The Improvement of the Prescision of an Inertial Navigation System for AUV based on the Neural Network

"URASHIMA" is an Autonomous Underwater Vehicle which was developed in Japan Agency for Marine-Earth Science and Technology in Japan. "URASHIMA" has the Inertial Navigation System in order to know its own current position, and it can cruise autonomously owing to the position data. However the position data includes some error which increases with the passage of time. In this paper, the method to improve the precision of the Inertial Navigation System based on the neural network is described. The network outputs the position data, which compensates the error of the position data calculated from it.

I. Introduction

Japan Agency for Marine-Earth Science and Technology (JAMSTEC) has some manned vehicles, Remotely Operated Vehicles and Autonomous Underwater Vehicles (AUVs). One of the AUVs is the Deep-sea Cruising AUV that is referred to as “URASHlMA” [1]. Its missions are the investigation of the seafloor in the wide area and the measurement of sea data in various areas. Therefore “URASHIMA” has the ability to cruise for several tens of hours. Since “URASHIMA” executes such missions continuously, it must be equipped with functions that it cruise autonomously for the long time. And it is impossible for AUVs like “URASHIMA” to cruise for the long time autonomously if it doesn't get current its own current position. Therefore AUVs are generally equipped with an Inertial Navigation System (INS), and the same is true in URASHIMA's case. The INS calculates the absolute position data (latitude/ongitude) of a moving object in real time. However the position data includes some error because the sensors, which are set inside the INS, have drift/bias error. And the error increases with the passage of time. Therefore it is very difficult that “URASHIMA” cruises for the long time dependent only on the INS without communications from its external environment.