Slam with corner features from a novel curvature-based local map representation

This paper presents a solution to the Simultaneous Localization and Map Building (SLAM) problem for a mobile agent which navigates in an indoor environment and it is equipped with a conventional laser range finder. The approach is based on the stochastic paradigm and it employs a novel feature-based approach for the map representation. Stochastic SLAM is performed by storing the robot pose and landmark locations in a single state vector, and estimating it by means of a recursive process. In our case, this estimation process is based on an extended Kalman filter (EKF). The main novelty of the described system is the efficient approach for natural feature extraction. This approach employs the curvature information associated to every planar scan provided by the laser range finder. In this work, corner feature has been considered. Real experiments carried out with a mobile robot show that the proposed approach acquires corners of the environment in a fast and accurate way. These landmarks permit to simultaneously localize the robot and build a corner-based map of the environment.