Using Regularized Softmax Regression in the GNSS/INS Integrated Navigation System with Nonholonomic Constraints

Abstract

The integration of global navigation satellite system (GNSS) and Inertial navigation system (INS) is widely implemented in land-vehicle navigation applications. However, the satellite signal is vulnerable in some special urban scenarios, consequently errors in terms of position, velocity and attitude grow rapidly in stand-alone mode especially for low-cost MEMS-based INS. In the conventional tight combination navigation schemes, system works on predicting model during the GNSS signal outage and the positioning accuracy is determined by the precision of the inertial navigation. Besides the lack of observation makes the estimate of inertial navigation error with GNSS information less reliable due to the satellite signal loss. In this paper, an improved non-holonomic constraints (NHC) method based on regularized softmax regression is proposed to enhance navigation precision when the number of visible satellite is insufficient. The velocity constraint condition is applied to simplify the system calculating equations of MEMS-based INS. Furthermore, a regularization softmax regression model based on the collected data is trained to recognize the vehicle motion pattern so as to realize deeper constraints. Simulation and field-test results indicate that the method is beneficial to raise the precision of low-cost GNSS/INS integrated navigation receiver by efficiently reduce the navigation errors.