USBL in extreme proximity: positioning lessons for subsea inspection

A formation of robots working closer than before

A group of marine robots is heading north this summer to work in formation and operate closer to glacier faces than such vehicles have previously gone. The GIANT project, a climate-science mission led by the British Antarctic Survey and backed by ARIA, is being supported by Sonardyne underwater positioning technology, including its Ranger 2 and Mini-Ranger 2 systems. The work, running from July into August in Kangerlussuaq Fjord, is staged from the research vessel RRS Sir David Attenborough and its daughter craft Erebus, and involves AUVs including the AutoSub Long Range (Boaty McBoatface), a Teledyne Gavia and ecoSUB Robotics vehicles. Those are the facts the announcement gives us, and they are enough to frame an engineering problem worth taking seriously.

The headline is the science, but the enabling condition is positioning. Putting several autonomous vehicles into coordinated close-quarters work against an ice front is, from a geodesy and navigation standpoint, one of the harder acoustic positioning problems anyone attempts. It is also the same problem – nearly component for component – that confronts a subsea inspection team asked to hold an inspection-class vehicle within a few metres of a jacket, a monopile foundation or a manifold. The surfaces differ, but the physics does not.

That is why this mission is worth the attention of survey and inspection managers who will never go near a glacier. The lessons about how you build position confidence at short standoff, in a hostile acoustic channel, with more than one vehicle in the water, transfer directly to structure inspection.

Why proximity is where the positioning budget gets spent

Proximity work inverts the usual comfort of open-water survey. On a lawnmower survey at altitude over featureless seabed, a position error of a metre or two is often tolerable against the deliverable. Bring the vehicle to within a few metres of a hard object it must not touch, and the same error becomes a collision risk, an aborted line, or an inspection finding that cannot be relocated on a return visit.

The governing relationship is simple and unforgiving. The allowable position uncertainty at the point of work is set by the smaller of two tolerances: the standoff you must keep to avoid contact, and the relocation accuracy the deliverable demands. A close-visual inspection that must let a later campaign return to the exact weld or the exact section of coating breakdown imposes a relocation tolerance measured in tens of centimetres. If your positioning solution cannot beat that at the working geometry, the data is a photograph without coordinates.

Across the wider industry the accuracy target is usually pinned to a recognised classification appropriate to the deliverable. Hydrographic survey standards such as IHO S-44 frame horizontal accuracy for bathymetric deliverables, but even their tightest orders are far looser than the tens-of-centimetres relocation tolerance that close-visual structure inspection demands. For structure-relative inspection work the binding figure is the client’s relocation requirement, not any open-water survey order. The point for a decision-maker is that the positioning architecture is not a line item to be minimised. It is the constraint that determines whether the mission produces usable, defensible data at all.

What actually drives the architecture choice

The decision in front of a project team is not “USBL or not USBL”. It is how to assemble an aiding architecture whose combined error at the working point stays inside the tolerance, given the acoustic environment and the number of vehicles. Four factors dominate.

The acoustic environment sets the error floor. USBL bearing error translates into a lateral position error that scales with slant range, which is why USBL accuracy is properly expressed as a percentage of slant range, not a fixed number of metres. Calibrated modern wideband systems achieve on the order of a fraction of a percent of slant range, but that figure assumes a well-behaved sound-speed field and a clean line of sight. Near a glacier front you have neither. Cold, fresh meltwater layered over warmer saline water produces steep sound-speed gradients that bend acoustic rays; sediment-laden plumes attenuate and scatter; and the ice face itself is a large, near-vertical reflector that generates multipath and can shadow the direct path entirely. A steel structure does the same thing to acoustics that an ice cliff does.

Absolute versus relative accuracy is a real fork, not a semantic one. If the vehicle only needs to hold station relative to the object it is inspecting, a relative solution – vehicle-referenced sensing, terrain or structure-relative aiding – can be tighter and simpler than a georeferenced one. If findings must be tied into a fixed geodetic frame for cross-campaign comparison, you carry the additional error of the surface positioning and the vertical reference all the way down. Deciding which one the deliverable truly requires, early, changes the whole sensor list.

The number of vehicles rations the acoustic channel. Formation work means several transponders sharing one water column. The acoustic channel is a scarce, half-duplex, latency-heavy resource, and simultaneous interrogation causes cross-talk. Practical systems resolve this with time-division schemes, in which each vehicle’s absolute position update rate is roughly the master cycle divided by the number of vehicles, or with code-division (wideband) schemes that allow concurrent interrogation and largely preserve the per-vehicle update rate. Between those updates, the vehicle is flying on inertial dead reckoning. The formation size therefore sets a hard requirement on how well the onboard navigator must coast.

The inertial core carries the gaps. Every credible proximity solution is an INS aided by acoustics, not raw acoustics. The inertial navigation system, disciplined by a Doppler velocity log in bottom-lock and reset by USBL or LBL fixes, is what holds position between acoustic updates and through short line-of-sight dropouts. The quality of that coast – grade of the IMU, availability of DVL bottom-lock, tightness of the integration – determines how much acoustic degradation the system can absorb before the solution drifts outside tolerance. Where acoustics fail entirely at very short standoff, optical and terrain-relative methods take over, and it is worth understanding where visual navigation buys position confidence that acoustics cannot before assuming a single sensor will carry terminal approach.

