Research On Simultaneous Localization And Mapping Of Mobile Robot

To autonomously explore in unknown environments, mobile robot needs abilities of sensing and interpreting its surroundings, and localizing itself in the environment. This is so called simultaneous localization and mapping(SLAM),which is the hot spot of research in the mobile robot community and the key prerequisite to make a truly autonomous mobile robot. In this paper, some SLAM methods of mobile robot under static unknown environment are studied.Firstly, localizing and mapping technologies of mobile robot are surveyed. The evolutions of SLAM are described and the major issues in SLAM are pointed.Secondly, a systematic framework of SLAM is established. Each part of the framework is analyzed in detail and designed. It provides the guidances for the follow-up theoretical studies and experiment designs.Thirdly, in the established framework the SLAM based on extended Kalman filter(EKF-SLAM) is studied. The implementation procedure of the EKF-SLAM algorithm is presented. The EKF-SLAM simulation experiments based on point features are conducted. The simulation results prove the integrity and realizability of the framework. In order to solve the problem of the Jacobian matrices being calculated during each sampling period and the truncation errors being introduced when linearizing the state model and the observation model in the EKF-SLAM algorithm, the SLAM algorithm based on unscented Kalman filter(UKF-SLAM) is studied. The comparative experiments are carried out to prove the validity of the UKF-SLAM algorithm.Finally, to deal with the weakness of the inaccurate mathematical model, which is led by ignoring ground condition, sensor precision and others uncertainty factors during kinematics model deduction of mobile robot, an approach of model errors compensation based on neural network with EKF is proposed. In the proposed algorithm, the errors are replaced with neural network function and the network's weight coefficients are augmented to the system state. The augmented state is predicted and updated by the EKF algorithm continuously. Then the network train on line and system state prediction can be realized simultaneously. The dynamic compensation ability and robustness of the proposed algorithm are verified by the SLAM experiments based on the model error compensation with EKF neural network.