Inspiration
In any sport, coaches tend to say "be consistent," but how do you measure that in this sport? Brake points sound like a good place to start.
What it does
Brake Point Drift Detector analyzes GR Cup telemetry to visualize brake point scatter and technique quality. It shows:
- Where drivers brake (GPS coordinates) and how hard
- Consistency metrics (dispersion in meters per zone)
- Brake pressure curves
- Interactive dashboard comparing any driver against fastest lap reference
Key insight: Winners brake 26m earlier with 23% less pressure and 24% smoother inputs.
How we built it
Python pipeline processing 4,462 brake events:
- Track centerline generation: GPS smoothing with Savitzky-Golay filter (periodic wrapping to eliminate start/finish kinks)
- Brake onset detection: Rising-edge logic on
max(pbrake_f, pbrake_r)with noise filtering - Zone clustering: Discovered brake zones from actual data (not predefined corners)
- Dispersion analysis: $\sigma = \sqrt{\frac{1}{n}\sum_{i=1}^{n}(d_i - \bar{d})^2}$ where $d_i$ is distance from zone centroid
- Plotly dashboard: Interactive track map with driver comparison modes
Challenges we ran into
GPS noise and track closure: Raw GPS had spikes (>10m jumps) and start/finish discontinuities. Solved with distance-based resampling and periodic smoothing that wraps the track as a continuous loop.
Brake curve extraction: Aligning variable-length pressure curves for comparison required time normalization and careful interpolation to preserve shape characteristics.
Finding the signal: Initial analysis showed podium drivers braking later with more pressure—opposite of reality. We had mixed up coordinate systems. Fixed by using proper distance calculations.
Accomplishments that we're proud of
Quantified what coaches feel: "Smooth beats hard" is now measurable—podium drivers are 23.6% smoother with 7.3 bar less peak pressure. They are also more consistent than the rest in where they break.
Real training value: Identified specific zones (3, 4, 8) where technique matters most (39-56% smoothness advantage). Drivers now know exactly where to focus practice time.
Production-ready tool: Clean pipeline, cached outputs, interactive dashboard deployed at edulizanay.github.io/14.toyota-hackathon
What we learned
Brake technique is measurable: Smoothness, peak pressure, and release control can be quantified from telemetry—opening the door for objective driver development.
Clustering beats assumptions: Letting data reveal brake zones (vs predefined corners) captured where drivers actually brake, making dispersion metrics meaningful.
What's next for Brake Point Drift Detector
I can see a world where this tool can take inputs almost in real time, to let drivers know exactly when and how how they should break to achieve near perfect consistency, gaining a few seconds per competition.
Log in or sign up for Devpost to join the conversation.