Novel Improved UKF Algorithm and Its Application in AUV Navigation System

Abstract

Through the complex underwater environment, Autonomous Underwater Vehicle (AUV) must use an effective and accurate navigation algorithm to ensure precise localization and autonomous navigation without satellite navigation system. As a development of Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF) has the advantages of simple implementation, high filtering accuracy and good convergence. Nevertheless, unknown system noise statistics will cause decreased filtering accuracy or even divergence. This paper proposed a navigation algorithm combined the Extreme Learning Machine (ELM) and UKF (ELM-UKF) for the purpose of improving the navigation accuracy. The proposed algorithm uses ELM to establish a system model including error terms to correct the system error. Through the sea trial in Menlou Reservoir using own AUV platform, Sailfish, the performance of AUV navigation based on ELM-UKF algorithm is significantly superior to the standard EKF and standard UKF, revealing that the proposed algorithm successfully improving the accuracy of navigation system.