Research On Distributed SLAM Based On Analysis And Optimization Of Performance

Simultaneous localization and mapping(SLAM) technology is the key link for autonomous mobile robot navigation, the traditional SLAM method has inherent flaws when dealing with dynamic change of observation information and large amount of calculation. Concerning this system structure flaw, distributed SLAM method establishes the corresponding distributed subsystem based on observation information to keep the system in dynamic changes according to the change of the observation information while reduce the amount of calculation, which ensures the stability of the system. The contents of this paper including:Firstly, based on the basic principle of SLAM, the mathematical and systematic models of distributed SLAM method are established according to the actual dynamic changes of the observation information, the distributed calculation method is explained in accordance with modified model, also the distributed calculation is combined with major filtering algorithm, in order to summarize the process of the suitable distributed method frame. Eventually, the advantages and disadvantages of distributed SLAM method and traditional SLAM method are analyzed through their comparison.Secondly, based on the dynamic reconfiguration problem in the realization of the SLAM system, a distributed SLAM method is proposed based on spatial domain theory, this method combines the distributed SLAM structure and characteristics of particle filter, and the following improvements are proposed: The method divides the landmarks of the whole space according to the angle between links of two landmarks to the robot to ensure that each subspace contains two non-collinear landmarks, according to the observation model established based on the landmarks combinations, also the dynamic adjustment process of particle distribution in the particle filter is improved to enhance the accuracy and stability of the system in the dynamic reconfiguration process. Eventually, the reliability of the method is proved through the simulation experiment based on the actual data.Thirdly, considering the real-time requirement as well as calculation amount caused by the distributed particle filter, the distributed particle filter is replaced by the distributed extended Kalman filter(D-EKF), a distributed SLAM method based on the heading-aid(DH-EKF) is proposed. Through the analysis on the convergence and consistency of the D-EKF-SLAM, the system model is updated through the filtering method considering the heading-aid and the form to realize. Eventually, the accuracy and convergence of the algorithm is verified through the actual mobile platform.Finally, based on the above-mentioned distributed SLAM method, this paper focuses on the improving the precision of the distributed formula filter, then a method to improve the distribution formula filter linearization is proposed according to the analysis on the limitation of the distribution formula filter estimation precision. At the same time, considering the consistency of system and the improvement on distributed system integration, the covariance matrix module and matching information module, as well as the contribution of distributed observation information ensure the system can make the reasonable integration judgment with high precision or high consistency. Eventually, the modified method can improve the precision and stability of the distributed SLAM method on special environment changes.