EM and Vision-Aided Multisensor Fusion Homing Guidance System (MSF-HGS) for Intelligent AUVs

Autonomous underwater vehicle (AUV) is an efficient technological tool for ocean exploration in the mid-21st century. Submerged docking stations (DSs) serve as charging stations to enhance vehicle endurance and enable long-range operations. In this regard, precise homing guidance techniques are essential for navigating toward homing and DS in unpredictable ocean environments. This article proposes an unsupervised multisensor fusion homing guidance system (MSF-HGS) based on Kalman filter (KF), extended KF (EKF), and advanced filter-like cubature KF (CKF) techniques. The system uses data from K-nearest neighbors (KNNs) techniques for electromagnetic (EM) field and deep learning (DL)-based underwater images. The novelty of this work lies in the fusion of different sampling frequency data (EM+ vision) in time series. Performance was validated by integrating multirate sensor data through field experiments, with EM field measurements at 2 Hz and optical images at ten frames per second collected onboard an AUV, referenced to the DS. Real-time measurements and computed AUV kinematics are fused to determine the optimal guidance for intelligent homing of AUV toward the DS. The performance of unsupervised sensor fusion of KF and EKF is demonstrated in real-time in different water bodies. The DL system is resistant to stray magnetic fields and turbid water conditions within range of 7 m and terminal homing precision within ±0.2 m from the center of dock. The success probability is 99.65% using the proposed mechanism, while it is 90% with conventional KF methods, 95% with EKF technique, and 96% with CKF technique.