Inspiration
I've always been captivated by motorsports because of the extreme strategy and accuracy needed to succeed. With its identically prepared GR86 cars, the Toyota GR Cup provides an ideal setting for investigating the potential of data-driven performance optimization. In addition to analyzing actual race telemetry, I wanted to develop a tool that could simulate race results and forecast driver performance in a research-based manner.
What it does
RaceIQ is a GR Cup racing analytics and simulation platform driven by AI.
- It consumes data from official results, weather, lap and sector times and telemetry.
- Uses machine learning to analyze vehicle performance and driver behavior.
- Forecasts race results under various tactics and circumstances.
- Produces useful information for race preparation and driver training.
- Offers visualizations such as predictive performance charts, sector heatmaps and speed maps.
How we built it
Data Processing: Several datasets, including lap times, telemetry, weather and results were cleaned, merged and standardized using Python's pandas and numpy functions.
Key features such as average sector speeds, lap consistency, throttle/brake patterns and weather-adjusted lap times were extracted through the process of feature engineering.
Machine Learning: Predictive models were implemented using xgboost for performance ranking and
scikit-learn for outcome prediction.
Simulation Engine: Developed a race simulator that uses Monte Carlo simulations and historical data to probabilistically model lap-by-lap performance.
Visualization: Performance heatmaps and race dashboards were created using matplotlib and plotly.
Docker and containerization were not used; all code and execution took place within Visual Studio Code.
Challenges we ran into
Dataset Complexity: Careful alignment was needed to handle several sources with disparate time formats, anonymized driver IDs and sector-level telemetry.
Given the scarcity of GR Cup historical data, it was difficult to strike a balance between overfitting and model complexity.
It took several iterations to convert raw telemetry into realistic lap-by-lap probabilistic simulations.
Accomplishments that we're proud of
Six distinct GR Cup datasets were successfully combined into a single framework for analysis.
- Developed a predictive model that predicted the top 5 positions with >85% accuracy.
- Produced visualizations that show each driver's and sector's performance bottlenecks.
- Without using third-party deployment tools, a fully functional VS Code-native platform was developed.
What we learned
- A thorough comprehension of how environmental variables, lap times and telemetry affect racing results.
- Hands-on experience with feature engineering for time-series motorsport data.
- Proficiency in creating machine learning pipelines that integrate visualization, simulation and prediction.
- Iterative development and comparing simulations to real-world data are crucial.
What's next for RaceIQ: AI-Driven GR Cup Performance Simulator
Incorporate real-time data feeds to support live race strategies.
Extend the analytics for driver and vehicle performance trends to include multi-race season data.
To improve driver tactics in a variety of scenarios, incorporate reinforcement learning.
Research findings on AI-driven predictive analytics in spec-series motorsport will be published in a research paper.
Log in or sign up for Devpost to join the conversation.