Third Place Solution in Spring 2025 Simulation Racing Series
Written by Ryan Chow, a rising senior at Sacred Heart Cathedral in San Francisco.
Table of Contents
Overview
My solution builds off of Derek Chen’s Fall 2024 team’s solution. This final solution incorporates a lot of the previous logic, while improving on throttle control and reworking waypoints to model real F1 racers and their race lines and techniques. The key innovation lies in optimizing waypoint selection and racing lines based on analysis of Lewis Hamilton’s fastest lap on Monza, combined with enhanced throttle control algorithms that better handle the dynamic requirements of professional racing lines.
Optimizations
I primarily focused on optimizing the throttle control, including tweaking values in specific variables and adding more conditions that will make the throttle controller more sensitive. I also tried optimizing brake-to-throttle transitions, but this didn’t fully work as intended. Some changes include expanding the friction coefficient system to also require optimal racing line optimization.
Beyond that, I changed the waypoints the car should follow, drawing inspiration from real F1 drivers racing the map. I noticed that the car wasn’t following the “optimal” racing line, and attempted to change it through optimizing the waypoints. This simple fix took a lot of trial and error. The biggest changes were made on the last, long curve, where the car doesn’t take unnecessarily sharp turns any more.
Future Improvements
Some future improvements that I didn’t have time to add but will look into in the future include:
- F1-Style Corner Categorization: Implement a more sophisticated corner classification system based on F1 engineering principles
Neural Network Training on Professional Telemetry: Develop ML models trained on extensive F1 telemetry databases to predict optimal racing lines for new track configurations: - Live Telemetry Analysis: Develop capabilities to analyze current F1 sessions and automatically update racing line strategies:
While these will be difficult to implement, it could make the AI racer extremely smart and be able to adapt to its environment effectively.
Conclusion
This solution represents a significant evolution in autonomous racing by bridging the gap between algorithmic control and professional racing technique. Building upon Derek Chen’s proven foundation, the integration of Lewis Hamilton’s F1 racing lines transforms the approach from purely computational to racing-intelligent.
The F1-inspired waypoint optimization demonstrates that studying professional driver technique can yield measurable performance improvements in autonomous racing. The late apex strategies, trail braking integration, and geometric racing lines borrowed from Formula 1 create a system that doesn’t just navigate the track safely, but actively pursues the fastest possible lap times using proven racing principles.
This hybrid approach—combining Chen’s robust algorithmic framework with Hamilton’s racing intelligence—validates the potential for professional motorsport knowledge to enhance autonomous racing performance. The modular design ensures that future F1 insights can be easily integrated, creating a pathway for continued performance evolution.
I’d like to thank Huo Chao Kuan, Dr. Allen Yang, and the rest of the Berkeley team for creating such a competitive and practical competition. It’s been really enjoyable to compete and I’ve learned a lot from this experience.