ABS, Polaris Shipping, HHI and AVIKUS Sign Agreement to Develop Conditional Unmanned Bridge for VLOCs

The Analysis

Four companies signed a piece of paper in June, and the cargo world should read it like a system tuning log, not a press release. ABS, Polaris Shipping, HD Hyundai Heavy Industries and AVIKUS agreed to develop a conditional unmanned bridge for Polaris Shipping's 325,000-ton Very Large Ore Carriers. The concept lets the bridge run without people during deep-sea open-ocean passages, under defined low-risk conditions. ABS runs structured safety evaluations aligned with the IMO MASS Code. Polaris feeds real operational data to define where humans stay in the loop. HHI works out the design changes on the hull and bridge side. AVIKUS writes the autonomous navigation system requirements. That is four workstreams, one boundary problem: under exactly which conditions does the machine hold the watch, and when does a human take it back.

I read agreements like this the way I read a system design doc. The interesting part is never the headline goal. The interesting part is the interface definitions, because every boundary drawn in this document becomes a constraint that ripples down to shippers, charterers and port callers for the next twenty years of these ships' lives. A VLOC is not a downtown delivery van. It is a slow, heavy, predictable queue item in the bulk trade, and that makes it the perfect first candidate for conditional autonomy. The open ocean leg of an ore run is long, monotonous and low-traffic. If autonomy enters commercial shipping anywhere first, it enters there.

Here is the part most coverage misses. You, the importer or exporter moving bulk, ore, steel, or project cargo on tonnage like this, are a stakeholder in that boundary definition whether you were at the table or not. The conditions written into the MASS Code alignment work will decide how these vessels are certified, crewed, insured and scheduled. So this piece does what I always do: model the problem first, translate the announcement into cost and timeline terms, run one explicit calculation, and then lay out what to actually do in the coming weeks and quarters, with thresholds and owners attached to each action.

Let me put the problem in queue terms, because that is how I see everything. A vessel voyage is a queue of states: loading, departure, coastal passage, open-ocean transit, approach, berthing, discharge. Each state has an arrival rate of unusual events and a service rate of human attention available to handle them. On a conventional ship, human attention is provisioned for the worst state, which is coastal approach in traffic. That means for the majority of a deep-sea transit, you are paying full staffing for a state that rarely generates demand. The unmanned bridge concept is, in essence, a decision to split that queue: keep humans staffed for the high-event states, and let the machine run the low-event state under conditions where the event arrival rate is provably low. Nobody is removing the queue. They are re-partitioning it.

Now look at the four parties as four modules in a system design, and you can see why the architecture makes sense. Polaris Shipping owns the operational data, so they own the ground truth. Their job, per the agreement, is to use real operational data to define human-automation boundaries. In my language, they are fitting the event distribution: how often, on real ore runs, does something happen on the open-ocean leg that would need a human decision, and what does that something look like. HHI owns the platform modifications, meaning they handle the physical and design-layer changes a bridge needs before it can be unmanned even conditionally. AVIKUS writes the navigation system requirements, which is the software layer. ABS sits above all three as the verifier, running structured safety evaluations aligned with the IMO MASS Code, which is the standard that regulators will eventually use to decide whether any of this is legal at scale.

When I see a consortium split like that, with a verifier built in from day one instead of bolted on at the end, I take the timeline more seriously than I otherwise would. I have watched plenty of autonomy announcements where the regulator shows up after the demo. Here the classification society is a signatory. The summary of the agreement says the four parties intend to build technical and operational readiness for what is expected to be the first autonomous shipping concept to reach commercial deployment. Expected is doing real work in that sentence, but the structure is right: verification is not an afterthought, it is a signing party.

Why does a bulk ship lead rather than a container ship or a tanker? Model the traffic. A 325,000-ton VLOC on an ore run spends most of its life on long, low-density open-ocean legs between a handful of fixed mine ports and fixed discharge ports. The event arrival rate on those legs is about as low as it gets in commercial shipping. Compare that with a feeder container ship weaving through coastal traffic, or a tanker maneuvering near terminals, and you see the conditional part of conditional unmanned is doing a lot of honest work. The concept explicitly does not claim unmanned operation everywhere. It claims unmanned operation during deep-sea open-ocean passages under defined low-risk conditions. That is a boundary, drawn tightly, on the leg where the machine's odds are best.

Now the part that actually touches your P&L if you charter or operate tonnage like this, or move cargo on ships that will inherit this technology. The first-order effect is not on ocean freight rates, and I want to be careful here not to promise a number the announcement does not support. Crew cost is a real line item on a bulk carrier, but the agreement does not say crews disappear; it says the bridge goes unmanned on a defined leg. What changes first is not cost, it is evidence. Every voyage these ships sail with instrumented bridges generates operational data that becomes the training and certification base for the next generation of conditions. Whoever charters or ships on those early vessels is, knowingly or not, inside the data collection program.

