Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles, Volume 2: Guidance (2018)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

38 This chapter presents critical considerations related to accounting for CAVs in trip-based models. Along with the subsequent chapters on disaggregate/dynamic models and strategic models, Chapter 6 pro- vides the context for the modeling adaptations and then an approach for implementing them. These chapters do not provide prescriptive rules for including CAVs in models but, rather, ways to use existing models for quantitative visualization of feasible alternative outcomes. Overview Trip-based models are long-range travel demand models that follow the conventional four-step process of trip generation, trip distribution, mode choice, and traffic assignment. Additional steps, feedback loops, and postprocessing often enhance trip-based models. These models have been calibrated, validated, and tested throughout the world, and they are used extensively across most MPOs and state DOTs in the United States. Table 3 summarizes potential changes to the trip-based modeling system from CAV impacts. Successfully modeling CAVs will require several changes to existing modeling processes, including • New modes or submodes: – CAVs, – SAVs, and – SAV access to transit (submode); • Additional submodels: – Auto availability models that reflect the level of market penetration of CAVs and – Market penetration models to determine fleet composition changes over time; • New algorithms and processes: – Routing routines to model dynamic ridesharing (e.g., uberPOOL), – Coordinated multimodal mobility services modeling (e.g., MaaS; automated tour plan- ning), and – Network flow coordination (real-time speed governing and predicted arrival rates); and • New supply models to reflect CAV impacts on roadway space. Applying Exploratory Models in the Context of Stable Travel Behavior When models are being applied in an exploratory manner, it is important for modeling ana- lysts to note the stability of trip rates and lengths in the United States. Adjustment of parameters T3 C H A P T E R 6 Adapting Trip-Based Models to Address CAVs Chapter Highlights • Provides high-level guidance on accounting for CAVs in trip-based models. • Identifies potential modeling changes. • Discusses the context and approaches for – Land use modeling, – Auto availability and mobility choices, – Trip generation, – Trip distribution, – Mode choice, and – Routing and traffic assignment.

Adapting Trip-Based Models to Address CAVs 39 such as trip rates and lengths, per person, needs to be done in the context of reasonableness as compared with historical trends. While trip rates and lengths may vary because of an urban area’s characteristics and size, as a whole, trip making and the time spent in mobility activities is relatively stable. Table 4 shows the change in travel statistics in the National Household Travel Survey between 1969 and 2009. While trip making (person trips per person per day) more than doubled between 1969 and 1995—most likely from women entering the workforce—the two most recent surveys show similar rates of trip generation per person (3.74 and 3.79 in 2001 and 2009, respectively). Average person and vehicle trip length per trip also shows relative stability, in miles, although travel times have most likely increased as urban areas have become more congested and travel times are more unreliable because of congested conditions. Table 5 summarizes travel trend excerpts from several American Time Use Surveys conducted annually by the Bureau of Labor Statistics (n.d.). Total time spent traveling for various activities, on average, varied by less than 3.6 minutes per day from 2003 to 2016. Other interim years of 2010 and 2015 are also shown and display little variation. Because travel activity on average has remained relatively stable over time, particularly in the recent decade, modeling analysts must use reasonable judgment when adjusting travel behavior parameters in travel demand models. Model Component Trip-Based Model Improvement Sociodemographics Land use/demographic model Adjust accessibility measures Land use/demographic model Account for parking reuse Land use/demographic model Estimate levels of expanded mobile populations Market/Fleet Fleet composition models Estimate and forecast types of vehicles and technology Auto Ownership Auto ownership model Estimate and forecast CAV or manual vehicle ownership Auto availability model Estimate and forecast availability of SAVs and carsharing Trip Generation Trip rates Estimate and forecast rates for expanded mobile populations Trip rates Account for zero-occupant vehicle trip generation Trip rates Adjust rates within reason for improved accessibility Trip Distribution Impedance to travel Estimate network cost matrices reflecting CAVs Impedance to travel Estimate new friction factor matrices if CAVs affect trip lengths Mode Choice Mode choice model Design new nesting structure including CAVs, SAVs, and SAV access to transit Mode choice model Account for MaaS impacts on multimodal tour plans Value of time Account for improved value of time for CAV modes Network Assignment Supply models Estimate CAV-enhanced capacity on signalized arterial systems Network capacity Estimate CAV-enhanced capacity on grade-separated facilities Path costs; pricing and tolling Estimate value of time including discounts for CAV passengers Table 3. Potential trip-based modeling changes.

