Optimal Karting Line

Blogs

August 12, 2026

TL;DR

GoPros telemetry is not always so reliable and a PWA can prove it

Intro

WHY Im writting this post: Want to share with my friends my refreshed gopro telemetry setup
What Ive learnt with it: Ive ended up making my first PWA around ffmpeg

Coming from here and here

The GD was the one that worked best so far

Go Pro Telemetry Overlay

Tinker around your gopro gps telemetry and the Spa and Nurburgring circuits:

Py Route Tracker ↗ Optimum Path | Github ↗

git clone https://github.com/JAlcocerT/optimum-path
cd ./optimum-path/overlay

For a quick check before rendering video:

uv run python gopro_h13_hud_fastlap.py \
  --config ../configs/GX010021_spa.json \
  --yes \
  --png 5

when happy, just:

uv run python gopro_h13_hud_fastlap.py \
    --config ../configs/GX010021_spa.json \
    --yes \
    --embed

Spa

With the VW Golf:

uv run python gopro_h9_racing_hud.py --config ../configs/GH010437_GH020437_spa_francorchamps.json --yes --embed

With the laguna mk2:

make -C /home/jalcocert/Desktop/optimum-path spa-francorchamps-hud

Yep, sth was terribly wrong with the GPS here…

This is a great reason to finally get a separated GPS module for my FPVs and other experiments

Karting and Theoretical Optimum Path

Method	Result	Status	Notes
Real best lap from GoPro/HUD	`61.27s`	valid reference	From generated HUD output `flying_lap_61.27s.mp4`.
GPS best lap used for track generation	`61.20s`	valid reference	From raw Hero 13 GPS rows after DOP filtering.
Centerline theoretical	`59.71s`	valid model result	Point-mass velocity model on generated Spa centerline.
Gradient/SLSQP racing line	`59.25s`	valid model result	Direct lap-time optimizer, same vehicle parameters.
Minimum-curvature path + velocity profile	`60.68s`	valid model result	Optimizes smoothness first, then computes speed profile.
GA direct-control quick run	`1001.80s`	DNF / invalid comparison	Quick 24-pop, 8-generation pass did not complete a lap.
RL/PPO evaluation	n/a	not run	`stable_baselines3` is not installed locally; existing model was trained on a different track.

For gradient-descent artifacts and the driver-action/circuit-position video:

make spa-gd-all

#Or step by step:

make spa-track
make spa-gd
make spa-gd-video

Where are these videos from? An European roadtrip!

What I learnt prior to this about real vs simulated results 📌

Under the honest κ-fix physics, the real driver is 3.35 s faster than the model says is theoretically possible.

The exact numbers:

Real fastest lap (GX020010 lap 2): 78.20 s
κ-fix N=400 honest theoretical optimum: 81.55 s
Gap: −3.35 s (real beats the calibrated model)

That doesn’t mean the driver is magically beating physics — it means the model is under-stating what the kart can actually do. Two consistent stories from today’s analyses:

The driver is faster in both corners and straights, but more on straights:

corners: −1.08 s (driver gains 1.1 s net)
straights: −2.26 s (driver gains 2.3 s)
total: −3.35 s

The sensitivity sweep says the model’s max_lateral_g = 0.95 is the most sensitive parameter at −27.5 s/g. Bumping it to 1.07 g (+12%) would bring the sim down to ~78 s and match your fast lap. That’s a believable correction: the calibrated 0.95 g came from yaw-rate-κ-based lap-time matching, which slightly under-counts true peaks. The sustained-vs-transient-peak distinction is exactly what Phase 4’s load-transfer model would capture.
The pro’s 75 s sits another ~3 s below your 78.20, which the sensitivity sweep says needs another +0.12 g of lateral — total 1.19 g. That’s high-but-plausible for harder/fresher tires (the pro likely had different equipment too).

So the honest, today-as-of-now reading is:

he calibrated point-mass model is slightly conservative — about 1.07 g of “true effective” lateral grip vs. the 0.95 g it has calibrated.

Once Phase 4’s bicycle model captures the transient grip from load transfer, the gap to your real laps should close (and possibly close further toward the pro).

The driver isn’t superhuman; the model is leaving 3 s on the table because it can’t represent trail-braking and load transfer.

