Retrieving Autonomous Underwater Vehicles (AUVs) using Unmanned Aerial Vehicles (UAVs) presents significant challenges due to wave-induced motion and wind disturbances during recovery. While prior research has primarily addressed air-to-air refueling and ground-to-air deployments, this work introduces an eagle-inspired UAV trajectory generation and control system specifically designed for AUV retrieval. Drawing inspiration from avian flight dynamics, the proposed approach mitigates the effects of downward airflow and environmental disturbances, enabling a stable and efficient recovery process. We propose a novel trajectory planning method that minimizes snap, integrates a cost function to account for the UAV’s airflow effects on the target, and adapts dynamically to wave-induced movements. A specialized connection mechanism, consisting of a hook-equipped UAV and a buoy-rope assembly attached to the AUV, is developed and validated through reduced-scale in-water field experiments. Post-connection, a suspended load controller ensures stability by compensating for shifts in inertia and external forces. Additionally, we present a Unity-based simulation environment that allows customization of vehicle parameters and mission scenarios. This work bridges the gap in maritime operations, offering a reliable and flexible solution for AUV recovery in dynamic marine environments.

Developing new functionality for underwater robots and testing them in the real world is time-consuming and resource-intensive. Simulation environments allow for rapid testing before field deployment. However, existing tools lack certain functionality for use cases in our project: i) developing learning-based methods for underwater vehicles; ii) creating teams of autonomous underwater, surface, and aerial vehicles; iii) integrating the simulation with mission planning for field experiments. A holistic solution to these problems presents great potential for bringing novel functionality into the underwater domain. In this paper we present SMaRCSim, a set of simulation packages that we have developed to help us address these issues.

Recent maritime incidents highlight the need for scalable underwater monitoring solutions. Traditional methods of ocean exploration face high costs and limited access, while collaborative teams of Autonomous Underwater Vehicle (AUV)s offer a promising alternative. However, underwater localization and multi-agent coordination remain challenging due to the lack of Global Navigation Satellite System (GNSS) signals underwater, the high cost of equipping all agents with advanced localization sensors, the limited acoustic communication bandwidth, and the susceptibility of acoustic communication to environmental interference. This work presents a hybrid decentralized framework for robust localization and formation control in teams of AUVs. Two leader agents equipped with GNSS or an advanced Inertial Navigation System (INS) guide a fleet consisting of an arbitrary number of followers in a circular arc formation using acoustic ranging for AUV localization. The framework is validated in a simulated environment and through field trials, demonstrating reliable acoustic localization and formation keeping. The results confirm the viability of the approach for scalable, high-precision underwater monitoring and event response.

Our planet is mostly covered in water, most of it still unexplored.In order to understand our environment better, oceanographers have been mapping and monitoring these waters using ship-mounted sensors and wired vehicles with limited range compared to the vastness of the oceans.The limited range and dependence on manned support vehicles has kept missions expensive and infrequent.To solve this problem, the sensors need to become independent of support vehicles, they need to venture into completely unexplored, unmapped regions of the seas by themselves and safely return with the data.This is where autonomous underwater vehicles (AUVs) have started to make a difference.In this thesis we investigate how multiple small AUVs can be utilized to efficiently and accurately sense very large volumes of water.

Water absorbs electromagnetic radiation, meaning satellite-based global positioning systems (we will use GPS to refer to any such system), wide-angle cameras and radio communications are infeasible.These constraints ultimately result in uncertain localization of  the vehicles.Furthermore, the vehicles are under constant disturbances from the water currents, fish and bio-fouling, which result in the dynamics of the vehicles being uncertain or even changing during the mission.

In the first part of this thesis, we focus on the large-scale sensing problem under localization uncertainties by examining the caging and coverage problems.In the coverage problem, each AUV is uncertain about its exact position while tasked with sensing a stationary area.We show that we can still guarantee complete coverage and formulate the efficiency characteristics of different approaches.Furthermore, we show that when the vehicles are equipped with sensors and low-bandwidth communication methods, we can increase the effective range of a team of AUVs considerably by utilizing loop-closures over shared pose-graphs. In the caging problem, the localization uncertainty is focused on the entity that is being caged, its location is unknown but bounded.We show that through a combination of algorithms, the caging problem can be solved and a solution can be guaranteed, while simultaneously producing a list of specifications for the mission.

