In recent years, lots and lots of pundits — including me — have speculated about how self-driving cars will change American cities. But almost all the talk has been just that — speculation. Because we can't collect data on the social effects of a technology that isn't available yet.
But a recent study does the next best thing: it starts with detailed data about today's traffic patterns, and then uses a computer simulation to predict what would happen if drivers switched to taking rides in self-driving taxis. The research, by University of Texas at Austin professor Kara Kockelman and the University of Utah's Daniel Fagnant, provides unprecedented insight into how self-driving vehicles will change the urban landscape.
The researchers found that a company running a hypothetical fleet of self-driving taxis can make big profits charging $1 per mile, a fraction of what American taxis cost today. Each new self-driving taxi added to the fleet eliminates the need for about 10 privately owned cars. And thanks in part to the possibility of on-the-fly carpooling, the system can gracefully handle the demand spikes that occur during rush hour.
Self-driving technology is going to dramatically change how people get to work, to school, and around town. It's going to make a car-free lifestyle a lot more attractive to a lot of people, make taxis affordable to a lot of people who can't afford them today, and help ease both traffic jams and parking shortages.
1) One self-driving taxi can replace around 10 private cars
Traffic planners collect detailed data about metropolitan traffic patterns to help them understand how changes to roads will affect traffic patterns. Kockelman and Fagnant realized they could use data from the Austin area (where UT is located) to build a precise simulation of how a self-driving taxi network would perform in the real world.
They started with existing traffic patterns and then created a computer model in which a small fraction of car commuters (1.3 percent in their main simulation) ditched their cars and relied on self-driving taxis to get around instead. Fagnant wrote a computer program that simulated the movement of each self-driving vehicle over the course of a 24-hour day.
In one 24-hour period, a fleet of 1,700 virtual self-driving vehicles completed 56,324 rides, which works out to a bit more than 30 rides per vehicle. Given that the typical privately owned car completes around three trips per day, this means each taxi does as much work as 10 private vehicles.
This finding is consistent with the experience of car-sharing services like Zipcar. One study found that each vehicle added to a car-sharing network corresponded to subscribers owning 9 to 13 fewer cars. This is a big social benefit, especially in cities that suffer from parking shortages.
2) Self-driving taxi rides are going to be really cheap
In the United States, the driver's labor is the single largest expense for today's taxi services. But self-driving taxis work 24 hours a day, seven days a week, for free. So taxi companies will be able to slash their fares.
How much cheaper could taxi service get? Fagnant and Kockelman assume self-driving taxis will cost around $70,000 — less than they cost now but a plausible guess for what they'll cost a decade from now. They also assume their virtual taxi company will charge $1 per mile, which is less than a third of the $3.65 per mile that conventional taxis cost in Austin. Despite these low fares, their simulated taxi service is highly profitable, earning a 13.9 percent annual return on investment.
There are several reasons to think fares will fall even lower than $1 per mile in the long run. Competition among self-driving taxi services should push prices — and therefore profits — down. Also, Fagnant and Kockelman assume self-driving taxis will have the same operating costs — about 50 cents per mile — as conventional vehicles. If self-driving taxis make super-efficient electric vehicles practical — as some people have predicted — then savings on energy should translate into lower fares. It's also likely that progress in sensor technology will eventually bring the cost of self-driving taxis below $70,000.
But even $1 per mile is cheap enough to make self-driving taxis a practical commuting option for many people, especially in urban areas. I live about two miles from Vox Media headquarters in downtown DC, so commuting by self-driving taxi would be a no-brainer for me — the $2 cab fare would be only slightly more than the $1.75 I currently pay to ride the bus to work, and it would be a lot more convenient.
3) Rush hour will work better than you think
When I wrote a few months ago that self-driving taxis will mean the end of mass car ownership, the most common objection I heard regarded rush hour. Sure, people said, you can take taxis everywhere the rest of the time. But during the morning commute, everyone is going to work at the same time, so there's not much room for car-sharing.
But the data tells a different story. Fagnant tells me a single self-driving taxi was able to complete several trips during each rush hour period.
