R-SLAM: Resilient localization and mapping in challenging environments

Abstract

Accurate pose estimation plays an important role in solution of simultaneous localization and mapping (SLAM) problem, required for many robotic applications. This paper presents a new approach called R-SLAM, primarily to overcome systematic and non-systematic odometry errors which are generally caused by uneven floors, unexpected objects on the floor or wheel-slippage due to skidding or fast turns. The hybrid approach presented here combines the strengths of feature based and grid based methods to produce globally consistent high resolution maps within various types of environments. The proposed R-SLAM algorithm improves the pose estimation by, positioning some of the particles relative to the features existing in the environment. So instead of assuming unimodal Gaussian distribution of particles, as it is in GMapping method, R-SLAM handles the odometry errors better without increasing the number of particles. R-SLAM approach does not necessitate any pre-defined landmarks, yet makes use of them whenever they exist in the environment. Restricted use of the features keeps the computational complexity suitable for real time applications in large scaled environments. A four wheeled, differential drive mobile robot platform and ROS framework have been used to implement and test the proposed method. Field test results reported in this article are obtained in historical Hagia Sophia museum, Istanbul, which occupies an area of 7500 square meters. To provide analogous results, another set of tests are performed on publicly available datasets and SLAM evaluation toolkit developed at the University of Freiburg is used to compare the results. (C) 2016 Elsevier B.V. All rights reserved.