In the second part, we focus on the individuals of the team and what they need to do in order for the team of AUVs to succeed.First, we identify that when there is a team of cooperative vehicles working together, conflicting goals rise.Each vehicle needs to pick between satisfying its own constraints and the constraints that come from being in a team. We propose a solution to this problem through a combination of Control Barrier Function (CBF) and Behavior Trees (BT).Secondly, we examine the possibility that a vehicle might undergo physical changes, like a broken thruster, that result in the vehicle being unable to complete the entire mission.Even in such a scenario, if the broken vehicle can still move to contact a normal one, the rest of the team can compensate through re-planning and the overall mission can still be completed.To do so, the broken vehicle must compensate for the change until a rendezvous.We propose a data-driven pipeline that can detect and plan around such a physical change within some bounds.

Vår planet är till största delen täckt av sjöar och hav, och en stor del av dessa är fortfarande outforskade. För att bättre förstå vår omvärld har oceanografer undersökt sjöar och hav med sensorer som antingen varit fast monterade på stora bemannade fartyg, eller på undervattensfarkoster som styrtsvia kablar från sådana fartyg. Detta beroende av fartyg har gjort uppdragen dyra och därför även få. För att lösa detta problem måste sensorplattformarna göras oberoende av kablar och bemannade fartyg, och ges förmåga att på egen hand bege sig in i outforskade områden och sedan komma tillbaka igen med ny information. Sådana sensorplattformar kallas AUV, Autonomous Underwater Vehicles, och i denna avhandling undersöker vi hur en grupp AUV:erkan användas för att söka av stora vattenvolymer på egen hand. Undervattensdomänen är förknippad med ett antal unika problem. Vatten absorberar elektromagnetisk strålning, vilket gör satellitnavigering (t.ex.GPS) och radiokommunikation oanvändbart i praktiken, vilket i sin tur leder till att det är svårt att bestämma sin egen position under vattnet. Vidare gör strömmar, vattendjur och s.k. biofouling (att t.ex. alger och smådjur växer där man inte vill ha dem) att en farkosts dynamik kan ändras över tid i vattnet.

I denna avhandlings första del studerar vi storskalig övervakning under positionsosäkerheter i form av instängnings- (caging) och yttäcknings-problem(coverage). I yttäckningsproblemet skall UAV:n söka av en given yta, samtidigt som UAV:n är osäker på sin egen position. Vi visar att det trots detta är möjligt att garantera total täckning av ytan, och beskriver hur effektiva olika sökmönster är för denna uppgift. Vidare visar vi hur en grupp av UAV:er utrustade med sensorer och akustiska modem med låg bandbredd kan lösadenna uppgift mer effektivt än enskilda UAV:er, genom s.k. loop closures och delande av tillståndsgrafer. I instängningsproblemet gäller positionsosäkerheten istället en inkräktare, t.ex. en fientlig ubåt, vars position är känd på ett ungefär, men inte exakt. Vi visar att man kan bestämma positionen exakt genom en kombination av algoritmer som först stänger in inkräktaren och sedan gradvis krymper den volym den kan uppehålla sig i. Vi beräknar också vilka prestanda systemet måste uppfylla för att kunna garantera instängning. I avhandlingens andra del fokuserar vi på de enskilda AUV:erna, och vad de måste göra för att hela gruppen skall lyckas med sitt uppdrag. Först noterar vi att när en grupp samarbetar så kommer det att uppstå målkonflikter. Vid vissa tidpunkter kommer individer att tvingas välja mellan egna mål, som t.ex. att inte få slut på batteri, och gruppens mål, som t.ex. att täcka av ett område. Vi föreslår en lösning till detta problem som bygger på kombinationen av s.k. Control Barrier Functions (CBF) och Beteendeträd (BT). Sedan undersöker vi fallet då en AUV påverkas av förändringar, så som en skadad propeller, som gör att den inte kan fullgöra den ursprungliga uppgiften. I ett sådant scenario vill man att den skadade farkosten skall försöka ta sig till en punkt där den kan kontakta resten av gruppen, så att de kan kompensera bortfallet. Vi föreslår en data-driven metod för som kan upptäcka och hantera vissa typer av sådana fel.

This paper outlines challenges and opportunities in operating underwater robots (so-called AUVs) on a seaweed farm. The need is driven by an emerging aquaculture industry on the Swedish west coast where large-scale seaweed farms are being developed. In this paper, the operational challenges are described and key technologies in using autonomous systems as a core part of the operation are developed and demonstrated. The paper presents a system and methods for operating an AUV in the seaweed farm, including initial localization of the farm based on a prior estimate and dead-reckoning navigation, and the subsequent scanning of the entire farm. Critical data from sidescan sonars for algorithm development are collected from real environments at a test site in the ocean, and the results are demonstrated in a simulated seaweed farm setup.

Multi agent coverage and robot navigation are two very important research fields within robotics. However, their intersection has received limited attention. In multi agent coverage, perfect navigation is often assumed, and in robot navigation,  the focus is often to minimize the localization error with the aid of stationary features from the environment.The need for integration of the two becomes clear in environments with very sparse features or landmarks, for example when a group of Autonomous Underwater Vehicles (AUVs) are to search a uniform seafloor for mines or other dangerous objects.In such environments, localization systems are often deprived of detectable features to use that could increase their accuracy.In this paper we propose an algorithm for doing navigation aware multi agent coverage in areas with no landmarks.Instead of using identical lawn mower patterns, we propose to mirror every other pattern to enable the agents to meet up and makeinter-agent measurements and share information regularly. This improves performance in two ways,global drift in relation to the area to be covered is reduced, and local coverage gaps between adjacent patterns are reduced.Further, we show that this can be accomplished within the constraints of very limited sensing, computing and communication resources that most AUVs have available.The effectiveness of our method is shown through statistically significant simulated experiments.

Underwater robotic exploration missions typically involve traveling long distances without any human contact.The robots that go on such missions risk getting damaged by the unknown environment, accruing great costs and missed opportunities.Thus it is important for the robot to be able to accommodate unknown changes to its dynamics as much as possible and attempt to finish the given mission, or at the very least move itself to a retrievable position.

In this paper, we show how we can detect physical changes to the robot reliably (79\% on real robot data) and then incorporate these changes through adapting the model to the data followed by automated control redesign. We adopt a piecewise-affine (PWA) modelling of the dynamics that is well suited for low data regime learning of the dynamics and provides a structure for computationally efficient control synthesis.We demonstrate the effectiveness of the proposed method on a combination of real robot data and simulated scenarios.

The number of mobile robots with constrained computing resources that need to execute complex machine learning models has been increasing during the past decade. Commonly, these robots rely on edge infrastructure accessible over wireless communication to execute heavy computational complex tasks. However, the edge might become unavailable and, consequently, oblige the execution of the tasks on the robot. This work focuses on making it possible to execute the tasks on the robots by reducing the complexity and the total number of parameters of pre-trained computer vision models. This is achieved by using model compression techniques such as Pruning and Knowledge Distillation. These compression techniques have strong theoretical and practical foundations, but their combined usage has not been widely explored in the literature. Therefore, this work especially focuses on investigating the effects of combining these two compression techniques. The results of this work reveal that up to 90% of the total number of parameters of a computer vision model can be removed without any considerable reduction in the model's accuracy.

Area coverage and robot navigation are two  important research fields within robotics. However, their intersection has received limited attention. In coverage problems, perfect navigation is often assumed, and in robot navigation, the focus is often to minimize the localization error while traveling a given trajectory.The need for integration of the two becomes clear in environments with very sparse features or landmarks, for example when an Autonomous Underwater Vehicle (AUV) is to search the seafloor for dangerous objects, such as mines.The potential consequences of missing a mine due to navigation errors can be catastrophic.If the localization error is large, a trajectory that was designed to guarantee complete coverage might have missed significant parts of the area. Thus, the coverage trajectory must be planned with the navigation performance in mind, applying a combination of using large enough planned overlaps of sensor footprints to account for the position uncertainty, and reducing this uncertainty by re-visiting the known sparse landmarks.

In this paper we compute trajectories that guarantee coverage for a given area under assumptions on worst case localization error growth.We further more compute upper bounds for how large areas can be covered using common coverage patterns and a single landmark, which leads to bounds on how sparse the landmarks can be in the regions to be covered.

Cyber-physical systems (CPSs) comprise a network of sensors and actuators that are integrated with a computing and communication core. Hydrobatic Autonomous Underwater Vehicles (AUVs) can be efficient and agile, offering new use cases in ocean production, environmental sensing and security. In this paper, a CPS concept for hydrobatic AUVs is validated in real-world field trials with the hydrobatic AUV SAM developed at the Swedish Maritime Robotics Center (SMaRC). We present system integration of hardware systems, software subsystems for mission planning using Neptus, mission execution using behavior trees, flight and trim control, navigation and dead reckoning. Together with the software systems, we show simulation environments in Simulink and Stonefish for virtual validation of the entire CPS. Extensive field validation of the different components of the CPS has been performed. Results of a field demonstration scenario involving the search and inspection of a submerged Mini Cooper using payload cameras on SAM in the Baltic Sea are presented. The full system including the mission planning interface, behavior tree, controllers, dead-reckoning and object detection algorithm is validated. The submerged target is successfully detected both in simulation and reality, and simulation tools show tight integration with target hardware.