Who’s Looking? The Bracelet Knows.

The attentive reader of these pages — and we are grateful for the attentive reader of these pages — has by now built up a fair picture of what the Galvanic Pulse bracelet does. Man-overboard detection by negative signal, covered in Schrödinger’s Watchkeeper. Fatigue and sleep-state inference. A gentle pulse on the wrist to deliver a watch reminder, or an acknowledgment gesture from the wearer back to the boat. A small, quiet thing on the wrist, doing several quiet jobs.

It may have escaped the less attentive reader, however — and we suspect the less attentive reader is most of us — that the bracelet does one more thing, and that this one more thing is, in many ways, what binds the rest of the system together. The bracelet is, in addition to everything else, the boat’s way of knowing where each person on board is, and who each person is, at every moment.

A Brief Recap — What the Bracelet Already Does

For the reader who has not yet read the other pieces — and as a quick context-setter for the rest of this post — the Galvanic Pulse already carries four well-documented jobs:

• Man-overboard detection by negative signal. The bracelet maintains a continuous low-power radio conversation with the boat; the conversation ending — because the wrist has left the vessel — is the MOB alarm. No button press required of the casualty.
• Fatigue and sleep-state inference. Accelerometry on the wrist drives a sleep / wake classifier; the long-term integral of activity informs the fatigue trace that the watch-keeping schedule respects.
• Notifications. A quiet vibration on the wrist when the Galvanic Voice has something to say — a watch reminder, a CAUTION-level event, an anchor note — addressed only to the relevant person rather than broadcast across the boat.
• Acknowledgments. A wrist gesture sent back to the system: the wearer has registered the alert and is dealing with it.

That is the version of the Galvanic Pulse story we have been telling. It is, deliberately, only part of the story.

The Quiet Fifth Job: Knowing Where Each Person Is

Every Galvanic Voice unit installed on the boat is, at the same time, a Bluetooth receiver. As each bracelet broadcasts its presence many times a second, each Voice unit in earshot measures the strength of the signal it receives — what radio engineers call the RSSI, the received-signal-strength indicator. A bracelet close to a Voice unit produces a loud RSSI on that unit; a bracelet two cabins away produces a quieter one; a bracelet on deck while the Voice unit is in the saloon produces something in between.

With multiple Voice units installed at known locations on the boat — the helm, the cockpit, the saloon, the forward cabin, the aft cabin — each one is reporting the RSSI of every bracelet at every moment. A fusion process running on the boat consumes those streams and asks the right question: given that this bracelet appears loudest on the aft-cabin Voice unit, second-loudest on the saloon unit, and very quiet on the forward-cabin unit, where is the bracelet? The answer is the aft cabin.

The technique is multi-station BLE-RSSI triangulation, and it produces, at every moment, a vessel-relative location for every bracelet on board — not a GPS position, but a categorical room or zone: helm, cockpit, saloon, forward cabin, aft cabin, head, foredeck. The infrastructure cost is zero, because the Voice units doing the receiving are the same Voice units already doing every other job on the boat. The Bluetooth receivers we needed for the MOB conversation are also, without any additional hardware, the localisation receivers. (The multi-station crew-localisation method is the subject of a pending patent application in the Galvanic Works portfolio.)

How It Actually Decides Who Is Where

There are three classifier variants in use, in increasing order of sophistication, picked according to how many Voice units are installed and how cleanly the boat’s interior geometry separates them:

• Nearest-station argmax. The bracelet is in the zone whose Voice unit reports the strongest current RSSI, provided the lead over the next-best unit exceeds a sensible margin (we use six decibels). This is the workhorse classifier — robust, computationally trivial, and sufficient for room-level resolution on boats with a Voice unit in each major zone.
• Path-loss multilateration. Each station’s RSSI is converted to a distance estimate using a calibrated log-distance path-loss model; three or more stations let the system fix a two-dimensional vessel-relative position by least-squares. Useful when zones are not naturally separated by bulkheads — on an open-plan motor yacht, for example.
• Fingerprinting. During a one-time walk-through at commissioning, the captain wears a bracelet and visits every named zone in turn; the system records the per-station RSSI signature of each zone. At runtime, a bracelet’s instantaneous RSSI vector is matched to the recorded signatures. This is the most accurate variant, and the one used on boats whose interior geometry is unusual enough that general-purpose classifiers struggle.

Whichever classifier is in use, the output is gently smoothed by a short mode filter — a brief walk through the galley does not redesignate the wearer as “in the galley” if they were sleeping forward two seconds before and will be sleeping forward two seconds after.

We will be honest about one thing here: localisation quality depends on the boat. A monohull with the Voice units in well-separated cabins, behind real bulkheads, produces clean, unambiguous classifications. An open-plan catamaran with one large saloon and a dozen sight-lines is a harder problem, and the system uses path-loss multilateration there to compensate. A very large yacht with many zones may need fingerprinting to disambiguate adjacent zones reliably. The system picks the right classifier for the topology, and tells the captain how confident it is.

Where You Sleep, and the System That Knows It

One of the most useful things the boat can do with the per-bracelet location stream is to answer a question the captain otherwise has to remember: which cabin is each crew member sleeping in tonight?

