A Runs Test-Based Kalman Filter With Both Adaptability and Robustness With Application to INS/GNSS Integration

Abstract

Kalman filter (KF) is one of the most efficient solutions applied for the integration of inertial navigation system (INS) and global navigation satellite system (GNSS). However, for standard KF derived from the ideal system model assumption, various disturbance factors inevitably decrease its estimation accuracy in practical scenarios. Therefore, it is necessary to suppress the interference factors such as inaccurate noise covariance or irregular process uncertainty during filtering, thus improving the integrated system’s stability and accuracy. Nevertheless, existing solutions usually target specific individual issues and lack compatibility with nonstationary scenarios in that multiple uncertain factors coexist. In this article, an improved KF with a switching scheme based on runs hypothesis test is designed to enhance the algorithm’s performance in nonstationary environments, and a process uncertainty robust filter based on mixture correntropy is proposed to suppress the potential diverge risks. By integrating with the variational Bayesian approach into the switching scheme, the noise covariance adaptability of the proposed scheme is further improved, and an improved KF with both adaptivity and robustness is obtained. The simulation and field test results demonstrate that the proposed filter achieved the expected superior performance in challenging nonstationary conditions.