GR PitSense Live – Real-Time Strategy Co-Driver

Real-time pit strategy engine for GR Cup, predicting the fastest race decisions lap-by-lap.

Comment

Inspiration

In the Toyota GR Cup, where cars are identical and talent is equally matched, strategy becomes the real weapon. A badly timed pit stop can destroy a race — a perfectly timed one can earn a podium. We wanted to take the incredible data Toyota provided — lap times, weather, telemetry — and turn it into something that actually helps race engineers make winning calls in the heat of competition.

What it does

GR PitSense Live is a real-time strategy console built for the GR Cup. It: • Predicts future lap times for any car in the race • Estimates time gained or lost for different pit stop windows • Shows the ideal attack/defend phases during a stint • Provides performance snapshots like gap to leader and pace trends • Replays the race as if live, enabling rapid what-if strategy testing In short: It’s a virtual race engineer helping teams shave seconds and gain positions.

How we built it

• Pre-processed Toyota GR Cup Circuit of the Americas – Race 1 dataset • Merged: o Lap times o Provisional results (gap to leader) o Weather data (track temperature) o Telemetry (speed, throttle, braking) • Feature engineering: o Tire age by lap o Predicted next laps if staying out vs. fresh tires after pit • ML model: o Gradient Boosting Regressor to predict future lap times • UI: o Streamlit real-time dashboard o Interactive controls for pit windows and scenarios • Deployment: o Always-on Linux systemd service at port 4088

Challenges we ran into

• Real race data is not clean — inconsistent column names & missing entries • Telemetry timestamps needed aggregation per lap • Pit behavior varies — required simplifying tire-age models to meet deadline • Real-time performance tuning while streaming model predictions • Ensuring UX stayed intuitive for fast decision-making

Accomplishments that we're proud of

• A full functional strategy engine built end-to-end in a short time • Extracted clean lap-level insights from messy telemetry sources • A UI that feels like sitting on the pit wall of a GR Cup race • Deployed a production-grade always-running service

What we learned

• How race strategy combines physics, driver behavior, and data science • ML needs context: tire degradation, heat cycles, traffic — not just numbers • Good visual analytics can help engineers trust and act on predictions • Toyota GR Cup data is incredibly rich and fun to explore!

What's next for GR PitSense Live – Real-Time Strategy Co-Driver

• Multi-car strategy & traffic modeling (undercut/overcut battles) • Tire degradation models learned from telemetry & track temperature • Fuel-load simulation for better stint predictions • Live data feed integration during real events • Voice-assistant: “Should we pit now?” → Yes/No with confidence • Mobile interface for engineers on-the-go

Built With

Share this project:

Updates