Predictive Wave Autopilot & 3D Wave Dynamics

An autopilot that sees the wave coming —
and steers before it hits.

Conventional autopilots react after the bow gets pushed off course. This system reconstructs the sea surface from three wireless motion pods, predicts the wave torque 10–15 seconds ahead, and applies counter-rudder before the hull ever slews. The feature racing systems sell for five figures — built open, on your own boat.

Reference kit: ≈ $650 · open-source core (pypilot) · your data stays on your boat

The software, running. Pick a sea below: each is one short recording, the twin switching into that sea and the pilot settling into it.

Recorded in demo mode. No sensors, dongle or helm were connected — every reading is produced by the vessel model inside the server, through the same code path the hardware will use. Nothing here is a sea trial. Rendered on the GPU at 60 frames per second of simulated time.

The cockpit digital twin: the hull you have, the sea state the engine measured, and where physics says you'll be in ten seconds.

§How it works, end to end

3 wireless IMU pods→ µs-class relative clock→ wave field estimator→ 10–15 s torque prediction→ feedforward into pypilot→ SeaStar hydraulic ram

50 Hzpod quaternion stream

<1 mspod → helm latency

10–15 sprediction horizon

200 mswatchdog → PID failover

30–50 huntil your boat's model is baked

§What makes it different

Predictive feedforward steering

Spatial phase between bow, midships and aft pods resolves wave direction and encounter timing. The rudder moves 1–2 seconds before peak torque — because a 4–8°/s ram can't win a fight it enters late.

Microsecond-class relative sync

Pods latch hardware timers the instant a radio sync beacon lands (Nordic PPI — silicon, no software path). No GPS, no NTP, no drift: one shared relative timeline, which is all wave math ever needed.

Five sea-state experts, not one model

Docking, flat motoring, quartering swell, short chop, heavy downwind — each regime gets its own specialized model and gain profile, hot-swapped by a supervisory classifier with hysteresis. No monolithic net averaging contradictory physics.

Fail-safe by architecture

The predictive engine injects into stock pypilot — it never replaces it. Engine hiccups? A 200 ms watchdog drops the helm into pypilot's native PID with rudder power uninterrupted. Three layers deep, each simpler than the one above.

The rudder is a sensor too

Motor current on the hydraulic pump reads as water load on the blade: wave impacts verified at the rudder before the hull turns, weather helm quantified from baseline drift. Hardware you already own becomes instrumentation.

Your boat learns itself

Start from a hull-archetype prior, shaped by the 3D “My Boat” configurator. Over the first 30–50 underway hours the cloud identifies this boat's response operators — then updates arrive as signed, per-regime artifacts that can never break a proven behavior.

§Supported hardware

Tier 1 — officially supported & tuned: Dometic SeaStar -3 reversible pump + balanced SeaStar ram on the pypilot high-current controller. Ships with pre-configured hydrodynamic profiles and calibrated slew tables.

Tier 2 — community: any pypilot-supported drive. The prediction engine emits standard heading/rudder deltas; you own the tuning.

§Buy Autopilot coming soon

The hardware packages — the three motion pods, the dongle and the boat computer, with pypilot and the prediction engine installed — will be sold here. The store is not open yet; until it is, the reference kit and its bill of materials are on the kit page, and an engagement can be booked from the front page.

Read the 4-page design document → See the kit & bill of materials → Follow the build log

An autopilot that sees the wave coming —and steers before it hits.