Dual Free-Size LS-SVM Assisted Maximum Correntropy Kalman Filtering for Seamless INS-Based Integrated Drone Localization

It is known that in seamless inertial navigation system (INS)-based integrated drone localization, once data outage occurs in auxiliary measurement technologies, such as global navigation satellite system and ultrawideband (UWB), traditional integrated navigation data fusion filters may give large errors. To overcome this issue, we develop a dual free-size least squares support vector machines (fLS-SVM) assisted maximum correntropy Kalman filter (KF) (mcKF) for fusing INS and UWB data. In the fLS-SVM structure, the interactive multiple models method is used to improve the adaptability of the training set of LS-SVM. Then, the fLS-SVM method is adopted to estimate position error from INS when the UWB data is outage and then used again to provide the estimation error of the previous LS-SVM algorithm. The mcKF is derived on the basis of the maximum correntropy criterion. One mcKF is to realize data fusion from INS and UWB, and another one is for estimating the INS's position error for the fLS-SVM's output. Experimental testing is done in INS/UWB-based indoor mobile drone localization environments. It is shown that the dual fLS-SVM assisted mcKF algorithm outperforms the KF, general KF with colored measurement noise, mcKF, KF+LS-SVM, and dual KF+LS-SVM it reduces the position errors by an average of about 93.64%, 93.63%, 93.63%, 39.99%, and 94.10%, respectively, in two tests.

Automated summary

This paper proposes a dual fLS-SVM assisted mcKF algorithm for INS and UWB data fusion, which reduces position errors when auxiliary measurement technologies are unavailable.