As the presence of Artificial Intelligence in our lives increases, Marilyn Mower explores a new era for sailing and the possibilities of autonomous yachts...
In April, a small aluminium boat will leave Plymouth, sailing west toward the New World. Its name is Mayflower Autonomous Ship (MAS) and it may succeed in having at least as significant an impact on mankind as its namesake did 400 years ago, but by entirely different means. While the first Mayflower’s voyage of discovery brought colonists across the Atlantic, forever changing the landscape and populace of a continent, this one sails unmanned, making its own navigating decisions and conducting research autonomously. When MAS crosses the Atlantic, it will make history as one of the most comprehensive applications of artificial intelligence (AI) yet, not only offering scientists a less expensive way to gather information, but altering the definition of “captain” in the process.
The Mayflower Autonomous Ship (MAS) may succeed in having at least as significant an impact on mankind as its namesake did 400 years ago, but by entirely different means.
Image Credits: IBM
MAS is a 15-metre solar-powered trimaran, built by Aluship in Gdańsk, Poland, and outfitted in Plymouth under the direction of ProMare, a non-profit marine research and exploration corporation founded in 2001. An amalgamation of the latest technology and staggering algorithms with two years of intensive machine learning puts the necessary capabilities for truly autonomous ship operation in one package. Make no mistake: this is not a robot or a drone.
The MAS400 project’s mission is to cruise non-stop to Plymouth, Massachusetts, in the most efficient and safest way possible without human interaction. It has six cameras, radar, sonar, AIS, GPS navigation programs and weather monitoring feeding 15 onboard edge computers (in which the data is processed by the device itself, rather than being transmitted to a data centre). Its captain, navigator, bosun and engineer are a rack of machines that will stand watch, determine course and speed, monitor power and conduct experiments.
MAS is scheduled to set sail from Plymouth Harbour on April 19, 2021.
Image credits: Tom Barnes for IBM
Designed to address the problem that marine science data is often difficult and expensive to obtain, MAS will conduct water quality sampling and listen for whales throughout its voyage. Its onboard computing power will monitor the electricity being stored or used in lithium-ion phosphate batteries from solar panels and, if necessary, turn on a small generator to keep the boat underway. It has been trained to recognise and track obstacles and to avoid them with manoeuvres consistent with COLREGs and navigation rules and its own capabilities as impacted by sea and weather conditions. Complex algorithms and decision trees weigh regulations and protocols against current and predicted weather and sea state, vessel operating status, obstacles and hazards. It will take about two weeks for MAS to complete its crossing, working 24/7, and during that time no human will have to tell it what to do next.
The MAS400 project began with a little bit of that indispensable ingredient – serendipity. Brett Phaneuf, co-founder of the non-profit research group ProMare and also head of the Plymouth-based submersible company M Subs, attended a meeting of the Plymouth Council in 2016 discussing how to recognise the 400th anniversary of the famous 1620 Mayflower voyage. There were proponents of sailing a replica to America, but Phaneuf piqued imaginations by saying the voyage should be about the future. He proposed a new kind of craft, solar powered and captained by AI; one that could conduct ocean research as it sailed to commemorate the voyage. The Council, community and several UK universities got behind the project, even though all the technology to make it happen wasn’t available at that point. The involvement of IBM provided the crucial link. While the development of the project and the design of the little ship by Rachel Nichols Lee at Whiskerstay in Falmouth and M Subs is fascinating – as is its ability to collect ocean data for the United Nation’s Decade of Ocean Science for Sustainable Development – our focus is on what MAS means for future yacht design and operations.
Brett Phaneuf, co-director of the Mayflower project.
Image Credits: Tom Barnes for IBM
Unmanned boats on the horizon
In commercial shipping, the idea of autonomous ships is huge, driven by the carrot of reduced costs. Consider the savings by eliminating crew costs, plus the accommodation space freed up for paying cargo.
“Taking the human factor out of MAS has allowed us to completely reimagine the boat’s design,” says Phaneuf. “Instead of thinking about eating, sleeping and sanitation, the design engineers were able to focus purely on the mechanics and function of the ship.”
The main hull of MAS under construction.
Image Credits: IBM
Of course, it also allows a greener ship with no hotel load. “The diesel generator is just 20kW and the boat can carry two tonnes of fuel if filled completely, but based on our CFD and trials, it won’t need it to cross the ocean,” says ProMare director Ayse Atauz Phaneuf. “In bright sun, it can operate nearly indefinitely on solar alone, albeit at reduced speed.”
