How realistic are the interpretations of several robo-taxi simulations made in the US, Sweden, and Portugal since 2012? Can we rely on these as discoveries to help indicate the size and range of robo fleets that might be possible?

A recent OECD-sponsored International Transport Forum report regarding a robotaxi simulation in Lisbon Portugal builds on several other similar simulation studies reaching back to the first ones done by Larry Burns at Columbia and Alain Kornhauser at Princeton starting in 2012. The OECD report outlines some of the history. There is considerable consistency in assumptions and findings, although including ride-sharing may also make a small difference (see the Austin study by Fagnant).

The basic observations that our cars are idle 95% of time, and that when not idle they have about 1.2 people in their seats begs the question: “how many fewer cars might be possible?” If all trips, people, times, and locations aligned perfectly on a 7/24 basis and everyone was willing to share vehicles and rides. About 1% of the current car population would be enough. As this is clearly absurd, students of robo fleets simulations answer this question by using real origin-destination (O-D) data and the assumption that all trips within the extent of the data sample would be replaced by a vehicle in a managed fleet of self-driving cars. With these more feasible constraints, simulators typically find that the trips under scrutiny could be handled with something around 10% of the vehicle population serving that same sample of trips. Does that 10% figure portend what would be possible in a world wherein no one—or almost no one—owned vehicles, but rather rented trips in robotic vehicles?

We believe not.

These simulations, constrained by the origin-destination (O-D) data available, to the simulators, generally make a few questionable assumptions, whether stated or implied. In the OECD study, for example, these are:

  1. Rather than cover the entire Lisbon Metropolitan Area (which generates over 5 million person-trips each day), this study focuses on the center city, the Lisbon municipality, with 1.2 million trips per day. Hence the simulation does not include the suburban commuters coming into the center, the suburb-to-suburb movement that is so car dependent, etc. While representative of European cities (says the report), the work provides insight to a somewhat limited set of urban environments within the worldwide context.
  2. Assumes replacement of personal vehicle trips first or at least concurrently with transit and taxi trips, even though people who already own personal vehicles will be natural holdouts.
  3. Repeats the common finding “only 10% of cars would be required”. This one-in-ten finding is generally arrived at using similar assumptions and simulation criteria as were used in other studies. This is not a fault of the simulation technique, but rather an artefact deriving from a constrained set of origins and destinations, per point 1.

We find all these recent simulations, including this one, to be exciting, non-scalable and somewhat dangerous:

  • Exciting because they promise such a miraculous solution;
  • Non-scalable because they are based on core, urban peak travel and do not account for many other travel purposes, distances, and especially does not account for many reasons people would still decide to own a vehicle. From our environmental perspective, owning increases vehicle miles traveled (VMT), and a myriad of additional footprint elements;
  • Dangerous because these simulation results invite complacency now, and promise disappointment later if reality falls significantly short of the studies’ implications.

At, our stated goal of “four times the 2010 PMT by 2050 with the same fleet size as in 2010” assumes a natural increase in demand for person miles traveled (PMT) of 400% by 2050 as predicted by current vehicle doubling time, which we have interpreted as PMT doubling time in order to re-think the problem given the SDC. Hence this implies that the world vehicle population would be the same in 2050 as it was in 2010, although turning over 400-500% faster due to heavy use, shared use, and the tragedy of the commons. By the last we refer to the tendency for humans to afford less regard to public or shared goods or resources than they might to personally-owned goods or resources.

In order to cap the vehicle count, a huge portion of the current vehicle count in the developed world must shift to the countries of new demand and currently lower vehicle counts. Hence, many countries with lower car populations would see increases. Other countries (in North America and Europe) would see drops and presumably would appreciate the safety and livability-recovery advantages outlined in the OECD study.

But crucially, there would not likely be a uniform vehicle-per-capita portion worldwide.

In 2010 the world vehicle population was at 0.148 of human population, the US was at 0.769. In 2050 the world vehicle population would have to be at 0.102 per capita to remain the same count as 2010. Why? World population was 6.9B in 2010 and is expected to be around 10B, reaching a plateau by 2050; hence 6.9/10 * 0.148 = 0.102. We do not expect the US/EU car population to drop to 0.102 per capita. Rather the aggregate world population would do so while the remaining vehicle count would be greatly re-distributed. Specifically, we would expect the developed-world motor vehicle per capita population to drop precipitously, but only by a factor of three or four rather than ten, since we believe the subject simulations are greatly underconstrained. That means many countries, although with growing vehicle populations, would still have lower per-capita ratios.

One of the things we seek to better understand at is what this potential new distribution might be and how to get there. This OECD study points to massive available CAPACITY in the system, but the assumptions regarding how that capacity can be accessed are still wanting.

Bern Grush and John Niles.

Leave a Reply