Abstract: For improving the estimation accuracy and the convergence speed of the unscented Kalman filter (UKF), a novel adaptive filter method is proposed. The error between the covariance matrices of innovation measurements and their corresponding estimations/predictions is utilized as the cost function. On the basis of the MIT rule, an adaptive algorithm is designed to update the covariance of the process uncertainties online by minimizing the cost function. The updated covariance is fed back into the normal UKF. Such an adaptive mechanism is intended to compensate the lack of a priori knowledge of the process uncertainty distribution and to improve the performance of UKF for the active state and parameter estimations. The asymptotic properties of this adaptive UKF are discussed. Simulations are conducted using an omni-directional mobile robot, and the results are compared with those obtained by normal UKF to demonstrate its effectiveness and advantage over the previous methods.