Machine learning-based electro-magnetic field guided localization technique for autonomous underwater vehicle homing

Abstract

Precise guidance is the primary requirement for the reliable homing of autonomous underwater vehicles (AUV). The necessity for accurate short-range position tracking arises due to the limitations of the inertial sensors’ cumulative integral drift, unreliable acoustic positioning system in very short ranges, and non-propagation of high-frequency electromagnetic signals of the GPS system. In this article, a very low-frequency electro-magnetic (EM) field is generated for very short-range guidance for the homing operation of underwater vehicles. The design and modeling of the 3-axis low-frequency EM-based shallow water submerged Docking Station (DS) at 10 m depth is explained and the EM field is generated for a 20 m range in the 3-axis plane using two perpendicular coils operated in distinct frequencies. The stochastic magnetometer measurement model is used to estimate the relative kinematics of the underwater vehicle with respect to the EM docking coil. The EM field datasets generated from EM field maps around DS are used for training and validating with different Machine Learning (ML) algorithms. The designed ML aided EM-based tracking technique guidance algorithms are validated for underwater vehicle localization that requires a precise and reliable homing process to the DS even in turbid waters, bio-fouling environment, and reflective boundaries though it has limitations in presence of ambient ferromagnetic sources. Numerical simulations have been carried out with conventional methods and ML algorithms on the generated EM data. The results indicate that the position accuracy achieved is within 1% for ML-aided EM based guidance techniques when compared to the conventional algorithm which has a position accuracy of 3%. ML aided techniques are advantageous as they are independent of the dynamic ocean environment and the underwater vehicle configuration and the Time to Home (ToH) to the DS is simulated in different scenarios, is 160 s–200 s with a vehicle speed of 0.25 m/s. It is observed that the vehicle can choose the shortest path for homing operation with the measured EM field values and with aid of ML techniques.