Long-Endurance AUVs: Rethinking Survey Planning for Multi-Week Dives

The demonstration, and what it actually proves

Cellula Robotics and Defence Research and Development Canada (DRDC) have reported a fully submerged endurance run by Cellula’s hydrogen fuel cell-powered Envoy AUV. On a representative mission profile the vehicle logged 385 hours on mission and covered 2,023km submerged, exceeding the platform’s published performance specification. The Envoy is owned by DRDC, and the two organisations have run a multi-year collaboration under a DRDC development contract aimed at maturing long-endurance autonomy. The fuel cell was developed with Infinity Fuel Cell and Hydrogen.

The detail that matters for survey planners sits in the mission profile rather than the headline hours. The run included more than 4,000 turns and manoeuvres, each drawing more energy than steady linear transit. Cellula’s leadership framed the result as a demonstration of what is operationally useful rather than a number achieved under optimised conditions, and we read it the same way. At 385 hours the vehicle was down for roughly sixteen days. That moves the Envoy out of the single-shift survey AUV category and into a class of platform that can stay submerged for more than a fortnight.

We treat this as a credible signal, not a marketing figure, precisely because the manoeuvring load was included. Endurance quoted from a straight-line transit at optimum speed tells you very little about a real survey, where line turns, depth changes and station-keeping against current dominate the hotel and propulsion budget. A figure earned across 4,000 turns is the kind of number you can actually plan against.

Why a sixteen-day dive rewrites the planning problem

For most of the survey-grade AUV fleet, endurance has been the first constraint you hit. A high-specification battery vehicle works a shift measured in tens of hours, surfaces or recovers, downloads, recharges and resets its position fix, then goes again. Every recovery is a natural break point: it is when you reset navigation drift against GNSS, when you quality-check the data you have just acquired, and when you decide whether the next dive runs as planned.

When the platform can stay down for two weeks, that rhythm disappears, and the binding constraint moves. Energy is no longer the first thing you run out of. The limits that bite instead are navigation integrity over a very long unaided run, the confidence you have in data you cannot inspect until recovery, and the reliability case for a system operating with no hands on it for days at a time.

This is the central judgement we want survey managers to take away: on a long-endurance platform, the assurance case becomes the binding constraint, not the vehicle’s energy budget. The hydrogen fuel cell solves the energy problem so thoroughly that it exposes the problems that energy scarcity used to mask. A 24-hour AUV that drifts or degrades costs you a shift. A 16-day AUV that drifts or degrades on day two can cost you the entire campaign before anyone knows.

What actually drives the decision

The choice to deploy a long-endurance AUV on a survey scope turns on a small number of engineering trade-offs. Get these right and the endurance is a genuine commercial advantage. Get them wrong and you have bought a very expensive way to acquire fourteen days of uncertain data.

Navigation aiding over an unaided run. A modern AUV holds position with an inertial navigation system aided by a Doppler velocity log, and the practical figure that governs everything is the aided drift rate, commonly quoted as a fraction of distance travelled. High-grade INS/DVL combinations are commonly cited as holding drift to a small fraction of distance travelled with bottom lock. Over 2,000km of submerged transit, even a fraction of a percent accumulates to a position error of the order of a kilometre if nothing resets it. That is meaningless for any survey product. The whole question becomes how you bound that error: periodic GNSS fixes at the surface, USBL aiding from a support vessel, an LBL array, or terrain-aided navigation against a known bathymetric prior. Each of those carries a cost in time, vessel presence or pre-survey effort, and a multi-week mission concept that does not state its aiding strategy explicitly is not a plan.

Data-quality latency. On a conventional AUV survey you find the bad line during the post-dive download and reacquire it the next day. On a sixteen-day mission you commit to two weeks of acquisition before anyone sees a sounding. A drifting sound velocity profile, a multibeam calibration that has crept, a slowly fouling sensor face, an INS that is not converging — any of these can quietly contaminate days of data. The vehicle’s onboard quality monitoring is no longer a convenience; it is the only thing standing between you and a fortnight of unusable acquisition.

Energy spent on manoeuvre versus transit. The Cellula result is useful because it accounts for this, but the general point holds for any platform you assess. A survey pattern that is mostly long straight lines spends energy very differently from one full of short lines, tight turns and frequent depth excursions. When you scale a published endurance figure to your own scope, scale it against your actual line plan, not against the transit speed that produced the brochure number.

Communications and intervention bandwidth. Subsea acoustic telemetry gives you a trickle: status pings, health summaries, perhaps a coarse position. It does not give you the data to judge survey quality, and it does not give you the bandwidth to re-plan a complex mission from the beach. A long-endurance concept lives or dies on how much autonomy it has been given to make sensible decisions without you, and on how trustworthy that autonomy’s decision logic is.

