BRENNEN

Industries / Technology / Mapping and geospatial

Street maps

Street mapping records every road, its name, how it joins other roads and the rules that apply on it, for printed road maps, navigation devices and phone apps.

7

Milestones achieved, from 1904 to 2015.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1904

    Road maps for motorists

    Before cars, most travel between cities went by rail, and Rand McNally's first map, printed in 1872, was a railroad guide. In 1904 Rand McNally printed its first automobile road map, of New York City and the country around it. Gulf Oil then began handing out free road maps at its new drive-in gas station in Pittsburgh: Hagerty, a magazine published by a car insurer, says it began in 1913, the year the station opened, and Maps of PA, a Pennsylvania map collector's site, says it began the year after. Within a decade most large oil companies had map programs, and Hagerty says they kept them going until the oil embargoes of the 1970s.

    Lessona map paid for by another business lasts only as long as that business needs it.

    hagerty.commapsofpa.com

    Credit

    Rand McNally; Gulf Oil

  2. 1985

    A car finds its street

    Etak, founded in 1983 by Stan Honey and five colleagues from the research institute SRI with seed money from Nolan Bushnell, the founder of Atari, began selling its Etak Navigator to drivers in the fall of 1985. It worked before GPS. The car kept track of itself by dead reckoning, measuring the distance and heading it had travelled, and removed the build-up of error by map matching: comparing the path driven with the roads on a digital map, on the reasoning that cars stay on roads. The map sat on cassette tapes holding 3.5 MB each. To build it, Etak scanned US Geological Survey maps and aerial photographs and used them to correct Census Bureau street files, whose coordinates were not accurate enough. Honey writes that the consumer market "would have to wait for 20 years to mature"; what paid was licensing the map, first to UPS, and News Corporation bought Etak in 1989 because it had "the highest quality digital roadmap database".

    Lessonthe digital map was worth more than the device built to show it.

    ethw.org

    Credit

    Stan Honey and the Etak team; Nolan Bushnell

  3. 1990

    A nationwide digital map

    For the 1990 census the US Census Bureau and the US Geological Survey built TIGER (Topologically Integrated Geographic Encoding and Referencing), a database of every road, railroad, river and boundary in the country that records which lines meet where, not only their shapes. The Census Bureau calls it "the first nationwide digital map of roads, boundaries and water." It placed addresses in the right census areas and produced the maps used for the count, and its files are updated every year and free to download. At TIGER's 25th anniversary in 2015, US Commerce Secretary Penny Pritzker said it "helped grow an entire industry — the GIS industry".

    census.govresources.data.gov

    Credit

    US Census Bureau; US Geological Survey

  4. 2009

    Faces blurred by machine

    Street-level imagery is photographs of public streets taken from cars that drive every road, as Google Street View does, and it captures faces and license plates along with the street. In a paper at the 2009 IEEE International Conference on Computer Vision, Andrea Frome, Luc Vincent and colleagues at Google described a fully automatic system that blurred more than 89% of faces and 94 to 96% of license plates in test sets drawn from Street View.

    Lessona map made from street photos has to remove people from them at the same scale as it collects them.

    research.google

    Credit

    Andrea Frome, Luc Vincent and colleagues, Google

  5. 2012

    A map launched too soon

    Apple launched its own new maps with iOS 6 in September 2012, and complaints about them followed at once. On 28 September Apple's chief executive, Tim Cook, published a letter saying "we fell short", and pointed customers to rivals while Apple fixed Maps: the Bing, MapQuest and Waze apps, or Google's and Nokia's maps on the web. In June 2018 Apple said it was rebuilding the map from its own data, gathered by iPhones and by its own fleet of cars carrying sensors and cameras, after four years of work; TechCrunch reported that reliance on other companies' data had been one of the map's main problems.

    Lessona navigation map assembled from other companies' data took Apple six years and its own survey fleet to replace.

    macstories.net9to5mac.comtechcrunch.com

    Credit

    Apple; Tim Cook

  6. 2014

    Machines read house numbers

    Geocoding means turning an address into a point on a map, and it depends on knowing which number belongs to which building. Ian Goodfellow and colleagues at Google trained a deep convolutional neural network, a kind of AI model for images, to read whole house numbers from Street View photographs in a single step. It read more than 96% of complete numbers correctly on SVHN, a public test set, and more than 90% on a harder Street View set with tens of millions of labeled numbers. The team wrote that it could transcribe every view of a street number it held for France in less than an hour, and that the system had helped extract close to 100 million street numbers worldwide.

    arxiv.orgarxiv.org

    Credit

    Ian J. Goodfellow, Yaroslav Bulatov, Julian Ibarz, Sacha Arnoud and Vinay Shet, Google

  7. 2015

    Carmakers buy a map

    Nokia paid $8.1 billion in 2008 for Navteq, a US digital map maker that became part of its map business, HERE. On 3 August 2015 Nokia agreed to sell HERE to Audi, BMW and Daimler for €2.8 billion ($3.1 billion), with the three German carmakers holding equal shares. AFP reported that HERE's maps were used in four of every five cars with built-in navigation in Europe and North America. Daimler's chief executive, Dieter Zetsche, said the aim was "to secure the independence of this central service for all vehicle manufacturers".

