Simultaneous Localization And Map-building For Mobile Robot

Autonomous system is required increasingly in the fields of industry, agriculture, scientific research and national defence along with the development of information technology. More and more research is focused on robot and robotization Robot Company and research institute are developing various service robots to work instead of human and developing autonomous land vehicle for the sake of safe drive or military affairs. According to working environment, it can be classified into indoor mobile robot and outdoor autonomous land vehicle (ALV). Autonomous navigation is necessary, and localization is a fundamental problem for autonomous navigation, and modern localization method is based on environment map with interoceptive and exteroceptive sensor. This dissertation is focused on environment map building, mobile robot localization, simultaneous localization and map building. Exciting results are achieved.The first subject of this dissertation is focused on environment map building for indoor environment and outdoor environment. To build indoor environment map, adaptive segmentation and weighted stochastic mean are used because the sample points are not distributing uniformly. Line segments are extracted from these sampling points and are used to approximate the shape of real environment objects. The environment map is called ordinal map if ordinal relationship between line segments and characteristic points is retained while merging local maps into global map. To build outdoor environment map, an approach based stochastic density is proposed for obstacle detection on the road and crosscountry. It is also used to extract environment land feature and build local map that consist of trees beside road. Two type approaches based respectively on assumption of independence and assumption of correlation are used to merge local maps into global map. The 2D road environment map is represented with the coordinates of trees and the 3D off-road environment map is represented with grids.The second subject of this dissertation is focused on mobile robot localization. A matching method based on ordinal map and relative relationship is proposed to locate mobile robot under global uncertainty. Global map and current local map are represented with directional line segments and characteristic points. Ordinal map improves the matching efficiency and relative position relationship avoids frequent coordinates transformation. Another recursive estimation method based on linear system models is proposed to track mobile robot position. The system state is redesigned and the global coordinates is considered as observation vector. It is free of linearization error due to linear approximate of nonlinear function. Finally, several localization methods, Linear Kalman filter, Extended Kalman Filter, Unscented Kalman Filter, Markov Localization, Monte Carlo Localization, Map Matching, are compared experimentally.The third subject of this dissertation is focused on SLAM (simultaneous localization and map-building). It is as the problem of "chicken and egg". In indoor environment, an optimization SLAM solution is proposed by reconstructing system state vector and reselecting observation vector. The blocks on diagonal of covariance matrix are two same symmetrical submatrix and the blocks under diagonal of covariance matrix are two opposite anti-symmetrical submatrix. The computation requirement is reduced without any approximation during covariance matrix updating. The optimization solution is consistent, convergent and computationally efficient for medium size environment. It fulfills with 3C requirement. Another approximation solution is proposed to reduce computation requirement much more by approximating blocks under diagonal as zero. Theresults of optimization and approximation solution are consistent to the result of full co variance solution. In outdoor environment, the augmented state prediction is modified to realize simultaneous localization and map building.