Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Efficient Detection and Control System for Underwater Docking using Machine Learning and Realistic Simulation: A Comprehensive Approach

Jalil Chavez-Galaviz, Jianwen Li, Matthew Bergman, Miras Mengdibayev, Nina Mahmoudian

TL;DR

This work compares different deep-learning architectures to perform underwater docking detection and classification, and shows the performance of the proposed approach by showing experimental results on the off-the-shelf AUV Iver3 and compared with classical vision methods.

Abstract

Underwater docking is critical to enable the persistent operation of Autonomous Underwater Vehicles (AUVs). For this, the AUV must be capable of detecting and localizing the docking station, which is complex due to the highly dynamic undersea environment. Image-based solutions offer a high acquisition rate and versatile alternative to adapt to this environment; however, the underwater environment presents challenges such as low visibility, high turbidity, and distortion. In addition to this, field experiments to validate underwater docking capabilities can be costly and dangerous due to the specialized equipment and safety considerations required to conduct the experiments. This work compares different deep-learning architectures to perform underwater docking detection and classification. The architecture with the best performance is then compressed using knowledge distillation under the teacher-student paradigm to reduce the network's memory footprint, allowing real-time implementation. To reduce the simulation-to-reality gap, a Generative Adversarial Network (GAN) is used to do image-to-image translation, converting the Gazebo simulation image into a realistic underwater-looking image. The obtained image is then processed using an underwater image formation model to simulate image attenuation over distance under different water types. The proposed method is finally evaluated according to the AUV docking success rate and compared with classical vision methods. The simulation results show an improvement of 20% in the high turbidity scenarios regardless of the underwater currents. Furthermore, we show the performance of the proposed approach by showing experimental results on the off-the-shelf AUV Iver3.

An Efficient Detection and Control System for Underwater Docking using Machine Learning and Realistic Simulation: A Comprehensive Approach

TL;DR

This work compares different deep-learning architectures to perform underwater docking detection and classification, and shows the performance of the proposed approach by showing experimental results on the off-the-shelf AUV Iver3 and compared with classical vision methods.

Abstract

Underwater docking is critical to enable the persistent operation of Autonomous Underwater Vehicles (AUVs). For this, the AUV must be capable of detecting and localizing the docking station, which is complex due to the highly dynamic undersea environment. Image-based solutions offer a high acquisition rate and versatile alternative to adapt to this environment; however, the underwater environment presents challenges such as low visibility, high turbidity, and distortion. In addition to this, field experiments to validate underwater docking capabilities can be costly and dangerous due to the specialized equipment and safety considerations required to conduct the experiments. This work compares different deep-learning architectures to perform underwater docking detection and classification. The architecture with the best performance is then compressed using knowledge distillation under the teacher-student paradigm to reduce the network's memory footprint, allowing real-time implementation. To reduce the simulation-to-reality gap, a Generative Adversarial Network (GAN) is used to do image-to-image translation, converting the Gazebo simulation image into a realistic underwater-looking image. The obtained image is then processed using an underwater image formation model to simulate image attenuation over distance under different water types. The proposed method is finally evaluated according to the AUV docking success rate and compared with classical vision methods. The simulation results show an improvement of 20% in the high turbidity scenarios regardless of the underwater currents. Furthermore, we show the performance of the proposed approach by showing experimental results on the off-the-shelf AUV Iver3.
Paper Structure (23 sections, 5 equations, 13 figures, 2 tables)

This paper contains 23 sections, 5 equations, 13 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Underwater Docking
  3. Docking Strategy
  4. Detection Architecture
  5. Dataset Generation
  6. Architecture Comparison
  7. CNN
  8. Residual Network
  9. Pyramidal Convolution
  10. Transformer
  11. Knowledge Distillation
  12. Underwater Simulation
  13. AUV Model
  14. Simulation Setup
  15. Underwater Imagery
  16. ...and 8 more sections

Figures (13)

  • Figure 1: Underwater docking setup used in this work consisting of the AUV platform (Iver3), and the dock (flat funnel-shaped docking station) with a light attached to aid with the optical based underwater. (a) shows an image from the AUV going towards the docking station in simulation. (b) depicts the AUV approaching the docking station during field experiments. (c) illustrates a CAD model of the AUV docked in the docking station.
  • Figure 2: Multiple stages of the docking system, during the approach stages, the AUV gets close to the docking stations. During the terminal homing it uses a combination of optical and acoustic guidance to quickly adapt to the docking station position.
  • Figure 3: Underwater docking strategy composed of several paths that guide the AUV towards the ASV carrying the docking station. Once the vehicle is close enough and the light beacon is in a visible range, the AUV is optically guided. If at any point the light is out of the frame, then the AUV switches back to follow the previously generated path.
  • Figure 4: Architecture of the docking system. During approach setup and approach stages, path planning and path follower components generate a path to align the vehicle towards the docking station using the information received through the acoustic modem. The AUV is equipped with a camera. The output of the camera is processed to generate an underwater-looking image fed into the underwater docking detection algorithm. This procedure is employed during the terminal homing stage to perform a fine adjustment of the vehicle pose to dock successfully. If a failure is detected the procedure is repeated in the missed approach stage.
  • Figure 5: Distribution of the docking position in the image frame on the original dataset (a), and the final dataset (b) after the generation of artificial images. As can be observed, the distribution of the docking position in (b) is more evenly distributed to avoid any bias in the detection.
  • ...and 8 more figures