Precise Positioning with Machine Learning based Kalman Filter using GNSS/IMU Measurements from Android Smartphone

This paper presents GNSS/INS integration Kalman filter for enhancement of positioning accuracy and robustness to surrounding environment. In the Kalman filter system, filter parameters such as process noise covariance and measurement noise covariance selected in the tuning process determine the characteristics of the overall system. Therefore, the empirical knowledge of the filter designer should be fully employed in the tuning process, and finding proper parameter values is still a challenging work. We adopt reinforcement learning to find the process noise covariance of the filter parameter. The experimental results show that the improvement of navigation performance is achieved by the efficient use of the learned process noise covariance matrix.