Real-beats-sim by 3 s, mostly explainable by transient grip the ellipse can’t represent is exactly what richer dynamics should fix, tinkered here.

Nurburgring

The kind of track day that you wouldnt miss with your friends for anything in the world

With the VW Golf:

With the laguna mk2:

Nurburgring World Record

And what do i get by adding the car parameters to my model?

Golf vs Laguna Fast Lap Comparison

Conclusions

Despite not having gps in my action cam…

I had a lot of fun in this circuit :)

PWA - GoPro telemetry extractor

Not enough with the ulm/ppl exam prep to get traffic at jalcocretech?

here is sth more: https://gopro-telemetry-overlay.pages.dev

cd ./optimum-path/overlay-pwa
make dev #make deploy
# npm cache clean --force

#npx wrangler pages deployment list 
#npx wrangler pages deploy dist

npx wrangler pages deploy dist --project-name=gopro-telemetry-overlay

acme.json
config.yml - Automatically created
traefik.yml

docker-compose.x300.yaml

.env

cf-token

FAQ

Linux 101

Download an OS like ZorinOS to get started.

Files Sync

Nextcloud: a personal cloud, one of the first setups for any homelab and working perfectly in Linux. Now with a native linux desktop client.
Proton CLI: finally arrived!

Download https://proton.me/download/drive/cli/index.html

https://proton.me/download/drive/cli/index.html

  cd /home/jalcocert/Downloads
  chmod +x proton-drive
  install -m 0755 proton-drive "$HOME/.local/bin/proton-drive"

You will auth with a link and get https://account.proton.me/auth-desktop

proton-drive auth login
proton-drive filesystem list /
proton-drive filesystem list /my-files

To upload one local file with the Proton Drive CLI:

proton-drive auth login proton-drive filesystem upload /path/to/local-file “/destination/folder”

Example, uploading ~/Downloads/report.pdf to the root of Proton Drive:

proton-drive filesystem upload ~/Downloads/report.pdf “/”

To see folders first:

proton-drive filesystem list “/”

To create a folder:

proton-drive filesystem create-folder “/” “Uploads” proton-drive filesystem upload ~/Downloads/report.pdf “/Uploads”

proton-drive filesystem upload ~/Desktop/DJI_20260726095445_0040_D.MP4 /my-files/Oa5Pro-Rysy

Rsync
You can play your own music and videos via: jellyfin and ultrasonic in android / ios / supersonic or sonixd (connects to subsonic or jellyfin) for desktop

Gonic | Docker Config 🐋 ↗ Jellyfin Media Server Tools 🐋 ↗

winget install sonixd

Jellyfin | Docker Config 🐋 ↗ QBittorrent Docker Config 🐋 ↗

OSS Photo Video

For photo:

Upscayl
RawTherapee

For video I tend to get away with ffmpeg

If you need more:

KDEnlive
OBSStudio for recording
Openshot

Gaming

You should be good with gaming via Steam in Linux for most games.

Just be careful with online ones as some anticheats dont work outofthebox.

I also tested SteamOS via a VM with gnome boxes:

Download the ~3gb and: https://store.steampowered.com/steamos/download?ver=steamdeck

#bunzip2 steamdeck-oobe-repair-20260707.10-3.8.14.img.bz2
gnome-boxes --open-uuid 0d42c90e-9c8c-4507-a166-2c82c41bccf2

#flatpak install flathub org.gnome.Boxes

Tools and Tech for Trips

Tech for a trip, like a travel router with VPN and some interesting

cd ./poc/z-poc-trip-planner-adk # cd ../z-poc-trip-planner-v4
make compose #http://192.168.1.2:3021

cd ./poc/z-poc-trip-planner-mama
#docker compose -f docker-compose.prod.yml up --build #3022
docker compose -f docker-compose.prod.yml up --build -d # http://trip-planner-web:3000

trip planner with weather time zone and costs

https://github.com/itskovacs/trip/releases/tag/1.47.0

My fav PWAs

Spliit, and a quick pwa poc i created around it

cd ./poc/pwa-spliit-analyzer
C:\Python312\python.exe -m http.server 5173 --bind 127.0.0.1

Some, like https://app.files.md/ allow you to interact with your local files!

Or with your fpv/drones hardware: https://app.betaflight.com/#

Batteries are a thing [AIoT] How to QnA Sensor Data