A digital twin to design, analyse, and visualise low-traffic neighbourhoods

AI:UK Research in action

Dustin Carlino

Outline

  1. Background of LTNs
  2. The tool
  3. Technical overview
  4. Next steps
  5. Discussion

Part 1: Background

The problem

  • Car-centric cities
    • Greenhouse gases
    • Noise pollution
    • Air quality
    • Space for parking
    • Collisions
    • Suburban sprawl and unsustainable land use
    • Lack of exercise

The problem

Rise of traffic on local streets from sat-nav

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/916749/road-traffic-estimates-in-great-britain-2019.pdf

The response

Modal filters

The response

Area of effect = Low traffic neighborhoods (LTNs)

The response

  • 2020 active travel fund
  • Mixed public response
    • miscommunication
    • lack of public education
    • hasty consultations
    • genuinely poorly designed LTNs

The current planning process

Communication by diagram

The current planning process

The current planning process

  • How’s this affect my trips?

Part 2: The LTN tool

Demo

  • Bristol and Camden

The LTN tool

  • http://ltn.abstreet.org
  • web browser, Mac, Windows, Linux
    • no mobile
  • free, open source
  • multiple audiences
    • local authorities / consultants
    • individuals
    • campaign groups

Scope

  • works anywhere, thanks to OpenStreetMap
    • most appropriate for cities

Credits

  • Dustin Carlino: project lead
  • Cindy Huang: UX designer (beginning of March)
  • Thanks
    • Robin Lovelace: product manager
    • Brian Deegan, Will Petty, Sustrans
    • Martin Lucas-Smith (CycleStreets)
    • Feedback and testing from many!

Credits

  • built on the A/B Street platform
  • alumni: Michael Kirk and Yuwen Li

Part 3: Technical overview

  • Neighborhood connectivity
  • Predicting rat-runs
  • Defining a neighborhood
  • Assessing overall impact
  • Heuristics for placing filters

Architecture

OpenStreetMap into a graph

  • Edges: road segments
  • Nodes: junctions
  • one-way streets, lane configuration

A neighborhood

  • the perimeter
    • usually “major” roads designed to handle more traffic
  • the interior
    • reduce traffic through here

Cells

Everywhere reachable by driving within the neighborhood, without leaving

Cells

Disconnected cells

  • A cell must touch the perimeter somewhere!
    • Otherwise, drivers can’t get in / out

Calculating cells

  • Pretty straightforward graph floodfill
  • Start from an unvisited road segment in the interior
    • expand
    • don’t cross modal filters, non-driveable roads, or the perimeter

Floodfill

Floodfill

Floodfill

Floodfill

The result

The graph of the neighborhood partitioned into strongly-connected components

Cells as areas

Can we match this?

Voronoi diagrams

(Balu Ertl on Wikipedia, CC BY-SA 4.0)

Not straightforward to apply to line segments

Approximating with grids

  • Diffusion on a 10 meter grid
  • Stop when there’s a collision
    • Record the adjacency for the 4-color theorem

Approximating with grids

Approximating with grids

Approximating with grids

Approximating with grids

Marching Squares to turn the grid into nice contours

Edge cases with calculating cells

One-way streets

Edge cases with calculating cells

Roads without motor vehicles

Edge cases with calculating cells

Is this one cell?

Edge cases with calculating cells

Does this path leave the neighborhood?

Edge cases with calculating cells

Decision: separate cells

Predicting rat-runs

Predicting rat-runs

  • What’s a rat-run?
  • Why are some streets quiet?

Predicting rat-runs

Rat-run definition

  • A shortest path starting and ending on the perimeter road
  • Not the bigger picture
    • Does this path save somebody time?
    • How many people might take this shortcut?

Rat-run definition

  • The start and end must be on different roads
  • Name changes

Rat-run definition

In freeflow conditions, the perimeter road is faster

Rat-run definition

Force the shortest path to stay inside the neighborhood

Rat-run results

Rat-run results

Rat-run results

Defining a neighborhood

Defining a neighborhood

Defining a neighborhood

  • This process will cross railroads, water
    • If it didn’t, we’re fine with heavy traffic on the nearest local road

Defining a neighborhood

Sometimes major roads just end

Defining a neighborhood

Spaces in between motorway loops or dual carriageways

Boundary adjustment

  • Plenty of reasons the heuristics aren’t perfect
  • Road classification varies regionally

Boundary adjustment

Maybe road classification is worth revisiting!

