An appearance-based map partition framework for submap SLAM

Recently, submap methods have been proposed to reduce the computing cost of simultaneous localization and mapping (SLAM) algorithm. Among others, this method is of special interest for efficiently managing large-scale environments, and is more suitable for further navigation task. For achieving that, the whole map is partitioned to geometric submaps. In this paper we present a novel framework for automatically obtaining those local submaps from observations. The framework, which imitate human behavior, utilize appearance-based criterion to decide whether mobile robot get into a new region. The difference of observations has been used as a sign of a new region for mobile robot. An important feature of our approach is that the partitioning is very effective without any delay. We depict an implementation which use 2D laser scans as a part of whole submap SLAM algorithm. The experimental results demonstrate the performance of the method. The partition results from our approach are close to the human concept; thus it is easy to add semantic information on the submap for further navigation.