Industries / Transport and vehicles / Ride-share and taxis
Matching riders and drivers
Matching is the work of pairing each rider with a nearby car and driver, by street hail, phone call or app.
6
Milestones achieved, from 2010 to 2026.
1
To watch for: announced, not yet confirmed outside.
1
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2010
A phone app books a car
UberCab launched in San Francisco on June 1, 2010, by the account of Conrad Whelan, its first engineer. At first it offered only black cars ordered through a phone app ("we were just limos at the time," early employee Austin Geidt recalled).
Credit
Garrett Camp, Travis Kalanick and the early Uber team
2011
Licensed cabs join the app
Hailo launched in London in the week of November 4, 2011, set up by three London cab drivers and three internet entrepreneurs. It showed riders the licensed black cabs nearby and sent each request to the nearest driver, then to the next nearest if the first declined.
Credit
Jay Bregman, Caspar Woolley, Ron Zeghibe, Russell Hall, Gary Jackson and Terry Runham
2014
Strangers share one car
Uber launched UberPool and Lyft launched Lyft Line in August 2014. Riders headed the same way share a car for a lower fare. A 2024 paper from the University of Illinois and Penn State says UberPool served over 100 million trips by 2016, and that Lyft Line made up 30% of Lyft rides, closer to 50% in New York and San Francisco.
Credit
Uber; Lyft
2016
A local rival wins China
On August 1, 2016, Uber sold its China business to Didi Chuxing in return for a 20% economic stake in the combined company, after losing more than $2 billion there.
LessonMatching gets better as more riders and drivers join, so the app with the most local users can outlast a richer rival from abroad.
Credit
Didi Chuxing (Cheng Wei); Uber
2024
Shared rides fade
Using New York City trip records, Pandey, Liu, Dong, Lehe and Gayah found that the share of app rides requested as shared fell from about 30% in February 2019 to 3% in January 2024. They tie the drop before COVID to Uber cutting its UberPool discounts, and the failure to recover to a loop: fewer sharers mean fewer good matches, which turns more riders away.
LessonSharing only works above a certain number of riders, and once it falls below that it struggles to come back.
Credit
Ayush Pandey and colleagues, University of Illinois Urbana-Champaign, Penn State and Jackson State University
2026
Robotaxis inside a rival's app
From September 9, 2026, Lyft riders in Nashville who order a regular ride may be matched with a driverless Waymo car at no extra cost if the trip starts and ends in Waymo's 60-square-mile service area. Waymo owns the more than 100 cars there, and Lyft's Flexdrive unit helps maintain them.
Credit
Waymo; Lyft
Watch for
One robot fleet, two rival apps
Baidu's Apollo Go began road tests of its RT6 robotaxi in the London borough of Brent, under deals with both Uber and Freenow by Lyft. Safety operators are on board, and paid rides are not expected before 2027, subject to approval.
Achieved whenLondon riders can book driverless Apollo Go rides through either app and a regulator or news report confirms it.
Claimed by
Baidu Apollo Go, Uber and Freenow by Lyft, July 2026
Open
01
What sharing could still save
A rerun of the 2017 MIT ride-sharing study on today's New York trips. Take recent public trip records, run an open matching algorithm, and publish what share of trips could have been shared with a set extra delay, next to the 3% that riders actually requested.
The briefCredit · unclaimed
Open 01 · the brief
What sharing could still save
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A rerun of the 2017 MIT ride-sharing study on today's New York trips. Take recent public trip records, run an open matching algorithm, and publish what share of trips could have been shared with a set extra delay, next to the 3% that riders actually requested.
Why it is still open
The MIT study used one week of 2013 yellow-taxi data and found that 3,000 four-seat cars could serve 98% of Manhattan taxi demand. Since then shared rides fell to about 3% of requests. This research found no published rerun on app trip data from after the pandemic.
Built with what exists
NYC Taxi and Limousine Commission (TLC) high-volume for-hire trip records (public monthly files); Python with DuckDB or pandas; an open map of the road network from OpenStreetMap; the algorithm described in the PNAS paper.
First step
Download one month of trip records, rebuild the 2013 test week from the yellow-taxi files, and check that your matcher gets close to the paper's 98%.
Done looks like
For at least four recent months, a published share of Manhattan trips that could be paired with no more than 5 minutes of added wait and detour, with code that reproduces the 2017 paper's main figure within 2 percentage points on its 2013 week.