Nerual Network Assisted Adaptive Unscented Kalman Filter for AUV

As one of the most important equipment for ocean exploration, Autonomous Underwater Vehicles (AUVs) has attracted wide attention. To collect the valid oceanic data in the complex underwater environment, autonomous navigation is an indispensable prerequisite for AUVs. Different from the traditional navigation algorithms, such as Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), the Adaptive Unscented Kalman Filter (AUKF) proposed by some scholars get better performance. However, there are only minor improvements for the navigation accuracy of AUKF. To improve the navigation accuracy, in this paper, we proposed a neural network assisted navigation method for AUV, which combines the AUKF with the neural networks. Firstly, the model proposed by this paper obtain the preliminary predictions of positions through AUKF. Secondly, the neural network is trained with the data calculated by AUKF and acquire the deviation. Finally, we correct the initial predictions of positions with the deviation. We compare the performance of the proposed methods to the AUKF using AUV real data. Experimental results show that neural network-assisted adaptive UKF filtering is more accurate for AUV navigation compared to the AUKF.