Global Localization of Mobile Robots for Indoor Environments Using Natural Landmarks

This paper introduces a global localization system based on natural landmarks for indoor mobile robots. The proposed approach is based on recognition of natural landmarks from laser scanner data. A previously built grid-based map is pre-processed off-line to obtain a model of landmarks and their attributes for each cell. The robot´s position and orientation are calculated by finding correspondence between the identified landmarks from robot´s current position and the landmarks associated to the model. This proposed approach called GL2 follows a two stage process. Initially a fast initial filter based on the number and type of landmarks is used to substantially reduce the search space. The second stage uses a modified discrete relaxation algorithm to perform a more detailed analysis and find the robot´s location and orientation. It is shown the robustness of the algorithm in complex and real office like environments, even in the presence of previously unknown obstacles