Autonomous robot navigation based on Simultaneous Localization and Mapping (SLAM), and particle filtering

This thesis provides a brief introduction to Simultaneous Localization and Mapping (SLAM) in extensive outdoor environments, a process that recently has made mobile robots truly autonomous. SLAM is a process in which localization and mapping are done simultaneously in an unknown environment without an access to a priori map. To solve the SLAM problem, sensor information and mathematical motion models are extensively used. This research introduces a probabilistic approach to a SLAM problem and under Gaussian and non-Gaussian conditions offers alternative solutions. Due to the non-linearity of the real systems, which are always subject to noise, an Extended Kalman Filter algorithm for the SLAM problem under Gaussian condition will be shown. Then an alternative way of dealing with SLAM problem with assumption of non-Gaussian and called FAST-SLAM will be analyzed. FAST-SLAM is an algorithm that using Rao-Blackwellised method for particle filtering, estimates the path of robot while the landmarks positions which are mutually independent and with no correlation, can be estimated by EKF. This process is doable using a factorization that fits very well in SLAM problem since there will be a representation of SLAM through out a Bayesian network. A solution to SLAM problem can be worthy if a proper representation of motion and sensor of a robot is accessible. In this thesis, a real outdoor autonomous robot is presented and the related equations are derived. Then simulated examples based on both filters are represented and results are discussed and compared.