And it will be a safer ship. Imagine if the 75 per cent of sea collisions caused by operator error could be eliminated. While radar, lidar (light detection and ranging) and AIS systems already exist, they tend to be merely alerts and aren’t integrated into a vessel’s command network or tied into other data such as course and speed, weather conditions and economy of operation.
Deck plans of the 14.9 metre MAS.
Image credits: Promare
First introduced at BOAT International’s Life Under Sail conference in November 2019, OSCAR is a collision-avoidance system of night vision and colour cameras and an artificial-intelligence processor that can tie into a vessel’s autopilot (or throttles, too, in a motor yacht). Invented by Raphaël Biancale, it spots floating, stationary or semi-submerged objects, uses AI to assess their potential danger, and can alert crew or even grab the helm to avoid collision.
AI is the ultimate “auto updater”, says Patrick Haebig of BSB Artificial Intelligence. “OSCAR never stops learning.” Learning consists of showing OSCAR a picture and telling what is important in that picture. For example, OSCAR has learned that buoys are likely to be stationary objects and should not be run into. “It took 7,000 images to teach OSCAR what a buoy is [considering they come in so many shapes, sizes and colours].” It must identify one in a split second and access the algorithm that goes with “buoy” to notify the skipper or initiate the evasion system if it is in the vessel’s path.
Raphaël Biancale, inventor of OSCAR.
Image Credits: Laurent Vilboux
“We are now up to 55 million images [from waves to birds, whales, ships, rafts and logs to rocks, beaches and breakwaters] in the visual recognition database,” continues Haebig. “We are comfortable that OSCAR is 90 per cent accurate at identifying things that could collide with a yacht.” The ultimate trial is happening right now with installations on 18 of the 32 IMOCA 60s currently slamming around the world in the Vendée Globe. One boat, Malizia, is testing the system tied into the autopilot.
MAS takes this collision avoidance to an automated extreme with dual-redundant AIS, Veripos GNSS, dual K-band RADAR from Wärtsilä and many camera systems. This data will be fed into the operating scenario and analysed by the onboard AI, which uses a “rules engine” to determine the COLREG-compliant options to navigate. Unlike that villainous computer HAL in 2001: A Space Odyssey, MAS transparently records all her decisions, and in the case of a course change required on encountering another vessel, broadcasts her intentions and new heading via VHS radio.
Malizia (in the foreground) is one of 18 IMOCA 60s trialling OSCAR in the Vendée Globe.
Image Credits: Yvan Zedda
All these technologies seem well designed for a single, small boat project, but how does AI factor into the world of superyachts? Is an autonomous yacht on the horizon?
Despite the fact that 2001 set us up for decades of machine distrust, AI and machine learning have grown exponentially in recent years. It’s what makes your Roomba, Alexa, Google Maps and even Tinder work.
OSCAR’s user interface.
Image Credits: BSB Marine
The most familiar autonomous developments are the self-driving cars currently being tested. Elements of this technology, such as automatic breaking, self-parking, front crash avoidance and lane drift warnings are already in use, and Mercedes-Benz, Tesla and Cadillac have more complex driver assist systems for limited applications. However, fully autonomous cars, originally promised for 2017, remain the stuff of movies. Fatalities during testing mean only controlled simulations at this time. The amount of training cars need to react safely has been vastly underestimated, so imagine the training necessary to autonomously operate something as complex as an 100-metre superyacht. There are, however, a growing number of superyacht-related operations in which AI can be remarkably useful.
Jeffrey Bowles, director of naval architecture studio DLBA, a division of Gibbs & Cox, sees AI as a tool for increasing owner happiness and making sure the yacht operates smoothly. “SAHM, or Self-Adaptive Health Monitoring, will allow artificial intelligence to monitor all the system data being created that no chief engineer or captain can possibly track,” he says. “It looks at all the data points coming from all systems and using algorithms, can predict if and when a piece of equipment is likely to fail, giving the crew time to address it before failure cancels the owner’s trip.” The one drawback at the moment is that there is no universal language for all the gear data.
OSCAR is a collision-avoidance system.
Image Credits: OSCAR
What if the ship could always pick the most economical route by itself? While studies have shown that AI could reduce fuel consumption by letting the computer do the routing, the next step will be creating cargo and tanker ships that can operate safely without crew (30 per cent of a voyage cost), at least during much of the voyage.
A new partnership between Rolls-Royce and the Google Cloud Machine Learning Engine promises self-learning ships thanks to advanced machine-learning algorithms. It will also bring the Rolls-Royce vision – first presented in 2013 – of a fully autonomous ship closer to reality. Step one is to accelerate the training of existing AI algorithms for image recognition and to define hazard rules using the software fuelling Google’s voice and image search applications. It’s cloud-based learning can be shared by any connected vessel or landside control centre.
