RRS — Real-time Race Strategist

From telemetry to victory — AI that tells you exactly when to pit, in 3 seconds flat.

Inspiration

I’m Siphe from South Africa, and I’ve always been obsessed with the split-second decisions that win races. In the GR Cup, pit strategy can turn P5 into P1 — but teams still guess tire wear using rules of thumb. When I saw the TRD GR Cup COTA 2024 telemetry dataset, I knew: this is the data that can end the guesswork.

What I Learned

• How to turn a 2.2 GB raw Kafka-style CSV into a clean, wide-format parquet in under 10 seconds.
  
• That distance = speed × time lets you reconstruct the entire COTA layout without GPS — pure telemetry magic.
  
• Gradient Boosting can achieve (R^2 = 0.74) predicting lap-time loss from tire wear, brake usage, and sector-specific stress.
  
• One-click Windows deployment with activate.bat is a game-changer for judges who hate pip install.

How I Built It

1. Data pipeline (data_preprocess.py):

   • Pivoted 500 000+ long-format rows into wide format.
     
   • Merged real lap times (meta_source == 'lap_time').
     
   • Computed cumulative distance → split COTA into 20 sectors.
     
   • Created a physics-based tire-wear model:
     [ \text{tire_wear_\%} = (lap \times 3.2) + (brake \times 8 \times \text{sector_wear_factor}) ]

2. AI model (train_model.py):

   • Trained GradientBoostingRegressor on tire_wear_%, lap, speed_avg, throttle_avg, brake.
     
   • Output: seconds lost per lap → exact pit window.

3. Dashboard (streamlit_app.py):

   • Sector Radar Chart reveals the top tire killers (T11 Esses = 2.5× wear).
     
   • Live AI advice: “PIT NOW → save 2.1 s”.
     
   • Deployed instantly on Streamlit Cloud.

4. One-click magic (activate.bat):

   streamlit run src\streamlit_app.py
   ## Accomplishments that we're proud of
   

What we learned

What's next for RRS(Real-time Race Strategist) — AI Pit Strategist

Built With

• joblib
• numpy
• pandas
• parquet
• plotly
• python
• scikit-learn
• streamlit