# Garmin Reactor 40 — PGN 126720 Reverse-Engineering Findings _Boat: Buttercup. Captured live on the NMEA 2000 bus (can0, 250 kbit/s) on 2026-06-20. Reactor 40 CCU + GHC control heads. All command encodings verified by button-press correlation; status stream structurally decoded for the first time._ ## Bus layout (this vessel) | N2K source | Device | Role | |---|---|---| | 2 | Reactor 40 CCU | Autopilot computer — **receives commands**, **broadcasts status** | | 4 | GPSMAP chartplotter | nav display — **sends the route/Go-To command (`0D`)** + nav PGNs 129284/129283/127258 | | 5 | GHC head | control head — **active controller; sends the hold-mode commands** (`02`/`05`/`11`) | | 0 | wind instrument | wind data (130306) | | 3 | GPS | position, COG/SOG | | 6, 7 | GHC heads / displays | inter-display sync | | 101 | SignalK (this plugin) | injects commands via canboatjs; also a registered controller (keepalive) | All autopilot traffic is **PGN 126720**, Garmin proprietary fast-packet, payload prefix `E5 98` = manufacturer code 229 (Garmin) + Marine industry. Inside that, the autopilot command/status group is marked by `10 17 04 04`. ## Commands (head → CCU, addressed to dst=2) — VERIFIED Container: `E5 98 10 17 04 04 ...` Heads continuously send a keepalive at ~2 Hz: subtype `15 03` (`E5 98 10 17 04 04 15 03 0X 00 C8 00`), where byte 8 differs per head (`02` from the active controller, `00` otherwise). Filter this out to isolate real commands. ### STATE / MODE — `E5 98 10 17 04 04 05 0A 00

`

The single `` byte selects the autopilot mode. Full map (verified live; the
pattern codes were confirmed in power-displacement vessel mode on 2026-06-21):

| Mode | code | Notes |
|---|---|---|
| Standby (disengage) | `02` | clutch released |
| Heading hold | `05` | clutch engages, display "Heading Hold" |
| Hard-turn / constant-rudder | `06` | **At the dock** it looks like heading hold (clutch in, "Heading Hold" shown). **Underway it pins the rudder hard over (~−27°) and circles the boat continuously** — NOT heading hold. Identity TBD (constant-rudder turn mode). |
| Wind hold (apparent) | `11` | sailboat; CCU then broadcasts wind-target field `00 0B`. Only **apparent** wind hold exists on this boat — no separate true-wind code (true wind needs a boat-speed source). |
| **Tack / Gybe** | `13` | sailing maneuver from wind/heading hold; preceded by selector `04 A2 00 ` (`dir` = turn direction). Tack and gybe send the **same** command — the CCU decides which from the wind geometry. |
| **Autopilot Setup Mode** | `07` | display "Autopilot in Setup Mode"; drives the rudder briefly (commissioning state), not a runtime steering mode |
| **Circles pattern** | `08` | preceded by selector `04 34 00 ` |
| **Zigzag pattern** | `09` | no direction selector (single engage) |
| **Williamson turn** | `0A` | preceded by selector `04 47 00 ` |
| **U-turn pattern** | `0B` | preceded by selector `04 6F 00 ` |

Codes `00,01,03,04` are inert (free wheel, no display change) and `0C`–`1F` read as
standby. After any engage, the active head emits a status ack `E5 98 10 17 04 04 25 02 00`.

> **Method note (2026-06-21):** codes `01`–`07` were verified live in power-displacement
> vessel mode by sending each through canboatjs and confirming by hand (clutch feel) + GHC
> display. This is the reliable method — the earlier read-only state-code probe mislabeled
> several codes (e.g. `06` and `07` looked inert in the telemetry but actually engage). The
> pilot was de-commissioned during these tests (compass-cal + speed-source alerts), so
> engaged modes pull the clutch but the steering interlock blocks active steering.

> ⚠️ Probing unknown state codes blind once drove the rudder into a stall and tripped a
> latched drive fault (cleared by an AP power-cycle). Don't fuzz state codes near the stops.

### STEERING PATTERNS (powerboat / power-displacement mode)

Each pattern = a **selector** frame then the **engage** state frame, sent together; the
`` byte (`00`/`01`) is the Port/Starboard choice offered on the head. Verified live
2026-06-21 on a calibrated pilot (patterns actually steer once the unit is commissioned).

| Pattern | Selector frame | Engage frame | Parameters |
|---|---|---|---|
| Zigzag | (none) | `... 05 0A 00 09` | amplitude + period (below) |
| Circles | `... 04 34 00 ` | `... 05 0A 00 08` | time (below) |
| U-turn | `... 04 6F 00 ` | `... 05 0A 00 0B` | none — direction only |
| Williamson turn | `... 04 47 00 ` | `... 05 0A 00 0A` | none — direction only |

(all frames prefixed `E5 98 10 17 04 04`)

**Pattern parameters** (set as their own frames before/while configuring):

| Parameter | Frame | Encoding | Captured |
|---|---|---|---|
| Zigzag amplitude | `... 00 1F 00 ` | radians (float) | `F3 66 DF 3E` ≈ 0.436 rad ≈ 25° |
| Zigzag period | `... 05 20 00 ` | **seconds** | `3C 00` = 60 s = 1 min |
| Circles time | `... 05 33 00 ` | **seconds** | `F0 00` = 240 s = 4 min |

The `00 1F` amplitude is the same float-field family as the wind target `00 0B`.

