Twin-screw docking practice in real San Juan Islands marinas.
At the end of August 2026 I'm supposed to captain Bonum Vitae — a Grand Banks 52 Heritage Motoryacht, about 58,000 pounds of her — through the San Juan Islands with family and friends aboard. I trained on similar boats about five years ago, but five years is enough time for confidence to turn back into theory, especially when the theory involves easing 29 tons into a narrow slip while a Pacific Northwest tidal current negotiates a different arrangement.
So I did what any reasonably anxious programmer might do: I built a simulator.
This was never meant to be a general boating game. It's rehearsal. The cruise runs Bellingham → Sucia → Stuart → Roche Harbor → Friday Harbor → Jones → Cypress → Bellingham, and the moments that occupied my mind were never the open-water legs — they were the slow minutes at either end: backing out of a side-tie, entering an unfamiliar basin, judging leftover momentum, making three small corrections before one big one becomes necessary.
A twin-screw motoryacht doesn't steer like a car at those speeds. Port engine ahead swings the bow to starboard. A handed propeller in reverse walks the stern sideways. The bow thruster shoves the bow, not the boat. And when the levers come back to neutral, 58,000 pounds does not come back to neutral with them. That inertia — not which lever, but how early and how long — is the thing this simulator exists to practice.
My first instinct was maximal fidelity: buy the real thing, a Glendinning CH2001 twin-lever control head — the hardware that wouldn't feel like a game controller because it isn't one. That idea did not survive contact with the integration details. Real marine control heads cost upward of $1,500 and speak CAN bus / J1939, not USB. Wiring one into a browser would have meant gutting a precision piece of boat equipment and rebuilding its electronics — at which point the project had drifted from "practice docking" to "learn embedded electronics by modifying safety-critical marine hardware." I was over my skis. The point was to reduce anxiety, not manufacture a new category of it.
Plan B was gloriously plausible: a $200 mechanical dual-lever marine throttle, housing cracked open, 10k potentiometers on the lever pivots, and a zero-solder Leo Bodnar USB board to present it all as a standard game controller. I still like that plan. It is also its own hardware project, and I wanted to know whether the simulator was useful before fabricating a helm for it.
For a while the helm was a Thrustmaster TWCS — a single flight-sim slider pressed into twin-screw service by mapping its one axis as a split throttle: mid-travel is dead neutral, above it drives the port engine ahead, below it the starboard engine astern. A ridiculous compromise (you can't command most port/starboard combinations), but it proved the sim out, and that split-slider mode still lives in the input layer.
Where it landed: a Thrustmaster TCA throttle quadrant. Designed for an Airbus, but it has the two things that matter — two independent physical levers, and a browser can read it through the Gamepad API with zero drivers. The sim handles the airplane-to-boat mismatch: swap-levers for reversed axes, idle-detent calibration so lever travel above the detent means ahead and below means astern, and the quadrant's switches double as engine masters and ignition.
No hardware? Keyboard works: W/S port throttle, I/K starboard,
A/D bow thruster. (Use Chrome for the quadrant — Safari never exposes
it.)
- Physics that punish impatience — per-engine thrust with prop walk, bow thruster at the bow, wind and current per marina, quadratic hull drag, and honest inertia. Telemetry shows SOG, heading, drift, and yaw rate.
- Seven real anchorages from the actual cruise plan, with layouts traced from harbor maps and OSM: Squalicum, Sucia (Fossil Bay), Reid Harbor, Roche Harbor, Friday Harbor, Jones Island, and Eagle Harbor. Each has arrival/departure exercises against real berths.
- Consequences — hit something hard enough and the hull takes damage: scuff, minor, major, severe, judged by closing speed at the point of contact. Gelcoat cracks stay on the hull; splintered timber stays on the dock. Restarting the exercise repairs everything, which is cheaper than the real arrangement.
- A lived-in marina — finger slips are two-thirds full of moored boats, the occasional small craft runs the fairway in or out, and the VHF panel murmurs synthesized harbor chatter if you leave it monitoring.
- Guidance that respects land — the wayline is a live shortest-water-path (A* over a per-marina navigation grid) from wherever you are to the berth, replanned as you drift. It goes around docks, pilings, and moored boats, never through them.
- Turbo mode — press
Tat full throttle. It is not seamanship.
npm install
npm run devOpen http://localhost:3000, pick a marina and an exercise, START ENGINES,
and try to dock without appearing in the incident log. H hides the panels.
The Electron desktop build (npm run app:dev / npm run app:build) wraps the
static export for offline use aboard.
Next.js (App Router) · React Three Fiber + drei · Rapier physics · WebAudio
for engine and radio sound · TypeScript. Marina layouts are plain data in
src/lib/marinas/; the boat's force model lives in
src/lib/sim/boat-physics.ts; collision damage in
src/lib/sim/collision-damage.ts; water pathfinding in
src/lib/sim/water-nav.ts.
Built with heavy assistance from coding agents, which turned out to be a fitting crew: they also needed to be told, repeatedly, which way the bow swings when the port engine goes ahead.