The second-order effect is schedule variance, and that is where I would put my attention. Autonomy on the open-ocean leg, done right, should tighten transit time variance rather than shorten mean transit time. A VLOC is not trying to go faster; it is trying to stop being late for reasons that live in the watchkeeping layer: fatigue, rotation mismatches, weather routing decisions taken by a tired officer at three in the morning. If the conditional unmanned leg executes as designed, the distribution of arrival times narrows. For bulk cargo, a narrower arrival distribution is worth real money even when the mean does not move, because the cost of a bulk laycan miss lands on the charterer in demurrage, storage, or downstream production scheduling. I will take a tighter distribution over a slightly faster mean any day of the week, and if you run plants or steel mills, so should you.

Let me run one explicit calculation so the variance point is not hand-waving. I want to stress that the agreement contains no voyage cost figures, so everything below is my own modeling with assumptions stated out loud, in the tradition of measure first, optimize later. Assume a long-haul ore trade where a vessel earns or costs, all-in, in the range of 20,000 US dollars per day to the owner, which is in line with recent capesize market history, and assume the discharge port is a bulk terminal where the berth window was booked for a specific laycan.

Assume a port-year where roughly one voyage in twenty arrives outside its laycan for reasons traceable to the ocean leg: delayed departure decisions, watchkeeping-driven speed profiles, or rerouting hesitations. Assume each miss costs the cargo owner 150,000 dollars between penalty clauses, re-booking, storage and plant disruption. On a fleet of, say, six vessels doing eight voyages a year each, that is 48 voyages, and 2.4 misses a year, or about 360,000 dollars of annual laycan-miss exposure for the cargo side alone. Now if conditional unmanned watchkeeping cuts the ocean-leg-attributable miss rate in half, that is 180,000 dollars a year recovered, per six ships, with zero change in mean speed. Scale that across a 300-ship capesize and VLOC fleet and you are looking at a nine-figure industry number that never shows up in a freight index. This is why I keep saying: watch the variance, not the mean.

Where is the bottleneck in this whole program, if you ask me? Not the algorithms. Open-ocean collision avoidance for low-traffic legs is a solved-ish problem in the lab. The bottleneck is the boundary definition work Polaris is signed up to do: enumerating, from real data, the exact set of conditions under which unmanned watch is acceptable, and the exact trigger set that hands the watch back to a human. Every condition you add shrinks the usable envelope; every condition you remove raises the risk of a handback the system cannot execute.

That trade is the whole project. Get it wrong in the conservative direction and the unmanned bridge never turns on, and the program becomes a very expensive data logger. Get it wrong in the aggressive direction and you get one bad handback that sets the regulatory conversation back five years. The MASS Code alignment work at ABS is what keeps this honest, because structured safety evaluation forces the boundary to be written down and tested rather than argued in a conference room.

There is also a handback problem that most people outside operations underestimate, so let me make it concrete in queue terms. When the system decides conditions are leaving the defined envelope, a human has to take the watch. But a human who has been out of the loop for nine days of open-ocean transit is not instantly at full situational awareness. The state that matters is not the machine's state, it is the human's ramp-up time. Any serious boundary definition has to model that ramp-up: how long from call to competent command, what the handover checklist contains, and what the ship does during those minutes.

I would want to see handback drills treated as a KPI with a threshold, not a training afterthought. My own working benchmark, and I state it as a benchmark rather than an industry standard: a competent officer should be able to go from called to fully commanding the watch in under ten minutes on a documented drill, and the system should be designed so that the envelope gives at least that much warning before conditions actually cross the line.

A few people will read the 325,000-ton figure and ask what deadweight has to do with autonomy. More than you would think, and it cuts both ways. Big ships are easier in some dimensions: their maneuvering cycles are slow, their open-ocean legs are long, and their route structures are simple, often a shuttle between one loading region and a small set of discharge ports. That is a friendly environment for a machine. But they are harder in others: the energy in a 325,000-ton hull means the margin for a late correction is thinner in distance terms, and the consequences of any failure scale with mass.

So the design question is not can the system steer the ship, it is can the system prove, to a verifier, that it detects envelope exits early enough for either itself or a human to act while the ship is still in a recoverable state. That framing, detect early, hand over with margin, is exactly the queue discipline I described above, applied to a hull where the physics punish sloppy timing.