**Pattern cancel behaviour (verified repeatedly):** while *any* pattern is engaged, a
heading-adjust key press (`26 00`/`26 02`, ∓1°) **cancels the pattern and reverts to
heading hold** at the adjusted heading. Standby (`05 0A 00 02`) also exits.

**Steering keys (heading & wind hold):** the left/right arrows send only `26 00` (−1°) /
`26 02` (+1°) — desired-target selection, *not* direct rudder control. The **same** `26`
command is used in both modes; the CCU interprets it by context — desired **heading** in
heading hold, desired **wind angle** in wind hold (verified live underway, both modes). Only
±1 appears (no `26 01`/`26 03` ±10) when the head shows just the two arrows.

### GPS / NAV STEERING (requires an active waypoint)

These need an active Go-To / route on the chartplotter. Captured live underway 2026-06-21:

| Mode | Frames | Notes |
|---|---|---|
| Nav / Go-To follow | `... 05 0A 00 0D` | **sent by the chartplotter (src 04)**, not the GHC. `0D` reads inert if there is no active waypoint — that's why the dock probe saw nothing. |
| Orbit | `... 04 5B 00 ` + `... 05 0A 00 0F` | port/stbd |
| Cloverleaf | `... 04 3E 00 ` + `... 05 0A 00 0E` | port/stbd; length param `00 3D` (below) |
| Search | `... 04 65 00 ` + `... 05 0A 00 10` | port/stbd; spacing param `00 66` (below) |

**The autopilot runs the GPS pattern, not the chartplotter.** Confirmed live: after engaging
cloverleaf, *stopping navigation on the chartplotter left the pattern running* — the Reactor
holds the waypoint and the geometry itself; the plotter only seeds the initial waypoint.

So the pattern parameters ARE sent to the CCU, as float fields (same family as `00 0B` wind
target / `00 1F` zigzag amplitude), in **SI meters** (the Garmin UI shows feet):

| Parameter | Frame | Captured |
|---|---|---|
| Cloverleaf length | `... 00 3D 00 ` | `67 66 98 43` = 304.8 m = **1000 ft** |
| Search spacing | `... 00 66 00 ` | `0A D7 73 42` = 60.96 m = **200 ft** |

(An earlier note here wrongly claimed these params stay in the chartplotter — they don't; the
mistake was scanning for the integer `1000` instead of the float `304.8 m`.)

**Direction-key (`26`) rule (general, verified):** heading hold and wind hold *adjust the
target* (heading / wind angle) and stay engaged; in **every other engaged mode** — all
patterns *and* nav/Go-To follow — a `26` key press **cancels to heading hold**.

**Underway verification (2026-06-21, open water, calibrated pilot, helm manned).** Every
engage command was driven through the plugin and confirmed by the boat's behaviour:
heading hold (`05`) held course; wind hold (`11`) held the apparent wind angle; zigzag (`09`)
wove; circles (`08`) circled at a steady rudder; u-turn (`0B`) made a clean 180°; williamson
(`0A`) turned out and back. **U-turn and williamson auto-revert to heading hold on completion.**
The patterns are now engaged by the plugin's `pattern` PUT (`steering.autopilot.rudder.pattern`,
value `":"`), which sends the selector then the engage frame.

### POWERBOAT RUDDER STEERING — `E5 98 10 17 04 04 04 2A 00 `

Emitted by the head only in a power hull profile (planing/displacement), when the steer
keys are used while engaged ("rudder steering"). `` `00`/`01`. This is the powerboat
NFU/dodge path; not available in sailboat vessel mode. Actuation unverified — captured while
the pilot was de-commissioned (compass-cal + speed-source alerts blocked the drive).

### HEADING ADJUST — `E5 98 10 17 04 04 26 `

(raw, padded: `26  00 FF FF FF FF`)

| Button | code |
|---|---|
| −1° | `00` |
| −10° | `01` |
| +1° | `02` |
| +10° | `03` |

Scheme: low bit = step size (small/large), high bit = direction (−/+). On jorgen-k's GHC-20 the
large step was labeled ±15°; the **code is identical** (`01`/`03`) — the degrees come from the
head's configured increment, not the wire code. So our ±10 confirms and generalizes his ±15.

These match and extend the only prior public artifact (jorgen-k/signalk-autopilot-garmin),
now verified on Reactor 40 hardware.

