This thesis examines the simultaneous localization and mapping (SLAM) problem for mobile robot navigation. To obviate the dependency on successful feature extraction, we developed an efficient and flexible genetic algorithmic map representation for view-based SLAM approaches. It does not rely on features and it is especially appropriate in a 3D environment. With this map representation, an efficient view-based SLAM approach: the Rao-Blackwellized Genetic Algorithmic Filter (RBGAF) SLAM is developed. Such a SLAM approach does not rely on features and it is capable of integrating arbitrary sensor and motion models. Further more, the approach can be implemented on a graphical processing unit with the development of a highly efficient parallel computing structure for RBGAF-SLAM. This significantly improves the processing speed so that real-time processing can be achieved. A set of simulation and experiments are presented to demonstrate its effectiveness and efficiency. The results verify that our approach achieved 3D real time SLAM in urban environment.