Robust Data Association In Smoothing SLAM

Simultaneous Localization and Mapping(SLAM) is the core technology of the au-tonomous robot navigation and a key issue of accomplished complex tasks inde-pendently in unknown environment, also it reflects the intelligent level of robots. Data association is a key of SLAM, it determines the positioning accuracy and precision maps, so this paper discusses the robust data association.SLAM is generally based on least-square optimization, however this kind of methods requires that all data associations are one hundred percent correct and the front-end of system has the high performance to process data associations. Algorithm would be failed if the system has false data associations. Switchable Constraints (SC) and Max-Mixture Models (M-M) methods made a preliminary research on this problem. SLAM is robust to deal with loops using those two methods, howerver those two meth-ods don’t deal with the data association in feature-landmarks maps, and they fail to ob-tain environmental information.According to the characteristic of feature maps and factor graphs with landmarks, this paper proposed a robust feature-landmarks map SLAM algorithm for incorrect da-ta-associations (RFM-SLAM) based on SC. This algorithm adds the factor nodes into topological and uses the historical information to modified the false data associations, so that SLAM could get the correct path of robot, location of landmarks and the topo-logical graph.According to the probability theory and the topological graph, switch variable is added into the topological map, and some extra constraints are added into the optimiza-tion formula, so that algorithm could converge to the optimum solution. The calculation efficiency of algorithm is ensured by the analysis of the spare matrix and g2o. Switch variable could remove the false data association and change the fixed topology in the back-end. This algorithm increases the robustness of SLAM and strengthens the con-nection between the front-end and the back-end. It also ensures the accuracy of localiza-tion and the precision of robot’s map. Finally, RFM-SLAM is proved by the experi-ments using some synthetic and real datasets, and it compares with other methods.