1 of 37

od2net: designing moonshot cycling networks

Dustin Carlino

11 November 2023

State of the Map EU 2023

2 of 37

Outline

Intro (10-15 mins)

What are we doing and why?
Demo
How it works

Workshop (30 mins)

Choose your own adventure, or follow along

Wrap-up (10 mins)

Optionally show other tools
Next steps & questions

3 of 37

About me

A/B Street and totally open source since 2018
The Alan Turing Institute since Dec 2021
Active Travel England since spring 2023
Mission/principles:

Rapidly slash use of cars in cities
Promote gov transparency on transportation planning and improve community engagement
Build cool stuff that anyone can use

4 of 37

5 of 37

6 of 37

Team

Chris Conlan for edge costs
Hadrien Salat for the logo
Thanks Andy Smith and Anna Zanchetta for assisting today
Colleagues at Active Travel England for work on safe routes to schools

7 of 37

Credits

All slides feature images with OSM data, attribution often clipped out

Apologies if color schemes are hard to see

8 of 37

Networks for drivers and cyclists aren’t equal

9 of 37

Networks for drivers and cyclists aren’t equal

10 of 37

LTS: Level of Traffic Stress

Using a mapping from OSM tags thanks to https://maps.bikeottawa.ca/lts/
Image from Alta Planning

11 of 37

How do we shift trips from driving to cycling?

Look for trips under 5-10km taken today by driving
Why don’t they cycle today?

Weather
Hills
Safe parking/storage
Culture
Need to carry lots of passengers or luggage
The roads aren’t safe

12 of 37

How do we shift trips from driving to cycling?

Look for trips under 5-10km taken today by driving
Why don’t they cycle today?

Weather (“no wrong weather, just wrong clothing”)
Hills (e-bikes)
Safe parking/storage
Culture
Need to carry lots of passengers or luggage (cargo bikes)
The roads aren’t safe (so let’s improve them)

13 of 37

How do we shift trips from driving to cycling?

Look for trips under 5-10km taken today by driving
Why don’t they cycle today?

Weather (“no wrong weather, just wrong clothing”)
Hills (e-bikes)
Safe parking/storage
Culture
Need to carry lots of passengers or luggage (cargo e-bikes notwithstanding)
The roads aren’t safe (so let’s improve them)

If those people did cycle, which roads would they take?
Which of those popular roads need improvements?

14 of 37

What is a route network?

One particular route request (A to B) gives a specific route, along different road segments…

15 of 37

What is a route network?

One particular route request (A to B) gives a specific route, along different road segments…

16 of 37

What is a route network?

“Sum up” many routes, seeing which roads are used the most

17 of 37

Why are route networks important?

Strategic network plans like https://content.tfl.gov.uk/strategic-cycling-analysis.pdf

18 of 37

Why are route networks important?

Propensity to Cycle Tool (https://pct.bike)

Robin Lovelace, Rachel Aldred, James Woodcock, et al
Incredible impact on cycle planning in England – an official Department for Transport tool, used by all strategic cycling network plans

Problems with the PCT (and followups CRUSE and NPT)

Over a day to calculate for a country
Dependent on external routing service
Specialized for one place, hence 3 projects

19 of 37

Why not use something existing?

Valhalla, OSRM, BRouter, OTP, R5, GraphHopper, etc
Even batch modes have performance problems
Maybe could in the future
od2net isn’t a user-facing router

20 of 37

od2net demo

(Be gentle, it’s only alpha)

Two very different interpretations

Plug gaps in existing quiet streets?
Or assume any existing road can be improved?

Tends to be motorways or big arterial roads – because those usually designed for directness

21 of 37

od2net goals

Reproduce results from PCT/NPT/etc much more scalably

25 minutes to route for ~13 million England-wide home-to-work trips
(About 90 minutes end-to-end including downloading input datasets and creating output PMTiles for viewing)

Lower the barrier of entry to create route networks anywhere

Start with your browser in a few minutes, no installs
Python examples for input data prep
Run country-wide examples on your own laptop in a few hours

Experiment with edge cost functions / routing profiles

22 of 37

Initial performance

https://github.com/dabreegster/od2net/blob/main/docs/performance.md

23 of 37

How to use od2net

Directly in your browser

Upload your own osm.pbf or extract from Overpass (small areas)
Simple origin/destination pattern
od2net.org/interactive.html
od2net.org/edge_cost.html

On your computer (compile it or Docker)

Use Python example scripts to prepare real OD data for different areas
Run od2net, then view the PMTiles output in od2net.org
Add your own area!

24 of 37

How to use od2net

You specify:

The study area
Origin/destination data

Zone to zone
Zone to point
Preprocessed LineStrings from somewhere else
Match nearest destination for every origin

Level of Traffic Stress definition
Edge cost function
Uptake model

You get:

A route network, expressed as GeoJSON, PMTiles, and CSV
A web app to explore output

25 of 37

How od2net works

OD data

config.json

osm.pbf

od2net import

network.bin

ch.bin

od2net route

output.geojson

tippecanoe

rnet.pmtiles

Svelte + MapLibre app

26 of 37

Part 2: Workshop

Go to docs.od2net.org (or slides.od2net.org)

Option A: use interactive “many to one” mode, tune routing parameters

No installs, no coding

Option B: prepare input for your own city, following the Python examples

Rust or Docker, tippecanoe, osmium, basic Python (or R or something else)

Option C: customize LTS or edge cost functions

Edit the existing Rust / JS code, or write new in Python / anything else

27 of 37

Option C: customize LTS or edge cost functions

Exclude some ways entirely

https://github.com/Urban-Analytics-Technology-Platform/od2net/blob/main/lts/src/allowed.rs

LTS definitions

28 of 37

Edge cost deep dive

Routing 101: minimize total “cost” (distance)
Multiply distance by possible penalties

LTS / safety from motor vehicle traffic
Other safety issues (canal tow-paths, unlit parks at night)
Noise (separate cycleway next to a motorway)
Cyclist speed: hilliness, surface
Greenery / scenicness
Proximity to shops (diversity of land uses?)
Directness / wayfinding
Delays at traffic signals

29 of 37

It’s down to preference

“The regular commuter with an e-bike”

Balance time and safety, but also be near bakeries on the way to work

“Training athlete”

Dislikes delays at traffic signals and too many turns
Prefers steep uphills
Doesn’t mind safety issues

“Going to school”

Safety and noise most important, but also greenery and shops

30 of 37

Step one: balance trade-offs

Penalty = LTS * [0, 1] + Amenities * [0, 1] + Greenspace * [0, 1]

0 is good, 1 is bad

Cost = Distance * Penalty

LTS + Amenities + Greenspace = 1.0

So max Penalty is 1

(Problems with penalty = 0…)

31 of 37

Step two: map each factor into [0, 1]

0 is good, 1 is bad

LTS

Amenities

0 if edge has >= threshold amenities nearby, 1 otherwise

Greenspace

TODO, but… separate [0, 1] sliders for “through greenspace, parallel to greenspace, nothing nearby”

32 of 37

Uptake model

Someone isn’t likely to cycle for 20 miles to work, even if the infrastructure is perfectly safe
So instead of counting 1 for each trip, apply a probability function based on distance