Environmentally Dependent Adaptive Parameterization of a GNSS-aided Tightly-Coupled Navigation Filter

Abstract

Parameterization of global navigation satellite system (GNSS)-aided navigation filter is an active research topic, because it is crucial for the state estimation accuracy and there is little theoretical guidance. This publication presents parameterization for extended Kalman filter (EKF) with the help of Bayesian optimization. Different ways to model and parameterize the measurement noise are discussed. An adaptive parameterization scheme is proposed, which maps the environment according to the dilution of precision (DOP) and signal-to-noise ratio (SNR). The new adaptive parameterization approach is evaluated with a test car in Aachen, Germany. Results are compared to a sigma-epsilon variance model and show a remarkable improvement of position estimation accuracy and preciseness. In average, the mean error along the validation data set is reduced by 2.5 m and the standard deviation is halved.