    Lessonrival carmakers chose to own one map together rather than depend on a technology company for it.

    au.finance.yahoo.comtech.eu

    Credit

    Nokia; Audi, BMW and Daimler

Watch for

  1. Map fixes within a day

    HERE says its map takes in what sensors on cars around the world record, including video, radar, lidar and driver reports, and compares it with the map. When they show something that does not match, such as new construction, changed lanes or a closed road, AI systems flag it, update the map and send the correction back to vehicles "typically within 24 hours".

    Achieved whenan outside test times a sample of real road changes from the day each happened to the day it appeared in HERE's delivered map.

    truckinginfo.com

    Claimed by

    Remco Timmer, senior vice president at HERE Technologies, at CES, as reported by Heavy Duty Trucking, January 2026

Open

  1. 01

    Speed limits against the signs

    A public, repeatable test of the speed limits that map providers give for roads, checked against the limits posted on the signs. For a random sample of road segments in several cities, the posted limit is read from a dated photo of the sign and compared with the limit in each map.

    The brief
    Credit · unclaimed
  2. 02

    Days until new roads appear

    A tracker that times how long newly opened roads take to appear in each major map. For each opening announced by a transport agency, it asks each map's routing service every week for a route that only makes sense over the new road, and records the first week the route uses it.

    The brief
    Credit · unclaimed

Open 01 · the brief

Speed limits against the signs

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public, repeatable test of the speed limits that map providers give for roads, checked against the limits posted on the signs. For a random sample of road segments in several cities, the posted limit is read from a dated photo of the sign and compared with the limit in each map.

Why it is still open

Google says AI models read speed-limit signs in Street View and other imagery to keep its limits current, but it publishes no error rate. Nexar, a dashcam company, compared its own sign readings in Phoenix, Arizona with the City of Phoenix's records and with OpenStreetMap, found 3 to 6% of limits wrong, and found 47% of limits missing from OpenStreetMap; it tested no commercial navigation map, and the readings it compared against are its own product. The World Bank's Global Road Safety Facility compared the limits in HERE, TomTom and OpenStreetMap for Bogotá, Buenos Aires, Rio de Janeiro and Mexico City and found the same street given anything from 20 km/h to 80 km/h depending on the platform, but its March 2025 report gives no error rate for each map against the signs. Public analyses on GitHub, such as vgXhc's study of tertiary roads, look at OpenStreetMap alone. No test across providers was found.

blog.googleinfo.getnexar.comblogs.worldbank.orggithub.com

Built with what exists

Mapillary street-level photos, which are open under the CC-BY-SA licence, and Panoramax, an open French alternative, both of which record when each photo was taken; the tester's own photos of signs, taken from the pavement and framed on the sign, not on passers-by; speed limits from TomTom's Snap to Roads API, the Mapbox Directions API's maxspeed annotation (in beta), OpenStreetMap's maxspeed tags and Overture's transportation data; QGIS and Python. Google's speed-limit service is open only to holders of its Asset Tracking licence. Read each provider's terms on publishing comparisons before starting.

help.mapillary.comdeveloper.tomtom.comdocs.mapbox.comdevelopers.google.com

  1. First step

    In one city, draw 200 road segments at random from OpenStreetMap, find a dated Mapillary photo of the speed sign on each, and record the posted limits.

  2. Done looks like

    2,000 sign-checked road segments in five cities in at least two countries, with each provider's share of exact matches, wrong limits and missing limits, published with links to the photos and the code.

Open 02 · the brief

Days until new roads appear

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A tracker that times how long newly opened roads take to appear in each major map. For each opening announced by a transport agency, it asks each map's routing service every week for a route that only makes sense over the new road, and records the first week the route uses it.

Why it is still open

Lyft's mapping team checked OpenStreetMap against imagery in 30 North American cities and found its core road features correct more than 95% of the time; it showed one case, the new terminal roads at New York's LaGuardia Airport in July 2020, that OpenStreetMap had "shortly after opening, faster than many other maps", but it timed no other map. HERE's 24-hour figure is its own (see Watch for). A 2026 University of Chicago student project with HERE on "map freshness" detects changes inside HERE's own data rather than comparing maps. No comparison of time to appear across providers was found.

api.openstreetmap.orggithub.com

Built with what exists

Road-opening notices from state transport departments and national road agencies; the routing services of Google (Routes API), TomTom and Mapbox, and OSRM or GraphHopper, two free route planners, running on current OpenStreetMap data; a scheduled Python job; a browser agent such as Browser Use 1.0 to gather opening notices from agency news pages (see the Note "Deleting 80% to start over"). Read each provider's terms on publishing comparisons before starting.

  1. First step

    List ten road openings announced for December 2026 and January 2027, and save each map's current route for every one before the road opens.

  2. Done looks like

    50 road openings in at least three countries, each with its official opening date and the first week each of five maps routed over it, published as a table with the median number of days for each map.