Autonomous Underwater Vehicles (AUVs) have become fundamental tools for marine scientists and industries to explore and monitor underwater areas. Non-predictable environmental conditions and sensor acquisitions make the design of AUV surveys challenging even for expert operators. Multiple attempts are required, and the collected data quality is not guaranteed: the AUV usually passively stores the sensors’ acquisitions that are then analyzed offline after its recovery by human operators in charge of identifying and localizing the so-called Objects of Potential Interest (OPIs). When it comes to acoustic images, the lack of features and low resolution make this task even more challenging. As a consequence of these statements, the marine community has sought robots able to meaningfully perceive and model the surroundings and autonomously conduct the assigned task, which are the foundations of fully autonomous vehicles. In this work, these topics have been investigated. To allow a compact and lightweight AUV to gather knowledge of the surroundings, an Automatic Target Recognition (ATR) strategy based on modern Convolutional Neural Networks (CNNs) for onboard online applications was developed. The ATR methodology was used to identify and localize potential targets of interest in Forward-Looking SONAR (FLS) imagery. Then, to avoid using pre-planned surveys and make an AUV actively considering the acquired data online, this thesis presents a probabilistic framework for FLS-driven seabed inspections. The realized sensor-driven Receding-Horizon Coverage Approach (RHCA) endows the AUV with the ability to autonomously conducting the survey and ensures adequate coverage of the target area. A Rapidly-exploring Random Tree (RRT) inspired view planning algorithm for underwater inspections was designed. Then, advancements for enhancing the performance of the algorithm have been carried out. In particular, a novel informed tree expansion methodology for guiding the vehicle towards the non-explored regions is proposed. Thanks to this solution based on the Kernel Density Estimation technique, the AUV learns the distribution of the discovered map. In addition, to explore other view planning strategies, a preliminary investigation about the exploitation of a Randomized Model Predictive Control (RMPC) approach for conducting autonomous seabed inspections is reported. Finally, the proposed ATR methodology and the RHCA have been combined to realize a target-aware planning solution for autonomously inspecting an area of interest. A probabilistic semantic map that includes the knowledge about the presence of the OPIs is created and updated by using the ATR findings. The semantic map enables the view planning algorithm to generate paths that cover the area of interest and simultaneously reduces the target localization uncertainty. Therefore, this methodology allows an AUV to meaningfully perceive and model the surroundings and autonomously conduct inspections surveys. The solutions proposed in this thesis have been firstly validated with realistic simulations made by means of the Unmanned Underwater Vehicle Simulator (UUV Simulator), where a dynamic model of FeelHippo AUV was implemented. Moreover, the ATR strategy, the RHCA framework, and the proposed view planning advances have been tested in real experimental campaigns at sea.

Sensor-driven autonomous inspections and CNN-based Automatic Target Recognition strategies towards underwater full autonomy / LEONARDO ZACCHINI. - (2022).

Sensor-driven autonomous inspections and CNN-based Automatic Target Recognition strategies towards underwater full autonomy

LEONARDO ZACCHINI
2022

Abstract

Autonomous Underwater Vehicles (AUVs) have become fundamental tools for marine scientists and industries to explore and monitor underwater areas. Non-predictable environmental conditions and sensor acquisitions make the design of AUV surveys challenging even for expert operators. Multiple attempts are required, and the collected data quality is not guaranteed: the AUV usually passively stores the sensors’ acquisitions that are then analyzed offline after its recovery by human operators in charge of identifying and localizing the so-called Objects of Potential Interest (OPIs). When it comes to acoustic images, the lack of features and low resolution make this task even more challenging. As a consequence of these statements, the marine community has sought robots able to meaningfully perceive and model the surroundings and autonomously conduct the assigned task, which are the foundations of fully autonomous vehicles. In this work, these topics have been investigated. To allow a compact and lightweight AUV to gather knowledge of the surroundings, an Automatic Target Recognition (ATR) strategy based on modern Convolutional Neural Networks (CNNs) for onboard online applications was developed. The ATR methodology was used to identify and localize potential targets of interest in Forward-Looking SONAR (FLS) imagery. Then, to avoid using pre-planned surveys and make an AUV actively considering the acquired data online, this thesis presents a probabilistic framework for FLS-driven seabed inspections. The realized sensor-driven Receding-Horizon Coverage Approach (RHCA) endows the AUV with the ability to autonomously conducting the survey and ensures adequate coverage of the target area. A Rapidly-exploring Random Tree (RRT) inspired view planning algorithm for underwater inspections was designed. Then, advancements for enhancing the performance of the algorithm have been carried out. In particular, a novel informed tree expansion methodology for guiding the vehicle towards the non-explored regions is proposed. Thanks to this solution based on the Kernel Density Estimation technique, the AUV learns the distribution of the discovered map. In addition, to explore other view planning strategies, a preliminary investigation about the exploitation of a Randomized Model Predictive Control (RMPC) approach for conducting autonomous seabed inspections is reported. Finally, the proposed ATR methodology and the RHCA have been combined to realize a target-aware planning solution for autonomously inspecting an area of interest. A probabilistic semantic map that includes the knowledge about the presence of the OPIs is created and updated by using the ATR findings. The semantic map enables the view planning algorithm to generate paths that cover the area of interest and simultaneously reduces the target localization uncertainty. Therefore, this methodology allows an AUV to meaningfully perceive and model the surroundings and autonomously conduct inspections surveys. The solutions proposed in this thesis have been firstly validated with realistic simulations made by means of the Unmanned Underwater Vehicle Simulator (UUV Simulator), where a dynamic model of FeelHippo AUV was implemented. Moreover, the ATR strategy, the RHCA framework, and the proposed view planning advances have been tested in real experimental campaigns at sea.
2022
Bendetto Allotta, Alessandro Ridolfi
LEONARDO ZACCHINI
File in questo prodotto:
File Dimensione Formato  
Phd_Thesis_ZacchiniLeonardo_compressed.pdf

accesso aperto

Descrizione: PhD thesis
Tipologia: Tesi di dottorato
Licenza: Open Access
Dimensione 1.71 MB
Formato Adobe PDF
1.71 MB Adobe PDF

I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/1264913
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact