Risk models may provide information about the risk associated with the operation of technical systems, and their output may be used as input to decision-making, e.g., by human operators. For highly autonomous systems, it is desirable to use risk models to enable the system itself to make risk-aware decisions. This impacts the requirements to the output of the risk model, as it must be directly useful for the decisions made by a control system. This study investigates how risk metrics can be used for supporting decisions by an autonomous surface vehicle (ASV) following an autonomous underwater vehicle. The feasibility of three different risk metrics is evaluated: expected economic loss, individual risk, and hazardous event probability. The metrics are evaluated in an analysis focused on ASV path planning. Both simulations, and experimental field trials with a real ASV in the Trondheimsfjord, Norway, were performed to investigate if and how the choice of risk metric affects the decisions made by the ASV. The results show the influence of each metric on the ASV’s planned path and illustrate the importance of considering the potential impact of the risk metric on safety when designing control algorithms and supervisory risk controllers for autonomous systems.

Operations of multiagent systems consisting of autonomous underwater vehicles (AUVs) and autonomous surface vessels (ASVs) offer a cost-effective solution to a wide range of marine applications, such as mapping and monitoring of the oceans. This article proposes a switching controller for tracking one or multiple AUVs with an ASV. Using three control modes, the controller ensures that: 1) each AUV eventually gets tracked and aided; 2) collisions with the AUVs are avoided; and 3) the ASV stops active propulsion whenever practical in order to conserve energy and reduce acoustic noise caused by the propulsion system. The switching of modes is based on the relative geometry between the ASV and the AUVs and backward reachable sets (BRSs). The switching controller is experimentally demonstrated with satisfying results in a series of field experiments in the Trondheim Fjord, where up to three AUVs executing missions simultaneously are tracked and aided by one ASV.

Risk awareness and assessment are fundamental aspects of human cognition and situational awareness, and play crucial roles in problem solving and decision-making. In this article, we present a novel methodology for integrated risk modeling and path planning in robotics mimicking these human processes. This approach creates a holistic geospatial data structure of risk, showing what may go wrong, where and when it is more likely, and the potential causes and consequences; all of which may be used as input to planning and decision-making algorithms for improved robotic autonomy. First, a hazard analysis of the operation is performed, with the objective of analyzing possible hazardous events, their causal factors, and potential consequences. This knowledge is then incorporated into a Bayesian belief network for estimating the risk at a particular point in space and time. Two methods for path planning taking these results as input are proposed: first, the risk-based path planner, and second, the risk-based traveling salesperson, both of which can balance the tradeoffs between risk and reward related to the mission objectives. We demonstrate the novel methodology with a real case study: seabed survey of the Tautra coral reef in Norway using an autonomous underwater vehicle (AUV), capitalizing on data from previous field operations. The case study shows that the AUV adapts its mission based on the perceived and assessed risk. By combining methods from robotics, artificial intelligence, risk science, and geoinformatics this work provides an interdisciplinary and novel contribution to enhanced robotic autonomy.

Optical imaging for identifying targets of interest is an important operational phase in many applications of autonomous underwater vehicles (AUVs). However, underwater optical imaging is challenging due to limited visibility and rugged terrain. To counter the low visibility, the AUV needs to fly at a low altitude over the targets. This results in a small optical footprint and corresponding small margins for error in maneuvering and navigation accuracy, in addition to a significant risk of collision with the seabed. In this paper, we propose a planning algorithm that adapts the mission to in-situ knowledge for safe and robust optical inspection of seafloor objects. By using automatic target recognition on data from a multibeam echosounder with a wider field of view, the algorithm checks whether the target was within the camera footprint. For targets that were deemed outside, i.e., missed, the plan is updated with the corrected object positions. This significantly increases the probability of successful optical inspection in a reduced amount of time, as well as reduces the risk of collision. The proposed method has been demonstrated in sea experiments using a HUGIN AUV.

Since 2017, NTNU’s Applied Underwater Robotics Laboratory has been developing an infrastructure for remote marine/subsea operations in Trondheim Fjord. The infrastructure, named the OceanLab subsea node, allows remote experimentation for three groups of assets: seabed infrastructure, surface or subsea vehicles/robots, and assets at remote experimentation sites. To achieve this task, a shoreside control room serves as a hub that enables efficient and diverse communication with assets in the field as well as with remote participants/operators. Remote experimentation has become more popular in recent years due to technological developments and convenience, the COVID-19 pandemic, and travel restrictions that were imposed. This situation has shown us that physical presence at the experimentation site is not necessarily the only option. Sharing of the infrastructure among different experts, which are geographically distributed, but participating in a single, local, real-time experiment, increases the level of expertise available and the efficiency of the operations. This paper also elaborates on the development of a virtual experimentation environment that includes simulators and digital twins of various marine vehicles, infrastructures, and the operational marine environment. By leveraging remote and virtual experimentation technologies, users and experts can achieve relevant results in a shorter time frame and at a reduced cost.

