抄録
In this paper, we propose an automatically learning technique to solve tweaking parameter problems of Unscented Kalman Filter (UKF) for accurate localization. The parameters consist of three kinds of parameters: I) Covariance matrix of input noise. II) Covariance matrix of measurement noise. III) Hyper-parameter of UKF. For the parameter learning. one of discriminative training methods is adopted to obtain the optimal parameters automatically. Simulation and experimental results in an outdoor environment are presented. We demonstrate the effectiveness of our proposed learning method for localization.