40 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles Travel Statistic 1969 1977 1983 1990 1995 2001 2009 95% CI Per person Daily person trips 2.02 2.92 2.89 3.76 4.3 3.74 3.79 0.03 Daily PMT 19.51 25.95 25.05 34.91 38.67 36.89 36.13 1.35 Per driver Daily vehicle trips 2.32 2.34 2.36 3.26 3.57 3.35 3.02 0.03 Daily VMT 20.64 19.49 18.68 28.49 32.14 32.73 28.97 0.71 Per household Daily person trips 6.36 7.69 7.2 8.94 10.49 9.66 9.5 0.09 Daily PMT 61.55 68.27 62.47 83.06 94.41 95.24 90.42 3.38 Daily vehicle trips 3.83 3.95 4.07 5.69 6.36 5.95 5.66 0.06 Daily VMT 34.01 32.97 32.16 49.76 57.25 58.05 54.38 1.34 Per trip Average person trip length (miles) 9.67 8.87 8.68 9.47 9.13 10.04 9.75 0.36 Average vehicle trip length (miles) 8.89 8.34 7.9 8.85 9.06 9.87 9.72 0.22 Source: Santos et al. 2011 (https://nhts.ornl.gov/2009/pub/stt.pdf). Note: CI = confidence interval. Table 4. Summary of travel statistics, National Household Travel Survey, 1969–2009. Activity 2016 2015 2010 2003 Travel related to personal care 1.8 1.2 1.2 0.6 Travel related to eating and drinking 6.6 6.6 7.2 7.2 Travel related to household activities 3 3 2.4 2.4 Travel related to purchasing goods and services 17.4 16.8 16.2 17.4 Travel related to caring for and helping household members 4.8 4.8 4.8 5.4 Travel related to caring for and helping nonhousehold members 3.6 3.6 3.6 5.4 Travel related to work 16.2 16.2 16.8 17.4 Travel related to education 1.8 1.8 1.8 2.4 Travel related to organizational, civic, and religious activities 2.4 2.4 2.4 2.4 Travel related to leisure and sports 12.6 12.6 13.2 13.8 Travel related to telephone calls 0.6 0 0 0 Total 70.8 69 69.6 74.4 Source: Bureau of Labor Statistics. Table 5. Travel minutes of activity from American Time Use Survey.

Adapting Trip-Based Models to Address CAVs 41 Land Use Modeling Context A land use/demographic allocation process is usually added prior to running the four-step trip-based modeling process. Linkage to the travel demand process, if it is done, uses measures of accessibility derived from the trip tables and roadway/transit networks. CAVs would present changes to accessibility and affect land use modeling in this way. In transportation modeling, land use modeling may be a misnomer for this important step in the process. Land use usually refers to the type of activity occurring or allowable, while typical travel demand models use socioeconomic data directly. Population, households by size, income, and other categories, and employment by category represent the level of activity in a location. These inputs, aggregated to traffic analysis zones (TAZs), become the independent variables in the trip generation step. Land use models and methods vary considerably between U.S. planning agencies. In many planning regions, ad hoc methods ranging from local stakeholder consensus meetings to formal scenario generation processes are used. Generalized growth patterns in residential and employ- ment locations are gleaned from workshops and converted to rational allocation of growth based on densities. Quantitative methods for forecasting the level of activity in TAZs are gaining in application. Aggregate quantitative methods are based on rational densities and predeter- mined allowable or desired uses, usually from a municipal comprehensive plan for future land use. Aggregate allocation of activity to TAZs may also use accessibility calculated from a travel demand model process as an indicator of readiness for a parcel to be developed. Disaggregate residential and employment models are applied in many larger metropolitan regions. Bid rent models that simulate competitive bidding for land among residents and employers are applied by using discrete choice models. These models also use transportation accessibility from travel demand modeling processes to represent valuation (i.e., cost) of land. Approach Accessibility is a measure that uses the aggregate relative distance, time, or cost of a TAZ to separate it from other TAZs. Accessibility could change after widespread introduction of CAVs to the transportation system. Accessibility is a common parameter used in both aggregate and disaggregate land use models to represent the relative cost of travel from residence to work or other activities. As the relative accessibility of a parcel or TAZ of land increases, it becomes more desirable for development. Measures of accessibility are calculated from existing and forecast distribution of trips or other activity at each TAZ combined with network costs. CAVs will potentially have significant impacts on travel costs and therefore would change the input accessibilities in most quantitative land use models. Travel costs can be categorized into personal travel time costs and vehicle operating costs. Vehicle operating may be affected by CAVs through depreciation, changes in insurance costs, changes in vehicle technology, and behavioral shifts toward vehicle sharing. Insur- ance costs for CAVs may decline because of a reduction in crashes. If CAV fleet service providers move toward electrification, both fuel and maintenance costs may decrease. Additionally, cost sharing from shared vehicle usage would divide operating costs, including tolls, among riders. Finally, under a shared usage CAV scenario, the capital cost of vehicles would decline significantly because CAVs may be used for many more hours compared with the typical privately owned vehicles that sit idle up to 23 hours per day. Travel demand model network shortest-path procedures are used to determine trip cost and travel time between TAZs. Composite measures that include both auto and transit costs are also CAVs will potentially have significant impacts on travel costs and therefore would change the input accessibilities in most quantitative land use models.

