DOI: 10.1017/S000000000000000 Printed in the United Kingdom Autonomous SLAM driven by Monte Carlo uncertainty maps- based navigation

This paper addresses the problem of implementing a SLAM algorithm combined with a non-reactive controller (such as trajectory following or path following). A general study showing the advantages of using predictors to avoid mapping inconsistences in autonomous SLAM architectures is presented. In addition, this paper presents a priority based uncertainty map construction method of the environment by a mobile robot when executing a SLAM algorithm. The SLAM algorithm is implemented with a Extended Kalman Filter (EKF) and extracts corners (convex and concave) and lines (associated with walls) from the surrounding environment. A navigation approach directs the robot motion to the regions of the environment with the higher uncertainty and the higher priority. The uncertainty of a region is specified by a probability characterization computed at the corresponding representative points. These points are obtained by a Monte Carlo experiment and their probability is estimated by the sum of Gaussians method, avoiding the time-consuming map-gridding procedure. The priority is determined by the frame in which the uncertainty region was detected (either local or global to the vehicle’s pose). The mobile robot has a non-reactive trajectory following controller implemented on it to drive the vehicle to the uncertainty points. SLAM real time experiments in real environment, navigation examples, uncertainty maps constructions along with algorithm strategies and architectures are also included in this work.