Akademik Veri Yönetim Sistemi
3rd International Conference on Artificial Intelligence and Applied Mathematics in Engineering, Antalya, Türkiye, 01 Ekim 2021, ss.44
Simultaneous Localization and Mapping (SLAM) deals with simultaneous mapping
and estimates the position of a robot moving in an environment. One of the most common
method used in solving the SLAM problem is Extended Kalman Filter (EKF). EKF is an
approach that gives acceptable results in the ideal simulation environment when the system
model is known accurately. However, in real life applications, EKF may have problems due
to the variable nature of environment noise. In this study, Adaptive EKF (AEKF) method with
an adaptive-based approach is used to solve this problem. AEKF aims to estimate the noise
statistics and covariance matrices in the classical EKF at each time step. During the estimation process, Maximum Likelihood Estimation (MLE) and Expectation-Maximization (EM)
Creation methods are used. Although this solution gives a high quality result, there is a risk
of non-positive definite matrices in the noise statistics obtained in the algorithm. Innovation
Covariance Estimation (ICE) was included to minimize this possibility. The mapping performances and the Root Mean Square Error (RMSE) values of AEKF are compared with EKF
and Uncented Kalman Filter (UKF) and it is seen that AEKF gives better results.