42 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles used (logsums from the mode choice step). These times and costs, combined with the level of population and/or employment in each TAZ, are used as inputs to land use allocation models. Often, land uses are part of comprehensive municipal city plans, in which land use is tied to zoning ordinances. These city plans usually indicate the expected land use in all parcels within the extraterritorial jurisdiction of the city. These can be a guide to the location of future popula- tion and employment and can be an input to a land use allocation model used as part of a travel demand modeling process. Factors to Consider for Land Use Modeling and CAVs Factors to consider when forecasting land use and demographic allocation that includes CAV scenarios include • Geographic distribution of growth—densification and/or urban sprawl, • Reuse of land formerly dedicated to parking, and • Demographic changes (aging and household composition). Changes in relative accessibility between TAZs derived from changes in transportation cost and travel time owing to CAVs may result in either densification or sprawl or a combination of both. The relief of the driving task may change the value of time for drivers commuting to work or for other trip purposes. Time spent driving can be viewed as wasteful. Drivers would be converted to passengers with time to perform other activities such as sleeping, reading, working remotely, or engaging in social activity. At a minimum, drivers would need to retain the function of piloting the CAV and monitoring systems, even at Level 4 technologies. Modelers may need to estimate and locate the amount of space dedicated to parking and forecast the conversion of the space to other uses or to open space. In the early years of CAV adoption, parking will remain a needed land use. In the longer term, land needed for parking may decline. Analysts will be able to model the trends in the reuse of parking facilities over time on the basis of observations, but in early years, scenarios of changes will need to be envisioned. Modelers will also need to account for more detailed demographic characteristics to model the potential increase in availability of CAVs to population segments not currently considered for trip-based models. CAVs may provide independent mobility for younger people not cur- rently eligible to legally operate a vehicle, probably between ages 12 and 18, through ridesharing and vehicle sharing. Levels of elderly and mobility-limited persons will need to be forecast in TAZs as ride and auto availability increase from the introduction of CAV technology. House- hold composition, including age categories and mobility limitations, will be important in the estimation of trip making, so these variables will need to be included in demographic forecast- ing models. Alternate Work Locations Alternate work locations are places where business can be conducted outside of home resi- dences but are not a usual place of work. These locations include formal alternate work locations such as branch business offices, informal offices where meeting rooms can be leased, or even coffee shops. Future peer-to-peer sharing of commercial office space may also surface as signifi- cant alternate work locations for many commuters, much in the way that peer-to-peer residence sharing has entered the temporary and vacation housing marketplace. If trends point to alternate work locations becoming more prevalent, or analysts wish to create exploratory scenarios, including increased sharing of commercial office space, trip lengths for work trips may decline in the future. This trend would have an impact on trip generation. More

Adapting Trip-Based Models to Address CAVs 43 trips may be generated with reduced trip lengths. The ability to work closer to home and other daily activities may spawn increased work trip rates as the alternative workplace option competes with telecommuting. Efficiencies created by CAVs may enhance or take from these options. A reduced impedance to travel created by CAVs could compete with alternate work locations, just as it would compete with the choice to remain working at home. Gaps in Current Land Use Models Regions will need to gather input for addressing long-range considerations for land use mod- eling with impacts from CAVs. Forecasting based on current trends in the growth of land use patterns may lead to omissions about changes in residential and commercial location decisions. Growth should be assessed in at least four scenarios: a continuation of past trends, more densifi- cation than current trends, more significant sprawl to exurban areas, and a combination of infill and sprawl growth in focused geographic subareas of a metropolitan region. Efficiencies gained from CAVs and related technologies such as comprehensive mobility sourcing (i.e., MaaS) could lead to choices in residential location that fill in and densify urban cores. The market for urban living space could grow if mobility services create a quality of life that some residents may be searching for in place of long commutes, separation from enter- tainment options, relief from yard maintenance, and a desire for greater active transportation options. Researching and quantifying current land developable as infill is important when increased densification impacts from CAVs are being considered. Also important is quantifying the amount of potential infill development that could occur if parking facilities become avail- able for residential and commercial redevelopment. Studies of convertible parking space versus space that would require more costly demolition would further enable forecasters to quantify redevelopment properties for input to land use models. If operating costs decline as a result of CAV deployment, and the share of household budgets dedicated to mobility declines in parallel, the effect could be a greater share of household budgets dedicated to housing. This may make inner-city housing more affordable and create densification. When inputs for land use modeling scenarios are being designed, careful consideration should exercised in speculation about commuting time increasing as a result of the driving task being relieved by CAV. As shown in Table 5, the time spent in travel has on average remained stable for many years. Although CAVs may have an impact on activity that can be performed while in transit, time spent traveling will not necessarily increase. Relief of the driving task indicates that the in-vehicle time may be spent more productively or used for entertainment purposes. However, the strong desire to minimize travel time may remain a common behavioral attribute into the future. Auto Availability and Mobility Choices Context The availability of autos to household members is used in trip-based models to indicate a propensity for use of the vehicles because they are costly, and most people would want to use something that takes a large share of household budgets. Households with no owned vehicles are used as an indicator of a requirement for public transportation or ridesharing. Some models use a ratio of number of household members of driving age to the number of autos available for use in a household as a measure of these modal propensities. Auto ownership can also be used as a factor in calculating vehicle trip frequency. More often, household income has been used as a variable indicating that no impediment to trip generation

