A MAPAEKF-SLAM ALGORITHM WITH RECURSIVE MEAN AND COVARIANCE OF PROCESS AND MEASUREMENT NOISE STATISTIC

Abstract

The most popular filtering method used for solving a Simultaneous Localization and Mapping is the Extended Kalman Filter. Essentially, it requires prior stochastic knowledge both the process and measurement noise statistic. In order to avoid this requirement, these noise statistics have been defined at the beginning and kept to be fixed for the whole process. Indeed, it will satisfy the desired robustness in the case of simulation. Oppositely, due to the continuous uncertainty affected by the dynamic system under time integration, this manner is strongly not recommended. The reason is, improperly defined noise will not only degrade the filter performance but also might lead the filter to divergence condition. For this reason, there has been a strong manner well-termed as an adaptive-based strategy that commonly used to equip the classical filter for having an ability to approximate the noise statistic. Of course, by knowing the closely responsive noise statistic, the robustness and accuracy of an EKF can increase. However, most of the existed Adaptive-EKF only considered that the process and measurement noise statistic are characteristically zero-mean and responsive covariances. Accordingly, the robustness of EKF can still be enhanced. This paper presents a proposed method named as a MAPAEKF-SLAM algorithm used for solving the SLAM problem of a mobile robot, Turtlebot2. Sequentially, a classical EKF was estimated using Maximum a Posteriori. However, due to the existence of unobserved value, EKF was also smoothed one time based on the fixed-interval smoothing method. This smoothing step aims to keep-up the derivation process under MAP creation. Realistically, this proposed method was simulated and compared to the conventional one. Finally, it has been showing better accuracy in terms of Root Mean Square Error (RMSE) of both Estimated Map Coordinate (EMC) and Estimated Path Coordinate (EPC).