NTNU’s Applied Underwater Robotics Lab (AU-RLab) has been developing a full-scale subsea testing and validation facility in the Trondheim fjord since 2017. These activities are the result of NTNU’s long-term collaboration with Equinor that is related to the development of a test and validation site for underwater technology, including the autonomous/resident vehicles. Infrastructure has been further developed in the scope of Subsea node of the OceanLab project. This paper describes OceanLab infrastructure, docking station, instrumentation and subsea assets, and modes of remote access to the infrastructure. Different use-cases such as: docking, remote experimentation, subsea residency and testing and validation of Underwater Intervention Drones are briefly elaborated. The paper also presents work that has been done in the OceanLab within the various EU Horizon projects. The aim of developing OceanLab infrastructure is to contribute to the development of new technology for observing the ocean, both for academia and industry; and to be part of the overall measurement system in the Trondheim fjord, allowing better understanding of the interplay in the sea.

Since few or no human operators are directly involved in the operation of an autonomous marine system (AMS), an online risk model is necessary to enhance the intelligence of the AMS, its situation awareness, and decision making. The current study combines the system-theoretic process analysis (STPA) with Bayesian belief networks (BBNs) to develop online risk models for an AMS. Furthermore, fuzzy discretization is introduced to deal with evidence uncertainty. The proposed risk model can update the risk level as the operating conditions change, providing a basis for AMS supervisory risk control (SRC). A two-level SRC is proposed in this study. Using the operation of an autonomous underwater vehicle (AUV) under sea ice as an example, the current work presents an online risk model and a corresponding SRC system, focusing on the navigation hazards to the AUV and its potential loss. The results of simulation studies show that the model enables the AUV to be informed of the risk level and to make risk-based decisions accordingly, thereby improving its intelligence. The importance of the evidence uncertainty in online risk models and SRC is analyzed and discussed. The results and conclusions of this analysis can be adapted to other AMSs.

The spring season was the target for the Nansen Legacy cruise organized in late April and first half of May 2021 following the transect defined for this series of cruises to capture the variations of the year sampling physical, biological and chemical conditions in the ice and the sea. The transect went through both open water and ice. Seven process stations were visited (P1 through P7) together with smaller NLEG stations according to the program for the seasonal investigations. The first station (P1) was in open waters, while the remaining six main station had ice coverage of varying degree. Each of the process stations lasted 24 hours or more to allow a full diurnal cycle. Sampling included ice physics, ice samples, phytoplankton, zooplankton, marine chemistry and eco toxicology using acoustic, optical and robotics methods together with lab analyses of physical samples. Remote sensing data were also matched with in situ observations of both sea and ice conditions.

Autonomous underwater vehicles (AUVs) rely on surface support for communication with operators and position fixes to bound inertial navigation errors. By installing an acoustic modem on an autonomous surface vehicle (ASV), the ASV can carry out these tasks, replacing more expensive and less flexible manned research vessels. This paper proposes a hybrid tracking controller for an ASV providing mission support for an AUV. The proposed controller keeps the ASV in a donut-shaped safety domain about the AUV defined by the risk of collision (inner boundary) and the risk of communication loss (outer boundary). At the same time, the hybrid controller reduces power consumption and acoustic signal noise by going into standby mode when it is within the safety domain. Results from a simulation study and field trials are presented to demonstrate and validate the controller's performance. The results show that the controller performed well in the tested cases.

Autonomous underwater vehicles (AUVs) are efficient sensor-carrying platforms for mapping and monitoring undersea ice. However, under-ice operations impose demanding requirements to the system, as it must deal with uncertain and unstructured environments, harsh environmental conditions, and reduced capabilities of the navigational sensors. This paper proposes a Bayesian approach to supervisory risk control, with the objective of providing risk management capabilities to the control system. First, an altitude guidance law for following a contour of an ice surface via pitch control using measurements from a Doppler velocity log (DVL) is proposed. Furthermore, a Bayesian network (BN) for probabilistic reasoning over the current state of risk during the operation is developed. This is then extended to a decision network (DN) for autonomously adapting the behavior of the AUV in order to maximize the mission utility, subject to a constraint on the predicted risk from the risk model. The vehicle is thus able to autonomously adapt its behavior in response to its current belief about the risk. The goal of this work is to improve the AUV performance and likelihood of mission success. Results from a simulation study are presented in order to demonstrate the performance of the proposed method.

This paper presents a framework for optimization-based informative planning and control with applications to adaptive sampling with AUVs under sea ice. A spatial model of the information of interest is approximated as a Gaussian process (GP), which is learned online from in-situ sensor data. The planner uses a two-layer model predictive control (MPC) scheme on a low-fidelity model of the vehicle for exploration and exploitation of the GP, subject to safety constraints. The planner trajectories are then tracked using a constant bearing based guidance law, aligning the desired orientation of the AUV toward the planned trajectory. The proposed framework enables the vehicle to plan and replan its mission as new data is obtained, while ensuring tracking of the planned trajectories and safety constraint satisfaction. Simulation results of a case study are presented for demonstrating the performance of the proposed method. An AUV is tasked with finding and tracking concentrations of marine biomass in 3D under sea ice while avoiding collisions.

