Autonomous subsea intervention (SEAVENTION)

There is potential for further value creation in the subsea petroleum industry, but there are significant challenges to be met. Subsea processing facilities are being developed for more distant, deeper and colder areas, and the industry projects that the next generation of autonomous Unmanned Underwater Vehicles (UUVs) will be available next to such subsea installations as resident underwater vehicles. A right level of autonomy for such UUVs can increase efficiency and safety of operations, and such a level may vary between and during operations. Hence, full autonomy / operator independence is in many cases not a goal it in self. Instead, the goal is to solve the right tasks in the right way. The goal of the SEAVENTION (Autonomous subSEA interVENTION) project is to deliver research and competence building on key areas for enabling UUVs to understand their environment and operate at a right level of autonomy while keeping the operator in the loop. This will provide a fundament for increased efficiency and safety of subsea operations. To achieve this, the goal was to use subsea 3D sensors (sensors that can give 3-dimentional data about the UUVs' environment) and develop methods for UUVs' to utilize this data to understand their environment. Moreover, we aimed to develop techniques for automated planning such that UUVs' can plan operations by themselves based on their understanding of their environment. Also, we targeted to investigate the usage of augmented reality for UUV operator decision support. With augmented reality it is possible to merge video and images from the real world with virtual images (e.g., we could display the status of a pump on top of an image of the pump). With the competence building and technology in SEAVENTION we set out to meet the industry's need for more cost-efficient subsea solutions, increased competitiveness of the Norwegian petroleum-related industry, increased safety, and competence building adapted to the needs of the industry. Machine learning and classical control: We have researched machine learning and classical control methods for UUV control for station-keeping and intervention. Moreover, we have also researched how to combine machine vision and control. To ensure safe UUV movement, we have proposed a method for collision avoidance and optimal path-replanning around obstacles. Moreover, we have developed methods for autonomous underwater grasping using a novel vision-based distance estimator. Automated planning and acting: In order for autonomous UUVs to carry out missions efficiently, there is a need for high-level task planning and coordination. To this end, we have developed a mission planning for a system of cooperative UUVs tested with experiments in the Trondheim fjord. We have also used two planning frameworks for automated planning and acting - also called AI planning - to implement AI planning for a subsea inspection scenario analyzed and formulated in collaboration with the industry partners. Underwater perception: UUVs benefit from perception to understand and interact with its environment. However, there are many challenges to underwater perception such as turbidity (i.e., "amount of dirt") in the water which causes degraded visibility and reflections. To achieve robust 6D pose estimation of objects that the UUVs could do intervention operations on, we have developed and tested marker-less and marker-based approaches. We have carried out controlled experiments with varying turbidity to evaluate the proposed systems provide full 6 degrees-of-free pose estimates in challenging conditions. To facilitate the training of machine learning-based perception systems, we have also implemented an approach to collect and automatically annotate underwater 6D pose estimation datasets. Autocalibration: Calibration quality of sensors onboard UUVs is directly tied to its precision of navigation. To this end, we have developed an algorithm for calibrating the relative pose and time delay between sensors on a robotic vehicle. Augmented reality for operator support: Augmented Reality (AR) can increase the situational awareness of human operators and support keeping them in the loop during operations with UUVs operating with varying levels of autonomy. To this end, we have designed and implemented several concepts for how to use AR in intervention operations. We have also evaluated the proposed approaches with mock-ups according to a System Usability Scale (SUS). Equinor, TechnipFMC, Oceaneering, IKM Technology, NTNU and SINTEF are partners in the project. The project is led by SINTEF. Project website: https://www.sintef.no/seavention

The background for the SEAVENTION project was a need in the oil and gas industry for higher level of autonomy in subsea inspection, maintenance and repair operations with unmanned underwater vehicles (UUVs). To meet this need, we have developed competence and methods for autonomous intervention and widely disseminated results. Thus, the results are in the public domain and may affect implementation of future industrial autonomous UUVs. The proposed methods are also applicable for UUVs in renewable energy (e.g., offshore wind), aquaculture (e.g., fish net inspection) and fisheries. A further effect of the project is an increased preparedness in the industry for uptake of autonomous UUV technologies. Moreover, the R&D partners can further steer research and teaching even better toward current and future industry needs. Manned surface vehicles involved in UUV operations can give high costs and environment footprint. The use of autonomous UUVs can reduce the use of such vehicles.

SEAVENTION addresses the goals of the PETROMAKS2 programme by researching and developing competence, novel methods, algorithms and technology for autonomous subsea intervention. The project will develop technology for advanced perception and planning with deep learning capabilities that enables Unmanned Underwater Vehicles (UUV) to perform light intervention tasks semi- or fully autonomously (e.g., cleaning, valve manipulation), while ensuring that operators are kept in the loop. To this end, SEAVENTION will deliver research and competence building on robust subsea perception with active 3D sensors, automated planning for UUV intervention procedures and augmented reality for UUV operator decision support. This is to meet the subsea industry's need for 1) more cost-efficient subsea solutions (autonomous UUVs, reduced need for surface support vessels), 2) increased competitiveness of Norwegian petroleum-related industry (enable them to offer the solutions of the future), 3) increased safety (less personnel offshore with autonomous UUVs, safe UUV operations, less dependence on highly skilled/experienced operators through new decision support systems) and 4) competence-building adapted to the needs of the industry. The main research challenges encompass 1) how to achieve robust 3D object detection and localization subsea, 2) how to merge high-level AI with low-level control for robust subsea intervention, 3) how to achieve dynamic UUV motion planning and decision making under uncertainty, and 4) how to achieve augmented reality for shared intervention control - in order to ensure the right level of autonomy and operator involvement in different subsea operations. Hence full autonomy / operator independence is in many cases not a goal it in self - the goal is to solve the right tasks in the right way. Project results will be tested and demonstrated in state-of-the-art lab facilities and we target also to do sea trials and demonstrations.