Significant Feature Selection in Range Scan Data for Geometrical Mobile Robot Mapping

Abstract

Simultaneous Mapping and Localization (SLAM) has become a very important task for mobile robots. Different approaches have been proposed over the last years. Most of them use directly raw data or simple features as geometric representa- tions. Recent works on SLAM have been proposed different geometric representations, which captures more context than other features, permitting an additional cognitive and reasoning mapping. In this paper, we propose a method to select significant features used for constructing polygonal maps for indoor mobile robot navigation. To segment and group raw laser range scan data in polygonal curves (polylines) a discrete curve evolution method (DCE) has been applied. Relevance measure computa- tions derived from DCE, together with length of line segments and turn angles are used to select relevant features. Features selected for a single scan St at time t are matched against a sub- group gt-1 of features from the global map Gt-1 at t - 1, obtained from the last known position. A least squares method is applied to find optimal translation and rotation between partial map St and the global map Gt-1. Finally, maps are merged to get the actualized map Gt.