street-sim-swarm

sim swarm

fillmore-preview

A place twin of San Francisco’s busiest permitted event block: O’Farrell Street between Fillmore and Steiner (Fillmore Farmers Market, 64 scheduled market days in current SFMTA permits). It combines real buildings and official zoning with an agent-based foot-traffic model to show how a gathering changes a block.

https://github.com/user-attachments/assets/27048955-93ff-409f-8c42-bee898c4512d

What’s in here

| Path | What it does | |—|—| | procgen/ | Python pipeline: OSM + zoning -> low-poly city (.glb), lot manifest (.geojson), preview (.png) | | web/index.html | Place twin + block impact report (three.js): traffic, spillover, data-source layers | | web/model-viewer/ | Simple orbit viewer for any generated .glb | | data/ | Fillmore-area OSM extract and DataSF zoning used for the example |

Quick start

pip install -r requirements.txt

# 1. Build the city (real OSM + official SF zoning)
python -m procgen.build --osm data/fillmore_real.osm.json --sf-zoning data/fillmore_zoning.geojson --out out/fillmore

# 2. Export for the web twin
python -m procgen.export_web --osm data/fillmore_real.osm.json --lots out/fillmore.lots.geojson \
  --event "O'Farrell Street" "Fillmore Street" "Steiner Street" --out web/city.json

# 3. View
python -m http.server 8000      # then open http://localhost:8000/web/

Any other SF area: replace --osm ... with --bbox S W N E and --sf-zoning with no value to fetch both live. Use --zones procgen/zones/fillmore.json for hand-drawn zoning splines (sketching, non-SF cities). Add --no-infill to show only real buildings.

Pipeline

  1. Parse OSM (osm_parse.py): buildings, parks, roads, projected to local meters.
  2. Zoning (zoning_sf.py, zoning.py): official DataSF districts mapped to generator rules, or Catmull-Rom zoning splines.
  3. Generate (generate.py): street surfaces, blocks, extruded OSM footprints, lot subdivision of empty land, low-poly kit (box, stepped, parapet, gable, bay), park trees. Seeded and deterministic.
  4. Simulate (web/index.html): 1,500 cohort agents, gravity-model destinations, street-network routing, hourly activity curve; market day vs. normal Saturday.

What is real vs. estimated

| Layer | Status | Source | |—|—|—| | Building footprints | Real | OpenStreetMap | | Building heights | Real where tagged, otherwise estimated from zoning | OSM tags / DataSF zoning | | Infill buildings | Not real (procedural, shown grey-olive) | Generator | | Zoning | Real | DataSF Zoning Districts (3i4a-hu95) | | Event block and days | Real | SFMTA Temporary Street Closures (7p5y-sxmu) | | Foot traffic, spillover | Estimated, not yet calibrated | Agent model |

Every building in *.lots.geojson carries height_source: osm_tag, zone_estimate, or procedural.

Privacy

Agents are aggregate cohorts, never individuals. No personal data is collected or modeled.

See DATA_SOURCES.md for attribution.

This site is open source. Improve this page.