Accurate localization is required for autonomous robots to navigate in cluttered environments safely. Therefore, Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF), which incorporate probabilistic concepts as localization methods, have been researched up to now. It should be noted, however, that the errors of kinematic parameters such as wheel diameter, tread, and mounting sensor position are not considered in the previous works. The present research proposes an Augmented UKF (AUKF), which is an extension of the UKF and can estimate the kinematic parameters together with the localization. The UKF and the AUKF are compared through some simulations to show that the proposed AUKF is much more accurate than the UKF. Additionally, localization experiments with only odometry are conducted using a real robot. It is shown from the experimental results that the localization using kinematic parameters estimated by the AUKF is more accurate than that using values measured by hand in advance.