## Status broadcast (CCU source 2 → 255) — NEW (nobody had decoded any of this)

The CCU broadcasts PGN 126720 at ~24 Hz. It is **not** a single mode enum; it is a stream of
**field sub-messages** in the `10 17 04 04` family:

`E5 98 10 17 04 04   0A  ...`

byte 6 = field id, byte 7 = type/length selector. Values are little-endian ints/IEEE-754 floats
(heading, rudder, gains, wind angle, etc.). There are also two other message types from the CCU:
`6C 07 02 02 ...` (low-rate summary) and head broadcasts `E7 08 0A 0A ...`.

### Engaged vs standby (from CCU status) — reliable

The CCU broadcasts engaged-markers `00 A2` / `02 74` / `00 72` at ~2 Hz **only while engaged**.
The provider counts them in a **3 s** window (≥2 ⇒ engaged); absence ⇒ standby. A tighter 1.3 s
window flapped between marker bursts (route↔standby every few seconds) and was widened.

The wind-tracking field `00 0B` (`E5 98 10 17 04 04 00 0B 00 `, the held apparent-wind
angle in radians) is present in **wind hold AND route follow**, absent in heading hold / standby.
It supplies the wind-angle value and serves as a fallback mode hint before the first command is seen.

### Reading the live AP mode — SOLVED (2026-06-21, verified live)

**The CCU never broadcasts its discrete steering mode.** A full PGN inventory of the CCU
(src 2) shows only 126720, 127245 (rudder), 127250 (heading), 127251 (rate of turn),
127257 (attitude), 61184 (sporadic proprietary), 60928 (address claim). The standard
mode-bearing PGN **127237 (Heading/Track Control) is not sent**. Inside 126720, a controlled
standby→heading→wind→route capture (route loaded the whole time) found **no field that
separates wind hold from route follow** — see the corrected note below.

**The mode is on the bus as the mode-set COMMANDS, not as status.** Every mode change is an
explicit command to the CCU (the boat forces a standby→engage for each change), so the latest
command *is* the current mode:

`E5 98 10 17 04 04 05 0A 00 `

| code | mode | sent by |
|---|---|---|
| `02` | standby | GHC (src 5) |
| `05` | heading / auto | GHC (src 5) |
| `11` | wind | GHC (src 5) |
| `0D` | route / nav | **chartplotter (src 4)** |
| `08/09/0A/0B/0E/0F/10` | patterns | whoever engages |

Captured live from src 5: `05 0A 00 02 / 11 / 05`. The provider now watches these frames from
**any** source and takes the latest as the mode, gated by the engaged-markers above for the
engaged/standby ground truth. Verified live across a full `auto → wind → standby → route → wind`
cycle with a route loaded throughout — every mode reads correctly, no flapping. This resolves the
old "can't tell pattern/wind/heading apart" gap.

### Corrected: what `00 0B` and `0E 02` actually mean

An earlier version of this doc claimed `00 0B` presence ⇒ wind mode, and (briefly) `0E 02` ≠
sentinel `71 17` ⇒ route. **Both are wrong on a sailboat with nav active.** `00 0B` streams in
wind **and** route (it is the apparent-wind angle the system tracks/displays, absent only in pure
heading hold). `0E 02` (and the plotter's 129284 / 129283, which *are* on the bus — an earlier
"no nav PGNs" claim was a hex typo, 129284 = `0x1F904`) track **a route being loaded on the
plotter**, which persists in every AP mode until you cancel nav on the plotter — so they flag
"route loaded," not "AP steering the route." Mode must come from the command stream above.

### Still to map (future)

Remaining 126720 status field-ids (heading, rudder angle, target heading) are live LE int/float
values; mapping them would let the plugin publish full `target.headingMagnetic` /
`target.windAngleApparent`. Not required for state/mode, which the command stream now covers.

## Artifacts (preserved on the Pi)

`~/dev/signalk-autopilot-provider-garmin/re/`
- `captures/cap_*.log` — raw candump for baseline, each button press, and each steady mode
- `apre.py` — fast-packet reassembler + `catalog`/`diff`
- `apcap.sh` — per-button capture+diff helper
- `status*.py` — status-stream field analyzers
- `modecap.py` / `combine6.py` — per-mode 126720 property tables + 6-way diff (the wind-vs-route study)
- `pgnsrc.py` — per-source PGN inventory (confirmed 127237 absent; found 129284 from src 4)
- `cmddump.py` — dumps AP-container commands from non-CCU sources (captured the `05 0A 00 ` mode-sets)

## Contribution plan

1. Open a Garmin-specific canboat issue (none exists) with an annotated visual-analyzer recording +
   the command table above + the status-stream structure.
2. Once the status field-ids are confirmed, PR canboat PGN definitions (follow the #633 pattern).
3. Upstream verified commands + status decode into jorgen-k's plugin and/or finish the V2 provider.

Thanks to Jörgen Karlsson (jorgen-k) for the original command encodings, and to Kees Verruijt
(canboat) and Scott Bender (signalk-autopilot) for the tooling and methodology.