Towards a non-probabilistic approach to hybrid geometry-topological SLAM

This article introduces a new approach to Simultaneous Localization and Mapping (SLAM) which pursues robustness and accuracy in large-scale environments. Unlike most successful works on SLAM, we use non-probability method to build the local geometry map, and let the loop closure data fusion module to deal with the inconsistency of global map which is the most difficult problem in SLAM. We also design other modules which are involved in our framework. In the Map partition module, we present a novel concept, which is called reinforced Sensed-Space Overlap, and use it in graph-based segmentation of the local region segmentation algorithm. The assumption-proven strategies were used in the loop closure data fusion module. In the meantime, two different algorithms were presented. The performance of framework and each module are testified by the experimental results.