44 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles from the lack of availability of vehicles exists. Changes in ownership patterns under CAV sce- narios indicate the need for a different schema to address vehicle availability and its indication of modal preference and ease of trip generation. Approach to CAV Auto Availability Estimation Although auto ownership is the preferred parameter in many travel demand models, CAVs could potentially shift the current pattern of individual ownership of vehicles. If the cost per trip—or mile traveled—decreases with the introduction of efficiently shared CAVs, many people may move away from current auto ownership patterns. Schemas for household trans- portation choices resulting from CAVs could include the following: • A household has roughly one car per household member of driving age, continuing the cur- rent pattern. • A household keeps a mix of owned CAVs and non-CAVs or vehicles equipped with only Level 1 or 2 technologies. • A household is within an area that has high accessibility to shared CAV services, and vehicle sharing is common at the household income level or in the geographic location. Several combinations of auto availability can be designed around observed data as they become available. Assumptions about auto availability will need to be made when exploratory modeling scenarios are being designed. Table 6 is an example of an auto availability table that could be developed for scenarios or from observed data over time. Auto availability should become a common parameter to supple- ment auto ownership for most travel demand models that include CAV forecast scenarios. These types of household mobility classifications will need to be determined or theorized as part of scenario modeling to dissect the complexity of mobility options that will become available in the future. Mobility as a Service The objective for measuring auto availability for trip-based models is to reflect the propensity for trip making and modal choice with higher levels of vehicle availability. CAVs could also be implemented in a MaaS transportation environment. With MaaS, the cost of transportation could be paid through a single system, most likely wireless and automated, such that a combina- tion of trips into a daily pattern could be planned and optimized to minimize cost and involve easy transfer between autos, public transportation, and bicycling or walking. A single transporta- tion card that pays for all modes seamlessly could be issued. Vehicle Ownership CAV Sharing Household Size Household Income Group CAV-Enabled Non-CAV Ridesharing Potential Vehicle-Sharing Potential 1 2 3 1 2 3 Low High Low High 1 Low Number or percentage of regional household totals High 2 Low High 3 Low High Table 6. Example CAV availability schema.

Adapting Trip-Based Models to Address CAVs 45 Trip Generation Context Trip generation is the process used in trip-based travel demand models to estimate and fore- cast the number of trip ends generated by residential or commercial activity in TAZs. Trip ends refer to each end of a single trip: the origin or the destination. In travel demand modeling, each end of a trip, before the ends are joined or linked together in the trip distribution step, is clas- sified as either a production or attraction trip end. Persons in households generate trips, and therefore production trip ends are calculated from the number of households in a TAZ. Attrac- tion trip ends are calculated from employment levels in TAZs with commercial activity—retail, office and other work locations, schools, and entertainment venues. Approach to Capturing CAV Effects on Trip Generation As shown in Table 5, total time spent traveling for different activities varied, on average, by less than 3.6 minutes per day between 2003 and 2016 in the United States. Because travel activity has remained relatively stable over time—particularly in the recent decade—modeling analysts must use reasonable judgment when adjusting travel behavior parameters in travel demand models. As Table 4 illustrated, trip rates per person and time spent in travel activity have also remained stable. Although CAVs may provide a more efficient and less costly mode of travel, analysts should be cognizant of the trend toward stabilizing and minimizing total per person time spent in travel activities. Overall, CAVs may change the need for travel, the efficiency with which it is performed, and the aggregate total travel by enabling trip making by populations whose mobility currently is limited. Telepresence and CAV Scenarios Several factors should be considered when modeling futures are being forecast, including the effects of telecommuting, alternate work locations, and other telepresence. These factors include the • Cost of travel, • Availability of vehicles or rides, • Efficiency and reliability of travel, • Quality of telepresence technology, and • Culture of workplaces. The greater efficiency, lower cost, and improved availability of travel under CAV scenarios creates an impetus to increase trip generation rates. The effects of the quality of telepresence technology and work- place culture can have either positive or negative impacts on the decision to telecommute or perform other activities through telepresence. CAVs may have an impact on cost, availability, and efficiency of travel— factors that commuters and others will weigh against the quality of tele- presence technology and workplace culture when making the decision to telecommute, teleshop, or physically travel. As CAV and tele presence technology develop in parallel, modelers will need to assess the cost, travel time, convenience, and other issues that will be used to compare the choice of satisfying work, shopping, and entertainment activities through mobility or telepresence. In a future CAV scenario in which efficiency of travel is enhanced and less costly, in terms of the share of household budget spent on transportation, and telepresence technology is The greater efficiency, lower cost, and improved availability of travel under CAV scenarios creates an impetus to increase trip generation.

