Improved Performance of Adaptive Ukf Slam with Scaling Parameter

Yalçın K., Karaçam S., Navruz T. S.

3rd International Conference on Artificial Intelligence and Applied Mathematics in Engineering, Antalya, Turkey, 01 October 2021, pp.44

  • Publication Type: Conference Paper / Summary Text
  • City: Antalya
  • Country: Turkey
  • Page Numbers: pp.44
  • Gazi University Affiliated: Yes


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.