The honest classifier here is a particularly elegant piece of the system. The bracelet’s instantaneous location can flicker — a person briefly visits the galley, walks through the saloon, fetches a jacket from a cabin that is not theirs. Sampling at every moment would produce a noisy, useless “current cabin” output. The right question to ask is not “where is this person now?” but “where does this person actually sleep?” — and the right answer is the time-weighted majority of nearest-Voice classifications gated by the bracelet’s own sleep-state signal. A person who lies down in a cabin and sustains the sleeping classification for hours is sleeping in that cabin; a person who briefly steps into the cabin to fetch a jumper is not.

Once the assignment is established, it is used. When the Galvanic Voice needs to wake the watch-on-call for the next shift, it does not ask the captain to specify which cabin to address — the system already knows. When an emergency-level alert escalates from the wrist to the whole boat, it speaks first into the cabin of the relevant crew member rather than broadcasting indiscriminately into the off-watch. When the captain needs to be woken — and only the captain — the alert is routed to the cabin the captain has actually been sleeping in tonight, not the cabin labelled “captain” on the architect’s drawings.

You may even change cabins every night. The system will track it. (We know you don’t. But it would, if you did.) (The auto-cabin-assignment method, gated by the bracelet’s sleep state, is also the subject of a pending patent application.)

Identity — Who Did That, and When

A second use of the bracelet that we want to be plain about is identity attribution. The Voice unit that delivers an alert knows which bracelets are present in its zone at the moment the alert is acknowledged. When somebody at the helm clears an anchor-watch CAUTION with a wrist gesture, the system records not only “acknowledged” but “acknowledged by the wearer of bracelet X, at the helm, at this timestamp”. Every interaction with the system is therefore attached to a person and a moment.

Two practical consequences flow from this.

One is forensic. When something has gone wrong on a passage and the question afterwards is “who cleared the anchor alarm at half past two that famous night?”, the answer is in the log. Not as accusation — as record. Boats are crewed by humans, humans make decisions, and the decisions ought to be retrievable afterwards by the people who care about them. The chartplotter, the autopilot, the alarm panel of the average modern cruiser have, on this score, no memory at all.

The other is operational. The captain can, at any moment, see who is on deck and who is not. Who is in the cockpit at three in the morning — the watch, presumably, but the system shows it rather than infers it — and who is below in their cabin, asleep, with their fatigue trace continuing to recover. The captain who is the only person up at first light, three hours before anyone else stirs, can know, with no further investigation, that nobody else is up. The skipper who has just heard a thump on deck can know whether somebody is up there or whether the boat is making the noise on its own.

Crews on a long passage talk about a particular kind of ambient awareness that takes years to develop — “I knew someone was up there before I heard the winch”. The bracelet gives the boat that awareness directly, without the years.

The Bigger the Boat, the Sharper the Picture

Worth saying out loud: the more Galvanic Voice units a boat carries, the sharper the localisation becomes. A single Voice unit on a forty-foot cruiser still distinguishes on deck from below deck reliably; a Voice at the helm and a Voice in the saloon distinguish helm-from-cockpit-from-saloon-from- cabin; a yacht with a Voice in every named cabin produces zone-level localisation with very little ambiguity.

The geometry of the boat matters as much as the count. Bulkheads attenuate Bluetooth signals; sight-lines do not. A monohull’s natural compartmentalisation does the classifier’s work for it. An open-plan catamaran with one large saloon and no walls between zones is a harder problem, solved by path-loss multilateration plus, where useful, a one-time fingerprinting walk-through at commissioning. The system picks the right classifier for the topology of the boat it is installed on, and the captain sees a confidence indicator in the app.

Why This Matters to the People on Board

For the captain, it means a boat that understands where the crew is — and therefore who is the watch, who is on call, who is in the cabin that needs to be woken next, who has just stepped on deck and who is still below. Decisions about routing, escalation and attention are made against a real picture of the crew’s whereabouts, not a guess.

For the crew member, it means a boat that addresses them by name and role, with no menu to navigate and no profile to set up — and that bothers them only when the alert is for them. The off-watch sleep is genuinely the off-watch sleep. The watch-on is woken to take over and the off-watch is not.

For the family or charter guest, it means a boat whose alerts are credible because they arrive at the relevant person rather than broadcasting through every cabin. The child asleep forward is not woken by a WARNING addressed to the captain aft. The system respects sleep where sleep should be respected, and breaks it only where it must.

For the fleet manager or operator, it means a tamper-evident record of who interacted with the system, when, and from where on the boat. The post-incident question is answerable from a log, not a memory.

The bracelet is, in the end, the boat’s way of paying attention to each person aboard — and the rest of the system uses that attention to make every other thing it does more accurate, more targeted, and more respectful of the people it is built to serve.

References

1. Bluetooth SIG. Bluetooth Core Specification, version 5.x. The standard governing the BLE advertising and RSSI measurement on which the localisation pipeline rests.
2. Rappaport, T.S. Wireless Communications: Principles and Practice. 2nd ed. Prentice Hall, 2002. (The standard reference for the log-distance path-loss model underlying the multilateration classifier.)
3. Kalman, R.E. “A New Approach to Linear Filtering and Prediction Problems.” Transactions of the ASME — Journal of Basic Engineering, 1960. (The recursive estimator used to smooth the per-bracelet location stream over time.)
4. International Maritime Organization. Resolution MSC.302(87): Adoption of Performance Standards for Bridge Alert Management. (The IMO framework governing alert routing and acknowledgment — the principle of targeted rather than broadcast alerting the localisation enables.)