One more thing about the data side, because it decides who benefits first. Polaris defining human-automation boundaries with real operational data means the boundary quality is a direct function of the data's quality and coverage. Data from a single trade loop, a single ship class, and a limited set of weather regimes produces a boundary that is honest but narrow. Widen the data and the envelope can widen with evidence behind it. For a cargo owner, this suggests a simple posture: the value of these programs compounds with every instrumented voyage, so the interesting question to ask an owner is not whether the system works, but how many hours of boundary-relevant operational data they are accumulating per quarter and who audits the labeling. That is a question any analytics-literate shipper can ask, and the quality of the answer tells you more about program maturity than any press release. Here is my action list, and I attach owners and thresholds because a recommendation without an owner is a wish.

Within the next two weeks, if you charter VLOC or capesize tonnage on long-haul ore, steel or bulk routes, task your chartering desk with a one-page map of which vessels in your book belong to owners publicly engaged in autonomy programs. Polaris Shipping is now the named one; there will be others. The output is a list with voyage counts attached, so you know what share of your program touches instrumented, boundary-testing tonnage. Owner: chartering desk. Success measure: the map exists, covers at least the next two quarters of fixture plans, and is reviewed at the monthly trade meeting.

Within the next month, add a data and connectivity clause to your chartering checklist for any new fixture on tonnage known to be running autonomy trials. You do not need to renegotiate existing deals. You need the standard clause ready, covering what operational data is logged, who owns it, what incident reporting the owner owes you when the unmanned mode is engaged or disengaged, and how a mode transition is communicated on the voyage. Owner: chartering plus legal. Success measure: a pre-approved clause sits in the template library before your next fixture round, not drafted after the fixture is agreed.

Within the same month, ask your operations team for a laycan variance report on your bulk program, six quarters back, split into ocean-leg-attributable delays and everything else. This is the measure-first discipline. You cannot price the benefit of tightened variance if you have never measured the variance. If the ocean-leg share is under five percent of your delay budget, autonomy on the open leg is a nice-to-have for you and your attention belongs elsewhere, probably at port congestion. If it is above ten percent, this technology is aimed directly at your biggest timing loss. Owner: operations analytics. Success measure: the split exists with a defined attribution rule, and the threshold decision is recorded in writing.

Over the next quarter, if you are a shipper with volume commitments on routes these vessels serve, open a quiet conversation with the owners involved about how mode transitions will be handled in voyage instructions. The question to put in writing: when the bridge moves between manned and unmanned modes, what does the charterer see, when, and who is accountable for the watch during the transition. Owner: commercial lead. Success measure: a written answer, even a preliminary one, on file before the concept study phase ends and the design phase begins.

And at the quarterly review, put a standing agenda item on autonomy readiness: track the ABS safety evaluation milestones, the MASS Code alignment outputs, and any published boundary definitions. The point is not to chase the news. The point is that certification milestones on these four-party programs are leading indicators for when conditional unmanned tonnage starts affecting charter markets, insurance terms and port acceptance, and those lead times are long. Owner: supply chain strategy. Success measure: a one-paragraph status with a go or no-go on monitoring intensity, every quarter, in the minutes.

Now the alternatives and the traps, because there are always traps. The alternative to engaging now is engaging later at charter-market prices, when instrumented tonnage starts carrying a premium or, the other way, when un-instrumented tonnage starts carrying a discount and you discover your preferred owners are on the wrong side of that line. The cost of monitoring now is a few analyst hours a quarter. The cost of discovering the market has split under you in 2028 is a repricing event on your whole bulk book. On the trap side, the big ones: first, do not assume unmanned means crewless; the agreement is about a bridge mode on a defined leg, and misreading that leads to fantasy cost models.

Do not build business cases on mean speed improvements; the evidence base points at variance, and a variance case dies if your delay data cannot attribute causes. Watch the insurance interface: if a mode transition is ambiguous in the voyage record, a laycan dispute becomes an insurance dispute, and that is where months disappear. Resist the urge to demand autonomy clauses into every fixture; over-requesting data rights on tonnage that is not in any trial program just marks you as a difficult counterparty.

One more trap, and it is the one I care about most. Do not let anyone in your organization tell you this technology is either magic or nonsense. It is neither. It is a boundary problem being worked with real data, real design changes and a real verifier, on the easiest leg in shipping, by four parties who each own one layer of the stack. The honest position is conditional optimism: the open-ocean leg will likely be the first commercial autonomy case in deep-sea shipping, exactly as the agreement's framing suggests, and it will arrive first as a variance improvement, second as a crewing model change, and much later as anything resembling an unmanned fleet.

If you have measured your variance, mapped your exposure, and put the clauses and the monitoring in place, you will feel this transition as a schedule that gets easier to plan. If you have not, you will read about it in a rate report after the market has already repriced. The queue does not care which one you are. Measure it, and you get to pick.

↑ Back to the key points

— By Ravi Chandran