  • Different priorities about moving vehicles through an area
  • Should that high street also be a through-route?
  • Lots of resistance to LTNs is really resistance to the chosen boundary

Boundary adjustment

Don’t be too prescriptive; let users adjust

Boundary adjustment

Boundary adjustment

  • Per block
  • Partitioning into contiguous neighborhoods
  • Adding a block to one neighborhood removes it from another

Blockfinding

  • Trace around the edge of a road
  • Uses the shape of roads and intersections, inferred from OpenStreetMap
  • A block internally tracks a list of (road, left/right)

Merging two blocks

  • Find the common slice of the perimeters
  • “Rotate” the perimeter until the common part matches up
  • Slice and stitch together

Merging two blocks

Merging two blocks

  • Collapsing dead-ends
  • Clockwise and counter-clockwise blocks
    • Winding order

Blockfinding limitations

Only traces roads, not natural features

Blockfinding limitations

Edge of the study area

Blockfinding limitations

  • Bridges / tunnels
    • Trace the 2D area of the planar graph?
    • Skip them?

Blockfinding limitations

Blockfinding limitations

You can’t always draw the boundaries you want

Assessing overall impact

Assessing overall impact

  • Show all rat-runs at once
  • Rank neighborhoods by number of internal rat-runs

Assessing overall impact

  • Is the rat-run realistic?
  • How many people will try to take it?
  • With new filters, how will traffic detour overall?
    • Maybe spillover to another neighborhood

Assessing overall impact

  • Where do driving trips begin and end?
  • For neighborhood-scale detours, MSOA zones are too large

Assessing overall impact

  • Calculate all driving routes before and after filters
  • Look for quieter and busier streets

Assessing overall impact

  • 2011 census data, commuting for work only
  • Urban Observatories traffic counts
    • Calibrate / validate
    • Where do the 5,000 vehicles/hour along a road start and end?
  • Bring your own demand model (doesn’t need to be public data)

Heuristics for placing filters

Heuristics for placing filters

  • Human judgment
    • Prioritize near parks, areas with prior problems
    • Can the bin collection truck turn around?
  • The use of automation
    • Seed ideas
    • Choice overload

Where should the filter go?

Where should the filter go?

  • Greedy: the road with the most rat-runs
  • Whack-a-mole

Where should the filter go?

  • Only one entrance per cell
    • Expensive, likely unpopular
    • Very different results for residents
    • Simpler crossings along the perimeter

Heuristics for placing filters

Split large cells (intuition)

Heuristics for placing filters

Minimum cut of the graph

Heuristics for placing filters

  • Trivial solutions near the perimeter
  • How large are the two resulting cells?

Heuristics for placing filters

Gridded streets?

Heuristics for placing filters

Part 4: Next steps

  • polish, continue testing with local authorities
  • share proposals online
  • training material (user guide, video tutorials)
  • popular feature requests
    • bus gates
    • one-way streets
    • demographics – who lives near interventions?
    • mode shift / traffic dissipation

Current usage

  • Nottingham
  • Campaigning group in Lyon
  • Initial testing in ~10 other places

Engagement model

  • I can help import your city
  • Do you have travel demand data?
  • What else should this tool do?
  • I’m looking for help from software engineers & UX designers

LTNs are just the start

  • interventions on perimeter roads
    • safe crossings
    • bus/cycle lanes
    • traffic signal timing

LTNs are just the start

  • other parts of A/B Street can help
    • road space reallocation
    • 15-minute neighborhoods / land use patterns
    • traffic simulation (rapid prototyping)

Where’s the AI?

  • Classic computer science graph algorithms vs machine learning
  • You can’t argue with the results (barring data quality / bugs)
    • The rat-runs show specific shortcuts!
    • The cells show if it’s possible to cut through or not!
  • Explainable systems are perceived as more trustworthy

Engaging communities

  • This tool is meant for everyone
  • People are visual, understand maps
  • Build consensus in live workshops

Conclusion