Ask a fleet technical superintendent what they get from an in-water hull survey and the honest answer is often: several hours of dark video and a surveyor's written impression. The footage is real evidence. The problem is that almost none of it is measured, positioned, or comparable to last year's.
The economics sit in the drydock planning meeting
Here is where the value actually lands, and it is not where most vendors point.
A drydock slot is booked months ahead with a defined scope. If the true condition of the hull — coating breakdown extent, corrosion severity, anode depletion, propeller condition, sea chest state — is established before that scope is fixed, the yard period is planned against reality. If it is established after the vessel is already in the dock, the scope changes late, and late scope change in a shipyard is the most expensive category of maintenance cost there is.
So the question a hull inspection system should be judged on is not image quality. It is whether the findings arrive early enough, and in a form specific enough, to change a plan that can still be changed.
Three things that make findings actionable
Measurement rather than description
"Significant coating breakdown observed on the port side aft" cannot be quoted, scheduled or budgeted. "Coating breakdown across 14.2 square metres, panels P42 to P47, mean severity grade 3" can. Turning observation into measurement requires stereo optics and a photogrammetric surface model, so that a defect has a real dimension and a real position rather than a timestamp in a video file.
Coverage you can prove
The first question any experienced surveyor asks of an inspection dataset is which parts of the hull were actually looked at. Without a coverage map, an absence of findings is ambiguous — it could mean the area is sound or that the vehicle never went there. A coverage map converts that ambiguity into information, and it is the single most underrated output of a robotic survey.
Comparison against the last survey
A single survey is a snapshot. Two surveys of the same hull, spatially registered to each other, are a corrosion rate and a coating degradation curve — which is what maintenance planning actually needs. This only works if findings are positioned on a persistent hull model rather than described in prose, which is why the photogrammetry matters more than it first appears.
The deliverable is not footage. It is a hull model with measured, positioned, graded findings that can be differenced against the last one.
Where AI genuinely helps, and where it does not
Computer vision is well suited to the triage layer of this problem. Coating breakdown, corrosion patterns, fouling severity, weld anomalies, missing anodes and mechanical damage are all visually distinctive, and a model can review every frame of a multi-hour survey without the attention decay a human reviewer inevitably experiences around hour two.
Biofouling grading is a particularly good fit. The Level of Fouling scale is a defined visual rubric, which makes it a well-posed classification problem, and consistency of grading between surveys matters more than any individual grade — exactly the property automated assessment is good at delivering.
What AI does not do is replace the surveyor. Structural significance is a judgement about consequence, not appearance. A two-millimetre crack is trivial in one location and a class item in another, and nothing in the image distinguishes those cases. Our systems triage and measure. The surveyor decides, and nothing enters a class report without that confirmation.
The regulatory picture is moving
Robotic in-water survey has moved from novelty to accepted practice with the major classification societies, and high-quality robotic video with supporting non-destructive testing data is increasingly accepted for in-water survey in lieu of drydocking, subject to each society's requirements. That acceptance is what converts an inspection improvement into a commercial one: it changes the survey regime, not just the survey.
Separately, biofouling has acquired regulatory weight of its own through IMO biofouling guidance and national biosecurity regimes, which raises the value of consistent, documented, comparable fouling assessment independent of any structural finding.
The fuel argument, which usually closes the deal
Fouling increases hydrodynamic drag. Drag increases fuel consumption. Consumption is now a rated and reported number under carbon-intensity regulation, and fuel remains the single largest variable cost in most operating budgets.
This is why hull condition monitoring frequently gets approved on the fuel case when it would not get approved on the safety case alone. A fouling curve that lets an operator schedule a cleaning at the right time — rather than too early, wasting a mobilisation, or too late, having already paid for the drag — tends to pay for the inspection programme on its own.
What we would tell a technical superintendent
Do not buy an inspection system on image resolution. Buy it on whether it produces a coverage map, a measured finding list positioned on a hull model, and an automatic comparison against the previous survey. Ask what happens to a finding the model is unsure about, and be suspicious of any answer that does not involve a human. Then ask whether the report is in a form your class surveyor will accept without a covering explanation.
If the answers are good, the system will change your drydock planning. If they are not, you have bought a better camera and the same meeting.