Fastest-lap Simulation
Head-to-Head Analysis

Project Story — Carma: Fastest Lap Intelligence

About the Project

Inspiration

Modern motorsport engineering blends telemetry, racing-line analysis, and sector timing to understand how and why lap times differ. We wanted to bring that experience into a browser-based tool that anyone can use without needing specialized engineering software.

Working with raw Toyota GR data pushed us to build something deeper than a lap-time viewer. We wanted to explain the fastest lap: where time was gained, how the racing line changed, and how throttle and braking influenced the result.

What We Built

Carma: Fastest Lap Intelligence is an interactive Streamlit platform for analyzing fastest laps at Circuit of the Americas (COTA) and Barber Motorsports Park. It merges endurance timing, telemetry, and racing-line reconstructions into a unified analysis suite.

Core Features

Fastest-lap extraction and ranking
Micro-sector delta analysis
Animated fastest-lap “ghost-car” simulation
Live throttle & brake overlays
Driver-vs-driver alignment on a shared station grid
Racing-line comparison and turn insights
Driver profiling via statistics and radar charts
Turn detection, coasting analysis, and telemetry heatmaps

How We Built It

Data Processing Pipeline

We designed a unified workflow for both tracks:

Parsed endurance CSVs to compute LAP_TIME_SEC, sector splits, and per-driver rankings.
Constructed Barber’s centerline using a GPS reference lap smoothed with cubic interpolation.
Mapped telemetry (x, y) → station distance (s) using nearest-point projection on each track’s centerline.
Built fastest-lap datasets (CSV/Parquet) by slicing telemetry to each lap.
Merged endurance + telemetry through synchronized timestamps and lap triggers.
Created micro-sectors by dividing the station axis into fine-grained segments (≈200–400 per lap).

How We Reconstructed the COTA Track Using Pixel Calculations

Because COTA lacked GPS-accurate centerline data, we engineered our own using a pixel-based reconstruction:

Selected a high-resolution COTA diagram.
Manually traced a racing-line polyline by capturing pixel coordinates for key points (T1 apex, esses, hairpins, stadium section, final corner).
Extracted each point’s pixel position ( (p_x, p_y) ).
Converted pixel distances to real-world scale by normalizing total arc length to COTA’s actual 5.513 km track length.
Applied B-spline smoothing to reduce jitter.
Resampled the smoothed polyline uniformly (≈1 m resolution).
Projected telemetry GPS points onto this pixel-derived centerline using nearest-point orthogonal projection.

This gave us a stable reference for:

Ghost-car animation
Sector visualization
Racing-line overlays
Driver comparisons
Accurate micro-sector delta computation

The pixel-derived COTA track was proportionally accurate enough that all telemetry mapped smoothly onto it, with clean animation and consistent station alignment.

Visualization & Interaction

We used Plotly for dynamic visuals and Streamlit for UI:

Sector-colored track maps
Speed traces and delta bar charts
Animated fastest-lap simulation
Racing-line overlays
Throttle/brake heatmaps
Turn zoom panels

Simulations followed the transformation: $$ t \rightarrow \text{elapsed} \rightarrow \text{speed} \rightarrow \text{distance} \rightarrow \text{station} \rightarrow (x, y) $$

Analysis Modules

Sector Analysis

Computes per-micro-sector delta times
Provides cumulative S1/S2/S3 deltas
Displays overlays and annotated track maps

Ghost Simulation

Rebuilds car motion frame-by-frame
Animates progress along the centerline
Color-labels by speed with sector transitions

Playback (Barber)

Scrubbable slider positions both cars at identical station locations
Shows synchronized throttle, brake, speed, and gear

Driver Profiling

Quantifies braking %, throttle usage, lift patterns, consistency
Displays a radar chart of behavior metrics

Advanced Insights

Turn detection (curvature-based)
Coasting analysis
Dual-line racing-line comparison
Telemetry heatmaps (throttle/brake)

Challenges We Faced

Telemetry Alignment

Telemetry was unevenly sampled and noisy. The (x, y) → s projection required smoothing to prevent jumps in delta calculations and ghost animations.

No GPS-Accurate COTA Centerline

We reconstructed COTA from pixels, which required careful:

scaling
smoothing
curvature validation
projection testing

Even tiny scaling errors caused major drift in the animations.

Sector Timing Issues

Some endurance files lacked valid sector fractions. To ensure stable visuals, we used a fallback:

$$ \text{If sector fractions are invalid, S1/S2/S3 = equal thirds of total track.} $$

Managing Multi-GB Telemetry

Raw telemetry exceeded GitHub limits. We condensed CSVs by filtering to only the fastest laps for each driver per circuit for their telemetry data, which allowed us to stay within the GitHub limits. However, for a proper analysis, using the original CSV telemetry files provides a better, more holistic view across all drivers and sections.