By now you have probably seen many stories with headlines like these:
- “Autonomous Taxis: Why You May Never Own a Self-Driving Car”;
- “Self-Driving Cars Will Transform the Human Environment”; and,
- “Shared Mobility Solutions Improve Social Inclusion.”
The last of these is from a news release announcing the latest iteration of a series of simulation modeling runs on how traffic in Lisbon, Portugal might be changed by a hypothetical autonomous vehicle revolution. Other simulations of this kind have been run on cities ranging from Austin to Singapore. Since the devil is always in the details, I decided to do a careful assessment of the assumptions made in this newest simulation, done under the auspices of the OECD’s International Transport Forum (ITF), several of whose conferences I’ve participated in.
The report in question, announced in ITF’s July newsletter, is called “Shared Mobility: Innovation for Liveable Cities.” The news release says the simulation demonstrated that replacing all conventional car and bus trips in Lisbon with automatically dispatched door-to-door services “would provide the same level of mobility to citizens using only 3% of the current number of vehicles.” It would also “cut emissions by one-third and on-street parking would become superfluous.”
Those are pretty amazing claims, so I downloaded the report on which they are based. The modeling uses an extensive database of origin-destination trips for Lisbon. It assumes that private cars are replaced by autonomous 6-seat shared taxis, and buses are replaced by autonomous 8- and 16-passenger taxi-buses. Commuter rail and heavy rail would continue as at present, though rail transit would lose half its current passengers.
What would it take to bring about this new state of affairs? First of all, built into the model is that privately owned cars are banned. The 45% of Lisbon residents currently driving themselves could walk or use rail—or be assigned to either a shared taxi or a taxi-bus. The assignment would be made by a central dispatch system, following an algorithm. The report asserts that this would provide former drivers with the “same level of mobility” as they now enjoy, but that seems hard to reconcile with reality. A shared taxi or taxi-bus would in many or most cases need to make multiple stops to pick up others and multiple stops to drop others off. (Think of this as giving up your car to use Super Shuttle everywhere.)
Also, at least in the United States, large numbers of people engage in trip-chaining, either on the way to work or on the way home, and that does not appear to be included in the algorithm. How would a shared-vehicle trip allow you to stop at the dry cleaners, and then the supermarket, on the way home? Or to stop at Starbucks on the way to work?
Moreover, it turns out that the simulation covers only the 37 sq. mi. of central Lisbon, not the suburbs. That would appear to mean that it covers only those who both live and work within the central area. I don’t know enough about the Lisbon metro area to guess what fraction of the trips that excludes, but this model would be a very poor fit for the typical suburbanized U.S. metro area, where suburb-to suburb commuting is the predominant pattern. And despite eliminating all Lisbon’s current bus passengers (14% mode share) and half the 14% who use rail, the report waves away potential labor opposition by stating that “labor issues could be mitigated by the fact that more Taxi-Buses than conventional buses will be needed.” How is that supposed to help preserve driver jobs, when the Taxi-Buses will be driverless?
In digging through my files to write this article, I discovered a 2015 article from Bern Grush and John Niles’ website, http://endofdriving.org, “What Do Robo-Taxi Simulations Tell Us?” Reviewing an earlier ITF Lisbon simulation, they noted some of the same points I arrived at above:
- Not taking seriously people’s desire to retain and use their presently owned vehicles;
- Not including the suburbs, which Grush and Niles say account for the large majority of the metro area’s 5 million daily trips (vs. 1.2 million in the central city modeled); and,
- Misleadingly concluding that only 10% of the current number of vehicles (in that earlier Lisbon simulation) will be needed, due to leaving out the longer trips within, and to and from, the suburbs.
Grush and Niles conclude that such simulations, while exciting, are both non-scalable and dangerous. They are non-scalable because urban core peak travel is different from the more diverse array of trips people make for many purposes and at all times of day. And they are dangerous because these simulation results “promise disappointment later if [when] reality falls significantly short of the studies’ implications.”
(This article first ran in the August 2016 issue of Surface Transportation Innovations)