Unmanned Systems Technology 042 | Mayflower Autonomous Ship | Embedded Computing | ElevonX Sierra VTOL | UUVs insight | Flygas Engineering GAS418S | Ocean Business 2021 report | Electric motors | Priva Kompano

26 Dossier | Mayflower Autonomous Ship autonomous marine vehicles) and Oliver Thompson, senior engineer at MSubs, as well as several other key team members, who have been indispensable to the development process, which he himself has overseen as director since its inception. AI on the Atlantic Phaneuf contacted Scott immediately after his meeting with Plymouth Council, and the two began brainstorming how the architecture of such a large vessel would look when no space or provisions for crew were needed – including interiors, hull forms and morphologies. “We’d built unmanned systems before, so we were pretty well-versed when it came to ship architecture and system integration,” Phaneuf says. “What we didn’t know so well and really wanted was autonomy, not just automation and traffic management but a ship that could genuinely think for itself and keep itself going. “We knew developing the necessary intelligence would take a long time, that the AI would need models, and the models would need data. We didn’t want to rely on other people’s data to train our neural networks, because that couldn’t be relied upon to help our ship work in its own unique environments and situations. As the old saying goes, if you put garbage in, you’ll get garbage out.” To gather imagery for pairing with their convolutional neural networks (CNNs) and hence training their computer vision systems, the team first built a number of stationary sensors – including cameras, radars, anemometers, AIS units and more – around Plymouth Sound, a deep inlet off the coast of the town. These amassed photos, videos and other data of offshore activities on beaches, offshore structures and jetties, as well as onshore locations and from moving vessels. “We’ve collected terabytes and terabytes of data over the past four years, and through our partnerships we used that ‘lake’ of data to make useful models, maybe the first of which revolved around categorisation of things visible in our data,” Phaneuf explains. “You can’t have a model of literally every ship, but we can have a model for every kind of ship – that’s what we aim to do, and the ship classification works great now. We’re continuing to refine and optimise it too, validating our recognition and auto-labelling tools in missions at sea. We ingest new data, tag it, produce the model, stress the model, refine it, add more data, and repeat. We don’t expect to stop improving our models.” Around the same time that the Mayflower team began collecting data, they also started designing the ship, looking at different concepts and consulting numerous naval architects while also raising funds. The hull form was settled and ordered in late 2019, with its shape and volume a compromise between affordability and what Phaneuf and Scott knew from experience to be necessary in terms of space for subsystems used to run a large unmanned vehicle. “We designed it to have the right volumes for componentry, and worked with a naval architect to validate the design before passing it to the mechanical engineering team, who actually continued and finalised the design while they were cutting the metal to build it,” Phaneuf recalls. The hull was delivered by Polish company Aluship some 9 months before its first test voyage, which was due in autumn 2020. Before that, the Mayflower’s engineers worked to integrate all the subsystems that had been procured, optimised or developed off-board for power, processing, navigation, comms and so on. “Quite a bit was also donated by suppliers who saw the value of our project early on. For example, Veripos gave us the GNSS equipment, iXblue gave us our AHRS, and Collins gave us numerous MEMS gyro systems,” Phaneuf adds. “And of course IBM gave us the technological platform that our AI and machine learning stand on, not to mention an incredible technical reachback to their huge team of experts, as well as building the front end of our web portal. That’s done wonders for spreading awareness and appreciation for our project: IBM estimate that at least 1.7 billion people have heard of the MAS400.” As with so many other recent projects, the MAS400’s first trial voyage was postponed until 2021 owing to the Covid-19 pandemic. When it finally took place, this first attempt at autonomously crossing the Atlantic failed because of a mechanical failure (not a software failure, as some people believed). Although this outcome might be surprising to the uninitiated, anyone familiar with the Microtransat Challenge – an annual competition to cross the Atlantic autonomously by USV – will know that February/March 2022 | Unmanned Systems Technology The ship’s interior features numerous power systems, computers and spaces dedicated to the installation of up to 1000 kg of research payloads