Underwater Robotic Operations Without Connectivity

1. Regulatory and Compliance Framework

Subsea robotic operations are regulated through an overlay of maritime, environmental, and sectoral regimes. In offshore energy, the Bureau of Safety and Environmental Enforcement (BSEE) administers Outer Continental Shelf inspection requirements under 30 CFR Part 250, and equivalent regulators — the UK Health and Safety Executive's Offshore Major Accident Regulations, Norway's Petroleum Safety Authority, Brazil's ANP — impose pipeline integrity, riser inspection, and subsea isolation valve verification cycles whose evidence must be auditable. The International Maritime Organization (IMO) MASS framework and the relevant flag-state rules for unmanned and autonomous maritime systems are converging on requirements for evidence-grade mission records, governed deviation from approved mission profiles, and demonstrable safety-case compliance.

In ocean science, the UN Convention on the Law of the Sea (UNCLOS), the International Seabed Authority's exploration regulations, and national marine protected area rules require that scientific sampling be governed, documented, and reconstructible, with particular sensitivity in areas beyond national jurisdiction (BBNJ Agreement) and in seabed-mining contract zones. In defense, mission-recording obligations under each operating navy's autonomy-doctrine framework — the U.S. Navy's Unmanned Campaign Framework, the UK Royal Navy's NavyX program — require that autonomous decisions be traceable to the operating commander's intent and to a recorded delegation of authority.

What unifies these regimes is the structural expectation that an autonomous system operating beyond direct human supervision must be able to produce a credentialed record of what it observed, what it considered, what it decided, what authority constrained the decision, and what it did, even when the operating environment makes contemporaneous reporting physically impossible. Underwater is the canonical case: the regulator's expectation cannot be met by streaming telemetry, because the bandwidth does not exist. The expectation can only be met by a vehicle that carries its own governance with it and surfaces the lineage on recovery.

2. The Architectural Requirement

Underwater communication is one of the most challenging environments in engineering. Radio frequencies that enable megabit communication in air attenuate within meters of seawater. Acoustic modems provide communication at ranges of several kilometers but at bandwidths of single-digit kilobits per second with multi-second propagation delays. Optical communication works at short range in clear water but fails in turbid conditions.

The practical consequence is that autonomous underwater vehicles (AUVs) operating at depth are fundamentally disconnected from surface operators. A subsea infrastructure inspection AUV at several hundred meters depth on a multi-hour mission has, at best, intermittent acoustic contact with its support vessel. The contact provides enough bandwidth for status telemetry but not enough for real-time guidance or mission replanning. The vehicle is, for practical purposes, on its own from launch to recovery.

The architectural requirement that follows is that the vehicle must carry, in resident memory, not merely its mission script but its mission semantics: the objectives, the governance constraints, the authority taxonomy under which deviations are permitted, the criteria for promoting an opportunistic observation to an investigation, and the lineage discipline that makes the mission record auditable on recovery. This is not a matter of better firmware. It is a matter of architectural shape: the vehicle must execute its mission as a persistent, self-evaluating semantic object rather than as a fixed waypoint program with conditional jumps.

Three properties define the requirement. First, persistence — the execution context survives intermittent power events, sensor faults, and acoustic dropout without losing its continuity. Second, self-evaluation — the vehicle continuously evaluates its own state against the mission's governance criteria and proposes mutations rather than waiting for external commands. Third, governed mutation — proposed deviations from the planned profile pass through a credentialed admissibility check whose authority taxonomy is itself resident on the vehicle, so that "is this deviation permitted" is an architectural question, not an operator question.

3. Why Procedural Compliance Fails

Current AUVs handle the connectivity problem through pre-programmed survey patterns. The vehicle follows a lawnmower path over the survey area, collects data at predetermined intervals, and surfaces for data offload and mission update. This approach works for uniform survey tasks but fails for inspection missions where the AUV needs to deviate from the pre-planned path to investigate anomalies detected during the survey. Pre-planned survey patterns collect data uniformly regardless of what the data reveals.

An AUV inspecting a subsea pipeline that detects a potential anomaly cannot pause to investigate it further unless the pre-planned mission included contingency behaviors for that exact scenario. Instead, the anomaly is flagged in the collected data, the vehicle completes its pre-planned path, surfaces, offloads data, and a human operator reviews the data and programs a follow-up mission to investigate the anomaly. This cycle of survey, surface, review, and re-deploy wastes hours or days of operational time and vessel cost for each iteration. In offshore energy operations where vessel day-rates exceed six figures, these delays have direct financial impact. In defense applications where underwater survey results are time-sensitive, the delays have operational impact.

The procedural compliance answer to this — more comprehensive pre-mission risk assessments, more elaborate contingency tables, more conservative mission profiles — does not solve the architectural problem. It compounds it. Each additional contingency table adds branches to a static program that still cannot synthesize new branches in response to genuinely novel observations. The program is more elaborate; it is not more cognitive. And every regulator-mandated contingency that the vehicle did not exercise still consumes mission-design effort and post-mission documentation effort.

The deeper procedural failure is that mission records produced by static-program AUVs do not exhibit the structural property regulators are converging on. The post-mission file shows what the vehicle did. It does not show what the vehicle considered and rejected, what governance constraint applied, or what authority would have permitted a different choice. A regulator asking "why did the AUV not investigate the pipeline anomaly when it was within fifty meters of it" gets the answer "it was not in the pre-planned profile" — which is procedurally complete and architecturally vacuous.