Accurate underwater navigation systems are required for closed-loop guidance and control of unmanned underwater vehicles (UUV). This paper proposes a sensor-based hybrid translational observer concept for underwater navigation using the hybrid dynamical systems framework, accounting for noisy, asynchronous and sporadic sensor measurements. Sensor measurements from an acoustic positioning system, a Doppler Velocity Log (DVL), an Inertial Measurement Unit (IMU) and a pressure gauge are used in the proposed observer. A method for filtering high-frequency noise is proposed, where the estimated states are obtained by taking a weighted discounted average of a finite number of previous measurements predicted forwards to the current time. The attitude of the vehicle is assumed known, and the acceleration measurements are assumed to be continuously available. Measurements of position, depth and linear velocity are assumed to be asynchronous and sporadically available, that is, they do not arrive at the same time, and their sampling rates are not constant. Uniform global asymptotic stability (UGAS) is established using Lyapunov theory for hybrid systems. Results from simulations are presented in order to demonstrate the performance of the proposed method.

Accurate underwater navigation systems are required for closed-loop guidance and control of unmanned underwater vehicles (UUV). This paper proposes a method for the design of a six degrees-of-freedom (DOF) sensor-based hybrid translational observer concept for underwater navigation using the hybrid dynamical systems framework, accounting for noisy, asynchronous and sporadically available sensor measurements. Sensor measurements from an acoustic positioning system, a Doppler Velocity Log (DVL), an Inertial Measurement Unit (IMU) and a pressure gauge are used in the proposed observer. A method for filtering high-frequency noise is proposed, where the estimated states are obtained by taking a weighted discounted average of a finite number of previous measurements predicted forwards to the current time. The attitude of the vehicle is assumed known, and acceleration measurements are assumed to be continuously available. Measurements of position, depth and linear velocity are assumed to be available sporadically with asynchronous sampling rates. Results from simulations are presented to demonstrate the performance of the proposed method.

Autonomous underwater vehicles (AUVs) are effective platforms for mapping and monitoring under the sea ice. However, under-ice operations impose demanding requirements to the system, as it must deal with uncertain and unstructured environments, harsh environmental conditions and reduced capabilities of the navigational sensors. This paper proposes a method for intelligent risk-based under-ice altitude control for AUVs. Firstly, an altitude guidance law for following a contour of an ice surface via pitch control using measurements from a Doppler velocity log (DVL) is proposed. Furthermore, a Bayesian network for probabilistic reasoning over the current state of risk during the operation is developed. This network is then extended to a decision network for autonomous risk-based selection and reselection of the setpoint for the altitude controller, balancing the trade-off between the reward of the setpoint and the risk involved. This will improve the system safety and reliability. Results from a simulation study are presented in order to demonstrate the performance of the proposed method.

This Master thesis investigates tools and methods for developing more robust autonomy solutions for autonomous underwater vehicles (AUV) in Arctic under-ice operations. AUVs are effective platforms for multi-disciplinary scientific research in the polar oceans. However, operations of AUVs in these areas involve a substantial risk of losing the vehicle. The environment under the sea ice is unstructured and unknown, and is further complicated by harsh environmental conditions and reduced capabilities of navigation sensors. Under the sea ice AUVs are reliant on an acoustic positioning system, as they are not able to surface for position fixes from a global positioning system (GPS). Position measurements from acoustic positioning systems are noisy and may drop out temporarily. Other navigation sensors, such as pressure gauges and Doppler velocity logs (DVL) may also temporarily drop out, and the signals are obtained at an asynchronous sampling rate. Given the nature of asynchronicity and sporadic availability of the navigational signals, the framework of hybrid dynamical systems is used to design and analyze a sensor-based hybrid observer. A method for the design of a sensor-based hybrid translational observer for underwater navigation is proposed, accounting for noisy, asynchronous and sporadically available sensor measurements. Here, acceleration measurements are assumed readily available. Position and velocity estimates are updated discretely, asynchronous and sporadically as new measurements are obtained. Between measurements, the estimates are continuously predicted forwards in time by integration of kinematic relationships. High-frequency noisy is filtered by taking a weighted discounted average of a finite number of previous states predicted forwards to the current time. Simulations of a six degree of freedom observer which relies on acceleration, velocity and position measurements are conducted. There is often a trade-off between the scientific reward of a mission plan and the risk involved. For instance, optical mapping of algae and phytoplankton closer to the surface yields higher quality of the acquired data, however, also increases the risk of losing the vehicle under the ice. By including an online risk model, the vehicle may take calculated risks by weighing the potential rewards of an action and the risk involved. A method for intelligent risk-based under-ice altitude control for AUVs is presented. An altitude guidance law for following a contour of the ice surface via pitch control using measurements from a Doppler velocity log (DVL) is proposed. Furthermore, an online risk model for probabilistic reasoning of the risk of vehicle loss is developed using the framework of Bayesian networks. This network is extended to a decision network for online autonomous risk-based selection and reselection of the setpoint for the altitude controller, and if necessary, send a signal to abort the mission. This will improve safety and robustness of under-ice operations for AUVs.