A Robust Back-end Based on Feature Maps in SLAM

Abstract

Simultaneous Localization and Map (SLAM) Building belongs to the category of the autonomous robot navigation. The solution of SLAM makes robots move by themselves. The current SLAM is based on least square optimization, while it requires all data associate correct. This paper proposed a robust algorithm based on feature maps. According to the relation of topological and mathematical model, a kind of switchable variable were added into the topological graph, and also some constraints were added into the back-end formulation. Our algorithm not only is able to recognize the false data association, but also could modify the data association in feature maps. The feature maps could be corrected by the robust back-end. The evaluation shows that the approach can deal with up to 500 false data association constraints on the datasets. This approach makes the back-end more powerful and the front-end would be easier than before.