Fragmented
Tracking, wind, sensors, video and coaching are typically analysed independently.
Doctoral Research Concept · Competitive Sailing
Make the Winning Boat
Can a sailing system learn how to win?
A doctoral research concept combining 60,000+ tracked races, real-time sensing, computer vision and artificial intelligence to understand the environment, improve decisions, optimize the boat and turn every race into the next training input.
- 60,000+
- Tracked races
- 6
- Research modules
- 1
- Closed learning loop
The research system
Six questions. One learning loop.
Each module creates evidence for the next. The final module turns measured gains and losses into the next training objective—and new data for the entire system.
- 01Know the venue
What normally happens here?
- 02Understand now
What is happening right now?
- 03Make the right decision
Where should we go?
- 04Make the boat faster
How do we maximise performance?
- 05Execute better
How do we minimise losses?
- 06Learn↺
What should we improve next?
- ↺
Learn returns evidence to modules 01–05. Every race becomes input for the next performance improvement cycle.
Research gap
Sailing generates data. It rarely learns from it.
Historical knowledge is underused
Thousands of previous races contain environmental and tactical patterns that are rarely systematically reused.
No closed learning loop
Most systems explain what happened. Few determine what should change and validate whether that change actually improves performance.
Research framework
Six work packages connect knowledge to performance.
WP01
Know the Venue
What normally happens here?
Analyse a potential corpus of 60,000+ historical tracked races together with weather, geography and environmental context.
- Data
- Race Tracks · Weather · Wind · Geography · Results
- Methods
- Trajectory Analysis · Spatial Modelling · Pattern Recognition
- Output
- Venue Intelligence Model
- Validation
- Prediction of venue-specific patterns on unseen races.
WP02
Understand Now
What is happening right now?
Estimate the live environmental and boat state from noisy measurements collected on moving platforms.
- Data
- Wind · GNSS · IMU · Waves · Current · Boat State
- Methods
- Sensor Fusion · State Estimation · Signal Processing
- Output
- Live Environmental Model
- Validation
- Comparison with high-quality reference measurements.
- True wind
- 12.4 kn / 218°
- GNSS
- 7.2 kn · COG 031°
- Heel
- 17.2°
- Pitch
- 2.8°
- Wave
- 0.6 m · 4.2 s
- Current
- 0.4 kn / 164°
WP03
Make the Right Decision
Who is actually winning—and what should happen next?
Combine historical races, the current environment and fleet geometry to estimate tactical advantage rather than geometric position alone.
- Data
- Race Tracks · Fleet Geometry · Environment · Opponent Behaviour
- Methods
- Trajectory Prediction · Machine Learning · Sequence Models · AI / LLM Reasoning Layer
- Output
- Tactical Intelligence Model
- Validation
- Prediction of future relative fleet advantage.
Boat A Closest · current rank 1Boat C Expected advantage +18 m
WP04
Make the Boat Faster
Which configuration produces maximum performance?
Connect computer-vision sail geometry, structural loads, crew position and boat state with measured speed and VMG.
- Data
- Sail Shape · Loads · Heel · Speed · VMG · Crew Position · Wind
- Methods
- Computer Vision · Multimodal Learning · Performance Modelling
- Output
- Boat Performance & Trim Model
- Validation
- Measurable improvement in boatspeed or VMG.
- Draft
- 42%
- Depth
- 11.8%
- Twist
- 14.2°
- Forestay
- 1.84 kN
- Shroud
- 2.31 kN
- Mainsheet
- 0.71 kN
- Heel
- 17.4°
- Speed / VMG
- 7.1 / 5.6 kn
Sail shape+Loads+Boat state+Environment
Performance modelIllustrative recommendation Target heel 16–18° · Reduce forestay sag
WP05
Execute Better
Where do we lose metres?
Automatically detect and benchmark tacks, gybes, starts, acceleration, mark roundings, hoists and drops.
- Data
- GNSS · IMU · Video · Rudder · Crew Movement
- Methods
- Event Detection · Time-Series Analysis · Benchmarking
- Output
- Boat Handling Model
- Validation
- Reduction of time or distance loss during manoeuvres.
- Average loss
- 7.2 m
- Best
- 4.1 m
- Recovery
- 8.4 s
- Recoverable
- 3.1 m
WP06
Learn
What should we train next?
Combine venue, environment, tactics, boat performance and handling into a measurable coaching cycle.
- Input
- Venue · Environment · Tactics · Boat · Handling
- Methods
- Explainable AI · Human–AI Interaction · Longitudinal Evaluation
- Output
- Integrated AI Coaching System
- Validation
- Measurable performance improvement over repeated training cycles.
Illustrative example
VenueEnvironmentTacticsBoatHandlingAI coaching
- DiagnosisTack loss exceeds benchmark
- PriorityTacking under 10–14 kn
- TargetReduce average below 5.0 m
TrainingNew dataRe-evaluation↺
Historical evidence base
60,000+ races to learn from.
The historical corpus can reveal recurring venue, fleet and tactical patterns at a scale that isolated training sessions cannot.
Available datasets will be evaluated for consistency, sampling rate, class differences, environmental context and suitability before model development.60,000+ races
- Race tracks
- Positions
- Starts
- Decision situations
- Upwind legs
- Tactical patterns
- Mark roundings
- Boat handling
- Venues
- Local patterns
Shared architecture
One multimodal research foundation.
Every work package uses a synchronized representation of the race rather than another isolated data silo.
Input layer
TracksWeatherSensorsVideoFleet
Data layerSynchronized Race StateTime · Position · Environment · Boat · Fleet
Model layer
Venue intelligenceEnvironmental stateTactical intelligencePerformanceHandling
Application layerIntegrated coaching
Scientific novelty
Evaluation
How do we know it works?
Model accuracy matters, but the final standard is real sailing performance.
Work packagePrimary evaluation
VenuePrediction on unseen races
EnvironmentAccuracy against reference sensors
TacticsFuture fleet-advantage prediction
Boat performanceSpeed or VMG improvement
Boat handlingReduced time or distance loss
CoachingImprovement over repeated training cycles
The final metric is not AI accuracy. It is better sailing performance.
Publication strategy
Four-year research programme.
- Year 1Venue intelligence
Data foundation · Paper 1
- Year 2Environment & tactics
Field studies · Papers 2–3
- Year 3Boat performance & handling
Experiments · Papers 4–5
- Year 4Integrated coaching
Validation · Paper 6
Existing foundation
Ambitious—but not starting from zero.
DataDrivenSailing and related prototype work provide practical experience with heterogeneous sailing data, synchronization, onboard video, field constraints and coaching workflows. They are a technical starting point, not claimed research results.
Read the DDS documentation- Existing work
- Sailing-data pipelines, synchronized media and applied analysis prototypes
- Technical background
- Data science, engineering problem solving and competitive sailing
- Potential collaboration
- Researchers, sailing teams, federations, coaches and technology providers
Research collaboration
Let's research how to win.
The project is designed as a modular doctoral research programme with independent scientific contributions and opportunities for collaboration across sailing, engineering, data science and artificial intelligence.
Discuss the Research