Switching reference image-based visual servoing forunderwater docking of autonomous underwater vehicle
Loading...
Date
2019-03-01
Authors
Mohd Faid Yahya
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Autonomous underwater vehicle (AUV) has great potential to dive deep under the
ocean and perform various tasks. However, the AUV operates on a limited amount of
battery capacity. To overcome this limitation, underwater docking is required so that
when the AUV docks, it is able to recharge the battery to full capacity. One of the
ways to achieve underwater docking is by means of vision-based robot control or
visual servoing. There are many methods to perform visual servoing such as positionbased
visual servoing (PBVS), image-based visual servoing (IBVS), and 2-½-D visual
servoing. Nevertheless, these methods failed when there is no resemblance of target
features between acquired and desired images. Such problem arises when the target
features from acquired image could be out of image plane or disoriented due to AUV’s
skewed position or appeared to be disfigured due to harsh underwater conditions. To
resolve this problem, a switching reference IBVS method is proposed in this study. To
realize the proposed method, a control system based on Proportional-Integral and
Sliding-Mode controllers are developed to control the AUV movement. Then, vision
system is developed based on deep learning to detect and classify targets installed on
the docking station. Subsequently, the switching reference IBVS method is developed
for guiding the AUV into the docking station. The underlying concept of the proposed
method is to switch the desired target features to match the currently acquired target
features. The method also enables the AUV to switch between two modes of operation
which are homing and docking. In addition, an AUV and a docking station prototypes
have been developed to verify simulation results. Simulation wise, the developed
control system has responsiveness to track desired trajectory by 93.89% and robust
under the effect of lateral water current up to 0.1 meter per second. As for the
developed vision system, the detection and classification accuracies of targets based
on confusion matrix are 96.68% and 99.72% respectively. Then, for switching
reference IBVS, when benchmarked, the proposed method excelled in reliability to
avoid collision between AUV and docking station by 83.33% and more robust under
the effect of missing target features when compared to normal IBVS method and IBVS
with attitude keeping control method by 100%. Finally, from experimental result, the
number of successful trials for underwater docking using the proposed method is 20
out of 24 or 83.33%.