46 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles technically cumbersome and work culture unsupportive, trip rates per worker could increase. The increase in trip making would be from fewer persons working, shopping, or completing other activities at home. Workers have many reasons for working at home, such as travel efficiency, reliability, and value of time. Other reasons may be less associated with transportation, such as tending to children or taking care of medical needs. Surveys and data will be needed to determine the reasons behind working from home so that modelers can create future scenarios of increased or decreased work trip generation. For instance, if telecommuting currently accounts for 5% to 15% of work trips on a typical day, analysts may modify trip rates under future CAV scenarios between those values by using either data-supported evidence of a trend or by creating an exploratory scenario. If alternate work locations are included in a scenario, work trip rates may increase as workers who used to work at home begin to access high-quality workspaces closer to their homes. However, analysis is needed to factor in other trip making if a time window becomes available due to the alternate work location. If a scenario includes increased working from home, trip rates should be reduced. Teleshopping is a trend that is becoming more prevalent. Studies of this trend and its impact on trip generation remain inconclusive, however. The stability of national trip rates shown in Table 4 indicates that trip making has not been in decline because of the increase in online shopping in the past decade. Nonetheless, some shopping activity is clearly being replaced by teleshopping, as indicated by increased revenues for online retailers and closing of many large retail store brick-and-mortar locations throughout the United States. Perhaps teleshopping is replacing many types of purchases, but trips are still being generated to satisfy other purchases. Trips may also be made to showrooms and then purchases made later online, effectively negat- ing the potential reduction in trip generation from teleshopping. Expanded Mobile Populations and CAVs CAVs may bring about robotic vehicles that can operate without a human driver or passen- gers. People whose mobility options are limited today may find greater mobility independence with CAVs. The ability to independently call for a vehicle and control the schedule at a lower cost than today’s demand-responsive public transportation may lead to increased trip generation by current mobility-limited populations. Modelers wishing to estimate the impact of mobility- limited populations can look to existing data from on-demand transit services. Agencies that provide these services usually collect detailed trip and user data. Analysts could create explor- atory scenarios in which a portion of these mobility-limited trips moves to CAV modes and modestly increases their overall trip frequency—trips that otherwise would have been satisfied through ridesharing with another household member or through public on-demand transporta- tion services. Similarly, populations currently under the legal driving age could create additional CAV trips. Travel currently satisfied by household carpooling (usually called serve-passenger in travel sur- veys) could be a source of data for analysts to use in determining the level of increase in trip generation from younger populations. School-age children may either ride school buses or have a parent give them a ride to and from school. Parents often transport a child to school daily, and the task is often split between two parents. Many trip-based models include a specific trip purpose for school trip generation. Most of these trips are generated as serve-passenger trips and become part of either school bus modes or are treated as household carpools. Analysts may wish to consider current travel surveys to deter- mine the proportion of these trips to reassign to CAV modes. In robo-taxi scenarios, school-aged

