OPTIMAL EKF FOR QUASI-TIGHTLY COUPLED GNSS/INS INTEGRATION

In this paper, we investigate a navigation technique based on a modified optimization method for the operation of an extended Kalman filter. The algorithm presented recursively adapts the noise covariance matrices that characterize the model at each iteration. This optimization technique is intended to be an add-on for classical localization processing, in order to fulfill demanding accuracy specifications even in the presence of large dynamics. More specifically, we target the development of a particular quasi-tightly coupled GNSS/INS integration, which enables an cost-effective positioning solution, regardless of the receiver internal architecture. In this respect, we also describe a novel realization of this coupling in the form of the model used for Kalman filtering. Our goal is to provide the precision gain that is required to address the critical constraints of modern automotive applications in real-time. We finally show that the development of such a navigation method could be appropriate for a real-world scenario by evaluating the algorithm performance in tracking a fast-moving vehicle. A set of numerical simulations compares the results achieved either with or without optimization, testing different common racetracks. This comparison exhibits that the position estimates RMSE can be reduced at least by a factor of about 4.