Fuzzy logic docking control for underwater vehicles based on deep learning feature selection

In this paper, a fuzzy logic controller based on sonar and optical camera sensing will be developed and deployed. At long range, deep learning is used for detection and localization of the docking station in sonar images to find the direct way to the docking station. In the close range deep learning is applied to identify features of the docking station in optical camera images, which are then used by the fuzzy system to orient and navigate the vehicle into the docking station. The navigation system is formed of three individual systems. The first one is the sensing, detection and localization system. The second one is fuzzy logic controller for guidance and the third one is a fuzzy logic unit for the thruster control. A robust docking guidance system is identified as the most and crucial part of the homing process to ensure successful underwater vehicle homing and docking. Therefore, inspite of the end-to-end solutions, we propose a methodology based on the combination of deep learning for identification and the well-known robust fuzzy control, which can cover up some uncertainties in the object detection and localization component. The fuzzy logic guidance controller in the second layer is designed to incrementally decrease the cross track error (XTE) between the vehicle and the docking station and then maintain it at zero. The input of the fuzzy logic guidance controller is produced by the detection and localization system, which provide range and bearing. The system was tested using data collected from the underwater test basin. The detection system showed a detection rate of 80% and a localization rate of 98%. The preliminary results show that appropriate fuzzy rules are fired for different vehicle situations as if an expert is navigating the vehicle to and into the docking station. For the 100 observation points in the test basin, the fuzzy guidance controller navigated the vehicle with an average heading error of 0.01 degrees.

I. Introduction

Due to several factors, such as battery and payload data memory capacities, underwater vehicle cannot stay much longer under water. Dependency on sea conditions for launch and recovery, increased misson costs due to the requirement of a ship for launch and recovery are other factors. With docking capabilities, an underwater vehicle can stay longer underwater, reducing the time and the costs incurring through launch and recovery and the necessary ship. One of the major issues when docking in the underwater environment is the lack of precision localization system such as GPS on land. This makes relative navigation by orienting on underwater structure a very important feature.