Our UBC Solar Race Strategy Team met this week to align on project direction and consolidate concept ideas for the Hack the Track presented by Toyota GR competition.
Brainstorming
Across our team, the recurring themes were:
Real-time performance feedback
Vehicle dynamics + hybrid physics/ML models
Racing line optimization
Pit stop detection + strategy
Weather + tire strategy (if data is available)
Deployable, production-ready data pipeline
Highlighted Concept Directions
Optimal Racing Line + Driver Feedback
Use hybrid physics-ML trajectory optimization to derive the ideal line + control inputs per track section, then compare to driver telemetry to generate actionable feedback.Pit Event Detection
Detect pitting using vehicle stoppage + location; track stint timing & race impact.Post-lap Analytics
Segment-by-segment insights on braking, acceleration, steering, and gear usage, with suggestions for next-lap improvement.Real-Time Strategy Simulation
Evaluate race situations using live telemetry from all cars to recommend decisions (e.g., passing likelihood, pit opportunities).Tech-Stack Alignment with Toyota Racing Development (TRD)
Inspired by race engineering job postings, focus on modern data tooling, simulation, and DevOps.
So, we will build a real-time racing intelligence stack:
Core Features
Hybrid physics-ML vehicle model
Optimal-trajectory generation per track section
Real-time telemetry ingestion + processing
Driver deviation detection (line, steering, accel, gear)
Feedback + actionable coaching
Optional: pit detection & stint tracking
High-Level Architecture
Pipeline: Kafka + Faust (stream + ETL)
DB: PostgreSQL
Modeling: JAX / PyTorch + JAXopt / CasADi
Backend: FastAPI / Flask
Frontend: React / Vue / Angular
Deployment: Docker, CI/CD
We chose this stack because it aligns well with TRD judging themes of: strong use of data, performance modeling grounded in physics, and real-time race engineering support.
Open Questions:
Do we have tire compound + wear data?
Do we have pit time annotations, or must we detect pitting?
Should the model support new tracks with limited prior data?
Our solution:
Scales to real-time engineering usage
Grounds ML in interpretable physics
Provides clear driver coaching + strategic insights
Reflects real tools + workflows used at TRD
This direction leverages our experience from designing strategy tools for solar racing, while expanding into advanced data-driven motorsports engineering. We seek to win with this solution.
More updates are coming soon!
Log in or sign up for Devpost to join the conversation.