The Verge of Change
Automakers are pursuing aggressive timetables to bring "Level 4" autonomous vehicles to market, where no input from a human driver is necessary. Volkswagen and Audi are aiming for 2019-2020 IEEE, GM in 2020 Wall Street Journal, and Ford and Toyota in 2021 Reuters. Though these may be the optimistic forecasts of futurists speaking to company shareholders, there are many reasons to believe that a fully-autonomous vehicle (AV) could be available to the public for use on general roads in the very near future. How can transportation planners help cities prepare for this technological change?
By the book, we as transportation planners would investigate future AV scenarios using a travel model after collecting travel behavior data with an AV-oriented survey. Given the long runways required of this kind of survey, if the data needed to model AV behavior have not already been collected today, then AV's will arrive before the models designed to study their effects.
This same phenomenon, where change happens faster than our textbook planning methods can react to, occurred from 1960 to 1980 when the participation rate of women in the workforce almost doubled. The plot below shows the labor force participation rate in the United States for men and women over the last half-century.
Planning activities and travel models in the 1960's before the shift did not predict the profound impacts that women entering the workforce would have on the labor market, vehicle ownership, daily activity patterns, and traffic congestion. A behavioral survey in the late 1960's could not have captured the imminent shift. Most importantly, the typical ten-year planning cycle in place caused planners to miss the rapid change altogether, instead building reactive models rather than predictive ones.
Like in the 1960's, planners today are not prepared to respond quickly when data does become available. Planners have tried to solve these problems in different ways:
- Sketch planning models Some agencies use small-capability, quick response frameworks to consider high-level scenarios. Though these tools can provide useful information, they have limited use for rigorous alternatives analyses.
- Existing trip-based or activity-based models A number of planners and researchers have adjusted model parameters — trip rates in trip-based models, daily activity patterns in activity-based models, or link capacities in highway assignment models — based on assumed responses to AV's. These studies can be useful, but often simply validate the analysts' initial assumptions due to the way these survey-based models are built.
Let us consider a third path: estimate demand with a passive data model and couple it with the evolutionary replanning algorithms in the MATSim open source project. This combination allows us (1) to respond quickly when passive data become available from AV's and (2) to examine many different ways that the transportation networks might be affected in the near-term by AV's without needing observed behavior.
Passive Data Model
The image below shows an idealized data flow for travel demand models. Trip-based and activity-based models both begin with small sample survey data that inform econometric models of travel behavior. With our passive data model, a large sample of passive data are merged using a simulation to generate daily trip patterns. For more details on this demand modeling approach, read about CityCast.
MATSim and AV's
Autonomous vehicles will surely change travel behavior and land use patterns in unforeseeable ways in the medium- to long-term. In the nearer term, however, people are likely to use AV's to aid in their existing daily patterns, making trips between their existing home and work locations. Knowing which of these trips are most likely to be replaced or supplemented with an AV is the important and difficult question for near-term planning. When we couple MATSim with our passive data model in CityCast, we have a powerful, flexible tool for studying the particular trips that AV's are likely to carry.
Open source MATSim allows us to examine AV's in a flexible way because of its evolutionary nature. It is a regional-level microsimulation that optimizes individual people's daily activity pattern, mode choices, and route choices. It optimizes each person's daily plan by iterating through random mutations of an initial set of activities and travel plans, which in CityCast come from a passive data demand model. At the end of each iteration in MATSim, each person receives a score (positive for time spent doing activities and negative for time spent traveling).
To calculate the scores, every travel mode, including AV's, are assigned different disutilities, or negative costs per minute. For example, time spent waiting for a bus counts more negatively than time walking to the bus or time inside an AV. At the end of each iteration, people keep the plan that maximizes their score. Eventually, after several dozen iterations, the people within the system have their individual best plan, whether than includes AV's trips or not.
Planning for AV's in the Near-Term
We have CityCast running in Asheville, NC, where we will be testing several near-term AV scenarios. Our aim will be to explain the assumptions we have made in plain language and to explain what the results would mean for policy and infrastructure decisions.
We have set up our first set of analyses to test one aspect of AV's that has generated much debate: AV ownership. Lauren Isaac, who has spent a lot of time thinking about the future impacts of AV's, presents a conceptual framework like this (2016):
- Utopia Shared, affordable AVs will reduce parking needs, improve mobility for disadvantaged populations, and lower greenhouse gas emissions.
- Nightmare Privately owned AVs will lower the burden of travel, enabling longer commutes, facilitating urban sprawl, and exacerbating traffic congestion.
Our second set of analyses will explore the rates of disutility that we give AV's as compared to conventional vehicles and transit. Stay tuned in the coming weeks for the results of our tests. Watch the AV category.