Adapting Trip-Based Models to Address CAVs 47 children may be able to call a CAV and be transported independently to and from school or other activities. These trips may add to vehicle trip generation. Zero-Occupant Vehicle Trip Generation CAV technology will bring about conditions in which completely autonomous vehicles (Level 5) may become prevalent as part of ridesharing fleets and owned CAVs. These vehicles can be called zero-occupant vehicles (ZOVs). Local bus transit trips could be satisfied with on-demand ZOV services that become single-occupant vehicle (SOV) or shared-ride vehicle trips. Analysts may wish to create exploratory scenarios to account for ZOV-generated VMT. ZOV trips could be promulgated by factoring trip ends in trip tables produced by trip-based models. Analysts would need to either measure or assume the proportion of trips made by ZOVs relative to other modes, such as owned vehicles or shared vehicle services (which could also be called initial-state ZOVs). The initial state of a ZOV at the beginning of each day would need to be established in the model design by assuming a geographic distribution of ZOVs, which would probably remain in a secure storage area until called upon by the next day’s travel activity. ZOVs would be connected to a centralized control system that would reposition the vehicle for optimal use for the next passenger call. This technology would learn from the previous day’s trip patterns and position ZOVs as efficiently as possible at the beginning of each day. A distribution of call wait times, possibly stratified by area type, could either be measured or assumed as part of exploratory modeling. In this way, estimates of ZOV trips could be made by area type, because denser parts of an urban region will most likely require a greater number of AVs, and market demand will require short response times for passenger pickup. Simulations of ZOVs, ridesharing, and dynamic en route ridesharing could be done as part of a research study and then summarized for aggregate trip-based models by area type. Trip Distribution Context Trip distribution in trip-based travel demand models is the process of joining production and attraction trip ends to form trips from an origin to a destination. Trip-based models do not link trips together into tours; instead, individual trip purposes are maintained independently but in proportion to each other. CAVs may affect trip distribution in significant ways. Improvements in automation and connectivity of vehicles will influence the trip distribution step by influencing impedance to travel in two main areas: (a) improved system operational efficiency and reliability and (b) better convenience and trip planning. Both factors would also affect trip cost. Automation will bring about operational improvements to vehicles and improve the effi- ciency and reliability of traffic flow. The result will be more reliably predicted travel times and reduced impedance to travel. Connectivity will allow route plans to be shared as travelers input their destinations into CAVs. Route plans can then be shared with traffic management centers that will be able to reliably predict demand at critical points in the network, such as traffic signals at major intersections and bridges across major geographic obstacles. V2I technology will then be able to adjust signal timings, and vehicles will be able to adjust approach speeds, creating harmonized, predictable arrival patterns. These types of advancements in CAVs will reduce impedance to travel. Impedance to travel is used as a separation component in gravity analogy trip distribution models that are commonly used as part of trip-based models. The separation between TAZs is measured by using shortest- path algorithms and modeling networks. Travel speeds in most trip-based models are gathered

48 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles from feedback loops from the traffic assignment step to reflect the effect of congestion resulting from inefficient networks. These inefficiencies could be reduced with CAVs. Approach Impedance to travel is usually measured in trip-based models by using travel time and, less fre- quently, travel distance. Some more-advanced trip-based models include composite measures of separation from mode choice modeling steps that include transit travel times, fares, and operating costs in one measure (logsums). Given that CAVs could be an additional mode, and that there could be various mixes of modal technologies (and pricing schema for various policy reasons) in shorter-term planning years, analysts may wish to consider using composite cost matrices. Another more specific method of reflecting the gains in network performance from CAVs would be to lower the weights applied to impedance in gravity models. These weights are known as “friction factors.” Friction factors are distributed generically across travel times and are strati- fied by trip purpose. Friction factors allow for the calibration of gravity models to observed pat- terns of trip distribution. The objective function for gravity calibration is trip length frequency distribution by trip purpose. If CAVs are expected to affect trip lengths—how far people are willing to travel for specific activities—then adjustment to friction factor curves to match pre- supposed trip length frequency distributions would be the method used to reflect the impact of CAVs on trip length. However, trip distribution and willingness to travel specific distances for work versus food shopping or other activities are predicated on a relatively stable pattern of land use. Grocery stores are dispersed among residential areas, while work locations are often more centralized and concentrated in urban cores. If the pattern of land use is expected to remain stable into the future, even with the widespread adoption of CAVs, then friction factors should be adjusted with caution. Mode Choice Context Mode choice is the process of determining the mode of travel for each person trip by trip pur- pose. Matrices of person trips by trip purpose are created in the previous step in the trip-based modeling process—trip distribution. Each origin-to-destination pair of TAZs is coded for connec- tivity to available modes; all TAZs are connected by auto, but transit service may not extend to all TAZs in a modeled region. Mode choice model parameters define three essential characteristics to compare the relative attractiveness of each mode for use: the characteristics of the user in a house- hold, characteristics of the trip itself, and characteristics associated with the destination TAZs. The most common mode choice modeling structure used in trip-based models is nested logit. In this design, options are nested logically into subdivisions. Each subdivided modal element has an associated disutility function that defines the probability of a user choosing that mode relative to the disutility of all other modes. Disutility is measured with many variables, and equations are estimated with multinomial regression against observed behavior. For exploratory model designs, analysts would need to presume coefficients and constants by using reasonable judg- ment drawn from other calibrated models. Approach For CAV modeling, mode choice is an important step. Figure 4 displays an example mode choice structure for CAVs. In this simple design, the primary nest of the choice model is travel

