Localization of mobile robots in indoor environments is challenging due to measurement noise, obstacles, and error accumulation, which can significantly affect positioning accuracy. This paper investigates the application of the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) for estimating the position and orientation of a wheeled mobile robot using a nonlinear kinematic model. A sensor fusion method is proposed to integrate Attitude and Heading Reference System (AHRS) measurements with an encoder-based kinematic model to improve localization accuracy.



In the proposed approach, wheel encoder measurements are used by the robot kinematic model to estimate its position and orientation during motion. The robot is commanded to follow predefined trajectories, and its position and traveled distance are estimated from the encoder data. AHRS measurements are then incorporated as external observations in the Kalman filtering framework to correct the predicted states and reduce the effects of accumulated errors. Both EKF and UKF are experimentally implemented and evaluated under the same operating conditions.



To assess the localization performance, the actual robot trajectory is recorded using an optical tracking system, which provides reference position measurements. The position estimates obtained from the encoder-based model, EKF, and UKF are compared with the reference trajectory using quantitative error measures. The experimental results show that, at the beginning of the trajectories, the position estimates obtained using both EKF and UKF closely follow the reference path. However, as the traveled distance increases, the encoder-based estimation error gradually grows because of cumulative measurement and modeling errors. The proposed sensor fusion approach effectively reduces these accumulated errors by combining encoder and AHRS measurements.



The comparison of the two filtering methods demonstrates that the UKF achieves slightly better localization accuracy than the EKF in both experimental trajectories and provides a greater reduction in the overall positioning error. Nevertheless, because the robot motion model exhibits relatively limited nonlinearity, the performance difference between EKF and UKF remains relatively small.