Autonomous vehicles: Forecasting

Since much of the technology and adoption of autonomous vehicles is highly unknown, confidently forecasting future impacts will be extremely hard, if not impossible. There are some assumptions and extrapolations that can be explored for certain models, but much of the work of preparing for a CAV future likely requires a different approach to traditional forecasting that pivots from today’s reality, in reaction to slight and predictable changes in demand and network supply. Given the numerous unknown impacts of CAVs, in terms of vehicle adoption, pricing, efficiency, policies, and behavioral response, a simple pivoting framework may no longer be valid.

Modelers should be prepared to produce forecasts as CAVs appear on the market and unknowns drop away, but in early planning stages, modelers may need to focus on enumerating and understanding the impacts of individual unknowns. For instance, what are the implications of 20% of regional households owning CAVs, versus an 80% adoption rate? What are the VMT, emissions, and equity impacts at the upper and lower bounds of this single parameter change? Separately, what are the impacts of changing all these other assumptions at their upper and lower bounds? This method is not explicitly forecasting what the future will be, but seeking to understand the potential impacts of individual factors on outcomes. Testing a variety of factors may help define a likely scenario of combined assumptions that represent specific policies or potential futures. In the absence of forecasts, we still have a clear understanding of potential impacts on travel behaviors as a reaction to different future choice sets, so models are still worthwhile and necessary tools to explore CAVs at such an early phase in their development.

As CAVs become a more tangible concept in the future, it may be possible to refine assumptions and even to begin producing forecast ranges with increased confidence. Even today, drawing similarities between TNC/rideshare users and future CAV early adopters, for instance, might allow for preliminary forecasting.

# Near Term Forecast

The assumption-driven sensitivity testing process above is useful for preliminary understanding of long-range impacts of CAVs, in line with a traditional long-range transportation plan at an MPO, for instance. However, there may be a need to understand more near-term impacts of CAVs, given that a transition period may look quite different from both today and a future 30 years away, and may exist for a relatively long period of time. Short-term forecasts, even as short out as 5 to 10 years may be necessary to describe early adoption of mixed CAV and conventional vehicle scenarios.

# Risk and Uncertainty

The transportation planning profession is no stranger to uncertainty, but the range of uncertain outcomes has in the past been either a narrower range of potential outcomes (e.g. longer commutes) or an unforeseen outcome (e.g. women entering the workforce). The potential influence of autonomous vehicles and mobility as a service may produce positive or negative impacts to society depending on the range of policies governing the use of these technologies. The risk of not addressing this uncertainty in forecasting is significant, as future projects or investments may have substantial unforeseen consequences. Forecasting, therefore, must adapt to take uncertainty into consideration more directly.

The uncertainty surrounding the adoption of autonomous and connected vehicles can be addressed by selecting a set of key input assumptions in the travel demand model to vary and the levels that should be tested for each input assumption. Then an experimental design can be applied to design a set of model runs to test these assumption ranges. One method to quantify the risk analysis applies regression analysis to model the key outputs as a function of the input assumption levels and produce a probability distribution of the key model outputs.

A significant challenge to implementing these types of sound forecasting schemes is communicating the results with decision-makers, who often serve in many different capacities and may not have the time or the appetite to discuss dozens of potential hypothetical futures. More research and experimentation is needed into how to effectively engage busy decision makers in to this type of information consumption.

# Visioning and Strategic Planning

In its simplest form, strategic planning begins with a (desirable) vision for the future, setting goals, outlining an approach, and designing a course of action to achieve that vision. In the urban and transportation planning context, visioning is the act of involving the public and other stakeholders in crafting a collectively agreed vision for the future.

In turn, a strategic plan must complement the vision so that the goals, approach, and course of action evolve in anticipation of and in response to those emergent forces. Strategic planning is suitable in the context of metropolitan planning, where the MPO wields considerable influence over outcomes through the levers of funding, prioritization, and zoning. On the contrary, in the CAV context, the MPO is one of many actors playing a part in shaping the future with CAVs, including sensing and communication manufacturers, standard-setting professional organizations like SAE International, regulatory bodies like the National Highway Traffic Safety Administration (NHTSA), and the public that ultimately will or will not accept or embrace a technology. These are the sources of deep uncertainty.

That being said, there may well be merit to choosing an approach to this uncertainty based on visioning and strategic planning if the MPO is able to leverage the cooperation and coordination of the aforementioned stakeholders toward achieving a desired vision for the future. Additionally, if by proving through credible microsimulation modeling that one or more CAV concepts, strategies, or technologies can deliver compelling benefits (e.g. reduced delay), then it may be possible to incentivize the investment and research that will realize those concepts, strategies, or technologies.

Whether strategic planning is the selected approach, there are practices from strategic planning that should be a part of any approach to CAV uncertainty. These include:

  • Collecting data: when data become available from published CAV studies or research, assimilate that data into existing models to improve their validity
  • Staying current: update assumptions as new information about CAV operations emerges
  • Forecast based on trends: observe trends in collected data and the direction of change in updated assumptions to refine future assumptions
  • Model frequently: as assumptions and models improve, run model(s) of the future, adjust approach and action plans accordingly to those outcomes that appear more likely

If stakeholders are continually involved and invested in the vision and in the strategic plan through this process, the MPO will be as well-positioned as it can be to manage and reduce uncertainty and to make better-informed decisions.

# Scenario Planning

Scenario planning has been in the transportation planning toolbox for many decades, but is recently enjoying a resurgence due to the significant uncertainty in the delivery of new technologies. Scenario planning provides a structured forecasting process where multiple scenarios (ranging from a few to a few hundred) are evaluated to understand the range of possible outcomes.