Adapting Trip-Based Models to Address CAVs 49 by either auto or transit. Within the transit nest, several access mode options exist, including walking, driving, park-and-ride, and drop-offs, and SAVs. On the auto side, vehicles are either owned or shared, as estimated by an auto availability modeling step. Shared vehicles could be further subdivided into rideshare vehicles or carshare club vehicles if there is a difference in the utility of those modes significant enough to warrant further nesting. The nesting structure in Figure 4 is but one example. Other structures may treat CAVs as an entirely new mode and as a third option to auto and transit. Because these are exploratory models until data become avail- able, the design of nesting structures is experimental. On the owned side, vehicles are divided into manually driven vehicles or CAVs (denoted as CAV in the chart). This division of owned vehicles is made because of the presumption that CAVs will have better operating characteristics and provide amenities that manually driven vehicles do not possess. Both classes of vehicles could be driven (or ridden in) or shared with other household members. Carpool formation could take place outside of the household as part of a ridesharing plan for those who choose to operate their vehicle as a rideshare vehicle, thereby lowering overall operating cost for each trip. For estimation of the disutility of each mode, several variables could be used. First, the char- acteristics of the chooser could be (a) the income level of the household, (b) whether the house- hold has any owned vehicles (manual or CAV equipped), and (c) whether the home location is in an area served by CAV fleet services (ridesharing through SAVs or carsharing through a CAV subscription service). The service level in an area served by CAV fleet services may indicate a variable for wait times. Further, the disutility of each mode may include characteristics of the trip. In a mixed vehicle environment of both CAVs and manually driven vehicles, the full benefit of CAV technology may not be fully realized if manually driven vehicles in the traffic stream degrade the potential efficiency gains from CAVs. However, exclusive lanes or geographic areas dedicated to efficient CAV operation may overcome these impediments. Travel time would be a fundamental param- eter in the disutility equations, and either a level-of-comfort variable or a binary variable could be estimated to indicate subjective qualities of CAV modes. Vehicles that are owned would still require parking at the destination end of each trip. As parking becomes scarcer and less needed in an urban CAV transportation environment, the cost for parking may rise, since land owners and managers could convert space formerly dedicated to parking to other, more valuable uses. The cost of parking may become a major factor of disutility Figure 4. Example of simplified CAV mode choice structure (PNR = park and ride; DA = drive alone; SR = shared ride).

50 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles for ownership of either CAVs or manually operated vehicles for trips with destinations in downtowns and dense areas of the city. If an environment of MaaS comes about in the future, modelers may want to consider the composite impact of multimodal tours rather than individual modal options for each trip in the mode choice step. A com- bination of modes may lower overall trip cost, ease the disutility associ- ated with transfers, and possibly be supplemented with incentives for use of multiple modes as public policy to reduce congestion. Modeling analysts may wish to estimate the impacts of MaaS and adjust penalties for modal transfers, parking cost, terminal time, or other parameters to reflect the ease of travel that MaaS brings to commuters and for other nonwork trip activities. For mode choice modeling and a future with CAV and other technologies, many options can be considered. Although the focus is primarily on efficiency and reliability of roadway perfor- mance to be enabled by CAVs, it is important for modelers to consider changes that may come in parking technology, route planning, MaaS, and the ride- and vehicle-sharing environment. Routing and Traffic Assignment Context Network assignment is the step in the process used to route trips by mode through networks. Network performance is then updated on the basis of the usage of each facility by comparing the loadings on each facility to its theoretical capacity. Trip-based models are usually applied by using a static user equilibrium (UE) traffic assignment model. Trips are loaded onto a network in aggregate (by TAZ) by time period or for an entire day. The initial state of the network, in terms of speed or cost or both, is updated by using the aggregate loadings of trips from and to each TAZ. The application is then run over many iterations until a predefined state of equilib- rium is reached—where no user can improve his or her travel time by changing routes—within a threshold parameter. As explained in greater detail in later sections of this report, regional dynamic traffic assign- ments (DTAs) are being tested for use with trip-based models to improve the limitations of static UE assignment methods. Static methods tend to overload facilities beyond available capacity, do not account for queuing and spill-back, and do not have sensitivity to intersection controls or other metering infrastructure such as ramp meters. DTA is used by taking aggregate loading from TAZs and breaking it into individual trips. The individual vehicles are then routed through a network in a simulation in which each vehicle (user) chooses an optimal path on the basis of narrow time slices of 15 seconds or less. Each link in the network is updated, and queues can be formed in response to signalized intersections and other controls. Approach to Modeling Network System Performance and CAVs Detailed information on system performance will be needed to model CAVs through a net- work accurately. Improved operational performance of CAVs is expected to be achieved from either automation or connectivity or both. Automation may result in closer headways, quicker response times, improved acceleration and deceleration profiles, and robotic controls that can use more narrow lanes. Connectivity may result in better formation of homogeneous platoons (those that have similar route plan profiles), and better coordination of signal timing in response to expected demand. Additionally, CAVs could improve network performance by optimizing departure times, speeds, and arrivals at signalized intersections or other constrained points in a If an environment of MaaS comes about in the future, modelers may want to consider the composite impact of multimodal tours.