"You don't have a uniform peak where everybody leaves their house at 7:30 and arrives to work at 8," he says. "You have some people who leave earlier, some who leave later, some who do reverse commuting."
According to Fagnant, an autonomous taxi might pick up someone in an outer-ring suburb and take him to his job in an inner-ring suburb. Then it could pick up someone in the inner-ring suburb and take her to her job downtown. It could then pick up someone who lives downtown and take him on a reverse commute out to the suburbs, before turning around to take another customer into downtown. It's true that some taxis will have to travel empty out to the suburbs to get the next passenger, which isn't efficient. But this doesn't have to happen as much as many people think.
4) Self-driving taxis can boost carpooling
The self-driving taxi network in the simulation was able to organize just-in-time carpooling arrangements, matching passengers who happened to be traveling in the same direction. This is similar to the UberPool and Lyft Line carpooling services that exist today.
Carpooling was still relatively rare, with about 4.5 percent of trips involving shared rides. However, this was still significant because it was concentrated during morning and evening rush hour, when cars and space on the roads were scarce.
And surprisingly, passengers who agreed to the carpooling option didn't see their average trip time go up. Carpooling does slightly increase the average distance each passenger travels, since sometimes one passenger has to go out of his way so another can be dropped off. But this increased travel time was offset by the fact that carpooling reduced the amount of time people spent waiting for a taxi to arrive. This effect was most dramatic at rush hour, when carpooling slashed the average time passengers spent waiting for a car from nine minutes to four and a half minutes.
5) The economics of self-driving taxis get better as the network grows
Due largely to computational limitations, the study focused on scenarios in which only a small percentage of trips in the Austin area — about 1.3 percent — are taken in self-driving taxis. But Fagnant and Kockelman also did some simulations with higher (but still single-digit) market share for self-driving taxis, and found that the economics of self-driving taxis improve as the size of the network grows.
If a network has more cars and passengers, it becomes easier for the system to find pairs of passengers who are a good fit for carpooling. The data reflects this. As the market share of self-driving taxis increased from 1.3 percent to about 6.5 percent, the fraction of carpooling trips rose from 4.5 percent to 5.9 percent.
With more taxis on the road — including many taxis whose passengers are willing to carpool — the average wait time for a vehicle fell. The length of the average trip — including time spent waiting for the car to show up — declined by 30 seconds from 14 and a half minutes to 14 minutes.
Finally, economies of scale should make self-driving taxi fleets more profitable as they get larger, as cars will spend less time waiting around (or driving back and forth) to get the next passenger. Assuming healthy competition, this should translate to even cheaper fares.
The simulation didn't capture every benefit of self-driving taxis
This research provides a lot of insight into how self-driving taxis will change the transportation market. However, some likely effects weren't captured by the simulation.
One is changes in housing patterns. Self-driving taxis will make housing that's relatively close to downtown but not well-served by transit relatively more attractive. At the same time, because self-driving taxis won't need parking, cities will be able to devote more space to housing people and less to housing cars. It's hard to predict exactly how this will change cities, but it seems likely that the parts of the city that are made newly convenient will experience a building boom as nearly empty parking lots are converted to housing.
Another important question not addressed in the research is how self-driving taxis will interact with transit. For example, self-driving taxis could allow increased use of the subway in the suburbs. Instead of driving to the subway station and paying several dollars to park in the park-and-ride (daily park-and-ride rates for subways in the Washington, DC, area, for example, average about $5), someone who lived two miles from the subway could take a taxi, paying $2 or or less each way. That could make subway stations practical in parts of the city that are currently too sparse to justify it.
Of course, the model may not work for everyone. Fagnant's simulation focused on a 12-by-24-mile rectangle enclosing the densest parts of the Austin metro area, excluding trips that went into outlying areas. People in these areas have long commutes, which leaves fewer opportunities for vehicle sharing. For them, owning a car outright may make more sense for the foreseeable future.
But there are millions of people who will find self-driving taxis to be a more convenient and affordable way to get to work than the alternatives.