Research On Distributed Fast Simultaneous Localization And Mapping Algorithm

Simultaneous localization and mapping(SLAM) technology is the fundamental guarantee for vehicle working steadily in an unknown environment and localization and mapping are the key prerequisite for the truly autonomous robot navigation. Fast SLAM is an efficient algorithm for SLAM problems. In order to overcome the limitations such as the low precision, the large amount of calculation, the severe sample degeneracy and difficulties in isolating potential faults of centralized Fast SLAM, the research of this thesis is a new distributed Fast SLAM method. The contents of this paper including:Firstly, based on autonomous navigation technology in an unknown environment, the basic principle of SLAM, EKF-SLAM and the centralized Fast SLAM is introduced in this paper. Fast SLAM 1.0 and Fast SLAM 2.0, as the widely used methods, the basic properties, theorems and the system model are reviewed.Secondly, based on the distributed particle filter and the federated Kalman filter, an improved distributed Fast SLAM model, based on the distributed structure, is provided. Furthermore, the consistency of the distributed Fast SLAM is analyzed by Monte Carlo tests and the feasibility of this algorithm is been proved by simulation result, and the limit of the uncertainty for the landmark estimation is discussed.Thirdly, according to the distributed Fast SLAM model, the robot state estimation flow and the landmarks states estimation flow are presented. In order to overcome the limitations of the low precision and the severe sample degeneracy of the distributed Fast SLAM algorithm, the distributed particle filter(DPF) is instead by the distributed unscented particle filter(DUPF) and both the innovation and the number of effective particle method are considered to redistribute the proportion of the sub filters in the new DUFast SLAM algorithm. Furthermore, the improved algorithm is proved by the simulation result and the mean square convergence of it is been discussed also.Finally, since the computational cost is increased, a distributed marginalized unscented Fast SLAM(DMUFast SLAM) algorithm is developed in this paper. In the proposed method, the other history information is neglected and the marginal distribution of the UPF is optimized to reduce the computational. The simulation results show that the improved DMUFast SLAM has a higher precision and smaller computational complexity.