real environment, the navigational tasks such as path planning and trajectory tracking need continuous position and orientitaion information of the robot. However, when the robot enters an unknown environment without maps, how to build an accurate map of the environment via sensor data while simultaneously using this map to determine its accurate position and orientation is a key and technically challenging problem, which is called simultaneous localization and mapping (SLAM). SLAM has attracted significant attention since it was introduced and a wide range of techniques have been reported for sovling this problem. However, most of these existing methods still have many unsolved problems when they are used in unknown large-scale environments. This dissertation focuses on several SLAM problems in unknown large-sclae environments. The main contribution of this dissertation can be concluded as follows.The data association in unknown environments has been solved. This disseration discusses the data association issues of the FastSLAM (Factored Solution to SLAM) algorithm in unknown large-scale environments. In normal FastSLAM, it always assumes that the data association is certain. But in real world, the data association is always uncertain. This disseration adopts a data association solution for FastSLAM algorithm in unknown environment (uncertain data association) by using the united technology of per-particle maximum likelihood and negative information approach. The advantage of the per-particle maximum likelihood approach is that the motion noise has no influence on the accuracy of the latter data association. In addition, the aim-lost phenomnen does not appear in the per-particle maximum likelihood approach either. The negative information makes sure that the errorous landmarks added in the map can be removed in time. The new method solves the data association of unknown large-sclae environments and improves the performance of SLAM greatly. The adaptive grid map has been constructed. This disseration addresses the problems of data storage and compuation of data association for large maps. Probabilistic occupancy grids approach is the most common and well-proven spatial representation used in SLAM mapping, but it requires huge memory and complex computation of data association in unknown large-scale environment. This disseration proposes an adaptive grids approach for mapping. This method unilizes the theory of quadtree, a well-known data structure able to achieve compact representations of large two-dimensional binary arrays. The adaptive grids approach automatically adjusts its grid size by the obstacle density of different zones in maps. On the one hand, this method improves precision of maps. On the other hand, it also reduces the computation of data association because of the reduced number of grids. At the same time, this disseration gives a brief and intuitionistic solution to compute the occupancy probability of adaptive grids based on the Bayesian theory.The precision of localization and mapping has been improved. In order to improve the poor precision and real-time performance of localization and mapping for the exixting SLAM algorithms, this dissertation introduces an improved PF-SLAM approach. The improved method first lets the robot run about ten steps and finish localizition based on the existing maps; and then uses laser rangefinder to observe environments and updates maps. So the the precision of localization and mapping can be improved. In this method, particles are distributed only around the robot, so it promotes the real-time performance.