EKF, AEKF and FastSLAM applications for improved solution of slam problem

Simultaneous Localization and Mapping (SLAM) problem is that a mobile robot maps the environment by using only sensor data during its movement in an unknown place and determines its position by using this map. SLAM is a problem that needs to be solved in order to ensure the autonomous motion of robot. Particle Filters (PF) and Kalman Filters (KF), which are probability-based filters, are used to solve this problem. Extended Kalman Filter (EKF) is one of the most widely used probability-based filter. Provided a properly defined system model and known fixed noise statics, EKF is an approach that gives acceptable results. However, these conditions can be done in ideal simulation environments. It is known that in real life, noise cannot be considered constant due to its characteristics, and even noise has a constantly increasing structure. Therefore, Adaptive Extended Kalman Filter (AEKF) is used to estimate noise statics recursively. In order to maintain the positive define characteristics of noise covariance Innovation Covariance Estimation (ICE) was used together with AGKF. One of the most important step of FastSLAM applications where PF is used to solve the SLAM problem is resampling. With resampling, degenerating possibility of particles is avoided. Five of the most used resampling methods in the literature are used for comparisons in this study. In this study, EKF, AEKF, AEKF-ICE and PF solutions are compared under varying environmental conditions. The effects of varianciesin the initial value of velocity and steering component of process noise, and range and bearing component of measurement noise on the performance of the filters are investigated separately. During these investigations, cases where data association, which is the state of establishing correlation between observation and prediction made by the robot, is given to the robot before motion or not, has been taken into consideration. Comparisons are made over the RMSE values of differences between true and predicted positions of robot and landmarks. As a result of the comparisons, it is seen that AEKF gave better results than other methods, and ICE provided improvements in these results.