Localization and mapping approximation for autonomous ground platforms, implementing SLAM algorithms

Abstract

Solving the SLAM (Simultaneous Localization And Mapping) problem increases the robot autonomy allowing the robot to build a map of the environment and simultaneously, estimate its pose (position and orientation) using this map. Usually, the robot pose is computed by means of odometry. This method uses the sensor information, nevertheless, these sensors introduce noise on the measurements, generating uncertainty in robot localization. For this reason, probabilistic methods as Kalman and Particle filters have been implemented in robotics to estimate the robot and landmark positions, manipulating the system errors. In this paper SLAM algorithms based Rao-Blackwellized particle filters were implemented in a mobile robotic platform to perform two-dimensional and three-dimensional unknown environments reconstructions. The SURF (Speed Up Robust Features) and RANSAC (Random Sample Consensus) methods were implemented for the visual feature extractions and the transformation estimations respectively. Furthermore, ROS (Robot Operating System) was used for platform position estimation and map building. Finally, a Monte Carlo algorithm was employed for autonomous navigation. This algorithm computes the shortest path between two points within the reconstructed map and executes it avoiding obstacles that may arise in its travel.