Adapting Trip-Based Models to Address CAVs 51 network. Gains from reduced crash rates would improve reliability and trip plans. Finally, costs and pricing of portions of a network may be quickly tabulated and communicated to CAVs. Efficiency of fuel use, including electric charging rates at specific times of the day, could result in lowered operating costs. Reductions in crashes would lead to reduced insurance costs. Shared travel would spread trip cost among several users. These details will require a detailed traffic assignment method. Current static UE models may remain in use for limited applications, but enhanced DTA models will be more functional. Current DTA models will also need enhancements to take advantage of speed harmonization, platooning, and CAV operating characteristics. As technology advances, many options will surface to optimize the efficiency of network performance, requiring new supply model designs. Options for pricing transportation facilities or incentivizing use through discounting could become simple once vehicles are automated and con- nected. Users may be able to select from several options as they plan their daily activities, and pricing may incentivize efficiency in the transporta- tion system. Controlled Intersection Facilities CAV technology is being researched and developed to take advantage of robotic control of vehicle spacing. Tight headway combined with coordinated acceleration and deceleration has been shown in simulations to enhance throughput at intersections. On the arterial system, intersections have the greatest impact on throughput capacity in addi- tion to side friction and curb cuts for business access. CAVs can take advantage of traffic signal coordination by forming platoons with tight spacing and controlling speed. However, prob- lems exist with this theory when platoons need to dissolve and intersections are tightly spaced. Research is continuing to resolve some of these issues to gain the most capacity possible at signal- ized intersections. Random arrivals and excess demand remain an issue. Another aspect of CAVs that may be useful in the future is coordinated arrivals at signal- ized intersections. This may be accomplished from connectivity and speed control far upstream from signalized intersections along a planned route. Vehicles could arrive in expected platoons or groups, and dynamic signals could anticipate loadings. A key component of this concept is accurate route plan information retrieval into a centralized, automated management controller. Currently, traffic management centers do not have this technical capability. For models, these technologies hold the promise to increase capacity at signalized arterial inter- sections. In static models, factoring arterial link capacity is all that is needed. DTA models would need to be enhanced to include simulation of route plans and speed harmonization profiles. Also, capability of simulating dynamic signalization in response to expected demand would need to be added. Modeling CAVs on Free-Flow Facilities CAVs are expected to improve traffic flow and total throughput capacity on freeways. Gains in capacity have been shown in experiments and in simulations that use close headway spacing and coordinated speed control to form platoons. In a mixed traffic stream of CAVs and manu- ally operated vehicles, fewer gains in capacity are expected as platoons form and dissolve and weaving occurs. Some speculation about exclusive use of managed lanes for CAVs exists. Using a physical separation from manually operated vehicles could enable CAVs to take advantage As technology advances, many options will surface to optimize the efficiency of network performance, requiring new supply model design.

52 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles of capacity-enhancing operation. Traffic assignment modeling of capacity improvements from CAVs in a static UE assignment procedure requires relatively simple change to link capacity. To take advantage of multiple types of per-trip SAVs, pricing analysts will need to estimate trip matrices by class, possibly categorizing SAV services into premium and nonpremium price classes. Pricing Considerations Tolling and pricing a network, either for financing or congestion pricing, may become much more complex in a CAV transportation environment. Managed lanes are currently set up to charge tolls for all vehicles of the same class, such as SOV or HOV 2+, the same during a prede- termined time period. CAVs may enable a more dynamic and disaggregate pricing system that incentivizes the use of AVs and high-capacity vehicles. In a dynamic pricing scheme, vehicles using the lanes at the same time may be charged differently on the basis of a policy to incentivize vehicle, user, passenger, and other potential characteristics. As connectivity becomes the norm, the potential for creating multiple user classes of vehicles is created. SAVs with higher person capacity may be favored by policy makers to discourage single-occupant CAVs and reduce VMT. The occupancy of an SAV could be communicated to managed lane management systems, and each vehicle could be charged appropriately regardless of the time of day. Prices could be set dynamically in relation to congestion on the facility. Static modeling of these types of systems can be performed with multiple class assignment techniques already available in some software packages. Further development of software that can model dynamic pricing schema is needed.