4. What the AQ Memory-Resident-Execution Primitive Provides

The Adaptive Query memory-resident-execution primitive disclosed under USPTO provisional 64/049,409 specifies the mission as a persistent semantic object that resides in the vehicle's working memory throughout the dive, continuously self-evaluates against its objectives and governance constraints, and proposes governed mutations of its own execution plan in response to observations. The primitive is technology-neutral with respect to the underlying compute substrate (any onboard processor, any persistence mechanism) and composes with the vehicle's existing navigation, sensor processing, and control stacks.

When the AUV detects a potential anomaly, the execution object evaluates the observation against its mission objectives, proposes a plan mutation to investigate the anomaly, validates the mutation against its governance policy including time constraints, battery reserves, and safety margins, and executes the investigation if the mutation is approved. The AUV does not need to surface for human approval. The governance that constrains the investigation, how much time to spend, how close to approach, what safety margins to maintain, is embedded in the execution object. The decision is governed, auditable, and fully recorded in lineage.

For multi-day missions, the execution object accumulates observations across the entire mission, identifying patterns that would only be visible to a human analyst reviewing the complete dataset after retrieval. The AUV can refine its survey strategy as it collects data, focusing attention on areas of interest and reducing time spent on areas that initial passes reveal to be unremarkable. The persistence property ensures that an intermittent fault — a sensor reset, a brief power excursion, an acoustic dropout — does not collapse the accumulated mission context, because the execution object's state is preserved as a credentialed continuity rather than reconstructed from scratch on each cycle.

The governed-mutation property is what distinguishes memory-resident execution from a sophisticated state machine. Each proposed deviation passes through a credentialed admissibility check against the resident authority taxonomy. The taxonomy is itself a signed object placed onboard at mission start by the authorizing operator, so the operator's intent is structurally present in every autonomous decision the vehicle makes during the dive. On recovery, the vehicle's lineage record shows every observation, every proposed mutation, every admissibility evaluation, and every actuation, with the credential of the authority under which each step occurred.

5. Compliance Mapping

Memory-resident execution maps directly onto the structural expectations of the major subsea regulatory regimes. For BSEE OCS pipeline inspection, the recovered lineage record demonstrates not only that the inspection was performed but that the vehicle considered each detected anomaly, evaluated it against its inspection-priority taxonomy, and either investigated or deferred under a credentialed authority. The integrity-management file is generated by the architecture; it is not assembled after the fact from sensor logs and mission-design documents.

For IMO MASS-aligned operations, the resident authority taxonomy is the structural embodiment of the operating commander's intent and the safety case. Deviations from the approved mission profile that fall within the taxonomy's permitted scope are governed autonomous decisions, recorded as such; deviations that would fall outside the taxonomy are structurally refused or downgraded, again recorded as such. The flag-state regulator inspecting the post-mission record sees a chain of credentialed decisions rather than a list of waypoints.

For ISA seabed exploration and BBNJ-relevant scientific sampling, the lineage record satisfies the convention's expectation that beyond-jurisdiction sampling be governed and reconstructible. For defense missions, the resident authority taxonomy carries the delegation of authority from the operating commander, and the lineage record on recovery is the structural artifact that demonstrates compliance with the mission directive without requiring real-time communication that would compromise operational security.

6. Adoption Pathway

Adoption of memory-resident execution does not require replacing existing AUV vehicle stacks, mission-planning tools, or post-mission analysis pipelines. The primitive sits between the mission planner and the vehicle's autonomy stack as a persistent execution substrate. Existing planners (NaviSuite, Greensea, Mission Oriented Operating Suite, vendor-proprietary tools) continue to be the operator-facing surface for authoring the mission semantics. Existing post-mission analysis tools continue to consume the recovered data. What changes is that the mission is loaded onto the vehicle as a self-evaluating semantic object rather than as a waypoint program, and the recovered record is a credentialed lineage rather than a sensor log.

The pragmatic adoption sequence begins with a single inspection class — most plausibly subsea pipeline integrity inspection in offshore energy, where the economic case is strongest because each avoided re-deployment cycle is a six-figure vessel-day saving. An operator (a pipeline owner, an inspection contractor, a vehicle OEM) instruments one AUV class with the memory-resident-execution substrate, runs a campaign cycle, and compares the recovered lineage records and the dive-cycle count against a comparable prior campaign. The integrity-management file emerging from the architecture is the artifact the regulator and the operator's own assurance function can both read.

For offshore energy operators, this reduces the number of dive cycles needed to complete an inspection from multiple pre-planned dives to a single adaptive dive. The AUV investigates anomalies during the survey rather than flagging them for follow-up. The operational cost reduction is proportional to the number of avoided re-deployment cycles. For ocean science, memory-resident execution enables AUVs to conduct adaptive sampling missions where the sampling strategy evolves based on what the instruments detect. An oceanographic AUV that detects an unexpected temperature gradient can investigate the gradient's extent and structure during the same dive rather than completing a pre-planned grid and returning for a targeted follow-up. For defense, underwater survey missions gain the ability to adapt to discovered conditions without surfacing to communicate with operators, maintaining operational security while increasing the quality and completeness of the survey data collected per mission.

The honest framing — the AQ primitive does not replace AUV autonomy stacks. It gives subsea autonomy the execution substrate the regulatory environment is converging on and that pre-programmed survey patterns cannot structurally provide.