Where survey teams get this wrong

1. Buying the endurance figure as the spec

Endurance is the most quotable number and the least useful in isolation. What you are actually buying is productive, survey-grade bottom time, which is endurance minus transit, minus any time the vehicle spends off-specification, minus the data you have to reject. We have seen the industry repeatedly conflate “hours submerged” with “hours of acceptable data,” and the gap between the two widens as missions get longer. Interrogate the manoeuvring profile behind any endurance claim and ask what fraction of it was on-survey at acquisition speed.

2. Under-resourcing the navigation aiding strategy

The single most common planning error on extended autonomous missions is treating the INS/DVL as if it bounds position error on its own. It does not — it bounds the rate at which error grows, and over a multi-day run that growth is unbounded without an external reset. Teams that have only ever run short dives, where surface GNSS resets came naturally every shift, carry that assumption into a two-week concept where it no longer holds. The aiding architecture — surface fixes, USBL, LBL, terrain-aided navigation — has to be designed to the survey’s positioning requirement, not bolted on afterwards.

3. Assuming sensor performance is stable across the whole dive

Calibrations drift, sound velocity structure changes with location and time of day, and sensor faces foul. On a short dive these effects are small and caught at download. Across sixteen days they are first-order error sources. The mistake is to plan for the survey’s instrument performance on day one and assume it holds to day sixteen, with no onboard mechanism to detect or correct the drift.

4. Treating the abort decision as an afterthought

With no operator watching in real time, the vehicle has to decide for itself when to continue, when to hold, and when to abort to a safe recovery. Those thresholds — on navigation confidence, data quality, energy reserve and system health — are an engineering deliverable, and the audit trail of why the vehicle made each integrity decision is part of the data product. Many programmes specify the survey sensors in detail and leave the integrity-decision logic vague. That is the wrong way round for an unattended platform.

5. Planning recovery and reserve as if it were a short dive

A long-endurance vehicle still has to come home, and the energy and navigation margin to reach a recovery point or a docking station is not a rounding error. The reserve has to be carried as a hard constraint, not consumed by an optimistic last survey line. The temptation to squeeze one more line out of a platform that has performed well all fortnight is exactly how a successful campaign turns into a search-and-recovery operation.

How to decide before you commit weeks of offshore time

We would not deploy a long-endurance AUV on a survey scope without working through the following, in writing, against the project’s acceptance criteria.

• Set the navigation budget against the survey order, then design aiding to meet it. Fix the required total horizontal uncertainty from the relevant IHO S-44 (Ed. 6.0) order — Exclusive Order being the tightest, followed by Special Order, with limits scaled to water depth and loosening through the lower orders — and work backwards. Specify the aiding cadence (surface GNSS interval, USBL fix rate, LBL coverage or terrain-aided navigation prior) needed to keep accumulated INS/DVL drift inside that figure for the full mission duration, not just per leg.

• Require an onboard data-quality watchdog with defined thresholds. The vehicle should monitor coverage, sounding density, sound velocity consistency and sensor health in real time, and flag or hold when acquisition falls outside acceptance limits. Treat its logs as part of the deliverable and review the threshold settings before mobilisation.

• Document the integrity-decision and abort logic, and audit it. Define the navigation-confidence, energy-reserve and system-health thresholds that trigger hold or abort, and require the vehicle to record the basis for each decision. This is the offshore-autonomy equivalent of an FMEA’s failure response, and class guidance for autonomous and remotely operated vessels (for example from DNV) provides a reasonable framework for structuring it.

• Scale endurance against your actual line plan. Take the published figure, apply your survey pattern’s manoeuvring fraction, add transit to and from the area, and carry a hard recovery reserve. Plan the productive bottom time from what is left, and resist consuming the reserve on the last line.

• Run a representative shakedown before the full mission. Before committing a sixteen-day deployment, run a shorter mission with the same sensor suite, line geometry and aiding architecture, recover it, and confirm the navigation solution and data quality held. The cost of a one- or two-day proving run is trivial against the cost of discovering a drift problem on day fourteen of a campaign.

• Specify the comms expectation honestly. Agree with the client what status information will be available during the dive, and make clear that survey-quality judgement happens at recovery, not in real time. Manage the expectation that an unattended platform cannot be steered like an ROV.

The Cellula and DRDC result is a real step, and the inclusion of 4,000 turns in the endurance figure is exactly the kind of honesty about usable performance the sector needs more of. For survey managers, the message is not that the platforms are nearly ready — it is that platform endurance is ceasing to be your problem, and the assurance case for what the vehicle does with all that bottom time is becoming the thing you have to engineer.

---

Based on: Cellula Robotics advances long-endurance AUV capability through Canadian defence collaboration