Scenario planning is probably the most prevalent approach to planning under uncertainty and so is the most easily adopted and adapted in the CAV context. Scenario planning involves making assumptions about those variables whose values are uncertain and modeling a variety of future conditions to better understand what outcomes are possible. Hence, in contrast with visioning and strategic planning, scenario planning more explicitly acknowledges the sources of uncertainty outside the planning organization’s control and prepares the organization for a multitude of possible outcomes. This approach is particularly attractive in the CAV context because, in the presence of deep uncertainty, scenario planning recognizes that the future cannot be known, and so preparation for one of multiple possible futures is prudent. By modeling a range of plausible futures and considering their outcomes and the implications and sensitivities of various assumptions, an organization is more likely to be prepared for the future that does emerge.

Constructing a manageable scenario planning approach hinges on making informed judgments about which variables are critical – i.e. that have the most influence on the outcomes – and what set of assumptions about those variables are reasonable and plausible. Better assumptions will be made through effective involvement from stakeholders most knowledgeable about the source of the uncertainty, for instance manufacturers of CAV technology.

Strategic modeling tools were developed to directly support scenario planning. These tools can be developed and applied quickly, focusing instead on the characteristics of the population and firms, the interaction between these agents and their responses to policies. They are limited in spatial and network detail, supporting the evaluation of the impacts to demand, rather than supply.

Applications

FDOT took a probabilistic approach to scenario planning by quantifying probabilities in variables of uncertain value at a later stage, in keeping with the strategic planning practices of collecting data and staying current.

Another notable application of scenario planning was done by ICF for FHWA in the Twaddell 2018 presentation. That presentation includes useful discussion of what goes into creating scenarios including examples of external forces (e.g. technologies or the environment); levers including trade policy, tax incentives, government mandates, consumer preferences, social-economic factors (e.g. population, workforce trends, market forces); and desired outcomes or goals such as equitable access, reduced congestion or environmental sustainability.

Other state DOTs have taken a scenario planning approach to CAVs. The Iowa DOT used scenario planning in a study of the impacts of AVs and advanced technologies on I-80 – specifically smart truck parking along the interstate – and varying levels of AV adoption in different forecast years to evaluate different possible futures (Iowa DOT, 2017). Millennial traveler behavior and the aging population were also part of the assumptions in the scenario construction. Through scenario planning for AVs, the Iowa DOT arrived at a set of recommendations for infrastructure improvements on the corridor.

# Exploratory Modeling and Analysis

EMA has seen increasing interest across a multitude of disciplines over the past decade as an alternative to scenario planning where deep uncertainty is present in the system under study. One of the primary challenges in scenario planning is the difficulty in deciding a manageable number of scenarios that captures the full breadth of uncertainty. According to the Society for Decision Making under Deep Uncertainty (DMDU), deep uncertainty exists where there is no clear consensus among stakeholders about:

  • the structure of the model that relates inputs and assumptions to outcomes,
  • the probability distributions of the system variables about which the parties are uncertain, or
  • which system behaviors are most important

This is an apt description of the current stage of evolution of CAVs, where there is no clear agreement which CAV concepts, strategies, or technologies will be realized; how quickly or whether they will be adopted; or how they will operate in the field if or when they are adopted. Indeed, the literature research summarized in this report confirms that the future with CAVs amounts to little more than speculation as to the benefits, with the consensus surrounding only the inevitability of CAV technology in some form or fashion. (DMDU Society, 2018)

EMA seeks to structure the approach to scenario planning in a systematic way that uses sensitivity analysis to explore patterns in model results to reduce the range and number of asserted model input values and scenarios (Bankes, 1993). Whereas scenario planning can produce a picture of a relatively small number of possible futures, it does not necessarily illuminate the relationship between the different assumptions and possible futures. EMA seeks to uncover the patterns or relationships in the system so as to provide more guidance to decision-makers about how their decisions might shape the future.

There is recent precedence of EMA in planning, specifically in relation to CAV uncertainty. An FHWA study used an integrated dynamic traffic assignment (DTA) model and activity-based model (ABM) to explore the relationships between AV adoption, traveler behavior, and the operational benefits of AVs and served as a demonstration of a framework for using EMA in regional transportation planning (Stabler, Bradley, Morgan, Slavin, & Haque, 2018).

# References

CDM Smith (2019). CAV Traffic Simulation Literature Review

Bankes, S. (1993, May-June). Exploratory Modeling for Policy Analysis. Operations Research, 41(3), 435-449.

DMDU Society. (2018). About Us. Retrieved from DMDU Society: http://www.deepuncertainty.org/about-us/

Iowa DOT. (2017). Interstate 80 Planning Study (PEL). Office of Location and Environment, Automated Corridors. Iowa DOT.

Minowitz, A. (2013). Visioning In Urban Planning- A Critical Review and Synthesis. A Thesis Presented in Partial Fulfillment of the Requirements for the Degree Master of Urban and Environmental Planning, Arizona State University.

Stabler, B., Bradley, M., Morgan, D., Slavin, H., & Haque, K. (2018). Volume 2: Model Impacts of Connected and Autonomous/Automated Vehicles (CAVs) and Ride-Hailing with an Activity-Based Model (ABM) and Dynamic Traffic Assignment (DTA) - An Experiment. Washington, DC: Federal Highway Administration.

Zegras, C., Sussman, J., & Conklin, C. (2004, March). Scenario Planning for Strategic Regional Transportation Planning. Journal of Urban Planning and Development, 130(1), 2-13.

Content Charrette: Autonomous Vehicles