Every decision, documented.
This page is our open engineering notebook: the requirements, trade studies, analyses, manufacturing and test results behind KH-01 — the reasoning, not just the render.
Design drivers.
Before a single curve was drawn, the mission was translated into hard engineering requirements.
| Requirement | Driver | Target |
|---|---|---|
| Sprint speed | Speed challenge task | ≥ 4 m/s |
| Course keeping | Channel navigation & gates | ± 0.5 m lateral error |
| Station keeping | Precision / docking tasks | ± 0.3 m, heading ± 10° |
| Obstacle avoidance | Buoy field task | Detect & avoid at ≥ 15 m |
| Runtime | Full mission without swap | ≥ 60 min |
| Transportability | Shipping to venue | 2-person carry, airline-checkable crate |
| Watertight integrity | Electronics survival | IP68 bay, leak alarm < 5 s |
Why a catamaran?
Three hull configurations were scored against the requirements that matter most for RoboBoat: sensor stability and usable deck area, not just raw speed.
| Criterion (weight) | Monohull | Catamaran | Trimaran |
|---|---|---|---|
| At-rest stability (×3) | 2 | 5 | 4 |
| Deck area (×3) | 2 | 5 | 3 |
| Drag at sprint speed (×2) | 4 | 3 | 4 |
| Manufacturability (×2) | 5 | 4 | 2 |
| Handling & transport (×1) | 4 | 4 | 2 |
| Weighted total | 31 | 46 | 31 |
Scores 1–5 (5 = best). The catamaran wins because perception quality and mission payload — which drive scoring across nearly every task — dominate over top speed.
Hulls shaped by CFD, not by fashion.
The demihull geometry was parametrized and iterated in CFD: length-to-beam ratio, entry angle and demihull spacing were swept to minimize total resistance at mission speed. Wave interference between the hulls was mapped to pick a spacing that keeps drag flat through the sprint regime.
[Add: solver used, mesh independence study, and the measured drag reduction versus the first concept — with the drag-vs-speed curve opposite.]
A sandwich built like an aircraft part.
The hulls are carbon-fiber skins over a PVC foam core, vacuum-infused in female molds and post-cured in-house. The layup is tuned per region — heavy where loads concentrate, minimal where weight is free savings.
| Region | Layup | Core |
|---|---|---|
| Hull skin (general) | 2 × 200 g/m² twill carbon [0/90] | 5 mm PVC, 80 kg/m³ |
| Hull skin (impact zone) | 3 × 200 g/m² twill carbon | 5 mm PVC, 100 kg/m³ |
| Deck | 2 × 200 g/m² twill carbon | 8 mm PVC, 60 kg/m³ |
| Cross-beam | 4 × 200 g/m² twill carbon ±45° | 15 mm PVC, 80 kg/m³ |
| Thruster mounts | 5 × 200 g/m² twill carbon + unidirectional tapes | Local solid carbon |
Thrust budgeted, power measured.
Thrusters were selected from bench thrust tests, not datasheets — and the power budget below is verified with logged current draw from lake runs, so endurance claims survive contact with reality.
| Consumer | Typical draw | Peak draw |
|---|---|---|
| 2 × thrusters (cruise) | 180 W | 600 W |
| Jetson Orin NX + peripherals | 25 W | 45 W |
| Sensors (camera, LiDAR, GNSS, IMU) | 15 W | 20 W |
| Network & telemetry | 8 W | 10 W |
| Total | 228 W | 675 W |
[Add: battery capacity, measured endurance at mission speed, and the thrust-vs-throttle bench curve. All figures above are draft placeholders.]
Simulate first. Swim second.
The autonomy stack is a layered pipeline on ROS 2: every layer — estimation, perception, planning, control — runs identically in simulation and on the real vessel. New mission logic is proven in sim, then on the bench with hardware-in-the-loop, and only then on the water.
- Perception: camera + LiDAR fusion for buoys, gates and obstacles
- Planning: behavior-tree mission execution, A*-based routing
- Control: LOS guidance, PID heading loop, thrust allocation
- Simulation: full-mission rehearsal before every water test
The test matrix doesn't lie.
Every KH-01 subsystem passes a written test matrix before it is allowed near a mission run. Results are logged, versioned and tied to the hardware revision that produced them.
| Test | Environment | Pass criteria | Status |
|---|---|---|---|
| Buoyancy & trim check | Test basin | Design waterline ± 10 mm | Scheduled |
| Max speed run | Lake | ≥ 4 m/s over 50 m | Scheduled |
| Waypoint navigation | Lake | ± 0.5 m at all gates | Scheduled |
| Obstacle avoidance | Lake (buoy field) | Zero contacts, ≥ 15 m detection | Scheduled |
| Station keeping | Lake (wind ≥ 3 m/s) | ± 0.3 m, heading ± 10° | Scheduled |
| Remote e-stop | bench + lake | Cutoff < 1 s at 30 m range | Scheduled |
| Full endurance run | Lake | ≥ 60 min mission profile | Scheduled |
Versions, not accidents.
| Revision | Change | Reason / outcome |
|---|---|---|
| V0.1 — foam concept | CNC-machined foam hull, hand tools | Float-and-fit checks; validated general geometry |
| V0.2 — glass prototype | Glass-fiber layup, first wet electronics | Exposed water-ingress paths at cable pass-throughs |
| V1.0 — carbon competition hull | Vacuum-infused carbon sandwich, sealed bay redesign | Target platform for RoboBoat 2027 (in build) |
[Extend this table as revisions land — judges read iteration history as engineering maturity.]
Problems we met, and how they died.
Compass deviation near thrusters
Symptom: heading drift under load.
Fix: IMU relocated on a carbon mast, soft-iron calibration added, motor cables twisted & shielded.
Water ingress at cable glands
Symptom: moisture in the electronics bay after hard runs.
Fix: machined gland plates, potting compound, and a pressurized dunk test before every event.
Vision failure in sun glare
Symptom: buoy detections dropped on low-angle light.
Fix: polarizing filter, exposure control, and LiDAR-weighted fusion fallback when camera confidence drops.
[Replace these with the team's real debugging stories — specific, honest problems read as strength, not weakness, to technical judges.]