Imagine New York City With 3,000 Taxis Instead of 13,000

Taxis in New York City
Photo: Getty Images

Large-capacity ride-sharing services could replace 98 percent of taxi service in Manhattan, researchers report this week in Proceedings of the National Academy of Sciences.

“We could drastically reduce the number of vehicles” with a “minor impact to users,” says Javier Alonso-Mora, a computer scientist at Delft Technical University in the Netherlands, who worked on the project while at the Massachusetts Institute of Technology.

Today, there are about 13,000 taxis in use in New York City every day—but by design they usually pick up and drop off a single passenger or group. Some popular transportation startups, such as Uber and Lyft, offer ride-sharing options, but vehicles typically have space for only two passengers at most. 

Research published in Proceedings of the National Academy of Sciences in 2014 found that 80 percent of Manhattan taxi trips could be shared by two riders, but the work didn’t take into account new riders joining after a trip has already begun. In addition, the 2014 work and other studies of ride sharing either limit the number of riders or they don’t study the effects of letting customers choose different pick-up and drop-off locations from each other, Alonso-Mora says. So the real benefits for large-capacity vehicles haven’t been determined before. 

Using a randomly picked week of New York City taxi data as input, the researchers created a computer program that produces a route for ride-sharing vehicles that minimizes passenger delay—both the time riders spend waiting for a ride and the delay caused by deviating from their route so their vehicles can pick up new passengers. The program penalized requests not completed.

It works like this: After a set time interval such as 30 seconds, the program checks for a new ride request and adds the request to a queue. By considering unfulfilled requests and trips already in progress, optimization algorithms compute which passenger should be picked up by which vehicle and where each vehicle should go.

The researchers tested the simulation with vehicle capacities of one person (traditional taxi), two (UberPool or Lyft), four (car), and 10 (minivan)—stopping at that number because of the extra computational power needed for simulating even higher capacities. They used a maximum of 2 minutes of waiting with a 4-minute delay, 5 minutes of waiting with a 10-minute delay, or 7 minutes of waiting with a 14-minute delay—similar to the 5- to 10-minute wait it would take to park a car in a busy area like New York City, Alonso-Mora says.

They found that it takes only 2,000 ten-passenger vehicles or 3,000 four-passenger vehicles to meet 98 percent of Manhattan taxi demand every day. (The leftover 2 percent would be lost because of the set delay constraint.) The mean wait would be 2.8 minutes, and the mean trip delay would be 3.5 minutes.

“You allow drivers the possibility to make the same amount of money working less,” says MIT computer scientist and collaborator Daniela Rus. So, she says, instead of taking jobs away from taxi drivers, this would let the same number of workers make the same amount of money in fewer shifts. (The New York Taxi Workers Alliance did not respond to a request for comment.)

Alonso-Mora believes the work is evidence that companies should expand large-capacity ride-sharing options, or buses should switch to more flexible, on-demand schedules. (Which option is better isn’t quite clear yet, he adds.)

Alan Erera, an industrial engineer at Georgia Institute of Technology who was not involved in the research but studies ride sharing, writes in an email: “I am very optimistic about the tremendous value that real-time ride-sharing systems hold for dramatically improving roadway and vehicle fleet utilization for moving passengers.”

However, he says, “these results are optimistic, since they assume that everyone is willing to share rides with anyone else, and will sacrifice their own convenience for system optimality. In reality, some will always prefer to ride alone.” He adds that the calculations don’t take into account the extra waiting time it takes for each passenger to get in and out of the vehicle and the amount of time the car is idle, nor any variability in travel time, which “skews the results somewhat away from conservatism.”

“Building new forms of transit by extending the Uber/Lyft model with automated vehicles and/or better ride-matching optimization,” he writes, “will actually only increase congestion and total vehicle-miles travel unless we can find approaches that ensure many riders will pool together and share trips.”

Alonso-Mora says that to avoid riding with strangers, users could set a preference for that in a theoretical app. He also notes that the algorithm can be tuned to factor in extra waiting time and travel-time variability, but the group found only “small differences” in system performance with different travel times. To deal with another potential challenge—tariffs—drivers could estimate the savings from multiple passengers and incorporate them into calculations. 

He says the next steps could be to take into account predictions of future requests to improve the algorithms. They may also analyze more cities and explore the implications of autonomous vehicles in such a system. 

There are already several companies around the world offering large-capacity ride-sharing services.

“We absolutely can replace 98 percent of taxi service,” says Matthew George, CEO of Bridj, a Boston-based startup that offers 13- to 15-seater bus ride sharing services in several U.S. cities. In March 2016, the company began partnering with the local transport authorities to provide customized public transport in Kansas City, Mo: Unionized city employees are at the wheels of its buses. He says there is a waitlist of 30 cities that want to start a similar program. 

Instead of using bus stops, when a user requests a ride, algorithms consider all nearby requests for similarity. A driver goes to common pick-up and drop-off points within a 5- to 7-minute walk of all rider starting locations and final destinations, respectively.

He says that right now, the average user spends about US $85 a month on its services. He won’t disclose exactly how many users there are but only that buses have gone “millions of millions” of passenger miles—about 80 percent of customers use the buses as their main daily transportation. Some use them in place of taxis—and he expects those kinds of customers to increase in 2017.

Advertisement

Cars That Think

IEEE Spectrum’s blog about the sensors, software, and systems that are making cars smarter, more entertaining, and ultimately, autonomous.
Contact us:  p.ross@ieee.org

Senior Editor
Philip E. Ross
New York City
Assistant Editor
Willie D. Jones
New York City
 
Senior Writer
Evan Ackerman
Berkeley, Calif.
Contributor
Lucas Laursen
Madrid
 

Newsletter Sign Up

Sign up for the Cars That Think newsletter and get biweekly updates, all delivered directly to your inbox.

Advertisement