AI CAPTAIN! What autonomous vessel UK sea trials will mean for the future of shipping
There are reported to be more than 1,000 autonomous vessels operating in international waters and 53 odd organisations contributing to various regulatory groups working to help MASS co-exist with manned shipping.
One autonomous ship making a transatlantic crossing this September is the Mayflower, a crewless solar powered trimaran which will use artificial intelligence (AI) to plot its course. The experiment marks 400 years since the original Mayflower set sail from Plymouth England to Plymouth Massachusetts.
To help the modern Mayflower learn to navigate, a tech portal has been operating in the Plymouth Sound to give the vessel’s “brain” an understanding of what sea hazards might look like.
In Portland Harbour meanwhile, a second trial, managed by Solis Marine for the MCA, is using live and historic datasets to help make the UK waters safe for the testing of remote and unmanned vessels as they navigate busy manned shipping lanes and harbours.
What can MASS learn from both projects?
The Modern Mayflower
Brett Phaneuf of the Submergence Group explains how the ship has learned to navigate and why MASS is a forcing function for maritime regulators
Over the past three years, my team has been installing a system of cameras, radars, AIS and other instruments around Plymouth Sound, on shore and on offshore structures. We operate a private wifi network that backhauls all the data and we process it on IBM super computers using their software, such as Power Vision AI. From this we make models and have thereby given the ship an “understanding” of what hazards at sea may be, like other ships, buoys, debris, land, docks, etc. This has been a painstaking process.
Then we use a product from IBM called Operational Decision Manager that works with rule-sets to determine possible outcomes to remain within a set of rules, in our case COLREGS.
All the sensors use our models to classify hazards and then that data, along with own-ship data, is fed into ODM and considered with regard to the ‘rules’ and then a set of possible actions to avoid a collision and remain COLREG complaint are output to the higher level autonomy system, or mission manager, on the ship.
It communicates with lower level microprocessors that take action, like turning the rudder, or hailing a vessel, or changing radar settings or slowing down/speeding up as needed to keep safe. All of this is decision making is kept in an immutable record on the ship and in the cloud so that we know what it did when and why……just in case. And also, we can learn from this…in fact, we can let the ship learn from its own successful behaviour in real-time if we so choose.
So, we just give it a goal like ‘Go to Plymouth Massachusetts’ and it finds the right chart information (loaded on the boat), understands the ship’s health and capability, checks with the Weather Company with onboard APIs provided by IBM and then plots the most efficient, safest course all by itself. And as it encounters obstacles or hazards, it acts to dynamically re-plan its mission while remaining safe.
We can also reach through our satcom and take action should we wish the ship to behave in a different manner than that which is has chosen.
For me, this is a forcing function for regulators. There isn’t any rule that prevents what we’re doing and in our opinion we are compliant with COLREGS, SOLAS and IMO – so this will be a study in how we manage this and learn what a suitable set of regulations should be, if any. I personally think that the current regulations are sufficient and we just need to accept ‘keeping a watch’ is not necessarily with a ‘human’ on the ship. It could be a human on shore with good communications, or AR/VR representations, or an AI captain – and that’s what we’re doing.
How we manage this sort of innovation in ports and harbours and coastal waterways will also be something that needs to be addressed and through Mayflower, we’re doing it.
Rosalind Blazejczyk, co-founder of Solis Marine explains how the current MARLab data experiment is helping stakeholders plot UK maritime’s course towards MASS adoption
Modern day ships are sophisticated computers with multiple standard sensors and inputs that collect different data sets to provide the captain and crew with automated support.
To understand what autonomous ships require to become fully automated from a navigational point of view, we have been gathering and testing a range of different data drawn from Government and commercial sources. We have focused on assembling in one place the information MASS operators say they require to allow autonomous surface and sub surface vessels to operate safely in UK waters and in line with COLREGS.
Since January, more than 18 stakeholders have been testing the MARLab data hub which we have developed with MariTrace on behalf of the MCA. The prototype aggregates live and historical data which is local to Portland Harbour, an active location for the testing of MASS vessels for several years.
The advantage of Portland is the ease of access to open, relatively quiet water, but with an option to test in busy waters if needed. By comparison, other locations requires MASS vessels to be towed to the test site or require longer transits before reaching test areas. In Portland, MASS vessels can operate immediately from jetty.
Our goal is the establish if MASS operators require this sort of platform and if so, in what format and where. What is missing, what other systems do we need to link into and how can a wider data led platform which covers all UK waters best integrate with existing ship systems to move towards a COLREG compliant system?
Currently this doesn’t exist, but the MARLab data experiment has made significant strides towards that goal simply by engaging with as many interested parties, operators and academics as possible.
Solis Marine will be submitting a full report on the MARLab test to the MCA at the end of March together with a series of industry informed recommendations on how best to take the data hub concept forward.
You can see the prototype in its first iteration here: