Auto-tuning extended Kalman filters to improve state estimation

In this a paper, an auto-tuning Extend Kalman filter (EKF) approach is developed. The objective is to design an algorithm to find the optimal values of the covariance matrices Q and R. Manual tuning of those parameters is hard and time-consuming. Besides, wrong combinations of their values can lead to filter divergence and inconsistency. The proposed approach combines several metrics derived from the filter requirements especially the filter consistency. A weighted cost function is established based on the defined metrics. The approach effectiveness is tested and verified on sensor fusion problems for drone indoor localization where good results are achieved using five (5) different numerical optimization solvers.