A new partnership between Rolls-Royce and the Google Cloud Machine Learning Engine promises self-learning ships thanks to advanced machine-learning algorithms
Image Credits: Boris Hermann / Team Seaexplorer
The cybersecurity threat
In June 2020, shipping giant Maersk was crippled for days and its orders disrupted for weeks when its IT system was wiped by malware. Maersk wasn’t even the target, but the cyberattack spread to 50,000 users and 600 sites in 130 countries. The attack cost Maersk an estimated $300 million (£210m).
AI is the perfect solution for the rise of cyber threats such as this because it reacts far quicker than human ever could, says Nick Pomponio, co-founder and operations director of CSS Platinum. The UK-based security firm has partnered with Darktrace, a scalable AI app that runs in the background of a communications network. “It learns habits, patterns, equipment and users on the network,” says Pomponio. “If someone or something, even latent malware, accesses the system and operates outside the application’s learned patterns, it isolates the communication and that user point and notifies the administrator before damage is done.”
The International Maritime Organization (IMO) mandated that all commercial vessels over 500GT, including superyachts, had a workable cybersecurity plan in place after January 1, 2021.
CSS Platinum gathers data from its network of users to predict cybersecurity threats.
Image credits: DLBA
Grand designs for AI
The design of the boats themselves could soon be given over to AI. “Design is an interesting area of development for artificial intelligence and some of the things in the works I cannot reveal,” says Bjørn-Johan Vartdal, head of the Maritime Incubator at Maersk DNV GL. “Basically you state the allowable boundaries and other details and ask the computer what is the best shape to achieve the goal. It might be the entire hull or it might be something as specific as the shape of a propeller. It gets the designer out of the box of preconceived notions. Maybe the solution isn’t buildable with traditional [manufacturing], but that’s when you have to think about 3D printing. It’s really opening up design.”
Vartdal cites a range of products developed via software from Autodesk which have been in existence since the turn of the century. “There are bicycles, aeroplane seats, fans and more that were designed by computers as the best solution for a given problem. Artificial intelligence is designing things that haven’t been possible before,” he says.
The processing demands for artificial intelligence are huge. To meet the demands of machine learning and computer-vision technology for MAS’s AI captain, the team used an IBM Power AC922 fuelled by IBM Power9 CPUs and NVIDIA GPUs, the same technologies behind the world’s smartest AI supercomputers.
Image Credits: CSS Platinum
Giedo Loeff, senior team leader of R&D at Feadship’s De Voogt Naval Architects, reveals the company is making use of AI in research while also looking to develop its capabilities. “We will need artificial intelligence more and more in the future to be able to digest vast quantities of data,” he says, citing the current stumbling block being the lack of digitisation of existing data due to challenges as broad as management, connectivity and legal ownership.
“For our work, AI will become an enabler for the digital progress of applications,” Loeff says. Describing a yacht as “a collection of systems spitting out vast quantities of data”, he notes that 100,000 parameters/samples per second is easily reached. A yacht’s HVAC data points alone can reach 15,000. “Crew can’t make good use of the avalanche of information with manual spreadsheets; it requires smart algorithms. They are currently primarily using monitor and alarm functions. Artificial intelligence could provide predictive insight.” One example is engine exhaust temperatures. “High temperatures could mean in-engine wear, exhaust filter clogging, bad weather, excessive hull/prop fouling or something else entirely. In order to determine and address the cause, you need data from various sources and diagnostics incorporating, all in real time.
Another yacht featuring OSCAR technology
Image Credits: Yann Riou
“For R&D purposes, we have three yachts [and one underway] fully wired up and high capacity data storage installed,” continues Loeff. “For example, we are currently using fibre-optic sensors to detect the influence of thermal, wave and sailing loads on the ship’s structure. These systems, at a moderate setting, sample at 20kHz, so every second [there are] 20,000 measurement points per sensor, and there are 40 of them. We run automated algorithms on board to reduce this data to 200Hz and send three-minute derived statistics to the cloud.”
While it’s easy to see how these AI examples could improve boat design and operation in the near future, it’s hard to imagine it replacing experienced superyacht crew when it comes to service, despite several recent studies revealing that millennials actually prefer interacting with robots. If so, there might be just the boat for them…
This feature is taken from the March 2021 issue of BOAT International. Get this magazine sent straight to your door, or subscribe and never miss an issue.