Smoother Aided Neuro Fuzzy Kalman Filter Approach to SLAM Problem

Haydar Ankışhan


Neuro Fuzzy, SLAM


Many Kalman based filters have been used for solving simultaneous localization and mapping problem for mobile robot and vehicles. These filter performances usually depend on the knowledge of priori information about process and measurement noises, state parameters and design matrices (Q and R). Unlike the previous studies, improved unscented Kalman filter (IUKF) is used in this work which provides an alternative solution to estimate state of model and increase the performance of conventional unscented Kalman filter. Some improvements have been applied for the filter. The first improvement is to use Q and R matrices tuned by adaptive neuro fuzzy inference system (ANFIS). The second is to use Rauch-Tung-Striebel (RTS) smoother that improves the correct a priori knowledge of filter. Some scenarios with varying land marks are studied by employing the mobile robot or vehicle simultaneously to localize and to map the environment using our improved Kalman filters. The performances of the improved filter have been evaluated. It has been successfully shown that the modified filter is able to improve performance of conventional UKF and can provide robust and accurate solutions.

Important Links:

Go Back