Vancouver SkyTrain PCB

A printed circuit board that is a map of Metro Vancouver's SkyTrain network, with one addressable RGB LED per station. Based on Idea022 Translink Map PCB.

Version	Where	What
v3 (current)	ai-v3/	200 × 109 mm board, real geography from the BC Freshwater Atlas, station positions from OpenStreetMap, 68 WS2812B LEDs (every current and under-construction station + SeaBus), XIAO controller footprint on the back, generated end-to-end from a data file
v2	v2/	100 × 83 mm schematic-style board, artwork drawn in Inkscape and imported with svg2shenzhen
v1	v1/	320 × 210 mm board with two LEDs per station (one per direction), ESP32

Front — exposed-copper water, silkscreen routes and station rings, one LED per station:

Back — XIAO controller footprint, four buttons, LED chain in/out test points, M3 mounting holes:

Renders are KiCad 10 kicad-cli pcb render output (ai-v3/tools/render_board.py); the geography-only preview is ai-v3/test-results/geo-preview.png.

v3 in one paragraph

Everything on the v3 board — LED placement, the exposed-copper water, the silkscreen routes, station rings and labels, the copper city names and wordmark — is generated from one data file, ai-v3/input/vancouver.json, by the scripts in ai-v3/tools/. The data file is itself built from public sources (ai-v3/input/build_from_fwa.py): the province's Freshwater Atlas for land, sea, rivers and lakes; OpenStreetMap for station positions; the province's ABMS layer for the UBC boundary. Labels are laid out by an optimiser and then finished by hand in KiCad, and the hand edits are pulled back into the data file so nothing is ever lost to a regeneration.

Pipeline (v3)

All commands run from ai-v3/tools/ unless noted. Python 3 with shapely, numpy, scipy, pillow, matplotlib; KiCad 10 (its kicad-cli is used for DRC and renders).

1. Data

cd ai-v3/input
python geo_fwa.py                 # (optional) preview the geography alone -> ../test-results/geo-preview.png
python build_from_fwa.py          # geography + stations + city labels + wordmark -> vancouver.json
cd ../tools
python test_vancouver_data.py     # LED count, chain, pitch, margins

build_from_fwa.py keeps every station's label and the suppressed_art list, so it is safe to re-run after hand edits.

2. Labels

python auto_label.py                       # greedy placement + simulated annealing
python preview_map.py ../input/vancouver.json ../test-results/preview.png --collisions
python check_fit.py ../input/vancouver.json --scale 2.0 -v

preview_map.py draws the board exactly as the exporter will (matplotlib, no KiCad); the red overlay marks collisions.

3. Board

python export_placement.py ../input/vancouver.json ../hardware --scale 2.0
python reposition_board.py ../hardware/vancouver-skytrain-pcb.kicad_pcb ../hardware/leds.csv --board 200 --height 109
python export_art.py ../input/vancouver.json ../hardware/vancouver-skytrain-pcb.kicad_pcb --scale 2.0
python render_board.py ../hardware/vancouver-skytrain-pcb.kicad_pcb ../test-results

export_art.py is idempotent: it records every block it writes (uuid → key) in vancouver-skytrain-pcb.kicad_pcb.art-manifest.json and removes them before writing fresh ones.

4. Hand edits in KiCad → back into the data (import before export)

Station labels, leader lines and copper text are ordinary gr_text / gr_line items. Move, rotate, re-justify, retext or delete them in KiCad, save, then:

python import_labels.py ../input/vancouver.json ../hardware/vancouver-skytrain-pcb.kicad_pcb

Each item is found by its deterministic uuid, so the import is exact:

You did in KiCad	Lands in vancouver.json as
moved / rotated / re-justified a station label	label.dx dy angle anchor (start / end / center)
deleted a leader line	leader dropped (leader absent)
edited a label's text (line break, shorter name)	station wrap list, or label.short
moved a city name or the wordmark	annotations[i].x / y
deleted a piece of generated art (SeaBus dash, route segment, boundary)	key added to suppressed_art; the exporter skips it from then on

Always run import_labels.py before export_art.py — the exporter regenerates every label from the data file and would otherwise overwrite the board's hand edits. import → export reproduces the board exactly.

Lessons learned

See lessons-learned.md — what to carry into the next board of this kind, and the mistakes not to repeat.