Where proximity positioning goes wrong

The failures in close-standoff work are consistent, and most of them are specification failures made long before mobilisation.

1. Treating the USBL accuracy figure as a fixed number

The most common error is reading a datasheet accuracy – a tidy figure in metres or a percentage – and carrying it into the error budget as a constant. USBL error is geometry-dependent. It grows with slant range and degrades badly at grazing angles, where the array resolves bearing poorly and ray bending is worst. A number quoted for a vertical, short-range geometry over a calibration site tells you little about performance when the vehicle is 200 m out and 20 m down against a reflective face. Build the budget from the actual working geometry, not the brochure.

2. Under-resourcing the sound-velocity profile

Acoustic positioning is only ever as good as the sound-speed model applied to it. In a strongly stratified column – exactly what a meltwater front or a summer thermocline produces – a single harmonic-mean sound speed is not good enough, because it ignores the ray bending that a real gradient imposes. Teams that take one SVP cast at the start of the day and treat it as valid for the shift are quietly injecting a slowly varying bias into every fix. The environment near a glacier can change the profile within hours; near a structure it changes with tide and freshwater outflow.

3. Ignoring channel contention until it bites

Project plans built around single-vehicle update rates fall over when the formation goes into the water. If the acoustic scheduling has not been designed – which vehicle interrogates when, at what rate, with what code – the vehicles will steal each other’s slots and the effective update rate per vehicle collapses. That is not a field problem to solve on deck; it is a design decision to make when the concept of operations is written.

4. Confusing absolute and relative requirements

Specifying full georeferenced accuracy when the task only needs structure-relative station-keeping over-engineers the surface and vertical reference chain and inflates cost. The reverse – specifying a relative solution when the client actually needs findings tied to a fixed frame for trend monitoring – produces data that cannot be compared campaign to campaign. Both are avoidable by interrogating the deliverable rather than the vehicle.

5. Assuming aiding will always be available

DVL bottom-lock disappears when the vehicle rises off the seabed into mid-water against an ice keel or a structure, and USBL line of sight is lost in the acoustic shadow of the object being inspected. A plan that has not defined what the navigator does during those dropouts – how long it can coast, what triggers a standoff-abort – is a plan that relies on nothing going wrong at the most demanding moment of the dive.

How to specify it, and how to test that you got it right

The positioning architecture for extreme-proximity work should be settled on paper and proven at mobilisation, not discovered on site. The following are the decisions and checks we would hold a project to.

• Write the position error budget before choosing sensors. State the required relocation tolerance and the minimum safe standoff, take the smaller, and treat it as the ceiling. Then work backwards: at the planned slant range and depression angle, does the USBL contribution alone consume more than a fraction of that ceiling? If it does, the answer is closer working geometry, LBL or sparse seabed transponders for a tighter local frame, or a shift to relative aiding – decided now, not after the first failed line.

• Calibrate and prove the acoustic system to recognised guidance. Run a full USBL calibration and offset verification on mobilisation in line with IMCA S 017, Guidance on vessel USBL systems for use in offshore survey, positioning and DP operations, and confirm the result with an independent check against a known point. A box-in that closes on paper but has never been validated against an external reference is not assurance.

• Match SVP cadence to the stratification, not the clock. In a strongly layered column, plan multiple casts per shift and consider a through-water sound-speed sensor on the vehicle and a moving-vessel profiler on the surface unit. Apply a ray-traced solution at grazing geometry rather than a harmonic mean. Treat the sound-speed field as a live input, not a mobilisation formality.

• Design the acoustic schedule for the full formation. Fix the TDMA or wideband code plan for the actual vehicle count, compute the resulting per-vehicle update interval, and confirm the INS holds inside tolerance across that interval at the expected current and manoeuvre rate. If it does not, either the formation shrinks or the inertial grade goes up.

• Define degraded-mode behaviour explicitly. Specify the coast time the navigator is allowed on inertial-only before an update is mandatory, the covariance threshold at which inspection data is flagged or rejected, and the standoff-abort criteria for DVL and USBL dropout. Log the integrity decision – the moment the system judged its own position good enough to keep working – so the assurance case is auditable after the fact.

• Gate the data on position confidence. Set an acceptance threshold on the filtered position covariance and reject or re-fly inspection data acquired outside it. Position uncertainty that is not recorded alongside every finding is a liability the moment a client asks how confident you are in a relocation.

The strategic reading for a survey or inspection manager is that this decision is about capability, not hardware. The temptation is always to specify a vehicle and a positioning box and assume the rest follows, but the harder and more valuable question is whether you are buying an AUV or the assured capability to deliver a result – and in proximity work the assurance case is the binding constraint. A climate mission flying robots into a glacier front and a contractor inspecting a wind-farm foundation are, at the level of positioning engineering, solving the same problem: how to keep a moving vehicle’s position trustworthy when the acoustic channel is fighting you and there is a hard surface a few metres away. Specify for that, prove it before you sail, and the science – or the inspection report – takes care of itself.

---

Based on: Marine robots in formation, closer to glaciers than ever, with Sonardyne positioning