Trajectory-Based Analysis in Nonlinear Navigation Using Extended Kalman Filter with Different Covariance Matrix Acquisition Methods

This paper focuses on the effect of extended Kalman filter (EKF) implementation in dealing with the navigation uncertainties for Turtlebot3 Burger mobile robot considering different initial covariance matrices implemented on different trajectory patterns. EKF is one of the most famous and simple algorithms that has been widely known for its ability to improve accuracy in nonlinear applications. One of the requirements in implementing EKF is to have a reliable covariance value that acts as an initializer before performing any navigation. Previous studies tend to ignore the importance of initial covariances by only assuming the value as an identity matrix, using existing datasets or using random values rather than obtaining it technically. Therefore, in this study, the initial covariance is formulated from experimental setup as well as simulation setup, and being evaluated on square, curve and diamond routes. The trajectory results proved that initial covariance from experimental setup is considered reliable as it improved the uncertainties of the mobile robot with small Euclidean distance errors of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{2 6. 5 9 \%, ~} \mathbf{7. 0 9 \%}$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1. 7 3 \%}$</tex> for square, curve and diamond routes respectively. Hence, the proposed method shows that the navigation performance can be improved by acquiring data from experimental setup for formulating initial covariance of Turtlebot3 Burger mobile robot in future applications.